1 // SPDX-License-Identifier: GPL-2.0
2
3 #include "bcachefs.h"
4 #include "alloc_foreground.h"
5 #include "btree_gc.h"
6 #include "btree_io.h"
7 #include "btree_iter.h"
8 #include "btree_journal_iter.h"
9 #include "btree_key_cache.h"
10 #include "btree_update_interior.h"
11 #include "btree_write_buffer.h"
12 #include "buckets.h"
13 #include "disk_accounting.h"
14 #include "enumerated_ref.h"
15 #include "errcode.h"
16 #include "error.h"
17 #include "journal.h"
18 #include "journal_io.h"
19 #include "journal_reclaim.h"
20 #include "replicas.h"
21 #include "snapshot.h"
22
23 #include <linux/prefetch.h>
24 #include <linux/string_helpers.h>
25
26 static const char * const trans_commit_flags_strs[] = {
27 #define x(n, ...) #n,
28 BCH_TRANS_COMMIT_FLAGS()
29 #undef x
30 NULL
31 };
32
bch2_trans_commit_flags_to_text(struct printbuf * out,enum bch_trans_commit_flags flags)33 void bch2_trans_commit_flags_to_text(struct printbuf *out, enum bch_trans_commit_flags flags)
34 {
35 enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
36
37 prt_printf(out, "watermark=%s", bch2_watermarks[watermark]);
38
39 flags >>= BCH_WATERMARK_BITS;
40 if (flags) {
41 prt_char(out, ' ');
42 bch2_prt_bitflags(out, trans_commit_flags_strs, flags);
43 }
44 }
45
verify_update_old_key(struct btree_trans * trans,struct btree_insert_entry * i)46 static void verify_update_old_key(struct btree_trans *trans, struct btree_insert_entry *i)
47 {
48 #ifdef CONFIG_BCACHEFS_DEBUG
49 struct bch_fs *c = trans->c;
50 struct bkey u;
51 struct bkey_s_c k = bch2_btree_path_peek_slot_exact(trans->paths + i->path, &u);
52
53 if (unlikely(trans->journal_replay_not_finished)) {
54 struct bkey_i *j_k =
55 bch2_journal_keys_peek_slot(c, i->btree_id, i->level, i->k->k.p);
56
57 if (j_k)
58 k = bkey_i_to_s_c(j_k);
59 }
60
61 u = *k.k;
62 u.needs_whiteout = i->old_k.needs_whiteout;
63
64 BUG_ON(memcmp(&i->old_k, &u, sizeof(struct bkey)));
65 BUG_ON(i->old_v != k.v);
66 #endif
67 }
68
insert_l(struct btree_trans * trans,struct btree_insert_entry * i)69 static inline struct btree_path_level *insert_l(struct btree_trans *trans, struct btree_insert_entry *i)
70 {
71 return (trans->paths + i->path)->l + i->level;
72 }
73
same_leaf_as_prev(struct btree_trans * trans,struct btree_insert_entry * i)74 static inline bool same_leaf_as_prev(struct btree_trans *trans,
75 struct btree_insert_entry *i)
76 {
77 return i != trans->updates &&
78 insert_l(trans, &i[0])->b == insert_l(trans, &i[-1])->b;
79 }
80
same_leaf_as_next(struct btree_trans * trans,struct btree_insert_entry * i)81 static inline bool same_leaf_as_next(struct btree_trans *trans,
82 struct btree_insert_entry *i)
83 {
84 return i + 1 < trans->updates + trans->nr_updates &&
85 insert_l(trans, &i[0])->b == insert_l(trans, &i[1])->b;
86 }
87
bch2_btree_node_prep_for_write(struct btree_trans * trans,struct btree_path * path,struct btree * b)88 inline void bch2_btree_node_prep_for_write(struct btree_trans *trans,
89 struct btree_path *path,
90 struct btree *b)
91 {
92 struct bch_fs *c = trans->c;
93
94 if (unlikely(btree_node_just_written(b)) &&
95 bch2_btree_post_write_cleanup(c, b))
96 bch2_trans_node_reinit_iter(trans, b);
97
98 /*
99 * If the last bset has been written, or if it's gotten too big - start
100 * a new bset to insert into:
101 */
102 if (want_new_bset(c, b))
103 bch2_btree_init_next(trans, b);
104 }
105
trans_lock_write_fail(struct btree_trans * trans,struct btree_insert_entry * i)106 static noinline int trans_lock_write_fail(struct btree_trans *trans, struct btree_insert_entry *i)
107 {
108 while (--i >= trans->updates) {
109 if (same_leaf_as_prev(trans, i))
110 continue;
111
112 bch2_btree_node_unlock_write(trans, trans->paths + i->path, insert_l(trans, i)->b);
113 }
114
115 trace_and_count(trans->c, trans_restart_would_deadlock_write, trans);
116 return btree_trans_restart(trans, BCH_ERR_transaction_restart_would_deadlock_write);
117 }
118
bch2_trans_lock_write(struct btree_trans * trans)119 static inline int bch2_trans_lock_write(struct btree_trans *trans)
120 {
121 EBUG_ON(trans->write_locked);
122
123 trans_for_each_update(trans, i) {
124 if (same_leaf_as_prev(trans, i))
125 continue;
126
127 if (bch2_btree_node_lock_write(trans, trans->paths + i->path, &insert_l(trans, i)->b->c))
128 return trans_lock_write_fail(trans, i);
129
130 if (!i->cached)
131 bch2_btree_node_prep_for_write(trans, trans->paths + i->path, insert_l(trans, i)->b);
132 }
133
134 trans->write_locked = true;
135 return 0;
136 }
137
bch2_trans_unlock_updates_write(struct btree_trans * trans)138 static inline void bch2_trans_unlock_updates_write(struct btree_trans *trans)
139 {
