1feeb2ccf5a187f2ce0f43cfefc16e99db05f753
[sfrench/cifs-2.6.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/module.h>
10 #include <linux/device-mapper.h>
11 #include <linux/dm-io.h>
12 #include <linux/vmalloc.h>
13 #include <linux/sort.h>
14 #include <linux/rbtree.h>
15 #include <linux/delay.h>
16 #include <linux/random.h>
17 #include <crypto/hash.h>
18 #include <crypto/skcipher.h>
19 #include <linux/async_tx.h>
20 #include "dm-bufio.h"
21
22 #define DM_MSG_PREFIX "integrity"
23
24 #define DEFAULT_INTERLEAVE_SECTORS      32768
25 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
26 #define DEFAULT_BUFFER_SECTORS          128
27 #define DEFAULT_JOURNAL_WATERMARK       50
28 #define DEFAULT_SYNC_MSEC               10000
29 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
30 #define MIN_LOG2_INTERLEAVE_SECTORS     3
31 #define MAX_LOG2_INTERLEAVE_SECTORS     31
32 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
33
34 /*
35  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
36  * so it should not be enabled in the official kernel
37  */
38 //#define DEBUG_PRINT
39 //#define INTERNAL_VERIFY
40
41 /*
42  * On disk structures
43  */
44
45 #define SB_MAGIC                        "integrt"
46 #define SB_VERSION                      1
47 #define SB_SECTORS                      8
48 #define MAX_SECTORS_PER_BLOCK           8
49
50 struct superblock {
51         __u8 magic[8];
52         __u8 version;
53         __u8 log2_interleave_sectors;
54         __u16 integrity_tag_size;
55         __u32 journal_sections;
56         __u64 provided_data_sectors;    /* userspace uses this value */
57         __u32 flags;
58         __u8 log2_sectors_per_block;
59 };
60
61 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
62
63 #define JOURNAL_ENTRY_ROUNDUP           8
64
65 typedef __u64 commit_id_t;
66 #define JOURNAL_MAC_PER_SECTOR          8
67
68 struct journal_entry {
69         union {
70                 struct {
71                         __u32 sector_lo;
72                         __u32 sector_hi;
73                 } s;
74                 __u64 sector;
75         } u;
76         commit_id_t last_bytes[0];
77         /* __u8 tag[0]; */
78 };
79
80 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
81
82 #if BITS_PER_LONG == 64
83 #define journal_entry_set_sector(je, x)         do { smp_wmb(); ACCESS_ONCE((je)->u.sector) = cpu_to_le64(x); } while (0)
84 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
85 #elif defined(CONFIG_LBDAF)
86 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); ACCESS_ONCE((je)->u.s.sector_hi) = cpu_to_le32((x) >> 32); } while (0)
87 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
88 #else
89 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); ACCESS_ONCE((je)->u.s.sector_hi) = cpu_to_le32(0); } while (0)
90 #define journal_entry_get_sector(je)            le32_to_cpu((je)->u.s.sector_lo)
91 #endif
92 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
93 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
94 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
95 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
96
97 #define JOURNAL_BLOCK_SECTORS           8
98 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
99 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
100
101 struct journal_sector {
102         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
103         __u8 mac[JOURNAL_MAC_PER_SECTOR];
104         commit_id_t commit_id;
105 };
106
107 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
108
109 #define METADATA_PADDING_SECTORS        8
110
111 #define N_COMMIT_IDS                    4
112
113 static unsigned char prev_commit_seq(unsigned char seq)
114 {
115         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
116 }
117
118 static unsigned char next_commit_seq(unsigned char seq)
119 {
120         return (seq + 1) % N_COMMIT_IDS;
121 }
122
123 /*
124  * In-memory structures
125  */
126
127 struct journal_node {
128         struct rb_node node;
129         sector_t sector;
130 };
131
132 struct alg_spec {
133         char *alg_string;
134         char *key_string;
135         __u8 *key;
136         unsigned key_size;
137 };
138
139 struct dm_integrity_c {
140         struct dm_dev *dev;
141         unsigned tag_size;
142         __s8 log2_tag_size;
143         sector_t start;
144         mempool_t *journal_io_mempool;
145         struct dm_io_client *io;
146         struct dm_bufio_client *bufio;
147         struct workqueue_struct *metadata_wq;
148         struct superblock *sb;
149         unsigned journal_pages;
150         struct page_list *journal;
151         struct page_list *journal_io;
152         struct page_list *journal_xor;
153
154         struct crypto_skcipher *journal_crypt;
155         struct scatterlist **journal_scatterlist;
156         struct scatterlist **journal_io_scatterlist;
157         struct skcipher_request **sk_requests;
158
159         struct crypto_shash *journal_mac;
160
161         struct journal_node *journal_tree;
162         struct rb_root journal_tree_root;
163
164         sector_t provided_data_sectors;
165
166         unsigned short journal_entry_size;
167         unsigned char journal_entries_per_sector;
168         unsigned char journal_section_entries;
169         unsigned short journal_section_sectors;
170         unsigned journal_sections;
171         unsigned journal_entries;
172         sector_t device_sectors;
173         unsigned initial_sectors;
174         unsigned metadata_run;
175         __s8 log2_metadata_run;
176         __u8 log2_buffer_sectors;
177         __u8 sectors_per_block;
178
179         unsigned char mode;
180         bool suspending;
181
182         int failed;
183
184         struct crypto_shash *internal_hash;
185
186         /* these variables are locked with endio_wait.lock */
187         struct rb_root in_progress;
188         wait_queue_head_t endio_wait;
189         struct workqueue_struct *wait_wq;
190
191         unsigned char commit_seq;
192         commit_id_t commit_ids[N_COMMIT_IDS];
193
194         unsigned committed_section;
195         unsigned n_committed_sections;
196
197         unsigned uncommitted_section;
198         unsigned n_uncommitted_sections;
199
200         unsigned free_section;
201         unsigned char free_section_entry;
202         unsigned free_sectors;
203
204         unsigned free_sectors_threshold;
205
206         struct workqueue_struct *commit_wq;
207         struct work_struct commit_work;
208
209         struct workqueue_struct *writer_wq;
210         struct work_struct writer_work;
211
212         struct bio_list flush_bio_list;
213
214         unsigned long autocommit_jiffies;
215         struct timer_list autocommit_timer;
216         unsigned autocommit_msec;
217
218         wait_queue_head_t copy_to_journal_wait;
219
220         struct completion crypto_backoff;
221
222         bool journal_uptodate;
223         bool just_formatted;
224
225         struct alg_spec internal_hash_alg;
226         struct alg_spec journal_crypt_alg;
227         struct alg_spec journal_mac_alg;
228 };
229
230 struct dm_integrity_range {
231         sector_t logical_sector;
232         unsigned n_sectors;
233         struct rb_node node;
234 };
235
236 struct dm_integrity_io {
237         struct work_struct work;
238
239         struct dm_integrity_c *ic;
240         bool write;
241         bool fua;
242
243         struct dm_integrity_range range;
244
245         sector_t metadata_block;
246         unsigned metadata_offset;
247
248         atomic_t in_flight;
249         int bi_error;
250
251         struct completion *completion;
252
253         struct block_device *orig_bi_bdev;
254         bio_end_io_t *orig_bi_end_io;
255         struct bio_integrity_payload *orig_bi_integrity;
256         struct bvec_iter orig_bi_iter;
257 };
258
259 struct journal_completion {
260         struct dm_integrity_c *ic;
261         atomic_t in_flight;
262         struct completion comp;
263 };
264
265 struct journal_io {
266         struct dm_integrity_range range;
267         struct journal_completion *comp;
268 };
269
270 static struct kmem_cache *journal_io_cache;
271
272 #define JOURNAL_IO_MEMPOOL      32
273
274 #ifdef DEBUG_PRINT
275 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
276 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
277 {
278         va_list args;
279         va_start(args, msg);
280         vprintk(msg, args);
281         va_end(args);
282         if (len)
283                 pr_cont(":");
284         while (len) {
285                 pr_cont(" %02x", *bytes);
286                 bytes++;
287                 len--;
288         }
289         pr_cont("\n");
290 }
291 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
292 #else
293 #define DEBUG_print(x, ...)                     do { } while (0)
294 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
295 #endif
296
297 /*
298  * DM Integrity profile, protection is performed layer above (dm-crypt)
299  */
300 static struct blk_integrity_profile dm_integrity_profile = {
301         .name                   = "DM-DIF-EXT-TAG",
302         .generate_fn            = NULL,
303         .verify_fn              = NULL,
304 };
305
306 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
307 static void integrity_bio_wait(struct work_struct *w);
308 static void dm_integrity_dtr(struct dm_target *ti);
309
310 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
311 {
312         if (!cmpxchg(&ic->failed, 0, err))
313                 DMERR("Error on %s: %d", msg, err);
314 }
315
316 static int dm_integrity_failed(struct dm_integrity_c *ic)
317 {
318         return ACCESS_ONCE(ic->failed);
319 }
320
321 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
322                                           unsigned j, unsigned char seq)
323 {
324         /*
325          * Xor the number with section and sector, so that if a piece of
326          * journal is written at wrong place, it is detected.
