Skip to contentSkip to conversation

netfilter: bpf: support prog update

RFC Superseded D. Wythe 2 participants 9 messages Dec 20, 2023 – Dec 28, 2023

Subsystems (6)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
author
#8 lore raw

RFC v3 2/2 selftests/bpf: Add netfilter link prog update test

In reply to D. Wythe

From: "D. Wythe" <[email protected]>

Update prog for active links and verify whether
the prog has been successfully replaced.

Expected output:

./test_progs -t netfilter_link_update_prog
Summary: 1/0 PASSED, 0 SKIPPED, 0 FAILED

---
 .../bpf/prog_tests/netfilter_link_update_prog.c    | 83 ++++++++++++++++++++++
 .../bpf/progs/test_netfilter_link_update_prog.c    | 24 +++++++
 2 files changed, 107 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/netfilter_link_update_prog.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_netfilter_link_update_prog.c
tools/testing/selftests/bpf/prog_tests/netfilter_link_update_prog.c new file +83 -0
new file mode 100644
index 00000000..d23b544
@@ -0,0 +1,83 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include <test_progs.h>
+#include <linux/netfilter.h>
+#include <network_helpers.h>
+#include "test_netfilter_link_update_prog.skel.h"
+
+#define SERVER_ADDR "127.0.0.1"
+#define SERVER_PORT 12345
+
+static const char dummy_message[] = "A dummy message";
+
+static int send_dummy(int client_fd)
+{
+	struct sockaddr_storage saddr;
+	struct sockaddr *saddr_p;
+	socklen_t saddr_len;
+	int err;
+
+	saddr_p = (struct sockaddr *)&saddr;
+	err = make_sockaddr(AF_INET, SERVER_ADDR, SERVER_PORT, &saddr, &saddr_len);
+	if (!ASSERT_OK(err, "make_sockaddr"))
+		return -1;
+
+	err = sendto(client_fd, dummy_message, sizeof(dummy_message) - 1, 0, saddr_p, saddr_len);
+	if (!ASSERT_GE(err, 0, "sendto"))
+		return -1;
+
+	return 0;
+}
+
+void test_netfilter_link_update_prog(void)
+{
+	LIBBPF_OPTS(bpf_netfilter_opts, opts,
+		.pf = NFPROTO_IPV4,
+		.hooknum = NF_INET_LOCAL_OUT,
+		.priority = 100);
+	struct test_netfilter_link_update_prog *skel;
+	struct bpf_program *prog;
+	int server_fd, client_fd;
+	int err;
+
+	skel = test_netfilter_link_update_prog__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "test_netfilter_link_update_prog__open_and_load"))
+		goto out;
+
+	prog = skel->progs.nf_link_prog;
+
+	if (!ASSERT_OK_PTR(prog, "load program"))
+		goto out;
+
+	skel->links.nf_link_prog = bpf_program__attach_netfilter(prog, &opts);
+	if (!ASSERT_OK_PTR(skel->links.nf_link_prog, "attach netfilter program"))
+		goto out;
+
+	server_fd = start_server(AF_INET, SOCK_DGRAM, SERVER_ADDR, SERVER_PORT, 0);
+	if (!ASSERT_GE(server_fd, 0, "start_server"))
+		goto out;
+
+	client_fd = connect_to_fd(server_fd, 0);
+	if (!ASSERT_GE(client_fd, 0, "connect_to_fd"))
+		goto out;
+
+	send_dummy(client_fd);
+
+	ASSERT_EQ(skel->bss->counter, 0, "counter should be zero");
+
+	err = bpf_link__update_program(skel->links.nf_link_prog, skel->progs.nf_link_prog_new);
+	if (!ASSERT_OK(err, "bpf_link__update_program"))
+		goto out;
+
+	send_dummy(client_fd);
+	ASSERT_GE(skel->bss->counter, 0, "counter should be greater than zero");
+out:
+	if (client_fd > 0)
+		close(client_fd);
+	if (server_fd > 0)
+		close(server_fd);
+
+	test_netfilter_link_update_prog__destroy(skel);
+}
+
+
tools/testing/selftests/bpf/progs/test_netfilter_link_update_prog.c new file +24 -0
new file mode 100644
index 00000000..42ae332
@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "vmlinux.h"
+#include <bpf/bpf_helpers.h>
+
+#define NF_ACCEPT 1
+
+SEC("netfilter")
+int nf_link_prog(struct bpf_nf_ctx *ctx)
+{
+	return NF_ACCEPT;
+}
+
+u64 counter = 0;
+
+SEC("netfilter")
+int nf_link_prog_new(struct bpf_nf_ctx *ctx)
+{
+	counter++;
+	return NF_ACCEPT;
+}
+
+char _license[] SEC("license") = "GPL";
+
Signature · 1 line
1.8.3.1

D. Wythe

author
#0 lore raw

v3 cover netfilter: bpf: support prog update

From: "D. Wythe" <[email protected]>

This patches attempt to implements updating of progs within bpf netfilter link, allowing user update their ebpf netfilter prog in hot update manner.

