1 // SPDX-License-Identifier: GPL-2.0
4 #include "btree_key_cache.h"
5 #include "btree_update.h"
10 #include "subvolume.h"
12 #include <linux/random.h>
14 static int bch2_subvolume_delete(struct btree_trans *, u32);
16 static int check_subvol(struct btree_trans *trans,
17 struct btree_iter *iter,
20 struct bch_fs *c = trans->c;
21 struct bkey_s_c_subvolume subvol;
22 struct bch_snapshot snapshot;
26 if (k.k->type != KEY_TYPE_subvolume)
29 subvol = bkey_s_c_to_subvolume(k);
30 snapid = le32_to_cpu(subvol.v->snapshot);
31 ret = bch2_snapshot_lookup(trans, snapid, &snapshot);
33 if (bch2_err_matches(ret, ENOENT))
34 bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
35 k.k->p.offset, snapid);
39 if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
42 ret = bch2_subvolume_delete(trans, iter->pos.offset);
44 bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
45 return ret ?: -BCH_ERR_transaction_restart_nested;
48 if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
49 u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
51 struct bch_snapshot_tree st;
54 snapshot_tree = snapshot_t(c, snapshot_root)->tree;
57 ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
59 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
60 "%s: snapshot tree %u not found", __func__, snapshot_tree);
65 if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset,
66 c, subvol_not_master_and_not_snapshot,
67 "subvolume %llu is not set as snapshot but is not master subvolume",
69 struct bkey_i_subvolume *s =
70 bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
71 ret = PTR_ERR_OR_ZERO(s);
75 SET_BCH_SUBVOLUME_SNAP(&s->v, true);
83 int bch2_check_subvols(struct bch_fs *c)
85 struct btree_iter iter;
89 ret = bch2_trans_run(c,
90 for_each_btree_key_commit(trans, iter,
91 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
92 NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
93 check_subvol(trans, &iter, k)));
101 int bch2_subvolume_invalid(struct bch_fs *c, struct bkey_s_c k,
102 enum bkey_invalid_flags flags, struct printbuf *err)
106 bkey_fsck_err_on(bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
107 bkey_gt(k.k->p, SUBVOL_POS_MAX), c, err,
114 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
117 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
119 prt_printf(out, "root %llu snapshot id %u",
120 le64_to_cpu(s.v->inode),
121 le32_to_cpu(s.v->snapshot));
123 if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, parent))
124 prt_printf(out, " parent %u", le32_to_cpu(s.v->parent));
127 static __always_inline int
128 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
129 bool inconsistent_if_not_found,
131 struct bch_subvolume *s)
133 int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol),
134 iter_flags, subvolume, s);
135 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT) &&
136 inconsistent_if_not_found,
137 trans->c, "missing subvolume %u", subvol);
141 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
142 bool inconsistent_if_not_found,
144 struct bch_subvolume *s)
146 return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
149 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
150 struct bch_subvolume *subvol)
152 struct bch_snapshot snap;
154 return bch2_snapshot_lookup(trans, snapshot, &snap) ?:
155 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
158 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
161 struct btree_iter iter;
162 struct bkey_s_c_subvolume subvol;
165 subvol = bch2_bkey_get_iter_typed(trans, &iter,
166 BTREE_ID_subvolumes, POS(0, subvolid),
167 BTREE_ITER_CACHED|BTREE_ITER_WITH_UPDATES,
169 ret = bkey_err(subvol);
170 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
171 "missing subvolume %u", subvolid);
174 *snapid = le32_to_cpu(subvol.v->snapshot);
175 bch2_trans_iter_exit(trans, &iter);
179 static int bch2_subvolume_reparent(struct btree_trans *trans,
180 struct btree_iter *iter,
182 u32 old_parent, u32 new_parent)
184 struct bkey_i_subvolume *s;
187 if (k.k->type != KEY_TYPE_subvolume)
190 if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, parent) &&
191 le32_to_cpu(bkey_s_c_to_subvolume(k).v->parent) != old_parent)
194 s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume);
195 ret = PTR_ERR_OR_ZERO(s);
199 s->v.parent = cpu_to_le32(new_parent);
204 * Separate from the snapshot tree in the snapshots btree, we record the tree
205 * structure of how snapshot subvolumes were created - the parent subvolume of
206 * each snapshot subvolume.
