xref: /openbmc/linux/drivers/md/dm-integrity.c (revision 82e6fdd6)
1 /*
2  * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
3  * Copyright (C) 2016-2017 Milan Broz
4  * Copyright (C) 2016-2017 Mikulas Patocka
5  *
6  * This file is released under the GPL.
7  */
8 
9 #include <linux/compiler.h>
10 #include <linux/module.h>
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/vmalloc.h>
14 #include <linux/sort.h>
15 #include <linux/rbtree.h>
16 #include <linux/delay.h>
17 #include <linux/random.h>
18 #include <crypto/hash.h>
19 #include <crypto/skcipher.h>
20 #include <linux/async_tx.h>
21 #include "dm-bufio.h"
22 
23 #define DM_MSG_PREFIX "integrity"
24 
25 #define DEFAULT_INTERLEAVE_SECTORS	32768
26 #define DEFAULT_JOURNAL_SIZE_FACTOR	7
27 #define DEFAULT_BUFFER_SECTORS		128
28 #define DEFAULT_JOURNAL_WATERMARK	50
29 #define DEFAULT_SYNC_MSEC		10000
30 #define DEFAULT_MAX_JOURNAL_SECTORS	131072
31 #define MIN_LOG2_INTERLEAVE_SECTORS	3
32 #define MAX_LOG2_INTERLEAVE_SECTORS	31
33 #define METADATA_WORKQUEUE_MAX_ACTIVE	16
34 
35 /*
36  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
37  * so it should not be enabled in the official kernel
38  */
39 //#define DEBUG_PRINT
40 //#define INTERNAL_VERIFY
41 
42 /*
43  * On disk structures
44  */
45 
46 #define SB_MAGIC			"integrt"
47 #define SB_VERSION			1
48 #define SB_SECTORS			8
49 #define MAX_SECTORS_PER_BLOCK		8
50 
51 struct superblock {
52 	__u8 magic[8];
53 	__u8 version;
54 	__u8 log2_interleave_sectors;
55 	__u16 integrity_tag_size;
56 	__u32 journal_sections;
57 	__u64 provided_data_sectors;	/* userspace uses this value */
58 	__u32 flags;
59 	__u8 log2_sectors_per_block;
60 };
61 
62 #define SB_FLAG_HAVE_JOURNAL_MAC	0x1
63 
64 #define	JOURNAL_ENTRY_ROUNDUP		8
65 
66 typedef __u64 commit_id_t;
67 #define JOURNAL_MAC_PER_SECTOR		8
68 
69 struct journal_entry {
70 	union {
71 		struct {
72 			__u32 sector_lo;
73 			__u32 sector_hi;
74 		} s;
75 		__u64 sector;
76 	} u;
77 	commit_id_t last_bytes[0];
78 	/* __u8 tag[0]; */
79 };
80 
81 #define journal_entry_tag(ic, je)		((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
82 
83 #if BITS_PER_LONG == 64
84 #define journal_entry_set_sector(je, x)		do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
85 #define journal_entry_get_sector(je)		le64_to_cpu((je)->u.sector)
86 #elif defined(CONFIG_LBDAF)
87 #define journal_entry_set_sector(je, x)		do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
88 #define journal_entry_get_sector(je)		le64_to_cpu((je)->u.sector)
89 #else
90 #define journal_entry_set_sector(je, x)		do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32(0)); } while (0)
91 #define journal_entry_get_sector(je)		le32_to_cpu((je)->u.s.sector_lo)
92 #endif
93 #define journal_entry_is_unused(je)		((je)->u.s.sector_hi == cpu_to_le32(-1))
94 #define journal_entry_set_unused(je)		do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
95 #define journal_entry_is_inprogress(je)		((je)->u.s.sector_hi == cpu_to_le32(-2))
96 #define journal_entry_set_inprogress(je)	do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
97 
98 #define JOURNAL_BLOCK_SECTORS		8
99 #define JOURNAL_SECTOR_DATA		((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
100 #define JOURNAL_MAC_SIZE		(JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
101 
102 struct journal_sector {
103 	__u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
104 	__u8 mac[JOURNAL_MAC_PER_SECTOR];
105 	commit_id_t commit_id;
106 };
107 
108 #define MAX_TAG_SIZE			(JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
109 
110 #define METADATA_PADDING_SECTORS	8
111 
112 #define N_COMMIT_IDS			4
113 
114 static unsigned char prev_commit_seq(unsigned char seq)
115 {
116 	return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
117 }
118 
119 static unsigned char next_commit_seq(unsigned char seq)
120 {
121 	return (seq + 1) % N_COMMIT_IDS;
122 }
123 
124 /*
125  * In-memory structures
126  */
127 
128 struct journal_node {
129 	struct rb_node node;
130 	sector_t sector;
131 };
132 
133 struct alg_spec {
134 	char *alg_string;
135 	char *key_string;
136 	__u8 *key;
137 	unsigned key_size;
138 };
139 
140 struct dm_integrity_c {
141 	struct dm_dev *dev;
142 	unsigned tag_size;
143 	__s8 log2_tag_size;
144 	sector_t start;
145 	mempool_t *journal_io_mempool;
146 	struct dm_io_client *io;
147 	struct dm_bufio_client *bufio;
148 	struct workqueue_struct *metadata_wq;
149 	struct superblock *sb;
150 	unsigned journal_pages;
151 	struct page_list *journal;
152 	struct page_list *journal_io;
153 	struct page_list *journal_xor;
154 
155 	struct crypto_skcipher *journal_crypt;
156 	struct scatterlist **journal_scatterlist;
157 	struct scatterlist **journal_io_scatterlist;
158 	struct skcipher_request **sk_requests;
159 
160 	struct crypto_shash *journal_mac;
161 
162 	struct journal_node *journal_tree;
163 	struct rb_root journal_tree_root;
164 
165 	sector_t provided_data_sectors;
166 
167 	unsigned short journal_entry_size;
168 	unsigned char journal_entries_per_sector;
169 	unsigned char journal_section_entries;
170 	unsigned short journal_section_sectors;
171 	unsigned journal_sections;
172 	unsigned journal_entries;
173 	sector_t device_sectors;
174 	unsigned initial_sectors;
175 	unsigned metadata_run;
176 	__s8 log2_metadata_run;
177 	__u8 log2_buffer_sectors;
178 	__u8 sectors_per_block;
179 
180 	unsigned char mode;
181 	bool suspending;
182 
183 	int failed;
184 
185 	struct crypto_shash *internal_hash;
186 
187 	/* these variables are locked with endio_wait.lock */
188 	struct rb_root in_progress;
189 	wait_queue_head_t endio_wait;
190 	struct workqueue_struct *wait_wq;
191 
192 	unsigned char commit_seq;
193 	commit_id_t commit_ids[N_COMMIT_IDS];
194 
195 	unsigned committed_section;
196 	unsigned n_committed_sections;
197 
198 	unsigned uncommitted_section;
199 	unsigned n_uncommitted_sections;
200 
201 	unsigned free_section;
202 	unsigned char free_section_entry;
203 	unsigned free_sectors;
204 
205 	unsigned free_sectors_threshold;
206 
207 	struct workqueue_struct *commit_wq;
208 	struct work_struct commit_work;
209 
210 	struct workqueue_struct *writer_wq;
211 	struct work_struct writer_work;
212 
213 	struct bio_list flush_bio_list;
214 
215 	unsigned long autocommit_jiffies;
216 	struct timer_list autocommit_timer;
217 	unsigned autocommit_msec;
218 
219 	wait_queue_head_t copy_to_journal_wait;
220 
221 	struct completion crypto_backoff;
222 
223 	bool journal_uptodate;
224 	bool just_formatted;
225 
226 	struct alg_spec internal_hash_alg;
227 	struct alg_spec journal_crypt_alg;
228 	struct alg_spec journal_mac_alg;
229 
230 	atomic64_t number_of_mismatches;
231 };
232 
233 struct dm_integrity_range {
234 	sector_t logical_sector;
235 	unsigned n_sectors;
236 	struct rb_node node;
237 };
238 
239 struct dm_integrity_io {
240 	struct work_struct work;
241 
242 	struct dm_integrity_c *ic;
243 	bool write;
244 	bool fua;
245 
246 	struct dm_integrity_range range;
247 
248 	sector_t metadata_block;
249 	unsigned metadata_offset;
250 
251 	atomic_t in_flight;
252 	blk_status_t bi_status;
253 
254 	struct completion *completion;
255 
256 	struct gendisk *orig_bi_disk;
257 	u8 orig_bi_partno;
258 	bio_end_io_t *orig_bi_end_io;
259 	struct bio_integrity_payload *orig_bi_integrity;
260 	struct bvec_iter orig_bi_iter;
261 };
262 
263 struct journal_completion {
264 	struct dm_integrity_c *ic;
265 	atomic_t in_flight;
266 	struct completion comp;
267 };
268 
269 struct journal_io {
270 	struct dm_integrity_range range;
271 	struct journal_completion *comp;
272 };
273 
274 static struct kmem_cache *journal_io_cache;
275 
276 #define JOURNAL_IO_MEMPOOL	32
277 
278 #ifdef DEBUG_PRINT
279 #define DEBUG_print(x, ...)	printk(KERN_DEBUG x, ##__VA_ARGS__)
280 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
281 {
282 	va_list args;
283 	va_start(args, msg);
284 	vprintk(msg, args);
285 	va_end(args);
286 	if (len)
287 		pr_cont(":");
288 	while (len) {
289 		pr_cont(" %02x", *bytes);
290 		bytes++;
291 		len--;
292 	}
293 	pr_cont("\n");
294 }
295 #define DEBUG_bytes(bytes, len, msg, ...)	__DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
296 #else
297 #define DEBUG_print(x, ...)			do { } while (0)
298 #define DEBUG_bytes(bytes, len, msg, ...)	do { } while (0)
299 #endif
300 
301 /*
302  * DM Integrity profile, protection is performed layer above (dm-crypt)
303  */
304 static const struct blk_integrity_profile dm_integrity_profile = {
305 	.name			= "DM-DIF-EXT-TAG",
306 	.generate_fn		= NULL,
307 	.verify_fn		= NULL,
308 };
309 
310 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
311 static void integrity_bio_wait(struct work_struct *w);
312 static void dm_integrity_dtr(struct dm_target *ti);
313 
314 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
315 {
316 	if (err == -EILSEQ)
317 		atomic64_inc(&ic->number_of_mismatches);
318 	if (!cmpxchg(&ic->failed, 0, err))
319 		DMERR("Error on %s: %d", msg, err);
320 }
321 
322 static int dm_integrity_failed(struct dm_integrity_c *ic)
323 {
324 	return READ_ONCE(ic->failed);
325 }
326 
327 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
328 					  unsigned j, unsigned char seq)
329 {
330 	/*
331 	 * Xor the number with section and sector, so that if a piece of
332 	 * journal is written at wrong place, it is detected.
