Skip to contentSkip to conversation

tracing: Add dynamic strings for synthetic events

PATCH Superseded Tom Zanussi 3 participants 10 messages Sep 30, 2020 – Oct 2, 2020

Subsystems (2)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover tracing: Add dynamic strings for synthetic events

Hi,

This is v2 of the dynamic string support for synthetic events.

v1 was missing some very important code in the in-kernel API functions that should have reserved room for dynamic strings. This adds it along also some additional test code in the synth_event_gen_test module that tests the dynamic strings for those APIs.

I also added Axel's changelog text to the 'tracing: Fix parse_synth_field() error handling' patch and used ssize_t as he also suggested. Thanks, Axel! I also noticed that the original patch didn't use <= as it should have, so changed that as well.

Neither my command-line tests or the synth_event_gen_test module are showing any problems at this point for me...

Tom

Original v1 text:

This patchset adds support for dynamic strings for synthetic events, as requested by Axel Rasmussen.

Actually, the first two patches should be applied in any case - the first just changes the current max string length and the second fixes a bug I found while testing.

It works fine for my simple test cases, but I still need to do a lot more testing, especially of the in-kernel API parts, which don't affect Axel's use case.

Anyway, Axel, please try it out and send me your Tested-by: if it works for you, and your broken testcase(s) if it doesn't. ;-)

Thanks,

Tom

The following changes since commit 106c076d23cca67c959a6fd1ccadb5b3ef01ddc9:

mmap_lock: add tracepoints around lock acquisition (2020-09-23 08:48:08 -0500)

are available in the Git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/zanussi/linux-trace.git ftrace/synth-dynstring-v2

Tom Zanussi (3):
tracing: Change STR_VAR_MAX_LEN
tracing: Fix parse_synth_field() error handling
tracing: Add support for dynamic strings to synthetic events

 Documentation/trace/events.rst      |  15 +-
 Documentation/trace/histogram.rst   |  18 +++
 kernel/trace/synth_event_gen_test.c |  18 ++-
 kernel/trace/trace_events_hist.c    |   9 ++
 kernel/trace/trace_events_synth.c   | 243 ++++++++++++++++++++++++----
 kernel/trace/trace_synth.h          |   6 +-
 6 files changed, 269 insertions(+), 40 deletions(-)
Signature · 1 line
2.17.1
#4 lore raw

PATCH v2 2/3 tracing: Fix parse_synth_field() error handling

synth_field_size() returns either a positive size or an error (zero or a negative value). However, the existing code assumes the only error value is 0. It doesn't handle negative error codes, as it assigns directly to field->size (a size_t; unsigned), thereby interpreting the error code as a valid size instead.

Do the test before assignment to field->size.

[ [email protected]: changelog addition, first paragraph above ]

---
 kernel/trace/trace_events_synth.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
kernel/trace/trace_events_synth.c +4 -2
index a9cd7793f7ea..fa8a99828f41 100644
@@ -465,6 +465,7 @@ static struct synth_field *parse_synth_field(int argc, const char **argv,
 	struct synth_field *field;
 	const char *prefix = NULL, *field_type = argv[0], *field_name, *array;
 	int len, ret = 0;
+	ssize_t size;
 
 	if (field_type[0] == ';')
 		field_type++;
@@ -520,11 +521,12 @@ static struct synth_field *parse_synth_field(int argc, const char **argv,
 			field->type[len - 1] = '\0';
 	}
 
-	field->size = synth_field_size(field->type);
-	if (!field->size) {
+	size = synth_field_size(field->type);
+	if (size <= 0) {
 		ret = -EINVAL;
 		goto free;
 	}
+	field->size = size;
 
 	if (synth_field_is_string(field->type))
 		field->is_string = true;
Signature · 1 line
2.17.1
#7 lore raw

PATCH v2 3/3 tracing: Add support for dynamic strings to synthetic events

In reply to Tom Zanussi

Currently, sythetic events only support static string fields such as:

  # echo 'test_latency u64 lat; char somename[32]' > /sys/kernel/debug/tracing/synthetic_events

Which is fine, but wastes a lot of space in the event.

