Skip to contentSkip to conversation

tracing/uprobe: Fix double perf_event linking on multiprobe uprobe

PATCH In review Masami Hiramatsu 4 participants 13 messages Jan 10, 2020 – Jan 22, 2020 Tracing

lore
Share & tools
lore

Masami Hiramatsu

author maintainer
#0 lore raw

PATCH 1/1 tracing/uprobe: Fix double perf_event linking on multiprobe uprobe

Fix double perf_event linking to trace_uprobe_filter on multiple uprobe event by moving trace_uprobe_filter under trace_probe_event.

In uprobe perf event, trace_uprobe_filter data structure is managing target mm filters (in perf_event) related to each uprobe event.

Since commit 60d53e2c3b75 ("tracing/probe: Split trace_event related data from trace_probe") left the trace_uprobe_filter data structure in trace_uprobe, if a trace_probe_event has multiple trace_uprobe (multi-probe event), a perf_event is added to different trace_uprobe_filter on each trace_uprobe. This leads a linked list corruption.

To fix this issue, move trace_uprobe_filter to trace_probe_event and link it once on each event instead of each probe.

Fixes 60d53e2c3b75 ("tracing/probe: Split trace_event related data from trace_probe")

URL: https://lkml.kernel.org/r/[email protected]

---
 kernel/trace/trace_kprobe.c |    2 -
 kernel/trace/trace_probe.c  |    5 +-
 kernel/trace/trace_probe.h  |    3 +
 kernel/trace/trace_uprobe.c |  124 ++++++++++++++++++++++++++++---------------
 4 files changed, 86 insertions(+), 48 deletions(-)
kernel/trace/trace_kprobe.c +1 -1
index 7f890262c8a3..3e5f9c7d939c 100644
@@ -290,7 +290,7 @@ static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 	INIT_HLIST_NODE(&tk->rp.kp.hlist);
 	INIT_LIST_HEAD(&tk->rp.kp.list);
 
-	ret = trace_probe_init(&tk->tp, event, group);
+	ret = trace_probe_init(&tk->tp, event, group, 0);
 	if (ret < 0)
 		goto error;
 
kernel/trace/trace_probe.c +3 -2
index 905b10af5d5c..bba18cf44a30 100644
@@ -984,7 +984,7 @@ void trace_probe_cleanup(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group)
+		     const char *group, size_t event_data_size)
 {
 	struct trace_event_call *call;
 	int ret = 0;
@@ -992,7 +992,8 @@ int trace_probe_init(struct trace_probe *tp, const char *event,
 	if (!event || !group)
 		return -EINVAL;
 
-	tp->event = kzalloc(sizeof(struct trace_probe_event), GFP_KERNEL);
+	tp->event = kzalloc(sizeof(struct trace_probe_event) + event_data_size,
+			    GFP_KERNEL);
 	if (!tp->event)
 		return -ENOMEM;
 
kernel/trace/trace_probe.h +2 -1
index 4ee703728aec..03e4e180058d 100644
@@ -230,6 +230,7 @@ struct trace_probe_event {
 	struct trace_event_call		call;
 	struct list_head 		files;
 	struct list_head		probes;
+	char				data[0];
 };
 
 struct trace_probe {
@@ -322,7 +323,7 @@ static inline bool trace_probe_has_single_file(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group);
+		     const char *group, size_t event_data_size);
 void trace_probe_cleanup(struct trace_probe *tp);
 int trace_probe_append(struct trace_probe *tp, struct trace_probe *to);
 void trace_probe_unlink(struct trace_probe *tp);
kernel/trace/trace_uprobe.c +80 -44

229 lines not on this page · raw message

author maintainer
#1 lore raw
Quoted 15 lines · On Fri, 10 Jan 2020 10:45:39 +0900 Masami Hiramatsu <[email protected]> wrote:

Fix double perf_event linking to trace_uprobe_filter on multiple uprobe event by moving trace_uprobe_filter under trace_probe_event.

In uprobe perf event, trace_uprobe_filter data structure is managing target mm filters (in perf_event) related to each uprobe event.

Since commit 60d53e2c3b75 ("tracing/probe: Split trace_event related data from trace_probe") left the trace_uprobe_filter data structure in trace_uprobe, if a trace_probe_event has multiple trace_uprobe (multi-probe event), a perf_event is added to different trace_uprobe_filter on each trace_uprobe. This leads a linked list corruption.

Quoted 6 lines

To fix this issue, move trace_uprobe_filter to trace_probe_event and link it once on each event instead of each probe.

Fixes 60d53e2c3b75 ("tracing/probe: Split trace_event related data from trace_probe")

URL: https://lkml.kernel.org/r/[email protected]

Oops, s/URL/Link/

Thanks,

Quoted 309 lines
---
 kernel/trace/trace_kprobe.c |    2 -
 kernel/trace/trace_probe.c  |    5 +-
 kernel/trace/trace_probe.h  |    3 +
 kernel/trace/trace_uprobe.c |  124 ++++++++++++++++++++++++++++---------------
 4 files changed, 86 insertions(+), 48 deletions(-)
kernel/trace/trace_kprobe.c +1 -1
index 7f890262c8a3..3e5f9c7d939c 100644
@@ -290,7 +290,7 @@ static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 	INIT_HLIST_NODE(&tk->rp.kp.hlist);
 	INIT_LIST_HEAD(&tk->rp.kp.list);
 
-	ret = trace_probe_init(&tk->tp, event, group);
+	ret = trace_probe_init(&tk->tp, event, group, 0);
 	if (ret < 0)
 		goto error;
 
kernel/trace/trace_probe.c +3 -2
index 905b10af5d5c..bba18cf44a30 100644
@@ -984,7 +984,7 @@ void trace_probe_cleanup(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group)
+		     const char *group, size_t event_data_size)
 {
 	struct trace_event_call *call;
 	int ret = 0;
@@ -992,7 +992,8 @@ int trace_probe_init(struct trace_probe *tp, const char *event,
 	if (!event || !group)
 		return -EINVAL;
 
-	tp->event = kzalloc(sizeof(struct trace_probe_event), GFP_KERNEL);
+	tp->event = kzalloc(sizeof(struct trace_probe_event) + event_data_size,
+			    GFP_KERNEL);
 	if (!tp->event)
 		return -ENOMEM;
 
kernel/trace/trace_probe.h +2 -1
index 4ee703728aec..03e4e180058d 100644
@@ -230,6 +230,7 @@ struct trace_probe_event {
 	struct trace_event_call		call;
 	struct list_head 		files;
 	struct list_head		probes;
+	char				data[0];
 };
 
 struct trace_probe {
@@ -322,7 +323,7 @@ static inline bool trace_probe_has_single_file(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group);
+		     const char *group, size_t event_data_size);
 void trace_probe_cleanup(struct trace_probe *tp);
 int trace_probe_append(struct trace_probe *tp, struct trace_probe *to);
 void trace_probe_unlink(struct trace_probe *tp);
kernel/trace/trace_uprobe.c +80 -44
index 352073d36585..f66e202fec13 100644
@@ -60,7 +60,6 @@ static struct dyn_event_operations trace_uprobe_ops = {
  */
 struct trace_uprobe {
 	struct dyn_event		devent;
-	struct trace_uprobe_filter	filter;
 	struct uprobe_consumer		consumer;
 	struct path			path;
 	struct inode			*inode;
@@ -264,6 +263,14 @@ process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 }
 NOKPROBE_SYMBOL(process_fetch_insn)
 
+static struct trace_uprobe_filter *
+trace_uprobe_get_filter(struct trace_uprobe *tu)
+{
+	struct trace_probe_event *event = tu->tp.event;
+
+	return (struct trace_uprobe_filter *)&event->data[0];
+}
+
 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 {
 	rwlock_init(&filter->rwlock);
@@ -351,7 +358,8 @@ alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 	if (!tu)
 		return ERR_PTR(-ENOMEM);
 
-	ret = trace_probe_init(&tu->tp, event, group);
+	ret = trace_probe_init(&tu->tp, event, group,
+				sizeof(struct trace_uprobe_filter));
 	if (ret < 0)
 		goto error;
 
