| Message ID | 20190919095903.19370-2-christian.brauner@ubuntu.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | seccomp: continue syscall from notifier | expand |
On Thu, Sep 19, 2019 at 11:59 AM Christian Brauner <christian.brauner@ubuntu.com> wrote: > This allows the seccomp notifier to continue a syscall. [...] > Recently we landed seccomp support for SECCOMP_RET_USER_NOTIF (cf. [4]) > which enables a process (watchee) to retrieve an fd for its seccomp > filter. This fd can then be handed to another (usually more privileged) > process (watcher). The watcher will then be able to receive seccomp > messages about the syscalls having been performed by the watchee. [...] > This can be solved by > telling seccomp to resume the syscall. [...] > @@ -780,8 +783,14 @@ static void seccomp_do_user_notification(int this_syscall, > list_del(&n.list); > out: > mutex_unlock(&match->notify_lock); > + > + /* Userspace requests to continue the syscall. */ > + if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) > + return 0; > + > syscall_set_return_value(current, task_pt_regs(current), > err, ret); > + return -1; > } Seccomp currently expects the various seccomp return values to be fully ordered based on how much action the kernel should take against the requested syscall. Currently, the range of return values is basically divided into three regions: "block syscall in some way" (from SECCOMP_RET_KILL_PROCESS to SECCOMP_RET_USER_NOTIF), "let ptrace decide" (SECCOMP_RET_TRACE) and "allow" (SECCOMP_RET_LOG and SECCOMP_RET_ALLOW). If SECCOMP_RET_USER_NOTIF becomes able to allow syscalls, it will be able to override a negative decision from SECCOMP_RET_TRACE. In practice, that's probably not a big deal, since I'm not aware of anyone actually using SECCOMP_RET_TRACE for security purposes, and on top of that, you'd have to allow ioctl(..., SECCOMP_IOCTL_NOTIF_SEND, ...) and seccomp() with SECCOMP_FILTER_FLAG_NEW_LISTENER in your seccomp policy for this to work. More interestingly, what about the case where two SECCOMP_RET_USER_NOTIF filters are installed? The most recently installed filter takes precedence if the return values's action parts are the same (and this is also documented in the manpage); so if a container engine installs a filter that always intercepts sys_foobar() (and never uses SECCOMP_USER_NOTIF_FLAG_CONTINUE), and then something inside the container also installs a filter that always intercepts sys_foobar() (and always uses SECCOMP_USER_NOTIF_FLAG_CONTINUE), the container engine's filter will become ineffective. With my tendency to overcomplicate things, I'm thinking that maybe it might be a good idea to: - collect a list of all filters that returned SECCOMP_RET_USER_NOTIF, as well as the highest-precedence return value that was less strict than SECCOMP_RET_USER_NOTIF - sequentially send notifications to all of the SECCOMP_RET_USER_NOTIF filters until one doesn't return SECCOMP_USER_NOTIF_FLAG_CONTINUE - if all returned SECCOMP_USER_NOTIF_FLAG_CONTINUE, go with the highest-precedence return value that was less strict than SECCOMP_RET_USER_NOTIF, or allow if no such return value was encountered But perhaps, for now, it would also be enough to just expand the big fat warning note and tell people that if they allow the use of SECCOMP_IOCTL_NOTIF_SEND and SECCOMP_FILTER_FLAG_NEW_LISTENER in their filter, SECCOMP_RET_USER_NOTIF is bypassable. And if someone actually has a usecase where SECCOMP_RET_USER_NOTIF should be secure and nested SECCOMP_RET_USER_NOTIF support is needed, that more complicated logic could be added later?