It also prevents the most commonly-defined strings in the existing trace events e.g. those defined using __string(), from being passed to synthetic events via the trace() action.

With this change, synthetic events with dynamic fields can be defined:

  # echo 'test_latency u64 lat; char somename[]' > /sys/kernel/debug/tracing/synthetic_events

And the trace() action can be used to generate events using either dynamic or static strings:

  # echo 'hist:keys=name:lat=common_timestamp.usecs-$ts0:onmatch(sys.event).test_latency($lat,name)' > /sys/kernel/debug/tracing/events

The synthetic event dynamic strings are implemented in the same way as the existing __data_loc strings and appear as such in the format file.

---
 Documentation/trace/events.rst      |  15 +-
 Documentation/trace/histogram.rst   |  18 +++
 kernel/trace/synth_event_gen_test.c |  18 ++-
 kernel/trace/trace_events_hist.c    |   9 ++
 kernel/trace/trace_events_synth.c   | 239 ++++++++++++++++++++++++----
 kernel/trace/trace_synth.h          |   4 +
 6 files changed, 265 insertions(+), 38 deletions(-)
Documentation/trace/events.rst +13 -2
index f792b1959a33..2a5aa48eff6c 100644
@@ -589,8 +589,19 @@ name::
         { .type = "int",                .name = "my_int_field" },
   };
 
-See synth_field_size() for available types. If field_name contains [n]
-the field is considered to be an array.
+See synth_field_size() for available types.
+
+If field_name contains [n], the field is considered to be a static array.
+
+If field_names contains[] (no subscript), the field is considered to
+be a dynamic array, which will only take as much space in the event as
+is required to hold the array.
+
+Because space for an event is reserved before assigning field values
+to the event, using dynamic arrays implies that the piecewise
+in-kernel API described below can't be used with dynamic arrays.  The
+other non-piecewise in-kernel APIs can, however, be used with dynamic
+arrays.
 
 If the event is created from within a module, a pointer to the module
 must be passed to synth_event_create().  This will ensure that the
Documentation/trace/histogram.rst +18 -0
index 8408670d0328..b573604deabd 100644
@@ -1776,6 +1776,24 @@ consisting of the name of the new event along with one or more
 variables and their types, which can be any valid field type,
 separated by semicolons, to the tracing/synthetic_events file.
 
+See synth_field_size() for available types.
+
+If field_name contains [n], the field is considered to be a static array.
+
+If field_names contains[] (no subscript), the field is considered to
+be a dynamic array, which will only take as much space in the event as
+is required to hold the array.
+
+A string field can be specified using either the static notation:
+
+  char name[32];
+
+Or the dynamic:
+
+  char name[];
+
+The size limit for either is 256.
+
 For instance, the following creates a new event named 'wakeup_latency'
 with 3 fields: lat, pid, and prio.  Each of those fields is simply a
 variable reference to a variable on another event::
kernel/trace/synth_event_gen_test.c +12 -6
index 7d56d621ffea..edd912cd14aa 100644
@@ -242,9 +242,11 @@ static struct synth_field_desc create_synth_test_fields[] = {
 	{ .type = "pid_t",		.name = "next_pid_field" },
 	{ .type = "char[16]",		.name = "next_comm_field" },
 	{ .type = "u64",		.name = "ts_ns" },
+	{ .type = "char[]",		.name = "dynstring_field_1" },
 	{ .type = "u64",		.name = "ts_ms" },
 	{ .type = "unsigned int",	.name = "cpu" },
 	{ .type = "char[64]",		.name = "my_string_field" },
+	{ .type = "char[]",		.name = "dynstring_field_2" },
 	{ .type = "int",		.name = "my_int_field" },
 };
 
