[PATCH net-next] trace_events_filter: conditional trace event (tcp_probe full=0)
From: Md. Islam
Date: Mon Feb 12 2018 - 00:08:54 EST
Recently tcp_probe kernel module has been replaced by trace_event. Old
tcp_probe had full=0 option where it only takes a snapshot only when
congestion window is changed. However I did not find such
functionality in trace_event. This is why I implemented this
"conditional trace_event" where a snapshot is taken only if some field
is changed. For instance, following filter will record a snapshot only
if snd_cwnd field is changed from the previous snapshot:
cd /sys/kernel/debug/tracing
echo 1 > events/tcp/tcp_probe/enable
echo "(sport == 8554 && snd_cwnd =< 0)" > events/tcp/tcp_probe/filter &
cat trace_pipe > /home/tamim/net-next/trace.data
This will record a snapshot where source port is 8554 and snd_cwnd is
not same as the snd_cwnd in previous snapshot. For lack of better
operator, I used "=<" to represent the field on which the filter will
be applied.
The way it works is, it updates the predicate with new value every
time. So the next time, if the field is same as in the predicate, the
snapshot is ignored. Initially it checks if the field is 0 or not.
diff --git a/kernel/trace/trace_events_filter.c
b/kernel/trace/trace_events_filter.c
index 61e7f06..8d733fa 100644
--- a/kernel/trace/trace_events_filter.c
+++ b/kernel/trace/trace_events_filter.c
@@ -39,6 +39,7 @@ enum filter_op_ids
OP_AND,
OP_GLOB,
OP_NE,
+ OP_NE_PREV,
OP_EQ,
OP_LT,
OP_LE,
@@ -62,6 +63,7 @@ static struct filter_op filter_ops[] = {
{ OP_AND, "&&", 2 },
{ OP_GLOB, "~", 4 },
{ OP_NE, "!=", 4 },
+ { OP_NE_PREV, "=<", 4 },
{ OP_EQ, "==", 4 },
{ OP_LT, "<", 5 },
{ OP_LE, "<=", 5 },
@@ -202,6 +204,21 @@ static int filter_pred_##size(struct filter_pred
*pred, void *event) \
return match; \
}
+#define DEFINE_NE_PREV_PRED(size) \
+static int filter_ne_prev_pred_##size(struct filter_pred *pred, void
*event) \
+{ \
+ u##size *addr = (u##size *)(event + pred->offset); \
+ u##size val = (u##size)pred->val; \
+ int match; \
+ \
+ match = (val == *addr) ^ pred->not; \
+ \
+ if(match == 1) \
+ pred->val = *addr; \
+ return match; \
+}
+
+
DEFINE_COMPARISON_PRED(s64);
DEFINE_COMPARISON_PRED(u64);
DEFINE_COMPARISON_PRED(s32);
@@ -216,6 +233,11 @@ DEFINE_EQUALITY_PRED(32);
DEFINE_EQUALITY_PRED(16);
DEFINE_EQUALITY_PRED(8);
+DEFINE_NE_PREV_PRED(64);
+DEFINE_NE_PREV_PRED(32);
+DEFINE_NE_PREV_PRED(16);
+DEFINE_NE_PREV_PRED(8);
+
/* Filter predicate for fixed sized arrays of characters */
static int filter_pred_string(struct filter_pred *pred, void *event)
{
@@ -980,7 +1002,7 @@ int filter_assign_type(const char *type)
static bool is_legal_op(struct ftrace_event_field *field, enum
filter_op_ids op)
{
if (is_string_field(field) &&
- (op != OP_EQ && op != OP_NE && op != OP_GLOB))
+ (op != OP_EQ && op != OP_NE && op != OP_GLOB && op != OP_NE_PREV))
return false;
if (!is_string_field(field) && op == OP_GLOB)
return false;
@@ -997,6 +1019,8 @@ static filter_pred_fn_t select_comparison_fn(enum
filter_op_ids op,
case 8:
if (op == OP_EQ || op == OP_NE)
fn = filter_pred_64;
+ else if (op == OP_NE_PREV)
+ fn = filter_ne_prev_pred_64;
else if (field_is_signed)
fn = pred_funcs_s64[op - PRED_FUNC_START];
else
@@ -1005,6 +1029,8 @@ static filter_pred_fn_t
select_comparison_fn(enum filter_op_ids op,
case 4:
if (op == OP_EQ || op == OP_NE)
fn = filter_pred_32;
+ else if (op == OP_NE_PREV)
+ fn = filter_ne_prev_pred_32;
else if (field_is_signed)
fn = pred_funcs_s32[op - PRED_FUNC_START];
else
@@ -1013,6 +1039,8 @@ static filter_pred_fn_t
select_comparison_fn(enum filter_op_ids op,
case 2:
if (op == OP_EQ || op == OP_NE)
fn = filter_pred_16;
+ else if (op == OP_NE_PREV)
+ fn = filter_ne_prev_pred_16;
else if (field_is_signed)
fn = pred_funcs_s16[op - PRED_FUNC_START];
else
@@ -1021,6 +1049,8 @@ static filter_pred_fn_t
select_comparison_fn(enum filter_op_ids op,
case 1:
if (op == OP_EQ || op == OP_NE)
fn = filter_pred_8;
+ else if (op == OP_NE_PREV)
+ fn = filter_ne_prev_pred_8;
else if (field_is_signed)
fn = pred_funcs_s8[op - PRED_FUNC_START];
else
@@ -1088,7 +1118,7 @@ static int init_pred(struct filter_parse_state *ps,
}
}
- if (pred->op == OP_NE)
+ if (pred->op == OP_NE || pred->op == OP_NE_PREV)
pred->not ^= 1;
pred->fn = fn;
@@ -2197,7 +2227,7 @@ static int ftrace_function_check_pred(struct
filter_pred *pred, int leaf)
* - only '==' and '!=' is used
* - the 'ip' field is used
*/
- if ((pred->op != OP_EQ) && (pred->op != OP_NE))
+ if ((pred->op != OP_EQ) && (pred->op != OP_NE) && (pred->op
!= OP_NE_PREV))
return -EINVAL;
if (strcmp(field->name, "ip"))
I'm new to upstream kernel development. Please let me any suggestion.
