Skip to contentSkip to conversation

bpf, sockmap: disallow sockmap mutation from tc, xdp and flow_dissector

PATCH Superseded Sechang Lim 5 participants 9 messages Jun 29 – Jun 30

Subsystems (6)

v3 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v3 cover bpf, sockmap: disallow sockmap mutation from tc, xdp and flow_dissector

A tc, xdp or flow_dissector program updating or deleting a sockmap deadlocks on stab->lock vs sk_callback_lock and has no reason to. Patch 1 disallows it in may_update_sockmap(); patch 2 drops the selftests that exercised it.

v3:
- drop the broken selftests (Jiayuan Chen)
- drop the Fixes tag and target bpf-next (Jiayuan Chen)

v2:
- https://lore.kernel.org/all/[email protected]/

v1:
- https://lore.kernel.org/all/[email protected]/

Sechang Lim (2):
bpf, sockmap: disallow update and delete from tc, xdp and
flow_dissector
selftests/bpf: drop tc/xdp/flow_dissector sockmap mutation tests

 kernel/bpf/verifier.c                         |  4 --
 .../selftests/bpf/prog_tests/fexit_bpf2bpf.c  | 13 -----
 .../selftests/bpf/prog_tests/sockmap_basic.c  | 52 -------------------
 .../bpf/progs/freplace_cls_redirect.c         | 34 ------------
 .../selftests/bpf/progs/test_sockmap_update.c | 48 -----------------
 .../bpf/progs/verifier_sockmap_mutate.c       | 10 ++--
 6 files changed, 5 insertions(+), 156 deletions(-)
 delete mode 100644 tools/testing/selftests/bpf/progs/freplace_cls_redirect.c
 delete mode 100644 tools/testing/selftests/bpf/progs/test_sockmap_update.c
Signature · 1 line
2.43.0
#1 lore raw

PATCH v3 1/2 bpf, sockmap: disallow update and delete from tc, xdp and flow_dissector

sock_map_update_common() and __sock_map_delete() hold stab->lock and call sock_map_unref() -> sock_map_del_link(), which takes sk_callback_lock for write. That gives the order stab->lock -> sk_callback_lock.

The reverse order comes from the SK_SKB stream parser. sk_psock_strp_data_ready() holds sk_callback_lock for read, and after the verdict tcp_bpf_strp_read_sock() acks the consumed data inline via __tcp_cleanup_rbuf(). The ACK goes out egress, where a sched_cls program deletes from the sockmap and takes stab->lock:

WARNING: possible circular locking dependency detected
------------------------------------------------------
syz.9.8824 is trying to acquire lock:
(&stab->lock){+.-.}-{3:3}, at: __sock_map_delete net/core/sock_map.c:421 but task is already holding lock:
(clock-AF_INET){++.-}-{3:3}, at: sk_psock_strp_data_ready net/core/skmsg.c:1173

-> #1 (clock-AF_INET){++.-}-{3:3}:

         _raw_write_lock_bh
         sock_map_del_link net/core/sock_map.c:167
         sock_map_unref net/core/sock_map.c:184
         sock_map_update_common net/core/sock_map.c:509
         sock_map_update_elem_sys net/core/sock_map.c:588
         map_update_elem kernel/bpf/syscall.c:1805

-> #0 (&stab->lock){+.-.}-{3:3}:

         _raw_spin_lock_bh
         __sock_map_delete net/core/sock_map.c:421
         sock_map_delete_elem net/core/sock_map.c:452
         bpf_prog_06044d24140080b6
         tcx_run net/core/dev.c:4451
         sch_handle_egress net/core/dev.c:4541
         __dev_queue_xmit net/core/dev.c:4808
         ...
         tcp_bpf_strp_read_sock net/ipv4/tcp_bpf.c:701
         strp_data_ready net/strparser/strparser.c:402
         sk_psock_strp_data_ready net/core/skmsg.c:1174
         tcp_data_queue net/ipv4/tcp_input.c:5661
  Possible unsafe locking scenario:
         CPU0                    CPU1
         ----                    ----
    rlock(clock-AF_INET);
                                 lock(&stab->lock);
                                 lock(clock-AF_INET);
    lock(&stab->lock);

*** DEADLOCK ***

A tc, xdp or flow_dissector program has no reason to update or delete a sockmap, and redirect does not go through here. Drop them from may_update_sockmap() so the verifier rejects it. It also closes the matching sockhash inversion.

