xref: /linux/fs/bcachefs/data_update.c (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "alloc_foreground.h"
5 #include "bkey_buf.h"
6 #include "btree_update.h"
7 #include "buckets.h"
8 #include "data_update.h"
9 #include "ec.h"
10 #include "error.h"
11 #include "extents.h"
12 #include "io_write.h"
13 #include "keylist.h"
14 #include "move.h"
15 #include "nocow_locking.h"
16 #include "rebalance.h"
17 #include "subvolume.h"
18 #include "trace.h"
19 
20 static void trace_move_extent_finish2(struct bch_fs *c, struct bkey_s_c k)
21 {
22 	if (trace_move_extent_finish_enabled()) {
23 		struct printbuf buf = PRINTBUF;
24 
25 		bch2_bkey_val_to_text(&buf, c, k);
26 		trace_move_extent_finish(c, buf.buf);
27 		printbuf_exit(&buf);
28 	}
29 }
30 
31 static void trace_move_extent_fail2(struct data_update *m,
32 			 struct bkey_s_c new,
33 			 struct bkey_s_c wrote,
34 			 struct bkey_i *insert,
35 			 const char *msg)
36 {
37 	struct bch_fs *c = m->op.c;
38 	struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
39 	const union bch_extent_entry *entry;
40 	struct bch_extent_ptr *ptr;
41 	struct extent_ptr_decoded p;
42 	struct printbuf buf = PRINTBUF;
43 	unsigned i, rewrites_found = 0;
44 
45 	if (!trace_move_extent_fail_enabled())
46 		return;
47 
48 	prt_str(&buf, msg);
49 
50 	if (insert) {
51 		i = 0;
52 		bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry) {
53 			if (((1U << i) & m->data_opts.rewrite_ptrs) &&
54 			    (ptr = bch2_extent_has_ptr(old, p, bkey_i_to_s(insert))) &&
55 			    !ptr->cached)
56 				rewrites_found |= 1U << i;
57 			i++;
58 		}
59 	}
60 
61 	prt_printf(&buf, "\nrewrite ptrs:   %u%u%u%u",
62 		   (m->data_opts.rewrite_ptrs & (1 << 0)) != 0,
63 		   (m->data_opts.rewrite_ptrs & (1 << 1)) != 0,
64 		   (m->data_opts.rewrite_ptrs & (1 << 2)) != 0,
65 		   (m->data_opts.rewrite_ptrs & (1 << 3)) != 0);
66 
67 	prt_printf(&buf, "\nrewrites found: %u%u%u%u",
68 		   (rewrites_found & (1 << 0)) != 0,
69 		   (rewrites_found & (1 << 1)) != 0,
70 		   (rewrites_found & (1 << 2)) != 0,
71 		   (rewrites_found & (1 << 3)) != 0);
72 
73 	prt_str(&buf, "\nold:    ");
74 	bch2_bkey_val_to_text(&buf, c, old);
75 
76 	prt_str(&buf, "\nnew:    ");
77 	bch2_bkey_val_to_text(&buf, c, new);
78 
79 	prt_str(&buf, "\nwrote:  ");
80 	bch2_bkey_val_to_text(&buf, c, wrote);
81 
82 	if (insert) {
83 		prt_str(&buf, "\ninsert: ");
84 		bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
85 	}
86 
87 	trace_move_extent_fail(c, buf.buf);
88 	printbuf_exit(&buf);
89 }
90 
91 static int __bch2_data_update_index_update(struct btree_trans *trans,
92 					   struct bch_write_op *op)
93 {
94 	struct bch_fs *c = op->c;
95 	struct btree_iter iter;
96 	struct data_update *m =
97 		container_of(op, struct data_update, op);
98 	struct keylist *keys = &op->insert_keys;
99 	struct bkey_buf _new, _insert;
100 	int ret = 0;
101 
102 	bch2_bkey_buf_init(&_new);
103 	bch2_bkey_buf_init(&_insert);
104 	bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
105 
