]>
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> | |
7 | :Author: Masami Hiramatsu <mhiramat@redhat.com> | |
8 | ||
9 | .. CONTENTS | |
10 | ||
11 | 1. Concepts: Kprobes, Jprobes, Return Probes | |
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 | |
19 | 9. Jprobes Example | |
20 | 10. Kretprobes Example | |
21 | Appendix A: The kprobes debugfs interface | |
22 | Appendix B: The kprobes sysctl interface | |
23 | ||
24 | Concepts: Kprobes, Jprobes, Return Probes | |
25 | ========================================= | |
d27a4ddd JK |
26 | |
27 | Kprobes enables you to dynamically break into any kernel routine and | |
28 | collect debugging and performance information non-disruptively. You | |
a1dac767 | 29 | can trap at almost any kernel code address [1]_, specifying a handler |
d27a4ddd | 30 | routine to be invoked when the breakpoint is hit. |
a1dac767 MCC |
31 | |
32 | .. [1] some parts of the kernel code can not be trapped, see | |
33 | :ref:`kprobes_blacklist`) | |
d27a4ddd JK |
34 | |
35 | There are currently three types of probes: kprobes, jprobes, and | |
36 | kretprobes (also called return probes). A kprobe can be inserted | |
37 | on virtually any instruction in the kernel. A jprobe is inserted at | |
38 | the entry to a kernel function, and provides convenient access to the | |
39 | function's arguments. A return probe fires when a specified function | |
40 | returns. | |
41 | ||
42 | In the typical case, Kprobes-based instrumentation is packaged as | |
43 | a kernel module. The module's init function installs ("registers") | |
44 | one or more probes, and the exit function unregisters them. A | |
45 | registration function such as register_kprobe() specifies where | |
46 | the probe is to be inserted and what handler is to be called when | |
47 | the probe is hit. | |
48 | ||
a1dac767 MCC |
49 | There are also ``register_/unregister_*probes()`` functions for batch |
50 | registration/unregistration of a group of ``*probes``. These functions | |
3b0cb4ca MH |
51 | can speed up unregistration process when you have to unregister |
52 | a lot of probes at once. | |
53 | ||
b26486bf MH |
54 | The next four subsections explain how the different types of |
55 | probes work and how jump optimization works. They explain certain | |
56 | things that you'll need to know in order to make the best use of | |
57 | Kprobes -- e.g., the difference between a pre_handler and | |
58 | a post_handler, and how to use the maxactive and nmissed fields of | |
59 | a kretprobe. But if you're in a hurry to start using Kprobes, you | |
a1dac767 | 60 | can skip ahead to :ref:`kprobes_archs_supported`. |
d27a4ddd | 61 | |
a1dac767 MCC |
62 | How Does a Kprobe Work? |
63 | ----------------------- | |
d27a4ddd JK |
64 | |
65 | When a kprobe is registered, Kprobes makes a copy of the probed | |
66 | instruction and replaces the first byte(s) of the probed instruction | |
67 | with a breakpoint instruction (e.g., int3 on i386 and x86_64). | |
68 | ||
69 | When a CPU hits the breakpoint instruction, a trap occurs, the CPU's | |
70 | registers are saved, and control passes to Kprobes via the | |
71 | notifier_call_chain mechanism. Kprobes executes the "pre_handler" | |
72 | associated with the kprobe, passing the handler the addresses of the | |
73 | kprobe struct and the saved registers. | |
74 | ||
75 | Next, Kprobes single-steps its copy of the probed instruction. | |
76 | (It would be simpler to single-step the actual instruction in place, | |
77 | but then Kprobes would have to temporarily remove the breakpoint | |
78 | instruction. This would open a small time window when another CPU | |
79 | could sail right past the probepoint.) | |
80 | ||
81 | After the instruction is single-stepped, Kprobes executes the | |
82 | "post_handler," if any, that is associated with the kprobe. | |
83 | Execution then continues with the instruction following the probepoint. | |
84 | ||
a1dac767 MCC |
85 | How Does a Jprobe Work? |
86 | ----------------------- | |
d27a4ddd JK |
87 | |
88 | A jprobe is implemented using a kprobe that is placed on a function's | |
89 | entry point. It employs a simple mirroring principle to allow | |
90 | seamless access to the probed function's arguments. The jprobe | |
91 | handler routine should have the same signature (arg list and return | |
92 | type) as the function being probed, and must always end by calling | |
93 | the Kprobes function jprobe_return(). | |
94 | ||
95 | Here's how it works. When the probe is hit, Kprobes makes a copy of | |
96 | the saved registers and a generous portion of the stack (see below). | |
97 | Kprobes then points the saved instruction pointer at the jprobe's | |
98 | handler routine, and returns from the trap. As a result, control | |
99 | passes to the handler, which is presented with the same register and | |
100 | stack contents as the probed function. When it is done, the handler | |
101 | calls jprobe_return(), which traps again to restore the original stack | |
102 | contents and processor state and switch to the probed function. | |
103 | ||
104 | By convention, the callee owns its arguments, so gcc may produce code | |
105 | that unexpectedly modifies that portion of the stack. This is why | |
106 | Kprobes saves a copy of the stack and restores it after the jprobe | |
