#include "bcachefs.h"
#include "btree_key_cache.h"
#include "btree_update.h"
+#include "errcode.h"
#include "error.h"
#include "fs.h"
#include "subvolume.h"
/* Snapshot tree: */
-static void bch2_delete_dead_snapshots_work(struct work_struct *);
-static void bch2_delete_dead_snapshots(struct bch_fs *);
-
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);
- pr_buf(out, "is_subvol %llu deleted %llu parent %u children %u %u subvol %u",
+ 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),
}
int bch2_snapshot_invalid(const struct bch_fs *c, struct bkey_s_c k,
- int rw, struct printbuf *err)
+ unsigned flags, 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) {
- pr_buf(err, "bad pos");
- return -EINVAL;
+ if (bkey_gt(k.k->p, POS(0, U32_MAX)) ||
+ bkey_lt(k.k->p, POS(0, 1))) {
+ prt_printf(err, "bad pos");
+ return -BCH_ERR_invalid_bkey;
}
if (bkey_val_bytes(k.k) != sizeof(struct bch_snapshot)) {
- pr_buf(err, "bad val size (%zu != %zu)",
+ prt_printf(err, "bad val size (%zu != %zu)",
bkey_val_bytes(k.k), sizeof(struct bch_snapshot));
- return -EINVAL;
+ return -BCH_ERR_invalid_bkey;
}
s = bkey_s_c_to_snapshot(k);
id = le32_to_cpu(s.v->parent);
if (id && id <= k.k->p.offset) {
- pr_buf(err, "bad parent node (%u <= %llu)",
+ prt_printf(err, "bad parent node (%u <= %llu)",
id, k.k->p.offset);
- return -EINVAL;
+ return -BCH_ERR_invalid_bkey;
}
if (le32_to_cpu(s.v->children[0]) < le32_to_cpu(s.v->children[1])) {
- pr_buf(err, "children not normalized");
- return -EINVAL;
+ prt_printf(err, "children not normalized");
+ return -BCH_ERR_invalid_bkey;
}
if (s.v->children[0] &&
s.v->children[0] == s.v->children[1]) {
- pr_buf(err, "duplicate child nodes");
- return -EINVAL;
+ prt_printf(err, "duplicate child nodes");
+ return -BCH_ERR_invalid_bkey;
}
for (i = 0; i < 2; i++) {
id = le32_to_cpu(s.v->children[i]);
if (id >= k.k->p.offset) {
- pr_buf(err, "bad child node (%u >= %llu)",
+ prt_printf(err, "bad child node (%u >= %llu)",
id, k.k->p.offset);
- return -EINVAL;
+ return -BCH_ERR_invalid_bkey;
}
}
if (!id)
return 0;
- ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
+ ret = snapshot_lookup(trans, id, &v);
if (ret == -ENOENT)
bch_err(trans->c, "snapshot node %u not found", id);
if (ret)
return !BCH_SNAPSHOT_DELETED(&v);
}
-static int bch2_snapshots_set_equiv(struct btree_trans *trans)
+static int bch2_snapshot_set_equiv(struct btree_trans *trans, struct bkey_s_c k)
{
struct bch_fs *c = trans->c;
- struct btree_iter iter;
- struct bkey_s_c k;
+ unsigned i, nr_live = 0, live_idx = 0;
struct bkey_s_c_snapshot snap;
- unsigned i;
- int ret;
+ u32 id = k.k->p.offset, child[2];
- for_each_btree_key(trans, iter, BTREE_ID_snapshots,
- POS_MIN, 0, k, ret) {
- u32 id = k.k->p.offset, child[2];
- unsigned nr_live = 0, live_idx = 0;
+ if (k.k->type != KEY_TYPE_snapshot)
+ return 0;
- if (k.k->type != KEY_TYPE_snapshot)
- continue;
+ snap = bkey_s_c_to_snapshot(k);
- 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]);
+ child[0] = le32_to_cpu(snap.v->children[0]);
+ child[1] = le32_to_cpu(snap.v->children[1]);
- for (i = 0; i < 2; i++) {
- ret = snapshot_live(trans, child[i]);
- if (ret < 0)
- goto err;
+ for (i = 0; i < 2; i++) {
+ int ret = snapshot_live(trans, child[i]);
- if (ret)
- live_idx = i;
- nr_live += ret;
- }
+ if (ret < 0)