106 	bch2_trans_iter_init(trans, &iter, m->btree_id,
107 			     bkey_start_pos(&bch2_keylist_front(keys)->k),
108 			     BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
109 
110 	while (1) {
111 		struct bkey_s_c k;
112 		struct bkey_s_c old = bkey_i_to_s_c(m->k.k);
113 		struct bkey_i *insert = NULL;
114 		struct bkey_i_extent *new;
115 		const union bch_extent_entry *entry_c;
116 		union bch_extent_entry *entry;
117 		struct extent_ptr_decoded p;
118 		struct bch_extent_ptr *ptr;
119 		const struct bch_extent_ptr *ptr_c;
120 		struct bpos next_pos;
121 		bool should_check_enospc;
122 		s64 i_sectors_delta = 0, disk_sectors_delta = 0;
123 		unsigned rewrites_found = 0, durability, i;
124 
125 		bch2_trans_begin(trans);
126 
127 		k = bch2_btree_iter_peek_slot(&iter);
128 		ret = bkey_err(k);
129 		if (ret)
130 			goto err;
131 
132 		new = bkey_i_to_extent(bch2_keylist_front(keys));
133 
134 		if (!bch2_extents_match(k, old)) {
135 			trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i),
136 						NULL, "no match:");
137 			goto nowork;
138 		}
139 
140 		bkey_reassemble(_insert.k, k);
141 		insert = _insert.k;
142 
143 		bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
144 		new = bkey_i_to_extent(_new.k);
145 		bch2_cut_front(iter.pos, &new->k_i);
146 
147 		bch2_cut_front(iter.pos,	insert);
148 		bch2_cut_back(new->k.p,		insert);
149 		bch2_cut_back(insert->k.p,	&new->k_i);
150 
151 		/*
152 		 * @old: extent that we read from
153 		 * @insert: key that we're going to update, initialized from
154 		 * extent currently in btree - same as @old unless we raced with
155 		 * other updates
156 		 * @new: extent with new pointers that we'll be adding to @insert
157 		 *
158 		 * Fist, drop rewrite_ptrs from @new:
159 		 */
160 		i = 0;
161 		bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs_c(old), p, entry_c) {
162 			if (((1U << i) & m->data_opts.rewrite_ptrs) &&
163 			    (ptr = bch2_extent_has_ptr(old, p, bkey_i_to_s(insert))) &&
164 			    !ptr->cached) {
165 				bch2_extent_ptr_set_cached(bkey_i_to_s(insert), ptr);
166 				rewrites_found |= 1U << i;
167 			}
168 			i++;
169 		}
170 
171 		if (m->data_opts.rewrite_ptrs &&
172 		    !rewrites_found &&
173 		    bch2_bkey_durability(c, k) >= m->op.opts.data_replicas) {
174 			trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "no rewrites found:");
175 			goto nowork;
176 		}
177 
178 		/*
179 		 * A replica that we just wrote might conflict with a replica
180 		 * that we want to keep, due to racing with another move:
181 		 */
182 restart_drop_conflicting_replicas:
183 		extent_for_each_ptr(extent_i_to_s(new), ptr)
184 			if ((ptr_c = bch2_bkey_has_device_c(bkey_i_to_s_c(insert), ptr->dev)) &&
185 			    !ptr_c->cached) {
186 				bch2_bkey_drop_ptr_noerror(bkey_i_to_s(&new->k_i), ptr);
187 				goto restart_drop_conflicting_replicas;
188 			}
189 
190 		if (!bkey_val_u64s(&new->k)) {
191 			trace_move_extent_fail2(m, k, bkey_i_to_s_c(&new->k_i), insert, "new replicas conflicted:");
192 			goto nowork;