107 | handler has run. Up to MAX_STACK_SIZE bytes are copied -- e.g., | |
108 | 64 bytes on i386. | |
109 | ||
110 | Note that the probed function's args may be passed on the stack | |
b5606c2d HH |
111 | or in registers. The jprobe will work in either case, so long as the |
112 | handler's prototype matches that of the probed function. | |
d27a4ddd | 113 | |
7a9011db DL |
114 | Note that in some architectures (e.g.: arm64 and sparc64) the stack |
115 | copy is not done, as the actual location of stacked parameters may be | |
116 | outside of a reasonable MAX_STACK_SIZE value and because that location | |
117 | cannot be determined by the jprobes code. In this case the jprobes | |
118 | user must be careful to make certain the calling signature of the | |
119 | function does not cause parameters to be passed on the stack (e.g.: | |
120 | more than eight function arguments, an argument of more than sixteen | |
121 | bytes, or more than 64 bytes of argument data, depending on | |
122 | architecture). | |
123 | ||
a1dac767 MCC |
124 | Return Probes |
125 | ------------- | |
f47cd9b5 | 126 | |
a1dac767 MCC |
127 | How Does a Return Probe Work? |
128 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ | |
d27a4ddd JK |
129 | |
130 | When you call register_kretprobe(), Kprobes establishes a kprobe at | |
131 | the entry to the function. When the probed function is called and this | |
132 | probe is hit, Kprobes saves a copy of the return address, and replaces | |
133 | the return address with the address of a "trampoline." The trampoline | |
134 | is an arbitrary piece of code -- typically just a nop instruction. | |
135 | At boot time, Kprobes registers a kprobe at the trampoline. | |
136 | ||
137 | When the probed function executes its return instruction, control | |
138 | passes to the trampoline and that probe is hit. Kprobes' trampoline | |
f47cd9b5 AS |
139 | handler calls the user-specified return handler associated with the |
140 | kretprobe, then sets the saved instruction pointer to the saved return | |
141 | address, and that's where execution resumes upon return from the trap. | |
d27a4ddd JK |
142 | |
143 | While the probed function is executing, its return address is | |
144 | stored in an object of type kretprobe_instance. Before calling | |
145 | register_kretprobe(), the user sets the maxactive field of the | |
146 | kretprobe struct to specify how many instances of the specified | |
147 | function can be probed simultaneously. register_kretprobe() | |
148 | pre-allocates the indicated number of kretprobe_instance objects. | |
149 | ||
150 | For example, if the function is non-recursive and is called with a | |
151 | spinlock held, maxactive = 1 should be enough. If the function is | |
152 | non-recursive and can never relinquish the CPU (e.g., via a semaphore | |
153 | or preemption), NR_CPUS should be enough. If maxactive <= 0, it is | |
154 | set to a default value. If CONFIG_PREEMPT is enabled, the default | |
155 | is max(10, 2*NR_CPUS). Otherwise, the default is NR_CPUS. | |
156 | ||
157 | It's not a disaster if you set maxactive too low; you'll just miss | |
158 | some probes. In the kretprobe struct, the nmissed field is set to | |
159 | zero when the return probe is registered, and is incremented every | |
160 | time the probed function is entered but there is no kretprobe_instance | |
161 | object available for establishing the return probe. | |
162 | ||
a1dac767 MCC |
163 | Kretprobe entry-handler |
164 | ^^^^^^^^^^^^^^^^^^^^^^^ | |
f47cd9b5 AS |
165 | |
166 | Kretprobes also provides an optional user-specified handler which runs | |
167 | on function entry. This handler is specified by setting the entry_handler | |
168 | field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the | |
169 | function entry is hit, the user-defined entry_handler, if any, is invoked. | |
170 | If the entry_handler returns 0 (success) then a corresponding return handler | |
171 | is guaranteed to be called upon function return. If the entry_handler | |
172 | returns a non-zero error then Kprobes leaves the return address as is, and | |
173 | the kretprobe has no further effect for that particular function instance. | |
174 | ||
175 | Multiple entry and return handler invocations are matched using the unique | |
176 | kretprobe_instance object associated with them. Additionally, a user | |
177 | may also specify per return-instance private data to be part of each | |
178 | kretprobe_instance object. This is especially useful when sharing private | |
179 | data between corresponding user entry and return handlers. The size of each | |
180 | private data object can be specified at kretprobe registration time by | |
181 | setting the data_size field of the kretprobe struct. This data can be | |
182 | accessed through the data field of each kretprobe_instance object. | |
183 | ||
184 | In case probed function is entered but there is no kretprobe_instance | |
185 | object available, then in addition to incrementing the nmissed count, | |
186 | the user entry_handler invocation is also skipped. | |
187 | ||
a1dac767 MCC |
188 | .. _kprobes_jump_optimization: |
189 | ||
190 | How Does Jump Optimization Work? | |
191 | -------------------------------- | |
b26486bf | 192 | |
5cc718b9 MH |
193 | If your kernel is built with CONFIG_OPTPROBES=y (currently this flag |
