#include "errcode.h"
#include "error.h"
#include "fs.h"
+#include "snapshot.h"
#include "subvolume.h"
-/* Snapshot tree: */
+#include <linux/random.h>
-void bch2_snapshot_to_text(struct printbuf *out, struct bch_fs *c,
- struct bkey_s_c k)
-{
- struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(k);
-
- prt_printf(out, "is_subvol %llu deleted %llu parent %10u children %10u %10u subvol %u",
- BCH_SNAPSHOT_SUBVOL(s.v),
- BCH_SNAPSHOT_DELETED(s.v),
- le32_to_cpu(s.v->parent),
- le32_to_cpu(s.v->children[0]),
- le32_to_cpu(s.v->children[1]),
- le32_to_cpu(s.v->subvol));
-}
-
-int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
- int rw, struct printbuf *err)
-{
- struct bkey_s_c_snapshot s;
- u32 i, id;
-
- if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0 ||
- bkey_cmp(k.k->p, POS(0, 1)) < 0) {
- prt_printf(err, "bad pos");
- return -EINVAL;
- }
-
- if (bkey_val_bytes(k.k) != sizeof(struct bch_snapshot)) {
- prt_printf(err, "bad val size (%zu != %zu)",
- bkey_val_bytes(k.k), sizeof(struct bch_snapshot));
- return -EINVAL;
- }
-
- s = bkey_s_c_to_snapshot(k);
-
- id = le32_to_cpu(s.v->parent);
- if (id && id <= k.k->p.offset) {
- prt_printf(err, "bad parent node (%u <= %llu)",
- id, k.k->p.offset);
- return -EINVAL;
- }
-
- if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
- prt_printf(err, "children not normalized");
- return -EINVAL;
- }
-
- if (s.v->children[0] &&
- s.v->children[0] == s.v->children[1]) {
- prt_printf(err, "duplicate child nodes");
- return -EINVAL;
- }
-
- for (i = 0; i < 2; i++) {
- id = le32_to_cpu(s.v->children[i]);
-
- if (id >= k.k->p.offset) {
- prt_printf(err, "bad child node (%u >= %llu)",
- id, k.k->p.offset);
- return -EINVAL;
- }
- }
-
- return 0;
-}
-
-int bch2_mark_snapshot(struct btree_trans *trans,
- struct bkey_s_c old, struct bkey_s_c new,
- unsigned flags)
-{
- struct bch_fs *c = trans->c;
- struct snapshot_t *t;
-
- t = genradix_ptr_alloc(&c->snapshots,
- U32_MAX - new.k->p.offset,
- GFP_KERNEL);
- if (!t)
- return -ENOMEM;
-
- if (new.k->type == KEY_TYPE_snapshot) {
- struct bkey_s_c_snapshot s = bkey_s_c_to_snapshot(new);
-
- t->parent = le32_to_cpu(s.v->parent);
- t->children[0] = le32_to_cpu(s.v->children[0]);
- t->children[1] = le32_to_cpu(s.v->children[1]);
- t->subvol = BCH_SNAPSHOT_SUBVOL(s.v) ? le32_to_cpu(s.v->subvol) : 0;
- } else {
- t->parent = 0;
- t->children[0] = 0;
- t->children[1] = 0;
- t->subvol = 0;
- }
-
- return 0;
-}
-
-static int snapshot_lookup(struct btree_trans *trans, u32 id,
- struct bch_snapshot *s)
-{
- struct btree_iter iter;
- struct bkey_s_c k;
- int ret;
-
- bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
- BTREE_ITER_WITH_UPDATES);
- k = bch2_btree_iter_peek_slot(&iter);
- ret = bkey_err(k) ?: k.k->type == KEY_TYPE_snapshot ? 0 : -ENOENT;
-
- if (!ret)
- *s = *bkey_s_c_to_snapshot(k).v;
-
- bch2_trans_iter_exit(trans, &iter);
- return ret;
-}
-
-static int snapshot_live(struct btree_trans *trans, u32 id)
-{
- struct bch_snapshot v;
- int ret;
-
- if (!id)
- return 0;
-
- ret = snapshot_lookup(trans, id, &v);
- if (ret == -ENOENT)
- bch_err(trans->c, "snapshot node %u not found", id);
- if (ret)
- return ret;
-
- return !BCH_SNAPSHOT_DELETED(&v);
-}
-
-static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
-{
- struct bch_fs *c = trans->c;
- unsigned i, nr_live = 0, live_idx = 0;
- struct bkey_s_c_snapshot snap;
- u32 id = k.k->p.offset, child[2];
-
- if (k.k->type != KEY_TYPE_snapshot)
- return 0;
-
- snap = bkey_s_c_to_snapshot(k);
-
- child[0] = le32_to_cpu(snap.v->children[0]);
- child[1] = le32_to_cpu(snap.v->children[1]);
-
- for (i = 0; i < 2; i++) {
- int ret = snapshot_live(trans, child[i]);
- if (ret < 0)
- return ret;
-
- if (ret)
- live_idx = i;
- nr_live += ret;
- }
-
- snapshot_t(c, id)->equiv = nr_live == 1
- ? snapshot_t(c, child[live_idx])->equiv
- : id;
- return 0;
-}
-
-/* fsck: */
-static int check_snapshot(struct btree_trans *trans,
- struct btree_iter *iter,
- struct bkey_s_c k)
-{
- struct bch_fs *c = trans->c;
- struct bkey_s_c_snapshot s;
