| Message ID | 20201112104512.17472-1-lhenriques@suse.de (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | ceph: fix race in concurrent __ceph_remove_cap invocations | expand |
On Thu, 2020-11-12 at 10:45 +0000, Luis Henriques wrote: > A NULL pointer dereference may occur in __ceph_remove_cap with some of the > callbacks used in ceph_iterate_session_caps, namely trim_caps_cb and > remove_session_caps_cb. These aren't protected against the concurrent > execution of __ceph_remove_cap. > > Since the callers of this function hold the i_ceph_lock, the fix is simply > a matter of returning immediately if caps->ci is NULL. > > Based on a patch from Jeff Layton. > > Cc: stable@vger.kernel.org > URL: https://tracker.ceph.com/issues/43272 > Link: https://www.spinics.net/lists/ceph-devel/msg47064.html > Signed-off-by: Luis Henriques <lhenriques@suse.de> > --- > fs/ceph/caps.c | 11 +++++++++-- > 1 file changed, 9 insertions(+), 2 deletions(-) > > diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c > index ded4229c314a..443f164760d5 100644 > --- a/fs/ceph/caps.c > +++ b/fs/ceph/caps.c > @@ -1140,12 +1140,19 @@ void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release) > { > struct ceph_mds_session *session = cap->session; > struct ceph_inode_info *ci = cap->ci; > - struct ceph_mds_client *mdsc = > - ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; > + struct ceph_mds_client *mdsc; > int removed = 0; > > > + /* 'ci' being NULL means he remove have already occurred */ > + if (!ci) { > + dout("%s: cap inode is NULL\n", __func__); > + return; > + } > + > dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); > > > + mdsc = ceph_inode_to_client(&ci->vfs_inode)->mdsc; > + > /* remove from inode's cap rbtree, and clear auth cap */ > rb_erase(&cap->ci_node, &ci->i_caps); > if (ci->i_auth_cap == cap) { Merged into testing branch (with a minor fix to the comment).
On Thu, Nov 12, 2020 at 6:48 PM Luis Henriques <lhenriques@suse.de> wrote: > > A NULL pointer dereference may occur in __ceph_remove_cap with some of the > callbacks used in ceph_iterate_session_caps, namely trim_caps_cb and > remove_session_caps_cb. These aren't protected against the concurrent > execution of __ceph_remove_cap. > they are protected by session mutex, never get executed concurrently > Since the callers of this function hold the i_ceph_lock, the fix is simply > a matter of returning immediately if caps->ci is NULL. > > Based on a patch from Jeff Layton. > > Cc: stable@vger.kernel.org > URL: https://tracker.ceph.com/issues/43272 > Link: https://www.spinics.net/lists/ceph-devel/msg47064.html > Signed-off-by: Luis Henriques <lhenriques@suse.de> > --- > fs/ceph/caps.c | 11 +++++++++-- > 1 file changed, 9 insertions(+), 2 deletions(-) > > diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c > index ded4229c314a..443f164760d5 100644 > --- a/fs/ceph/caps.c > +++ b/fs/ceph/caps.c > @@ -1140,12 +1140,19 @@ void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release) > { > struct ceph_mds_session *session = cap->session; > struct ceph_inode_info *ci = cap->ci; > - struct ceph_mds_client *mdsc = > - ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; > + struct ceph_mds_client *mdsc; > int removed = 0; > > + /* 'ci' being NULL means he remove have already occurred */ > + if (!ci) { > + dout("%s: cap inode is NULL\n", __func__); > + return; > + } > + > dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); > > + mdsc = ceph_inode_to_client(&ci->vfs_inode)->mdsc; > + > /* remove from inode's cap rbtree, and clear auth cap */ > rb_erase(&cap->ci_node, &ci->i_caps); > if (ci->i_auth_cap == cap) {