327          */
328         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
329 }
330
331 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
332                                 sector_t *area, sector_t *offset)
333 {
334         __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
335
336         *area = data_sector >> log2_interleave_sectors;
337         *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
338 }
339
340 #define sector_to_block(ic, n)                                          \
341 do {                                                                    \
342         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
343         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
344 } while (0)
345
346 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
347                                             sector_t offset, unsigned *metadata_offset)
348 {
349         __u64 ms;
350         unsigned mo;
351
352         ms = area << ic->sb->log2_interleave_sectors;
353         if (likely(ic->log2_metadata_run >= 0))
354                 ms += area << ic->log2_metadata_run;
355         else
356                 ms += area * ic->metadata_run;
357         ms >>= ic->log2_buffer_sectors;
358
359         sector_to_block(ic, offset);
360
361         if (likely(ic->log2_tag_size >= 0)) {
362                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
363                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
364         } else {
365                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
366                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
367         }
368         *metadata_offset = mo;
369         return ms;
370 }
371
372 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
373 {
374         sector_t result;
375
376         result = area << ic->sb->log2_interleave_sectors;
377         if (likely(ic->log2_metadata_run >= 0))
378                 result += (area + 1) << ic->log2_metadata_run;
379         else
380                 result += (area + 1) * ic->metadata_run;
381
382         result += (sector_t)ic->initial_sectors + offset;
383         return result;
384 }
385
386 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
387 {
388         if (unlikely(*sec_ptr >= ic->journal_sections))
389                 *sec_ptr -= ic->journal_sections;
390 }
391
392 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
393 {
394         struct dm_io_request io_req;
395         struct dm_io_region io_loc;
396
397         io_req.bi_op = op;
398         io_req.bi_op_flags = op_flags;
399         io_req.mem.type = DM_IO_KMEM;
400         io_req.mem.ptr.addr = ic->sb;
401         io_req.notify.fn = NULL;
402         io_req.client = ic->io;
403         io_loc.bdev = ic->dev->bdev;
404         io_loc.sector = ic->start;
405         io_loc.count = SB_SECTORS;
406
407         return dm_io(&io_req, 1, &io_loc, NULL);
408 }
409
410 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
411                                  bool e, const char *function)
412 {
413 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
414         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
415
416         if (unlikely(section >= ic->journal_sections) ||
417             unlikely(offset >= limit)) {
418                 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
419                         function, section, offset, ic->journal_sections, limit);
420                 BUG();
421         }
422 #endif
423 }
424
425 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
426                                unsigned *pl_index, unsigned *pl_offset)
427 {
428         unsigned sector;
429
430         access_journal_check(ic, section, offset, false, "page_list_location");
431
432         sector = section * ic->journal_section_sectors + offset;
433
434         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
435         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
436 }
437
438 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
439                                                unsigned section, unsigned offset, unsigned *n_sectors)
440 {
441         unsigned pl_index, pl_offset;
442         char *va;
443
444         page_list_location(ic, section, offset, &pl_index, &pl_offset);
445
446         if (n_sectors)
447                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
448
449         va = lowmem_page_address(pl[pl_index].page);
450
451         return (struct journal_sector *)(va + pl_offset);
452 }
453
454 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
455 {
456         return access_page_list(ic, ic->journal, section, offset, NULL);
457 }
458
459 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
460 {
461         unsigned rel_sector, offset;
462         struct journal_sector *js;
463
464         access_journal_check(ic, section, n, true, "access_journal_entry");
465
466         rel_sector = n % JOURNAL_BLOCK_SECTORS;
467         offset = n / JOURNAL_BLOCK_SECTORS;
468
469         js = access_journal(ic, section, rel_sector);
470         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
471 }
472
473 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
474 {
475         n <<= ic->sb->log2_sectors_per_block;
476
477         n += JOURNAL_BLOCK_SECTORS;
478
479         access_journal_check(ic, section, n, false, "access_journal_data");
480
481         return access_journal(ic, section, n);
482 }
483
484 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
485 {
486         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
487         int r;
488         unsigned j, size;
489
490         desc->tfm = ic->journal_mac;
491         desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
492
493         r = crypto_shash_init(desc);
494         if (unlikely(r)) {
495                 dm_integrity_io_error(ic, "crypto_shash_init", r);
496                 goto err;
497         }
498
499         for (j = 0; j < ic->journal_section_entries; j++) {
500                 struct journal_entry *je = access_journal_entry(ic, section, j);
501                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
502                 if (unlikely(r)) {
503                         dm_integrity_io_error(ic, "crypto_shash_update", r);
504                         goto err;
505                 }
506         }
507
508         size = crypto_shash_digestsize(ic->journal_mac);
509
510         if (likely(size <= JOURNAL_MAC_SIZE)) {
511                 r = crypto_shash_final(desc, result);
512                 if (unlikely(r)) {
513                         dm_integrity_io_error(ic, "crypto_shash_final", r);
514                         goto err;
515                 }
516                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
517         } else {
518                 __u8 digest[size];
519                 r = crypto_shash_final(desc, digest);
520                 if (unlikely(r)) {
521                         dm_integrity_io_error(ic, "crypto_shash_final", r);
522                         goto err;
523                 }
524                 memcpy(result, digest, JOURNAL_MAC_SIZE);
525         }
526
527         return;
528 err:
529         memset(result, 0, JOURNAL_MAC_SIZE);
530 }
531
532 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
533 {
534         __u8 result[JOURNAL_MAC_SIZE];
535         unsigned j;
536
537         if (!ic->journal_mac)
538                 return;
539
540         section_mac(ic, section, result);
541
542         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
543                 struct journal_sector *js = access_journal(ic, section, j);
544
545                 if (likely(wr))
546                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
547                 else {
548                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
549                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
550                 }
551         }
552 }
553
554 static void complete_journal_op(void *context)
555 {
556         struct journal_completion *comp = context;
557         BUG_ON(!atomic_read(&comp->in_flight));
558         if (likely(atomic_dec_and_test(&comp->in_flight)))
559                 complete(&comp->comp);
560 }
561
562 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
563                         unsigned n_sections, struct journal_completion *comp)
564 {
565         struct async_submit_ctl submit;
566         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
567         unsigned pl_index, pl_offset, section_index;
568         struct page_list *source_pl, *target_pl;
569
570         if (likely(encrypt)) {
571                 source_pl = ic->journal;
572                 target_pl = ic->journal_io;
573         } else {
574                 source_pl = ic->journal_io;
575                 target_pl = ic->journal;
576         }
577
578         page_list_location(ic, section, 0, &pl_index, &pl_offset);
579
580         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
581
582         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
583
584         section_index = pl_index;
585
586         do {
587                 size_t this_step;
588                 struct page *src_pages[2];
589                 struct page *dst_page;
590
591                 while (unlikely(pl_index == section_index)) {
592                         unsigned dummy;
593                         if (likely(encrypt))
594                                 rw_section_mac(ic, section, true);
595                         section++;
596                         n_sections--;
597                         if (!n_sections)
598                                 break;
599                         page_list_location(ic, section, 0, &section_index, &dummy);
600                 }
601
602                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
603                 dst_page = target_pl[pl_index].page;
604                 src_pages[0] = source_pl[pl_index].page;
605                 src_pages[1] = ic->journal_xor[pl_index].page;
606
607                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
608
609                 pl_index++;
610                 pl_offset = 0;
611                 n_bytes -= this_step;
612         } while (n_bytes);
613
614         BUG_ON(n_sections);
615
616         async_tx_issue_pending_all();
617 }
618
619 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
620 {
621         struct journal_completion *comp = req->data;
622         if (unlikely(err)) {
623                 if (likely(err == -EINPROGRESS)) {
624                         complete(&comp->ic->crypto_backoff);
625                         return;
626                 }
627                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
628         }
629         complete_journal_op(comp);
630 }
631
632 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
633 {
634         int r;
635         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
636                                       complete_journal_encrypt, comp);
637         if (likely(encrypt))
638                 r = crypto_skcipher_encrypt(req);
639         else
640                 r = crypto_skcipher_decrypt(req);
641         if (likely(!r))
642                 return false;
643         if (likely(r == -EINPROGRESS))
644                 return true;
645         if (likely(r == -EBUSY)) {
646                 wait_for_completion(&comp->ic->crypto_backoff);
647                 reinit_completion(&comp->ic->crypto_backoff);
648                 return true;
649         }
650         dm_integrity_io_error(comp->ic, "encrypt", r);
651         return false;
652 }
653
654 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
655                           unsigned n_sections, struct journal_completion *comp)
656 {
657         struct scatterlist **source_sg;
658         struct scatterlist **target_sg;
659
660         atomic_add(2, &comp->in_flight);
661
662         if (likely(encrypt)) {
663                 source_sg = ic->journal_scatterlist;
664                 target_sg = ic->journal_io_scatterlist;
665         } else {
666                 source_sg = ic->journal_io_scatterlist;
667                 target_sg = ic->journal_scatterlist;
668         }
669
670         do {
671                 struct skcipher_request *req;
672                 unsigned ivsize;
673                 char *iv;
674
675                 if (likely(encrypt))
676                         rw_section_mac(ic, section, true);
677
678                 req = ic->sk_requests[section];
679                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
680                 iv = req->iv;
681
682                 memcpy(iv, iv + ivsize, ivsize);
683
684                 req->src = source_sg[section];
685                 req->dst = target_sg[section];
686
687                 if (unlikely(do_crypt(encrypt, req, comp)))
688                         atomic_inc(&comp->in_flight);
689
690                 section++;
691                 n_sections--;
692         } while (n_sections);
693
694         atomic_dec(&comp->in_flight);
695         complete_journal_op(comp);
696 }
697
698 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
699                             unsigned n_sections, struct journal_completion *comp)
700 {
701         if (ic->journal_xor)
702                 return xor_journal(ic, encrypt, section, n_sections, comp);
703         else
704                 return crypt_journal(ic, encrypt, section, n_sections, comp);
705 }
706
707 static void complete_journal_io(unsigned long error, void *context)
708 {
709         struct journal_completion *comp = context;
710         if (unlikely(error != 0))
711                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
712         complete_journal_op(comp);
713 }
714
715 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
716                        unsigned n_sections, struct journal_completion *comp)
717 {
718         struct dm_io_request io_req;
719         struct dm_io_region io_loc;
720         unsigned sector, n_sectors, pl_index, pl_offset;
721         int r;
722
723         if (unlikely(dm_integrity_failed(ic))) {
724                 if (comp)
725                         complete_journal_io(-1UL, comp);
726                 return;
727         }
728
729         sector = section * ic->journal_section_sectors;
730         n_sectors = n_sections * ic->journal_section_sectors;
731
732         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
733         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
734
735         io_req.bi_op = op;
736         io_req.bi_op_flags = op_flags;
737         io_req.mem.type = DM_IO_PAGE_LIST;
738         if (ic->journal_io)
739                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
740         else
741                 io_req.mem.ptr.pl = &ic->journal[pl_index];
742         io_req.mem.offset = pl_offset;
743         if (likely(comp != NULL)) {
744                 io_req.notify.fn = complete_journal_io;
745                 io_req.notify.context = comp;
746         } else {
747                 io_req.notify.fn = NULL;
748         }
749         io_req.client = ic->io;
750         io_loc.bdev = ic->dev->bdev;
751         io_loc.sector = ic->start + SB_SECTORS + sector;
752         io_loc.count = n_sectors;
753
754         r = dm_io(&io_req, 1, &io_loc, NULL);
755         if (unlikely(r)) {
756                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
757                 if (comp) {
758                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
759                         complete_journal_io(-1UL, comp);
760                 }
761         }
762 }
763
764 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
765 {
766         struct journal_completion io_comp;
767         struct journal_completion crypt_comp_1;
768         struct journal_completion crypt_comp_2;
769         unsigned i;
770
771         io_comp.ic = ic;
772         io_comp.comp = COMPLETION_INITIALIZER_ONSTACK(io_comp.comp);
773
774         if (commit_start + commit_sections <= ic->journal_sections) {
775                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
776                 if (ic->journal_io) {
777                         crypt_comp_1.ic = ic;
778                         crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
779                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
780                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
781                         wait_for_completion_io(&crypt_comp_1.comp);
782                 } else {
783                         for (i = 0; i < commit_sections; i++)
784                                 rw_section_mac(ic, commit_start + i, true);
785                 }
786                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, commit_sections, &io_comp);
787         } else {
788                 unsigned to_end;
789                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
790                 to_end = ic->journal_sections - commit_start;
791                 if (ic->journal_io) {
792                         crypt_comp_1.ic = ic;
793                         crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
794                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
795                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
796                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
797                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
798                                 crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
799                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
800                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
801                                 wait_for_completion_io(&crypt_comp_1.comp);
802                         } else {
803                                 crypt_comp_2.ic = ic;
804                                 crypt_comp_2.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_2.comp);
805                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
806                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
807                                 wait_for_completion_io(&crypt_comp_1.comp);
808                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
809                                 wait_for_completion_io(&crypt_comp_2.comp);
810                         }
811                 } else {
812                         for (i = 0; i < to_end; i++)
813                                 rw_section_mac(ic, commit_start + i, true);
814                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
815                         for (i = 0; i < commit_sections - to_end; i++)
816                                 rw_section_mac(ic, i, true);
817                 }
818                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
819         }
820
821         wait_for_completion_io(&io_comp.comp);
822 }
823
824 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
825                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
826 {
827         struct dm_io_request io_req;
828         struct dm_io_region io_loc;
829         int r;
830         unsigned sector, pl_index, pl_offset;
831
832         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