- struct bch_subvolume subvol;
- struct bch_snapshot v;
- struct printbuf buf = PRINTBUF;
- bool should_have_subvol;
- u32 i, id;
- int ret = 0;
-
- if (k.k->type != KEY_TYPE_snapshot)
- return 0;
-
- s = bkey_s_c_to_snapshot(k);
- id = le32_to_cpu(s.v->parent);
- if (id) {
- ret = snapshot_lookup(trans, id, &v);
- if (ret == -ENOENT)
- bch_err(c, "snapshot with nonexistent parent:\n %s",
- (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
- if (ret)
- goto err;
-
- if (le32_to_cpu(v.children[0]) != s.k->p.offset &&
- le32_to_cpu(v.children[1]) != s.k->p.offset) {
- bch_err(c, "snapshot parent %u missing pointer to child %llu",
- id, s.k->p.offset);
- ret = -EINVAL;
- goto err;
- }
- }
-
- for (i = 0; i < 2 && s.v->children[i]; i++) {
- id = le32_to_cpu(s.v->children[i]);
-
- ret = snapshot_lookup(trans, id, &v);
- if (ret == -ENOENT)
- bch_err(c, "snapshot node %llu has nonexistent child %u",
- s.k->p.offset, id);
- if (ret)
- goto err;
-
- if (le32_to_cpu(v.parent) != s.k->p.offset) {
- bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
- id, le32_to_cpu(v.parent), s.k->p.offset);
- ret = -EINVAL;
- goto err;
- }
- }
-
- should_have_subvol = BCH_SNAPSHOT_SUBVOL(s.v) &&
- !BCH_SNAPSHOT_DELETED(s.v);
-
- if (should_have_subvol) {
- id = le32_to_cpu(s.v->subvol);
- ret = bch2_subvolume_get(trans, id, 0, false, &subvol);
- if (ret == -ENOENT)
- bch_err(c, "snapshot points to nonexistent subvolume:\n %s",
- (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
- if (ret)
- goto err;
-
- if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
- bch_err(c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
- s.k->p.offset);
- ret = -EINVAL;
- goto err;
- }
- } else {
- if (fsck_err_on(s.v->subvol, c, "snapshot should not point to subvol:\n %s",
- (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf))) {
- struct bkey_i_snapshot *u = bch2_trans_kmalloc(trans, sizeof(*u));
-
- ret = PTR_ERR_OR_ZERO(u);
- if (ret)
- goto err;
-
- bkey_reassemble(&u->k_i, s.s_c);
- u->v.subvol = 0;
- ret = bch2_trans_update(trans, iter, &u->k_i, 0);
- if (ret)
- goto err;
- }
- }
-
- if (BCH_SNAPSHOT_DELETED(s.v))
- set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
-err:
-fsck_err:
- printbuf_exit(&buf);
- return ret;
-}
-
-int bch2_fs_check_snapshots(struct bch_fs *c)
-{
- struct btree_trans trans;
- struct btree_iter iter;
- struct bkey_s_c k;
- int ret;
-
- bch2_trans_init(&trans, c, 0, 0);
-
- ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_snapshots,
- POS(BCACHEFS_ROOT_INO, 0),
- BTREE_ITER_PREFETCH, k,
- NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
- check_snapshot(&trans, &iter, k));
-
- if (ret)
- bch_err(c, "error %i checking snapshots", ret);
-
- bch2_trans_exit(&trans);
- return ret;
-}
+static int bch2_subvolume_delete(struct btree_trans *, u32);
static int check_subvol(struct btree_trans *trans,
struct btree_iter *iter,
struct bkey_s_c k)
{
+ struct bch_fs *c = trans->c;
struct bkey_s_c_subvolume subvol;
struct bch_snapshot snapshot;
unsigned snapid;
- int ret;
+ int ret = 0;
if (k.k->type != KEY_TYPE_subvolume)
return 0;
subvol = bkey_s_c_to_subvolume(k);
snapid = le32_to_cpu(subvol.v->snapshot);
- ret = snapshot_lookup(trans, snapid, &snapshot);
+ ret = bch2_snapshot_lookup(trans, snapid, &snapshot);
- if (ret == -ENOENT)
- bch_err(trans->c, "subvolume %llu points to nonexistent snapshot %u",
+ if (bch2_err_matches(ret, ENOENT))
+ bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
k.k->p.offset, snapid);
if (ret)
return ret;
if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
ret = bch2_subvolume_delete(trans, iter->pos.offset);
- if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
- bch_err(trans->c, "error deleting subvolume %llu: %s",
- iter->pos.offset, bch2_err_str(ret));
- if (ret)
- return ret;
+ bch_err_msg(c, ret, "deleting subvolume %llu", iter->pos.offset);
+ return ret ?: -BCH_ERR_transaction_restart_nested;
}
- return 0;
-}
-
-int bch2_fs_check_subvols(struct bch_fs *c)
-{
- struct btree_trans trans;
