-4c58f10d4897ff15849e6c028a9f665e03b8d4c0
+b1899a0bd9af8040b592cfdfe2df3c0c1869b3bb
#include "quota_types.h"
#include "rebalance_types.h"
#include "replicas_types.h"
+#include "subvolume_types.h"
#include "super_types.h"
/* Number of nodes btree coalesce will try to coalesce at once */
struct bch_snapshot_table __rcu *snapshot_table;
struct mutex snapshot_table_lock;
struct work_struct snapshot_delete_work;
+ struct work_struct snapshot_wait_for_pagecache_and_delete_work;
+ struct snapshot_id_list snapshots_unlinked;
+ struct mutex snapshots_unlinked_lock;
/* BTREE CACHE */
struct bio_set btree_bio;
* can delete it (or whether it should just be rm -rf'd)
*/
LE32_BITMASK(BCH_SUBVOLUME_SNAP, struct bch_subvolume, flags, 1, 2)
+LE32_BITMASK(BCH_SUBVOLUME_UNLINKED, struct bch_subvolume, flags, 2, 3)
/* Snapshots */
struct bkey_i *n;
__le64 *refcount;
int add = !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1;
+ char buf[200];
int ret;
bch2_trans_iter_init(trans, &iter, BTREE_ID_reflink, POS(0, *idx),
refcount = bkey_refcount(n);
if (!refcount) {
+ bch2_bkey_val_to_text(&PBUF(buf), c, p.s_c);
bch2_fs_inconsistent(c,
- "%llu:%llu len %u points to nonexistent indirect extent %llu",
- p.k->p.inode, p.k->p.offset, p.k->size, *idx);
+ "nonexistent indirect extent at %llu while marking\n %s",
+ *idx, buf);
ret = -EIO;
goto err;
}
if (!*refcount && (flags & BTREE_TRIGGER_OVERWRITE)) {
+ bch2_bkey_val_to_text(&PBUF(buf), c, p.s_c);
bch2_fs_inconsistent(c,
- "%llu:%llu len %u idx %llu indirect extent refcount underflow",
- p.k->p.inode, p.k->p.offset, p.k->size, *idx);
+ "indirect extent refcount underflow at %llu while marking\n %s",
+ *idx, buf);
ret = -EIO;
goto err;
}
struct bch_inode_unpacked *dir_u,
struct bch_inode_unpacked *inode_u,
const struct qstr *name,
- int deleting_snapshot)
+ bool deleting_snapshot)
{
struct bch_fs *c = trans->c;
struct btree_iter dir_iter = { NULL };
if (ret)
goto err;
- if (deleting_snapshot <= 0 && S_ISDIR(inode_u->bi_mode)) {
+ if (!deleting_snapshot && S_ISDIR(inode_u->bi_mode)) {
ret = bch2_empty_dir_trans(trans, inum);
if (ret)
goto err;
}
- if (deleting_snapshot < 0 &&
- inode_u->bi_subvol) {
- struct bch_subvolume s;
-
- ret = bch2_subvolume_get(trans, inode_u->bi_subvol, true,
- BTREE_ITER_CACHED|
- BTREE_ITER_WITH_UPDATES,
- &s);
- if (ret)
- goto err;
-
- if (BCH_SUBVOLUME_SNAP(&s))
- deleting_snapshot = 1;
+ if (deleting_snapshot && !inode_u->bi_subvol) {
+ ret = -ENOENT;
+ goto err;
}
- if (deleting_snapshot == 1) {
- if (!inode_u->bi_subvol) {
- ret = -ENOENT;
- goto err;
- }
-
- ret = bch2_subvolume_delete(trans, inode_u->bi_subvol,
- deleting_snapshot);
+ if (deleting_snapshot || inode_u->bi_subvol) {
+ ret = bch2_subvolume_unlink(trans, inode_u->bi_subvol);
if (ret)
goto err;
int bch2_unlink_trans(struct btree_trans *, subvol_inum,
struct bch_inode_unpacked *,
struct bch_inode_unpacked *,
- const struct qstr *, int);
+ const struct qstr *, bool);
int bch2_rename_trans(struct btree_trans *,
subvol_inum, struct bch_inode_unpacked *,
dir = path.dentry->d_parent->d_inode;