@@ -254,7 +256,7 @@ static struct synth_field_desc create_synth_test_fields[] = {
  */
 static int __init test_create_synth_event(void)
 {
-	u64 vals[7];
+	u64 vals[9];
 	int ret;
 
 	/* Create the create_synth_test event with the fields above */
@@ -292,10 +294,12 @@ static int __init test_create_synth_event(void)
 	vals[0] = 777;			/* next_pid_field */
 	vals[1] = (u64)(long)"tiddlywinks";	/* next_comm_field */
 	vals[2] = 1000000;		/* ts_ns */
-	vals[3] = 1000;			/* ts_ms */
-	vals[4] = raw_smp_processor_id(); /* cpu */
-	vals[5] = (u64)(long)"thneed";	/* my_string_field */
-	vals[6] = 398;			/* my_int_field */
+	vals[3] = (u64)(long)"xrayspecs";	/* dynstring_field_1 */
+	vals[4] = 1000;			/* ts_ms */
+	vals[5] = raw_smp_processor_id(); /* cpu */
+	vals[6] = (u64)(long)"thneed";	/* my_string_field */
+	vals[7] = (u64)(long)"kerplunk";	/* dynstring_field_2 */
+	vals[8] = 398;			/* my_int_field */
 
 	/* Now generate a create_synth_test event */
 	ret = synth_event_trace_array(create_synth_test, vals, ARRAY_SIZE(vals));
@@ -422,13 +426,15 @@ static int __init test_trace_synth_event(void)
 	int ret;
 
 	/* Trace some bogus values just for testing */
-	ret = synth_event_trace(create_synth_test, 7,	/* number of values */
+	ret = synth_event_trace(create_synth_test, 9,	/* number of values */
 				(u64)444,		/* next_pid_field */
 				(u64)(long)"clackers",	/* next_comm_field */
 				(u64)1000000,		/* ts_ns */
+				(u64)(long)"viewmaster",/* dynstring_field_1 */
 				(u64)1000,		/* ts_ms */
 				(u64)raw_smp_processor_id(), /* cpu */
 				(u64)(long)"Thneed",	/* my_string_field */
+				(u64)(long)"yoyos",	/* dynstring_field_2 */
 				(u64)999);		/* my_int_field */
 	return ret;
 }
kernel/trace/trace_events_hist.c +9 -0
index 1b2ef6490229..ee6b27c82350 100644
@@ -3279,6 +3279,15 @@ static int check_synth_field(struct synth_event *event,
 
 	field = event->fields[field_pos];
 
+	/*
+	 * A dynamic string synth field can accept static or
+	 * dynamic. A static string synth field can only accept a
+	 * same-sized static string, which is checked for later.
+	 */
+	if (strstr(hist_field->type, "char[") && field->is_string
+	    && field->is_dynamic)
+		return 0;
+
 	if (strcmp(field->type, hist_field->type) != 0) {
 		if (field->size != hist_field->size ||
 		    field->is_signed != hist_field->is_signed)
kernel/trace/trace_events_synth.c +209 -30

413 lines not on this page · raw message

kernel/trace/trace_synth.h +4 -0
index 5166705d1556..6e146b959dcd 100644
@@ -16,6 +16,8 @@ struct synth_field {
 	unsigned int offset;
 	bool is_signed;
 	bool is_string;
+	bool is_dynamic;
+	bool field_pos;
 };
 
 struct synth_event {
@@ -24,6 +26,8 @@ struct synth_event {
 	char					*name;
 	struct synth_field			**fields;
 	unsigned int				n_fields;
+	struct synth_field			**dynamic_fields;
+	unsigned int				n_dynamic_fields;
 	unsigned int				n_u64;
 	struct trace_event_class		class;
 	struct trace_event_call			call;
Signature · 1 line
2.17.1
#1 lore raw

PATCH v2 1/3 tracing: Change STR_VAR_MAX_LEN

In reply to Tom Zanussi

32 is too small for this value, and anyway it makes more sense to use MAX_FILTER_STR_VAL, as this is also the value used for variable-length __strings.

