staging: android: ashmem: Fix lockdep issue during llseek
Subsystems (2)
PATCH v2 1/1 staging: android: ashmem: Fix lockdep issue during llseek
ashmem_mutex create a chain of dependencies like so:
(1)
mmap syscall ->
mmap_sem -> (acquired)
ashmem_mmap
ashmem_mutex (try to acquire)
(block)
(2)
llseek syscall ->
ashmem_llseek ->
ashmem_mutex -> (acquired)
inode_lock ->
inode->i_rwsem (try to acquire)
(block)
(3)
getdents ->
iterate_dir ->
inode_lock ->
inode->i_rwsem (acquired)
copy_to_user ->
mmap_sem (try to acquire)
There is a lock ordering created between mmap_sem and inode->i_rwsem causing a lockdep splat [2] during a syzcaller test, this patch fixes the issue by unlocking the mutex earlier. Functionally that's Ok since we don't need to protect vfs_llseek.
[1] https://patchwork.kernel.org/patch/10185031/
[2] https://lkml.org/lkml/2018/1/10/48
---
Changes since first version:
Don't relock after vfs call since its not needed. Only reason we lock is to protect races with asma->file.
https://patchwork.kernel.org/patch/10185031/
drivers/staging/android/ashmem.c | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-)
drivers/staging/android/ashmem.c +7 -8
@@ -326,24 +326,23 @@ static loff_t ashmem_llseek(struct file *file, loff_t offset, int origin)mutex_lock (&ashmem_mutex);if (asma->size ==0 ) { - ret = -EINVAL ; - goto out; +mutex_unlock (&ashmem_mutex); +return -EINVAL ; }if (!asma->file) { - ret = -EBADF ; - goto out; +mutex_unlock (&ashmem_mutex); +return -EBADF ; } +mutex_unlock (&ashmem_mutex); + ret =vfs_llseek (asma->file, offset, origin);if (ret <0 ) - goto out; +return ret;/** Copy f_pos from backing file, since f_ops->llseek() sets it */ file->f_pos = asma->file->f_pos; - -out: -mutex_unlock (&ashmem_mutex);return ret; }
Signature · 1 line
2.16.1.291.g4437f3f132-goog
Greg Kroah-Hartman
maintainerQuoted 40 lines · Joel Fernandes
ashmem_mutexcreate a chain of dependencies like so:(1)
mmap syscall ->
mmap_sem-> (acquired)
ashmem_mmap
ashmem_mutex(try to acquire)
(block)(2)
llseek syscall ->
ashmem_llseek->
ashmem_mutex-> (acquired)
inode_lock->
inode->i_rwsem(try to acquire)
(block)(3)
getdents ->
iterate_dir->
inode_lock->
inode->i_rwsem(acquired)
copy_to_user->
mmap_sem(try to acquire)There is a lock ordering created between
mmap_semandinode->i_rwsemcausing a lockdep splat [2] during a syzcaller test, this patch fixes the issue by unlocking the mutex earlier. Functionally that's Ok since we don't need to protectvfs_llseek.[1] https://patchwork.kernel.org/patch/10185031/
[2] https://lkml.org/lkml/2018/1/10/48Cc Todd Kjos <[email protected]>Cc Arve Hjonnevag <[email protected]>Cc Greg Hackmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Reported-by [email protected]Signed-off-by Joel Fernandes <[email protected]>
Quoted 5 lines
---
Changes since first version:
Don't relock after vfs call since its not needed. Only reason we lock is to protect races withasma->file.
https://patchwork.kernel.org/patch/10185031/
I'd like some acks from others before I take this patch.