193 		}
194 
195 		/* Now, drop pointers that conflict with what we just wrote: */
196 		extent_for_each_ptr_decode(extent_i_to_s(new), p, entry)
197 			if ((ptr = bch2_bkey_has_device(bkey_i_to_s(insert), p.ptr.dev)))
198 				bch2_bkey_drop_ptr_noerror(bkey_i_to_s(insert), ptr);
199 
200 		durability = bch2_bkey_durability(c, bkey_i_to_s_c(insert)) +
201 			bch2_bkey_durability(c, bkey_i_to_s_c(&new->k_i));
202 
203 		/* Now, drop excess replicas: */
204 restart_drop_extra_replicas:
205 		bkey_for_each_ptr_decode(old.k, bch2_bkey_ptrs(bkey_i_to_s(insert)), p, entry) {
206 			unsigned ptr_durability = bch2_extent_ptr_durability(c, &p);
207 
208 			if (!p.ptr.cached &&
209 			    durability - ptr_durability >= m->op.opts.data_replicas) {
210 				durability -= ptr_durability;
211 
212 				bch2_extent_ptr_set_cached(bkey_i_to_s(insert), &entry->ptr);
213 				goto restart_drop_extra_replicas;
214 			}
215 		}
216 
217 		/* Finally, add the pointers we just wrote: */
218 		extent_for_each_ptr_decode(extent_i_to_s(new), p, entry)
219 			bch2_extent_ptr_decoded_append(insert, &p);
220 
221 		bch2_bkey_narrow_crcs(insert, (struct bch_extent_crc_unpacked) { 0 });
222 		bch2_extent_normalize(c, bkey_i_to_s(insert));
223 
224 		ret = bch2_sum_sector_overwrites(trans, &iter, insert,
225 						 &should_check_enospc,
226 						 &i_sectors_delta,
227 						 &disk_sectors_delta);
228 		if (ret)
229 			goto err;
230 
231 		if (disk_sectors_delta > (s64) op->res.sectors) {
232 			ret = bch2_disk_reservation_add(c, &op->res,
233 						disk_sectors_delta - op->res.sectors,
234 						!should_check_enospc
235 						? BCH_DISK_RESERVATION_NOFAIL : 0);
236 			if (ret)
237 				goto out;
238 		}
239 
240 		next_pos = insert->k.p;
241 
242 		/*
243 		 * Check for nonce offset inconsistency:
244 		 * This is debug code - we've been seeing this bug rarely, and
245 		 * it's been hard to reproduce, so this should give us some more
246 		 * information when it does occur:
247 		 */
248 		struct printbuf err = PRINTBUF;
249 		int invalid = bch2_bkey_invalid(c, bkey_i_to_s_c(insert), __btree_node_type(0, m->btree_id), 0, &err);
250 		printbuf_exit(&err);
251 
252 		if (invalid) {
253 			struct printbuf buf = PRINTBUF;
254 
255 			prt_str(&buf, "about to insert invalid key in data update path");
256 			prt_str(&buf, "\nold: ");
257 			bch2_bkey_val_to_text(&buf, c, old);
258 			prt_str(&buf, "\nk:   ");
259 			bch2_bkey_val_to_text(&buf, c, k);
260 			prt_str(&buf, "\nnew: ");
261 			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
262 
263 			bch2_print_string_as_lines(KERN_ERR, buf.buf);
264 			printbuf_exit(&buf);
265 
266 			bch2_fatal_error(c);
267 			goto out;
268 		}
269 
270 		if (trace_data_update_enabled()) {
271 			struct printbuf buf = PRINTBUF;
272 
273 			prt_str(&buf, "\nold: ");
274 			bch2_bkey_val_to_text(&buf, c, old);
275 			prt_str(&buf, "\nk:   ");
276 			bch2_bkey_val_to_text(&buf, c, k);
277 			prt_str(&buf, "\nnew: ");
278 			bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(insert));
