Commit | Line | Data |
---|---|---|
a1dac767 MCC |
1 | ======================= |
2 | Kernel Probes (Kprobes) | |
3 | ======================= | |
4 | ||
5 | :Author: Jim Keniston <jkenisto@us.ibm.com> | |
6 | :Author: Prasanna S Panchamukhi <prasanna.panchamukhi@gmail.com> | |
b65a6b44 | 7 | :Author: Masami Hiramatsu <mhiramat@kernel.org> |
a1dac767 MCC |
8 | |
9 | .. CONTENTS | |
10 | ||
9b17374e | 11 | 1. Concepts: Kprobes, and Return Probes |
a1dac767 MCC |
12 | 2. Architectures Supported |
13 | 3. Configuring Kprobes | |
14 | 4. API Reference | |
15 | 5. Kprobes Features and Limitations | |
16 | 6. Probe Overhead | |
17 | 7. TODO | |
18 | 8. Kprobes Example | |
9b17374e MH |
19 | 9. Kretprobes Example |
20 | 10. Deprecated Features | |
a1dac767 MCC |
21 | Appendix A: The kprobes debugfs interface |
22 | Appendix B: The kprobes sysctl interface | |
0c75f127 | 23 | Appendix C: References |
a1dac767 | 24 | |
9b17374e | 25 | Concepts: Kprobes and Return Probes |
a1dac767 | 26 | ========================================= |
d27a4ddd JK |
27 | |
28 | Kprobes enables you to dynamically break into any kernel routine and | |
29 | collect debugging and performance information non-disruptively. You | |
a1dac767 | 30 | can trap at almost any kernel code address [1]_, specifying a handler |
d27a4ddd | 31 | routine to be invoked when the breakpoint is hit. |
a1dac767 MCC |
32 | |
33 | .. [1] some parts of the kernel code can not be trapped, see | |
34 | :ref:`kprobes_blacklist`) | |
d27a4ddd | 35 | |
9b17374e MH |
36 | There are currently two types of probes: kprobes, and kretprobes |
37 | (also called return probes). A kprobe can be inserted on virtually | |
38 | any instruction in the kernel. A return probe fires when a specified | |
39 | function returns. | |
d27a4ddd JK |
40 | |
41 | In the typical case, Kprobes-based instrumentation is packaged as | |
42 | a kernel module. The module's init function installs ("registers") | |
43 | one or more probes, and the exit function unregisters them. A | |
44 | registration function such as register_kprobe() specifies where | |
45 | the probe is to be inserted and what handler is to be called when | |
46 | the probe is hit. | |
47 | ||
a1dac767 MCC |
48 | There are also ``register_/unregister_*probes()`` functions for batch |
49 | registration/unregistration of a group of ``*probes``. These functions | |
3b0cb4ca MH |
50 | can speed up unregistration process when you have to unregister |
51 | a lot of probes at once. | |
52 | ||
b26486bf MH |
53 | The next four subsections explain how the different types of |
54 | probes work and how jump optimization works. They explain certain | |
55 | things that you'll need to know in order to make the best use of | |
56 | Kprobes -- e.g., the difference between a pre_handler and | |
57 | a post_handler, and how to use the maxactive and nmissed fields of | |
58 | a kretprobe. But if you're in a hurry to start using Kprobes, you | |
a1dac767 | 59 | can skip ahead to :ref:`kprobes_archs_supported`. |
d27a4ddd | 60 | |
a1dac767 MCC |
61 | How Does a Kprobe Work? |
62 | ----------------------- | |
d27a4ddd JK |
63 | |
64 | When a kprobe is registered, Kprobes makes a copy of the probed | |
65 | instruction and replaces the first byte(s) of the probed instruction | |
66 | with a breakpoint instruction (e.g., int3 on i386 and x86_64). | |
67 | ||
68 | When a CPU hits the breakpoint instruction, a trap occurs, the CPU's | |
69 | registers are saved, and control passes to Kprobes via the | |
70 | notifier_call_chain mechanism. Kprobes executes the "pre_handler" | |
71 | associated with the kprobe, passing the handler the addresses of the | |
72 | kprobe struct and the saved registers. | |
73 | ||
74 | Next, Kprobes single-steps its copy of the probed instruction. | |
75 | (It would be simpler to single-step the actual instruction in place, | |
76 | but then Kprobes would have to temporarily remove the breakpoint | |
77 | instruction. This would open a small time window when another CPU | |
78 | could sail right past the probepoint.) | |
79 | ||
80 | After the instruction is single-stepped, Kprobes executes the | |
81 | "post_handler," if any, that is associated with the kprobe. | |
82 | Execution then continues with the instruction following the probepoint. | |
83 | ||
dcce32d9 MH |
84 | Changing Execution Path |
85 | ----------------------- | |
86 | ||
01bdee64 MH |
87 | Since kprobes can probe into a running kernel code, it can change the |
88 | register set, including instruction pointer. This operation requires | |
89 | maximum care, such as keeping the stack frame, recovering the execution | |
90 | path etc. Since it operates on a running kernel and needs deep knowledge | |
91 | of computer architecture and concurrent computing, you can easily shoot | |
92 | your foot. | |
dcce32d9 MH |
93 | |
94 | If you change the instruction pointer (and set up other related | |
01bdee64 MH |
95 | registers) in pre_handler, you must return !0 so that kprobes stops |
96 | single stepping and just returns to the given address. | |
