1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_key_cache.h"
5 #include "btree_update.h"
12 static void bch2_delete_dead_snapshots_work(struct work_struct *);
13 static void bch2_delete_dead_snapshots(struct bch_fs *);
15 void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
18 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
20 pr_buf(out, "is_subvol %llu deleted %llu parent %u children %u %u subvol %u",
21 BCH_SNAPSHOT_SUBVOL(s.v),
22 BCH_SNAPSHOT_DELETED(s.v),
23 le32_to_cpu(s.v->parent),
24 le32_to_cpu(s.v->children[0]),
25 le32_to_cpu(s.v->children[1]),
26 le32_to_cpu(s.v->subvol));
29 int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
30 int rw, struct printbuf *err)
32 struct bkey_s_c_snapshot s;
35 if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0 ||
36 bkey_cmp(k.k->p, POS(0, 1)) < 0) {
37 pr_buf(err, "bad pos");
41 if (bkey_val_bytes(k.k) != sizeof(struct bch_snapshot)) {
42 pr_buf(err, "bad val size (%zu != %zu)",
43 bkey_val_bytes(k.k), sizeof(struct bch_snapshot));
47 s = bkey_s_c_to_snapshot(k);
49 id = le32_to_cpu(s.v->parent);
50 if (id && id <= k.k->p.offset) {
51 pr_buf(err, "bad parent node (%u <= %llu)",
56 if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
57 pr_buf(err, "children not normalized");
61 if (s.v->children[0] &&
62 s.v->children[0] == s.v->children[1]) {
63 pr_buf(err, "duplicate child nodes");
67 for (i = 0; i < 2; i++) {
68 id = le32_to_cpu(s.v->children[i]);
70 if (id >= k.k->p.offset) {
71 pr_buf(err, "bad child node (%u >= %llu)",
80 int bch2_mark_snapshot(struct btree_trans *trans,
81 struct bkey_s_c old, struct bkey_s_c new,
84 struct bch_fs *c = trans->c;
87 t = genradix_ptr_alloc(&c->snapshots,
88 U32_MAX - new.k->p.offset,
93 if (new.k->type == KEY_TYPE_snapshot) {
94 struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
96 t->parent = le32_to_cpu(s.v->parent);
97 t->children[0] = le32_to_cpu(s.v->children[0]);
98 t->children[1] = le32_to_cpu(s.v->children[1]);
99 t->subvol = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
110 static int snapshot_lookup(struct btree_trans *trans, u32 id,
111 struct bch_snapshot *s)
113 struct btree_iter iter;
117 bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
118 BTREE_ITER_WITH_UPDATES);
119 k = bch2_btree_iter_peek_slot(&iter);
120 ret = bkey_err(k) ?: k.k->type == KEY_TYPE_snapshot ? 0 : -ENOENT;
123 *s = *bkey_s_c_to_snapshot(k).v;
125 bch2_trans_iter_exit(trans, &iter);
129 static int snapshot_live(struct btree_trans *trans, u32 id)
131 struct bch_snapshot v;
137 ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
139 bch_err(trans->c, "snapshot node %u not found", id);
143 return !BCH_SNAPSHOT_DELETED(&v);
146 static int bch2_snapshots_set_equiv(struct btree_trans *trans)
148 struct bch_fs *c = trans->c;
149 struct btree_iter iter;
151 struct bkey_s_c_snapshot snap;
155 for_each_btree_key(trans, iter, BTREE_ID_snapshots,
156 POS_MIN, 0, k, ret) {
157 u32 id = k.k->p.offset, child[2];
158 unsigned nr_live = 0, live_idx = 0;
160 if (k.k->type != KEY_TYPE_snapshot)
163 snap = bkey_s_c_to_snapshot(k);
164 child[0] = le32_to_cpu(snap.v->children[0]);
165 child[1] = le32_to_cpu(snap.v->children[1]);
167 for (i = 0; i < 2; i++) {
168 ret = snapshot_live(trans, child[i]);
177 snapshot_t(c, id)->equiv = nr_live == 1
178 ? snapshot_t(c, child[live_idx])->equiv
182 bch2_trans_iter_exit(trans, &iter);
185 bch_err(c, "error walking snapshots: %i", ret);
191 static int bch2_snapshot_check(struct btree_trans *trans,