333 	 */
334 	return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
335 }
336 
337 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
338 				sector_t *area, sector_t *offset)
339 {
340 	__u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
341 
342 	*area = data_sector >> log2_interleave_sectors;
343 	*offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
344 }
345 
346 #define sector_to_block(ic, n)						\
347 do {									\
348 	BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));		\
349 	(n) >>= (ic)->sb->log2_sectors_per_block;			\
350 } while (0)
351 
352 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
353 					    sector_t offset, unsigned *metadata_offset)
354 {
355 	__u64 ms;
356 	unsigned mo;
357 
358 	ms = area << ic->sb->log2_interleave_sectors;
359 	if (likely(ic->log2_metadata_run >= 0))
360 		ms += area << ic->log2_metadata_run;
361 	else
362 		ms += area * ic->metadata_run;
363 	ms >>= ic->log2_buffer_sectors;
364 
365 	sector_to_block(ic, offset);
366 
367 	if (likely(ic->log2_tag_size >= 0)) {
368 		ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
369 		mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
370 	} else {
371 		ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
372 		mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
373 	}
374 	*metadata_offset = mo;
375 	return ms;
376 }
377 
378 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
379 {
380 	sector_t result;
381 
382 	result = area << ic->sb->log2_interleave_sectors;
383 	if (likely(ic->log2_metadata_run >= 0))
384 		result += (area + 1) << ic->log2_metadata_run;
385 	else
386 		result += (area + 1) * ic->metadata_run;
387 
388 	result += (sector_t)ic->initial_sectors + offset;
389 	return result;
390 }
391 
392 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
393 {
394 	if (unlikely(*sec_ptr >= ic->journal_sections))
395 		*sec_ptr -= ic->journal_sections;
396 }
397 
398 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
399 {
400 	struct dm_io_request io_req;
401 	struct dm_io_region io_loc;
402 
403 	io_req.bi_op = op;
404 	io_req.bi_op_flags = op_flags;
405 	io_req.mem.type = DM_IO_KMEM;
406 	io_req.mem.ptr.addr = ic->sb;
407 	io_req.notify.fn = NULL;
408 	io_req.client = ic->io;
409 	io_loc.bdev = ic->dev->bdev;
410 	io_loc.sector = ic->start;
411 	io_loc.count = SB_SECTORS;
412 
413 	return dm_io(&io_req, 1, &io_loc, NULL);
414 }
415 
416 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
417 				 bool e, const char *function)
418 {
419 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
420 	unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
421 
422 	if (unlikely(section >= ic->journal_sections) ||
423 	    unlikely(offset >= limit)) {
424 		printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
425 			function, section, offset, ic->journal_sections, limit);
426 		BUG();
427 	}
428 #endif
429 }
430 
431 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
432 			       unsigned *pl_index, unsigned *pl_offset)
433 {
434 	unsigned sector;
435 
436 	access_journal_check(ic, section, offset, false, "page_list_location");
437 
438 	sector = section * ic->journal_section_sectors + offset;
439 
440 	*pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
441 	*pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
442 }
443 
444 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
445 					       unsigned section, unsigned offset, unsigned *n_sectors)
446 {
447 	unsigned pl_index, pl_offset;
448 	char *va;
449 
450 	page_list_location(ic, section, offset, &pl_index, &pl_offset);
451 
452 	if (n_sectors)
453 		*n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
454 
455 	va = lowmem_page_address(pl[pl_index].page);
456 
457 	return (struct journal_sector *)(va + pl_offset);
458 }
459 
460 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
461 {
462 	return access_page_list(ic, ic->journal, section, offset, NULL);
463 }
464 
465 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
466 {
467 	unsigned rel_sector, offset;
468 	struct journal_sector *js;
469 
470 	access_journal_check(ic, section, n, true, "access_journal_entry");
471 
472 	rel_sector = n % JOURNAL_BLOCK_SECTORS;
473 	offset = n / JOURNAL_BLOCK_SECTORS;
474 
475 	js = access_journal(ic, section, rel_sector);
476 	return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
477 }
478 
479 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
480 {
481 	n <<= ic->sb->log2_sectors_per_block;
482 
483 	n += JOURNAL_BLOCK_SECTORS;
484 
485 	access_journal_check(ic, section, n, false, "access_journal_data");
486 
487 	return access_journal(ic, section, n);
488 }
489 
490 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
491 {
492 	SHASH_DESC_ON_STACK(desc, ic->journal_mac);
493 	int r;
494 	unsigned j, size;
495 
496 	desc->tfm = ic->journal_mac;
497 	desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
498 
499 	r = crypto_shash_init(desc);
500 	if (unlikely(r)) {
501 		dm_integrity_io_error(ic, "crypto_shash_init", r);
502 		goto err;
503 	}
504 
505 	for (j = 0; j < ic->journal_section_entries; j++) {
506 		struct journal_entry *je = access_journal_entry(ic, section, j);
507 		r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
508 		if (unlikely(r)) {
509 			dm_integrity_io_error(ic, "crypto_shash_update", r);
510 			goto err;
511 		}
512 	}
513 
514 	size = crypto_shash_digestsize(ic->journal_mac);
515 
516 	if (likely(size <= JOURNAL_MAC_SIZE)) {
517 		r = crypto_shash_final(desc, result);
518 		if (unlikely(r)) {
519 			dm_integrity_io_error(ic, "crypto_shash_final", r);
520 			goto err;
521 		}
522 		memset(result + size, 0, JOURNAL_MAC_SIZE - size);
523 	} else {
524 		__u8 digest[size];
525 		r = crypto_shash_final(desc, digest);
526 		if (unlikely(r)) {
527 			dm_integrity_io_error(ic, "crypto_shash_final", r);
528 			goto err;
529 		}
530 		memcpy(result, digest, JOURNAL_MAC_SIZE);
531 	}
532 
533 	return;
534 err:
535 	memset(result, 0, JOURNAL_MAC_SIZE);
536 }
537 
538 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
539 {
540 	__u8 result[JOURNAL_MAC_SIZE];
541 	unsigned j;
542 
543 	if (!ic->journal_mac)
544 		return;
545 
546 	section_mac(ic, section, result);
547 
548 	for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
549 		struct journal_sector *js = access_journal(ic, section, j);
550 
551 		if (likely(wr))
552 			memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
553 		else {
554 			if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
555 				dm_integrity_io_error(ic, "journal mac", -EILSEQ);
556 		}
557 	}
558 }
559 
560 static void complete_journal_op(void *context)
561 {
562 	struct journal_completion *comp = context;
563 	BUG_ON(!atomic_read(&comp->in_flight));
564 	if (likely(atomic_dec_and_test(&comp->in_flight)))
565 		complete(&comp->comp);
566 }
567 
568 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
569 			unsigned n_sections, struct journal_completion *comp)
570 {
571 	struct async_submit_ctl submit;
572 	size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
573 	unsigned pl_index, pl_offset, section_index;
574 	struct page_list *source_pl, *target_pl;
575 
576 	if (likely(encrypt)) {
577 		source_pl = ic->journal;
578 		target_pl = ic->journal_io;
579 	} else {
580 		source_pl = ic->journal_io;
581 		target_pl = ic->journal;
582 	}
583 
584 	page_list_location(ic, section, 0, &pl_index, &pl_offset);
585 
586 	atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
587 
588 	init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
589 
590 	section_index = pl_index;
591 
592 	do {
593 		size_t this_step;
594 		struct page *src_pages[2];
595 		struct page *dst_page;
596 
597 		while (unlikely(pl_index == section_index)) {
598 			unsigned dummy;
599 			if (likely(encrypt))
600 				rw_section_mac(ic, section, true);
601 			section++;
602 			n_sections--;
603 			if (!n_sections)
604 				break;
605 			page_list_location(ic, section, 0, &section_index, &dummy);
606 		}
607 
608 		this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
609 		dst_page = target_pl[pl_index].page;
610 		src_pages[0] = source_pl[pl_index].page;
611 		src_pages[1] = ic->journal_xor[pl_index].page;
612 
613 		async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
614 
615 		pl_index++;
616 		pl_offset = 0;
617 		n_bytes -= this_step;
618 	} while (n_bytes);
619 
620 	BUG_ON(n_sections);
621 
622 	async_tx_issue_pending_all();
623 }
624 
625 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
626 {
627 	struct journal_completion *comp = req->data;
628 	if (unlikely(err)) {
629 		if (likely(err == -EINPROGRESS)) {
630 			complete(&comp->ic->crypto_backoff);
631 			return;
632 		}
633 		dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
634 	}
635 	complete_journal_op(comp);
636 }
637 
638 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
639 {
640 	int r;
641 	skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
642 				      complete_journal_encrypt, comp);
643 	if (likely(encrypt))
644 		r = crypto_skcipher_encrypt(req);
645 	else
646 		r = crypto_skcipher_decrypt(req);
647 	if (likely(!r))
648 		return false;
649 	if (likely(r == -EINPROGRESS))
650 		return true;
651 	if (likely(r == -EBUSY)) {
652 		wait_for_completion(&comp->ic->crypto_backoff);
653 		reinit_completion(&comp->ic->crypto_backoff);
654 		return true;
655 	}
656 	dm_integrity_io_error(comp->ic, "encrypt", r);
657 	return false;
658 }
659 
660 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
661 			  unsigned n_sections, struct journal_completion *comp)
662 {
663 	struct scatterlist **source_sg;
664 	struct scatterlist **target_sg;
665 
666 	atomic_add(2, &comp->in_flight);
667 
668 	if (likely(encrypt)) {
669 		source_sg = ic->journal_scatterlist;
670 		target_sg = ic->journal_io_scatterlist;
671 	} else {
672 		source_sg = ic->journal_io_scatterlist;
673 		target_sg = ic->journal_scatterlist;
674 	}
675 
676 	do {
677 		struct skcipher_request *req;
678 		unsigned ivsize;
679 		char *iv;
680 
681 		if (likely(encrypt))
682 			rw_section_mac(ic, section, true);
683 
684 		req = ic->sk_requests[section];
685 		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
686 		iv = req->iv;
687 
688 		memcpy(iv, iv + ivsize, ivsize);
689 
690 		req->src = source_sg[section];
691 		req->dst = target_sg[section];
692 
693 		if (unlikely(do_crypt(encrypt, req, comp)))
694 			atomic_inc(&comp->in_flight);
695 
696 		section++;
697 		n_sections--;
698 	} while (n_sections);
699 
700 	atomic_dec(&comp->in_flight);
701 	complete_journal_op(comp);
702 }
703 
704 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
705 			    unsigned n_sections, struct journal_completion *comp)
706 {
707 	if (ic->journal_xor)
708 		return xor_journal(ic, encrypt, section, n_sections, comp);
709 	else
710 		return crypt_journal(ic, encrypt, section, n_sections, comp);
711 }
712 
713 static void complete_journal_io(unsigned long error, void *context)
714 {
715 	struct journal_completion *comp = context;
716 	if (unlikely(error != 0))
717 		dm_integrity_io_error(comp->ic, "writing journal", -EIO);
718 	complete_journal_op(comp);
719 }
720 
721 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
722 		       unsigned n_sections, struct journal_completion *comp)
723 {
724 	struct dm_io_request io_req;
725 	struct dm_io_region io_loc;
726 	unsigned sector, n_sectors, pl_index, pl_offset;
727 	int r;
728 
729 	if (unlikely(dm_integrity_failed(ic))) {
730 		if (comp)
731 			complete_journal_io(-1UL, comp);
732 		return;
733 	}
734 
735 	sector = section * ic->journal_section_sectors;
736 	n_sectors = n_sections * ic->journal_section_sectors;
737 
738 	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
739 	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
740 
741 	io_req.bi_op = op;
742 	io_req.bi_op_flags = op_flags;
743 	io_req.mem.type = DM_IO_PAGE_LIST;
744 	if (ic->journal_io)
745 		io_req.mem.ptr.pl = &ic->journal_io[pl_index];
746 	else
747 		io_req.mem.ptr.pl = &ic->journal[pl_index];
748 	io_req.mem.offset = pl_offset;
749 	if (likely(comp != NULL)) {
750 		io_req.notify.fn = complete_journal_io;
751 		io_req.notify.context = comp;
752 	} else {
753 		io_req.notify.fn = NULL;
754 	}
755 	io_req.client = ic->io;
756 	io_loc.bdev = ic->dev->bdev;
757 	io_loc.sector = ic->start + SB_SECTORS + sector;
758 	io_loc.count = n_sectors;
759 
760 	r = dm_io(&io_req, 1, &io_loc, NULL);
761 	if (unlikely(r)) {
762 		dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
763 		if (comp) {
764 			WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
765 			complete_journal_io(-1UL, comp);
766 		}
767 	}
768 }
769 
770 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
771 {
772 	struct journal_completion io_comp;
773 	struct journal_completion crypt_comp_1;
774 	struct journal_completion crypt_comp_2;
775 	unsigned i;
776 
777 	io_comp.ic = ic;
778 	init_completion(&io_comp.comp);
779 
780 	if (commit_start + commit_sections <= ic->journal_sections) {
781 		io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
782 		if (ic->journal_io) {
783 			crypt_comp_1.ic = ic;
784 			init_completion(&crypt_comp_1.comp);
785 			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
786 			encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
787 			wait_for_completion_io(&crypt_comp_1.comp);
788 		} else {
789 			for (i = 0; i < commit_sections; i++)
790 				rw_section_mac(ic, commit_start + i, true);
791 		}
792 		rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
793 			   commit_sections, &io_comp);
794 	} else {
795 		unsigned to_end;
796 		io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
797 		to_end = ic->journal_sections - commit_start;
798 		if (ic->journal_io) {
799 			crypt_comp_1.ic = ic;
800 			init_completion(&crypt_comp_1.comp);
801 			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
802 			encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
803 			if (try_wait_for_completion(&crypt_comp_1.comp)) {
804 				rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
805 				reinit_completion(&crypt_comp_1.comp);
806 				crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
807 				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
808 				wait_for_completion_io(&crypt_comp_1.comp);
809 			} else {
810 				crypt_comp_2.ic = ic;
811 				init_completion(&crypt_comp_2.comp);
812 				crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
813 				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
814 				wait_for_completion_io(&crypt_comp_1.comp);
815 				rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
816 				wait_for_completion_io(&crypt_comp_2.comp);
817 			}
818 		} else {
819 			for (i = 0; i < to_end; i++)
820 				rw_section_mac(ic, commit_start + i, true);
821 			rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
822 			for (i = 0; i < commit_sections - to_end; i++)
823 				rw_section_mac(ic, i, true);
824 		}
825 		rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
826 	}
827 
828 	wait_for_completion_io(&io_comp.comp);
829 }
830 
831 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
832 			      unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
833 {
834 	struct dm_io_request io_req;
835 	struct dm_io_region io_loc;
836 	int r;