---
 kernel/trace/trace_synth.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/trace/trace_synth.h +1 -1
index ac35c45207c4..5166705d1556 100644
@@ -7,7 +7,7 @@
 #define SYNTH_SYSTEM		"synthetic"
 #define SYNTH_FIELDS_MAX	32
 
-#define STR_VAR_LEN_MAX		32 /* must be multiple of sizeof(u64) */
+#define STR_VAR_LEN_MAX		MAX_FILTER_STR_VAL /* must be multiple of sizeof(u64) */
 
 struct synth_field {
 	char *type;
Signature · 1 line
2.17.1
Tested-by on v2 1/3
#2 lore raw

Re: [PATCH v2 1/3] tracing: Change STR_VAR_MAX_LEN

I still get the same WARN_ON behavior when trying to use bpftrace. But, I spent some time today reading through that call path, and at this point I'm convinced that it's the version of bpftrace I'm using which is slightly broken, not the kernel in this case. (To be fair, I'm trying to use some unreleased tip-of-tree version of it, with some not-yet-merged patches, and some extra hacks on top of all that, so it's not particularly surprising...)

In my experiments with just the synthetic event + histogram triggers, this patchset works as expected for my use case.

So (for the whole series, not just this one patch):

Quoted 26 lines · Tom Zanussi

32 is too small for this value, and anyway it makes more sense to use MAX_FILTER_STR_VAL, as this is also the value used for variable-length __strings.

---
 kernel/trace/trace_synth.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/trace/trace_synth.h +1 -2
index ac35c45207c4..5166705d1556 100644
@@ -7,7 +7,7 @@
 #define SYNTH_SYSTEM           "synthetic"
 #define SYNTH_FIELDS_MAX       32
 
-#define STR_VAR_LEN_MAX                32 /* must be multiple of sizeof(u64) */
+#define STR_VAR_LEN_MAX                MAX_FILTER_STR_VAL /* must be multiple of sizeof(u64) */
 
 struct synth_field {
        char *type;
--

2.17.1

maintainer Reviewed-by on v2 2/3
#5 lore raw

Re: [PATCH v2 2/3] tracing: Fix parse_synth_field() error handling

In reply to Tom Zanussi

Hi Tom,

Quoted 7 lines · On Wed, 30 Sep 2020 13:40:51 -0500 Tom Zanussi <[email protected]> wrote:

synth_field_size() returns either a positive size or an error (zero or a negative value). However, the existing code assumes the only error value is 0. It doesn't handle negative error codes, as it assigns directly to field->size (a size_t; unsigned), thereby interpreting the error code as a valid size instead.

Do the test before assignment to field->size.

This looks good to me. And I guess it is a kind of bugfix which should be backported? (doesn't this need a Fixes tag?)

Thank you,

Quoted 38 lines

[ [email protected]: changelog addition, first paragraph above ]

---
 kernel/trace/trace_events_synth.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
kernel/trace/trace_events_synth.c +4 -2
index a9cd7793f7ea..fa8a99828f41 100644
@@ -465,6 +465,7 @@ static struct synth_field *parse_synth_field(int argc, const char **argv,
 	struct synth_field *field;
 	const char *prefix = NULL, *field_type = argv[0], *field_name, *array;
 	int len, ret = 0;
+	ssize_t size;
 
 	if (field_type[0] == ';')
 		field_type++;
@@ -520,11 +521,12 @@ static struct synth_field *parse_synth_field(int argc, const char **argv,
 			field->type[len - 1] = '\0';
 	}
 