On Thu, 2020-11-12 at 20:43 +0800, Yan, Zheng wrote: > On Thu, Nov 12, 2020 at 6:48 PM Luis Henriques <lhenriques@suse.de> wrote: > > > > A NULL pointer dereference may occur in __ceph_remove_cap with some of the > > callbacks used in ceph_iterate_session_caps, namely trim_caps_cb and > > remove_session_caps_cb. These aren't protected against the concurrent > > execution of __ceph_remove_cap. > > > > they are protected by session mutex, never get executed concurrently > Maybe not concurrently with one another, but the s_mutex is _not_ held when __ceph_remove_caps is called from ceph_evict_inode. We can't rely on it to protect this. > > Since the callers of this function hold the i_ceph_lock, the fix is simply > > a matter of returning immediately if caps->ci is NULL. > > > > Based on a patch from Jeff Layton. > > > > Cc: stable@vger.kernel.org > > URL: https://tracker.ceph.com/issues/43272 > > Link: https://www.spinics.net/lists/ceph-devel/msg47064.html > > Signed-off-by: Luis Henriques <lhenriques@suse.de> > > --- > > fs/ceph/caps.c | 11 +++++++++-- > > 1 file changed, 9 insertions(+), 2 deletions(-) > > > > diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c > > index ded4229c314a..443f164760d5 100644 > > --- a/fs/ceph/caps.c > > +++ b/fs/ceph/caps.c > > @@ -1140,12 +1140,19 @@ void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release) > > { > > struct ceph_mds_session *session = cap->session; > > struct ceph_inode_info *ci = cap->ci; > > - struct ceph_mds_client *mdsc = > > - ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; > > + struct ceph_mds_client *mdsc; > > int removed = 0; > > > > + /* 'ci' being NULL means he remove have already occurred */ > > + if (!ci) { > > + dout("%s: cap inode is NULL\n", __func__); > > + return; > > + } > > + > > dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); > > > > + mdsc = ceph_inode_to_client(&ci->vfs_inode)->mdsc; > > + > > /* remove from inode's cap rbtree, and clear auth cap */ > > rb_erase(&cap->ci_node, &ci->i_caps); > > if (ci->i_auth_cap == cap) {
Jeff Layton <jlayton@kernel.org> writes: > On Thu, 2020-11-12 at 20:43 +0800, Yan, Zheng wrote: >> On Thu, Nov 12, 2020 at 6:48 PM Luis Henriques <lhenriques@suse.de> wrote: >> > >> > A NULL pointer dereference may occur in __ceph_remove_cap with some of the >> > callbacks used in ceph_iterate_session_caps, namely trim_caps_cb and >> > remove_session_caps_cb. These aren't protected against the concurrent >> > execution of __ceph_remove_cap. >> > >> >> they are protected by session mutex, never get executed concurrently >> > > Maybe not concurrently with one another, but the s_mutex is _not_ held > when __ceph_remove_caps is called from ceph_evict_inode. We can't rely > on it to protect this. Hmm, yeah. I guess the changelog could mention that. Thanks, Jeff. Cheers,
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c index ded4229c314a..443f164760d5 100644 --- a/fs/ceph/caps.c +++ b/fs/ceph/caps.c @@ -1140,12 +1140,19 @@ void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release) { struct ceph_mds_session *session = cap->session; struct ceph_inode_info *ci = cap->ci; - struct ceph_mds_client *mdsc = - ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc; + struct ceph_mds_client *mdsc; int removed = 0; + /* 'ci' being NULL means he remove have already occurred */ + if (!ci) { + dout("%s: cap inode is NULL\n", __func__); + return; + } + dout("__ceph_remove_cap %p from %p\n", cap, &ci->vfs_inode); + mdsc = ceph_inode_to_client(&ci->vfs_inode)->mdsc; + /* remove from inode's cap rbtree, and clear auth cap */ rb_erase(&cap->ci_node, &ci->i_caps); if (ci->i_auth_cap == cap) {
A NULL pointer dereference may occur in __ceph_remove_cap with some of the callbacks used in ceph_iterate_session_caps, namely trim_caps_cb and remove_session_caps_cb. These aren't protected against the concurrent execution of __ceph_remove_cap. Since the callers of this function hold the i_ceph_lock, the fix is simply a matter of returning immediately if caps->ci is NULL. Based on a patch from Jeff Layton. Cc: stable@vger.kernel.org URL: https://tracker.ceph.com/issues/43272 Link: https://www.spinics.net/lists/ceph-devel/msg47064.html Signed-off-by: Luis Henriques <lhenriques@suse.de> --- fs/ceph/caps.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)