837 	unsigned sector, pl_index, pl_offset;
838 
839 	BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
840 
841 	if (unlikely(dm_integrity_failed(ic))) {
842 		fn(-1UL, data);
843 		return;
844 	}
845 
846 	sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
847 
848 	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
849 	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
850 
851 	io_req.bi_op = REQ_OP_WRITE;
852 	io_req.bi_op_flags = 0;
853 	io_req.mem.type = DM_IO_PAGE_LIST;
854 	io_req.mem.ptr.pl = &ic->journal[pl_index];
855 	io_req.mem.offset = pl_offset;
856 	io_req.notify.fn = fn;
857 	io_req.notify.context = data;
858 	io_req.client = ic->io;
859 	io_loc.bdev = ic->dev->bdev;
860 	io_loc.sector = ic->start + target;
861 	io_loc.count = n_sectors;
862 
863 	r = dm_io(&io_req, 1, &io_loc, NULL);
864 	if (unlikely(r)) {
865 		WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
866 		fn(-1UL, data);
867 	}
868 }
869 
870 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
871 {
872 	struct rb_node **n = &ic->in_progress.rb_node;
873 	struct rb_node *parent;
874 
875 	BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
876 
877 	parent = NULL;
878 
879 	while (*n) {
880 		struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
881 
882 		parent = *n;
883 		if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
884 			n = &range->node.rb_left;
885 		} else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
886 			n = &range->node.rb_right;
887 		} else {
888 			return false;
889 		}
890 	}
891 
892 	rb_link_node(&new_range->node, parent, n);
893 	rb_insert_color(&new_range->node, &ic->in_progress);
894 
895 	return true;
896 }
897 
898 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
899 {
900 	rb_erase(&range->node, &ic->in_progress);
901 	wake_up_locked(&ic->endio_wait);
902 }
903 
904 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
905 {
906 	unsigned long flags;
907 
908 	spin_lock_irqsave(&ic->endio_wait.lock, flags);
909 	remove_range_unlocked(ic, range);
910 	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
911 }
912 
913 static void init_journal_node(struct journal_node *node)
914 {
915 	RB_CLEAR_NODE(&node->node);
916 	node->sector = (sector_t)-1;
917 }
918 
919 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
920 {
921 	struct rb_node **link;
922 	struct rb_node *parent;
923 
924 	node->sector = sector;
925 	BUG_ON(!RB_EMPTY_NODE(&node->node));
926 
927 	link = &ic->journal_tree_root.rb_node;
928 	parent = NULL;
929 
930 	while (*link) {
931 		struct journal_node *j;
932 		parent = *link;
933 		j = container_of(parent, struct journal_node, node);
934 		if (sector < j->sector)
935 			link = &j->node.rb_left;
936 		else
937 			link = &j->node.rb_right;
938 	}
939 
940 	rb_link_node(&node->node, parent, link);
941 	rb_insert_color(&node->node, &ic->journal_tree_root);
942 }
943 
944 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
945 {
946 	BUG_ON(RB_EMPTY_NODE(&node->node));
947 	rb_erase(&node->node, &ic->journal_tree_root);
948 	init_journal_node(node);
949 }
950 
951 #define NOT_FOUND	(-1U)
952 
953 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
954 {
955 	struct rb_node *n = ic->journal_tree_root.rb_node;
956 	unsigned found = NOT_FOUND;
957 	*next_sector = (sector_t)-1;
958 	while (n) {
959 		struct journal_node *j = container_of(n, struct journal_node, node);
960 		if (sector == j->sector) {
961 			found = j - ic->journal_tree;
962 		}
963 		if (sector < j->sector) {
964 			*next_sector = j->sector;
965 			n = j->node.rb_left;
966 		} else {
967 			n = j->node.rb_right;
968 		}
969 	}
970 
971 	return found;
972 }
973 
974 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
975 {
976 	struct journal_node *node, *next_node;
977 	struct rb_node *next;
978 
979 	if (unlikely(pos >= ic->journal_entries))
980 		return false;
981 	node = &ic->journal_tree[pos];
982 	if (unlikely(RB_EMPTY_NODE(&node->node)))
983 		return false;
984 	if (unlikely(node->sector != sector))
985 		return false;
986 
987 	next = rb_next(&node->node);
988 	if (unlikely(!next))
989 		return true;
990 
991 	next_node = container_of(next, struct journal_node, node);
992 	return next_node->sector != sector;
993 }
994 
995 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
996 {
997 	struct rb_node *next;
998 	struct journal_node *next_node;
999 	unsigned next_section;
1000 
1001 	BUG_ON(RB_EMPTY_NODE(&node->node));
1002 
1003 	next = rb_next(&node->node);
1004 	if (unlikely(!next))
1005 		return false;
1006 
1007 	next_node = container_of(next, struct journal_node, node);
1008 
1009 	if (next_node->sector != node->sector)
1010 		return false;
1011 
1012 	next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1013 	if (next_section >= ic->committed_section &&
1014 	    next_section < ic->committed_section + ic->n_committed_sections)
1015 		return true;
1016 	if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1017 		return true;
1018 
1019 	return false;
1020 }
1021 
1022 #define TAG_READ	0
1023 #define TAG_WRITE	1
1024 #define TAG_CMP		2
1025 
1026 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1027 			       unsigned *metadata_offset, unsigned total_size, int op)
1028 {
1029 	do {
1030 		unsigned char *data, *dp;
1031 		struct dm_buffer *b;
1032 		unsigned to_copy;
1033 		int r;
1034 
1035 		r = dm_integrity_failed(ic);
1036 		if (unlikely(r))
1037 			return r;
1038 
1039 		data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1040 		if (unlikely(IS_ERR(data)))
1041 			return PTR_ERR(data);
1042 
1043 		to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1044 		dp = data + *metadata_offset;
1045 		if (op == TAG_READ) {
1046 			memcpy(tag, dp, to_copy);
1047 		} else if (op == TAG_WRITE) {
1048 			memcpy(dp, tag, to_copy);
1049 			dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1050 		} else  {
1051 			/* e.g.: op == TAG_CMP */
1052 			if (unlikely(memcmp(dp, tag, to_copy))) {
1053 				unsigned i;
1054 
1055 				for (i = 0; i < to_copy; i++) {
1056 					if (dp[i] != tag[i])
1057 						break;
1058 					total_size--;
1059 				}
1060 				dm_bufio_release(b);
1061 				return total_size;
1062 			}
1063 		}
1064 		dm_bufio_release(b);
1065 
1066 		tag += to_copy;
1067 		*metadata_offset += to_copy;
1068 		if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1069 			(*metadata_block)++;
1070 			*metadata_offset = 0;
1071 		}
1072 		total_size -= to_copy;
1073 	} while (unlikely(total_size));
1074 
1075 	return 0;
1076 }
1077 
1078 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1079 {
1080 	int r;
1081 	r = dm_bufio_write_dirty_buffers(ic->bufio);
1082 	if (unlikely(r))
1083 		dm_integrity_io_error(ic, "writing tags", r);
1084 }
1085 
1086 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1087 {
1088 	DECLARE_WAITQUEUE(wait, current);
1089 	__add_wait_queue(&ic->endio_wait, &wait);
1090 	__set_current_state(TASK_UNINTERRUPTIBLE);
1091 	spin_unlock_irq(&ic->endio_wait.lock);
1092 	io_schedule();
1093 	spin_lock_irq(&ic->endio_wait.lock);
1094 	__remove_wait_queue(&ic->endio_wait, &wait);
1095 }
1096 
1097 static void autocommit_fn(struct timer_list *t)
1098 {
1099 	struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1100 
1101 	if (likely(!dm_integrity_failed(ic)))
1102 		queue_work(ic->commit_wq, &ic->commit_work);
1103 }
1104 
1105 static void schedule_autocommit(struct dm_integrity_c *ic)
1106 {
1107 	if (!timer_pending(&ic->autocommit_timer))
1108 		mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1109 }
1110 
1111 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1112 {
1113 	struct bio *bio;
1114 	unsigned long flags;
1115 
1116 	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1117 	bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1118 	bio_list_add(&ic->flush_bio_list, bio);
1119 	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1120 
1121 	queue_work(ic->commit_wq, &ic->commit_work);
1122 }
1123 
1124 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1125 {
1126 	int r = dm_integrity_failed(ic);
1127 	if (unlikely(r) && !bio->bi_status)
1128 		bio->bi_status = errno_to_blk_status(r);
1129 	bio_endio(bio);
1130 }
1131 
1132 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1133 {
1134 	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1135 
1136 	if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1137 		submit_flush_bio(ic, dio);
1138 	else
1139 		do_endio(ic, bio);
1140 }
1141 
1142 static void dec_in_flight(struct dm_integrity_io *dio)
1143 {
1144 	if (atomic_dec_and_test(&dio->in_flight)) {
1145 		struct dm_integrity_c *ic = dio->ic;
1146 		struct bio *bio;
1147 
1148 		remove_range(ic, &dio->range);
1149 
1150 		if (unlikely(dio->write))
1151 			schedule_autocommit(ic);
1152 
1153 		bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1154 
1155 		if (unlikely(dio->bi_status) && !bio->bi_status)
1156 			bio->bi_status = dio->bi_status;
1157 		if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1158 			dio->range.logical_sector += dio->range.n_sectors;
1159 			bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1160 			INIT_WORK(&dio->work, integrity_bio_wait);
1161 			queue_work(ic->wait_wq, &dio->work);
1162 			return;
1163 		}
1164 		do_endio_flush(ic, dio);
1165 	}
1166 }
1167 
1168 static void integrity_end_io(struct bio *bio)
1169 {
1170 	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1171 
1172 	bio->bi_iter = dio->orig_bi_iter;
1173 	bio->bi_disk = dio->orig_bi_disk;
1174 	bio->bi_partno = dio->orig_bi_partno;
1175 	if (dio->orig_bi_integrity) {
1176 		bio->bi_integrity = dio->orig_bi_integrity;
1177 		bio->bi_opf |= REQ_INTEGRITY;
1178 	}
1179 	bio->bi_end_io = dio->orig_bi_end_io;
1180 
1181 	if (dio->completion)
1182 		complete(dio->completion);
1183 
1184 	dec_in_flight(dio);
1185 }
1186 
1187 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1188 				      const char *data, char *result)
1189 {
1190 	__u64 sector_le = cpu_to_le64(sector);
1191 	SHASH_DESC_ON_STACK(req, ic->internal_hash);
1192 	int r;
1193 	unsigned digest_size;
1194 
1195 	req->tfm = ic->internal_hash;
1196 	req->flags = 0;
1197 
1198 	r = crypto_shash_init(req);
1199 	if (unlikely(r < 0)) {
1200 		dm_integrity_io_error(ic, "crypto_shash_init", r);
1201 		goto failed;
1202 	}
1203 
1204 	r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1205 	if (unlikely(r < 0)) {
1206 		dm_integrity_io_error(ic, "crypto_shash_update", r);
1207 		goto failed;
1208 	}
1209 
1210 	r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1211 	if (unlikely(r < 0)) {
1212 		dm_integrity_io_error(ic, "crypto_shash_update", r);
1213 		goto failed;
1214 	}
1215 
1216 	r = crypto_shash_final(req, result);
1217 	if (unlikely(r < 0)) {
1218 		dm_integrity_io_error(ic, "crypto_shash_final", r);
1219 		goto failed;
1220 	}
1221 
1222 	digest_size = crypto_shash_digestsize(ic->internal_hash);
1223 	if (unlikely(digest_size < ic->tag_size))
1224 		memset(result + digest_size, 0, ic->tag_size - digest_size);
1225 
1226 	return;
1227 
1228 failed:
1229 	/* this shouldn't happen anyway, the hash functions have no reason to fail */
1230 	get_random_bytes(result, ic->tag_size);
1231 }
1232 
1233 static void integrity_metadata(struct work_struct *w)
1234 {
1235 	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1236 	struct dm_integrity_c *ic = dio->ic;
1237 
1238 	int r;
1239 
1240 	if (ic->internal_hash) {
1241 		struct bvec_iter iter;
1242 		struct bio_vec bv;
1243 		unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1244 		struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1245 		char *checksums;
1246 		unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1247 		char checksums_onstack[ic->tag_size + extra_space];
1248 		unsigned sectors_to_process = dio->range.n_sectors;
1249 		sector_t sector = dio->range.logical_sector;
1250 
1251 		if (unlikely(ic->mode == 'R'))
1252 			goto skip_io;
1253 
1254 		checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1255 				    GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1256 		if (!checksums)
1257 			checksums = checksums_onstack;
1258 
1259 		__bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1260 			unsigned pos;
1261 			char *mem, *checksums_ptr;
1262 
1263 again:
1264 			mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1265 			pos = 0;
1266 			checksums_ptr = checksums;
1267 			do {
1268 				integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1269 				checksums_ptr += ic->tag_size;
1270 				sectors_to_process -= ic->sectors_per_block;
1271 				pos += ic->sectors_per_block << SECTOR_SHIFT;
1272 				sector += ic->sectors_per_block;
1273 			} while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1274 			kunmap_atomic(mem);
1275 
1276 			r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1277 						checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1278 			if (unlikely(r)) {
1279 				if (r > 0) {
1280 					DMERR("Checksum failed at sector 0x%llx",
1281 					      (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1282 					r = -EILSEQ;
1283 					atomic64_inc(&ic->number_of_mismatches);
1284 				}
1285 				if (likely(checksums != checksums_onstack))
1286 					kfree(checksums);
1287 				goto error;
1288 			}
1289 
1290 			if (!sectors_to_process)
1291 				break;
1292 
1293 			if (unlikely(pos < bv.bv_len)) {
1294 				bv.bv_offset += pos;
1295 				bv.bv_len -= pos;
1296 				goto again;
1297 			}
1298 		}
1299 
1300 		if (likely(checksums != checksums_onstack))
1301 			kfree(checksums);
1302 	} else {
1303 		struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1304 
1305 		if (bip) {
1306 			struct bio_vec biv;
1307 			struct bvec_iter iter;
1308 			unsigned data_to_process = dio->range.n_sectors;
1309 			sector_to_block(ic, data_to_process);
1310 			data_to_process *= ic->tag_size;
1311 
1312 			bip_for_each_vec(biv, bip, iter) {
1313 				unsigned char *tag;
1314 				unsigned this_len;
1315 
1316 				BUG_ON(PageHighMem(biv.bv_page));
1317 				tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1318 				this_len = min(biv.bv_len, data_to_process);
1319 				r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1320 							this_len, !dio->write ? TAG_READ : TAG_WRITE);
1321 				if (unlikely(r))
1322 					goto error;
1323 				data_to_process -= this_len;
1324 				if (!data_to_process)
1325 					break;
1326 			}
1327 		}
1328 	}
1329 skip_io:
1330 	dec_in_flight(dio);
1331 	return;
1332 error:
1333 	dio->bi_status = errno_to_blk_status(r);
1334 	dec_in_flight(dio);
1335 }
1336 
1337 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1338 {
1339 	struct dm_integrity_c *ic = ti->private;
1340 	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1341 	struct bio_integrity_payload *bip;
1342 
1343 	sector_t area, offset;
1344 
1345 	dio->ic = ic;
1346 	dio->bi_status = 0;
1347 
1348 	if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1349 		submit_flush_bio(ic, dio);
1350 		return DM_MAPIO_SUBMITTED;
1351 	}
1352 
1353 	dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1354 	dio->write = bio_op(bio) == REQ_OP_WRITE;
1355 	dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1356 	if (unlikely(dio->fua)) {
1357 		/*
1358 		 * Don't pass down the FUA flag because we have to flush
1359 		 * disk cache anyway.
