[bpf-next v3 0/2] Fix kcm + sockmap by checking psock type

We check if the sk_user_data (the psock in skmsg) is in fact a sockmap
type to late, after we read the refcnt which is an error. This
series moves the check up before reading refcnt and also adds a test
to test_maps to test trying to add a KCM socket into a sockmap.

While reviewig this code I also found an issue with KCM and kTLS
where each uses sk_data_ready hooks and associated stream parser
breaking expectations in kcm, ktls or both. But that fix will need
to go to net.

Thanks to Eric for reporting.

v2: Fix up file +/- my scripts lost track of them
v3: return EBUSY if refcnt is zero

John Fastabend (2):
  bpf: skmsg, fix psock create on existing kcm/tls port
  bpf: test_maps add a test to catch kcm + sockmap

 include/linux/skmsg.h                     | 25 +++++++++---
 net/core/sock_map.c                       | 11 +++---
 tools/testing/selftests/bpf/Makefile      |  2 +-
 tools/testing/selftests/bpf/sockmap_kcm.c | 14 +++++++
 tools/testing/selftests/bpf/test_maps.c   | 64 ++++++++++++++++++++++++++++++-
 5 files changed, 103 insertions(+), 13 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/sockmap_kcm.c

-- 
1.9.1

Before using the psock returned by sk_psock_get() when adding it to a
sockmap we need to ensure it is actually a sockmap based psock.
Previously we were only checking this after incrementing the reference
counter which was an error. This resulted in a slab-out-of-bounds
error when the psock was not actually a sockmap type.

This moves the check up so the reference counter is only used
if it is a sockmap psock.

Eric reported the following KASAN BUG,

BUG: KASAN: slab-out-of-bounds in atomic_read include/asm-generic/atomic-instrumented.h:21 [inline]
BUG: KASAN: slab-out-of-bounds in refcount_inc_not_zero_checked+0x97/0x2f0 lib/refcount.c:120
Read of size 4 at addr ffff88019548be58 by task syz-executor4/22387

CPU: 1 PID: 22387 Comm: syz-executor4 Not tainted 4.19.0-rc7+ #264
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
Call Trace:
 __dump_stack lib/dump_stack.c:77 [inline]
 dump_stack+0x1c4/0x2b4 lib/dump_stack.c:113
 print_address_description.cold.8+0x9/0x1ff mm/kasan/report.c:256
 kasan_report_error mm/kasan/report.c:354 [inline]
 kasan_report.cold.9+0x242/0x309 mm/kasan/report.c:412
 check_memory_region_inline mm/kasan/kasan.c:260 [inline]
 check_memory_region+0x13e/0x1b0 mm/kasan/kasan.c:267
 kasan_check_read+0x11/0x20 mm/kasan/kasan.c:272
 atomic_read include/asm-generic/atomic-instrumented.h:21 [inline]
 refcount_inc_not_zero_checked+0x97/0x2f0 lib/refcount.c:120
 sk_psock_get include/linux/skmsg.h:379 [inline]
 sock_map_link.isra.6+0x41f/0xe30 net/core/sock_map.c:178
 sock_hash_update_common+0x19b/0x11e0 net/core/sock_map.c:669
 sock_hash_update_elem+0x306/0x470 net/core/sock_map.c:738
 map_update_elem+0x819/0xdf0 kernel/bpf/syscall.c:818

Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Reported-by: Eric Dumazet <redacted>
Fixes: 604326b41a6f ("bpf, sockmap: convert to generic sk_msg interface")
---
 include/linux/skmsg.h | 25 ++++++++++++++++++++-----
 net/core/sock_map.c   | 11 ++++++-----
 2 files changed, 26 insertions(+), 10 deletions(-)

diff --git a/include/linux/skmsg.h b/include/linux/skmsg.h
index 22347b0..84e1886 100644
--- a/include/linux/skmsg.h
+++ b/include/linux/skmsg.h

@@ -270,11 +270,6 @@ static inline struct sk_psock *sk_psock(const struct sock *sk)
 	return rcu_dereference_sk_user_data(sk);
 }
 
-static inline bool sk_has_psock(struct sock *sk)
-{
-	return sk_psock(sk) != NULL && sk->sk_prot->recvmsg == tcp_bpf_recvmsg;
-}
-
 static inline void sk_psock_queue_msg(struct sk_psock *psock,
 				      struct sk_msg *msg)
 {

@@ -374,6 +369,26 @@ static inline bool sk_psock_test_state(const struct sk_psock *psock,
 	return test_bit(bit, &psock->state);
 }
 
