Sasha Goldshtein | 7789318 | 2017-03-30 05:05:10 -0400 | [diff] [blame] | 1 | #!/usr/bin/python |
| 2 | # @lint-avoid-python-3-compatibility-imports |
| 3 | # |
| 4 | # funcslower Trace slow kernel or user function calls. |
| 5 | # For Linux, uses BCC, eBPF. |
| 6 | # |
| 7 | # USAGE: funcslower [-h] [-p PID] [-m MIN_MS] [-u MIN_US] [-a ARGUMENTS] |
| 8 | # [-T] [-t] [-v] function [function ...] |
| 9 | # |
| 10 | # WARNING: This tool traces function calls by instrumenting the entry and |
| 11 | # return from each function. For commonly-invoked functions like memory allocs |
| 12 | # or file writes, this can be extremely expensive. Mind the overhead. |
| 13 | # |
Ivan Babrou | 525f3e9 | 2017-11-21 16:03:57 -0800 | [diff] [blame] | 14 | # NOTE: This tool cannot trace nested functions in the same invocation |
| 15 | # due to instrumentation specifics, only innermost calls will be visible. |
| 16 | # |
Sasha Goldshtein | 7789318 | 2017-03-30 05:05:10 -0400 | [diff] [blame] | 17 | # By default, a minimum millisecond threshold of 1 is used. |
| 18 | # |
| 19 | # Copyright 2017, Sasha Goldshtein |
| 20 | # Licensed under the Apache License, Version 2.0 (the "License") |
| 21 | # |
| 22 | # 30-Mar-2017 Sasha Goldshtein Created this. |
| 23 | |
| 24 | from __future__ import print_function |
| 25 | from bcc import BPF |
| 26 | import argparse |
| 27 | import ctypes as ct |
| 28 | import time |
| 29 | |
| 30 | examples = """examples: |
| 31 | ./funcslower vfs_write # trace vfs_write calls slower than 1ms |
| 32 | ./funcslower -m 10 vfs_write # same, but slower than 10ms |
| 33 | ./funcslower -u 10 c:open # trace open calls slower than 10us |
| 34 | ./funcslower -p 135 c:open # trace pid 135 only |
| 35 | ./funcslower c:malloc c:free # trace both malloc and free slower than 1ms |
| 36 | ./funcslower -a 2 c:open # show first two arguments to open |
| 37 | """ |
| 38 | parser = argparse.ArgumentParser( |
| 39 | description="Trace slow kernel or user function calls.", |
| 40 | formatter_class=argparse.RawDescriptionHelpFormatter, |
| 41 | epilog=examples) |
| 42 | parser.add_argument("-p", "--pid", type=int, metavar="PID", dest="tgid", |
| 43 | help="trace this PID only") |
| 44 | parser.add_argument("-m", "--min-ms", type=float, dest="min_ms", |
| 45 | help="minimum duration to trace (ms)") |
| 46 | parser.add_argument("-u", "--min-us", type=float, dest="min_us", |
| 47 | help="minimum duration to trace (us)") |
| 48 | parser.add_argument("-a", "--arguments", type=int, |
| 49 | help="print this many entry arguments, as hex") |
| 50 | parser.add_argument("-T", "--time", action="store_true", |
| 51 | help="show HH:MM:SS timestamp") |
| 52 | parser.add_argument("-t", "--timestamp", action="store_true", |
| 53 | help="show timestamp in seconds at us resolution") |
| 54 | parser.add_argument("-v", "--verbose", action="store_true", |
| 55 | help="print the BPF program for debugging purposes") |
| 56 | parser.add_argument(metavar="function", nargs="+", dest="functions", |