Besides, a corresponding test case has been added to verify whether the update works.

--
v1:
1. remove unnecessary context, access the prog directly via rcu.
2. remove synchronize_rcu(), dealloc the nf_link via kfree_rcu.
3. check the dead flag during the update.
--
v1->v2:
1. remove unnecessary nf_prog, accessing nf_link->link.prog in direct. --
v2->v3:
1. access nf_link->link.prog via rcu_dereference_raw to avoid warning.

D. Wythe (2):
netfilter: bpf: support prog update selftests/bpf: Add netfilter link prog update test

 net/netfilter/nf_bpf_link.c                        | 63 ++++++++++++----
 .../bpf/prog_tests/netfilter_link_update_prog.c    | 83 ++++++++++++++++++++++
 .../bpf/progs/test_netfilter_link_update_prog.c    | 24 +++++++
 3 files changed, 155 insertions(+), 15 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/netfilter_link_update_prog.c
 create mode 100644 tools/testing/selftests/bpf/progs/test_netfilter_link_update_prog.c
Signature · 1 line
1.8.3.1
author
#1 lore raw

RFC v3 1/2 netfilter: bpf: support prog update

From: "D. Wythe" <[email protected]>

To support the prog update, we need to ensure that the prog seen within the hook is always valid. Considering that hooks are always protected by rcu_read_lock(), which provide us the ability to access the prog under rcu.

---
 net/netfilter/nf_bpf_link.c | 63 ++++++++++++++++++++++++++++++++++-----------
 1 file changed, 48 insertions(+), 15 deletions(-)
net/netfilter/nf_bpf_link.c +48 -15
index e502ec0..9bc91d1 100644
@@ -8,17 +8,8 @@
 #include <net/netfilter/nf_bpf_link.h>
 #include <uapi/linux/netfilter_ipv4.h>
 
-static unsigned int nf_hook_run_bpf(void *bpf_prog, struct sk_buff *skb,
-				    const struct nf_hook_state *s)
-{
-	const struct bpf_prog *prog = bpf_prog;
-	struct bpf_nf_ctx ctx = {
-		.state = s,
-		.skb = skb,
-	};
-
-	return bpf_prog_run(prog, &ctx);
-}
+/* protect link update in parallel */
+static DEFINE_MUTEX(bpf_nf_mutex);
 
 struct bpf_nf_link {
 	struct bpf_link link;
@@ -26,8 +17,20 @@ struct bpf_nf_link {
 	struct net *net;
 	u32 dead;
 	const struct nf_defrag_hook *defrag_hook;
+	struct rcu_head head;
 };
 
+static unsigned int nf_hook_run_bpf(void *bpf_link, struct sk_buff *skb,
+				    const struct nf_hook_state *s)
+{
+	const struct bpf_nf_link *nf_link = bpf_link;
+	struct bpf_nf_ctx ctx = {
+		.state = s,
+		.skb = skb,
+	};
+	return bpf_prog_run(rcu_dereference_raw(nf_link->link.prog), &ctx);
+}
+
 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) || IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 static const struct nf_defrag_hook *
 get_proto_defrag_hook(struct bpf_nf_link *link,
@@ -126,8 +129,7 @@ static void bpf_nf_link_release(struct bpf_link *link)
 static void bpf_nf_link_dealloc(struct bpf_link *link)
 {
 	struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
-
-	kfree(nf_link);
+	kfree_rcu(nf_link, head);
 }
 
 static int bpf_nf_link_detach(struct bpf_link *link)
@@ -162,7 +164,34 @@ static int bpf_nf_link_fill_link_info(const struct bpf_link *link,
 static int bpf_nf_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
 			      struct bpf_prog *old_prog)
 {
-	return -EOPNOTSUPP;
+	struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
+	int err = 0;
+
+	mutex_lock(&bpf_nf_mutex);
+
+	if (nf_link->dead) {
+		err = -EPERM;
+		goto out;
+	}
+
+	/* target old_prog mismatch */
+	if (old_prog && link->prog != old_prog) {
+		err = -EPERM;
+		goto out;
+	}
+
+	old_prog = link->prog;
+	if (old_prog == new_prog) {
+		/* don't need update */
+		bpf_prog_put(new_prog);
+		goto out;
+	}
+
+	old_prog = xchg(&link->prog, new_prog);
+	bpf_prog_put(old_prog);
+out:
+	mutex_unlock(&bpf_nf_mutex);
+	return err;
 }
 
