Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
To reproduce the issue the following script can be used:
#!/bin/bash
CGROOT=/sys/fs/cgroup
mkdir -p ${CGROOT}/A ${CGROOT}/B ${CGROOT}/A/C sleep 1
./test_cgrp2_attach ${CGROOT}/A egress & A_PID=$! ./test_cgrp2_attach ${CGROOT}/B egress & B_PID=$!
echo $$ > ${CGROOT}/A/C/cgroup.procs iperf -s & S_PID=$! iperf -c localhost -t 100 & C_PID=$!
sleep 1
echo $$ > ${CGROOT}/B/cgroup.procs echo ${S_PID} > ${CGROOT}/B/cgroup.procs echo ${C_PID} > ${CGROOT}/B/cgroup.procs
sleep 1
rmdir ${CGROOT}/A/C rmdir ${CGROOT}/A
sleep 1
kill -9 ${S_PID} ${C_PID} ${A_PID} ${B_PID}
test_cgrp2_attach is an example from samples/bpf with the following patch applied (required to close cgroup and bpf program file descriptors after attachment):
diff --git a/samples/bpf/test_cgrp2_attach.c b/samples/bpf/test_cgrp2_attach.c index 20fbd1241db3..7c7d0e91204d 100644 --- a/samples/bpf/test_cgrp2_attach.c +++ b/samples/bpf/test_cgrp2_attach.c @@ -111,6 +111,8 @@ static int attach_filter(int cg_fd, int type, int verdict) strerror(errno)); return EXIT_FAILURE; } + close(cg_fd); + close(prog_fd); while (1) { key = MAP_KEY_PACKETS; assert(bpf_map_lookup_elem(map_fd, &key, &pkt_cnt) == 0);
On the unpatched kernel the following stacktrace can be obtained:
[ 33.619799] BUG: unable to handle page fault for address: ffffbdb4801ab002 [ 33.620677] #PF: supervisor read access in kernel mode [ 33.621293] #PF: error_code(0x0000) - not-present page [ 33.621918] PGD 236d59067 P4D 236d59067 PUD 236d5c067 PMD 236d5d067 PTE 0 [ 33.622754] Oops: 0000 [#1] SMP NOPTI [ 33.623202] CPU: 0 PID: 601 Comm: iperf Not tainted 5.5.0-rc2+ #23 [ 33.623943] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20190727_073836-buildvm-ppc64le-16.ppc.f4 [ 33.625545] RIP: 0010:__cgroup_bpf_run_filter_skb+0x29f/0x3d0 [ 33.626231] Code: f6 0f 84 3a 01 00 00 49 8d 47 30 31 db 48 89 44 24 30 48 8b 45 08 65 48 89 05 4d 9d e0 64 48 8b d [ 33.628431] RSP: 0018:ffffbdb4802ffa90 EFLAGS: 00010246 [ 33.629051] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000034 [ 33.629906] RDX: 0000000000000000 RSI: ffff9ddf9d7a0000 RDI: ffff9ddf9b97f1c0 [ 33.630761] RBP: ffff9ddf9d4899d0 R08: ffff9ddfb67ddd80 R09: 0000000000010000 [ 33.631616] R10: 0000000000000070 R11: ffffbdb4802ffde8 R12: ffff9ddf9ba858e0 [ 33.632463] R13: 0000000000000001 R14: ffffbdb4801ab000 R15: ffff9ddf9ba858e0 [ 33.633306] FS: 00007f9d15ed9700(0000) GS:ffff9ddfb7c00000(0000) knlGS:0000000000000000 [ 33.634262] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 33.634945] CR2: ffffbdb4801ab002 CR3: 000000021b94e000 CR4: 00000000003406f0 [ 33.635809] Call Trace: [ 33.636118] ? __cgroup_bpf_run_filter_skb+0x2bf/0x3d0 [ 33.636728] ? __switch_to_asm+0x40/0x70 [ 33.637196] ip_finish_output+0x68/0xa0 [ 33.637654] ip_output+0x76/0xf0 [ 33.638046] ? __ip_finish_output+0x1c0/0x1c0 [ 33.638576] __ip_queue_xmit+0x157/0x410 [ 33.639049] __tcp_transmit_skb+0x535/0xaf0 [ 33.639557] tcp_write_xmit+0x378/0x1190 [ 33.640049] ? _copy_from_iter_full+0x8d/0x260 [ 33.640592] tcp_sendmsg_locked+0x2a2/0xdc0 [ 33.641098] ? sock_has_perm+0x10/0xa0 [ 33.641574] tcp_sendmsg+0x28/0x40 [ 33.641985] sock_sendmsg+0x57/0x60 [ 33.642411] sock_write_iter+0x97/0x100 [ 33.642876] new_sync_write+0x1b6/0x1d0 [ 33.643339] vfs_write+0xb6/0x1a0 [ 33.643752] ksys_write+0xa7/0xe0 [ 33.644156] do_syscall_64+0x5b/0x1b0 [ 33.644605] entry_SYSCALL_64_after_hwframe+0x44/0xa9
Fix this by grabbing a reference to the bpf structure of each ancestor on the initialization of the cgroup bpf structure, and dropping the reference at the end of releasing the cgroup bpf structure.