192 struct bkey_s_c_snapshot s)
194 struct bch_subvolume subvol;
195 struct bch_snapshot v;
199 id = le32_to_cpu(s.v->subvol);
200 ret = lockrestart_do(trans, bch2_subvolume_get(trans, id, 0, false, &subvol));
202 bch_err(trans->c, "snapshot node %llu has nonexistent subvolume %u",
207 if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
208 bch_err(trans->c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
213 id = le32_to_cpu(s.v->parent);
215 ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
217 bch_err(trans->c, "snapshot node %llu has nonexistent parent %u",
222 if (le32_to_cpu(v.children[0]) != s.k->p.offset &&
223 le32_to_cpu(v.children[1]) != s.k->p.offset) {
224 bch_err(trans->c, "snapshot parent %u missing pointer to child %llu",
230 for (i = 0; i < 2 && s.v->children[i]; i++) {
231 id = le32_to_cpu(s.v->children[i]);
233 ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
235 bch_err(trans->c, "snapshot node %llu has nonexistent child %u",
240 if (le32_to_cpu(v.parent) != s.k->p.offset) {
241 bch_err(trans->c, "snapshot child %u has wrong parent (got %u should be %llu)",
242 id, le32_to_cpu(v.parent), s.k->p.offset);
250 int bch2_fs_snapshots_check(struct bch_fs *c)
252 struct btree_trans trans;
253 struct btree_iter iter;
255 struct bch_snapshot s;
259 bch2_trans_init(&trans, c, 0, 0);
261 for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
262 POS_MIN, 0, k, ret) {
263 if (k.k->type != KEY_TYPE_snapshot)
266 ret = bch2_snapshot_check(&trans, bkey_s_c_to_snapshot(k));
270 bch2_trans_iter_exit(&trans, &iter);
273 bch_err(c, "error %i checking snapshots", ret);
277 for_each_btree_key(&trans, iter, BTREE_ID_subvolumes,
278 POS_MIN, 0, k, ret) {
279 if (k.k->type != KEY_TYPE_subvolume)
282 id = le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot);
283 ret = snapshot_lookup(&trans, id, &s);
286 k = bch2_btree_iter_peek(&iter);
288 } else if (ret == -ENOENT)
289 bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
294 bch2_trans_iter_exit(&trans, &iter);
296 bch2_trans_exit(&trans);
300 void bch2_fs_snapshots_exit(struct bch_fs *c)
302 genradix_free(&c->snapshots);
305 int bch2_fs_snapshots_start(struct bch_fs *c)
307 struct btree_trans trans;
308 struct btree_iter iter;
310 bool have_deleted = false;
313 bch2_trans_init(&trans, c, 0, 0);
315 for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
316 POS_MIN, 0, k, ret) {
317 if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0)
320 if (k.k->type != KEY_TYPE_snapshot) {
321 bch_err(c, "found wrong key type %u in snapshot node table",
326 if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
329 ret = bch2_mark_snapshot(&trans, bkey_s_c_null, k, 0);
333 bch2_trans_iter_exit(&trans, &iter);
338 ret = bch2_snapshots_set_equiv(&trans);
342 bch2_trans_exit(&trans);
344 if (!ret && have_deleted) {
345 bch_info(c, "restarting deletion of dead snapshots");
347 bch2_delete_dead_snapshots_work(&c->snapshot_delete_work);
349 bch2_delete_dead_snapshots(c);
357 * Mark a snapshot as deleted, for future cleanup:
359 static int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
361 struct btree_iter iter;
363 struct bkey_i_snapshot *s;
366 bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
368 k = bch2_btree_iter_peek_slot(&iter);
373 if (k.k->type != KEY_TYPE_snapshot) {
374 bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
379 /* already deleted? */
380 if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
