On 05/04/2023 19:42, Konstantin Meskhidze (A) wrote:
4/4/2023 7:42 PM, Mickaël Salaün пишет:
On 04/04/2023 11:31, Konstantin Meskhidze (A) wrote:
3/31/2023 8:24 PM, Mickaël Salaün пишет:
On 23/03/2023 09:52, Konstantin Meskhidze wrote:
This commit adds network rules support in the ruleset management
helpers and the landlock_create_ruleset syscall.
Refactor user space API to support network actions. Add new network
access flags, network rule and network attributes. Increment Landlock
ABI version. Expand access_masks_t to u32 to be sure network access
rights can be stored. Implement socket_bind() and socket_connect()
LSM hooks, which enable to restrict TCP socket binding and connection
to specific ports.
Signed-off-by: Konstantin Meskhidze <konstantin.meskhidze@xxxxxxxxxx>
---
Changes since v9:
* Changes UAPI port field to __u64.
* Moves shared code into check_socket_access().
* Adds get_raw_handled_net_accesses() and
get_current_net_domain() helpers.
* Minor fixes.
Changes since v8:
* Squashes commits.
* Refactors commit message.
* Changes UAPI port field to __be16.
* Changes logic of bind/connect hooks with AF_UNSPEC families.
* Adds address length checking.
* Minor fixes.
Changes since v7:
* Squashes commits.
* Increments ABI version to 4.
* Refactors commit message.
* Minor fixes.
Changes since v6:
* Renames landlock_set_net_access_mask() to landlock_add_net_access_mask()
because it OR values.
* Makes landlock_add_net_access_mask() more resilient incorrect values.
* Refactors landlock_get_net_access_mask().
* Renames LANDLOCK_MASK_SHIFT_NET to LANDLOCK_SHIFT_ACCESS_NET and use
LANDLOCK_NUM_ACCESS_FS as value.
* Updates access_masks_t to u32 to support network access actions.
* Refactors landlock internal functions to support network actions with
landlock_key/key_type/id types.
Changes since v5:
* Gets rid of partial revert from landlock_add_rule
syscall.
* Formats code with clang-format-14.
Changes since v4:
* Refactors landlock_create_ruleset() - splits ruleset and
masks checks.
* Refactors landlock_create_ruleset() and landlock mask
setters/getters to support two rule types.
* Refactors landlock_add_rule syscall add_rule_path_beneath
function by factoring out get_ruleset_from_fd() and
landlock_put_ruleset().
Changes since v3:
* Splits commit.
* Adds network rule support for internal landlock functions.
* Adds set_mask and get_mask for network.
* Adds rb_root root_net_port.
---
include/uapi/linux/landlock.h | 49 +++++
security/landlock/Kconfig | 1 +
security/landlock/Makefile | 2 +
security/landlock/limits.h | 6 +-
security/landlock/net.c | 198 +++++++++++++++++++
security/landlock/net.h | 26 +++
security/landlock/ruleset.c | 52 ++++-
security/landlock/ruleset.h | 63 +++++-
security/landlock/setup.c | 2 +
security/landlock/syscalls.c | 72 ++++++-
tools/testing/selftests/landlock/base_test.c | 2 +-
11 files changed, 450 insertions(+), 23 deletions(-)
create mode 100644 security/landlock/net.c
create mode 100644 security/landlock/net.h
[...]
diff --git a/security/landlock/net.c b/security/landlock/net.c
[...]
+static int check_addrlen(const struct sockaddr *const address, int addrlen)
const int addrlen
Got it.
+{
+ if (addrlen < offsetofend(struct sockaddr, sa_family))
+ return -EINVAL;
+ switch (address->sa_family) {
+ case AF_UNSPEC:
+ case AF_INET:
+ if (addrlen < sizeof(struct sockaddr_in))
+ return -EINVAL;
+ return 0;
+#if IS_ENABLED(CONFIG_IPV6)
+ case AF_INET6:
+ if (addrlen < SIN6_LEN_RFC2133)
+ return -EINVAL;
+ return 0;
+#endif
+ }
+ WARN_ON_ONCE(1);
+ return 0;
+}
+
+static u16 get_port(const struct sockaddr *const address)
+{
+ /* Gets port value in host byte order. */
+ switch (address->sa_family) {
+ case AF_UNSPEC:
+ case AF_INET: {
+ const struct sockaddr_in *const sockaddr =
+ (struct sockaddr_in *)address;
+ return ntohs(sockaddr->sin_port);
Storing ports in big endian (in rulesets) would avoid converting them
every time the kernel checks a socket port. The above comment should
then be updated too.
I thought we came to a conclusion to stick to host endianess and
let kernel do the checks under the hood:
https://lore.kernel.org/linux-security-module/278ab07f-7583-a4e0-3d37-1bacd091531d@xxxxxxxxxxx/
Did I misunderstand something?