Joel, did the original reporter say this patch solved their issue or not? For some reason I didn't think it worked properly...
thanks,
greg k-h
Quoted 47 lines · Greg Kroah-Hartman
Quoted 40 lines · Joel Fernandes
ashmem_mutexcreate a chain of dependencies like so:(1)
mmap syscall ->
mmap_sem-> (acquired)
ashmem_mmap
ashmem_mutex(try to acquire)
(block)(2)
llseek syscall ->
ashmem_llseek->
ashmem_mutex-> (acquired)
inode_lock->
inode->i_rwsem(try to acquire)
(block)(3)
getdents ->
iterate_dir->
inode_lock->
inode->i_rwsem(acquired)
copy_to_user->
mmap_sem(try to acquire)There is a lock ordering created between
mmap_semandinode->i_rwsemcausing a lockdep splat [2] during a syzcaller test, this patch fixes the issue by unlocking the mutex earlier. Functionally that's Ok since we don't need to protectvfs_llseek.[1] https://patchwork.kernel.org/patch/10185031/
[2] https://lkml.org/lkml/2018/1/10/48Cc Todd Kjos <[email protected]>Cc Arve Hjonnevag <[email protected]>Cc Greg Hackmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Reported-by [email protected]Signed-off-by Joel Fernandes <[email protected]>Quoted 5 lines
---
Changes since first version:
Don't relock after vfs call since its not needed. Only reason we lock is to protect races withasma->file.
https://patchwork.kernel.org/patch/10185031/
Quoted 1 line
I'd like some acks from others before I take this patch.
GregH, Todd, could you provide Acks?
Quoted 3 lines
Joel, did the original reporter say this patch solved their issue or not? For some reason I didn't think it worked properly...
That's a different but similar issue:
https://patchwork.kernel.org/patch/10202127/. That was related to RECLAIM_FS lockdep recursion. That was posted as an RFC unlike this one, and its still being investigated since Miles reported that the lockdep inode annotation doesn't fix the issue.
This one on the other hand has a straightforward fix, and was verified by the syzbot.
I can see why its easy to confuse the two issues.
Thanks,
- Joel
Quoted 4 lines
thanks,
greg k-h
(reposting in plain text, sorry for the previous HTML email, I should have not posted from the Phone)
Quoted 47 lines · Greg Kroah-Hartman
Quoted 40 lines · Joel Fernandes
ashmem_mutexcreate a chain of dependencies like so:(1)
mmap syscall ->
mmap_sem-> (acquired)
ashmem_mmap
ashmem_mutex(try to acquire)
(block)(2)
llseek syscall ->
ashmem_llseek->
ashmem_mutex-> (acquired)
inode_lock->
inode->i_rwsem(try to acquire)
(block)(3)
getdents ->
iterate_dir->
inode_lock->
inode->i_rwsem(acquired)
copy_to_user->
mmap_sem(try to acquire)There is a lock ordering created between
mmap_semandinode->i_rwsemcausing a lockdep splat [2] during a syzcaller test, this patch fixes the issue by unlocking the mutex earlier. Functionally that's Ok since we don't need to protectvfs_llseek.[1] https://patchwork.kernel.org/patch/10185031/
[2] https://lkml.org/lkml/2018/1/10/48Cc Todd Kjos <[email protected]>Cc Arve Hjonnevag <[email protected]>Cc Greg Hackmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Reported-by [email protected]Signed-off-by Joel Fernandes <[email protected]>Quoted 5 lines
---
Changes since first version:
Don't relock after vfs call since its not needed. Only reason we lock is to protect races withasma->file.
https://patchwork.kernel.org/patch/10185031/
Quoted 1 line
I'd like some acks from others before I take this patch.
GregH, Todd, could you provide Acks?
Quoted 3 lines
Joel, did the original reporter say this patch solved their issue or not? For some reason I didn't think it worked properly...
That's a different but similar issue:
https://patchwork.kernel.org/patch/10202127/. That was related to RECLAIM_FS lockdep recursion. That was posted as an RFC unlike this one, and its still being investigated since Miles reported that the lockdep inode annotation doesn't fix the issue.
This one on the other hand has a straightforward fix, and was verified by the syzbot.