This will restore the hierarchical order of cgroup bpf releasing, without adding any operations on hot paths.
Thanks to Josef Bacik for the debugging and the initial analysis of the problem.
Fixes: 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") Reported-by: Josef Bacik josef@toxicpanda.com Signed-off-by: Roman Gushchin guro@fb.com Cc: Alexei Starovoitov ast@kernel.org Cc: stable@vger.kernel.org --- kernel/bpf/cgroup.c | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-)
diff --git a/kernel/bpf/cgroup.c b/kernel/bpf/cgroup.c index 4fb20ab179fe..9e43b72eb619 100644 --- a/kernel/bpf/cgroup.c +++ b/kernel/bpf/cgroup.c @@ -35,8 +35,8 @@ void cgroup_bpf_offline(struct cgroup *cgrp) */ static void cgroup_bpf_release(struct work_struct *work) { - struct cgroup *cgrp = container_of(work, struct cgroup, - bpf.release_work); + struct cgroup *p, *cgrp = container_of(work, struct cgroup, + bpf.release_work); enum bpf_cgroup_storage_type stype; struct bpf_prog_array *old_array; unsigned int type; @@ -65,6 +65,9 @@ static void cgroup_bpf_release(struct work_struct *work)
mutex_unlock(&cgroup_mutex);
+ for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p)) + cgroup_bpf_put(p); + percpu_ref_exit(&cgrp->bpf.refcnt); cgroup_put(cgrp); } @@ -199,6 +202,7 @@ int cgroup_bpf_inherit(struct cgroup *cgrp) */ #define NR ARRAY_SIZE(cgrp->bpf.effective) struct bpf_prog_array *arrays[NR] = {}; + struct cgroup *p; int ret, i;
ret = percpu_ref_init(&cgrp->bpf.refcnt, cgroup_bpf_release_fn, 0, @@ -206,6 +210,9 @@ int cgroup_bpf_inherit(struct cgroup *cgrp) if (ret) return ret;
+ for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p)) + cgroup_bpf_get(p); + for (i = 0; i < NR; i++) INIT_LIST_HEAD(&cgrp->bpf.progs[i]);
On Fri, Dec 27, 2019 at 01:50:34PM -0800, Roman Gushchin wrote:
Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
To reproduce the issue the following script can be used:
#!/bin/bash
CGROOT=/sys/fs/cgroup
mkdir -p ${CGROOT}/A ${CGROOT}/B ${CGROOT}/A/C sleep 1
./test_cgrp2_attach ${CGROOT}/A egress & A_PID=$! ./test_cgrp2_attach ${CGROOT}/B egress & B_PID=$!
echo $$ > ${CGROOT}/A/C/cgroup.procs iperf -s & S_PID=$! iperf -c localhost -t 100 & C_PID=$!
