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)) {
40 ret = bch2_subvolume_delete(trans, iter->pos.offset);
41 bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
42 return ret ?: -BCH_ERR_transaction_restart_nested;
45 if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
46 u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
48 struct bch_snapshot_tree st;
51 snapshot_tree = snapshot_t(c, snapshot_root)->tree;
54 ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
56 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
57 "%s: snapshot tree %u not found", __func__, snapshot_tree);
62 if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset,
63 c, subvol_not_master_and_not_snapshot,
64 "subvolume %llu is not set as snapshot but is not master subvolume",
66 struct bkey_i_subvolume *s =
67 bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
68 ret = PTR_ERR_OR_ZERO(s);
72 SET_BCH_SUBVOLUME_SNAP(&s->v, true);
80 int bch2_check_subvols(struct bch_fs *c)
82 int ret = bch2_trans_run(c,
83 for_each_btree_key_commit(trans, iter,
84 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
85 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
86 check_subvol(trans, &iter, k)));
93 int bch2_subvolume_invalid(struct bch_fs *c, struct bkey_s_c k,
94 enum bkey_invalid_flags flags, struct printbuf *err)
98 bkey_fsck_err_on(bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
99 bkey_gt(k.k->p, SUBVOL_POS_MAX), c, err,
106 void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
109 struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
111 prt_printf(out, "root %llu snapshot id %u",
112 le64_to_cpu(s.v->inode),
113 le32_to_cpu(s.v->snapshot));
115 if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, parent))
116 prt_printf(out, " parent %u", le32_to_cpu(s.v->parent));
119 static __always_inline int
120 bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
121 bool inconsistent_if_not_found,
123 struct bch_subvolume *s)
125 int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol),
126 iter_flags, subvolume, s);
127 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT) &&
128 inconsistent_if_not_found,
129 trans->c, "missing subvolume %u", subvol);
133 int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
134 bool inconsistent_if_not_found,
136 struct bch_subvolume *s)
138 return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
141 int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
142 struct bch_subvolume *subvol)
144 struct bch_snapshot snap;
146 return bch2_snapshot_lookup(trans, snapshot, &snap) ?:
147 bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
150 int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
153 struct btree_iter iter;
154 struct bkey_s_c_subvolume subvol;
157 subvol = bch2_bkey_get_iter_typed(trans, &iter,
158 BTREE_ID_subvolumes, POS(0, subvolid),
159 BTREE_ITER_CACHED|BTREE_ITER_WITH_UPDATES,
161 ret = bkey_err(subvol);
162 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
163 "missing subvolume %u", subvolid);
166 *snapid = le32_to_cpu(subvol.v->snapshot);
167 bch2_trans_iter_exit(trans, &iter);
171 static int bch2_subvolume_reparent(struct btree_trans *trans,
172 struct btree_iter *iter,
174 u32 old_parent, u32 new_parent)
176 struct bkey_i_subvolume *s;
179 if (k.k->type != KEY_TYPE_subvolume)
182 if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, parent) &&
183 le32_to_cpu(bkey_s_c_to_subvolume(k).v->parent) != old_parent)
186 s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume);
187 ret = PTR_ERR_OR_ZERO(s);
191 s->v.parent = cpu_to_le32(new_parent);
196 * Separate from the snapshot tree in the snapshots btree, we record the tree
197 * structure of how snapshot subvolumes were created - the parent subvolume of
198 * each snapshot subvolume.
