1=========================================
2Uprobe-tracer: Uprobe-based Event Tracing
3=========================================
4
5:Author: Srikar Dronamraju
6
7
8Overview
9--------
10Uprobe based trace events are similar to kprobe based trace events.
11To enable this feature, build your kernel with CONFIG_UPROBE_EVENTS=y.
12
13Similar to the kprobe-event tracer, this doesn't need to be activated via
14current_tracer. Instead of that, add probe points via
15/sys/kernel/debug/tracing/uprobe_events, and enable it via
16/sys/kernel/debug/tracing/events/uprobes/<EVENT>/enable.
17
18However unlike kprobe-event tracer, the uprobe event interface expects the
19user to calculate the offset of the probepoint in the object.
20
21You can also use /sys/kernel/debug/tracing/dynamic_events instead of
22uprobe_events. That interface will provide unified access to other
23dynamic events too.
24
25Synopsis of uprobe_tracer
26-------------------------
27::
28
29  p[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS] : Set a uprobe
30  r[:[GRP/]EVENT] PATH:OFFSET [FETCHARGS] : Set a return uprobe (uretprobe)
31  -:[GRP/]EVENT                           : Clear uprobe or uretprobe event
32
33  GRP           : Group name. If omitted, "uprobes" is the default value.
34  EVENT         : Event name. If omitted, the event name is generated based
35                  on PATH+OFFSET.
36  PATH          : Path to an executable or a library.
37  OFFSET        : Offset where the probe is inserted.
38
39  FETCHARGS     : Arguments. Each probe can have up to 128 args.
40   %REG         : Fetch register REG
41   @ADDR	: Fetch memory at ADDR (ADDR should be in userspace)
42   @+OFFSET	: Fetch memory at OFFSET (OFFSET from same file as PATH)
43   $stackN	: Fetch Nth entry of stack (N >= 0)
44   $stack	: Fetch stack address.
45   $retval	: Fetch return value.(\*1)
46   $comm	: Fetch current task comm.
47   +|-[u]OFFS(FETCHARG) : Fetch memory at FETCHARG +|- OFFS address.(\*2)(\*3)
48   NAME=FETCHARG     : Set NAME as the argument name of FETCHARG.
49   FETCHARG:TYPE     : Set TYPE as the type of FETCHARG. Currently, basic types
50		       (u8/u16/u32/u64/s8/s16/s32/s64), hexadecimal types
51		       (x8/x16/x32/x64), "string" and bitfield are supported.
52
53  (\*1) only for return probe.
54  (\*2) this is useful for fetching a field of data structures.
55  (\*3) Unlike kprobe event, "u" prefix will just be ignored, becuse uprobe
56        events can access only user-space memory.
57
58Types
59-----
60Several types are supported for fetch-args. Uprobe tracer will access memory
61by given type. Prefix 's' and 'u' means those types are signed and unsigned
62respectively. 'x' prefix implies it is unsigned. Traced arguments are shown
63in decimal ('s' and 'u') or hexadecimal ('x'). Without type casting, 'x32'
64or 'x64' is used depends on the architecture (e.g. x86-32 uses x32, and
65x86-64 uses x64).
66String type is a special type, which fetches a "null-terminated" string from
67user space.
68Bitfield is another special type, which takes 3 parameters, bit-width, bit-
69offset, and container-size (usually 32). The syntax is::
70
71 b<bit-width>@<bit-offset>/<container-size>
72
73For $comm, the default type is "string"; any other type is invalid.
74
75
76Event Profiling
77---------------
78You can check the total number of probe hits per event via
79/sys/kernel/debug/tracing/uprobe_profile. The first column is the filename,
80the second is the event name, the third is the number of probe hits.