140 if (likely(trans->write_locked)) {
141 trans_for_each_update(trans, i)
142 if (btree_node_locked_type(trans->paths + i->path, i->level) ==
143 BTREE_NODE_WRITE_LOCKED)
144 bch2_btree_node_unlock_write_inlined(trans,
145 trans->paths + i->path, insert_l(trans, i)->b);
146 trans->write_locked = false;
147 }
148 }
149
150 /* Inserting into a given leaf node (last stage of insert): */
151
152 /* Handle overwrites and do insert, for non extents: */
bch2_btree_bset_insert_key(struct btree_trans * trans,struct btree_path * path,struct btree * b,struct btree_node_iter * node_iter,struct bkey_i * insert)153 bool bch2_btree_bset_insert_key(struct btree_trans *trans,
154 struct btree_path *path,
155 struct btree *b,
156 struct btree_node_iter *node_iter,
157 struct bkey_i *insert)
158 {
159 struct bkey_packed *k;
160 unsigned clobber_u64s = 0, new_u64s = 0;
161
162 EBUG_ON(btree_node_just_written(b));
163 EBUG_ON(bset_written(b, btree_bset_last(b)));
164 EBUG_ON(bkey_deleted(&insert->k) && bkey_val_u64s(&insert->k));
165 EBUG_ON(bpos_lt(insert->k.p, b->data->min_key));
166 EBUG_ON(bpos_gt(insert->k.p, b->data->max_key));
167 EBUG_ON(insert->k.u64s > bch2_btree_keys_u64s_remaining(b));
168 EBUG_ON(!b->c.level && !bpos_eq(insert->k.p, path->pos));
169 kmsan_check_memory(insert, bkey_bytes(&insert->k));
170
171 k = bch2_btree_node_iter_peek_all(node_iter, b);
172 if (k && bkey_cmp_left_packed(b, k, &insert->k.p))
173 k = NULL;
174
175 /* @k is the key being overwritten/deleted, if any: */
176 EBUG_ON(k && bkey_deleted(k));
177
178 /* Deleting, but not found? nothing to do: */
179 if (bkey_deleted(&insert->k) && !k)
180 return false;
181
182 if (bkey_deleted(&insert->k)) {
183 /* Deleting: */
184 btree_account_key_drop(b, k);
185 k->type = KEY_TYPE_deleted;
186
187 if (k->needs_whiteout)
188 push_whiteout(b, insert->k.p);
189 k->needs_whiteout = false;
190
191 if (k >= btree_bset_last(b)->start) {
192 clobber_u64s = k->u64s;
193 bch2_bset_delete(b, k, clobber_u64s);
194 goto fix_iter;
195 } else {
196 bch2_btree_path_fix_key_modified(trans, b, k);
197 }
198
199 return true;
200 }
201
202 if (k) {
203 /* Overwriting: */
204 btree_account_key_drop(b, k);
205 k->type = KEY_TYPE_deleted;
206
207 insert->k.needs_whiteout = k->needs_whiteout;
208 k->needs_whiteout = false;
209
210 if (k >= btree_bset_last(b)->start) {
211 clobber_u64s = k->u64s;
212 goto overwrite;
213 } else {
214 bch2_btree_path_fix_key_modified(trans, b, k);
215 }
216 }
217
218 k = bch2_btree_node_iter_bset_pos(node_iter, b, bset_tree_last(b));
219 overwrite:
220 bch2_bset_insert(b, k, insert, clobber_u64s);
221 new_u64s = k->u64s;
222 fix_iter:
223 if (clobber_u64s != new_u64s)
224 bch2_btree_node_iter_fix(trans, path, b, node_iter, k,
225 clobber_u64s, new_u64s);
226 return true;
227 }
228
__btree_node_flush(struct journal * j,struct journal_entry_pin * pin,unsigned i,u64 seq)229 static int __btree_node_flush(struct journal *j, struct journal_entry_pin *pin,
230 unsigned i, u64 seq)
231 {
232 struct bch_fs *c = container_of(j, struct bch_fs, journal);
233 struct btree_write *w = container_of(pin, struct btree_write, journal);
234 struct btree *b = container_of(w, struct btree, writes[i]);
235 struct btree_trans *trans = bch2_trans_get(c);
236 unsigned long old, new;
237 unsigned idx = w - b->writes;
238
239 btree_node_lock_nopath_nofail(trans, &b->c, SIX_LOCK_read);
240
241 old = READ_ONCE(b->flags);
242 do {
243 new = old;
244
245 if (!(old & (1 << BTREE_NODE_dirty)) ||
246 !!(old & (1 << BTREE_NODE_write_idx)) != idx ||
247 w->journal.seq != seq)
248 break;
249
250 new &= ~BTREE_WRITE_TYPE_MASK;
251 new |= BTREE_WRITE_journal_reclaim;
252 new |= 1 << BTREE_NODE_need_write;
253 } while (!try_cmpxchg(&b->flags, &old, new));
254
255 btree_node_write_if_need(trans, b, SIX_LOCK_read);
256 six_unlock_read(&b->c.lock);
257
258 bch2_trans_put(trans);
259 return 0;
260 }
261
bch2_btree_node_flush0(struct journal * j,struct journal_entry_pin * pin,u64 seq)262 int bch2_btree_node_flush0(struct journal *j, struct journal_entry_pin *pin, u64 seq)
263 {
264 return __btree_node_flush(j, pin, 0, seq);
265 }
266
bch2_btree_node_flush1(struct journal * j,struct journal_entry_pin * pin,u64 seq)267 int bch2_btree_node_flush1(struct journal *j, struct journal_entry_pin *pin, u64 seq)
268 {
269 return __btree_node_flush(j, pin, 1, seq);
270 }
271
bch2_btree_add_journal_pin(struct bch_fs * c,struct btree * b,u64 seq)272 inline void bch2_btree_add_journal_pin(struct bch_fs *c,
273 struct btree *b, u64 seq)
274 {
275 struct btree_write *w = btree_current_write(b);
276
277 bch2_journal_pin_add(&c->journal, seq, &w->journal,