- ret = __bch2_unlink(dir, path.dentry, 1);
+ ret = __bch2_unlink(dir, path.dentry, true);
if (!ret) {
fsnotify_rmdir(dir, path.dentry);
d_delete(path.dentry);
}
int __bch2_unlink(struct inode *vdir, struct dentry *dentry,
- int deleting_snapshot)
+ bool deleting_snapshot)
{
struct bch_fs *c = vdir->i_sb->s_fs_info;
struct bch_inode_info *dir = to_bch_ei(vdir);
static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
{
- return __bch2_unlink(vdir, dentry, -1);
+ return __bch2_unlink(vdir, dentry, false);
}
static int bch2_symlink(struct user_namespace *mnt_userns,
}
}
+void bch2_evict_subvolume_inodes(struct bch_fs *c,
+ struct snapshot_id_list *s)
+{
+ struct super_block *sb = c->vfs_sb;
+ struct inode *inode;
+
+ spin_lock(&sb->s_inode_list_lock);
+ list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
+ if (!snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) ||
+ (inode->i_state & I_FREEING))
+ continue;
+
+ d_mark_dontcache(inode);
+ d_prune_aliases(inode);
+ }
+ spin_unlock(&sb->s_inode_list_lock);
+again:
+ cond_resched();
+ spin_lock(&sb->s_inode_list_lock);
+ list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
+ if (!snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) ||
+ (inode->i_state & I_FREEING))
+ continue;
+
+ if (!(inode->i_state & I_DONTCACHE)) {
+ d_mark_dontcache(inode);
+ d_prune_aliases(inode);
+ }
+
+ spin_lock(&inode->i_lock);
+ if (snapshot_list_has_id(s, to_bch_ei(inode)->ei_subvol) &&
+ !(inode->i_state & I_FREEING)) {
+ wait_queue_head_t *wq = bit_waitqueue(&inode->i_state, __I_NEW);
+ DEFINE_WAIT_BIT(wait, &inode->i_state, __I_NEW);
+ prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
+ spin_unlock(&inode->i_lock);
+ spin_unlock(&sb->s_inode_list_lock);
+ schedule();
+ finish_wait(wq, &wait.wq_entry);
+ goto again;
+ }
+
+ spin_unlock(&inode->i_lock);
+ }
+ spin_unlock(&sb->s_inode_list_lock);
+}
+
static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
{
struct super_block *sb = dentry->d_sb;
int bch2_setattr_nonsize(struct user_namespace *,
struct bch_inode_info *,
struct iattr *);
-int __bch2_unlink(struct inode *, struct dentry *, int);
+int __bch2_unlink(struct inode *, struct dentry *, bool);
+
+void bch2_evict_subvolume_inodes(struct bch_fs *, struct snapshot_id_list *);
void bch2_vfs_exit(void);
int bch2_vfs_init(void);
#else
+static inline void bch2_evict_subvolume_inodes(struct bch_fs *c,
+ struct snapshot_id_list *s) {}
static inline void bch2_vfs_exit(void) {}
static inline int bch2_vfs_init(void) { return 0; }
/* Subvolume root? */
if (inode_u.bi_subvol) {
- ret = bch2_subvolume_delete(trans, inode_u.bi_subvol, -1);
+ ret = bch2_subvolume_delete(trans, inode_u.bi_subvol);
if (ret)
goto err;
}
struct btree_trans trans;
struct btree_iter iter;
struct bkey_s_c k;
+ struct bkey_s_c_subvolume subvol;
int ret;
bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
for_each_btree_key(&trans, iter, BTREE_ID_subvolumes, POS_MIN,
0, k, ret) {
+ if (k.k->type != KEY_TYPE_subvolume)
+ continue;
+
+ subvol = bkey_s_c_to_subvolume(k);
+