| 57 | help="function(s) to trace") |
| 58 | |
| 59 | args = parser.parse_args() |
| 60 | # fractions are allowed, but rounded to an integer nanosecond |
| 61 | if args.min_ms: |
| 62 | duration_ns = int(args.min_ms * 1000000) |
| 63 | elif args.min_us: |
| 64 | duration_ns = int(args.min_us * 1000) |
| 65 | else: |
| 66 | duration_ns = 1000000 # default to 1ms |
| 67 | |
| 68 | bpf_text = """ |
| 69 | #include <uapi/linux/ptrace.h> |
| 70 | #include <linux/sched.h> // for TASK_COMM_LEN |
| 71 | |
| 72 | struct entry_t { |
| 73 | u64 id; |
| 74 | u64 start_ns; |
| 75 | #ifdef GRAB_ARGS |
| 76 | u64 args[6]; |
| 77 | #endif |
| 78 | }; |
| 79 | |
| 80 | struct data_t { |
| 81 | u64 id; |
| 82 | u64 tgid_pid; |
| 83 | u64 start_ns; |
| 84 | u64 duration_ns; |
| 85 | u64 retval; |
| 86 | char comm[TASK_COMM_LEN]; |
| 87 | #ifdef GRAB_ARGS |
| 88 | u64 args[6]; |
| 89 | #endif |
| 90 | }; |
| 91 | |
| 92 | BPF_HASH(entryinfo, u64, struct entry_t); |
| 93 | BPF_PERF_OUTPUT(events); |
| 94 | |
| 95 | static int trace_entry(struct pt_regs *ctx, int id) |
| 96 | { |
| 97 | u64 tgid_pid = bpf_get_current_pid_tgid(); |
| 98 | u32 tgid = tgid_pid >> 32; |
| 99 | if (TGID_FILTER) |
| 100 | return 0; |
| 101 | |
| 102 | u32 pid = tgid_pid; |
| 103 | |
| 104 | struct entry_t entry = {}; |
| 105 | entry.start_ns = bpf_ktime_get_ns(); |
| 106 | entry.id = id; |
| 107 | #ifdef GRAB_ARGS |
| 108 | entry.args[0] = PT_REGS_PARM1(ctx); |
| 109 | entry.args[1] = PT_REGS_PARM2(ctx); |
| 110 | entry.args[2] = PT_REGS_PARM3(ctx); |
| 111 | entry.args[3] = PT_REGS_PARM4(ctx); |
| 112 | entry.args[4] = PT_REGS_PARM5(ctx); |
| 113 | entry.args[5] = PT_REGS_PARM6(ctx); |
| 114 | #endif |
| 115 | |
| 116 | entryinfo.update(&tgid_pid, &entry); |
| 117 | |
| 118 | return 0; |
| 119 | } |
| 120 | |
| 121 | int trace_return(struct pt_regs *ctx) |
| 122 | { |
| 123 | struct entry_t *entryp; |
| 124 | u64 tgid_pid = bpf_get_current_pid_tgid(); |
| 125 | |
| 126 | entryp = entryinfo.lookup(&tgid_pid); |
| 127 | if (entryp == 0) { |
| 128 | return 0; |
| 129 | } |
| 130 | |
| 131 | u64 delta_ns = bpf_ktime_get_ns() - entryp->start_ns; |
| 132 | entryinfo.delete(&tgid_pid); |
| 133 | |
| 134 | if (delta_ns < DURATION_NS) |
| 135 | return 0; |
| 136 | |
| 137 | struct data_t data = {}; |
| 138 | data.id = entryp->id; |
| 139 | data.tgid_pid = tgid_pid; |
| 140 | data.start_ns = entryp->start_ns; |
| 141 | data.duration_ns = delta_ns; |
| 142 | data.retval = PT_REGS_RC(ctx); |
| 143 | #ifdef GRAB_ARGS |
| 144 | bpf_probe_read(&data.args[0], sizeof(data.args), entryp->args); |
| 145 | #endif |
| 146 | bpf_get_current_comm(&data.comm, sizeof(data.comm)); |
| 147 | events.perf_submit(ctx, &data, sizeof(data)); |
| 148 | |
| 149 | return 0; |
| 150 | } |
| 151 | """ |
| 152 | |
| 153 | bpf_text = bpf_text.replace('DURATION_NS', str(duration_ns)) |
| 154 | if args.arguments: |
| 155 | bpf_text = "#define GRAB_ARGS\n" + bpf_text |
| 156 | if args.tgid: |
Ivan Babrou | 95c20b0 | 2017-11-21 15:35:42 -0800 | [diff] [blame] | 157 | bpf_text = bpf_text.replace('TGID_FILTER', 'tgid != %d' % args.tgid) |