 static const struct bpf_link_ops bpf_nf_link_lops = {
@@ -226,7 +255,11 @@ int bpf_nf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
 
 	link->hook_ops.hook = nf_hook_run_bpf;
 	link->hook_ops.hook_ops_type = NF_HOOK_OP_BPF;
-	link->hook_ops.priv = prog;
+
+	/* bpf_nf_link_release & bpf_nf_link_dealloc() can ensures that link remains
+	 * valid at all times within nf_hook_run_bpf().
+	 */
+	link->hook_ops.priv = link;
 
 	link->hook_ops.pf = attr->link_create.netfilter.pf;
 	link->hook_ops.priority = attr->link_create.netfilter.priority;
Signature · 1 line
1.8.3.1
#2 lore raw
Quoted 38 lines · D. Wythe

From: "D. Wythe" <[email protected]>

To support the prog update, we need to ensure that the prog seen within the hook is always valid. Considering that hooks are always protected by rcu_read_lock(), which provide us the ability to access the prog under rcu.

---
 net/netfilter/nf_bpf_link.c | 63 ++++++++++++++++++++++++++++++++++-----------
 1 file changed, 48 insertions(+), 15 deletions(-)
net/netfilter/nf_bpf_link.c +2 -11
index e502ec0..9bc91d1 100644
@@ -8,17 +8,8 @@
 #include <net/netfilter/nf_bpf_link.h>
 #include <uapi/linux/netfilter_ipv4.h>
 
-static unsigned int nf_hook_run_bpf(void *bpf_prog, struct sk_buff *skb,
-                                   const struct nf_hook_state *s)
-{
-       const struct bpf_prog *prog = bpf_prog;
-       struct bpf_nf_ctx ctx = {
-               .state = s,
-               .skb = skb,
-       };
-
-       return bpf_prog_run(prog, &ctx);
-}
+/* protect link update in parallel */
+static DEFINE_MUTEX(bpf_nf_mutex);
 
 struct bpf_nf_link {
        struct bpf_link link;

net/netfilter/nf_bpf_link.c struct bpf_nf_link { · line 20

Quoted 4 lines
net/netfilter/nf_bpf_link.c +1 -0
@@ -26,8 +17,20 @@ struct bpf_nf_link {
        struct net *net;
        u32 dead;
        const struct nf_defrag_hook *defrag_hook;
+       struct rcu_head head;

I have to point out the same issues as before, but will ask them differently...

Why do you think above rcu_head is necessary?

Quoted 17 lines
net/netfilter/nf_bpf_link.c +11 -0
 };
 
+static unsigned int nf_hook_run_bpf(void *bpf_link, struct sk_buff *skb,
+                                   const struct nf_hook_state *s)
+{
+       const struct bpf_nf_link *nf_link = bpf_link;
+       struct bpf_nf_ctx ctx = {
+               .state = s,
+               .skb = skb,
+       };
+       return bpf_prog_run(rcu_dereference_raw(nf_link->link.prog), &ctx);
+}
+
 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) || IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
 static const struct nf_defrag_hook *
 get_proto_defrag_hook(struct bpf_nf_link *link,

net/netfilter/nf_bpf_link.c static void bpf_nf_link_release(struct bpf_link *link) · line 132

Quoted 6 lines
net/netfilter/nf_bpf_link.c +1 -2
@@ -126,8 +129,7 @@ static void bpf_nf_link_release(struct bpf_link *link)
 static void bpf_nf_link_dealloc(struct bpf_link *link)
 {
        struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
-
-       kfree(nf_link);
+       kfree_rcu(nf_link, head);

Why is this needed ?
Have you looked at tcx_link_lops ?

net/netfilter/nf_bpf_link.c static int bpf_nf_link_fill_link_info(const struct bpf_link *link, · line 170

Quoted 12 lines
net/netfilter/nf_bpf_link.c +4 -1
 }
 
 static int bpf_nf_link_detach(struct bpf_link *link)
@@ -162,7 +164,34 @@ static int bpf_nf_link_fill_link_info(const struct bpf_link *link,
 static int bpf_nf_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
                              struct bpf_prog *old_prog)
 {
-       return -EOPNOTSUPP;
+       struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
+       int err = 0;
+
+       mutex_lock(&bpf_nf_mutex);

Why do you need this mutex?
What race does it solve?