+ if (BCH_SUBVOLUME_UNLINKED(subvol.v)) {
+ ret = __bch2_trans_do(&trans, NULL, NULL,
+ BTREE_INSERT_LAZY_RW,
+ bch2_subvolume_delete(&trans, iter.pos.offset));
+ if (ret) {
+ bch_err(c, "error deleting subvolume %llu: %i",
+ iter.pos.offset, ret);
+ break;
+ }
+ }
}
bch2_trans_iter_exit(&trans, &iter);
u->v.front_pad = 0;
u->v.back_pad = 0;
- return bch2_trans_update(trans, iter, &u->k_i, 0);
+ return bch2_trans_update(trans, iter, &u->k_i, BTREE_TRIGGER_NORUN);
}
static int fix_reflink_p(struct bch_fs *c)
bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u);
/* Subvolume root? */
- if (inode_u.bi_subvol) {
- ret = bch2_subvolume_delete(&trans, inode_u.bi_subvol, -1);
- if (ret)
- goto err;
- }
+ BUG_ON(inode_u.bi_subvol);
bkey_inode_generation_init(&delete.k_i);
delete.k.p = iter.pos;
set_bit(BCH_FS_REBUILD_REPLICAS, &c->flags);
}
- if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
- bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
- c->opts.version_upgrade = true;
- c->opts.fsck = true;
- c->opts.fix_errors = FSCK_OPT_YES;
- } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
- bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
- c->opts.version_upgrade = true;
- c->opts.fsck = true;
+ if (!c->opts.nochanges) {
+ if (c->sb.version < bcachefs_metadata_version_inode_backpointers) {
+ bch_info(c, "version prior to inode backpointers, upgrade and fsck required");
+ c->opts.version_upgrade = true;
+ c->opts.fsck = true;
+ c->opts.fix_errors = FSCK_OPT_YES;
+ } else if (c->sb.version < bcachefs_metadata_version_subvol_dirent) {
+ bch_info(c, "filesystem version is prior to subvol_dirent - upgrading");
+ c->opts.version_upgrade = true;
+ c->opts.fsck = true;
+ }
}
ret = bch2_blacklist_table_initialize(c);
#include "btree_key_cache.h"
#include "btree_update.h"
#include "error.h"
+#include "fs.h"
#include "subvolume.h"
/* Snapshot tree: */
return ret;
}
-/* List of snapshot IDs that are being deleted: */
-struct snapshot_id_list {
- u32 nr;
- u32 size;
- u32 *d;
-};
-
-static bool snapshot_list_has_id(struct snapshot_id_list *s, u32 id)
-{
- unsigned i;
-
- for (i = 0; i < s->nr; i++)
- if (id == s->d[i])
- return true;
- return false;
-}
-
static int snapshot_id_add(struct snapshot_id_list *s, u32 id)
{
BUG_ON(snapshot_list_has_id(s, id));
return ret;
}
-/* XXX: mark snapshot id for deletion, walk btree and delete: */
-int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid,
- int deleting_snapshot)
+/*
+ * Delete subvolume, mark snapshot ID as deleted, queue up snapshot
+ * deletion/cleanup:
+ */
+int bch2_subvolume_delete(struct btree_trans *trans, u32 subvolid)
{
struct btree_iter iter;
struct bkey_s_c k;
subvol = bkey_s_c_to_subvolume(k);
snapid = le32_to_cpu(subvol.v->snapshot);
- if (deleting_snapshot >= 0 &&
- deleting_snapshot != BCH_SUBVOLUME_SNAP(subvol.v)) {
- ret = -ENOENT;
- goto err;
- }
-
delete = bch2_trans_kmalloc(trans, sizeof(*delete));
ret = PTR_ERR_OR_ZERO(delete);
if (ret)
return ret;
}
+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);
+ struct snapshot_id_list s;
+ u32 *id;
+ int ret = 0;
