Skip to contentSkip to conversation

staging: android: ashmem: Fix lockdep issue during llseek

PATCH In review Joel Fernandes 4 participants 6 messages Feb 16, 2018 – Feb 27, 2018

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

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

Cc Todd Kjos <[email protected]>
Cc Arve Hjonnevag <[email protected]>
Cc Greg Hackmann <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Reported-by [email protected]

---
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
index bbdc53b686dd..b330e86b3a49 100644
@@ -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
#1 lore raw
Quoted 40 lines · Joel Fernandes

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

Cc Todd Kjos <[email protected]>
Cc Arve Hjonnevag <[email protected]>
Cc Greg Hackmann <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Reported-by [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 with asma->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

#2 lore raw
Quoted 47 lines · Greg Kroah-Hartman
Quoted 40 lines · Joel Fernandes

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

Cc Todd Kjos <[email protected]>
Cc Arve Hjonnevag <[email protected]>
Cc Greg Hackmann <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Reported-by [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 with asma->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

#3 lore raw
In reply to Greg Kroah-Hartman

(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_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

Cc Todd Kjos <[email protected]>
Cc Arve Hjonnevag <[email protected]>
Cc Greg Hackmann <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Reported-by [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 with asma->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

#4 lore raw

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_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

Cc Todd Kjos <[email protected]>
Cc Arve Hjonnevag <[email protected]>
Cc Greg Hackmann <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Reported-by [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 with asma->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

Acked-by
#5 lore raw
In reply to Joel Fernandes
Quoted 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_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

Cc Todd Kjos <[email protected]>
Cc Arve Hjonnevag <[email protected]>
Cc Greg Hackmann <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Reported-by [email protected]

---
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/

Quoted 3 lines
Quoted 1 line

I'd like some acks from others before I take this patch.

GregH, Todd, could you provide Acks?