#define LOG_WALK_REPLAY_ALL 3
static int btrfs_log_inode(struct btrfs_trans_handle *trans,
- struct btrfs_root *root, struct inode *inode,
- int inode_only);
+ struct btrfs_root *root, struct inode *inode,
+ int inode_only,
+ const loff_t start,
+ const loff_t end);
static int link_to_fixup_dir(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct btrfs_path *path, u64 objectid);
* This handles both files and directories.
*/
static int btrfs_log_inode(struct btrfs_trans_handle *trans,
- struct btrfs_root *root, struct inode *inode,
- int inode_only)
+ struct btrfs_root *root, struct inode *inode,
+ int inode_only,
+ const loff_t start,
+ const loff_t end)
{
struct btrfs_path *path;
struct btrfs_path *dst_path;
int ins_nr;
bool fast_search = false;
u64 ino = btrfs_ino(inode);
+ struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
path = btrfs_alloc_path();
if (!path)
if (ret < 0) {
err = ret;
goto out_unlock;
- } if (ret) {
+ }
+ if (ret) {
ins_nr = 0;
btrfs_release_path(path);
continue;
goto out_unlock;
}
} else if (inode_only == LOG_INODE_ALL) {
- struct extent_map_tree *tree = &BTRFS_I(inode)->extent_tree;
struct extent_map *em, *n;
- write_lock(&tree->lock);
- list_for_each_entry_safe(em, n, &tree->modified_extents, list)
- list_del_init(&em->list);
- write_unlock(&tree->lock);
+ write_lock(&em_tree->lock);
+ /*
+ * We can't just remove every em if we're called for a ranged
+ * fsync - that is, one that doesn't cover the whole possible
+ * file range (0 to LLONG_MAX). This is because we can have
+ * em's that fall outside the range we're logging and therefore
+ * their ordered operations haven't completed yet
+ * (btrfs_finish_ordered_io() not invoked yet). This means we
+ * didn't get their respective file extent item in the fs/subvol
+ * tree yet, and need to let the next fast fsync (one which
+ * consults the list of modified extent maps) find the em so
+ * that it logs a matching file extent item and waits for the
+ * respective ordered operation to complete (if it's still
+ * running).
+ *
+ * Removing every em outside the range we're logging would make
+ * the next fast fsync not log their matching file extent items,
+ * therefore making us lose data after a log replay.
+ */
+ list_for_each_entry_safe(em, n, &em_tree->modified_extents,
+ list) {
+ const u64 mod_end = em->mod_start + em->mod_len - 1;
+
+ if (em->mod_start >= start && mod_end <= end)
+ list_del_init(&em->list);
+ }
+ write_unlock(&em_tree->lock);
}
if (inode_only == LOG_INODE_ALL && S_ISDIR(inode->i_mode)) {
goto out_unlock;
}
}
+
BTRFS_I(inode)->logged_trans = trans->transid;
BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->last_sub_trans;
out_unlock:
*/
static int btrfs_log_inode_parent(struct btrfs_trans_handle *trans,
struct btrfs_root *root, struct inode *inode,
- struct dentry *parent, int exists_only,
+ struct dentry *parent,
+ const loff_t start,
+ const loff_t end,
+ int exists_only,
struct btrfs_log_ctx *ctx)
{
int inode_only = exists_only ? LOG_INODE_EXISTS : LOG_INODE_ALL;
if (ret)
goto end_no_trans;
- ret = btrfs_log_inode(trans, root, inode, inode_only);
+ ret = btrfs_log_inode(trans, root, inode, inode_only, start, end);
if (ret)
goto end_trans;
if (BTRFS_I(inode)->generation >
root->fs_info->last_trans_committed) {
- ret = btrfs_log_inode(trans, root, inode, inode_only);
+ ret = btrfs_log_inode(trans, root, inode, inode_only,
+ 0, LLONG_MAX);
if (ret)
goto end_trans;
}
*/
int btrfs_log_dentry_safe(struct btrfs_trans_handle *trans,
struct btrfs_root *root, struct dentry *dentry,
+ const loff_t start,
+ const loff_t end,
struct btrfs_log_ctx *ctx)
{
struct dentry *parent = dget_parent(dentry);
int ret;
ret = btrfs_log_inode_parent(trans, root, dentry->d_inode, parent,
- 0, ctx);
+ start, end, 0, ctx);
dput(parent);
return ret;
root->fs_info->last_trans_committed))
return 0;
- return btrfs_log_inode_parent(trans, root, inode, parent, 1, NULL);
+ return btrfs_log_inode_parent(trans, root, inode, parent, 0,
+ LLONG_MAX, 1, NULL);
}