194 | is automatically set 'y' on x86/x86-64, non-preemptive kernel) and | |
b26486bf MH |
195 | the "debug.kprobes_optimization" kernel parameter is set to 1 (see |
196 | sysctl(8)), Kprobes tries to reduce probe-hit overhead by using a jump | |
197 | instruction instead of a breakpoint instruction at each probepoint. | |
198 | ||
a1dac767 MCC |
199 | Init a Kprobe |
200 | ^^^^^^^^^^^^^ | |
b26486bf MH |
201 | |
202 | When a probe is registered, before attempting this optimization, | |
203 | Kprobes inserts an ordinary, breakpoint-based kprobe at the specified | |
204 | address. So, even if it's not possible to optimize this particular | |
205 | probepoint, there'll be a probe there. | |
206 | ||
a1dac767 MCC |
207 | Safety Check |
208 | ^^^^^^^^^^^^ | |
b26486bf MH |
209 | |
210 | Before optimizing a probe, Kprobes performs the following safety checks: | |
211 | ||
212 | - Kprobes verifies that the region that will be replaced by the jump | |
a1dac767 MCC |
213 | instruction (the "optimized region") lies entirely within one function. |
214 | (A jump instruction is multiple bytes, and so may overlay multiple | |
215 | instructions.) | |
b26486bf MH |
216 | |
217 | - Kprobes analyzes the entire function and verifies that there is no | |
a1dac767 MCC |
218 | jump into the optimized region. Specifically: |
219 | ||
b26486bf MH |
220 | - the function contains no indirect jump; |
221 | - the function contains no instruction that causes an exception (since | |
a1dac767 MCC |
222 | the fixup code triggered by the exception could jump back into the |
223 | optimized region -- Kprobes checks the exception tables to verify this); | |
b26486bf | 224 | - there is no near jump to the optimized region (other than to the first |
a1dac767 | 225 | byte). |
b26486bf MH |
226 | |
227 | - For each instruction in the optimized region, Kprobes verifies that | |
a1dac767 | 228 | the instruction can be executed out of line. |
b26486bf | 229 | |
a1dac767 MCC |
230 | Preparing Detour Buffer |
231 | ^^^^^^^^^^^^^^^^^^^^^^^ | |
b26486bf MH |
232 | |
233 | Next, Kprobes prepares a "detour" buffer, which contains the following | |
234 | instruction sequence: | |
a1dac767 | 235 | |
b26486bf MH |
236 | - code to push the CPU's registers (emulating a breakpoint trap) |
237 | - a call to the trampoline code which calls user's probe handlers. | |
238 | - code to restore registers | |
239 | - the instructions from the optimized region | |
240 | - a jump back to the original execution path. | |
241 | ||
a1dac767 MCC |
242 | Pre-optimization |
243 | ^^^^^^^^^^^^^^^^ | |
b26486bf MH |
244 | |
245 | After preparing the detour buffer, Kprobes verifies that none of the | |
246 | following situations exist: | |
a1dac767 | 247 | |
b26486bf | 248 | - The probe has either a break_handler (i.e., it's a jprobe) or a |
a1dac767 | 249 | post_handler. |
b26486bf MH |
250 | - Other instructions in the optimized region are probed. |
251 | - The probe is disabled. | |
a1dac767 | 252 | |
b26486bf MH |
253 | In any of the above cases, Kprobes won't start optimizing the probe. |
254 | Since these are temporary situations, Kprobes tries to start | |
255 | optimizing it again if the situation is changed. | |
256 | ||
257 | If the kprobe can be optimized, Kprobes enqueues the kprobe to an | |
258 | optimizing list, and kicks the kprobe-optimizer workqueue to optimize | |
259 | it. If the to-be-optimized probepoint is hit before being optimized, | |
260 | Kprobes returns control to the original instruction path by setting | |
261 | the CPU's instruction pointer to the copied code in the detour buffer | |
262 | -- thus at least avoiding the single-step. | |
263 | ||
a1dac767 MCC |
264 | Optimization |
265 | ^^^^^^^^^^^^ | |
b26486bf MH |
266 | |
267 | The Kprobe-optimizer doesn't insert the jump instruction immediately; | |
268 | rather, it calls synchronize_sched() for safety first, because it's | |
269 | possible for a CPU to be interrupted in the middle of executing the | |
a1dac767 | 270 | optimized region [3]_. As you know, synchronize_sched() can ensure |
b26486bf MH |
271 | that all interruptions that were active when synchronize_sched() |
272 | was called are done, but only if CONFIG_PREEMPT=n. So, this version | |
a1dac767 | 273 | of kprobe optimization supports only kernels with CONFIG_PREEMPT=n [4]_. |
b26486bf MH |
274 | |
275 | After that, the Kprobe-optimizer calls stop_machine() to replace | |
276 | the optimized region with a jump instruction to the detour buffer, | |
277 | using text_poke_smp(). | |
278 | ||
a1dac767 MCC |
279 | Unoptimization |
280 | ^^^^^^^^^^^^^^ | |
b26486bf MH |
281 | |
282 | When an optimized kprobe is unregistered, disabled, or blocked by | |
283 | another kprobe, it will be unoptimized. If this happens before | |
284 | the optimization is complete, the kprobe is just dequeued from the | |
285 | optimized list. If the optimization has been done, the jump is | |
286 | replaced with the original code (except for an int3 breakpoint in | |
287 | the first byte) by using text_poke_smp(). | |
288 | ||
a1dac767 MCC |
289 | .. [3] Please imagine that the 2nd instruction is interrupted and then |
290 | the optimizer replaces the 2nd instruction with the jump *address* | |
291 | while the interrupt handler is running. When the interrupt | |
