Merge branch 'fix-bpf-multi-uprobe-pid-filtering-logic'

Andrii Nakryiko says:

====================
Fix BPF multi-uprobe PID filtering logic

It turns out that current implementation of multi-uprobe PID filtering logic
is broken. It filters by thread, while the promise is filtering by process.
Patch #1 fixes the logic trivially. The rest is testing and mitigations that
are necessary for libbpf to not break users of USDT programs.

v1->v2:
  - fix selftest in last patch (CI);
  - use semicolon in patch #3 (Jiri).
====================

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Alexei Starovoitov <[email protected]>

4 files changed, 206 insertions, 19 deletions

diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
index f5154c051d2c..6249dac61701 100644
--- a/kernel/trace/bpf_trace.c
+++ b/kernel/trace/bpf_trace.c
@@ -3295,7 +3295,7 @@ static int uprobe_prog_run(struct bpf_uprobe *uprobe,
 	struct bpf_run_ctx *old_run_ctx;
 	int err = 0;
 
-	if (link->task && current != link->task)
+	if (link->task && current->mm != link->task->mm)
 		return 0;
 
 	if (sleepable)
@@ -3396,8 +3396,9 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr
 	upath = u64_to_user_ptr(attr->link_create.uprobe_multi.path);
 	uoffsets = u64_to_user_ptr(attr->link_create.uprobe_multi.offsets);
 	cnt = attr->link_create.uprobe_multi.cnt;
+	pid = attr->link_create.uprobe_multi.pid;
 
-	if (!upath || !uoffsets || !cnt)
+	if (!upath || !uoffsets || !cnt || pid < 0)
 		return -EINVAL;
 	if (cnt > MAX_UPROBE_MULTI_CNT)
 		return -E2BIG;
@@ -3421,11 +3422,8 @@ int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *pr
 		goto error_path_put;
 	}
 
-	pid = attr->link_create.uprobe_multi.pid;
 	if (pid) {
-		rcu_read_lock();
-		task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
-		rcu_read_unlock();
+		task = get_pid_task(find_vpid(pid), PIDTYPE_TGID);
 		if (!task) {
 			err = -ESRCH;
 			goto error_path_put;
diff --git a/tools/lib/bpf/features.c b/tools/lib/bpf/features.c
index a336786a22a3..3df0125ed5fa 100644
--- a/tools/lib/bpf/features.c
+++ b/tools/lib/bpf/features.c
@@ -392,11 +392,40 @@ static int probe_uprobe_multi_link(int token_fd)
 	link_fd = bpf_link_create(prog_fd, -1, BPF_TRACE_UPROBE_MULTI, &link_opts);
 	err = -errno; /* close() can clobber errno */
 
+	if (link_fd >= 0 || err != -EBADF) {
+		close(link_fd);
+		close(prog_fd);
+		return 0;
+	}
+
+	/* Initial multi-uprobe support in kernel didn't handle PID filtering
+	 * correctly (it was doing thread filtering, not process filtering).
+	 * So now we'll detect if PID filtering logic was fixed, and, if not,
+	 * we'll pretend multi-uprobes are not supported, if not.
+	 * Multi-uprobes are used in USDT attachment logic, and we need to be
+	 * conservative here, because multi-uprobe selection happens early at
+	 * load time, while the use of PID filtering is known late at
+	 * attachment time, at which point it's too late to undo multi-uprobe
+	 * selection.
+	 *
+	 * Creating uprobe with pid == -1 for (invalid) '/' binary will fail
+	 * early with -EINVAL on kernels with fixed PID filtering logic;
+	 * otherwise -ESRCH would be returned if passed correct binary path
+	 * (but we'll just get -BADF, of course).
+	 */
+	link_opts.uprobe_multi.pid = -1; /* invalid PID */
+	link_opts.uprobe_multi.path = "/"; /* invalid path */
+	link_opts.uprobe_multi.offsets = &offset;
+	link_opts.uprobe_multi.cnt = 1;
+
+	link_fd = bpf_link_create(prog_fd, -1, BPF_TRACE_UPROBE_MULTI, &link_opts);
+	err = -errno; /* close() can clobber errno */
+
 	if (link_fd >= 0)
 		close(link_fd);
 	close(prog_fd);
 