833
834         if (unlikely(dm_integrity_failed(ic))) {
835                 fn(-1UL, data);
836                 return;
837         }
838
839         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
840
841         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
842         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
843
844         io_req.bi_op = REQ_OP_WRITE;
845         io_req.bi_op_flags = 0;
846         io_req.mem.type = DM_IO_PAGE_LIST;
847         io_req.mem.ptr.pl = &ic->journal[pl_index];
848         io_req.mem.offset = pl_offset;
849         io_req.notify.fn = fn;
850         io_req.notify.context = data;
851         io_req.client = ic->io;
852         io_loc.bdev = ic->dev->bdev;
853         io_loc.sector = ic->start + target;
854         io_loc.count = n_sectors;
855
856         r = dm_io(&io_req, 1, &io_loc, NULL);
857         if (unlikely(r)) {
858                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
859                 fn(-1UL, data);
860         }
861 }
862
863 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
864 {
865         struct rb_node **n = &ic->in_progress.rb_node;
866         struct rb_node *parent;
867
868         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
869
870         parent = NULL;
871
872         while (*n) {
873                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
874
875                 parent = *n;
876                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
877                         n = &range->node.rb_left;
878                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
879                         n = &range->node.rb_right;
880                 } else {
881                         return false;
882                 }
883         }
884
885         rb_link_node(&new_range->node, parent, n);
886         rb_insert_color(&new_range->node, &ic->in_progress);
887
888         return true;
889 }
890
891 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
892 {
893         rb_erase(&range->node, &ic->in_progress);
894         wake_up_locked(&ic->endio_wait);
895 }
896
897 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
898 {
899         unsigned long flags;
900
901         spin_lock_irqsave(&ic->endio_wait.lock, flags);
902         remove_range_unlocked(ic, range);
903         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
904 }
905
906 static void init_journal_node(struct journal_node *node)
907 {
908         RB_CLEAR_NODE(&node->node);
909         node->sector = (sector_t)-1;
910 }
911
912 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
913 {
914         struct rb_node **link;
915         struct rb_node *parent;
916
917         node->sector = sector;
918         BUG_ON(!RB_EMPTY_NODE(&node->node));
919
920         link = &ic->journal_tree_root.rb_node;
921         parent = NULL;
922
923         while (*link) {
924                 struct journal_node *j;
925                 parent = *link;
926                 j = container_of(parent, struct journal_node, node);
927                 if (sector < j->sector)
928                         link = &j->node.rb_left;
929                 else
930                         link = &j->node.rb_right;
931         }
932
933         rb_link_node(&node->node, parent, link);
934         rb_insert_color(&node->node, &ic->journal_tree_root);
935 }
936
937 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
938 {
939         BUG_ON(RB_EMPTY_NODE(&node->node));
940         rb_erase(&node->node, &ic->journal_tree_root);
941         init_journal_node(node);
942 }
943
944 #define NOT_FOUND       (-1U)
945
946 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
947 {
948         struct rb_node *n = ic->journal_tree_root.rb_node;
949         unsigned found = NOT_FOUND;
950         *next_sector = (sector_t)-1;
951         while (n) {
952                 struct journal_node *j = container_of(n, struct journal_node, node);
953                 if (sector == j->sector) {
954                         found = j - ic->journal_tree;
955                 }
956                 if (sector < j->sector) {
957                         *next_sector = j->sector;
958                         n = j->node.rb_left;
959                 } else {
960                         n = j->node.rb_right;
961                 }
962         }
963
964         return found;
965 }
966
967 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
968 {
969         struct journal_node *node, *next_node;
970         struct rb_node *next;
971
972         if (unlikely(pos >= ic->journal_entries))
973                 return false;
974         node = &ic->journal_tree[pos];
975         if (unlikely(RB_EMPTY_NODE(&node->node)))
976                 return false;
977         if (unlikely(node->sector != sector))
978                 return false;
979
980         next = rb_next(&node->node);
981         if (unlikely(!next))
982                 return true;
983
984         next_node = container_of(next, struct journal_node, node);
985         return next_node->sector != sector;
986 }
987
988 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
989 {
990         struct rb_node *next;
991         struct journal_node *next_node;
992         unsigned next_section;
993
994         BUG_ON(RB_EMPTY_NODE(&node->node));
995
996         next = rb_next(&node->node);
997         if (unlikely(!next))
998                 return false;
999
1000         next_node = container_of(next, struct journal_node, node);
1001
1002         if (next_node->sector != node->sector)
1003                 return false;
1004
1005         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1006         if (next_section >= ic->committed_section &&
1007             next_section < ic->committed_section + ic->n_committed_sections)
1008                 return true;
1009         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1010                 return true;
1011
1012         return false;
1013 }
1014
1015 #define TAG_READ        0
1016 #define TAG_WRITE       1
1017 #define TAG_CMP         2
1018
1019 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1020                                unsigned *metadata_offset, unsigned total_size, int op)
1021 {
1022         do {
1023                 unsigned char *data, *dp;
1024                 struct dm_buffer *b;
1025                 unsigned to_copy;
1026                 int r;
1027
1028                 r = dm_integrity_failed(ic);
1029                 if (unlikely(r))
1030                         return r;
1031
1032                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1033                 if (unlikely(IS_ERR(data)))
1034                         return PTR_ERR(data);
1035
1036                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1037                 dp = data + *metadata_offset;
1038                 if (op == TAG_READ) {
1039                         memcpy(tag, dp, to_copy);
1040                 } else if (op == TAG_WRITE) {
1041                         memcpy(dp, tag, to_copy);
1042                         dm_bufio_mark_buffer_dirty(b);
1043                 } else  {
1044                         /* e.g.: op == TAG_CMP */
1045                         if (unlikely(memcmp(dp, tag, to_copy))) {
1046                                 unsigned i;
1047
1048                                 for (i = 0; i < to_copy; i++) {
1049                                         if (dp[i] != tag[i])
1050                                                 break;
1051                                         total_size--;
1052                                 }
1053                                 dm_bufio_release(b);
1054                                 return total_size;
1055                         }
1056                 }
1057                 dm_bufio_release(b);
1058
1059                 tag += to_copy;
1060                 *metadata_offset += to_copy;
1061                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1062                         (*metadata_block)++;
1063                         *metadata_offset = 0;
1064                 }
1065                 total_size -= to_copy;
1066         } while (unlikely(total_size));
1067
1068         return 0;
1069 }
1070
1071 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1072 {
1073         int r;
1074         r = dm_bufio_write_dirty_buffers(ic->bufio);
1075         if (unlikely(r))
1076                 dm_integrity_io_error(ic, "writing tags", r);
1077 }
1078
1079 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1080 {
1081         DECLARE_WAITQUEUE(wait, current);
1082         __add_wait_queue(&ic->endio_wait, &wait);
1083         __set_current_state(TASK_UNINTERRUPTIBLE);
1084         spin_unlock_irq(&ic->endio_wait.lock);
1085         io_schedule();
1086         spin_lock_irq(&ic->endio_wait.lock);
1087         __remove_wait_queue(&ic->endio_wait, &wait);
1088 }
1089
1090 static void autocommit_fn(unsigned long data)
1091 {
1092         struct dm_integrity_c *ic = (struct dm_integrity_c *)data;
1093
1094         if (likely(!dm_integrity_failed(ic)))
1095                 queue_work(ic->commit_wq, &ic->commit_work);
1096 }
1097
1098 static void schedule_autocommit(struct dm_integrity_c *ic)
1099 {
1100         if (!timer_pending(&ic->autocommit_timer))
1101                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1102 }
1103
1104 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1105 {
1106         struct bio *bio;
1107         spin_lock_irq(&ic->endio_wait.lock);
1108         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1109         bio_list_add(&ic->flush_bio_list, bio);
1110         spin_unlock_irq(&ic->endio_wait.lock);
1111         queue_work(ic->commit_wq, &ic->commit_work);
1112 }
1113
1114 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1115 {
1116         int r = dm_integrity_failed(ic);
1117         if (unlikely(r) && !bio->bi_error)
1118                 bio->bi_error = r;
1119         bio_endio(bio);
1120 }
1121
1122 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1123 {
1124         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1125
1126         if (unlikely(dio->fua) && likely(!bio->bi_error) && likely(!dm_integrity_failed(ic)))
1127                 submit_flush_bio(ic, dio);
1128         else
1129                 do_endio(ic, bio);
1130 }
1131
1132 static void dec_in_flight(struct dm_integrity_io *dio)
1133 {
1134         if (atomic_dec_and_test(&dio->in_flight)) {
1135                 struct dm_integrity_c *ic = dio->ic;
1136                 struct bio *bio;
1137
1138                 remove_range(ic, &dio->range);
1139
1140                 if (unlikely(dio->write))
1141                         schedule_autocommit(ic);
1142
1143                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1144
1145                 if (unlikely(dio->bi_error) && !bio->bi_error)
1146                         bio->bi_error = dio->bi_error;
1147                 if (likely(!bio->bi_error) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1148                         dio->range.logical_sector += dio->range.n_sectors;
1149                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1150                         INIT_WORK(&dio->work, integrity_bio_wait);
1151                         queue_work(ic->wait_wq, &dio->work);
1152                         return;
1153                 }
1154                 do_endio_flush(ic, dio);
1155         }
1156 }
1157
1158 static void integrity_end_io(struct bio *bio)
1159 {
1160         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1161
1162         bio->bi_iter = dio->orig_bi_iter;
1163         bio->bi_bdev = dio->orig_bi_bdev;
1164         if (dio->orig_bi_integrity) {
1165                 bio->bi_integrity = dio->orig_bi_integrity;
1166                 bio->bi_opf |= REQ_INTEGRITY;
1167         }
1168         bio->bi_end_io = dio->orig_bi_end_io;
1169
1170         if (dio->completion)
1171                 complete(dio->completion);
1172
1173         dec_in_flight(dio);
1174 }
1175
1176 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1177                                       const char *data, char *result)
1178 {
1179         __u64 sector_le = cpu_to_le64(sector);
1180         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1181         int r;
1182         unsigned digest_size;
1183
1184         req->tfm = ic->internal_hash;
1185         req->flags = 0;
1186
1187         r = crypto_shash_init(req);
1188         if (unlikely(r < 0)) {
1189                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1190                 goto failed;
1191         }
1192
1193         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1194         if (unlikely(r < 0)) {
1195                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1196                 goto failed;
1197         }
1198
1199         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1200         if (unlikely(r < 0)) {
1201                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1202                 goto failed;
1203         }
1204
1205         r = crypto_shash_final(req, result);
1206         if (unlikely(r < 0)) {
1207                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1208                 goto failed;
1209         }
1210
1211         digest_size = crypto_shash_digestsize(ic->internal_hash);
1212         if (unlikely(digest_size < ic->tag_size))
1213                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1214
1215         return;
1216
1217 failed:
1218         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1219         get_random_bytes(result, ic->tag_size);
1220 }
1221
1222 static void integrity_metadata(struct work_struct *w)
1223 {
1224         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1225         struct dm_integrity_c *ic = dio->ic;
1226
1227         int r;
1228
1229         if (ic->internal_hash) {
1230                 struct bvec_iter iter;
1231                 struct bio_vec bv;
1232                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1233                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1234                 char *checksums;
1235                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1236                 char checksums_onstack[ic->tag_size + extra_space];
1237                 unsigned sectors_to_process = dio->range.n_sectors;
1238                 sector_t sector = dio->range.logical_sector;
1239
1240                 if (unlikely(ic->mode == 'R'))
1241                         goto skip_io;
1242
1243                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1244                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1245                 if (!checksums)
1246                         checksums = checksums_onstack;
1247
1248                 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1249                         unsigned pos;
1250                         char *mem, *checksums_ptr;
1251
1252 again:
1253                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1254                         pos = 0;
1255                         checksums_ptr = checksums;
1256                         do {
1257                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1258                                 checksums_ptr += ic->tag_size;
1259                                 sectors_to_process -= ic->sectors_per_block;
1260                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1261                                 sector += ic->sectors_per_block;
1262                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1263                         kunmap_atomic(mem);
1264
1265                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1266                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1267                         if (unlikely(r)) {
1268                                 if (r > 0) {
1269                                         DMERR("Checksum failed at sector 0x%llx",
1270                                               (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1271                                         r = -EILSEQ;
1272                                 }
1273                                 if (likely(checksums != checksums_onstack))
1274                                         kfree(checksums);
1275                                 goto error;
1276                         }
1277
1278                         if (!sectors_to_process)
1279                                 break;
1280
1281                         if (unlikely(pos < bv.bv_len)) {
1282                                 bv.bv_offset += pos;
1283                                 bv.bv_len -= pos;
1284                                 goto again;
1285                         }
1286                 }
1287
1288                 if (likely(checksums != checksums_onstack))
1289                         kfree(checksums);
1290         } else {
1291                 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1292
1293                 if (bip) {
1294                         struct bio_vec biv;
1295                         struct bvec_iter iter;
1296                         unsigned data_to_process = dio->range.n_sectors;
1297                         sector_to_block(ic, data_to_process);
1298                         data_to_process *= ic->tag_size;
1299
1300                         bip_for_each_vec(biv, bip, iter) {
1301                                 unsigned char *tag;
1302                                 unsigned this_len;
1303
1304                                 BUG_ON(PageHighMem(biv.bv_page));
1305                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1306                                 this_len = min(biv.bv_len, data_to_process);
1307                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1308                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1309                                 if (unlikely(r))
1310                                         goto error;
1311                                 data_to_process -= this_len;
1312                                 if (!data_to_process)
1313                                         break;
1314                         }
1315                 }
1316         }
1317 skip_io:
1318         dec_in_flight(dio);
1319         return;
1320 error:
1321         dio->bi_error = r;
1322         dec_in_flight(dio);
1323 }
1324
1325 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1326 {
1327         struct dm_integrity_c *ic = ti->private;
1328         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1329         struct bio_integrity_payload *bip;
1330
1331         sector_t area, offset;
1332
1333         dio->ic = ic;
1334         dio->bi_error = 0;
1335
1336         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1337                 submit_flush_bio(ic, dio);
1338                 return DM_MAPIO_SUBMITTED;
1339         }
1340
1341         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1342         dio->write = bio_op(bio) == REQ_OP_WRITE;
1343         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1344         if (unlikely(dio->fua)) {
1345                 /*
1346                  * Don't pass down the FUA flag because we have to flush
1347                  * disk cache anyway.