sleep 1
echo $$ > ${CGROOT}/B/cgroup.procs echo ${S_PID} > ${CGROOT}/B/cgroup.procs echo ${C_PID} > ${CGROOT}/B/cgroup.procs
sleep 1
rmdir ${CGROOT}/A/C rmdir ${CGROOT}/A
sleep 1
kill -9 ${S_PID} ${C_PID} ${A_PID} ${B_PID}
test_cgrp2_attach is an example from samples/bpf with the following patch applied (required to close cgroup and bpf program file descriptors after attachment):
diff --git a/samples/bpf/test_cgrp2_attach.c b/samples/bpf/test_cgrp2_attach.c index 20fbd1241db3..7c7d0e91204d 100644 --- a/samples/bpf/test_cgrp2_attach.c +++ b/samples/bpf/test_cgrp2_attach.c @@ -111,6 +111,8 @@ static int attach_filter(int cg_fd, int type, int verdict) strerror(errno)); return EXIT_FAILURE; }
close(cg_fd);close(prog_fd); while (1) { key = MAP_KEY_PACKETS; assert(bpf_map_lookup_elem(map_fd, &key, &pkt_cnt) == 0);On the unpatched kernel the following stacktrace can be obtained:
[ 33.619799] BUG: unable to handle page fault for address: ffffbdb4801ab002 [ 33.620677] #PF: supervisor read access in kernel mode [ 33.621293] #PF: error_code(0x0000) - not-present page [ 33.621918] PGD 236d59067 P4D 236d59067 PUD 236d5c067 PMD 236d5d067 PTE 0 [ 33.622754] Oops: 0000 [#1] SMP NOPTI [ 33.623202] CPU: 0 PID: 601 Comm: iperf Not tainted 5.5.0-rc2+ #23 [ 33.623943] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ?-20190727_073836-buildvm-ppc64le-16.ppc.f4 [ 33.625545] RIP: 0010:__cgroup_bpf_run_filter_skb+0x29f/0x3d0 [ 33.626231] Code: f6 0f 84 3a 01 00 00 49 8d 47 30 31 db 48 89 44 24 30 48 8b 45 08 65 48 89 05 4d 9d e0 64 48 8b d [ 33.628431] RSP: 0018:ffffbdb4802ffa90 EFLAGS: 00010246 [ 33.629051] RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000034 [ 33.629906] RDX: 0000000000000000 RSI: ffff9ddf9d7a0000 RDI: ffff9ddf9b97f1c0 [ 33.630761] RBP: ffff9ddf9d4899d0 R08: ffff9ddfb67ddd80 R09: 0000000000010000 [ 33.631616] R10: 0000000000000070 R11: ffffbdb4802ffde8 R12: ffff9ddf9ba858e0 [ 33.632463] R13: 0000000000000001 R14: ffffbdb4801ab000 R15: ffff9ddf9ba858e0 [ 33.633306] FS: 00007f9d15ed9700(0000) GS:ffff9ddfb7c00000(0000) knlGS:0000000000000000 [ 33.634262] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 33.634945] CR2: ffffbdb4801ab002 CR3: 000000021b94e000 CR4: 00000000003406f0 [ 33.635809] Call Trace: [ 33.636118] ? __cgroup_bpf_run_filter_skb+0x2bf/0x3d0 [ 33.636728] ? __switch_to_asm+0x40/0x70 [ 33.637196] ip_finish_output+0x68/0xa0 [ 33.637654] ip_output+0x76/0xf0 [ 33.638046] ? __ip_finish_output+0x1c0/0x1c0 [ 33.638576] __ip_queue_xmit+0x157/0x410 [ 33.639049] __tcp_transmit_skb+0x535/0xaf0 [ 33.639557] tcp_write_xmit+0x378/0x1190 [ 33.640049] ? _copy_from_iter_full+0x8d/0x260 [ 33.640592] tcp_sendmsg_locked+0x2a2/0xdc0 [ 33.641098] ? sock_has_perm+0x10/0xa0 [ 33.641574] tcp_sendmsg+0x28/0x40 [ 33.641985] sock_sendmsg+0x57/0x60 [ 33.642411] sock_write_iter+0x97/0x100 [ 33.642876] new_sync_write+0x1b6/0x1d0 [ 33.643339] vfs_write+0xb6/0x1a0 [ 33.643752] ksys_write+0xa7/0xe0 [ 33.644156] do_syscall_64+0x5b/0x1b0 [ 33.644605] entry_SYSCALL_64_after_hwframe+0x44/0xa9
Fix this by grabbing a reference to the bpf structure of each ancestor on the initialization of the cgroup bpf structure, and dropping the reference at the end of releasing the cgroup bpf structure.
This will restore the hierarchical order of cgroup bpf releasing, without adding any operations on hot paths.
Thanks to Josef Bacik for the debugging and the initial analysis of the problem.
Fixes: 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") Reported-by: Josef Bacik josef@toxicpanda.com Signed-off-by: Roman Gushchin guro@fb.com Cc: Alexei Starovoitov ast@kernel.org Cc: stable@vger.kernel.org
A friendly ping!
Thanks!
On Fri, Dec 27, 2019 at 1:50 PM Roman Gushchin guro@fb.com wrote:
Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
[...]
Thanks to Josef Bacik for the debugging and the initial analysis of the problem.
Fixes: 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") Reported-by: Josef Bacik josef@toxicpanda.com Signed-off-by: Roman Gushchin guro@fb.com Cc: Alexei Starovoitov ast@kernel.org Cc: stable@vger.kernel.org
LGTM. Thanks for the fix!
Acked-by: Song Liu songliubraving@fb.com
On Fri, Dec 27, 2019 at 01:50:34PM -0800, Roman Gushchin wrote:
Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
...