+ return ret;
- snapshot_t(c, id)->equiv = nr_live == 1
- ? snapshot_t(c, child[live_idx])->equiv
- : id;
+ if (ret)
+ live_idx = i;
+ nr_live += ret;
}
-err:
- bch2_trans_iter_exit(trans, &iter);
-
- if (ret)
- bch_err(c, "error walking snapshots: %i", ret);
- return ret;
+ snapshot_t(c, id)->equiv = nr_live == 1
+ ? snapshot_t(c, child[live_idx])->equiv
+ : id;
+ return 0;
}
/* fsck: */
-static int bch2_snapshot_check(struct btree_trans *trans,
- struct bkey_s_c_snapshot s)
+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;
-
- id = le32_to_cpu(s.v->subvol);
- ret = lockrestart_do(trans, bch2_subvolume_get(trans, id, 0, false, &subvol));
- if (ret == -ENOENT)
- bch_err(trans->c, "snapshot node %llu has nonexistent subvolume %u",
- s.k->p.offset, id);
- if (ret)
- return ret;
+ int ret = 0;
- if (BCH_SNAPSHOT_SUBVOL(s.v) != (le32_to_cpu(subvol.snapshot) == s.k->p.offset)) {
- bch_err(trans->c, "snapshot node %llu has wrong BCH_SNAPSHOT_SUBVOL",
- s.k->p.offset);
- return -EINVAL;
- }
+ 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 = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
+ ret = snapshot_lookup(trans, id, &v);
if (ret == -ENOENT)
- bch_err(trans->c, "snapshot node %llu has nonexistent parent %u",
- s.k->p.offset, id);
+ bch_err(c, "snapshot with nonexistent parent:\n %s",
+ (bch2_bkey_val_to_text(&buf, c, s.s_c), buf.buf));
if (ret)
- return 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(trans->c, "snapshot parent %u missing pointer to child %llu",
+ bch_err(c, "snapshot parent %u missing pointer to child %llu",
id, s.k->p.offset);
- return -EINVAL;
+ ret = -EINVAL;
+ goto err;
}
}
for (i = 0; i < 2 && s.v->children[i]; i++) {
id = le32_to_cpu(s.v->children[i]);
- ret = lockrestart_do(trans, snapshot_lookup(trans, id, &v));
+ ret = snapshot_lookup(trans, id, &v);
if (ret == -ENOENT)
- bch_err(trans->c, "snapshot node %llu has nonexistent child %u",
+ bch_err(c, "snapshot node %llu has nonexistent child %u",
s.k->p.offset, id);
if (ret)
- return ret;
+ goto err;
if (le32_to_cpu(v.parent) != s.k->p.offset) {
- bch_err(trans->c, "snapshot child %u has wrong parent (got %u should be %llu)",
+ bch_err(c, "snapshot child %u has wrong parent (got %u should be %llu)",
id, le32_to_cpu(v.parent), s.k->p.offset);
- return -EINVAL;
+ ret = -EINVAL;
+ goto err;
}
}
- return 0;
+ 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_snapshots_check(struct bch_fs *c)
+int bch2_fs_check_snapshots(struct bch_fs *c)
{
struct btree_trans trans;
struct btree_iter iter;
struct bkey_s_c k;
- struct bch_snapshot s;
- unsigned id;
int ret;
bch2_trans_init(&trans, c, 0, 0);
- for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
- POS_MIN, 0, k, ret) {
- if (k.k->type != KEY_TYPE_snapshot)
- continue;
+ ret = for_each_btree_key_commit(&trans, iter,
+ BTREE_ID_snapshots, POS_MIN,
+ 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);
- ret = bch2_snapshot_check(&trans, bkey_s_c_to_snapshot(k));
+ bch2_trans_exit(&trans);
+ return ret;
+}
+
+static int check_subvol(struct btree_trans *trans,
+ struct btree_iter *iter,
+ struct bkey_s_c k)
+{
+ struct bkey_s_c_subvolume subvol;
+ struct bch_snapshot snapshot;
+ unsigned snapid;
+ int ret;
+
+ 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);
+
+ if (ret == -ENOENT)