We indeed stick to the host endianess for the UAPI/syscalls, but
internally the kernel has to do the conversion with as it is currently
done by calling ntohs(). To avoid calling ntohs() every time get_port()
is called, we can instead only call htons() when creating rules (i.e.
one-time htons call instead of multiple ntohs calls).
Do you mean we need to covert port in landlock_append_net_rule():
...
int err;
const struct landlock_id id = {
.key.data = ntohs(port),
.type = LANDLOCK_KEY_NET_PORT,
};
BUILD_BUG_ON(sizeof(port) > sizeof(id.key.data));
...
????
landlock_append_net_rule() takes a u16 (host endianess, which is the
case with this patch series) and should store a big endian 16-bit
integer. See my patch:
const struct landlock_id id = {
- .key.data = port,
+ .key.data = (__force uintptr_t)htons(port),
.type = LANDLOCK_KEY_NET_PORT,
};
Do you mean we need to do port converting __be16 -> u16 in
check_socket_access()???
Removing the ntohs() call from get_port() enables to return __be16
instead of u16, and check_socket_access() will then need to use the same
type.
Ok. I got it. Thanks.
+ }
+#if IS_ENABLED(CONFIG_IPV6)
+ case AF_INET6: {
+ const struct sockaddr_in6 *const sockaddr_ip6 =
+ (struct sockaddr_in6 *)address;
+ return ntohs(sockaddr_ip6->sin6_port);
+ }
+#endif
+ }
+ WARN_ON_ONCE(1);
+ return 0;
+}
+
+static int check_socket_access(struct socket *sock, struct sockaddr *address, int addrlen, u16 port,
+ access_mask_t access_request)
+{
+ int ret;
+ bool allowed = false;
+ layer_mask_t layer_masks[LANDLOCK_NUM_ACCESS_NET] = {};
+ const struct landlock_rule *rule;
+ access_mask_t handled_access;
+ const struct landlock_id id = {
+ .key.data = port,
+ .type = LANDLOCK_KEY_NET_PORT,
+ };
+ const struct landlock_ruleset *const domain = get_current_net_domain();
+
+ if (WARN_ON_ONCE(!domain))
+ return 0;
+ if (WARN_ON_ONCE(domain->num_layers < 1))
+ return -EACCES;
+ /* Check if it's a TCP socket. */
+ if (sock->type != SOCK_STREAM)
+ return 0;
+
+ ret = check_addrlen(address, addrlen);
+ if (ret)
+ return ret;
+
+ switch (address->sa_family) {
+ case AF_UNSPEC:
+ /*
+ * Connecting to an address with AF_UNSPEC dissolves the TCP
+ * association, which have the same effect as closing the
+ * connection while retaining the socket object (i.e., the file
+ * descriptor). As for dropping privileges, closing
+ * connections is always allowed.
+ */
+ if (access_request == LANDLOCK_ACCESS_NET_CONNECT_TCP)
+ return 0;
+
+ /*
+ * For compatibility reason, accept AF_UNSPEC for bind
+ * accesses (mapped to AF_INET) only if the address is
+ * INADDR_ANY (cf. __inet_bind). Checking the address is
+ * required to not wrongfully return -EACCES instead of
+ * -EAFNOSUPPORT.
+ */
+ if (access_request == LANDLOCK_ACCESS_NET_BIND_TCP) {
+ const struct sockaddr_in *const sockaddr =
+ (struct sockaddr_in *)address;
+
+ if (sockaddr->sin_addr.s_addr != htonl(INADDR_ANY))
+ return -EAFNOSUPPORT;
+ }
+
+ fallthrough;
+ case AF_INET:
+#if IS_ENABLED(CONFIG_IPV6)
+ case AF_INET6:
+#endif
+ rule = landlock_find_rule(domain, id);
+ handled_access = landlock_init_layer_masks(
+ domain, access_request, &layer_masks,
+ LANDLOCK_KEY_NET_PORT);
+ allowed = landlock_unmask_layers(rule, handled_access,
+ &layer_masks,
+ ARRAY_SIZE(layer_masks));
+ }
+ return allowed ? 0 : -EACCES;
+}
+
+static int hook_socket_bind(struct socket *sock, struct sockaddr *address,
+ int addrlen)
+{ >>> + return check_socket_access(sock, address, addrlen, get_port(address),
+ LANDLOCK_ACCESS_NET_BIND_TCP);
get_port() is called before check_addrlen(), which is an issue.
You'll find attached a patch for these fixes, please squash it in this
one for the next version.
I'll send other reviews by the end of the week.
+}
+
+static int hook_socket_connect(struct socket *sock, struct sockaddr *address,
+ int addrlen)
+{
+ return check_socket_access(sock, address, addrlen, get_port(address),
+ LANDLOCK_ACCESS_NET_CONNECT_TCP);
+}
[...]
.