Quoted 43 lines
net/netfilter/nf_bpf_link.c +29 -2
+
+       if (nf_link->dead) {
+               err = -EPERM;
+               goto out;
+       }
+
+       /* target old_prog mismatch */
+       if (old_prog && link->prog != old_prog) {
+               err = -EPERM;
+               goto out;
+       }
+
+       old_prog = link->prog;
+       if (old_prog == new_prog) {
+               /* don't need update */
+               bpf_prog_put(new_prog);
+               goto out;
+       }
+
+       old_prog = xchg(&link->prog, new_prog);
+       bpf_prog_put(old_prog);
+out:
+       mutex_unlock(&bpf_nf_mutex);
+       return err;
 }
 
 static const struct bpf_link_ops bpf_nf_link_lops = {
@@ -226,7 +255,11 @@ int bpf_nf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
 
        link->hook_ops.hook = nf_hook_run_bpf;
        link->hook_ops.hook_ops_type = NF_HOOK_OP_BPF;
-       link->hook_ops.priv = prog;
+
+       /* bpf_nf_link_release & bpf_nf_link_dealloc() can ensures that link remains
+        * valid at all times within nf_hook_run_bpf().
+        */
+       link->hook_ops.priv = link;
 
        link->hook_ops.pf = attr->link_create.netfilter.pf;
        link->hook_ops.priority = attr->link_create.netfilter.priority;
--

1.8.3.1

author
#3 lore raw
Quoted 64 lines · Alexei Starovoitov
Quoted 37 lines · D. Wythe

From: "D. Wythe" <[email protected]>

To support the prog update, we need to ensure that the prog seen within the hook is always valid. Considering that hooks are always protected by rcu_read_lock(), which provide us the ability to access the prog under rcu.

---
  net/netfilter/nf_bpf_link.c | 63 ++++++++++++++++++++++++++++++++++-----------
  1 file changed, 48 insertions(+), 15 deletions(-)
net/netfilter/nf_bpf_link.c +2 -11
index e502ec0..9bc91d1 100644
@@ -8,17 +8,8 @@
  #include <net/netfilter/nf_bpf_link.h>
  #include <uapi/linux/netfilter_ipv4.h>
 
-static unsigned int nf_hook_run_bpf(void *bpf_prog, struct sk_buff *skb,
-                                   const struct nf_hook_state *s)
-{
-       const struct bpf_prog *prog = bpf_prog;
-       struct bpf_nf_ctx ctx = {
-               .state = s,
-               .skb = skb,
-       };
-
-       return bpf_prog_run(prog, &ctx);
-}
+/* protect link update in parallel */
+static DEFINE_MUTEX(bpf_nf_mutex);
 
  struct bpf_nf_link {
         struct bpf_link link;

net/netfilter/nf_bpf_link.c struct bpf_nf_link { · line 20

Quoted 4 lines
net/netfilter/nf_bpf_link.c +1 -0
@@ -26,8 +17,20 @@ struct bpf_nf_link {
         struct net *net;
         u32 dead;
         const struct nf_defrag_hook *defrag_hook;
+       struct rcu_head head;

I have to point out the same issues as before, but will ask them differently...

Why do you think above rcu_head is necessary?

Quoted 17 lines
net/netfilter/nf_bpf_link.c +11 -0
  };
 
+static unsigned int nf_hook_run_bpf(void *bpf_link, struct sk_buff *skb,
+                                   const struct nf_hook_state *s)
+{
+       const struct bpf_nf_link *nf_link = bpf_link;
+       struct bpf_nf_ctx ctx = {
+               .state = s,
+               .skb = skb,
+       };
+       return bpf_prog_run(rcu_dereference_raw(nf_link->link.prog), &ctx);
+}
+
  #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) || IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
  static const struct nf_defrag_hook *
  get_proto_defrag_hook(struct bpf_nf_link *link,
@@ -126,8 +129,7 @@ static void bpf_nf_link_release(struct bpf_link *link)
Quoted 8 lines

net/netfilter/nf_bpf_link.c static void bpf_nf_link_release(struct bpf_link *link) · line 132

Quoted 6 lines
net/netfilter/nf_bpf_link.c +1 -2
  static void bpf_nf_link_dealloc(struct bpf_link *link)
  {
         struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
-
-       kfree(nf_link);
+       kfree_rcu(nf_link, head);

Why is this needed ?
Have you looked at tcx_link_lops ?

Introducing rcu_head/kfree_rcu is to address the situation where the netfilter hooks might still access the link after bpf_nf_link_dealloc.

                                                     nf_hook_run_bpf
                                                     const struct bpf_nf_link *nf_link = bpf_link;

bpf_nf_link_release nf_unregister_net_hook(nf_link->net, &nf_link->hook_ops);

bpf_nf_link_dealloc
    free(link)
bpf_prog_run(link->prog);

I had checked the tcx_link_lops ,it's seems it use the synchronize_rcu() to solve the same problem, which is also the way we used in the first version.

https://lore.kernel.org/bpf/[email protected]/

However, we have received some opposing views, believing that this is a bit overkill, so we decided to use kfree_rcu.

https://lore.kernel.org/bpf/[email protected]/

Quoted 14 lines

net/netfilter/nf_bpf_link.c static int bpf_nf_link_fill_link_info(const struct bpf_link *link, · line 170

Quoted 12 lines
net/netfilter/nf_bpf_link.c +4 -1
  }
 
  static int bpf_nf_link_detach(struct bpf_link *link)
@@ -162,7 +164,34 @@ static int bpf_nf_link_fill_link_info(const struct bpf_link *link,
  static int bpf_nf_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
                               struct bpf_prog *old_prog)
  {
-       return -EOPNOTSUPP;
+       struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
+       int err = 0;
+
+       mutex_lock(&bpf_nf_mutex);

Why do you need this mutex?
What race does it solve?