-	field->size = synth_field_size(field->type);
-	if (!field->size) {
+	size = synth_field_size(field->type);
+	if (size <= 0) {
 		ret = -EINVAL;
 		goto free;
 	}
+	field->size = size;
 
 	if (synth_field_is_string(field->type))
 		field->is_string = true;
Signature · 1 line
2.17.1
Signature · 1 line
Masami Hiramatsu <[email protected]>
#6 lore raw

Re: [PATCH v2 2/3] tracing: Fix parse_synth_field() error handling

Hi Masami,

Quoted 14 lines · Masami Hiramatsu

Hi Tom,

Quoted 9 lines · On Wed, 30 Sep 2020 13:40:51 -0500 Tom Zanussi <[email protected]> wrote:

synth_field_size() returns either a positive size or an error (zero or
a negative value). However, the existing code assumes the only error
value is 0. It doesn't handle negative error codes, as it assigns directly to field->size (a size_t; unsigned), thereby interpreting the
error code as a valid size instead.

Quoted 5 lines
Quoted 1 line

Do the test before assignment to field->size.

This looks good to me. And I guess it is a kind of bugfix which should be backported?
(doesn't this need a Fixes tag?)

Yep, thanks for pointing that out.

Quoted 3 lines

Thanks!

Tom

Quoted 47 lines

Thank you,

Quoted 43 lines

[ [email protected]: changelog addition, first paragraph above ]

---
 kernel/trace/trace_events_synth.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
kernel/trace/trace_events_synth.c +1 -0
index a9cd7793f7ea..fa8a99828f41 100644
@@ -465,6 +465,7 @@ static struct synth_field
 *parse_synth_field(int argc, const char **argv,
 	struct synth_field *field;
 	const char *prefix = NULL, *field_type = argv[0], *field_name,
 *array;
 	int len, ret = 0;
+	ssize_t size;
 
kernel/trace/trace_events_synth.c +3 -2
 	if (field_type[0] == ';')
 		field_type++;
@@ -520,11 +521,12 @@ static struct synth_field
 *parse_synth_field(int argc, const char **argv,
 			field->type[len - 1] = '\0';
 	}
 
-	field->size = synth_field_size(field->type);
-	if (!field->size) {
+	size = synth_field_size(field->type);
+	if (size <= 0) {
 		ret = -EINVAL;
 		goto free;
 	}
+	field->size = size;
 
 	if (synth_field_is_string(field->type))

field->is_string = true;

Signature · 1 line
2.17.1
#3 lore raw

Re: [PATCH v2 1/3] tracing: Change STR_VAR_MAX_LEN

In reply to Axel Rasmussen

Hi Axel,

Quoted 9 lines · Axel Rasmussen

I still get the same WARN_ON behavior when trying to use bpftrace. But, I spent some time today reading through that call path, and at this point I'm convinced that it's the version of bpftrace I'm using which is slightly broken, not the kernel in this case. (To be fair, I'm trying to use some unreleased tip-of-tree version of it, with some
not-yet-merged patches, and some extra hacks on top of all that, so it's not particularly surprising...)

Quoted 6 lines

In my experiments with just the synthetic event + histogram triggers, this patchset works as expected for my use case.

So (for the whole series, not just this one patch):

Thanks!

Tom

Quoted 34 lines
Quoted 31 lines · Tom Zanussi

32 is too small for this value, and anyway it makes more sense to use
MAX_FILTER_STR_VAL, as this is also the value used for variable- length
__strings.

---
 kernel/trace/trace_synth.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/trace/trace_synth.h +0 -1
index ac35c45207c4..5166705d1556 100644
@@ -7,7 +7,7 @@
 #define SYNTH_SYSTEM           "synthetic"
 #define SYNTH_FIELDS_MAX       32
 
-#define STR_VAR_LEN_MAX                32 /* must be multiple of

sizeof(u64) */
+#define STR_VAR_LEN_MAX MAX_FILTER_STR_VAL /* must
be multiple of sizeof(u64) */

 struct synth_field {
        char *type;

--
2.17.1

maintainer
#8 lore raw

Re: [PATCH v2 3/3] tracing: Add support for dynamic strings to synthetic events

In reply to Tom Zanussi

Hi Tom,

Quoted 6 lines · On Wed, 30 Sep 2020 13:40:52 -0500 Tom Zanussi <[email protected]> wrote:

Currently, sythetic events only support static string fields such as:

  # echo 'test_latency u64 lat; char somename[32]' > /sys/kernel/debug/tracing/synthetic_events

Which is fine, but wastes a lot of space in the event.