On Thu, Sep 19, 2019 at 09:37:06PM +0200, Jann Horn wrote: > On Thu, Sep 19, 2019 at 11:59 AM Christian Brauner > <christian.brauner@ubuntu.com> wrote: > > This allows the seccomp notifier to continue a syscall. > [...] > > Recently we landed seccomp support for SECCOMP_RET_USER_NOTIF (cf. [4]) > > which enables a process (watchee) to retrieve an fd for its seccomp > > filter. This fd can then be handed to another (usually more privileged) > > process (watcher). The watcher will then be able to receive seccomp > > messages about the syscalls having been performed by the watchee. > [...] > > This can be solved by > > telling seccomp to resume the syscall. > [...] > > @@ -780,8 +783,14 @@ static void seccomp_do_user_notification(int this_syscall, > > list_del(&n.list); > > out: > > mutex_unlock(&match->notify_lock); > > + > > + /* Userspace requests to continue the syscall. */ > > + if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) > > + return 0; > > + > > syscall_set_return_value(current, task_pt_regs(current), > > err, ret); > > + return -1; > > } > > Seccomp currently expects the various seccomp return values to be > fully ordered based on how much action the kernel should take against > the requested syscall. Currently, the range of return values is > basically divided into three regions: "block syscall in some way" > (from SECCOMP_RET_KILL_PROCESS to SECCOMP_RET_USER_NOTIF), "let ptrace > decide" (SECCOMP_RET_TRACE) and "allow" (SECCOMP_RET_LOG and > SECCOMP_RET_ALLOW). If SECCOMP_RET_USER_NOTIF becomes able to allow > syscalls, it will be able to override a negative decision from > SECCOMP_RET_TRACE. > > In practice, that's probably not a big deal, since I'm not aware of > anyone actually using SECCOMP_RET_TRACE for security purposes, and on > top of that, you'd have to allow ioctl(..., SECCOMP_IOCTL_NOTIF_SEND, > ...) and seccomp() with SECCOMP_FILTER_FLAG_NEW_LISTENER in your > seccomp policy for this to work. > > More interestingly, what about the case where two > SECCOMP_RET_USER_NOTIF filters are installed? The most recently > installed filter takes precedence if the return values's action parts > are the same (and this is also documented in the manpage); so if a > container engine installs a filter that always intercepts sys_foobar() > (and never uses SECCOMP_USER_NOTIF_FLAG_CONTINUE), and then something > inside the container also installs a filter that always intercepts > sys_foobar() (and always uses SECCOMP_USER_NOTIF_FLAG_CONTINUE), the > container engine's filter will become ineffective. Excellent point. We discussed the nested container case today. > > With my tendency to overcomplicate things, I'm thinking that maybe it > might be a good idea to: > - collect a list of all filters that returned SECCOMP_RET_USER_NOTIF, > as well as the highest-precedence return value that was less strict > than SECCOMP_RET_USER_NOTIF > - sequentially send notifications to all of the > SECCOMP_RET_USER_NOTIF filters until one doesn't return > SECCOMP_USER_NOTIF_FLAG_CONTINUE > - if all returned SECCOMP_USER_NOTIF_FLAG_CONTINUE, go with the > highest-precedence return value that was less strict than > SECCOMP_RET_USER_NOTIF, or allow if no such return value was > encountered > > But perhaps, for now, it would also be enough to just expand the big > fat warning note and tell people that if they allow the use of > SECCOMP_IOCTL_NOTIF_SEND and SECCOMP_FILTER_FLAG_NEW_LISTENER in their > filter, SECCOMP_RET_USER_NOTIF is bypassable. And if someone actually > has a usecase where SECCOMP_RET_USER_NOTIF should be secure and nested > SECCOMP_RET_USER_NOTIF support is needed, that more complicated logic > could be added later? Yes, I think that is the correct approach for now. Realistically, the most useful scenario is a host-privileged supervisor process and a user-namespaced supervised process (or to use a concrete example, a host-privileged container manager and an unprivileged container). Having a user-namespaced supervisor process supervising another nested user-namespaced process is for the most part useless because the supervisor can't do any of the interesting syscalls (e.g. mounting block devices that are deemed safe, faking mknod() etc.). So I expect seccomp with USER_NOTIF to be blocked just for good measure. Also - maybe I'm wrong - the warning we added points out that this is only safe if the supervised process can already rely on kernel (or other) restrictions, i.e. even if an attacker overwrites pointer syscall arguments with harmful ones the supervisor must be sure that they