---
 kernel/bpf/verifier.c | 4 ----
 1 file changed, 4 deletions(-)
kernel/bpf/verifier.c +0 -4
index 25aea4271cd0..58d766c34626 100644
@@ -8489,11 +8489,7 @@ static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id)
 			return true;
 		break;
 	case BPF_PROG_TYPE_SOCKET_FILTER:
-	case BPF_PROG_TYPE_SCHED_CLS:
-	case BPF_PROG_TYPE_SCHED_ACT:
-	case BPF_PROG_TYPE_XDP:
 	case BPF_PROG_TYPE_SK_REUSEPORT:
-	case BPF_PROG_TYPE_FLOW_DISSECTOR:
 	case BPF_PROG_TYPE_SK_LOOKUP:
 		return true;
 	default:
Signature · 1 line
2.43.0
#2 lore raw

Re: [PATCH bpf-next v3 1/2] bpf, sockmap: disallow update and delete from tc, xdp and flow_dissector

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Incomplete Deadlock Fix / Bypass Vector
--

commit 11c2a6f499580e1830b9a3c86513eb4f154d7710
Author: Sechang Lim <[email protected]>

bpf, sockmap: disallow update and delete from tc, xdp and flow_dissector

This commit aims to fix a circular locking dependency between sk_callback_lock and stab->lock by disallowing sockmap updates and deletes from tc, xdp, and flow_dissector programs.

Quoted 3 lines
kernel/bpf/verifier.c +0 -0

[ ... ]

kernel/bpf/verifier.c static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id) · line 8494

Quoted 11 lines
kernel/bpf/verifier.c +0 -4
@@ -8489,11 +8489,7 @@ static bool may_update_sockmap(struct bpf_verifier_env *env, int func_id)
 			return true;
 		break;
 	case BPF_PROG_TYPE_SOCKET_FILTER:
-	case BPF_PROG_TYPE_SCHED_CLS:
-	case BPF_PROG_TYPE_SCHED_ACT:
-	case BPF_PROG_TYPE_XDP:
 	case BPF_PROG_TYPE_SK_REUSEPORT:
-	case BPF_PROG_TYPE_FLOW_DISSECTOR:
 	case BPF_PROG_TYPE_SK_LOOKUP:
 		return true;

[Severity: High]
This is a pre-existing issue, but does leaving BPF_PROG_TYPE_SOCKET_FILTER and BPF_PROG_TYPE_SOCK_OPS (located just above this block) allowed to update/delete sockmaps leave the deadlock unresolved?

The commit message describes a deadlock where sk_callback_lock is held for read, a TCP ACK goes out egress, and a BPF program modifies the sockmap, acquiring stab->lock.

Can BPF_PROG_TYPE_SOCK_OPS execute synchronously during this same egress path and trigger the same issue?

sk_psock_strp_data_ready() // holds sk_callback_lock
__tcp_cleanup_rbuf()

      tcp_send_ack()
         __tcp_transmit_skb()
            bpf_skops_write_hdr_opt()  // runs BPF_PROG_TYPE_SOCK_OPS

Similarly, can BPF_PROG_TYPE_SOCKET_FILTER be executed synchronously during egress when dev_queue_xmit_nit() calls run_filter()?

If either of these programs calls a sockmap map update or delete helper, wouldn't they acquire stab->lock and recreate the exact same circular locking dependency?

