Commit 376ae5c0 authored by Sasha Goldshtein's avatar Sasha Goldshtein Committed by 4ast

argdist, trace: Native tracepoint support (#724)

* Remove tracepoint.py

The `Tracepoint` class which implements the necessary
support for the tracepoint kprobe-based hack is no
longer needed and can be removed.

* argdist: Native tracepoint support

This commit migrates argdist to use the native bcc/BPF
tracepoint support instead of the hackish kprobe-
based approach. The resulting programs are cleaner
and likely more efficient.

As a result of this change, there is a slight API
change in how argdist is used with tracepoints. To
access fields from the tracepoint structure, the user
is expected to use `args->field` directly. This
leverages most of the built-in bcc support for
generating the tracepoint probe function.

* trace: Native tracepoint support

This commit migrates trace to use the native bcc/BPF
tracepoint support instead of the hackish kprobe-
based approach. The resulting programs are cleaner
and likely more efficient.

As with argdist, users are now expected to use the
`args` structure pointer to access the tracepoint's
arguments.

For example:

```
trace 't:irq:irq_handler_entry (args->irq != 27) "irq %d", args->irq'
```
parent d2f4762a
......@@ -92,10 +92,9 @@ The expression(s) to capture.
These are the values that are assigned to the histogram or raw event collection.
You may use the parameters directly, or valid C expressions that involve the
parameters, such as "size % 10".
Tracepoints may access a special structure called "tp" that is formatted according
to the tracepoint format (which you can obtain using tplist). For example, the
block:block_rq_complete tracepoint can access tp.nr_sector. You may also use the
members of the "tp" struct directly, e.g. "nr_sector" instead of "tp.nr_sector".
Tracepoints may access a special structure called "args" that is formatted
according to the tracepoint format (which you can obtain using tplist).
For example, the block:block_rq_complete tracepoint can access args->nr_sector.
USDT probes may access the arguments defined by the tracing program in the
special arg1, arg2, ... variables. To obtain their types, use the tplist tool.
Return probes can use the argument values received by the
......
......@@ -93,11 +93,9 @@ format specifier replacements may be any C expressions, and may refer to the
same special keywords as in the predicate (arg1, arg2, etc.).
In tracepoints, both the predicate and the arguments may refer to the tracepoint
format structure, which is stored in the special "tp" variable. For example, the
block:block_rq_complete tracepoint can print or filter by tp.nr_sector. To
discover the format of your tracepoint, use the tplist tool. Note that you can
also use the members of the "tp" struct directly, e.g "nr_sector" instead of
"tp.nr_sector".
format structure, which is stored in the special "args" variable. For example, the
block:block_rq_complete tracepoint can print or filter by args->nr_sector. To
discover the format of your tracepoint, use the tplist tool.
In USDT probes, the arg1, ..., argN variables refer to the probe's arguments.
To determine which arguments your probe has, use the tplist tool.
......@@ -126,7 +124,7 @@ Trace returns from the readline function in bash and print the return value as a
.TP
Trace the block:block_rq_complete tracepoint and print the number of sectors completed:
#
.B trace 't:block:block_rq_complete """%d sectors"", nr_sector'
.B trace 't:block:block_rq_complete """%d sectors"", args->nr_sector'
.TP
Trace the pthread_create USDT probe from the pthread library and print the address of the thread's start function:
#
......
......@@ -27,7 +27,6 @@ basestring = (unicode if sys.version_info[0] < 3 else str)
from .libbcc import lib, _CB_TYPE, bcc_symbol
from .table import Table
from .tracepoint import Tracepoint
from .perf import Perf
from .usyms import ProcessSymbols
......
# Copyright 2016 Sasha Goldshtein
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import ctypes as ct
import multiprocessing
import os
import re
from .perf import Perf
class Tracepoint(object):
enabled_tracepoints = []
trace_root = "/sys/kernel/debug/tracing"
event_root = os.path.join(trace_root, "events")
@classmethod
def _any_tracepoints_enabled(cls):
return len(cls.enabled_tracepoints) > 0
@classmethod
def generate_decl(cls):