@@ -65,6 +65,9 @@ static void cgroup_bpf_release(struct work_struct *work) mutex_unlock(&cgroup_mutex);
- for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
cgroup_bpf_put(p);
The fix makes sense, but is it really safe to walk cgroup hierarchy without holding cgroup_mutex?
On Fri, Jan 03, 2020 at 04:35:25PM -0800, Alexei Starovoitov wrote:
On Fri, Dec 27, 2019 at 01:50:34PM -0800, Roman Gushchin wrote:
Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
...
@@ -65,6 +65,9 @@ static void cgroup_bpf_release(struct work_struct *work) mutex_unlock(&cgroup_mutex);
- for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
cgroup_bpf_put(p);The fix makes sense, but is it really safe to walk cgroup hierarchy without holding cgroup_mutex?
It is, because we're holding a reference to the original cgroup and going towards the root. On each level the cgroup is protected by a reference from their child cgroup.
Thanks!
On Sat, Jan 04, 2020 at 01:13:24AM +0000, Roman Gushchin wrote:
On Fri, Jan 03, 2020 at 04:35:25PM -0800, Alexei Starovoitov wrote:
On Fri, Dec 27, 2019 at 01:50:34PM -0800, Roman Gushchin wrote:
Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
...
@@ -65,6 +65,9 @@ static void cgroup_bpf_release(struct work_struct *work) mutex_unlock(&cgroup_mutex);
- for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
cgroup_bpf_put(p);The fix makes sense, but is it really safe to walk cgroup hierarchy without holding cgroup_mutex?
It is, because we're holding a reference to the original cgroup and going towards the root. On each level the cgroup is protected by a reference from their child cgroup.
cgroup_bpf_put(p) can make bpf.refcnt zero which may call cgroup_bpf_release() on another cpu which will do cgroup_put() and this cpu p = cgroup_parent(p) would be use-after-free? May be not due to the way work_queues are implemented. But it feels dangerous to have such delicate release logic. Why not to move the loop under the mutex and make things obvious?
On Fri, Jan 03, 2020 at 06:31:14PM -0800, Alexei Starovoitov wrote:
On Sat, Jan 04, 2020 at 01:13:24AM +0000, Roman Gushchin wrote:
On Fri, Jan 03, 2020 at 04:35:25PM -0800, Alexei Starovoitov wrote:
On Fri, Dec 27, 2019 at 01:50:34PM -0800, Roman Gushchin wrote:
Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
...
@@ -65,6 +65,9 @@ static void cgroup_bpf_release(struct work_struct *work) mutex_unlock(&cgroup_mutex);
- for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
cgroup_bpf_put(p);The fix makes sense, but is it really safe to walk cgroup hierarchy without holding cgroup_mutex?
It is, because we're holding a reference to the original cgroup and going towards the root. On each level the cgroup is protected by a reference from their child cgroup.
cgroup_bpf_put(p) can make bpf.refcnt zero which may call cgroup_bpf_release() on another cpu which will do cgroup_put() and this cpu p = cgroup_parent(p) would be use-after-free? May be not due to the way work_queues are implemented. But it feels dangerous to have such delicate release logic.
If I understand your concern correctly: you assume that parent's cgroup_bpf_release() can be finished prior to the child's one and the final cgroup_put() will release the parent?
If so, it's not possible, because the child hold a reference to the parent (independent to all cgroup bpf stuff), which exists at least until the final cgroup_put() in cgroup_bpf_release(). Please, look at css_free_rwork_fn() for details.
Why not to move the loop under the mutex and make things obvious?
Traversing the cgroup tree to the root cgroup without additional locking seems pretty common to me. You can find a ton of examples in mm/memcontrol.c. So it doesn't look scary or adventurous to me.
I think it doesn't matter that much here, so I'm ok with putting it under the mutex, but IMO it won't make the code any safer.
cc Tejun for the second opinion on cgroup locking
On Sat, Jan 04, 2020 at 03:00:46AM +0000, Roman Gushchin wrote:
On Fri, Jan 03, 2020 at 06:31:14PM -0800, Alexei Starovoitov wrote:
On Sat, Jan 04, 2020 at 01:13:24AM +0000, Roman Gushchin wrote:
On Fri, Jan 03, 2020 at 04:35:25PM -0800, Alexei Starovoitov wrote:
On Fri, Dec 27, 2019 at 01:50:34PM -0800, Roman Gushchin wrote:
Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
...
@@ -65,6 +65,9 @@ static void cgroup_bpf_release(struct work_struct *work) mutex_unlock(&cgroup_mutex);
- for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
cgroup_bpf_put(p);The fix makes sense, but is it really safe to walk cgroup hierarchy without holding cgroup_mutex?