-	return link_fd < 0 && err == -EBADF;
+	return link_fd < 0 && err == -EINVAL;
 }
 
 static int probe_kern_bpf_cookie(int token_fd)
diff --git a/tools/testing/selftests/bpf/prog_tests/uprobe_multi_test.c b/tools/testing/selftests/bpf/prog_tests/uprobe_multi_test.c
index 8269cdee33ae..bf6ca8e3eb13 100644
--- a/tools/testing/selftests/bpf/prog_tests/uprobe_multi_test.c
+++ b/tools/testing/selftests/bpf/prog_tests/uprobe_multi_test.c
@@ -1,12 +1,14 @@
 // SPDX-License-Identifier: GPL-2.0
 
 #include <unistd.h>
+#include <pthread.h>
 #include <test_progs.h>
 #include "uprobe_multi.skel.h"
 #include "uprobe_multi_bench.skel.h"
 #include "uprobe_multi_usdt.skel.h"
 #include "bpf/libbpf_internal.h"
 #include "testing_helpers.h"
+#include "../sdt.h"
 
 static char test_data[] = "test_data";
 
@@ -25,9 +27,17 @@ noinline void uprobe_multi_func_3(void)
 	asm volatile ("");
 }
 
+noinline void usdt_trigger(void)
+{
+	STAP_PROBE(test, pid_filter_usdt);
+}
+
 struct child {
 	int go[2];
+	int c2p[2]; /* child -> parent channel */
 	int pid;
+	int tid;
+	pthread_t thread;
 };
 
 static void release_child(struct child *child)
@@ -38,6 +48,10 @@ static void release_child(struct child *child)
 		return;
 	close(child->go[1]);
 	close(child->go[0]);
+	if (child->thread)
+		pthread_join(child->thread, NULL);
+	close(child->c2p[0]);
+	close(child->c2p[1]);
 	if (child->pid > 0)
 		waitpid(child->pid, &child_status, 0);
 }
@@ -63,7 +77,7 @@ static struct child *spawn_child(void)
 	if (pipe(child.go))
 		return NULL;
 
-	child.pid = fork();
+	child.pid = child.tid = fork();
 	if (child.pid < 0) {
 		release_child(&child);
 		errno = EINVAL;
@@ -82,6 +96,7 @@ static struct child *spawn_child(void)
 		uprobe_multi_func_1();
 		uprobe_multi_func_2();
 		uprobe_multi_func_3();
+		usdt_trigger();
 
 		exit(errno);
 	}
@@ -89,6 +104,67 @@ static struct child *spawn_child(void)
 	return &child;
 }
 
+static void *child_thread(void *ctx)
+{
+	struct child *child = ctx;
+	int c = 0, err;
+
+	child->tid = syscall(SYS_gettid);
+
+	/* let parent know we are ready */
+	err = write(child->c2p[1], &c, 1);
+	if (err != 1)
+		pthread_exit(&err);
+
+	/* wait for parent's kick */
+	err = read(child->go[0], &c, 1);
+	if (err != 1)
+		pthread_exit(&err);
+
+	uprobe_multi_func_1();
+	uprobe_multi_func_2();
+	uprobe_multi_func_3();
+	usdt_trigger();
+
+	err = 0;
+	pthread_exit(&err);
+}
+
+static struct child *spawn_thread(void)
+{
+	static struct child child;
+	int c, err;
+
+	/* pipe to notify child to execute the trigger functions */
+	if (pipe(child.go))
+		return NULL;
+	/* pipe to notify parent that child thread is ready */
+	if (pipe(child.c2p)) {
+		close(child.go[0]);
+		close(child.go[1]);
+		return NULL;
+	}
+
+	child.pid = getpid();
+
+	err = pthread_create(&child.thread, NULL, child_thread, &child);
+	if (err) {
+		err = -errno;
+		close(child.go[0]);
+		close(child.go[1]);
+		close(child.c2p[0]);
+		close(child.c2p[1]);
+		errno = -err;
+		return NULL;
+	}
+
+	err = read(child.c2p[0], &c, 1);
+	if (!ASSERT_EQ(err, 1, "child_thread_ready"))
+		return NULL;
+
+	return &child;
+}
+
 static void uprobe_multi_test_run(struct uprobe_multi *skel, struct child *child)
 {
 	skel->bss->uprobe_multi_func_1_addr = (__u64) uprobe_multi_func_1;
@@ -103,15 +179,23 @@ static void uprobe_multi_test_run(struct uprobe_multi *skel, struct child *child
 	 * passed at the probe attach.
 	 */
 	skel->bss->pid = child ? 0 : getpid();
+	skel->bss->expect_pid = child ? child->pid : 0;
+
+	/* trigger all probes, if we are testing child *process*, just to make
+	 * sure that PID filtering doesn't let through activations from wrong
+	 * PIDs; when we test child *thread*, we don't want to do this to
+	 * avoid double counting number of triggering events
+	 */
+	if (!child || !child->thread) {
+		uprobe_multi_func_1();
+		uprobe_multi_func_2();
+		uprobe_multi_func_3();
+		usdt_trigger();
+	}
 