are already blocked anyway. Which can be generalized to: if an unwanted syscall goes through in _some_ way then the supervisor must be sure that it is blocked. Iiuc, for your specific attack all the nested attacker can do is to never actually get the (outer) supervisor to fake the syscall for it. A more interesting case might be where the host-privileged supervising process wants to deny a syscall that would otherwise succeed. But if that's the case then the outer supervisor is trying to implement a security policy. But we explicitly point out that this is not possible with the notifier in general. But honestly, that is very advanced and it seems unlikely that someone would want this. So I'd say let's just point this out. Christian
diff --git a/include/uapi/linux/seccomp.h b/include/uapi/linux/seccomp.h index 90734aa5aa36..8a5dafed8a64 100644 --- a/include/uapi/linux/seccomp.h +++ b/include/uapi/linux/seccomp.h @@ -76,6 +76,26 @@ struct seccomp_notif { struct seccomp_data data; }; +/* + * Valid flags for struct seccomp_notif_resp + * + * Note, the SECCOMP_USER_NOTIF_FLAG_CONTINUE flag must be used with caution! + * If set by the process supervising the syscalls of another process the + * syscall will continue. This is problematic because of an inherent TOCTOU. + * An attacker can exploit the time while the supervised process is waiting on + * a response from the supervising process to rewrite syscall arguments which + * are passed as pointers of the intercepted syscall. + * It should be absolutely clear that this means that the seccomp notifier + * _cannot_ be used to implement a security policy! It should only ever be used + * in scenarios where a more privileged process supervises the syscalls of a + * lesser privileged process to get around kernel-enforced security + * restrictions when the privileged process deems this safe. In other words, + * in order to continue a syscall the supervising process should be sure that + * another security mechanism or the kernel itself will sufficiently block + * syscalls if arguments are rewritten to something unsafe. + */ +#define SECCOMP_USER_NOTIF_FLAG_CONTINUE BIT(0) + struct seccomp_notif_resp { __u64 id; __s64 val; diff --git a/kernel/seccomp.c b/kernel/seccomp.c index dba52a7db5e8..12d2227e5786 100644 --- a/kernel/seccomp.c +++ b/kernel/seccomp.c @@ -75,6 +75,7 @@ struct seccomp_knotif { /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */ int error; long val; + u32 flags; /* Signals when this has entered SECCOMP_NOTIFY_REPLIED */ struct completion ready; @@ -732,11 +733,12 @@ static u64 seccomp_next_notify_id(struct seccomp_filter *filter) return filter->notif->next_id++; } -static void seccomp_do_user_notification(int this_syscall, - struct seccomp_filter *match, - const struct seccomp_data *sd) +static int seccomp_do_user_notification(int this_syscall, + struct seccomp_filter *match, + const struct seccomp_data *sd) { int err; + u32 flags = 0; long ret = 0; struct seccomp_knotif n = {}; @@ -764,6 +766,7 @@ static void seccomp_do_user_notification(int this_syscall, if (err == 0) { ret = n.val; err = n.error; + flags = n.flags; } /* @@ -780,8 +783,14 @@ static void seccomp_do_user_notification(int this_syscall, list_del(&n.list); out: mutex_unlock(&match->notify_lock); + + /* Userspace requests to continue the syscall. */ + if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) + return 0; + syscall_set_return_value(current, task_pt_regs(current), err, ret); + return -1; } static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, @@ -867,8 +876,10 @@ static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, return 0; case SECCOMP_RET_USER_NOTIF: - seccomp_do_user_notification(this_syscall, match, sd); - goto skip; + if (seccomp_do_user_notification(this_syscall, match, sd)) + goto skip; + + return 0; case SECCOMP_RET_LOG: seccomp_log(this_syscall, 0, action, true); @@ -1087,7 +1098,11 @@ static long seccomp_notify_send(struct seccomp_filter *filter, if (copy_from_user(&resp, buf, sizeof(resp))) return -EFAULT; - if (resp.flags) + if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE) + return -EINVAL; + + if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) && + (resp.error || resp.val)) return -EINVAL; ret = mutex_lock_interruptible(&filter->notify_lock); @@ -1116,6 +1131,7 @@ static long seccomp_notify_send(struct seccomp_filter *filter, knotif->state = SECCOMP_NOTIFY_REPLIED; knotif->error = resp.error; knotif->val = resp.val; + knotif->flags = resp.flags; complete(&knotif->ready); out: mutex_unlock(&filter->notify_lock);
