[PATCH v2 0/4] seccomp: Add SECCOMP_FILTER_FLAG_KILL_PROCESS

[PATCH v2 0/4] seccomp: Add SECCOMP_FILTER_FLAG_KILL_PROCESS

[Kees Cook]

This series is the result of Fabricio and I going around a few times
on possible solutions for finding a way to enhance RET_KILL to kill
the process group. There's a lot of ways this could be done, but I
wanted something that felt cleanest. As it happens, Tyler's recent
patch series for logging improvement also needs to know a litte bit
more during filter runs, and the solution for both is to pass back
the matched filter. This lets us examine it here for RET_KILL and
in the future for logging changes.

The filter passing is patch 1, the new flag for RET_KILL is patch 2.
Some test refactoring is in patch 3 for the RET_DATA ordering, and
patch 4 is the test for the new RET_KILL flag.

Please take a look!

Thanks,

-Kees

v2:
- moved kill_process bool into struct padding gap (tyhicks)
- improved comments/docs in various places for clarify (tyhicks)
- use ASSERT_TRUE() for WIFEXITED and WIFSIGNALLED (tyhicks)
- adding Reviewed-bys from tyhicks

[PATCH v2 1/4] seccomp: Provide matching filter for introspection

[Kees Cook]

Both the upcoming logging improvements and changes to RET_KILL will need
to know which filter a given seccomp return value originated from. In
order to delay logic processing of result until after the seccomp loop,
this adds a single pointer assignment on matches. This will allow both
log and RET_KILL logic to work off the filter rather than doing more
expensive tests inside the time-critical run_filters loop.

Running tight cycles of getpid() with filters attached shows no measurable
difference in speed.

Suggested-by: Tyler Hicks <tyhicks@canonical.com>
Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Tyler Hicks <tyhicks@canonical.com>
---
 kernel/seccomp.c | 18 +++++++++++++++---
 1 file changed, 15 insertions(+), 3 deletions(-)

diff --git a/kernel/seccomp.c b/kernel/seccomp.c
index 98b59b5db90b..1f3347fc2605 100644
--- a/kernel/seccomp.c
+++ b/kernel/seccomp.c
@@ -171,10 +171,14 @@ static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
 /**
  * seccomp_run_filters - evaluates all seccomp filters against @sd
  * @sd: optional seccomp data to be passed to filters
+ * @match: stores struct seccomp_filter that resulted in the return value,
+ *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
+ *         be unchanged.
  *
  * Returns valid seccomp BPF response codes.
  */
-static u32 seccomp_run_filters(const struct seccomp_data *sd)
+static u32 seccomp_run_filters(const struct seccomp_data *sd,
+			       struct seccomp_filter **match)
 {
 	struct seccomp_data sd_local;
 	u32 ret = SECCOMP_RET_ALLOW;
@@ -198,8 +202,10 @@ static u32 seccomp_run_filters(const struct seccomp_data *sd)
 	for (; f; f = f->prev) {
 		u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
 
-		if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
+		if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) {
 			ret = cur_ret;
+			*match = f;
+		}
 	}
 	return ret;
 }
@@ -566,6 +572,7 @@ static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 			    const bool recheck_after_trace)
 {
 	u32 filter_ret, action;
+	struct seccomp_filter *match = NULL;
 	int data;
 
 	/*
@@ -574,7 +581,7 @@ static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 	 */
 	rmb();
 
-	filter_ret = seccomp_run_filters(sd);
+	filter_ret = seccomp_run_filters(sd, &match);
 	data = filter_ret & SECCOMP_RET_DATA;
 	action = filter_ret & SECCOMP_RET_ACTION;
 
@@ -638,6 +645,11 @@ static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 		return 0;
 
 	case SECCOMP_RET_ALLOW:
+		/*
+		 * Note that the "match" filter will always be NULL for
+		 * this action since SECCOMP_RET_ALLOW is the starting
+		 * state in seccomp_run_filters().
+		 */
 		return 0;
 
 	case SECCOMP_RET_KILL:
-- 
2.7.4

[PATCH v2 2/4] seccomp: Add SECCOMP_FILTER_FLAG_KILL_PROCESS

[Kees Cook]

Right now, SECCOMP_RET_KILL kills the current thread. There have been
a few requests for RET_KILL to kill the entire process (the thread
group), but since seccomp's u32 return values are ABI, and ordered by
lowest value, with RET_KILL as 0, there isn't a trivial way to provide
an even smaller value that would mean the more restrictive action of
killing the thread group.

