[PATCH resend] seccomp: Make duplicate listener detection non-racy

[Jann Horn]

Currently, init_listener() tries to prevent adding a filter with
SECCOMP_FILTER_FLAG_NEW_LISTENER if one of the existing filters already
has a listener. However, this check happens without holding any lock that
would prevent another thread from concurrently installing a new filter
(potentially with a listener) on top of the ones we already have.

Theoretically, this is also a data race: The plain load from
current->seccomp.filter can race with concurrent writes to the same
location.

Fix it by moving the check into the region that holds the siglock to guard
against concurrent TSYNC.

(I am not marking this for stable backport because I believe that this does
not have any implications beyond a theoretical data race and allowing
userspace to create seccomp filters with weird semantics if userspace
concurrently installs seccomp filters in a way no benign userspace workload
would.)

(The "Fixes" tag points to the commit that introduced the theoretical
data race; concurrent installation of another filter with TSYNC only
became possible later, in commit 51891498f2da ("seccomp: allow TSYNC and
USER_NOTIF together").)

Fixes: 6a21cc50f0c7 ("seccomp: add a return code to trap to userspace")
Reviewed-by: Tycho Andersen <tycho@tycho.pizza>
Signed-off-by: Jann Horn <jannh@xxxxxxxxxx>
---
NOTE: After Tycho gave his Reviewed-by, I've had to adjust the errno
to -EBUSY (my original patch broke UAPI, good thing we have selftests),
remove the unused "cur" from init_listener(), and remove the now
useless initialization of "ret".

resending because the first time I mangled the diff... sorry

 kernel/seccomp.c | 38 +++++++++++++++++++++++++++++++-------
 1 file changed, 31 insertions(+), 7 deletions(-)

diff --git a/kernel/seccomp.c b/kernel/seccomp.c
index 676d4af62103..c359ef4380ad 100644
--- a/kernel/seccomp.c
+++ b/kernel/seccomp.c
@@ -1472,13 +1472,7 @@ static const struct file_operations seccomp_notify_ops = {
 
 static struct file *init_listener(struct seccomp_filter *filter)
 {
-	struct file *ret = ERR_PTR(-EBUSY);
-	struct seccomp_filter *cur;
-
-	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
-		if (cur->notif)
-			goto out;
-	}
+	struct file *ret;
 
 	ret = ERR_PTR(-ENOMEM);
 	filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
@@ -1504,6 +1498,31 @@ static struct file *init_listener(struct seccomp_filter *filter)
 	return ret;
 }
 
+/*
+ * Does @new_child have a listener while an ancestor also has a listener?
+ * If so, we'll want to reject this filter.
+ * This only has to be tested for the current process, even in the TSYNC case,
+ * because TSYNC installs @child with the same parent on all threads.
+ * Note that @new_child is not hooked up to its parent at this point yet, so
+ * we use current->seccomp.filter.
+ */
+static bool has_duplicate_listener(struct seccomp_filter *new_child)
+{
+	struct seccomp_filter *cur;
+
+	/* must be protected against concurrent TSYNC */
+	lockdep_assert_held(&current->sighand->siglock);
+
+	if (!new_child->notif)
+		return false;
+	for (cur = current->seccomp.filter; cur; cur = cur->prev) {
+		if (cur->notif)
+			return true;
+	}
+
+	return false;
+}
+
 /**
  * seccomp_set_mode_filter: internal function for setting seccomp filter
  * @flags:  flags to change filter behavior
@@ -1575,6 +1594,11 @@ static long seccomp_set_mode_filter(unsigned int flags,
 	if (!seccomp_may_assign_mode(seccomp_mode))
 		goto out;
 
+	if (has_duplicate_listener(prepared)) {
+		ret = -EBUSY;
+		goto out;
+	}
+
 	ret = seccomp_attach_filter(flags, prepared);
 	if (ret)
 		goto out;

base-commit: fb0155a09b0224a7147cb07a4ce6034c8d29667f
-- 
2.28.0.806.g8561365e88-goog