Quoted 7 lines

It also prevents the most commonly-defined strings in the existing trace events e.g. those defined using __string(), from being passed to synthetic events via the trace() action.

With this change, synthetic events with dynamic fields can be defined:

  # echo 'test_latency u64 lat; char somename[]' > /sys/kernel/debug/tracing/synthetic_events

Could you add a testcase (and update existing one) of ftracetest for this new feature too?

Quoted 9 lines

And the trace() action can be used to generate events using either dynamic or static strings:

  # echo 'hist:keys=name:lat=common_timestamp.usecs-$ts0:onmatch(sys.event).test_latency($lat,name)' > /sys/kernel/debug/tracing/events

The synthetic event dynamic strings are implemented in the same way as the existing __data_loc strings and appear as such in the format file.

Quoted 8 lines
---
 Documentation/trace/events.rst      |  15 +-
 Documentation/trace/histogram.rst   |  18 +++
 kernel/trace/synth_event_gen_test.c |  18 ++-
 kernel/trace/trace_events_hist.c    |   9 ++
 kernel/trace/trace_events_synth.c   | 239 ++++++++++++++++++++++++----
 kernel/trace/trace_synth.h          |   4 +

And you might also need to update tracefs/README so that user can check whether the kernel supports dynamic string or not.

Thank you,

Signature · 1 line
Masami Hiramatsu <[email protected]>
#9 lore raw

Re: [PATCH v2 3/3] tracing: Add support for dynamic strings to synthetic events

Hi Masami,

Quoted 18 lines · Masami Hiramatsu

Hi Tom,

Quoted 13 lines · On Wed, 30 Sep 2020 13:40:52 -0500 Tom Zanussi <[email protected]> wrote:

Currently, sythetic events only support static string fields such as:

  # echo 'test_latency u64 lat; char somename[32]' >
/sys/kernel/debug/tracing/synthetic_events

Which is fine, but wastes a lot of space in the event.

It also prevents the most commonly-defined strings in the existing trace events e.g. those defined using __string(), from being passed to
synthetic events via the trace() action.

Quoted 9 lines
Quoted 5 lines

With this change, synthetic events with dynamic fields can be defined:

  # echo 'test_latency u64 lat; char somename[]' >
/sys/kernel/debug/tracing/synthetic_events

Could you add a testcase (and update existing one) of ftracetest for this new feature too?

Yes, I'll add it in the next version.

Quoted 23 lines
Quoted 14 lines

And the trace() action can be used to generate events using either dynamic or static strings:

  # echo 'hist:keys=name:lat=common_timestamp.usecs-
$ts0:onmatch(sys.event).test_latency($lat,name)' >
/sys/kernel/debug/tracing/events

The synthetic event dynamic strings are implemented in the same way as
the existing __data_loc strings and appear as such in the format file.

Quoted 8 lines
---
 Documentation/trace/events.rst      |  15 +-
 Documentation/trace/histogram.rst   |  18 +++
 kernel/trace/synth_event_gen_test.c |  18 ++-
 kernel/trace/trace_events_hist.c    |   9 ++
 kernel/trace/trace_events_synth.c   | 239
++++++++++++++++++++++++----
 kernel/trace/trace_synth.h          |   4 +
Quoted 3 lines

And you might also need to update tracefs/README so that user can check whether the kernel supports dynamic string or not.

Yeah, good to add regardless, will do.

Thanks,

Tom