]>
Commit | Line | Data |
---|---|---|
d8ec9185 MH |
1 | Kprobe-based Event Tracer |
2 | ========================= | |
3 | ||
4 | Documentation is written by Masami Hiramatsu | |
5 | ||
6 | ||
7 | Overview | |
8 | -------- | |
9 | This tracer is similar to the events tracer which is based on Tracepoint | |
10 | infrastructure. Instead of Tracepoint, this tracer is based on kprobes(kprobe | |
11 | and kretprobe). It probes anywhere where kprobes can probe(this means, all | |
12 | functions body except for __kprobes functions). | |
13 | ||
14 | Unlike the function tracer, this tracer can probe instructions inside of | |
15 | kernel functions. It allows you to check which instruction has been executed. | |
16 | ||
17 | Unlike the Tracepoint based events tracer, this tracer can add and remove | |
18 | probe points on the fly. | |
19 | ||
20 | Similar to the events tracer, this tracer doesn't need to be activated via | |
21 | current_tracer, instead of that, just set probe points via | |
22 | /sys/kernel/debug/tracing/kprobe_events. And you can set filters on each | |
23 | probe events via /sys/kernel/debug/tracing/events/kprobes/<EVENT>/filter. | |
24 | ||
25 | ||
26 | Synopsis of kprobe_events | |
27 | ------------------------- | |
28 | p[:EVENT] SYMBOL[+offs|-offs]|MEMADDR [FETCHARGS] : Set a probe | |
29 | r[:EVENT] SYMBOL[+0] [FETCHARGS] : Set a return probe | |
30 | ||
31 | EVENT : Event name. | |
32 | SYMBOL[+offs|-offs] : Symbol+offset where the probe is inserted. | |
33 | MEMADDR : Address where the probe is inserted. | |
34 | ||
35 | FETCHARGS : Arguments. | |
36 | %REG : Fetch register REG | |
37 | sN : Fetch Nth entry of stack (N >= 0) | |
38 | sa : Fetch stack address. | |
39 | @ADDR : Fetch memory at ADDR (ADDR should be in kernel) | |
40 | @SYM[+|-offs] : Fetch memory at SYM +|- offs (SYM should be a data symbol) | |
41 | aN : Fetch function argument. (N >= 0)(*) | |
42 | rv : Fetch return value.(**) | |
43 | ra : Fetch return address.(**) | |
44 | +|-offs(FETCHARG) : fetch memory at FETCHARG +|- offs address.(***) | |
45 | ||
46 | (*) aN may not correct on asmlinkaged functions and at the middle of | |
47 | function body. | |
48 | (**) only for return probe. | |
49 | (***) this is useful for fetching a field of data structures. | |
50 | ||
51 | ||
52 | Per-Probe Event Filtering | |
53 | ------------------------- | |
54 | Per-probe event filtering feature allows you to set different filter on each | |
55 | probe and gives you what arguments will be shown in trace buffer. If an event | |
56 | name is specified right after 'p:' or 'r:' in kprobe_events, the tracer adds | |
57 | an event under tracing/events/kprobes/<EVENT>, at the directory you can see | |
58 | 'id', 'enabled', 'format' and 'filter'. | |
59 | ||
60 | enabled: | |
61 | You can enable/disable the probe by writing 1 or 0 on it. | |
62 | ||
63 | format: | |
64 | It shows the format of this probe event. It also shows aliases of arguments | |
65 | which you specified to kprobe_events. | |
66 | ||
67 | filter: | |
68 | You can write filtering rules of this event. And you can use both of aliase | |
69 | names and field names for describing filters. | |
70 | ||
71 | ||
72 | Usage examples | |
73 | -------------- | |
74 | To add a probe as a new event, write a new definition to kprobe_events | |
75 | as below. | |
76 | ||
77 | echo p:myprobe do_sys_open a0 a1 a2 a3 > /sys/kernel/debug/tracing/kprobe_events | |
78 | ||
79 | This sets a kprobe on the top of do_sys_open() function with recording | |
80 | 1st to 4th arguments as "myprobe" event. | |
81 | ||
82 | echo r:myretprobe do_sys_open rv ra >> /sys/kernel/debug/tracing/kprobe_events | |
83 | ||
84 | This sets a kretprobe on the return point of do_sys_open() function with | |
85 | recording return value and return address as "myretprobe" event. | |
86 | You can see the format of these events via | |
87 | /sys/kernel/debug/tracing/events/kprobes/<EVENT>/format. | |
88 | ||
89 | cat /sys/kernel/debug/tracing/events/kprobes/myprobe/format | |
90 | name: myprobe | |
91 | ID: 23 | |
92 | format: | |
93 | field:unsigned short common_type; offset:0; size:2; | |
94 | field:unsigned char common_flags; offset:2; size:1; | |
95 | field:unsigned char common_preempt_count; offset:3; size:1; | |
96 | field:int common_pid; offset:4; size:4; | |
97 | field:int common_tgid; offset:8; size:4; | |
98 | ||
99 | field: unsigned long ip; offset:16;tsize:8; | |
100 | field: int nargs; offset:24;tsize:4; | |
101 | field: unsigned long arg0; offset:32;tsize:8; | |
102 | field: unsigned long arg1; offset:40;tsize:8; | |
103 | field: unsigned long arg2; offset:48;tsize:8; | |
104 | field: unsigned long arg3; offset:56;tsize:8; | |
105 | ||
106 | alias: a0; original: arg0; | |
107 | alias: a1; original: arg1; | |
108 | alias: a2; original: arg2; | |
109 | alias: a3; original: arg3; | |
110 | ||
111 | print fmt: "%lx: 0x%lx 0x%lx 0x%lx 0x%lx", ip, arg0, arg1, arg2, arg3 | |
112 | ||
113 | ||
114 | You can see that the event has 4 arguments and alias expressions | |
115 | corresponding to it. | |
116 | ||
117 | echo > /sys/kernel/debug/tracing/kprobe_events | |
118 | ||
119 | This clears all probe points. and you can see the traced information via | |
120 | /sys/kernel/debug/tracing/trace. | |
121 | ||
122 | cat /sys/kernel/debug/tracing/trace | |
123 | # tracer: nop | |
124 | # | |
125 | # TASK-PID CPU# TIMESTAMP FUNCTION | |
126 | # | | | | | | |
127 | <...>-1447 [001] 1038282.286875: do_sys_open+0x0/0xd6: 0x3 0x7fffd1ec4440 0x8000 0x0 | |
128 | <...>-1447 [001] 1038282.286878: sys_openat+0xc/0xe <- do_sys_open: 0xfffffffffffffffe 0xffffffff81367a3a | |
129 | <...>-1447 [001] 1038282.286885: do_sys_open+0x0/0xd6: 0xffffff9c 0x40413c 0x8000 0x1b6 | |
130 | <...>-1447 [001] 1038282.286915: sys_open+0x1b/0x1d <- do_sys_open: 0x3 0xffffffff81367a3a | |
131 | <...>-1447 [001] 1038282.286969: do_sys_open+0x0/0xd6: 0xffffff9c 0x4041c6 0x98800 0x10 | |
132 | <...>-1447 [001] 1038282.286976: sys_open+0x1b/0x1d <- do_sys_open: 0x3 0xffffffff81367a3a | |
133 | ||
134 | ||
135 | Each line shows when the kernel hits a probe, and <- SYMBOL means kernel | |
136 | returns from SYMBOL(e.g. "sys_open+0x1b/0x1d <- do_sys_open" means kernel | |
137 | returns from do_sys_open to sys_open+0x1b). | |
138 | ||
139 |