[PATCH bpf-next 0/2] bpf: fix kernel page fault in lpm map trie_get_next_key

[PATCH bpf-next 0/2] bpf: fix kernel page fault in lpm map trie_get_next_key

[Yonghong Song]

A kernel page fault which happens in lpm map trie_get_next_key is reported
by syzbot and Eric. The issue was introduced by commit b471f2f1de8b
("bpf: implement MAP_GET_NEXT_KEY command for LPM_TRIE map").
Patch #1 fixed the issue in the kernel and patch #2 adds a multithreaded
test case in tools/testing/selftests/bpf/test_lpm_map.

Yonghong Song (2):
  bpf: fix kernel page fault in lpm map trie_get_next_key
  tools/bpf: add a multithreaded stress test in bpf selftests
    test_lpm_map

 kernel/bpf/lpm_trie.c                      | 26 ++++----
 tools/testing/selftests/bpf/Makefile       |  2 +-
 tools/testing/selftests/bpf/test_lpm_map.c | 95 ++++++++++++++++++++++++++++++
 3 files changed, 107 insertions(+), 16 deletions(-)

-- 
2.9.5

[PATCH bpf-next 1/2] bpf: fix kernel page fault in lpm map trie_get_next_key

[Yonghong Song]

Commit b471f2f1de8b ("bpf: implement MAP_GET_NEXT_KEY command
for LPM_TRIE map") introduces a bug likes below:

    if (!rcu_dereference(trie->root))
        return -ENOENT;
    if (!key || key->prefixlen > trie->max_prefixlen) {
        root = &trie->root;
        goto find_leftmost;
    }
    ......
  find_leftmost:
    for (node = rcu_dereference(*root); node;) {

In the code after label find_leftmost, it is assumed
that *root should not be NULL, but it is not true as
it is possbile trie->root is changed to NULL by an
asynchronous delete operation.

The issue is reported by syzbot and Eric Dumazet with the
below error log:
  ......
  kasan: CONFIG_KASAN_INLINE enabled
  kasan: GPF could be caused by NULL-ptr deref or user memory access
  general protection fault: 0000 [#1] SMP KASAN
  Dumping ftrace buffer:
     (ftrace buffer empty)
  Modules linked in:
  CPU: 1 PID: 8033 Comm: syz-executor3 Not tainted 4.15.0-rc8+ #4
  Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
  RIP: 0010:trie_get_next_key+0x3c2/0xf10 kernel/bpf/lpm_trie.c:682
  ......

This patch fixed the issue by use local rcu_dereferenced
pointer instead of *(&trie->root) later on.

Fixes: b471f2f1de8b ("bpf: implement MAP_GET_NEXT_KEY command or LPM_TRIE map")
Reported-by: syzbot <syzkaller@googlegroups.com>
Reported-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: Yonghong Song <yhs@fb.com>
---
 kernel/bpf/lpm_trie.c | 26 +++++++++++---------------
 1 file changed, 11 insertions(+), 15 deletions(-)

diff --git a/kernel/bpf/lpm_trie.c b/kernel/bpf/lpm_trie.c
index 8f083ea..7b469d1 100644
--- a/kernel/bpf/lpm_trie.c
+++ b/kernel/bpf/lpm_trie.c
@@ -593,11 +593,10 @@ static void trie_free(struct bpf_map *map)
 
 static int trie_get_next_key(struct bpf_map *map, void *_key, void *_next_key)
 {
+	struct lpm_trie_node *node, *next_node = NULL, *parent, *search_root;
 	struct lpm_trie *trie = container_of(map, struct lpm_trie, map);
 	struct bpf_lpm_trie_key *key = _key, *next_key = _next_key;
-	struct lpm_trie_node *node, *next_node = NULL, *parent;
 	struct lpm_trie_node **node_stack = NULL;
-	struct lpm_trie_node __rcu **root;
 	int err = 0, stack_ptr = -1;
 	unsigned int next_bit;
 	size_t matchlen;
@@ -614,14 +613,13 @@ static int trie_get_next_key(struct bpf_map *map, void *_key, void *_next_key)
 	 */
 
 	/* Empty trie */
-	if (!rcu_dereference(trie->root))
+	search_root = rcu_dereference(trie->root);
+	if (!search_root)
 		return -ENOENT;
 
 	/* For invalid key, find the leftmost node in the trie */
-	if (!key || key->prefixlen > trie->max_prefixlen) {
-		root = &trie->root;
+	if (!key || key->prefixlen > trie->max_prefixlen)
 		goto find_leftmost;
-	}
 
 	node_stack = kmalloc(trie->max_prefixlen * sizeof(struct lpm_trie_node *),
 			     GFP_ATOMIC | __GFP_NOWARN);
@@ -629,7 +627,7 @@ static int trie_get_next_key(struct bpf_map *map, void *_key, void *_next_key)
 		return -ENOMEM;
 