+static inline struct sk_psock *sk_psock_get_checked(struct sock *sk)
+{
+	struct sk_psock *psock;
+
+	rcu_read_lock();
+	psock = sk_psock(sk);
+	if (psock) {
+		if (sk->sk_prot->recvmsg != tcp_bpf_recvmsg) {
+			psock = ERR_PTR(-EBUSY);
+			goto out;
+		}
+
+		if (!refcount_inc_not_zero(&psock->refcnt))
+			psock = ERR_PTR(-EBUSY);
+	}
+out:
+	rcu_read_unlock();
+	return psock;
+}
+
 static inline struct sk_psock *sk_psock_get(struct sock *sk)
 {
 	struct sk_psock *psock;

diff --git a/net/core/sock_map.c b/net/core/sock_map.c
index 3c0e44c..be6092a 100644
--- a/net/core/sock_map.c
+++ b/net/core/sock_map.c

@@ -175,12 +175,13 @@ static int sock_map_link(struct bpf_map *map, struct sk_psock_progs *progs,
 		}
 	}
 
-	psock = sk_psock_get(sk);
+	psock = sk_psock_get_checked(sk);
+	if (IS_ERR(psock)) {
+		ret = PTR_ERR(psock);
+		goto out_progs;
+	}
+
 	if (psock) {
-		if (!sk_has_psock(sk)) {
-			ret = -EBUSY;
-			goto out_progs;
-		}
 		if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) ||
 		    (skb_progs  && READ_ONCE(psock->progs.skb_parser))) {
 			sk_psock_put(sk, psock);

-- 
1.9.1

Adding a socket to both sockmap and kcm is not supported due to
collision on sk_user_data usage.

If selftests is run without KCM support we will issue a warning
and continue with the tests.

Signed-off-by: John Fastabend <john.fastabend@gmail.com>
---
 tools/testing/selftests/bpf/Makefile      |  2 +-
 tools/testing/selftests/bpf/sockmap_kcm.c | 14 +++++++
 tools/testing/selftests/bpf/test_maps.c   | 64 ++++++++++++++++++++++++++++++-
 3 files changed, 77 insertions(+), 3 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/sockmap_kcm.c

diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile
index d99dd6f..f290554 100644
--- a/tools/testing/selftests/bpf/Makefile
+++ b/tools/testing/selftests/bpf/Makefile

@@ -28,7 +28,7 @@ TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test
 
 TEST_GEN_FILES = test_pkt_access.o test_xdp.o test_l4lb.o test_tcp_estats.o test_obj_id.o \
 	test_pkt_md_access.o test_xdp_redirect.o test_xdp_meta.o sockmap_parse_prog.o     \
-	sockmap_verdict_prog.o dev_cgroup.o sample_ret0.o test_tracepoint.o \
+	sockmap_verdict_prog.o sockmap_kcm.o dev_cgroup.o sample_ret0.o test_tracepoint.o \
 	test_l4lb_noinline.o test_xdp_noinline.o test_stacktrace_map.o \
 	sample_map_ret0.o test_tcpbpf_kern.o test_stacktrace_build_id.o \
 	sockmap_tcp_msg_prog.o connect4_prog.o connect6_prog.o test_adjust_tail.o \

diff --git a/tools/testing/selftests/bpf/sockmap_kcm.c b/tools/testing/selftests/bpf/sockmap_kcm.c
new file mode 100644
index 0000000..4377adc
--- /dev/null
+++ b/tools/testing/selftests/bpf/sockmap_kcm.c

@@ -0,0 +1,14 @@
+#include <linux/bpf.h>
+#include "bpf_helpers.h"
+#include "bpf_util.h"
+#include "bpf_endian.h"
+
+int _version SEC("version") = 1;
+
+SEC("socket_kcm")
+int bpf_prog1(struct __sk_buff *skb)
+{
+	return skb->len;
+}
+
+char _license[] SEC("license") = "GPL";

diff --git a/tools/testing/selftests/bpf/test_maps.c b/tools/testing/selftests/bpf/test_maps.c
index 9b552c0..be20f1d 100644
--- a/tools/testing/selftests/bpf/test_maps.c
+++ b/tools/testing/selftests/bpf/test_maps.c

@@ -20,6 +20,7 @@
 #include <sys/socket.h>
 #include <netinet/in.h>
 #include <linux/bpf.h>
+#include <linux/kcm.h>
 