To avoid user update a link with differ prog at the same time. I noticed that sys_bpf() doesn't seem to prevent being invoked by user at the same time. Have I missed something?

Best wishes,
D. Wythe

Quoted 43 lines
Quoted 43 lines
net/netfilter/nf_bpf_link.c +29 -2
+
+       if (nf_link->dead) {
+               err = -EPERM;
+               goto out;
+       }
+
+       /* target old_prog mismatch */
+       if (old_prog && link->prog != old_prog) {
+               err = -EPERM;
+               goto out;
+       }
+
+       old_prog = link->prog;
+       if (old_prog == new_prog) {
+               /* don't need update */
+               bpf_prog_put(new_prog);
+               goto out;
+       }
+
+       old_prog = xchg(&link->prog, new_prog);
+       bpf_prog_put(old_prog);
+out:
+       mutex_unlock(&bpf_nf_mutex);
+       return err;
  }
 
  static const struct bpf_link_ops bpf_nf_link_lops = {
@@ -226,7 +255,11 @@ int bpf_nf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
 
         link->hook_ops.hook = nf_hook_run_bpf;
         link->hook_ops.hook_ops_type = NF_HOOK_OP_BPF;
-       link->hook_ops.priv = prog;
+
+       /* bpf_nf_link_release & bpf_nf_link_dealloc() can ensures that link remains
+        * valid at all times within nf_hook_run_bpf().
+        */
+       link->hook_ops.priv = link;
 
         link->hook_ops.pf = attr->link_create.netfilter.pf;
         link->hook_ops.priority = attr->link_create.netfilter.priority;
--

1.8.3.1

#4 lore raw
Quoted 77 lines · D. Wythe
Quoted 64 lines · Alexei Starovoitov
Quoted 37 lines · D. Wythe

From: "D. Wythe" <[email protected]>

To support the prog update, we need to ensure that the prog seen within the hook is always valid. Considering that hooks are always protected by rcu_read_lock(), which provide us the ability to access the prog under rcu.

---
  net/netfilter/nf_bpf_link.c | 63 ++++++++++++++++++++++++++++++++++-----------
  1 file changed, 48 insertions(+), 15 deletions(-)
net/netfilter/nf_bpf_link.c +2 -11
index e502ec0..9bc91d1 100644
@@ -8,17 +8,8 @@
  #include <net/netfilter/nf_bpf_link.h>
  #include <uapi/linux/netfilter_ipv4.h>
 
-static unsigned int nf_hook_run_bpf(void *bpf_prog, struct sk_buff *skb,
-                                   const struct nf_hook_state *s)
-{
-       const struct bpf_prog *prog = bpf_prog;
-       struct bpf_nf_ctx ctx = {
-               .state = s,
-               .skb = skb,
-       };
-
-       return bpf_prog_run(prog, &ctx);
-}
+/* protect link update in parallel */
+static DEFINE_MUTEX(bpf_nf_mutex);
 
  struct bpf_nf_link {
         struct bpf_link link;

net/netfilter/nf_bpf_link.c struct bpf_nf_link { · line 20

Quoted 4 lines
net/netfilter/nf_bpf_link.c +1 -0
@@ -26,8 +17,20 @@ struct bpf_nf_link {
         struct net *net;
         u32 dead;
         const struct nf_defrag_hook *defrag_hook;
+       struct rcu_head head;

I have to point out the same issues as before, but will ask them differently...

Why do you think above rcu_head is necessary?

Quoted 17 lines
net/netfilter/nf_bpf_link.c +11 -0
  };
 
+static unsigned int nf_hook_run_bpf(void *bpf_link, struct sk_buff *skb,
+                                   const struct nf_hook_state *s)
+{
+       const struct bpf_nf_link *nf_link = bpf_link;
+       struct bpf_nf_ctx ctx = {
+               .state = s,
+               .skb = skb,
+       };
+       return bpf_prog_run(rcu_dereference_raw(nf_link->link.prog), &ctx);
+}
+
  #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) || IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
  static const struct nf_defrag_hook *
  get_proto_defrag_hook(struct bpf_nf_link *link,
@@ -126,8 +129,7 @@ static void bpf_nf_link_release(struct bpf_link *link)
Quoted 8 lines

net/netfilter/nf_bpf_link.c static void bpf_nf_link_release(struct bpf_link *link) · line 132

Quoted 6 lines
net/netfilter/nf_bpf_link.c +1 -2
  static void bpf_nf_link_dealloc(struct bpf_link *link)
  {
         struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
-
-       kfree(nf_link);
+       kfree_rcu(nf_link, head);

Why is this needed ?
Have you looked at tcx_link_lops ?