Instead, create a filter flag that indicates that a RET_KILL from this
filter must kill the process rather than the thread. This can be set
(and not cleared) via the new SECCOMP_FILTER_FLAG_KILL_PROCESS flag.

Pros:
 - the logic for the filter action is contained in the filter.
 - userspace can detect support for the feature since earlier kernels
   will reject the new flag.
Cons:
 - depends on adding an assignment to the seccomp_run_filters() loop
   (previous patch).

Alternatives to this approach with pros/cons:

- Use a new test during seccomp_run_filters() that treats the RET_DATA
  mask of a RET_KILL action as special. If a new bit is set in the data,
  then treat the return value as -1 (lower than 0).
  Pros:
   - the logic for the filter action is contained in the filter.
  Cons:
   - added complexity to time-sensitive seccomp_run_filters() loop.
   - there isn't a trivial way for userspace to detect if the kernel
     supports the feature (earlier kernels will silently ignore the
     RET_DATA and only kill the thread).

- Have SECCOMP_FILTER_FLAG_KILL_PROCESS attach to the seccomp struct
  rather than the filter.
  Pros:
   - no change needed to seccomp_run_filters() loop.
  Cons:
   - the change in behavior technically originates external to the
     filter, which allows for later filters to "enhance" a previously
     applied filter's RET_KILL to kill the entire process, which may
     be unexpected.

Signed-off-by: Kees Cook <keescook@chromium.org>
---
 include/linux/seccomp.h      |  3 ++-
 include/uapi/linux/seccomp.h |  3 ++-
 kernel/seccomp.c             | 22 +++++++++++++++++++++-
 3 files changed, 25 insertions(+), 3 deletions(-)

diff --git a/include/linux/seccomp.h b/include/linux/seccomp.h
index ecc296c137cd..59d001ba655c 100644
--- a/include/linux/seccomp.h
+++ b/include/linux/seccomp.h
@@ -3,7 +3,8 @@
 
 #include <uapi/linux/seccomp.h>
 
-#define SECCOMP_FILTER_FLAG_MASK	(SECCOMP_FILTER_FLAG_TSYNC)
+#define SECCOMP_FILTER_FLAG_MASK	(SECCOMP_FILTER_FLAG_TSYNC | \
+					 SECCOMP_FILTER_FLAG_KILL_PROCESS)
 
 #ifdef CONFIG_SECCOMP
 
diff --git a/include/uapi/linux/seccomp.h b/include/uapi/linux/seccomp.h
index 0f238a43ff1e..4b75d8c297b6 100644
--- a/include/uapi/linux/seccomp.h
+++ b/include/uapi/linux/seccomp.h
@@ -15,7 +15,8 @@
 #define SECCOMP_SET_MODE_FILTER	1
 
 /* Valid flags for SECCOMP_SET_MODE_FILTER */
-#define SECCOMP_FILTER_FLAG_TSYNC	1
+#define SECCOMP_FILTER_FLAG_TSYNC		1
+#define SECCOMP_FILTER_FLAG_KILL_PROCESS	2
 
 /*
  * All BPF programs must return a 32-bit value.
diff --git a/kernel/seccomp.c b/kernel/seccomp.c
index 1f3347fc2605..297f8bfc3b72 100644
--- a/kernel/seccomp.c
+++ b/kernel/seccomp.c
@@ -44,6 +44,7 @@
  *         is only needed for handling filters shared across tasks.
  * @prev: points to a previously installed, or inherited, filter
  * @prog: the BPF program to evaluate
+ * @kill_process: if true, RET_KILL will kill process rather than thread.
  *
  * seccomp_filter objects are organized in a tree linked via the @prev
  * pointer.  For any task, it appears to be a singly-linked list starting
@@ -57,6 +58,7 @@
  */
 struct seccomp_filter {
 	refcount_t usage;
+	bool kill_process;
 	struct seccomp_filter *prev;
 	struct bpf_prog *prog;
 };
@@ -450,6 +452,10 @@ static long seccomp_attach_filter(unsigned int flags,
 			return ret;
 	}
 