383 s = bch2_trans_kmalloc(trans, sizeof(*s));
384 ret = PTR_ERR_OR_ZERO(s);
388 bkey_reassemble(&s->k_i, k);
390 SET_BCH_SNAPSHOT_DELETED(&s->v, true);
391 ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
395 bch2_trans_iter_exit(trans, &iter);
399 static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
401 struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
403 struct bkey_s_c_snapshot s;
404 struct bkey_i_snapshot *parent;
409 bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
411 k = bch2_btree_iter_peek_slot(&iter);
416 if (k.k->type != KEY_TYPE_snapshot) {
417 bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
422 s = bkey_s_c_to_snapshot(k);
424 BUG_ON(!BCH_SNAPSHOT_DELETED(s.v));
425 parent_id = le32_to_cpu(s.v->parent);
428 bch2_trans_iter_init(trans, &p_iter, BTREE_ID_snapshots,
431 k = bch2_btree_iter_peek_slot(&p_iter);
436 if (k.k->type != KEY_TYPE_snapshot) {
437 bch2_fs_inconsistent(trans->c, "missing snapshot %u", parent_id);
442 parent = bch2_trans_kmalloc(trans, sizeof(*parent));
443 ret = PTR_ERR_OR_ZERO(parent);
447 bkey_reassemble(&parent->k_i, k);
449 for (i = 0; i < 2; i++)
450 if (le32_to_cpu(parent->v.children[i]) == id)
454 bch_err(trans->c, "snapshot %u missing child pointer to %u",
457 parent->v.children[i] = 0;
459 if (le32_to_cpu(parent->v.children[0]) <
460 le32_to_cpu(parent->v.children[1]))
461 swap(parent->v.children[0],
462 parent->v.children[1]);
464 ret = bch2_trans_update(trans, &p_iter, &parent->k_i, 0);
469 ret = bch2_btree_delete_at(trans, &iter, 0);
471 bch2_trans_iter_exit(trans, &p_iter);
472 bch2_trans_iter_exit(trans, &iter);
476 int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
478 u32 *snapshot_subvols,
481 struct btree_iter iter;
482 struct bkey_i_snapshot *n;
487 bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
488 POS_MIN, BTREE_ITER_INTENT);
489 k = bch2_btree_iter_peek(&iter);
494 for (i = 0; i < nr_snapids; i++) {
495 k = bch2_btree_iter_prev_slot(&iter);
500 if (!k.k || !k.k->p.offset) {
505 n = bch2_trans_kmalloc(trans, sizeof(*n));
506 ret = PTR_ERR_OR_ZERO(n);
510 bkey_snapshot_init(&n->k_i);
513 n->v.parent = cpu_to_le32(parent);
514 n->v.subvol = cpu_to_le32(snapshot_subvols[i]);
516 SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
518 ret = bch2_trans_update(trans, &iter, &n->k_i, 0) ?:
519 bch2_mark_snapshot(trans, bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0);
523 new_snapids[i] = iter.pos.offset;
527 bch2_btree_iter_set_pos(&iter, POS(0, parent));
528 k = bch2_btree_iter_peek(&iter);
533 if (k.k->type != KEY_TYPE_snapshot) {
534 bch_err(trans->c, "snapshot %u not found", parent);
539 n = bch2_trans_kmalloc(trans, sizeof(*n));
540 ret = PTR_ERR_OR_ZERO(n);
544 bkey_reassemble(&n->k_i, k);
546 if (n->v.children[0] || n->v.children[1]) {
547 bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
552 n->v.children[0] = cpu_to_le32(new_snapids[0]);
553 n->v.children[1] = cpu_to_le32(new_snapids[1]);
554 SET_BCH_SNAPSHOT_SUBVOL(&n->v, false);
555 ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
560 bch2_trans_iter_exit(trans, &iter);
564 static int snapshot_id_add(snapshot_id_list *s, u32 id)
566 BUG_ON(snapshot_list_has_id(s, id));
568 return darray_push(s, id);
571 static int bch2_snapshot_delete_keys_btree(struct btree_trans *trans,
572 snapshot_id_list *deleted,
573 enum btree_id btree_id)
575 struct bch_fs *c = trans->c;
576 struct btree_iter iter;
578 snapshot_id_list equiv_seen = { 0 };
579 struct bpos last_pos = POS_MIN;