+
+ while (!ret) {
+ mutex_lock(&c->snapshots_unlinked_lock);
+ s = c->snapshots_unlinked;
+ memset(&c->snapshots_unlinked, 0, sizeof(c->snapshots_unlinked));
+ mutex_unlock(&c->snapshots_unlinked_lock);
+
+ if (!s.nr)
+ break;
+
+ bch2_evict_subvolume_inodes(c, &s);
+
+ for (id = s.d; id < s.d + s.nr; id++) {
+ 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);
+ break;
+ }
+ }
+
+ kfree(s.d);
+ }
+
+ percpu_ref_put(&c->writes);
+}
+
+struct subvolume_unlink_hook {
+ struct btree_trans_commit_hook h;
+ u32 subvol;
+};
+
+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);
+ struct bch_fs *c = trans->c;
+ int ret = 0;
+
+ 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);
+ mutex_unlock(&c->snapshots_unlinked_lock);
+
+ if (ret)
+ return ret;
+
+ if (unlikely(!percpu_ref_tryget(&c->writes)))
+ return -EROFS;
+
+ if (!queue_work(system_long_wq, &c->snapshot_wait_for_pagecache_and_delete_work))
+ percpu_ref_put(&c->writes);
+ 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;
+
+ h->h.fn = bch2_subvolume_wait_for_pagecache_and_delete_hook;
+ h->subvol = subvolid;
+ bch2_trans_commit_hook(trans, &h->h);
+err:
+ bch2_trans_iter_exit(trans, &iter);
+ return ret;
+}
+
int bch2_subvolume_create(struct btree_trans *trans, u64 inode,
u32 src_subvolid,
u32 *new_subvolid,
int bch2_fs_subvolumes_init(struct bch_fs *c)
{
INIT_WORK(&c->snapshot_delete_work, bch2_delete_dead_snapshots_work);
+ INIT_WORK(&c->snapshot_wait_for_pagecache_and_delete_work,
+ bch2_subvolume_wait_for_pagecache_and_delete);
+ mutex_init(&c->snapshots_unlinked_lock);
return 0;
}
#ifndef _BCACHEFS_SUBVOLUME_H
#define _BCACHEFS_SUBVOLUME_H
+#include "subvolume_types.h"
+
void bch2_snapshot_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
const char *bch2_snapshot_invalid(const struct bch_fs *, struct bkey_s_c);
return 0;
}
+static inline bool snapshot_list_has_id(struct snapshot_id_list *s, u32 id)
+{
+ unsigned i;
+
+ for (i = 0; i < s->nr; i++)
+ if (id == s->d[i])
+ return true;
+ return false;
+}
+
int bch2_fs_snapshots_check(struct bch_fs *);
void bch2_fs_snapshots_exit(struct bch_fs *);
int bch2_fs_snapshots_start(struct bch_fs *);
bool, int, struct bch_subvolume *);
int bch2_subvolume_get_snapshot(struct btree_trans *, u32, u32 *);
-int bch2_subvolume_delete(struct btree_trans *, u32, int);
+int bch2_subvolume_delete(struct btree_trans *, u32);
+int bch2_subvolume_unlink(struct btree_trans *, u32);
int bch2_subvolume_create(struct btree_trans *, u64, u32,
u32 *, u32 *, bool);
--- /dev/null
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _BCACHEFS_SUBVOLUME_TYPES_H
+#define _BCACHEFS_SUBVOLUME_TYPES_H
+
+struct snapshot_id_list {
+ u32 nr;
+ u32 size;
+ u32 *d;
+};
+
+#endif /* _BCACHEFS_SUBVOLUME_TYPES_H */
if (ret)
goto err;
- /*
- * We need to write out a journal entry before we start doing btree
- * updates, to ensure that on unclean shutdown new journal blacklist
- * entries are created:
- */
- bch2_journal_meta(&c->journal);
-
clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
for_each_rw_member(ca, c, i)