xref: /linux/drivers/md/dm-integrity.c (revision 8e07e0e3964ca4e23ce7b68e2096fe660a888942)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
4  * Copyright (C) 2016-2017 Milan Broz
5  * Copyright (C) 2016-2017 Mikulas Patocka
6  *
7  * This file is released under the GPL.
8  */
9 
10 #include "dm-bio-record.h"
11 
12 #include <linux/compiler.h>
13 #include <linux/module.h>
14 #include <linux/device-mapper.h>
15 #include <linux/dm-io.h>
16 #include <linux/vmalloc.h>
17 #include <linux/sort.h>
18 #include <linux/rbtree.h>
19 #include <linux/delay.h>
20 #include <linux/random.h>
21 #include <linux/reboot.h>
22 #include <crypto/hash.h>
23 #include <crypto/skcipher.h>
24 #include <linux/async_tx.h>
25 #include <linux/dm-bufio.h>
26 
27 #include "dm-audit.h"
28 
29 #define DM_MSG_PREFIX "integrity"
30 
31 #define DEFAULT_INTERLEAVE_SECTORS	32768
32 #define DEFAULT_JOURNAL_SIZE_FACTOR	7
33 #define DEFAULT_SECTORS_PER_BITMAP_BIT	32768
34 #define DEFAULT_BUFFER_SECTORS		128
35 #define DEFAULT_JOURNAL_WATERMARK	50
36 #define DEFAULT_SYNC_MSEC		10000
37 #define DEFAULT_MAX_JOURNAL_SECTORS	(IS_ENABLED(CONFIG_64BIT) ? 131072 : 8192)
38 #define MIN_LOG2_INTERLEAVE_SECTORS	3
39 #define MAX_LOG2_INTERLEAVE_SECTORS	31
40 #define METADATA_WORKQUEUE_MAX_ACTIVE	16
41 #define RECALC_SECTORS			(IS_ENABLED(CONFIG_64BIT) ? 32768 : 2048)
42 #define RECALC_WRITE_SUPER		16
43 #define BITMAP_BLOCK_SIZE		4096	/* don't change it */
44 #define BITMAP_FLUSH_INTERVAL		(10 * HZ)
45 #define DISCARD_FILLER			0xf6
46 #define SALT_SIZE			16
47 
48 /*
49  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
50  * so it should not be enabled in the official kernel
51  */
52 //#define DEBUG_PRINT
53 //#define INTERNAL_VERIFY
54 
55 /*
56  * On disk structures
57  */
58 
59 #define SB_MAGIC			"integrt"
60 #define SB_VERSION_1			1
61 #define SB_VERSION_2			2
62 #define SB_VERSION_3			3
63 #define SB_VERSION_4			4
64 #define SB_VERSION_5			5
65 #define SB_SECTORS			8
66 #define MAX_SECTORS_PER_BLOCK		8
67 
68 struct superblock {
69 	__u8 magic[8];
70 	__u8 version;
71 	__u8 log2_interleave_sectors;
72 	__le16 integrity_tag_size;
73 	__le32 journal_sections;
74 	__le64 provided_data_sectors;	/* userspace uses this value */
75 	__le32 flags;
76 	__u8 log2_sectors_per_block;
77 	__u8 log2_blocks_per_bitmap_bit;
78 	__u8 pad[2];
79 	__le64 recalc_sector;
80 	__u8 pad2[8];
81 	__u8 salt[SALT_SIZE];
82 };
83 
84 #define SB_FLAG_HAVE_JOURNAL_MAC	0x1
85 #define SB_FLAG_RECALCULATING		0x2
86 #define SB_FLAG_DIRTY_BITMAP		0x4
87 #define SB_FLAG_FIXED_PADDING		0x8
88 #define SB_FLAG_FIXED_HMAC		0x10
89 
90 #define	JOURNAL_ENTRY_ROUNDUP		8
91 
92 typedef __le64 commit_id_t;
93 #define JOURNAL_MAC_PER_SECTOR		8
94 
95 struct journal_entry {
96 	union {
97 		struct {
98 			__le32 sector_lo;
99 			__le32 sector_hi;
100 		} s;
101 		__le64 sector;
102 	} u;
103 	commit_id_t last_bytes[];
104 	/* __u8 tag[0]; */
105 };
106 
107 #define journal_entry_tag(ic, je)		((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
108 
109 #if BITS_PER_LONG == 64
110 #define journal_entry_set_sector(je, x)		do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
111 #else
112 #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)
113 #endif
114 #define journal_entry_get_sector(je)		le64_to_cpu((je)->u.sector)
115 #define journal_entry_is_unused(je)		((je)->u.s.sector_hi == cpu_to_le32(-1))
116 #define journal_entry_set_unused(je)		((je)->u.s.sector_hi = cpu_to_le32(-1))
117 #define journal_entry_is_inprogress(je)		((je)->u.s.sector_hi == cpu_to_le32(-2))
118 #define journal_entry_set_inprogress(je)	((je)->u.s.sector_hi = cpu_to_le32(-2))
119 
120 #define JOURNAL_BLOCK_SECTORS		8
121 #define JOURNAL_SECTOR_DATA		((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
122 #define JOURNAL_MAC_SIZE		(JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
123 
124 struct journal_sector {
125 	struct_group(sectors,
126 		__u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
127 		__u8 mac[JOURNAL_MAC_PER_SECTOR];
128 	);
129 	commit_id_t commit_id;
130 };
131 
132 #define MAX_TAG_SIZE			(JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
133 
134 #define METADATA_PADDING_SECTORS	8
135 
136 #define N_COMMIT_IDS			4
137 
138 static unsigned char prev_commit_seq(unsigned char seq)
139 {
140 	return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
141 }
142 
143 static unsigned char next_commit_seq(unsigned char seq)
144 {
145 	return (seq + 1) % N_COMMIT_IDS;
146 }
147 
148 /*
149  * In-memory structures
150  */
151 
152 struct journal_node {
153 	struct rb_node node;
154 	sector_t sector;
155 };
156 
157 struct alg_spec {
158 	char *alg_string;
159 	char *key_string;
160 	__u8 *key;
161 	unsigned int key_size;
162 };
163 
164 struct dm_integrity_c {
165 	struct dm_dev *dev;
166 	struct dm_dev *meta_dev;
167 	unsigned int tag_size;
168 	__s8 log2_tag_size;
169 	sector_t start;
170 	mempool_t journal_io_mempool;
171 	struct dm_io_client *io;
172 	struct dm_bufio_client *bufio;
173 	struct workqueue_struct *metadata_wq;
174 	struct superblock *sb;
175 	unsigned int journal_pages;
176 	unsigned int n_bitmap_blocks;
177 
178 	struct page_list *journal;
179 	struct page_list *journal_io;
180 	struct page_list *journal_xor;
181 	struct page_list *recalc_bitmap;
182 	struct page_list *may_write_bitmap;
183 	struct bitmap_block_status *bbs;
184 	unsigned int bitmap_flush_interval;
185 	int synchronous_mode;
186 	struct bio_list synchronous_bios;
187 	struct delayed_work bitmap_flush_work;
188 
189 	struct crypto_skcipher *journal_crypt;
190 	struct scatterlist **journal_scatterlist;
191 	struct scatterlist **journal_io_scatterlist;
192 	struct skcipher_request **sk_requests;
193 
194 	struct crypto_shash *journal_mac;
195 
196 	struct journal_node *journal_tree;
197 	struct rb_root journal_tree_root;
198 
199 	sector_t provided_data_sectors;
200 
201 	unsigned short journal_entry_size;
202 	unsigned char journal_entries_per_sector;
203 	unsigned char journal_section_entries;
204 	unsigned short journal_section_sectors;
205 	unsigned int journal_sections;
206 	unsigned int journal_entries;
207 	sector_t data_device_sectors;
208 	sector_t meta_device_sectors;
209 	unsigned int initial_sectors;
210 	unsigned int metadata_run;
211 	__s8 log2_metadata_run;
212 	__u8 log2_buffer_sectors;
213 	__u8 sectors_per_block;
214 	__u8 log2_blocks_per_bitmap_bit;
215 
216 	unsigned char mode;
217 
218 	int failed;
219 
220 	struct crypto_shash *internal_hash;
221 
222 	struct dm_target *ti;
223 
224 	/* these variables are locked with endio_wait.lock */
225 	struct rb_root in_progress;
226 	struct list_head wait_list;
227 	wait_queue_head_t endio_wait;
228 	struct workqueue_struct *wait_wq;
229 	struct workqueue_struct *offload_wq;
230 
231 	unsigned char commit_seq;
232 	commit_id_t commit_ids[N_COMMIT_IDS];
233 
234 	unsigned int committed_section;
235 	unsigned int n_committed_sections;
236 
237 	unsigned int uncommitted_section;
238 	unsigned int n_uncommitted_sections;
239 
240 	unsigned int free_section;
241 	unsigned char free_section_entry;
242 	unsigned int free_sectors;
243 
244 	unsigned int free_sectors_threshold;
245 
246 	struct workqueue_struct *commit_wq;
247 	struct work_struct commit_work;
248 
249 	struct workqueue_struct *writer_wq;
250 	struct work_struct writer_work;
251 
252 	struct workqueue_struct *recalc_wq;
253 	struct work_struct recalc_work;
254 
255 	struct bio_list flush_bio_list;
256 
257 	unsigned long autocommit_jiffies;
258 	struct timer_list autocommit_timer;
259 	unsigned int autocommit_msec;
260 
261 	wait_queue_head_t copy_to_journal_wait;
262 
263 	struct completion crypto_backoff;
264 
265 	bool wrote_to_journal;
266 	bool journal_uptodate;
267 	bool just_formatted;
268 	bool recalculate_flag;
269 	bool reset_recalculate_flag;
270 	bool discard;
271 	bool fix_padding;
272 	bool fix_hmac;
273 	bool legacy_recalculate;
274 
275 	struct alg_spec internal_hash_alg;
276 	struct alg_spec journal_crypt_alg;
277 	struct alg_spec journal_mac_alg;
278 
279 	atomic64_t number_of_mismatches;
280 
281 	struct notifier_block reboot_notifier;
282 };
283 
284 struct dm_integrity_range {
285 	sector_t logical_sector;
286 	sector_t n_sectors;
287 	bool waiting;
288 	union {
289 		struct rb_node node;
290 		struct {
291 			struct task_struct *task;
292 			struct list_head wait_entry;
293 		};
294 	};
295 };
296 
297 struct dm_integrity_io {
298 	struct work_struct work;
299 
300 	struct dm_integrity_c *ic;
301 	enum req_op op;
302 	bool fua;
303 
304 	struct dm_integrity_range range;
305 
306 	sector_t metadata_block;
307 	unsigned int metadata_offset;
308 
309 	atomic_t in_flight;
310 	blk_status_t bi_status;
311 
312 	struct completion *completion;
313 
314 	struct dm_bio_details bio_details;
315 };
316 
317 struct journal_completion {
318 	struct dm_integrity_c *ic;
319 	atomic_t in_flight;
320 	struct completion comp;
321 };
322 
323 struct journal_io {
324 	struct dm_integrity_range range;
325 	struct journal_completion *comp;
326 };
327 
328 struct bitmap_block_status {
329 	struct work_struct work;
330 	struct dm_integrity_c *ic;
331 	unsigned int idx;
332 	unsigned long *bitmap;
333 	struct bio_list bio_queue;
334 	spinlock_t bio_queue_lock;
335 
336 };
337 
338 static struct kmem_cache *journal_io_cache;
339 
340 #define JOURNAL_IO_MEMPOOL	32
341 
342 #ifdef DEBUG_PRINT
343 #define DEBUG_print(x, ...)			printk(KERN_DEBUG x, ##__VA_ARGS__)
344 #define DEBUG_bytes(bytes, len, msg, ...)	printk(KERN_DEBUG msg "%s%*ph\n", ##__VA_ARGS__, \
345 						       len ? ": " : "", len, bytes)
346 #else
347 #define DEBUG_print(x, ...)			do { } while (0)
348 #define DEBUG_bytes(bytes, len, msg, ...)	do { } while (0)
349 #endif
350 
351 static void dm_integrity_prepare(struct request *rq)
352 {
353 }
354 
355 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
356 {
357 }
358 
359 /*
360  * DM Integrity profile, protection is performed layer above (dm-crypt)
361  */
362 static const struct blk_integrity_profile dm_integrity_profile = {
363 	.name			= "DM-DIF-EXT-TAG",
364 	.generate_fn		= NULL,
365 	.verify_fn		= NULL,
366 	.prepare_fn		= dm_integrity_prepare,
367 	.complete_fn		= dm_integrity_complete,
368 };
369 
370 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
371 static void integrity_bio_wait(struct work_struct *w);
372 static void dm_integrity_dtr(struct dm_target *ti);
373 
374 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
375 {
376 	if (err == -EILSEQ)
377 		atomic64_inc(&ic->number_of_mismatches);
378 	if (!cmpxchg(&ic->failed, 0, err))
379 		DMERR("Error on %s: %d", msg, err);
380 }
381 
382 static int dm_integrity_failed(struct dm_integrity_c *ic)
383 {
384 	return READ_ONCE(ic->failed);
385 }
386 
387 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
388 {
389 	if (ic->legacy_recalculate)
390 		return false;
391 	if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
392 	    ic->internal_hash_alg.key || ic->journal_mac_alg.key :
393 	    ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
394 		return true;
395 	return false;
396 }
397 
398 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned int i,
399 					  unsigned int j, unsigned char seq)
400 {
401 	/*
402 	 * Xor the number with section and sector, so that if a piece of
403 	 * journal is written at wrong place, it is detected.