278 btree_node_write_idx(b) == 0
279 ? bch2_btree_node_flush0
280 : bch2_btree_node_flush1);
281 }
282
283 /**
284 * bch2_btree_insert_key_leaf() - insert a key one key into a leaf node
285 * @trans: btree transaction object
286 * @path: path pointing to @insert's pos
287 * @insert: key to insert
288 * @journal_seq: sequence number of journal reservation
289 */
bch2_btree_insert_key_leaf(struct btree_trans * trans,struct btree_path * path,struct bkey_i * insert,u64 journal_seq)290 inline void bch2_btree_insert_key_leaf(struct btree_trans *trans,
291 struct btree_path *path,
292 struct bkey_i *insert,
293 u64 journal_seq)
294 {
295 struct bch_fs *c = trans->c;
296 struct btree *b = path_l(path)->b;
297 struct bset_tree *t = bset_tree_last(b);
298 struct bset *i = bset(b, t);
299 int old_u64s = bset_u64s(t);
300 int old_live_u64s = b->nr.live_u64s;
301 int live_u64s_added, u64s_added;
302
303 if (unlikely(!bch2_btree_bset_insert_key(trans, path, b,
304 &path_l(path)->iter, insert)))
305 return;
306
307 i->journal_seq = cpu_to_le64(max(journal_seq, le64_to_cpu(i->journal_seq)));
308
309 bch2_btree_add_journal_pin(c, b, journal_seq);
310
311 if (unlikely(!btree_node_dirty(b))) {
312 EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags));
313 set_btree_node_dirty_acct(c, b);
314 }
315
316 live_u64s_added = (int) b->nr.live_u64s - old_live_u64s;
317 u64s_added = (int) bset_u64s(t) - old_u64s;
318
319 if (b->sib_u64s[0] != U16_MAX && live_u64s_added < 0)
320 b->sib_u64s[0] = max(0, (int) b->sib_u64s[0] + live_u64s_added);
321 if (b->sib_u64s[1] != U16_MAX && live_u64s_added < 0)
322 b->sib_u64s[1] = max(0, (int) b->sib_u64s[1] + live_u64s_added);
323
324 if (u64s_added > live_u64s_added &&
325 bch2_maybe_compact_whiteouts(c, b))
326 bch2_trans_node_reinit_iter(trans, b);
327 }
328
329 /* Cached btree updates: */
330
331 /* Normal update interface: */
332
btree_insert_entry_checks(struct btree_trans * trans,struct btree_insert_entry * i)333 static inline void btree_insert_entry_checks(struct btree_trans *trans,
334 struct btree_insert_entry *i)
335 {
336 struct btree_path *path = trans->paths + i->path;
337
338 BUG_ON(!bpos_eq(i->k->k.p, path->pos));
339 BUG_ON(i->cached != path->cached);
340 BUG_ON(i->level != path->level);
341 BUG_ON(i->btree_id != path->btree_id);
342 BUG_ON(i->bkey_type != __btree_node_type(path->level, path->btree_id));
343 EBUG_ON(!i->level &&
344 btree_type_has_snapshots(i->btree_id) &&
345 !(i->flags & BTREE_UPDATE_internal_snapshot_node) &&
346 test_bit(JOURNAL_replay_done, &trans->c->journal.flags) &&
347 i->k->k.p.snapshot &&
348 bch2_snapshot_is_internal_node(trans->c, i->k->k.p.snapshot) > 0);
349 }
350
bch2_trans_journal_res_get(struct btree_trans * trans,unsigned flags)351 static __always_inline int bch2_trans_journal_res_get(struct btree_trans *trans,
352 unsigned flags)
353 {
354 return bch2_journal_res_get(&trans->c->journal, &trans->journal_res,
355 trans->journal_u64s, flags, trans);
356 }
357
358 #define JSET_ENTRY_LOG_U64s 4
359
journal_transaction_name(struct btree_trans * trans)360 static noinline void journal_transaction_name(struct btree_trans *trans)
361 {
362 struct bch_fs *c = trans->c;
363 struct journal *j = &c->journal;
364 struct jset_entry *entry =
365 bch2_journal_add_entry(j, &trans->journal_res,
366 BCH_JSET_ENTRY_log, 0, 0,
367 JSET_ENTRY_LOG_U64s);
368 struct jset_entry_log *l =
369 container_of(entry, struct jset_entry_log, entry);
370
371 memcpy_and_pad(l->d, JSET_ENTRY_LOG_U64s * sizeof(u64),
372 trans->fn, strlen(trans->fn), 0);
373 }
374
btree_key_can_insert(struct btree_trans * trans,struct btree * b,unsigned u64s)375 static inline int btree_key_can_insert(struct btree_trans *trans,
376 struct btree *b, unsigned u64s)
377 {
378 if (!bch2_btree_node_insert_fits(b, u64s))
379 return bch_err_throw(trans->c, btree_insert_btree_node_full);
380
381 return 0;
382 }
383
384 noinline static int
btree_key_can_insert_cached_slowpath(struct btree_trans * trans,unsigned flags,struct btree_path * path,unsigned new_u64s)385 btree_key_can_insert_cached_slowpath(struct btree_trans *trans, unsigned flags,
386 struct btree_path *path, unsigned new_u64s)
387 {
388 struct bkey_cached *ck = (void *) path->l[0].b;
389 struct bkey_i *new_k;
390 int ret;
391
392 bch2_trans_unlock_updates_write(trans);
393 bch2_trans_unlock(trans);
394
395 new_k = kmalloc(new_u64s * sizeof(u64), GFP_KERNEL);
396 if (!new_k) {
397 struct bch_fs *c = trans->c;
398 bch_err(c, "error allocating memory for key cache key, btree %s u64s %u",
399 bch2_btree_id_str(path->btree_id), new_u64s);
400 return bch_err_throw(c, ENOMEM_btree_key_cache_insert);
