xref: /linux/fs/bcachefs/journal_io.c (revision 0e8655b4e852ef97655648b91ce780384a073ff4)
1 // SPDX-License-Identifier: GPL-2.0
2 #include "bcachefs.h"
3 #include "alloc_background.h"
4 #include "alloc_foreground.h"
5 #include "btree_io.h"
6 #include "btree_update_interior.h"
7 #include "btree_write_buffer.h"
8 #include "buckets.h"
9 #include "checksum.h"
10 #include "disk_groups.h"
11 #include "error.h"
12 #include "journal.h"
13 #include "journal_io.h"
14 #include "journal_reclaim.h"
15 #include "journal_seq_blacklist.h"
16 #include "replicas.h"
17 #include "sb-clean.h"
18 #include "trace.h"
19 
20 void bch2_journal_pos_from_member_info_set(struct bch_fs *c)
21 {
22 	lockdep_assert_held(&c->sb_lock);
23 
24 	for_each_member_device(c, ca) {
25 		struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
26 
27 		m->last_journal_bucket = cpu_to_le32(ca->journal.cur_idx);
28 		m->last_journal_bucket_offset = cpu_to_le32(ca->mi.bucket_size - ca->journal.sectors_free);
29 	}
30 }
31 
32 void bch2_journal_pos_from_member_info_resume(struct bch_fs *c)
33 {
34 	mutex_lock(&c->sb_lock);
35 	for_each_member_device(c, ca) {
36 		struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, ca->dev_idx);
37 
38 		unsigned idx = le32_to_cpu(m.last_journal_bucket);
39 		if (idx < ca->journal.nr)
40 			ca->journal.cur_idx = idx;
41 		unsigned offset = le32_to_cpu(m.last_journal_bucket_offset);
42 		if (offset <= ca->mi.bucket_size)
43 			ca->journal.sectors_free = ca->mi.bucket_size - offset;
44 	}
45 	mutex_unlock(&c->sb_lock);
46 }
47 
48 void bch2_journal_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
49 			       struct journal_replay *j)
50 {
51 	darray_for_each(j->ptrs, i) {
52 		if (i != j->ptrs.data)
53 			prt_printf(out, " ");
54 		prt_printf(out, "%u:%u:%u (sector %llu)",
55 			   i->dev, i->bucket, i->bucket_offset, i->sector);
56 	}
57 }
58 
59 static void bch2_journal_replay_to_text(struct printbuf *out, struct bch_fs *c,
60 					struct journal_replay *j)
61 {
62 	prt_printf(out, "seq %llu ", le64_to_cpu(j->j.seq));
63 
64 	bch2_journal_ptrs_to_text(out, c, j);
65 
66 	for_each_jset_entry_type(entry, &j->j, BCH_JSET_ENTRY_datetime) {
67 		struct jset_entry_datetime *datetime =
68 			container_of(entry, struct jset_entry_datetime, entry);
69 		bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
70 		break;
71 	}
72 }
73 
74 static struct nonce journal_nonce(const struct jset *jset)
75 {
76 	return (struct nonce) {{
77 		[0] = 0,
78 		[1] = ((__le32 *) &jset->seq)[0],
79 		[2] = ((__le32 *) &jset->seq)[1],
80 		[3] = BCH_NONCE_JOURNAL,
81 	}};
82 }
83 
84 static bool jset_csum_good(struct bch_fs *c, struct jset *j, struct bch_csum *csum)
85 {
86 	if (!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(j))) {
87 		*csum = (struct bch_csum) {};
88 		return false;
89 	}
90 
91 	*csum = csum_vstruct(c, JSET_CSUM_TYPE(j), journal_nonce(j), j);
92 	return !bch2_crc_cmp(j->csum, *csum);
93 }
94 
95 static inline u32 journal_entry_radix_idx(struct bch_fs *c, u64 seq)
96 {
97 	return (seq - c->journal_entries_base_seq) & (~0U >> 1);
98 }
99 
100 static void __journal_replay_free(struct bch_fs *c,
101 				  struct journal_replay *i)
102 {
103 	struct journal_replay **p =
104 		genradix_ptr(&c->journal_entries,
105 			     journal_entry_radix_idx(c, le64_to_cpu(i->j.seq)));
106 
107 	BUG_ON(*p != i);
108 	*p = NULL;
109 	kvfree(i);
110 }
111 
112 static void journal_replay_free(struct bch_fs *c, struct journal_replay *i, bool blacklisted)
113 {
114 	if (blacklisted)
115 		i->ignore_blacklisted = true;
116 	else
117 		i->ignore_not_dirty = true;
118 
119 	if (!c->opts.read_entire_journal)
120 		__journal_replay_free(c, i);
121 }
122 
123 struct journal_list {
124 	struct closure		cl;
125 	u64			last_seq;
126 	struct mutex		lock;
127 	int			ret;
128 };
129 
130 #define JOURNAL_ENTRY_ADD_OK		0
131 #define JOURNAL_ENTRY_ADD_OUT_OF_RANGE	5
132 
133 /*
134  * Given a journal entry we just read, add it to the list of journal entries to
135  * be replayed:
136  */
137 static int journal_entry_add(struct bch_fs *c, struct bch_dev *ca,
138 			     struct journal_ptr entry_ptr,
139 			     struct journal_list *jlist, struct jset *j)
140 {
141 	struct genradix_iter iter;
142 	struct journal_replay **_i, *i, *dup;
143 	size_t bytes = vstruct_bytes(j);
144 	u64 last_seq = !JSET_NO_FLUSH(j) ? le64_to_cpu(j->last_seq) : 0;
145 	struct printbuf buf = PRINTBUF;
146 	int ret = JOURNAL_ENTRY_ADD_OK;
147 
148 	if (!c->journal.oldest_seq_found_ondisk ||
149 	    le64_to_cpu(j->seq) < c->journal.oldest_seq_found_ondisk)
150 		c->journal.oldest_seq_found_ondisk = le64_to_cpu(j->seq);
151 
152 	/* Is this entry older than the range we need? */
153 	if (!c->opts.read_entire_journal &&
154 	    le64_to_cpu(j->seq) < jlist->last_seq)
155 		return JOURNAL_ENTRY_ADD_OUT_OF_RANGE;
156 
157 	/*
158 	 * genradixes are indexed by a ulong, not a u64, so we can't index them
159 	 * by sequence number directly: Assume instead that they will all fall
160 	 * within the range of +-2billion of the filrst one we find.
161 	 */
162 	if (!c->journal_entries_base_seq)
163 		c->journal_entries_base_seq = max_t(s64, 1, le64_to_cpu(j->seq) - S32_MAX);
164 
165 	/* Drop entries we don't need anymore */
166 	if (last_seq > jlist->last_seq && !c->opts.read_entire_journal) {
167 		genradix_for_each_from(&c->journal_entries, iter, _i,
168 				       journal_entry_radix_idx(c, jlist->last_seq)) {
169 			i = *_i;
170 
171 			if (journal_replay_ignore(i))
172 				continue;
173 
174 			if (le64_to_cpu(i->j.seq) >= last_seq)
175 				break;
176 
177 			journal_replay_free(c, i, false);
178 		}
179 	}
180 
181 	jlist->last_seq = max(jlist->last_seq, last_seq);
182 
183 	_i = genradix_ptr_alloc(&c->journal_entries,
184 				journal_entry_radix_idx(c, le64_to_cpu(j->seq)),
185 				GFP_KERNEL);
186 	if (!_i)
187 		return -BCH_ERR_ENOMEM_journal_entry_add;
188 
189 	/*
190 	 * Duplicate journal entries? If so we want the one that didn't have a
191 	 * checksum error:
192 	 */
193 	dup = *_i;
194 	if (dup) {
195 		bool identical = bytes == vstruct_bytes(&dup->j) &&
196 			!memcmp(j, &dup->j, bytes);
197 		bool not_identical = !identical &&
198 			entry_ptr.csum_good &&
199 			dup->csum_good;
200 
201 		bool same_device = false;
202 		darray_for_each(dup->ptrs, ptr)
203 			if (ptr->dev == ca->dev_idx)
204 				same_device = true;
205 
206 		ret = darray_push(&dup->ptrs, entry_ptr);
207 		if (ret)
208 			goto out;
209 
210 		bch2_journal_replay_to_text(&buf, c, dup);
211 
212 		fsck_err_on(same_device,
213 			    c, journal_entry_dup_same_device,
214 			    "duplicate journal entry on same device\n  %s",
215 			    buf.buf);
216 
217 		fsck_err_on(not_identical,
218 			    c, journal_entry_replicas_data_mismatch,
219 			    "found duplicate but non identical journal entries\n  %s",
220 			    buf.buf);
221 
222 		if (entry_ptr.csum_good && !identical)
223 			goto replace;
224 
225 		goto out;
226 	}
227 replace:
228 	i = kvmalloc(offsetof(struct journal_replay, j) + bytes, GFP_KERNEL);
229 	if (!i)
230 		return -BCH_ERR_ENOMEM_journal_entry_add;
231 
232 	darray_init(&i->ptrs);
233 	i->csum_good		= entry_ptr.csum_good;
234 	i->ignore_blacklisted	= false;
235 	i->ignore_not_dirty	= false;
236 	unsafe_memcpy(&i->j, j, bytes, "embedded variable length struct");
237 
238 	if (dup) {
239 		/* The first ptr should represent the jset we kept: */
240 		darray_for_each(dup->ptrs, ptr)
241 			darray_push(&i->ptrs, *ptr);
242 		__journal_replay_free(c, dup);
243 	} else {
244 		darray_push(&i->ptrs, entry_ptr);
245 	}
246 
247 	*_i = i;
248 out:
249 fsck_err:
250 	printbuf_exit(&buf);
251 	return ret;
252 }
253 
254 /* this fills in a range with empty jset_entries: */
255 static void journal_entry_null_range(void *start, void *end)
256 {
257 	struct jset_entry *entry;
258 
259 	for (entry = start; entry != end; entry = vstruct_next(entry))
260 		memset(entry, 0, sizeof(*entry));
261 }
262 
263 #define JOURNAL_ENTRY_REREAD	5
264 #define JOURNAL_ENTRY_NONE	6