dcce32d9 MH |
97 | This also means post_handler should not be called anymore. |
98 | ||
01bdee64 MH |
99 | Note that this operation may be harder on some architectures which use |
100 | TOC (Table of Contents) for function call, since you have to setup a new | |
101 | TOC for your function in your module, and recover the old one after | |
102 | returning from it. | |
dcce32d9 | 103 | |
a1dac767 MCC |
104 | Return Probes |
105 | ------------- | |
f47cd9b5 | 106 | |
a1dac767 MCC |
107 | How Does a Return Probe Work? |
108 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
d27a4ddd JK |
109 | |
110 | When you call register_kretprobe(), Kprobes establishes a kprobe at | |
111 | the entry to the function. When the probed function is called and this | |
112 | probe is hit, Kprobes saves a copy of the return address, and replaces | |
113 | the return address with the address of a "trampoline." The trampoline | |
114 | is an arbitrary piece of code -- typically just a nop instruction. | |
115 | At boot time, Kprobes registers a kprobe at the trampoline. | |
116 | ||
117 | When the probed function executes its return instruction, control | |
118 | passes to the trampoline and that probe is hit. Kprobes' trampoline | |
f47cd9b5 AS |
119 | handler calls the user-specified return handler associated with the |
120 | kretprobe, then sets the saved instruction pointer to the saved return | |
121 | address, and that's where execution resumes upon return from the trap. | |
d27a4ddd JK |
122 | |
123 | While the probed function is executing, its return address is | |
124 | stored in an object of type kretprobe_instance. Before calling | |
125 | register_kretprobe(), the user sets the maxactive field of the | |
126 | kretprobe struct to specify how many instances of the specified | |
127 | function can be probed simultaneously. register_kretprobe() | |
128 | pre-allocates the indicated number of kretprobe_instance objects. | |
129 | ||
130 | For example, if the function is non-recursive and is called with a | |
131 | spinlock held, maxactive = 1 should be enough. If the function is | |
132 | non-recursive and can never relinquish the CPU (e.g., via a semaphore | |
133 | or preemption), NR_CPUS should be enough. If maxactive <= 0, it is | |
3b7ddab8 | 134 | set to a default value: max(10, 2*NR_CPUS). |
d27a4ddd JK |
135 | |
136 | It's not a disaster if you set maxactive too low; you'll just miss | |
137 | some probes. In the kretprobe struct, the nmissed field is set to | |
138 | zero when the return probe is registered, and is incremented every | |
139 | time the probed function is entered but there is no kretprobe_instance | |
140 | object available for establishing the return probe. | |
141 | ||
a1dac767 MCC |
142 | Kretprobe entry-handler |
143 | ^^^^^^^^^^^^^^^^^^^^^^^ | |
f47cd9b5 AS |
144 | |
145 | Kretprobes also provides an optional user-specified handler which runs | |
146 | on function entry. This handler is specified by setting the entry_handler | |
147 | field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the | |
148 | function entry is hit, the user-defined entry_handler, if any, is invoked. | |
149 | If the entry_handler returns 0 (success) then a corresponding return handler | |
150 | is guaranteed to be called upon function return. If the entry_handler | |
151 | returns a non-zero error then Kprobes leaves the return address as is, and | |
152 | the kretprobe has no further effect for that particular function instance. | |
153 | ||
154 | Multiple entry and return handler invocations are matched using the unique | |
155 | kretprobe_instance object associated with them. Additionally, a user | |
156 | may also specify per return-instance private data to be part of each | |
157 | kretprobe_instance object. This is especially useful when sharing private | |
158 | data between corresponding user entry and return handlers. The size of each | |
159 | private data object can be specified at kretprobe registration time by | |
160 | setting the data_size field of the kretprobe struct. This data can be | |
161 | accessed through the data field of each kretprobe_instance object. | |
162 | ||
163 | In case probed function is entered but there is no kretprobe_instance | |
164 | object available, then in addition to incrementing the nmissed count, | |
165 | the user entry_handler invocation is also skipped. | |
166 | ||
a1dac767 MCC |
167 | .. _kprobes_jump_optimization: |
168 | ||
169 | How Does Jump Optimization Work? | |
170 | -------------------------------- | |
b26486bf | 171 | |
5cc718b9 MH |
172 | If your kernel is built with CONFIG_OPTPROBES=y (currently this flag |
173 | is automatically set 'y' on x86/x86-64, non-preemptive kernel) and | |
b26486bf MH |
174 | the "debug.kprobes_optimization" kernel parameter is set to 1 (see |
175 | sysctl(8)), Kprobes tries to reduce probe-hit overhead by using a jump | |
176 | instruction instead of a breakpoint instruction at each probepoint. | |
177 | ||
a1dac767 MCC |
178 | Init a Kprobe |