 	/* Try to find the exact node for the given key */
-	for (node = rcu_dereference(trie->root); node;) {
+	for (node = search_root; node;) {
 		node_stack[++stack_ptr] = node;
 		matchlen = longest_prefix_match(trie, node, key);
 		if (node->prefixlen != matchlen ||
@@ -640,10 +638,8 @@ static int trie_get_next_key(struct bpf_map *map, void *_key, void *_next_key)
 		node = rcu_dereference(node->child[next_bit]);
 	}
 	if (!node || node->prefixlen != key->prefixlen ||
-	    (node->flags & LPM_TREE_NODE_FLAG_IM)) {
-		root = &trie->root;
+	    (node->flags & LPM_TREE_NODE_FLAG_IM))
 		goto find_leftmost;
-	}
 
 	/* The node with the exactly-matching key has been found,
 	 * find the first node in postorder after the matched node.
@@ -651,10 +647,10 @@ static int trie_get_next_key(struct bpf_map *map, void *_key, void *_next_key)
 	node = node_stack[stack_ptr];
 	while (stack_ptr > 0) {
 		parent = node_stack[stack_ptr - 1];
-		if (rcu_dereference(parent->child[0]) == node &&
-		    rcu_dereference(parent->child[1])) {
-			root = &parent->child[1];
-			goto find_leftmost;
+		if (rcu_dereference(parent->child[0]) == node) {
+			search_root = rcu_dereference(parent->child[1]);
+			if (search_root)
+				goto find_leftmost;
 		}
 		if (!(parent->flags & LPM_TREE_NODE_FLAG_IM)) {
 			next_node = parent;
@@ -673,7 +669,7 @@ static int trie_get_next_key(struct bpf_map *map, void *_key, void *_next_key)
 	/* Find the leftmost non-intermediate node, all intermediate nodes
 	 * have exact two children, so this function will never return NULL.
 	 */
-	for (node = rcu_dereference(*root); node;) {
+	for (node = search_root; node;) {
 		if (!(node->flags & LPM_TREE_NODE_FLAG_IM))
 			next_node = node;
 		node = rcu_dereference(node->child[0]);
-- 
2.9.5

[PATCH bpf-next 2/2] tools/bpf: add a multithreaded stress test in bpf selftests test_lpm_map

[Yonghong Song]

The new test will spawn four threads, doing map update, delete, lookup
and get_next_key in parallel. It is able to reproduce the issue in the
previous commit found by syzbot and Eric Dumazet.

Signed-off-by: Yonghong Song <yhs@fb.com>
---
 tools/testing/selftests/bpf/Makefile       |  2 +-
 tools/testing/selftests/bpf/test_lpm_map.c | 95 ++++++++++++++++++++++++++++++
 2 files changed, 96 insertions(+), 1 deletion(-)

diff --git a/tools/testing/selftests/bpf/Makefile b/tools/testing/selftests/bpf/Makefile
index 9868835..bf05bc5 100644
--- a/tools/testing/selftests/bpf/Makefile
+++ b/tools/testing/selftests/bpf/Makefile
@@ -11,7 +11,7 @@ ifneq ($(wildcard $(GENHDR)),)
 endif
 
 CFLAGS += -Wall -O2 -I$(APIDIR) -I$(LIBDIR) -I$(GENDIR) $(GENFLAGS) -I../../../include
-LDLIBS += -lcap -lelf -lrt
+LDLIBS += -lcap -lelf -lrt -lpthread
 
 TEST_GEN_PROGS = test_verifier test_tag test_maps test_lru_map test_lpm_map test_progs \
 	test_align test_verifier_log test_dev_cgroup test_tcpbpf_user
diff --git a/tools/testing/selftests/bpf/test_lpm_map.c b/tools/testing/selftests/bpf/test_lpm_map.c
index 0815108..2be87e9 100644
--- a/tools/testing/selftests/bpf/test_lpm_map.c
+++ b/tools/testing/selftests/bpf/test_lpm_map.c
@@ -14,6 +14,7 @@
 #include <errno.h>
 #include <inttypes.h>
 #include <linux/bpf.h>
+#include <pthread.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -641,6 +642,98 @@ static void test_lpm_get_next_key(void)
 	close(map_fd);
 }
 