265 #define JOURNAL_ENTRY_BAD	7
266 
267 static void journal_entry_err_msg(struct printbuf *out,
268 				  u32 version,
269 				  struct jset *jset,
270 				  struct jset_entry *entry)
271 {
272 	prt_str(out, "invalid journal entry, version=");
273 	bch2_version_to_text(out, version);
274 
275 	if (entry) {
276 		prt_str(out, " type=");
277 		bch2_prt_jset_entry_type(out, entry->type);
278 	}
279 
280 	if (!jset) {
281 		prt_printf(out, " in superblock");
282 	} else {
283 
284 		prt_printf(out, " seq=%llu", le64_to_cpu(jset->seq));
285 
286 		if (entry)
287 			prt_printf(out, " offset=%zi/%u",
288 				   (u64 *) entry - jset->_data,
289 				   le32_to_cpu(jset->u64s));
290 	}
291 
292 	prt_str(out, ": ");
293 }
294 
295 #define journal_entry_err(c, version, jset, entry, _err, msg, ...)	\
296 ({									\
297 	struct printbuf _buf = PRINTBUF;				\
298 									\
299 	journal_entry_err_msg(&_buf, version, jset, entry);		\
300 	prt_printf(&_buf, msg, ##__VA_ARGS__);				\
301 									\
302 	switch (flags & BCH_VALIDATE_write) {				\
303 	case READ:							\
304 		mustfix_fsck_err(c, _err, "%s", _buf.buf);		\
305 		break;							\
306 	case WRITE:							\
307 		bch2_sb_error_count(c, BCH_FSCK_ERR_##_err);		\
308 		bch_err(c, "corrupt metadata before write: %s\n", _buf.buf);\
309 		if (bch2_fs_inconsistent(c)) {				\
310 			ret = -BCH_ERR_fsck_errors_not_fixed;		\
311 			goto fsck_err;					\
312 		}							\
313 		break;							\
314 	}								\
315 									\
316 	printbuf_exit(&_buf);						\
317 	true;								\
318 })
319 
320 #define journal_entry_err_on(cond, ...)					\
321 	((cond) ? journal_entry_err(__VA_ARGS__) : false)
322 
323 #define FSCK_DELETED_KEY	5
324 
325 static int journal_validate_key(struct bch_fs *c,
326 				struct jset *jset,
327 				struct jset_entry *entry,
328 				unsigned level, enum btree_id btree_id,
329 				struct bkey_i *k,
330 				unsigned version, int big_endian,
331 				enum bch_validate_flags flags)
332 {
333 	int write = flags & BCH_VALIDATE_write;
334 	void *next = vstruct_next(entry);
335 	struct printbuf buf = PRINTBUF;
336 	int ret = 0;
337 
338 	if (journal_entry_err_on(!k->k.u64s,
339 				 c, version, jset, entry,
340 				 journal_entry_bkey_u64s_0,
341 				 "k->u64s 0")) {
342 		entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
343 		journal_entry_null_range(vstruct_next(entry), next);
344 		return FSCK_DELETED_KEY;
345 	}
346 
347 	if (journal_entry_err_on((void *) bkey_next(k) >
348 				 (void *) vstruct_next(entry),
349 				 c, version, jset, entry,
350 				 journal_entry_bkey_past_end,
351 				 "extends past end of journal entry")) {
352 		entry->u64s = cpu_to_le16((u64 *) k - entry->_data);
353 		journal_entry_null_range(vstruct_next(entry), next);
354 		return FSCK_DELETED_KEY;
355 	}
356 
357 	if (journal_entry_err_on(k->k.format != KEY_FORMAT_CURRENT,
358 				 c, version, jset, entry,
359 				 journal_entry_bkey_bad_format,
360 				 "bad format %u", k->k.format)) {
361 		le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
362 		memmove(k, bkey_next(k), next - (void *) bkey_next(k));
363 		journal_entry_null_range(vstruct_next(entry), next);
364 		return FSCK_DELETED_KEY;
365 	}
366 
367 	if (!write)
368 		bch2_bkey_compat(level, btree_id, version, big_endian,
369 				 write, NULL, bkey_to_packed(k));
370 
371 	if (bch2_bkey_invalid(c, bkey_i_to_s_c(k),
372 			      __btree_node_type(level, btree_id), write, &buf)) {
373 		printbuf_reset(&buf);
374 		journal_entry_err_msg(&buf, version, jset, entry);
375 		prt_newline(&buf);
376 		printbuf_indent_add(&buf, 2);
377 
378 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(k));
379 		prt_newline(&buf);
380 		bch2_bkey_invalid(c, bkey_i_to_s_c(k),
381 				  __btree_node_type(level, btree_id), write, &buf);
382 
383 		mustfix_fsck_err(c, journal_entry_bkey_invalid,
384 				 "%s", buf.buf);
385 
386 		le16_add_cpu(&entry->u64s, -((u16) k->k.u64s));
387 		memmove(k, bkey_next(k), next - (void *) bkey_next(k));
388 		journal_entry_null_range(vstruct_next(entry), next);
389 
390 		printbuf_exit(&buf);
391 		return FSCK_DELETED_KEY;
392 	}
393 
394 	if (write)
395 		bch2_bkey_compat(level, btree_id, version, big_endian,
396 				 write, NULL, bkey_to_packed(k));
397 fsck_err:
398 	printbuf_exit(&buf);
399 	return ret;
400 }
401 
402 static int journal_entry_btree_keys_validate(struct bch_fs *c,
403 				struct jset *jset,
404 				struct jset_entry *entry,
405 				unsigned version, int big_endian,
406 				enum bch_validate_flags flags)
407 {
408 	struct bkey_i *k = entry->start;
409 
410 	while (k != vstruct_last(entry)) {
411 		int ret = journal_validate_key(c, jset, entry,
412 					       entry->level,
413 					       entry->btree_id,
414 					       k, version, big_endian,
415 					       flags|BCH_VALIDATE_journal);
416 		if (ret == FSCK_DELETED_KEY)
417 			continue;
418 
419 		k = bkey_next(k);
420 	}
421 
422 	return 0;
423 }
424 
425 static void journal_entry_btree_keys_to_text(struct printbuf *out, struct bch_fs *c,
426 					     struct jset_entry *entry)
427 {
428 	bool first = true;
429 
430 	jset_entry_for_each_key(entry, k) {
431 		if (!first) {
432 			prt_newline(out);
433 			bch2_prt_jset_entry_type(out, entry->type);
434 			prt_str(out, ": ");
435 		}
436 		prt_printf(out, "btree=%s l=%u ", bch2_btree_id_str(entry->btree_id), entry->level);
437 		bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(k));
438 		first = false;
439 	}
440 }
441 
442 static int journal_entry_btree_root_validate(struct bch_fs *c,
443 				struct jset *jset,
444 				struct jset_entry *entry,
445 				unsigned version, int big_endian,
446 				enum bch_validate_flags flags)
447 {
448 	struct bkey_i *k = entry->start;
449 	int ret = 0;
450 
451 	if (journal_entry_err_on(!entry->u64s ||
452 				 le16_to_cpu(entry->u64s) != k->k.u64s,
453 				 c, version, jset, entry,
454 				 journal_entry_btree_root_bad_size,
455 				 "invalid btree root journal entry: wrong number of keys")) {
456 		void *next = vstruct_next(entry);
457 		/*
458 		 * we don't want to null out this jset_entry,
459 		 * just the contents, so that later we can tell
460 		 * we were _supposed_ to have a btree root
461 		 */
462 		entry->u64s = 0;
463 		journal_entry_null_range(vstruct_next(entry), next);
464 		return 0;
465 	}
466 
467 	ret = journal_validate_key(c, jset, entry, 1, entry->btree_id, k,
468 				   version, big_endian, flags);
469 	if (ret == FSCK_DELETED_KEY)
470 		ret = 0;
471 fsck_err:
472 	return ret;
473 }
474 
475 static void journal_entry_btree_root_to_text(struct printbuf *out, struct bch_fs *c,
476 					     struct jset_entry *entry)
477 {
478 	journal_entry_btree_keys_to_text(out, c, entry);
479 }
480 
481 static int journal_entry_prio_ptrs_validate(struct bch_fs *c,
482 				struct jset *jset,
483 				struct jset_entry *entry,
484 				unsigned version, int big_endian,
485 				enum bch_validate_flags flags)
486 {
487 	/* obsolete, don't care: */
488 	return 0;
489 }
490 
491 static void journal_entry_prio_ptrs_to_text(struct printbuf *out, struct bch_fs *c,
492 					    struct jset_entry *entry)
493 {
494 }
495 
496 static int journal_entry_blacklist_validate(struct bch_fs *c,
497 				struct jset *jset,
498 				struct jset_entry *entry,
499 				unsigned version, int big_endian,
500 				enum bch_validate_flags flags)
501 {
502 	int ret = 0;
503 
504 	if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 1,
505 				 c, version, jset, entry,
506 				 journal_entry_blacklist_bad_size,
507 		"invalid journal seq blacklist entry: bad size")) {
508 		journal_entry_null_range(entry, vstruct_next(entry));
509 	}
510 fsck_err:
511 	return ret;
512 }
513 
514 static void journal_entry_blacklist_to_text(struct printbuf *out, struct bch_fs *c,
515 					    struct jset_entry *entry)
516 {
517 	struct jset_entry_blacklist *bl =
518 		container_of(entry, struct jset_entry_blacklist, entry);
519 
520 	prt_printf(out, "seq=%llu", le64_to_cpu(bl->seq));
521 }
522 
523 static int journal_entry_blacklist_v2_validate(struct bch_fs *c,