@@ -359,7 +367,7 @@ alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 	tu->consumer.handler = uprobe_dispatcher;
 	if (is_ret)
 		tu->consumer.ret_handler = uretprobe_dispatcher;
-	init_trace_uprobe_filter(&tu->filter);
+	init_trace_uprobe_filter(trace_uprobe_get_filter(tu));
 	return tu;
 
 error:
@@ -1067,13 +1075,14 @@ static void __probe_event_disable(struct trace_probe *tp)
 	struct trace_probe *pos;
 	struct trace_uprobe *tu;
 
+	tu = container_of(tp, struct trace_uprobe, tp);
+	WARN_ON(!uprobe_filter_is_empty(trace_uprobe_get_filter(tu)));
+
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 		tu = container_of(pos, struct trace_uprobe, tp);
 		if (!tu->inode)
 			continue;
 
-		WARN_ON(!uprobe_filter_is_empty(&tu->filter));
-
 		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 		tu->inode = NULL;
 	}
@@ -1108,7 +1117,7 @@ static int probe_event_enable(struct trace_event_call *call,
 	}
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
+	WARN_ON(!uprobe_filter_is_empty(trace_uprobe_get_filter(tu)));
 
 	if (enabled)
 		return 0;
@@ -1205,39 +1214,39 @@ __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
 }
 
 static inline bool
-uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
+trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
+			  struct perf_event *event)
 {
-	return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
+	return __uprobe_perf_filter(filter, event->hw.target->mm);
 }
 
-static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
+static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
+				       struct perf_event *event)
 {
 	bool done;
 
-	write_lock(&tu->filter.rwlock);
+	write_lock(&filter->rwlock);
 	if (event->hw.target) {
 		list_del(&event->hw.tp_list);
-		done = tu->filter.nr_systemwide ||
+		done = filter->nr_systemwide ||
 			(event->hw.target->flags & PF_EXITING) ||
-			uprobe_filter_event(tu, event);
+			trace_uprobe_filter_event(filter, event);
 	} else {
-		tu->filter.nr_systemwide--;
-		done = tu->filter.nr_systemwide;
+		filter->nr_systemwide--;
+		done = filter->nr_systemwide;
 	}
-	write_unlock(&tu->filter.rwlock);
-
-	if (!done)
-		return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
+	write_unlock(&filter->rwlock);
 
-	return 0;
+	return done;
 }
 
-static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
+/* This returns true if the filter always covers target mm */
+static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
+				    struct perf_event *event)
 {
 	bool done;
-	int err;
 
-	write_lock(&tu->filter.rwlock);
+	write_lock(&filter->rwlock);
 	if (event->hw.target) {
 		/*
 		 * event->parent != NULL means copy_process(), we can avoid
@@ -1247,28 +1256,21 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 		 * attr.enable_on_exec means that exec/mmap will install the
 		 * breakpoints we need.
 		 */
-		done = tu->filter.nr_systemwide ||
+		done = filter->nr_systemwide ||
 			event->parent || event->attr.enable_on_exec ||
-			uprobe_filter_event(tu, event);
-		list_add(&event->hw.tp_list, &tu->filter.perf_events);
+			trace_uprobe_filter_event(filter, event);
+		list_add(&event->hw.tp_list, &filter->perf_events);
 	} else {
-		done = tu->filter.nr_systemwide;
-		tu->filter.nr_systemwide++;
+		done = filter->nr_systemwide;
+		filter->nr_systemwide++;
 	}
-	write_unlock(&tu->filter.rwlock);
+	write_unlock(&filter->rwlock);
 
-	err = 0;
-	if (!done) {
-		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
-		if (err)
-			uprobe_perf_close(tu, event);
-	}
-	return err;
+	return done;
 }
 
-static int uprobe_perf_multi_call(struct trace_event_call *call,
-				  struct perf_event *event,
-		int (*op)(struct trace_uprobe *tu, struct perf_event *event))
+static int uprobe_perf_close(struct trace_event_call *call,
+			     struct perf_event *event)
 {
 	struct trace_probe *pos, *tp;
 	struct trace_uprobe *tu;
@@ -1278,25 +1280,59 @@ static int uprobe_perf_multi_call(struct trace_event_call *call,
 	if (WARN_ON_ONCE(!tp))
 		return -ENODEV;
 
+	tu = container_of(tp, struct trace_uprobe, tp);
+	if (trace_uprobe_filter_remove(trace_uprobe_get_filter(tu), event))
+		return 0;
+
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 		tu = container_of(pos, struct trace_uprobe, tp);
-		ret = op(tu, event);
+		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 		if (ret)
 			break;
 	}
 
 	return ret;
 }
+
+static int uprobe_perf_open(struct trace_event_call *call,
+			    struct perf_event *event)
+{
+	struct trace_probe *pos, *tp;
+	struct trace_uprobe *tu;
+	int err = 0;
+
+	tp = trace_probe_primary_from_call(call);
+	if (WARN_ON_ONCE(!tp))
+		return -ENODEV;
+
+	tu = container_of(tp, struct trace_uprobe, tp);
+	if (trace_uprobe_filter_add(trace_uprobe_get_filter(tu), event))
+		return 0;
+
+	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
+		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+		if (err) {
+			uprobe_perf_close(call, event);
+			break;
+		}
+	}
+
+	return err;
+}
+
 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
 				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
 {
+	struct trace_uprobe_filter *filter;
 	struct trace_uprobe *tu;
 	int ret;
 
 	tu = container_of(uc, struct trace_uprobe, consumer);
-	read_lock(&tu->filter.rwlock);
-	ret = __uprobe_perf_filter(&tu->filter, mm);
-	read_unlock(&tu->filter.rwlock);
+	filter = trace_uprobe_get_filter(tu);
+
+	read_lock(&filter->rwlock);
+	ret = __uprobe_perf_filter(filter, mm);
+	read_unlock(&filter->rwlock);
 
 	return ret;
 }
@@ -1419,10 +1455,10 @@ trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
 		return 0;
 
 	case TRACE_REG_PERF_OPEN:
-		return uprobe_perf_multi_call(event, data, uprobe_perf_open);
+		return uprobe_perf_open(event, data);
 
 	case TRACE_REG_PERF_CLOSE:
-		return uprobe_perf_multi_call(event, data, uprobe_perf_close);
+		return uprobe_perf_close(event, data);
 
 #endif
 	default:
Signature · 1 line
Masami Hiramatsu <[email protected]>
Tested-by on v1 1/1
#2 lore raw
In reply to Masami Hiramatsu

Em Fri, Jan 10, 2020 at 10:45:39AM +0900, Masami Hiramatsu escreveu:

Quoted 15 lines

Fix double perf_event linking to trace_uprobe_filter on multiple uprobe event by moving trace_uprobe_filter under trace_probe_event.

In uprobe perf event, trace_uprobe_filter data structure is managing target mm filters (in perf_event) related to each uprobe event.

Since commit 60d53e2c3b75 ("tracing/probe: Split trace_event related data from trace_probe") left the trace_uprobe_filter data structure in trace_uprobe, if a trace_probe_event has multiple trace_uprobe (multi-probe event), a perf_event is added to different trace_uprobe_filter on each trace_uprobe. This leads a linked list corruption.

Quoted 7 lines

To fix this issue, move trace_uprobe_filter to trace_probe_event and link it once on each event instead of each probe.