if not cls._any_tracepoints_enabled():
return ""
return "\nBPF_HASH(__trace_di, u64, u64);\n"
@classmethod
def generate_entry_probe(cls):
if not cls._any_tracepoints_enabled():
return ""
return """
int __trace_entry_update(struct pt_regs *ctx)
{
u64 tid = bpf_get_current_pid_tgid();
u64 val = PT_REGS_PARM1(ctx);
__trace_di.update(&tid, &val);
return 0;
}
"""
def __init__(self, category, event, tp_id):
self.category = category
self.event = event
self.tp_id = tp_id
self._retrieve_struct_fields()
def _retrieve_struct_fields(self):
self.struct_fields = []
format_lines = Tracepoint.get_tpoint_format(self.category,
self.event)
for line in format_lines:
match = re.search(r'field:([^;]*);.*size:(\d+);', line)
if match is None:
continue
parts = match.group(1).split()
field_name = parts[-1:][0]
field_type = " ".join(parts[:-1])
field_size = int(match.group(2))
if "__data_loc" in field_type:
continue
if field_name.startswith("common_"):
continue
self.struct_fields.append((field_type, field_name))
def _generate_struct_fields(self):
text = ""
for field_type, field_name in self.struct_fields:
text += " %s %s;\n" % (field_type, field_name)
return text
def generate_struct(self):
self.struct_name = self.event + "_trace_entry"
return """
struct %s {
u64 __do_not_use__;
%s
};
""" % (self.struct_name, self._generate_struct_fields())
def _generate_struct_locals(self):
text = ""
for field_type, field_name in self.struct_fields:
if field_type == "char" and field_name.endswith(']'):
# Special case for 'char whatever[N]', should
# be assigned to a 'char *'
field_type = "char *"
field_name = re.sub(r'\[\d+\]$', '', field_name)
text += " %s %s = tp.%s;\n" % (
field_type, field_name, field_name)
return text
def generate_get_struct(self):
return """
u64 tid = bpf_get_current_pid_tgid();
u64 *di = __trace_di.lookup(&tid);
if (di == 0) { return 0; }
struct %s tp = {};
bpf_probe_read(&tp, sizeof(tp), (void *)*di);
%s
""" % (self.struct_name, self._generate_struct_locals())
@classmethod
def enable_tracepoint(cls, category, event):
tp_id = cls.get_tpoint_id(category, event)
if tp_id == -1:
raise ValueError("no such tracepoint found: %s:%s" %
(category, event))
Perf.perf_event_open(tp_id, ptype=Perf.PERF_TYPE_TRACEPOINT)
new_tp = Tracepoint(category, event, tp_id)
cls.enabled_tracepoints.append(new_tp)
return new_tp
@staticmethod
def get_tpoint_id(category, event):
evt_dir = os.path.join(Tracepoint.event_root, category, event)
try:
return int(
open(os.path.join(evt_dir, "id")).read().strip())
except:
return -1
@staticmethod
def get_tpoint_format(category, event):
evt_dir = os.path.join(Tracepoint.event_root, category, event)
try:
return open(os.path.join(evt_dir, "format")).readlines()
except:
return ""
@classmethod
def attach(cls, bpf):
if cls._any_tracepoints_enabled():
bpf.attach_kprobe(event="tracing_generic_entry_update",
fn_name="__trace_entry_update")
......@@ -12,7 +12,7 @@
# Licensed under the Apache License, Version 2.0 (the "License")
# Copyright (C) 2016 Sasha Goldshtein.
from bcc import BPF, Tracepoint, Perf, USDT
from bcc import BPF, USDT
from time import sleep, strftime
import argparse
import re
......@@ -195,9 +195,6 @@ u64 __time = bpf_ktime_get_ns();
self.library = "" # kernel
self.tp_category = parts[1]
self.tp_event = self.function
self.tp = Tracepoint.enable_tracepoint(
self.tp_category, self.tp_event)
self.function = "perf_trace_" + self.function
elif self.probe_type == "u":
self.library = parts[1]
self.probe_func_name = "%s_probe%d" % \
......@@ -329,8 +326,10 @@ u64 __time = bpf_ktime_get_ns();
program = ""
probe_text = """
DATA_DECL
int PROBENAME(struct pt_regs *ctx SIGNATURE)
""" + (
"TRACEPOINT_PROBE(%s, %s)" % (self.tp_category, self.tp_event) \
if self.probe_type == "t" \
else "int PROBENAME(struct pt_regs *ctx SIGNATURE)") + """
{
PID_FILTER
PREFIX
......@@ -352,10 +351,7 @@ int PROBENAME(struct pt_regs *ctx SIGNATURE)
# value we collected when entering the function:
self._replace_entry_exprs()
if self.probe_type == "t":
program += self.tp.generate_struct()
prefix += self.tp.generate_get_struct()
elif self.probe_type == "p" and len(self.signature) > 0:
if self.probe_type == "p" and len(self.signature) > 0:
# Only entry uprobes/kprobes can have user-specified
# signatures. Other probes force it to ().
signature = ", " + self.signature
......@@ -396,7 +392,9 @@ int PROBENAME(struct pt_regs *ctx SIGNATURE)