179 | ^^^^^^^^^^^^^ | |
b26486bf MH |
180 | |
181 | When a probe is registered, before attempting this optimization, | |
182 | Kprobes inserts an ordinary, breakpoint-based kprobe at the specified | |
183 | address. So, even if it's not possible to optimize this particular | |
184 | probepoint, there'll be a probe there. | |
185 | ||
a1dac767 MCC |
186 | Safety Check |
187 | ^^^^^^^^^^^^ | |
b26486bf MH |
188 | |
189 | Before optimizing a probe, Kprobes performs the following safety checks: | |
190 | ||
191 | - Kprobes verifies that the region that will be replaced by the jump | |
a1dac767 MCC |
192 | instruction (the "optimized region") lies entirely within one function. |
193 | (A jump instruction is multiple bytes, and so may overlay multiple | |
194 | instructions.) | |
b26486bf MH |
195 | |
196 | - Kprobes analyzes the entire function and verifies that there is no | |
a1dac767 MCC |
197 | jump into the optimized region. Specifically: |
198 | ||
b26486bf MH |
199 | - the function contains no indirect jump; |
200 | - the function contains no instruction that causes an exception (since | |
a1dac767 MCC |
201 | the fixup code triggered by the exception could jump back into the |
202 | optimized region -- Kprobes checks the exception tables to verify this); | |
b26486bf | 203 | - there is no near jump to the optimized region (other than to the first |
a1dac767 | 204 | byte). |
b26486bf MH |
205 | |
206 | - For each instruction in the optimized region, Kprobes verifies that | |
a1dac767 | 207 | the instruction can be executed out of line. |
b26486bf | 208 | |
a1dac767 MCC |
209 | Preparing Detour Buffer |
210 | ^^^^^^^^^^^^^^^^^^^^^^^ | |
b26486bf MH |
211 | |
212 | Next, Kprobes prepares a "detour" buffer, which contains the following | |
213 | instruction sequence: | |
a1dac767 | 214 | |
b26486bf MH |
215 | - code to push the CPU's registers (emulating a breakpoint trap) |
216 | - a call to the trampoline code which calls user's probe handlers. | |
217 | - code to restore registers | |
218 | - the instructions from the optimized region | |
219 | - a jump back to the original execution path. | |
220 | ||
a1dac767 MCC |
221 | Pre-optimization |
222 | ^^^^^^^^^^^^^^^^ | |
b26486bf MH |
223 | |
224 | After preparing the detour buffer, Kprobes verifies that none of the | |
225 | following situations exist: | |
a1dac767 | 226 | |
9b17374e | 227 | - The probe has a post_handler. |
b26486bf MH |
228 | - Other instructions in the optimized region are probed. |
229 | - The probe is disabled. | |
a1dac767 | 230 | |
b26486bf MH |
231 | In any of the above cases, Kprobes won't start optimizing the probe. |
232 | Since these are temporary situations, Kprobes tries to start | |
233 | optimizing it again if the situation is changed. | |
234 | ||
235 | If the kprobe can be optimized, Kprobes enqueues the kprobe to an | |
236 | optimizing list, and kicks the kprobe-optimizer workqueue to optimize | |
237 | it. If the to-be-optimized probepoint is hit before being optimized, | |
238 | Kprobes returns control to the original instruction path by setting | |
239 | the CPU's instruction pointer to the copied code in the detour buffer | |
240 | -- thus at least avoiding the single-step. | |
241 | ||
a1dac767 MCC |
242 | Optimization |
243 | ^^^^^^^^^^^^ | |
b26486bf MH |
244 | |
245 | The Kprobe-optimizer doesn't insert the jump instruction immediately; | |
1755eced | 246 | rather, it calls synchronize_rcu() for safety first, because it's |
b26486bf | 247 | possible for a CPU to be interrupted in the middle of executing the |
1755eced PM |
248 | optimized region [3]_. As you know, synchronize_rcu() can ensure |
249 | that all interruptions that were active when synchronize_rcu() | |
b26486bf | 250 | was called are done, but only if CONFIG_PREEMPT=n. So, this version |
a1dac767 | 251 | of kprobe optimization supports only kernels with CONFIG_PREEMPT=n [4]_. |
b26486bf MH |
252 | |
253 | After that, the Kprobe-optimizer calls stop_machine() to replace | |
254 | the optimized region with a jump instruction to the detour buffer, | |
255 | using text_poke_smp(). | |
256 | ||
a1dac767 MCC |
257 | Unoptimization |
258 | ^^^^^^^^^^^^^^ | |
b26486bf MH |
259 | |
260 | When an optimized kprobe is unregistered, disabled, or blocked by | |
261 | another kprobe, it will be unoptimized. If this happens before | |
262 | the optimization is complete, the kprobe is just dequeued from the | |
263 | optimized list. If the optimization has been done, the jump is | |
264 | replaced with the original code (except for an int3 breakpoint in | |
265 | the first byte) by using text_poke_smp(). | |
266 | ||
a1dac767 MCC |
267 | .. [3] Please imagine that the 2nd instruction is interrupted and then |
268 | the optimizer replaces the 2nd instruction with the jump *address* | |
269 | while the interrupt handler is running. When the interrupt | |
270 | returns to original address, there is no valid instruction, | |
271 | and it causes an unexpected result. | |
b26486bf | 272 | |
a1dac767 MCC |