+	/* Set process-killing flag, if present. */
+	if (flags & SECCOMP_FILTER_FLAG_KILL_PROCESS)
+		filter->kill_process = true;
+
 	/*
 	 * If there is an existing filter, make it the prev and don't drop its
 	 * task reference.
@@ -665,7 +671,21 @@ static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 			seccomp_init_siginfo(&info, this_syscall, data);
 			do_coredump(&info);
 		}
-		do_exit(SIGSYS);
+		/*
+		 * The only way match can be NULL here is if something
+		 * went very wrong in seccomp_run_filters() (e.g. a NULL
+		 * filter list in struct seccomp) and the return action
+		 * falls back to failing closed. In this case, take the
+		 * strongest possible action.
+		 *
+		 * If we get here with match->kill_process set, we need
+		 * to kill the entire thread group. Otherwise, kill only
+		 * the offending thread.
+		 */
+		if (!match || match->kill_process)
+			do_group_exit(SIGSYS);
+		else
+			do_exit(SIGSYS);
 	}
 
 	unreachable();
-- 
2.7.4

[PATCH v2 3/4] selftests/seccomp: Refactor RET_ERRNO tests

[Kees Cook]

This refactors the errno tests (since they all use the same pattern for
their filter) and adds a RET_DATA field ordering test.

Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Tyler Hicks <tyhicks@canonical.com>
---
 tools/testing/selftests/seccomp/seccomp_bpf.c | 95 ++++++++++++++++-----------
 1 file changed, 58 insertions(+), 37 deletions(-)

diff --git a/tools/testing/selftests/seccomp/seccomp_bpf.c b/tools/testing/selftests/seccomp/seccomp_bpf.c
index 73f5ea6778ce..ee78a53da5d1 100644
--- a/tools/testing/selftests/seccomp/seccomp_bpf.c
+++ b/tools/testing/selftests/seccomp/seccomp_bpf.c
@@ -136,7 +136,7 @@ TEST(no_new_privs_support)
 	}
 }
 
-/* Tests kernel support by checking for a copy_from_user() fault on * NULL. */
+/* Tests kernel support by checking for a copy_from_user() fault on NULL. */
 TEST(mode_filter_support)
 {
 	long ret;
@@ -541,26 +541,30 @@ TEST(arg_out_of_range)
 	EXPECT_EQ(EINVAL, errno);
 }
 
+#define ERRNO_FILTER(name, errno)					\
+	struct sock_filter _read_filter_##name[] = {			\
+		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,				\
+			offsetof(struct seccomp_data, nr)),		\
+		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),	\
+		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | errno),	\
+		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),		\
+	};								\
+	struct sock_fprog prog_##name = {				\
+		.len = (unsigned short)ARRAY_SIZE(_read_filter_##name),	\
+		.filter = _read_filter_##name,				\
+	}
+
+/* Make sure basic errno values are correctly passed through a filter. */
 TEST(ERRNO_valid)
 {
-	struct sock_filter filter[] = {
-		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
-			offsetof(struct seccomp_data, nr)),
-		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
-		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | E2BIG),
-		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
-	};
-	struct sock_fprog prog = {
-		.len = (unsigned short)ARRAY_SIZE(filter),
-		.filter = filter,
-	};
+	ERRNO_FILTER(valid, E2BIG);
 	long ret;
 	pid_t parent = getppid();
 
 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 	ASSERT_EQ(0, ret);
 
-	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
+	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_valid);
 	ASSERT_EQ(0, ret);
 
 	EXPECT_EQ(parent, syscall(__NR_getppid));
@@ -568,26 +572,17 @@ TEST(ERRNO_valid)
 	EXPECT_EQ(E2BIG, errno);
 }
 
+/* Make sure an errno of zero is correctly handled by the arch code. */
 TEST(ERRNO_zero)
 {
-	struct sock_filter filter[] = {
-		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
-			offsetof(struct seccomp_data, nr)),
-		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
-		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 0),
-		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
-	};
-	struct sock_fprog prog = {
-		.len = (unsigned short)ARRAY_SIZE(filter),
-		.filter = filter,
-	};
+	ERRNO_FILTER(zero, 0);
 	long ret;
 	pid_t parent = getppid();
 
 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 	ASSERT_EQ(0, ret);
 
-	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
+	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_zero);
 	ASSERT_EQ(0, ret);
 
