forked from iovisor/bcc
-
Notifications
You must be signed in to change notification settings - Fork 0
/
funclatency.py
executable file
·281 lines (250 loc) · 8.3 KB
/
funclatency.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
#!/usr/bin/python
# @lint-avoid-python-3-compatibility-imports
#
# funclatency Time functions and print latency as a histogram.
# For Linux, uses BCC, eBPF.
#
# USAGE: funclatency [-h] [-p PID] [-i INTERVAL] [-T] [-u] [-m] [-F] [-r] [-v]
# pattern
#
# Run "funclatency -h" for full usage.
#
# The pattern is a string with optional '*' wildcards, similar to file
# globbing. If you'd prefer to use regular expressions, use the -r option.
#
# Currently nested or recursive functions are not supported properly, and
# timestamps will be overwritten, creating dubious output. Try to match single
# functions, or groups of functions that run at the same stack layer, and
# don't ultimately call each other.
#
# Copyright (c) 2015 Brendan Gregg.
# Licensed under the Apache License, Version 2.0 (the "License")
#
# 20-Sep-2015 Brendan Gregg Created this.
# 06-Oct-2016 Sasha Goldshtein Added user function support.
from __future__ import print_function
from bcc import BPF
from time import sleep, strftime
import argparse
import signal
# arguments
examples = """examples:
./funclatency do_sys_open # time the do_sys_open() kernel function
./funclatency c:read # time the read() C library function
./funclatency -u vfs_read # time vfs_read(), in microseconds
./funclatency -m do_nanosleep # time do_nanosleep(), in milliseconds
./funclatency -i 2 -d 10 c:open # output every 2 seconds, for duration 10s
./funclatency -mTi 5 vfs_read # output every 5 seconds, with timestamps
./funclatency -p 181 vfs_read # time process 181 only
./funclatency 'vfs_fstat*' # time both vfs_fstat() and vfs_fstatat()
./funclatency 'c:*printf' # time the *printf family of functions
./funclatency -F 'vfs_r*' # show one histogram per matched function
"""
parser = argparse.ArgumentParser(
description="Time functions and print latency as a histogram",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog=examples)
parser.add_argument("-p", "--pid", type=int,
help="trace this PID only")
parser.add_argument("-i", "--interval", type=int,
help="summary interval, in seconds")
parser.add_argument("-d", "--duration", type=int,
help="total duration of trace, in seconds")
parser.add_argument("-T", "--timestamp", action="store_true",
help="include timestamp on output")
parser.add_argument("-u", "--microseconds", action="store_true",
help="microsecond histogram")
parser.add_argument("-m", "--milliseconds", action="store_true",
help="millisecond histogram")
parser.add_argument("-F", "--function", action="store_true",
help="show a separate histogram per function")
parser.add_argument("-r", "--regexp", action="store_true",
help="use regular expressions. Default is \"*\" wildcards only.")
parser.add_argument("-v", "--verbose", action="store_true",
help="print the BPF program (for debugging purposes)")
parser.add_argument("pattern",
help="search expression for functions")
parser.add_argument("--ebpf", action="store_true",
help=argparse.SUPPRESS)
args = parser.parse_args()
if args.duration and not args.interval:
args.interval = args.duration
if not args.interval:
args.interval = 99999999
def bail(error):
print("Error: " + error)
exit(1)
parts = args.pattern.split(':')
if len(parts) == 1:
library = None
pattern = args.pattern
elif len(parts) == 2:
library = parts[0]
libpath = BPF.find_library(library) or BPF.find_exe(library)
if not libpath:
bail("can't resolve library %s" % library)
library = libpath
pattern = parts[1]
else:
bail("unrecognized pattern format '%s'" % pattern)
if not args.regexp:
pattern = pattern.replace('*', '.*')
pattern = '^' + pattern + '$'
# define BPF program
bpf_text = """
#include <uapi/linux/ptrace.h>
typedef struct ip_pid {
u64 ip;
u64 pid;
} ip_pid_t;
typedef struct hist_key {
ip_pid_t key;
u64 slot;
} hist_key_t;
BPF_HASH(start, u32);
BPF_ARRAY(avg, u64, 2);
STORAGE
int trace_func_entry(struct pt_regs *ctx)
{
u64 pid_tgid = bpf_get_current_pid_tgid();
u32 pid = pid_tgid;
u32 tgid = pid_tgid >> 32;
u64 ts = bpf_ktime_get_ns();
FILTER
ENTRYSTORE
start.update(&pid, &ts);
return 0;
}
int trace_func_return(struct pt_regs *ctx)
{
u64 *tsp, delta;
u64 pid_tgid = bpf_get_current_pid_tgid();
u32 pid = pid_tgid;
u32 tgid = pid_tgid >> 32;
// calculate delta time
tsp = start.lookup(&pid);
if (tsp == 0) {
return 0; // missed start
}
delta = bpf_ktime_get_ns() - *tsp;
start.delete(&pid);
u32 lat = 0;
u32 cnt = 1;
u64 *sum = avg.lookup(&lat);
if (sum) lock_xadd(sum, delta);
u64 *cnts = avg.lookup(&cnt);
if (cnts) lock_xadd(cnts, 1);
FACTOR
// store as histogram
STORE
return 0;
}
"""