583 * XXX: We should also delete whiteouts that no longer overwrite
587 bch2_trans_iter_init(trans, &iter, btree_id, POS_MIN,
590 BTREE_ITER_NOT_EXTENTS|
591 BTREE_ITER_ALL_SNAPSHOTS);
593 while ((bch2_trans_begin(trans),
594 (k = bch2_btree_iter_peek(&iter)).k) &&
595 !(ret = bkey_err(k))) {
596 u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
598 if (bkey_cmp(k.k->p, last_pos))
602 if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
603 snapshot_list_has_id(&equiv_seen, equiv)) {
604 if (btree_id == BTREE_ID_inodes &&
605 bch2_btree_key_cache_flush(trans, btree_id, iter.pos))
608 ret = __bch2_trans_do(trans, NULL, NULL,
610 bch2_btree_iter_traverse(&iter) ?:
611 bch2_btree_delete_at(trans, &iter,
612 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
616 ret = snapshot_id_add(&equiv_seen, equiv);
621 bch2_btree_iter_advance(&iter);
623 bch2_trans_iter_exit(trans, &iter);
625 darray_exit(&equiv_seen);
630 static void bch2_delete_dead_snapshots_work(struct work_struct *work)
632 struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
633 struct btree_trans trans;
634 struct btree_iter iter;
636 struct bkey_s_c_snapshot snap;
637 snapshot_id_list deleted = { 0 };
638 u32 i, id, children[2];
641 bch2_trans_init(&trans, c, 0, 0);
644 * For every snapshot node: If we have no live children and it's not
645 * pointed to by a subvolume, delete it:
647 for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
648 POS_MIN, 0, k, ret) {
649 if (k.k->type != KEY_TYPE_snapshot)
652 snap = bkey_s_c_to_snapshot(k);
653 if (BCH_SNAPSHOT_DELETED(snap.v) ||
654 BCH_SNAPSHOT_SUBVOL(snap.v))
657 children[0] = le32_to_cpu(snap.v->children[0]);
658 children[1] = le32_to_cpu(snap.v->children[1]);
660 ret = snapshot_live(&trans, children[0]) ?:
661 snapshot_live(&trans, children[1]);
667 ret = __bch2_trans_do(&trans, NULL, NULL, 0,
668 bch2_snapshot_node_set_deleted(&trans, iter.pos.offset));
670 bch_err(c, "error deleting snapshot %llu: %i", iter.pos.offset, ret);
674 bch2_trans_iter_exit(&trans, &iter);
677 bch_err(c, "error walking snapshots: %i", ret);
681 ret = bch2_snapshots_set_equiv(&trans);
685 for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
686 POS_MIN, 0, k, ret) {
687 if (k.k->type != KEY_TYPE_snapshot)
690 snap = bkey_s_c_to_snapshot(k);
691 if (BCH_SNAPSHOT_DELETED(snap.v)) {
692 ret = snapshot_id_add(&deleted, k.k->p.offset);
697 bch2_trans_iter_exit(&trans, &iter);
700 bch_err(c, "error walking snapshots: %i", ret);
704 for (id = 0; id < BTREE_ID_NR; id++) {
705 if (!btree_type_has_snapshots(id))
708 ret = bch2_snapshot_delete_keys_btree(&trans, &deleted, id);
710 bch_err(c, "error deleting snapshot keys: %i", ret);
715 for (i = 0; i < deleted.nr; i++) {
716 ret = __bch2_trans_do(&trans, NULL, NULL, 0,
717 bch2_snapshot_node_delete(&trans, deleted.data[i]));
719 bch_err(c, "error deleting snapshot %u: %i",
720 deleted.data[i], ret);
725 darray_exit(&deleted);
726 bch2_trans_exit(&trans);
727 percpu_ref_put(&c->writes);
730 static void bch2_delete_dead_snapshots(struct bch_fs *c)
732 if (unlikely(!percpu_ref_tryget(&c->writes)))
735 if (!queue_work(system_long_wq, &c->snapshot_delete_work))
736 percpu_ref_put(&c->writes);
739 static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
740 struct btree_trans_commit_hook *h)
742 bch2_delete_dead_snapshots(trans->c);
748 int bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k,
749 int rw, struct printbuf *err)
751 if (bkey_cmp(k.k->p, SUBVOL_POS_MIN) < 0 ||