279 
280 			trace_data_update(c, buf.buf);
281 			printbuf_exit(&buf);
282 		}
283 
284 		ret =   bch2_insert_snapshot_whiteouts(trans, m->btree_id,
285 						k.k->p, bkey_start_pos(&insert->k)) ?:
286 			bch2_insert_snapshot_whiteouts(trans, m->btree_id,
287 						k.k->p, insert->k.p) ?:
288 			bch2_bkey_set_needs_rebalance(c, insert, &op->opts) ?:
289 			bch2_trans_update(trans, &iter, insert,
290 				BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
291 			bch2_trans_commit(trans, &op->res,
292 				NULL,
293 				BCH_TRANS_COMMIT_no_check_rw|
294 				BCH_TRANS_COMMIT_no_enospc|
295 				m->data_opts.btree_insert_flags);
296 		if (!ret) {
297 			bch2_btree_iter_set_pos(&iter, next_pos);
298 
299 			this_cpu_add(c->counters[BCH_COUNTER_move_extent_finish], new->k.size);
300 			trace_move_extent_finish2(c, bkey_i_to_s_c(&new->k_i));
301 		}
302 err:
303 		if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
304 			ret = 0;
305 		if (ret)
306 			break;
307 next:
308 		while (bkey_ge(iter.pos, bch2_keylist_front(keys)->k.p)) {
309 			bch2_keylist_pop_front(keys);
310 			if (bch2_keylist_empty(keys))
311 				goto out;
312 		}
313 		continue;
314 nowork:
315 		if (m->stats) {
316 			BUG_ON(k.k->p.offset <= iter.pos.offset);
317 			atomic64_inc(&m->stats->keys_raced);
318 			atomic64_add(k.k->p.offset - iter.pos.offset,
319 				     &m->stats->sectors_raced);
320 		}
321 
322 		count_event(c, move_extent_fail);
323 
324 		bch2_btree_iter_advance(&iter);
325 		goto next;
326 	}
327 out:
328 	bch2_trans_iter_exit(trans, &iter);
329 	bch2_bkey_buf_exit(&_insert, c);
330 	bch2_bkey_buf_exit(&_new, c);
331 	BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
332 	return ret;
333 }
334 
335 int bch2_data_update_index_update(struct bch_write_op *op)
336 {
337 	return bch2_trans_run(op->c, __bch2_data_update_index_update(trans, op));
338 }
339 
340 void bch2_data_update_read_done(struct data_update *m,
341 				struct bch_extent_crc_unpacked crc)
342 {
343 	/* write bio must own pages: */
344 	BUG_ON(!m->op.wbio.bio.bi_vcnt);
345 
346 	m->op.crc = crc;
347 	m->op.wbio.bio.bi_iter.bi_size = crc.compressed_size << 9;
348 
349 	closure_call(&m->op.cl, bch2_write, NULL, NULL);
350 }
351 
352 void bch2_data_update_exit(struct data_update *update)
353 {
354 	struct bch_fs *c = update->op.c;
355 	struct bkey_ptrs_c ptrs =
356 		bch2_bkey_ptrs_c(bkey_i_to_s_c(update->k.k));
357 
358 	bkey_for_each_ptr(ptrs, ptr) {
359 		if (c->opts.nocow_enabled)
360 			bch2_bucket_nocow_unlock(&c->nocow_locks,
361 						 PTR_BUCKET_POS(c, ptr), 0);
362 		percpu_ref_put(&bch_dev_bkey_exists(c, ptr->dev)->ref);
363 	}
364 
365 	bch2_bkey_buf_exit(&update->k, c);
366 	bch2_disk_reservation_put(c, &update->op.res);
367 	bch2_bio_free_pages_pool(c, &update->op.wbio.bio);
368 }
369 
370 static void bch2_update_unwritten_extent(struct btree_trans *trans,
371 				  struct data_update *update)
372 {
373 	struct bch_fs *c = update->op.c;
374 	struct bio *bio = &update->op.wbio.bio;
