Skip to contentSkip to conversation

Btrfs: fix send dealing with file renames and directory moves

PATCH In review Filipe David Borba Manana 1 participant 2 messages Feb 1, 2014 – Feb 15, 2014 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Btrfs: fix send dealing with file renames and directory moves

This fixes a case that the commit titled:

Btrfs: fix infinite path build loops in incremental send

didn't cover. If the parent-child relationship between 2 directories is inverted, both get renamed, and the former parent has a file that got renamed too (but remains a child of that directory), the incremental send operation would use the file's old path after sending an unlink operation for that old path, causing receive to fail on future operations like changing owner, permissions or utimes of the corresponding inode.

This is not a regression from the commit mentioned before, as without that commit we would fall into the issues that commit fixed, so it's just one case that wasn't covered before.

Simple steps to reproduce this issue are:

      $ mkfs.btrfs -f /dev/sdb3
      $ mount /dev/sdb3 /mnt/btrfs
      $ mkdir -p /mnt/btrfs/a/b/c/d
      $ touch /mnt/btrfs/a/b/c/d/file
      $ mkdir -p /mnt/btrfs/a/b/x
      $ btrfs subvol snapshot -r /mnt/btrfs /mnt/btrfs/snap1
      $ mv /mnt/btrfs/a/b/x /mnt/btrfs/a/b/c/x2
      $ mv /mnt/btrfs/a/b/c/d /mnt/btrfs/a/b/c/x2/d2
      $ mv /mnt/btrfs/a/b/c/x2/d2/file /mnt/btrfs/a/b/c/x2/d2/file2
      $ btrfs subvol snapshot -r /mnt/btrfs /mnt/btrfs/snap2
      $ btrfs send -p /mnt/btrfs/snap1 /mnt/btrfs/snap2 > /tmp/incremental.send

A patch to update the test btrfs/030 from xfstests, so that it covers this case, will be submitted soon.

---
 fs/btrfs/send.c |   10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)
fs/btrfs/send.c +1 -9
index 7250d86..e0b49f6 100644
@@ -2121,8 +2121,6 @@ static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
 	u64 parent_inode = 0;
 	u64 parent_gen = 0;
 	int stop = 0;
-	u64 start_ino = ino;
-	u64 start_gen = gen;
 	int skip_name_cache = 0;
 
 	name = fs_path_alloc();
@@ -2134,7 +2132,6 @@ static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
 	if (is_waiting_for_move(sctx, ino))
 		skip_name_cache = 1;
 
-again:
 	dest->reversed = 1;
 	fs_path_reset(dest);
 
@@ -2149,13 +2146,8 @@ again:
 			stop = 1;
 
 		if (!skip_name_cache &&
-		    is_waiting_for_move(sctx, parent_inode)) {
-			ino = start_ino;
-			gen = start_gen;
-			stop = 0;
+		    is_waiting_for_move(sctx, parent_inode))
 			skip_name_cache = 1;
-			goto again;
-		}
 
 		ret = fs_path_add_path(dest, name);
 		if (ret < 0)
Signature · 1 line
1.7.9.5
#1 lore raw
Quoted 29 lines · Filipe David Borba Manana

This fixes a case that the commit titled:

Btrfs: fix infinite path build loops in incremental send

didn't cover. If the parent-child relationship between 2 directories is inverted, both get renamed, and the former parent has a file that got renamed too (but remains a child of that directory), the incremental send operation would use the file's old path after sending an unlink operation for that old path, causing receive to fail on future operations like changing owner, permissions or utimes of the corresponding inode.

This is not a regression from the commit mentioned before, as without that commit we would fall into the issues that commit fixed, so it's just one case that wasn't covered before.

Simple steps to reproduce this issue are:

      $ mkfs.btrfs -f /dev/sdb3
      $ mount /dev/sdb3 /mnt/btrfs
      $ mkdir -p /mnt/btrfs/a/b/c/d
      $ touch /mnt/btrfs/a/b/c/d/file
      $ mkdir -p /mnt/btrfs/a/b/x
      $ btrfs subvol snapshot -r /mnt/btrfs /mnt/btrfs/snap1
      $ mv /mnt/btrfs/a/b/x /mnt/btrfs/a/b/c/x2
      $ mv /mnt/btrfs/a/b/c/d /mnt/btrfs/a/b/c/x2/d2
      $ mv /mnt/btrfs/a/b/c/x2/d2/file /mnt/btrfs/a/b/c/x2/d2/file2
      $ btrfs subvol snapshot -r /mnt/btrfs /mnt/btrfs/snap2
      $ btrfs send -p /mnt/btrfs/snap1 /mnt/btrfs/snap2 > /tmp/incremental.send
Quoted 4 lines

A patch to update the test btrfs/030 from xfstests, so that it covers this case, will be submitted soon.

Hi,

Should this go to 3.14? It is covered by the test case btrfs/030 for xfstests.

thanks

Quoted 43 lines
---
 fs/btrfs/send.c |   10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)
fs/btrfs/send.c +1 -10
index 7250d86..e0b49f6 100644
@@ -2121,8 +2121,6 @@ static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
        u64 parent_inode = 0;
        u64 parent_gen = 0;
        int stop = 0;
-       u64 start_ino = ino;
-       u64 start_gen = gen;
        int skip_name_cache = 0;
 
        name = fs_path_alloc();
@@ -2134,7 +2132,6 @@ static int get_cur_path(struct send_ctx *sctx, u64 ino, u64 gen,
        if (is_waiting_for_move(sctx, ino))
                skip_name_cache = 1;
 
-again:
        dest->reversed = 1;
        fs_path_reset(dest);
 
@@ -2149,13 +2146,8 @@ again:
                        stop = 1;
 
                if (!skip_name_cache &&
-                   is_waiting_for_move(sctx, parent_inode)) {
-                       ino = start_ino;
-                       gen = start_gen;
-                       stop = 0;
+                   is_waiting_for_move(sctx, parent_inode))
                        skip_name_cache = 1;
-                       goto again;
-               }
 
                ret = fs_path_add_path(dest, name);
                if (ret < 0)
--

1.7.9.5

Signature · 5 lines
Filipe David Manana,

"Reasonable men adapt themselves to the world.
 Unreasonable men adapt the world to themselves.
 That's why all progress depends on unreasonable men."