Quoted 3 lines

Introducing rcu_head/kfree_rcu is to address the situation where the netfilter hooks might
still access the link after bpf_nf_link_dealloc.

Why do you think so?

Quoted 8 lines

nf_hook_run_bpf const struct
bpf_nf_link *nf_link = bpf_link;

bpf_nf_link_release

     nf_unregister_net_hook(nf_link->net, &nf_link->hook_ops);
Quoted 7 lines

bpf_nf_link_dealloc
free(link)
bpf_prog_run(link->prog);

I had checked the tcx_link_lops ,it's seems it use the synchronize_rcu() to solve the

Where do you see such code in tcx_link_lops ?

Quoted 25 lines

same problem, which is also the way we used in the first version.

https://lore.kernel.org/bpf/[email protected]/

However, we have received some opposing views, believing that this is a bit overkill,
so we decided to use kfree_rcu.

https://lore.kernel.org/bpf/[email protected]/

Quoted 14 lines

net/netfilter/nf_bpf_link.c static int bpf_nf_link_fill_link_info(const struct bpf_link *link, · line 170

Quoted 12 lines
net/netfilter/nf_bpf_link.c +4 -1
  }
 
  static int bpf_nf_link_detach(struct bpf_link *link)
@@ -162,7 +164,34 @@ static int bpf_nf_link_fill_link_info(const struct bpf_link *link,
  static int bpf_nf_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
                               struct bpf_prog *old_prog)
  {
-       return -EOPNOTSUPP;
+       struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
+       int err = 0;
+
+       mutex_lock(&bpf_nf_mutex);

Why do you need this mutex?
What race does it solve?

Quoted 4 lines

To avoid user update a link with differ prog at the same time. I noticed that sys_bpf()
doesn't seem to prevent being invoked by user at the same time. Have I missed something?

You're correct that sys_bpf() doesn't lock anything. But what are you serializing in this bpf_nf_link_update() ? What will happen if multiple bpf_nf_link_update() without mutex run on different CPUs in parallel ?

author
#5 lore raw
Quoted 76 lines · Alexei Starovoitov
Quoted 75 lines · D. Wythe
Quoted 72 lines · Alexei Starovoitov
Quoted 37 lines · D. Wythe

From: "D. Wythe" <[email protected]>

To support the prog update, we need to ensure that the prog seen within the hook is always valid. Considering that hooks are always protected by rcu_read_lock(), which provide us the ability to access the prog under rcu.

---
   net/netfilter/nf_bpf_link.c | 63 ++++++++++++++++++++++++++++++++++-----------
   1 file changed, 48 insertions(+), 15 deletions(-)
net/netfilter/nf_bpf_link.c +2 -11
index e502ec0..9bc91d1 100644
@@ -8,17 +8,8 @@
   #include <net/netfilter/nf_bpf_link.h>
   #include <uapi/linux/netfilter_ipv4.h>
 
-static unsigned int nf_hook_run_bpf(void *bpf_prog, struct sk_buff *skb,
-                                   const struct nf_hook_state *s)
-{
-       const struct bpf_prog *prog = bpf_prog;
-       struct bpf_nf_ctx ctx = {
-               .state = s,
-               .skb = skb,
-       };
-
-       return bpf_prog_run(prog, &ctx);
-}
+/* protect link update in parallel */
+static DEFINE_MUTEX(bpf_nf_mutex);
 
   struct bpf_nf_link {
          struct bpf_link link;

net/netfilter/nf_bpf_link.c struct bpf_nf_link { · line 20

Quoted 4 lines
net/netfilter/nf_bpf_link.c +1 -0
@@ -26,8 +17,20 @@ struct bpf_nf_link {
          struct net *net;
          u32 dead;
          const struct nf_defrag_hook *defrag_hook;
+       struct rcu_head head;

I have to point out the same issues as before, but will ask them differently...

Why do you think above rcu_head is necessary?