Fixes 60d53e2c3b75 ("tracing/probe: Split trace_event related data from trace_probe")

URL: https://lkml.kernel.org/r/[email protected]

Thanks, now it works as expected:

If you don't mind I can put it into my next perf/urgent pull req to Ingo/Thomas,

And here is the sequence to check that libbpf routines linked statically with perf are in fact being used when using 'perf trace' with one of its events being a .c file, which gets interpreted as a bpf program, gets compiled, and then the .o is handed over to libbpf on its way to the kernel, in this specific case it'll collect, among other things, the open, openat, open* pathnames, then we see that perf stat counts how many times those routines were called:

[root@quaco ~]# perf probe -x ~/bin/perf bpf_[op]*__*
Added new events:
probe_perf:bpf_output__fprintf (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_output__printer (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__bpil_offs_to_addr (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__bpil_addr_to_offs (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__get_prog_info_linear (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__attach_trace (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__attach_raw_tracepoint (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__attach_tracepoint (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__attach_uprobe (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__attach_kprobe (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__attach_perf_event (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__find_map_by_offset (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__find_map_fd_by_name (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__find_map_by_name (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_expected_attach_type (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__get_expected_attach_type (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_tracing (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_tracing (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_perf_event (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_perf_event (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_xdp (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_xdp (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_raw_tracepoint (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_raw_tracepoint (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_tracepoint (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_tracepoint (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_sched_act (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_sched_act (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_sched_cls (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_sched_cls (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_kprobe (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_kprobe (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_socket_filter (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_socket_filter (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_type (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_type (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__get_type (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__nth_fd (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_prep (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__size (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__fd (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__title (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_ifindex (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__priv (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__set_priv (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__prev (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__next (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__priv (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__set_priv (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__btf_fd (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__btf (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__kversion (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__name (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__next (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__close (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__pin (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__unpin_programs (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__pin_programs (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__unpin_maps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__pin_maps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__unpin (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__pin (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__unpin_instance (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__pin_instance (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__load (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__load_xattr (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__unload (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__open_buffer (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__open_mem (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__open_file (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__open (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__open_xattr (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__load_progs (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__is_function_storage (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__load (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__collect_reloc (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__relocate (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__relocate (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__reloc_text (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__relocate_core (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__create_maps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__populate_internal_map (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__reuse_map (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__probe_caps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__probe_array_mmap (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__probe_btf_datasec (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__probe_btf_func (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__probe_global_data (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__probe_name (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__collect_reloc (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__record_reloc (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__section_to_libbpf_map_type (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__shndx_is_maps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__shndx_is_data (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__find_program_by_title (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__find_prog_by_idx (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__elf_collect (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__sanitize_and_load_btf (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_btf (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__is_btf_mandatory (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__sanitize_btf_ext (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__sanitize_btf (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_maps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_user_btf_maps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_user_btf_map (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_user_maps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_global_data_maps (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_internal_map (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__add_map (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__variable_offset (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__section_size (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_kversion (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_license (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__check_endianness (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__elf_init (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__elf_finish (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__new (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__init_prog_names (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_object__add_program (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__init (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__exit (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_program__unload (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_prog_linfo__lfind (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_prog_linfo__lfind_addr_func (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_prog_linfo__new (on bpf_[op]*__* in /home/acme/bin/perf) probe_perf:bpf_prog_linfo__free (on bpf_[op]*__* in /home/acme/bin/perf)

You can now use it in all perf tools, such as:

perf record -e probe_perf:bpf_prog_linfo__free -aR sleep 1

[root@quaco ~]# #perf stat -e probe_perf:bpf* perf trace
[root@quaco ~]# cat ~/.perfconfig.OFF
[llvm]
dump-obj = true
clang-opt = -g
[trace]
add_events = /home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.o #add_events = /home/acme/augmented_raw_syscalls.c
#add_events = /home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.c #add_events = /wb/augmented_raw_syscalls.o
show_zeros = yes
show_duration = no
no_inherit = yes
#show_timestamp = no
#show_arg_names = no
args_alignment = 40
#show_prefix = yes

[annotate]

hide_src_code = false
[root@quaco ~]# mv ~/.perfconfig ~/.perfconfig.OFF
mv: cannot stat '/root/.perfconfig': No such file or directory
[root@quaco ~]#
[root@quaco ~]# perf stat -e probe_perf:bpf* perf trace -e /home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.c,open* sleep 1

     0.000 ( 0.009 ms): sleep/16207 openat(dfd: CWD, filename: "/etc/ld.so.cache", flags: RDONLY|CLOEXEC) = 3
     0.031 ( 0.005 ms): sleep/16207 openat(dfd: CWD, filename: "/lib64/libc.so.6", flags: RDONLY|CLOEXEC) = 3
     0.308 ( 0.008 ms): sleep/16207 openat(dfd: CWD, filename: "", flags: RDONLY|CLOEXEC)                 = 3

Performance counter stats for 'perf trace -e /home/acme/git/perf/tools/perf/examples/bpf/augmented_raw_syscalls.c,open* sleep 1':

                 0      probe_perf:bpf_prog_linfo__free
                 0      probe_perf:bpf_prog_linfo__new
                 0      probe_perf:bpf_prog_linfo__lfind_addr_func
                 0      probe_perf:bpf_prog_linfo__lfind
                18      probe_perf:bpf_program__unload
                 9      probe_perf:bpf_program__exit
                 9      probe_perf:bpf_program__init
                 9      probe_perf:bpf_object__add_program
                 1      probe_perf:bpf_object__init_prog_names
                 1      probe_perf:bpf_object__new
                 2      probe_perf:bpf_object__elf_finish
                 1      probe_perf:bpf_object__elf_init
                 1      probe_perf:bpf_object__check_endianness
                 1      probe_perf:bpf_object__init_license
                 1      probe_perf:bpf_object__init_kversion
                 0      probe_perf:bpf_object__section_size
                 0      probe_perf:bpf_object__variable_offset
                 6      probe_perf:bpf_object__add_map
                 0      probe_perf:bpf_object__init_internal_map
                 1      probe_perf:bpf_object__init_global_data_maps
                 1      probe_perf:bpf_object__init_user_maps
                 0      probe_perf:bpf_object__init_user_btf_map
                 1      probe_perf:bpf_object__init_user_btf_maps
                 1      probe_perf:bpf_object__init_maps
                 0      probe_perf:bpf_object__sanitize_btf
                 0      probe_perf:bpf_object__sanitize_btf_ext
                 1      probe_perf:bpf_object__is_btf_mandatory
                 1      probe_perf:bpf_object__init_btf
                 1      probe_perf:bpf_object__sanitize_and_load_btf
                 1      probe_perf:bpf_object__elf_collect
                 8      probe_perf:bpf_object__find_prog_by_idx
                15      probe_perf:bpf_object__find_program_by_title
                 0      probe_perf:bpf_object__shndx_is_data
                17      probe_perf:bpf_object__shndx_is_maps
                17      probe_perf:bpf_object__section_to_libbpf_map_type
                17      probe_perf:bpf_program__record_reloc
                 8      probe_perf:bpf_program__collect_reloc
                 1      probe_perf:bpf_object__probe_name
                 1      probe_perf:bpf_object__probe_global_data
                 1      probe_perf:bpf_object__probe_btf_func
                 1      probe_perf:bpf_object__probe_btf_datasec
                 1      probe_perf:bpf_object__probe_array_mmap
                 1      probe_perf:bpf_object__probe_caps
                 0      probe_perf:bpf_object__reuse_map
                 0      probe_perf:bpf_object__populate_internal_map
                 1      probe_perf:bpf_object__create_maps
                 0      probe_perf:bpf_object__relocate_core
                 0      probe_perf:bpf_program__reloc_text
                 9      probe_perf:bpf_program__relocate
                 1      probe_perf:bpf_object__relocate
                 1      probe_perf:bpf_object__collect_reloc
                 9      probe_perf:bpf_program__load
               154      probe_perf:bpf_program__is_function_storage
                 1      probe_perf:bpf_object__load_progs
                 0      probe_perf:bpf_object__open_xattr
                 0      probe_perf:bpf_object__open
                 0      probe_perf:bpf_object__open_file
                 1      probe_perf:bpf_object__open_mem
                 1      probe_perf:bpf_object__open_buffer
                 1      probe_perf:bpf_object__unload
                 1      probe_perf:bpf_object__load_xattr
                 1      probe_perf:bpf_object__load
                 0      probe_perf:bpf_program__pin_instance
                 0      probe_perf:bpf_program__unpin_instance
                 0      probe_perf:bpf_program__pin
                 0      probe_perf:bpf_program__unpin
                 0      probe_perf:bpf_object__pin_maps
                 0      probe_perf:bpf_object__unpin_maps
                 0      probe_perf:bpf_object__pin_programs
                 0      probe_perf:bpf_object__unpin_programs
                 0      probe_perf:bpf_object__pin
                 1      probe_perf:bpf_object__close
                17      probe_perf:bpf_object__next
                 0      probe_perf:bpf_object__name
                 0      probe_perf:bpf_object__kversion
                 0      probe_perf:bpf_object__btf
                 0      probe_perf:bpf_object__btf_fd
                 0      probe_perf:bpf_object__set_priv
                 0      probe_perf:bpf_object__priv
               159      probe_perf:bpf_program__next
                 0      probe_perf:bpf_program__prev
                 9      probe_perf:bpf_program__set_priv
                27      probe_perf:bpf_program__priv
                 0      probe_perf:bpf_program__set_ifindex
                 9      probe_perf:bpf_program__title
                15      probe_perf:bpf_program__fd
                 0      probe_perf:bpf_program__size
                 0      probe_perf:bpf_program__set_prep
                15      probe_perf:bpf_program__nth_fd
                 0      probe_perf:bpf_program__get_type
                 9      probe_perf:bpf_program__set_type
                 0      probe_perf:bpf_program__is_type
                 0      probe_perf:bpf_program__set_socket_filter
                 0      probe_perf:bpf_program__is_socket_filter
                 0      probe_perf:bpf_program__set_kprobe
                 0      probe_perf:bpf_program__is_kprobe
                 0      probe_perf:bpf_program__set_sched_cls
                 0      probe_perf:bpf_program__is_sched_cls
                 0      probe_perf:bpf_program__set_sched_act
                 0      probe_perf:bpf_program__is_sched_act
                 9      probe_perf:bpf_program__set_tracepoint
                 0      probe_perf:bpf_program__is_tracepoint
                 0      probe_perf:bpf_program__set_raw_tracepoint
                 0      probe_perf:bpf_program__is_raw_tracepoint
                 0      probe_perf:bpf_program__set_xdp
                 0      probe_perf:bpf_program__is_xdp
                 0      probe_perf:bpf_program__set_perf_event
                 0      probe_perf:bpf_program__is_perf_event
                 0      probe_perf:bpf_program__set_tracing
                 0      probe_perf:bpf_program__is_tracing
                 0      probe_perf:bpf_program__get_expected_attach_type
                 0      probe_perf:bpf_program__set_expected_attach_type
                 4      probe_perf:bpf_object__find_map_by_name
                 0      probe_perf:bpf_object__find_map_fd_by_name
                 0      probe_perf:bpf_object__find_map_by_offset
                 0      probe_perf:bpf_program__attach_perf_event
                 0      probe_perf:bpf_program__attach_kprobe
                 0      probe_perf:bpf_program__attach_uprobe
                 0      probe_perf:bpf_program__attach_tracepoint
                 0      probe_perf:bpf_program__attach_raw_tracepoint
                 0      probe_perf:bpf_program__attach_trace
                 0      probe_perf:bpf_program__get_prog_info_linear
                 0      probe_perf:bpf_program__bpil_addr_to_offs
                 0      probe_perf:bpf_program__bpil_offs_to_addr
                 0      probe_perf:bpf_output__printer
                 0      probe_perf:bpf_output__fprintf
       2.375277776 seconds time elapsed

