[PATCH net-next] tools: ynl: check for overflow of constructed messages

[PATCH net-next] tools: ynl: check for overflow of constructed messages

[Jakub Kicinski]

Donald points out that we don't check for overflows.
Stash the length of the message on nlmsg_pid (nlmsg_seq would
do as well). This allows the attribute helpers to remain
self-contained (no extra arguments). Also let the put
helpers continue to return nothing. The error is checked
only in (newly introduced) ynl_msg_end().

Signed-off-by: Jakub Kicinski <kuba@kernel.org>
---
This was first discussed when I posted the libmnl replacement:
https://lore.kernel.org/all/CAD4GDZzF55bkoZ_o0S784PmfW4+L_QrG2ofWg6CeQk4FCWTUiw@mail.gmail.com/
---
 tools/net/ynl/lib/ynl-priv.h | 34 ++++++++++++++++++++++++++++++----
 tools/net/ynl/lib/ynl.c      | 36 ++++++++++++++++++++++++++++++++++++
 tools/net/ynl/lib/ynl.h      |  2 ++
 3 files changed, 68 insertions(+), 4 deletions(-)

diff --git a/tools/net/ynl/lib/ynl-priv.h b/tools/net/ynl/lib/ynl-priv.h
index a8099fab035d..6cf890080dc0 100644
--- a/tools/net/ynl/lib/ynl-priv.h
+++ b/tools/net/ynl/lib/ynl-priv.h
@@ -135,6 +135,8 @@ int ynl_error_parse(struct ynl_parse_arg *yarg, const char *msg);
 
 /* Netlink message handling helpers */
 
+#define YNL_MSG_OVERFLOW	1
+
 static inline struct nlmsghdr *ynl_nlmsg_put_header(void *buf)
 {
 	struct nlmsghdr *nlh = buf;
@@ -239,11 +241,29 @@ ynl_attr_first(const void *start, size_t len, size_t skip)
 	return ynl_attr_if_good(start + len, attr);
 }
 
+static inline bool
+__ynl_attr_put_overflow(struct nlmsghdr *nlh, size_t size)
+{
+	bool o;
+
+	/* ynl_msg_start() stashed buffer length in nlmsg_pid. */
+	o = nlh->nlmsg_len + NLA_HDRLEN + NLMSG_ALIGN(size) > nlh->nlmsg_pid;
+	if (o)
+		/* YNL_MSG_OVERFLOW is < NLMSG_HDRLEN, all subsequent checks
+		 * are guaranteed to fail.
+		 */
+		nlh->nlmsg_pid = YNL_MSG_OVERFLOW;
+	return o;
+}
+
 static inline struct nlattr *
 ynl_attr_nest_start(struct nlmsghdr *nlh, unsigned int attr_type)
 {
 	struct nlattr *attr;
 
+	if (__ynl_attr_put_overflow(nlh, 0))
+		return ynl_nlmsg_end_addr(nlh) - NLA_HDRLEN;
+
 	attr = ynl_nlmsg_end_addr(nlh);
 	attr->nla_type = attr_type | NLA_F_NESTED;
 	nlh->nlmsg_len += NLA_HDRLEN;
@@ -263,6 +283,9 @@ ynl_attr_put(struct nlmsghdr *nlh, unsigned int attr_type,
 {
 	struct nlattr *attr;
 
+	if (__ynl_attr_put_overflow(nlh, size))
+		return;
+
 	attr = ynl_nlmsg_end_addr(nlh);
 	attr->nla_type = attr_type;
 	attr->nla_len = NLA_HDRLEN + size;
@@ -276,14 +299,17 @@ static inline void
 ynl_attr_put_str(struct nlmsghdr *nlh, unsigned int attr_type, const char *str)
 {
 	struct nlattr *attr;
-	const char *end;
+	size_t len;
+
+	len = strlen(str);
+	if (__ynl_attr_put_overflow(nlh, len))
+		return;
 
 	attr = ynl_nlmsg_end_addr(nlh);
 	attr->nla_type = attr_type;
 
-	end = stpcpy(ynl_attr_data(attr), str);
-	attr->nla_len =
-		NLA_HDRLEN + NLA_ALIGN(end - (char *)ynl_attr_data(attr));
+	strcpy(ynl_attr_data(attr), str);
+	attr->nla_len = NLA_HDRLEN + NLA_ALIGN(len);
 
 	nlh->nlmsg_len += NLMSG_ALIGN(attr->nla_len);
 }
diff --git a/tools/net/ynl/lib/ynl.c b/tools/net/ynl/lib/ynl.c
index 484070492b17..5c9d955d0f22 100644
--- a/tools/net/ynl/lib/ynl.c
+++ b/tools/net/ynl/lib/ynl.c
@@ -404,9 +404,33 @@ struct nlmsghdr *ynl_msg_start(struct ynl_sock *ys, __u32 id, __u16 flags)
 	nlh->nlmsg_flags = flags;
 	nlh->nlmsg_seq = ++ys->seq;
 