1348                  */
1349                 bio->bi_opf &= ~REQ_FUA;
1350         }
1351         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1352                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1353                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1354                       (unsigned long long)ic->provided_data_sectors);
1355                 return -EIO;
1356         }
1357         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1358                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1359                       ic->sectors_per_block,
1360                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1361                 return -EIO;
1362         }
1363
1364         if (ic->sectors_per_block > 1) {
1365                 struct bvec_iter iter;
1366                 struct bio_vec bv;
1367                 bio_for_each_segment(bv, bio, iter) {
1368                         if (unlikely((bv.bv_offset | bv.bv_len) & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1369                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1370                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1371                                 return -EIO;
1372                         }
1373                 }
1374         }
1375
1376         bip = bio_integrity(bio);
1377         if (!ic->internal_hash) {
1378                 if (bip) {
1379                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1380                         if (ic->log2_tag_size >= 0)
1381                                 wanted_tag_size <<= ic->log2_tag_size;
1382                         else
1383                                 wanted_tag_size *= ic->tag_size;
1384                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1385                                 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1386                                 return -EIO;
1387                         }
1388                 }
1389         } else {
1390                 if (unlikely(bip != NULL)) {
1391                         DMERR("Unexpected integrity data when using internal hash");
1392                         return -EIO;
1393                 }
1394         }
1395
1396         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1397                 return -EIO;
1398
1399         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1400         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1401         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1402
1403         dm_integrity_map_continue(dio, true);
1404         return DM_MAPIO_SUBMITTED;
1405 }
1406
1407 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1408                                  unsigned journal_section, unsigned journal_entry)
1409 {
1410         struct dm_integrity_c *ic = dio->ic;
1411         sector_t logical_sector;
1412         unsigned n_sectors;
1413
1414         logical_sector = dio->range.logical_sector;
1415         n_sectors = dio->range.n_sectors;
1416         do {
1417                 struct bio_vec bv = bio_iovec(bio);
1418                 char *mem;
1419
1420                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1421                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1422                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1423                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1424 retry_kmap:
1425                 mem = kmap_atomic(bv.bv_page);
1426                 if (likely(dio->write))
1427                         flush_dcache_page(bv.bv_page);
1428
1429                 do {
1430                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1431
1432                         if (unlikely(!dio->write)) {
1433                                 struct journal_sector *js;
1434                                 char *mem_ptr;
1435                                 unsigned s;
1436
1437                                 if (unlikely(journal_entry_is_inprogress(je))) {
1438                                         flush_dcache_page(bv.bv_page);
1439                                         kunmap_atomic(mem);
1440
1441                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1442                                         goto retry_kmap;
1443                                 }
1444                                 smp_rmb();
1445                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1446                                 js = access_journal_data(ic, journal_section, journal_entry);
1447                                 mem_ptr = mem + bv.bv_offset;
1448                                 s = 0;
1449                                 do {
1450                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1451                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1452                                         js++;
1453                                         mem_ptr += 1 << SECTOR_SHIFT;
1454                                 } while (++s < ic->sectors_per_block);
1455 #ifdef INTERNAL_VERIFY
1456                                 if (ic->internal_hash) {
1457                                         char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1458
1459                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1460                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1461                                                 DMERR("Checksum failed when reading from journal, at sector 0x%llx",
1462                                                       (unsigned long long)logical_sector);
1463                                         }
1464                                 }
1465 #endif
1466                         }
1467
1468                         if (!ic->internal_hash) {
1469                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1470                                 unsigned tag_todo = ic->tag_size;
1471                                 char *tag_ptr = journal_entry_tag(ic, je);
1472
1473                                 if (bip) do {
1474                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1475                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1476                                         char *tag_addr;
1477                                         BUG_ON(PageHighMem(biv.bv_page));
1478                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1479                                         if (likely(dio->write))
1480                                                 memcpy(tag_ptr, tag_addr, tag_now);
1481                                         else
1482                                                 memcpy(tag_addr, tag_ptr, tag_now);
1483                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1484                                         tag_ptr += tag_now;
1485                                         tag_todo -= tag_now;
1486                                 } while (unlikely(tag_todo)); else {
1487                                         if (likely(dio->write))
1488                                                 memset(tag_ptr, 0, tag_todo);
1489                                 }
1490                         }
1491
1492                         if (likely(dio->write)) {
1493                                 struct journal_sector *js;
1494                                 unsigned s;
1495
1496                                 js = access_journal_data(ic, journal_section, journal_entry);
1497                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1498
1499                                 s = 0;
1500                                 do {
1501                                         je->last_bytes[s] = js[s].commit_id;
1502                                 } while (++s < ic->sectors_per_block);
1503
1504                                 if (ic->internal_hash) {
1505                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1506                                         if (unlikely(digest_size > ic->tag_size)) {
1507                                                 char checksums_onstack[digest_size];
1508                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1509                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1510                                         } else
1511                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1512                                 }
1513
1514                                 journal_entry_set_sector(je, logical_sector);
1515                         }
1516                         logical_sector += ic->sectors_per_block;
1517
1518                         journal_entry++;
1519                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1520                                 journal_entry = 0;
1521                                 journal_section++;
1522                                 wraparound_section(ic, &journal_section);
1523                         }
1524
1525                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1526                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1527
1528                 if (unlikely(!dio->write))
1529                         flush_dcache_page(bv.bv_page);
1530                 kunmap_atomic(mem);
1531         } while (n_sectors);
1532
1533         if (likely(dio->write)) {
1534                 smp_mb();
1535                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1536                         wake_up(&ic->copy_to_journal_wait);
1537                 if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1538                         queue_work(ic->commit_wq, &ic->commit_work);
1539                 } else {
1540                         schedule_autocommit(ic);
1541                 }
1542         } else {
1543                 remove_range(ic, &dio->range);
1544         }
1545
1546         if (unlikely(bio->bi_iter.bi_size)) {
1547                 sector_t area, offset;
1548
1549                 dio->range.logical_sector = logical_sector;
1550                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1551                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1552                 return true;
1553         }
1554
1555         return false;
1556 }
1557
1558 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1559 {
1560         struct dm_integrity_c *ic = dio->ic;
1561         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1562         unsigned journal_section, journal_entry;
1563         unsigned journal_read_pos;
1564         struct completion read_comp;
1565         bool need_sync_io = ic->internal_hash && !dio->write;
1566
1567         if (need_sync_io && from_map) {
1568                 INIT_WORK(&dio->work, integrity_bio_wait);
1569                 queue_work(ic->metadata_wq, &dio->work);
1570                 return;
1571         }
1572
1573 lock_retry:
1574         spin_lock_irq(&ic->endio_wait.lock);
1575 retry:
1576         if (unlikely(dm_integrity_failed(ic))) {
1577                 spin_unlock_irq(&ic->endio_wait.lock);
1578                 do_endio(ic, bio);
1579                 return;
1580         }
1581         dio->range.n_sectors = bio_sectors(bio);
1582         journal_read_pos = NOT_FOUND;
1583         if (likely(ic->mode == 'J')) {
1584                 if (dio->write) {
1585                         unsigned next_entry, i, pos;
1586                         unsigned ws, we;
1587
1588                         dio->range.n_sectors = min(dio->range.n_sectors, ic->free_sectors);
1589                         if (unlikely(!dio->range.n_sectors))
1590                                 goto sleep;
1591                         ic->free_sectors -= dio->range.n_sectors;
1592                         journal_section = ic->free_section;
1593                         journal_entry = ic->free_section_entry;
1594
1595                         next_entry = ic->free_section_entry + dio->range.n_sectors;
1596                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1597                         ic->free_section += next_entry / ic->journal_section_entries;
1598                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1599                         wraparound_section(ic, &ic->free_section);
1600
1601                         pos = journal_section * ic->journal_section_entries + journal_entry;
1602                         ws = journal_section;
1603                         we = journal_entry;
1604                         i = 0;
1605                         do {
1606                                 struct journal_entry *je;
1607
1608                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1609                                 pos++;
1610                                 if (unlikely(pos >= ic->journal_entries))
1611                                         pos = 0;
1612
1613                                 je = access_journal_entry(ic, ws, we);
1614                                 BUG_ON(!journal_entry_is_unused(je));
1615                                 journal_entry_set_inprogress(je);
1616                                 we++;
1617                                 if (unlikely(we == ic->journal_section_entries)) {
1618                                         we = 0;
1619                                         ws++;
1620                                         wraparound_section(ic, &ws);
1621                                 }
1622                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1623
1624                         spin_unlock_irq(&ic->endio_wait.lock);
1625                         goto journal_read_write;
1626                 } else {
1627                         sector_t next_sector;
1628                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1629                         if (likely(journal_read_pos == NOT_FOUND)) {
1630                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1631                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1632                         } else {
1633                                 unsigned i;
1634                                 unsigned jp = journal_read_pos + 1;
1635                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1636                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1637                                                 break;
1638                                 }
1639                                 dio->range.n_sectors = i;
1640                         }
1641                 }
1642         }
1643         if (unlikely(!add_new_range(ic, &dio->range))) {
1644                 /*
1645                  * We must not sleep in the request routine because it could
1646                  * stall bios on current->bio_list.
1647                  * So, we offload the bio to a workqueue if we have to sleep.