+ bch_err(trans->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)
- break;
+ return ret;
}
- bch2_trans_iter_exit(&trans, &iter);
- if (ret) {
- bch_err(c, "error %i checking snapshots", ret);
- goto err;
- }
+ 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));
- for_each_btree_key(&trans, iter, BTREE_ID_subvolumes,
- POS_MIN, 0, k, ret) {
- if (k.k->type != KEY_TYPE_subvolume)
- continue;
-again_2:
- id = le32_to_cpu(bkey_s_c_to_subvolume(k).v->snapshot);
- ret = snapshot_lookup(&trans, id, &s);
-
- if (ret == -EINTR) {
- k = bch2_btree_iter_peek(&iter);
- goto again_2;
- } else if (ret == -ENOENT)
- bch_err(c, "subvolume %llu points to nonexistent snapshot %u",
- k.k->p.offset, id);
- else if (ret)
- break;
- }
- bch2_trans_iter_exit(&trans, &iter);
-err:
bch2_trans_exit(&trans);
+
return ret;
}
struct btree_trans trans;
struct btree_iter iter;
struct bkey_s_c k;
- bool have_deleted = false;
int ret = 0;
bch2_trans_init(&trans, c, 0, 0);
- for_each_btree_key(&trans, iter, BTREE_ID_snapshots,
- POS_MIN, 0, k, ret) {
- if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0)
- break;
-
- if (k.k->type != KEY_TYPE_snapshot) {
- bch_err(c, "found wrong key type %u in snapshot node table",
- k.k->type);
- continue;
- }
-
- if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
- have_deleted = true;
+ 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));
- ret = bch2_mark_snapshot(&trans, bkey_s_c_null, k, 0);
- if (ret)
- break;
- }
- bch2_trans_iter_exit(&trans, &iter);
-
- if (ret)
- goto err;
-
- ret = bch2_snapshots_set_equiv(&trans);
- if (ret)
- goto err;
-err:
bch2_trans_exit(&trans);
- if (!ret && have_deleted) {
- bch_info(c, "restarting deletion of dead snapshots");
- if (c->opts.fsck) {
- bch2_delete_dead_snapshots_work(&c->snapshot_delete_work);
- } else {
- bch2_delete_dead_snapshots(c);
- }
- }
-
+ if (ret)
+ bch_err(c, "error starting snapshots: %s", bch2_err_str(ret));
return ret;
}
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;
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;
+ s = bch2_bkey_get_mut_typed(trans, &iter, snapshot);
+ ret = PTR_ERR_OR_ZERO(s);
+ if (unlikely(ret)) {
+ bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing snapshot %u", id);
goto err;
}
/* already deleted? */
- if (BCH_SNAPSHOT_DELETED(bkey_s_c_to_snapshot(k).v))
+ if (BCH_SNAPSHOT_DELETED(&s->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;
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;
parent_id = le32_to_cpu(s.v->parent);
if (parent_id) {
+ struct bkey_i_snapshot *parent;
+
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));
+ parent = bch2_bkey_get_mut_typed(trans, &p_iter, snapshot);
ret = PTR_ERR_OR_ZERO(parent);
- if (ret)
+ if (unlikely(ret)) {
+ bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing snapshot %u", parent_id);
goto err;
-
- bkey_reassemble(&parent->k_i, k);
+ }
for (i = 0; i < 2; i++)
if (le32_to_cpu(parent->v.children[i]) == id)
goto err;
if (!k.k || !k.k->p.offset) {
- ret = -ENOSPC;
+ ret = -BCH_ERR_ENOSPC_snapshot_create;
goto err;
}
- n = bch2_trans_kmalloc(trans, sizeof(*n));
+ n = bch2_bkey_alloc(trans, &iter, snapshot);
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);
+ ret = bch2_trans_update(trans, &iter, &n->k_i, 0);
if (ret)
goto err;
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));