375 	struct bkey_i_extent *e;
376 	struct write_point *wp;
377 	struct closure cl;
378 	struct btree_iter iter;
379 	struct bkey_s_c k;
380 	int ret;
381 
382 	closure_init_stack(&cl);
383 	bch2_keylist_init(&update->op.insert_keys, update->op.inline_keys);
384 
385 	while (bio_sectors(bio)) {
386 		unsigned sectors = bio_sectors(bio);
387 
388 		bch2_trans_iter_init(trans, &iter, update->btree_id, update->op.pos,
389 				     BTREE_ITER_SLOTS);
390 		ret = lockrestart_do(trans, ({
391 			k = bch2_btree_iter_peek_slot(&iter);
392 			bkey_err(k);
393 		}));
394 		bch2_trans_iter_exit(trans, &iter);
395 
396 		if (ret || !bch2_extents_match(k, bkey_i_to_s_c(update->k.k)))
397 			break;
398 
399 		e = bkey_extent_init(update->op.insert_keys.top);
400 		e->k.p = update->op.pos;
401 
402 		ret = bch2_alloc_sectors_start_trans(trans,
403 				update->op.target,
404 				false,
405 				update->op.write_point,
406 				&update->op.devs_have,
407 				update->op.nr_replicas,
408 				update->op.nr_replicas,
409 				update->op.watermark,
410 				0, &cl, &wp);
411 		if (bch2_err_matches(ret, BCH_ERR_operation_blocked)) {
412 			bch2_trans_unlock(trans);
413 			closure_sync(&cl);
414 			continue;
415 		}
416 
417 		bch_err_fn_ratelimited(c, ret);
418 
419 		if (ret)
420 			return;
421 
422 		sectors = min(sectors, wp->sectors_free);
423 
424 		bch2_key_resize(&e->k, sectors);
425 
426 		bch2_open_bucket_get(c, wp, &update->op.open_buckets);
427 		bch2_alloc_sectors_append_ptrs(c, wp, &e->k_i, sectors, false);
428 		bch2_alloc_sectors_done(c, wp);
429 
430 		bio_advance(bio, sectors << 9);
431 		update->op.pos.offset += sectors;
432 
433 		extent_for_each_ptr(extent_i_to_s(e), ptr)
434 			ptr->unwritten = true;
435 		bch2_keylist_push(&update->op.insert_keys);
436 
437 		ret = __bch2_data_update_index_update(trans, &update->op);
438 
439 		bch2_open_buckets_put(c, &update->op.open_buckets);
440 
441 		if (ret)
442 			break;
443 	}
444 
445 	if (closure_nr_remaining(&cl) != 1) {
446 		bch2_trans_unlock(trans);
447 		closure_sync(&cl);
448 	}
449 }
450 
451 int bch2_extent_drop_ptrs(struct btree_trans *trans,
452 			  struct btree_iter *iter,
453 			  struct bkey_s_c k,
454 			  struct data_update_opts data_opts)
455 {
456 	struct bch_fs *c = trans->c;
457 	struct bkey_i *n;
458 	int ret;
459 
460 	n = bch2_bkey_make_mut_noupdate(trans, k);
461 	ret = PTR_ERR_OR_ZERO(n);
462 	if (ret)
463 		return ret;
464 
465 	while (data_opts.kill_ptrs) {
466 		unsigned i = 0, drop = __fls(data_opts.kill_ptrs);
467 		struct bch_extent_ptr *ptr;
468 
469 		bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, i++ == drop);
470 		data_opts.kill_ptrs ^= 1U << drop;
471 	}
472 
473 	/*
474 	 * If the new extent no longer has any pointers, bch2_extent_normalize()
475 	 * will do the appropriate thing with it (turning it into a
476 	 * KEY_TYPE_error key, or just a discard if it was a cached extent)
477 	 */
478 	bch2_extent_normalize(c, bkey_i_to_s(n));
479 
480 	/*