       0.579446000 seconds user
       0.733851000 seconds sys

[root@quaco ~]#

Thanks,

- Arnaldo

Quoted 307 lines
 kernel/trace/trace_kprobe.c |    2 -
 kernel/trace/trace_probe.c  |    5 +-
 kernel/trace/trace_probe.h  |    3 +
 kernel/trace/trace_uprobe.c |  124 ++++++++++++++++++++++++++++---------------
 4 files changed, 86 insertions(+), 48 deletions(-)
kernel/trace/trace_kprobe.c +1 -1
index 7f890262c8a3..3e5f9c7d939c 100644
@@ -290,7 +290,7 @@ static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 	INIT_HLIST_NODE(&tk->rp.kp.hlist);
 	INIT_LIST_HEAD(&tk->rp.kp.list);
 
-	ret = trace_probe_init(&tk->tp, event, group);
+	ret = trace_probe_init(&tk->tp, event, group, 0);
 	if (ret < 0)
 		goto error;
 
kernel/trace/trace_probe.c +3 -2
index 905b10af5d5c..bba18cf44a30 100644
@@ -984,7 +984,7 @@ void trace_probe_cleanup(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group)
+		     const char *group, size_t event_data_size)
 {
 	struct trace_event_call *call;
 	int ret = 0;
@@ -992,7 +992,8 @@ int trace_probe_init(struct trace_probe *tp, const char *event,
 	if (!event || !group)
 		return -EINVAL;
 
-	tp->event = kzalloc(sizeof(struct trace_probe_event), GFP_KERNEL);
+	tp->event = kzalloc(sizeof(struct trace_probe_event) + event_data_size,
+			    GFP_KERNEL);
 	if (!tp->event)
 		return -ENOMEM;
 
kernel/trace/trace_probe.h +2 -1
index 4ee703728aec..03e4e180058d 100644
@@ -230,6 +230,7 @@ struct trace_probe_event {
 	struct trace_event_call		call;
 	struct list_head 		files;
 	struct list_head		probes;
+	char				data[0];
 };
 
 struct trace_probe {
@@ -322,7 +323,7 @@ static inline bool trace_probe_has_single_file(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group);
+		     const char *group, size_t event_data_size);
 void trace_probe_cleanup(struct trace_probe *tp);
 int trace_probe_append(struct trace_probe *tp, struct trace_probe *to);
 void trace_probe_unlink(struct trace_probe *tp);
kernel/trace/trace_uprobe.c +80 -44
index 352073d36585..f66e202fec13 100644
@@ -60,7 +60,6 @@ static struct dyn_event_operations trace_uprobe_ops = {
  */
 struct trace_uprobe {
 	struct dyn_event		devent;
-	struct trace_uprobe_filter	filter;
 	struct uprobe_consumer		consumer;
 	struct path			path;
 	struct inode			*inode;
@@ -264,6 +263,14 @@ process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 }
 NOKPROBE_SYMBOL(process_fetch_insn)
 
+static struct trace_uprobe_filter *
+trace_uprobe_get_filter(struct trace_uprobe *tu)
+{
+	struct trace_probe_event *event = tu->tp.event;
+
+	return (struct trace_uprobe_filter *)&event->data[0];
+}
+
 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 {
 	rwlock_init(&filter->rwlock);
@@ -351,7 +358,8 @@ alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 	if (!tu)
 		return ERR_PTR(-ENOMEM);
 
-	ret = trace_probe_init(&tu->tp, event, group);
+	ret = trace_probe_init(&tu->tp, event, group,
+				sizeof(struct trace_uprobe_filter));
 	if (ret < 0)
 		goto error;
 
@@ -359,7 +367,7 @@ alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 	tu->consumer.handler = uprobe_dispatcher;
 	if (is_ret)
 		tu->consumer.ret_handler = uretprobe_dispatcher;
-	init_trace_uprobe_filter(&tu->filter);
+	init_trace_uprobe_filter(trace_uprobe_get_filter(tu));
 	return tu;
 
 error:
@@ -1067,13 +1075,14 @@ static void __probe_event_disable(struct trace_probe *tp)
 	struct trace_probe *pos;
 	struct trace_uprobe *tu;
 
+	tu = container_of(tp, struct trace_uprobe, tp);
+	WARN_ON(!uprobe_filter_is_empty(trace_uprobe_get_filter(tu)));
+
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 		tu = container_of(pos, struct trace_uprobe, tp);
 		if (!tu->inode)
 			continue;
 
-		WARN_ON(!uprobe_filter_is_empty(&tu->filter));
-
 		uprobe_unregister(tu->inode, tu->offset, &tu->consumer);
 		tu->inode = NULL;
 	}
@@ -1108,7 +1117,7 @@ static int probe_event_enable(struct trace_event_call *call,
 	}
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	WARN_ON(!uprobe_filter_is_empty(&tu->filter));
+	WARN_ON(!uprobe_filter_is_empty(trace_uprobe_get_filter(tu)));
 
 	if (enabled)
 		return 0;
@@ -1205,39 +1214,39 @@ __uprobe_perf_filter(struct trace_uprobe_filter *filter, struct mm_struct *mm)
 }
 
 static inline bool
-uprobe_filter_event(struct trace_uprobe *tu, struct perf_event *event)
+trace_uprobe_filter_event(struct trace_uprobe_filter *filter,
+			  struct perf_event *event)
 {
-	return __uprobe_perf_filter(&tu->filter, event->hw.target->mm);
+	return __uprobe_perf_filter(filter, event->hw.target->mm);
 }
 
-static int uprobe_perf_close(struct trace_uprobe *tu, struct perf_event *event)
+static bool trace_uprobe_filter_remove(struct trace_uprobe_filter *filter,
+				       struct perf_event *event)
 {
 	bool done;
 