404 	 */
405 	return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
406 }
407 
408 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
409 				sector_t *area, sector_t *offset)
410 {
411 	if (!ic->meta_dev) {
412 		__u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
413 		*area = data_sector >> log2_interleave_sectors;
414 		*offset = (unsigned int)data_sector & ((1U << log2_interleave_sectors) - 1);
415 	} else {
416 		*area = 0;
417 		*offset = data_sector;
418 	}
419 }
420 
421 #define sector_to_block(ic, n)						\
422 do {									\
423 	BUG_ON((n) & (unsigned int)((ic)->sectors_per_block - 1));		\
424 	(n) >>= (ic)->sb->log2_sectors_per_block;			\
425 } while (0)
426 
427 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
428 					    sector_t offset, unsigned int *metadata_offset)
429 {
430 	__u64 ms;
431 	unsigned int mo;
432 
433 	ms = area << ic->sb->log2_interleave_sectors;
434 	if (likely(ic->log2_metadata_run >= 0))
435 		ms += area << ic->log2_metadata_run;
436 	else
437 		ms += area * ic->metadata_run;
438 	ms >>= ic->log2_buffer_sectors;
439 
440 	sector_to_block(ic, offset);
441 
442 	if (likely(ic->log2_tag_size >= 0)) {
443 		ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
444 		mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
445 	} else {
446 		ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
447 		mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
448 	}
449 	*metadata_offset = mo;
450 	return ms;
451 }
452 
453 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
454 {
455 	sector_t result;
456 
457 	if (ic->meta_dev)
458 		return offset;
459 
460 	result = area << ic->sb->log2_interleave_sectors;
461 	if (likely(ic->log2_metadata_run >= 0))
462 		result += (area + 1) << ic->log2_metadata_run;
463 	else
464 		result += (area + 1) * ic->metadata_run;
465 
466 	result += (sector_t)ic->initial_sectors + offset;
467 	result += ic->start;
468 
469 	return result;
470 }
471 
472 static void wraparound_section(struct dm_integrity_c *ic, unsigned int *sec_ptr)
473 {
474 	if (unlikely(*sec_ptr >= ic->journal_sections))
475 		*sec_ptr -= ic->journal_sections;
476 }
477 
478 static void sb_set_version(struct dm_integrity_c *ic)
479 {
480 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
481 		ic->sb->version = SB_VERSION_5;
482 	else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
483 		ic->sb->version = SB_VERSION_4;
484 	else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
485 		ic->sb->version = SB_VERSION_3;
486 	else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
487 		ic->sb->version = SB_VERSION_2;
488 	else
489 		ic->sb->version = SB_VERSION_1;
490 }
491 
492 static int sb_mac(struct dm_integrity_c *ic, bool wr)
493 {
494 	SHASH_DESC_ON_STACK(desc, ic->journal_mac);
495 	int r;
496 	unsigned int mac_size = crypto_shash_digestsize(ic->journal_mac);
497 	__u8 *sb = (__u8 *)ic->sb;
498 	__u8 *mac = sb + (1 << SECTOR_SHIFT) - mac_size;
499 
500 	if (sizeof(struct superblock) + mac_size > 1 << SECTOR_SHIFT) {
501 		dm_integrity_io_error(ic, "digest is too long", -EINVAL);
502 		return -EINVAL;
503 	}
504 
505 	desc->tfm = ic->journal_mac;
506 
507 	if (likely(wr)) {
508 		r = crypto_shash_digest(desc, sb, mac - sb, mac);
509 		if (unlikely(r < 0)) {
510 			dm_integrity_io_error(ic, "crypto_shash_digest", r);
511 			return r;
512 		}
513 	} else {
514 		__u8 actual_mac[HASH_MAX_DIGESTSIZE];
515 
516 		r = crypto_shash_digest(desc, sb, mac - sb, actual_mac);
517 		if (unlikely(r < 0)) {
518 			dm_integrity_io_error(ic, "crypto_shash_digest", r);
519 			return r;
520 		}
521 		if (memcmp(mac, actual_mac, mac_size)) {
522 			dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
523 			dm_audit_log_target(DM_MSG_PREFIX, "mac-superblock", ic->ti, 0);
524 			return -EILSEQ;
525 		}
526 	}
527 
528 	return 0;
529 }
530 
531 static int sync_rw_sb(struct dm_integrity_c *ic, blk_opf_t opf)
532 {
533 	struct dm_io_request io_req;
534 	struct dm_io_region io_loc;
535 	const enum req_op op = opf & REQ_OP_MASK;
536 	int r;
537 
538 	io_req.bi_opf = opf;
539 	io_req.mem.type = DM_IO_KMEM;
540 	io_req.mem.ptr.addr = ic->sb;
541 	io_req.notify.fn = NULL;
542 	io_req.client = ic->io;
543 	io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
544 	io_loc.sector = ic->start;
545 	io_loc.count = SB_SECTORS;
546 
547 	if (op == REQ_OP_WRITE) {
548 		sb_set_version(ic);
549 		if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
550 			r = sb_mac(ic, true);
551 			if (unlikely(r))
552 				return r;
553 		}
554 	}
555 
556 	r = dm_io(&io_req, 1, &io_loc, NULL);
557 	if (unlikely(r))
558 		return r;
559 
560 	if (op == REQ_OP_READ) {
561 		if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
562 			r = sb_mac(ic, false);
563 			if (unlikely(r))
564 				return r;
565 		}
566 	}
567 
568 	return 0;
569 }
570 
571 #define BITMAP_OP_TEST_ALL_SET		0
572 #define BITMAP_OP_TEST_ALL_CLEAR	1
573 #define BITMAP_OP_SET			2
574 #define BITMAP_OP_CLEAR			3
575 
576 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
577 			    sector_t sector, sector_t n_sectors, int mode)
578 {
579 	unsigned long bit, end_bit, this_end_bit, page, end_page;
580 	unsigned long *data;
581 
582 	if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
583 		DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
584 			sector,
585 			n_sectors,
586 			ic->sb->log2_sectors_per_block,
587 			ic->log2_blocks_per_bitmap_bit,
588 			mode);
589 		BUG();
590 	}
591 
592 	if (unlikely(!n_sectors))
593 		return true;
594 
595 	bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
596 	end_bit = (sector + n_sectors - 1) >>
597 		(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
598 
599 	page = bit / (PAGE_SIZE * 8);
600 	bit %= PAGE_SIZE * 8;
601 
602 	end_page = end_bit / (PAGE_SIZE * 8);
603 	end_bit %= PAGE_SIZE * 8;
604 
605 repeat:
606 	if (page < end_page)
607 		this_end_bit = PAGE_SIZE * 8 - 1;
608 	else
609 		this_end_bit = end_bit;
610 
611 	data = lowmem_page_address(bitmap[page].page);
612 
613 	if (mode == BITMAP_OP_TEST_ALL_SET) {
614 		while (bit <= this_end_bit) {
615 			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
616 				do {
617 					if (data[bit / BITS_PER_LONG] != -1)
618 						return false;
619 					bit += BITS_PER_LONG;
620 				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
621 				continue;
622 			}
623 			if (!test_bit(bit, data))
624 				return false;
625 			bit++;
626 		}
627 	} else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
628 		while (bit <= this_end_bit) {
629 			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
630 				do {
631 					if (data[bit / BITS_PER_LONG] != 0)
632 						return false;
633 					bit += BITS_PER_LONG;
634 				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
635 				continue;
636 			}
637 			if (test_bit(bit, data))
638 				return false;
639 			bit++;
640 		}
641 	} else if (mode == BITMAP_OP_SET) {
642 		while (bit <= this_end_bit) {
643 			if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
644 				do {
645 					data[bit / BITS_PER_LONG] = -1;
646 					bit += BITS_PER_LONG;
647 				} while (this_end_bit >= bit + BITS_PER_LONG - 1);
648 				continue;
649 			}
650 			__set_bit(bit, data);
651 			bit++;
652 		}
653 	} else if (mode == BITMAP_OP_CLEAR) {
654 		if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
655 			clear_page(data);
656 		else {
657 			while (bit <= this_end_bit) {
658 				if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
659 					do {
660 						data[bit / BITS_PER_LONG] = 0;
661 						bit += BITS_PER_LONG;
662 					} while (this_end_bit >= bit + BITS_PER_LONG - 1);
663 					continue;
664 				}
665 				__clear_bit(bit, data);
666 				bit++;
667 			}
668 		}
669 	} else {
670 		BUG();
671 	}
672 
673 	if (unlikely(page < end_page)) {
674 		bit = 0;
675 		page++;
676 		goto repeat;
677 	}
678 
679 	return true;
680 }
681 
682 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
683 {
684 	unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
685 	unsigned int i;
686 
687 	for (i = 0; i < n_bitmap_pages; i++) {
688 		unsigned long *dst_data = lowmem_page_address(dst[i].page);
689 		unsigned long *src_data = lowmem_page_address(src[i].page);
690 
691 		copy_page(dst_data, src_data);
692 	}
693 }
694 
695 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
696 {
697 	unsigned int bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
698 	unsigned int bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
699 
700 	BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
701 	return &ic->bbs[bitmap_block];
702 }
703 
704 static void access_journal_check(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
705 				 bool e, const char *function)
706 {
707 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
708 	unsigned int limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
709 
710 	if (unlikely(section >= ic->journal_sections) ||
711 	    unlikely(offset >= limit)) {
712 		DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
713 		       function, section, offset, ic->journal_sections, limit);
714 		BUG();
715 	}
716 #endif
717 }
718 
719 static void page_list_location(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
720 			       unsigned int *pl_index, unsigned int *pl_offset)
721 {
722 	unsigned int sector;
723 
724 	access_journal_check(ic, section, offset, false, "page_list_location");
725 
726 	sector = section * ic->journal_section_sectors + offset;
727 
728 	*pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
729 	*pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
730 }
731 
732 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
733 					       unsigned int section, unsigned int offset, unsigned int *n_sectors)
734 {
735 	unsigned int pl_index, pl_offset;
736 	char *va;
737 
738 	page_list_location(ic, section, offset, &pl_index, &pl_offset);
739 
740 	if (n_sectors)
741 		*n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
742 
743 	va = lowmem_page_address(pl[pl_index].page);
744 
745 	return (struct journal_sector *)(va + pl_offset);
746 }
747 
748 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset)
749 {
750 	return access_page_list(ic, ic->journal, section, offset, NULL);
751 }
752 
753 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
754 {
755 	unsigned int rel_sector, offset;
756 	struct journal_sector *js;
757 
758 	access_journal_check(ic, section, n, true, "access_journal_entry");
759 
760 	rel_sector = n % JOURNAL_BLOCK_SECTORS;
761 	offset = n / JOURNAL_BLOCK_SECTORS;
762 
763 	js = access_journal(ic, section, rel_sector);
764 	return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
765 }
766 
767 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
768 {
769 	n <<= ic->sb->log2_sectors_per_block;
770 
771 	n += JOURNAL_BLOCK_SECTORS;
772 
773 	access_journal_check(ic, section, n, false, "access_journal_data");
774 
775 	return access_journal(ic, section, n);
776 }
777 
778 static void section_mac(struct dm_integrity_c *ic, unsigned int section, __u8 result[JOURNAL_MAC_SIZE])
779 {
780 	SHASH_DESC_ON_STACK(desc, ic->journal_mac);
781 	int r;
782 	unsigned int j, size;
783 
784 	desc->tfm = ic->journal_mac;
785 
786 	r = crypto_shash_init(desc);
787 	if (unlikely(r < 0)) {
788 		dm_integrity_io_error(ic, "crypto_shash_init", r);
789 		goto err;
790 	}
791 
792 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
793 		__le64 section_le;
794 
795 		r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
796 		if (unlikely(r < 0)) {
797 			dm_integrity_io_error(ic, "crypto_shash_update", r);
798 			goto err;
799 		}
800 
801 		section_le = cpu_to_le64(section);
802 		r = crypto_shash_update(desc, (__u8 *)&section_le, sizeof(section_le));
803 		if (unlikely(r < 0)) {
804 			dm_integrity_io_error(ic, "crypto_shash_update", r);
805 			goto err;
806 		}
807 	}
808 
809 	for (j = 0; j < ic->journal_section_entries; j++) {
810 		struct journal_entry *je = access_journal_entry(ic, section, j);
811 
812 		r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof(je->u.sector));
813 		if (unlikely(r < 0)) {
814 			dm_integrity_io_error(ic, "crypto_shash_update", r);
815 			goto err;
816 		}
817 	}
818 
819 	size = crypto_shash_digestsize(ic->journal_mac);
820 
821 	if (likely(size <= JOURNAL_MAC_SIZE)) {
822 		r = crypto_shash_final(desc, result);
823 		if (unlikely(r < 0)) {
824 			dm_integrity_io_error(ic, "crypto_shash_final", r);
825 			goto err;
826 		}
827 		memset(result + size, 0, JOURNAL_MAC_SIZE - size);
828 	} else {
829 		__u8 digest[HASH_MAX_DIGESTSIZE];
830 
831 		if (WARN_ON(size > sizeof(digest))) {
832 			dm_integrity_io_error(ic, "digest_size", -EINVAL);
833 			goto err;
834 		}
835 		r = crypto_shash_final(desc, digest);
836 		if (unlikely(r < 0)) {
837 			dm_integrity_io_error(ic, "crypto_shash_final", r);
838 			goto err;
839 		}
840 		memcpy(result, digest, JOURNAL_MAC_SIZE);
841 	}
842 
843 	return;
844 err:
845 	memset(result, 0, JOURNAL_MAC_SIZE);
846 }
847 
848 static void rw_section_mac(struct dm_integrity_c *ic, unsigned int section, bool wr)
849 {
850 	__u8 result[JOURNAL_MAC_SIZE];
851 	unsigned int j;
852 
853 	if (!ic->journal_mac)
854 		return;
855 
856 	section_mac(ic, section, result);
857 
858 	for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
859 		struct journal_sector *js = access_journal(ic, section, j);
860 
861 		if (likely(wr))
862 			memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
863 		else {
864 			if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR)) {
865 				dm_integrity_io_error(ic, "journal mac", -EILSEQ);
866 				dm_audit_log_target(DM_MSG_PREFIX, "mac-journal", ic->ti, 0);
867 			}
868 		}
869 	}
870 }
871 
872 static void complete_journal_op(void *context)
873 {
874 	struct journal_completion *comp = context;
875 
876 	BUG_ON(!atomic_read(&comp->in_flight));
877 	if (likely(atomic_dec_and_test(&comp->in_flight)))
878 		complete(&comp->comp);
879 }
880 
881 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
882 			unsigned int n_sections, struct journal_completion *comp)
883 {
884 	struct async_submit_ctl submit;
885 	size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
886 	unsigned int pl_index, pl_offset, section_index;
887 	struct page_list *source_pl, *target_pl;
888 
889 	if (likely(encrypt)) {
890 		source_pl = ic->journal;
891 		target_pl = ic->journal_io;
892 	} else {
893 		source_pl = ic->journal_io;
894 		target_pl = ic->journal;
895 	}
896 
897 	page_list_location(ic, section, 0, &pl_index, &pl_offset);
898 
899 	atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
900 
901 	init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
902 
903 	section_index = pl_index;
904 
905 	do {
906 		size_t this_step;
907 		struct page *src_pages[2];
908 		struct page *dst_page;
909 
910 		while (unlikely(pl_index == section_index)) {
911 			unsigned int dummy;
912 
913 			if (likely(encrypt))
914 				rw_section_mac(ic, section, true);
915 			section++;
916 			n_sections--;
917 			if (!n_sections)
918 				break;
919 			page_list_location(ic, section, 0, &section_index, &dummy);
920 		}
921 
922 		this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
923 		dst_page = target_pl[pl_index].page;
924 		src_pages[0] = source_pl[pl_index].page;
925 		src_pages[1] = ic->journal_xor[pl_index].page;
926 
927 		async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
928 
929 		pl_index++;
930 		pl_offset = 0;
931 		n_bytes -= this_step;
932 	} while (n_bytes);
933 
934 	BUG_ON(n_sections);
935 
936 	async_tx_issue_pending_all();
937 }
938 
939 static void complete_journal_encrypt(void *data, int err)
940 {
941 	struct journal_completion *comp = data;
942 
943 	if (unlikely(err)) {
944 		if (likely(err == -EINPROGRESS)) {
945 			complete(&comp->ic->crypto_backoff);
946 			return;
947 		}
948 		dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
949 	}
950 	complete_journal_op(comp);
951 }
952 
953 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
954 {
955 	int r;
956 
957 	skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
958 				      complete_journal_encrypt, comp);
959 	if (likely(encrypt))
960 		r = crypto_skcipher_encrypt(req);
961 	else
962 		r = crypto_skcipher_decrypt(req);
963 	if (likely(!r))
964 		return false;
965 	if (likely(r == -EINPROGRESS))
966 		return true;
967 	if (likely(r == -EBUSY)) {
968 		wait_for_completion(&comp->ic->crypto_backoff);
969 		reinit_completion(&comp->ic->crypto_backoff);
970 		return true;
971 	}
972 	dm_integrity_io_error(comp->ic, "encrypt", r);
973 	return false;
974 }
975 
976 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
977 			  unsigned int n_sections, struct journal_completion *comp)
978 {
979 	struct scatterlist **source_sg;
980 	struct scatterlist **target_sg;
981 
982 	atomic_add(2, &comp->in_flight);
983 
984 	if (likely(encrypt)) {
985 		source_sg = ic->journal_scatterlist;
986 		target_sg = ic->journal_io_scatterlist;
987 	} else {
988 		source_sg = ic->journal_io_scatterlist;
989 		target_sg = ic->journal_scatterlist;
990 	}
991 
992 	do {
993 		struct skcipher_request *req;
994 		unsigned int ivsize;
995 		char *iv;
996 
997 		if (likely(encrypt))
998 			rw_section_mac(ic, section, true);
999 
1000 		req = ic->sk_requests[section];
1001 		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1002 		iv = req->iv;
1003 
1004 		memcpy(iv, iv + ivsize, ivsize);
1005 
1006 		req->src = source_sg[section];
1007 		req->dst = target_sg[section];
1008 
1009 		if (unlikely(do_crypt(encrypt, req, comp)))
1010 			atomic_inc(&comp->in_flight);
1011 
1012 		section++;
1013 		n_sections--;
1014 	} while (n_sections);
1015 
1016 	atomic_dec(&comp->in_flight);
1017 	complete_journal_op(comp);
1018 }
1019 
1020 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
1021 			    unsigned int n_sections, struct journal_completion *comp)
1022 {
1023 	if (ic->journal_xor)
1024 		return xor_journal(ic, encrypt, section, n_sections, comp);
1025 	else
1026 		return crypt_journal(ic, encrypt, section, n_sections, comp);
1027 }
1028 
1029 static void complete_journal_io(unsigned long error, void *context)
1030 {
1031 	struct journal_completion *comp = context;
1032 
1033 	if (unlikely(error != 0))
1034 		dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1035 	complete_journal_op(comp);
1036 }
1037 
1038 static void rw_journal_sectors(struct dm_integrity_c *ic, blk_opf_t opf,
1039 			       unsigned int sector, unsigned int n_sectors,
1040 			       struct journal_completion *comp)
1041 {
1042 	struct dm_io_request io_req;
1043 	struct dm_io_region io_loc;
1044 	unsigned int pl_index, pl_offset;
1045 	int r;
1046 
1047 	if (unlikely(dm_integrity_failed(ic))) {
1048 		if (comp)
1049 			complete_journal_io(-1UL, comp);
1050 		return;
1051 	}
1052 
1053 	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1054 	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1055 
1056 	io_req.bi_opf = opf;
1057 	io_req.mem.type = DM_IO_PAGE_LIST;
1058 	if (ic->journal_io)
1059 		io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1060 	else
1061 		io_req.mem.ptr.pl = &ic->journal[pl_index];
1062 	io_req.mem.offset = pl_offset;
1063 	if (likely(comp != NULL)) {
1064 		io_req.notify.fn = complete_journal_io;
1065 		io_req.notify.context = comp;
1066 	} else {
1067 		io_req.notify.fn = NULL;
1068 	}
1069 	io_req.client = ic->io;
1070 	io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1071 	io_loc.sector = ic->start + SB_SECTORS + sector;
1072 	io_loc.count = n_sectors;
1073 
1074 	r = dm_io(&io_req, 1, &io_loc, NULL);
1075 	if (unlikely(r)) {
1076 		dm_integrity_io_error(ic, (opf & REQ_OP_MASK) == REQ_OP_READ ?