752 bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0) {
753 pr_buf(err, "invalid pos");
757 if (bkey_val_bytes(k.k) != sizeof(struct bch_subvolume)) {
758 pr_buf(err, "incorrect value size (%zu != %zu)",
759 bkey_val_bytes(k.k), sizeof(struct bch_subvolume));
766 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
769 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
771 pr_buf(out, "root %llu snapshot id %u",
772 le64_to_cpu(s.v->inode),
773 le32_to_cpu(s.v->snapshot));
776 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
777 bool inconsistent_if_not_found,
779 struct bch_subvolume *s)
781 struct btree_iter iter;
785 bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
787 k = bch2_btree_iter_peek_slot(&iter);
788 ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -ENOENT;
790 if (ret == -ENOENT && inconsistent_if_not_found)
791 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
793 *s = *bkey_s_c_to_subvolume(k).v;
795 bch2_trans_iter_exit(trans, &iter);
799 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
800 struct bch_subvolume *subvol)
802 struct bch_snapshot snap;
804 return snapshot_lookup(trans, snapshot, &snap) ?:
805 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
808 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
811 struct bch_subvolume s;
814 ret = bch2_subvolume_get(trans, subvol, true,
816 BTREE_ITER_WITH_UPDATES,
819 *snapid = le32_to_cpu(s.snapshot);
824 * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
827 int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
829 struct btree_iter iter;
831 struct bkey_s_c_subvolume subvol;
832 struct btree_trans_commit_hook *h;
833 struct bkey_i *delete;
837 bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
841 k = bch2_btree_iter_peek_slot(&iter);
846 if (k.k->type != KEY_TYPE_subvolume) {
847 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
852 subvol = bkey_s_c_to_subvolume(k);
853 snapid = le32_to_cpu(subvol.v->snapshot);
855 delete = bch2_trans_kmalloc(trans, sizeof(*delete));
856 ret = PTR_ERR_OR_ZERO(delete);
860 bkey_init(&delete->k);
861 delete->k.p = iter.pos;
862 ret = bch2_trans_update(trans, &iter, delete, 0);
866 ret = bch2_snapshot_node_set_deleted(trans, snapid);
868 h = bch2_trans_kmalloc(trans, sizeof(*h));
869 ret = PTR_ERR_OR_ZERO(h);
873 h->fn = bch2_delete_dead_snapshots_hook;
874 bch2_trans_commit_hook(trans, h);
876 bch2_trans_iter_exit(trans, &iter);
880 void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
882 struct bch_fs *c = container_of(work, struct bch_fs,
883 snapshot_wait_for_pagecache_and_delete_work);
889 mutex_lock(&c->snapshots_unlinked_lock);
890 s = c->snapshots_unlinked;
891 darray_init(&c->snapshots_unlinked);
892 mutex_unlock(&c->snapshots_unlinked_lock);
897 bch2_evict_subvolume_inodes(c, &s);
899 for (id = s.data; id < s.data + s.nr; id++) {
900 ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL,
901 bch2_subvolume_delete(&trans, *id));
903 bch_err(c, "error %i deleting subvolume %u", ret, *id);
911 percpu_ref_put(&c->writes);
914 struct subvolume_unlink_hook {
915 struct btree_trans_commit_hook h;
919 int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
920 struct btree_trans_commit_hook *_h)
922 struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
923 struct bch_fs *c = trans->c;
926 mutex_lock(&c->snapshots_unlinked_lock);
927 if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
928 ret = snapshot_id_add(&c->snapshots_unlinked, h->subvol);