481 	 * Since we're not inserting through an extent iterator
482 	 * (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
483 	 * we aren't using the extent overwrite path to delete, we're
484 	 * just using the normal key deletion path:
485 	 */
486 	if (bkey_deleted(&n->k) && !(iter->flags & BTREE_ITER_IS_EXTENTS))
487 		n->k.size = 0;
488 
489 	return bch2_trans_relock(trans) ?:
490 		bch2_trans_update(trans, iter, n, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
491 		bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
492 }
493 
494 int bch2_data_update_init(struct btree_trans *trans,
495 			  struct btree_iter *iter,
496 			  struct moving_context *ctxt,
497 			  struct data_update *m,
498 			  struct write_point_specifier wp,
499 			  struct bch_io_opts io_opts,
500 			  struct data_update_opts data_opts,
501 			  enum btree_id btree_id,
502 			  struct bkey_s_c k)
503 {
504 	struct bch_fs *c = trans->c;
505 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
506 	const union bch_extent_entry *entry;
507 	struct extent_ptr_decoded p;
508 	unsigned i, reserve_sectors = k.k->size * data_opts.extra_replicas;
509 	unsigned ptrs_locked = 0;
510 	int ret = 0;
511 
512 	bch2_bkey_buf_init(&m->k);
513 	bch2_bkey_buf_reassemble(&m->k, c, k);
514 	m->btree_id	= btree_id;
515 	m->data_opts	= data_opts;
516 	m->ctxt		= ctxt;
517 	m->stats	= ctxt ? ctxt->stats : NULL;
518 
519 	bch2_write_op_init(&m->op, c, io_opts);
520 	m->op.pos	= bkey_start_pos(k.k);
521 	m->op.version	= k.k->version;
522 	m->op.target	= data_opts.target;
523 	m->op.write_point = wp;
524 	m->op.nr_replicas = 0;
525 	m->op.flags	|= BCH_WRITE_PAGES_STABLE|
526 		BCH_WRITE_PAGES_OWNED|
527 		BCH_WRITE_DATA_ENCODED|
528 		BCH_WRITE_MOVE|
529 		m->data_opts.write_flags;
530 	m->op.compression_opt	= background_compression(io_opts);
531 	m->op.watermark		= m->data_opts.btree_insert_flags & BCH_WATERMARK_MASK;
532 
533 	bkey_for_each_ptr(ptrs, ptr)
534 		percpu_ref_get(&bch_dev_bkey_exists(c, ptr->dev)->ref);
535 
536 	unsigned durability_have = 0, durability_removing = 0;
537 
538 	i = 0;
539 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
540 		bool locked;
541 
542 		if (((1U << i) & m->data_opts.rewrite_ptrs)) {
543 			BUG_ON(p.ptr.cached);
544 
545 			if (crc_is_compressed(p.crc))
546 				reserve_sectors += k.k->size;
547 
548 			m->op.nr_replicas += bch2_extent_ptr_desired_durability(c, &p);
549 			durability_removing += bch2_extent_ptr_desired_durability(c, &p);
550 		} else if (!p.ptr.cached &&
551 			   !((1U << i) & m->data_opts.kill_ptrs)) {
552 			bch2_dev_list_add_dev(&m->op.devs_have, p.ptr.dev);
553 			durability_have += bch2_extent_ptr_durability(c, &p);
554 		}
555 
556 		/*
557 		 * op->csum_type is normally initialized from the fs/file's
558 		 * current options - but if an extent is encrypted, we require
559 		 * that it stays encrypted:
560 		 */
561 		if (bch2_csum_type_is_encryption(p.crc.csum_type)) {
562 			m->op.nonce	= p.crc.nonce + p.crc.offset;
563 			m->op.csum_type = p.crc.csum_type;