I can see why its easy to confuse the two issues.
Thanks,
- Joel
Ack. I confirmed that the syzbot could not reproduce the issue with this patch.
Quoted 74 lines · Joel Fernandes
(reposting in plain text, sorry for the previous HTML email, I should have not posted from the Phone)
Quoted 47 lines · Greg Kroah-Hartman
Quoted 40 lines · Joel Fernandes
ashmem_mutexcreate a chain of dependencies like so:(1)
mmap syscall ->
mmap_sem-> (acquired)
ashmem_mmap
ashmem_mutex(try to acquire)
(block)(2)
llseek syscall ->
ashmem_llseek->
ashmem_mutex-> (acquired)
inode_lock->
inode->i_rwsem(try to acquire)
(block)(3)
getdents ->
iterate_dir->
inode_lock->
inode->i_rwsem(acquired)
copy_to_user->
mmap_sem(try to acquire)There is a lock ordering created between
mmap_semandinode->i_rwsemcausing a lockdep splat [2] during a syzcaller test, this patch fixes the issue by unlocking the mutex earlier. Functionally that's Ok since we don't need to protectvfs_llseek.[1] https://patchwork.kernel.org/patch/10185031/
[2] https://lkml.org/lkml/2018/1/10/48Cc Todd Kjos <[email protected]>Cc Arve Hjonnevag <[email protected]>Cc Greg Hackmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Reported-by [email protected]Signed-off-by Joel Fernandes <[email protected]>Quoted 5 lines
---
Changes since first version:
Don't relock after vfs call since its not needed. Only reason we lock is to protect races withasma->file.
https://patchwork.kernel.org/patch/10185031/Quoted 1 line
I'd like some acks from others before I take this patch.
GregH, Todd, could you provide Acks?
Quoted 3 lines
Joel, did the original reporter say this patch solved their issue or not? For some reason I didn't think it worked properly...
That's a different but similar issue:
https://patchwork.kernel.org/patch/10202127/. That was related toRECLAIM_FSlockdep recursion. That was posted as an RFC unlike this one, and its still being investigated since Miles reported that the lockdep inode annotation doesn't fix the issue.This one on the other hand has a straightforward fix, and was verified by the syzbot.
I can see why its easy to confuse the two issues.
Thanks,
- Joel
Greg Hackmann
Acked-byQuoted 52 lines · Joel Fernandes
(reposting in plain text, sorry for the previous HTML email, I should have not posted from the Phone)
Quoted 47 lines · Greg Kroah-Hartman
Quoted 45 lines · Joel Fernandes
ashmem_mutexcreate a chain of dependencies like so:(1)
mmap syscall ->
mmap_sem-> (acquired)
ashmem_mmap
ashmem_mutex(try to acquire)
(block)(2)
llseek syscall ->
ashmem_llseek->
ashmem_mutex-> (acquired)
inode_lock->
inode->i_rwsem(try to acquire)
(block)(3)
getdents ->
iterate_dir->
inode_lock->
inode->i_rwsem(acquired)
copy_to_user->
mmap_sem(try to acquire)There is a lock ordering created between
mmap_semandinode->i_rwsemcausing a lockdep splat [2] during a syzcaller test, this patch fixes the issue by unlocking the mutex earlier. Functionally that's Ok since we don't need to protectvfs_llseek.[1] https://patchwork.kernel.org/patch/10185031/
[2] https://lkml.org/lkml/2018/1/10/48Cc Todd Kjos <[email protected]>Cc Arve Hjonnevag <[email protected]>Cc Greg Hackmann <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Reported-by [email protected]Signed-off-by Joel Fernandes <[email protected]>---
Changes since first version:
Don't relock after vfs call since its not needed. Only reason we lock is to protect races withasma->file.
https://patchwork.kernel.org/patch/10185031/
Quoted 3 lines
Quoted 1 line
I'd like some acks from others before I take this patch.
GregH, Todd, could you provide Acks?