It is, because we're holding a reference to the original cgroup and going towards the root. On each level the cgroup is protected by a reference from their child cgroup.
cgroup_bpf_put(p) can make bpf.refcnt zero which may call cgroup_bpf_release() on another cpu which will do cgroup_put() and this cpu p = cgroup_parent(p) would be use-after-free? May be not due to the way work_queues are implemented. But it feels dangerous to have such delicate release logic.
If I understand your concern correctly: you assume that parent's cgroup_bpf_release() can be finished prior to the child's one and the final cgroup_put() will release the parent?
If so, it's not possible, because the child hold a reference to the parent (independent to all cgroup bpf stuff), which exists at least until the final cgroup_put() in cgroup_bpf_release(). Please, look at css_free_rwork_fn() for details.
Why not to move the loop under the mutex and make things obvious?
Traversing the cgroup tree to the root cgroup without additional locking seems pretty common to me. You can find a ton of examples in mm/memcontrol.c. So it doesn't look scary or adventurous to me.
I think it doesn't matter that much here, so I'm ok with putting it under the mutex, but IMO it won't make the code any safer.
cc Tejun for the second opinion on cgroup locking
Checked with TJ offline. This seems fine.
I tweaked commit log: - extra 'diff' lines were confusing 'git am' - commit description shouldn't be split into multiline
And applied to bpf tree. Thanks
On Mon, Jan 06, 2020 at 02:07:48PM -0800, Alexei Starovoitov wrote:
On Sat, Jan 04, 2020 at 03:00:46AM +0000, Roman Gushchin wrote:
On Fri, Jan 03, 2020 at 06:31:14PM -0800, Alexei Starovoitov wrote:
On Sat, Jan 04, 2020 at 01:13:24AM +0000, Roman Gushchin wrote:
On Fri, Jan 03, 2020 at 04:35:25PM -0800, Alexei Starovoitov wrote:
On Fri, Dec 27, 2019 at 01:50:34PM -0800, Roman Gushchin wrote:
Before commit 4bfc0bb2c60e ("bpf: decouple the lifetime of cgroup_bpf from cgroup itself") cgroup bpf structures were released with corresponding cgroup structures. It guaranteed the hierarchical order of destruction: children were always first. It preserved attached programs from being released before their propagated copies.
But with cgroup auto-detachment there are no such guarantees anymore: cgroup bpf is released as soon as the cgroup is offline and there are no live associated sockets. It means that an attached program can be detached and released, while its propagated copy is still living in the cgroup subtree. This will obviously lead to an use-after-free bug.
...
@@ -65,6 +65,9 @@ static void cgroup_bpf_release(struct work_struct *work) mutex_unlock(&cgroup_mutex);
- for (p = cgroup_parent(cgrp); p; p = cgroup_parent(p))
cgroup_bpf_put(p);The fix makes sense, but is it really safe to walk cgroup hierarchy without holding cgroup_mutex?
It is, because we're holding a reference to the original cgroup and going towards the root. On each level the cgroup is protected by a reference from their child cgroup.
cgroup_bpf_put(p) can make bpf.refcnt zero which may call cgroup_bpf_release() on another cpu which will do cgroup_put() and this cpu p = cgroup_parent(p) would be use-after-free? May be not due to the way work_queues are implemented. But it feels dangerous to have such delicate release logic.
If I understand your concern correctly: you assume that parent's cgroup_bpf_release() can be finished prior to the child's one and the final cgroup_put() will release the parent?
If so, it's not possible, because the child hold a reference to the parent (independent to all cgroup bpf stuff), which exists at least until the final cgroup_put() in cgroup_bpf_release(). Please, look at css_free_rwork_fn() for details.
Why not to move the loop under the mutex and make things obvious?
Traversing the cgroup tree to the root cgroup without additional locking seems pretty common to me. You can find a ton of examples in mm/memcontrol.c. So it doesn't look scary or adventurous to me.
I think it doesn't matter that much here, so I'm ok with putting it under the mutex, but IMO it won't make the code any safer.
cc Tejun for the second opinion on cgroup locking
Checked with TJ offline. This seems fine.
I tweaked commit log:
- extra 'diff' lines were confusing 'git am'
- commit description shouldn't be split into multiline
Hm, I thought we don't break it only on the "Fixes:" line. Maybe it's subtree-dependent :)
And applied to bpf tree. Thanks
Thank you!
linux-stable-mirror@lists.linaro.org
-
Alexei Starovoitov -
Roman Gushchin -
Song Liu