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#3 lore raw

Re: [PATCH bpf-next v3 1/2] bpf, sockmap: disallow update and delete from tc, xdp and flow_dissector

In reply to Sechang Lim
This fixes the behaviour added by commit 0126240f448d ("bpf: sockmap: Allow update from BPF"), which introduced may_update_sockmap() and added the four BPF_PROG_TYPE_* cases that this change…
Quoted 10 lines

sock_map_update_common() and __sock_map_delete() hold stab->lock and call sock_map_unref() -> sock_map_del_link(), which takes sk_callback_lock for write. That gives the order stab->lock -> sk_callback_lock.

The reverse order comes from the SK_SKB stream parser. sk_psock_strp_data_ready() holds sk_callback_lock for read, and after the verdict tcp_bpf_strp_read_sock() acks the consumed data inline via __tcp_cleanup_rbuf(). The ACK goes out egress, where a sched_cls program deletes from the sockmap and takes stab->lock:

Quoted 7 lines

A tc, xdp or flow_dissector program has no reason to update or delete a sockmap, and redirect does not go through here. Drop them from may_update_sockmap() so the verifier rejects it. It also closes the matching sockhash inversion.

This fixes the behaviour added by commit 0126240f448d ("bpf: sockmap: Allow update from BPF"), which introduced may_update_sockmap() and added the four BPF_PROG_TYPE_* cases that this change removes. Should it carry:

Fixes 0126240f448d ("bpf: sockmap: Allow update from BPF")

---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md

CI run summary: https://github.com/kernel-patches/bpf/actions/runs/28391303635

#4 lore raw

Re: [PATCH bpf-next v3 1/2] bpf, sockmap: disallow update and delete from tc, xdp and flow_dissector

In reply to Sechang Lim
Quoted 50 lines · Sechang Lim

sock_map_update_common() and __sock_map_delete() hold stab->lock and call sock_map_unref() -> sock_map_del_link(), which takes sk_callback_lock for write. That gives the order stab->lock -> sk_callback_lock.

The reverse order comes from the SK_SKB stream parser. sk_psock_strp_data_ready() holds sk_callback_lock for read, and after the verdict tcp_bpf_strp_read_sock() acks the consumed data inline via __tcp_cleanup_rbuf(). The ACK goes out egress, where a sched_cls program deletes from the sockmap and takes stab->lock:

WARNING: possible circular locking dependency detected
------------------------------------------------------
syz.9.8824 is trying to acquire lock:
(&stab->lock){+.-.}-{3:3}, at: __sock_map_delete net/core/sock_map.c:421 but task is already holding lock:
(clock-AF_INET){++.-}-{3:3}, at: sk_psock_strp_data_ready net/core/skmsg.c:1173

-> #1 (clock-AF_INET){++.-}-{3:3}:

         _raw_write_lock_bh
         sock_map_del_link net/core/sock_map.c:167
         sock_map_unref net/core/sock_map.c:184
         sock_map_update_common net/core/sock_map.c:509
         sock_map_update_elem_sys net/core/sock_map.c:588
         map_update_elem kernel/bpf/syscall.c:1805

-> #0 (&stab->lock){+.-.}-{3:3}:

         _raw_spin_lock_bh
         __sock_map_delete net/core/sock_map.c:421
         sock_map_delete_elem net/core/sock_map.c:452
         bpf_prog_06044d24140080b6
         tcx_run net/core/dev.c:4451
         sch_handle_egress net/core/dev.c:4541
         __dev_queue_xmit net/core/dev.c:4808
         ...
         tcp_bpf_strp_read_sock net/ipv4/tcp_bpf.c:701
         strp_data_ready net/strparser/strparser.c:402
         sk_psock_strp_data_ready net/core/skmsg.c:1174
         tcp_data_queue net/ipv4/tcp_input.c:5661
  Possible unsafe locking scenario:
         CPU0                    CPU1
         ----                    ----
    rlock(clock-AF_INET);
                                 lock(&stab->lock);
                                 lock(clock-AF_INET);
    lock(&stab->lock);
   *** DEADLOCK ***
Quoted 7 lines

A tc, xdp or flow_dissector program has no reason to update or delete a sockmap, and redirect does not go through here. Drop them from may_update_sockmap() so the verifier rejects it. It also closes the matching sockhash inversion.