524 				struct jset *jset,
525 				struct jset_entry *entry,
526 				unsigned version, int big_endian,
527 				enum bch_validate_flags flags)
528 {
529 	struct jset_entry_blacklist_v2 *bl_entry;
530 	int ret = 0;
531 
532 	if (journal_entry_err_on(le16_to_cpu(entry->u64s) != 2,
533 				 c, version, jset, entry,
534 				 journal_entry_blacklist_v2_bad_size,
535 		"invalid journal seq blacklist entry: bad size")) {
536 		journal_entry_null_range(entry, vstruct_next(entry));
537 		goto out;
538 	}
539 
540 	bl_entry = container_of(entry, struct jset_entry_blacklist_v2, entry);
541 
542 	if (journal_entry_err_on(le64_to_cpu(bl_entry->start) >
543 				 le64_to_cpu(bl_entry->end),
544 				 c, version, jset, entry,
545 				 journal_entry_blacklist_v2_start_past_end,
546 		"invalid journal seq blacklist entry: start > end")) {
547 		journal_entry_null_range(entry, vstruct_next(entry));
548 	}
549 out:
550 fsck_err:
551 	return ret;
552 }
553 
554 static void journal_entry_blacklist_v2_to_text(struct printbuf *out, struct bch_fs *c,
555 					       struct jset_entry *entry)
556 {
557 	struct jset_entry_blacklist_v2 *bl =
558 		container_of(entry, struct jset_entry_blacklist_v2, entry);
559 
560 	prt_printf(out, "start=%llu end=%llu",
561 	       le64_to_cpu(bl->start),
562 	       le64_to_cpu(bl->end));
563 }
564 
565 static int journal_entry_usage_validate(struct bch_fs *c,
566 				struct jset *jset,
567 				struct jset_entry *entry,
568 				unsigned version, int big_endian,
569 				enum bch_validate_flags flags)
570 {
571 	struct jset_entry_usage *u =
572 		container_of(entry, struct jset_entry_usage, entry);
573 	unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
574 	int ret = 0;
575 
576 	if (journal_entry_err_on(bytes < sizeof(*u),
577 				 c, version, jset, entry,
578 				 journal_entry_usage_bad_size,
579 				 "invalid journal entry usage: bad size")) {
580 		journal_entry_null_range(entry, vstruct_next(entry));
581 		return ret;
582 	}
583 
584 fsck_err:
585 	return ret;
586 }
587 
588 static void journal_entry_usage_to_text(struct printbuf *out, struct bch_fs *c,
589 					struct jset_entry *entry)
590 {
591 	struct jset_entry_usage *u =
592 		container_of(entry, struct jset_entry_usage, entry);
593 
594 	prt_str(out, "type=");
595 	bch2_prt_fs_usage_type(out, u->entry.btree_id);
596 	prt_printf(out, " v=%llu", le64_to_cpu(u->v));
597 }
598 
599 static int journal_entry_data_usage_validate(struct bch_fs *c,
600 				struct jset *jset,
601 				struct jset_entry *entry,
602 				unsigned version, int big_endian,
603 				enum bch_validate_flags flags)
604 {
605 	struct jset_entry_data_usage *u =
606 		container_of(entry, struct jset_entry_data_usage, entry);
607 	unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
608 	struct printbuf err = PRINTBUF;
609 	int ret = 0;
610 
611 	if (journal_entry_err_on(bytes < sizeof(*u) ||
612 				 bytes < sizeof(*u) + u->r.nr_devs,
613 				 c, version, jset, entry,
614 				 journal_entry_data_usage_bad_size,
615 				 "invalid journal entry usage: bad size")) {
616 		journal_entry_null_range(entry, vstruct_next(entry));
617 		goto out;
618 	}
619 
620 	if (journal_entry_err_on(bch2_replicas_entry_validate(&u->r, c->disk_sb.sb, &err),
621 				 c, version, jset, entry,
622 				 journal_entry_data_usage_bad_size,
623 				 "invalid journal entry usage: %s", err.buf)) {
624 		journal_entry_null_range(entry, vstruct_next(entry));
625 		goto out;
626 	}
627 out:
628 fsck_err:
629 	printbuf_exit(&err);
630 	return ret;
631 }
632 
633 static void journal_entry_data_usage_to_text(struct printbuf *out, struct bch_fs *c,
634 					     struct jset_entry *entry)
635 {
636 	struct jset_entry_data_usage *u =
637 		container_of(entry, struct jset_entry_data_usage, entry);
638 
639 	bch2_replicas_entry_to_text(out, &u->r);
640 	prt_printf(out, "=%llu", le64_to_cpu(u->v));
641 }
642 
643 static int journal_entry_clock_validate(struct bch_fs *c,
644 				struct jset *jset,
645 				struct jset_entry *entry,
646 				unsigned version, int big_endian,
647 				enum bch_validate_flags flags)
648 {
649 	struct jset_entry_clock *clock =
650 		container_of(entry, struct jset_entry_clock, entry);
651 	unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
652 	int ret = 0;
653 
654 	if (journal_entry_err_on(bytes != sizeof(*clock),
655 				 c, version, jset, entry,
656 				 journal_entry_clock_bad_size,
657 				 "bad size")) {
658 		journal_entry_null_range(entry, vstruct_next(entry));
659 		return ret;
660 	}
661 
662 	if (journal_entry_err_on(clock->rw > 1,
663 				 c, version, jset, entry,
664 				 journal_entry_clock_bad_rw,
665 				 "bad rw")) {
666 		journal_entry_null_range(entry, vstruct_next(entry));
667 		return ret;
668 	}
669 
670 fsck_err:
671 	return ret;
672 }
673 
674 static void journal_entry_clock_to_text(struct printbuf *out, struct bch_fs *c,
675 					struct jset_entry *entry)
676 {
677 	struct jset_entry_clock *clock =
678 		container_of(entry, struct jset_entry_clock, entry);
679 
680 	prt_printf(out, "%s=%llu", clock->rw ? "write" : "read", le64_to_cpu(clock->time));
681 }
682 
683 static int journal_entry_dev_usage_validate(struct bch_fs *c,
684 				struct jset *jset,
685 				struct jset_entry *entry,
686 				unsigned version, int big_endian,
687 				enum bch_validate_flags flags)
688 {
689 	struct jset_entry_dev_usage *u =
690 		container_of(entry, struct jset_entry_dev_usage, entry);
691 	unsigned bytes = jset_u64s(le16_to_cpu(entry->u64s)) * sizeof(u64);
692 	unsigned expected = sizeof(*u);
693 	int ret = 0;
694 
695 	if (journal_entry_err_on(bytes < expected,
696 				 c, version, jset, entry,
697 				 journal_entry_dev_usage_bad_size,
698 				 "bad size (%u < %u)",
699 				 bytes, expected)) {
700 		journal_entry_null_range(entry, vstruct_next(entry));
701 		return ret;
702 	}
703 
704 	if (journal_entry_err_on(u->pad,
705 				 c, version, jset, entry,
706 				 journal_entry_dev_usage_bad_pad,
707 				 "bad pad")) {
708 		journal_entry_null_range(entry, vstruct_next(entry));
709 		return ret;
710 	}
711 
712 fsck_err:
713 	return ret;
714 }
715 
716 static void journal_entry_dev_usage_to_text(struct printbuf *out, struct bch_fs *c,
717 					    struct jset_entry *entry)
718 {
719 	struct jset_entry_dev_usage *u =
720 		container_of(entry, struct jset_entry_dev_usage, entry);
721 	unsigned i, nr_types = jset_entry_dev_usage_nr_types(u);
722 
723 	prt_printf(out, "dev=%u", le32_to_cpu(u->dev));
724 
725 	for (i = 0; i < nr_types; i++) {
726 		bch2_prt_data_type(out, i);
727 		prt_printf(out, ": buckets=%llu sectors=%llu fragmented=%llu",
728 		       le64_to_cpu(u->d[i].buckets),
729 		       le64_to_cpu(u->d[i].sectors),
730 		       le64_to_cpu(u->d[i].fragmented));
731 	}
732 }
733 
734 static int journal_entry_log_validate(struct bch_fs *c,
735 				struct jset *jset,
736 				struct jset_entry *entry,
737 				unsigned version, int big_endian,
738 				enum bch_validate_flags flags)
739 {
740 	return 0;
741 }
742 
743 static void journal_entry_log_to_text(struct printbuf *out, struct bch_fs *c,
744 				      struct jset_entry *entry)
745 {
746 	struct jset_entry_log *l = container_of(entry, struct jset_entry_log, entry);
747 	unsigned bytes = vstruct_bytes(entry) - offsetof(struct jset_entry_log, d);
748 
749 	prt_printf(out, "%.*s", bytes, l->d);
750 }
751 
752 static int journal_entry_overwrite_validate(struct bch_fs *c,
753 				struct jset *jset,
754 				struct jset_entry *entry,
755 				unsigned version, int big_endian,
756 				enum bch_validate_flags flags)
757 {
758 	return journal_entry_btree_keys_validate(c, jset, entry,
759 				version, big_endian, READ);
760 }
761 
762 static void journal_entry_overwrite_to_text(struct printbuf *out, struct bch_fs *c,
763 					    struct jset_entry *entry)
764 {
765 	journal_entry_btree_keys_to_text(out, c, entry);
766 }
767 
768 static int journal_entry_write_buffer_keys_validate(struct bch_fs *c,
769 				struct jset *jset,
770 				struct jset_entry *entry,