1648                  */
1649 sleep:
1650                 if (from_map) {
1651                         spin_unlock_irq(&ic->endio_wait.lock);
1652                         INIT_WORK(&dio->work, integrity_bio_wait);
1653                         queue_work(ic->wait_wq, &dio->work);
1654                         return;
1655                 } else {
1656                         sleep_on_endio_wait(ic);
1657                         goto retry;
1658                 }
1659         }
1660         spin_unlock_irq(&ic->endio_wait.lock);
1661
1662         if (unlikely(journal_read_pos != NOT_FOUND)) {
1663                 journal_section = journal_read_pos / ic->journal_section_entries;
1664                 journal_entry = journal_read_pos % ic->journal_section_entries;
1665                 goto journal_read_write;
1666         }
1667
1668         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1669
1670         if (need_sync_io) {
1671                 read_comp = COMPLETION_INITIALIZER_ONSTACK(read_comp);
1672                 dio->completion = &read_comp;
1673         } else
1674                 dio->completion = NULL;
1675
1676         dio->orig_bi_iter = bio->bi_iter;
1677
1678         dio->orig_bi_bdev = bio->bi_bdev;
1679         bio->bi_bdev = ic->dev->bdev;
1680
1681         dio->orig_bi_integrity = bio_integrity(bio);
1682         bio->bi_integrity = NULL;
1683         bio->bi_opf &= ~REQ_INTEGRITY;
1684
1685         dio->orig_bi_end_io = bio->bi_end_io;
1686         bio->bi_end_io = integrity_end_io;
1687
1688         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1689         bio->bi_iter.bi_sector += ic->start;
1690         generic_make_request(bio);
1691
1692         if (need_sync_io) {
1693                 wait_for_completion_io(&read_comp);
1694                 integrity_metadata(&dio->work);
1695         } else {
1696                 INIT_WORK(&dio->work, integrity_metadata);
1697                 queue_work(ic->metadata_wq, &dio->work);
1698         }
1699
1700         return;
1701
1702 journal_read_write:
1703         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1704                 goto lock_retry;
1705
1706         do_endio_flush(ic, dio);
1707 }
1708
1709
1710 static void integrity_bio_wait(struct work_struct *w)
1711 {
1712         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1713
1714         dm_integrity_map_continue(dio, false);
1715 }
1716
1717 static void pad_uncommitted(struct dm_integrity_c *ic)
1718 {
1719         if (ic->free_section_entry) {
1720                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1721                 ic->free_section_entry = 0;
1722                 ic->free_section++;
1723                 wraparound_section(ic, &ic->free_section);
1724                 ic->n_uncommitted_sections++;
1725         }
1726 }
1727
1728 static void integrity_commit(struct work_struct *w)
1729 {
1730         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1731         unsigned commit_start, commit_sections;
1732         unsigned i, j, n;
1733         struct bio *flushes;
1734
1735         del_timer(&ic->autocommit_timer);
1736
1737         spin_lock_irq(&ic->endio_wait.lock);
1738         flushes = bio_list_get(&ic->flush_bio_list);
1739         if (unlikely(ic->mode != 'J')) {
1740                 spin_unlock_irq(&ic->endio_wait.lock);
1741                 dm_integrity_flush_buffers(ic);
1742                 goto release_flush_bios;
1743         }
1744
1745         pad_uncommitted(ic);
1746         commit_start = ic->uncommitted_section;
1747         commit_sections = ic->n_uncommitted_sections;
1748         spin_unlock_irq(&ic->endio_wait.lock);
1749
1750         if (!commit_sections)
1751                 goto release_flush_bios;
1752
1753         i = commit_start;
1754         for (n = 0; n < commit_sections; n++) {
1755                 for (j = 0; j < ic->journal_section_entries; j++) {
1756                         struct journal_entry *je;
1757                         je = access_journal_entry(ic, i, j);
1758                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1759                 }
1760                 for (j = 0; j < ic->journal_section_sectors; j++) {
1761                         struct journal_sector *js;
1762                         js = access_journal(ic, i, j);
1763                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1764                 }
1765                 i++;
1766                 if (unlikely(i >= ic->journal_sections))
1767                         ic->commit_seq = next_commit_seq(ic->commit_seq);
1768                 wraparound_section(ic, &i);
1769         }
1770         smp_rmb();
1771
1772         write_journal(ic, commit_start, commit_sections);
1773
1774         spin_lock_irq(&ic->endio_wait.lock);
1775         ic->uncommitted_section += commit_sections;
1776         wraparound_section(ic, &ic->uncommitted_section);
1777         ic->n_uncommitted_sections -= commit_sections;
1778         ic->n_committed_sections += commit_sections;
1779         spin_unlock_irq(&ic->endio_wait.lock);
1780
1781         if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1782                 queue_work(ic->writer_wq, &ic->writer_work);
1783
1784 release_flush_bios:
1785         while (flushes) {
1786                 struct bio *next = flushes->bi_next;
1787                 flushes->bi_next = NULL;
1788                 do_endio(ic, flushes);
1789                 flushes = next;
1790         }
1791 }
1792
1793 static void complete_copy_from_journal(unsigned long error, void *context)
1794 {
1795         struct journal_io *io = context;
1796         struct journal_completion *comp = io->comp;
1797         struct dm_integrity_c *ic = comp->ic;
1798         remove_range(ic, &io->range);
1799         mempool_free(io, ic->journal_io_mempool);
1800         if (unlikely(error != 0))
1801                 dm_integrity_io_error(ic, "copying from journal", -EIO);
1802         complete_journal_op(comp);
1803 }
1804
1805 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1806                                struct journal_entry *je)
1807 {
1808         unsigned s = 0;
1809         do {
1810                 js->commit_id = je->last_bytes[s];
1811                 js++;
1812         } while (++s < ic->sectors_per_block);
1813 }
1814
1815 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1816                              unsigned write_sections, bool from_replay)
1817 {
1818         unsigned i, j, n;
1819         struct journal_completion comp;
1820
1821         comp.ic = ic;
1822         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1823         comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
1824
1825         i = write_start;
1826         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1827 #ifndef INTERNAL_VERIFY
1828                 if (unlikely(from_replay))
1829 #endif
1830                         rw_section_mac(ic, i, false);
1831                 for (j = 0; j < ic->journal_section_entries; j++) {
1832                         struct journal_entry *je = access_journal_entry(ic, i, j);
1833                         sector_t sec, area, offset;
1834                         unsigned k, l, next_loop;
1835                         sector_t metadata_block;
1836                         unsigned metadata_offset;
1837                         struct journal_io *io;
1838
1839                         if (journal_entry_is_unused(je))
1840                                 continue;
1841                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1842                         sec = journal_entry_get_sector(je);
1843                         if (unlikely(from_replay)) {
1844                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1845                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1846                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
1847                                 }
1848                         }
1849                         get_area_and_offset(ic, sec, &area, &offset);
1850                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1851                         for (k = j + 1; k < ic->journal_section_entries; k++) {
1852                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1853                                 sector_t sec2, area2, offset2;
1854                                 if (journal_entry_is_unused(je2))
1855                                         break;
1856                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1857                                 sec2 = journal_entry_get_sector(je2);
1858                                 get_area_and_offset(ic, sec2, &area2, &offset2);
1859                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1860                                         break;
1861                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1862                         }
1863                         next_loop = k - 1;
1864
1865                         io = mempool_alloc(ic->journal_io_mempool, GFP_NOIO);
1866                         io->comp = &comp;
1867                         io->range.logical_sector = sec;
1868                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1869
1870                         spin_lock_irq(&ic->endio_wait.lock);
1871                         while (unlikely(!add_new_range(ic, &io->range)))
1872                                 sleep_on_endio_wait(ic);
1873
1874                         if (likely(!from_replay)) {
1875                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1876
1877                                 /* don't write if there is newer committed sector */
1878                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
1879                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
1880
1881                                         journal_entry_set_unused(je2);
1882                                         remove_journal_node(ic, &section_node[j]);
1883                                         j++;
1884                                         sec += ic->sectors_per_block;
1885                                         offset += ic->sectors_per_block;
1886                                 }
1887                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
1888                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
1889
1890                                         journal_entry_set_unused(je2);
1891                                         remove_journal_node(ic, &section_node[k - 1]);
1892                                         k--;
1893                                 }
1894                                 if (j == k) {
1895                                         remove_range_unlocked(ic, &io->range);
1896                                         spin_unlock_irq(&ic->endio_wait.lock);
1897                                         mempool_free(io, ic->journal_io_mempool);
1898                                         goto skip_io;
1899                                 }
1900                                 for (l = j; l < k; l++) {
1901                                         remove_journal_node(ic, &section_node[l]);
1902                                 }
1903                         }
1904                         spin_unlock_irq(&ic->endio_wait.lock);
1905
1906                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
1907                         for (l = j; l < k; l++) {
1908                                 int r;
1909                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
1910
1911                                 if (
1912 #ifndef INTERNAL_VERIFY
1913                                     unlikely(from_replay) &&
1914 #endif
1915                                     ic->internal_hash) {
1916                                         char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1917
1918                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
1919                                                                   (char *)access_journal_data(ic, i, l), test_tag);
1920                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
1921                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
1922                                 }
1923
1924                                 journal_entry_set_unused(je2);
1925                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
1926                                                         ic->tag_size, TAG_WRITE);
1927                                 if (unlikely(r)) {
1928                                         dm_integrity_io_error(ic, "reading tags", r);
1929                                 }
1930                         }
1931
1932                         atomic_inc(&comp.in_flight);
1933                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
1934                                           (k - j) << ic->sb->log2_sectors_per_block,
1935                                           get_data_sector(ic, area, offset),
1936                                           complete_copy_from_journal, io);
1937 skip_io:
1938                         j = next_loop;
1939                 }
1940         }
1941
1942         dm_bufio_write_dirty_buffers_async(ic->bufio);
1943
1944         complete_journal_op(&comp);
1945         wait_for_completion_io(&comp.comp);
1946
1947         dm_integrity_flush_buffers(ic);
1948 }
1949
1950 static void integrity_writer(struct work_struct *w)
1951 {
1952         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
1953         unsigned write_start, write_sections;
1954
1955         unsigned prev_free_sectors;
1956
1957         /* the following test is not needed, but it tests the replay code */
1958         if (ACCESS_ONCE(ic->suspending))
1959                 return;
1960
1961         spin_lock_irq(&ic->endio_wait.lock);
1962         write_start = ic->committed_section;
1963         write_sections = ic->n_committed_sections;
1964         spin_unlock_irq(&ic->endio_wait.lock);
1965
1966         if (!write_sections)
1967                 return;
1968
1969         do_journal_write(ic, write_start, write_sections, false);
1970
1971         spin_lock_irq(&ic->endio_wait.lock);
1972
1973         ic->committed_section += write_sections;
1974         wraparound_section(ic, &ic->committed_section);
1975         ic->n_committed_sections -= write_sections;
1976
1977         prev_free_sectors = ic->free_sectors;
1978         ic->free_sectors += write_sections * ic->journal_section_entries;
1979         if (unlikely(!prev_free_sectors))
1980                 wake_up_locked(&ic->endio_wait);
1981
1982         spin_unlock_irq(&ic->endio_wait.lock);
1983 }
1984
1985 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
1986                          unsigned n_sections, unsigned char commit_seq)
1987 {
1988         unsigned i, j, n;
1989
1990         if (!n_sections)
1991                 return;
1992
1993         for (n = 0; n < n_sections; n++) {
1994                 i = start_section + n;
1995                 wraparound_section(ic, &i);
1996                 for (j = 0; j < ic->journal_section_sectors; j++) {
1997                         struct journal_sector *js = access_journal(ic, i, j);
1998                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
1999                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2000                 }
2001                 for (j = 0; j < ic->journal_section_entries; j++) {
2002                         struct journal_entry *je = access_journal_entry(ic, i, j);
2003                         journal_entry_set_unused(je);
2004                 }
2005         }
2006
2007         write_journal(ic, start_section, n_sections);
2008 }
2009
2010 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2011 {
2012         unsigned char k;
2013         for (k = 0; k < N_COMMIT_IDS; k++) {
2014                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2015                         return k;
2016         }
2017         dm_integrity_io_error(ic, "journal commit id", -EIO);
2018         return -EIO;
2019 }
2020
2021 static void replay_journal(struct dm_integrity_c *ic)
2022 {
2023         unsigned i, j;
2024         bool used_commit_ids[N_COMMIT_IDS];
2025         unsigned max_commit_id_sections[N_COMMIT_IDS];
2026         unsigned write_start, write_sections;
2027         unsigned continue_section;
2028         bool journal_empty;
2029         unsigned char unused, last_used, want_commit_seq;
2030
2031         if (ic->mode == 'R')
2032                 return;
2033
2034         if (ic->journal_uptodate)
2035                 return;
2036
2037         last_used = 0;
2038         write_start = 0;
2039
2040         if (!ic->just_formatted) {
2041                 DEBUG_print("reading journal\n");
2042                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2043                 if (ic->journal_io)
2044                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2045                 if (ic->journal_io) {
2046                         struct journal_completion crypt_comp;
2047                         crypt_comp.ic = ic;
2048                         crypt_comp.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp.comp);
2049                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2050                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2051                         wait_for_completion(&crypt_comp.comp);
2052                 }
2053                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2054         }
2055
2056         if (dm_integrity_failed(ic))
2057                 goto clear_journal;
2058
2059         journal_empty = true;
2060         memset(used_commit_ids, 0, sizeof used_commit_ids);
2061         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2062         for (i = 0; i < ic->journal_sections; i++) {
2063                 for (j = 0; j < ic->journal_section_sectors; j++) {
2064                         int k;
2065                         struct journal_sector *js = access_journal(ic, i, j);
2066                         k = find_commit_seq(ic, i, j, js->commit_id);
2067                         if (k < 0)
2068                                 goto clear_journal;
2069                         used_commit_ids[k] = true;
2070                         max_commit_id_sections[k] = i;
2071                 }
2072                 if (journal_empty) {
2073                         for (j = 0; j < ic->journal_section_entries; j++) {
2074                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2075                                 if (!journal_entry_is_unused(je)) {
2076                                         journal_empty = false;
2077                                         break;
2078                                 }
2079                         }
2080                 }
2081         }
2082
2083         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2084                 unused = N_COMMIT_IDS - 1;
2085                 while (unused && !used_commit_ids[unused - 1])
2086                         unused--;
2087         } else {
2088                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2089                         if (!used_commit_ids[unused])
2090                                 break;
2091                 if (unused == N_COMMIT_IDS) {
2092                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2093                         goto clear_journal;
2094                 }
2095         }
2096         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2097                     unused, used_commit_ids[0], used_commit_ids[1],
2098                     used_commit_ids[2], used_commit_ids[3]);
2099
2100         last_used = prev_commit_seq(unused);
2101         want_commit_seq = prev_commit_seq(last_used);
2102
2103         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2104                 journal_empty = true;
2105
2106         write_start = max_commit_id_sections[last_used] + 1;
2107         if (unlikely(write_start >= ic->journal_sections))
2108                 want_commit_seq = next_commit_seq(want_commit_seq);
2109         wraparound_section(ic, &write_start);
2110
2111         i = write_start;
2112         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2113                 for (j = 0; j < ic->journal_section_sectors; j++) {
2114                         struct journal_sector *js = access_journal(ic, i, j);
2115
2116                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2117                                 /*
2118                                  * This could be caused by crash during writing.