John,

please ack.

#5 lore raw

Re: [PATCH bpf-next v3 1/2] bpf, sockmap: disallow update and delete from tc, xdp and flow_dissector

Quoted 56 lines · Alexei Starovoitov
Quoted 55 lines · Sechang Lim

sock_map_update_common() and __sock_map_delete() hold stab->lock and call sock_map_unref() -> sock_map_del_link(), which takes sk_callback_lock for write. That gives the order stab->lock -> sk_callback_lock.

The reverse order comes from the SK_SKB stream parser. sk_psock_strp_data_ready() holds sk_callback_lock for read, and after the verdict tcp_bpf_strp_read_sock() acks the consumed data inline via __tcp_cleanup_rbuf(). The ACK goes out egress, where a sched_cls program deletes from the sockmap and takes stab->lock:

WARNING: possible circular locking dependency detected
------------------------------------------------------
syz.9.8824 is trying to acquire lock:
(&stab->lock){+.-.}-{3:3}, at: __sock_map_delete net/core/sock_map.c:421 but task is already holding lock:
(clock-AF_INET){++.-}-{3:3}, at: sk_psock_strp_data_ready net/core/skmsg.c:1173

-> #1 (clock-AF_INET){++.-}-{3:3}:

         _raw_write_lock_bh
         sock_map_del_link net/core/sock_map.c:167
         sock_map_unref net/core/sock_map.c:184
         sock_map_update_common net/core/sock_map.c:509
         sock_map_update_elem_sys net/core/sock_map.c:588
         map_update_elem kernel/bpf/syscall.c:1805

-> #0 (&stab->lock){+.-.}-{3:3}:

         _raw_spin_lock_bh
         __sock_map_delete net/core/sock_map.c:421
         sock_map_delete_elem net/core/sock_map.c:452
         bpf_prog_06044d24140080b6
         tcx_run net/core/dev.c:4451
         sch_handle_egress net/core/dev.c:4541
         __dev_queue_xmit net/core/dev.c:4808
         ...
         tcp_bpf_strp_read_sock net/ipv4/tcp_bpf.c:701
         strp_data_ready net/strparser/strparser.c:402
         sk_psock_strp_data_ready net/core/skmsg.c:1174
         tcp_data_queue net/ipv4/tcp_input.c:5661
  Possible unsafe locking scenario:
         CPU0                    CPU1
         ----                    ----
    rlock(clock-AF_INET);
                                 lock(&stab->lock);
                                 lock(clock-AF_INET);
    lock(&stab->lock);
   *** DEADLOCK ***

A tc, xdp or flow_dissector program has no reason to update or delete a sockmap, and redirect does not go through here. Drop them from may_update_sockmap() so the verifier rejects it. It also closes the matching sockhash inversion.

Quoted 6 lines
Quoted 2 lines

John,

please ack.

Hi Sechang,

I think we additionally need to also block BPF_PROG_TYPE_SOCKET_FILTER? Did you check this case I guess the same case is possible there?

Then another patch needs to restrict BPF_SOCK_OPS users. For that we need to block BPF_SOCK_OPS_HDR_OPT_LEN_CB and BPF_SOCK_OPS_WRITE_*. Let me know if you want to do those as well. Let me know if you want to do both patches or just the prog blocking above with the possible addition of SOCKET_FILTER. I didn't search very hard so probably need to check all the BPF_SOCK_OPS_* to find the valid cases.

Thanks,
John

#6 lore raw

Re: [PATCH bpf-next v3 1/2] bpf, sockmap: disallow update and delete from tc, xdp and flow_dissector

Quoted 2 lines · John Fastabend

I think we additionally need to also block BPF_PROG_TYPE_SOCKET_FILTER? Did you check this case I guess the same case is possible there?

Yes, same inversion. I'll block BPF_PROG_TYPE_SOCKET_FILTER in v4.