401 }
402
403 ret = bch2_trans_relock(trans) ?:
404 bch2_trans_lock_write(trans);
405 if (unlikely(ret)) {
406 kfree(new_k);
407 return ret;
408 }
409
410 memcpy(new_k, ck->k, ck->u64s * sizeof(u64));
411
412 trans_for_each_update(trans, i)
413 if (i->old_v == &ck->k->v)
414 i->old_v = &new_k->v;
415
416 kfree(ck->k);
417 ck->u64s = new_u64s;
418 ck->k = new_k;
419 return 0;
420 }
421
btree_key_can_insert_cached(struct btree_trans * trans,unsigned flags,struct btree_path * path,unsigned u64s)422 static int btree_key_can_insert_cached(struct btree_trans *trans, unsigned flags,
423 struct btree_path *path, unsigned u64s)
424 {
425 struct bch_fs *c = trans->c;
426 struct bkey_cached *ck = (void *) path->l[0].b;
427 unsigned new_u64s;
428 struct bkey_i *new_k;
429 unsigned watermark = flags & BCH_WATERMARK_MASK;
430
431 EBUG_ON(path->level);
432
433 if (watermark < BCH_WATERMARK_reclaim &&
434 !test_bit(BKEY_CACHED_DIRTY, &ck->flags) &&
435 bch2_btree_key_cache_must_wait(c))
436 return bch_err_throw(c, btree_insert_need_journal_reclaim);
437
438 /*
439 * bch2_varint_decode can read past the end of the buffer by at most 7
440 * bytes (it won't be used):
441 */
442 u64s += 1;
443
444 if (u64s <= ck->u64s)
445 return 0;
446
447 new_u64s = roundup_pow_of_two(u64s);
448 new_k = krealloc(ck->k, new_u64s * sizeof(u64), GFP_NOWAIT|__GFP_NOWARN);
449 if (unlikely(!new_k))
450 return btree_key_can_insert_cached_slowpath(trans, flags, path, new_u64s);
451
452 trans_for_each_update(trans, i)
453 if (i->old_v == &ck->k->v)
454 i->old_v = &new_k->v;
455
456 ck->u64s = new_u64s;
457 ck->k = new_k;
458 return 0;
459 }
460
461 /* Triggers: */
462
run_one_mem_trigger(struct btree_trans * trans,struct btree_insert_entry * i,unsigned flags)463 static int run_one_mem_trigger(struct btree_trans *trans,
464 struct btree_insert_entry *i,
465 unsigned flags)
466 {
467 verify_update_old_key(trans, i);
468
469 if (unlikely(flags & BTREE_TRIGGER_norun))
470 return 0;
471
472 struct bkey_s_c old = { &i->old_k, i->old_v };
473 struct bkey_i *new = i->k;
474 const struct bkey_ops *old_ops = bch2_bkey_type_ops(old.k->type);
475 const struct bkey_ops *new_ops = bch2_bkey_type_ops(i->k->k.type);
476
477 if (old_ops->trigger == new_ops->trigger)
478 return bch2_key_trigger(trans, i->btree_id, i->level,
479 old, bkey_i_to_s(new),
480 BTREE_TRIGGER_insert|BTREE_TRIGGER_overwrite|flags);
481 else
482 return bch2_key_trigger_new(trans, i->btree_id, i->level,
483 bkey_i_to_s(new), flags) ?:
484 bch2_key_trigger_old(trans, i->btree_id, i->level,
485 old, flags);
486 }
487
run_one_trans_trigger(struct btree_trans * trans,struct btree_insert_entry * i)488 static int run_one_trans_trigger(struct btree_trans *trans, struct btree_insert_entry *i)
489 {
490 verify_update_old_key(trans, i);
491
492 if ((i->flags & BTREE_TRIGGER_norun) ||
493 !btree_node_type_has_trans_triggers(i->bkey_type))
494 return 0;
495
496 /*
497 * Transactional triggers create new btree_insert_entries, so we can't
498 * pass them a pointer to a btree_insert_entry, that memory is going to
499 * move:
500 */
501 struct bkey old_k = i->old_k;
502 struct bkey_s_c old = { &old_k, i->old_v };
503 const struct bkey_ops *old_ops = bch2_bkey_type_ops(old.k->type);
504 const struct bkey_ops *new_ops = bch2_bkey_type_ops(i->k->k.type);
505 unsigned flags = i->flags|BTREE_TRIGGER_transactional;
506
507 if (!i->insert_trigger_run &&
508 !i->overwrite_trigger_run &&
509 old_ops->trigger == new_ops->trigger) {
510 i->overwrite_trigger_run = true;
511 i->insert_trigger_run = true;
512 return bch2_key_trigger(trans, i->btree_id, i->level, old, bkey_i_to_s(i->k),
513 BTREE_TRIGGER_insert|
514 BTREE_TRIGGER_overwrite|flags) ?: 1;
515 } else if (!i->overwrite_trigger_run) {
516 i->overwrite_trigger_run = true;
517 return bch2_key_trigger_old(trans, i->btree_id, i->level, old, flags) ?: 1;
518 } else if (!i->insert_trigger_run) {
519 i->insert_trigger_run = true;
520 return bch2_key_trigger_new(trans, i->btree_id, i->level, bkey_i_to_s(i->k), flags) ?: 1;
521 } else {
522 return 0;
523 }
524 }
525
bch2_trans_commit_run_triggers(struct btree_trans * trans)526 static int bch2_trans_commit_run_triggers(struct btree_trans *trans)
527 {
528 unsigned sort_id_start = 0;
529
530 while (sort_id_start < trans->nr_updates) {
531 unsigned i, sort_id = trans->updates[sort_id_start].sort_order;
532 bool trans_trigger_run;
533
534 /*
535 * For a given btree, this algorithm runs insert triggers before
536 * overwrite triggers: this is so that when extents are being
537 * moved (e.g. by FALLOCATE_FL_INSERT_RANGE), we don't drop
538 * references before they are re-added.