771 				unsigned version, int big_endian,
772 				enum bch_validate_flags flags)
773 {
774 	return journal_entry_btree_keys_validate(c, jset, entry,
775 				version, big_endian, READ);
776 }
777 
778 static void journal_entry_write_buffer_keys_to_text(struct printbuf *out, struct bch_fs *c,
779 					    struct jset_entry *entry)
780 {
781 	journal_entry_btree_keys_to_text(out, c, entry);
782 }
783 
784 static int journal_entry_datetime_validate(struct bch_fs *c,
785 				struct jset *jset,
786 				struct jset_entry *entry,
787 				unsigned version, int big_endian,
788 				enum bch_validate_flags flags)
789 {
790 	unsigned bytes = vstruct_bytes(entry);
791 	unsigned expected = 16;
792 	int ret = 0;
793 
794 	if (journal_entry_err_on(vstruct_bytes(entry) < expected,
795 				 c, version, jset, entry,
796 				 journal_entry_dev_usage_bad_size,
797 				 "bad size (%u < %u)",
798 				 bytes, expected)) {
799 		journal_entry_null_range(entry, vstruct_next(entry));
800 		return ret;
801 	}
802 fsck_err:
803 	return ret;
804 }
805 
806 static void journal_entry_datetime_to_text(struct printbuf *out, struct bch_fs *c,
807 					    struct jset_entry *entry)
808 {
809 	struct jset_entry_datetime *datetime =
810 		container_of(entry, struct jset_entry_datetime, entry);
811 
812 	bch2_prt_datetime(out, le64_to_cpu(datetime->seconds));
813 }
814 
815 struct jset_entry_ops {
816 	int (*validate)(struct bch_fs *, struct jset *,
817 			struct jset_entry *, unsigned, int,
818 			enum bch_validate_flags);
819 	void (*to_text)(struct printbuf *, struct bch_fs *, struct jset_entry *);
820 };
821 
822 static const struct jset_entry_ops bch2_jset_entry_ops[] = {
823 #define x(f, nr)						\
824 	[BCH_JSET_ENTRY_##f]	= (struct jset_entry_ops) {	\
825 		.validate	= journal_entry_##f##_validate,	\
826 		.to_text	= journal_entry_##f##_to_text,	\
827 	},
828 	BCH_JSET_ENTRY_TYPES()
829 #undef x
830 };
831 
832 int bch2_journal_entry_validate(struct bch_fs *c,
833 				struct jset *jset,
834 				struct jset_entry *entry,
835 				unsigned version, int big_endian,
836 				enum bch_validate_flags flags)
837 {
838 	return entry->type < BCH_JSET_ENTRY_NR
839 		? bch2_jset_entry_ops[entry->type].validate(c, jset, entry,
840 				version, big_endian, flags)
841 		: 0;
842 }
843 
844 void bch2_journal_entry_to_text(struct printbuf *out, struct bch_fs *c,
845 				struct jset_entry *entry)
846 {
847 	bch2_prt_jset_entry_type(out, entry->type);
848 
849 	if (entry->type < BCH_JSET_ENTRY_NR) {
850 		prt_str(out, ": ");
851 		bch2_jset_entry_ops[entry->type].to_text(out, c, entry);
852 	}
853 }
854 
855 static int jset_validate_entries(struct bch_fs *c, struct jset *jset,
856 				 enum bch_validate_flags flags)
857 {
858 	unsigned version = le32_to_cpu(jset->version);
859 	int ret = 0;
860 
861 	vstruct_for_each(jset, entry) {
862 		if (journal_entry_err_on(vstruct_next(entry) > vstruct_last(jset),
863 				c, version, jset, entry,
864 				journal_entry_past_jset_end,
865 				"journal entry extends past end of jset")) {
866 			jset->u64s = cpu_to_le32((u64 *) entry - jset->_data);
867 			break;
868 		}
869 
870 		ret = bch2_journal_entry_validate(c, jset, entry,
871 					version, JSET_BIG_ENDIAN(jset), flags);
872 		if (ret)
873 			break;
874 	}
875 fsck_err:
876 	return ret;
877 }
878 
879 static int jset_validate(struct bch_fs *c,
880 			 struct bch_dev *ca,
881 			 struct jset *jset, u64 sector,
882 			 enum bch_validate_flags flags)
883 {
884 	unsigned version;
885 	int ret = 0;
886 
887 	if (le64_to_cpu(jset->magic) != jset_magic(c))
888 		return JOURNAL_ENTRY_NONE;
889 
890 	version = le32_to_cpu(jset->version);
891 	if (journal_entry_err_on(!bch2_version_compatible(version),
892 			c, version, jset, NULL,
893 			jset_unsupported_version,
894 			"%s sector %llu seq %llu: incompatible journal entry version %u.%u",
895 			ca ? ca->name : c->name,
896 			sector, le64_to_cpu(jset->seq),
897 			BCH_VERSION_MAJOR(version),
898 			BCH_VERSION_MINOR(version))) {
899 		/* don't try to continue: */
900 		return -EINVAL;
901 	}
902 
903 	if (journal_entry_err_on(!bch2_checksum_type_valid(c, JSET_CSUM_TYPE(jset)),
904 			c, version, jset, NULL,
905 			jset_unknown_csum,
906 			"%s sector %llu seq %llu: journal entry with unknown csum type %llu",
907 			ca ? ca->name : c->name,
908 			sector, le64_to_cpu(jset->seq),
909 			JSET_CSUM_TYPE(jset)))
910 		ret = JOURNAL_ENTRY_BAD;
911 
912 	/* last_seq is ignored when JSET_NO_FLUSH is true */
913 	if (journal_entry_err_on(!JSET_NO_FLUSH(jset) &&
914 				 le64_to_cpu(jset->last_seq) > le64_to_cpu(jset->seq),
915 				 c, version, jset, NULL,
916 				 jset_last_seq_newer_than_seq,
917 				 "invalid journal entry: last_seq > seq (%llu > %llu)",
918 				 le64_to_cpu(jset->last_seq),
919 				 le64_to_cpu(jset->seq))) {
920 		jset->last_seq = jset->seq;
921 		return JOURNAL_ENTRY_BAD;
922 	}
923 
924 	ret = jset_validate_entries(c, jset, flags);
925 fsck_err:
926 	return ret;
927 }
928 
929 static int jset_validate_early(struct bch_fs *c,
930 			 struct bch_dev *ca,
931 			 struct jset *jset, u64 sector,
932 			 unsigned bucket_sectors_left,
933 			 unsigned sectors_read)
934 {
935 	size_t bytes = vstruct_bytes(jset);
936 	unsigned version;
937 	enum bch_validate_flags flags = BCH_VALIDATE_journal;
938 	int ret = 0;
939 
940 	if (le64_to_cpu(jset->magic) != jset_magic(c))
941 		return JOURNAL_ENTRY_NONE;
942 
943 	version = le32_to_cpu(jset->version);
944 	if (journal_entry_err_on(!bch2_version_compatible(version),
945 			c, version, jset, NULL,
946 			jset_unsupported_version,
947 			"%s sector %llu seq %llu: unknown journal entry version %u.%u",
948 			ca ? ca->name : c->name,
949 			sector, le64_to_cpu(jset->seq),
950 			BCH_VERSION_MAJOR(version),
951 			BCH_VERSION_MINOR(version))) {
952 		/* don't try to continue: */
953 		return -EINVAL;
954 	}
955 
956 	if (bytes > (sectors_read << 9) &&
957 	    sectors_read < bucket_sectors_left)
958 		return JOURNAL_ENTRY_REREAD;
959 
960 	if (journal_entry_err_on(bytes > bucket_sectors_left << 9,
961 			c, version, jset, NULL,
962 			jset_past_bucket_end,
963 			"%s sector %llu seq %llu: journal entry too big (%zu bytes)",
964 			ca ? ca->name : c->name,
965 			sector, le64_to_cpu(jset->seq), bytes))
966 		le32_add_cpu(&jset->u64s,
967 			     -((bytes - (bucket_sectors_left << 9)) / 8));
968 fsck_err:
969 	return ret;
970 }
971 
972 struct journal_read_buf {
973 	void		*data;
974 	size_t		size;
975 };
976 
977 static int journal_read_buf_realloc(struct journal_read_buf *b,
978 				    size_t new_size)
979 {
980 	void *n;
981 
982 	/* the bios are sized for this many pages, max: */
983 	if (new_size > JOURNAL_ENTRY_SIZE_MAX)
984 		return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
985 
986 	new_size = roundup_pow_of_two(new_size);
987 	n = kvmalloc(new_size, GFP_KERNEL);
988 	if (!n)
989 		return -BCH_ERR_ENOMEM_journal_read_buf_realloc;
990 
991 	kvfree(b->data);
992 	b->data = n;
993 	b->size = new_size;
994 	return 0;
995 }
996 
997 static int journal_read_bucket(struct bch_dev *ca,
998 			       struct journal_read_buf *buf,
999 			       struct journal_list *jlist,
1000 			       unsigned bucket)
1001 {
1002 	struct bch_fs *c = ca->fs;
1003 	struct journal_device *ja = &ca->journal;
1004 	struct jset *j = NULL;
1005 	unsigned sectors, sectors_read = 0;
1006 	u64 offset = bucket_to_sector(ca, ja->buckets[bucket]),
1007 	    end = offset + ca->mi.bucket_size;
1008 	bool saw_bad = false, csum_good;
1009 	struct printbuf err = PRINTBUF;
1010 	int ret = 0;
1011 
1012 	pr_debug("reading %u", bucket);
1013 
1014 	while (offset < end) {
1015 		if (!sectors_read) {
1016 			struct bio *bio;
1017 			unsigned nr_bvecs;
1018 reread:
1019 			sectors_read = min_t(unsigned,
1020 				end - offset, buf->size >> 9);
1021 			nr_bvecs = buf_pages(buf->data, sectors_read << 9);
1022 
1023 			bio = bio_kmalloc(nr_bvecs, GFP_KERNEL);
1024 			bio_init(bio, ca->disk_sb.bdev, bio->bi_inline_vecs, nr_bvecs, REQ_OP_READ);
1025 