273 | .. [4] This optimization-safety checking may be replaced with the |
274 | stop-machine method that ksplice uses for supporting a CONFIG_PREEMPT=y | |
275 | kernel. | |
b26486bf MH |
276 | |
277 | NOTE for geeks: | |
278 | The jump optimization changes the kprobe's pre_handler behavior. | |
279 | Without optimization, the pre_handler can change the kernel's execution | |
280 | path by changing regs->ip and returning 1. However, when the probe | |
281 | is optimized, that modification is ignored. Thus, if you want to | |
282 | tweak the kernel's execution path, you need to suppress optimization, | |
283 | using one of the following techniques: | |
a1dac767 | 284 | |
059053a2 | 285 | - Specify an empty function for the kprobe's post_handler. |
a1dac767 MCC |
286 | |
287 | or | |
288 | ||
b26486bf MH |
289 | - Execute 'sysctl -w debug.kprobes_optimization=n' |
290 | ||
a1dac767 MCC |
291 | .. _kprobes_blacklist: |
292 | ||
293 | Blacklist | |
294 | --------- | |
376e2424 MH |
295 | |
296 | Kprobes can probe most of the kernel except itself. This means | |
297 | that there are some functions where kprobes cannot probe. Probing | |
298 | (trapping) such functions can cause a recursive trap (e.g. double | |
299 | fault) or the nested probe handler may never be called. | |
300 | Kprobes manages such functions as a blacklist. | |
301 | If you want to add a function into the blacklist, you just need | |
302 | to (1) include linux/kprobes.h and (2) use NOKPROBE_SYMBOL() macro | |
303 | to specify a blacklisted function. | |
304 | Kprobes checks the given probe address against the blacklist and | |
305 | rejects registering it, if the given address is in the blacklist. | |
306 | ||
a1dac767 MCC |
307 | .. _kprobes_archs_supported: |
308 | ||
309 | Architectures Supported | |
310 | ======================= | |
d27a4ddd | 311 | |
9b17374e | 312 | Kprobes and return probes are implemented on the following |
d27a4ddd JK |
313 | architectures: |
314 | ||
b26486bf MH |
315 | - i386 (Supports jump optimization) |
316 | - x86_64 (AMD-64, EM64T) (Supports jump optimization) | |
d27a4ddd | 317 | - ppc64 |
d27a4ddd | 318 | - sparc64 (Return probes not yet implemented.) |
5de865b4 | 319 | - arm |
f8279621 | 320 | - ppc |
9bb4d9df | 321 | - mips |
369e8c35 | 322 | - s390 |
376e5fd7 | 323 | - parisc |
d27a4ddd | 324 | |
a1dac767 MCC |
325 | Configuring Kprobes |
326 | =================== | |
d27a4ddd JK |
327 | |
328 | When configuring the kernel using make menuconfig/xconfig/oldconfig, | |
e94102e5 TY |
329 | ensure that CONFIG_KPROBES is set to "y", look for "Kprobes" under |
330 | "General architecture-dependent options". | |
8861da31 JK |
331 | |
332 | So that you can load and unload Kprobes-based instrumentation modules, | |
333 | make sure "Loadable module support" (CONFIG_MODULES) and "Module | |
334 | unloading" (CONFIG_MODULE_UNLOAD) are set to "y". | |
d27a4ddd | 335 | |
09b18203 AM |
336 | Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL |
337 | are set to "y", since kallsyms_lookup_name() is used by the in-kernel | |
338 | kprobe address resolution code. | |
d27a4ddd JK |
339 | |
340 | If you need to insert a probe in the middle of a function, you may find | |
341 | it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO), | |
342 | so you can use "objdump -d -l vmlinux" to see the source-to-object | |
343 | code mapping. | |
344 | ||
a1dac767 MCC |
345 | API Reference |
346 | ============= | |
d27a4ddd JK |
347 | |
348 | The Kprobes API includes a "register" function and an "unregister" | |
3b0cb4ca MH |
349 | function for each type of probe. The API also includes "register_*probes" |
350 | and "unregister_*probes" functions for (un)registering arrays of probes. | |
351 | Here are terse, mini-man-page specifications for these functions and | |
352 | the associated probe handlers that you'll write. See the files in the | |
353 | samples/kprobes/ sub-directory for examples. | |
d27a4ddd | 354 | |
a1dac767 MCC |
355 | register_kprobe |
356 | --------------- | |
d27a4ddd | 357 | |
a1dac767 MCC |
358 | :: |
359 | ||
360 | #include <linux/kprobes.h> | |
361 | int register_kprobe(struct kprobe *kp); | |
d27a4ddd | 362 | |
ec6aba3d PZ |
363 | Sets a breakpoint at the address kp->addr. When the breakpoint is hit, Kprobes |
364 | calls kp->pre_handler. After the probed instruction is single-stepped, Kprobe | |
365 | calls kp->post_handler. Any or all handlers can be NULL. If kp->flags is set | |
366 | KPROBE_FLAG_DISABLED, that kp will be registered but disabled, so, its handlers | |
367 | aren't hit until calling enable_kprobe(kp). | |
d27a4ddd | 368 | |
a1dac767 MCC |
369 | .. note:: |
370 | ||
371 | 1. With the introduction of the "symbol_name" field to struct kprobe, | |
372 | the probepoint address resolution will now be taken care of by the kernel. | |
373 | The following will now work:: | |
09b18203 AM |
374 | |
375 | kp.symbol_name = "symbol_name"; | |
376 | ||
a1dac767 MCC |
377 | (64-bit powerpc intricacies such as function descriptors are handled |
378 | transparently) | |