+ n = bch2_bkey_get_mut_typed(trans, &iter, snapshot);
ret = PTR_ERR_OR_ZERO(n);
- if (ret)
+ if (unlikely(ret)) {
+ if (ret == -ENOENT)
+ bch_err(trans->c, "snapshot %u not found", parent);
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");
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)
return ret;
}
-static int snapshot_id_add(snapshot_id_list *s, u32 id)
+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)
{
- BUG_ON(snapshot_list_has_id(s, id));
+ struct bch_fs *c = trans->c;
+ u32 equiv = snapshot_t(c, k.k->p.snapshot)->equiv;
- return darray_push(*s, id);
+ if (!bkey_eq(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_snapshot_delete_keys_btree(struct btree_trans *trans,
- snapshot_id_list *deleted,
- enum btree_id btree_id)
+static int bch2_delete_redundant_snapshot(struct btree_trans *trans, struct btree_iter *iter,
+ struct bkey_s_c k)
{
- struct bch_fs *c = trans->c;
- struct btree_iter iter;
- struct bkey_s_c k;
- snapshot_id_list equiv_seen = { 0 };
- struct bpos last_pos = POS_MIN;
- int ret = 0;
+ struct bkey_s_c_snapshot snap;
+ u32 children[2];
+ int ret;
- /*
- * XXX: We should also delete whiteouts that no longer overwrite
- * anything
- */
+ if (k.k->type != KEY_TYPE_snapshot)
+ return 0;
- bch2_trans_iter_init(trans, &iter, btree_id, POS_MIN,
- BTREE_ITER_INTENT|
- BTREE_ITER_PREFETCH|
- BTREE_ITER_NOT_EXTENTS|
- BTREE_ITER_ALL_SNAPSHOTS);
-
- while ((bch2_trans_begin(trans),
- (k = bch2_btree_iter_peek(&iter)).k) &&
- !(ret = bkey_err(k))) {
- 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)) {
- if (btree_id == BTREE_ID_inodes &&
- bch2_btree_key_cache_flush(trans, btree_id, iter.pos))
- continue;
-
- ret = __bch2_trans_do(trans, NULL, NULL,
- BTREE_INSERT_NOFAIL,
- bch2_btree_iter_traverse(&iter) ?:
- bch2_btree_delete_at(trans, &iter,
- BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE));
- if (ret)
- break;
- } else {
- ret = snapshot_id_add(&equiv_seen, equiv);
- if (ret)
- break;
- }
+ snap = bkey_s_c_to_snapshot(k);
+ if (BCH_SNAPSHOT_DELETED(snap.v) ||
+ BCH_SNAPSHOT_SUBVOL(snap.v))
+ return 0;
- bch2_btree_iter_advance(&iter);
- }
- bch2_trans_iter_exit(trans, &iter);
+ children[0] = le32_to_cpu(snap.v->children[0]);
+ children[1] = le32_to_cpu(snap.v->children[1]);
- darray_exit(equiv_seen);
+ ret = snapshot_live(trans, children[0]) ?:
+ snapshot_live(trans, children[1]);
+ if (ret < 0)
+ return ret;
- return ret;
+ if (!ret)
+ return bch2_snapshot_node_set_deleted(trans, k.k->p.offset);
+ return 0;
}
-static void bch2_delete_dead_snapshots_work(struct work_struct *work)
+int bch2_delete_dead_snapshots(struct bch_fs *c)
{
- struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
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, children[2];
+ 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:
*/
- 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) ||
- BCH_SNAPSHOT_SUBVOL(snap.v))
- continue;
-
- children[0] = le32_to_cpu(snap.v->children[0]);
- children[1] = le32_to_cpu(snap.v->children[1]);
-
- ret = snapshot_live(&trans, children[0]) ?:
- snapshot_live(&trans, children[1]);
- if (ret < 0)
- break;
- if (ret)
- continue;
-
- ret = __bch2_trans_do(&trans, NULL, NULL, 0,
- bch2_snapshot_node_set_deleted(&trans, iter.pos.offset));
- if (ret) {