- struct btree_iter iter;
- struct bkey_s_c k;
- int ret;
-
- bch2_trans_init(&trans, c, 0, 0);
-
- ret = for_each_btree_key_commit(&trans, iter,
- BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
- NULL, NULL, BTREE_INSERT_LAZY_RW|BTREE_INSERT_NOFAIL,
- check_subvol(&trans, &iter, k));
-
- bch2_trans_exit(&trans);
-
- return ret;
-}
+ if (!BCH_SUBVOLUME_SNAP(subvol.v)) {
+ u32 snapshot_root = bch2_snapshot_root(c, le32_to_cpu(subvol.v->snapshot));
+ u32 snapshot_tree;
+ struct bch_snapshot_tree st;
-void bch2_fs_snapshots_exit(struct bch_fs *c)
-{
- genradix_free(&c->snapshots);
-}
+ rcu_read_lock();
+ snapshot_tree = snapshot_t(c, snapshot_root)->tree;
+ rcu_read_unlock();
-int bch2_fs_snapshots_start(struct bch_fs *c)
-{
- struct btree_trans trans;
- struct btree_iter iter;
- struct bkey_s_c k;
- int ret = 0;
+ ret = bch2_snapshot_tree_lookup(trans, snapshot_tree, &st);
- bch2_trans_init(&trans, c, 0, 0);
-
- for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
- POS_MIN, 0, k,
- bch2_mark_snapshot(&trans, bkey_s_c_null, k, 0) ?:
- bch2_snapshot_set_equiv(&trans, k));
-
- bch2_trans_exit(&trans);
-
- if (ret)
- bch_err(c, "error starting snapshots: %s", bch2_err_str(ret));
- return ret;
-}
+ bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
+ "%s: snapshot tree %u not found", __func__, snapshot_tree);
-/*
- * Mark a snapshot as deleted, for future cleanup:
- */
-static int bch2_snapshot_node_set_deleted(struct btree_trans *trans, u32 id)
-{
- struct btree_iter iter;
- struct bkey_s_c k;
- struct bkey_i_snapshot *s;
- int ret = 0;
+ if (ret)
+ return ret;
- bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
- BTREE_ITER_INTENT);
- k = bch2_btree_iter_peek_slot(&iter);
- ret = bkey_err(k);
- if (ret)
- goto err;
+ if (fsck_err_on(le32_to_cpu(st.master_subvol) != subvol.k->p.offset,
+ c, subvol_not_master_and_not_snapshot,
+ "subvolume %llu is not set as snapshot but is not master subvolume",
+ k.k->p.offset)) {
+ struct bkey_i_subvolume *s =
+ bch2_bkey_make_mut_typed(trans, iter, &subvol.s_c, 0, subvolume);
+ ret = PTR_ERR_OR_ZERO(s);
+ if (ret)
+ return ret;
- if (k.k->type != KEY_TYPE_snapshot) {
- bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
- ret = -ENOENT;
- goto err;
+ SET_BCH_SUBVOLUME_SNAP(&s->v, true);
+ }
}
- /* already deleted? */
- if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
- goto err;
-
- s = bch2_trans_kmalloc(trans, sizeof(*s));
- ret = PTR_ERR_OR_ZERO(s);
- if (ret)
- goto err;
-
- bkey_reassemble(&s->k_i, k);
- SET_BCH_SNAPSHOT_DELETED(&s->v, true);
- SET_BCH_SNAPSHOT_SUBVOL(&s->v, false);
- s->v.subvol = 0;
-
- ret = bch2_trans_update(trans, &iter, &s->k_i, 0);
- if (ret)
- goto err;
-err:
- bch2_trans_iter_exit(trans, &iter);
+fsck_err:
return ret;
}
-static int bch2_snapshot_node_delete(struct btree_trans *trans, u32 id)
+int bch2_check_subvols(struct bch_fs *c)
{
- struct btree_iter iter, p_iter = (struct btree_iter) { NULL };
- struct bkey_s_c k;
- struct bkey_s_c_snapshot s;
- struct bkey_i_snapshot *parent;
- u32 parent_id;
- unsigned i;
- int ret = 0;
-
- bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots, POS(0, id),
- BTREE_ITER_INTENT);
- k = bch2_btree_iter_peek_slot(&iter);
- ret = bkey_err(k);
- if (ret)
- goto err;
-
- if (k.k->type != KEY_TYPE_snapshot) {
- bch2_fs_inconsistent(trans->c, "missing snapshot %u", id);
- ret = -ENOENT;
- goto err;
- }
-
- s = bkey_s_c_to_snapshot(k);
-
- BUG_ON(!BCH_SNAPSHOT_DELETED(s.v));
- parent_id = le32_to_cpu(s.v->parent);
-
- if (parent_id) {
- bch2_trans_iter_init(trans, &p_iter, BTREE_ID_snapshots,
- POS(0, parent_id),
- BTREE_ITER_INTENT);
- k = bch2_btree_iter_peek_slot(&p_iter);
- ret = bkey_err(k);
- if (ret)
- goto err;
-
- if (k.k->type != KEY_TYPE_snapshot) {
- bch2_fs_inconsistent(trans->c, "missing snapshot %u", parent_id);
- ret = -ENOENT;
- goto err;
- }
-
- parent = bch2_trans_kmalloc(trans, sizeof(*parent));
- ret = PTR_ERR_OR_ZERO(parent);
- if (ret)
- goto err;
-
- bkey_reassemble(&parent->k_i, k);
-