292 | returns to original address, there is no valid instruction, | |
293 | and it causes an unexpected result. | |
b26486bf | 294 | |
a1dac767 MCC |
295 | .. [4] This optimization-safety checking may be replaced with the |
296 | stop-machine method that ksplice uses for supporting a CONFIG_PREEMPT=y | |
297 | kernel. | |
b26486bf MH |
298 | |
299 | NOTE for geeks: | |
300 | The jump optimization changes the kprobe's pre_handler behavior. | |
301 | Without optimization, the pre_handler can change the kernel's execution | |
302 | path by changing regs->ip and returning 1. However, when the probe | |
303 | is optimized, that modification is ignored. Thus, if you want to | |
304 | tweak the kernel's execution path, you need to suppress optimization, | |
305 | using one of the following techniques: | |
a1dac767 | 306 | |
b26486bf | 307 | - Specify an empty function for the kprobe's post_handler or break_handler. |
a1dac767 MCC |
308 | |
309 | or | |
310 | ||
b26486bf MH |
311 | - Execute 'sysctl -w debug.kprobes_optimization=n' |
312 | ||
a1dac767 MCC |
313 | .. _kprobes_blacklist: |
314 | ||
315 | Blacklist | |
316 | --------- | |
376e2424 MH |
317 | |
318 | Kprobes can probe most of the kernel except itself. This means | |
319 | that there are some functions where kprobes cannot probe. Probing | |
320 | (trapping) such functions can cause a recursive trap (e.g. double | |
321 | fault) or the nested probe handler may never be called. | |
322 | Kprobes manages such functions as a blacklist. | |
323 | If you want to add a function into the blacklist, you just need | |
324 | to (1) include linux/kprobes.h and (2) use NOKPROBE_SYMBOL() macro | |
325 | to specify a blacklisted function. | |
326 | Kprobes checks the given probe address against the blacklist and | |
327 | rejects registering it, if the given address is in the blacklist. | |
328 | ||
a1dac767 MCC |
329 | .. _kprobes_archs_supported: |
330 | ||
331 | Architectures Supported | |
332 | ======================= | |
d27a4ddd JK |
333 | |
334 | Kprobes, jprobes, and return probes are implemented on the following | |
335 | architectures: | |
336 | ||
b26486bf MH |
337 | - i386 (Supports jump optimization) |
338 | - x86_64 (AMD-64, EM64T) (Supports jump optimization) | |
d27a4ddd | 339 | - ppc64 |
8861da31 | 340 | - ia64 (Does not support probes on instruction slot1.) |
d27a4ddd | 341 | - sparc64 (Return probes not yet implemented.) |
5de865b4 | 342 | - arm |
f8279621 | 343 | - ppc |
9bb4d9df | 344 | - mips |
369e8c35 | 345 | - s390 |
d27a4ddd | 346 | |
a1dac767 MCC |
347 | Configuring Kprobes |
348 | =================== | |
d27a4ddd JK |
349 | |
350 | When configuring the kernel using make menuconfig/xconfig/oldconfig, | |
080684c8 LB |
351 | ensure that CONFIG_KPROBES is set to "y". Under "General setup", look |
352 | for "Kprobes". | |
8861da31 JK |
353 | |
354 | So that you can load and unload Kprobes-based instrumentation modules, | |
355 | make sure "Loadable module support" (CONFIG_MODULES) and "Module | |
356 | unloading" (CONFIG_MODULE_UNLOAD) are set to "y". | |
d27a4ddd | 357 | |
09b18203 AM |
358 | Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL |
359 | are set to "y", since kallsyms_lookup_name() is used by the in-kernel | |
360 | kprobe address resolution code. | |
d27a4ddd JK |
361 | |
362 | If you need to insert a probe in the middle of a function, you may find | |
363 | it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO), | |
364 | so you can use "objdump -d -l vmlinux" to see the source-to-object | |
365 | code mapping. | |
366 | ||
a1dac767 MCC |
367 | API Reference |
368 | ============= | |
d27a4ddd JK |
369 | |
370 | The Kprobes API includes a "register" function and an "unregister" | |
3b0cb4ca MH |
371 | function for each type of probe. The API also includes "register_*probes" |
372 | and "unregister_*probes" functions for (un)registering arrays of probes. | |
373 | Here are terse, mini-man-page specifications for these functions and | |
374 | the associated probe handlers that you'll write. See the files in the | |
375 | samples/kprobes/ sub-directory for examples. | |
d27a4ddd | 376 | |
a1dac767 MCC |
377 | register_kprobe |
378 | --------------- | |
d27a4ddd | 379 | |
a1dac767 MCC |
380 | :: |
381 | ||
382 | #include <linux/kprobes.h> | |
383 | int register_kprobe(struct kprobe *kp); | |
d27a4ddd JK |
384 | |
385 | Sets a breakpoint at the address kp->addr. When the breakpoint is | |
386 | hit, Kprobes calls kp->pre_handler. After the probed instruction | |
387 | is single-stepped, Kprobe calls kp->post_handler. If a fault | |
388 | occurs during execution of kp->pre_handler or kp->post_handler, | |
389 | or during single-stepping of the probed instruction, Kprobes calls | |
de5bd88d MH |
390 | kp->fault_handler. Any or all handlers can be NULL. If kp->flags |
391 | is set KPROBE_FLAG_DISABLED, that kp will be registered but disabled, | |
a33f3224 | 392 | so, its handlers aren't hit until calling enable_kprobe(kp). |
d27a4ddd | 393 | |
a1dac767 MCC |
394 | .. note:: |
395 | ||
396 | 1. With the introduction of the "symbol_name" field to struct kprobe, | |
397 | the probepoint address resolution will now be taken care of by the kernel. | |
398 | The following will now work:: | |
09b18203 AM |
399 | |
400 | kp.symbol_name = "symbol_name"; | |