- bch_err(c, "error deleting snapshot %llu: %i", iter.pos.offset, ret);
- break;
- }
- }
- bch2_trans_iter_exit(&trans, &iter);
-
+ 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 walking snapshots: %i", ret);
+ bch_err(c, "error deleting redundant snapshots: %s", bch2_err_str(ret));
goto err;
}
- ret = bch2_snapshots_set_equiv(&trans);
- if (ret)
+ 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) {
snap = bkey_s_c_to_snapshot(k);
if (BCH_SNAPSHOT_DELETED(snap.v)) {
- ret = snapshot_id_add(&deleted, k.k->p.offset);
+ 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: %i", 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 = bch2_snapshot_delete_keys_btree(&trans, &deleted, id);
+ 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: %i", ret);
+ bch_err(c, "error deleting snapshot keys: %s", bch2_err_str(ret));
goto err;
}
}
for (i = 0; i < deleted.nr; i++) {
- ret = __bch2_trans_do(&trans, NULL, NULL, 0,
+ ret = commit_do(&trans, NULL, NULL, 0,
bch2_snapshot_node_delete(&trans, deleted.data[i]));
if (ret) {
- bch_err(c, "error deleting snapshot %u: %i",
- deleted.data[i], 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);
+ darray_exit(&deleted);
bch2_trans_exit(&trans);
- percpu_ref_put(&c->writes);
+ return ret;
}
-static void bch2_delete_dead_snapshots(struct bch_fs *c)
+static void bch2_delete_dead_snapshots_work(struct work_struct *work)
{
- if (unlikely(!percpu_ref_tryget(&c->writes)))
- return;
+ struct bch_fs *c = container_of(work, struct bch_fs, snapshot_delete_work);
- if (!queue_work(system_long_wq, &c->snapshot_delete_work))
- percpu_ref_put(&c->writes);
+ bch2_delete_dead_snapshots(c);
+ bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
+}
+
+void bch2_delete_dead_snapshots_async(struct bch_fs *c)
+{
+ if (bch2_write_ref_tryget(c, BCH_WRITE_REF_delete_dead_snapshots) &&
+ !queue_work(system_long_wq, &c->snapshot_delete_work))
+ bch2_write_ref_put(c, BCH_WRITE_REF_delete_dead_snapshots);
}
static int bch2_delete_dead_snapshots_hook(struct btree_trans *trans,
struct btree_trans_commit_hook *h)
{
- bch2_delete_dead_snapshots(trans->c);
+ 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;
+
+ bch2_delete_dead_snapshots_async(c);
return 0;
}
/* Subvolumes: */
int bch2_subvolume_invalid(const struct bch_fs *c, struct bkey_s_c k,
- int rw, struct printbuf *err)
+ unsigned flags, struct printbuf *err)
{
- if (bkey_cmp(k.k->p, SUBVOL_POS_MIN) < 0 ||
- bkey_cmp(k.k->p, SUBVOL_POS_MAX) > 0) {
- pr_buf(err, "invalid pos");
- return -EINVAL;
+ if (bkey_lt(k.k->p, SUBVOL_POS_MIN) ||
+ bkey_gt(k.k->p, SUBVOL_POS_MAX)) {
+ prt_printf(err, "invalid pos");
+ return -BCH_ERR_invalid_bkey;
}
if (bkey_val_bytes(k.k) != sizeof(struct bch_subvolume)) {
- pr_buf(err, "incorrect value size (%zu != %zu)",
+ prt_printf(err, "incorrect value size (%zu != %zu)",
bkey_val_bytes(k.k), sizeof(struct bch_subvolume));
- return -EINVAL;
+ return -BCH_ERR_invalid_bkey;
}
return 0;
{
struct bkey_s_c_subvolume s = bkey_s_c_to_subvolume(k);
- pr_buf(out, "root %llu snapshot id %u",
+ prt_printf(out, "root %llu snapshot id %u",
le64_to_cpu(s.v->inode),
le32_to_cpu(s.v->snapshot));
}
-int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
- bool inconsistent_if_not_found,
- int iter_flags,
- struct bch_subvolume *s)
