In the Linux kernel, the following vulnerability has been resolved:
libfs: fix infinite directory reads for offset dir
After we switch tmpfs dir operations from simple_dir_operations to
simple_offset_dir_operations, every rename happened will fill new dentry
to dest dir's maple tree(&SHMEM_I(inode)->dir_offsets->mt) with a free
key starting with octx->newx_offset, and then set newx_offset equals to
free key + 1. This will lead to infinite readdir combine with rename
happened at the same time, which fail generic/736 in xfstests(detail show
as below).
- create 5000 files(1 2 3...) under one dir
- call readdir(man 3 readdir) once, and get one entry
- rename(entry, "TEMPFILE"), then rename("TEMPFILE", entry)
- loop 2~3, until readdir return nothing or we loop too many
times(tmpfs break test with the second condition)
We choose the same logic what commit 9b378f6ad48cf ("btrfs: fix infinite
directory reads") to fix it, record the last_index when we open dir, and
do not emit the entry which index >= last_index. The file->private_data
now used in offset dir can use directly to do this, and we also update
the last_index when we llseek the dir file.
[brauner: only update last_index after seek when offset is zero like Jan suggested]
References
In the Linux kernel, the following vulnerability has been resolved:
libfs: fix infinite directory reads for offset dir
After we switch tmpfs dir operations from simple_dir_operations to
simple_offset_dir_operations, every rename happened will fill new dentry
to dest dir's maple tree(&SHMEM_I(inode)->dir_offsets->mt) with a free
key starting with octx->newx_offset, and then set newx_offset equals to
free key + 1. This will lead to infinite readdir combine with rename
happened at the same time, which fail generic/736 in xfstests(detail show
as below).
times(tmpfs break test with the second condition)
We choose the same logic what commit 9b378f6ad48cf ("btrfs: fix infinite
directory reads") to fix it, record the last_index when we open dir, and
do not emit the entry which index >= last_index. The file->private_data
now used in offset dir can use directly to do this, and we also update
the last_index when we llseek the dir file.
[brauner: only update last_index after seek when offset is zero like Jan suggested]
References