1360 		 */
1361 		bio->bi_opf &= ~REQ_FUA;
1362 	}
1363 	if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1364 		DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1365 		      (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1366 		      (unsigned long long)ic->provided_data_sectors);
1367 		return DM_MAPIO_KILL;
1368 	}
1369 	if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1370 		DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1371 		      ic->sectors_per_block,
1372 		      (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1373 		return DM_MAPIO_KILL;
1374 	}
1375 
1376 	if (ic->sectors_per_block > 1) {
1377 		struct bvec_iter iter;
1378 		struct bio_vec bv;
1379 		bio_for_each_segment(bv, bio, iter) {
1380 			if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1381 				DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1382 					bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1383 				return DM_MAPIO_KILL;
1384 			}
1385 		}
1386 	}
1387 
1388 	bip = bio_integrity(bio);
1389 	if (!ic->internal_hash) {
1390 		if (bip) {
1391 			unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1392 			if (ic->log2_tag_size >= 0)
1393 				wanted_tag_size <<= ic->log2_tag_size;
1394 			else
1395 				wanted_tag_size *= ic->tag_size;
1396 			if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1397 				DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1398 				return DM_MAPIO_KILL;
1399 			}
1400 		}
1401 	} else {
1402 		if (unlikely(bip != NULL)) {
1403 			DMERR("Unexpected integrity data when using internal hash");
1404 			return DM_MAPIO_KILL;
1405 		}
1406 	}
1407 
1408 	if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1409 		return DM_MAPIO_KILL;
1410 
1411 	get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1412 	dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1413 	bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1414 
1415 	dm_integrity_map_continue(dio, true);
1416 	return DM_MAPIO_SUBMITTED;
1417 }
1418 
1419 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1420 				 unsigned journal_section, unsigned journal_entry)
1421 {
1422 	struct dm_integrity_c *ic = dio->ic;
1423 	sector_t logical_sector;
1424 	unsigned n_sectors;
1425 
1426 	logical_sector = dio->range.logical_sector;
1427 	n_sectors = dio->range.n_sectors;
1428 	do {
1429 		struct bio_vec bv = bio_iovec(bio);
1430 		char *mem;
1431 
1432 		if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1433 			bv.bv_len = n_sectors << SECTOR_SHIFT;
1434 		n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1435 		bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1436 retry_kmap:
1437 		mem = kmap_atomic(bv.bv_page);
1438 		if (likely(dio->write))
1439 			flush_dcache_page(bv.bv_page);
1440 
1441 		do {
1442 			struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1443 
1444 			if (unlikely(!dio->write)) {
1445 				struct journal_sector *js;
1446 				char *mem_ptr;
1447 				unsigned s;
1448 
1449 				if (unlikely(journal_entry_is_inprogress(je))) {
1450 					flush_dcache_page(bv.bv_page);
1451 					kunmap_atomic(mem);
1452 
1453 					__io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1454 					goto retry_kmap;
1455 				}
1456 				smp_rmb();
1457 				BUG_ON(journal_entry_get_sector(je) != logical_sector);
1458 				js = access_journal_data(ic, journal_section, journal_entry);
1459 				mem_ptr = mem + bv.bv_offset;
1460 				s = 0;
1461 				do {
1462 					memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1463 					*(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1464 					js++;
1465 					mem_ptr += 1 << SECTOR_SHIFT;
1466 				} while (++s < ic->sectors_per_block);
1467 #ifdef INTERNAL_VERIFY
1468 				if (ic->internal_hash) {
1469 					char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1470 
1471 					integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1472 					if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1473 						DMERR("Checksum failed when reading from journal, at sector 0x%llx",
1474 						      (unsigned long long)logical_sector);
1475 					}
1476 				}
1477 #endif
1478 			}
1479 
1480 			if (!ic->internal_hash) {
1481 				struct bio_integrity_payload *bip = bio_integrity(bio);
1482 				unsigned tag_todo = ic->tag_size;
1483 				char *tag_ptr = journal_entry_tag(ic, je);
1484 
1485 				if (bip) do {
1486 					struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1487 					unsigned tag_now = min(biv.bv_len, tag_todo);
1488 					char *tag_addr;
1489 					BUG_ON(PageHighMem(biv.bv_page));
1490 					tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1491 					if (likely(dio->write))
1492 						memcpy(tag_ptr, tag_addr, tag_now);
1493 					else
1494 						memcpy(tag_addr, tag_ptr, tag_now);
1495 					bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1496 					tag_ptr += tag_now;
1497 					tag_todo -= tag_now;
1498 				} while (unlikely(tag_todo)); else {
1499 					if (likely(dio->write))
1500 						memset(tag_ptr, 0, tag_todo);
1501 				}
1502 			}
1503 
1504 			if (likely(dio->write)) {
1505 				struct journal_sector *js;
1506 				unsigned s;
1507 
1508 				js = access_journal_data(ic, journal_section, journal_entry);
1509 				memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1510 
1511 				s = 0;
1512 				do {
1513 					je->last_bytes[s] = js[s].commit_id;
1514 				} while (++s < ic->sectors_per_block);
1515 
1516 				if (ic->internal_hash) {
1517 					unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1518 					if (unlikely(digest_size > ic->tag_size)) {
1519 						char checksums_onstack[digest_size];
1520 						integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1521 						memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1522 					} else
1523 						integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1524 				}
1525 
1526 				journal_entry_set_sector(je, logical_sector);
1527 			}
1528 			logical_sector += ic->sectors_per_block;
1529 
1530 			journal_entry++;
1531 			if (unlikely(journal_entry == ic->journal_section_entries)) {
1532 				journal_entry = 0;
1533 				journal_section++;
1534 				wraparound_section(ic, &journal_section);
1535 			}
1536 
1537 			bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1538 		} while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1539 
1540 		if (unlikely(!dio->write))
1541 			flush_dcache_page(bv.bv_page);
1542 		kunmap_atomic(mem);
1543 	} while (n_sectors);
1544 
1545 	if (likely(dio->write)) {
1546 		smp_mb();
1547 		if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1548 			wake_up(&ic->copy_to_journal_wait);
1549 		if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1550 			queue_work(ic->commit_wq, &ic->commit_work);
1551 		} else {
1552 			schedule_autocommit(ic);
1553 		}
1554 	} else {
1555 		remove_range(ic, &dio->range);
1556 	}
1557 
1558 	if (unlikely(bio->bi_iter.bi_size)) {
1559 		sector_t area, offset;
1560 
1561 		dio->range.logical_sector = logical_sector;
1562 		get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1563 		dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1564 		return true;
1565 	}
1566 
1567 	return false;
1568 }
1569 
1570 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1571 {
1572 	struct dm_integrity_c *ic = dio->ic;
1573 	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1574 	unsigned journal_section, journal_entry;
1575 	unsigned journal_read_pos;
1576 	struct completion read_comp;
1577 	bool need_sync_io = ic->internal_hash && !dio->write;
1578 
1579 	if (need_sync_io && from_map) {
1580 		INIT_WORK(&dio->work, integrity_bio_wait);
1581 		queue_work(ic->metadata_wq, &dio->work);
1582 		return;
1583 	}
1584 
1585 lock_retry:
1586 	spin_lock_irq(&ic->endio_wait.lock);
1587 retry:
1588 	if (unlikely(dm_integrity_failed(ic))) {
1589 		spin_unlock_irq(&ic->endio_wait.lock);
1590 		do_endio(ic, bio);
1591 		return;
1592 	}
1593 	dio->range.n_sectors = bio_sectors(bio);
1594 	journal_read_pos = NOT_FOUND;
1595 	if (likely(ic->mode == 'J')) {
1596 		if (dio->write) {
1597 			unsigned next_entry, i, pos;
1598 			unsigned ws, we, range_sectors;
1599 
1600 			dio->range.n_sectors = min(dio->range.n_sectors,
1601 						   ic->free_sectors << ic->sb->log2_sectors_per_block);
1602 			if (unlikely(!dio->range.n_sectors))
1603 				goto sleep;
1604 			range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1605 			ic->free_sectors -= range_sectors;
1606 			journal_section = ic->free_section;
1607 			journal_entry = ic->free_section_entry;
1608 
1609 			next_entry = ic->free_section_entry + range_sectors;
1610 			ic->free_section_entry = next_entry % ic->journal_section_entries;
1611 			ic->free_section += next_entry / ic->journal_section_entries;
1612 			ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1613 			wraparound_section(ic, &ic->free_section);
1614 
1615 			pos = journal_section * ic->journal_section_entries + journal_entry;
1616 			ws = journal_section;
1617 			we = journal_entry;
1618 			i = 0;
1619 			do {
1620 				struct journal_entry *je;
1621 
1622 				add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1623 				pos++;
1624 				if (unlikely(pos >= ic->journal_entries))
1625 					pos = 0;
1626 
1627 				je = access_journal_entry(ic, ws, we);
1628 				BUG_ON(!journal_entry_is_unused(je));
1629 				journal_entry_set_inprogress(je);
1630 				we++;
1631 				if (unlikely(we == ic->journal_section_entries)) {
1632 					we = 0;
1633 					ws++;
1634 					wraparound_section(ic, &ws);
1635 				}
1636 			} while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1637 
1638 			spin_unlock_irq(&ic->endio_wait.lock);
1639 			goto journal_read_write;
1640 		} else {
1641 			sector_t next_sector;
1642 			journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1643 			if (likely(journal_read_pos == NOT_FOUND)) {
1644 				if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1645 					dio->range.n_sectors = next_sector - dio->range.logical_sector;
1646 			} else {
1647 				unsigned i;
1648 				unsigned jp = journal_read_pos + 1;
1649 				for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1650 					if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1651 						break;
1652 				}
1653 				dio->range.n_sectors = i;
1654 			}
1655 		}
1656 	}
1657 	if (unlikely(!add_new_range(ic, &dio->range))) {
1658 		/*
1659 		 * We must not sleep in the request routine because it could
1660 		 * stall bios on current->bio_list.
1661 		 * So, we offload the bio to a workqueue if we have to sleep.