208 * When a subvolume is deleted, we scan for child subvolumes and reparant them,
209 * to avoid dangling references:
211 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
213 struct btree_iter iter;
215 struct bch_subvolume s;
217 return lockrestart_do(trans,
218 bch2_subvolume_get(trans, subvolid_to_delete, true,
219 BTREE_ITER_CACHED, &s)) ?:
220 for_each_btree_key_commit(trans, iter,
221 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
222 NULL, NULL, BTREE_INSERT_NOFAIL,
223 bch2_subvolume_reparent(trans, &iter, k,
224 subvolid_to_delete, le32_to_cpu(s.parent)));
228 * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
231 static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
233 struct btree_iter iter;
234 struct bkey_s_c_subvolume subvol;
238 subvol = bch2_bkey_get_iter_typed(trans, &iter,
239 BTREE_ID_subvolumes, POS(0, subvolid),
240 BTREE_ITER_CACHED|BTREE_ITER_INTENT,
242 ret = bkey_err(subvol);
243 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
244 "missing subvolume %u", subvolid);
248 snapid = le32_to_cpu(subvol.v->snapshot);
250 ret = bch2_btree_delete_at(trans, &iter, 0) ?:
251 bch2_snapshot_node_set_deleted(trans, snapid);
252 bch2_trans_iter_exit(trans, &iter);
256 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
258 return bch2_subvolumes_reparent(trans, subvolid) ?:
259 commit_do(trans, NULL, NULL, BTREE_INSERT_NOFAIL,
260 __bch2_subvolume_delete(trans, subvolid));
263 static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
265 struct bch_fs *c = container_of(work, struct bch_fs,
266 snapshot_wait_for_pagecache_and_delete_work);
272 mutex_lock(&c->snapshots_unlinked_lock);
273 s = c->snapshots_unlinked;
274 darray_init(&c->snapshots_unlinked);
275 mutex_unlock(&c->snapshots_unlinked_lock);
280 bch2_evict_subvolume_inodes(c, &s);
282 for (id = s.data; id < s.data + s.nr; id++) {
283 ret = bch2_trans_run(c, bch2_subvolume_delete(trans, *id));
285 bch_err_msg(c, ret, "deleting subvolume %u", *id);
293 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
296 struct subvolume_unlink_hook {
297 struct btree_trans_commit_hook h;
301 static int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
302 struct btree_trans_commit_hook *_h)
304 struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
305 struct bch_fs *c = trans->c;
308 mutex_lock(&c->snapshots_unlinked_lock);
309 if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
310 ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
311 mutex_unlock(&c->snapshots_unlinked_lock);
316 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
319 if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
320 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
324 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
326 struct btree_iter iter;
327 struct bkey_i_subvolume *n;
328 struct subvolume_unlink_hook *h;
331 h = bch2_trans_kmalloc(trans, sizeof(*h));
332 ret = PTR_ERR_OR_ZERO(h);
336 h->h.fn = bch2_subvolume_wait_for_pagecache_and_delete_hook;
337 h->subvol = subvolid;
338 bch2_trans_commit_hook(trans, &h->h);
340 n = bch2_bkey_get_mut_typed(trans, &iter,
341 BTREE_ID_subvolumes, POS(0, subvolid),
342 BTREE_ITER_CACHED, subvolume);
343 ret = PTR_ERR_OR_ZERO(n);
345 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
346 "missing subvolume %u", subvolid);
350 SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
351 bch2_trans_iter_exit(trans, &iter);
355 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
361 struct bch_fs *c = trans->c;
362 struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
363 struct bkey_i_subvolume *new_subvol = NULL;
364 struct bkey_i_subvolume *src_subvol = NULL;
365 u32 parent = 0, new_nodes[2], snapshot_subvols[2];
368 ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
369 BTREE_ID_subvolumes, POS(0, U32_MAX));
370 if (ret == -BCH_ERR_ENOSPC_btree_slot)
371 ret = -BCH_ERR_ENOSPC_subvolume_create;
375 snapshot_subvols[0] = dst_iter.pos.offset;
376 snapshot_subvols[1] = src_subvolid;
379 /* Creating a snapshot: */
381 src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter,
382 BTREE_ID_subvolumes, POS(0, src_subvolid),
383 BTREE_ITER_CACHED, subvolume);
384 ret = PTR_ERR_OR_ZERO(src_subvol);
386 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
387 "subvolume %u not found", src_subvolid);
391 parent = le32_to_cpu(src_subvol->v.snapshot);
394 ret = bch2_snapshot_node_create(trans, parent, new_nodes,
396 src_subvolid ? 2 : 1);
401 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
402 ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
407 new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume);
408 ret = PTR_ERR_OR_ZERO(new_subvol);
412 new_subvol->v.flags = 0;
413 new_subvol->v.snapshot = cpu_to_le32(new_nodes[0]);
414 new_subvol->v.inode = cpu_to_le64(inode);
415 new_subvol->v.parent = cpu_to_le32(src_subvolid);
416 new_subvol->v.otime.lo = cpu_to_le64(bch2_current_time(c));
417 new_subvol->v.otime.hi = 0;
419 SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
420 SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
422 *new_subvolid = new_subvol->k.p.offset;
423 *new_snapshotid = new_nodes[0];
425 bch2_trans_iter_exit(trans, &src_iter);
426 bch2_trans_iter_exit(trans, &dst_iter);
430 int bch2_fs_subvolumes_init(struct bch_fs *c)
432 INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
433 INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
434 bch2_subvolume_wait_for_pagecache_and_delete);
435 mutex_init(&c->snapshots_unlinked_lock);