perf: tracing: Have perf system calls read user space

Allow some of the system call events to read user space buffers. Instead
of just showing the pointer into user space, allow perf events to also
record the content of those pointers. For example:

# perf record -e syscalls:sys_enter_openat ls /usr/bin
[..]
# perf script
ls 1024 [005] 52.902721: syscalls:sys_enter_openat: dfd: 0xffffff9c, filename: 0x7fc1dbae321c "/etc/ld.so.cache", flags: 0x00080000, mode: 0x00000000
ls 1024 [005] 52.902899: syscalls:sys_enter_openat: dfd: 0xffffff9c, filename: 0x7fc1dbaae140 "/lib/x86_64-linux-gnu/libselinux.so.1", flags: 0x00080000, mode: 0x00000000
ls 1024 [005] 52.903471: syscalls:sys_enter_openat: dfd: 0xffffff9c, filename: 0x7fc1dbaae690 "/lib/x86_64-linux-gnu/libcap.so.2", flags: 0x00080000, mode: 0x00000000
ls 1024 [005] 52.903946: syscalls:sys_enter_openat: dfd: 0xffffff9c, filename: 0x7fc1dbaaebe0 "/lib/x86_64-linux-gnu/libc.so.6", flags: 0x00080000, mode: 0x00000000
ls 1024 [005] 52.904629: syscalls:sys_enter_openat: dfd: 0xffffff9c, filename: 0x7fc1dbaaf110 "/lib/x86_64-linux-gnu/libpcre2-8.so.0", flags: 0x00080000, mode: 0x00000000
ls 1024 [005] 52.906985: syscalls:sys_enter_openat: dfd: 0xffffffffffffff9c, filename: 0x7fc1dba92904 "/proc/filesystems", flags: 0x00080000, mode: 0x00000000
ls 1024 [005] 52.907323: syscalls:sys_enter_openat: dfd: 0xffffff9c, filename: 0x7fc1dba19490 "/usr/lib/locale/locale-archive", flags: 0x00080000, mode: 0x00000000
ls 1024 [005] 52.907746: syscalls:sys_enter_openat: dfd: 0xffffff9c, filename: 0x556fb888dcd0 "/usr/bin", flags: 0x00090800, mode: 0x00000000

Cc: Masami Hiramatsu <[e-mail]>
Cc: Mark Rutland <[e-mail]>
Cc: Mathieu Desnoyers <[e-mail]>
Cc: Andrew Morton <[e-mail]>
Cc: Peter Zijlstra <[e-mail]>
Cc: Namhyung Kim <[e-mail]>
Cc: Takaya Saeki <[e-mail]>
Cc: Tom Zanussi <[e-mail]>
Cc: Thomas Gleixner <[e-mail]>
Cc: Ian Rogers <[e-mail]>
Cc: Douglas Raillard <[e-mail]>
Cc: Arnaldo Carvalho de Melo <[e-mail]>
Cc: Jiri Olsa <[e-mail]>
Cc: Adrian Hunter <[e-mail]>
Cc: Ingo Molnar <[e-mail]>
Link: https://lore.kernel.org/[e-mail]
Signed-off-by: Steven Rostedt (Google) <[e-mail]>

kernel/trace/trace_syscalls.c

···
 	return buf;
 }
 
+static int
+syscall_get_data(struct syscall_metadata *sys_data, unsigned long *args,
+		 char **buffer, int *size, int *user_size)
+{
+	struct syscall_user_buffer *sbuf;
+
+	/* If the syscall_buffer is NULL, tracing is being shutdown */
+	sbuf = READ_ONCE(syscall_buffer);
+	if (!sbuf)
+		return -1;
+
+	*buffer = sys_fault_user(sys_data, sbuf, args, user_size);
+	/*
+	 * user_size is the amount of data to append.
+	 * Need to add 4 for the meta field that points to
+	 * the user memory at the end of the event and also
+	 * stores its size.
+	 */
+	*size = 4 + *user_size;
+	return 0;
+}
+
+static void syscall_put_data(struct syscall_metadata *sys_data,
+			     struct syscall_trace_enter *entry,
+			     char *buffer, int size)
+{
+	void *ptr;
+	int val;
+
+	/*
+	 * Set the pointer to point to the meta data of the event
+	 * that has information about the stored user space memory.
+	 */
+	ptr = (void *)entry->args + sizeof(unsigned long) * sys_data->nb_args;
+
+	/*
+	 * The meta data will store the offset of the user data from
+	 * the beginning of the event.
+	 */
+	val  = (ptr - (void *)entry) + 4;
+
+	/* Store the offset and the size into the meta data */
+	*(int *)ptr = val | (size << 16);
+
+	/* Nothing to do if the user space was empty or faulted */
+	if (size) {
+		/* Now store the user space data into the event */
+		ptr += 4;
+		memcpy(ptr, buffer, size);
+	}
+}
+
 static void ftrace_syscall_enter(void *data, struct pt_regs *regs, long id)
 {
 	struct trace_array *tr = data;
···
 	syscall_get_arguments(current, regs, args);
 