1662 		 */
1663 sleep:
1664 		if (from_map) {
1665 			spin_unlock_irq(&ic->endio_wait.lock);
1666 			INIT_WORK(&dio->work, integrity_bio_wait);
1667 			queue_work(ic->wait_wq, &dio->work);
1668 			return;
1669 		} else {
1670 			sleep_on_endio_wait(ic);
1671 			goto retry;
1672 		}
1673 	}
1674 	spin_unlock_irq(&ic->endio_wait.lock);
1675 
1676 	if (unlikely(journal_read_pos != NOT_FOUND)) {
1677 		journal_section = journal_read_pos / ic->journal_section_entries;
1678 		journal_entry = journal_read_pos % ic->journal_section_entries;
1679 		goto journal_read_write;
1680 	}
1681 
1682 	dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1683 
1684 	if (need_sync_io) {
1685 		init_completion(&read_comp);
1686 		dio->completion = &read_comp;
1687 	} else
1688 		dio->completion = NULL;
1689 
1690 	dio->orig_bi_iter = bio->bi_iter;
1691 
1692 	dio->orig_bi_disk = bio->bi_disk;
1693 	dio->orig_bi_partno = bio->bi_partno;
1694 	bio_set_dev(bio, ic->dev->bdev);
1695 
1696 	dio->orig_bi_integrity = bio_integrity(bio);
1697 	bio->bi_integrity = NULL;
1698 	bio->bi_opf &= ~REQ_INTEGRITY;
1699 
1700 	dio->orig_bi_end_io = bio->bi_end_io;
1701 	bio->bi_end_io = integrity_end_io;
1702 
1703 	bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1704 	bio->bi_iter.bi_sector += ic->start;
1705 	generic_make_request(bio);
1706 
1707 	if (need_sync_io) {
1708 		wait_for_completion_io(&read_comp);
1709 		if (likely(!bio->bi_status))
1710 			integrity_metadata(&dio->work);
1711 		else
1712 			dec_in_flight(dio);
1713 
1714 	} else {
1715 		INIT_WORK(&dio->work, integrity_metadata);
1716 		queue_work(ic->metadata_wq, &dio->work);
1717 	}
1718 
1719 	return;
1720 
1721 journal_read_write:
1722 	if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1723 		goto lock_retry;
1724 
1725 	do_endio_flush(ic, dio);
1726 }
1727 
1728 
1729 static void integrity_bio_wait(struct work_struct *w)
1730 {
1731 	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1732 
1733 	dm_integrity_map_continue(dio, false);
1734 }
1735 
1736 static void pad_uncommitted(struct dm_integrity_c *ic)
1737 {
1738 	if (ic->free_section_entry) {
1739 		ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1740 		ic->free_section_entry = 0;
1741 		ic->free_section++;
1742 		wraparound_section(ic, &ic->free_section);
1743 		ic->n_uncommitted_sections++;
1744 	}
1745 	WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1746 		(ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1747 }
1748 
1749 static void integrity_commit(struct work_struct *w)
1750 {
1751 	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1752 	unsigned commit_start, commit_sections;
1753 	unsigned i, j, n;
1754 	struct bio *flushes;
1755 
1756 	del_timer(&ic->autocommit_timer);
1757 
1758 	spin_lock_irq(&ic->endio_wait.lock);
1759 	flushes = bio_list_get(&ic->flush_bio_list);
1760 	if (unlikely(ic->mode != 'J')) {
1761 		spin_unlock_irq(&ic->endio_wait.lock);
1762 		dm_integrity_flush_buffers(ic);
1763 		goto release_flush_bios;
1764 	}
1765 
1766 	pad_uncommitted(ic);
1767 	commit_start = ic->uncommitted_section;
1768 	commit_sections = ic->n_uncommitted_sections;
1769 	spin_unlock_irq(&ic->endio_wait.lock);
1770 
1771 	if (!commit_sections)
1772 		goto release_flush_bios;
1773 
1774 	i = commit_start;
1775 	for (n = 0; n < commit_sections; n++) {
1776 		for (j = 0; j < ic->journal_section_entries; j++) {
1777 			struct journal_entry *je;
1778 			je = access_journal_entry(ic, i, j);
1779 			io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1780 		}
1781 		for (j = 0; j < ic->journal_section_sectors; j++) {
1782 			struct journal_sector *js;
1783 			js = access_journal(ic, i, j);
1784 			js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1785 		}
1786 		i++;
1787 		if (unlikely(i >= ic->journal_sections))
1788 			ic->commit_seq = next_commit_seq(ic->commit_seq);
1789 		wraparound_section(ic, &i);
1790 	}
1791 	smp_rmb();
1792 
1793 	write_journal(ic, commit_start, commit_sections);
1794 
1795 	spin_lock_irq(&ic->endio_wait.lock);
1796 	ic->uncommitted_section += commit_sections;
1797 	wraparound_section(ic, &ic->uncommitted_section);
1798 	ic->n_uncommitted_sections -= commit_sections;
1799 	ic->n_committed_sections += commit_sections;
1800 	spin_unlock_irq(&ic->endio_wait.lock);
1801 
1802 	if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1803 		queue_work(ic->writer_wq, &ic->writer_work);
1804 
1805 release_flush_bios:
1806 	while (flushes) {
1807 		struct bio *next = flushes->bi_next;
1808 		flushes->bi_next = NULL;
1809 		do_endio(ic, flushes);
1810 		flushes = next;
1811 	}
1812 }
1813 
1814 static void complete_copy_from_journal(unsigned long error, void *context)
1815 {
1816 	struct journal_io *io = context;
1817 	struct journal_completion *comp = io->comp;
1818 	struct dm_integrity_c *ic = comp->ic;
1819 	remove_range(ic, &io->range);
1820 	mempool_free(io, ic->journal_io_mempool);
1821 	if (unlikely(error != 0))
1822 		dm_integrity_io_error(ic, "copying from journal", -EIO);
1823 	complete_journal_op(comp);
1824 }
1825 
1826 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1827 			       struct journal_entry *je)
1828 {
1829 	unsigned s = 0;
1830 	do {
1831 		js->commit_id = je->last_bytes[s];
1832 		js++;
1833 	} while (++s < ic->sectors_per_block);
1834 }
1835 
1836 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1837 			     unsigned write_sections, bool from_replay)
1838 {
1839 	unsigned i, j, n;
1840 	struct journal_completion comp;
1841 	struct blk_plug plug;
1842 
1843 	blk_start_plug(&plug);
1844 
1845 	comp.ic = ic;
1846 	comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1847 	init_completion(&comp.comp);
1848 
1849 	i = write_start;
1850 	for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1851 #ifndef INTERNAL_VERIFY
1852 		if (unlikely(from_replay))
1853 #endif
1854 			rw_section_mac(ic, i, false);
1855 		for (j = 0; j < ic->journal_section_entries; j++) {
1856 			struct journal_entry *je = access_journal_entry(ic, i, j);
1857 			sector_t sec, area, offset;
1858 			unsigned k, l, next_loop;
1859 			sector_t metadata_block;
1860 			unsigned metadata_offset;
1861 			struct journal_io *io;
1862 
1863 			if (journal_entry_is_unused(je))
1864 				continue;
1865 			BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1866 			sec = journal_entry_get_sector(je);
1867 			if (unlikely(from_replay)) {
1868 				if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1869 					dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1870 					sec &= ~(sector_t)(ic->sectors_per_block - 1);
1871 				}
1872 			}
1873 			get_area_and_offset(ic, sec, &area, &offset);
1874 			restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1875 			for (k = j + 1; k < ic->journal_section_entries; k++) {
1876 				struct journal_entry *je2 = access_journal_entry(ic, i, k);
1877 				sector_t sec2, area2, offset2;
1878 				if (journal_entry_is_unused(je2))
1879 					break;
1880 				BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1881 				sec2 = journal_entry_get_sector(je2);
1882 				get_area_and_offset(ic, sec2, &area2, &offset2);
1883 				if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1884 					break;
1885 				restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1886 			}
1887 			next_loop = k - 1;
1888 
1889 			io = mempool_alloc(ic->journal_io_mempool, GFP_NOIO);
1890 			io->comp = &comp;
1891 			io->range.logical_sector = sec;
1892 			io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1893 
1894 			spin_lock_irq(&ic->endio_wait.lock);
1895 			while (unlikely(!add_new_range(ic, &io->range)))
1896 				sleep_on_endio_wait(ic);
1897 
1898 			if (likely(!from_replay)) {
1899 				struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1900 
1901 				/* don't write if there is newer committed sector */
1902 				while (j < k && find_newer_committed_node(ic, &section_node[j])) {
1903 					struct journal_entry *je2 = access_journal_entry(ic, i, j);
1904 
1905 					journal_entry_set_unused(je2);
1906 					remove_journal_node(ic, &section_node[j]);
1907 					j++;
1908 					sec += ic->sectors_per_block;
1909 					offset += ic->sectors_per_block;
1910 				}
1911 				while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
1912 					struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
1913 
1914 					journal_entry_set_unused(je2);
1915 					remove_journal_node(ic, &section_node[k - 1]);
1916 					k--;
1917 				}
1918 				if (j == k) {
1919 					remove_range_unlocked(ic, &io->range);
1920 					spin_unlock_irq(&ic->endio_wait.lock);
1921 					mempool_free(io, ic->journal_io_mempool);
1922 					goto skip_io;
1923 				}
1924 				for (l = j; l < k; l++) {
1925 					remove_journal_node(ic, &section_node[l]);
1926 				}
1927 			}
1928 			spin_unlock_irq(&ic->endio_wait.lock);
1929 
1930 			metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
1931 			for (l = j; l < k; l++) {
1932 				int r;
1933 				struct journal_entry *je2 = access_journal_entry(ic, i, l);
1934 
1935 				if (
1936 #ifndef INTERNAL_VERIFY
1937 				    unlikely(from_replay) &&
1938 #endif
1939 				    ic->internal_hash) {
1940 					char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1941 
1942 					integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
1943 								  (char *)access_journal_data(ic, i, l), test_tag);
1944 					if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
1945 						dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
1946 				}
1947 
1948 				journal_entry_set_unused(je2);
1949 				r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
1950 							ic->tag_size, TAG_WRITE);
1951 				if (unlikely(r)) {
1952 					dm_integrity_io_error(ic, "reading tags", r);
1953 				}
1954 			}
1955 
1956 			atomic_inc(&comp.in_flight);
1957 			copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
1958 					  (k - j) << ic->sb->log2_sectors_per_block,
1959 					  get_data_sector(ic, area, offset),
1960 					  complete_copy_from_journal, io);
1961 skip_io:
1962 			j = next_loop;
1963 		}
1964 	}
1965 
1966 	dm_bufio_write_dirty_buffers_async(ic->bufio);
1967 
1968 	blk_finish_plug(&plug);
1969 
1970 	complete_journal_op(&comp);
1971 	wait_for_completion_io(&comp.comp);
1972 
1973 	dm_integrity_flush_buffers(ic);
1974 }
1975 
1976 static void integrity_writer(struct work_struct *w)
1977 {
1978 	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
1979 	unsigned write_start, write_sections;
1980 
1981 	unsigned prev_free_sectors;
1982 
1983 	/* the following test is not needed, but it tests the replay code */
1984 	if (READ_ONCE(ic->suspending))
1985 		return;
1986 
1987 	spin_lock_irq(&ic->endio_wait.lock);
1988 	write_start = ic->committed_section;
1989 	write_sections = ic->n_committed_sections;
1990 	spin_unlock_irq(&ic->endio_wait.lock);
1991 
1992 	if (!write_sections)
1993 		return;
1994 
1995 	do_journal_write(ic, write_start, write_sections, false);
1996 
1997 	spin_lock_irq(&ic->endio_wait.lock);
1998 
1999 	ic->committed_section += write_sections;
2000 	wraparound_section(ic, &ic->committed_section);
2001 	ic->n_committed_sections -= write_sections;
2002 
2003 	prev_free_sectors = ic->free_sectors;
2004 	ic->free_sectors += write_sections * ic->journal_section_entries;
2005 	if (unlikely(!prev_free_sectors))
2006 		wake_up_locked(&ic->endio_wait);
2007 
2008 	spin_unlock_irq(&ic->endio_wait.lock);
2009 }
2010 
2011 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2012 			 unsigned n_sections, unsigned char commit_seq)
2013 {
2014 	unsigned i, j, n;
2015 
2016 	if (!n_sections)
2017 		return;
2018 
2019 	for (n = 0; n < n_sections; n++) {
2020 		i = start_section + n;
2021 		wraparound_section(ic, &i);
2022 		for (j = 0; j < ic->journal_section_sectors; j++) {
2023 			struct journal_sector *js = access_journal(ic, i, j);
2024 			memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2025 			js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2026 		}
2027 		for (j = 0; j < ic->journal_section_entries; j++) {
2028 			struct journal_entry *je = access_journal_entry(ic, i, j);
2029 			journal_entry_set_unused(je);
2030 		}
2031 	}
2032 
2033 	write_journal(ic, start_section, n_sections);
2034 }
2035 
2036 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2037 {
2038 	unsigned char k;
2039 	for (k = 0; k < N_COMMIT_IDS; k++) {
2040 		if (dm_integrity_commit_id(ic, i, j, k) == id)
2041 			return k;
2042 	}
2043 	dm_integrity_io_error(ic, "journal commit id", -EIO);
2044 	return -EIO;
2045 }
2046 
2047 static void replay_journal(struct dm_integrity_c *ic)
2048 {
2049 	unsigned i, j;
2050 	bool used_commit_ids[N_COMMIT_IDS];
2051 	unsigned max_commit_id_sections[N_COMMIT_IDS];
2052 	unsigned write_start, write_sections;
2053 	unsigned continue_section;
2054 	bool journal_empty;
2055 	unsigned char unused, last_used, want_commit_seq;
2056 
2057 	if (ic->mode == 'R')
2058 		return;
2059 
2060 	if (ic->journal_uptodate)
2061 		return;
2062 
2063 	last_used = 0;
2064 	write_start = 0;
2065 
2066 	if (!ic->just_formatted) {
2067 		DEBUG_print("reading journal\n");
2068 		rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2069 		if (ic->journal_io)
2070 			DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2071 		if (ic->journal_io) {
2072 			struct journal_completion crypt_comp;
2073 			crypt_comp.ic = ic;
2074 			init_completion(&crypt_comp.comp);
2075 			crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2076 			encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2077 			wait_for_completion(&crypt_comp.comp);
2078 		}
2079 		DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2080 	}
2081 
2082 	if (dm_integrity_failed(ic))
2083 		goto clear_journal;
2084 
2085 	journal_empty = true;
2086 	memset(used_commit_ids, 0, sizeof used_commit_ids);
2087 	memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2088 	for (i = 0; i < ic->journal_sections; i++) {
2089 		for (j = 0; j < ic->journal_section_sectors; j++) {
2090 			int k;
2091 			struct journal_sector *js = access_journal(ic, i, j);
2092 			k = find_commit_seq(ic, i, j, js->commit_id);
2093 			if (k < 0)
2094 				goto clear_journal;
2095 			used_commit_ids[k] = true;
2096 			max_commit_id_sections[k] = i;
2097 		}
2098 		if (journal_empty) {
2099 			for (j = 0; j < ic->journal_section_entries; j++) {
2100 				struct journal_entry *je = access_journal_entry(ic, i, j);
2101 				if (!journal_entry_is_unused(je)) {
2102 					journal_empty = false;
2103 					break;
2104 				}
2105 			}
2106 		}
2107 	}
2108 
2109 	if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2110 		unused = N_COMMIT_IDS - 1;
2111 		while (unused && !used_commit_ids[unused - 1])
2112 			unused--;
2113 	} else {
2114 		for (unused = 0; unused < N_COMMIT_IDS; unused++)
2115 			if (!used_commit_ids[unused])
2116 				break;
2117 		if (unused == N_COMMIT_IDS) {
2118 			dm_integrity_io_error(ic, "journal commit ids", -EIO);
2119 			goto clear_journal;
2120 		}
2121 	}
2122 	DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2123 		    unused, used_commit_ids[0], used_commit_ids[1],
2124 		    used_commit_ids[2], used_commit_ids[3]);
2125 
2126 	last_used = prev_commit_seq(unused);
2127 	want_commit_seq = prev_commit_seq(last_used);
2128 
2129 	if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2130 		journal_empty = true;
2131 
2132 	write_start = max_commit_id_sections[last_used] + 1;
2133 	if (unlikely(write_start >= ic->journal_sections))
2134 		want_commit_seq = next_commit_seq(want_commit_seq);
2135 	wraparound_section(ic, &write_start);
2136 
2137 	i = write_start;
2138 	for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2139 		for (j = 0; j < ic->journal_section_sectors; j++) {
2140 			struct journal_sector *js = access_journal(ic, i, j);
2141 
2142 			if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2143 				/*
2144 				 * This could be caused by crash during writing.