539 *
540 * Running triggers will append more updates to the list of
541 * updates as we're walking it:
542 */
543 do {
544 trans_trigger_run = false;
545
546 for (i = sort_id_start;
547 i < trans->nr_updates && trans->updates[i].sort_order <= sort_id;
548 i++) {
549 if (trans->updates[i].sort_order < sort_id) {
550 sort_id_start = i;
551 continue;
552 }
553
554 int ret = run_one_trans_trigger(trans, trans->updates + i);
555 if (ret < 0)
556 return ret;
557 if (ret)
558 trans_trigger_run = true;
559 }
560 } while (trans_trigger_run);
561
562 sort_id_start = i;
563 }
564
565 #ifdef CONFIG_BCACHEFS_DEBUG
566 trans_for_each_update(trans, i)
567 BUG_ON(!(i->flags & BTREE_TRIGGER_norun) &&
568 btree_node_type_has_trans_triggers(i->bkey_type) &&
569 (!i->insert_trigger_run || !i->overwrite_trigger_run));
570 #endif
571 return 0;
572 }
573
bch2_trans_commit_run_gc_triggers(struct btree_trans * trans)574 static noinline int bch2_trans_commit_run_gc_triggers(struct btree_trans *trans)
575 {
576 trans_for_each_update(trans, i)
577 if (btree_node_type_has_triggers(i->bkey_type) &&
578 gc_visited(trans->c, gc_pos_btree(i->btree_id, i->level, i->k->k.p))) {
579 int ret = run_one_mem_trigger(trans, i, i->flags|BTREE_TRIGGER_gc);
580 if (ret)
581 return ret;
582 }
583
584 return 0;
585 }
586
587 static inline int
bch2_trans_commit_write_locked(struct btree_trans * trans,unsigned flags,struct btree_insert_entry ** stopped_at,unsigned long trace_ip)588 bch2_trans_commit_write_locked(struct btree_trans *trans, unsigned flags,
589 struct btree_insert_entry **stopped_at,
590 unsigned long trace_ip)
591 {
592 struct bch_fs *c = trans->c;
593 struct btree_trans_commit_hook *h;
594 unsigned u64s = 0;
595 int ret = 0;
596
597 bch2_trans_verify_not_unlocked_or_in_restart(trans);
598 #if 0
599 /* todo: bring back dynamic fault injection */
600 if (race_fault()) {
601 trace_and_count(c, trans_restart_fault_inject, trans, trace_ip);
602 return btree_trans_restart(trans, BCH_ERR_transaction_restart_fault_inject);
603 }
604 #endif
605 /*
606 * Check if the insert will fit in the leaf node with the write lock
607 * held, otherwise another thread could write the node changing the
608 * amount of space available:
609 */
610
611 prefetch(&trans->c->journal.flags);
612
613 trans_for_each_update(trans, i) {
614 /* Multiple inserts might go to same leaf: */
615 if (!same_leaf_as_prev(trans, i))
616 u64s = 0;
617
618 u64s += i->k->k.u64s;
619 ret = !i->cached
620 ? btree_key_can_insert(trans, insert_l(trans, i)->b, u64s)
621 : btree_key_can_insert_cached(trans, flags, trans->paths + i->path, u64s);
622 if (ret) {
623 *stopped_at = i;
624 return ret;
625 }
626
627 i->k->k.needs_whiteout = false;
628 }
629
630 /*
631 * Don't get journal reservation until after we know insert will
632 * succeed:
633 */
634 if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res))) {
635 ret = bch2_trans_journal_res_get(trans,
636 (flags & BCH_WATERMARK_MASK)|
637 JOURNAL_RES_GET_NONBLOCK);
638 if (ret)
639 return ret;
640
641 if (unlikely(trans->journal_transaction_names))
642 journal_transaction_name(trans);
643 }
644
645 /*
646 * Not allowed to fail after we've gotten our journal reservation - we
647 * have to use it:
648 */
649
650 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) &&
651 !(flags & BCH_TRANS_COMMIT_no_journal_res)) {
652 if (static_branch_unlikely(&bch2_journal_seq_verify))
653 trans_for_each_update(trans, i)
654 i->k->k.bversion.lo = trans->journal_res.seq;
655 else if (static_branch_unlikely(&bch2_inject_invalid_keys))
656 trans_for_each_update(trans, i)
657 i->k->k.bversion = MAX_VERSION;
658 }
659
660 h = trans->hooks;
661 while (h) {
662 ret = h->fn(trans, h);
663 if (ret)
664 return ret;
665 h = h->next;
666 }
667
668 struct bkey_i *accounting;
669
670 percpu_down_read(&c->mark_lock);
671 for (accounting = btree_trans_subbuf_base(trans, &trans->accounting);
672 accounting != btree_trans_subbuf_top(trans, &trans->accounting);
673 accounting = bkey_next(accounting)) {
674 ret = bch2_accounting_trans_commit_hook(trans,
675 bkey_i_to_accounting(accounting), flags);
676 if (ret)
677 goto revert_fs_usage;
678 }
679 percpu_up_read(&c->mark_lock);
680