 	EXPECT_EQ(parent, syscall(__NR_getppid));
@@ -595,26 +590,21 @@ TEST(ERRNO_zero)
 	EXPECT_EQ(0, read(0, NULL, 0));
 }
 
+/*
+ * The SECCOMP_RET_DATA mask is 16 bits wide, but errno is smaller.
+ * This tests that the errno value gets capped correctly, fixed by
+ * 580c57f10768 ("seccomp: cap SECCOMP_RET_ERRNO data to MAX_ERRNO").
+ */
 TEST(ERRNO_capped)
 {
-	struct sock_filter filter[] = {
-		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
-			offsetof(struct seccomp_data, nr)),
-		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_read, 0, 1),
-		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ERRNO | 4096),
-		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
-	};
-	struct sock_fprog prog = {
-		.len = (unsigned short)ARRAY_SIZE(filter),
-		.filter = filter,
-	};
+	ERRNO_FILTER(capped, 4096);
 	long ret;
 	pid_t parent = getppid();
 
 	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
 	ASSERT_EQ(0, ret);
 
-	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog);
+	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_capped);
 	ASSERT_EQ(0, ret);
 
 	EXPECT_EQ(parent, syscall(__NR_getppid));
@@ -622,6 +612,37 @@ TEST(ERRNO_capped)
 	EXPECT_EQ(4095, errno);
 }
 
+/*
+ * Filters are processed in reverse order: last applied is executed first.
+ * Since only the SECCOMP_RET_ACTION mask is tested for return values, the
+ * SECCOMP_RET_DATA mask results will follow the most recently applied
+ * matching filter return (and not the lowest or highest value).
+ */
+TEST(ERRNO_order)
+{
+	ERRNO_FILTER(first,  11);
+	ERRNO_FILTER(second, 13);
+	ERRNO_FILTER(third,  12);
+	long ret;
+	pid_t parent = getppid();
+
+	ret = prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);
+	ASSERT_EQ(0, ret);
+
+	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_first);
+	ASSERT_EQ(0, ret);
+
+	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_second);
+	ASSERT_EQ(0, ret);
+
+	ret = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog_third);
+	ASSERT_EQ(0, ret);
+
+	EXPECT_EQ(parent, syscall(__NR_getppid));
+	EXPECT_EQ(-1, read(0, NULL, 0));
+	EXPECT_EQ(12, errno);
+}
+
 FIXTURE_DATA(TRAP) {
 	struct sock_fprog prog;
 };
-- 
2.7.4

[PATCH v2 4/4] selftests/seccomp: Test thread vs process killing

[Kees Cook]

SECCOMP_RET_KILL is supposed to kill the current thread (and userspace
depends on this), so test for this, distinct from killing the entire
process. This also tests killing the entire process with the new
SECCOMP_FILTER_FLAG_KILL_PROCESS flag. (This also moves a bunch of
defines up earlier in the file to use them earlier.)

Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Tyler Hicks <tyhicks@canonical.com>
---
 tools/testing/selftests/seccomp/seccomp_bpf.c | 185 ++++++++++++++++++++------
 1 file changed, 144 insertions(+), 41 deletions(-)

diff --git a/tools/testing/selftests/seccomp/seccomp_bpf.c b/tools/testing/selftests/seccomp/seccomp_bpf.c
index ee78a53da5d1..d0a9bebf21f3 100644
--- a/tools/testing/selftests/seccomp/seccomp_bpf.c
+++ b/tools/testing/selftests/seccomp/seccomp_bpf.c
@@ -87,6 +87,51 @@ struct seccomp_data {
 };
 #endif
 