2119                                  * We won't replay the inconsistent part of the
2120                                  * journal.
2121                                  */
2122                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2123                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2124                                 goto brk;
2125                         }
2126                 }
2127                 i++;
2128                 if (unlikely(i >= ic->journal_sections))
2129                         want_commit_seq = next_commit_seq(want_commit_seq);
2130                 wraparound_section(ic, &i);
2131         }
2132 brk:
2133
2134         if (!journal_empty) {
2135                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2136                             write_sections, write_start, want_commit_seq);
2137                 do_journal_write(ic, write_start, write_sections, true);
2138         }
2139
2140         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2141                 continue_section = write_start;
2142                 ic->commit_seq = want_commit_seq;
2143                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2144         } else {
2145                 unsigned s;
2146                 unsigned char erase_seq;
2147 clear_journal:
2148                 DEBUG_print("clearing journal\n");
2149
2150                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2151                 s = write_start;
2152                 init_journal(ic, s, 1, erase_seq);
2153                 s++;
2154                 wraparound_section(ic, &s);
2155                 if (ic->journal_sections >= 2) {
2156                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2157                         s += ic->journal_sections - 2;
2158                         wraparound_section(ic, &s);
2159                         init_journal(ic, s, 1, erase_seq);
2160                 }
2161
2162                 continue_section = 0;
2163                 ic->commit_seq = next_commit_seq(erase_seq);
2164         }
2165
2166         ic->committed_section = continue_section;
2167         ic->n_committed_sections = 0;
2168
2169         ic->uncommitted_section = continue_section;
2170         ic->n_uncommitted_sections = 0;
2171
2172         ic->free_section = continue_section;
2173         ic->free_section_entry = 0;
2174         ic->free_sectors = ic->journal_entries;
2175
2176         ic->journal_tree_root = RB_ROOT;
2177         for (i = 0; i < ic->journal_entries; i++)
2178                 init_journal_node(&ic->journal_tree[i]);
2179 }
2180
2181 static void dm_integrity_postsuspend(struct dm_target *ti)
2182 {
2183         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2184
2185         del_timer_sync(&ic->autocommit_timer);
2186
2187         ic->suspending = true;
2188
2189         queue_work(ic->commit_wq, &ic->commit_work);
2190         drain_workqueue(ic->commit_wq);
2191
2192         if (ic->mode == 'J') {
2193                 drain_workqueue(ic->writer_wq);
2194                 dm_integrity_flush_buffers(ic);
2195         }
2196
2197         ic->suspending = false;
2198
2199         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2200
2201         ic->journal_uptodate = true;
2202 }
2203
2204 static void dm_integrity_resume(struct dm_target *ti)
2205 {
2206         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2207
2208         replay_journal(ic);
2209 }
2210
2211 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2212                                 unsigned status_flags, char *result, unsigned maxlen)
2213 {
2214         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2215         unsigned arg_count;
2216         size_t sz = 0;
2217
2218         switch (type) {
2219         case STATUSTYPE_INFO:
2220                 result[0] = '\0';
2221                 break;
2222
2223         case STATUSTYPE_TABLE: {
2224                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2225                 watermark_percentage += ic->journal_entries / 2;
2226                 do_div(watermark_percentage, ic->journal_entries);
2227                 arg_count = 5;
2228                 arg_count += ic->sectors_per_block != 1;
2229                 arg_count += !!ic->internal_hash_alg.alg_string;
2230                 arg_count += !!ic->journal_crypt_alg.alg_string;
2231                 arg_count += !!ic->journal_mac_alg.alg_string;
2232                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2233                        ic->tag_size, ic->mode, arg_count);
2234                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2235                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2236                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2237                 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2238                 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2239                 if (ic->sectors_per_block != 1)
2240                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2241
2242 #define EMIT_ALG(a, n)                                                  \
2243                 do {                                                    \
2244                         if (ic->a.alg_string) {                         \
2245                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2246                                 if (ic->a.key_string)                   \
2247                                         DMEMIT(":%s", ic->a.key_string);\
2248                         }                                               \
2249                 } while (0)
2250                 EMIT_ALG(internal_hash_alg, "internal_hash");
2251                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2252                 EMIT_ALG(journal_mac_alg, "journal_mac");
2253                 break;
2254         }
2255         }
2256 }
2257
2258 static int dm_integrity_iterate_devices(struct dm_target *ti,
2259                                         iterate_devices_callout_fn fn, void *data)
2260 {
2261         struct dm_integrity_c *ic = ti->private;
2262
2263         return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2264 }
2265
2266 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2267 {
2268         struct dm_integrity_c *ic = ti->private;
2269
2270         if (ic->sectors_per_block > 1) {
2271                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2272                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2273                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2274         }
2275 }
2276
2277 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2278 {
2279         unsigned sector_space = JOURNAL_SECTOR_DATA;
2280
2281         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2282         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2283                                          JOURNAL_ENTRY_ROUNDUP);
2284
2285         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2286                 sector_space -= JOURNAL_MAC_PER_SECTOR;
2287         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2288         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2289         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2290         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2291 }
2292
2293 static int calculate_device_limits(struct dm_integrity_c *ic)
2294 {
2295         __u64 initial_sectors;
2296         sector_t last_sector, last_area, last_offset;
2297
2298         calculate_journal_section_size(ic);
2299         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2300         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->device_sectors || initial_sectors > UINT_MAX)
2301                 return -EINVAL;
2302         ic->initial_sectors = initial_sectors;
2303
2304         ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2305                                    (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2306         if (!(ic->metadata_run & (ic->metadata_run - 1)))
2307                 ic->log2_metadata_run = __ffs(ic->metadata_run);
2308         else
2309                 ic->log2_metadata_run = -1;
2310
2311         get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2312         last_sector = get_data_sector(ic, last_area, last_offset);
2313
2314         if (ic->start + last_sector < last_sector || ic->start + last_sector >= ic->device_sectors)
2315                 return -EINVAL;
2316
2317         return 0;
2318 }
2319
2320 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2321 {
2322         unsigned journal_sections;
2323         int test_bit;
2324
2325         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2326         memcpy(ic->sb->magic, SB_MAGIC, 8);
2327         ic->sb->version = SB_VERSION;
2328         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2329         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2330         if (ic->journal_mac_alg.alg_string)
2331                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2332
2333         calculate_journal_section_size(ic);
2334         journal_sections = journal_sectors / ic->journal_section_sectors;
2335         if (!journal_sections)
2336                 journal_sections = 1;
2337         ic->sb->journal_sections = cpu_to_le32(journal_sections);
2338
2339         if (!interleave_sectors)
2340                 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2341         ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2342         ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2343         ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2344
2345         ic->provided_data_sectors = 0;
2346         for (test_bit = fls64(ic->device_sectors) - 1; test_bit >= 3; test_bit--) {
2347                 __u64 prev_data_sectors = ic->provided_data_sectors;
2348
2349                 ic->provided_data_sectors |= (sector_t)1 << test_bit;
2350                 if (calculate_device_limits(ic))
2351                         ic->provided_data_sectors = prev_data_sectors;
2352         }
2353
2354         if (!ic->provided_data_sectors)
2355                 return -EINVAL;
2356
2357         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2358
2359         return 0;
2360 }
2361
2362 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2363 {
2364         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2365         struct blk_integrity bi;
2366
2367         memset(&bi, 0, sizeof(bi));
2368         bi.profile = &dm_integrity_profile;
2369         bi.tuple_size = ic->tag_size;
2370         bi.tag_size = bi.tuple_size;
2371         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2372
2373         blk_integrity_register(disk, &bi);
2374         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2375 }
2376
2377 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2378 {
2379         unsigned i;
2380
2381         if (!pl)
2382                 return;
2383         for (i = 0; i < ic->journal_pages; i++)
2384                 if (pl[i].page)
2385                         __free_page(pl[i].page);
2386         kvfree(pl);
2387 }
2388
2389 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2390 {
2391         size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2392         struct page_list *pl;
2393         unsigned i;
2394
2395         pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2396         if (!pl)
2397                 return NULL;
2398
2399         for (i = 0; i < ic->journal_pages; i++) {
2400                 pl[i].page = alloc_page(GFP_KERNEL);
2401                 if (!pl[i].page) {
2402                         dm_integrity_free_page_list(ic, pl);
2403                         return NULL;
2404                 }
2405                 if (i)
2406                         pl[i - 1].next = &pl[i];
2407         }
2408
2409         return pl;
2410 }
2411
2412 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2413 {
2414         unsigned i;
2415         for (i = 0; i < ic->journal_sections; i++)
2416                 kvfree(sl[i]);
2417         kfree(sl);
2418 }
2419
2420 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2421 {
2422         struct scatterlist **sl;
2423         unsigned i;
2424
2425         sl = kvmalloc(ic->journal_sections * sizeof(struct scatterlist *), GFP_KERNEL | __GFP_ZERO);
2426         if (!sl)
2427                 return NULL;
2428
2429         for (i = 0; i < ic->journal_sections; i++) {
2430                 struct scatterlist *s;
2431                 unsigned start_index, start_offset;
2432                 unsigned end_index, end_offset;
2433                 unsigned n_pages;
2434                 unsigned idx;
2435
2436                 page_list_location(ic, i, 0, &start_index, &start_offset);
2437                 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2438
2439                 n_pages = (end_index - start_index + 1);
2440
2441                 s = kvmalloc(n_pages * sizeof(struct scatterlist), GFP_KERNEL);
2442                 if (!s) {
2443                         dm_integrity_free_journal_scatterlist(ic, sl);
2444                         return NULL;
2445                 }
2446
2447                 sg_init_table(s, n_pages);
2448                 for (idx = start_index; idx <= end_index; idx++) {
2449                         char *va = lowmem_page_address(pl[idx].page);
2450                         unsigned start = 0, end = PAGE_SIZE;
2451                         if (idx == start_index)
2452                                 start = start_offset;
2453                         if (idx == end_index)
2454                                 end = end_offset + (1 << SECTOR_SHIFT);
2455                         sg_set_buf(&s[idx - start_index], va + start, end - start);
2456                 }
2457
2458                 sl[i] = s;
2459         }
2460
2461         return sl;
2462 }
2463
2464 static void free_alg(struct alg_spec *a)
2465 {
2466         kzfree(a->alg_string);
2467         kzfree(a->key);
2468         memset(a, 0, sizeof *a);
2469 }
2470
2471 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2472 {
2473         char *k;
2474
2475         free_alg(a);
2476
2477         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2478         if (!a->alg_string)
2479                 goto nomem;
2480
2481         k = strchr(a->alg_string, ':');
2482         if (k) {
2483                 *k = 0;
2484                 a->key_string = k + 1;
2485                 if (strlen(a->key_string) & 1)
2486                         goto inval;
2487
2488                 a->key_size = strlen(a->key_string) / 2;
2489                 a->key = kmalloc(a->key_size, GFP_KERNEL);
2490                 if (!a->key)
2491                         goto nomem;
2492                 if (hex2bin(a->key, a->key_string, a->key_size))
2493                         goto inval;
2494         }
2495
2496         return 0;
2497 inval:
2498         *error = error_inval;
2499         return -EINVAL;
2500 nomem:
2501         *error = "Out of memory for an argument";
2502         return -ENOMEM;
2503 }
2504
2505 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2506                    char *error_alg, char *error_key)
2507 {
2508         int r;
2509
2510         if (a->alg_string) {
2511                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2512                 if (IS_ERR(*hash)) {
2513                         *error = error_alg;
2514                         r = PTR_ERR(*hash);
2515                         *hash = NULL;
2516                         return r;
2517                 }
2518
2519                 if (a->key) {
2520                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
2521                         if (r) {
2522                                 *error = error_key;
2523                                 return r;
2524                         }
2525                 }
2526         }
2527
2528         return 0;
2529 }
2530
2531 static int create_journal(struct dm_integrity_c *ic, char **error)
2532 {
2533         int r = 0;
2534         unsigned i;
2535         __u64 journal_pages, journal_desc_size, journal_tree_size;
2536         unsigned char *crypt_data = NULL;
2537
2538         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2539         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2540         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2541         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2542
2543         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2544                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2545         journal_desc_size = journal_pages * sizeof(struct page_list);
2546         if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2547                 *error = "Journal doesn't fit into memory";
2548                 r = -ENOMEM;
2549                 goto bad;
2550         }
2551         ic->journal_pages = journal_pages;
2552
2553         ic->journal = dm_integrity_alloc_page_list(ic);
2554         if (!ic->journal) {
2555                 *error = "Could not allocate memory for journal";
2556                 r = -ENOMEM;
2557                 goto bad;
2558         }
2559         if (ic->journal_crypt_alg.alg_string) {
2560                 unsigned ivsize, blocksize;
2561                 struct journal_completion comp;
2562
2563                 comp.ic = ic;
2564                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2565                 if (IS_ERR(ic->journal_crypt)) {
2566                         *error = "Invalid journal cipher";