- for (i = 0; i < 2; i++)
- if (le32_to_cpu(parent->v.children[i]) == id)
- break;
-
- if (i == 2)
- bch_err(trans->c, "snapshot %u missing child pointer to %u",
- parent_id, id);
- else
- parent->v.children[i] = 0;
-
- if (le32_to_cpu(parent->v.children[0]) <
- le32_to_cpu(parent->v.children[1]))
- swap(parent->v.children[0],
- parent->v.children[1]);
-
- ret = bch2_trans_update(trans, &p_iter, &parent->k_i, 0);
- if (ret)
- goto err;
- }
-
- ret = bch2_btree_delete_at(trans, &iter, 0);
-err:
- bch2_trans_iter_exit(trans, &p_iter);
- bch2_trans_iter_exit(trans, &iter);
+ int ret = bch2_trans_run(c,
+ for_each_btree_key_commit(trans, iter,
+ BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
+ NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+ check_subvol(trans, &iter, k)));
+ bch_err_fn(c, ret);
return ret;
}
-int bch2_snapshot_node_create(struct btree_trans *trans, u32 parent,
- u32 *new_snapids,
- u32 *snapshot_subvols,
- unsigned nr_snapids)
+/* Subvolumes: */
+
+int bch2_subvolume_invalid(struct bch_fs *c, struct bkey_s_c k,
+ enum bkey_invalid_flags flags, struct printbuf *err)
{
- struct btree_iter iter;
- struct bkey_i_snapshot *n;
- struct bkey_s_c k;
- unsigned i;
int ret = 0;
- bch2_trans_iter_init(trans, &iter, BTREE_ID_snapshots,
- POS_MIN, BTREE_ITER_INTENT);
- k = bch2_btree_iter_peek(&iter);
- ret = bkey_err(k);
- if (ret)
- goto err;
-
- for (i = 0; i < nr_snapids; i++) {
- k = bch2_btree_iter_prev_slot(&iter);
- ret = bkey_err(k);
- if (ret)
- goto err;
-
- if (!k.k || !k.k->p.offset) {
- ret = -ENOSPC;
- goto err;
- }
-
- n = bch2_trans_kmalloc(trans, sizeof(*n));
- ret = PTR_ERR_OR_ZERO(n);
- if (ret)
- goto err;
-
- bkey_snapshot_init(&n->k_i);
- n->k.p = iter.pos;
- n->v.flags = 0;
- n->v.parent = cpu_to_le32(parent);
- n->v.subvol = cpu_to_le32(snapshot_subvols[i]);
- n->v.pad = 0;
- SET_BCH_SNAPSHOT_SUBVOL(&n->v, true);
-
- ret = bch2_trans_update(trans, &iter, &n->k_i, 0) ?:
- bch2_mark_snapshot(trans, bkey_s_c_null, bkey_i_to_s_c(&n->k_i), 0);
- if (ret)
- goto err;
-
- new_snapids[i] = iter.pos.offset;
- }
-
- if (parent) {
- bch2_btree_iter_set_pos(&iter, POS(0, parent));
- k = bch2_btree_iter_peek(&iter);
- ret = bkey_err(k);
- if (ret)
- goto err;
-
- if (k.k->type != KEY_TYPE_snapshot) {
- bch_err(trans->c, "snapshot %u not found", parent);
- ret = -ENOENT;
- goto err;
- }
-
- n = bch2_trans_kmalloc(trans, sizeof(*n));
- ret = PTR_ERR_OR_ZERO(n);
- if (ret)
- goto err;
-
- bkey_reassemble(&n->k_i, k);
-
- if (n->v.children[0] || n->v.children[1]) {
- bch_err(trans->c, "Trying to add child snapshot nodes to parent that already has children");
- ret = -EINVAL;
- goto err;
- }
-
- n->v.children[0] = cpu_to_le32(new_snapids[0]);
- n->v.children[1] = cpu_to_le32(new_snapids[1]);
- n->v.subvol = 0;
- SET_BCH_SNAPSHOT_SUBVOL(&n->v, false);
- ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
- if (ret)
- goto err;
- }
-err:
- bch2_trans_iter_exit(trans, &iter);
+ bkey_fsck_err_on(bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
+ bkey_gt(k.k->p, SUBVOL_POS_MAX), c, err,
+ subvol_pos_bad,
+ "invalid pos");
+fsck_err:
return ret;
}
-static int snapshot_delete_key(struct btree_trans *trans,
- struct btree_iter *iter,
- struct bkey_s_c k,
- snapshot_id_list *deleted,
- snapshot_id_list *equiv_seen,
- struct bpos *last_pos)
-{
- struct bch_fs *c = trans->c;
- u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
-
- if (bkey_cmp(k.k->p, *last_pos))
- equiv_seen->nr = 0;
- *last_pos = k.k->p;
-
- if (snapshot_list_has_id(deleted, k.k->p.snapshot) ||
- snapshot_list_has_id(equiv_seen, equiv)) {
- return bch2_btree_delete_at(trans, iter,
- BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
- } else {
- return snapshot_list_add(c, equiv_seen, equiv);
- }
-}
-
-static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btree_iter *iter,
- struct bkey_s_c k)
+void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
+ struct bkey_s_c k)
{
- struct bkey_s_c_snapshot snap;
- u32 children[2];
- int ret;
-
- if (k.k->type != KEY_TYPE_snapshot)