Sasha Goldshtein | 7789318 | 2017-03-30 05:05:10 -0400 | [diff] [blame] | 158 | else: |
| 159 | bpf_text = bpf_text.replace('TGID_FILTER', '0') |
| 160 | |
| 161 | for i in range(len(args.functions)): |
| 162 | bpf_text += """ |
| 163 | int trace_%d(struct pt_regs *ctx) { |
| 164 | return trace_entry(ctx, %d); |
| 165 | } |
| 166 | """ % (i, i) |
| 167 | |
| 168 | if args.verbose: |
| 169 | print(bpf_text) |
| 170 | |
| 171 | b = BPF(text=bpf_text) |
| 172 | |
| 173 | for i, function in enumerate(args.functions): |
| 174 | if ":" in function: |
| 175 | library, func = function.split(":") |
| 176 | b.attach_uprobe(name=library, sym=func, fn_name="trace_%d" % i) |
| 177 | b.attach_uretprobe(name=library, sym=func, fn_name="trace_return") |
| 178 | else: |
| 179 | b.attach_kprobe(event=function, fn_name="trace_%d" % i) |
| 180 | b.attach_kretprobe(event=function, fn_name="trace_return") |
| 181 | |
| 182 | TASK_COMM_LEN = 16 # linux/sched.h |
| 183 | |
| 184 | class Data(ct.Structure): |
| 185 | _fields_ = [ |
| 186 | ("id", ct.c_ulonglong), |
| 187 | ("tgid_pid", ct.c_ulonglong), |
| 188 | ("start_ns", ct.c_ulonglong), |
| 189 | ("duration_ns", ct.c_ulonglong), |
| 190 | ("retval", ct.c_ulonglong), |
| 191 | ("comm", ct.c_char * TASK_COMM_LEN) |
| 192 | ] + ([("args", ct.c_ulonglong * 6)] if args.arguments else []) |
| 193 | |
| 194 | time_designator = "us" if args.min_us else "ms" |
| 195 | time_value = args.min_us or args.min_ms or 1 |
| 196 | time_multiplier = 1000 if args.min_us else 1000000 |
| 197 | time_col = args.time or args.timestamp |
| 198 | print("Tracing function calls slower than %g %s... Ctrl+C to quit." % |
| 199 | (time_value, time_designator)) |
| 200 | print((("%-10s " % "TIME" if time_col else "") + "%-14s %-6s %7s %16s %s") % |
| 201 | ("COMM", "PID", "LAT(%s)" % time_designator, "RVAL", |
| 202 | "FUNC" + (" ARGS" if args.arguments else ""))) |
| 203 | |
| 204 | earliest_ts = 0 |
| 205 | |
| 206 | def time_str(event): |
| 207 | if args.time: |
| 208 | return "%-10s " % time.strftime("%H:%M:%S") |
| 209 | if args.timestamp: |
| 210 | global earliest_ts |
| 211 | if earliest_ts == 0: |
| 212 | earliest_ts = event.start_ns |
| 213 | return "%-10.6f " % ((event.start_ns - earliest_ts) / 1000000000.0) |
| 214 | return "" |
| 215 | |
| 216 | def args_str(event): |
| 217 | if not args.arguments: |
| 218 | return "" |
| 219 | return str.join(" ", ["0x%x" % arg for arg in event.args[:args.arguments]]) |
| 220 | |
| 221 | def print_event(cpu, data, size): |
| 222 | event = ct.cast(data, ct.POINTER(Data)).contents |
| 223 | ts = float(event.duration_ns) / time_multiplier |
| 224 | print((time_str(event) + "%-14.14s %-6s %7.2f %16x %s %s") % |
| 225 | (event.comm.decode(), event.tgid_pid >> 32, |
| 226 | ts, event.retval, args.functions[event.id], args_str(event))) |
| 227 | |
| 228 | b["events"].open_perf_buffer(print_event, page_cnt=64) |
| 229 | while True: |
| 230 | b.kprobe_poll() |