2567                         r = PTR_ERR(ic->journal_crypt);
2568                         ic->journal_crypt = NULL;
2569                         goto bad;
2570                 }
2571                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2572                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2573
2574                 if (ic->journal_crypt_alg.key) {
2575                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2576                                                    ic->journal_crypt_alg.key_size);
2577                         if (r) {
2578                                 *error = "Error setting encryption key";
2579                                 goto bad;
2580                         }
2581                 }
2582                 DEBUG_print("cipher %s, block size %u iv size %u\n",
2583                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2584
2585                 ic->journal_io = dm_integrity_alloc_page_list(ic);
2586                 if (!ic->journal_io) {
2587                         *error = "Could not allocate memory for journal io";
2588                         r = -ENOMEM;
2589                         goto bad;
2590                 }
2591
2592                 if (blocksize == 1) {
2593                         struct scatterlist *sg;
2594                         SKCIPHER_REQUEST_ON_STACK(req, ic->journal_crypt);
2595                         unsigned char iv[ivsize];
2596                         skcipher_request_set_tfm(req, ic->journal_crypt);
2597
2598                         ic->journal_xor = dm_integrity_alloc_page_list(ic);
2599                         if (!ic->journal_xor) {
2600                                 *error = "Could not allocate memory for journal xor";
2601                                 r = -ENOMEM;
2602                                 goto bad;
2603                         }
2604
2605                         sg = kvmalloc((ic->journal_pages + 1) * sizeof(struct scatterlist), GFP_KERNEL);
2606                         if (!sg) {
2607                                 *error = "Unable to allocate sg list";
2608                                 r = -ENOMEM;
2609                                 goto bad;
2610                         }
2611                         sg_init_table(sg, ic->journal_pages + 1);
2612                         for (i = 0; i < ic->journal_pages; i++) {
2613                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
2614                                 clear_page(va);
2615                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
2616                         }
2617                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2618                         memset(iv, 0x00, ivsize);
2619
2620                         skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, iv);
2621                         comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
2622                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2623                         if (do_crypt(true, req, &comp))
2624                                 wait_for_completion(&comp.comp);
2625                         kvfree(sg);
2626                         r = dm_integrity_failed(ic);
2627                         if (r) {
2628                                 *error = "Unable to encrypt journal";
2629                                 goto bad;
2630                         }
2631                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2632
2633                         crypto_free_skcipher(ic->journal_crypt);
2634                         ic->journal_crypt = NULL;
2635                 } else {
2636                         SKCIPHER_REQUEST_ON_STACK(req, ic->journal_crypt);
2637                         unsigned char iv[ivsize];
2638                         unsigned crypt_len = roundup(ivsize, blocksize);
2639
2640                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2641                         if (!crypt_data) {
2642                                 *error = "Unable to allocate crypt data";
2643                                 r = -ENOMEM;
2644                                 goto bad;
2645                         }
2646
2647                         skcipher_request_set_tfm(req, ic->journal_crypt);
2648
2649                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2650                         if (!ic->journal_scatterlist) {
2651                                 *error = "Unable to allocate sg list";
2652                                 r = -ENOMEM;
2653                                 goto bad;
2654                         }
2655                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2656                         if (!ic->journal_io_scatterlist) {
2657                                 *error = "Unable to allocate sg list";
2658                                 r = -ENOMEM;
2659                                 goto bad;
2660                         }
2661                         ic->sk_requests = kvmalloc(ic->journal_sections * sizeof(struct skcipher_request *), GFP_KERNEL | __GFP_ZERO);
2662                         if (!ic->sk_requests) {
2663                                 *error = "Unable to allocate sk requests";
2664                                 r = -ENOMEM;
2665                                 goto bad;
2666                         }
2667                         for (i = 0; i < ic->journal_sections; i++) {
2668                                 struct scatterlist sg;
2669                                 struct skcipher_request *section_req;
2670                                 __u32 section_le = cpu_to_le32(i);
2671
2672                                 memset(iv, 0x00, ivsize);
2673                                 memset(crypt_data, 0x00, crypt_len);
2674                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
2675
2676                                 sg_init_one(&sg, crypt_data, crypt_len);
2677                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, iv);
2678                                 comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
2679                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2680                                 if (do_crypt(true, req, &comp))
2681                                         wait_for_completion(&comp.comp);
2682
2683                                 r = dm_integrity_failed(ic);
2684                                 if (r) {
2685                                         *error = "Unable to generate iv";
2686                                         goto bad;
2687                                 }
2688
2689                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2690                                 if (!section_req) {
2691                                         *error = "Unable to allocate crypt request";
2692                                         r = -ENOMEM;
2693                                         goto bad;
2694                                 }
2695                                 section_req->iv = kmalloc(ivsize * 2, GFP_KERNEL);
2696                                 if (!section_req->iv) {
2697                                         skcipher_request_free(section_req);
2698                                         *error = "Unable to allocate iv";
2699                                         r = -ENOMEM;
2700                                         goto bad;
2701                                 }
2702                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
2703                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
2704                                 ic->sk_requests[i] = section_req;
2705                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
2706                         }
2707                 }
2708         }
2709
2710         for (i = 0; i < N_COMMIT_IDS; i++) {
2711                 unsigned j;
2712 retest_commit_id:
2713                 for (j = 0; j < i; j++) {
2714                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
2715                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
2716                                 goto retest_commit_id;
2717                         }
2718                 }
2719                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
2720         }
2721
2722         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
2723         if (journal_tree_size > ULONG_MAX) {
2724                 *error = "Journal doesn't fit into memory";
2725                 r = -ENOMEM;
2726                 goto bad;
2727         }
2728         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
2729         if (!ic->journal_tree) {
2730                 *error = "Could not allocate memory for journal tree";
2731                 r = -ENOMEM;
2732         }
2733 bad:
2734         kfree(crypt_data);
2735         return r;
2736 }
2737
2738 /*
2739  * Construct a integrity mapping
2740  *
2741  * Arguments:
2742  *      device
2743  *      offset from the start of the device
2744  *      tag size
2745  *      D - direct writes, J - journal writes, R - recovery mode
2746  *      number of optional arguments
2747  *      optional arguments:
2748  *              journal_sectors
2749  *              interleave_sectors
2750  *              buffer_sectors
2751  *              journal_watermark
2752  *              commit_time
2753  *              internal_hash
2754  *              journal_crypt
2755  *              journal_mac
2756  *              block_size
2757  */
2758 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
2759 {
2760         struct dm_integrity_c *ic;
2761         char dummy;
2762         int r;
2763         unsigned extra_args;
2764         struct dm_arg_set as;
2765         static struct dm_arg _args[] = {
2766                 {0, 9, "Invalid number of feature args"},
2767         };
2768         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
2769         bool should_write_sb;
2770         __u64 threshold;
2771         unsigned long long start;
2772
2773 #define DIRECT_ARGUMENTS        4
2774
2775         if (argc <= DIRECT_ARGUMENTS) {
2776                 ti->error = "Invalid argument count";
2777                 return -EINVAL;
2778         }
2779
2780         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
2781         if (!ic) {
2782                 ti->error = "Cannot allocate integrity context";
2783                 return -ENOMEM;
2784         }
2785         ti->private = ic;
2786         ti->per_io_data_size = sizeof(struct dm_integrity_io);
2787
2788         ic->in_progress = RB_ROOT;
2789         init_waitqueue_head(&ic->endio_wait);
2790         bio_list_init(&ic->flush_bio_list);
2791         init_waitqueue_head(&ic->copy_to_journal_wait);
2792         init_completion(&ic->crypto_backoff);
2793
2794         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
2795         if (r) {
2796                 ti->error = "Device lookup failed";
2797                 goto bad;
2798         }
2799
2800         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
2801                 ti->error = "Invalid starting offset";
2802                 r = -EINVAL;
2803                 goto bad;
2804         }
2805         ic->start = start;
2806
2807         if (strcmp(argv[2], "-")) {
2808                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
2809                         ti->error = "Invalid tag size";
2810                         r = -EINVAL;
2811                         goto bad;
2812                 }
2813         }
2814
2815         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
2816                 ic->mode = argv[3][0];
2817         else {
2818                 ti->error = "Invalid mode (expecting J, D, R)";
2819                 r = -EINVAL;
2820                 goto bad;
2821         }
2822
2823         ic->device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
2824         journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
2825                         ic->device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
2826         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2827         buffer_sectors = DEFAULT_BUFFER_SECTORS;
2828         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
2829         sync_msec = DEFAULT_SYNC_MSEC;
2830         ic->sectors_per_block = 1;
2831
2832         as.argc = argc - DIRECT_ARGUMENTS;
2833         as.argv = argv + DIRECT_ARGUMENTS;
2834         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
2835         if (r)
2836                 goto bad;
2837
2838         while (extra_args--) {
2839                 const char *opt_string;
2840                 unsigned val;
2841                 opt_string = dm_shift_arg(&as);
2842                 if (!opt_string) {
2843                         r = -EINVAL;
2844                         ti->error = "Not enough feature arguments";
2845                         goto bad;
2846                 }
2847                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
2848                         journal_sectors = val;
2849                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
2850                         interleave_sectors = val;
2851                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
2852                         buffer_sectors = val;
2853                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
2854                         journal_watermark = val;
2855                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
2856                         sync_msec = val;
2857                 else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
2858                         if (val < 1 << SECTOR_SHIFT ||
2859                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
2860                             (val & (val -1))) {
2861                                 r = -EINVAL;
2862                                 ti->error = "Invalid block_size argument";
2863                                 goto bad;
2864                         }
2865                         ic->sectors_per_block = val >> SECTOR_SHIFT;
2866                 } else if (!memcmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
2867                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
2868                                             "Invalid internal_hash argument");
2869                         if (r)
2870                                 goto bad;
2871                 } else if (!memcmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
2872                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
2873                                             "Invalid journal_crypt argument");
2874                         if (r)
2875                                 goto bad;
2876                 } else if (!memcmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
2877                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
2878                                             "Invalid journal_mac argument");
2879                         if (r)
2880                                 goto bad;
2881                 } else {
2882                         r = -EINVAL;
2883                         ti->error = "Invalid argument";
2884                         goto bad;
2885                 }
2886         }
2887
2888         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
2889                     "Invalid internal hash", "Error setting internal hash key");
2890         if (r)
2891                 goto bad;
2892
2893         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
2894                     "Invalid journal mac", "Error setting journal mac key");
2895         if (r)
2896                 goto bad;
2897
2898         if (!ic->tag_size) {
2899                 if (!ic->internal_hash) {
2900                         ti->error = "Unknown tag size";
2901                         r = -EINVAL;
2902                         goto bad;
2903                 }
2904                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
2905         }
2906         if (ic->tag_size > MAX_TAG_SIZE) {
2907                 ti->error = "Too big tag size";
2908                 r = -EINVAL;
2909                 goto bad;
2910         }
2911         if (!(ic->tag_size & (ic->tag_size - 1)))
2912                 ic->log2_tag_size = __ffs(ic->tag_size);
2913         else
2914                 ic->log2_tag_size = -1;
2915
2916         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
2917         ic->autocommit_msec = sync_msec;
2918         setup_timer(&ic->autocommit_timer, autocommit_fn, (unsigned long)ic);
2919
2920         ic->io = dm_io_client_create();
2921         if (IS_ERR(ic->io)) {
2922                 r = PTR_ERR(ic->io);
2923                 ic->io = NULL;
2924                 ti->error = "Cannot allocate dm io";
2925                 goto bad;
2926         }
2927
2928         ic->journal_io_mempool = mempool_create_slab_pool(JOURNAL_IO_MEMPOOL, journal_io_cache);
2929         if (!ic->journal_io_mempool) {
2930                 r = -ENOMEM;
2931                 ti->error = "Cannot allocate mempool";
2932                 goto bad;
2933         }
2934
2935         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
2936                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
2937         if (!ic->metadata_wq) {
2938                 ti->error = "Cannot allocate workqueue";
2939                 r = -ENOMEM;
2940                 goto bad;
2941         }
2942
2943         /*
2944          * If this workqueue were percpu, it would cause bio reordering
2945          * and reduced performance.