1026 			bio->bi_iter.bi_sector = offset;
1027 			bch2_bio_map(bio, buf->data, sectors_read << 9);
1028 
1029 			ret = submit_bio_wait(bio);
1030 			kfree(bio);
1031 
1032 			if (bch2_dev_io_err_on(ret, ca, BCH_MEMBER_ERROR_read,
1033 					       "journal read error: sector %llu",
1034 					       offset) ||
1035 			    bch2_meta_read_fault("journal")) {
1036 				/*
1037 				 * We don't error out of the recovery process
1038 				 * here, since the relevant journal entry may be
1039 				 * found on a different device, and missing or
1040 				 * no journal entries will be handled later
1041 				 */
1042 				goto out;
1043 			}
1044 
1045 			j = buf->data;
1046 		}
1047 
1048 		ret = jset_validate_early(c, ca, j, offset,
1049 				    end - offset, sectors_read);
1050 		switch (ret) {
1051 		case 0:
1052 			sectors = vstruct_sectors(j, c->block_bits);
1053 			break;
1054 		case JOURNAL_ENTRY_REREAD:
1055 			if (vstruct_bytes(j) > buf->size) {
1056 				ret = journal_read_buf_realloc(buf,
1057 							vstruct_bytes(j));
1058 				if (ret)
1059 					goto err;
1060 			}
1061 			goto reread;
1062 		case JOURNAL_ENTRY_NONE:
1063 			if (!saw_bad)
1064 				goto out;
1065 			/*
1066 			 * On checksum error we don't really trust the size
1067 			 * field of the journal entry we read, so try reading
1068 			 * again at next block boundary:
1069 			 */
1070 			sectors = block_sectors(c);
1071 			goto next_block;
1072 		default:
1073 			goto err;
1074 		}
1075 
1076 		if (le64_to_cpu(j->seq) > ja->highest_seq_found) {
1077 			ja->highest_seq_found = le64_to_cpu(j->seq);
1078 			ja->cur_idx = bucket;
1079 			ja->sectors_free = ca->mi.bucket_size -
1080 				bucket_remainder(ca, offset) - sectors;
1081 		}
1082 
1083 		/*
1084 		 * This happens sometimes if we don't have discards on -
1085 		 * when we've partially overwritten a bucket with new
1086 		 * journal entries. We don't need the rest of the
1087 		 * bucket:
1088 		 */
1089 		if (le64_to_cpu(j->seq) < ja->bucket_seq[bucket])
1090 			goto out;
1091 
1092 		ja->bucket_seq[bucket] = le64_to_cpu(j->seq);
1093 
1094 		enum bch_csum_type csum_type = JSET_CSUM_TYPE(j);
1095 		struct bch_csum csum;
1096 		csum_good = jset_csum_good(c, j, &csum);
1097 
1098 		if (bch2_dev_io_err_on(!csum_good, ca, BCH_MEMBER_ERROR_checksum,
1099 				       "%s",
1100 				       (printbuf_reset(&err),
1101 					prt_str(&err, "journal "),
1102 					bch2_csum_err_msg(&err, csum_type, j->csum, csum),
1103 					err.buf)))
1104 			saw_bad = true;
1105 
1106 		ret = bch2_encrypt(c, JSET_CSUM_TYPE(j), journal_nonce(j),
1107 			     j->encrypted_start,
1108 			     vstruct_end(j) - (void *) j->encrypted_start);
1109 		bch2_fs_fatal_err_on(ret, c, "decrypting journal entry: %s", bch2_err_str(ret));
1110 
1111 		mutex_lock(&jlist->lock);
1112 		ret = journal_entry_add(c, ca, (struct journal_ptr) {
1113 					.csum_good	= csum_good,
1114 					.dev		= ca->dev_idx,
1115 					.bucket		= bucket,
1116 					.bucket_offset	= offset -
1117 						bucket_to_sector(ca, ja->buckets[bucket]),
1118 					.sector		= offset,
1119 					}, jlist, j);
1120 		mutex_unlock(&jlist->lock);
1121 
1122 		switch (ret) {
1123 		case JOURNAL_ENTRY_ADD_OK:
1124 			break;
1125 		case JOURNAL_ENTRY_ADD_OUT_OF_RANGE:
1126 			break;
1127 		default:
1128 			goto err;
1129 		}
1130 next_block:
1131 		pr_debug("next");
1132 		offset		+= sectors;
1133 		sectors_read	-= sectors;
1134 		j = ((void *) j) + (sectors << 9);
1135 	}
1136 
1137 out:
1138 	ret = 0;
1139 err:
1140 	printbuf_exit(&err);
1141 	return ret;
1142 }
1143 
1144 static CLOSURE_CALLBACK(bch2_journal_read_device)
1145 {
1146 	closure_type(ja, struct journal_device, read);
1147 	struct bch_dev *ca = container_of(ja, struct bch_dev, journal);
1148 	struct bch_fs *c = ca->fs;
1149 	struct journal_list *jlist =
1150 		container_of(cl->parent, struct journal_list, cl);
1151 	struct journal_read_buf buf = { NULL, 0 };
1152 	unsigned i;
1153 	int ret = 0;
1154 
1155 	if (!ja->nr)
1156 		goto out;
1157 
1158 	ret = journal_read_buf_realloc(&buf, PAGE_SIZE);
1159 	if (ret)
1160 		goto err;
1161 
1162 	pr_debug("%u journal buckets", ja->nr);
1163 
1164 	for (i = 0; i < ja->nr; i++) {
1165 		ret = journal_read_bucket(ca, &buf, jlist, i);
1166 		if (ret)
1167 			goto err;
1168 	}
1169 
1170 	/*
1171 	 * Set dirty_idx to indicate the entire journal is full and needs to be
1172 	 * reclaimed - journal reclaim will immediately reclaim whatever isn't
1173 	 * pinned when it first runs:
1174 	 */
1175 	ja->discard_idx = ja->dirty_idx_ondisk =
1176 		ja->dirty_idx = (ja->cur_idx + 1) % ja->nr;
1177 out:
1178 	bch_verbose(c, "journal read done on device %s, ret %i", ca->name, ret);
1179 	kvfree(buf.data);
1180 	percpu_ref_put(&ca->io_ref);
1181 	closure_return(cl);
1182 	return;
1183 err:
1184 	mutex_lock(&jlist->lock);
1185 	jlist->ret = ret;
1186 	mutex_unlock(&jlist->lock);
1187 	goto out;
1188 }
1189 
1190 int bch2_journal_read(struct bch_fs *c,
1191 		      u64 *last_seq,
1192 		      u64 *blacklist_seq,
1193 		      u64 *start_seq)
1194 {
1195 	struct journal_list jlist;
1196 	struct journal_replay *i, **_i, *prev = NULL;
1197 	struct genradix_iter radix_iter;
1198 	struct printbuf buf = PRINTBUF;
1199 	bool degraded = false, last_write_torn = false;
1200 	u64 seq;
1201 	int ret = 0;
1202 
1203 	closure_init_stack(&jlist.cl);
1204 	mutex_init(&jlist.lock);
1205 	jlist.last_seq = 0;
1206 	jlist.ret = 0;
1207 
1208 	for_each_member_device(c, ca) {
1209 		if (!c->opts.fsck &&
1210 		    !(bch2_dev_has_data(c, ca) & (1 << BCH_DATA_journal)))
1211 			continue;
1212 
1213 		if ((ca->mi.state == BCH_MEMBER_STATE_rw ||
1214 		     ca->mi.state == BCH_MEMBER_STATE_ro) &&
1215 		    percpu_ref_tryget(&ca->io_ref))
1216 			closure_call(&ca->journal.read,
1217 				     bch2_journal_read_device,
1218 				     system_unbound_wq,
1219 				     &jlist.cl);
1220 		else
1221 			degraded = true;
1222 	}
1223 
1224 	closure_sync(&jlist.cl);
1225 
1226 	if (jlist.ret)
1227 		return jlist.ret;
1228 
1229 	*last_seq	= 0;
1230 	*start_seq	= 0;
1231 	*blacklist_seq	= 0;
1232 
1233 	/*
1234 	 * Find most recent flush entry, and ignore newer non flush entries -
1235 	 * those entries will be blacklisted:
1236 	 */
1237 	genradix_for_each_reverse(&c->journal_entries, radix_iter, _i) {
1238 		enum bch_validate_flags flags = BCH_VALIDATE_journal;
1239 
1240 		i = *_i;
1241 
1242 		if (journal_replay_ignore(i))
1243 			continue;
1244 
1245 		if (!*start_seq)
1246 			*blacklist_seq = *start_seq = le64_to_cpu(i->j.seq) + 1;
1247 
1248 		if (JSET_NO_FLUSH(&i->j)) {
1249 			i->ignore_blacklisted = true;
1250 			continue;
1251 		}
1252 
1253 		if (!last_write_torn && !i->csum_good) {
1254 			last_write_torn = true;
1255 			i->ignore_blacklisted = true;
1256 			continue;
1257 		}
1258 
1259 		if (journal_entry_err_on(le64_to_cpu(i->j.last_seq) > le64_to_cpu(i->j.seq),
1260 					 c, le32_to_cpu(i->j.version), &i->j, NULL,
1261 					 jset_last_seq_newer_than_seq,
1262 					 "invalid journal entry: last_seq > seq (%llu > %llu)",
1263 					 le64_to_cpu(i->j.last_seq),
1264 					 le64_to_cpu(i->j.seq)))
1265 			i->j.last_seq = i->j.seq;
1266 
1267 		*last_seq	= le64_to_cpu(i->j.last_seq);
1268 		*blacklist_seq	= le64_to_cpu(i->j.seq) + 1;
1269 		break;
1270 	}
1271 
1272 	if (!*start_seq) {
1273 		bch_info(c, "journal read done, but no entries found");
1274 		return 0;
1275 	}
1276 
1277 	if (!*last_seq) {
1278 		fsck_err(c, dirty_but_no_journal_entries_post_drop_nonflushes,
1279 			 "journal read done, but no entries found after dropping non-flushes");
1280 		return 0;
1281 	}
1282 
1283 	bch_info(c, "journal read done, replaying entries %llu-%llu",
1284 		 *last_seq, *blacklist_seq - 1);
1285 
1286 	if (*start_seq != *blacklist_seq)
1287 		bch_info(c, "dropped unflushed entries %llu-%llu",
1288 			 *blacklist_seq, *start_seq - 1);
1289 
1290 	/* Drop blacklisted entries and entries older than last_seq: */
1291 	genradix_for_each(&c->journal_entries, radix_iter, _i) {
1292 		i = *_i;
1293 