+#ifndef __NR_seccomp
+# if defined(__i386__)
+#  define __NR_seccomp 354
+# elif defined(__x86_64__)
+#  define __NR_seccomp 317
+# elif defined(__arm__)
+#  define __NR_seccomp 383
+# elif defined(__aarch64__)
+#  define __NR_seccomp 277
+# elif defined(__hppa__)
+#  define __NR_seccomp 338
+# elif defined(__powerpc__)
+#  define __NR_seccomp 358
+# elif defined(__s390__)
+#  define __NR_seccomp 348
+# else
+#  warning "seccomp syscall number unknown for this architecture"
+#  define __NR_seccomp 0xffff
+# endif
+#endif
+
+#ifndef SECCOMP_SET_MODE_STRICT
+#define SECCOMP_SET_MODE_STRICT 0
+#endif
+
+#ifndef SECCOMP_SET_MODE_FILTER
+#define SECCOMP_SET_MODE_FILTER 1
+#endif
+
+#ifndef SECCOMP_FILTER_FLAG_TSYNC
+#define SECCOMP_FILTER_FLAG_TSYNC 1
+#endif
+
+#ifndef SECCOMP_FILTER_FLAG_KILL_PROCESS
+#define SECCOMP_FILTER_FLAG_KILL_PROCESS 2
+#endif
+
+#ifndef seccomp
+int seccomp(unsigned int op, unsigned int flags, void *args)
+{
+	errno = 0;
+	return syscall(__NR_seccomp, op, flags, args);
+}
+#endif
+
 #if __BYTE_ORDER == __LITTLE_ENDIAN
 #define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
 #elif __BYTE_ORDER == __BIG_ENDIAN
@@ -520,6 +565,105 @@ TEST_SIGNAL(KILL_one_arg_six, SIGSYS)
 	close(fd);
 }
 
+/* This is a thread task to die via seccomp filter violation. */
+void *kill_thread(void *data)
+{
+	bool die = (bool)data;
+
+	if (die) {
+		prctl(PR_GET_SECCOMP, 0, 0, 0, 0);
+		return (void *)SIBLING_EXIT_FAILURE;
+	}
+
+	return (void *)SIBLING_EXIT_UNKILLED;
+}
+
+/* Prepare a thread that will kill itself or both of us. */
+void kill_thread_or_group(struct __test_metadata *_metadata, bool kill_process)
+{
+	pthread_t thread;
+	void *status;
+	unsigned int flags;
+	/* Kill only when calling __NR_prctl. */
+	struct sock_filter filter[] = {
+		BPF_STMT(BPF_LD|BPF_W|BPF_ABS,
+			offsetof(struct seccomp_data, nr)),
+		BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, __NR_prctl, 0, 1),
+		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_KILL),
+		BPF_STMT(BPF_RET|BPF_K, SECCOMP_RET_ALLOW),
+	};
+	struct sock_fprog prog = {
+		.len = (unsigned short)ARRAY_SIZE(filter),
+		.filter = filter,
+	};
+
+	ASSERT_EQ(0, prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
+		TH_LOG("Kernel does not support PR_SET_NO_NEW_PRIVS!");
+	}
+
+	flags = kill_process ? SECCOMP_FILTER_FLAG_KILL_PROCESS : 0;
+	ASSERT_EQ(0, seccomp(SECCOMP_SET_MODE_FILTER, flags, &prog)) {
+		if (kill_process)
+			TH_LOG("Kernel does not support SECCOMP_FILTER_FLAG_KILL_PROCESS");
+		else
+			TH_LOG("Kernel does not support seccomp syscall");
+	}
+
+	/* Start a thread that will exit immediately. */
+	ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)false));
+	ASSERT_EQ(0, pthread_join(thread, &status));
+	ASSERT_EQ(SIBLING_EXIT_UNKILLED, (unsigned long)status);
+
+	/* Start a thread that will die immediately. */
+	ASSERT_EQ(0, pthread_create(&thread, NULL, kill_thread, (void *)true));
+	ASSERT_EQ(0, pthread_join(thread, &status));
+	ASSERT_NE(SIBLING_EXIT_FAILURE, (unsigned long)status);
+
+	/*
+	 * If we get here, only the spawned thread died. Let the parent know
+	 * this entire process (all threads including this one) didn't die.
+	 */
+	exit(42);
+}
+
+TEST(KILL_thread)
+{
+	int status;
+	pid_t child_pid;
+
+	child_pid = fork();
+	ASSERT_LE(0, child_pid);
+	if (child_pid == 0) {
+		kill_thread_or_group(_metadata, false);
+		_exit(38);
+	}
+
+	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
+
+	/* If only the thread was killed, we'll see exit 42. */
+	ASSERT_TRUE(WIFEXITED(status));
+	ASSERT_EQ(42, WEXITSTATUS(status));
+}
+
+TEST(KILL_process)
+{
+	int status;
+	pid_t child_pid;
+
+	child_pid = fork();
+	ASSERT_LE(0, child_pid);
+	if (child_pid == 0) {
+		kill_thread_or_group(_metadata, true);
+		_exit(38);
+	}
+
+	ASSERT_EQ(child_pid, waitpid(child_pid, &status, 0));
+
+	/* If the entire process was killed, we'll see SIGSYS. */
+	ASSERT_TRUE(WIFSIGNALED(status));
+	ASSERT_EQ(SIGSYS, WTERMSIG(status));
+}
+
 /* TODO(wad) add 64-bit versus 32-bit arg tests. */
 TEST(arg_out_of_range)
 {
@@ -1675,47 +1819,6 @@ TEST_F_SIGNAL(TRACE_syscall, kill_after_ptrace, SIGSYS)
 	EXPECT_NE(self->mypid, syscall(__NR_getpid));
 }
 