401 | ||
a1dac767 MCC |
402 | (64-bit powerpc intricacies such as function descriptors are handled |
403 | transparently) | |
09b18203 | 404 | |
a1dac767 MCC |
405 | 2. Use the "offset" field of struct kprobe if the offset into the symbol |
406 | to install a probepoint is known. This field is used to calculate the | |
407 | probepoint. | |
09b18203 | 408 | |
a1dac767 MCC |
409 | 3. Specify either the kprobe "symbol_name" OR the "addr". If both are |
410 | specified, kprobe registration will fail with -EINVAL. | |
09b18203 | 411 | |
a1dac767 MCC |
412 | 4. With CISC architectures (such as i386 and x86_64), the kprobes code |
413 | does not validate if the kprobe.addr is at an instruction boundary. | |
414 | Use "offset" with caution. | |
09b18203 | 415 | |
d27a4ddd JK |
416 | register_kprobe() returns 0 on success, or a negative errno otherwise. |
417 | ||
a1dac767 MCC |
418 | User's pre-handler (kp->pre_handler):: |
419 | ||
420 | #include <linux/kprobes.h> | |
421 | #include <linux/ptrace.h> | |
422 | int pre_handler(struct kprobe *p, struct pt_regs *regs); | |
d27a4ddd JK |
423 | |
424 | Called with p pointing to the kprobe associated with the breakpoint, | |
425 | and regs pointing to the struct containing the registers saved when | |
426 | the breakpoint was hit. Return 0 here unless you're a Kprobes geek. | |
427 | ||
a1dac767 MCC |
428 | User's post-handler (kp->post_handler):: |
429 | ||
430 | #include <linux/kprobes.h> | |
431 | #include <linux/ptrace.h> | |
432 | void post_handler(struct kprobe *p, struct pt_regs *regs, | |
433 | unsigned long flags); | |
d27a4ddd JK |
434 | |
435 | p and regs are as described for the pre_handler. flags always seems | |
436 | to be zero. | |
437 | ||
a1dac767 MCC |
438 | User's fault-handler (kp->fault_handler):: |
439 | ||
440 | #include <linux/kprobes.h> | |
441 | #include <linux/ptrace.h> | |
442 | int fault_handler(struct kprobe *p, struct pt_regs *regs, int trapnr); | |
d27a4ddd JK |
443 | |
444 | p and regs are as described for the pre_handler. trapnr is the | |
445 | architecture-specific trap number associated with the fault (e.g., | |
446 | on i386, 13 for a general protection fault or 14 for a page fault). | |
447 | Returns 1 if it successfully handled the exception. | |
448 | ||
a1dac767 MCC |
449 | register_jprobe |
450 | --------------- | |
451 | ||
452 | :: | |
d27a4ddd | 453 | |
a1dac767 MCC |
454 | #include <linux/kprobes.h> |
455 | int register_jprobe(struct jprobe *jp) | |
d27a4ddd JK |
456 | |
457 | Sets a breakpoint at the address jp->kp.addr, which must be the address | |
458 | of the first instruction of a function. When the breakpoint is hit, | |
459 | Kprobes runs the handler whose address is jp->entry. | |
460 | ||
461 | The handler should have the same arg list and return type as the probed | |
462 | function; and just before it returns, it must call jprobe_return(). | |
463 | (The handler never actually returns, since jprobe_return() returns | |
b5606c2d HH |
464 | control to Kprobes.) If the probed function is declared asmlinkage |
465 | or anything else that affects how args are passed, the handler's | |
466 | declaration must match. | |
d27a4ddd JK |
467 | |
468 | register_jprobe() returns 0 on success, or a negative errno otherwise. | |
469 | ||
a1dac767 MCC |
470 | register_kretprobe |
471 | ------------------ | |
d27a4ddd | 472 | |
a1dac767 MCC |
473 | :: |
474 | ||
475 | #include <linux/kprobes.h> | |
476 | int register_kretprobe(struct kretprobe *rp); | |
d27a4ddd JK |
477 | |
478 | Establishes a return probe for the function whose address is | |
479 | rp->kp.addr. When that function returns, Kprobes calls rp->handler. | |
480 | You must set rp->maxactive appropriately before you call | |
481 | register_kretprobe(); see "How Does a Return Probe Work?" for details. | |
482 | ||
483 | register_kretprobe() returns 0 on success, or a negative errno | |
484 | otherwise. | |
485 | ||
a1dac767 MCC |
486 | User's return-probe handler (rp->handler):: |
487 | ||
488 | #include <linux/kprobes.h> | |
489 | #include <linux/ptrace.h> | |
490 | int kretprobe_handler(struct kretprobe_instance *ri, | |
491 | struct pt_regs *regs); | |
d27a4ddd JK |
492 | |
493 | regs is as described for kprobe.pre_handler. ri points to the | |
494 | kretprobe_instance object, of which the following fields may be | |
495 | of interest: | |
a1dac767 | 496 | |
d27a4ddd JK |
497 | - ret_addr: the return address |
498 | - rp: points to the corresponding kretprobe object | |
499 | - task: points to the corresponding task struct | |
f47cd9b5 AS |
500 | - data: points to per return-instance private data; see "Kretprobe |
501 | entry-handler" for details. | |
09b18203 AM |
502 | |
503 | The regs_return_value(regs) macro provides a simple abstraction to | |
504 | extract the return value from the appropriate register as defined by | |
505 | the architecture's ABI. | |
506 | ||
d27a4ddd JK |
507 | The handler's return value is currently ignored. |
508 | ||
a1dac767 MCC |
509 | unregister_*probe |
510 | ------------------ | |
d27a4ddd | 511 | |
a1dac767 MCC |
512 | :: |
513 | ||
514 | #include <linux/kprobes.h> | |
515 | void unregister_kprobe(struct kprobe *kp); | |
516 | void unregister_jprobe(struct jprobe *jp); | |
517 | void unregister_kretprobe(struct kretprobe *rp); | |