81
82Usage examples
83--------------
84 * Add a probe as a new uprobe event, write a new definition to uprobe_events
85   as below (sets a uprobe at an offset of 0x4245c0 in the executable /bin/bash)::
86
87    echo 'p /bin/bash:0x4245c0' > /sys/kernel/debug/tracing/uprobe_events
88
89 * Add a probe as a new uretprobe event::
90
91    echo 'r /bin/bash:0x4245c0' > /sys/kernel/debug/tracing/uprobe_events
92
93 * Unset registered event::
94
95    echo '-:p_bash_0x4245c0' >> /sys/kernel/debug/tracing/uprobe_events
96
97 * Print out the events that are registered::
98
99    cat /sys/kernel/debug/tracing/uprobe_events
100
101 * Clear all events::
102
103    echo > /sys/kernel/debug/tracing/uprobe_events
104
105Following example shows how to dump the instruction pointer and %ax register
106at the probed text address. Probe zfree function in /bin/zsh::
107
108    # cd /sys/kernel/debug/tracing/
109    # cat /proc/`pgrep zsh`/maps | grep /bin/zsh | grep r-xp
110    00400000-0048a000 r-xp 00000000 08:03 130904 /bin/zsh
111    # objdump -T /bin/zsh | grep -w zfree
112    0000000000446420 g    DF .text  0000000000000012  Base        zfree
113
1140x46420 is the offset of zfree in object /bin/zsh that is loaded at
1150x00400000. Hence the command to uprobe would be::
116
117    # echo 'p:zfree_entry /bin/zsh:0x46420 %ip %ax' > uprobe_events
118
119And the same for the uretprobe would be::
120
121    # echo 'r:zfree_exit /bin/zsh:0x46420 %ip %ax' >> uprobe_events
122
123.. note:: User has to explicitly calculate the offset of the probe-point
124	in the object.
125
126We can see the events that are registered by looking at the uprobe_events file.
127::
128
129    # cat uprobe_events
130    p:uprobes/zfree_entry /bin/zsh:0x00046420 arg1=%ip arg2=%ax
131    r:uprobes/zfree_exit /bin/zsh:0x00046420 arg1=%ip arg2=%ax
132
133Format of events can be seen by viewing the file events/uprobes/zfree_entry/format.
134::
135
136    # cat events/uprobes/zfree_entry/format
137    name: zfree_entry
138    ID: 922
139    format:
140         field:unsigned short common_type;         offset:0;  size:2; signed:0;
141         field:unsigned char common_flags;         offset:2;  size:1; signed:0;
142         field:unsigned char common_preempt_count; offset:3;  size:1; signed:0;
143         field:int common_pid;                     offset:4;  size:4; signed:1;
144         field:int common_padding;                 offset:8;  size:4; signed:1;
145
146         field:unsigned long __probe_ip;           offset:12; size:4; signed:0;
147         field:u32 arg1;                           offset:16; size:4; signed:0;
148         field:u32 arg2;                           offset:20; size:4; signed:0;
149
150    print fmt: "(%lx) arg1=%lx arg2=%lx", REC->__probe_ip, REC->arg1, REC->arg2
151
152Right after definition, each event is disabled by default. For tracing these
153events, you need to enable it by::
154
155    # echo 1 > events/uprobes/enable
156
157Lets start tracing, sleep for some time and stop tracing.
158::
159
160    # echo 1 > tracing_on
161    # sleep 20
162    # echo 0 > tracing_on
163
164Also, you can disable the event by::
165
166    # echo 0 > events/uprobes/enable
167
168And you can see the traced information via /sys/kernel/debug/tracing/trace.
169::
170
171    # cat trace
172    # tracer: nop
173    #
174    #           TASK-PID    CPU#    TIMESTAMP  FUNCTION
175    #              | |       |          |         |
176                 zsh-24842 [006] 258544.995456: zfree_entry: (0x446420) arg1=446420 arg2=79
177                 zsh-24842 [007] 258545.000270: zfree_exit:  (0x446540 <- 0x446420) arg1=446540 arg2=0
178                 zsh-24842 [002] 258545.043929: zfree_entry: (0x446420) arg1=446420 arg2=79
179                 zsh-24842 [004] 258547.046129: zfree_exit:  (0x446540 <- 0x446420) arg1=446540 arg2=0
180
181Output shows us uprobe was triggered for a pid 24842 with ip being 0x446420
182and contents of ax register being 79. And uretprobe was triggered with ip at
1830x446540 with counterpart function entry at 0x446420.
184