-#ifndef __NR_seccomp
-# if defined(__i386__)
-#  define __NR_seccomp 354
-# elif defined(__x86_64__)
-#  define __NR_seccomp 317
-# elif defined(__arm__)
-#  define __NR_seccomp 383
-# elif defined(__aarch64__)
-#  define __NR_seccomp 277
-# elif defined(__hppa__)
-#  define __NR_seccomp 338
-# elif defined(__powerpc__)
-#  define __NR_seccomp 358
-# elif defined(__s390__)
-#  define __NR_seccomp 348
-# else
-#  warning "seccomp syscall number unknown for this architecture"
-#  define __NR_seccomp 0xffff
-# endif
-#endif
-
-#ifndef SECCOMP_SET_MODE_STRICT
-#define SECCOMP_SET_MODE_STRICT 0
-#endif
-
-#ifndef SECCOMP_SET_MODE_FILTER
-#define SECCOMP_SET_MODE_FILTER 1
-#endif
-
-#ifndef SECCOMP_FILTER_FLAG_TSYNC
-#define SECCOMP_FILTER_FLAG_TSYNC 1
-#endif
-
-#ifndef seccomp
-int seccomp(unsigned int op, unsigned int flags, void *args)
-{
-	errno = 0;
-	return syscall(__NR_seccomp, op, flags, args);
-}
-#endif
-
 TEST(seccomp_syscall)
 {
 	struct sock_filter filter[] = {
-- 
2.7.4

Re: [PATCH v2 2/4] seccomp: Add SECCOMP_FILTER_FLAG_KILL_PROCESS

[Tyler Hicks]

[-- Attachment #1.1: Type: text/plain, Size: 5261 bytes --]

On 08/07/2017 08:59 PM, Kees Cook wrote:
> Right now, SECCOMP_RET_KILL kills the current thread. There have been
> a few requests for RET_KILL to kill the entire process (the thread
> group), but since seccomp's u32 return values are ABI, and ordered by
> lowest value, with RET_KILL as 0, there isn't a trivial way to provide
> an even smaller value that would mean the more restrictive action of
> killing the thread group.
> 
> Instead, create a filter flag that indicates that a RET_KILL from this
> filter must kill the process rather than the thread. This can be set
> (and not cleared) via the new SECCOMP_FILTER_FLAG_KILL_PROCESS flag.
> 
> Pros:
>  - the logic for the filter action is contained in the filter.
>  - userspace can detect support for the feature since earlier kernels
>    will reject the new flag.
> Cons:
>  - depends on adding an assignment to the seccomp_run_filters() loop
>    (previous patch).
> 
> Alternatives to this approach with pros/cons:
> 
> - Use a new test during seccomp_run_filters() that treats the RET_DATA
>   mask of a RET_KILL action as special. If a new bit is set in the data,
>   then treat the return value as -1 (lower than 0).
>   Pros:
>    - the logic for the filter action is contained in the filter.
>   Cons:
>    - added complexity to time-sensitive seccomp_run_filters() loop.
>    - there isn't a trivial way for userspace to detect if the kernel
>      supports the feature (earlier kernels will silently ignore the
>      RET_DATA and only kill the thread).
> 
> - Have SECCOMP_FILTER_FLAG_KILL_PROCESS attach to the seccomp struct
>   rather than the filter.
>   Pros:
>    - no change needed to seccomp_run_filters() loop.
>   Cons:
>    - the change in behavior technically originates external to the
>      filter, which allows for later filters to "enhance" a previously
>      applied filter's RET_KILL to kill the entire process, which may
>      be unexpected.
> 
> Signed-off-by: Kees Cook <keescook@chromium.org>

v2 of these patches all look good to me.