2145 				 * We won't replay the inconsistent part of the
2146 				 * journal.
2147 				 */
2148 				DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2149 					    i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2150 				goto brk;
2151 			}
2152 		}
2153 		i++;
2154 		if (unlikely(i >= ic->journal_sections))
2155 			want_commit_seq = next_commit_seq(want_commit_seq);
2156 		wraparound_section(ic, &i);
2157 	}
2158 brk:
2159 
2160 	if (!journal_empty) {
2161 		DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2162 			    write_sections, write_start, want_commit_seq);
2163 		do_journal_write(ic, write_start, write_sections, true);
2164 	}
2165 
2166 	if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2167 		continue_section = write_start;
2168 		ic->commit_seq = want_commit_seq;
2169 		DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2170 	} else {
2171 		unsigned s;
2172 		unsigned char erase_seq;
2173 clear_journal:
2174 		DEBUG_print("clearing journal\n");
2175 
2176 		erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2177 		s = write_start;
2178 		init_journal(ic, s, 1, erase_seq);
2179 		s++;
2180 		wraparound_section(ic, &s);
2181 		if (ic->journal_sections >= 2) {
2182 			init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2183 			s += ic->journal_sections - 2;
2184 			wraparound_section(ic, &s);
2185 			init_journal(ic, s, 1, erase_seq);
2186 		}
2187 
2188 		continue_section = 0;
2189 		ic->commit_seq = next_commit_seq(erase_seq);
2190 	}
2191 
2192 	ic->committed_section = continue_section;
2193 	ic->n_committed_sections = 0;
2194 
2195 	ic->uncommitted_section = continue_section;
2196 	ic->n_uncommitted_sections = 0;
2197 
2198 	ic->free_section = continue_section;
2199 	ic->free_section_entry = 0;
2200 	ic->free_sectors = ic->journal_entries;
2201 
2202 	ic->journal_tree_root = RB_ROOT;
2203 	for (i = 0; i < ic->journal_entries; i++)
2204 		init_journal_node(&ic->journal_tree[i]);
2205 }
2206 
2207 static void dm_integrity_postsuspend(struct dm_target *ti)
2208 {
2209 	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2210 
2211 	del_timer_sync(&ic->autocommit_timer);
2212 
2213 	ic->suspending = true;
2214 
2215 	queue_work(ic->commit_wq, &ic->commit_work);
2216 	drain_workqueue(ic->commit_wq);
2217 
2218 	if (ic->mode == 'J') {
2219 		drain_workqueue(ic->writer_wq);
2220 		dm_integrity_flush_buffers(ic);
2221 	}
2222 
2223 	ic->suspending = false;
2224 
2225 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2226 
2227 	ic->journal_uptodate = true;
2228 }
2229 
2230 static void dm_integrity_resume(struct dm_target *ti)
2231 {
2232 	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2233 
2234 	replay_journal(ic);
2235 }
2236 
2237 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2238 				unsigned status_flags, char *result, unsigned maxlen)
2239 {
2240 	struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2241 	unsigned arg_count;
2242 	size_t sz = 0;
2243 
2244 	switch (type) {
2245 	case STATUSTYPE_INFO:
2246 		DMEMIT("%llu", (unsigned long long)atomic64_read(&ic->number_of_mismatches));
2247 		break;
2248 
2249 	case STATUSTYPE_TABLE: {
2250 		__u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2251 		watermark_percentage += ic->journal_entries / 2;
2252 		do_div(watermark_percentage, ic->journal_entries);
2253 		arg_count = 5;
2254 		arg_count += ic->sectors_per_block != 1;
2255 		arg_count += !!ic->internal_hash_alg.alg_string;
2256 		arg_count += !!ic->journal_crypt_alg.alg_string;
2257 		arg_count += !!ic->journal_mac_alg.alg_string;
2258 		DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2259 		       ic->tag_size, ic->mode, arg_count);
2260 		DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2261 		DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2262 		DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2263 		DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2264 		DMEMIT(" commit_time:%u", ic->autocommit_msec);
2265 		if (ic->sectors_per_block != 1)
2266 			DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2267 
2268 #define EMIT_ALG(a, n)							\
2269 		do {							\
2270 			if (ic->a.alg_string) {				\
2271 				DMEMIT(" %s:%s", n, ic->a.alg_string);	\
2272 				if (ic->a.key_string)			\
2273 					DMEMIT(":%s", ic->a.key_string);\
2274 			}						\
2275 		} while (0)
2276 		EMIT_ALG(internal_hash_alg, "internal_hash");
2277 		EMIT_ALG(journal_crypt_alg, "journal_crypt");
2278 		EMIT_ALG(journal_mac_alg, "journal_mac");
2279 		break;
2280 	}
2281 	}
2282 }
2283 
2284 static int dm_integrity_iterate_devices(struct dm_target *ti,
2285 					iterate_devices_callout_fn fn, void *data)
2286 {
2287 	struct dm_integrity_c *ic = ti->private;
2288 
2289 	return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2290 }
2291 
2292 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2293 {
2294 	struct dm_integrity_c *ic = ti->private;
2295 
2296 	if (ic->sectors_per_block > 1) {
2297 		limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2298 		limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2299 		blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2300 	}
2301 }
2302 
2303 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2304 {
2305 	unsigned sector_space = JOURNAL_SECTOR_DATA;
2306 
2307 	ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2308 	ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2309 					 JOURNAL_ENTRY_ROUNDUP);
2310 
2311 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2312 		sector_space -= JOURNAL_MAC_PER_SECTOR;
2313 	ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2314 	ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2315 	ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2316 	ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2317 }
2318 
2319 static int calculate_device_limits(struct dm_integrity_c *ic)
2320 {
2321 	__u64 initial_sectors;
2322 	sector_t last_sector, last_area, last_offset;
2323 
2324 	calculate_journal_section_size(ic);
2325 	initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2326 	if (initial_sectors + METADATA_PADDING_SECTORS >= ic->device_sectors || initial_sectors > UINT_MAX)
2327 		return -EINVAL;
2328 	ic->initial_sectors = initial_sectors;
2329 
2330 	ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2331 				   (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2332 	if (!(ic->metadata_run & (ic->metadata_run - 1)))
2333 		ic->log2_metadata_run = __ffs(ic->metadata_run);
2334 	else
2335 		ic->log2_metadata_run = -1;
2336 
2337 	get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2338 	last_sector = get_data_sector(ic, last_area, last_offset);
2339 
2340 	if (ic->start + last_sector < last_sector || ic->start + last_sector >= ic->device_sectors)
2341 		return -EINVAL;
2342 
2343 	return 0;
2344 }
2345 
2346 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2347 {
2348 	unsigned journal_sections;
2349 	int test_bit;
2350 
2351 	memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2352 	memcpy(ic->sb->magic, SB_MAGIC, 8);
2353 	ic->sb->version = SB_VERSION;
2354 	ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2355 	ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2356 	if (ic->journal_mac_alg.alg_string)
2357 		ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2358 
2359 	calculate_journal_section_size(ic);
2360 	journal_sections = journal_sectors / ic->journal_section_sectors;
2361 	if (!journal_sections)
2362 		journal_sections = 1;
2363 	ic->sb->journal_sections = cpu_to_le32(journal_sections);
2364 
2365 	if (!interleave_sectors)
2366 		interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2367 	ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2368 	ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2369 	ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2370 
2371 	ic->provided_data_sectors = 0;
2372 	for (test_bit = fls64(ic->device_sectors) - 1; test_bit >= 3; test_bit--) {
2373 		__u64 prev_data_sectors = ic->provided_data_sectors;
2374 
2375 		ic->provided_data_sectors |= (sector_t)1 << test_bit;
2376 		if (calculate_device_limits(ic))
2377 			ic->provided_data_sectors = prev_data_sectors;
2378 	}
2379 
2380 	if (!ic->provided_data_sectors)
2381 		return -EINVAL;
2382 
2383 	ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2384 
2385 	return 0;
2386 }
2387 
2388 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2389 {
2390 	struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2391 	struct blk_integrity bi;
2392 
2393 	memset(&bi, 0, sizeof(bi));
2394 	bi.profile = &dm_integrity_profile;
2395 	bi.tuple_size = ic->tag_size;
2396 	bi.tag_size = bi.tuple_size;
2397 	bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2398 
2399 	blk_integrity_register(disk, &bi);
2400 	blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2401 }
2402 
2403 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2404 {
2405 	unsigned i;
2406 
2407 	if (!pl)
2408 		return;
2409 	for (i = 0; i < ic->journal_pages; i++)
2410 		if (pl[i].page)
2411 			__free_page(pl[i].page);
2412 	kvfree(pl);
2413 }
2414 
2415 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2416 {
2417 	size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2418 	struct page_list *pl;
2419 	unsigned i;
2420 
2421 	pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2422 	if (!pl)
2423 		return NULL;
2424 
2425 	for (i = 0; i < ic->journal_pages; i++) {
2426 		pl[i].page = alloc_page(GFP_KERNEL);
2427 		if (!pl[i].page) {
2428 			dm_integrity_free_page_list(ic, pl);
2429 			return NULL;
2430 		}
2431 		if (i)
2432 			pl[i - 1].next = &pl[i];
2433 	}
2434 
2435 	return pl;
2436 }
2437 
2438 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2439 {
2440 	unsigned i;
2441 	for (i = 0; i < ic->journal_sections; i++)
2442 		kvfree(sl[i]);
2443 	kfree(sl);
2444 }
2445 
2446 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2447 {
2448 	struct scatterlist **sl;
2449 	unsigned i;
2450 
2451 	sl = kvmalloc(ic->journal_sections * sizeof(struct scatterlist *), GFP_KERNEL | __GFP_ZERO);
2452 	if (!sl)
2453 		return NULL;
2454 
2455 	for (i = 0; i < ic->journal_sections; i++) {
2456 		struct scatterlist *s;
2457 		unsigned start_index, start_offset;
2458 		unsigned end_index, end_offset;
2459 		unsigned n_pages;
2460 		unsigned idx;
2461 
2462 		page_list_location(ic, i, 0, &start_index, &start_offset);
2463 		page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2464 
2465 		n_pages = (end_index - start_index + 1);
2466 
2467 		s = kvmalloc(n_pages * sizeof(struct scatterlist), GFP_KERNEL);
2468 		if (!s) {
2469 			dm_integrity_free_journal_scatterlist(ic, sl);
2470 			return NULL;
2471 		}
2472 
2473 		sg_init_table(s, n_pages);
2474 		for (idx = start_index; idx <= end_index; idx++) {
2475 			char *va = lowmem_page_address(pl[idx].page);
2476 			unsigned start = 0, end = PAGE_SIZE;
2477 			if (idx == start_index)
2478 				start = start_offset;
2479 			if (idx == end_index)
2480 				end = end_offset + (1 << SECTOR_SHIFT);
2481 			sg_set_buf(&s[idx - start_index], va + start, end - start);
2482 		}
2483 
2484 		sl[i] = s;
2485 	}
2486 
2487 	return sl;
2488 }
2489 
2490 static void free_alg(struct alg_spec *a)
2491 {
2492 	kzfree(a->alg_string);
2493 	kzfree(a->key);
2494 	memset(a, 0, sizeof *a);
2495 }
2496 
2497 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2498 {
2499 	char *k;
2500 
2501 	free_alg(a);
2502 
2503 	a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2504 	if (!a->alg_string)
2505 		goto nomem;
2506 
2507 	k = strchr(a->alg_string, ':');
2508 	if (k) {
2509 		*k = 0;
2510 		a->key_string = k + 1;
2511 		if (strlen(a->key_string) & 1)
2512 			goto inval;
2513 
2514 		a->key_size = strlen(a->key_string) / 2;
2515 		a->key = kmalloc(a->key_size, GFP_KERNEL);
2516 		if (!a->key)
2517 			goto nomem;
2518 		if (hex2bin(a->key, a->key_string, a->key_size))
2519 			goto inval;
2520 	}
2521 
2522 	return 0;
2523 inval:
2524 	*error = error_inval;
2525 	return -EINVAL;
2526 nomem:
2527 	*error = "Out of memory for an argument";
2528 	return -ENOMEM;
2529 }
2530 
2531 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2532 		   char *error_alg, char *error_key)
2533 {
2534 	int r;
2535 
2536 	if (a->alg_string) {
2537 		*hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2538 		if (IS_ERR(*hash)) {
2539 			*error = error_alg;
2540 			r = PTR_ERR(*hash);
2541 			*hash = NULL;
2542 			return r;
2543 		}
2544 
2545 		if (a->key) {
2546 			r = crypto_shash_setkey(*hash, a->key, a->key_size);
2547 			if (r) {
2548 				*error = error_key;
2549 				return r;
2550 			}
2551 		}
2552 	}
2553 
2554 	return 0;
2555 }
2556 
2557 static int create_journal(struct dm_integrity_c *ic, char **error)
2558 {
2559 	int r = 0;
2560 	unsigned i;
2561 	__u64 journal_pages, journal_desc_size, journal_tree_size;
2562 	unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2563 	struct skcipher_request *req = NULL;
2564 
2565 	ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2566 	ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2567 	ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2568 	ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2569 
2570 	journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2571 				PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2572 	journal_desc_size = journal_pages * sizeof(struct page_list);
2573 	if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2574 		*error = "Journal doesn't fit into memory";