1077 				      "reading journal" : "writing journal", r);
1078 		if (comp) {
1079 			WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1080 			complete_journal_io(-1UL, comp);
1081 		}
1082 	}
1083 }
1084 
1085 static void rw_journal(struct dm_integrity_c *ic, blk_opf_t opf,
1086 		       unsigned int section, unsigned int n_sections,
1087 		       struct journal_completion *comp)
1088 {
1089 	unsigned int sector, n_sectors;
1090 
1091 	sector = section * ic->journal_section_sectors;
1092 	n_sectors = n_sections * ic->journal_section_sectors;
1093 
1094 	rw_journal_sectors(ic, opf, sector, n_sectors, comp);
1095 }
1096 
1097 static void write_journal(struct dm_integrity_c *ic, unsigned int commit_start, unsigned int commit_sections)
1098 {
1099 	struct journal_completion io_comp;
1100 	struct journal_completion crypt_comp_1;
1101 	struct journal_completion crypt_comp_2;
1102 	unsigned int i;
1103 
1104 	io_comp.ic = ic;
1105 	init_completion(&io_comp.comp);
1106 
1107 	if (commit_start + commit_sections <= ic->journal_sections) {
1108 		io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1109 		if (ic->journal_io) {
1110 			crypt_comp_1.ic = ic;
1111 			init_completion(&crypt_comp_1.comp);
1112 			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1113 			encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1114 			wait_for_completion_io(&crypt_comp_1.comp);
1115 		} else {
1116 			for (i = 0; i < commit_sections; i++)
1117 				rw_section_mac(ic, commit_start + i, true);
1118 		}
1119 		rw_journal(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, commit_start,
1120 			   commit_sections, &io_comp);
1121 	} else {
1122 		unsigned int to_end;
1123 
1124 		io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1125 		to_end = ic->journal_sections - commit_start;
1126 		if (ic->journal_io) {
1127 			crypt_comp_1.ic = ic;
1128 			init_completion(&crypt_comp_1.comp);
1129 			crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1130 			encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1131 			if (try_wait_for_completion(&crypt_comp_1.comp)) {
1132 				rw_journal(ic, REQ_OP_WRITE | REQ_FUA,
1133 					   commit_start, to_end, &io_comp);
1134 				reinit_completion(&crypt_comp_1.comp);
1135 				crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1136 				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1137 				wait_for_completion_io(&crypt_comp_1.comp);
1138 			} else {
1139 				crypt_comp_2.ic = ic;
1140 				init_completion(&crypt_comp_2.comp);
1141 				crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1142 				encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1143 				wait_for_completion_io(&crypt_comp_1.comp);
1144 				rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1145 				wait_for_completion_io(&crypt_comp_2.comp);
1146 			}
1147 		} else {
1148 			for (i = 0; i < to_end; i++)
1149 				rw_section_mac(ic, commit_start + i, true);
1150 			rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1151 			for (i = 0; i < commit_sections - to_end; i++)
1152 				rw_section_mac(ic, i, true);
1153 		}
1154 		rw_journal(ic, REQ_OP_WRITE | REQ_FUA, 0, commit_sections - to_end, &io_comp);
1155 	}
1156 
1157 	wait_for_completion_io(&io_comp.comp);
1158 }
1159 
1160 static void copy_from_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
1161 			      unsigned int n_sectors, sector_t target, io_notify_fn fn, void *data)
1162 {
1163 	struct dm_io_request io_req;
1164 	struct dm_io_region io_loc;
1165 	int r;
1166 	unsigned int sector, pl_index, pl_offset;
1167 
1168 	BUG_ON((target | n_sectors | offset) & (unsigned int)(ic->sectors_per_block - 1));
1169 
1170 	if (unlikely(dm_integrity_failed(ic))) {
1171 		fn(-1UL, data);
1172 		return;
1173 	}
1174 
1175 	sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1176 
1177 	pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1178 	pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1179 
1180 	io_req.bi_opf = REQ_OP_WRITE;
1181 	io_req.mem.type = DM_IO_PAGE_LIST;
1182 	io_req.mem.ptr.pl = &ic->journal[pl_index];
1183 	io_req.mem.offset = pl_offset;
1184 	io_req.notify.fn = fn;
1185 	io_req.notify.context = data;
1186 	io_req.client = ic->io;
1187 	io_loc.bdev = ic->dev->bdev;
1188 	io_loc.sector = target;
1189 	io_loc.count = n_sectors;
1190 
1191 	r = dm_io(&io_req, 1, &io_loc, NULL);
1192 	if (unlikely(r)) {
1193 		WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1194 		fn(-1UL, data);
1195 	}
1196 }
1197 
1198 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1199 {
1200 	return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1201 	       range1->logical_sector + range1->n_sectors > range2->logical_sector;
1202 }
1203 
1204 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1205 {
1206 	struct rb_node **n = &ic->in_progress.rb_node;
1207 	struct rb_node *parent;
1208 
1209 	BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned int)(ic->sectors_per_block - 1));
1210 
1211 	if (likely(check_waiting)) {
1212 		struct dm_integrity_range *range;
1213 
1214 		list_for_each_entry(range, &ic->wait_list, wait_entry) {
1215 			if (unlikely(ranges_overlap(range, new_range)))
1216 				return false;
1217 		}
1218 	}
1219 
1220 	parent = NULL;
1221 
1222 	while (*n) {
1223 		struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1224 
1225 		parent = *n;
1226 		if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector)
1227 			n = &range->node.rb_left;
1228 		else if (new_range->logical_sector >= range->logical_sector + range->n_sectors)
1229 			n = &range->node.rb_right;
1230 		else
1231 			return false;
1232 	}
1233 
1234 	rb_link_node(&new_range->node, parent, n);
1235 	rb_insert_color(&new_range->node, &ic->in_progress);
1236 
1237 	return true;
1238 }
1239 
1240 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1241 {
1242 	rb_erase(&range->node, &ic->in_progress);
1243 	while (unlikely(!list_empty(&ic->wait_list))) {
1244 		struct dm_integrity_range *last_range =
1245 			list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1246 		struct task_struct *last_range_task;
1247 
1248 		last_range_task = last_range->task;
1249 		list_del(&last_range->wait_entry);
1250 		if (!add_new_range(ic, last_range, false)) {
1251 			last_range->task = last_range_task;
1252 			list_add(&last_range->wait_entry, &ic->wait_list);
1253 			break;
1254 		}
1255 		last_range->waiting = false;
1256 		wake_up_process(last_range_task);
1257 	}
1258 }
1259 
1260 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1261 {
1262 	unsigned long flags;
1263 
1264 	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1265 	remove_range_unlocked(ic, range);
1266 	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1267 }
1268 
1269 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1270 {
1271 	new_range->waiting = true;
1272 	list_add_tail(&new_range->wait_entry, &ic->wait_list);
1273 	new_range->task = current;
1274 	do {
1275 		__set_current_state(TASK_UNINTERRUPTIBLE);
1276 		spin_unlock_irq(&ic->endio_wait.lock);
1277 		io_schedule();
1278 		spin_lock_irq(&ic->endio_wait.lock);
1279 	} while (unlikely(new_range->waiting));
1280 }
1281 
1282 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1283 {
1284 	if (unlikely(!add_new_range(ic, new_range, true)))
1285 		wait_and_add_new_range(ic, new_range);
1286 }
1287 
1288 static void init_journal_node(struct journal_node *node)
1289 {
1290 	RB_CLEAR_NODE(&node->node);
1291 	node->sector = (sector_t)-1;
1292 }
1293 
1294 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1295 {
1296 	struct rb_node **link;
1297 	struct rb_node *parent;
1298 
1299 	node->sector = sector;
1300 	BUG_ON(!RB_EMPTY_NODE(&node->node));
1301 
1302 	link = &ic->journal_tree_root.rb_node;
1303 	parent = NULL;
1304 
1305 	while (*link) {
1306 		struct journal_node *j;
1307 
1308 		parent = *link;
1309 		j = container_of(parent, struct journal_node, node);
1310 		if (sector < j->sector)
1311 			link = &j->node.rb_left;
1312 		else
1313 			link = &j->node.rb_right;
1314 	}
1315 
1316 	rb_link_node(&node->node, parent, link);
1317 	rb_insert_color(&node->node, &ic->journal_tree_root);
1318 }
1319 
1320 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1321 {
1322 	BUG_ON(RB_EMPTY_NODE(&node->node));
1323 	rb_erase(&node->node, &ic->journal_tree_root);
1324 	init_journal_node(node);
1325 }
1326 
1327 #define NOT_FOUND	(-1U)
1328 
1329 static unsigned int find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1330 {
1331 	struct rb_node *n = ic->journal_tree_root.rb_node;
1332 	unsigned int found = NOT_FOUND;
1333 
1334 	*next_sector = (sector_t)-1;
1335 	while (n) {
1336 		struct journal_node *j = container_of(n, struct journal_node, node);
1337 
1338 		if (sector == j->sector)
1339 			found = j - ic->journal_tree;
1340 
1341 		if (sector < j->sector) {
1342 			*next_sector = j->sector;
1343 			n = j->node.rb_left;
1344 		} else
1345 			n = j->node.rb_right;
1346 	}
1347 
1348 	return found;
1349 }
1350 
1351 static bool test_journal_node(struct dm_integrity_c *ic, unsigned int pos, sector_t sector)
1352 {
1353 	struct journal_node *node, *next_node;
1354 	struct rb_node *next;
1355 
1356 	if (unlikely(pos >= ic->journal_entries))
1357 		return false;
1358 	node = &ic->journal_tree[pos];
1359 	if (unlikely(RB_EMPTY_NODE(&node->node)))
1360 		return false;
1361 	if (unlikely(node->sector != sector))
1362 		return false;
1363 
1364 	next = rb_next(&node->node);
1365 	if (unlikely(!next))
1366 		return true;
1367 
1368 	next_node = container_of(next, struct journal_node, node);
1369 	return next_node->sector != sector;
1370 }
1371 
1372 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1373 {
1374 	struct rb_node *next;
1375 	struct journal_node *next_node;
1376 	unsigned int next_section;
1377 
1378 	BUG_ON(RB_EMPTY_NODE(&node->node));
1379 
1380 	next = rb_next(&node->node);
1381 	if (unlikely(!next))
1382 		return false;
1383 
1384 	next_node = container_of(next, struct journal_node, node);
1385 
1386 	if (next_node->sector != node->sector)
1387 		return false;
1388 
1389 	next_section = (unsigned int)(next_node - ic->journal_tree) / ic->journal_section_entries;
1390 	if (next_section >= ic->committed_section &&
1391 	    next_section < ic->committed_section + ic->n_committed_sections)
1392 		return true;
1393 	if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1394 		return true;
1395 
1396 	return false;
1397 }
1398 
1399 #define TAG_READ	0
1400 #define TAG_WRITE	1
1401 #define TAG_CMP		2
1402 
1403 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1404 			       unsigned int *metadata_offset, unsigned int total_size, int op)
1405 {
1406 #define MAY_BE_FILLER		1
1407 #define MAY_BE_HASH		2
1408 	unsigned int hash_offset = 0;
1409 	unsigned int may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1410 
1411 	do {
1412 		unsigned char *data, *dp;
1413 		struct dm_buffer *b;
1414 		unsigned int to_copy;
1415 		int r;
1416 
1417 		r = dm_integrity_failed(ic);
1418 		if (unlikely(r))
1419 			return r;
1420 
1421 		data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1422 		if (IS_ERR(data))
1423 			return PTR_ERR(data);
1424 
1425 		to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1426 		dp = data + *metadata_offset;
1427 		if (op == TAG_READ) {
1428 			memcpy(tag, dp, to_copy);
1429 		} else if (op == TAG_WRITE) {
1430 			if (memcmp(dp, tag, to_copy)) {
1431 				memcpy(dp, tag, to_copy);
1432 				dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1433 			}
1434 		} else {
1435 			/* e.g.: op == TAG_CMP */
1436 
1437 			if (likely(is_power_of_2(ic->tag_size))) {
1438 				if (unlikely(memcmp(dp, tag, to_copy)))
1439 					if (unlikely(!ic->discard) ||
1440 					    unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1441 						goto thorough_test;
1442 				}
1443 			} else {
1444 				unsigned int i, ts;
1445 thorough_test:
1446 				ts = total_size;
1447 
1448 				for (i = 0; i < to_copy; i++, ts--) {
1449 					if (unlikely(dp[i] != tag[i]))
1450 						may_be &= ~MAY_BE_HASH;
1451 					if (likely(dp[i] != DISCARD_FILLER))
1452 						may_be &= ~MAY_BE_FILLER;
1453 					hash_offset++;
1454 					if (unlikely(hash_offset == ic->tag_size)) {
1455 						if (unlikely(!may_be)) {
1456 							dm_bufio_release(b);
1457 							return ts;
1458 						}
1459 						hash_offset = 0;
1460 						may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1461 					}
1462 				}
1463 			}
1464 		}
1465 		dm_bufio_release(b);
1466 
1467 		tag += to_copy;
1468 		*metadata_offset += to_copy;
1469 		if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1470 			(*metadata_block)++;
1471 			*metadata_offset = 0;
1472 		}
1473 
1474 		if (unlikely(!is_power_of_2(ic->tag_size)))
1475 			hash_offset = (hash_offset + to_copy) % ic->tag_size;
1476 
1477 		total_size -= to_copy;
1478 	} while (unlikely(total_size));
1479 
1480 	return 0;
1481 #undef MAY_BE_FILLER
1482 #undef MAY_BE_HASH
1483 }
1484 
1485 struct flush_request {
1486 	struct dm_io_request io_req;
1487 	struct dm_io_region io_reg;
1488 	struct dm_integrity_c *ic;
1489 	struct completion comp;
1490 };
1491 
1492 static void flush_notify(unsigned long error, void *fr_)
1493 {
1494 	struct flush_request *fr = fr_;
1495 
1496 	if (unlikely(error != 0))
1497 		dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1498 	complete(&fr->comp);
1499 }
1500 
1501 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1502 {
1503 	int r;
1504 	struct flush_request fr;
1505 
1506 	if (!ic->meta_dev)
1507 		flush_data = false;
1508 	if (flush_data) {
1509 		fr.io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC,
1510 		fr.io_req.mem.type = DM_IO_KMEM,
1511 		fr.io_req.mem.ptr.addr = NULL,
1512 		fr.io_req.notify.fn = flush_notify,
1513 		fr.io_req.notify.context = &fr;
1514 		fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1515 		fr.io_reg.bdev = ic->dev->bdev,
1516 		fr.io_reg.sector = 0,
1517 		fr.io_reg.count = 0,
1518 		fr.ic = ic;
1519 		init_completion(&fr.comp);
1520 		r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1521 		BUG_ON(r);
1522 	}
1523 
1524 	r = dm_bufio_write_dirty_buffers(ic->bufio);
1525 	if (unlikely(r))
1526 		dm_integrity_io_error(ic, "writing tags", r);
1527 
1528 	if (flush_data)
1529 		wait_for_completion(&fr.comp);
1530 }
1531 
1532 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1533 {
1534 	DECLARE_WAITQUEUE(wait, current);
1535 
1536 	__add_wait_queue(&ic->endio_wait, &wait);
1537 	__set_current_state(TASK_UNINTERRUPTIBLE);
1538 	spin_unlock_irq(&ic->endio_wait.lock);
1539 	io_schedule();
1540 	spin_lock_irq(&ic->endio_wait.lock);
1541 	__remove_wait_queue(&ic->endio_wait, &wait);
1542 }
1543 
1544 static void autocommit_fn(struct timer_list *t)
1545 {
1546 	struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1547 
1548 	if (likely(!dm_integrity_failed(ic)))
1549 		queue_work(ic->commit_wq, &ic->commit_work);
1550 }
1551 
1552 static void schedule_autocommit(struct dm_integrity_c *ic)
1553 {
1554 	if (!timer_pending(&ic->autocommit_timer))
1555 		mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1556 }
1557 
1558 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1559 {
1560 	struct bio *bio;
1561 	unsigned long flags;
1562 
1563 	spin_lock_irqsave(&ic->endio_wait.lock, flags);
1564 	bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1565 	bio_list_add(&ic->flush_bio_list, bio);
1566 	spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1567 
1568 	queue_work(ic->commit_wq, &ic->commit_work);
1569 }
1570 
1571 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1572 {
1573 	int r;
1574 
1575 	r = dm_integrity_failed(ic);
1576 	if (unlikely(r) && !bio->bi_status)
1577 		bio->bi_status = errno_to_blk_status(r);
1578 	if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1579 		unsigned long flags;
1580 
1581 		spin_lock_irqsave(&ic->endio_wait.lock, flags);
1582 		bio_list_add(&ic->synchronous_bios, bio);
1583 		queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1584 		spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1585 		return;
1586 	}
1587 	bio_endio(bio);
1588 }
1589 
1590 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1591 {
1592 	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1593 
1594 	if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1595 		submit_flush_bio(ic, dio);
1596 	else
1597 		do_endio(ic, bio);
1598 }
1599 
1600 static void dec_in_flight(struct dm_integrity_io *dio)
1601 {
1602 	if (atomic_dec_and_test(&dio->in_flight)) {
1603 		struct dm_integrity_c *ic = dio->ic;
1604 		struct bio *bio;
1605 
1606 		remove_range(ic, &dio->range);
1607 
1608 		if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1609 			schedule_autocommit(ic);
1610 
1611 		bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1612 		if (unlikely(dio->bi_status) && !bio->bi_status)
1613 			bio->bi_status = dio->bi_status;
1614 		if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1615 			dio->range.logical_sector += dio->range.n_sectors;
1616 			bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1617 			INIT_WORK(&dio->work, integrity_bio_wait);
1618 			queue_work(ic->offload_wq, &dio->work);
1619 			return;
1620 		}
1621 		do_endio_flush(ic, dio);
1622 	}
1623 }
1624 
1625 static void integrity_end_io(struct bio *bio)
1626 {
1627 	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1628 
1629 	dm_bio_restore(&dio->bio_details, bio);
1630 	if (bio->bi_integrity)
1631 		bio->bi_opf |= REQ_INTEGRITY;
1632 
1633 	if (dio->completion)
1634 		complete(dio->completion);
1635 
1636 	dec_in_flight(dio);
1637 }
1638 
1639 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1640 				      const char *data, char *result)
1641 {
1642 	__le64 sector_le = cpu_to_le64(sector);
1643 	SHASH_DESC_ON_STACK(req, ic->internal_hash);
1644 	int r;
1645 	unsigned int digest_size;
1646 
1647 	req->tfm = ic->internal_hash;
1648 
1649 	r = crypto_shash_init(req);
1650 	if (unlikely(r < 0)) {
1651 		dm_integrity_io_error(ic, "crypto_shash_init", r);
1652 		goto failed;
1653 	}
1654 
1655 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1656 		r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1657 		if (unlikely(r < 0)) {
1658 			dm_integrity_io_error(ic, "crypto_shash_update", r);
1659 			goto failed;
1660 		}
1661 	}
1662 
1663 	r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof(sector_le));
1664 	if (unlikely(r < 0)) {
1665 		dm_integrity_io_error(ic, "crypto_shash_update", r);
1666 		goto failed;
1667 	}
1668 
1669 	r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1670 	if (unlikely(r < 0)) {
1671 		dm_integrity_io_error(ic, "crypto_shash_update", r);
1672 		goto failed;
1673 	}
1674 
1675 	r = crypto_shash_final(req, result);
1676 	if (unlikely(r < 0)) {
1677 		dm_integrity_io_error(ic, "crypto_shash_final", r);
1678 		goto failed;
1679 	}
1680 
1681 	digest_size = crypto_shash_digestsize(ic->internal_hash);
1682 	if (unlikely(digest_size < ic->tag_size))
1683 		memset(result + digest_size, 0, ic->tag_size - digest_size);
1684 
1685 	return;
1686 
1687 failed:
1688 	/* this shouldn't happen anyway, the hash functions have no reason to fail */
1689 	get_random_bytes(result, ic->tag_size);
1690 }
1691 
1692 static void integrity_metadata(struct work_struct *w)
1693 {
1694 	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1695 	struct dm_integrity_c *ic = dio->ic;
1696 
1697 	int r;
1698 
1699 	if (ic->internal_hash) {
1700 		struct bvec_iter iter;
1701 		struct bio_vec bv;
1702 		unsigned int digest_size = crypto_shash_digestsize(ic->internal_hash);
1703 		struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1704 		char *checksums;
1705 		unsigned int extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1706 		char checksums_onstack[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1707 		sector_t sector;
1708 		unsigned int sectors_to_process;
1709 
1710 		if (unlikely(ic->mode == 'R'))
1711 			goto skip_io;
1712 
1713 		if (likely(dio->op != REQ_OP_DISCARD))
1714 			checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1715 					    GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1716 		else
1717 			checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1718 		if (!checksums) {
1719 			checksums = checksums_onstack;
1720 			if (WARN_ON(extra_space &&
1721 				    digest_size > sizeof(checksums_onstack))) {
1722 				r = -EINVAL;
1723 				goto error;
1724 			}
1725 		}
1726 
1727 		if (unlikely(dio->op == REQ_OP_DISCARD)) {
1728 			unsigned int bi_size = dio->bio_details.bi_iter.bi_size;
1729 			unsigned int max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1730 			unsigned int max_blocks = max_size / ic->tag_size;
1731 
1732 			memset(checksums, DISCARD_FILLER, max_size);
1733 
1734 			while (bi_size) {
1735 				unsigned int this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1736 
1737 				this_step_blocks = min(this_step_blocks, max_blocks);
1738 				r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1739 							this_step_blocks * ic->tag_size, TAG_WRITE);
1740 				if (unlikely(r)) {
1741 					if (likely(checksums != checksums_onstack))
1742 						kfree(checksums);
1743 					goto error;
1744 				}
1745 
1746 				bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1747 			}
1748 
1749 			if (likely(checksums != checksums_onstack))
1750 				kfree(checksums);
1751 			goto skip_io;
1752 		}
1753 
1754 		sector = dio->range.logical_sector;
1755 		sectors_to_process = dio->range.n_sectors;
1756 
1757 		__bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1758 			struct bio_vec bv_copy = bv;
1759 			unsigned int pos;
1760 			char *mem, *checksums_ptr;
1761 
1762 again:
1763 			mem = bvec_kmap_local(&bv_copy);
1764 			pos = 0;
1765 			checksums_ptr = checksums;
1766 			do {
1767 				integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1768 				checksums_ptr += ic->tag_size;
1769 				sectors_to_process -= ic->sectors_per_block;
1770 				pos += ic->sectors_per_block << SECTOR_SHIFT;
1771 				sector += ic->sectors_per_block;
1772 			} while (pos < bv_copy.bv_len && sectors_to_process && checksums != checksums_onstack);
1773 			kunmap_local(mem);
1774 
1775 			r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1776 						checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1777 			if (unlikely(r)) {
1778 				if (r > 0) {
1779 					sector_t s;
1780 
1781 					s = sector - ((r + ic->tag_size - 1) / ic->tag_size);
1782 					DMERR_LIMIT("%pg: Checksum failed at sector 0x%llx",
1783 						    bio->bi_bdev, s);
1784 					r = -EILSEQ;
1785 					atomic64_inc(&ic->number_of_mismatches);
1786 					dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
1787 							 bio, s, 0);
1788 				}
1789 				if (likely(checksums != checksums_onstack))
1790 					kfree(checksums);
1791 				goto error;
1792 			}
1793 
1794 			if (!sectors_to_process)
1795 				break;
1796 
1797 			if (unlikely(pos < bv_copy.bv_len)) {
1798 				bv_copy.bv_offset += pos;
1799 				bv_copy.bv_len -= pos;
1800 				goto again;
1801 			}
1802 		}
1803 
1804 		if (likely(checksums != checksums_onstack))
1805 			kfree(checksums);
1806 	} else {
1807 		struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1808 
1809 		if (bip) {
1810 			struct bio_vec biv;
1811 			struct bvec_iter iter;
1812 			unsigned int data_to_process = dio->range.n_sectors;
1813 
1814 			sector_to_block(ic, data_to_process);
1815 			data_to_process *= ic->tag_size;
1816 
1817 			bip_for_each_vec(biv, bip, iter) {
1818 				unsigned char *tag;
1819 				unsigned int this_len;
1820 
1821 				BUG_ON(PageHighMem(biv.bv_page));
1822 				tag = bvec_virt(&biv);
1823 				this_len = min(biv.bv_len, data_to_process);
1824 				r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1825 							this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1826 				if (unlikely(r))
1827 					goto error;
1828 				data_to_process -= this_len;
1829 				if (!data_to_process)
1830 					break;
1831 			}
1832 		}
1833 	}
1834 skip_io:
1835 	dec_in_flight(dio);
1836 	return;
1837 error:
1838 	dio->bi_status = errno_to_blk_status(r);
1839 	dec_in_flight(dio);
1840 }
1841 
1842 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1843 {
1844 	struct dm_integrity_c *ic = ti->private;
1845 	struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1846 	struct bio_integrity_payload *bip;
1847 
1848 	sector_t area, offset;
1849 
1850 	dio->ic = ic;
1851 	dio->bi_status = 0;
1852 	dio->op = bio_op(bio);
1853 
1854 	if (unlikely(dio->op == REQ_OP_DISCARD)) {
1855 		if (ti->max_io_len) {
1856 			sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1857 			unsigned int log2_max_io_len = __fls(ti->max_io_len);
1858 			sector_t start_boundary = sec >> log2_max_io_len;
1859 			sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1860 
1861 			if (start_boundary < end_boundary) {
1862 				sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1863 
1864 				dm_accept_partial_bio(bio, len);
1865 			}
1866 		}
1867 	}
1868 
1869 	if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1870 		submit_flush_bio(ic, dio);
1871 		return DM_MAPIO_SUBMITTED;
1872 	}
1873 
1874 	dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1875 	dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1876 	if (unlikely(dio->fua)) {
1877 		/*
1878 		 * Don't pass down the FUA flag because we have to flush
1879 		 * disk cache anyway.