Quoted 7 lines

Then another patch needs to restrict BPF_SOCK_OPS users. For that we need to block BPF_SOCK_OPS_HDR_OPT_LEN_CB and BPF_SOCK_OPS_WRITE_*. Let me know if you want to do those as well. Let me know if you want to do both patches or just the prog blocking above with the possible addition of SOCKET_FILTER. I didn't search very hard so probably need to check all the BPF_SOCK_OPS_* to find the valid cases.

Just the prog blocking with BPF_PROG_TYPE_SOCKET_FILTER added. I'll leave SOCK_OPS out of this series.

Bests,
Sechang

#7 lore raw

PATCH v3 2/2 selftests/bpf: drop tc/xdp/flow_dissector sockmap mutation tests

In reply to Sechang Lim

tc, xdp and flow_dissector programs can no longer update or delete a sockmap. Adjust the tests:

- verifier_sockmap_mutate: the tc, xdp and flow_dissector cases now expect __failure with "cannot update sockmap in this context".
- sockmap_basic: drop "sockmap update" / "sockhash update", which load a SEC("tc") program that copies a sock between maps.
- fexit_bpf2bpf: drop "func_sockmap_update", whose freplace program updates a sockmap in the tc cls_redirect context.

Remove the now-unused test_sockmap_update.c and freplace_cls_redirect.c.

---
 .../selftests/bpf/prog_tests/fexit_bpf2bpf.c  | 13 -----
 .../selftests/bpf/prog_tests/sockmap_basic.c  | 52 -------------------
 .../bpf/progs/freplace_cls_redirect.c         | 34 ------------
 .../selftests/bpf/progs/test_sockmap_update.c | 48 -----------------
 .../bpf/progs/verifier_sockmap_mutate.c       | 10 ++--
 5 files changed, 5 insertions(+), 152 deletions(-)
 delete mode 100644 tools/testing/selftests/bpf/progs/freplace_cls_redirect.c
 delete mode 100644 tools/testing/selftests/bpf/progs/test_sockmap_update.c
tools/testing/selftests/bpf/prog_tests/fexit_bpf2bpf.c +0 -13
index 92c20803ea76..d3a954158c33 100644
@@ -336,17 +336,6 @@ static void test_fmod_ret_freplace(void)
 }
 
 
-static void test_func_sockmap_update(void)
-{
-	const char *prog_name[] = {
-		"freplace/cls_redirect",
-	};
-	test_fexit_bpf2bpf_common("./freplace_cls_redirect.bpf.o",
-				  "./test_cls_redirect.bpf.o",
-				  ARRAY_SIZE(prog_name),
-				  prog_name, false, NULL);
-}
-
 static void test_func_replace_void(void)
 {
 	const char *prog_name[] = {
@@ -599,8 +588,6 @@ void serial_test_fexit_bpf2bpf(void)
 		test_func_replace();
 	if (test__start_subtest("func_replace_verify"))
 		test_func_replace_verify();
-	if (test__start_subtest("func_sockmap_update"))
-		test_func_sockmap_update();
 	if (test__start_subtest("func_replace_return_code"))
 		test_func_replace_return_code();
 	if (test__start_subtest("func_map_prog_compatibility"))
tools/testing/selftests/bpf/prog_tests/sockmap_basic.c +0 -52