- return 0;
-
- snap = bkey_s_c_to_snapshot(k);
- if (BCH_SNAPSHOT_DELETED(snap.v) ||
- BCH_SNAPSHOT_SUBVOL(snap.v))
- return 0;
-
- children[0] = le32_to_cpu(snap.v->children[0]);
- children[1] = le32_to_cpu(snap.v->children[1]);
+ struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
- ret = snapshot_live(trans, children[0]) ?:
- snapshot_live(trans, children[1]);
- if (ret < 0)
- return ret;
+ prt_printf(out, "root %llu snapshot id %u",
+ le64_to_cpu(s.v->inode),
+ le32_to_cpu(s.v->snapshot));
- if (!ret)
- return bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
- return 0;
+ if (bkey_val_bytes(s.k) > offsetof(struct bch_subvolume, parent))
+ prt_printf(out, " parent %u", le32_to_cpu(s.v->parent));
}
-int bch2_delete_dead_snapshots(struct bch_fs *c)
+static __always_inline int
+bch2_subvolume_get_inlined(struct btree_trans *trans, unsigned subvol,
+ bool inconsistent_if_not_found,
+ int iter_flags,
+ struct bch_subvolume *s)
{
- struct btree_trans trans;
- struct btree_iter iter;
- struct bkey_s_c k;
- struct bkey_s_c_snapshot snap;
- snapshot_id_list deleted = { 0 };
- u32 i, id;
- int ret = 0;
-
- if (!test_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags))
- return 0;
-
- if (!test_bit(BCH_FS_STARTED, &c->flags)) {
- ret = bch2_fs_read_write_early(c);
- if (ret) {
- bch_err(c, "error deleleting dead snapshots: error going rw: %s", bch2_err_str(ret));
- return ret;
- }
- }
-
- bch2_trans_init(&trans, c, 0, 0);
-
- /*
- * For every snapshot node: If we have no live children and it's not
- * pointed to by a subvolume, delete it:
- */
- ret = for_each_btree_key_commit(&trans, iter, BTREE_ID_snapshots,
- POS_MIN, 0, k,
- NULL, NULL, 0,
- bch2_delete_redundant_snapshot(&trans, &iter, k));
- if (ret) {
- bch_err(c, "error deleting redundant snapshots: %s", bch2_err_str(ret));
- goto err;
- }
-
- for_each_btree_key2(&trans, iter, BTREE_ID_snapshots,
- POS_MIN, 0, k,
- bch2_snapshot_set_equiv(&trans, k));
- if (ret) {
- bch_err(c, "error in bch2_snapshots_set_equiv: %s", bch2_err_str(ret));
- goto err;
- }
-
- for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
- POS_MIN, 0, k, ret) {
- if (k.k->type != KEY_TYPE_snapshot)
- continue;
-
- snap = bkey_s_c_to_snapshot(k);
- if (BCH_SNAPSHOT_DELETED(snap.v)) {
- ret = snapshot_list_add(c, &deleted, k.k->p.offset);
- if (ret)
- break;
- }
- }
- bch2_trans_iter_exit(&trans, &iter);
-
- if (ret) {
- bch_err(c, "error walking snapshots: %s", bch2_err_str(ret));
- goto err;
- }
-
- for (id = 0; id < BTREE_ID_NR; id++) {
- struct bpos last_pos = POS_MIN;
- snapshot_id_list equiv_seen = { 0 };
-
- if (!btree_type_has_snapshots(id))
- continue;
-
- ret = for_each_btree_key_commit(&trans, iter,
- id, POS_MIN,
- BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
- NULL, NULL, BTREE_INSERT_NOFAIL,
- snapshot_delete_key(&trans, &iter, k, &deleted, &equiv_seen, &last_pos));
-
- darray_exit(&equiv_seen);
-
- if (ret) {
- bch_err(c, "error deleting snapshot keys: %s", bch2_err_str(ret));
- goto err;
- }
- }
-
- for (i = 0; i < deleted.nr; i++) {
- ret = commit_do(&trans, NULL, NULL, 0,
- bch2_snapshot_node_delete(&trans, deleted.data[i]));
- if (ret) {
- bch_err(c, "error deleting snapshot %u: %s",
- deleted.data[i], bch2_err_str(ret));
- goto err;
- }
- }
-
- clear_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
-err:
- darray_exit(&deleted);
- bch2_trans_exit(&trans);
+ int ret = bch2_bkey_get_val_typed(trans, BTREE_ID_subvolumes, POS(0, subvol),
+ iter_flags, subvolume, s);
+ bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT) &&
+ inconsistent_if_not_found,
+ trans->c, "missing subvolume %u", subvol);
return ret;
}
-static void bch2_delete_dead_snapshots_work(struct work_struct *work)
-{
- struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
-
- bch2_delete_dead_snapshots(c);
- percpu_ref_put(&c->writes);
-}
-
-void bch2_delete_dead_snapshots_async(struct bch_fs *c)
+int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
+ bool inconsistent_if_not_found,
+ int iter_flags,