d27a4ddd JK |
518 | |
519 | Removes the specified probe. The unregister function can be called | |
520 | at any time after the probe has been registered. | |
521 | ||
a1dac767 MCC |
522 | .. note:: |
523 | ||
524 | If the functions find an incorrect probe (ex. an unregistered probe), | |
525 | they clear the addr field of the probe. | |
3b0cb4ca | 526 | |
a1dac767 MCC |
527 | register_*probes |
528 | ---------------- | |
3b0cb4ca | 529 | |
a1dac767 MCC |
530 | :: |
531 | ||
532 | #include <linux/kprobes.h> | |
533 | int register_kprobes(struct kprobe **kps, int num); | |
534 | int register_kretprobes(struct kretprobe **rps, int num); | |
535 | int register_jprobes(struct jprobe **jps, int num); | |
3b0cb4ca MH |
536 | |
537 | Registers each of the num probes in the specified array. If any | |
538 | error occurs during registration, all probes in the array, up to | |
539 | the bad probe, are safely unregistered before the register_*probes | |
540 | function returns. | |
a1dac767 MCC |
541 | |
542 | - kps/rps/jps: an array of pointers to ``*probe`` data structures | |
3b0cb4ca MH |
543 | - num: the number of the array entries. |
544 | ||
a1dac767 MCC |
545 | .. note:: |
546 | ||
547 | You have to allocate(or define) an array of pointers and set all | |
548 | of the array entries before using these functions. | |
549 | ||
550 | unregister_*probes | |
551 | ------------------ | |
3b0cb4ca | 552 | |
a1dac767 | 553 | :: |
3b0cb4ca | 554 | |
a1dac767 MCC |
555 | #include <linux/kprobes.h> |
556 | void unregister_kprobes(struct kprobe **kps, int num); | |
557 | void unregister_kretprobes(struct kretprobe **rps, int num); | |
558 | void unregister_jprobes(struct jprobe **jps, int num); | |
3b0cb4ca MH |
559 | |
560 | Removes each of the num probes in the specified array at once. | |
561 | ||
a1dac767 | 562 | .. note:: |
3b0cb4ca | 563 | |
a1dac767 MCC |
564 | If the functions find some incorrect probes (ex. unregistered |
565 | probes) in the specified array, they clear the addr field of those | |
566 | incorrect probes. However, other probes in the array are | |
567 | unregistered correctly. | |
de5bd88d | 568 | |
a1dac767 MCC |
569 | disable_*probe |
570 | -------------- | |
de5bd88d | 571 | |
a1dac767 MCC |
572 | :: |
573 | ||
574 | #include <linux/kprobes.h> | |
575 | int disable_kprobe(struct kprobe *kp); | |
576 | int disable_kretprobe(struct kretprobe *rp); | |
577 | int disable_jprobe(struct jprobe *jp); | |
578 | ||
579 | Temporarily disables the specified ``*probe``. You can enable it again by using | |
8f9b1528 | 580 | enable_*probe(). You must specify the probe which has been registered. |
de5bd88d | 581 | |
a1dac767 MCC |
582 | enable_*probe |
583 | ------------- | |
de5bd88d | 584 | |
a1dac767 | 585 | :: |
de5bd88d | 586 | |
a1dac767 MCC |
587 | #include <linux/kprobes.h> |
588 | int enable_kprobe(struct kprobe *kp); | |
589 | int enable_kretprobe(struct kretprobe *rp); | |
590 | int enable_jprobe(struct jprobe *jp); | |
591 | ||
592 | Enables ``*probe`` which has been disabled by disable_*probe(). You must specify | |
8f9b1528 | 593 | the probe which has been registered. |
de5bd88d | 594 | |
a1dac767 MCC |
595 | Kprobes Features and Limitations |
596 | ================================ | |
d27a4ddd | 597 | |
8861da31 JK |
598 | Kprobes allows multiple probes at the same address. Currently, |
599 | however, there cannot be multiple jprobes on the same function at | |
b26486bf MH |
600 | the same time. Also, a probepoint for which there is a jprobe or |
601 | a post_handler cannot be optimized. So if you install a jprobe, | |
602 | or a kprobe with a post_handler, at an optimized probepoint, the | |
603 | probepoint will be unoptimized automatically. | |
d27a4ddd JK |
604 | |
605 | In general, you can install a probe anywhere in the kernel. | |
606 | In particular, you can probe interrupt handlers. Known exceptions | |
607 | are discussed in this section. | |
608 | ||
8861da31 JK |
609 | The register_*probe functions will return -EINVAL if you attempt |
610 | to install a probe in the code that implements Kprobes (mostly | |
a1dac767 | 611 | kernel/kprobes.c and ``arch/*/kernel/kprobes.c``, but also functions such |
8861da31 | 612 | as do_page_fault and notifier_call_chain). |
d27a4ddd JK |
613 | |
614 | If you install a probe in an inline-able function, Kprobes makes | |
615 | no attempt to chase down all inline instances of the function and | |
616 | install probes there. gcc may inline a function without being asked, | |
617 | so keep this in mind if you're not seeing the probe hits you expect. | |
618 | ||
619 | A probe handler can modify the environment of the probed function | |
620 | -- e.g., by modifying kernel data structures, or by modifying the | |
621 | contents of the pt_regs struct (which are restored to the registers | |
622 | upon return from the breakpoint). So Kprobes can be used, for example, | |
623 | to install a bug fix or to inject faults for testing. Kprobes, of | |
624 | course, has no way to distinguish the deliberately injected faults | |
625 | from the accidental ones. Don't drink and probe. | |
626 | ||
627 | Kprobes makes no attempt to prevent probe handlers from stepping on | |
628 | each other -- e.g., probing printk() and then calling printk() from a | |
8861da31 JK |