1294 		if (journal_replay_ignore(i))
1295 			continue;
1296 
1297 		seq = le64_to_cpu(i->j.seq);
1298 		if (seq < *last_seq) {
1299 			journal_replay_free(c, i, false);
1300 			continue;
1301 		}
1302 
1303 		if (bch2_journal_seq_is_blacklisted(c, seq, true)) {
1304 			fsck_err_on(!JSET_NO_FLUSH(&i->j), c,
1305 				    jset_seq_blacklisted,
1306 				    "found blacklisted journal entry %llu", seq);
1307 			i->ignore_blacklisted = true;
1308 		}
1309 	}
1310 
1311 	/* Check for missing entries: */
1312 	seq = *last_seq;
1313 	genradix_for_each(&c->journal_entries, radix_iter, _i) {
1314 		i = *_i;
1315 
1316 		if (journal_replay_ignore(i))
1317 			continue;
1318 
1319 		BUG_ON(seq > le64_to_cpu(i->j.seq));
1320 
1321 		while (seq < le64_to_cpu(i->j.seq)) {
1322 			u64 missing_start, missing_end;
1323 			struct printbuf buf1 = PRINTBUF, buf2 = PRINTBUF;
1324 
1325 			while (seq < le64_to_cpu(i->j.seq) &&
1326 			       bch2_journal_seq_is_blacklisted(c, seq, false))
1327 				seq++;
1328 
1329 			if (seq == le64_to_cpu(i->j.seq))
1330 				break;
1331 
1332 			missing_start = seq;
1333 
1334 			while (seq < le64_to_cpu(i->j.seq) &&
1335 			       !bch2_journal_seq_is_blacklisted(c, seq, false))
1336 				seq++;
1337 
1338 			if (prev) {
1339 				bch2_journal_ptrs_to_text(&buf1, c, prev);
1340 				prt_printf(&buf1, " size %zu", vstruct_sectors(&prev->j, c->block_bits));
1341 			} else
1342 				prt_printf(&buf1, "(none)");
1343 			bch2_journal_ptrs_to_text(&buf2, c, i);
1344 
1345 			missing_end = seq - 1;
1346 			fsck_err(c, journal_entries_missing,
1347 				 "journal entries %llu-%llu missing! (replaying %llu-%llu)\n"
1348 				 "  prev at %s\n"
1349 				 "  next at %s, continue?",
1350 				 missing_start, missing_end,
1351 				 *last_seq, *blacklist_seq - 1,
1352 				 buf1.buf, buf2.buf);
1353 
1354 			printbuf_exit(&buf1);
1355 			printbuf_exit(&buf2);
1356 		}
1357 
1358 		prev = i;
1359 		seq++;
1360 	}
1361 
1362 	genradix_for_each(&c->journal_entries, radix_iter, _i) {
1363 		struct bch_replicas_padded replicas = {
1364 			.e.data_type = BCH_DATA_journal,
1365 			.e.nr_required = 1,
1366 		};
1367 
1368 		i = *_i;
1369 		if (journal_replay_ignore(i))
1370 			continue;
1371 
1372 		darray_for_each(i->ptrs, ptr) {
1373 			struct bch_dev *ca = bch2_dev_have_ref(c, ptr->dev);
1374 
1375 			if (!ptr->csum_good)
1376 				bch_err_dev_offset(ca, ptr->sector,
1377 						   "invalid journal checksum, seq %llu%s",
1378 						   le64_to_cpu(i->j.seq),
1379 						   i->csum_good ? " (had good copy on another device)" : "");
1380 		}
1381 
1382 		ret = jset_validate(c,
1383 				    bch2_dev_have_ref(c, i->ptrs.data[0].dev),
1384 				    &i->j,
1385 				    i->ptrs.data[0].sector,
1386 				    READ);
1387 		if (ret)
1388 			goto err;
1389 
1390 		darray_for_each(i->ptrs, ptr)
1391 			replicas.e.devs[replicas.e.nr_devs++] = ptr->dev;
1392 
1393 		bch2_replicas_entry_sort(&replicas.e);
1394 
1395 		printbuf_reset(&buf);
1396 		bch2_replicas_entry_to_text(&buf, &replicas.e);
1397 
1398 		if (!degraded &&
1399 		    !bch2_replicas_marked(c, &replicas.e) &&
1400 		    (le64_to_cpu(i->j.seq) == *last_seq ||
1401 		     fsck_err(c, journal_entry_replicas_not_marked,
1402 			      "superblock not marked as containing replicas for journal entry %llu\n  %s",
1403 			      le64_to_cpu(i->j.seq), buf.buf))) {
1404 			ret = bch2_mark_replicas(c, &replicas.e);
1405 			if (ret)
1406 				goto err;
1407 		}
1408 	}
1409 err:
1410 fsck_err:
1411 	printbuf_exit(&buf);
1412 	return ret;
1413 }
1414 
1415 /* journal write: */
1416 
1417 static void __journal_write_alloc(struct journal *j,
1418 				  struct journal_buf *w,
1419 				  struct dev_alloc_list *devs_sorted,
1420 				  unsigned sectors,
1421 				  unsigned *replicas,
1422 				  unsigned replicas_want)
1423 {
1424 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1425 	struct journal_device *ja;
1426 	struct bch_dev *ca;
1427 	unsigned i;
1428 
1429 	if (*replicas >= replicas_want)
1430 		return;
1431 
1432 	for (i = 0; i < devs_sorted->nr; i++) {
1433 		ca = rcu_dereference(c->devs[devs_sorted->devs[i]]);
1434 		if (!ca)
1435 			continue;
1436 
1437 		ja = &ca->journal;
1438 
1439 		/*
1440 		 * Check that we can use this device, and aren't already using
1441 		 * it:
1442 		 */
1443 		if (!ca->mi.durability ||
1444 		    ca->mi.state != BCH_MEMBER_STATE_rw ||
1445 		    !ja->nr ||
1446 		    bch2_bkey_has_device_c(bkey_i_to_s_c(&w->key), ca->dev_idx) ||
1447 		    sectors > ja->sectors_free)
1448 			continue;
1449 
1450 		bch2_dev_stripe_increment(ca, &j->wp.stripe);
1451 
1452 		bch2_bkey_append_ptr(&w->key,
1453 			(struct bch_extent_ptr) {
1454 				  .offset = bucket_to_sector(ca,
1455 					ja->buckets[ja->cur_idx]) +
1456 					ca->mi.bucket_size -
1457 					ja->sectors_free,
1458 				  .dev = ca->dev_idx,
1459 		});
1460 
1461 		ja->sectors_free -= sectors;
1462 		ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1463 
1464 		*replicas += ca->mi.durability;
1465 
1466 		if (*replicas >= replicas_want)
1467 			break;
1468 	}
1469 }
1470 
1471 /**
1472  * journal_write_alloc - decide where to write next journal entry
1473  *
1474  * @j:		journal object
1475  * @w:		journal buf (entry to be written)
1476  *
1477  * Returns: 0 on success, or -EROFS on failure
1478  */
1479 static int journal_write_alloc(struct journal *j, struct journal_buf *w)
1480 {
1481 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1482 	struct bch_devs_mask devs;
1483 	struct journal_device *ja;
1484 	struct bch_dev *ca;
1485 	struct dev_alloc_list devs_sorted;
1486 	unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1487 	unsigned target = c->opts.metadata_target ?:
1488 		c->opts.foreground_target;
1489 	unsigned i, replicas = 0, replicas_want =
1490 		READ_ONCE(c->opts.metadata_replicas);
1491 	unsigned replicas_need = min_t(unsigned, replicas_want,
1492 				       READ_ONCE(c->opts.metadata_replicas_required));
1493 
1494 	rcu_read_lock();
1495 retry:
1496 	devs = target_rw_devs(c, BCH_DATA_journal, target);
1497 
1498 	devs_sorted = bch2_dev_alloc_list(c, &j->wp.stripe, &devs);
1499 
1500 	__journal_write_alloc(j, w, &devs_sorted,
1501 			      sectors, &replicas, replicas_want);
1502 
1503 	if (replicas >= replicas_want)
1504 		goto done;
1505 
1506 	for (i = 0; i < devs_sorted.nr; i++) {
1507 		ca = rcu_dereference(c->devs[devs_sorted.devs[i]]);
1508 		if (!ca)
1509 			continue;
1510 
1511 		ja = &ca->journal;
1512 
1513 		if (sectors > ja->sectors_free &&
1514 		    sectors <= ca->mi.bucket_size &&
1515 		    bch2_journal_dev_buckets_available(j, ja,
1516 					journal_space_discarded)) {
1517 			ja->cur_idx = (ja->cur_idx + 1) % ja->nr;
1518 			ja->sectors_free = ca->mi.bucket_size;
1519 
1520 			/*
1521 			 * ja->bucket_seq[ja->cur_idx] must always have
1522 			 * something sensible:
1523 			 */
1524 			ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1525 		}
1526 	}
1527 
1528 	__journal_write_alloc(j, w, &devs_sorted,
1529 			      sectors, &replicas, replicas_want);
1530 
1531 	if (replicas < replicas_want && target) {
1532 		/* Retry from all devices: */
1533 		target = 0;
1534 		goto retry;
1535 	}
1536 done:
1537 	rcu_read_unlock();
1538 
1539 	BUG_ON(bkey_val_u64s(&w->key.k) > BCH_REPLICAS_MAX);
1540 
1541 	return replicas >= replicas_need ? 0 : -EROFS;
1542 }
1543 
1544 static void journal_buf_realloc(struct journal *j, struct journal_buf *buf)
1545 {
1546 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1547 
1548 	/* we aren't holding j->lock: */
1549 	unsigned new_size = READ_ONCE(j->buf_size_want);
1550 	void *new_buf;
1551 
1552 	if (buf->buf_size >= new_size)
1553 		return;
1554 
1555 	size_t btree_write_buffer_size = new_size / 64;
1556 
1557 	if (bch2_btree_write_buffer_resize(c, btree_write_buffer_size))
1558 		return;
1559 
1560 	new_buf = kvmalloc(new_size, GFP_NOFS|__GFP_NOWARN);
1561 	if (!new_buf)
1562 		return;
1563 
1564 	memcpy(new_buf, buf->data, buf->buf_size);
1565 
1566 	spin_lock(&j->lock);
1567 	swap(buf->data,		new_buf);