-	write_lock(&tu->filter.rwlock);
+	write_lock(&filter->rwlock);
 	if (event->hw.target) {
 		list_del(&event->hw.tp_list);
-		done = tu->filter.nr_systemwide ||
+		done = filter->nr_systemwide ||
 			(event->hw.target->flags & PF_EXITING) ||
-			uprobe_filter_event(tu, event);
+			trace_uprobe_filter_event(filter, event);
 	} else {
-		tu->filter.nr_systemwide--;
-		done = tu->filter.nr_systemwide;
+		filter->nr_systemwide--;
+		done = filter->nr_systemwide;
 	}
-	write_unlock(&tu->filter.rwlock);
-
-	if (!done)
-		return uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
+	write_unlock(&filter->rwlock);
 
-	return 0;
+	return done;
 }
 
-static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
+/* This returns true if the filter always covers target mm */
+static bool trace_uprobe_filter_add(struct trace_uprobe_filter *filter,
+				    struct perf_event *event)
 {
 	bool done;
-	int err;
 
-	write_lock(&tu->filter.rwlock);
+	write_lock(&filter->rwlock);
 	if (event->hw.target) {
 		/*
 		 * event->parent != NULL means copy_process(), we can avoid
@@ -1247,28 +1256,21 @@ static int uprobe_perf_open(struct trace_uprobe *tu, struct perf_event *event)
 		 * attr.enable_on_exec means that exec/mmap will install the
 		 * breakpoints we need.
 		 */
-		done = tu->filter.nr_systemwide ||
+		done = filter->nr_systemwide ||
 			event->parent || event->attr.enable_on_exec ||
-			uprobe_filter_event(tu, event);
-		list_add(&event->hw.tp_list, &tu->filter.perf_events);
+			trace_uprobe_filter_event(filter, event);
+		list_add(&event->hw.tp_list, &filter->perf_events);
 	} else {
-		done = tu->filter.nr_systemwide;
-		tu->filter.nr_systemwide++;
+		done = filter->nr_systemwide;
+		filter->nr_systemwide++;
 	}
-	write_unlock(&tu->filter.rwlock);
+	write_unlock(&filter->rwlock);
 
-	err = 0;
-	if (!done) {
-		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
-		if (err)
-			uprobe_perf_close(tu, event);
-	}
-	return err;
+	return done;
 }
 
-static int uprobe_perf_multi_call(struct trace_event_call *call,
-				  struct perf_event *event,
-		int (*op)(struct trace_uprobe *tu, struct perf_event *event))
+static int uprobe_perf_close(struct trace_event_call *call,
+			     struct perf_event *event)
 {
 	struct trace_probe *pos, *tp;
 	struct trace_uprobe *tu;
@@ -1278,25 +1280,59 @@ static int uprobe_perf_multi_call(struct trace_event_call *call,
 	if (WARN_ON_ONCE(!tp))
 		return -ENODEV;
 
+	tu = container_of(tp, struct trace_uprobe, tp);
+	if (trace_uprobe_filter_remove(trace_uprobe_get_filter(tu), event))
+		return 0;
+
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 		tu = container_of(pos, struct trace_uprobe, tp);
-		ret = op(tu, event);
+		ret = uprobe_apply(tu->inode, tu->offset, &tu->consumer, false);
 		if (ret)
 			break;
 	}
 
 	return ret;
 }
+
+static int uprobe_perf_open(struct trace_event_call *call,
+			    struct perf_event *event)
+{
+	struct trace_probe *pos, *tp;
+	struct trace_uprobe *tu;
+	int err = 0;
+
+	tp = trace_probe_primary_from_call(call);
+	if (WARN_ON_ONCE(!tp))
+		return -ENODEV;
+
+	tu = container_of(tp, struct trace_uprobe, tp);
+	if (trace_uprobe_filter_add(trace_uprobe_get_filter(tu), event))
+		return 0;
+
+	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
+		err = uprobe_apply(tu->inode, tu->offset, &tu->consumer, true);
+		if (err) {
+			uprobe_perf_close(call, event);
+			break;
+		}
+	}
+
+	return err;
+}
+
 static bool uprobe_perf_filter(struct uprobe_consumer *uc,
 				enum uprobe_filter_ctx ctx, struct mm_struct *mm)
 {
+	struct trace_uprobe_filter *filter;
 	struct trace_uprobe *tu;
 	int ret;
 
 	tu = container_of(uc, struct trace_uprobe, consumer);
-	read_lock(&tu->filter.rwlock);
-	ret = __uprobe_perf_filter(&tu->filter, mm);
-	read_unlock(&tu->filter.rwlock);
+	filter = trace_uprobe_get_filter(tu);
+
+	read_lock(&filter->rwlock);
+	ret = __uprobe_perf_filter(filter, mm);
+	read_unlock(&filter->rwlock);
 
 	return ret;
 }
@@ -1419,10 +1455,10 @@ trace_uprobe_register(struct trace_event_call *event, enum trace_reg type,
 		return 0;
 
 	case TRACE_REG_PERF_OPEN:
-		return uprobe_perf_multi_call(event, data, uprobe_perf_open);
+		return uprobe_perf_open(event, data);
 
 	case TRACE_REG_PERF_CLOSE:
-		return uprobe_perf_multi_call(event, data, uprobe_perf_close);
+		return uprobe_perf_close(event, data);
 
 #endif
 	default:
Signature · 1 line
- Arnaldo
maintainer
#3 lore raw
Quoted 2 lines · On Tue, 14 Jan 2020 14:35:35 -0300 Arnaldo Carvalho de Melo…

If you don't mind I can put it into my next perf/urgent pull req to Ingo/Thomas,

I was going to pull this into my urgent tree when my tests finish with my for-next code.

I have one or two other patches I need to apply to urgent as well, so it's not an issue for me to take this.

-- Steve

#4 lore raw

Em Tue, Jan 14, 2020 at 12:44:04PM -0500, Steven Rostedt escreveu:

Quoted 11 lines
Quoted 2 lines · On Tue, 14 Jan 2020 14:35:35 -0300 Arnaldo Carvalho de Melo…

If you don't mind I can put it into my next perf/urgent pull req to Ingo/Thomas,

I was going to pull this into my urgent tree when my tests finish with my for-next code.

I have one or two other patches I need to apply to urgent as well, so it's not an issue for me to take this.

Go ahead, I still need some more time for my tests with the other patches, please just collect my Tested-by,

Thanks,

- Arnaldo

maintainer
#5 lore raw
Quoted 7 lines · On Tue, 14 Jan 2020 14:45:03 -0300 Arnaldo Carvalho de Melo…

Em Tue, Jan 14, 2020 at 12:44:04PM -0500, Steven Rostedt escreveu:

Quoted 6 lines
Quoted 2 lines · On Tue, 14 Jan 2020 14:35:35 -0300 Arnaldo Carvalho de Melo…

If you don't mind I can put it into my next perf/urgent pull req to Ingo/Thomas,

Quoted 9 lines
Quoted 5 lines

I was going to pull this into my urgent tree when my tests finish with my for-next code.

I have one or two other patches I need to apply to urgent as well, so it's not an issue for me to take this.

Go ahead, I still need some more time for my tests with the other patches, please just collect my Tested-by,

Will do!

This also needs to run through my test suite which will take 13 hours more (after my previous one finishes, which took more than 13 because of all the patches I was testing).

-- Steve

author maintainer
#6 lore raw
Quoted 20 lines · On Tue, 14 Jan 2020 13:32:46 -0500 Steven Rostedt <[email protected]> wrote:
Quoted 7 lines · On Tue, 14 Jan 2020 14:45:03 -0300 Arnaldo Carvalho de Melo…

Em Tue, Jan 14, 2020 at 12:44:04PM -0500, Steven Rostedt escreveu:

Quoted 6 lines
Quoted 2 lines · On Tue, 14 Jan 2020 14:35:35 -0300 Arnaldo Carvalho de Melo…

If you don't mind I can put it into my next perf/urgent pull req to Ingo/Thomas,

Quoted 9 lines
Quoted 5 lines

I was going to pull this into my urgent tree when my tests finish with my for-next code.

I have one or two other patches I need to apply to urgent as well, so it's not an issue for me to take this.