2575 		r = -ENOMEM;
2576 		goto bad;
2577 	}
2578 	ic->journal_pages = journal_pages;
2579 
2580 	ic->journal = dm_integrity_alloc_page_list(ic);
2581 	if (!ic->journal) {
2582 		*error = "Could not allocate memory for journal";
2583 		r = -ENOMEM;
2584 		goto bad;
2585 	}
2586 	if (ic->journal_crypt_alg.alg_string) {
2587 		unsigned ivsize, blocksize;
2588 		struct journal_completion comp;
2589 
2590 		comp.ic = ic;
2591 		ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2592 		if (IS_ERR(ic->journal_crypt)) {
2593 			*error = "Invalid journal cipher";
2594 			r = PTR_ERR(ic->journal_crypt);
2595 			ic->journal_crypt = NULL;
2596 			goto bad;
2597 		}
2598 		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2599 		blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2600 
2601 		if (ic->journal_crypt_alg.key) {
2602 			r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2603 						   ic->journal_crypt_alg.key_size);
2604 			if (r) {
2605 				*error = "Error setting encryption key";
2606 				goto bad;
2607 			}
2608 		}
2609 		DEBUG_print("cipher %s, block size %u iv size %u\n",
2610 			    ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2611 
2612 		ic->journal_io = dm_integrity_alloc_page_list(ic);
2613 		if (!ic->journal_io) {
2614 			*error = "Could not allocate memory for journal io";
2615 			r = -ENOMEM;
2616 			goto bad;
2617 		}
2618 
2619 		if (blocksize == 1) {
2620 			struct scatterlist *sg;
2621 
2622 			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2623 			if (!req) {
2624 				*error = "Could not allocate crypt request";
2625 				r = -ENOMEM;
2626 				goto bad;
2627 			}
2628 
2629 			crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2630 			if (!crypt_iv) {
2631 				*error = "Could not allocate iv";
2632 				r = -ENOMEM;
2633 				goto bad;
2634 			}
2635 
2636 			ic->journal_xor = dm_integrity_alloc_page_list(ic);
2637 			if (!ic->journal_xor) {
2638 				*error = "Could not allocate memory for journal xor";
2639 				r = -ENOMEM;
2640 				goto bad;
2641 			}
2642 
2643 			sg = kvmalloc((ic->journal_pages + 1) * sizeof(struct scatterlist), GFP_KERNEL);
2644 			if (!sg) {
2645 				*error = "Unable to allocate sg list";
2646 				r = -ENOMEM;
2647 				goto bad;
2648 			}
2649 			sg_init_table(sg, ic->journal_pages + 1);
2650 			for (i = 0; i < ic->journal_pages; i++) {
2651 				char *va = lowmem_page_address(ic->journal_xor[i].page);
2652 				clear_page(va);
2653 				sg_set_buf(&sg[i], va, PAGE_SIZE);
2654 			}
2655 			sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2656 			memset(crypt_iv, 0x00, ivsize);
2657 
2658 			skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2659 			init_completion(&comp.comp);
2660 			comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2661 			if (do_crypt(true, req, &comp))
2662 				wait_for_completion(&comp.comp);
2663 			kvfree(sg);
2664 			r = dm_integrity_failed(ic);
2665 			if (r) {
2666 				*error = "Unable to encrypt journal";
2667 				goto bad;
2668 			}
2669 			DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2670 
2671 			crypto_free_skcipher(ic->journal_crypt);
2672 			ic->journal_crypt = NULL;
2673 		} else {
2674 			unsigned crypt_len = roundup(ivsize, blocksize);
2675 
2676 			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2677 			if (!req) {
2678 				*error = "Could not allocate crypt request";
2679 				r = -ENOMEM;
2680 				goto bad;
2681 			}
2682 
2683 			crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2684 			if (!crypt_iv) {
2685 				*error = "Could not allocate iv";
2686 				r = -ENOMEM;
2687 				goto bad;
2688 			}
2689 
2690 			crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2691 			if (!crypt_data) {
2692 				*error = "Unable to allocate crypt data";
2693 				r = -ENOMEM;
2694 				goto bad;
2695 			}
2696 
2697 			ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2698 			if (!ic->journal_scatterlist) {
2699 				*error = "Unable to allocate sg list";
2700 				r = -ENOMEM;
2701 				goto bad;
2702 			}
2703 			ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2704 			if (!ic->journal_io_scatterlist) {
2705 				*error = "Unable to allocate sg list";
2706 				r = -ENOMEM;
2707 				goto bad;
2708 			}
2709 			ic->sk_requests = kvmalloc(ic->journal_sections * sizeof(struct skcipher_request *), GFP_KERNEL | __GFP_ZERO);
2710 			if (!ic->sk_requests) {
2711 				*error = "Unable to allocate sk requests";
2712 				r = -ENOMEM;
2713 				goto bad;
2714 			}
2715 			for (i = 0; i < ic->journal_sections; i++) {
2716 				struct scatterlist sg;
2717 				struct skcipher_request *section_req;
2718 				__u32 section_le = cpu_to_le32(i);
2719 
2720 				memset(crypt_iv, 0x00, ivsize);
2721 				memset(crypt_data, 0x00, crypt_len);
2722 				memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
2723 
2724 				sg_init_one(&sg, crypt_data, crypt_len);
2725 				skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
2726 				init_completion(&comp.comp);
2727 				comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2728 				if (do_crypt(true, req, &comp))
2729 					wait_for_completion(&comp.comp);
2730 
2731 				r = dm_integrity_failed(ic);
2732 				if (r) {
2733 					*error = "Unable to generate iv";
2734 					goto bad;
2735 				}
2736 
2737 				section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2738 				if (!section_req) {
2739 					*error = "Unable to allocate crypt request";
2740 					r = -ENOMEM;
2741 					goto bad;
2742 				}
2743 				section_req->iv = kmalloc(ivsize * 2, GFP_KERNEL);
2744 				if (!section_req->iv) {
2745 					skcipher_request_free(section_req);
2746 					*error = "Unable to allocate iv";
2747 					r = -ENOMEM;
2748 					goto bad;
2749 				}
2750 				memcpy(section_req->iv + ivsize, crypt_data, ivsize);
2751 				section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
2752 				ic->sk_requests[i] = section_req;
2753 				DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
2754 			}
2755 		}
2756 	}
2757 
2758 	for (i = 0; i < N_COMMIT_IDS; i++) {
2759 		unsigned j;
2760 retest_commit_id:
2761 		for (j = 0; j < i; j++) {
2762 			if (ic->commit_ids[j] == ic->commit_ids[i]) {
2763 				ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
2764 				goto retest_commit_id;
2765 			}
2766 		}
2767 		DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
2768 	}
2769 
2770 	journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
2771 	if (journal_tree_size > ULONG_MAX) {
2772 		*error = "Journal doesn't fit into memory";
2773 		r = -ENOMEM;
2774 		goto bad;
2775 	}
2776 	ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
2777 	if (!ic->journal_tree) {
2778 		*error = "Could not allocate memory for journal tree";
2779 		r = -ENOMEM;
2780 	}
2781 bad:
2782 	kfree(crypt_data);
2783 	kfree(crypt_iv);
2784 	skcipher_request_free(req);
2785 
2786 	return r;
2787 }
2788 
2789 /*
2790  * Construct a integrity mapping
2791  *
2792  * Arguments:
2793  *	device
2794  *	offset from the start of the device
2795  *	tag size
2796  *	D - direct writes, J - journal writes, R - recovery mode
2797  *	number of optional arguments
2798  *	optional arguments:
2799  *		journal_sectors
2800  *		interleave_sectors
2801  *		buffer_sectors
2802  *		journal_watermark
2803  *		commit_time
2804  *		internal_hash
2805  *		journal_crypt
2806  *		journal_mac
2807  *		block_size
2808  */
2809 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
2810 {
2811 	struct dm_integrity_c *ic;
2812 	char dummy;
2813 	int r;
2814 	unsigned extra_args;
2815 	struct dm_arg_set as;
2816 	static const struct dm_arg _args[] = {
2817 		{0, 9, "Invalid number of feature args"},
2818 	};
2819 	unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
2820 	bool should_write_sb;
2821 	__u64 threshold;
2822 	unsigned long long start;
2823 
2824 #define DIRECT_ARGUMENTS	4
2825 
2826 	if (argc <= DIRECT_ARGUMENTS) {
2827 		ti->error = "Invalid argument count";
2828 		return -EINVAL;
2829 	}
2830 
2831 	ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
2832 	if (!ic) {
2833 		ti->error = "Cannot allocate integrity context";
2834 		return -ENOMEM;
2835 	}
2836 	ti->private = ic;
2837 	ti->per_io_data_size = sizeof(struct dm_integrity_io);
2838 
2839 	ic->in_progress = RB_ROOT;
2840 	init_waitqueue_head(&ic->endio_wait);
2841 	bio_list_init(&ic->flush_bio_list);
2842 	init_waitqueue_head(&ic->copy_to_journal_wait);
2843 	init_completion(&ic->crypto_backoff);
2844 	atomic64_set(&ic->number_of_mismatches, 0);
2845 
2846 	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
2847 	if (r) {
2848 		ti->error = "Device lookup failed";
2849 		goto bad;
2850 	}
2851 
2852 	if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
2853 		ti->error = "Invalid starting offset";
2854 		r = -EINVAL;
2855 		goto bad;
2856 	}
2857 	ic->start = start;
2858 
2859 	if (strcmp(argv[2], "-")) {
2860 		if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
2861 			ti->error = "Invalid tag size";
2862 			r = -EINVAL;
2863 			goto bad;
2864 		}
2865 	}
2866 
2867 	if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
2868 		ic->mode = argv[3][0];
2869 	else {
2870 		ti->error = "Invalid mode (expecting J, D, R)";
2871 		r = -EINVAL;
2872 		goto bad;
2873 	}
2874 
2875 	ic->device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
2876 	journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
2877 			ic->device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
2878 	interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2879 	buffer_sectors = DEFAULT_BUFFER_SECTORS;
2880 	journal_watermark = DEFAULT_JOURNAL_WATERMARK;
2881 	sync_msec = DEFAULT_SYNC_MSEC;
2882 	ic->sectors_per_block = 1;
2883 
2884 	as.argc = argc - DIRECT_ARGUMENTS;
2885 	as.argv = argv + DIRECT_ARGUMENTS;
2886 	r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
2887 	if (r)
2888 		goto bad;
2889 
2890 	while (extra_args--) {
2891 		const char *opt_string;
2892 		unsigned val;
2893 		opt_string = dm_shift_arg(&as);
2894 		if (!opt_string) {
2895 			r = -EINVAL;
2896 			ti->error = "Not enough feature arguments";
2897 			goto bad;
2898 		}
2899 		if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
2900 			journal_sectors = val;
2901 		else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
2902 			interleave_sectors = val;
2903 		else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
2904 			buffer_sectors = val;
2905 		else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
2906 			journal_watermark = val;
2907 		else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
2908 			sync_msec = val;
2909 		else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
2910 			if (val < 1 << SECTOR_SHIFT ||
2911 			    val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
2912 			    (val & (val -1))) {
2913 				r = -EINVAL;
2914 				ti->error = "Invalid block_size argument";
2915 				goto bad;
2916 			}
2917 			ic->sectors_per_block = val >> SECTOR_SHIFT;
2918 		} else if (!memcmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
2919 			r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
2920 					    "Invalid internal_hash argument");
2921 			if (r)
2922 				goto bad;
2923 		} else if (!memcmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
2924 			r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
2925 					    "Invalid journal_crypt argument");
2926 			if (r)
2927 				goto bad;
2928 		} else if (!memcmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
2929 			r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
2930 					    "Invalid journal_mac argument");
2931 			if (r)
2932 				goto bad;
2933 		} else {
2934 			r = -EINVAL;
2935 			ti->error = "Invalid argument";
2936 			goto bad;
2937 		}
2938 	}
2939 
2940 	r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
2941 		    "Invalid internal hash", "Error setting internal hash key");
2942 	if (r)
2943 		goto bad;
2944 
2945 	r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
2946 		    "Invalid journal mac", "Error setting journal mac key");
2947 	if (r)
2948 		goto bad;
2949 
2950 	if (!ic->tag_size) {
2951 		if (!ic->internal_hash) {
2952 			ti->error = "Unknown tag size";
2953 			r = -EINVAL;
2954 			goto bad;
2955 		}
2956 		ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
2957 	}
2958 	if (ic->tag_size > MAX_TAG_SIZE) {
2959 		ti->error = "Too big tag size";
2960 		r = -EINVAL;
2961 		goto bad;
2962 	}
2963 	if (!(ic->tag_size & (ic->tag_size - 1)))
2964 		ic->log2_tag_size = __ffs(ic->tag_size);
2965 	else
2966 		ic->log2_tag_size = -1;
2967 
2968 	ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
2969 	ic->autocommit_msec = sync_msec;
2970 	timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
2971 
2972 	ic->io = dm_io_client_create();
2973 	if (IS_ERR(ic->io)) {
2974 		r = PTR_ERR(ic->io);
2975 		ic->io = NULL;
2976 		ti->error = "Cannot allocate dm io";
2977 		goto bad;
2978 	}
2979 
2980 	ic->journal_io_mempool = mempool_create_slab_pool(JOURNAL_IO_MEMPOOL, journal_io_cache);
2981 	if (!ic->journal_io_mempool) {
2982 		r = -ENOMEM;
2983 		ti->error = "Cannot allocate mempool";
2984 		goto bad;
2985 	}
2986 
2987 	ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
2988 					  WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
2989 	if (!ic->metadata_wq) {
2990 		ti->error = "Cannot allocate workqueue";
2991 		r = -ENOMEM;
2992 		goto bad;
2993 	}
2994 
2995 	/*
2996 	 * If this workqueue were percpu, it would cause bio reordering
2997 	 * and reduced performance.