Quoted 17 lines
net/netfilter/nf_bpf_link.c +11 -0
   };
 
+static unsigned int nf_hook_run_bpf(void *bpf_link, struct sk_buff *skb,
+                                   const struct nf_hook_state *s)
+{
+       const struct bpf_nf_link *nf_link = bpf_link;
+       struct bpf_nf_ctx ctx = {
+               .state = s,
+               .skb = skb,
+       };
+       return bpf_prog_run(rcu_dereference_raw(nf_link->link.prog), &ctx);
+}
+
   #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV4) || IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
   static const struct nf_defrag_hook *
   get_proto_defrag_hook(struct bpf_nf_link *link,

net/netfilter/nf_bpf_link.c static void bpf_nf_link_release(struct bpf_link *link) · line 132

Quoted 6 lines
net/netfilter/nf_bpf_link.c +1 -2
@@ -126,8 +129,7 @@ static void bpf_nf_link_release(struct bpf_link *link)
   static void bpf_nf_link_dealloc(struct bpf_link *link)
   {
          struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
-
-       kfree(nf_link);
+       kfree_rcu(nf_link, head);

Why is this needed ?
Have you looked at tcx_link_lops ?

Quoted 5 lines
Quoted 3 lines

Introducing rcu_head/kfree_rcu is to address the situation where the netfilter hooks might
still access the link after bpf_nf_link_dealloc.

Why do you think so?

Hi Alexei,

IMMO, nf_unregister_net_hook does not wait for the completion of the execution of the hook that is being removed, instead, it allocates a new array without the very hook to replace the old arrayvia rcu_assign_pointer() (in __nf_hook_entries_try_shrink), then it use call_rcu() to release the old one.

You can find more details in commit 8c873e2199700c2de7dbd5eedb9d90d5f109462b.

In other words, when nf_unregister_net_hook returns, there may still be contexts executing hooks on the old array, which means that the link may still be accessed after nf_unregister_net_hook returns.

And that's the reason why we use kfree_rcu() to release the link.

Quoted 7 lines
Quoted 7 lines

nf_hook_run_bpf const struct
bpf_nf_link *nf_link = bpf_link;

bpf_nf_link_release

      nf_unregister_net_hook(nf_link->net, &nf_link->hook_ops);
Quoted 8 lines
Quoted 7 lines

bpf_nf_link_dealloc
free(link)
bpf_prog_run(link->prog);

I had checked the tcx_link_lops ,it's seems it use the synchronize_rcu() to solve the

Where do you see such code in tcx_link_lops ?

I'm not certain if the reason that it choose to use synchronize_rcu()is the same as mine, but I did see it here:

tcx_link_release() -> tcx_entry_sync()

static inline void tcx_entry_sync(void)
{
    /* bpf_mprog_entry got a/b swapped, therefore ensure that
     * there are no inflight users on the old one anymore. */
    synchronize_rcu();
}

Quoted 24 lines
Quoted 24 lines

same problem, which is also the way we used in the first version.

https://lore.kernel.org/bpf/[email protected]/

However, we have received some opposing views, believing that this is a bit overkill,
so we decided to use kfree_rcu.

https://lore.kernel.org/bpf/[email protected]/

Quoted 14 lines

net/netfilter/nf_bpf_link.c static int bpf_nf_link_fill_link_info(const struct bpf_link *link, · line 170

Quoted 12 lines
net/netfilter/nf_bpf_link.c +4 -1
   }
 
   static int bpf_nf_link_detach(struct bpf_link *link)
@@ -162,7 +164,34 @@ static int bpf_nf_link_fill_link_info(const struct bpf_link *link,
   static int bpf_nf_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
                                struct bpf_prog *old_prog)
   {
-       return -EOPNOTSUPP;
+       struct bpf_nf_link *nf_link = container_of(link, struct bpf_nf_link, link);
+       int err = 0;
+
+       mutex_lock(&bpf_nf_mutex);

Why do you need this mutex?
What race does it solve?

Quoted 8 lines
Quoted 4 lines

To avoid user update a link with differ prog at the same time. I noticed that sys_bpf()
doesn't seem to prevent being invoked by user at the same time. Have I missed something?

You're correct that sys_bpf() doesn't lock anything. But what are you serializing in this bpf_nf_link_update() ? What will happen if multiple bpf_nf_link_update() without mutex run on different CPUs in parallel ?

I must admit that it is indeed feasible if we eliminate the mutex and use cmpxchg to swap the prog (we need to ensure that there is only one bpf_prog_put() on the old prog). However, when cmpxchg fails, it means that this context has not outcompeted the other one, and we have to return a failure. Maybe something like this:

if (!cmpxchg(&link->prog, old_prog, new_prog)) {
    /* already replaced by another link_update */
    return -xxx;
}

As a comparison, The version with the mutex wouldn't encounter this error, every update would succeed. I think that it's too harsh for the user to receive a failure in that case since they haven't done anything wrong.

Best wishes,
D. Wythe

#6 lore raw
Quoted 14 lines · D. Wythe

Hi Alexei,

IMMO, nf_unregister_net_hook does not wait for the completion of the execution of the hook that is being removed,
instead, it allocates a new array without the very hook to replace the old arrayvia rcu_assign_pointer() (in __nf_hook_entries_try_shrink), then it use call_rcu() to release the old one.