1880 		 */
1881 		bio->bi_opf &= ~REQ_FUA;
1882 	}
1883 	if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1884 		DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1885 		      dio->range.logical_sector, bio_sectors(bio),
1886 		      ic->provided_data_sectors);
1887 		return DM_MAPIO_KILL;
1888 	}
1889 	if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned int)(ic->sectors_per_block - 1))) {
1890 		DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1891 		      ic->sectors_per_block,
1892 		      dio->range.logical_sector, bio_sectors(bio));
1893 		return DM_MAPIO_KILL;
1894 	}
1895 
1896 	if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1897 		struct bvec_iter iter;
1898 		struct bio_vec bv;
1899 
1900 		bio_for_each_segment(bv, bio, iter) {
1901 			if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1902 				DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1903 					bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1904 				return DM_MAPIO_KILL;
1905 			}
1906 		}
1907 	}
1908 
1909 	bip = bio_integrity(bio);
1910 	if (!ic->internal_hash) {
1911 		if (bip) {
1912 			unsigned int wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1913 
1914 			if (ic->log2_tag_size >= 0)
1915 				wanted_tag_size <<= ic->log2_tag_size;
1916 			else
1917 				wanted_tag_size *= ic->tag_size;
1918 			if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1919 				DMERR("Invalid integrity data size %u, expected %u",
1920 				      bip->bip_iter.bi_size, wanted_tag_size);
1921 				return DM_MAPIO_KILL;
1922 			}
1923 		}
1924 	} else {
1925 		if (unlikely(bip != NULL)) {
1926 			DMERR("Unexpected integrity data when using internal hash");
1927 			return DM_MAPIO_KILL;
1928 		}
1929 	}
1930 
1931 	if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1932 		return DM_MAPIO_KILL;
1933 
1934 	get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1935 	dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1936 	bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1937 
1938 	dm_integrity_map_continue(dio, true);
1939 	return DM_MAPIO_SUBMITTED;
1940 }
1941 
1942 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1943 				 unsigned int journal_section, unsigned int journal_entry)
1944 {
1945 	struct dm_integrity_c *ic = dio->ic;
1946 	sector_t logical_sector;
1947 	unsigned int n_sectors;
1948 
1949 	logical_sector = dio->range.logical_sector;
1950 	n_sectors = dio->range.n_sectors;
1951 	do {
1952 		struct bio_vec bv = bio_iovec(bio);
1953 		char *mem;
1954 
1955 		if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1956 			bv.bv_len = n_sectors << SECTOR_SHIFT;
1957 		n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1958 		bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1959 retry_kmap:
1960 		mem = kmap_local_page(bv.bv_page);
1961 		if (likely(dio->op == REQ_OP_WRITE))
1962 			flush_dcache_page(bv.bv_page);
1963 
1964 		do {
1965 			struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1966 
1967 			if (unlikely(dio->op == REQ_OP_READ)) {
1968 				struct journal_sector *js;
1969 				char *mem_ptr;
1970 				unsigned int s;
1971 
1972 				if (unlikely(journal_entry_is_inprogress(je))) {
1973 					flush_dcache_page(bv.bv_page);
1974 					kunmap_local(mem);
1975 
1976 					__io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1977 					goto retry_kmap;
1978 				}
1979 				smp_rmb();
1980 				BUG_ON(journal_entry_get_sector(je) != logical_sector);
1981 				js = access_journal_data(ic, journal_section, journal_entry);
1982 				mem_ptr = mem + bv.bv_offset;
1983 				s = 0;
1984 				do {
1985 					memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1986 					*(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1987 					js++;
1988 					mem_ptr += 1 << SECTOR_SHIFT;
1989 				} while (++s < ic->sectors_per_block);
1990 #ifdef INTERNAL_VERIFY
1991 				if (ic->internal_hash) {
1992 					char checksums_onstack[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1993 
1994 					integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1995 					if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1996 						DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1997 							    logical_sector);
1998 						dm_audit_log_bio(DM_MSG_PREFIX, "journal-checksum",
1999 								 bio, logical_sector, 0);
2000 					}
2001 				}
2002 #endif
2003 			}
2004 
2005 			if (!ic->internal_hash) {
2006 				struct bio_integrity_payload *bip = bio_integrity(bio);
2007 				unsigned int tag_todo = ic->tag_size;
2008 				char *tag_ptr = journal_entry_tag(ic, je);
2009 
2010 				if (bip) {
2011 					do {
2012 						struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2013 						unsigned int tag_now = min(biv.bv_len, tag_todo);
2014 						char *tag_addr;
2015 
2016 						BUG_ON(PageHighMem(biv.bv_page));
2017 						tag_addr = bvec_virt(&biv);
2018 						if (likely(dio->op == REQ_OP_WRITE))
2019 							memcpy(tag_ptr, tag_addr, tag_now);
2020 						else
2021 							memcpy(tag_addr, tag_ptr, tag_now);
2022 						bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2023 						tag_ptr += tag_now;
2024 						tag_todo -= tag_now;
2025 					} while (unlikely(tag_todo));
2026 				} else if (likely(dio->op == REQ_OP_WRITE))
2027 					memset(tag_ptr, 0, tag_todo);
2028 			}
2029 
2030 			if (likely(dio->op == REQ_OP_WRITE)) {
2031 				struct journal_sector *js;
2032 				unsigned int s;
2033 
2034 				js = access_journal_data(ic, journal_section, journal_entry);
2035 				memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2036 
2037 				s = 0;
2038 				do {
2039 					je->last_bytes[s] = js[s].commit_id;
2040 				} while (++s < ic->sectors_per_block);
2041 
2042 				if (ic->internal_hash) {
2043 					unsigned int digest_size = crypto_shash_digestsize(ic->internal_hash);
2044 
2045 					if (unlikely(digest_size > ic->tag_size)) {
2046 						char checksums_onstack[HASH_MAX_DIGESTSIZE];
2047 
2048 						integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2049 						memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2050 					} else
2051 						integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2052 				}
2053 
2054 				journal_entry_set_sector(je, logical_sector);
2055 			}
2056 			logical_sector += ic->sectors_per_block;
2057 
2058 			journal_entry++;
2059 			if (unlikely(journal_entry == ic->journal_section_entries)) {
2060 				journal_entry = 0;
2061 				journal_section++;
2062 				wraparound_section(ic, &journal_section);
2063 			}
2064 
2065 			bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2066 		} while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2067 
2068 		if (unlikely(dio->op == REQ_OP_READ))
2069 			flush_dcache_page(bv.bv_page);
2070 		kunmap_local(mem);
2071 	} while (n_sectors);
2072 
2073 	if (likely(dio->op == REQ_OP_WRITE)) {
2074 		smp_mb();
2075 		if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2076 			wake_up(&ic->copy_to_journal_wait);
2077 		if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2078 			queue_work(ic->commit_wq, &ic->commit_work);
2079 		else
2080 			schedule_autocommit(ic);
2081 	} else
2082 		remove_range(ic, &dio->range);
2083 
2084 	if (unlikely(bio->bi_iter.bi_size)) {
2085 		sector_t area, offset;
2086 
2087 		dio->range.logical_sector = logical_sector;
2088 		get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2089 		dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2090 		return true;
2091 	}
2092 
2093 	return false;
2094 }
2095 
2096 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2097 {
2098 	struct dm_integrity_c *ic = dio->ic;
2099 	struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2100 	unsigned int journal_section, journal_entry;
2101 	unsigned int journal_read_pos;
2102 	struct completion read_comp;
2103 	bool discard_retried = false;
2104 	bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2105 
2106 	if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2107 		need_sync_io = true;
2108 
2109 	if (need_sync_io && from_map) {
2110 		INIT_WORK(&dio->work, integrity_bio_wait);
2111 		queue_work(ic->offload_wq, &dio->work);
2112 		return;
2113 	}
2114 
2115 lock_retry:
2116 	spin_lock_irq(&ic->endio_wait.lock);
2117 retry:
2118 	if (unlikely(dm_integrity_failed(ic))) {
2119 		spin_unlock_irq(&ic->endio_wait.lock);
2120 		do_endio(ic, bio);
2121 		return;
2122 	}
2123 	dio->range.n_sectors = bio_sectors(bio);
2124 	journal_read_pos = NOT_FOUND;
2125 	if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2126 		if (dio->op == REQ_OP_WRITE) {
2127 			unsigned int next_entry, i, pos;
2128 			unsigned int ws, we, range_sectors;
2129 
2130 			dio->range.n_sectors = min(dio->range.n_sectors,
2131 						   (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2132 			if (unlikely(!dio->range.n_sectors)) {
2133 				if (from_map)
2134 					goto offload_to_thread;
2135 				sleep_on_endio_wait(ic);
2136 				goto retry;
2137 			}
2138 			range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2139 			ic->free_sectors -= range_sectors;
2140 			journal_section = ic->free_section;
2141 			journal_entry = ic->free_section_entry;
2142 
2143 			next_entry = ic->free_section_entry + range_sectors;
2144 			ic->free_section_entry = next_entry % ic->journal_section_entries;
2145 			ic->free_section += next_entry / ic->journal_section_entries;
2146 			ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2147 			wraparound_section(ic, &ic->free_section);
2148 
2149 			pos = journal_section * ic->journal_section_entries + journal_entry;
2150 			ws = journal_section;
2151 			we = journal_entry;
2152 			i = 0;
2153 			do {
2154 				struct journal_entry *je;
2155 
2156 				add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2157 				pos++;
2158 				if (unlikely(pos >= ic->journal_entries))
2159 					pos = 0;
2160 
2161 				je = access_journal_entry(ic, ws, we);
2162 				BUG_ON(!journal_entry_is_unused(je));
2163 				journal_entry_set_inprogress(je);
2164 				we++;
2165 				if (unlikely(we == ic->journal_section_entries)) {
2166 					we = 0;
2167 					ws++;
2168 					wraparound_section(ic, &ws);
2169 				}
2170 			} while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2171 
2172 			spin_unlock_irq(&ic->endio_wait.lock);
2173 			goto journal_read_write;
2174 		} else {
2175 			sector_t next_sector;
2176 
2177 			journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2178 			if (likely(journal_read_pos == NOT_FOUND)) {
2179 				if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2180 					dio->range.n_sectors = next_sector - dio->range.logical_sector;
2181 			} else {
2182 				unsigned int i;
2183 				unsigned int jp = journal_read_pos + 1;
2184 
2185 				for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2186 					if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2187 						break;
2188 				}
2189 				dio->range.n_sectors = i;
2190 			}
2191 		}
2192 	}
2193 	if (unlikely(!add_new_range(ic, &dio->range, true))) {
2194 		/*
2195 		 * We must not sleep in the request routine because it could
2196 		 * stall bios on current->bio_list.
2197 		 * So, we offload the bio to a workqueue if we have to sleep.
2198 		 */
2199 		if (from_map) {
2200 offload_to_thread:
2201 			spin_unlock_irq(&ic->endio_wait.lock);
2202 			INIT_WORK(&dio->work, integrity_bio_wait);
2203 			queue_work(ic->wait_wq, &dio->work);
2204 			return;
2205 		}
2206 		if (journal_read_pos != NOT_FOUND)
2207 			dio->range.n_sectors = ic->sectors_per_block;
2208 		wait_and_add_new_range(ic, &dio->range);
2209 		/*
2210 		 * wait_and_add_new_range drops the spinlock, so the journal
2211 		 * may have been changed arbitrarily. We need to recheck.
2212 		 * To simplify the code, we restrict I/O size to just one block.