09b18203 | 379 | |
a1dac767 MCC |
380 | 2. Use the "offset" field of struct kprobe if the offset into the symbol |
381 | to install a probepoint is known. This field is used to calculate the | |
382 | probepoint. | |
09b18203 | 383 | |
a1dac767 MCC |
384 | 3. Specify either the kprobe "symbol_name" OR the "addr". If both are |
385 | specified, kprobe registration will fail with -EINVAL. | |
09b18203 | 386 | |
a1dac767 MCC |
387 | 4. With CISC architectures (such as i386 and x86_64), the kprobes code |
388 | does not validate if the kprobe.addr is at an instruction boundary. | |
389 | Use "offset" with caution. | |
09b18203 | 390 | |
d27a4ddd JK |
391 | register_kprobe() returns 0 on success, or a negative errno otherwise. |
392 | ||
a1dac767 MCC |
393 | User's pre-handler (kp->pre_handler):: |
394 | ||
395 | #include <linux/kprobes.h> | |
396 | #include <linux/ptrace.h> | |
397 | int pre_handler(struct kprobe *p, struct pt_regs *regs); | |
d27a4ddd JK |
398 | |
399 | Called with p pointing to the kprobe associated with the breakpoint, | |
400 | and regs pointing to the struct containing the registers saved when | |
401 | the breakpoint was hit. Return 0 here unless you're a Kprobes geek. | |
402 | ||
a1dac767 MCC |
403 | User's post-handler (kp->post_handler):: |
404 | ||
405 | #include <linux/kprobes.h> | |
406 | #include <linux/ptrace.h> | |
407 | void post_handler(struct kprobe *p, struct pt_regs *regs, | |
408 | unsigned long flags); | |
d27a4ddd JK |
409 | |
410 | p and regs are as described for the pre_handler. flags always seems | |
411 | to be zero. | |
412 | ||
a1dac767 MCC |
413 | register_kretprobe |
414 | ------------------ | |
d27a4ddd | 415 | |
a1dac767 MCC |
416 | :: |
417 | ||
418 | #include <linux/kprobes.h> | |
419 | int register_kretprobe(struct kretprobe *rp); | |
d27a4ddd JK |
420 | |
421 | Establishes a return probe for the function whose address is | |
422 | rp->kp.addr. When that function returns, Kprobes calls rp->handler. | |
423 | You must set rp->maxactive appropriately before you call | |
424 | register_kretprobe(); see "How Does a Return Probe Work?" for details. | |
425 | ||
426 | register_kretprobe() returns 0 on success, or a negative errno | |
427 | otherwise. | |
428 | ||
a1dac767 MCC |
429 | User's return-probe handler (rp->handler):: |
430 | ||
431 | #include <linux/kprobes.h> | |
432 | #include <linux/ptrace.h> | |
433 | int kretprobe_handler(struct kretprobe_instance *ri, | |
434 | struct pt_regs *regs); | |
d27a4ddd JK |
435 | |
436 | regs is as described for kprobe.pre_handler. ri points to the | |
437 | kretprobe_instance object, of which the following fields may be | |
438 | of interest: | |
a1dac767 | 439 | |
d27a4ddd JK |
440 | - ret_addr: the return address |
441 | - rp: points to the corresponding kretprobe object | |
442 | - task: points to the corresponding task struct | |
f47cd9b5 AS |
443 | - data: points to per return-instance private data; see "Kretprobe |
444 | entry-handler" for details. | |
09b18203 AM |
445 | |
446 | The regs_return_value(regs) macro provides a simple abstraction to | |
447 | extract the return value from the appropriate register as defined by | |
448 | the architecture's ABI. | |
449 | ||
d27a4ddd JK |
450 | The handler's return value is currently ignored. |
451 | ||
a1dac767 MCC |
452 | unregister_*probe |
453 | ------------------ | |
d27a4ddd | 454 | |
a1dac767 MCC |
455 | :: |
456 | ||
457 | #include <linux/kprobes.h> | |
458 | void unregister_kprobe(struct kprobe *kp); | |
a1dac767 | 459 | void unregister_kretprobe(struct kretprobe *rp); |
d27a4ddd JK |
460 | |
461 | Removes the specified probe. The unregister function can be called | |
462 | at any time after the probe has been registered. | |
463 | ||
a1dac767 MCC |
464 | .. note:: |
465 | ||
466 | If the functions find an incorrect probe (ex. an unregistered probe), | |
467 | they clear the addr field of the probe. | |
3b0cb4ca | 468 | |
a1dac767 MCC |
469 | register_*probes |
470 | ---------------- | |
3b0cb4ca | 471 | |
a1dac767 MCC |
472 | :: |
473 | ||
474 | #include <linux/kprobes.h> | |
475 | int register_kprobes(struct kprobe **kps, int num); | |
476 | int register_kretprobes(struct kretprobe **rps, int num); | |
3b0cb4ca MH |
477 | |
478 | Registers each of the num probes in the specified array. If any | |
479 | error occurs during registration, all probes in the array, up to | |
480 | the bad probe, are safely unregistered before the register_*probes | |
481 | function returns. | |
a1dac767 | 482 | |
bc8c9da5 | 483 | - kps/rps: an array of pointers to ``*probe`` data structures |
3b0cb4ca MH |
484 | - num: the number of the array entries. |
485 | ||
a1dac767 MCC |
486 | .. note:: |
487 | ||
488 | You have to allocate(or define) an array of pointers and set all | |
489 | of the array entries before using these functions. | |
490 | ||
491 | unregister_*probes | |
492 | ------------------ | |
3b0cb4ca | 493 | |
a1dac767 | 494 | :: |
3b0cb4ca | 495 | |
a1dac767 MCC |
496 | #include <linux/kprobes.h> |