+ struct bch_subvolume *s)
{
- if (!percpu_ref_tryget_live(&c->writes))
- return;
-
- if (!queue_work(system_long_wq, &c->snapshot_delete_work))
- percpu_ref_put(&c->writes);
+ return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
}
-static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
- struct btree_trans_commit_hook *h)
+int bch2_subvol_is_ro_trans(struct btree_trans *trans, u32 subvol)
{
- struct bch_fs *c = trans->c;
-
- set_bit(BCH_FS_HAVE_DELETED_SNAPSHOTS, &c->flags);
-
- if (!test_bit(BCH_FS_FSCK_DONE, &c->flags))
- return 0;
+ struct bch_subvolume s;
+ int ret = bch2_subvolume_get_inlined(trans, subvol, true, 0, &s);
+ if (ret)
+ return ret;
- bch2_delete_dead_snapshots_async(c);
+ if (BCH_SUBVOLUME_RO(&s))
+ return -EROFS;
return 0;
}
-/* Subvolumes: */
-
-int bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k,
- int rw, struct printbuf *err)
+int bch2_subvol_is_ro(struct bch_fs *c, u32 subvol)
{
- if (bkey_cmp(k.k->p, SUBVOL_POS_MIN) < 0 ||
- bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0) {
- prt_printf(err, "invalid pos");
- return -EINVAL;
- }
-
- if (bkey_val_bytes(k.k) != sizeof(struct bch_subvolume)) {
- prt_printf(err, "incorrect value size (%zu != %zu)",
- bkey_val_bytes(k.k), sizeof(struct bch_subvolume));
- return -EINVAL;
- }
-
- return 0;
+ return bch2_trans_do(c, NULL, NULL, 0,
+ bch2_subvol_is_ro_trans(trans, subvol));
}
-void bch2_subvolume_to_text(struct printbuf *out, struct bch_fs *c,
- struct bkey_s_c k)
+int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
+ struct bch_subvolume *subvol)
{
- struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
+ struct bch_snapshot snap;
- prt_printf(out, "root %llu snapshot id %u",
- le64_to_cpu(s.v->inode),
- le32_to_cpu(s.v->snapshot));
+ return bch2_snapshot_lookup(trans, snapshot, &snap) ?:
+ bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
}
-int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
- bool inconsistent_if_not_found,
- int iter_flags,
- struct bch_subvolume *s)
+int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvolid,
+ u32 *snapid)
{
struct btree_iter iter;
- struct bkey_s_c k;
+ struct bkey_s_c_subvolume subvol;
int ret;
- bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes, POS(0, subvol),
- iter_flags);
- k = bch2_btree_iter_peek_slot(&iter);
- ret = bkey_err(k) ?: k.k->type == KEY_TYPE_subvolume ? 0 : -ENOENT;
-
- if (ret == -ENOENT && inconsistent_if_not_found)
- bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvol);
- if (!ret)
- *s = *bkey_s_c_to_subvolume(k).v;
+ subvol = bch2_bkey_get_iter_typed(trans, &iter,
+ BTREE_ID_subvolumes, POS(0, subvolid),
+ BTREE_ITER_CACHED|BTREE_ITER_WITH_UPDATES,
+ subvolume);
+ ret = bkey_err(subvol);
+ bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
+ "missing subvolume %u", subvolid);
+ if (likely(!ret))
+ *snapid = le32_to_cpu(subvol.v->snapshot);
bch2_trans_iter_exit(trans, &iter);
return ret;
}
-int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
- struct bch_subvolume *subvol)
+static int bch2_subvolume_reparent(struct btree_trans *trans,
+ struct btree_iter *iter,
+ struct bkey_s_c k,
+ u32 old_parent, u32 new_parent)
{
- struct bch_snapshot snap;
+ struct bkey_i_subvolume *s;
+ int ret;
- return snapshot_lookup(trans, snapshot, &snap) ?:
- bch2_subvolume_get(trans, le32_to_cpu(snap.subvol), true, 0, subvol);
+ if (k.k->type != KEY_TYPE_subvolume)
+ return 0;
+
+ if (bkey_val_bytes(k.k) > offsetof(struct bch_subvolume, parent) &&
+ le32_to_cpu(bkey_s_c_to_subvolume(k).v->parent) != old_parent)
+ return 0;
+
+ s = bch2_bkey_make_mut_typed(trans, iter, &k, 0, subvolume);
+ ret = PTR_ERR_OR_ZERO(s);
+ if (ret)
+ return ret;
+
+ s->v.parent = cpu_to_le32(new_parent);
+ return 0;
}
-int bch2_subvolume_get_snapshot(struct btree_trans *trans, u32 subvol,
- u32 *snapid)
+/*
+ * Separate from the snapshot tree in the snapshots btree, we record the tree
+ * structure of how snapshot subvolumes were created - the parent subvolume of
+ * each snapshot subvolume.