+	/* This is a local YNL hack for length checking, we put the buffer
+	 * length in nlmsg_pid, since messages sent to the kernel always use
+	 * PID 0. Message needs to be terminated with ynl_msg_end().
+	 */
+	nlh->nlmsg_pid = YNL_SOCKET_BUFFER_SIZE;
+
 	return nlh;
 }
 
+static int ynl_msg_end(struct ynl_sock *ys, struct nlmsghdr *nlh)
+{
+	/* We stash buffer length in nlmsg_pid. */
+	if (nlh->nlmsg_pid == 0) {
+		yerr(ys, YNL_ERROR_INPUT_INVALID,
+		     "Unknwon input buffer length");
+		return -EINVAL;
+	}
+	if (nlh->nlmsg_pid == YNL_MSG_OVERFLOW) {
+		yerr(ys, YNL_ERROR_INPUT_TOO_BIG,
+		     "Constructred message longer than internal buffer");
+		return -EMSGSIZE;
+	}
+
+	nlh->nlmsg_pid = 0;
+	return 0;
+}
+
 struct nlmsghdr *
 ynl_gemsg_start(struct ynl_sock *ys, __u32 id, __u16 flags,
 		__u8 cmd, __u8 version)
@@ -607,6 +631,10 @@ static int ynl_sock_read_family(struct ynl_sock *ys, const char *family_name)
 	nlh = ynl_gemsg_start_req(ys, GENL_ID_CTRL, CTRL_CMD_GETFAMILY, 1);
 	ynl_attr_put_str(nlh, CTRL_ATTR_FAMILY_NAME, family_name);
 
+	err = ynl_msg_end(ys, nlh);
+	if (err < 0)
+		return err;
+
 	err = send(ys->socket, nlh, nlh->nlmsg_len, 0);
 	if (err < 0) {
 		perr(ys, "failed to request socket family info");
@@ -868,6 +896,10 @@ int ynl_exec(struct ynl_sock *ys, struct nlmsghdr *req_nlh,
 {
 	int err;
 
+	err = ynl_msg_end(ys, req_nlh);
+	if (err < 0)
+		return err;
+
 	err = send(ys->socket, req_nlh, req_nlh->nlmsg_len, 0);
 	if (err < 0)
 		return err;
@@ -921,6 +953,10 @@ int ynl_exec_dump(struct ynl_sock *ys, struct nlmsghdr *req_nlh,
 {
 	int err;
 
+	err = ynl_msg_end(ys, req_nlh);
+	if (err < 0)
+		return err;
+
 	err = send(ys->socket, req_nlh, req_nlh->nlmsg_len, 0);
 	if (err < 0)
 		return err;
diff --git a/tools/net/ynl/lib/ynl.h b/tools/net/ynl/lib/ynl.h
index dbeeef8ce91a..9842e85a8c57 100644
--- a/tools/net/ynl/lib/ynl.h
+++ b/tools/net/ynl/lib/ynl.h
@@ -20,6 +20,8 @@ enum ynl_error_code {
 	YNL_ERROR_ATTR_INVALID,
 	YNL_ERROR_UNKNOWN_NTF,
 	YNL_ERROR_INV_RESP,
+	YNL_ERROR_INPUT_INVALID,
+	YNL_ERROR_INPUT_TOO_BIG,
 };
 
 /**
-- 
2.44.0

Re: [PATCH net-next] tools: ynl: check for overflow of constructed messages

[Donald Hunter]

Jakub Kicinski <kuba@kernel.org> writes:

> Donald points out that we don't check for overflows.
> Stash the length of the message on nlmsg_pid (nlmsg_seq would
> do as well). This allows the attribute helpers to remain
> self-contained (no extra arguments). Also let the put
> helpers continue to return nothing. The error is checked
> only in (newly introduced) ynl_msg_end().
>
> Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Reviewed-by: Donald Hunter <donald.hunter@gmail.com>

Re: [PATCH net-next] tools: ynl: check for overflow of constructed messages

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Tue,  5 Mar 2024 10:50:00 -0800 you wrote:
> Donald points out that we don't check for overflows.
> Stash the length of the message on nlmsg_pid (nlmsg_seq would
> do as well). This allows the attribute helpers to remain
> self-contained (no extra arguments). Also let the put
> helpers continue to return nothing. The error is checked
> only in (newly introduced) ynl_msg_end().
> 
> [...]

Here is the summary with links:
  - [net-next] tools: ynl: check for overflow of constructed messages
    https://git.kernel.org/netdev/net-next/c/15d2540e0d62

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html