perf/core: Implement the 'perf_uprobe' PMU
authorSong Liu <songliubraving@fb.com>
Wed, 6 Dec 2017 22:45:16 +0000 (14:45 -0800)
committerIngo Molnar <mingo@kernel.org>
Tue, 6 Feb 2018 10:29:28 +0000 (11:29 +0100)
This patch adds perf_uprobe support with similar pattern as previous
patch (for kprobe).

Two functions, create_local_trace_uprobe() and
destroy_local_trace_uprobe(), are created so a uprobe can be created
and attached to the file descriptor created by perf_event_open().

Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Yonghong Song <yhs@fb.com>
Reviewed-by: Josef Bacik <jbacik@fb.com>
Cc: <daniel@iogearbox.net>
Cc: <davem@davemloft.net>
Cc: <kernel-team@fb.com>
Cc: <rostedt@goodmis.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20171206224518.3598254-7-songliubraving@fb.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
include/linux/trace_events.h
kernel/events/core.c
kernel/trace/trace_event_perf.c
kernel/trace/trace_probe.h
kernel/trace/trace_uprobe.c

index 21c5d43a21af1e94eed0b3c7fae506ba35564acb..0d9d6cb454b1f1934728cc053a59c0faf423c511 100644 (file)
@@ -537,6 +537,10 @@ extern void perf_trace_del(struct perf_event *event, int flags);
 extern int  perf_kprobe_init(struct perf_event *event, bool is_retprobe);
 extern void perf_kprobe_destroy(struct perf_event *event);
 #endif
+#ifdef CONFIG_UPROBE_EVENTS
+extern int  perf_uprobe_init(struct perf_event *event, bool is_retprobe);
+extern void perf_uprobe_destroy(struct perf_event *event);
+#endif
 extern int  ftrace_profile_set_filter(struct perf_event *event, int event_id,
                                     char *filter_str);
 extern void ftrace_profile_free_filter(struct perf_event *event);
index 333735531968cddaab85bea86a5cc1a56e0e7a1b..5a54630db86bb79bd535a5774fe2aaf0dd11237a 100644 (file)
@@ -7992,7 +7992,7 @@ static struct pmu perf_tracepoint = {
        .read           = perf_swevent_read,
 };
 
-#ifdef CONFIG_KPROBE_EVENTS
+#if defined(CONFIG_KPROBE_EVENTS) || defined(CONFIG_UPROBE_EVENTS)
 /*
  * Flags in config, used by dynamic PMU kprobe and uprobe
  * The flags should match following PMU_FORMAT_ATTR().
@@ -8020,7 +8020,9 @@ static const struct attribute_group *probe_attr_groups[] = {
        &probe_format_group,
        NULL,
 };
+#endif
 
+#ifdef CONFIG_KPROBE_EVENTS
 static int perf_kprobe_event_init(struct perf_event *event);
 static struct pmu perf_kprobe = {
        .task_ctx_nr    = perf_sw_context,
@@ -8057,12 +8059,52 @@ static int perf_kprobe_event_init(struct perf_event *event)
 }
 #endif /* CONFIG_KPROBE_EVENTS */
 
+#ifdef CONFIG_UPROBE_EVENTS
+static int perf_uprobe_event_init(struct perf_event *event);
+static struct pmu perf_uprobe = {
+       .task_ctx_nr    = perf_sw_context,
+       .event_init     = perf_uprobe_event_init,
+       .add            = perf_trace_add,
+       .del            = perf_trace_del,
+       .start          = perf_swevent_start,
+       .stop           = perf_swevent_stop,
+       .read           = perf_swevent_read,
+       .attr_groups    = probe_attr_groups,
+};
+
+static int perf_uprobe_event_init(struct perf_event *event)
+{
+       int err;
+       bool is_retprobe;
+
+       if (event->attr.type != perf_uprobe.type)
+               return -ENOENT;
+       /*
+        * no branch sampling for probe events
+        */
+       if (has_branch_stack(event))
+               return -EOPNOTSUPP;
+
+       is_retprobe = event->attr.config & PERF_PROBE_CONFIG_IS_RETPROBE;
+       err = perf_uprobe_init(event, is_retprobe);
+       if (err)
+               return err;
+
+       event->destroy = perf_uprobe_destroy;
+
+       return 0;
+}
+#endif /* CONFIG_UPROBE_EVENTS */
+
 static inline void perf_tp_register(void)
 {
        perf_pmu_register(&perf_tracepoint, "tracepoint", PERF_TYPE_TRACEPOINT);
 #ifdef CONFIG_KPROBE_EVENTS
        perf_pmu_register(&perf_kprobe, "kprobe", -1);
 #endif
+#ifdef CONFIG_UPROBE_EVENTS
+       perf_pmu_register(&perf_uprobe, "uprobe", -1);
+#endif
 }
 