681 /* XXX: we only want to run this if deltas are nonzero */
682 bch2_trans_account_disk_usage_change(trans);
683
684 trans_for_each_update(trans, i)
685 if (btree_node_type_has_atomic_triggers(i->bkey_type)) {
686 ret = run_one_mem_trigger(trans, i, BTREE_TRIGGER_atomic|i->flags);
687 if (ret)
688 goto fatal_err;
689 }
690
691 if (unlikely(c->gc_pos.phase)) {
692 ret = bch2_trans_commit_run_gc_triggers(trans);
693 if (ret)
694 goto fatal_err;
695 }
696
697 struct bkey_validate_context validate_context = { .from = BKEY_VALIDATE_commit };
698
699 if (!(flags & BCH_TRANS_COMMIT_no_journal_res))
700 validate_context.flags = BCH_VALIDATE_write|BCH_VALIDATE_commit;
701
702 for (struct jset_entry *i = btree_trans_journal_entries_start(trans);
703 i != btree_trans_journal_entries_top(trans);
704 i = vstruct_next(i)) {
705 ret = bch2_journal_entry_validate(c, NULL, i,
706 bcachefs_metadata_version_current,
707 CPU_BIG_ENDIAN, validate_context);
708 if (unlikely(ret)) {
709 bch2_trans_inconsistent(trans, "invalid journal entry on insert from %s\n",
710 trans->fn);
711 goto fatal_err;
712 }
713 }
714
715 trans_for_each_update(trans, i) {
716 validate_context.level = i->level;
717 validate_context.btree = i->btree_id;
718
719 ret = bch2_bkey_validate(c, bkey_i_to_s_c(i->k), validate_context);
720 if (unlikely(ret)){
721 bch2_trans_inconsistent(trans, "invalid bkey on insert from %s -> %ps\n",
722 trans->fn, (void *) i->ip_allocated);
723 goto fatal_err;
724 }
725 btree_insert_entry_checks(trans, i);
726 }
727
728 if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res))) {
729 struct journal *j = &c->journal;
730 struct jset_entry *entry;
731
732 trans_for_each_update(trans, i) {
733 if (i->key_cache_already_flushed)
734 continue;
735
736 if (i->flags & BTREE_UPDATE_nojournal)
737 continue;
738
739 verify_update_old_key(trans, i);
740
741 if (trans->journal_transaction_names) {
742 entry = bch2_journal_add_entry(j, &trans->journal_res,
743 BCH_JSET_ENTRY_overwrite,
744 i->btree_id, i->level,
745 i->old_k.u64s);
746 bkey_reassemble((struct bkey_i *) entry->start,
747 (struct bkey_s_c) { &i->old_k, i->old_v });
748 }
749
750 entry = bch2_journal_add_entry(j, &trans->journal_res,
751 BCH_JSET_ENTRY_btree_keys,
752 i->btree_id, i->level,
753 i->k->k.u64s);
754 bkey_copy((struct bkey_i *) entry->start, i->k);
755 }
756
757 memcpy_u64s_small(journal_res_entry(&c->journal, &trans->journal_res),
758 btree_trans_journal_entries_start(trans),
759 trans->journal_entries.u64s);
760
761 EBUG_ON(trans->journal_res.u64s < trans->journal_entries.u64s);
762
763 trans->journal_res.offset += trans->journal_entries.u64s;
764 trans->journal_res.u64s -= trans->journal_entries.u64s;
765
766 memcpy_u64s_small(bch2_journal_add_entry(j, &trans->journal_res,
767 BCH_JSET_ENTRY_write_buffer_keys,
768 BTREE_ID_accounting, 0,
769 trans->accounting.u64s)->_data,
770 btree_trans_subbuf_base(trans, &trans->accounting),
771 trans->accounting.u64s);
772
773 if (trans->journal_seq)
774 *trans->journal_seq = trans->journal_res.seq;
775 }
776
777 trans_for_each_update(trans, i) {
778 struct btree_path *path = trans->paths + i->path;
779
780 if (!i->cached)
781 bch2_btree_insert_key_leaf(trans, path, i->k, trans->journal_res.seq);
782 else if (!i->key_cache_already_flushed)
783 bch2_btree_insert_key_cached(trans, flags, i);
784 else
785 bch2_btree_key_cache_drop(trans, path);
786 }
787
788 return 0;
789 fatal_err:
790 bch2_fs_fatal_error(c, "fatal error in transaction commit: %s", bch2_err_str(ret));
791 percpu_down_read(&c->mark_lock);
792 revert_fs_usage:
793 for (struct bkey_i *i = btree_trans_subbuf_base(trans, &trans->accounting);
794 i != accounting;
795 i = bkey_next(i))
796 bch2_accounting_trans_commit_revert(trans, bkey_i_to_accounting(i), flags);
797 percpu_up_read(&c->mark_lock);
798 return ret;
799 }
800
bch2_drop_overwrites_from_journal(struct btree_trans * trans)801 static noinline void bch2_drop_overwrites_from_journal(struct btree_trans *trans)
802 {
803 /*
804 * Accounting keys aren't deduped in the journal: we have to compare
805 * each individual update against what's in the btree to see if it has
806 * been applied yet, and accounting updates also don't overwrite,
807 * they're deltas that accumulate.