Reviewed-by: Tyler Hicks <tyhicks@canonical.com>

Thanks!

Tyler

> ---
>  include/linux/seccomp.h      |  3 ++-
>  include/uapi/linux/seccomp.h |  3 ++-
>  kernel/seccomp.c             | 22 +++++++++++++++++++++-
>  3 files changed, 25 insertions(+), 3 deletions(-)
> 
> diff --git a/include/linux/seccomp.h b/include/linux/seccomp.h
> index ecc296c137cd..59d001ba655c 100644
> --- a/include/linux/seccomp.h
> +++ b/include/linux/seccomp.h
> @@ -3,7 +3,8 @@
>  
>  #include <uapi/linux/seccomp.h>
>  
> -#define SECCOMP_FILTER_FLAG_MASK	(SECCOMP_FILTER_FLAG_TSYNC)
> +#define SECCOMP_FILTER_FLAG_MASK	(SECCOMP_FILTER_FLAG_TSYNC | \
> +					 SECCOMP_FILTER_FLAG_KILL_PROCESS)
>  
>  #ifdef CONFIG_SECCOMP
>  
> diff --git a/include/uapi/linux/seccomp.h b/include/uapi/linux/seccomp.h
> index 0f238a43ff1e..4b75d8c297b6 100644
> --- a/include/uapi/linux/seccomp.h
> +++ b/include/uapi/linux/seccomp.h
> @@ -15,7 +15,8 @@
>  #define SECCOMP_SET_MODE_FILTER	1
>  
>  /* Valid flags for SECCOMP_SET_MODE_FILTER */
> -#define SECCOMP_FILTER_FLAG_TSYNC	1
> +#define SECCOMP_FILTER_FLAG_TSYNC		1
> +#define SECCOMP_FILTER_FLAG_KILL_PROCESS	2
>  
>  /*
>   * All BPF programs must return a 32-bit value.
> diff --git a/kernel/seccomp.c b/kernel/seccomp.c
> index 1f3347fc2605..297f8bfc3b72 100644
> --- a/kernel/seccomp.c
> +++ b/kernel/seccomp.c
> @@ -44,6 +44,7 @@
>   *         is only needed for handling filters shared across tasks.
>   * @prev: points to a previously installed, or inherited, filter
>   * @prog: the BPF program to evaluate
> + * @kill_process: if true, RET_KILL will kill process rather than thread.
>   *
>   * seccomp_filter objects are organized in a tree linked via the @prev
>   * pointer.  For any task, it appears to be a singly-linked list starting
> @@ -57,6 +58,7 @@
>   */
>  struct seccomp_filter {
>  	refcount_t usage;
> +	bool kill_process;
>  	struct seccomp_filter *prev;
>  	struct bpf_prog *prog;
>  };
> @@ -450,6 +452,10 @@ static long seccomp_attach_filter(unsigned int flags,
>  			return ret;
>  	}
>  
> +	/* Set process-killing flag, if present. */
> +	if (flags & SECCOMP_FILTER_FLAG_KILL_PROCESS)
> +		filter->kill_process = true;
> +
>  	/*
>  	 * If there is an existing filter, make it the prev and don't drop its
>  	 * task reference.
> @@ -665,7 +671,21 @@ static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
>  			seccomp_init_siginfo(&info, this_syscall, data);
>  			do_coredump(&info);
>  		}
> -		do_exit(SIGSYS);
> +		/*
> +		 * The only way match can be NULL here is if something
> +		 * went very wrong in seccomp_run_filters() (e.g. a NULL
> +		 * filter list in struct seccomp) and the return action
> +		 * falls back to failing closed. In this case, take the
> +		 * strongest possible action.
> +		 *
> +		 * If we get here with match->kill_process set, we need
> +		 * to kill the entire thread group. Otherwise, kill only
> +		 * the offending thread.
> +		 */
> +		if (!match || match->kill_process)
> +			do_group_exit(SIGSYS);
> +		else
> +			do_exit(SIGSYS);
>  	}
>  
>  	unreachable();
> 



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 801 bytes --]