Many thanks
Tamim
PhD Candidate,
Kent State University
diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c
index 61e7f06..8d733fa 100644
--- a/kernel/trace/trace_events_filter.c
+++ b/kernel/trace/trace_events_filter.c
@@ -39,6 +39,7 @@ enum filter_op_ids
OP_AND,
OP_GLOB,
OP_NE,
+ OP_NE_PREV,
OP_EQ,
OP_LT,
OP_LE,
@@ -62,6 +63,7 @@ static struct filter_op filter_ops[] = {
{ OP_AND, "&&", 2 },
{ OP_GLOB, "~", 4 },
{ OP_NE, "!=", 4 },
+ { OP_NE_PREV, "=<", 4 },
{ OP_EQ, "==", 4 },
{ OP_LT, "<", 5 },
{ OP_LE, "<=", 5 },
@@ -202,6 +204,21 @@ static int filter_pred_##size(struct filter_pred *pred, void *event) \
return match; \
}
+#define DEFINE_NE_PREV_PRED(size) \
+static int filter_ne_prev_pred_##size(struct filter_pred *pred, void *event) \
+{ \
+ u##size *addr = (u##size *)(event + pred->offset); \
+ u##size val = (u##size)pred->val; \
+ int match; \
+ \
+ match = (val == *addr) ^ pred->not; \
+ \
+ if(match == 1) \
+ pred->val = *addr; \
+ return match; \
+}
+
+
DEFINE_COMPARISON_PRED(s64);
DEFINE_COMPARISON_PRED(u64);
DEFINE_COMPARISON_PRED(s32);
@@ -216,6 +233,11 @@ DEFINE_EQUALITY_PRED(32);
DEFINE_EQUALITY_PRED(16);
DEFINE_EQUALITY_PRED(8);
+DEFINE_NE_PREV_PRED(64);
+DEFINE_NE_PREV_PRED(32);
+DEFINE_NE_PREV_PRED(16);
+DEFINE_NE_PREV_PRED(8);
+
/* Filter predicate for fixed sized arrays of characters */
static int filter_pred_string(struct filter_pred *pred, void *event)
{
@@ -980,7 +1002,7 @@ int filter_assign_type(const char *type)
static bool is_legal_op(struct ftrace_event_field *field, enum filter_op_ids op)
{
if (is_string_field(field) &&
- (op != OP_EQ && op != OP_NE && op != OP_GLOB))
+ (op != OP_EQ && op != OP_NE && op != OP_GLOB && op != OP_NE_PREV))
return false;
if (!is_string_field(field) && op == OP_GLOB)
return false;
@@ -997,6 +1019,8 @@ static filter_pred_fn_t select_comparison_fn(enum filter_op_ids op,
case 8:
if (op == OP_EQ || op == OP_NE)
fn = filter_pred_64;
+ else if (op == OP_NE_PREV)
+ fn = filter_ne_prev_pred_64;
else if (field_is_signed)
fn = pred_funcs_s64[op - PRED_FUNC_START];
else
@@ -1005,6 +1029,8 @@ static filter_pred_fn_t select_comparison_fn(enum filter_op_ids op,
case 4:
if (op == OP_EQ || op == OP_NE)
fn = filter_pred_32;
+ else if (op == OP_NE_PREV)
+ fn = filter_ne_prev_pred_32;
else if (field_is_signed)
fn = pred_funcs_s32[op - PRED_FUNC_START];
else
@@ -1013,6 +1039,8 @@ static filter_pred_fn_t select_comparison_fn(enum filter_op_ids op,
case 2:
if (op == OP_EQ || op == OP_NE)
fn = filter_pred_16;
+ else if (op == OP_NE_PREV)
+ fn = filter_ne_prev_pred_16;
else if (field_is_signed)
fn = pred_funcs_s16[op - PRED_FUNC_START];
else
@@ -1021,6 +1049,8 @@ static filter_pred_fn_t select_comparison_fn(enum filter_op_ids op,
case 1:
if (op == OP_EQ || op == OP_NE)
fn = filter_pred_8;
+ else if (op == OP_NE_PREV)
+ fn = filter_ne_prev_pred_8;
else if (field_is_signed)
fn = pred_funcs_s8[op - PRED_FUNC_START];
else
@@ -1088,7 +1118,7 @@ static int init_pred(struct filter_parse_state *ps,
}
}
- if (pred->op == OP_NE)
+ if (pred->op == OP_NE || pred->op == OP_NE_PREV)
pred->not ^= 1;
pred->fn = fn;
@@ -2197,7 +2227,7 @@ static int ftrace_function_check_pred(struct filter_pred *pred, int leaf)
* - only '==' and '!=' is used
* - the 'ip' field is used
*/
- if ((pred->op != OP_EQ) && (pred->op != OP_NE))
+ if ((pred->op != OP_EQ) && (pred->op != OP_NE) && (pred->op != OP_NE_PREV))
return -EINVAL;
if (strcmp(field->name, "ip"))