808 */
809 trans_for_each_update(trans, i)
810 if (i->k->k.type != KEY_TYPE_accounting)
811 bch2_journal_key_overwritten(trans->c, i->btree_id, i->level, i->k->k.p);
812 }
813
bch2_trans_commit_journal_pin_flush(struct journal * j,struct journal_entry_pin * _pin,u64 seq)814 static int bch2_trans_commit_journal_pin_flush(struct journal *j,
815 struct journal_entry_pin *_pin, u64 seq)
816 {
817 return 0;
818 }
819
820 /*
821 * Get journal reservation, take write locks, and attempt to do btree update(s):
822 */
do_bch2_trans_commit(struct btree_trans * trans,unsigned flags,struct btree_insert_entry ** stopped_at,unsigned long trace_ip)823 static inline int do_bch2_trans_commit(struct btree_trans *trans, unsigned flags,
824 struct btree_insert_entry **stopped_at,
825 unsigned long trace_ip)
826 {
827 struct bch_fs *c = trans->c;
828 int ret = 0, u64s_delta = 0;
829
830 for (unsigned idx = 0; idx < trans->nr_updates; idx++) {
831 struct btree_insert_entry *i = trans->updates + idx;
832 if (i->cached)
833 continue;
834
835 u64s_delta += !bkey_deleted(&i->k->k) ? i->k->k.u64s : 0;
836 u64s_delta -= i->old_btree_u64s;
837
838 if (!same_leaf_as_next(trans, i)) {
839 if (u64s_delta <= 0) {
840 ret = bch2_foreground_maybe_merge(trans, i->path,
841 i->level, flags);
842 if (unlikely(ret))
843 return ret;
844 }
845
846 u64s_delta = 0;
847 }
848 }
849
850 ret = bch2_trans_lock_write(trans);
851 if (unlikely(ret))
852 return ret;
853
854 ret = bch2_trans_commit_write_locked(trans, flags, stopped_at, trace_ip);
855
856 if (!ret && unlikely(trans->journal_replay_not_finished))
857 bch2_drop_overwrites_from_journal(trans);
858
859 bch2_trans_unlock_updates_write(trans);
860
861 if (!ret && trans->journal_pin)
862 bch2_journal_pin_add(&c->journal, trans->journal_res.seq,
863 trans->journal_pin,
864 bch2_trans_commit_journal_pin_flush);
865
866 /*
867 * Drop journal reservation after dropping write locks, since dropping
868 * the journal reservation may kick off a journal write:
869 */
870 if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res)))
871 bch2_journal_res_put(&c->journal, &trans->journal_res);
872
873 return ret;
874 }
875
journal_reclaim_wait_done(struct bch_fs * c)876 static int journal_reclaim_wait_done(struct bch_fs *c)
877 {
878 int ret = bch2_journal_error(&c->journal) ?:
879 bch2_btree_key_cache_wait_done(c);
880
881 if (!ret)
882 journal_reclaim_kick(&c->journal);
883 return ret;
884 }
885
886 static noinline
bch2_trans_commit_error(struct btree_trans * trans,unsigned flags,struct btree_insert_entry * i,int ret,unsigned long trace_ip)887 int bch2_trans_commit_error(struct btree_trans *trans, unsigned flags,
888 struct btree_insert_entry *i,
889 int ret, unsigned long trace_ip)
890 {
891 struct bch_fs *c = trans->c;
892 enum bch_watermark watermark = flags & BCH_WATERMARK_MASK;
893
894 if (bch2_err_matches(ret, BCH_ERR_journal_res_blocked)) {
895 /*
896 * XXX: this should probably be a separate BTREE_INSERT_NONBLOCK
897 * flag
898 */
899 if ((flags & BCH_TRANS_COMMIT_journal_reclaim) &&
900 watermark < BCH_WATERMARK_reclaim) {
901 ret = bch_err_throw(c, journal_reclaim_would_deadlock);
902 goto out;
903 }
904
905 ret = drop_locks_do(trans,
906 bch2_trans_journal_res_get(trans,
907 (flags & BCH_WATERMARK_MASK)|
908 JOURNAL_RES_GET_CHECK));
909 goto out;
910 }
911
912 switch (ret) {
913 case -BCH_ERR_btree_insert_btree_node_full:
914 ret = bch2_btree_split_leaf(trans, i->path, flags);
915 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
916 trace_and_count(c, trans_restart_btree_node_split, trans,
917 trace_ip, trans->paths + i->path);
918 break;
919 case -BCH_ERR_btree_insert_need_mark_replicas:
920 ret = drop_locks_do(trans,
921 bch2_accounting_update_sb(trans));
922 break;
923 case -BCH_ERR_btree_insert_need_journal_reclaim:
924 bch2_trans_unlock(trans);
925
926 trace_and_count(c, trans_blocked_journal_reclaim, trans, trace_ip);
927 track_event_change(&c->times[BCH_TIME_blocked_key_cache_flush], true);
928
929 wait_event_freezable(c->journal.reclaim_wait,
930 (ret = journal_reclaim_wait_done(c)));
931
932 track_event_change(&c->times[BCH_TIME_blocked_key_cache_flush], false);
933
934 if (ret < 0)
935 break;
936
937 ret = bch2_trans_relock(trans);
938 break;
939 default:
940 BUG_ON(ret >= 0);
941 break;
942 }
943 out:
944 BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart) != !!trans->restarted);
945
946 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOSPC) &&
947 (flags & BCH_TRANS_COMMIT_no_enospc), c,
948 "%s: incorrectly got %s\n", __func__, bch2_err_str(ret));
949
950 return ret;
951 }
952
953 /*
954 * This is for updates done in the early part of fsck - btree_gc - before we've
955 * gone RW. we only add the new key to the list of keys for journal replay to
956 * do.