Go ahead, I still need some more time for my tests with the other patches, please just collect my Tested-by,

Quoted 5 lines

Will do!

This also needs to run through my test suite which will take 13 hours more (after my previous one finishes, which took more than 13 because of all the patches I was testing).

Thanks Steve and Arnaldo!

This may not take a long time to be tested but requires perf to enable multiple perf event.

Thank you,

Signature · 1 line
Masami Hiramatsu <[email protected]>
#7 lore raw
In reply to Masami Hiramatsu

kernel/trace/trace_probe.h struct trace_probe_event { · line 234

Quoted 10 lines · Masami Hiramatsu
kernel/trace/trace_probe.h +1 -0
index 4ee703728aec..03e4e180058d 100644
@@ -230,6 +230,7 @@ struct trace_probe_event {
 	struct trace_event_call		call;
 	struct list_head 		files;
 	struct list_head		probes;
+	char				data[0];
 };

Would it make sense to make the above:

	struct trace_uprobe_filter	filter[0];

instead? That would ensure that alignment is respected. While I think the current code works by accident.

Quoted 11 lines
kernel/trace/trace_probe.h +7 -0
@@ -264,6 +263,14 @@ process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 }
 NOKPROBE_SYMBOL(process_fetch_insn)
 
+static struct trace_uprobe_filter *
+trace_uprobe_get_filter(struct trace_uprobe *tu)
+{
+	struct trace_probe_event *event = tu->tp.event;
+
+	return (struct trace_uprobe_filter *)&event->data[0];
+}
author maintainer
#8 lore raw
Quoted 15 lines · On Mon, 20 Jan 2020 13:40:22 +0100 Peter Zijlstra <[email protected]> wrote:

kernel/trace/trace_probe.h struct trace_probe_event { · line 234

Quoted 10 lines · Masami Hiramatsu
kernel/trace/trace_probe.h +1 -0
index 4ee703728aec..03e4e180058d 100644
@@ -230,6 +230,7 @@ struct trace_probe_event {
 	struct trace_event_call		call;
 	struct list_head 		files;
 	struct list_head		probes;
+	char				data[0];
 };

Would it make sense to make the above:

Quoted 4 lines
	struct trace_uprobe_filter	filter[0];

instead? That would ensure that alignment is respected. While I think the current code works by accident.

Hmm, if we consider the alignment, shouldn't we allocate the structure with the alignment gap? Currently it just added the sizeof(struct trace_uprobe_filter) when kzalloc().

In this case, I think we should introduce a new data structure, trace_uprobe_event.

Thank you,

Quoted 14 lines
Quoted 12 lines
kernel/trace/trace_probe.h +7 -0
@@ -264,6 +263,14 @@ process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 }
 NOKPROBE_SYMBOL(process_fetch_insn)
 
+static struct trace_uprobe_filter *
+trace_uprobe_get_filter(struct trace_uprobe *tu)
+{
+	struct trace_probe_event *event = tu->tp.event;
+
+	return (struct trace_uprobe_filter *)&event->data[0];
+}
Signature · 1 line
Masami Hiramatsu <[email protected]>
author maintainer
#9 lore raw

PATCH 1/1 tracing/uprobe: Fix to make trace_uprobe_filter alignment safe

In reply to Peter Zijlstra

Commit 99c9a923e97a ("tracing/uprobe: Fix double perf_event linking on multiprobe uprobe") moved trace_uprobe_filter on trace_probe_event. However, since it introduced a flexible data structure with char array and type casting, the alignment of trace_uprobe_filter can be broken.

This changes the type of the array to trace_uprobe_filter data strucure to fix it.

Fixes 99c9a923e97a ("tracing/uprobe: Fix double perf_event linking on multiprobe uprobe")
---
 kernel/trace/trace_kprobe.c |    2 +-
 kernel/trace/trace_probe.c  |    9 ++++++---
 kernel/trace/trace_probe.h  |   10 ++++++++--
 kernel/trace/trace_uprobe.c |   29 +++++++----------------------
 4 files changed, 22 insertions(+), 28 deletions(-)
kernel/trace/trace_kprobe.c +1 -1
index 3e5f9c7d939c..3f54dc2f6e1c 100644
@@ -290,7 +290,7 @@ static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 	INIT_HLIST_NODE(&tk->rp.kp.hlist);
 	INIT_LIST_HEAD(&tk->rp.kp.list);
 
-	ret = trace_probe_init(&tk->tp, event, group, 0);
+	ret = trace_probe_init(&tk->tp, event, group, false);
 	if (ret < 0)
 		goto error;
 
kernel/trace/trace_probe.c +6 -3
index bba18cf44a30..9ae87be422f2 100644
@@ -984,16 +984,19 @@ void trace_probe_cleanup(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group, size_t event_data_size)
+		     const char *group, bool alloc_filter)
 {
 	struct trace_event_call *call;
+	size_t size = sizeof(struct trace_probe_event);
 	int ret = 0;
 
 	if (!event || !group)
 		return -EINVAL;
 
-	tp->event = kzalloc(sizeof(struct trace_probe_event) + event_data_size,
-			    GFP_KERNEL);
+	if (alloc_filter)
+		size += sizeof(struct trace_uprobe_filter);
+
+	tp->event = kzalloc(size, GFP_KERNEL);
 	if (!tp->event)
 		return -ENOMEM;
 
kernel/trace/trace_probe.h +8 -2
index 03e4e180058d..a0ff9e200ef6 100644
@@ -223,6 +223,12 @@ struct probe_arg {
 	const struct fetch_type	*type;	/* Type of this argument */
 };
 
+struct trace_uprobe_filter {
+	rwlock_t		rwlock;
+	int			nr_systemwide;
+	struct list_head	perf_events;
+};
+
 /* Event call and class holder */
 struct trace_probe_event {
 	unsigned int			flags;	/* For TP_FLAG_* */
@@ -230,7 +236,7 @@ struct trace_probe_event {
 	struct trace_event_call		call;
 	struct list_head 		files;
 	struct list_head		probes;
-	char				data[0];
+	struct trace_uprobe_filter	filter[0];
 };
 
 struct trace_probe {
@@ -323,7 +329,7 @@ static inline bool trace_probe_has_single_file(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group, size_t event_data_size);
+		     const char *group, bool alloc_filter);
 void trace_probe_cleanup(struct trace_probe *tp);
 int trace_probe_append(struct trace_probe *tp, struct trace_probe *to);
 void trace_probe_unlink(struct trace_probe *tp);
kernel/trace/trace_uprobe.c +7 -22
index f66e202fec13..21fb512d6a66 100644
@@ -34,12 +34,6 @@ struct uprobe_trace_entry_head {
 #define DATAOF_TRACE_ENTRY(entry, is_return)		\
 	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
 
-struct trace_uprobe_filter {
-	rwlock_t		rwlock;
-	int			nr_systemwide;
-	struct list_head	perf_events;
-};
-
 static int trace_uprobe_create(int argc, const char **argv);
 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
 static int trace_uprobe_release(struct dyn_event *ev);
@@ -263,14 +257,6 @@ process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 }
 NOKPROBE_SYMBOL(process_fetch_insn)
 
-static struct trace_uprobe_filter *
-trace_uprobe_get_filter(struct trace_uprobe *tu)
-{
-	struct trace_probe_event *event = tu->tp.event;
-
-	return (struct trace_uprobe_filter *)&event->data[0];
-}
-
 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 {
 	rwlock_init(&filter->rwlock);
@@ -358,8 +344,7 @@ alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 	if (!tu)
 		return ERR_PTR(-ENOMEM);
 
-	ret = trace_probe_init(&tu->tp, event, group,
-				sizeof(struct trace_uprobe_filter));
+	ret = trace_probe_init(&tu->tp, event, group, true);
 	if (ret < 0)
 		goto error;
 
@@ -367,7 +352,7 @@ alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 	tu->consumer.handler = uprobe_dispatcher;
 	if (is_ret)
 		tu->consumer.ret_handler = uretprobe_dispatcher;
-	init_trace_uprobe_filter(trace_uprobe_get_filter(tu));
+	init_trace_uprobe_filter(tu->tp.event.filter);
 	return tu;
 
 error:
@@ -1076,7 +1061,7 @@ static void __probe_event_disable(struct trace_probe *tp)
 	struct trace_uprobe *tu;
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	WARN_ON(!uprobe_filter_is_empty(trace_uprobe_get_filter(tu)));
+	WARN_ON(!uprobe_filter_is_empty(tu->tp.event.filter));
 