 static void perf_event_free_filter(struct perf_event *event)
@@ -8150,6 +8192,10 @@ static inline bool perf_event_is_tracing(struct perf_event *event)
 #ifdef CONFIG_KPROBE_EVENTS
        if (event->pmu == &perf_kprobe)
                return true;
+#endif
+#ifdef CONFIG_UPROBE_EVENTS
+       if (event->pmu == &perf_uprobe)
+               return true;
 #endif
        return false;
 }
index 779baade91140e965f4f587d25f076ca0c631a32..2c416509b834a728e1c1bd8d5564fa1505cf2ea1 100644 (file)
@@ -286,6 +286,59 @@ void perf_kprobe_destroy(struct perf_event *p_event)
 }
 #endif /* CONFIG_KPROBE_EVENTS */
 
+#ifdef CONFIG_UPROBE_EVENTS
+int perf_uprobe_init(struct perf_event *p_event, bool is_retprobe)
+{
+       int ret;
+       char *path = NULL;
+       struct trace_event_call *tp_event;
+
+       if (!p_event->attr.uprobe_path)
+               return -EINVAL;
+       path = kzalloc(PATH_MAX, GFP_KERNEL);
+       if (!path)
+               return -ENOMEM;
+       ret = strncpy_from_user(
+               path, u64_to_user_ptr(p_event->attr.uprobe_path), PATH_MAX);
+       if (ret < 0)
+               goto out;
+       if (path[0] == '\0') {
+               ret = -EINVAL;
+               goto out;
+       }
+
+       tp_event = create_local_trace_uprobe(
+               path, p_event->attr.probe_offset, is_retprobe);
+       if (IS_ERR(tp_event)) {
+               ret = PTR_ERR(tp_event);
+               goto out;
+       }
+
+       /*
+        * local trace_uprobe need to hold event_mutex to call
+        * uprobe_buffer_enable() and uprobe_buffer_disable().
+        * event_mutex is not required for local trace_kprobes.
+        */
+       mutex_lock(&event_mutex);
+       ret = perf_trace_event_init(tp_event, p_event);
+       if (ret)
+               destroy_local_trace_uprobe(tp_event);
+       mutex_unlock(&event_mutex);
+out:
+       kfree(path);
+       return ret;
+}
+
+void perf_uprobe_destroy(struct perf_event *p_event)
+{
+       mutex_lock(&event_mutex);
+       perf_trace_event_close(p_event);
+       perf_trace_event_unreg(p_event);
+       mutex_unlock(&event_mutex);
+       destroy_local_trace_uprobe(p_event->tp_event);
+}
+#endif /* CONFIG_UPROBE_EVENTS */
+
 int perf_trace_add(struct perf_event *p_event, int flags)
 {
        struct trace_event_call *tp_event = p_event->tp_event;
index e3d940a49dcdebfbf7382a89a658949cee477e2d..27de3174050a98ee32f7fc6f1cbf2573d0758d07 100644 (file)
@@ -410,4 +410,8 @@ extern struct trace_event_call *
 create_local_trace_kprobe(char *func, void *addr, unsigned long offs,
                          bool is_return);
 extern void destroy_local_trace_kprobe(struct trace_event_call *event_call);
+
+extern struct trace_event_call *
+create_local_trace_uprobe(char *name, unsigned long offs, bool is_return);
+extern void destroy_local_trace_uprobe(struct trace_event_call *event_call);
 #endif
index 40592e7b3568bcfd41220f579fc05066e3675b33..e3cbae702a53da96e7bea192ead16aec92d3412e 100644 (file)
@@ -1292,16 +1292,25 @@ static struct trace_event_functions uprobe_funcs = {
        .trace          = print_uprobe_event
 };
 