# do we need to store the IP and pid for each invocation?
need_key = args.function or (library and not args.pid)
# code substitutions
if args.pid:
bpf_text = bpf_text.replace('FILTER',
'if (tgid != %d) { return 0; }' % args.pid)
else:
bpf_text = bpf_text.replace('FILTER', '')
if args.milliseconds:
bpf_text = bpf_text.replace('FACTOR', 'delta /= 1000000;')
label = "msecs"
elif args.microseconds:
bpf_text = bpf_text.replace('FACTOR', 'delta /= 1000;')
label = "usecs"
else:
bpf_text = bpf_text.replace('FACTOR', '')
label = "nsecs"
if need_key:
bpf_text = bpf_text.replace('STORAGE', 'BPF_HASH(ipaddr, u32);\n' +
'BPF_HISTOGRAM(dist, hist_key_t);')
# stash the IP on entry, as on return it's kretprobe_trampoline:
bpf_text = bpf_text.replace('ENTRYSTORE',
'u64 ip = PT_REGS_IP(ctx); ipaddr.update(&pid, &ip);')
pid = '-1' if not library else 'tgid'
bpf_text = bpf_text.replace('STORE',
"""
u64 ip, *ipp = ipaddr.lookup(&pid);
if (ipp) {
ip = *ipp;
hist_key_t key;
key.key.ip = ip;
key.key.pid = %s;
key.slot = bpf_log2l(delta);
dist.increment(key);
ipaddr.delete(&pid);
}
""" % pid)
else:
bpf_text = bpf_text.replace('STORAGE', 'BPF_HISTOGRAM(dist);')
bpf_text = bpf_text.replace('ENTRYSTORE', '')
bpf_text = bpf_text.replace('STORE',
'dist.increment(bpf_log2l(delta));')
if args.verbose or args.ebpf:
print(bpf_text)
if args.ebpf:
exit()
# signal handler
def signal_ignore(signal, frame):
print()
# load BPF program
b = BPF(text=bpf_text)
# attach probes
if not library:
b.attach_kprobe(event_re=pattern, fn_name="trace_func_entry")
b.attach_kretprobe(event_re=pattern, fn_name="trace_func_return")
matched = b.num_open_kprobes()
else:
b.attach_uprobe(name=library, sym_re=pattern, fn_name="trace_func_entry",
pid=args.pid or -1)
b.attach_uretprobe(name=library, sym_re=pattern,
fn_name="trace_func_return", pid=args.pid or -1)
matched = b.num_open_uprobes()
if matched == 0:
print("0 functions matched by \"%s\". Exiting." % args.pattern)
exit()
# header
print("Tracing %d functions for \"%s\"... Hit Ctrl-C to end." %
(matched / 2, args.pattern))
# output
def print_section(key):
if not library:
return BPF.sym(key[0], -1)
else:
return "%s [%d]" % (BPF.sym(key[0], key[1]), key[1])
exiting = 0 if args.interval else 1
seconds = 0
dist = b.get_table("dist")
while (1):
try:
sleep(args.interval)
seconds += args.interval
except KeyboardInterrupt:
exiting = 1
# as cleanup can take many seconds, trap Ctrl-C:
signal.signal(signal.SIGINT, signal_ignore)
if args.duration and seconds >= args.duration:
exiting = 1
print()
if args.timestamp:
print("%-8s\n" % strftime("%H:%M:%S"), end="")
if need_key:
dist.print_log2_hist(label, "Function", section_print_fn=print_section,
bucket_fn=lambda k: (k.ip, k.pid))
else:
dist.print_log2_hist(label)
dist.clear()
total = b['avg'][0].value
counts = b['avg'][1].value
if counts > 0:
if label == 'msecs':
total /= 1000000
elif label == 'usecs':
total /= 1000
avg = total/counts
print("\navg = %ld %s, total: %ld %s, count: %ld\n" %(total/counts, label, total, label, counts))
if exiting:
print("Detaching...")
exit()