 	if (child)
 		kick_child(child);
 
-	/* trigger all probes */
-	uprobe_multi_func_1();
-	uprobe_multi_func_2();
-	uprobe_multi_func_3();
-
 	/*
 	 * There are 2 entry and 2 exit probe called for each uprobe_multi_func_[123]
 	 * function and each slepable probe (6) increments uprobe_multi_sleep_result.
@@ -126,8 +210,12 @@ static void uprobe_multi_test_run(struct uprobe_multi *skel, struct child *child
 
 	ASSERT_EQ(skel->bss->uprobe_multi_sleep_result, 6, "uprobe_multi_sleep_result");
 
-	if (child)
+	ASSERT_FALSE(skel->bss->bad_pid_seen, "bad_pid_seen");
+
+	if (child) {
 		ASSERT_EQ(skel->bss->child_pid, child->pid, "uprobe_multi_child_pid");
+		ASSERT_EQ(skel->bss->child_tid, child->tid, "uprobe_multi_child_tid");
+	}
 }
 
 static void test_skel_api(void)
@@ -190,8 +278,24 @@ __test_attach_api(const char *binary, const char *pattern, struct bpf_uprobe_mul
 	if (!ASSERT_OK_PTR(skel->links.uprobe_extra, "bpf_program__attach_uprobe_multi"))
 		goto cleanup;
 
+	/* Attach (uprobe-backed) USDTs */
+	skel->links.usdt_pid = bpf_program__attach_usdt(skel->progs.usdt_pid, pid, binary,
+							"test", "pid_filter_usdt", NULL);
+	if (!ASSERT_OK_PTR(skel->links.usdt_pid, "attach_usdt_pid"))
+		goto cleanup;
+
+	skel->links.usdt_extra = bpf_program__attach_usdt(skel->progs.usdt_extra, -1, binary,
+							  "test", "pid_filter_usdt", NULL);
+	if (!ASSERT_OK_PTR(skel->links.usdt_extra, "attach_usdt_extra"))
+		goto cleanup;
+
 	uprobe_multi_test_run(skel, child);
 
+	ASSERT_FALSE(skel->bss->bad_pid_seen_usdt, "bad_pid_seen_usdt");
+	if (child) {
+		ASSERT_EQ(skel->bss->child_pid_usdt, child->pid, "usdt_multi_child_pid");
+		ASSERT_EQ(skel->bss->child_tid_usdt, child->tid, "usdt_multi_child_tid");
+	}
 cleanup:
 	uprobe_multi__destroy(skel);
 }
@@ -210,6 +314,13 @@ test_attach_api(const char *binary, const char *pattern, struct bpf_uprobe_multi
 		return;
 