629 | probe handler. If a probe handler hits a probe, that second probe's |
630 | handlers won't be run in that instance, and the kprobe.nmissed member | |
631 | of the second probe will be incremented. | |
632 | ||
633 | As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of | |
634 | the same handler) may run concurrently on different CPUs. | |
635 | ||
636 | Kprobes does not use mutexes or allocate memory except during | |
d27a4ddd JK |
637 | registration and unregistration. |
638 | ||
639 | Probe handlers are run with preemption disabled. Depending on the | |
0f55a2f3 MH |
640 | architecture and optimization state, handlers may also run with |
641 | interrupts disabled (e.g., kretprobe handlers and optimized kprobe | |
642 | handlers run without interrupt disabled on x86/x86-64). In any case, | |
643 | your handler should not yield the CPU (e.g., by attempting to acquire | |
644 | a semaphore). | |
d27a4ddd JK |
645 | |
646 | Since a return probe is implemented by replacing the return | |
647 | address with the trampoline's address, stack backtraces and calls | |
648 | to __builtin_return_address() will typically yield the trampoline's | |
649 | address instead of the real return address for kretprobed functions. | |
650 | (As far as we can tell, __builtin_return_address() is used only | |
651 | for instrumentation and error reporting.) | |
652 | ||
8861da31 JK |
653 | If the number of times a function is called does not match the number |
654 | of times it returns, registering a return probe on that function may | |
bf8f6e5b AM |
655 | produce undesirable results. In such a case, a line: |
656 | kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c | |
657 | gets printed. With this information, one will be able to correlate the | |
658 | exact instance of the kretprobe that caused the problem. We have the | |
659 | do_exit() case covered. do_execve() and do_fork() are not an issue. | |
660 | We're unaware of other specific cases where this could be a problem. | |
8861da31 JK |
661 | |
662 | If, upon entry to or exit from a function, the CPU is running on | |
663 | a stack other than that of the current task, registering a return | |
664 | probe on that function may produce undesirable results. For this | |
665 | reason, Kprobes doesn't support return probes (or kprobes or jprobes) | |
666 | on the x86_64 version of __switch_to(); the registration functions | |
667 | return -EINVAL. | |
d27a4ddd | 668 | |
b26486bf MH |
669 | On x86/x86-64, since the Jump Optimization of Kprobes modifies |
670 | instructions widely, there are some limitations to optimization. To | |
671 | explain it, we introduce some terminology. Imagine a 3-instruction | |
672 | sequence consisting of a two 2-byte instructions and one 3-byte | |
673 | instruction. | |
674 | ||
a1dac767 MCC |
675 | :: |
676 | ||
677 | IA | |
678 | | | |
679 | [-2][-1][0][1][2][3][4][5][6][7] | |
680 | [ins1][ins2][ ins3 ] | |
681 | [<- DCR ->] | |
682 | [<- JTPR ->] | |
b26486bf | 683 | |
a1dac767 MCC |
684 | ins1: 1st Instruction |
685 | ins2: 2nd Instruction | |
686 | ins3: 3rd Instruction | |
687 | IA: Insertion Address | |
688 | JTPR: Jump Target Prohibition Region | |
689 | DCR: Detoured Code Region | |
b26486bf MH |
690 | |
691 | The instructions in DCR are copied to the out-of-line buffer | |
692 | of the kprobe, because the bytes in DCR are replaced by | |
693 | a 5-byte jump instruction. So there are several limitations. | |
694 | ||
695 | a) The instructions in DCR must be relocatable. | |
696 | b) The instructions in DCR must not include a call instruction. | |
697 | c) JTPR must not be targeted by any jump or call instruction. | |
b595076a | 698 | d) DCR must not straddle the border between functions. |
b26486bf MH |
699 | |
700 | Anyway, these limitations are checked by the in-kernel instruction | |
701 | decoder, so you don't need to worry about that. | |
702 | ||
a1dac767 MCC |
703 | Probe Overhead |
704 | ============== | |
d27a4ddd JK |
705 | |
706 | On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0 | |
707 | microseconds to process. Specifically, a benchmark that hits the same | |
708 | probepoint repeatedly, firing a simple handler each time, reports 1-2 | |
709 | million hits per second, depending on the architecture. A jprobe or | |
710 | return-probe hit typically takes 50-75% longer than a kprobe hit. | |
711 | When you have a return probe set on a function, adding a kprobe at | |
712 | the entry to that function adds essentially no overhead. | |
713 | ||
a1dac767 | 714 | Here are sample overhead figures (in usec) for different architectures:: |
d27a4ddd | 715 | |
a1dac767 MCC |
716 | k = kprobe; j = jprobe; r = return probe; kr = kprobe + return probe |
717 | on same function; jr = jprobe + return probe on same function:: | |
d27a4ddd | 718 | |
a1dac767 MCC |
719 | i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips |
720 | k = 0.57 usec; j = 1.00; r = 0.92; kr = 0.99; jr = 1.40 | |
d27a4ddd | 721 | |
a1dac767 MCC |
722 | x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips |
723 | k = 0.49 usec; j = 0.76; r = 0.80; kr = 0.82; jr = 1.07 | |
d27a4ddd | 724 | |
a1dac767 MCC |
725 | ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU) |
726 | k = 0.77 usec; j = 1.31; r = 1.26; kr = 1.45; jr = 1.99 | |
727 | ||
728 | Optimized Probe Overhead | |