2998 	 */
2999 	ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3000 	if (!ic->wait_wq) {
3001 		ti->error = "Cannot allocate workqueue";
3002 		r = -ENOMEM;
3003 		goto bad;
3004 	}
3005 
3006 	ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3007 	if (!ic->commit_wq) {
3008 		ti->error = "Cannot allocate workqueue";
3009 		r = -ENOMEM;
3010 		goto bad;
3011 	}
3012 	INIT_WORK(&ic->commit_work, integrity_commit);
3013 
3014 	if (ic->mode == 'J') {
3015 		ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3016 		if (!ic->writer_wq) {
3017 			ti->error = "Cannot allocate workqueue";
3018 			r = -ENOMEM;
3019 			goto bad;
3020 		}
3021 		INIT_WORK(&ic->writer_work, integrity_writer);
3022 	}
3023 
3024 	ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3025 	if (!ic->sb) {
3026 		r = -ENOMEM;
3027 		ti->error = "Cannot allocate superblock area";
3028 		goto bad;
3029 	}
3030 
3031 	r = sync_rw_sb(ic, REQ_OP_READ, 0);
3032 	if (r) {
3033 		ti->error = "Error reading superblock";
3034 		goto bad;
3035 	}
3036 	should_write_sb = false;
3037 	if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3038 		if (ic->mode != 'R') {
3039 			if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3040 				r = -EINVAL;
3041 				ti->error = "The device is not initialized";
3042 				goto bad;
3043 			}
3044 		}
3045 
3046 		r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3047 		if (r) {
3048 			ti->error = "Could not initialize superblock";
3049 			goto bad;
3050 		}
3051 		if (ic->mode != 'R')
3052 			should_write_sb = true;
3053 	}
3054 
3055 	if (ic->sb->version != SB_VERSION) {
3056 		r = -EINVAL;
3057 		ti->error = "Unknown version";
3058 		goto bad;
3059 	}
3060 	if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3061 		r = -EINVAL;
3062 		ti->error = "Tag size doesn't match the information in superblock";
3063 		goto bad;
3064 	}
3065 	if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3066 		r = -EINVAL;
3067 		ti->error = "Block size doesn't match the information in superblock";
3068 		goto bad;
3069 	}
3070 	if (!le32_to_cpu(ic->sb->journal_sections)) {
3071 		r = -EINVAL;
3072 		ti->error = "Corrupted superblock, journal_sections is 0";
3073 		goto bad;
3074 	}
3075 	/* make sure that ti->max_io_len doesn't overflow */
3076 	if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3077 	    ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3078 		r = -EINVAL;
3079 		ti->error = "Invalid interleave_sectors in the superblock";
3080 		goto bad;
3081 	}
3082 	ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3083 	if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3084 		/* test for overflow */
3085 		r = -EINVAL;
3086 		ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3087 		goto bad;
3088 	}
3089 	if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3090 		r = -EINVAL;
3091 		ti->error = "Journal mac mismatch";
3092 		goto bad;
3093 	}
3094 	r = calculate_device_limits(ic);
3095 	if (r) {
3096 		ti->error = "The device is too small";
3097 		goto bad;
3098 	}
3099 	if (ti->len > ic->provided_data_sectors) {
3100 		r = -EINVAL;
3101 		ti->error = "Not enough provided sectors for requested mapping size";
3102 		goto bad;
3103 	}
3104 
3105 	if (!buffer_sectors)
3106 		buffer_sectors = 1;
3107 	ic->log2_buffer_sectors = min3((int)__fls(buffer_sectors), (int)__ffs(ic->metadata_run), 31 - SECTOR_SHIFT);
3108 
3109 	threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3110 	threshold += 50;
3111 	do_div(threshold, 100);
3112 	ic->free_sectors_threshold = threshold;
3113 
3114 	DEBUG_print("initialized:\n");
3115 	DEBUG_print("	integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3116 	DEBUG_print("	journal_entry_size %u\n", ic->journal_entry_size);
3117 	DEBUG_print("	journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3118 	DEBUG_print("	journal_section_entries %u\n", ic->journal_section_entries);
3119 	DEBUG_print("	journal_section_sectors %u\n", ic->journal_section_sectors);
3120 	DEBUG_print("	journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3121 	DEBUG_print("	journal_entries %u\n", ic->journal_entries);
3122 	DEBUG_print("	log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3123 	DEBUG_print("	device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
3124 	DEBUG_print("	initial_sectors 0x%x\n", ic->initial_sectors);
3125 	DEBUG_print("	metadata_run 0x%x\n", ic->metadata_run);
3126 	DEBUG_print("	log2_metadata_run %d\n", ic->log2_metadata_run);
3127 	DEBUG_print("	provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3128 		    (unsigned long long)ic->provided_data_sectors);
3129 	DEBUG_print("	log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3130 
3131 	ic->bufio = dm_bufio_client_create(ic->dev->bdev, 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors),
3132 					   1, 0, NULL, NULL);
3133 	if (IS_ERR(ic->bufio)) {
3134 		r = PTR_ERR(ic->bufio);
3135 		ti->error = "Cannot initialize dm-bufio";
3136 		ic->bufio = NULL;
3137 		goto bad;
3138 	}
3139 	dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3140 
3141 	if (ic->mode != 'R') {
3142 		r = create_journal(ic, &ti->error);
3143 		if (r)
3144 			goto bad;
3145 	}
3146 
3147 	if (should_write_sb) {
3148 		int r;
3149 
3150 		init_journal(ic, 0, ic->journal_sections, 0);
3151 		r = dm_integrity_failed(ic);
3152 		if (unlikely(r)) {
3153 			ti->error = "Error initializing journal";
3154 			goto bad;
3155 		}
3156 		r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3157 		if (r) {
3158 			ti->error = "Error initializing superblock";
3159 			goto bad;
3160 		}
3161 		ic->just_formatted = true;
3162 	}
3163 
3164 	r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3165 	if (r)
3166 		goto bad;
3167 
3168 	if (!ic->internal_hash)
3169 		dm_integrity_set(ti, ic);
3170 
3171 	ti->num_flush_bios = 1;
3172 	ti->flush_supported = true;
3173 
3174 	return 0;
3175 bad:
3176 	dm_integrity_dtr(ti);
3177 	return r;
3178 }
3179 
3180 static void dm_integrity_dtr(struct dm_target *ti)
3181 {
3182 	struct dm_integrity_c *ic = ti->private;
3183 
3184 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3185 
3186 	if (ic->metadata_wq)
3187 		destroy_workqueue(ic->metadata_wq);
3188 	if (ic->wait_wq)
3189 		destroy_workqueue(ic->wait_wq);
3190 	if (ic->commit_wq)
3191 		destroy_workqueue(ic->commit_wq);
3192 	if (ic->writer_wq)
3193 		destroy_workqueue(ic->writer_wq);
3194 	if (ic->bufio)
3195 		dm_bufio_client_destroy(ic->bufio);
3196 	mempool_destroy(ic->journal_io_mempool);
3197 	if (ic->io)
3198 		dm_io_client_destroy(ic->io);
3199 	if (ic->dev)
3200 		dm_put_device(ti, ic->dev);
3201 	dm_integrity_free_page_list(ic, ic->journal);
3202 	dm_integrity_free_page_list(ic, ic->journal_io);
3203 	dm_integrity_free_page_list(ic, ic->journal_xor);
3204 	if (ic->journal_scatterlist)
3205 		dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3206 	if (ic->journal_io_scatterlist)
3207 		dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3208 	if (ic->sk_requests) {
3209 		unsigned i;
3210 
3211 		for (i = 0; i < ic->journal_sections; i++) {
3212 			struct skcipher_request *req = ic->sk_requests[i];
3213 			if (req) {
3214 				kzfree(req->iv);
3215 				skcipher_request_free(req);
3216 			}
3217 		}
3218 		kvfree(ic->sk_requests);
3219 	}
3220 	kvfree(ic->journal_tree);
3221 	if (ic->sb)
3222 		free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3223 
3224 	if (ic->internal_hash)
3225 		crypto_free_shash(ic->internal_hash);
3226 	free_alg(&ic->internal_hash_alg);
3227 
3228 	if (ic->journal_crypt)
3229 		crypto_free_skcipher(ic->journal_crypt);
3230 	free_alg(&ic->journal_crypt_alg);
3231 
3232 	if (ic->journal_mac)
3233 		crypto_free_shash(ic->journal_mac);
3234 	free_alg(&ic->journal_mac_alg);
3235 
3236 	kfree(ic);
3237 }
3238 
3239 static struct target_type integrity_target = {
3240 	.name			= "integrity",
3241 	.version		= {1, 1, 0},
3242 	.module			= THIS_MODULE,
3243 	.features		= DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3244 	.ctr			= dm_integrity_ctr,
3245 	.dtr			= dm_integrity_dtr,
3246 	.map			= dm_integrity_map,
3247 	.postsuspend		= dm_integrity_postsuspend,
3248 	.resume			= dm_integrity_resume,
3249 	.status			= dm_integrity_status,
3250 	.iterate_devices	= dm_integrity_iterate_devices,
3251 	.io_hints		= dm_integrity_io_hints,
3252 };
3253 
3254 int __init dm_integrity_init(void)
3255 {
3256 	int r;
3257 
3258 	journal_io_cache = kmem_cache_create("integrity_journal_io",
3259 					     sizeof(struct journal_io), 0, 0, NULL);
3260 	if (!journal_io_cache) {
3261 		DMERR("can't allocate journal io cache");
3262 		return -ENOMEM;
3263 	}
3264 
3265 	r = dm_register_target(&integrity_target);
3266 
3267 	if (r < 0)
3268 		DMERR("register failed %d", r);
3269 
3270 	return r;
3271 }
3272 
3273 void dm_integrity_exit(void)
3274 {
3275 	dm_unregister_target(&integrity_target);
3276 	kmem_cache_destroy(journal_io_cache);
3277 }
3278 
3279 module_init(dm_integrity_init);
3280 module_exit(dm_integrity_exit);
3281 
3282 MODULE_AUTHOR("Milan Broz");
3283 MODULE_AUTHOR("Mikulas Patocka");
3284 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3285 MODULE_LICENSE("GPL");
3286