2213 		 */
2214 		if (journal_read_pos != NOT_FOUND) {
2215 			sector_t next_sector;
2216 			unsigned int new_pos;
2217 
2218 			new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2219 			if (unlikely(new_pos != journal_read_pos)) {
2220 				remove_range_unlocked(ic, &dio->range);
2221 				goto retry;
2222 			}
2223 		}
2224 	}
2225 	if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2226 		sector_t next_sector;
2227 		unsigned int new_pos;
2228 
2229 		new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2230 		if (unlikely(new_pos != NOT_FOUND) ||
2231 		    unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2232 			remove_range_unlocked(ic, &dio->range);
2233 			spin_unlock_irq(&ic->endio_wait.lock);
2234 			queue_work(ic->commit_wq, &ic->commit_work);
2235 			flush_workqueue(ic->commit_wq);
2236 			queue_work(ic->writer_wq, &ic->writer_work);
2237 			flush_workqueue(ic->writer_wq);
2238 			discard_retried = true;
2239 			goto lock_retry;
2240 		}
2241 	}
2242 	spin_unlock_irq(&ic->endio_wait.lock);
2243 
2244 	if (unlikely(journal_read_pos != NOT_FOUND)) {
2245 		journal_section = journal_read_pos / ic->journal_section_entries;
2246 		journal_entry = journal_read_pos % ic->journal_section_entries;
2247 		goto journal_read_write;
2248 	}
2249 
2250 	if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2251 		if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2252 				     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2253 			struct bitmap_block_status *bbs;
2254 
2255 			bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2256 			spin_lock(&bbs->bio_queue_lock);
2257 			bio_list_add(&bbs->bio_queue, bio);
2258 			spin_unlock(&bbs->bio_queue_lock);
2259 			queue_work(ic->writer_wq, &bbs->work);
2260 			return;
2261 		}
2262 	}
2263 
2264 	dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2265 
2266 	if (need_sync_io) {
2267 		init_completion(&read_comp);
2268 		dio->completion = &read_comp;
2269 	} else
2270 		dio->completion = NULL;
2271 
2272 	dm_bio_record(&dio->bio_details, bio);
2273 	bio_set_dev(bio, ic->dev->bdev);
2274 	bio->bi_integrity = NULL;
2275 	bio->bi_opf &= ~REQ_INTEGRITY;
2276 	bio->bi_end_io = integrity_end_io;
2277 	bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2278 
2279 	if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2280 		integrity_metadata(&dio->work);
2281 		dm_integrity_flush_buffers(ic, false);
2282 
2283 		dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2284 		dio->completion = NULL;
2285 
2286 		submit_bio_noacct(bio);
2287 
2288 		return;
2289 	}
2290 
2291 	submit_bio_noacct(bio);
2292 
2293 	if (need_sync_io) {
2294 		wait_for_completion_io(&read_comp);
2295 		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2296 		    dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2297 			goto skip_check;
2298 		if (ic->mode == 'B') {
2299 			if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2300 					     dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2301 				goto skip_check;
2302 		}
2303 
2304 		if (likely(!bio->bi_status))
2305 			integrity_metadata(&dio->work);
2306 		else
2307 skip_check:
2308 			dec_in_flight(dio);
2309 	} else {
2310 		INIT_WORK(&dio->work, integrity_metadata);
2311 		queue_work(ic->metadata_wq, &dio->work);
2312 	}
2313 
2314 	return;
2315 
2316 journal_read_write:
2317 	if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2318 		goto lock_retry;
2319 
2320 	do_endio_flush(ic, dio);
2321 }
2322 
2323 
2324 static void integrity_bio_wait(struct work_struct *w)
2325 {
2326 	struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2327 
2328 	dm_integrity_map_continue(dio, false);
2329 }
2330 
2331 static void pad_uncommitted(struct dm_integrity_c *ic)
2332 {
2333 	if (ic->free_section_entry) {
2334 		ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2335 		ic->free_section_entry = 0;
2336 		ic->free_section++;
2337 		wraparound_section(ic, &ic->free_section);
2338 		ic->n_uncommitted_sections++;
2339 	}
2340 	if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2341 		    (ic->n_uncommitted_sections + ic->n_committed_sections) *
2342 		    ic->journal_section_entries + ic->free_sectors)) {
2343 		DMCRIT("journal_sections %u, journal_section_entries %u, "
2344 		       "n_uncommitted_sections %u, n_committed_sections %u, "
2345 		       "journal_section_entries %u, free_sectors %u",
2346 		       ic->journal_sections, ic->journal_section_entries,
2347 		       ic->n_uncommitted_sections, ic->n_committed_sections,
2348 		       ic->journal_section_entries, ic->free_sectors);
2349 	}
2350 }
2351 
2352 static void integrity_commit(struct work_struct *w)
2353 {
2354 	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2355 	unsigned int commit_start, commit_sections;
2356 	unsigned int i, j, n;
2357 	struct bio *flushes;
2358 
2359 	del_timer(&ic->autocommit_timer);
2360 
2361 	spin_lock_irq(&ic->endio_wait.lock);
2362 	flushes = bio_list_get(&ic->flush_bio_list);
2363 	if (unlikely(ic->mode != 'J')) {
2364 		spin_unlock_irq(&ic->endio_wait.lock);
2365 		dm_integrity_flush_buffers(ic, true);
2366 		goto release_flush_bios;
2367 	}
2368 
2369 	pad_uncommitted(ic);
2370 	commit_start = ic->uncommitted_section;
2371 	commit_sections = ic->n_uncommitted_sections;
2372 	spin_unlock_irq(&ic->endio_wait.lock);
2373 
2374 	if (!commit_sections)
2375 		goto release_flush_bios;
2376 
2377 	ic->wrote_to_journal = true;
2378 
2379 	i = commit_start;
2380 	for (n = 0; n < commit_sections; n++) {
2381 		for (j = 0; j < ic->journal_section_entries; j++) {
2382 			struct journal_entry *je;
2383 
2384 			je = access_journal_entry(ic, i, j);
2385 			io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2386 		}
2387 		for (j = 0; j < ic->journal_section_sectors; j++) {
2388 			struct journal_sector *js;
2389 
2390 			js = access_journal(ic, i, j);
2391 			js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2392 		}
2393 		i++;
2394 		if (unlikely(i >= ic->journal_sections))
2395 			ic->commit_seq = next_commit_seq(ic->commit_seq);
2396 		wraparound_section(ic, &i);
2397 	}
2398 	smp_rmb();
2399 
2400 	write_journal(ic, commit_start, commit_sections);
2401 
2402 	spin_lock_irq(&ic->endio_wait.lock);
2403 	ic->uncommitted_section += commit_sections;
2404 	wraparound_section(ic, &ic->uncommitted_section);
2405 	ic->n_uncommitted_sections -= commit_sections;
2406 	ic->n_committed_sections += commit_sections;
2407 	spin_unlock_irq(&ic->endio_wait.lock);
2408 
2409 	if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2410 		queue_work(ic->writer_wq, &ic->writer_work);
2411 
2412 release_flush_bios:
2413 	while (flushes) {
2414 		struct bio *next = flushes->bi_next;
2415 
2416 		flushes->bi_next = NULL;
2417 		do_endio(ic, flushes);
2418 		flushes = next;
2419 	}
2420 }
2421 
2422 static void complete_copy_from_journal(unsigned long error, void *context)
2423 {
2424 	struct journal_io *io = context;
2425 	struct journal_completion *comp = io->comp;
2426 	struct dm_integrity_c *ic = comp->ic;
2427 
2428 	remove_range(ic, &io->range);
2429 	mempool_free(io, &ic->journal_io_mempool);
2430 	if (unlikely(error != 0))
2431 		dm_integrity_io_error(ic, "copying from journal", -EIO);
2432 	complete_journal_op(comp);
2433 }
2434 
2435 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2436 			       struct journal_entry *je)
2437 {
2438 	unsigned int s = 0;
2439 
2440 	do {
2441 		js->commit_id = je->last_bytes[s];
2442 		js++;
2443 	} while (++s < ic->sectors_per_block);
2444 }
2445 
2446 static void do_journal_write(struct dm_integrity_c *ic, unsigned int write_start,
2447 			     unsigned int write_sections, bool from_replay)
2448 {
2449 	unsigned int i, j, n;
2450 	struct journal_completion comp;
2451 	struct blk_plug plug;
2452 
2453 	blk_start_plug(&plug);
2454 
2455 	comp.ic = ic;
2456 	comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2457 	init_completion(&comp.comp);
2458 
2459 	i = write_start;
2460 	for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2461 #ifndef INTERNAL_VERIFY
2462 		if (unlikely(from_replay))
2463 #endif
2464 			rw_section_mac(ic, i, false);
2465 		for (j = 0; j < ic->journal_section_entries; j++) {
2466 			struct journal_entry *je = access_journal_entry(ic, i, j);
2467 			sector_t sec, area, offset;
2468 			unsigned int k, l, next_loop;
2469 			sector_t metadata_block;
2470 			unsigned int metadata_offset;
2471 			struct journal_io *io;
2472 
2473 			if (journal_entry_is_unused(je))
2474 				continue;
2475 			BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2476 			sec = journal_entry_get_sector(je);
2477 			if (unlikely(from_replay)) {
2478 				if (unlikely(sec & (unsigned int)(ic->sectors_per_block - 1))) {
2479 					dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2480 					sec &= ~(sector_t)(ic->sectors_per_block - 1);
2481 				}
2482 				if (unlikely(sec >= ic->provided_data_sectors)) {
2483 					journal_entry_set_unused(je);
2484 					continue;
2485 				}
2486 			}
2487 			get_area_and_offset(ic, sec, &area, &offset);
2488 			restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2489 			for (k = j + 1; k < ic->journal_section_entries; k++) {
2490 				struct journal_entry *je2 = access_journal_entry(ic, i, k);
2491 				sector_t sec2, area2, offset2;
2492 
2493 				if (journal_entry_is_unused(je2))
2494 					break;
2495 				BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2496 				sec2 = journal_entry_get_sector(je2);
2497 				if (unlikely(sec2 >= ic->provided_data_sectors))
2498 					break;
2499 				get_area_and_offset(ic, sec2, &area2, &offset2);
2500 				if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2501 					break;
2502 				restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2503 			}
2504 			next_loop = k - 1;
2505 
2506 			io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2507 			io->comp = &comp;
2508 			io->range.logical_sector = sec;
2509 			io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2510 
2511 			spin_lock_irq(&ic->endio_wait.lock);
2512 			add_new_range_and_wait(ic, &io->range);
2513 
2514 			if (likely(!from_replay)) {
2515 				struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2516 
2517 				/* don't write if there is newer committed sector */
2518 				while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2519 					struct journal_entry *je2 = access_journal_entry(ic, i, j);
2520 
2521 					journal_entry_set_unused(je2);
2522 					remove_journal_node(ic, &section_node[j]);
2523 					j++;
2524 					sec += ic->sectors_per_block;
2525 					offset += ic->sectors_per_block;
2526 				}
2527 				while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2528 					struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2529 
2530 					journal_entry_set_unused(je2);
2531 					remove_journal_node(ic, &section_node[k - 1]);
2532 					k--;
2533 				}
2534 				if (j == k) {
2535 					remove_range_unlocked(ic, &io->range);
2536 					spin_unlock_irq(&ic->endio_wait.lock);
2537 					mempool_free(io, &ic->journal_io_mempool);
2538 					goto skip_io;
2539 				}
2540 				for (l = j; l < k; l++)
2541 					remove_journal_node(ic, &section_node[l]);
2542 			}
2543 			spin_unlock_irq(&ic->endio_wait.lock);
2544 
2545 			metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2546 			for (l = j; l < k; l++) {
2547 				int r;
2548 				struct journal_entry *je2 = access_journal_entry(ic, i, l);
2549 
2550 				if (
2551 #ifndef INTERNAL_VERIFY
2552 				    unlikely(from_replay) &&
2553 #endif
2554 				    ic->internal_hash) {
2555 					char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2556 
2557 					integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2558 								  (char *)access_journal_data(ic, i, l), test_tag);
2559 					if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size))) {
2560 						dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2561 						dm_audit_log_target(DM_MSG_PREFIX, "integrity-replay-journal", ic->ti, 0);
2562 					}
2563 				}
2564 
2565 				journal_entry_set_unused(je2);
2566 				r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2567 							ic->tag_size, TAG_WRITE);
2568 				if (unlikely(r))
2569 					dm_integrity_io_error(ic, "reading tags", r);
2570 			}
2571 
2572 			atomic_inc(&comp.in_flight);
2573 			copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2574 					  (k - j) << ic->sb->log2_sectors_per_block,
2575 					  get_data_sector(ic, area, offset),
2576 					  complete_copy_from_journal, io);
2577 skip_io:
2578 			j = next_loop;
2579 		}
2580 	}
2581 
2582 	dm_bufio_write_dirty_buffers_async(ic->bufio);
2583 
2584 	blk_finish_plug(&plug);
2585 
2586 	complete_journal_op(&comp);
2587 	wait_for_completion_io(&comp.comp);
2588 
2589 	dm_integrity_flush_buffers(ic, true);
2590 }
2591 
2592 static void integrity_writer(struct work_struct *w)
2593 {
2594 	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2595 	unsigned int write_start, write_sections;
2596 	unsigned int prev_free_sectors;
2597 
2598 	spin_lock_irq(&ic->endio_wait.lock);
2599 	write_start = ic->committed_section;
2600 	write_sections = ic->n_committed_sections;
2601 	spin_unlock_irq(&ic->endio_wait.lock);
2602 
2603 	if (!write_sections)
2604 		return;
2605 
2606 	do_journal_write(ic, write_start, write_sections, false);
2607 
2608 	spin_lock_irq(&ic->endio_wait.lock);
2609 
2610 	ic->committed_section += write_sections;
2611 	wraparound_section(ic, &ic->committed_section);
2612 	ic->n_committed_sections -= write_sections;
2613 
2614 	prev_free_sectors = ic->free_sectors;
2615 	ic->free_sectors += write_sections * ic->journal_section_entries;
2616 	if (unlikely(!prev_free_sectors))
2617 		wake_up_locked(&ic->endio_wait);
2618 
2619 	spin_unlock_irq(&ic->endio_wait.lock);
2620 }
2621 
2622 static void recalc_write_super(struct dm_integrity_c *ic)
2623 {
2624 	int r;
2625 
2626 	dm_integrity_flush_buffers(ic, false);
2627 	if (dm_integrity_failed(ic))
2628 		return;
2629 
2630 	r = sync_rw_sb(ic, REQ_OP_WRITE);
2631 	if (unlikely(r))
2632 		dm_integrity_io_error(ic, "writing superblock", r);
2633 }
2634 
2635 static void integrity_recalc(struct work_struct *w)
2636 {
2637 	struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2638 	size_t recalc_tags_size;
2639 	u8 *recalc_buffer = NULL;
2640 	u8 *recalc_tags = NULL;
2641 	struct dm_integrity_range range;
2642 	struct dm_io_request io_req;
2643 	struct dm_io_region io_loc;
2644 	sector_t area, offset;
2645 	sector_t metadata_block;
2646 	unsigned int metadata_offset;
2647 	sector_t logical_sector, n_sectors;
2648 	__u8 *t;
2649 	unsigned int i;
2650 	int r;
2651 	unsigned int super_counter = 0;
2652 	unsigned recalc_sectors = RECALC_SECTORS;
2653 
2654 retry:
2655 	recalc_buffer = __vmalloc(recalc_sectors << SECTOR_SHIFT, GFP_NOIO);
2656 	if (!recalc_buffer) {
2657 oom:
2658 		recalc_sectors >>= 1;
2659 		if (recalc_sectors >= 1U << ic->sb->log2_sectors_per_block)
2660 			goto retry;
2661 		DMCRIT("out of memory for recalculate buffer - recalculation disabled");
2662 		goto free_ret;
2663 	}
2664 	recalc_tags_size = (recalc_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
2665 	if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
2666 		recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
2667 	recalc_tags = kvmalloc(recalc_tags_size, GFP_NOIO);
2668 	if (!recalc_tags) {
2669 		vfree(recalc_buffer);
2670 		recalc_buffer = NULL;
2671 		goto oom;
2672 	}
2673 
2674 	DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2675 
2676 	spin_lock_irq(&ic->endio_wait.lock);
2677 
2678 next_chunk:
2679 
2680 	if (unlikely(dm_post_suspending(ic->ti)))
2681 		goto unlock_ret;
2682 
2683 	range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2684 	if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2685 		if (ic->mode == 'B') {
2686 			block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2687 			DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2688 			queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2689 		}
2690 		goto unlock_ret;
2691 	}
2692 
2693 	get_area_and_offset(ic, range.logical_sector, &area, &offset);
2694 	range.n_sectors = min((sector_t)recalc_sectors, ic->provided_data_sectors - range.logical_sector);
2695 	if (!ic->meta_dev)
2696 		range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned int)offset);
2697 
2698 	add_new_range_and_wait(ic, &range);
2699 	spin_unlock_irq(&ic->endio_wait.lock);
2700 	logical_sector = range.logical_sector;
2701 	n_sectors = range.n_sectors;
2702 
2703 	if (ic->mode == 'B') {
2704 		if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2705 			goto advance_and_next;
2706 
2707 		while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2708 				       ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2709 			logical_sector += ic->sectors_per_block;
2710 			n_sectors -= ic->sectors_per_block;
2711 			cond_resched();
2712 		}
2713 		while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2714 				       ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2715 			n_sectors -= ic->sectors_per_block;
2716 			cond_resched();
2717 		}
2718 		get_area_and_offset(ic, logical_sector, &area, &offset);
2719 	}
2720 
2721 	DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2722 
2723 	if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2724 		recalc_write_super(ic);
2725 		if (ic->mode == 'B')
2726 			queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2727 
2728 		super_counter = 0;
2729 	}
2730 
2731 	if (unlikely(dm_integrity_failed(ic)))
2732 		goto err;
2733 
2734 	io_req.bi_opf = REQ_OP_READ;
2735 	io_req.mem.type = DM_IO_VMA;
2736 	io_req.mem.ptr.addr = recalc_buffer;
2737 	io_req.notify.fn = NULL;
2738 	io_req.client = ic->io;
2739 	io_loc.bdev = ic->dev->bdev;
2740 	io_loc.sector = get_data_sector(ic, area, offset);
2741 	io_loc.count = n_sectors;
2742 
2743 	r = dm_io(&io_req, 1, &io_loc, NULL);
2744 	if (unlikely(r)) {
2745 		dm_integrity_io_error(ic, "reading data", r);
2746 		goto err;
2747 	}
2748 
2749 	t = recalc_tags;
2750 	for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2751 		integrity_sector_checksum(ic, logical_sector + i, recalc_buffer + (i << SECTOR_SHIFT), t);
2752 		t += ic->tag_size;
2753 	}
2754 
2755 	metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2756 
2757 	r = dm_integrity_rw_tag(ic, recalc_tags, &metadata_block, &metadata_offset, t - recalc_tags, TAG_WRITE);
2758 	if (unlikely(r)) {
2759 		dm_integrity_io_error(ic, "writing tags", r);
2760 		goto err;
2761 	}
2762 
2763 	if (ic->mode == 'B') {
2764 		sector_t start, end;
2765 
2766 		start = (range.logical_sector >>
2767 			 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2768 			(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2769 		end = ((range.logical_sector + range.n_sectors) >>
2770 		       (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2771 			(ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2772 		block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2773 	}
2774 
2775 advance_and_next:
2776 	cond_resched();
2777 
2778 	spin_lock_irq(&ic->endio_wait.lock);
2779 	remove_range_unlocked(ic, &range);
2780 	ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2781 	goto next_chunk;
2782 
2783 err:
2784 	remove_range(ic, &range);
2785 	goto free_ret;
2786 
2787 unlock_ret:
2788 	spin_unlock_irq(&ic->endio_wait.lock);
2789 
2790 	recalc_write_super(ic);
2791 
2792 free_ret:
2793 	vfree(recalc_buffer);
2794 	kvfree(recalc_tags);
2795 }
2796 
2797 static void bitmap_block_work(struct work_struct *w)
2798 {
2799 	struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2800 	struct dm_integrity_c *ic = bbs->ic;
2801 	struct bio *bio;
2802 	struct bio_list bio_queue;
2803 	struct bio_list waiting;
2804 
2805 	bio_list_init(&waiting);
2806 
2807 	spin_lock(&bbs->bio_queue_lock);
2808 	bio_queue = bbs->bio_queue;
2809 	bio_list_init(&bbs->bio_queue);
2810 	spin_unlock(&bbs->bio_queue_lock);
2811 
2812 	while ((bio = bio_list_pop(&bio_queue))) {
2813 		struct dm_integrity_io *dio;
2814 
2815 		dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2816 
2817 		if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2818 				    dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2819 			remove_range(ic, &dio->range);
2820 			INIT_WORK(&dio->work, integrity_bio_wait);
2821 			queue_work(ic->offload_wq, &dio->work);
2822 		} else {
2823 			block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2824 					dio->range.n_sectors, BITMAP_OP_SET);
2825 			bio_list_add(&waiting, bio);
2826 		}
2827 	}
2828 
2829 	if (bio_list_empty(&waiting))
2830 		return;
2831 
2832 	rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC,
2833 			   bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2834 			   BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2835 
2836 	while ((bio = bio_list_pop(&waiting))) {
2837 		struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2838 
2839 		block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2840 				dio->range.n_sectors, BITMAP_OP_SET);
2841 
2842 		remove_range(ic, &dio->range);
2843 		INIT_WORK(&dio->work, integrity_bio_wait);
2844 		queue_work(ic->offload_wq, &dio->work);
2845 	}
2846 
2847 	queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2848 }
2849 
2850 static void bitmap_flush_work(struct work_struct *work)
2851 {
2852 	struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2853 	struct dm_integrity_range range;
2854 	unsigned long limit;
2855 	struct bio *bio;
2856 
2857 	dm_integrity_flush_buffers(ic, false);
2858 
2859 	range.logical_sector = 0;
2860 	range.n_sectors = ic->provided_data_sectors;
2861 
2862 	spin_lock_irq(&ic->endio_wait.lock);
2863 	add_new_range_and_wait(ic, &range);
2864 	spin_unlock_irq(&ic->endio_wait.lock);
2865 
2866 	dm_integrity_flush_buffers(ic, true);
2867 
2868 	limit = ic->provided_data_sectors;
2869 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2870 		limit = le64_to_cpu(ic->sb->recalc_sector)
2871 			>> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2872 			<< (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2873 	}
2874 	/*DEBUG_print("zeroing journal\n");*/
2875 	block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2876 	block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2877 
2878 	rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
2879 			   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2880 
2881 	spin_lock_irq(&ic->endio_wait.lock);
2882 	remove_range_unlocked(ic, &range);
2883 	while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2884 		bio_endio(bio);
2885 		spin_unlock_irq(&ic->endio_wait.lock);
2886 		spin_lock_irq(&ic->endio_wait.lock);
2887 	}
2888 	spin_unlock_irq(&ic->endio_wait.lock);
2889 }
2890 
2891 
2892 static void init_journal(struct dm_integrity_c *ic, unsigned int start_section,
2893 			 unsigned int n_sections, unsigned char commit_seq)
2894 {
2895 	unsigned int i, j, n;
2896 
2897 	if (!n_sections)
2898 		return;
2899 
2900 	for (n = 0; n < n_sections; n++) {
2901 		i = start_section + n;
2902 		wraparound_section(ic, &i);
2903 		for (j = 0; j < ic->journal_section_sectors; j++) {
2904 			struct journal_sector *js = access_journal(ic, i, j);
2905 
2906 			BUILD_BUG_ON(sizeof(js->sectors) != JOURNAL_SECTOR_DATA);
2907 			memset(&js->sectors, 0, sizeof(js->sectors));
2908 			js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2909 		}
2910 		for (j = 0; j < ic->journal_section_entries; j++) {
2911 			struct journal_entry *je = access_journal_entry(ic, i, j);
2912 
2913 			journal_entry_set_unused(je);
2914 		}
2915 	}
2916 
2917 	write_journal(ic, start_section, n_sections);
2918 }
2919 
2920 static int find_commit_seq(struct dm_integrity_c *ic, unsigned int i, unsigned int j, commit_id_t id)
2921 {
2922 	unsigned char k;
2923 
2924 	for (k = 0; k < N_COMMIT_IDS; k++) {
2925 		if (dm_integrity_commit_id(ic, i, j, k) == id)
2926 			return k;
2927 	}
2928 	dm_integrity_io_error(ic, "journal commit id", -EIO);
2929 	return -EIO;
2930 }
2931 
2932 static void replay_journal(struct dm_integrity_c *ic)
2933 {
2934 	unsigned int i, j;
2935 	bool used_commit_ids[N_COMMIT_IDS];
2936 	unsigned int max_commit_id_sections[N_COMMIT_IDS];
2937 	unsigned int write_start, write_sections;
2938 	unsigned int continue_section;
2939 	bool journal_empty;
2940 	unsigned char unused, last_used, want_commit_seq;
2941 
2942 	if (ic->mode == 'R')
2943 		return;
2944 
2945 	if (ic->journal_uptodate)
2946 		return;
2947 
2948 	last_used = 0;
2949 	write_start = 0;
2950 
2951 	if (!ic->just_formatted) {
2952 		DEBUG_print("reading journal\n");
2953 		rw_journal(ic, REQ_OP_READ, 0, ic->journal_sections, NULL);
2954 		if (ic->journal_io)
2955 			DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2956 		if (ic->journal_io) {
2957 			struct journal_completion crypt_comp;
2958 
2959 			crypt_comp.ic = ic;
2960 			init_completion(&crypt_comp.comp);
2961 			crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2962 			encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2963 			wait_for_completion(&crypt_comp.comp);
2964 		}
2965 		DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2966 	}
2967 
2968 	if (dm_integrity_failed(ic))
2969 		goto clear_journal;
2970 
2971 	journal_empty = true;
2972 	memset(used_commit_ids, 0, sizeof(used_commit_ids));
2973 	memset(max_commit_id_sections, 0, sizeof(max_commit_id_sections));
2974 	for (i = 0; i < ic->journal_sections; i++) {
2975 		for (j = 0; j < ic->journal_section_sectors; j++) {
2976 			int k;
2977 			struct journal_sector *js = access_journal(ic, i, j);
2978 
2979 			k = find_commit_seq(ic, i, j, js->commit_id);
2980 			if (k < 0)
2981 				goto clear_journal;
2982 			used_commit_ids[k] = true;
2983 			max_commit_id_sections[k] = i;
2984 		}
2985 		if (journal_empty) {
2986 			for (j = 0; j < ic->journal_section_entries; j++) {
2987 				struct journal_entry *je = access_journal_entry(ic, i, j);
2988 
2989 				if (!journal_entry_is_unused(je)) {
2990 					journal_empty = false;
2991 					break;
2992 				}
2993 			}
2994 		}
2995 	}
2996 
2997 	if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2998 		unused = N_COMMIT_IDS - 1;
2999 		while (unused && !used_commit_ids[unused - 1])
3000 			unused--;
3001 	} else {
3002 		for (unused = 0; unused < N_COMMIT_IDS; unused++)
3003 			if (!used_commit_ids[unused])
3004 				break;
3005 		if (unused == N_COMMIT_IDS) {
3006 			dm_integrity_io_error(ic, "journal commit ids", -EIO);
3007 			goto clear_journal;
3008 		}
3009 	}
3010 	DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
3011 		    unused, used_commit_ids[0], used_commit_ids[1],
3012 		    used_commit_ids[2], used_commit_ids[3]);
3013 
3014 	last_used = prev_commit_seq(unused);
3015 	want_commit_seq = prev_commit_seq(last_used);
3016 
3017 	if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
3018 		journal_empty = true;
3019 
3020 	write_start = max_commit_id_sections[last_used] + 1;
3021 	if (unlikely(write_start >= ic->journal_sections))
3022 		want_commit_seq = next_commit_seq(want_commit_seq);
3023 	wraparound_section(ic, &write_start);
3024 
3025 	i = write_start;
3026 	for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
3027 		for (j = 0; j < ic->journal_section_sectors; j++) {
3028 			struct journal_sector *js = access_journal(ic, i, j);
3029 
3030 			if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
3031 				/*
3032 				 * This could be caused by crash during writing.