+ *
+ * When a subvolume is deleted, we scan for child subvolumes and reparant them,
+ * to avoid dangling references:
+ */
+static int bch2_subvolumes_reparent(struct btree_trans *trans, u32 subvolid_to_delete)
{
struct bch_subvolume s;
- int ret;
- ret = bch2_subvolume_get(trans, subvol, true,
- BTREE_ITER_CACHED|
- BTREE_ITER_WITH_UPDATES,
- &s);
-
- *snapid = le32_to_cpu(s.snapshot);
- return ret;
+ return lockrestart_do(trans,
+ bch2_subvolume_get(trans, subvolid_to_delete, true,
+ BTREE_ITER_CACHED, &s)) ?:
+ for_each_btree_key_commit(trans, iter,
+ BTREE_ID_subvolumes, POS_MIN, BTREE_ITER_PREFETCH, k,
+ NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+ bch2_subvolume_reparent(trans, &iter, k,
+ subvolid_to_delete, le32_to_cpu(s.parent)));
}
/*
* Delete subvolume, mark snapshot ID as deleted, queue up snapshot
* deletion/cleanup:
*/
-int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
+static int __bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
{
struct btree_iter iter;
- struct bkey_s_c k;
struct bkey_s_c_subvolume subvol;
- struct btree_trans_commit_hook *h;
u32 snapid;
int ret = 0;
- bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
- POS(0, subvolid),
- BTREE_ITER_CACHED|
- BTREE_ITER_INTENT);
- k = bch2_btree_iter_peek_slot(&iter);
- ret = bkey_err(k);
+ subvol = bch2_bkey_get_iter_typed(trans, &iter,
+ BTREE_ID_subvolumes, POS(0, subvolid),
+ BTREE_ITER_CACHED|BTREE_ITER_INTENT,
+ subvolume);
+ ret = bkey_err(subvol);
+ bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
+ "missing subvolume %u", subvolid);
if (ret)
- goto err;
-
- if (k.k->type != KEY_TYPE_subvolume) {
- bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
- ret = -EIO;
- goto err;
- }
+ return ret;
- subvol = bkey_s_c_to_subvolume(k);
snapid = le32_to_cpu(subvol.v->snapshot);
- ret = bch2_btree_delete_at(trans, &iter, 0);
- if (ret)
- goto err;
-
- ret = bch2_snapshot_node_set_deleted(trans, snapid);
- if (ret)
- goto err;
-
- h = bch2_trans_kmalloc(trans, sizeof(*h));
- ret = PTR_ERR_OR_ZERO(h);
- if (ret)
- goto err;
-
- h->fn = bch2_delete_dead_snapshots_hook;
- bch2_trans_commit_hook(trans, h);
-err:
+ ret = bch2_btree_delete_at(trans, &iter, 0) ?:
+ bch2_snapshot_node_set_deleted(trans, snapid);
bch2_trans_iter_exit(trans, &iter);
return ret;
}
-void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
+static int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
+{
+ return bch2_subvolumes_reparent(trans, subvolid) ?:
+ commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
+ __bch2_subvolume_delete(trans, subvolid));
+}
+
+static void bch2_subvolume_wait_for_pagecache_and_delete(struct work_struct *work)
{
struct bch_fs *c = container_of(work, struct bch_fs,
snapshot_wait_for_pagecache_and_delete_work);
bch2_evict_subvolume_inodes(c, &s);
for (id = s.data; id < s.data + s.nr; id++) {
- ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL,
- bch2_subvolume_delete(&trans, *id));
- if (ret) {
- bch_err(c, "error deleting subvolume %u: %s", *id, bch2_err_str(ret));
+ ret = bch2_trans_run(c, bch2_subvolume_delete(trans, *id));
+ bch_err_msg(c, ret, "deleting subvolume %u", *id);
+ if (ret)
break;
- }
}
darray_exit(&s);
}
- percpu_ref_put(&c->writes);
+ bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
}
struct subvolume_unlink_hook {
u32 subvol;
};
-int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
+static int bch2_subvolume_wait_for_pagecache_and_delete_hook(struct btree_trans *trans,
struct btree_trans_commit_hook *_h)
{
struct subvolume_unlink_hook *h = container_of(_h, struct subvolume_unlink_hook, h);
if (ret)
return ret;
- if (unlikely(!percpu_ref_tryget_live(&c->writes)))
+ if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_snapshot_delete_pagecache))
return -EROFS;
- if (!queue_work(system_long_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
- percpu_ref_put(&c->writes);
+ if (!queue_work(c->write_ref_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
+ bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
return 0;
}
int bch2_subvolume_unlink(struct btree_trans *trans, u32 subvolid)
