xref: /linux/Documentation/trace/fprobe.rst (revision f96a974170b749e3a56844e25b31d46a7233b6f6)
1.. SPDX-License-Identifier: GPL-2.0
2
3==================================
4Fprobe - Function entry/exit probe
5==================================
6
7.. Author: Masami Hiramatsu <mhiramat@kernel.org>
8
9Introduction
10============
11
12Fprobe is a function entry/exit probe based on the function-graph tracing
13feature in ftrace.
14Instead of tracing all functions, if you want to attach callbacks on specific
15function entry and exit, similar to the kprobes and kretprobes, you can
16use fprobe. Compared with kprobes and kretprobes, fprobe gives faster
17instrumentation for multiple functions with single handler. This document
18describes how to use fprobe.
19
20The usage of fprobe
21===================
22
23The fprobe is a wrapper of ftrace (+ kretprobe-like return callback) to
24attach callbacks to multiple function entry and exit. User needs to set up
25the `struct fprobe` and pass it to `register_fprobe()`.
26
27Typically, `fprobe` data structure is initialized with the `entry_handler`
28and/or `exit_handler` as below.
29
30.. code-block:: c
31
32 struct fprobe fp = {
33        .entry_handler  = my_entry_callback,
34        .exit_handler   = my_exit_callback,
35 };
36
37To enable the fprobe, call one of register_fprobe(), register_fprobe_ips(), and
38register_fprobe_syms(). These functions register the fprobe with different types
39of parameters.
40
41The register_fprobe() enables a fprobe by function-name filters.
42E.g. this enables @fp on "func*()" function except "func2()".::
43
44  register_fprobe(&fp, "func*", "func2");
45
46The register_fprobe_ips() enables a fprobe by ftrace-location addresses.
47E.g.
48
49.. code-block:: c
50
51  unsigned long ips[] = { 0x.... };
52
53  register_fprobe_ips(&fp, ips, ARRAY_SIZE(ips));
54
55And the register_fprobe_syms() enables a fprobe by symbol names.
56E.g.
57
58.. code-block:: c
59
60  char syms[] = {"func1", "func2", "func3"};
61
62  register_fprobe_syms(&fp, syms, ARRAY_SIZE(syms));
63
64To disable (remove from functions) this fprobe, call::
65
66  unregister_fprobe(&fp);
67
68You can temporally (soft) disable the fprobe by::
69
70  disable_fprobe(&fp);
71
72and resume by::
73
74  enable_fprobe(&fp);
75
76The above is defined by including the header::
77
78  #include <linux/fprobe.h>
79
80Same as ftrace, the registered callbacks will start being called some time
81after the register_fprobe() is called and before it returns. See
82:file:`Documentation/trace/ftrace.rst`.
83
84Also, the unregister_fprobe() will guarantee that the both enter and exit
85handlers are no longer being called by functions after unregister_fprobe()
86returns as same as unregister_ftrace_function().
87
88The fprobe entry/exit handler
89=============================
90
91The prototype of the entry/exit callback function are as follows:
92
93.. code-block:: c
94
95 int entry_callback(struct fprobe *fp, unsigned long entry_ip, unsigned long ret_ip, struct ftrace_regs *fregs, void *entry_data);
96
97 void exit_callback(struct fprobe *fp, unsigned long entry_ip, unsigned long ret_ip, struct ftrace_regs *fregs, void *entry_data);
98
99Note that the @entry_ip is saved at function entry and passed to exit
100handler.
101If the entry callback function returns !0, the corresponding exit callback
102will be cancelled.
103
104@fp
105        This is the address of `fprobe` data structure related to this handler.
106        You can embed the `fprobe` to your data structure and get it by
107        container_of() macro from @fp. The @fp must not be NULL.
108
109@entry_ip
110        This is the ftrace address of the traced function (both entry and exit).
111        Note that this may not be the actual entry address of the function but
112        the address where the ftrace is instrumented.
113
114@ret_ip
115        This is the return address that the traced function will return to,
116        somewhere in the caller. This can be used at both entry and exit.
117
118@fregs
119        This is the `ftrace_regs` data structure at the entry and exit. This
120        includes the function parameters, or the return values. So user can
121        access thos values via appropriate `ftrace_regs_*` APIs.
122
123@entry_data
124        This is a local storage to share the data between entry and exit handlers.
125        This storage is NULL by default. If the user specify `exit_handler` field
126        and `entry_data_size` field when registering the fprobe, the storage is
127        allocated and passed to both `entry_handler` and `exit_handler`.
128
129Entry data size and exit handlers on the same function
130======================================================
131
132Since the entry data is passed via per-task stack and it has limited size,
133the entry data size per probe is limited to `15 * sizeof(long)`. You also need
134to take care that the different fprobes are probing on the same function, this
135limit becomes smaller. The entry data size is aligned to `sizeof(long)` and
136each fprobe which has exit handler uses a `sizeof(long)` space on the stack,
137you should keep the number of fprobes on the same function as small as
138possible.
139
140Share the callbacks with kprobes
141================================
142
143Since the recursion safeness of the fprobe (and ftrace) is a bit different
144from the kprobes, this may cause an issue if user wants to run the same
145code from the fprobe and the kprobes.
146
147Kprobes has per-cpu 'current_kprobe' variable which protects the kprobe
148handler from recursion in all cases. On the other hand, fprobe uses
149only ftrace_test_recursion_trylock(). This allows interrupt context to
150call another (or same) fprobe while the fprobe user handler is running.
151
152This is not a matter if the common callback code has its own recursion
153detection, or it can handle the recursion in the different contexts
154(normal/interrupt/NMI.)
155But if it relies on the 'current_kprobe' recursion lock, it has to check
156kprobe_running() and use kprobe_busy_*() APIs.
157
158Fprobe has FPROBE_FL_KPROBE_SHARED flag to do this. If your common callback
159code will be shared with kprobes, please set FPROBE_FL_KPROBE_SHARED
160*before* registering the fprobe, like:
161
162.. code-block:: c
163
164 fprobe.flags = FPROBE_FL_KPROBE_SHARED;
165
166 register_fprobe(&fprobe, "func*", NULL);
167
168This will protect your common callback from the nested call.
169
170The missed counter
171==================
172
173The `fprobe` data structure has `fprobe::nmissed` counter field as same as
174kprobes.
175This counter counts up when;
176
177 - fprobe fails to take ftrace_recursion lock. This usually means that a function
178   which is traced by other ftrace users is called from the entry_handler.
179
180 - fprobe fails to setup the function exit because of failing to allocate the
181   data buffer from the per-task shadow stack.
182
183The `fprobe::nmissed` field counts up in both cases. Therefore, the former
184skips both of entry and exit callback and the latter skips the exit
185callback, but in both case the counter will increase by 1.
186
187Note that if you set the FTRACE_OPS_FL_RECURSION and/or FTRACE_OPS_FL_RCU to
188`fprobe::ops::flags` (ftrace_ops::flags) when registering the fprobe, this
189counter may not work correctly, because ftrace skips the fprobe function which
190increase the counter.
191
192
193Functions and structures
194========================
195
196.. kernel-doc:: include/linux/fprobe.h
197.. kernel-doc:: kernel/trace/fprobe.c
198
199