pid=self.pid or -1)
def _attach_k(self):
if self.probe_type == "r" or self.probe_type == "t":
if self.probe_type == "t":
pass # Nothing to do for tracepoints
elif self.probe_type == "r":
self.bpf.attach_kretprobe(event=self.function,
fn_name=self.probe_func_name)
else:
......@@ -537,10 +535,10 @@ argdist -C 'p:c:fork()#fork calls'
Count fork() calls in libc across all processes
Can also use funccount.py, which is easier and more flexible
argdist -H 't:block:block_rq_complete():u32:tp.nr_sector'
argdist -H 't:block:block_rq_complete():u32:args->nr_sector'
Print histogram of number of sectors in completing block I/O requests
argdist -C 't:irq:irq_handler_entry():int:tp.irq'
argdist -C 't:irq:irq_handler_entry():int:args->irq'
Aggregate interrupts by interrupt request (IRQ)
argdist -C 'u:pthread:pthread_start():u64:arg2' -p 1337
......@@ -613,8 +611,6 @@ struct __string_t { char s[%d]; };
bpf_source += "#include <%s>\n" % include
bpf_source += BPF.generate_auto_includes(
map(lambda p: p.raw_spec, self.probes))
bpf_source += Tracepoint.generate_decl()
bpf_source += Tracepoint.generate_entry_probe()
for probe in self.probes:
bpf_source += probe.generate_text()
if self.args.verbose:
......@@ -627,7 +623,6 @@ struct __string_t { char s[%d]; };
self.bpf = BPF(text=bpf_source, usdt_contexts=usdt_contexts)
def _attach(self):
Tracepoint.attach(self.bpf)
for probe in self.probes:
probe.attach(self.bpf)
if self.args.verbose:
......
......@@ -264,24 +264,18 @@ certain kinds of allocations or visually group them together.
argdist also has basic support for kernel tracepoints. It is sometimes more
convenient to use tracepoints because they are documented and don't vary a lot
between kernel versions like function signatures tend to. For example, let's
trace the net:net_dev_start_xmit tracepoint and print the interface name that
is transmitting:
between kernel versions. For example, let's trace the net:net_dev_start_xmit
tracepoint and print out the protocol field from the tracepoint structure:
# argdist -c -C 't:net:net_dev_start_xmit(void *a, void *b, struct net_device *c):char*:c->name' -n 2
[05:01:10]
t:net:net_dev_start_xmit(void *a, void *b, struct net_device *c):char*:c->name
# argdist -C 't:net:net_dev_start_xmit():u16:args->protocol'
[13:01:49]
t:net:net_dev_start_xmit():u16:args->protocol
COUNT EVENT
4 c->name = eth0
[05:01:11]
t:net:net_dev_start_xmit(void *a, void *b, struct net_device *c):char*:c->name
COUNT EVENT
6 c->name = lo
92 c->name = eth0
8 args->protocol = 2048
^C
Note that to determine the necessary function signature you need to look at the
TP_PROTO declaration in the kernel headers. For example, the net_dev_start_xmit
tracepoint is defined in the include/trace/events/net.h header file.
Note that to discover the format of the net:net_dev_start_xmit tracepoint, you
use the tplist tool (tplist -v net:net_dev_start_xmit).
Here's a final example that finds how many write() system calls are performed
by each process on the system:
......@@ -388,10 +382,10 @@ argdist -C 'p:c:fork()#fork calls'
Count fork() calls in libc across all processes
Can also use funccount.py, which is easier and more flexible
argdist -H 't:block:block_rq_complete():u32:tp.nr_sector'
argdist -H 't:block:block_rq_complete():u32:args->nr_sector'
Print histogram of number of sectors in completing block I/O requests
argdist -C 't:irq:irq_handler_entry():int:tp.irq'
argdist -C 't:irq:irq_handler_entry():int:args->irq'
Aggregate interrupts by interrupt request (IRQ)
argdist -C 'u:pthread:pthread_start():u64:arg2' -p 1337
......
......@@ -9,7 +9,7 @@
# Licensed under the Apache License, Version 2.0 (the "License")
# Copyright (C) 2016 Sasha Goldshtein.
from bcc import BPF, Tracepoint, Perf, USDT
from bcc import BPF, USDT
from functools import partial
from time import sleep, strftime
import argparse
......@@ -138,10 +138,8 @@ class Probe(object):
if self.probe_type == "t":
self.tp_category = parts[1]
self.tp_event = parts[2]
self.tp = Tracepoint.enable_tracepoint(
self.tp_category, self.tp_event)
self.library = "" # kernel
self.function = "perf_trace_%s" % self.tp_event
self.function = "" # generated from TRACEPOINT_PROBE
elif self.probe_type == "u":
self.library = parts[1]
self.usdt_name = parts[2]
......@@ -357,9 +355,6 @@ BPF_PERF_OUTPUT(%s);
prefix = ""
signature = "struct pt_regs *ctx"