2946          */
2947         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
2948         if (!ic->wait_wq) {
2949                 ti->error = "Cannot allocate workqueue";
2950                 r = -ENOMEM;
2951                 goto bad;
2952         }
2953
2954         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
2955         if (!ic->commit_wq) {
2956                 ti->error = "Cannot allocate workqueue";
2957                 r = -ENOMEM;
2958                 goto bad;
2959         }
2960         INIT_WORK(&ic->commit_work, integrity_commit);
2961
2962         if (ic->mode == 'J') {
2963                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
2964                 if (!ic->writer_wq) {
2965                         ti->error = "Cannot allocate workqueue";
2966                         r = -ENOMEM;
2967                         goto bad;
2968                 }
2969                 INIT_WORK(&ic->writer_work, integrity_writer);
2970         }
2971
2972         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
2973         if (!ic->sb) {
2974                 r = -ENOMEM;
2975                 ti->error = "Cannot allocate superblock area";
2976                 goto bad;
2977         }
2978
2979         r = sync_rw_sb(ic, REQ_OP_READ, 0);
2980         if (r) {
2981                 ti->error = "Error reading superblock";
2982                 goto bad;
2983         }
2984         should_write_sb = false;
2985         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
2986                 if (ic->mode != 'R') {
2987                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
2988                                 r = -EINVAL;
2989                                 ti->error = "The device is not initialized";
2990                                 goto bad;
2991                         }
2992                 }
2993
2994                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
2995                 if (r) {
2996                         ti->error = "Could not initialize superblock";
2997                         goto bad;
2998                 }
2999                 if (ic->mode != 'R')
3000                         should_write_sb = true;
3001         }
3002
3003         if (ic->sb->version != SB_VERSION) {
3004                 r = -EINVAL;
3005                 ti->error = "Unknown version";
3006                 goto bad;
3007         }
3008         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3009                 r = -EINVAL;
3010                 ti->error = "Tag size doesn't match the information in superblock";
3011                 goto bad;
3012         }
3013         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3014                 r = -EINVAL;
3015                 ti->error = "Block size doesn't match the information in superblock";
3016                 goto bad;
3017         }
3018         /* make sure that ti->max_io_len doesn't overflow */
3019         if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3020             ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3021                 r = -EINVAL;
3022                 ti->error = "Invalid interleave_sectors in the superblock";
3023                 goto bad;
3024         }
3025         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3026         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3027                 /* test for overflow */
3028                 r = -EINVAL;
3029                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3030                 goto bad;
3031         }
3032         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3033                 r = -EINVAL;
3034                 ti->error = "Journal mac mismatch";
3035                 goto bad;
3036         }
3037         r = calculate_device_limits(ic);
3038         if (r) {
3039                 ti->error = "The device is too small";
3040                 goto bad;
3041         }
3042
3043         if (!buffer_sectors)
3044                 buffer_sectors = 1;
3045         ic->log2_buffer_sectors = min3((int)__fls(buffer_sectors), (int)__ffs(ic->metadata_run), 31 - SECTOR_SHIFT);
3046
3047         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3048         threshold += 50;
3049         do_div(threshold, 100);
3050         ic->free_sectors_threshold = threshold;
3051
3052         DEBUG_print("initialized:\n");
3053         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3054         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
3055         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3056         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
3057         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
3058         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3059         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
3060         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3061         DEBUG_print("   device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
3062         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
3063         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
3064         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
3065         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3066                     (unsigned long long)ic->provided_data_sectors);
3067         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3068
3069         ic->bufio = dm_bufio_client_create(ic->dev->bdev, 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors),
3070                                            1, 0, NULL, NULL);
3071         if (IS_ERR(ic->bufio)) {
3072                 r = PTR_ERR(ic->bufio);
3073                 ti->error = "Cannot initialize dm-bufio";
3074                 ic->bufio = NULL;
3075                 goto bad;
3076         }
3077         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3078
3079         if (ic->mode != 'R') {
3080                 r = create_journal(ic, &ti->error);
3081                 if (r)
3082                         goto bad;
3083         }
3084
3085         if (should_write_sb) {
3086                 int r;
3087
3088                 init_journal(ic, 0, ic->journal_sections, 0);
3089                 r = dm_integrity_failed(ic);
3090                 if (unlikely(r)) {
3091                         ti->error = "Error initializing journal";
3092                         goto bad;
3093                 }
3094                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3095                 if (r) {
3096                         ti->error = "Error initializing superblock";
3097                         goto bad;
3098                 }
3099                 ic->just_formatted = true;
3100         }
3101
3102         r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3103         if (r)
3104                 goto bad;
3105
3106         if (!ic->internal_hash)
3107                 dm_integrity_set(ti, ic);
3108
3109         ti->num_flush_bios = 1;
3110         ti->flush_supported = true;
3111
3112         return 0;
3113 bad:
3114         dm_integrity_dtr(ti);
3115         return r;
3116 }
3117
3118 static void dm_integrity_dtr(struct dm_target *ti)
3119 {
3120         struct dm_integrity_c *ic = ti->private;
3121
3122         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3123
3124         if (ic->metadata_wq)
3125                 destroy_workqueue(ic->metadata_wq);
3126         if (ic->wait_wq)
3127                 destroy_workqueue(ic->wait_wq);
3128         if (ic->commit_wq)
3129                 destroy_workqueue(ic->commit_wq);
3130         if (ic->writer_wq)
3131                 destroy_workqueue(ic->writer_wq);
3132         if (ic->bufio)
3133                 dm_bufio_client_destroy(ic->bufio);
3134         mempool_destroy(ic->journal_io_mempool);
3135         if (ic->io)
3136                 dm_io_client_destroy(ic->io);
3137         if (ic->dev)
3138                 dm_put_device(ti, ic->dev);
3139         dm_integrity_free_page_list(ic, ic->journal);
3140         dm_integrity_free_page_list(ic, ic->journal_io);
3141         dm_integrity_free_page_list(ic, ic->journal_xor);
3142         if (ic->journal_scatterlist)
3143                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3144         if (ic->journal_io_scatterlist)
3145                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3146         if (ic->sk_requests) {
3147                 unsigned i;
3148
3149                 for (i = 0; i < ic->journal_sections; i++) {
3150                         struct skcipher_request *req = ic->sk_requests[i];
3151                         if (req) {
3152                                 kzfree(req->iv);
3153                                 skcipher_request_free(req);
3154                         }
3155                 }
3156                 kvfree(ic->sk_requests);
3157         }
3158         kvfree(ic->journal_tree);
3159         if (ic->sb)
3160                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3161
3162         if (ic->internal_hash)
3163                 crypto_free_shash(ic->internal_hash);
3164         free_alg(&ic->internal_hash_alg);
3165
3166         if (ic->journal_crypt)
3167                 crypto_free_skcipher(ic->journal_crypt);
3168         free_alg(&ic->journal_crypt_alg);
3169
3170         if (ic->journal_mac)
3171                 crypto_free_shash(ic->journal_mac);
3172         free_alg(&ic->journal_mac_alg);
3173
3174         kfree(ic);
3175 }
3176
3177 static struct target_type integrity_target = {
3178         .name                   = "integrity",
3179         .version                = {1, 0, 0},
3180         .module                 = THIS_MODULE,
3181         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3182         .ctr                    = dm_integrity_ctr,
3183         .dtr                    = dm_integrity_dtr,
3184         .map                    = dm_integrity_map,
3185         .postsuspend            = dm_integrity_postsuspend,
3186         .resume                 = dm_integrity_resume,
3187         .status                 = dm_integrity_status,
3188         .iterate_devices        = dm_integrity_iterate_devices,
3189         .io_hints               = dm_integrity_io_hints,
3190 };
3191
3192 int __init dm_integrity_init(void)
3193 {
3194         int r;
3195
3196         journal_io_cache = kmem_cache_create("integrity_journal_io",
3197                                              sizeof(struct journal_io), 0, 0, NULL);
3198         if (!journal_io_cache) {
3199                 DMERR("can't allocate journal io cache");
3200                 return -ENOMEM;
3201         }
3202
3203         r = dm_register_target(&integrity_target);
3204
3205         if (r < 0)
3206                 DMERR("register failed %d", r);
3207
3208         return r;
3209 }
3210
3211 void dm_integrity_exit(void)
3212 {
3213         dm_unregister_target(&integrity_target);
3214         kmem_cache_destroy(journal_io_cache);
3215 }
3216
3217 module_init(dm_integrity_init);
3218 module_exit(dm_integrity_exit);
3219
3220 MODULE_AUTHOR("Milan Broz");
3221 MODULE_AUTHOR("Mikulas Patocka");
3222 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3223 MODULE_LICENSE("GPL");