+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_iter iter;
struct bkey_s_c k;
return ret;
}
+int bch2_subvolume_get(struct btree_trans *trans, unsigned subvol,
+ bool inconsistent_if_not_found,
+ int iter_flags,
+ struct bch_subvolume *s)
+{
+ return bch2_subvolume_get_inlined(trans, subvol, inconsistent_if_not_found, iter_flags, s);
+}
+
int bch2_snapshot_get_subvol(struct btree_trans *trans, u32 snapshot,
struct bch_subvolume *subvol)
{
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);
+ ret = bch2_subvolume_get_inlined(trans, subvol, true,
+ BTREE_ITER_CACHED|
+ BTREE_ITER_WITH_UPDATES,
+ &s);
+ if (!ret)
+ *snapid = le32_to_cpu(s.snapshot);
return ret;
}
struct bkey_s_c k;
struct bkey_s_c_subvolume subvol;
struct btree_trans_commit_hook *h;
- struct bkey_i *delete;
u32 snapid;
int ret = 0;
subvol = bkey_s_c_to_subvolume(k);
snapid = le32_to_cpu(subvol.v->snapshot);
- delete = bch2_trans_kmalloc(trans, sizeof(*delete));
- ret = PTR_ERR_OR_ZERO(delete);
+ ret = bch2_btree_delete_at(trans, &iter, 0);
if (ret)
goto err;
- bkey_init(&delete->k);
- delete->k.p = iter.pos;
- ret = bch2_trans_update(trans, &iter, delete, 0);
+ ret = bch2_snapshot_node_set_deleted(trans, snapid);
if (ret)
goto err;
- ret = bch2_snapshot_node_set_deleted(trans, snapid);
-
h = bch2_trans_kmalloc(trans, sizeof(*h));
ret = PTR_ERR_OR_ZERO(h);
if (ret)
while (!ret) {
mutex_lock(&c->snapshots_unlinked_lock);
s = c->snapshots_unlinked;
- darray_init(c->snapshots_unlinked);
+ darray_init(&c->snapshots_unlinked);
mutex_unlock(&c->snapshots_unlinked_lock);
if (!s.nr)
ret = bch2_trans_do(c, NULL, NULL, BTREE_INSERT_NOFAIL,
bch2_subvolume_delete(&trans, *id));
if (ret) {
- bch_err(c, "error %i deleting subvolume %u", ret, *id);
+ bch_err(c, "error deleting subvolume %u: %s", *id, bch2_err_str(ret));
break;
}
}
- darray_exit(s);
+ darray_exit(&s);
}
- percpu_ref_put(&c->writes);
+ bch2_write_ref_put(c, BCH_WRITE_REF_snapshot_delete_pagecache);
}
struct subvolume_unlink_hook {
mutex_lock(&c->snapshots_unlinked_lock);
if (!snapshot_list_has_id(&c->snapshots_unlinked, h->subvol))
- ret = snapshot_id_add(&c->snapshots_unlinked, h->subvol);
+ ret = snapshot_list_add(c, &c->snapshots_unlinked, h->subvol);
mutex_unlock(&c->snapshots_unlinked_lock);
if (ret)
return ret;
- if (unlikely(!percpu_ref_tryget(&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);
+ 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;
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));
+ n = bch2_bkey_get_mut_typed(trans, &iter, subvolume);
ret = PTR_ERR_OR_ZERO(n);
- if (ret)
+ if (unlikely(ret)) {
+ bch2_fs_inconsistent_on(ret == -ENOENT, trans->c, "missing subvolume %u", subvolid);
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);
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)
+ if (bkey_gt(k.k->p, SUBVOL_POS_MAX))
break;
/*
}
if (!ret)
- ret = -ENOSPC;
+ ret = -BCH_ERR_ENOSPC_subvolume_create;
goto err;
found_slot:
snapshot_subvols[0] = dst_iter.pos.offset;
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, subvolume);
+ ret = PTR_ERR_OR_ZERO(src_subvol);
+ if (unlikely(ret)) {
+ bch2_fs_inconsistent_on(ret == -ENOENT, trans->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, 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);
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;