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 		tu = container_of(pos, struct trace_uprobe, tp);
@@ -1117,7 +1102,7 @@ static int probe_event_enable(struct trace_event_call *call,
 	}
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	WARN_ON(!uprobe_filter_is_empty(trace_uprobe_get_filter(tu)));
+	WARN_ON(!uprobe_filter_is_empty(tu->tp.event.filter));
 
 	if (enabled)
 		return 0;
@@ -1281,7 +1266,7 @@ static int uprobe_perf_close(struct trace_event_call *call,
 		return -ENODEV;
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	if (trace_uprobe_filter_remove(trace_uprobe_get_filter(tu), event))
+	if (trace_uprobe_filter_remove(tu->tp.event.filter, event))
 		return 0;
 
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
@@ -1306,7 +1291,7 @@ static int uprobe_perf_open(struct trace_event_call *call,
 		return -ENODEV;
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	if (trace_uprobe_filter_add(trace_uprobe_get_filter(tu), event))
+	if (trace_uprobe_filter_add(tu->tp.event.filter, event))
 		return 0;
 
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
@@ -1328,7 +1313,7 @@ static bool uprobe_perf_filter(struct uprobe_consumer *uc,
 	int ret;
 
 	tu = container_of(uc, struct trace_uprobe, consumer);
-	filter = trace_uprobe_get_filter(tu);
+	filter = tu->tp.event.filter;
 
 	read_lock(&filter->rwlock);
 	ret = __uprobe_perf_filter(filter, mm);
maintainer
#10 lore raw

Re: [PATCH] tracing/uprobe: Fix to make trace_uprobe_filter alignment safe

Quoted 12 lines · On Tue, 21 Jan 2020 16:46:19 +0900 Masami Hiramatsu <[email protected]> wrote:

Commit 99c9a923e97a ("tracing/uprobe: Fix double perf_event linking on multiprobe uprobe") moved trace_uprobe_filter on trace_probe_event. However, since it introduced a flexible data structure with char array and type casting, the alignment of trace_uprobe_filter can be broken.

This changes the type of the array to trace_uprobe_filter data strucure to fix it.

Fixes 99c9a923e97a ("tracing/uprobe: Fix double perf_event linking on multiprobe uprobe")
Quoted 6 lines
---
 kernel/trace/trace_kprobe.c |    2 +-
 kernel/trace/trace_probe.c  |    9 ++++++---
 kernel/trace/trace_probe.h  |   10 ++++++++--
 kernel/trace/trace_uprobe.c |   29 +++++++----------------------
 4 files changed, 22 insertions(+), 28 deletions(-)

I hit the following build error with this:

/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘alloc_trace_uprobe’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:355:39: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?

  init_trace_uprobe_filter(tu->tp.event.filter);
                                       ^
                                       ->
In file included from /work/git/linux-trace.git/arch/x86/include/asm/bug.h:83,
                 from /work/git/linux-trace.git/include/linux/bug.h:5,
                 from /work/git/linux-trace.git/arch/x86/include/asm/paravirt.h:15,
                 from /work/git/linux-trace.git/arch/x86/include/asm/irqflags.h:72,
                 from /work/git/linux-trace.git/include/linux/irqflags.h:16,
                 from /work/git/linux-trace.git/include/linux/rcupdate.h:26,
                 from /work/git/linux-trace.git/include/linux/rbtree.h:22,
                 from /work/git/linux-trace.git/include/linux/key.h:15,
                 from /work/git/linux-trace.git/include/linux/security.h:26,
                 from /work/git/linux-trace.git/kernel/trace/trace_uprobe.c:10:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘__probe_event_disable’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1064:46: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  WARN_ON(!uprobe_filter_is_empty(tu->tp.event.filter));
                                              ^
/work/git/linux-trace.git/include/asm-generic/bug.h:113:25: note: in definition of macro ‘WARN_ON’
  int __ret_warn_on = !!(condition);    \
                         ^~~~~~~~~
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘probe_event_enable’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1105:46: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  WARN_ON(!uprobe_filter_is_empty(tu->tp.event.filter));
                                              ^
/work/git/linux-trace.git/include/asm-generic/bug.h:113:25: note: in definition of macro ‘WARN_ON’
  int __ret_warn_on = !!(condition);    \
                         ^~~~~~~~~
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘uprobe_perf_close’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1269:45: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  if (trace_uprobe_filter_remove(tu->tp.event.filter, event))
                                             ^
                                             ->
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘uprobe_perf_open’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1294:42: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  if (trace_uprobe_filter_add(tu->tp.event.filter, event))
                                          ^
                                          ->
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘uprobe_perf_filter’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1316:23: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  filter = tu->tp.event.filter;
                       ^
                       ->

make[3]: *** [/work/git/linux-trace.git/scripts/Makefile.build:266: kernel/trace/trace_uprobe.o] Error 1 make[3]: *** Waiting for unfinished jobs....

Config attached.

-- Steve

author maintainer
#11 lore raw

Re: [PATCH] tracing/uprobe: Fix to make trace_uprobe_filter alignment safe

Quoted 23 lines · On Tue, 21 Jan 2020 10:50:55 -0500 Steven Rostedt <[email protected]> wrote:
Quoted 12 lines · On Tue, 21 Jan 2020 16:46:19 +0900 Masami Hiramatsu <[email protected]> wrote:

Commit 99c9a923e97a ("tracing/uprobe: Fix double perf_event linking on multiprobe uprobe") moved trace_uprobe_filter on trace_probe_event. However, since it introduced a flexible data structure with char array and type casting, the alignment of trace_uprobe_filter can be broken.

This changes the type of the array to trace_uprobe_filter data strucure to fix it.

Fixes 99c9a923e97a ("tracing/uprobe: Fix double perf_event linking on multiprobe uprobe")
Quoted 6 lines
---
 kernel/trace/trace_kprobe.c |    2 +-
 kernel/trace/trace_probe.c  |    9 ++++++---
 kernel/trace/trace_probe.h  |   10 ++++++++--
 kernel/trace/trace_uprobe.c |   29 +++++++----------------------
 4 files changed, 22 insertions(+), 28 deletions(-)
Quoted 1 line

I hit the following build error with this:

Hmm, I missed something... let me check and update it.

Thank you,

Quoted 52 lines

/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘alloc_trace_uprobe’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:355:39: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?

  init_trace_uprobe_filter(tu->tp.event.filter);
                                       ^
                                       ->
In file included from /work/git/linux-trace.git/arch/x86/include/asm/bug.h:83,
                 from /work/git/linux-trace.git/include/linux/bug.h:5,
                 from /work/git/linux-trace.git/arch/x86/include/asm/paravirt.h:15,
                 from /work/git/linux-trace.git/arch/x86/include/asm/irqflags.h:72,
                 from /work/git/linux-trace.git/include/linux/irqflags.h:16,
                 from /work/git/linux-trace.git/include/linux/rcupdate.h:26,
                 from /work/git/linux-trace.git/include/linux/rbtree.h:22,
                 from /work/git/linux-trace.git/include/linux/key.h:15,
                 from /work/git/linux-trace.git/include/linux/security.h:26,
                 from /work/git/linux-trace.git/kernel/trace/trace_uprobe.c:10:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘__probe_event_disable’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1064:46: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  WARN_ON(!uprobe_filter_is_empty(tu->tp.event.filter));
                                              ^
/work/git/linux-trace.git/include/asm-generic/bug.h:113:25: note: in definition of macro ‘WARN_ON’
  int __ret_warn_on = !!(condition);    \
                         ^~~~~~~~~
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘probe_event_enable’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1105:46: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  WARN_ON(!uprobe_filter_is_empty(tu->tp.event.filter));
                                              ^
/work/git/linux-trace.git/include/asm-generic/bug.h:113:25: note: in definition of macro ‘WARN_ON’
  int __ret_warn_on = !!(condition);    \
                         ^~~~~~~~~
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘uprobe_perf_close’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1269:45: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  if (trace_uprobe_filter_remove(tu->tp.event.filter, event))
                                             ^
                                             ->
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘uprobe_perf_open’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1294:42: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  if (trace_uprobe_filter_add(tu->tp.event.filter, event))
                                          ^
                                          ->
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c: In function ‘uprobe_perf_filter’:
/work/git/linux-trace.git/kernel/trace/trace_uprobe.c:1316:23: error: ‘tu->tp.event’ is a pointer; did you mean to use ‘->’?
  filter = tu->tp.event.filter;
                       ^
                       ->

make[3]: *** [/work/git/linux-trace.git/scripts/Makefile.build:266: kernel/trace/trace_uprobe.o] Error 1 make[3]: *** Waiting for unfinished jobs....

Config attached.

-- Steve

Signature · 1 line
Masami Hiramatsu <[email protected]>
author maintainer
#12 lore raw

PATCH v2 1/1 tracing/uprobe: Fix to make trace_uprobe_filter alignment safe

Commit 99c9a923e97a ("tracing/uprobe: Fix double perf_event linking on multiprobe uprobe") moved trace_uprobe_filter on trace_probe_event. However, since it introduced a flexible data structure with char array and type casting, the alignment of trace_uprobe_filter can be broken.

This changes the type of the array to trace_uprobe_filter data strucure to fix it.

Fixes 99c9a923e97a ("tracing/uprobe: Fix double perf_event linking on multiprobe uprobe")

---
Changes in v2:
- Fix build errors against pointer-field mistaking.

---
 kernel/trace/trace_kprobe.c |    2 +-
 kernel/trace/trace_probe.c  |    9 ++++++---
 kernel/trace/trace_probe.h  |   10 ++++++++--
 kernel/trace/trace_uprobe.c |   29 +++++++----------------------
 4 files changed, 22 insertions(+), 28 deletions(-)
kernel/trace/trace_kprobe.c +1 -1
index 3e5f9c7d939c..3f54dc2f6e1c 100644
@@ -290,7 +290,7 @@ static struct trace_kprobe *alloc_trace_kprobe(const char *group,
 	INIT_HLIST_NODE(&tk->rp.kp.hlist);
 	INIT_LIST_HEAD(&tk->rp.kp.list);
 
-	ret = trace_probe_init(&tk->tp, event, group, 0);
+	ret = trace_probe_init(&tk->tp, event, group, false);
 	if (ret < 0)
 		goto error;
 
kernel/trace/trace_probe.c +6 -3
index bba18cf44a30..9ae87be422f2 100644
@@ -984,16 +984,19 @@ void trace_probe_cleanup(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group, size_t event_data_size)
+		     const char *group, bool alloc_filter)
 {
 	struct trace_event_call *call;
+	size_t size = sizeof(struct trace_probe_event);
 	int ret = 0;
 
 	if (!event || !group)
 		return -EINVAL;
 
-	tp->event = kzalloc(sizeof(struct trace_probe_event) + event_data_size,
-			    GFP_KERNEL);
+	if (alloc_filter)
+		size += sizeof(struct trace_uprobe_filter);
+
+	tp->event = kzalloc(size, GFP_KERNEL);
 	if (!tp->event)
 		return -ENOMEM;
 
kernel/trace/trace_probe.h +8 -2
index 03e4e180058d..a0ff9e200ef6 100644
@@ -223,6 +223,12 @@ struct probe_arg {
 	const struct fetch_type	*type;	/* Type of this argument */
 };
 
+struct trace_uprobe_filter {
+	rwlock_t		rwlock;
+	int			nr_systemwide;
+	struct list_head	perf_events;
+};
+
 /* Event call and class holder */
 struct trace_probe_event {
 	unsigned int			flags;	/* For TP_FLAG_* */
@@ -230,7 +236,7 @@ struct trace_probe_event {
 	struct trace_event_call		call;
 	struct list_head 		files;
 	struct list_head		probes;
-	char				data[0];
+	struct trace_uprobe_filter	filter[0];
 };
 
 struct trace_probe {
@@ -323,7 +329,7 @@ static inline bool trace_probe_has_single_file(struct trace_probe *tp)
 }
 
 int trace_probe_init(struct trace_probe *tp, const char *event,
-		     const char *group, size_t event_data_size);
+		     const char *group, bool alloc_filter);
 void trace_probe_cleanup(struct trace_probe *tp);
 int trace_probe_append(struct trace_probe *tp, struct trace_probe *to);
 void trace_probe_unlink(struct trace_probe *tp);
kernel/trace/trace_uprobe.c +7 -22
index f66e202fec13..2619bc5ed520 100644
@@ -34,12 +34,6 @@ struct uprobe_trace_entry_head {
 #define DATAOF_TRACE_ENTRY(entry, is_return)		\
 	((void*)(entry) + SIZEOF_TRACE_ENTRY(is_return))
 
-struct trace_uprobe_filter {
-	rwlock_t		rwlock;
-	int			nr_systemwide;
-	struct list_head	perf_events;
-};
-
 static int trace_uprobe_create(int argc, const char **argv);
 static int trace_uprobe_show(struct seq_file *m, struct dyn_event *ev);
 static int trace_uprobe_release(struct dyn_event *ev);
@@ -263,14 +257,6 @@ process_fetch_insn(struct fetch_insn *code, struct pt_regs *regs, void *dest,
 }
 NOKPROBE_SYMBOL(process_fetch_insn)
 
-static struct trace_uprobe_filter *
-trace_uprobe_get_filter(struct trace_uprobe *tu)
-{
-	struct trace_probe_event *event = tu->tp.event;
-
-	return (struct trace_uprobe_filter *)&event->data[0];
-}
-
 static inline void init_trace_uprobe_filter(struct trace_uprobe_filter *filter)
 {
 	rwlock_init(&filter->rwlock);
@@ -358,8 +344,7 @@ alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 	if (!tu)
 		return ERR_PTR(-ENOMEM);
 
-	ret = trace_probe_init(&tu->tp, event, group,
-				sizeof(struct trace_uprobe_filter));
+	ret = trace_probe_init(&tu->tp, event, group, true);
 	if (ret < 0)
 		goto error;
 
@@ -367,7 +352,7 @@ alloc_trace_uprobe(const char *group, const char *event, int nargs, bool is_ret)
 	tu->consumer.handler = uprobe_dispatcher;
 	if (is_ret)
 		tu->consumer.ret_handler = uretprobe_dispatcher;
-	init_trace_uprobe_filter(trace_uprobe_get_filter(tu));
+	init_trace_uprobe_filter(tu->tp.event->filter);
 	return tu;
 
 error:
@@ -1076,7 +1061,7 @@ static void __probe_event_disable(struct trace_probe *tp)
 	struct trace_uprobe *tu;
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	WARN_ON(!uprobe_filter_is_empty(trace_uprobe_get_filter(tu)));
+	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
 
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
 		tu = container_of(pos, struct trace_uprobe, tp);
@@ -1117,7 +1102,7 @@ static int probe_event_enable(struct trace_event_call *call,
 	}
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	WARN_ON(!uprobe_filter_is_empty(trace_uprobe_get_filter(tu)));
+	WARN_ON(!uprobe_filter_is_empty(tu->tp.event->filter));
 
 	if (enabled)
 		return 0;
@@ -1281,7 +1266,7 @@ static int uprobe_perf_close(struct trace_event_call *call,
 		return -ENODEV;
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	if (trace_uprobe_filter_remove(trace_uprobe_get_filter(tu), event))
+	if (trace_uprobe_filter_remove(tu->tp.event->filter, event))
 		return 0;
 
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
@@ -1306,7 +1291,7 @@ static int uprobe_perf_open(struct trace_event_call *call,
 		return -ENODEV;
 
 	tu = container_of(tp, struct trace_uprobe, tp);
-	if (trace_uprobe_filter_add(trace_uprobe_get_filter(tu), event))
+	if (trace_uprobe_filter_add(tu->tp.event->filter, event))
 		return 0;
 
 	list_for_each_entry(pos, trace_probe_probe_list(tp), list) {
@@ -1328,7 +1313,7 @@ static bool uprobe_perf_filter(struct uprobe_consumer *uc,
 	int ret;
 
 	tu = container_of(uc, struct trace_uprobe, consumer);
-	filter = trace_uprobe_get_filter(tu);
+	filter = tu->tp.event->filter;
 
 	read_lock(&filter->rwlock);
 	ret = __uprobe_perf_filter(filter, mm);