929 mutex_unlock(&c->snapshots_unlinked_lock);
934 if (unlikely(!percpu_ref_tryget(&c->writes)))
937 if (!queue_work(system_long_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
938 percpu_ref_put(&c->writes);
942 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
944 struct btree_iter iter;
946 struct bkey_i_subvolume *n;
947 struct subvolume_unlink_hook *h;
950 bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
954 k = bch2_btree_iter_peek_slot(&iter);
959 if (k.k->type != KEY_TYPE_subvolume) {
960 bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
965 n = bch2_trans_kmalloc(trans, sizeof(*n));
966 ret = PTR_ERR_OR_ZERO(n);
970 bkey_reassemble(&n->k_i, k);
971 SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
973 ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
977 h = bch2_trans_kmalloc(trans, sizeof(*h));
978 ret = PTR_ERR_OR_ZERO(h);
982 h->h.fn = bch2_subvolume_wait_for_pagecache_and_delete_hook;
983 h->subvol = subvolid;
984 bch2_trans_commit_hook(trans, &h->h);
986 bch2_trans_iter_exit(trans, &iter);
990 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
996 struct bch_fs *c = trans->c;
997 struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
998 struct bkey_i_subvolume *new_subvol = NULL;
999 struct bkey_i_subvolume *src_subvol = NULL;
1001 u32 parent = 0, new_nodes[2], snapshot_subvols[2];
1004 for_each_btree_key(trans, dst_iter, BTREE_ID_subvolumes, SUBVOL_POS_MIN,
1005 BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
1006 if (bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0)
1010 * bch2_subvolume_delete() doesn't flush the btree key cache -
1011 * ideally it would but that's tricky
1013 if (bkey_deleted(k.k) &&
1014 !bch2_btree_key_cache_find(c, BTREE_ID_subvolumes, dst_iter.pos))
1022 snapshot_subvols[0] = dst_iter.pos.offset;
1023 snapshot_subvols[1] = src_subvolid;
1026 /* Creating a snapshot: */
1027 src_subvol = bch2_trans_kmalloc(trans, sizeof(*src_subvol));
1028 ret = PTR_ERR_OR_ZERO(src_subvol);
1032 bch2_trans_iter_init(trans, &src_iter, BTREE_ID_subvolumes,
1033 POS(0, src_subvolid),
1036 k = bch2_btree_iter_peek_slot(&src_iter);
1041 if (k.k->type != KEY_TYPE_subvolume) {
1042 bch_err(c, "subvolume %u not found", src_subvolid);
1047 bkey_reassemble(&src_subvol->k_i, k);
1048 parent = le32_to_cpu(src_subvol->v.snapshot);
1051 ret = bch2_snapshot_node_create(trans, parent, new_nodes,
1053 src_subvolid ? 2 : 1);
1058 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
1059 ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
1064 new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
1065 ret = PTR_ERR_OR_ZERO(new_subvol);
1069 bkey_subvolume_init(&new_subvol->k_i);
1070 new_subvol->v.flags = 0;
1071 new_subvol->v.snapshot = cpu_to_le32(new_nodes[0]);
1072 new_subvol->v.inode = cpu_to_le64(inode);
1073 SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
1074 SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
1075 new_subvol->k.p = dst_iter.pos;
1076 ret = bch2_trans_update(trans, &dst_iter, &new_subvol->k_i, 0);
1080 *new_subvolid = new_subvol->k.p.offset;
1081 *new_snapshotid = new_nodes[0];
1083 bch2_trans_iter_exit(trans, &src_iter);
1084 bch2_trans_iter_exit(trans, &dst_iter);
1088 int bch2_fs_subvolumes_init(struct bch_fs *c)
1090 INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
1091 INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
1092 bch2_subvolume_wait_for_pagecache_and_delete);
1093 mutex_init(&c->snapshots_unlinked_lock);