957 */
958 static noinline int
do_bch2_trans_commit_to_journal_replay(struct btree_trans * trans)959 do_bch2_trans_commit_to_journal_replay(struct btree_trans *trans)
960 {
961 struct bch_fs *c = trans->c;
962
963 BUG_ON(current != c->recovery_task);
964
965 trans_for_each_update(trans, i) {
966 int ret = bch2_journal_key_insert(c, i->btree_id, i->level, i->k);
967 if (ret)
968 return ret;
969 }
970
971 for (struct jset_entry *i = btree_trans_journal_entries_start(trans);
972 i != btree_trans_journal_entries_top(trans);
973 i = vstruct_next(i)) {
974 if (i->type == BCH_JSET_ENTRY_btree_keys ||
975 i->type == BCH_JSET_ENTRY_write_buffer_keys) {
976 jset_entry_for_each_key(i, k) {
977 int ret = bch2_journal_key_insert(c, i->btree_id, i->level, k);
978 if (ret)
979 return ret;
980 }
981 }
982
983 if (i->type == BCH_JSET_ENTRY_btree_root) {
984 guard(mutex)(&c->btree_root_lock);
985
986 struct btree_root *r = bch2_btree_id_root(c, i->btree_id);
987
988 bkey_copy(&r->key, i->start);
989 r->level = i->level;
990 r->alive = true;
991 }
992 }
993
994 for (struct bkey_i *i = btree_trans_subbuf_base(trans, &trans->accounting);
995 i != btree_trans_subbuf_top(trans, &trans->accounting);
996 i = bkey_next(i)) {
997 int ret = bch2_journal_key_insert(c, BTREE_ID_accounting, 0, i);
998 if (ret)
999 return ret;
1000 }
1001
1002 return 0;
1003 }
1004
__bch2_trans_commit(struct btree_trans * trans,unsigned flags)1005 int __bch2_trans_commit(struct btree_trans *trans, unsigned flags)
1006 {
1007 struct btree_insert_entry *errored_at = NULL;
1008 struct bch_fs *c = trans->c;
1009 unsigned journal_u64s = 0;
1010 int ret = 0;
1011
1012 bch2_trans_verify_not_unlocked_or_in_restart(trans);
1013
1014 ret = trans_maybe_inject_restart(trans, _RET_IP_);
1015 if (unlikely(ret))
1016 goto out_reset;
1017
1018 if (!trans->nr_updates &&
1019 !trans->journal_entries.u64s &&
1020 !trans->accounting.u64s)
1021 goto out_reset;
1022
1023 ret = bch2_trans_commit_run_triggers(trans);
1024 if (ret)
1025 goto out_reset;
1026
1027 if (!(flags & BCH_TRANS_COMMIT_no_check_rw) &&
1028 unlikely(!enumerated_ref_tryget(&c->writes, BCH_WRITE_REF_trans))) {
1029 if (unlikely(!test_bit(BCH_FS_may_go_rw, &c->flags)))
1030 ret = do_bch2_trans_commit_to_journal_replay(trans);
1031 else
1032 ret = bch_err_throw(c, erofs_trans_commit);
1033 goto out_reset;
1034 }
1035
1036 EBUG_ON(test_bit(BCH_FS_clean_shutdown, &c->flags));
1037
1038 journal_u64s = jset_u64s(trans->accounting.u64s);
1039 trans->journal_transaction_names = READ_ONCE(c->opts.journal_transaction_names);
1040 if (trans->journal_transaction_names)
1041 journal_u64s += jset_u64s(JSET_ENTRY_LOG_U64s);
1042
1043 trans_for_each_update(trans, i) {
1044 struct btree_path *path = trans->paths + i->path;
1045
1046 EBUG_ON(!path->should_be_locked);
1047
1048 ret = bch2_btree_path_upgrade(trans, path, i->level + 1);
1049 if (unlikely(ret))
1050 goto out;
1051
1052 EBUG_ON(!btree_node_intent_locked(path, i->level));
1053
1054 if (i->key_cache_already_flushed)
1055 continue;
1056
1057 if (i->flags & BTREE_UPDATE_nojournal)
1058 continue;
1059
1060 /* we're going to journal the key being updated: */
1061 journal_u64s += jset_u64s(i->k->k.u64s);
1062
1063 /* and we're also going to log the overwrite: */
1064 if (trans->journal_transaction_names)
1065 journal_u64s += jset_u64s(i->old_k.u64s);
1066 }
1067
1068 if (trans->extra_disk_res) {
1069 ret = bch2_disk_reservation_add(c, trans->disk_res,
1070 trans->extra_disk_res,
1071 (flags & BCH_TRANS_COMMIT_no_enospc)
1072 ? BCH_DISK_RESERVATION_NOFAIL : 0);
1073 if (ret)
1074 goto err;
1075 }
1076 retry:
1077 errored_at = NULL;
1078 bch2_trans_verify_not_unlocked_or_in_restart(trans);
1079 if (likely(!(flags & BCH_TRANS_COMMIT_no_journal_res)))
1080 memset(&trans->journal_res, 0, sizeof(trans->journal_res));
1081 memset(&trans->fs_usage_delta, 0, sizeof(trans->fs_usage_delta));
1082
1083 trans->journal_u64s = journal_u64s + trans->journal_entries.u64s;
1084
1085 ret = do_bch2_trans_commit(trans, flags, &errored_at, _RET_IP_);
1086
1087 /* make sure we didn't drop or screw up locks: */
1088 bch2_trans_verify_locks(trans);
1089
1090 if (ret)
1091 goto err;
1092
1093 trace_and_count(c, transaction_commit, trans, _RET_IP_);
1094 out:
1095 if (likely(!(flags & BCH_TRANS_COMMIT_no_check_rw)))
1096 enumerated_ref_put(&c->writes, BCH_WRITE_REF_trans);
1097 out_reset:
1098 if (!ret)
1099 bch2_trans_downgrade(trans);
1100 bch2_trans_reset_updates(trans);
1101
1102 return ret;
1103 err:
1104 ret = bch2_trans_commit_error(trans, flags, errored_at, ret, _RET_IP_);
1105 if (ret)
1106 goto out;
1107
1108 /*
1109 * We might have done another transaction commit in the error path -
1110 * i.e. btree write buffer flush - which will have made use of
1111 * trans->journal_res, but with BCH_TRANS_COMMIT_no_journal_res that is
1112 * how the journal sequence number to pin is passed in - so we must
1113 * restart:
1114 */
1115 if (flags & BCH_TRANS_COMMIT_no_journal_res) {
1116 ret = bch_err_throw(c, transaction_restart_nested);
1117 goto out;
1118 }
1119
1120 goto retry;
1121 }
1122