1568 	swap(buf->buf_size,	new_size);
1569 	spin_unlock(&j->lock);
1570 
1571 	kvfree(new_buf);
1572 }
1573 
1574 static inline struct journal_buf *journal_last_unwritten_buf(struct journal *j)
1575 {
1576 	return j->buf + (journal_last_unwritten_seq(j) & JOURNAL_BUF_MASK);
1577 }
1578 
1579 static CLOSURE_CALLBACK(journal_write_done)
1580 {
1581 	closure_type(w, struct journal_buf, io);
1582 	struct journal *j = container_of(w, struct journal, buf[w->idx]);
1583 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1584 	struct bch_replicas_padded replicas;
1585 	union journal_res_state old, new;
1586 	u64 v, seq = le64_to_cpu(w->data->seq);
1587 	int err = 0;
1588 
1589 	bch2_time_stats_update(!JSET_NO_FLUSH(w->data)
1590 			       ? j->flush_write_time
1591 			       : j->noflush_write_time, j->write_start_time);
1592 
1593 	if (!w->devs_written.nr) {
1594 		bch_err(c, "unable to write journal to sufficient devices");
1595 		err = -EIO;
1596 	} else {
1597 		bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
1598 					 w->devs_written);
1599 		if (bch2_mark_replicas(c, &replicas.e))
1600 			err = -EIO;
1601 	}
1602 
1603 	if (err)
1604 		bch2_fatal_error(c);
1605 
1606 	closure_debug_destroy(cl);
1607 
1608 	spin_lock(&j->lock);
1609 	if (seq >= j->pin.front)
1610 		journal_seq_pin(j, seq)->devs = w->devs_written;
1611 	if (err && (!j->err_seq || seq < j->err_seq))
1612 		j->err_seq	= seq;
1613 	w->write_done = true;
1614 
1615 	bool completed = false;
1616 
1617 	for (seq = journal_last_unwritten_seq(j);
1618 	     seq <= journal_cur_seq(j);
1619 	     seq++) {
1620 		w = j->buf + (seq & JOURNAL_BUF_MASK);
1621 		if (!w->write_done)
1622 			break;
1623 
1624 		if (!j->err_seq && !JSET_NO_FLUSH(w->data)) {
1625 			j->flushed_seq_ondisk = seq;
1626 			j->last_seq_ondisk = w->last_seq;
1627 
1628 			bch2_do_discards(c);
1629 			closure_wake_up(&c->freelist_wait);
1630 			bch2_reset_alloc_cursors(c);
1631 		}
1632 
1633 		j->seq_ondisk = seq;
1634 
1635 		/*
1636 		 * Updating last_seq_ondisk may let bch2_journal_reclaim_work() discard
1637 		 * more buckets:
1638 		 *
1639 		 * Must come before signaling write completion, for
1640 		 * bch2_fs_journal_stop():
1641 		 */
1642 		if (j->watermark != BCH_WATERMARK_stripe)
1643 			journal_reclaim_kick(&c->journal);
1644 
1645 		v = atomic64_read(&j->reservations.counter);
1646 		do {
1647 			old.v = new.v = v;
1648 			BUG_ON(journal_state_count(new, new.unwritten_idx));
1649 			BUG_ON(new.unwritten_idx != (seq & JOURNAL_BUF_MASK));
1650 
1651 			new.unwritten_idx++;
1652 		} while ((v = atomic64_cmpxchg(&j->reservations.counter, old.v, new.v)) != old.v);
1653 
1654 		closure_wake_up(&w->wait);
1655 		completed = true;
1656 	}
1657 
1658 	if (completed) {
1659 		bch2_journal_reclaim_fast(j);
1660 		bch2_journal_space_available(j);
1661 
1662 		track_event_change(&c->times[BCH_TIME_blocked_journal_max_in_flight], false);
1663 
1664 		journal_wake(j);
1665 	}
1666 
1667 	if (journal_last_unwritten_seq(j) == journal_cur_seq(j) &&
1668 		   new.cur_entry_offset < JOURNAL_ENTRY_CLOSED_VAL) {
1669 		struct journal_buf *buf = journal_cur_buf(j);
1670 		long delta = buf->expires - jiffies;
1671 
1672 		/*
1673 		 * We don't close a journal entry to write it while there's
1674 		 * previous entries still in flight - the current journal entry
1675 		 * might want to be written now:
1676 		 */
1677 		mod_delayed_work(j->wq, &j->write_work, max(0L, delta));
1678 	}
1679 
1680 	/*
1681 	 * We don't typically trigger journal writes from her - the next journal
1682 	 * write will be triggered immediately after the previous one is
1683 	 * allocated, in bch2_journal_write() - but the journal write error path
1684 	 * is special:
1685 	 */
1686 	bch2_journal_do_writes(j);
1687 	spin_unlock(&j->lock);
1688 }
1689 
1690 static void journal_write_endio(struct bio *bio)
1691 {
1692 	struct journal_bio *jbio = container_of(bio, struct journal_bio, bio);
1693 	struct bch_dev *ca = jbio->ca;
1694 	struct journal *j = &ca->fs->journal;
1695 	struct journal_buf *w = j->buf + jbio->buf_idx;
1696 
1697 	if (bch2_dev_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write,
1698 			       "error writing journal entry %llu: %s",
1699 			       le64_to_cpu(w->data->seq),
1700 			       bch2_blk_status_to_str(bio->bi_status)) ||
1701 	    bch2_meta_write_fault("journal")) {
1702 		unsigned long flags;
1703 
1704 		spin_lock_irqsave(&j->err_lock, flags);
1705 		bch2_dev_list_drop_dev(&w->devs_written, ca->dev_idx);
1706 		spin_unlock_irqrestore(&j->err_lock, flags);
1707 	}
1708 
1709 	closure_put(&w->io);
1710 	percpu_ref_put(&ca->io_ref);
1711 }
1712 
1713 static CLOSURE_CALLBACK(journal_write_submit)
1714 {
1715 	closure_type(w, struct journal_buf, io);
1716 	struct journal *j = container_of(w, struct journal, buf[w->idx]);
1717 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1718 	unsigned sectors = vstruct_sectors(w->data, c->block_bits);
1719 
1720 	extent_for_each_ptr(bkey_i_to_s_extent(&w->key), ptr) {
1721 		struct bch_dev *ca = bch2_dev_get_ioref(c, ptr->dev, WRITE);
1722 		if (!ca) {
1723 			/* XXX: fix this */
1724 			bch_err(c, "missing device for journal write\n");
1725 			continue;
1726 		}
1727 
1728 		this_cpu_add(ca->io_done->sectors[WRITE][BCH_DATA_journal],
1729 			     sectors);
1730 
1731 		struct journal_device *ja = &ca->journal;
1732 		struct bio *bio = &ja->bio[w->idx]->bio;
1733 		bio_reset(bio, ca->disk_sb.bdev, REQ_OP_WRITE|REQ_SYNC|REQ_META);
1734 		bio->bi_iter.bi_sector	= ptr->offset;
1735 		bio->bi_end_io		= journal_write_endio;
1736 		bio->bi_private		= ca;
1737 
1738 		BUG_ON(bio->bi_iter.bi_sector == ca->prev_journal_sector);
1739 		ca->prev_journal_sector = bio->bi_iter.bi_sector;
1740 
1741 		if (!JSET_NO_FLUSH(w->data))
1742 			bio->bi_opf    |= REQ_FUA;
1743 		if (!JSET_NO_FLUSH(w->data) && !w->separate_flush)
1744 			bio->bi_opf    |= REQ_PREFLUSH;
1745 
1746 		bch2_bio_map(bio, w->data, sectors << 9);
1747 
1748 		trace_and_count(c, journal_write, bio);
1749 		closure_bio_submit(bio, cl);
1750 
1751 		ja->bucket_seq[ja->cur_idx] = le64_to_cpu(w->data->seq);
1752 	}
1753 
1754 	continue_at(cl, journal_write_done, j->wq);
1755 }
1756 
1757 static CLOSURE_CALLBACK(journal_write_preflush)
1758 {
1759 	closure_type(w, struct journal_buf, io);
1760 	struct journal *j = container_of(w, struct journal, buf[w->idx]);
1761 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1762 
1763 	if (j->seq_ondisk + 1 != le64_to_cpu(w->data->seq)) {
1764 		spin_lock(&j->lock);
1765 		closure_wait(&j->async_wait, cl);
1766 		spin_unlock(&j->lock);
1767 
1768 		continue_at(cl, journal_write_preflush, j->wq);
1769 		return;
1770 	}
1771 
1772 	if (w->separate_flush) {
1773 		for_each_rw_member(c, ca) {
1774 			percpu_ref_get(&ca->io_ref);
1775 
1776 			struct journal_device *ja = &ca->journal;
1777 			struct bio *bio = &ja->bio[w->idx]->bio;
1778 			bio_reset(bio, ca->disk_sb.bdev,
1779 				  REQ_OP_WRITE|REQ_SYNC|REQ_META|REQ_PREFLUSH);
1780 			bio->bi_end_io		= journal_write_endio;
1781 			bio->bi_private		= ca;
1782 			closure_bio_submit(bio, cl);
1783 		}
1784 
1785 		continue_at(cl, journal_write_submit, j->wq);
1786 	} else {
1787 		/*
1788 		 * no need to punt to another work item if we're not waiting on
1789 		 * preflushes
1790 		 */
1791 		journal_write_submit(&cl->work);
1792 	}
1793 }
1794 
1795 static int bch2_journal_write_prep(struct journal *j, struct journal_buf *w)
1796 {
1797 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1798 	struct jset_entry *start, *end;
1799 	struct jset *jset = w->data;
1800 	struct journal_keys_to_wb wb = { NULL };
1801 	unsigned sectors, bytes, u64s;
1802 	unsigned long btree_roots_have = 0;
1803 	bool validate_before_checksum = false;
1804 	u64 seq = le64_to_cpu(jset->seq);
1805 	int ret;
1806 
1807 	/*
1808 	 * Simple compaction, dropping empty jset_entries (from journal
1809 	 * reservations that weren't fully used) and merging jset_entries that
1810 	 * can be.