 	__test_attach_api(binary, pattern, opts, child);
+
+	/* pid filter (thread) */
+	child = spawn_thread();
+	if (!ASSERT_OK_PTR(child, "spawn_thread"))
+		return;
+
+	__test_attach_api(binary, pattern, opts, child);
 }
 
 static void test_attach_api_pattern(void)
@@ -397,7 +508,7 @@ static void test_attach_api_fails(void)
 	link_fd = bpf_link_create(prog_fd, 0, BPF_TRACE_UPROBE_MULTI, &opts);
 	if (!ASSERT_ERR(link_fd, "link_fd"))
 		goto cleanup;
-	ASSERT_EQ(link_fd, -ESRCH, "pid_is_wrong");
+	ASSERT_EQ(link_fd, -EINVAL, "pid_is_wrong");
 
 cleanup:
 	if (link_fd >= 0)
@@ -495,6 +606,13 @@ static void test_link_api(void)
 		return;
 
 	__test_link_api(child);
+
+	/* pid filter (thread) */
+	child = spawn_thread();
+	if (!ASSERT_OK_PTR(child, "spawn_thread"))
+		return;
+
+	__test_link_api(child);
 }
 
 static void test_bench_attach_uprobe(void)
diff --git a/tools/testing/selftests/bpf/progs/uprobe_multi.c b/tools/testing/selftests/bpf/progs/uprobe_multi.c
index 419d9aa28fce..44190efcdba2 100644
--- a/tools/testing/selftests/bpf/progs/uprobe_multi.c
+++ b/tools/testing/selftests/bpf/progs/uprobe_multi.c
@@ -1,8 +1,8 @@
 // SPDX-License-Identifier: GPL-2.0
-#include <linux/bpf.h>
+#include "vmlinux.h"
 #include <bpf/bpf_helpers.h>
 #include <bpf/bpf_tracing.h>
-#include <stdbool.h>
+#include <bpf/usdt.bpf.h>
 
 char _license[] SEC("license") = "GPL";
 
@@ -22,6 +22,13 @@ __u64 uprobe_multi_sleep_result = 0;
 
 int pid = 0;
 int child_pid = 0;
+int child_tid = 0;
+int child_pid_usdt = 0;
+int child_tid_usdt = 0;
+
+int expect_pid = 0;
+bool bad_pid_seen = false;
+bool bad_pid_seen_usdt = false;
 
 bool test_cookie = false;
 void *user_ptr = 0;
@@ -36,11 +43,19 @@ static __always_inline bool verify_sleepable_user_copy(void)
 
 static void uprobe_multi_check(void *ctx, bool is_return, bool is_sleep)
 {
-	child_pid = bpf_get_current_pid_tgid() >> 32;
+	__u64 cur_pid_tgid = bpf_get_current_pid_tgid();
+	__u32 cur_pid;
 
-	if (pid && child_pid != pid)
+	cur_pid = cur_pid_tgid >> 32;
+	if (pid && cur_pid != pid)
 		return;
 
+	if (expect_pid && cur_pid != expect_pid)
+		bad_pid_seen = true;
+
+	child_pid = cur_pid_tgid >> 32;
+	child_tid = (__u32)cur_pid_tgid;
+
 	__u64 cookie = test_cookie ? bpf_get_attach_cookie(ctx) : 0;
 	__u64 addr = bpf_get_func_ip(ctx);
 
@@ -97,5 +112,32 @@ int uretprobe_sleep(struct pt_regs *ctx)
 SEC("uprobe.multi//proc/self/exe:uprobe_multi_func_*")
 int uprobe_extra(struct pt_regs *ctx)
 {
+	/* we need this one just to mix PID-filtered and global uprobes */
+	return 0;
+}
+
+SEC("usdt")
+int usdt_pid(struct pt_regs *ctx)
+{
+	__u64 cur_pid_tgid = bpf_get_current_pid_tgid();
+	__u32 cur_pid;
+
+	cur_pid = cur_pid_tgid >> 32;
+	if (pid && cur_pid != pid)
+		return 0;
+
+	if (expect_pid && cur_pid != expect_pid)
+		bad_pid_seen_usdt = true;
+
+	child_pid_usdt = cur_pid_tgid >> 32;
+	child_tid_usdt = (__u32)cur_pid_tgid;
+
+	return 0;
+}
+
+SEC("usdt")
+int usdt_extra(struct pt_regs *ctx)
+{
+	/* we need this one just to mix PID-filtered and global USDT probes */
 	return 0;
 }