497 | void unregister_kprobes(struct kprobe **kps, int num); | |
498 | void unregister_kretprobes(struct kretprobe **rps, int num); | |
3b0cb4ca MH |
499 | |
500 | Removes each of the num probes in the specified array at once. | |
501 | ||
a1dac767 | 502 | .. note:: |
3b0cb4ca | 503 | |
a1dac767 MCC |
504 | If the functions find some incorrect probes (ex. unregistered |
505 | probes) in the specified array, they clear the addr field of those | |
506 | incorrect probes. However, other probes in the array are | |
507 | unregistered correctly. | |
de5bd88d | 508 | |
a1dac767 MCC |
509 | disable_*probe |
510 | -------------- | |
de5bd88d | 511 | |
a1dac767 MCC |
512 | :: |
513 | ||
514 | #include <linux/kprobes.h> | |
515 | int disable_kprobe(struct kprobe *kp); | |
516 | int disable_kretprobe(struct kretprobe *rp); | |
a1dac767 MCC |
517 | |
518 | Temporarily disables the specified ``*probe``. You can enable it again by using | |
8f9b1528 | 519 | enable_*probe(). You must specify the probe which has been registered. |
de5bd88d | 520 | |
a1dac767 MCC |
521 | enable_*probe |
522 | ------------- | |
de5bd88d | 523 | |
a1dac767 | 524 | :: |
de5bd88d | 525 | |
a1dac767 MCC |
526 | #include <linux/kprobes.h> |
527 | int enable_kprobe(struct kprobe *kp); | |
528 | int enable_kretprobe(struct kretprobe *rp); | |
a1dac767 MCC |
529 | |
530 | Enables ``*probe`` which has been disabled by disable_*probe(). You must specify | |
8f9b1528 | 531 | the probe which has been registered. |
de5bd88d | 532 | |
a1dac767 MCC |
533 | Kprobes Features and Limitations |
534 | ================================ | |
d27a4ddd | 535 | |
9b17374e MH |
536 | Kprobes allows multiple probes at the same address. Also, |
537 | a probepoint for which there is a post_handler cannot be optimized. | |
538 | So if you install a kprobe with a post_handler, at an optimized | |
539 | probepoint, the probepoint will be unoptimized automatically. | |
d27a4ddd JK |
540 | |
541 | In general, you can install a probe anywhere in the kernel. | |
542 | In particular, you can probe interrupt handlers. Known exceptions | |
543 | are discussed in this section. | |
544 | ||
8861da31 JK |
545 | The register_*probe functions will return -EINVAL if you attempt |
546 | to install a probe in the code that implements Kprobes (mostly | |
a1dac767 | 547 | kernel/kprobes.c and ``arch/*/kernel/kprobes.c``, but also functions such |
8861da31 | 548 | as do_page_fault and notifier_call_chain). |
d27a4ddd JK |
549 | |
550 | If you install a probe in an inline-able function, Kprobes makes | |
551 | no attempt to chase down all inline instances of the function and | |
552 | install probes there. gcc may inline a function without being asked, | |
553 | so keep this in mind if you're not seeing the probe hits you expect. | |
554 | ||
555 | A probe handler can modify the environment of the probed function | |
556 | -- e.g., by modifying kernel data structures, or by modifying the | |
557 | contents of the pt_regs struct (which are restored to the registers | |
558 | upon return from the breakpoint). So Kprobes can be used, for example, | |
559 | to install a bug fix or to inject faults for testing. Kprobes, of | |
560 | course, has no way to distinguish the deliberately injected faults | |
561 | from the accidental ones. Don't drink and probe. | |
562 | ||
563 | Kprobes makes no attempt to prevent probe handlers from stepping on | |
564 | each other -- e.g., probing printk() and then calling printk() from a | |
8861da31 JK |
565 | probe handler. If a probe handler hits a probe, that second probe's |
566 | handlers won't be run in that instance, and the kprobe.nmissed member | |
567 | of the second probe will be incremented. | |
568 | ||
569 | As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of | |
570 | the same handler) may run concurrently on different CPUs. | |
571 | ||
572 | Kprobes does not use mutexes or allocate memory except during | |
d27a4ddd JK |
573 | registration and unregistration. |
574 | ||
2bbda764 MH |
575 | Probe handlers are run with preemption disabled or interrupt disabled, |
576 | which depends on the architecture and optimization state. (e.g., | |
577 | kretprobe handlers and optimized kprobe handlers run without interrupt | |
578 | disabled on x86/x86-64). In any case, your handler should not yield | |
579 | the CPU (e.g., by attempting to acquire a semaphore, or waiting I/O). | |
d27a4ddd JK |
580 | |
581 | Since a return probe is implemented by replacing the return | |
582 | address with the trampoline's address, stack backtraces and calls | |
583 | to __builtin_return_address() will typically yield the trampoline's | |
584 | address instead of the real return address for kretprobed functions. | |
585 | (As far as we can tell, __builtin_return_address() is used only | |
586 | for instrumentation and error reporting.) | |
587 | ||
8861da31 JK |
588 | If the number of times a function is called does not match the number |
589 | of times it returns, registering a return probe on that function may | |
bf8f6e5b AM |