1811 	 *
1812 	 * If we wanted to be really fancy here, we could sort all the keys in
1813 	 * the jset and drop keys that were overwritten - probably not worth it:
1814 	 */
1815 	vstruct_for_each(jset, i) {
1816 		unsigned u64s = le16_to_cpu(i->u64s);
1817 
1818 		/* Empty entry: */
1819 		if (!u64s)
1820 			continue;
1821 
1822 		/*
1823 		 * New btree roots are set by journalling them; when the journal
1824 		 * entry gets written we have to propagate them to
1825 		 * c->btree_roots
1826 		 *
1827 		 * But, every journal entry we write has to contain all the
1828 		 * btree roots (at least for now); so after we copy btree roots
1829 		 * to c->btree_roots we have to get any missing btree roots and
1830 		 * add them to this journal entry:
1831 		 */
1832 		switch (i->type) {
1833 		case BCH_JSET_ENTRY_btree_root:
1834 			bch2_journal_entry_to_btree_root(c, i);
1835 			__set_bit(i->btree_id, &btree_roots_have);
1836 			break;
1837 		case BCH_JSET_ENTRY_write_buffer_keys:
1838 			EBUG_ON(!w->need_flush_to_write_buffer);
1839 
1840 			if (!wb.wb)
1841 				bch2_journal_keys_to_write_buffer_start(c, &wb, seq);
1842 
1843 			jset_entry_for_each_key(i, k) {
1844 				ret = bch2_journal_key_to_wb(c, &wb, i->btree_id, k);
1845 				if (ret) {
1846 					bch2_fs_fatal_error(c, "flushing journal keys to btree write buffer: %s",
1847 							    bch2_err_str(ret));
1848 					bch2_journal_keys_to_write_buffer_end(c, &wb);
1849 					return ret;
1850 				}
1851 			}
1852 			i->type = BCH_JSET_ENTRY_btree_keys;
1853 			break;
1854 		}
1855 	}
1856 
1857 	if (wb.wb)
1858 		bch2_journal_keys_to_write_buffer_end(c, &wb);
1859 
1860 	spin_lock(&c->journal.lock);
1861 	w->need_flush_to_write_buffer = false;
1862 	spin_unlock(&c->journal.lock);
1863 
1864 	start = end = vstruct_last(jset);
1865 
1866 	end	= bch2_btree_roots_to_journal_entries(c, end, btree_roots_have);
1867 
1868 	struct jset_entry_datetime *d =
1869 		container_of(jset_entry_init(&end, sizeof(*d)), struct jset_entry_datetime, entry);
1870 	d->entry.type	= BCH_JSET_ENTRY_datetime;
1871 	d->seconds	= cpu_to_le64(ktime_get_real_seconds());
1872 
1873 	bch2_journal_super_entries_add_common(c, &end, seq);
1874 	u64s	= (u64 *) end - (u64 *) start;
1875 
1876 	WARN_ON(u64s > j->entry_u64s_reserved);
1877 
1878 	le32_add_cpu(&jset->u64s, u64s);
1879 
1880 	sectors = vstruct_sectors(jset, c->block_bits);
1881 	bytes	= vstruct_bytes(jset);
1882 
1883 	if (sectors > w->sectors) {
1884 		bch2_fs_fatal_error(c, ": journal write overran available space, %zu > %u (extra %u reserved %u/%u)",
1885 				    vstruct_bytes(jset), w->sectors << 9,
1886 				    u64s, w->u64s_reserved, j->entry_u64s_reserved);
1887 		return -EINVAL;
1888 	}
1889 
1890 	jset->magic		= cpu_to_le64(jset_magic(c));
1891 	jset->version		= cpu_to_le32(c->sb.version);
1892 
1893 	SET_JSET_BIG_ENDIAN(jset, CPU_BIG_ENDIAN);
1894 	SET_JSET_CSUM_TYPE(jset, bch2_meta_checksum_type(c));
1895 
1896 	if (!JSET_NO_FLUSH(jset) && journal_entry_empty(jset))
1897 		j->last_empty_seq = seq;
1898 
1899 	if (bch2_csum_type_is_encryption(JSET_CSUM_TYPE(jset)))
1900 		validate_before_checksum = true;
1901 
1902 	if (le32_to_cpu(jset->version) < bcachefs_metadata_version_current)
1903 		validate_before_checksum = true;
1904 
1905 	if (validate_before_checksum &&
1906 	    (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1907 		return ret;
1908 
1909 	ret = bch2_encrypt(c, JSET_CSUM_TYPE(jset), journal_nonce(jset),
1910 		    jset->encrypted_start,
1911 		    vstruct_end(jset) - (void *) jset->encrypted_start);
1912 	if (bch2_fs_fatal_err_on(ret, c, "decrypting journal entry: %s", bch2_err_str(ret)))
1913 		return ret;
1914 
1915 	jset->csum = csum_vstruct(c, JSET_CSUM_TYPE(jset),
1916 				  journal_nonce(jset), jset);
1917 
1918 	if (!validate_before_checksum &&
1919 	    (ret = jset_validate(c, NULL, jset, 0, WRITE)))
1920 		return ret;
1921 
1922 	memset((void *) jset + bytes, 0, (sectors << 9) - bytes);
1923 	return 0;
1924 }
1925 
1926 static int bch2_journal_write_pick_flush(struct journal *j, struct journal_buf *w)
1927 {
1928 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1929 	int error = bch2_journal_error(j);
1930 
1931 	/*
1932 	 * If the journal is in an error state - we did an emergency shutdown -
1933 	 * we prefer to continue doing journal writes. We just mark them as
1934 	 * noflush so they'll never be used, but they'll still be visible by the
1935 	 * list_journal tool - this helps in debugging.
1936 	 *
1937 	 * There's a caveat: the first journal write after marking the
1938 	 * superblock dirty must always be a flush write, because on startup
1939 	 * from a clean shutdown we didn't necessarily read the journal and the
1940 	 * new journal write might overwrite whatever was in the journal
1941 	 * previously - we can't leave the journal without any flush writes in
1942 	 * it.
1943 	 *
1944 	 * So if we're in an error state, and we're still starting up, we don't
1945 	 * write anything at all.
1946 	 */
1947 	if (error && test_bit(JOURNAL_need_flush_write, &j->flags))
1948 		return -EIO;
1949 
1950 	if (error ||
1951 	    w->noflush ||
1952 	    (!w->must_flush &&
1953 	     (jiffies - j->last_flush_write) < msecs_to_jiffies(c->opts.journal_flush_delay) &&
1954 	     test_bit(JOURNAL_may_skip_flush, &j->flags))) {
1955 		w->noflush = true;
1956 		SET_JSET_NO_FLUSH(w->data, true);
1957 		w->data->last_seq	= 0;
1958 		w->last_seq		= 0;
1959 
1960 		j->nr_noflush_writes++;
1961 	} else {
1962 		w->must_flush = true;
1963 		j->last_flush_write = jiffies;
1964 		j->nr_flush_writes++;
1965 		clear_bit(JOURNAL_need_flush_write, &j->flags);
1966 	}
1967 
1968 	return 0;
1969 }
1970 
1971 CLOSURE_CALLBACK(bch2_journal_write)
1972 {
1973 	closure_type(w, struct journal_buf, io);
1974 	struct journal *j = container_of(w, struct journal, buf[w->idx]);
1975 	struct bch_fs *c = container_of(j, struct bch_fs, journal);
1976 	struct bch_replicas_padded replicas;
1977 	unsigned nr_rw_members = 0;
1978 	int ret;
1979 
1980 	for_each_rw_member(c, ca)
1981 		nr_rw_members++;
1982 
1983 	BUG_ON(BCH_SB_CLEAN(c->disk_sb.sb));
1984 	BUG_ON(!w->write_started);
1985 	BUG_ON(w->write_allocated);
1986 	BUG_ON(w->write_done);
1987 
1988 	j->write_start_time = local_clock();
1989 
1990 	spin_lock(&j->lock);
1991 	if (nr_rw_members > 1)
1992 		w->separate_flush = true;
1993 
1994 	ret = bch2_journal_write_pick_flush(j, w);
1995 	spin_unlock(&j->lock);
1996 	if (ret)
1997 		goto err;
1998 
1999 	mutex_lock(&j->buf_lock);
2000 	journal_buf_realloc(j, w);
2001 
2002 	ret = bch2_journal_write_prep(j, w);
2003 	mutex_unlock(&j->buf_lock);
2004 	if (ret)
2005 		goto err;
2006 
2007 	j->entry_bytes_written += vstruct_bytes(w->data);
2008 
2009 	while (1) {
2010 		spin_lock(&j->lock);
2011 		ret = journal_write_alloc(j, w);
2012 		if (!ret || !j->can_discard)
2013 			break;
2014 
2015 		spin_unlock(&j->lock);
2016 		bch2_journal_do_discards(j);
2017 	}
2018 
2019 	if (ret) {
2020 		struct printbuf buf = PRINTBUF;
2021 		buf.atomic++;
2022 
2023 		prt_printf(&buf, bch2_fmt(c, "Unable to allocate journal write: %s"),
2024 			   bch2_err_str(ret));
2025 		__bch2_journal_debug_to_text(&buf, j);
2026 		spin_unlock(&j->lock);
2027 		bch2_print_string_as_lines(KERN_ERR, buf.buf);
2028 		printbuf_exit(&buf);
2029 		goto err;
2030 	}
2031 
2032 	/*
2033 	 * write is allocated, no longer need to account for it in
2034 	 * bch2_journal_space_available():
2035 	 */
2036 	w->sectors = 0;
2037 	w->write_allocated = true;
2038 
2039 	/*
2040 	 * journal entry has been compacted and allocated, recalculate space
2041 	 * available:
2042 	 */
2043 	bch2_journal_space_available(j);
2044 	bch2_journal_do_writes(j);
2045 	spin_unlock(&j->lock);
2046 
2047 	w->devs_written = bch2_bkey_devs(bkey_i_to_s_c(&w->key));
2048 
2049 	if (c->opts.nochanges)
2050 		goto no_io;
2051 
2052 	/*
2053 	 * Mark journal replicas before we submit the write to guarantee
2054 	 * recovery will find the journal entries after a crash.
2055 	 */
2056 	bch2_devlist_to_replicas(&replicas.e, BCH_DATA_journal,
2057 				 w->devs_written);
2058 	ret = bch2_mark_replicas(c, &replicas.e);
2059 	if (ret)
2060 		goto err;
2061 
2062 	if (!JSET_NO_FLUSH(w->data))
2063 		continue_at(cl, journal_write_preflush, j->wq);
2064 	else
2065 		continue_at(cl, journal_write_submit, j->wq);
2066 	return;
2067 no_io:
2068 	continue_at(cl, journal_write_done, j->wq);
2069 	return;
2070 err:
2071 	bch2_fatal_error(c);
2072 	continue_at(cl, journal_write_done, j->wq);
2073 }
2074