70 lines not on this page · raw message

tools/testing/selftests/bpf/progs/freplace_cls_redirect.c deleted +0 -34
deleted file mode 100644
index 7e94412d47a5..000000000000
@@ -1,34 +0,0 @@
-// SPDX-License-Identifier: GPL-2.0
-// Copyright (c) 2020 Facebook
-
-#include <linux/stddef.h>
-#include <linux/bpf.h>
-#include <linux/pkt_cls.h>
-#include <bpf/bpf_endian.h>
-#include <bpf/bpf_helpers.h>
-
-struct {
-	__uint(type, BPF_MAP_TYPE_SOCKMAP);
-	__type(key, int);
-	__type(value, int);
-	__uint(max_entries, 2);
-} sock_map SEC(".maps");
-
-SEC("freplace/cls_redirect")
-int freplace_cls_redirect_test(struct __sk_buff *skb)
-{
-	int ret = 0;
-	const int zero = 0;
-	struct bpf_sock *sk;
-
-	sk = bpf_map_lookup_elem(&sock_map, &zero);
-	if (!sk)
-		return TC_ACT_SHOT;
-
-	ret = bpf_map_update_elem(&sock_map, &zero, sk, 0);
-	bpf_sk_release(sk);
-
-	return ret == 0 ? TC_ACT_OK : TC_ACT_SHOT;
-}
-
-char _license[] SEC("license") = "GPL";
tools/testing/selftests/bpf/progs/test_sockmap_update.c deleted +0 -48

48 lines not on this page · raw message

tools/testing/selftests/bpf/progs/verifier_sockmap_mutate.c +5 -5
index fe4b123187b8..b11026123690 100644
@@ -74,7 +74,7 @@ static __always_inline void test_sockmap_lookup_and_mutate(void)
 }
 
 SEC("action")
-__success
+__failure __msg("cannot update sockmap in this context")
 int test_sched_act(struct __sk_buff *skb)
 {
 	test_sockmap_mutate(skb->sk);
@@ -82,7 +82,7 @@ int test_sched_act(struct __sk_buff *skb)
 }
 
 SEC("classifier")
-__success
+__failure __msg("cannot update sockmap in this context")
 int test_sched_cls(struct __sk_buff *skb)
 {
 	test_sockmap_mutate(skb->sk);
@@ -90,7 +90,7 @@ int test_sched_cls(struct __sk_buff *skb)
 }
 
 SEC("flow_dissector")
-__success
+__failure __msg("cannot update sockmap in this context")
 int test_flow_dissector_delete(struct __sk_buff *skb __always_unused)
 {
 	test_sockmap_delete();
@@ -98,7 +98,7 @@ int test_flow_dissector_delete(struct __sk_buff *skb __always_unused)
 }
 
 SEC("flow_dissector")
-__failure __msg("program of this type cannot use helper bpf_sk_release")
+__failure __msg("cannot update sockmap in this context")
 int test_flow_dissector_update(struct __sk_buff *skb __always_unused)
 {
 	test_sockmap_lookup_and_update(); /* no access to skb->sk */
@@ -179,7 +179,7 @@ int test_sockops_update_dedicated(struct bpf_sock_ops *ctx)
 }
 
 SEC("xdp")
-__success
+__failure __msg("cannot update sockmap in this context")
 int test_xdp(struct xdp_md *ctx __always_unused)
 {
 	test_sockmap_lookup_and_mutate();
Signature · 1 line
2.43.0
#8 lore raw

Re: [PATCH bpf-next v3 2/2] selftests/bpf: drop tc/xdp/flow_dissector sockmap mutation tests

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] The commit deletes the selftest file test_sockmap_update.c but leaves broken references to it in…

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] The commit deletes the selftest file test_sockmap_update.c but leaves broken references to it in the sockmap documentation. --

commit ccad6ec599cc55c0fbdeda989132b9073df009d1
Author: Sechang Lim <[email protected]>

selftests/bpf: drop tc/xdp/flow_dissector sockmap mutation tests

tc, xdp and flow_dissector programs can no longer update or delete a sockmap. Adjust the tests and remove the now-unused test_sockmap_update.c and freplace_cls_redirect.c files.

tools/testing/selftests/bpf/progs/test_sockmap_update.c

Quoted 5 lines
tools/testing/selftests/bpf/progs/test_sockmap_update.c deleted +0 -0
deleted file mode 100644
index 6d64ea536e3d8..0000000000000

[Severity: Low]
Does this commit also need to update Documentation/bpf/map_sockmap.rst?

Since test_sockmap_update.c is being removed here, the literal references and hyperlinks to it in Documentation/bpf/map_sockmap.rst (around lines 352 and 498) are left as broken links.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=2