 	if (mayfault) {
-		struct syscall_user_buffer *sbuf;
-
-		/* If the syscall_buffer is NULL, tracing is being shutdown */
-		sbuf = READ_ONCE(syscall_buffer);
-		if (!sbuf)
+		if (syscall_get_data(sys_data, args, &user_ptr,
+				     &size, &user_size) < 0)
 			return;
-
-		user_ptr = sys_fault_user(sys_data, sbuf, args, &user_size);
-		/*
-		 * user_size is the amount of data to append.
-		 * Need to add 4 for the meta field that points to
-		 * the user memory at the end of the event and also
-		 * stores its size.
-		 */
-		size = 4 + user_size;
 	}
 
 	size += sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
···
 
 	memcpy(entry->args, args, sizeof(unsigned long) * sys_data->nb_args);
 
-	if (mayfault) {
-		void *ptr;
-		int val;
-
-		/*
-		 * Set the pointer to point to the meta data of the event
-		 * that has information about the stored user space memory.
-		 */
-		ptr = (void *)entry->args + sizeof(unsigned long) * sys_data->nb_args;
-
-		/*
-		 * The meta data will store the offset of the user data from
-		 * the beginning of the event.
-		 */
-		val  = (ptr - (void *)entry) + 4;
-
-		/* Store the offset and the size into the meta data */
-		*(int *)ptr = val | (user_size << 16);
-
-		/* Nothing to do if the user space was empty or faulted */
-		if (user_size) {
-			/* Now store the user space data into the event */
-			ptr += 4;
-			memcpy(ptr, user_ptr, user_size);
-		}
-	}
+	if (mayfault)
+		syscall_put_data(sys_data, entry, user_ptr, user_size);
 
 	trace_event_buffer_commit(&fbuffer);
 }
···
 	struct hlist_head *head;
 	unsigned long args[6];
 	bool valid_prog_array;
+	bool mayfault;
+	char *user_ptr;
 	int syscall_nr;
+	int user_size;
 	int rctx;
-	int size;
+	int size = 0;
 
 	/*
 	 * Syscall probe called with preemption enabled, but the ring
···
 	if (!sys_data)
 		return;
 
+	syscall_get_arguments(current, regs, args);
+
+	/* Check if this syscall event faults in user space memory */
+	mayfault = sys_data->user_mask != 0;
+
+	if (mayfault) {
+		if (syscall_get_data(sys_data, args, &user_ptr,
+				     &size, &user_size) < 0)
+			return;
+	}
+
 	head = this_cpu_ptr(sys_data->enter_event->perf_events);
 	valid_prog_array = bpf_prog_array_valid(sys_data->enter_event);
 	if (!valid_prog_array && hlist_empty(head))
 		return;
 
 	/* get the size after alignment with the u32 buffer size field */
-	size = sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
+	size += sizeof(unsigned long) * sys_data->nb_args + sizeof(*rec);
 	size = ALIGN(size + sizeof(u32), sizeof(u64));
 	size -= sizeof(u32);
 
···
 		return;
 
 	rec->nr = syscall_nr;
-	syscall_get_arguments(current, regs, args);
 	memcpy(&rec->args, args, sizeof(unsigned long) * sys_data->nb_args);
+
+	if (mayfault)
+		syscall_put_data(sys_data, rec, user_ptr, user_size);
 
 	if ((valid_prog_array &&
 	     !perf_call_bpf_enter(sys_data->enter_event, fake_regs, sys_data, rec)) ||
···
 
 static int perf_sysenter_enable(struct trace_event_call *call)
 {
+	struct syscall_metadata *sys_data = call->data;
 	int num;
+	int ret;
 
-	num = ((struct syscall_metadata *)call->data)->syscall_nr;
+	num = sys_data->syscall_nr;
 
 	guard(mutex)(&syscall_trace_lock);
+	if (sys_data->user_mask) {
+		ret = syscall_fault_buffer_enable();
+		if (ret < 0)
+			return ret;
+	}
 	if (!sys_perf_refcount_enter) {
-		int ret = register_trace_sys_enter(perf_syscall_enter, NULL);
+		ret = register_trace_sys_enter(perf_syscall_enter, NULL);
 		if (ret) {
 			pr_info("event trace: Could not activate syscall entry trace point");
+			if (sys_data->user_mask)
+				syscall_fault_buffer_disable();
 			return ret;
 		}
 	}
···
 
 static void perf_sysenter_disable(struct trace_event_call *call)
 {
+	struct syscall_metadata *sys_data = call->data;
 	int num;
 
-	num = ((struct syscall_metadata *)call->data)->syscall_nr;
+	num = sys_data->syscall_nr;
 
 	guard(mutex)(&syscall_trace_lock);
 	sys_perf_refcount_enter--;
 	clear_bit(num, enabled_perf_enter_syscalls);
 	if (!sys_perf_refcount_enter)
 		unregister_trace_sys_enter(perf_syscall_enter, NULL);
+	if (sys_data->user_mask)
+		syscall_fault_buffer_disable();
 }
 
 static int perf_call_bpf_exit(struct trace_event_call *call, struct pt_regs *regs,