This allows the seccomp notifier to continue a syscall. A positive discussion about this feature was triggered by a post to the ksummit-discuss mailing list (cf. [3]) and took place during KSummit (cf. [1]) and again at the containers/checkpoint-restore micro-conference at Linux Plumbers. Recently we landed seccomp support for SECCOMP_RET_USER_NOTIF (cf. [4]) which enables a process (watchee) to retrieve an fd for its seccomp filter. This fd can then be handed to another (usually more privileged) process (watcher). The watcher will then be able to receive seccomp messages about the syscalls having been performed by the watchee. This feature is heavily used in some userspace workloads. For example, it is currently used to intercept mknod() syscalls in user namespaces aka in containers. The mknod() syscall can be easily filtered based on dev_t. This allows us to only intercept a very specific subset of mknod() syscalls. Furthermore, mknod() is not possible in user namespaces toto coelo and so intercepting and denying syscalls that are not in the whitelist on accident is not a big deal. The watchee won't notice a difference. In contrast to mknod(), a lot of other syscall we intercept (e.g. setxattr()) cannot be easily filtered like mknod() because they have pointer arguments. Additionally, some of them might actually succeed in user namespaces (e.g. setxattr() for all "user.*" xattrs). Since we currently cannot tell seccomp to continue from a user notifier we are stuck with performing all of the syscalls in lieu of the container. This is a huge security liability since it is extremely difficult to correctly assume all of the necessary privileges of the calling task such that the syscall can be successfully emulated without escaping other additional security restrictions (think missing CAP_MKNOD for mknod(), or MS_NODEV on a filesystem etc.). This can be solved by telling seccomp to resume the syscall. One thing that came up in the discussion was the problem that another thread could change the memory after userspace has decided to let the syscall continue which is a well known TOCTOU with seccomp which is present in other ways already. The discussion showed that this feature is already very useful for any syscall without pointer arguments. For any accidentally intercepted non-pointer syscall it is safe to continue. For syscalls with pointer arguments there is a race but for any cautious userspace and the main usec cases the race doesn't matter. The notifier is intended to be used in a scenario where a more privileged watcher supervises the syscalls of lesser privileged watchee to allow it to get around kernel-enforced limitations by performing the syscall for it whenever deemed save by the watcher. Hence, if a user tricks the watcher into allowing a syscall they will either get a deny based on kernel-enforced restrictions later or they will have changed the arguments in such a way that they manage to perform a syscall with arguments that they would've been allowed to do anyway. In general, it is good to point out again, that the notifier fd was not intended to allow userspace to implement a security policy but rather to work around kernel security mechanisms in cases where the watcher knows that a given action is safe to perform. /* References */ [1]: https://linuxplumbersconf.org/event/4/contributions/560 [2]: https://linuxplumbersconf.org/event/4/contributions/477 [3]: https://lore.kernel.org/r/20190719093538.dhyopljyr5ns33qx@brauner.io [4]: commit 6a21cc50f0c7 ("seccomp: add a return code to trap to userspace") Co-developed-by: Kees Cook <keescook@chromium.org> Signed-off-by: Kees Cook <keescook@chromium.org> Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com> Reviewed-by: Tycho Andersen <tycho@tycho.ws> Cc: Andy Lutomirski <luto@amacapital.net> Cc: Will Drewry <wad@chromium.org> CC: Tyler Hicks <tyhicks@canonical.com> --- /* v1 */ - Kees Cook <keescook@chromium.org>, Tycho Andersen <tycho@tycho.ws>: - s/SECCOMP_RET_USER_NOTIF_ALLOW/SECCOMP_USER_NOTIF_FLAG_CONTINUE/g - Kees Cook <keescook@chromium.org>: - put giant warning about the dangers, and correct usage of the SECCOMP_USER_NOTIF_FLAG_CONTINUE flag - Kees Cook <keescook@chromium.org>: - change return type for seccomp_do_user_notification() to int to align with similar functions /* v0 */ Link: https://lore.kernel.org/r/20190918084833.9369-2-christian.brauner@ubuntu.com --- include/uapi/linux/seccomp.h | 20 ++++++++++++++++++++ kernel/seccomp.c | 28 ++++++++++++++++++++++------ 2 files changed, 42 insertions(+), 6 deletions(-)