 #include <bpf/bpf.h>
 #include <bpf/libbpf.h>

@@ -479,14 +480,16 @@ static void test_devmap(int task, void *data)
 #define SOCKMAP_PARSE_PROG "./sockmap_parse_prog.o"
 #define SOCKMAP_VERDICT_PROG "./sockmap_verdict_prog.o"
 #define SOCKMAP_TCP_MSG_PROG "./sockmap_tcp_msg_prog.o"
+#define KCM_PROG "./sockmap_kcm.o"
 static void test_sockmap(int tasks, void *data)
 {
 	struct bpf_map *bpf_map_rx, *bpf_map_tx, *bpf_map_msg, *bpf_map_break;
-	int map_fd_msg = 0, map_fd_rx = 0, map_fd_tx = 0, map_fd_break;
+	int map_fd_msg = 0, map_fd_rx = 0, map_fd_tx = 0, map_fd_break, kcm;
 	int ports[] = {50200, 50201, 50202, 50204};
 	int err, i, fd, udp, sfd[6] = {0xdeadbeef};
 	u8 buf[20] = {0x0, 0x5, 0x3, 0x2, 0x1, 0x0};
-	int parse_prog, verdict_prog, msg_prog;
+	int parse_prog, verdict_prog, msg_prog, kcm_prog;
+	struct kcm_attach attach_info;
 	struct sockaddr_in addr;
 	int one = 1, s, sc, rc;
 	struct bpf_object *obj;

@@ -744,6 +747,62 @@ static void test_sockmap(int tasks, void *data)
 		goto out_sockmap;
 	}
 
+	/* Test adding a KCM socket into map */
+#define AF_KCM 41
+	kcm = socket(AF_KCM, SOCK_DGRAM, KCMPROTO_CONNECTED);
+	if (kcm == -1) {
+		printf("Warning, KCM+Sockmap could not be tested.\n");
+		goto skip_kcm;
+	}
+
+	err = bpf_prog_load(KCM_PROG,
+			    BPF_PROG_TYPE_SOCKET_FILTER,
+			    &obj, &kcm_prog);
+	if (err) {
+		printf("Failed to load SK_SKB parse prog\n");
+		goto out_sockmap;
+	}
+
+	i = 2;
+	memset(&attach_info, 0, sizeof(attach_info));
+	attach_info.fd = sfd[i];
+	attach_info.bpf_fd = kcm_prog;
+	err = ioctl(kcm, SIOCKCMATTACH, &attach_info);
+	if (!err) {
+		perror("Failed KCM attached to sockmap fd: ");
+		goto out_sockmap;
+	}
+
+	err = bpf_map_delete_elem(fd, &i);
+	if (err) {
+		printf("Failed delete sockmap from empty map %i %i\n", err, errno);
+		goto out_sockmap;
+	}
+
+	err = ioctl(kcm, SIOCKCMATTACH, &attach_info);
+	if (err) {
+		perror("Failed KCM attach");
+		goto out_sockmap;
+	}
+
+	err = bpf_map_update_elem(fd, &i, &sfd[i], BPF_ANY);
+	if (!err) {
+		printf("Failed sockmap attached KCM sock!\n");
+		goto out_sockmap;
+	}
+	err = ioctl(kcm, SIOCKCMUNATTACH, &attach_info);
+	if (err) {
+		printf("Failed detach KCM sock!\n");
+		goto out_sockmap;
+	}
+
+	err = bpf_map_update_elem(fd, &i, &sfd[i], BPF_ANY);
+	if (err) {
+		printf("Failed post-kcm update sockmap '%i:%i'\n",
+		       i, sfd[i]);
+		goto out_sockmap;
+	}
+
 	/* Test map update elem afterwards fd lives in fd and map_fd */
 	for (i = 2; i < 6; i++) {
 		err = bpf_map_update_elem(map_fd_rx, &i, &sfd[i], BPF_ANY);

@@ -776,6 +835,7 @@ static void test_sockmap(int tasks, void *data)
 		}
 	}
 
+skip_kcm:
 	/* Put sfd[2] (sending fd below) into msg map to test sendmsg bpf */
 	i = 0;
 	err = bpf_map_update_elem(map_fd_msg, &i, &sfd[2], BPF_ANY);

-- 
1.9.1

On 10/18/2018 10:58 PM, John Fastabend wrote:

We check if the sk_user_data (the psock in skmsg) is in fact a sockmap
type to late, after we read the refcnt which is an error. This
series moves the check up before reading refcnt and also adds a test
to test_maps to test trying to add a KCM socket into a sockmap.

While reviewig this code I also found an issue with KCM and kTLS
where each uses sk_data_ready hooks and associated stream parser
breaking expectations in kcm, ktls or both. But that fix will need
to go to net.

Thanks to Eric for reporting.

v2: Fix up file +/- my scripts lost track of them
v3: return EBUSY if refcnt is zero

John Fastabend (2):
  bpf: skmsg, fix psock create on existing kcm/tls port
  bpf: test_maps add a test to catch kcm + sockmap

 include/linux/skmsg.h                     | 25 +++++++++---
 net/core/sock_map.c                       | 11 +++---
 tools/testing/selftests/bpf/Makefile      |  2 +-
 tools/testing/selftests/bpf/sockmap_kcm.c | 14 +++++++
 tools/testing/selftests/bpf/test_maps.c   | 64 ++++++++++++++++++++++++++++++-
 5 files changed, 103 insertions(+), 13 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/sockmap_kcm.c

Applied, thanks!