if self.probe_type == "t":
data_decl += self.tp.generate_struct()
prefix = self.tp.generate_get_struct()
data_fields = ""
for i, expr in enumerate(self.values):
......@@ -377,8 +372,14 @@ BPF_PERF_OUTPUT(%s);
ctx, BPF_F_REUSE_STACKID
);""" % self.stacks_name
text = """
int %s(%s)
if self.probe_type == "t":
heading = "TRACEPOINT_PROBE(%s, %s)" % \
(self.tp_category, self.tp_event)
ctx_name = "args"
else:
heading = "int %s(%s)" % (self.probe_name, signature)
ctx_name = "ctx"
text = heading + """
{
%s
%s
......@@ -391,15 +392,14 @@ int %s(%s)
bpf_get_current_comm(&__data.comm, sizeof(__data.comm));
%s
%s
%s.perf_submit(ctx, &__data, sizeof(__data));
%s.perf_submit(%s, &__data, sizeof(__data));
return 0;
}
"""
text = text % (self.probe_name, signature,
pid_filter, prefix,
text = text % (pid_filter, prefix,
self._generate_usdt_filter_read(), self.filter,
self.struct_name, data_fields,
stack_trace, self.events_name)
stack_trace, self.events_name, ctx_name)
return data_decl + "\n" + text
......@@ -464,9 +464,10 @@ int %s(%s)
if self.probe_type == "r":
bpf.attach_kretprobe(event=self.function,
fn_name=self.probe_name)
elif self.probe_type == "p" or self.probe_type == "t":
elif self.probe_type == "p":
bpf.attach_kprobe(event=self.function,
fn_name=self.probe_name)
# Note that tracepoints don't need an explicit attach
def _attach_u(self, bpf):
libpath = BPF.find_library(self.library)
......@@ -511,7 +512,7 @@ trace 'r::__kmalloc (retval == 0) "kmalloc failed!"
Trace returns from __kmalloc which returned a null pointer
trace 'r:c:malloc (retval) "allocated = %p", retval
Trace returns from malloc and print non-NULL allocated buffers
trace 't:block:block_rq_complete "sectors=%d", tp.nr_sector'
trace 't:block:block_rq_complete "sectors=%d", args->nr_sector'
Trace the block_rq_complete kernel tracepoint and print # of tx sectors
trace 'u:pthread:pthread_create (arg4 != 0)'
Trace the USDT probe pthread_create when its 4th argument is non-zero
......@@ -558,8 +559,6 @@ trace 'u:pthread:pthread_create (arg4 != 0)'
"""
self.program += BPF.generate_auto_includes(
map(lambda p: p.raw_probe, self.probes))
self.program += Tracepoint.generate_decl()
self.program += Tracepoint.generate_entry_probe()
for probe in self.probes:
self.program += probe.generate_program(
self.args.include_self)
......@@ -580,7 +579,6 @@ trace 'u:pthread:pthread_create (arg4 != 0)'
print(probe.usdt.get_text())
usdt_contexts.append(probe.usdt)
self.bpf = BPF(text=self.program, usdt_contexts=usdt_contexts)
Tracepoint.attach(self.bpf)
for probe in self.probes:
if self.args.verbose:
print(probe)
......
......@@ -84,15 +84,15 @@ trace has also some basic support for kernel tracepoints. For example, let's
trace the block:block_rq_complete tracepoint and print out the number of sectors
transferred:
# trace 't:block:block_rq_complete "sectors=%d", tp.nr_sector'
# trace 't:block:block_rq_complete "sectors=%d", args->nr_sector'
TIME PID COMM FUNC -
01:23:51 0 swapper/0 block_rq_complete sectors=8
01:23:55 10017 kworker/u64: block_rq_complete sectors=1
01:23:55 0 swapper/0 block_rq_complete sectors=8
^C
To discover the tracepoint structure format (which you can refer to as the "tp"
variable), use the tplist tool. For example:
To discover the tracepoint structure format (which you can refer to as the "args"
pointer variable), use the tplist tool. For example:
# tplist -v block:block_rq_complete
block:block_rq_complete
......@@ -102,7 +102,7 @@ block:block_rq_complete
int errors;
char rwbs[8];
This output tells you that you can use "tp.dev", "tp.sector", etc. in your
This output tells you that you can use "args->dev", "args->sector", etc. in your
predicate and trace arguments.
As a final example, let's trace open syscalls for a specific process. By
......@@ -169,7 +169,7 @@ trace 'r::__kmalloc (retval == 0) "kmalloc failed!"
Trace returns from __kmalloc which returned a null pointer
trace 'r:c:malloc (retval) "allocated = %p", retval
Trace returns from malloc and print non-NULL allocated buffers
trace 't:block:block_rq_complete "sectors=%d", tp.nr_sector'
trace 't:block:block_rq_complete "sectors=%d", args->nr_sector'
Trace the block_rq_complete kernel tracepoint and print # of tx sectors
trace 'u:pthread:pthread_create (arg4 != 0)'
Trace the USDT probe pthread_create when its 4th argument is non-zero
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment