Skip to contentSkip to conversation

io_uring: don't audit the capability check in io_uring_create()

PATCH Applied Ondrej Mosnacek 3 participants 4 messages Jul 18, 2023 – Jul 25, 2023

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 io_uring: don't audit the capability check in io_uring_create()

The check being unconditional may lead to unwanted denials reported by LSMs when a process has the capability granted by DAC, but denied by an LSM. In the case of SELinux such denials are a problem, since they can't be effectively filtered out via the policy and when not silenced, they produce noise that may hide a true problem or an attack.

Since not having the capability merely means that the created io_uring context will be accounted against the current user's RLIMIT_MEMLOCK limit, we can disable auditing of denials for this check by using ns_capable_noaudit() instead of capable().

Fixes 2b188cc1bb85 ("Add io_uring IO interface")
---
 io_uring/io_uring.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
io_uring/io_uring.c +1 -1
index 7505de2428e03..a9923676d16d6 100644
@@ -3870,7 +3870,7 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,
 		ctx->syscall_iopoll = 1;
 
 	ctx->compat = in_compat_syscall();
-	if (!capable(CAP_IPC_LOCK))
+	if (!ns_capable_noaudit(&init_user_ns, CAP_IPC_LOCK))
 		ctx->user = get_uid(current_user());
 
 	/*
Signature · 1 line
2.41.0
Reviewed-by on v1 1/1
#1 lore raw

Hi, Ondrej,

Quoted 10 lines · Ondrej Mosnacek

The check being unconditional may lead to unwanted denials reported by LSMs when a process has the capability granted by DAC, but denied by an LSM. In the case of SELinux such denials are a problem, since they can't be effectively filtered out via the policy and when not silenced, they produce noise that may hide a true problem or an attack.

Since not having the capability merely means that the created io_uring context will be accounted against the current user's RLIMIT_MEMLOCK limit, we can disable auditing of denials for this check by using ns_capable_noaudit() instead of capable().

Could you add a comment, or add some documentation to ns_capable_noaudit() about when it should be used? It wasn't apparent to me, at least, before this explanation.

Quoted 12 lines
Fixes 2b188cc1bb85 ("Add io_uring IO interface")
---
 io_uring/io_uring.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
io_uring/io_uring.c +0 -0
index 7505de2428e03..a9923676d16d6 100644

io_uring/io_uring.c static __cold int io_uring_create(unsigned entries, struct io_uring_params *p, · line 3876

Quoted 8 lines
io_uring/io_uring.c +1 -1
@@ -3870,7 +3870,7 @@ static __cold int io_uring_create(unsigned entries, struct io_uring_params *p,
 		ctx->syscall_iopoll = 1;
 
 	ctx->compat = in_compat_syscall();
-	if (!capable(CAP_IPC_LOCK))
+	if (!ns_capable_noaudit(&init_user_ns, CAP_IPC_LOCK))
 		ctx->user = get_uid(current_user());
 
 	/*
#2 lore raw
Quoted 11 lines · Jeff Moyer

Hi, Ondrej,

Quoted 6 lines · Ondrej Mosnacek

The check being unconditional may lead to unwanted denials reported by LSMs when a process has the capability granted by DAC, but denied by an LSM. In the case of SELinux such denials are a problem, since they can't be effectively filtered out via the policy and when not silenced, they produce noise that may hide a true problem or an attack.

Quoted 8 lines
Quoted 4 lines

Since not having the capability merely means that the created io_uring context will be accounted against the current user's RLIMIT_MEMLOCK limit, we can disable auditing of denials for this check by using ns_capable_noaudit() instead of capable().

Could you add a comment, or add some documentation to ns_capable_noaudit() about when it should be used? It wasn't apparent to me, at least, before this explanation.

This has been requested before, so I finally forced myself to look into it and only now I realized that there is a subtle difference between the has_capability and capable helpers. As the docstrings say, the former doesn't set the PF_SUPERPRIV on the task when the check succeeds, while the latter does. The problem is that I don't know what the exact implications are and thus I'm not able to document which helper should be used in what situation... It is possible some of the existing call sites use the wrong helper in the noaudit case (possibly including ones that I added/suggested).

The comment at its declaration says "Used super-user privileges" and it seems to be used only to propagate into the ASU flag in task accounting information. But in the case of capability checks that do not fail the syscall it is not easy to tell if "super-user privileges" were "used" or not (or, rather, whether the task should be accounted as such or not after a successful check).

If anyone is reading this and has a better understanding of the PF_SUPERPRIV flag semantics, I'd be thankful for a clarification so that we can sort out this mess :)

--
Ondrej Mosnacek
Senior Software Engineer, Linux Security - SELinux kernel Red Hat, Inc.

maintainer Applied
#3 lore raw
In reply to Ondrej Mosnacek
Quoted 6 lines · Ondrej Mosnacek

The check being unconditional may lead to unwanted denials reported by LSMs when a process has the capability granted by DAC, but denied by an LSM. In the case of SELinux such denials are a problem, since they can't be effectively filtered out via the policy and when not silenced, they produce noise that may hide a true problem or an attack.

Quoted 6 lines

Since not having the capability merely means that the created io_uring context will be accounted against the current user's RLIMIT_MEMLOCK limit, we can disable auditing of denials for this check by using ns_capable_noaudit() instead of capable().

[...]

Applied, thanks!

[1/1] io_uring: don't audit the capability check in io_uring_create()

      commit: 6adc2272aaaf84f34b652cf77f770c6fcc4b8336

Best regards,

Signature · 1 line
Jens Axboe