729 | ------------------------ | |
b26486bf MH |
730 | |
731 | Typically, an optimized kprobe hit takes 0.07 to 0.1 microseconds to | |
a1dac767 MCC |
732 | process. Here are sample overhead figures (in usec) for x86 architectures:: |
733 | ||
734 | k = unoptimized kprobe, b = boosted (single-step skipped), o = optimized kprobe, | |
735 | r = unoptimized kretprobe, rb = boosted kretprobe, ro = optimized kretprobe. | |
b26486bf | 736 | |
a1dac767 MCC |
737 | i386: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips |
738 | k = 0.80 usec; b = 0.33; o = 0.05; r = 1.10; rb = 0.61; ro = 0.33 | |
b26486bf | 739 | |
a1dac767 MCC |
740 | x86-64: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips |
741 | k = 0.99 usec; b = 0.43; o = 0.06; r = 1.24; rb = 0.68; ro = 0.30 | |
b26486bf | 742 | |
a1dac767 MCC |
743 | TODO |
744 | ==== | |
d27a4ddd | 745 | |
8861da31 | 746 | a. SystemTap (http://sourceware.org/systemtap): Provides a simplified |
a1dac767 | 747 | programming interface for probe-based instrumentation. Try it out. |
8861da31 JK |
748 | b. Kernel return probes for sparc64. |
749 | c. Support for other architectures. | |
750 | d. User-space probes. | |
751 | e. Watchpoint probes (which fire on data references). | |
d27a4ddd | 752 | |
a1dac767 MCC |
753 | Kprobes Example |
754 | =============== | |
d27a4ddd | 755 | |
804defea | 756 | See samples/kprobes/kprobe_example.c |
d27a4ddd | 757 | |
a1dac767 MCC |
758 | Jprobes Example |
759 | =============== | |
d27a4ddd | 760 | |
804defea | 761 | See samples/kprobes/jprobe_example.c |
d27a4ddd | 762 | |
a1dac767 MCC |
763 | Kretprobes Example |
764 | ================== | |
d27a4ddd | 765 | |
804defea | 766 | See samples/kprobes/kretprobe_example.c |
d27a4ddd JK |
767 | |
768 | For additional information on Kprobes, refer to the following URLs: | |
bf8f6e5b | 769 | |
a1dac767 MCC |
770 | - http://www-106.ibm.com/developerworks/library/l-kprobes.html?ca=dgr-lnxw42Kprobe |
771 | - http://www.redhat.com/magazine/005mar05/features/kprobes/ | |
772 | - http://www-users.cs.umn.edu/~boutcher/kprobes/ | |
773 | - http://www.linuxsymposium.org/2006/linuxsymposium_procv2.pdf (pages 101-115) | |
774 | ||
775 | ||
776 | The kprobes debugfs interface | |
777 | ============================= | |
bf8f6e5b | 778 | |
bf8f6e5b AM |
779 | |
780 | With recent kernels (> 2.6.20) the list of registered kprobes is visible | |
156f5a78 | 781 | under the /sys/kernel/debug/kprobes/ directory (assuming debugfs is mounted at //sys/kernel/debug). |
bf8f6e5b | 782 | |
a1dac767 | 783 | /sys/kernel/debug/kprobes/list: Lists all registered probes on the system:: |
bf8f6e5b | 784 | |
a1dac767 MCC |
785 | c015d71a k vfs_read+0x0 |
786 | c011a316 j do_fork+0x0 | |
787 | c03dedc5 r tcp_v4_rcv+0x0 | |
bf8f6e5b AM |
788 | |
789 | The first column provides the kernel address where the probe is inserted. | |
790 | The second column identifies the type of probe (k - kprobe, r - kretprobe | |
791 | and j - jprobe), while the third column specifies the symbol+offset of | |
792 | the probe. If the probed function belongs to a module, the module name | |
e8386a0c MH |
793 | is also specified. Following columns show probe status. If the probe is on |
794 | a virtual address that is no longer valid (module init sections, module | |
795 | virtual addresses that correspond to modules that've been unloaded), | |
de5bd88d | 796 | such probes are marked with [GONE]. If the probe is temporarily disabled, |
b26486bf | 797 | such probes are marked with [DISABLED]. If the probe is optimized, it is |
9ed330d3 WL |
798 | marked with [OPTIMIZED]. If the probe is ftrace-based, it is marked with |
799 | [FTRACE]. | |
bf8f6e5b | 800 | |
156f5a78 | 801 | /sys/kernel/debug/kprobes/enabled: Turn kprobes ON/OFF forcibly. |
bf8f6e5b | 802 | |
de5bd88d MH |
803 | Provides a knob to globally and forcibly turn registered kprobes ON or OFF. |
804 | By default, all kprobes are enabled. By echoing "0" to this file, all | |
805 | registered probes will be disarmed, till such time a "1" is echoed to this | |
806 | file. Note that this knob just disarms and arms all kprobes and doesn't | |
807 | change each probe's disabling state. This means that disabled kprobes (marked | |
808 | [DISABLED]) will be not enabled if you turn ON all kprobes by this knob. | |
b26486bf MH |
809 | |
810 | ||
a1dac767 MCC |
811 | The kprobes sysctl interface |
812 | ============================ | |
b26486bf MH |
813 | |
814 | /proc/sys/debug/kprobes-optimization: Turn kprobes optimization ON/OFF. | |
815 | ||
816 | When CONFIG_OPTPROBES=y, this sysctl interface appears and it provides | |
817 | a knob to globally and forcibly turn jump optimization (see section | |
a1dac767 MCC |
818 | :ref:`kprobes_jump_optimization`) ON or OFF. By default, jump optimization |
819 | is allowed (ON). If you echo "0" to this file or set | |
820 | "debug.kprobes_optimization" to 0 via sysctl, all optimized probes will be | |
821 | unoptimized, and any new probes registered after that will not be optimized. | |
43e5f7e1 MCC |
822 | |
823 | Note that this knob *changes* the optimized state. This means that optimized | |
a1dac767 | 824 | probes (marked [OPTIMIZED]) will be unoptimized ([OPTIMIZED] tag will be |
b26486bf MH |
825 | removed). If the knob is turned on, they will be optimized again. |
826 |