Comment 2 for bug 1674868

Revision history for this message
Andy Caldwell (andy-m-caldwell) wrote :

[Impact]

 * Any attempt to create/overwrite symlinks in a partition mounted by fuseext2 causes the fuseext2 process to deadlock, preventing the mounted filesystem from being used at all until the userspace process is killed and the filesystem is remounted.

[Test Case]

 * dd if=/dev/zero of=partition bs=1M count=200
 * mkfs.ext2 partition
 * mkdir mount
 * fuseext2 partition mount
 * cd mount
 * touch test
 * ln -s test link - Doesn't complete
 * In another shell try to do anything else in the mount directory (ls, touch, rm) - Don't complete

[Regression Potential]

 * Since this changes the locking strategy of the code to hold locks less, the main risk is exposing a case where data can be accessed in a non-threadsafe manner, leading to unexpected behaviour.
 * The places where the change has been made are at the exit points of the function, and match where similar unlocks are made in other places in the code.
 * Since the program at risk is (by design) a userspace program, the risk of kernel data leakage is minimal.

[Other Info]

 * I can't actually work out where the upstream code for this project lives, the only upstream I could find is https://github.com/alperakcan/fuse-ext2 but that doesn't have any locking code at all (even in the reentrant branch) so I can't see where this issue came from.
 * This issue seems to apply to all versions of fuseext2 in the ubuntu repos (well, at least it's there in trusty, xenial and zesty)