590 | produce undesirable results. In such a case, a line: |
591 | kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c | |
592 | gets printed. With this information, one will be able to correlate the | |
593 | exact instance of the kretprobe that caused the problem. We have the | |
594 | do_exit() case covered. do_execve() and do_fork() are not an issue. | |
595 | We're unaware of other specific cases where this could be a problem. | |
8861da31 JK |
596 | |
597 | If, upon entry to or exit from a function, the CPU is running on | |
598 | a stack other than that of the current task, registering a return | |
599 | probe on that function may produce undesirable results. For this | |
9b17374e | 600 | reason, Kprobes doesn't support return probes (or kprobes) |
8861da31 JK |
601 | on the x86_64 version of __switch_to(); the registration functions |
602 | return -EINVAL. | |
d27a4ddd | 603 | |
b26486bf MH |
604 | On x86/x86-64, since the Jump Optimization of Kprobes modifies |
605 | instructions widely, there are some limitations to optimization. To | |
606 | explain it, we introduce some terminology. Imagine a 3-instruction | |
607 | sequence consisting of a two 2-byte instructions and one 3-byte | |
608 | instruction. | |
609 | ||
a1dac767 MCC |
610 | :: |
611 | ||
612 | IA | |
613 | | | |
614 | [-2][-1][0][1][2][3][4][5][6][7] | |
615 | [ins1][ins2][ ins3 ] | |
616 | [<- DCR ->] | |
617 | [<- JTPR ->] | |
b26486bf | 618 | |
a1dac767 MCC |
619 | ins1: 1st Instruction |
620 | ins2: 2nd Instruction | |
621 | ins3: 3rd Instruction | |
622 | IA: Insertion Address | |
623 | JTPR: Jump Target Prohibition Region | |
624 | DCR: Detoured Code Region | |
b26486bf MH |
625 | |
626 | The instructions in DCR are copied to the out-of-line buffer | |
627 | of the kprobe, because the bytes in DCR are replaced by | |
628 | a 5-byte jump instruction. So there are several limitations. | |
629 | ||
630 | a) The instructions in DCR must be relocatable. | |
631 | b) The instructions in DCR must not include a call instruction. | |
632 | c) JTPR must not be targeted by any jump or call instruction. | |
b595076a | 633 | d) DCR must not straddle the border between functions. |
b26486bf MH |
634 | |
635 | Anyway, these limitations are checked by the in-kernel instruction | |
636 | decoder, so you don't need to worry about that. | |
637 | ||
a1dac767 MCC |
638 | Probe Overhead |
639 | ============== | |
d27a4ddd JK |
640 | |
641 | On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0 | |
642 | microseconds to process. Specifically, a benchmark that hits the same | |
643 | probepoint repeatedly, firing a simple handler each time, reports 1-2 | |
9b17374e MH |
644 | million hits per second, depending on the architecture. A return-probe |
645 | hit typically takes 50-75% longer than a kprobe hit. | |
d27a4ddd JK |
646 | When you have a return probe set on a function, adding a kprobe at |
647 | the entry to that function adds essentially no overhead. | |
648 | ||
a1dac767 | 649 | Here are sample overhead figures (in usec) for different architectures:: |
d27a4ddd | 650 | |
9b17374e MH |
651 | k = kprobe; r = return probe; kr = kprobe + return probe |
652 | on same function | |
d27a4ddd | 653 | |
a1dac767 | 654 | i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips |
9b17374e | 655 | k = 0.57 usec; r = 0.92; kr = 0.99 |
d27a4ddd | 656 | |
a1dac767 | 657 | x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips |
9b17374e | 658 | k = 0.49 usec; r = 0.80; kr = 0.82 |
d27a4ddd | 659 | |
a1dac767 | 660 | ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU) |
9b17374e | 661 | k = 0.77 usec; r = 1.26; kr = 1.45 |
a1dac767 MCC |
662 | |
663 | Optimized Probe Overhead | |
664 | ------------------------ | |
b26486bf MH |
665 | |
666 | Typically, an optimized kprobe hit takes 0.07 to 0.1 microseconds to | |
a1dac767 MCC |
667 | process. Here are sample overhead figures (in usec) for x86 architectures:: |
668 | ||
669 | k = unoptimized kprobe, b = boosted (single-step skipped), o = optimized kprobe, | |
670 | r = unoptimized kretprobe, rb = boosted kretprobe, ro = optimized kretprobe. | |
b26486bf | 671 | |
a1dac767 MCC |
672 | i386: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips |
673 | k = 0.80 usec; b = 0.33; o = 0.05; r = 1.10; rb = 0.61; ro = 0.33 | |
b26486bf | 674 | |
a1dac767 MCC |
675 | x86-64: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips |
676 | k = 0.99 usec; b = 0.43; o = 0.06; r = 1.24; rb = 0.68; ro = 0.30 | |
b26486bf | 677 | |
a1dac767 MCC |
678 | TODO |
679 | ==== | |
d27a4ddd | 680 | |
8861da31 | 681 | a. SystemTap (http://sourceware.org/systemtap): Provides a simplified |
a1dac767 | 682 | programming interface for probe-based instrumentation. Try it out. |
8861da31 JK |
683 | b. Kernel return probes for sparc64. |
684 | c. Support for other architectures. | |
685 | d. User-space probes. | |
686 | e. Watchpoint probes (which fire on data references). | |
d27a4ddd | 687 | |
a1dac767 MCC |
688 | Kprobes Example |
689 | =============== | |
