Skip to contentSkip to conversation

bisected: 3.8 -> 3.9 cannot umount after using fuse

REGRESSION Michael Leun 2 participants 4 messages May 4, 2013 VFS

lore
Share & tools
lore
#0 lore raw

Mount some filesystem, below that mountpoint as user mount some fuse fs (tried that with sshfs and fuseiso, both same error), fusermount -u the fuse fs, try to unmount the filesystem -> busy.

ml@lara:~> id -u
500
ml@lara:~> grep test /etc/fstab
/dev/vg1/test /mnt2 ext4acl,user_xattr,user,noauto 0 0
ml@lara:~> mount /mnt2
ml@lara:~> cd /mnt2
ml@lara:/mnt2> ls -l
insgesamt 24

drwx------ 2 root root  16384  4. Mai 11:24 lost+found
drwxr-xr-x 2 root root   4096  4. Mai 11:44 test
drwxr-xr-x 2 ml   users  4096  4. Mai 11:25 xenia

ml@lara:/mnt2> sshfs xenia: xenia
Password:
ml@lara:/mnt2> ls xenia/bla
xenia/bla
ml@lara:/mnt2> fusermount -u xenia
ml@lara:/mnt2> cd
ml@lara:~> umount /mnt2
umount: /mnt2: target is busy.
(In some cases useful info about processes that use
the device is found by lsof(8) or fuser(1))
ml@lara:~> fuser -m /mnt2
ml@lara:~>

It does'nt matter if the mounting and unmounting of the /mnt2 filesystem is done as user or as root, same result.

This works in 3.8(.x). Bisecting yields

57eccb830f1cc93d4b506ba306d8dfa685e0c88f is the first bad commit commit 57eccb830f1cc93d4b506ba306d8dfa685e0c88f

Author: Al Viro <[email protected]>
Date:   Fri Feb 22 22:49:10 2013 -0500

mount: consolidate permission checks

... and ask for global CAP_SYS_ADMIN only for superblock-level remounts

:040000 040000 4e926c083f6ede645158f02714d73cdd3b233a05 584611744bad96544578259a6fa176577df0ebbd M fs

After reverting 57eccb830f1cc93d4b506ba306d8dfa685e0c88f from 3.9 that umount above works (not busy).

I've seen that this patch also was identified as problem in relation with user ns and module loading, see lkml

CAGXu5jLciqXcuhzDVPKPqA+mwv8e28zGAMScatzov3U07BuBaQ@mail.gmail.com

After fixing this upstream the fix should be an stabile candidate.

Signature · 3 lines
MfG,

Michael Leun
maintainer
#1 lore raw
Quoted 2 lines · Michael Leun

After reverting 57eccb830f1cc93d4b506ba306d8dfa685e0c88f from 3.9 that umount above works (not busy).

Sigh... Check if the following fix works for your testcase:

fs/namespace.c +2 -3
index b4f96a5..b68eef2d 100644
@@ -2284,12 +2284,11 @@ long do_mount(const char *dev_name, const char *dir_name,
 
 	retval = security_sb_mount(dev_name, &path,
 				   type_page, flags, data_page);
+	if (!retval && !may_mount())
+		retval = -EPERM;
 	if (retval)
 		goto dput_out;
 
-	if (!may_mount())
-		return -EPERM;
-
 	/* Default to relatime unless overriden */
 	if (!(flags & MS_NOATIME))
 		mnt_flags |= MNT_RELATIME;
#2 lore raw

fs/namespace.c

Quoted 10 lines · On Sat, 4 May 2013 16:52:14 +0100 Al Viro <[email protected]> wrote:
Quoted 2 lines · Michael Leun

After reverting 57eccb830f1cc93d4b506ba306d8dfa685e0c88f from 3.9 that umount above works (not busy).

Sigh... Check if the following fix works for your testcase:

fs/namespace.c +0 -0
index b4f96a5..b68eef2d 100644

[...]

Yup, that fixes it.

Signature · 3 lines
MfG,

Michael Leun
maintainer
#3 lore raw
Quoted 8 lines · Michael Leun
Quoted 5 lines · On Sat, 4 May 2013 16:52:14 +0100 Al Viro <[email protected]> wrote:
Quoted 2 lines · Michael Leun

After reverting 57eccb830f1cc93d4b506ba306d8dfa685e0c88f from 3.9 that umount above works (not busy).

Quoted 8 lines

fs/namespace.c

Quoted 5 lines

Sigh... Check if the following fix works for your testcase:

fs/namespace.c +0 -0
index b4f96a5..b68eef2d 100644

[...]

Yup, that fixes it.

OK, committed, will go to Linus today, with Cc: stable