564 		}
565 
566 		if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible)
567 			m->op.incompressible = true;
568 
569 		if (c->opts.nocow_enabled) {
570 			if (ctxt) {
571 				move_ctxt_wait_event(ctxt,
572 						(locked = bch2_bucket_nocow_trylock(&c->nocow_locks,
573 									  PTR_BUCKET_POS(c, &p.ptr), 0)) ||
574 						(!atomic_read(&ctxt->read_sectors) &&
575 						 !atomic_read(&ctxt->write_sectors)));
576 
577 				if (!locked)
578 					bch2_bucket_nocow_lock(&c->nocow_locks,
579 							       PTR_BUCKET_POS(c, &p.ptr), 0);
580 			} else {
581 				if (!bch2_bucket_nocow_trylock(&c->nocow_locks,
582 							       PTR_BUCKET_POS(c, &p.ptr), 0)) {
583 					ret = -BCH_ERR_nocow_lock_blocked;
584 					goto err;
585 				}
586 			}
587 			ptrs_locked |= (1U << i);
588 		}
589 
590 		i++;
591 	}
592 
593 	/*
594 	 * If current extent durability is less than io_opts.data_replicas,
595 	 * we're not trying to rereplicate the extent up to data_replicas here -
596 	 * unless extra_replicas was specified
597 	 *
598 	 * Increasing replication is an explicit operation triggered by
599 	 * rereplicate, currently, so that users don't get an unexpected -ENOSPC
600 	 */
601 	if (!(m->data_opts.write_flags & BCH_WRITE_CACHED) &&
602 	    durability_have >= io_opts.data_replicas) {
603 		m->data_opts.kill_ptrs |= m->data_opts.rewrite_ptrs;
604 		m->data_opts.rewrite_ptrs = 0;
605 		/* if iter == NULL, it's just a promote */
606 		if (iter)
607 			ret = bch2_extent_drop_ptrs(trans, iter, k, m->data_opts);
608 		goto done;
609 	}
610 
611 	m->op.nr_replicas = min(durability_removing, io_opts.data_replicas - durability_have) +
612 		m->data_opts.extra_replicas;
613 	m->op.nr_replicas_required = m->op.nr_replicas;
614 
615 	BUG_ON(!m->op.nr_replicas);
616 
617 	if (reserve_sectors) {
618 		ret = bch2_disk_reservation_add(c, &m->op.res, reserve_sectors,
619 				m->data_opts.extra_replicas
620 				? 0
621 				: BCH_DISK_RESERVATION_NOFAIL);
622 		if (ret)
623 			goto err;
624 	}
625 
626 	if (bkey_extent_is_unwritten(k)) {
627 		bch2_update_unwritten_extent(trans, m);
628 		goto done;
629 	}
630 
631 	return 0;
632 err:
633 	i = 0;
634 	bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
635 		if ((1U << i) & ptrs_locked)
636 			bch2_bucket_nocow_unlock(&c->nocow_locks,
637 						 PTR_BUCKET_POS(c, &p.ptr), 0);
638 		percpu_ref_put(&bch_dev_bkey_exists(c, p.ptr.dev)->ref);
639 		i++;
640 	}
641 
642 	bch2_bkey_buf_exit(&m->k, c);
643 	bch2_bio_free_pages_pool(c, &m->op.wbio.bio);
644 	return ret;
645 done:
646 	bch2_data_update_exit(m);
647 	return ret ?: -BCH_ERR_data_update_done;
648 }
649 
650 void bch2_data_update_opts_normalize(struct bkey_s_c k, struct data_update_opts *opts)
651 {
652 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
653 	unsigned i = 0;
654 
655 	bkey_for_each_ptr(ptrs, ptr) {
656 		if ((opts->rewrite_ptrs & (1U << i)) && ptr->cached) {
657 			opts->kill_ptrs |= 1U << i;
658 			opts->rewrite_ptrs ^= 1U << i;
659 		}
660 
661 		i++;
662 	}
663 }
664