+#define MAX_TEST_KEYS	4
+struct lpm_mt_test_info {
+	int cmd; /* 0: update, 1: delete, 2: lookup, 3: get_next_key */
+	int iter;
+	int map_fd;
+	struct {
+		__u32 prefixlen;
+		__u32 data;
+	} key[MAX_TEST_KEYS];
+};
+
+static void *lpm_test_command(void *arg)
+{
+	int i, j, ret, iter, key_size;
+	struct lpm_mt_test_info *info = arg;
+	struct bpf_lpm_trie_key *key_p;
+
+	key_size = sizeof(struct bpf_lpm_trie_key) + sizeof(__u32);
+	key_p = alloca(key_size);
+	for (iter = 0; iter < info->iter; iter++)
+		for (i = 0; i < MAX_TEST_KEYS; i++) {
+			/* first half of iterations in forward order,
+			 * and second half in backward order.
+			 */
+			j = (iter < (info->iter / 2)) ? i : MAX_TEST_KEYS - i - 1;
+			key_p->prefixlen = info->key[j].prefixlen;
+			memcpy(key_p->data, &info->key[j].data, sizeof(__u32));
+			if (info->cmd == 0) {
+				__u32 value = j;
+				/* update must succeed */
+				assert(bpf_map_update_elem(info->map_fd, key_p, &value, 0) == 0);
+			} else if (info->cmd == 1) {
+				ret = bpf_map_delete_elem(info->map_fd, key_p);
+				assert(ret == 0 || errno == ENOENT);
+			} else if (info->cmd == 2) {
+				__u32 value;
+				ret = bpf_map_lookup_elem(info->map_fd, key_p, &value);
+				assert(ret == 0 || errno == ENOENT);
+			} else {
+				struct bpf_lpm_trie_key *next_key_p = alloca(key_size);
+				ret = bpf_map_get_next_key(info->map_fd, key_p, next_key_p);
+				assert(ret == 0 || errno == ENOENT || errno == ENOMEM);
+			}
+		}
+
+	// Pass successful exit info back to the main thread
+	pthread_exit((void *)info);
+}
+
+static void setup_lpm_mt_test_info(struct lpm_mt_test_info *info, int map_fd)
+{
+	info->iter = 2000;
+	info->map_fd = map_fd;
+	info->key[0].prefixlen = 16;
+	inet_pton(AF_INET, "192.168.0.0", &info->key[0].data);
+	info->key[1].prefixlen = 24;
+	inet_pton(AF_INET, "192.168.0.0", &info->key[1].data);
+	info->key[2].prefixlen = 24;
+	inet_pton(AF_INET, "192.168.128.0", &info->key[2].data);
+	info->key[3].prefixlen = 24;
+	inet_pton(AF_INET, "192.168.1.0", &info->key[3].data);
+}
+
+static void test_lpm_multi_thread(void)
+{
+	struct lpm_mt_test_info info[4];
+	size_t key_size, value_size;
+	pthread_t thread_id[4];
+	int i, map_fd;
+	void *ret;
+
+	/* create a trie */
+	value_size = sizeof(__u32);
+	key_size = sizeof(struct bpf_lpm_trie_key) + value_size;
+	map_fd = bpf_create_map(BPF_MAP_TYPE_LPM_TRIE, key_size, value_size,
+				100, BPF_F_NO_PREALLOC);
+
+	/* create 4 threads to test update, delete, lookup and get_next_key */
+	setup_lpm_mt_test_info(&info[0], map_fd);
+	for (i = 0; i < 4; i++) {
+		if (i != 0)
+			memcpy(&info[i], &info[0], sizeof(info[i]));
+		info[i].cmd = i;
+		assert(pthread_create(&thread_id[i], NULL, &lpm_test_command, &info[i]) == 0);
+	}
+
+	for (i = 0; i < 4; i++)
+		assert(pthread_join(thread_id[i], &ret) == 0 && ret == (void *)&info[i]);
+
+	close(map_fd);
+}
+
 int main(void)
 {
 	struct rlimit limit  = { RLIM_INFINITY, RLIM_INFINITY };
@@ -667,6 +760,8 @@ int main(void)
 
 	test_lpm_get_next_key();
 
+	test_lpm_multi_thread();
+
 	printf("test_lpm: OK\n");
 	return 0;
 }
-- 
2.9.5

Re: [PATCH bpf-next 0/2] bpf: fix kernel page fault in lpm map trie_get_next_key

[Alexei Starovoitov]

On Fri, Jan 26, 2018 at 03:06:06PM -0800, Yonghong Song wrote:
> A kernel page fault which happens in lpm map trie_get_next_key is reported
> by syzbot and Eric. The issue was introduced by commit b471f2f1de8b
> ("bpf: implement MAP_GET_NEXT_KEY command for LPM_TRIE map").
> Patch #1 fixed the issue in the kernel and patch #2 adds a multithreaded
> test case in tools/testing/selftests/bpf/test_lpm_map.

Applied to bpf-next, thank you Yonghong!