On 10/20/2018 12:51 AM, Daniel Borkmann wrote:

On 10/18/2018 10:58 PM, John Fastabend wrote:

quoted

We check if the sk_user_data (the psock in skmsg) is in fact a sockmap
type to late, after we read the refcnt which is an error. This
series moves the check up before reading refcnt and also adds a test
to test_maps to test trying to add a KCM socket into a sockmap.

While reviewig this code I also found an issue with KCM and kTLS
where each uses sk_data_ready hooks and associated stream parser
breaking expectations in kcm, ktls or both. But that fix will need
to go to net.

Thanks to Eric for reporting.

v2: Fix up file +/- my scripts lost track of them
v3: return EBUSY if refcnt is zero

John Fastabend (2):
  bpf: skmsg, fix psock create on existing kcm/tls port
  bpf: test_maps add a test to catch kcm + sockmap

 include/linux/skmsg.h                     | 25 +++++++++---
 net/core/sock_map.c                       | 11 +++---
 tools/testing/selftests/bpf/Makefile      |  2 +-
 tools/testing/selftests/bpf/sockmap_kcm.c | 14 +++++++
 tools/testing/selftests/bpf/test_maps.c   | 64 ++++++++++++++++++++++++++++++-
 5 files changed, 103 insertions(+), 13 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/sockmap_kcm.c

Applied, thanks!

Fyi, I've only applied patch 1/2 for now to get the bug fixed. The patch 2/2 throws
a bunch of warnings that look like the below. Also, I think we leak kcm socket in
error paths and once we're done with testing, so would be good to close it once
unneeded. Please respin the test as a stand-alone commit, thanks:

[...]
bpf-next/tools/testing/selftests/bpf/libbpf.a -lcap -lelf -lrt -lpthread -o /home/darkstar/trees/bpf-next-ok/tools/testing/selftests/bpf/test_maps
test_maps.c: In function ‘test_sockmap’:
test_maps.c:869:0: warning: "AF_KCM" redefined
 #define AF_KCM 41

In file included from /usr/include/sys/socket.h:38:0,
                 from test_maps.c:21:
/usr/include/bits/socket.h:133:0: note: this is the location of the previous definition
 #define AF_KCM  PF_KCM

On 10/19/2018 03:57 PM, Daniel Borkmann wrote:

On 10/20/2018 12:51 AM, Daniel Borkmann wrote:

quoted

On 10/18/2018 10:58 PM, John Fastabend wrote:

quoted

We check if the sk_user_data (the psock in skmsg) is in fact a sockmap
type to late, after we read the refcnt which is an error. This
series moves the check up before reading refcnt and also adds a test
to test_maps to test trying to add a KCM socket into a sockmap.

While reviewig this code I also found an issue with KCM and kTLS
where each uses sk_data_ready hooks and associated stream parser
breaking expectations in kcm, ktls or both. But that fix will need
to go to net.

Thanks to Eric for reporting.

v2: Fix up file +/- my scripts lost track of them
v3: return EBUSY if refcnt is zero

John Fastabend (2):
  bpf: skmsg, fix psock create on existing kcm/tls port
  bpf: test_maps add a test to catch kcm + sockmap

 include/linux/skmsg.h                     | 25 +++++++++---
 net/core/sock_map.c                       | 11 +++---
 tools/testing/selftests/bpf/Makefile      |  2 +-
 tools/testing/selftests/bpf/sockmap_kcm.c | 14 +++++++
 tools/testing/selftests/bpf/test_maps.c   | 64 ++++++++++++++++++++++++++++++-
 5 files changed, 103 insertions(+), 13 deletions(-)
 create mode 100644 tools/testing/selftests/bpf/sockmap_kcm.c

Applied, thanks!

Fyi, I've only applied patch 1/2 for now to get the bug fixed. The patch 2/2 throws
a bunch of warnings that look like the below. Also, I think we leak kcm socket in
error paths and once we're done with testing, so would be good to close it once
unneeded. Please respin the test as a stand-alone commit, thanks:

Thanks, I didn't see the warnings below locally but will look
into spinning a good version tonight with the closing sock fix
as well.

John

[...]
bpf-next/tools/testing/selftests/bpf/libbpf.a -lcap -lelf -lrt -lpthread -o /home/darkstar/trees/bpf-next-ok/tools/testing/selftests/bpf/test_maps
test_maps.c: In function ‘test_sockmap’:
test_maps.c:869:0: warning: "AF_KCM" redefined
 #define AF_KCM 41

In file included from /usr/include/sys/socket.h:38:0,
                 from test_maps.c:21:
/usr/include/bits/socket.h:133:0: note: this is the location of the previous definition
 #define AF_KCM  PF_KCM