You can find more details in commit
8c873e2199700c2de7dbd5eedb9d90d5f109462b.

Quoted 16 lines

In other words, when nf_unregister_net_hook returns, there may still be contexts executing hooks on the
old array, which means that the link may still be accessed after nf_unregister_net_hook returns.

And that's the reason why we use kfree_rcu() to release the link.

Quoted 10 lines
Quoted 10 lines

nf_hook_run_bpf const struct
bpf_nf_link *nf_link = bpf_link;

bpf_nf_link_release

      nf_unregister_net_hook(nf_link->net, &nf_link->hook_ops);

bpf_nf_link_dealloc
free(link)
bpf_prog_run(link->prog);

Got it.
Sounds like it's an existing bug. If so it should be an independent patch with Fixes tag.

Also please craft a test case to demonstrate UAF.

Quoted 13 lines

I must admit that it is indeed feasible if we eliminate the mutex and use cmpxchg to swap the prog (we need to ensure that there is only one bpf_prog_put() on the old prog).
However, when cmpxchg fails, it means that this context has not outcompeted the other one, and we have to return a failure. Maybe something like this:

if (!cmpxchg(&link->prog, old_prog, new_prog)) {
     /* already replaced by another link_update */
     return -xxx;
}
Quoted 4 lines

As a comparison, The version with the mutex wouldn't encounter this error, every update would succeed. I think that it's too harsh for the user to receive a failure
in that case since they haven't done anything wrong.

Disagree. The mutex doesn't prevent this issue.
There is always a race.
It happens when link_update.old_prog_fd and BPF_F_REPLACE were specified.
One user space passes an FD of the old prog and
another user space doing the same. They both race and one of them gets

if (old_prog && link->prog != old_prog) {
               err = -EPERM;

it's no different with dropping the mutex and doing:

if (old_prog) {
    if (!cmpxchg(&link->prog, old_prog, new_prog))
      -EPERM
} else {

old_prog = xchg(&link->prog, new_prog);
}

author
#7 lore raw
Quoted 27 lines · Alexei Starovoitov
Quoted 26 lines · D. Wythe

Hi Alexei,

IMMO, nf_unregister_net_hook does not wait for the completion of the execution of the hook that is being removed,
instead, it allocates a new array without the very hook to replace the old arrayvia rcu_assign_pointer() (in __nf_hook_entries_try_shrink), then it use call_rcu() to release the old one.

You can find more details in commit
8c873e2199700c2de7dbd5eedb9d90d5f109462b.

In other words, when nf_unregister_net_hook returns, there may still be contexts executing hooks on the
old array, which means that the link may still be accessed after nf_unregister_net_hook returns.

And that's the reason why we use kfree_rcu() to release the link.

Quoted 7 lines
Quoted 7 lines

nf_hook_run_bpf const struct
bpf_nf_link *nf_link = bpf_link;

bpf_nf_link_release

       nf_unregister_net_hook(nf_link->net, &nf_link->hook_ops);
Quoted 9 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

bpf_nf_link_dealloc
free(link)
bpf_prog_run(link->prog);

Got it.
Sounds like it's an existing bug. If so it should be an independent patch with Fixes tag.

Also please craft a test case to demonstrate UAF.

It is not an existing bug... Accessing the link within the hook was something I introduced here to support updates😉, as previously there was no access to the link within the hook.

Quoted 26 lines
Quoted 12 lines

I must admit that it is indeed feasible if we eliminate the mutex and use cmpxchg to swap the prog (we need to ensure that there is only one bpf_prog_put() on the old prog).
However, when cmpxchg fails, it means that this context has not outcompeted the other one, and we have to return a failure. Maybe something like this:

if (!cmpxchg(&link->prog, old_prog, new_prog)) {
      /* already replaced by another link_update */
      return -xxx;
}
Quoted 4 lines

As a comparison, The version with the mutex wouldn't encounter this error, every update would succeed. I think that it's too harsh for the user to receive a failure
in that case since they haven't done anything wrong.

Disagree. The mutex doesn't prevent this issue.
There is always a race.
It happens when link_update.old_prog_fd and BPF_F_REPLACE were specified.
One user space passes an FD of the old prog and
another user space doing the same. They both race and one of them gets

if (old_prog && link->prog != old_prog) {
                err = -EPERM;
Quoted 7 lines

it's no different with dropping the mutex and doing:

if (old_prog) {
     if (!cmpxchg(&link->prog, old_prog, new_prog))
       -EPERM
} else {
    old_prog = xchg(&link->prog, new_prog);
}

Got it!  It's very helpful, Thanks very much! I will modify my patch accordingly.

Best wishes,
D. Wythe