200 * When a subvolume is deleted, we scan for child subvolumes and reparant them,
201 * to avoid dangling references:
203 static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
205 struct bch_subvolume s;
207 return lockrestart_do(trans,
208 bch2_subvolume_get(trans, subvolid_to_delete, true,
209 BTREE_ITER_CACHED, &s)) ?:
210 for_each_btree_key_commit(trans, iter,
211 BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
212 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
213 bch2_subvolume_reparent(trans, &iter, k,
214 subvolid_to_delete, le32_to_cpu(s.parent)));
218 * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
221 static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
223 struct btree_iter iter;
224 struct bkey_s_c_subvolume subvol;
228 subvol = bch2_bkey_get_iter_typed(trans, &iter,
229 BTREE_ID_subvolumes, POS(0, subvolid),
230 BTREE_ITER_CACHED|BTREE_ITER_INTENT,
232 ret = bkey_err(subvol);
233 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
234 "missing subvolume %u", subvolid);
238 snapid = le32_to_cpu(subvol.v->snapshot);
240 ret = bch2_btree_delete_at(trans, &iter, 0) ?:
241 bch2_snapshot_node_set_deleted(trans, snapid);
242 bch2_trans_iter_exit(trans, &iter);
246 static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
248 return bch2_subvolumes_reparent(trans, subvolid) ?:
249 commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
250 __bch2_subvolume_delete(trans, subvolid));
253 static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
255 struct bch_fs *c = container_of(work, struct bch_fs,
256 snapshot_wait_for_pagecache_and_delete_work);
262 mutex_lock(&c->snapshots_unlinked_lock);
263 s = c->snapshots_unlinked;
264 darray_init(&c->snapshots_unlinked);
265 mutex_unlock(&c->snapshots_unlinked_lock);
270 bch2_evict_subvolume_inodes(c, &s);
272 for (id = s.data; id < s.data + s.nr; id++) {
273 ret = bch2_trans_run(c, bch2_subvolume_delete(trans, *id));
274 bch_err_msg(c, ret, "deleting subvolume %u", *id);
282 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
285 struct subvolume_unlink_hook {
286 struct btree_trans_commit_hook h;
290 static int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
291 struct btree_trans_commit_hook *_h)
293 struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
294 struct bch_fs *c = trans->c;
297 mutex_lock(&c->snapshots_unlinked_lock);
298 if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
299 ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
300 mutex_unlock(&c->snapshots_unlinked_lock);
305 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
308 if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
309 bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
313 int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
315 struct btree_iter iter;
316 struct bkey_i_subvolume *n;
317 struct subvolume_unlink_hook *h;
320 h = bch2_trans_kmalloc(trans, sizeof(*h));
321 ret = PTR_ERR_OR_ZERO(h);
325 h->h.fn = bch2_subvolume_wait_for_pagecache_and_delete_hook;
326 h->subvol = subvolid;
327 bch2_trans_commit_hook(trans, &h->h);
329 n = bch2_bkey_get_mut_typed(trans, &iter,
330 BTREE_ID_subvolumes, POS(0, subvolid),
331 BTREE_ITER_CACHED, subvolume);
332 ret = PTR_ERR_OR_ZERO(n);
334 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
335 "missing subvolume %u", subvolid);
339 SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
340 bch2_trans_iter_exit(trans, &iter);
344 int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
350 struct bch_fs *c = trans->c;
351 struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
352 struct bkey_i_subvolume *new_subvol = NULL;
353 struct bkey_i_subvolume *src_subvol = NULL;
354 u32 parent = 0, new_nodes[2], snapshot_subvols[2];
357 ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
358 BTREE_ID_subvolumes, POS(0, U32_MAX));
359 if (ret == -BCH_ERR_ENOSPC_btree_slot)
360 ret = -BCH_ERR_ENOSPC_subvolume_create;
364 snapshot_subvols[0] = dst_iter.pos.offset;
365 snapshot_subvols[1] = src_subvolid;
368 /* Creating a snapshot: */
370 src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter,
371 BTREE_ID_subvolumes, POS(0, src_subvolid),
372 BTREE_ITER_CACHED, subvolume);
373 ret = PTR_ERR_OR_ZERO(src_subvol);
375 bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
376 "subvolume %u not found", src_subvolid);
380 parent = le32_to_cpu(src_subvol->v.snapshot);
383 ret = bch2_snapshot_node_create(trans, parent, new_nodes,
385 src_subvolid ? 2 : 1);
390 src_subvol->v.snapshot = cpu_to_le32(new_nodes[1]);
391 ret = bch2_trans_update(trans, &src_iter, &src_subvol->k_i, 0);
396 new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume);
397 ret = PTR_ERR_OR_ZERO(new_subvol);
401 new_subvol->v.flags = 0;
402 new_subvol->v.snapshot = cpu_to_le32(new_nodes[0]);
403 new_subvol->v.inode = cpu_to_le64(inode);
404 new_subvol->v.parent = cpu_to_le32(src_subvolid);
405 new_subvol->v.otime.lo = cpu_to_le64(bch2_current_time(c));
406 new_subvol->v.otime.hi = 0;
408 SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
409 SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
411 *new_subvolid = new_subvol->k.p.offset;
412 *new_snapshotid = new_nodes[0];
414 bch2_trans_iter_exit(trans, &src_iter);
415 bch2_trans_iter_exit(trans, &dst_iter);
419 int bch2_fs_subvolumes_init(struct bch_fs *c)
421 INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
422 INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
423 bch2_subvolume_wait_for_pagecache_and_delete);
424 mutex_init(&c->snapshots_unlinked_lock);