3033 				 * We won't replay the inconsistent part of the
3034 				 * journal.
3035 				 */
3036 				DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
3037 					    i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
3038 				goto brk;
3039 			}
3040 		}
3041 		i++;
3042 		if (unlikely(i >= ic->journal_sections))
3043 			want_commit_seq = next_commit_seq(want_commit_seq);
3044 		wraparound_section(ic, &i);
3045 	}
3046 brk:
3047 
3048 	if (!journal_empty) {
3049 		DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3050 			    write_sections, write_start, want_commit_seq);
3051 		do_journal_write(ic, write_start, write_sections, true);
3052 	}
3053 
3054 	if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3055 		continue_section = write_start;
3056 		ic->commit_seq = want_commit_seq;
3057 		DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3058 	} else {
3059 		unsigned int s;
3060 		unsigned char erase_seq;
3061 
3062 clear_journal:
3063 		DEBUG_print("clearing journal\n");
3064 
3065 		erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3066 		s = write_start;
3067 		init_journal(ic, s, 1, erase_seq);
3068 		s++;
3069 		wraparound_section(ic, &s);
3070 		if (ic->journal_sections >= 2) {
3071 			init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3072 			s += ic->journal_sections - 2;
3073 			wraparound_section(ic, &s);
3074 			init_journal(ic, s, 1, erase_seq);
3075 		}
3076 
3077 		continue_section = 0;
3078 		ic->commit_seq = next_commit_seq(erase_seq);
3079 	}
3080 
3081 	ic->committed_section = continue_section;
3082 	ic->n_committed_sections = 0;
3083 
3084 	ic->uncommitted_section = continue_section;
3085 	ic->n_uncommitted_sections = 0;
3086 
3087 	ic->free_section = continue_section;
3088 	ic->free_section_entry = 0;
3089 	ic->free_sectors = ic->journal_entries;
3090 
3091 	ic->journal_tree_root = RB_ROOT;
3092 	for (i = 0; i < ic->journal_entries; i++)
3093 		init_journal_node(&ic->journal_tree[i]);
3094 }
3095 
3096 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3097 {
3098 	DEBUG_print("%s\n", __func__);
3099 
3100 	if (ic->mode == 'B') {
3101 		ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3102 		ic->synchronous_mode = 1;
3103 
3104 		cancel_delayed_work_sync(&ic->bitmap_flush_work);
3105 		queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3106 		flush_workqueue(ic->commit_wq);
3107 	}
3108 }
3109 
3110 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3111 {
3112 	struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3113 
3114 	DEBUG_print("%s\n", __func__);
3115 
3116 	dm_integrity_enter_synchronous_mode(ic);
3117 
3118 	return NOTIFY_DONE;
3119 }
3120 
3121 static void dm_integrity_postsuspend(struct dm_target *ti)
3122 {
3123 	struct dm_integrity_c *ic = ti->private;
3124 	int r;
3125 
3126 	WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3127 
3128 	del_timer_sync(&ic->autocommit_timer);
3129 
3130 	if (ic->recalc_wq)
3131 		drain_workqueue(ic->recalc_wq);
3132 
3133 	if (ic->mode == 'B')
3134 		cancel_delayed_work_sync(&ic->bitmap_flush_work);
3135 
3136 	queue_work(ic->commit_wq, &ic->commit_work);
3137 	drain_workqueue(ic->commit_wq);
3138 
3139 	if (ic->mode == 'J') {
3140 		queue_work(ic->writer_wq, &ic->writer_work);
3141 		drain_workqueue(ic->writer_wq);
3142 		dm_integrity_flush_buffers(ic, true);
3143 		if (ic->wrote_to_journal) {
3144 			init_journal(ic, ic->free_section,
3145 				     ic->journal_sections - ic->free_section, ic->commit_seq);
3146 			if (ic->free_section) {
3147 				init_journal(ic, 0, ic->free_section,
3148 					     next_commit_seq(ic->commit_seq));
3149 			}
3150 		}
3151 	}
3152 
3153 	if (ic->mode == 'B') {
3154 		dm_integrity_flush_buffers(ic, true);
3155 #if 1
3156 		/* set to 0 to test bitmap replay code */
3157 		init_journal(ic, 0, ic->journal_sections, 0);
3158 		ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3159 		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3160 		if (unlikely(r))
3161 			dm_integrity_io_error(ic, "writing superblock", r);
3162 #endif
3163 	}
3164 
3165 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3166 
3167 	ic->journal_uptodate = true;
3168 }
3169 
3170 static void dm_integrity_resume(struct dm_target *ti)
3171 {
3172 	struct dm_integrity_c *ic = ti->private;
3173 	__u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3174 	int r;
3175 
3176 	DEBUG_print("resume\n");
3177 
3178 	ic->wrote_to_journal = false;
3179 
3180 	if (ic->provided_data_sectors != old_provided_data_sectors) {
3181 		if (ic->provided_data_sectors > old_provided_data_sectors &&
3182 		    ic->mode == 'B' &&
3183 		    ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3184 			rw_journal_sectors(ic, REQ_OP_READ, 0,
3185 					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3186 			block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3187 					ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3188 			rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3189 					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3190 		}
3191 
3192 		ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3193 		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3194 		if (unlikely(r))
3195 			dm_integrity_io_error(ic, "writing superblock", r);
3196 	}
3197 
3198 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3199 		DEBUG_print("resume dirty_bitmap\n");
3200 		rw_journal_sectors(ic, REQ_OP_READ, 0,
3201 				   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3202 		if (ic->mode == 'B') {
3203 			if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3204 			    !ic->reset_recalculate_flag) {
3205 				block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3206 				block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3207 				if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3208 						     BITMAP_OP_TEST_ALL_CLEAR)) {
3209 					ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3210 					ic->sb->recalc_sector = cpu_to_le64(0);
3211 				}
3212 			} else {
3213 				DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3214 					    ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3215 				ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3216 				block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3217 				block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3218 				block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3219 				rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3220 						   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3221 				ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3222 				ic->sb->recalc_sector = cpu_to_le64(0);
3223 			}
3224 		} else {
3225 			if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3226 			      block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3227 			    ic->reset_recalculate_flag) {
3228 				ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3229 				ic->sb->recalc_sector = cpu_to_le64(0);
3230 			}
3231 			init_journal(ic, 0, ic->journal_sections, 0);
3232 			replay_journal(ic);
3233 			ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3234 		}
3235 		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3236 		if (unlikely(r))
3237 			dm_integrity_io_error(ic, "writing superblock", r);
3238 	} else {
3239 		replay_journal(ic);
3240 		if (ic->reset_recalculate_flag) {
3241 			ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3242 			ic->sb->recalc_sector = cpu_to_le64(0);
3243 		}
3244 		if (ic->mode == 'B') {
3245 			ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3246 			ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3247 			r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3248 			if (unlikely(r))
3249 				dm_integrity_io_error(ic, "writing superblock", r);
3250 
3251 			block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3252 			block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3253 			block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3254 			if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3255 			    le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3256 				block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3257 						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3258 				block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3259 						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3260 				block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3261 						ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3262 			}
3263 			rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3264 					   ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3265 		}
3266 	}
3267 
3268 	DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3269 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3270 		__u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3271 
3272 		DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3273 		if (recalc_pos < ic->provided_data_sectors) {
3274 			queue_work(ic->recalc_wq, &ic->recalc_work);
3275 		} else if (recalc_pos > ic->provided_data_sectors) {
3276 			ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3277 			recalc_write_super(ic);
3278 		}
3279 	}
3280 
3281 	ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3282 	ic->reboot_notifier.next = NULL;
3283 	ic->reboot_notifier.priority = INT_MAX - 1;	/* be notified after md and before hardware drivers */
3284 	WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3285 
3286 #if 0
3287 	/* set to 1 to stress test synchronous mode */
3288 	dm_integrity_enter_synchronous_mode(ic);
3289 #endif
3290 }
3291 
3292 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3293 				unsigned int status_flags, char *result, unsigned int maxlen)
3294 {
3295 	struct dm_integrity_c *ic = ti->private;
3296 	unsigned int arg_count;
3297 	size_t sz = 0;
3298 
3299 	switch (type) {
3300 	case STATUSTYPE_INFO:
3301 		DMEMIT("%llu %llu",
3302 			(unsigned long long)atomic64_read(&ic->number_of_mismatches),
3303 			ic->provided_data_sectors);
3304 		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3305 			DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3306 		else
3307 			DMEMIT(" -");
3308 		break;
3309 
3310 	case STATUSTYPE_TABLE: {
3311 		__u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3312 
3313 		watermark_percentage += ic->journal_entries / 2;
3314 		do_div(watermark_percentage, ic->journal_entries);
3315 		arg_count = 3;
3316 		arg_count += !!ic->meta_dev;
3317 		arg_count += ic->sectors_per_block != 1;
3318 		arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3319 		arg_count += ic->reset_recalculate_flag;
3320 		arg_count += ic->discard;
3321 		arg_count += ic->mode == 'J';
3322 		arg_count += ic->mode == 'J';
3323 		arg_count += ic->mode == 'B';
3324 		arg_count += ic->mode == 'B';
3325 		arg_count += !!ic->internal_hash_alg.alg_string;
3326 		arg_count += !!ic->journal_crypt_alg.alg_string;
3327 		arg_count += !!ic->journal_mac_alg.alg_string;
3328 		arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3329 		arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3330 		arg_count += ic->legacy_recalculate;
3331 		DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3332 		       ic->tag_size, ic->mode, arg_count);
3333 		if (ic->meta_dev)
3334 			DMEMIT(" meta_device:%s", ic->meta_dev->name);
3335 		if (ic->sectors_per_block != 1)
3336 			DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3337 		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3338 			DMEMIT(" recalculate");
3339 		if (ic->reset_recalculate_flag)
3340 			DMEMIT(" reset_recalculate");
3341 		if (ic->discard)
3342 			DMEMIT(" allow_discards");
3343 		DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3344 		DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3345 		DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3346 		if (ic->mode == 'J') {
3347 			DMEMIT(" journal_watermark:%u", (unsigned int)watermark_percentage);
3348 			DMEMIT(" commit_time:%u", ic->autocommit_msec);
3349 		}
3350 		if (ic->mode == 'B') {
3351 			DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3352 			DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3353 		}
3354 		if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3355 			DMEMIT(" fix_padding");
3356 		if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3357 			DMEMIT(" fix_hmac");
3358 		if (ic->legacy_recalculate)
3359 			DMEMIT(" legacy_recalculate");
3360 
3361 #define EMIT_ALG(a, n)							\
3362 		do {							\
3363 			if (ic->a.alg_string) {				\
3364 				DMEMIT(" %s:%s", n, ic->a.alg_string);	\
3365 				if (ic->a.key_string)			\
3366 					DMEMIT(":%s", ic->a.key_string);\
3367 			}						\
3368 		} while (0)
3369 		EMIT_ALG(internal_hash_alg, "internal_hash");
3370 		EMIT_ALG(journal_crypt_alg, "journal_crypt");
3371 		EMIT_ALG(journal_mac_alg, "journal_mac");
3372 		break;
3373 	}
3374 	case STATUSTYPE_IMA:
3375 		DMEMIT_TARGET_NAME_VERSION(ti->type);
3376 		DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
3377 			ic->dev->name, ic->start, ic->tag_size, ic->mode);
3378 
3379 		if (ic->meta_dev)
3380 			DMEMIT(",meta_device=%s", ic->meta_dev->name);
3381 		if (ic->sectors_per_block != 1)
3382 			DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
3383 
3384 		DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
3385 		       'y' : 'n');
3386 		DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
3387 		DMEMIT(",fix_padding=%c",
3388 		       ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
3389 		DMEMIT(",fix_hmac=%c",
3390 		       ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
3391 		DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
3392 
3393 		DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
3394 		DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
3395 		DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
3396 		DMEMIT(";");
3397 		break;
3398 	}
3399 }
3400 
3401 static int dm_integrity_iterate_devices(struct dm_target *ti,
3402 					iterate_devices_callout_fn fn, void *data)
3403 {
3404 	struct dm_integrity_c *ic = ti->private;
3405 
3406 	if (!ic->meta_dev)
3407 		return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3408 	else
3409 		return fn(ti, ic->dev, 0, ti->len, data);
3410 }
3411 
3412 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3413 {
3414 	struct dm_integrity_c *ic = ti->private;
3415 
3416 	if (ic->sectors_per_block > 1) {
3417 		limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3418 		limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3419 		blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3420 		limits->dma_alignment = limits->logical_block_size - 1;
3421 	}
3422 }
3423 
3424 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3425 {
3426 	unsigned int sector_space = JOURNAL_SECTOR_DATA;
3427 
3428 	ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3429 	ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3430 					 JOURNAL_ENTRY_ROUNDUP);
3431 
3432 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3433 		sector_space -= JOURNAL_MAC_PER_SECTOR;
3434 	ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3435 	ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3436 	ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3437 	ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3438 }
3439 
3440 static int calculate_device_limits(struct dm_integrity_c *ic)
3441 {
3442 	__u64 initial_sectors;
3443 
3444 	calculate_journal_section_size(ic);
3445 	initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3446 	if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3447 		return -EINVAL;
3448 	ic->initial_sectors = initial_sectors;
3449 
3450 	if (!ic->meta_dev) {
3451 		sector_t last_sector, last_area, last_offset;
3452 
3453 		/* we have to maintain excessive padding for compatibility with existing volumes */
3454 		__u64 metadata_run_padding =
3455 			ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3456 			(__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3457 			(__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3458 
3459 		ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3460 					    metadata_run_padding) >> SECTOR_SHIFT;
3461 		if (!(ic->metadata_run & (ic->metadata_run - 1)))
3462 			ic->log2_metadata_run = __ffs(ic->metadata_run);
3463 		else
3464 			ic->log2_metadata_run = -1;
3465 
3466 		get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3467 		last_sector = get_data_sector(ic, last_area, last_offset);
3468 		if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3469 			return -EINVAL;
3470 	} else {
3471 		__u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3472 
3473 		meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3474 				>> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3475 		meta_size <<= ic->log2_buffer_sectors;
3476 		if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3477 		    ic->initial_sectors + meta_size > ic->meta_device_sectors)
3478 			return -EINVAL;
3479 		ic->metadata_run = 1;
3480 		ic->log2_metadata_run = 0;
3481 	}
3482 
3483 	return 0;
3484 }
3485 
3486 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3487 {
3488 	if (!ic->meta_dev) {
3489 		int test_bit;
3490 
3491 		ic->provided_data_sectors = 0;
3492 		for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3493 			__u64 prev_data_sectors = ic->provided_data_sectors;
3494 
3495 			ic->provided_data_sectors |= (sector_t)1 << test_bit;
3496 			if (calculate_device_limits(ic))
3497 				ic->provided_data_sectors = prev_data_sectors;
3498 		}
3499 	} else {
3500 		ic->provided_data_sectors = ic->data_device_sectors;
3501 		ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3502 	}
3503 }
3504 
3505 static int initialize_superblock(struct dm_integrity_c *ic,
3506 				 unsigned int journal_sectors, unsigned int interleave_sectors)
3507 {
3508 	unsigned int journal_sections;
3509 	int test_bit;
3510 
3511 	memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3512 	memcpy(ic->sb->magic, SB_MAGIC, 8);
3513 	ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3514 	ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3515 	if (ic->journal_mac_alg.alg_string)
3516 		ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3517 
3518 	calculate_journal_section_size(ic);
3519 	journal_sections = journal_sectors / ic->journal_section_sectors;
3520 	if (!journal_sections)
3521 		journal_sections = 1;
3522 
3523 	if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3524 		ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3525 		get_random_bytes(ic->sb->salt, SALT_SIZE);
3526 	}
3527 
3528 	if (!ic->meta_dev) {
3529 		if (ic->fix_padding)
3530 			ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3531 		ic->sb->journal_sections = cpu_to_le32(journal_sections);
3532 		if (!interleave_sectors)
3533 			interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3534 		ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3535 		ic->sb->log2_interleave_sectors = max_t(__u8, MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3536 		ic->sb->log2_interleave_sectors = min_t(__u8, MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3537 
3538 		get_provided_data_sectors(ic);
3539 		if (!ic->provided_data_sectors)
3540 			return -EINVAL;
3541 	} else {
3542 		ic->sb->log2_interleave_sectors = 0;
3543 
3544 		get_provided_data_sectors(ic);
3545 		if (!ic->provided_data_sectors)
3546 			return -EINVAL;
3547 
3548 try_smaller_buffer:
3549 		ic->sb->journal_sections = cpu_to_le32(0);
3550 		for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3551 			__u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3552 			__u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3553 
3554 			if (test_journal_sections > journal_sections)
3555 				continue;
3556 			ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3557 			if (calculate_device_limits(ic))
3558 				ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3559 
3560 		}
3561 		if (!le32_to_cpu(ic->sb->journal_sections)) {
3562 			if (ic->log2_buffer_sectors > 3) {
3563 				ic->log2_buffer_sectors--;
3564 				goto try_smaller_buffer;
3565 			}
3566 			return -EINVAL;
3567 		}
3568 	}
3569 
3570 	ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3571 
3572 	sb_set_version(ic);
3573 
3574 	return 0;
3575 }
3576 
3577 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3578 {
3579 	struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3580 	struct blk_integrity bi;
3581 
3582 	memset(&bi, 0, sizeof(bi));
3583 	bi.profile = &dm_integrity_profile;
3584 	bi.tuple_size = ic->tag_size;
3585 	bi.tag_size = bi.tuple_size;
3586 	bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3587 
3588 	blk_integrity_register(disk, &bi);
3589 	blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3590 }
3591 
3592 static void dm_integrity_free_page_list(struct page_list *pl)
3593 {
3594 	unsigned int i;
3595 
3596 	if (!pl)
3597 		return;
