CIFS: Fix NULL-pointer dereference in smb2_push_mandatory_locks
Subsystems (2)
PATCH 1/1 CIFS: Fix NULL-pointer dereference in smb2_push_mandatory_locks
Currently when the client creates a cifsFileInfo structure for a newly opened file, it allocates a list of byte-range locks with a pointer to the new cfile and attaches this list to the inode's lock list. The latter happens before initializing all other fields, e.g. cfile->tlink. Thus a partially initialized cifsFileInfo structure becomes available to other threads that walk through the inode's lock list. One example of such a thread may be an oplock break worker thread that tries to push all cached byte-range locks. This causes NULL-pointer dereference in smb2_push_mandatory_locks() when accessing cfile->tlink:
[598428.945633] BUG: kernel NULL pointer dereference, address: 0000000000000038 ... [598428.945749] Workqueue: cifsoplockd cifs_oplock_break [cifs] [598428.945793] RIP: 0010:smb2_push_mandatory_locks+0xd6/0x5a0 [cifs] ... [598428.945834] Call Trace: [598428.945870] ? cifs_revalidate_mapping+0x45/0x90 [cifs] [598428.945901] cifs_oplock_break+0x13d/0x450 [cifs] [598428.945909] process_one_work+0x1db/0x380 [598428.945914] worker_thread+0x4d/0x400 [598428.945921] kthread+0x104/0x140 [598428.945925] ? process_one_work+0x380/0x380 [598428.945931] ? kthread_park+0x80/0x80 [598428.945937] ret_from_fork+0x35/0x40
Fix this by reordering initialization steps of the cifsFileInfo
structure: initialize all the fields first and then add the new byte-range lock list to the inode's lock list.
--- fs/cifs/file.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
fs/cifs/file.c +4 -3
@@ -313,9 +313,6 @@ cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,INIT_LIST_HEAD (&fdlocks->locks); fdlocks->cfile = cfile; cfile->llist = fdlocks; -cifs_down_write (&cinode->lock_sem); -list_add (&fdlocks->llist, &cinode->llist); -up_write (&cinode->lock_sem); cfile->count =1 ; cfile->pid = current->tgid;
@@ -339,6 +336,10 @@ cifs_new_fileinfo(struct cifs_fid *fid, struct file *file, oplock =0 ; } +cifs_down_write (&cinode->lock_sem); +list_add (&fdlocks->llist, &cinode->llist); +up_write (&cinode->lock_sem); +spin_lock (&tcon->open_file_lock);if (fid->pending_open->oplock !=CIFS_OPLOCK_NO_CHANGE && oplock) oplock = fid->pending_open->oplock;
Signature · 1 line
2.17.1
Steve French
AppliedMerged into cifs-2.6.git for-next
Quoted 65 lines · Pavel Shilovsky
Currently when the client creates a cifsFileInfo structure for a newly opened file, it allocates a list of byte-range locks with a pointer to the new cfile and attaches this list to the inode's lock list. The latter happens before initializing all other fields, e.g.
cfile->tlink. Thus a partially initialized cifsFileInfo structure becomes available to other threads that walk through the inode's lock list. One example of such a thread may be an oplock break worker thread that tries to push all cached byte-range locks. This causes NULL-pointer dereference insmb2_push_mandatory_locks()when accessingcfile->tlink:[598428.945633] BUG: kernel NULL pointer dereference, address: 0000000000000038 ... [598428.945749] Workqueue: cifsoplockd cifs_oplock_break [cifs] [598428.945793] RIP: 0010:smb2_push_mandatory_locks+0xd6/0x5a0 [cifs] ... [598428.945834] Call Trace: [598428.945870] ? cifs_revalidate_mapping+0x45/0x90 [cifs] [598428.945901] cifs_oplock_break+0x13d/0x450 [cifs] [598428.945909] process_one_work+0x1db/0x380 [598428.945914] worker_thread+0x4d/0x400 [598428.945921] kthread+0x104/0x140 [598428.945925] ? process_one_work+0x380/0x380 [598428.945931] ? kthread_park+0x80/0x80 [598428.945937] ret_from_fork+0x35/0x40Fix this by reordering initialization steps of the cifsFileInfo
structure: initialize all the fields first and then add the new byte-range lock list to the inode's lock list.Cc Stable <[email protected]>Signed-off-by Pavel Shilovsky <[email protected]>--- fs/cifs/file.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)fs/cifs/file.c +4
-4@@ -313,9 +313,6 @@ cifs_new_fileinfo(struct cifs_fid *fid, struct file *file,INIT_LIST_HEAD (&fdlocks->locks); fdlocks->cfile = cfile; cfile->llist = fdlocks; -cifs_down_write (&cinode->lock_sem); -list_add (&fdlocks->llist, &cinode->llist); -up_write (&cinode->lock_sem); cfile->count =1 ; cfile->pid = current->tgid;@@ -339,6 +336,10 @@ cifs_new_fileinfo(struct cifs_fid *fid, struct file *file, oplock =0 ; } +cifs_down_write (&cinode->lock_sem); +list_add (&fdlocks->llist, &cinode->llist); +up_write (&cinode->lock_sem); +spin_lock (&tcon->open_file_lock);if (fid->pending_open->oplock !=CIFS_OPLOCK_NO_CHANGE && oplock) oplock = fid->pending_open->oplock; --2.17.1
Signature · 3 lines
Thanks, Steve
Quoted 10 lines · Pavel Shilovsky
Currently when the client creates a cifsFileInfo structure for a newly opened file, it allocates a list of byte-range locks with a pointer to the new cfile and attaches this list to the inode's lock list. The latter happens before initializing all other fields, e.g.
cfile->tlink. Thus a partially initialized cifsFileInfo structure becomes available to other threads that walk through the inode's lock list. One example of such a thread may be an oplock break worker thread that tries to push all cached byte-range locks. This causes NULL-pointer dereference insmb2_push_mandatory_locks()when accessingcfile->tlink:
reviewing late but this makes sense.
Signature · 4 lines
Aurélien Aptel / SUSE Labs Samba Team GPG: 1839 CB5F 9F5B FB9B AA97 8C99 03C8 A49B 521B D5D3 SUSE Software Solutions Germany GmbH, Maxfeldstr. 5, 90409 Nürnberg, DE GF: Felix Imendörffer, Mary Higgins, Sri Rasiah HRB 247165 (AG München)