d27a4ddd | 690 | |
804defea | 691 | See samples/kprobes/kprobe_example.c |
d27a4ddd | 692 | |
a1dac767 MCC |
693 | Kretprobes Example |
694 | ================== | |
d27a4ddd | 695 | |
804defea | 696 | See samples/kprobes/kretprobe_example.c |
d27a4ddd | 697 | |
9b17374e MH |
698 | Deprecated Features |
699 | =================== | |
700 | ||
701 | Jprobes is now a deprecated feature. People who are depending on it should | |
702 | migrate to other tracing features or use older kernels. Please consider to | |
703 | migrate your tool to one of the following options: | |
704 | ||
705 | - Use trace-event to trace target function with arguments. | |
706 | ||
707 | trace-event is a low-overhead (and almost no visible overhead if it | |
708 | is off) statically defined event interface. You can define new events | |
709 | and trace it via ftrace or any other tracing tools. | |
710 | ||
711 | See the following urls: | |
712 | ||
713 | - https://lwn.net/Articles/379903/ | |
714 | - https://lwn.net/Articles/381064/ | |
715 | - https://lwn.net/Articles/383362/ | |
716 | ||
717 | - Use ftrace dynamic events (kprobe event) with perf-probe. | |
718 | ||
719 | If you build your kernel with debug info (CONFIG_DEBUG_INFO=y), you can | |
720 | find which register/stack is assigned to which local variable or arguments | |
721 | by using perf-probe and set up new event to trace it. | |
722 | ||
723 | See following documents: | |
724 | ||
5fb94e9c MCC |
725 | - Documentation/trace/kprobetrace.rst |
726 | - Documentation/trace/events.rst | |
9b17374e MH |
727 | - tools/perf/Documentation/perf-probe.txt |
728 | ||
a1dac767 MCC |
729 | |
730 | The kprobes debugfs interface | |
731 | ============================= | |
bf8f6e5b | 732 | |
bf8f6e5b AM |
733 | |
734 | With recent kernels (> 2.6.20) the list of registered kprobes is visible | |
156f5a78 | 735 | under the /sys/kernel/debug/kprobes/ directory (assuming debugfs is mounted at //sys/kernel/debug). |
bf8f6e5b | 736 | |
a1dac767 | 737 | /sys/kernel/debug/kprobes/list: Lists all registered probes on the system:: |
bf8f6e5b | 738 | |
a1dac767 | 739 | c015d71a k vfs_read+0x0 |
a1dac767 | 740 | c03dedc5 r tcp_v4_rcv+0x0 |
bf8f6e5b AM |
741 | |
742 | The first column provides the kernel address where the probe is inserted. | |
9b17374e MH |
743 | The second column identifies the type of probe (k - kprobe and r - kretprobe) |
744 | while the third column specifies the symbol+offset of the probe. | |
745 | If the probed function belongs to a module, the module name is also | |
746 | specified. Following columns show probe status. If the probe is on | |
e8386a0c MH |
747 | a virtual address that is no longer valid (module init sections, module |
748 | virtual addresses that correspond to modules that've been unloaded), | |
de5bd88d | 749 | such probes are marked with [GONE]. If the probe is temporarily disabled, |
b26486bf | 750 | such probes are marked with [DISABLED]. If the probe is optimized, it is |
9ed330d3 WL |
751 | marked with [OPTIMIZED]. If the probe is ftrace-based, it is marked with |
752 | [FTRACE]. | |
bf8f6e5b | 753 | |
156f5a78 | 754 | /sys/kernel/debug/kprobes/enabled: Turn kprobes ON/OFF forcibly. |
bf8f6e5b | 755 | |
de5bd88d MH |
756 | Provides a knob to globally and forcibly turn registered kprobes ON or OFF. |
757 | By default, all kprobes are enabled. By echoing "0" to this file, all | |
758 | registered probes will be disarmed, till such time a "1" is echoed to this | |
759 | file. Note that this knob just disarms and arms all kprobes and doesn't | |
760 | change each probe's disabling state. This means that disabled kprobes (marked | |
761 | [DISABLED]) will be not enabled if you turn ON all kprobes by this knob. | |
b26486bf MH |
762 | |
763 | ||
a1dac767 MCC |
764 | The kprobes sysctl interface |
765 | ============================ | |
b26486bf MH |
766 | |
767 | /proc/sys/debug/kprobes-optimization: Turn kprobes optimization ON/OFF. | |
768 | ||
769 | When CONFIG_OPTPROBES=y, this sysctl interface appears and it provides | |
770 | a knob to globally and forcibly turn jump optimization (see section | |
a1dac767 MCC |
771 | :ref:`kprobes_jump_optimization`) ON or OFF. By default, jump optimization |
772 | is allowed (ON). If you echo "0" to this file or set | |
773 | "debug.kprobes_optimization" to 0 via sysctl, all optimized probes will be | |
774 | unoptimized, and any new probes registered after that will not be optimized. | |
43e5f7e1 MCC |
775 | |
776 | Note that this knob *changes* the optimized state. This means that optimized | |
a1dac767 | 777 | probes (marked [OPTIMIZED]) will be unoptimized ([OPTIMIZED] tag will be |
b26486bf MH |
778 | removed). If the knob is turned on, they will be optimized again. |
779 | ||
0c75f127 NR |
780 | References |
781 | ========== | |
782 | ||
783 | For additional information on Kprobes, refer to the following URLs: | |
784 | ||
438697a3 | 785 | - https://lwn.net/Articles/132196/ |
0c75f127 NR |
786 | - https://www.kernel.org/doc/ols/2006/ols2006v2-pages-109-124.pdf |
787 |