3598 	for (i = 0; pl[i].page; i++)
3599 		__free_page(pl[i].page);
3600 	kvfree(pl);
3601 }
3602 
3603 static struct page_list *dm_integrity_alloc_page_list(unsigned int n_pages)
3604 {
3605 	struct page_list *pl;
3606 	unsigned int i;
3607 
3608 	pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3609 	if (!pl)
3610 		return NULL;
3611 
3612 	for (i = 0; i < n_pages; i++) {
3613 		pl[i].page = alloc_page(GFP_KERNEL);
3614 		if (!pl[i].page) {
3615 			dm_integrity_free_page_list(pl);
3616 			return NULL;
3617 		}
3618 		if (i)
3619 			pl[i - 1].next = &pl[i];
3620 	}
3621 	pl[i].page = NULL;
3622 	pl[i].next = NULL;
3623 
3624 	return pl;
3625 }
3626 
3627 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3628 {
3629 	unsigned int i;
3630 
3631 	for (i = 0; i < ic->journal_sections; i++)
3632 		kvfree(sl[i]);
3633 	kvfree(sl);
3634 }
3635 
3636 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3637 								   struct page_list *pl)
3638 {
3639 	struct scatterlist **sl;
3640 	unsigned int i;
3641 
3642 	sl = kvmalloc_array(ic->journal_sections,
3643 			    sizeof(struct scatterlist *),
3644 			    GFP_KERNEL | __GFP_ZERO);
3645 	if (!sl)
3646 		return NULL;
3647 
3648 	for (i = 0; i < ic->journal_sections; i++) {
3649 		struct scatterlist *s;
3650 		unsigned int start_index, start_offset;
3651 		unsigned int end_index, end_offset;
3652 		unsigned int n_pages;
3653 		unsigned int idx;
3654 
3655 		page_list_location(ic, i, 0, &start_index, &start_offset);
3656 		page_list_location(ic, i, ic->journal_section_sectors - 1,
3657 				   &end_index, &end_offset);
3658 
3659 		n_pages = (end_index - start_index + 1);
3660 
3661 		s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3662 				   GFP_KERNEL);
3663 		if (!s) {
3664 			dm_integrity_free_journal_scatterlist(ic, sl);
3665 			return NULL;
3666 		}
3667 
3668 		sg_init_table(s, n_pages);
3669 		for (idx = start_index; idx <= end_index; idx++) {
3670 			char *va = lowmem_page_address(pl[idx].page);
3671 			unsigned int start = 0, end = PAGE_SIZE;
3672 
3673 			if (idx == start_index)
3674 				start = start_offset;
3675 			if (idx == end_index)
3676 				end = end_offset + (1 << SECTOR_SHIFT);
3677 			sg_set_buf(&s[idx - start_index], va + start, end - start);
3678 		}
3679 
3680 		sl[i] = s;
3681 	}
3682 
3683 	return sl;
3684 }
3685 
3686 static void free_alg(struct alg_spec *a)
3687 {
3688 	kfree_sensitive(a->alg_string);
3689 	kfree_sensitive(a->key);
3690 	memset(a, 0, sizeof(*a));
3691 }
3692 
3693 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3694 {
3695 	char *k;
3696 
3697 	free_alg(a);
3698 
3699 	a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3700 	if (!a->alg_string)
3701 		goto nomem;
3702 
3703 	k = strchr(a->alg_string, ':');
3704 	if (k) {
3705 		*k = 0;
3706 		a->key_string = k + 1;
3707 		if (strlen(a->key_string) & 1)
3708 			goto inval;
3709 
3710 		a->key_size = strlen(a->key_string) / 2;
3711 		a->key = kmalloc(a->key_size, GFP_KERNEL);
3712 		if (!a->key)
3713 			goto nomem;
3714 		if (hex2bin(a->key, a->key_string, a->key_size))
3715 			goto inval;
3716 	}
3717 
3718 	return 0;
3719 inval:
3720 	*error = error_inval;
3721 	return -EINVAL;
3722 nomem:
3723 	*error = "Out of memory for an argument";
3724 	return -ENOMEM;
3725 }
3726 
3727 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3728 		   char *error_alg, char *error_key)
3729 {
3730 	int r;
3731 
3732 	if (a->alg_string) {
3733 		*hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3734 		if (IS_ERR(*hash)) {
3735 			*error = error_alg;
3736 			r = PTR_ERR(*hash);
3737 			*hash = NULL;
3738 			return r;
3739 		}
3740 
3741 		if (a->key) {
3742 			r = crypto_shash_setkey(*hash, a->key, a->key_size);
3743 			if (r) {
3744 				*error = error_key;
3745 				return r;
3746 			}
3747 		} else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3748 			*error = error_key;
3749 			return -ENOKEY;
3750 		}
3751 	}
3752 
3753 	return 0;
3754 }
3755 
3756 static int create_journal(struct dm_integrity_c *ic, char **error)
3757 {
3758 	int r = 0;
3759 	unsigned int i;
3760 	__u64 journal_pages, journal_desc_size, journal_tree_size;
3761 	unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3762 	struct skcipher_request *req = NULL;
3763 
3764 	ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3765 	ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3766 	ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3767 	ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3768 
3769 	journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3770 				PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3771 	journal_desc_size = journal_pages * sizeof(struct page_list);
3772 	if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3773 		*error = "Journal doesn't fit into memory";
3774 		r = -ENOMEM;
3775 		goto bad;
3776 	}
3777 	ic->journal_pages = journal_pages;
3778 
3779 	ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3780 	if (!ic->journal) {
3781 		*error = "Could not allocate memory for journal";
3782 		r = -ENOMEM;
3783 		goto bad;
3784 	}
3785 	if (ic->journal_crypt_alg.alg_string) {
3786 		unsigned int ivsize, blocksize;
3787 		struct journal_completion comp;
3788 
3789 		comp.ic = ic;
3790 		ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3791 		if (IS_ERR(ic->journal_crypt)) {
3792 			*error = "Invalid journal cipher";
3793 			r = PTR_ERR(ic->journal_crypt);
3794 			ic->journal_crypt = NULL;
3795 			goto bad;
3796 		}
3797 		ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3798 		blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3799 
3800 		if (ic->journal_crypt_alg.key) {
3801 			r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3802 						   ic->journal_crypt_alg.key_size);
3803 			if (r) {
3804 				*error = "Error setting encryption key";
3805 				goto bad;
3806 			}
3807 		}
3808 		DEBUG_print("cipher %s, block size %u iv size %u\n",
3809 			    ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3810 
3811 		ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3812 		if (!ic->journal_io) {
3813 			*error = "Could not allocate memory for journal io";
3814 			r = -ENOMEM;
3815 			goto bad;
3816 		}
3817 
3818 		if (blocksize == 1) {
3819 			struct scatterlist *sg;
3820 
3821 			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3822 			if (!req) {
3823 				*error = "Could not allocate crypt request";
3824 				r = -ENOMEM;
3825 				goto bad;
3826 			}
3827 
3828 			crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3829 			if (!crypt_iv) {
3830 				*error = "Could not allocate iv";
3831 				r = -ENOMEM;
3832 				goto bad;
3833 			}
3834 
3835 			ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3836 			if (!ic->journal_xor) {
3837 				*error = "Could not allocate memory for journal xor";
3838 				r = -ENOMEM;
3839 				goto bad;
3840 			}
3841 
3842 			sg = kvmalloc_array(ic->journal_pages + 1,
3843 					    sizeof(struct scatterlist),
3844 					    GFP_KERNEL);
3845 			if (!sg) {
3846 				*error = "Unable to allocate sg list";
3847 				r = -ENOMEM;
3848 				goto bad;
3849 			}
3850 			sg_init_table(sg, ic->journal_pages + 1);
3851 			for (i = 0; i < ic->journal_pages; i++) {
3852 				char *va = lowmem_page_address(ic->journal_xor[i].page);
3853 
3854 				clear_page(va);
3855 				sg_set_buf(&sg[i], va, PAGE_SIZE);
3856 			}
3857 			sg_set_buf(&sg[i], &ic->commit_ids, sizeof(ic->commit_ids));
3858 
3859 			skcipher_request_set_crypt(req, sg, sg,
3860 						   PAGE_SIZE * ic->journal_pages + sizeof(ic->commit_ids), crypt_iv);
3861 			init_completion(&comp.comp);
3862 			comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3863 			if (do_crypt(true, req, &comp))
3864 				wait_for_completion(&comp.comp);
3865 			kvfree(sg);
3866 			r = dm_integrity_failed(ic);
3867 			if (r) {
3868 				*error = "Unable to encrypt journal";
3869 				goto bad;
3870 			}
3871 			DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3872 
3873 			crypto_free_skcipher(ic->journal_crypt);
3874 			ic->journal_crypt = NULL;
3875 		} else {
3876 			unsigned int crypt_len = roundup(ivsize, blocksize);
3877 
3878 			req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3879 			if (!req) {
3880 				*error = "Could not allocate crypt request";
3881 				r = -ENOMEM;
3882 				goto bad;
3883 			}
3884 
3885 			crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3886 			if (!crypt_iv) {
3887 				*error = "Could not allocate iv";
3888 				r = -ENOMEM;
3889 				goto bad;
3890 			}
3891 
3892 			crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3893 			if (!crypt_data) {
3894 				*error = "Unable to allocate crypt data";
3895 				r = -ENOMEM;
3896 				goto bad;
3897 			}
3898 
3899 			ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3900 			if (!ic->journal_scatterlist) {
3901 				*error = "Unable to allocate sg list";
3902 				r = -ENOMEM;
3903 				goto bad;
3904 			}
3905 			ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3906 			if (!ic->journal_io_scatterlist) {
3907 				*error = "Unable to allocate sg list";
3908 				r = -ENOMEM;
3909 				goto bad;
3910 			}
3911 			ic->sk_requests = kvmalloc_array(ic->journal_sections,
3912 							 sizeof(struct skcipher_request *),
3913 							 GFP_KERNEL | __GFP_ZERO);
3914 			if (!ic->sk_requests) {
3915 				*error = "Unable to allocate sk requests";
3916 				r = -ENOMEM;
3917 				goto bad;
3918 			}
3919 			for (i = 0; i < ic->journal_sections; i++) {
3920 				struct scatterlist sg;
3921 				struct skcipher_request *section_req;
3922 				__le32 section_le = cpu_to_le32(i);
3923 
3924 				memset(crypt_iv, 0x00, ivsize);
3925 				memset(crypt_data, 0x00, crypt_len);
3926 				memcpy(crypt_data, &section_le, min_t(size_t, crypt_len, sizeof(section_le)));
3927 
3928 				sg_init_one(&sg, crypt_data, crypt_len);
3929 				skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3930 				init_completion(&comp.comp);
3931 				comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3932 				if (do_crypt(true, req, &comp))
3933 					wait_for_completion(&comp.comp);
3934 
3935 				r = dm_integrity_failed(ic);
3936 				if (r) {
3937 					*error = "Unable to generate iv";
3938 					goto bad;
3939 				}
3940 
3941 				section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3942 				if (!section_req) {
3943 					*error = "Unable to allocate crypt request";
3944 					r = -ENOMEM;
3945 					goto bad;
3946 				}
3947 				section_req->iv = kmalloc_array(ivsize, 2,
3948 								GFP_KERNEL);
3949 				if (!section_req->iv) {
3950 					skcipher_request_free(section_req);
3951 					*error = "Unable to allocate iv";
3952 					r = -ENOMEM;
3953 					goto bad;
3954 				}
3955 				memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3956 				section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3957 				ic->sk_requests[i] = section_req;
3958 				DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3959 			}
3960 		}
3961 	}
3962 
3963 	for (i = 0; i < N_COMMIT_IDS; i++) {
3964 		unsigned int j;
3965 
3966 retest_commit_id:
3967 		for (j = 0; j < i; j++) {
3968 			if (ic->commit_ids[j] == ic->commit_ids[i]) {
3969 				ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3970 				goto retest_commit_id;
3971 			}
3972 		}
3973 		DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3974 	}
3975 
3976 	journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3977 	if (journal_tree_size > ULONG_MAX) {
3978 		*error = "Journal doesn't fit into memory";
3979 		r = -ENOMEM;
3980 		goto bad;
3981 	}
3982 	ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3983 	if (!ic->journal_tree) {
3984 		*error = "Could not allocate memory for journal tree";
3985 		r = -ENOMEM;
3986 	}
3987 bad:
3988 	kfree(crypt_data);
3989 	kfree(crypt_iv);
3990 	skcipher_request_free(req);
3991 
3992 	return r;
3993 }
3994 
3995 /*
3996  * Construct a integrity mapping
3997  *
3998  * Arguments:
3999  *	device
4000  *	offset from the start of the device
4001  *	tag size
4002  *	D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
4003  *	number of optional arguments
4004  *	optional arguments:
4005  *		journal_sectors
4006  *		interleave_sectors
4007  *		buffer_sectors
4008  *		journal_watermark
4009  *		commit_time
4010  *		meta_device
4011  *		block_size
4012  *		sectors_per_bit
4013  *		bitmap_flush_interval
4014  *		internal_hash
4015  *		journal_crypt
4016  *		journal_mac
4017  *		recalculate
4018  */
4019 static int dm_integrity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
4020 {
4021 	struct dm_integrity_c *ic;
4022 	char dummy;
4023 	int r;
4024 	unsigned int extra_args;
4025 	struct dm_arg_set as;
4026 	static const struct dm_arg _args[] = {
4027 		{0, 18, "Invalid number of feature args"},
4028 	};
4029 	unsigned int journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
4030 	bool should_write_sb;
4031 	__u64 threshold;
4032 	unsigned long long start;
4033 	__s8 log2_sectors_per_bitmap_bit = -1;
4034 	__s8 log2_blocks_per_bitmap_bit;
4035 	__u64 bits_in_journal;
4036 	__u64 n_bitmap_bits;
4037 
4038 #define DIRECT_ARGUMENTS	4
4039 
4040 	if (argc <= DIRECT_ARGUMENTS) {
4041 		ti->error = "Invalid argument count";
4042 		return -EINVAL;
4043 	}
4044 
4045 	ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
4046 	if (!ic) {
4047 		ti->error = "Cannot allocate integrity context";
4048 		return -ENOMEM;
4049 	}
4050 	ti->private = ic;
4051 	ti->per_io_data_size = sizeof(struct dm_integrity_io);
4052 	ic->ti = ti;
4053 
4054 	ic->in_progress = RB_ROOT;
4055 	INIT_LIST_HEAD(&ic->wait_list);
4056 	init_waitqueue_head(&ic->endio_wait);
4057 	bio_list_init(&ic->flush_bio_list);
4058 	init_waitqueue_head(&ic->copy_to_journal_wait);
4059 	init_completion(&ic->crypto_backoff);
4060 	atomic64_set(&ic->number_of_mismatches, 0);
4061 	ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
4062 
4063 	r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4064 	if (r) {
4065 		ti->error = "Device lookup failed";
4066 		goto bad;
4067 	}
4068 
4069 	if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4070 		ti->error = "Invalid starting offset";
4071 		r = -EINVAL;
4072 		goto bad;
4073 	}
4074 	ic->start = start;
4075 
4076 	if (strcmp(argv[2], "-")) {
4077 		if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4078 			ti->error = "Invalid tag size";
4079 			r = -EINVAL;
4080 			goto bad;
4081 		}
4082 	}
4083 
4084 	if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4085 	    !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
4086 		ic->mode = argv[3][0];
4087 	} else {
4088 		ti->error = "Invalid mode (expecting J, B, D, R)";
4089 		r = -EINVAL;
4090 		goto bad;
4091 	}
4092 
4093 	journal_sectors = 0;
4094 	interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4095 	buffer_sectors = DEFAULT_BUFFER_SECTORS;
4096 	journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4097 	sync_msec = DEFAULT_SYNC_MSEC;
4098 	ic->sectors_per_block = 1;
4099 
4100 	as.argc = argc - DIRECT_ARGUMENTS;
4101 	as.argv = argv + DIRECT_ARGUMENTS;
4102 	r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4103 	if (r)
4104 		goto bad;
4105 
4106 	while (extra_args--) {
4107 		const char *opt_string;
4108 		unsigned int val;
4109 		unsigned long long llval;
4110 
4111 		opt_string = dm_shift_arg(&as);
4112 		if (!opt_string) {
4113 			r = -EINVAL;
4114 			ti->error = "Not enough feature arguments";
4115 			goto bad;
4116 		}
4117 		if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4118 			journal_sectors = val ? val : 1;
4119 		else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4120 			interleave_sectors = val;
4121 		else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4122 			buffer_sectors = val;
4123 		else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4124 			journal_watermark = val;
4125 		else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4126 			sync_msec = val;
4127 		else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4128 			if (ic->meta_dev) {
4129 				dm_put_device(ti, ic->meta_dev);
4130 				ic->meta_dev = NULL;
4131 			}
4132 			r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4133 					  dm_table_get_mode(ti->table), &ic->meta_dev);
4134 			if (r) {
4135 				ti->error = "Device lookup failed";
4136 				goto bad;
4137 			}
4138 		} else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4139 			if (val < 1 << SECTOR_SHIFT ||
4140 			    val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4141 			    (val & (val - 1))) {
4142 				r = -EINVAL;
4143 				ti->error = "Invalid block_size argument";
4144 				goto bad;
4145 			}
4146 			ic->sectors_per_block = val >> SECTOR_SHIFT;
4147 		} else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4148 			log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4149 		} else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4150 			if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4151 				r = -EINVAL;
4152 				ti->error = "Invalid bitmap_flush_interval argument";
4153 				goto bad;
4154 			}
4155 			ic->bitmap_flush_interval = msecs_to_jiffies(val);
4156 		} else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4157 			r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4158 					    "Invalid internal_hash argument");
4159 			if (r)
4160 				goto bad;
4161 		} else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4162 			r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4163 					    "Invalid journal_crypt argument");
4164 			if (r)
4165 				goto bad;
4166 		} else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4167 			r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4168 					    "Invalid journal_mac argument");
4169 			if (r)
4170 				goto bad;
4171 		} else if (!strcmp(opt_string, "recalculate")) {
4172 			ic->recalculate_flag = true;
4173 		} else if (!strcmp(opt_string, "reset_recalculate")) {
4174 			ic->recalculate_flag = true;
4175 			ic->reset_recalculate_flag = true;
4176 		} else if (!strcmp(opt_string, "allow_discards")) {
4177 			ic->discard = true;
4178 		} else if (!strcmp(opt_string, "fix_padding")) {
4179 			ic->fix_padding = true;
4180 		} else if (!strcmp(opt_string, "fix_hmac")) {
4181 			ic->fix_hmac = true;
4182 		} else if (!strcmp(opt_string, "legacy_recalculate")) {
4183 			ic->legacy_recalculate = true;
4184 		} else {
4185 			r = -EINVAL;
4186 			ti->error = "Invalid argument";
4187 			goto bad;
4188 		}
4189 	}
4190 
4191 	ic->data_device_sectors = bdev_nr_sectors(ic->dev->bdev);
4192 	if (!ic->meta_dev)
4193 		ic->meta_device_sectors = ic->data_device_sectors;
4194 	else
4195 		ic->meta_device_sectors = bdev_nr_sectors(ic->meta_dev->bdev);
4196 
4197 	if (!journal_sectors) {
4198 		journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4199 				      ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4200 	}
4201 
4202 	if (!buffer_sectors)
4203 		buffer_sectors = 1;
4204 	ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4205 
4206 	r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4207 		    "Invalid internal hash", "Error setting internal hash key");
4208 	if (r)
4209 		goto bad;
4210 
4211 	r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4212 		    "Invalid journal mac", "Error setting journal mac key");
4213 	if (r)
4214 		goto bad;
4215 
4216 	if (!ic->tag_size) {
4217 		if (!ic->internal_hash) {
4218 			ti->error = "Unknown tag size";
4219 			r = -EINVAL;
4220 			goto bad;
4221 		}
4222 		ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4223 	}
4224 	if (ic->tag_size > MAX_TAG_SIZE) {
4225 		ti->error = "Too big tag size";
4226 		r = -EINVAL;
4227 		goto bad;
4228 	}
4229 	if (!(ic->tag_size & (ic->tag_size - 1)))
4230 		ic->log2_tag_size = __ffs(ic->tag_size);
4231 	else
4232 		ic->log2_tag_size = -1;
4233 
4234 	if (ic->mode == 'B' && !ic->internal_hash) {
4235 		r = -EINVAL;
4236 		ti->error = "Bitmap mode can be only used with internal hash";
4237 		goto bad;
4238 	}
4239 
4240 	if (ic->discard && !ic->internal_hash) {
4241 		r = -EINVAL;
4242 		ti->error = "Discard can be only used with internal hash";
4243 		goto bad;
4244 	}
4245 
4246 	ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4247 	ic->autocommit_msec = sync_msec;
4248 	timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4249 
4250 	ic->io = dm_io_client_create();
4251 	if (IS_ERR(ic->io)) {
4252 		r = PTR_ERR(ic->io);
4253 		ic->io = NULL;
4254 		ti->error = "Cannot allocate dm io";
4255 		goto bad;
4256 	}
4257 
4258 	r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4259 	if (r) {
4260 		ti->error = "Cannot allocate mempool";
4261 		goto bad;
4262 	}
4263 
4264 	ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4265 					  WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4266 	if (!ic->metadata_wq) {
4267 		ti->error = "Cannot allocate workqueue";
4268 		r = -ENOMEM;
4269 		goto bad;
4270 	}
4271 
4272 	/*
4273 	 * If this workqueue weren't ordered, it would cause bio reordering
4274 	 * and reduced performance.