-static int register_uprobe_event(struct trace_uprobe *tu)
+static inline void init_trace_event_call(struct trace_uprobe *tu,
+                                        struct trace_event_call *call)
 {
-       struct trace_event_call *call = &tu->tp.call;
-       int ret;
-
-       /* Initialize trace_event_call */
        INIT_LIST_HEAD(&call->class->fields);
        call->event.funcs = &uprobe_funcs;
        call->class->define_fields = uprobe_event_define_fields;
 
+       call->flags = TRACE_EVENT_FL_UPROBE;
+       call->class->reg = trace_uprobe_register;
+       call->data = tu;
+}
+
+static int register_uprobe_event(struct trace_uprobe *tu)
+{
+       struct trace_event_call *call = &tu->tp.call;
+       int ret = 0;
+
+       init_trace_event_call(tu, call);
+
        if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0)
                return -ENOMEM;
 
@@ -1311,9 +1320,6 @@ static int register_uprobe_event(struct trace_uprobe *tu)
                return -ENODEV;
        }
 
-       call->flags = TRACE_EVENT_FL_UPROBE;
-       call->class->reg = trace_uprobe_register;
-       call->data = tu;
        ret = trace_add_event_call(call);
 
        if (ret) {
@@ -1339,6 +1345,70 @@ static int unregister_uprobe_event(struct trace_uprobe *tu)
        return 0;
 }
 
+#ifdef CONFIG_PERF_EVENTS
+struct trace_event_call *
+create_local_trace_uprobe(char *name, unsigned long offs, bool is_return)
+{
+       struct trace_uprobe *tu;
+       struct inode *inode;
+       struct path path;
+       int ret;
+
+       ret = kern_path(name, LOOKUP_FOLLOW, &path);
+       if (ret)
+               return ERR_PTR(ret);
+
+       inode = igrab(d_inode(path.dentry));
+       path_put(&path);
+
+       if (!inode || !S_ISREG(inode->i_mode)) {
+               iput(inode);
+               return ERR_PTR(-EINVAL);
+       }
+
+       /*
+        * local trace_kprobes are not added to probe_list, so they are never
+        * searched in find_trace_kprobe(). Therefore, there is no concern of
+        * duplicated name "DUMMY_EVENT" here.
+        */
+       tu = alloc_trace_uprobe(UPROBE_EVENT_SYSTEM, "DUMMY_EVENT", 0,
+                               is_return);
+
+       if (IS_ERR(tu)) {
+               pr_info("Failed to allocate trace_uprobe.(%d)\n",
+                       (int)PTR_ERR(tu));
+               return ERR_CAST(tu);
+       }
+
+       tu->offset = offs;
+       tu->inode = inode;
+       tu->filename = kstrdup(name, GFP_KERNEL);
+       init_trace_event_call(tu, &tu->tp.call);
+
+       if (set_print_fmt(&tu->tp, is_ret_probe(tu)) < 0) {
+               ret = -ENOMEM;
+               goto error;
+       }
+
+       return &tu->tp.call;
+error:
+       free_trace_uprobe(tu);
+       return ERR_PTR(ret);
+}
+
+void destroy_local_trace_uprobe(struct trace_event_call *event_call)
+{
+       struct trace_uprobe *tu;
+
+       tu = container_of(event_call, struct trace_uprobe, tp.call);
+
+       kfree(tu->tp.call.print_fmt);
+       tu->tp.call.print_fmt = NULL;
+
+       free_trace_uprobe(tu);
+}
+#endif /* CONFIG_PERF_EVENTS */
+
 /* Make a trace interface for controling probe points */
 static __init int init_uprobe_trace(void)
 {