{
struct btree_iter iter;
- struct bkey_s_c k;
struct bkey_i_subvolume *n;
struct subvolume_unlink_hook *h;
int ret = 0;
- bch2_trans_iter_init(trans, &iter, BTREE_ID_subvolumes,
- POS(0, subvolid),
- BTREE_ITER_CACHED|
- BTREE_ITER_INTENT);
- k = bch2_btree_iter_peek_slot(&iter);
- ret = bkey_err(k);
- if (ret)
- goto err;
-
- if (k.k->type != KEY_TYPE_subvolume) {
- bch2_fs_inconsistent(trans->c, "missing subvolume %u", subvolid);
- ret = -EIO;
- goto err;
- }
-
- n = bch2_trans_kmalloc(trans, sizeof(*n));
- ret = PTR_ERR_OR_ZERO(n);
- if (ret)
- goto err;
-
- bkey_reassemble(&n->k_i, k);
- SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
-
- ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
- if (ret)
- goto err;
-
h = bch2_trans_kmalloc(trans, sizeof(*h));
ret = PTR_ERR_OR_ZERO(h);
if (ret)
- goto err;
+ return ret;
h->h.fn = bch2_subvolume_wait_for_pagecache_and_delete_hook;
h->subvol = subvolid;
bch2_trans_commit_hook(trans, &h->h);
-err:
+
+ n = bch2_bkey_get_mut_typed(trans, &iter,
+ BTREE_ID_subvolumes, POS(0, subvolid),
+ BTREE_ITER_CACHED, subvolume);
+ ret = PTR_ERR_OR_ZERO(n);
+ if (unlikely(ret)) {
+ bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), trans->c,
+ "missing subvolume %u", subvolid);
+ return ret;
+ }
+
+ SET_BCH_SUBVOLUME_UNLINKED(&n->v, true);
bch2_trans_iter_exit(trans, &iter);
return ret;
}
struct btree_iter dst_iter, src_iter = (struct btree_iter) { NULL };
struct bkey_i_subvolume *new_subvol = NULL;
struct bkey_i_subvolume *src_subvol = NULL;
- struct bkey_s_c k;
u32 parent = 0, new_nodes[2], snapshot_subvols[2];
int ret = 0;
- for_each_btree_key(trans, dst_iter, BTREE_ID_subvolumes, SUBVOL_POS_MIN,
- BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
- if (bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0)
- break;
-
- /*
- * bch2_subvolume_delete() doesn't flush the btree key cache -
- * ideally it would but that's tricky
- */
- if (bkey_deleted(k.k) &&
- !bch2_btree_key_cache_find(c, BTREE_ID_subvolumes, dst_iter.pos))
- goto found_slot;
- }
+ ret = bch2_bkey_get_empty_slot(trans, &dst_iter,
+ BTREE_ID_subvolumes, POS(0, U32_MAX));
+ if (ret == -BCH_ERR_ENOSPC_btree_slot)
+ ret = -BCH_ERR_ENOSPC_subvolume_create;
+ if (ret)
+ return ret;
- if (!ret)
- ret = -ENOSPC;
- goto err;
-found_slot:
snapshot_subvols[0] = dst_iter.pos.offset;
snapshot_subvols[1] = src_subvolid;
if (src_subvolid) {
/* Creating a snapshot: */
- src_subvol = bch2_trans_kmalloc(trans, sizeof(*src_subvol));
- ret = PTR_ERR_OR_ZERO(src_subvol);
- if (ret)
- goto err;
-
- bch2_trans_iter_init(trans, &src_iter, BTREE_ID_subvolumes,
- POS(0, src_subvolid),
- BTREE_ITER_CACHED|
- BTREE_ITER_INTENT);
- k = bch2_btree_iter_peek_slot(&src_iter);
- ret = bkey_err(k);
- if (ret)
- goto err;
- if (k.k->type != KEY_TYPE_subvolume) {
- bch_err(c, "subvolume %u not found", src_subvolid);
- ret = -ENOENT;
+ src_subvol = bch2_bkey_get_mut_typed(trans, &src_iter,
+ BTREE_ID_subvolumes, POS(0, src_subvolid),
+ BTREE_ITER_CACHED, subvolume);
+ ret = PTR_ERR_OR_ZERO(src_subvol);
+ if (unlikely(ret)) {
+ bch2_fs_inconsistent_on(bch2_err_matches(ret, ENOENT), c,
+ "subvolume %u not found", src_subvolid);
goto err;
}
- bkey_reassemble(&src_subvol->k_i, k);
parent = le32_to_cpu(src_subvol->v.snapshot);
}
goto err;
}
- new_subvol = bch2_trans_kmalloc(trans, sizeof(*new_subvol));
+ new_subvol = bch2_bkey_alloc(trans, &dst_iter, 0, subvolume);
ret = PTR_ERR_OR_ZERO(new_subvol);
if (ret)
goto err;
- bkey_subvolume_init(&new_subvol->k_i);
new_subvol->v.flags = 0;
new_subvol->v.snapshot = cpu_to_le32(new_nodes[0]);
new_subvol->v.inode = cpu_to_le64(inode);
+ new_subvol->v.parent = cpu_to_le32(src_subvolid);
+ new_subvol->v.otime.lo = cpu_to_le64(bch2_current_time(c));
+ new_subvol->v.otime.hi = 0;
+
SET_BCH_SUBVOLUME_RO(&new_subvol->v, ro);
SET_BCH_SUBVOLUME_SNAP(&new_subvol->v, src_subvolid != 0);
- new_subvol->k.p = dst_iter.pos;
- ret = bch2_trans_update(trans, &dst_iter, &new_subvol->k_i, 0);
- if (ret)
- goto err;
*new_subvolid = new_subvol->k.p.offset;
*new_snapshotid = new_nodes[0];
projects / bcachefs-tools-debian / blobdiff