4275 	 */
4276 	ic->wait_wq = alloc_ordered_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM);
4277 	if (!ic->wait_wq) {
4278 		ti->error = "Cannot allocate workqueue";
4279 		r = -ENOMEM;
4280 		goto bad;
4281 	}
4282 
4283 	ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4284 					  METADATA_WORKQUEUE_MAX_ACTIVE);
4285 	if (!ic->offload_wq) {
4286 		ti->error = "Cannot allocate workqueue";
4287 		r = -ENOMEM;
4288 		goto bad;
4289 	}
4290 
4291 	ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4292 	if (!ic->commit_wq) {
4293 		ti->error = "Cannot allocate workqueue";
4294 		r = -ENOMEM;
4295 		goto bad;
4296 	}
4297 	INIT_WORK(&ic->commit_work, integrity_commit);
4298 
4299 	if (ic->mode == 'J' || ic->mode == 'B') {
4300 		ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4301 		if (!ic->writer_wq) {
4302 			ti->error = "Cannot allocate workqueue";
4303 			r = -ENOMEM;
4304 			goto bad;
4305 		}
4306 		INIT_WORK(&ic->writer_work, integrity_writer);
4307 	}
4308 
4309 	ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4310 	if (!ic->sb) {
4311 		r = -ENOMEM;
4312 		ti->error = "Cannot allocate superblock area";
4313 		goto bad;
4314 	}
4315 
4316 	r = sync_rw_sb(ic, REQ_OP_READ);
4317 	if (r) {
4318 		ti->error = "Error reading superblock";
4319 		goto bad;
4320 	}
4321 	should_write_sb = false;
4322 	if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4323 		if (ic->mode != 'R') {
4324 			if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4325 				r = -EINVAL;
4326 				ti->error = "The device is not initialized";
4327 				goto bad;
4328 			}
4329 		}
4330 
4331 		r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4332 		if (r) {
4333 			ti->error = "Could not initialize superblock";
4334 			goto bad;
4335 		}
4336 		if (ic->mode != 'R')
4337 			should_write_sb = true;
4338 	}
4339 
4340 	if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4341 		r = -EINVAL;
4342 		ti->error = "Unknown version";
4343 		goto bad;
4344 	}
4345 	if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4346 		r = -EINVAL;
4347 		ti->error = "Tag size doesn't match the information in superblock";
4348 		goto bad;
4349 	}
4350 	if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4351 		r = -EINVAL;
4352 		ti->error = "Block size doesn't match the information in superblock";
4353 		goto bad;
4354 	}
4355 	if (!le32_to_cpu(ic->sb->journal_sections)) {
4356 		r = -EINVAL;
4357 		ti->error = "Corrupted superblock, journal_sections is 0";
4358 		goto bad;
4359 	}
4360 	/* make sure that ti->max_io_len doesn't overflow */
4361 	if (!ic->meta_dev) {
4362 		if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4363 		    ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4364 			r = -EINVAL;
4365 			ti->error = "Invalid interleave_sectors in the superblock";
4366 			goto bad;
4367 		}
4368 	} else {
4369 		if (ic->sb->log2_interleave_sectors) {
4370 			r = -EINVAL;
4371 			ti->error = "Invalid interleave_sectors in the superblock";
4372 			goto bad;
4373 		}
4374 	}
4375 	if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4376 		r = -EINVAL;
4377 		ti->error = "Journal mac mismatch";
4378 		goto bad;
4379 	}
4380 
4381 	get_provided_data_sectors(ic);
4382 	if (!ic->provided_data_sectors) {
4383 		r = -EINVAL;
4384 		ti->error = "The device is too small";
4385 		goto bad;
4386 	}
4387 
4388 try_smaller_buffer:
4389 	r = calculate_device_limits(ic);
4390 	if (r) {
4391 		if (ic->meta_dev) {
4392 			if (ic->log2_buffer_sectors > 3) {
4393 				ic->log2_buffer_sectors--;
4394 				goto try_smaller_buffer;
4395 			}
4396 		}
4397 		ti->error = "The device is too small";
4398 		goto bad;
4399 	}
4400 
4401 	if (log2_sectors_per_bitmap_bit < 0)
4402 		log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4403 	if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4404 		log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4405 
4406 	bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4407 	if (bits_in_journal > UINT_MAX)
4408 		bits_in_journal = UINT_MAX;
4409 	while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4410 		log2_sectors_per_bitmap_bit++;
4411 
4412 	log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4413 	ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4414 	if (should_write_sb)
4415 		ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4416 
4417 	n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4418 				+ (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4419 	ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4420 
4421 	if (!ic->meta_dev)
4422 		ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4423 
4424 	if (ti->len > ic->provided_data_sectors) {
4425 		r = -EINVAL;
4426 		ti->error = "Not enough provided sectors for requested mapping size";
4427 		goto bad;
4428 	}
4429 
4430 
4431 	threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4432 	threshold += 50;
4433 	do_div(threshold, 100);
4434 	ic->free_sectors_threshold = threshold;
4435 
4436 	DEBUG_print("initialized:\n");
4437 	DEBUG_print("	integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4438 	DEBUG_print("	journal_entry_size %u\n", ic->journal_entry_size);
4439 	DEBUG_print("	journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4440 	DEBUG_print("	journal_section_entries %u\n", ic->journal_section_entries);
4441 	DEBUG_print("	journal_section_sectors %u\n", ic->journal_section_sectors);
4442 	DEBUG_print("	journal_sections %u\n", (unsigned int)le32_to_cpu(ic->sb->journal_sections));
4443 	DEBUG_print("	journal_entries %u\n", ic->journal_entries);
4444 	DEBUG_print("	log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4445 	DEBUG_print("	data_device_sectors 0x%llx\n", bdev_nr_sectors(ic->dev->bdev));
4446 	DEBUG_print("	initial_sectors 0x%x\n", ic->initial_sectors);
4447 	DEBUG_print("	metadata_run 0x%x\n", ic->metadata_run);
4448 	DEBUG_print("	log2_metadata_run %d\n", ic->log2_metadata_run);
4449 	DEBUG_print("	provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4450 	DEBUG_print("	log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4451 	DEBUG_print("	bits_in_journal %llu\n", bits_in_journal);
4452 
4453 	if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4454 		ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4455 		ic->sb->recalc_sector = cpu_to_le64(0);
4456 	}
4457 
4458 	if (ic->internal_hash) {
4459 		ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4460 		if (!ic->recalc_wq) {
4461 			ti->error = "Cannot allocate workqueue";
4462 			r = -ENOMEM;
4463 			goto bad;
4464 		}
4465 		INIT_WORK(&ic->recalc_work, integrity_recalc);
4466 	} else {
4467 		if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4468 			ti->error = "Recalculate can only be specified with internal_hash";
4469 			r = -EINVAL;
4470 			goto bad;
4471 		}
4472 	}
4473 
4474 	if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4475 	    le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4476 	    dm_integrity_disable_recalculate(ic)) {
4477 		ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4478 		r = -EOPNOTSUPP;
4479 		goto bad;
4480 	}
4481 
4482 	ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4483 			1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL, 0);
4484 	if (IS_ERR(ic->bufio)) {
4485 		r = PTR_ERR(ic->bufio);
4486 		ti->error = "Cannot initialize dm-bufio";
4487 		ic->bufio = NULL;
4488 		goto bad;
4489 	}
4490 	dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4491 
4492 	if (ic->mode != 'R') {
4493 		r = create_journal(ic, &ti->error);
4494 		if (r)
4495 			goto bad;
4496 
4497 	}
4498 
4499 	if (ic->mode == 'B') {
4500 		unsigned int i;
4501 		unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4502 
4503 		ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4504 		if (!ic->recalc_bitmap) {
4505 			r = -ENOMEM;
4506 			goto bad;
4507 		}
4508 		ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4509 		if (!ic->may_write_bitmap) {
4510 			r = -ENOMEM;
4511 			goto bad;
4512 		}
4513 		ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4514 		if (!ic->bbs) {
4515 			r = -ENOMEM;
4516 			goto bad;
4517 		}
4518 		INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4519 		for (i = 0; i < ic->n_bitmap_blocks; i++) {
4520 			struct bitmap_block_status *bbs = &ic->bbs[i];
4521 			unsigned int sector, pl_index, pl_offset;
4522 
4523 			INIT_WORK(&bbs->work, bitmap_block_work);
4524 			bbs->ic = ic;
4525 			bbs->idx = i;
4526 			bio_list_init(&bbs->bio_queue);
4527 			spin_lock_init(&bbs->bio_queue_lock);
4528 
4529 			sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4530 			pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4531 			pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4532 
4533 			bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4534 		}
4535 	}
4536 
4537 	if (should_write_sb) {
4538 		init_journal(ic, 0, ic->journal_sections, 0);
4539 		r = dm_integrity_failed(ic);
4540 		if (unlikely(r)) {
4541 			ti->error = "Error initializing journal";
4542 			goto bad;
4543 		}
4544 		r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
4545 		if (r) {
4546 			ti->error = "Error initializing superblock";
4547 			goto bad;
4548 		}
4549 		ic->just_formatted = true;
4550 	}
4551 
4552 	if (!ic->meta_dev) {
4553 		r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4554 		if (r)
4555 			goto bad;
4556 	}
4557 	if (ic->mode == 'B') {
4558 		unsigned int max_io_len;
4559 
4560 		max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4561 		if (!max_io_len)
4562 			max_io_len = 1U << 31;
4563 		DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4564 		if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4565 			r = dm_set_target_max_io_len(ti, max_io_len);
4566 			if (r)
4567 				goto bad;
4568 		}
4569 	}
4570 
4571 	if (!ic->internal_hash)
4572 		dm_integrity_set(ti, ic);
4573 
4574 	ti->num_flush_bios = 1;
4575 	ti->flush_supported = true;
4576 	if (ic->discard)
4577 		ti->num_discard_bios = 1;
4578 
4579 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
4580 	return 0;
4581 
4582 bad:
4583 	dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
4584 	dm_integrity_dtr(ti);
4585 	return r;
4586 }
4587 
4588 static void dm_integrity_dtr(struct dm_target *ti)
4589 {
4590 	struct dm_integrity_c *ic = ti->private;
4591 
4592 	BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4593 	BUG_ON(!list_empty(&ic->wait_list));
4594 
4595 	if (ic->mode == 'B')
4596 		cancel_delayed_work_sync(&ic->bitmap_flush_work);
4597 	if (ic->metadata_wq)
4598 		destroy_workqueue(ic->metadata_wq);
4599 	if (ic->wait_wq)
4600 		destroy_workqueue(ic->wait_wq);
4601 	if (ic->offload_wq)
4602 		destroy_workqueue(ic->offload_wq);
4603 	if (ic->commit_wq)
4604 		destroy_workqueue(ic->commit_wq);
4605 	if (ic->writer_wq)
4606 		destroy_workqueue(ic->writer_wq);
4607 	if (ic->recalc_wq)
4608 		destroy_workqueue(ic->recalc_wq);
4609 	kvfree(ic->bbs);
4610 	if (ic->bufio)
4611 		dm_bufio_client_destroy(ic->bufio);
4612 	mempool_exit(&ic->journal_io_mempool);
4613 	if (ic->io)
4614 		dm_io_client_destroy(ic->io);
4615 	if (ic->dev)
4616 		dm_put_device(ti, ic->dev);
4617 	if (ic->meta_dev)
4618 		dm_put_device(ti, ic->meta_dev);
4619 	dm_integrity_free_page_list(ic->journal);
4620 	dm_integrity_free_page_list(ic->journal_io);
4621 	dm_integrity_free_page_list(ic->journal_xor);
4622 	dm_integrity_free_page_list(ic->recalc_bitmap);
4623 	dm_integrity_free_page_list(ic->may_write_bitmap);
4624 	if (ic->journal_scatterlist)
4625 		dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4626 	if (ic->journal_io_scatterlist)
4627 		dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4628 	if (ic->sk_requests) {
4629 		unsigned int i;
4630 
4631 		for (i = 0; i < ic->journal_sections; i++) {
4632 			struct skcipher_request *req;
4633 
4634 			req = ic->sk_requests[i];
4635 			if (req) {
4636 				kfree_sensitive(req->iv);
4637 				skcipher_request_free(req);
4638 			}
4639 		}
4640 		kvfree(ic->sk_requests);
4641 	}
4642 	kvfree(ic->journal_tree);
4643 	if (ic->sb)
4644 		free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4645 
4646 	if (ic->internal_hash)
4647 		crypto_free_shash(ic->internal_hash);
4648 	free_alg(&ic->internal_hash_alg);
4649 
4650 	if (ic->journal_crypt)
4651 		crypto_free_skcipher(ic->journal_crypt);
4652 	free_alg(&ic->journal_crypt_alg);
4653 
4654 	if (ic->journal_mac)
4655 		crypto_free_shash(ic->journal_mac);
4656 	free_alg(&ic->journal_mac_alg);
4657 
4658 	kfree(ic);
4659 	dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
4660 }
4661 
4662 static struct target_type integrity_target = {
4663 	.name			= "integrity",
4664 	.version		= {1, 10, 0},
4665 	.module			= THIS_MODULE,
4666 	.features		= DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4667 	.ctr			= dm_integrity_ctr,
4668 	.dtr			= dm_integrity_dtr,
4669 	.map			= dm_integrity_map,
4670 	.postsuspend		= dm_integrity_postsuspend,
4671 	.resume			= dm_integrity_resume,
4672 	.status			= dm_integrity_status,
4673 	.iterate_devices	= dm_integrity_iterate_devices,
4674 	.io_hints		= dm_integrity_io_hints,
4675 };
4676 
4677 static int __init dm_integrity_init(void)
4678 {
4679 	int r;
4680 
4681 	journal_io_cache = kmem_cache_create("integrity_journal_io",
4682 					     sizeof(struct journal_io), 0, 0, NULL);
4683 	if (!journal_io_cache) {
4684 		DMERR("can't allocate journal io cache");
4685 		return -ENOMEM;
4686 	}
4687 
4688 	r = dm_register_target(&integrity_target);
4689 	if (r < 0) {
4690 		kmem_cache_destroy(journal_io_cache);
4691 		return r;
4692 	}
4693 
4694 	return 0;
4695 }
4696 
4697 static void __exit dm_integrity_exit(void)
4698 {
4699 	dm_unregister_target(&integrity_target);
4700 	kmem_cache_destroy(journal_io_cache);
4701 }
4702 
4703 module_init(dm_integrity_init);
4704 module_exit(dm_integrity_exit);
4705 
4706 MODULE_AUTHOR("Milan Broz");
4707 MODULE_AUTHOR("Mikulas Patocka");
4708 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4709 MODULE_LICENSE("GPL");
4710