1 Title : Kernel Probes (Kprobes)
2 Authors : Jim Keniston <jkenisto@us.ibm.com>
3 : Prasanna S Panchamukhi <prasanna@in.ibm.com>
7 1. Concepts: Kprobes, Jprobes, Return Probes
8 2. Architectures Supported
11 5. Kprobes Features and Limitations
16 10. Kretprobes Example
17 Appendix A: The kprobes debugfs interface
19 1. Concepts: Kprobes, Jprobes, Return Probes
21 Kprobes enables you to dynamically break into any kernel routine and
22 collect debugging and performance information non-disruptively. You
23 can trap at almost any kernel code address, specifying a handler
24 routine to be invoked when the breakpoint is hit.
26 There are currently three types of probes: kprobes, jprobes, and
27 kretprobes (also called return probes). A kprobe can be inserted
28 on virtually any instruction in the kernel. A jprobe is inserted at
29 the entry to a kernel function, and provides convenient access to the
30 function's arguments. A return probe fires when a specified function
33 In the typical case, Kprobes-based instrumentation is packaged as
34 a kernel module. The module's init function installs ("registers")
35 one or more probes, and the exit function unregisters them. A
36 registration function such as register_kprobe() specifies where
37 the probe is to be inserted and what handler is to be called when
40 There are also register_/unregister_*probes() functions for batch
41 registration/unregistration of a group of *probes. These functions
42 can speed up unregistration process when you have to unregister
43 a lot of probes at once.
45 The next three subsections explain how the different types of
46 probes work. They explain certain things that you'll need to
47 know in order to make the best use of Kprobes -- e.g., the
48 difference between a pre_handler and a post_handler, and how
49 to use the maxactive and nmissed fields of a kretprobe. But
50 if you're in a hurry to start using Kprobes, you can skip ahead
53 1.1 How Does a Kprobe Work?
55 When a kprobe is registered, Kprobes makes a copy of the probed
56 instruction and replaces the first byte(s) of the probed instruction
57 with a breakpoint instruction (e.g., int3 on i386 and x86_64).
59 When a CPU hits the breakpoint instruction, a trap occurs, the CPU's
60 registers are saved, and control passes to Kprobes via the
61 notifier_call_chain mechanism. Kprobes executes the "pre_handler"
62 associated with the kprobe, passing the handler the addresses of the
63 kprobe struct and the saved registers.
65 Next, Kprobes single-steps its copy of the probed instruction.
66 (It would be simpler to single-step the actual instruction in place,
67 but then Kprobes would have to temporarily remove the breakpoint
68 instruction. This would open a small time window when another CPU
69 could sail right past the probepoint.)
71 After the instruction is single-stepped, Kprobes executes the
72 "post_handler," if any, that is associated with the kprobe.
73 Execution then continues with the instruction following the probepoint.
75 1.2 How Does a Jprobe Work?
77 A jprobe is implemented using a kprobe that is placed on a function's
78 entry point. It employs a simple mirroring principle to allow
79 seamless access to the probed function's arguments. The jprobe
80 handler routine should have the same signature (arg list and return
81 type) as the function being probed, and must always end by calling
82 the Kprobes function jprobe_return().
84 Here's how it works. When the probe is hit, Kprobes makes a copy of
85 the saved registers and a generous portion of the stack (see below).
86 Kprobes then points the saved instruction pointer at the jprobe's
87 handler routine, and returns from the trap. As a result, control
88 passes to the handler, which is presented with the same register and
89 stack contents as the probed function. When it is done, the handler
90 calls jprobe_return(), which traps again to restore the original stack
91 contents and processor state and switch to the probed function.
93 By convention, the callee owns its arguments, so gcc may produce code
94 that unexpectedly modifies that portion of the stack. This is why
95 Kprobes saves a copy of the stack and restores it after the jprobe
96 handler has run. Up to MAX_STACK_SIZE bytes are copied -- e.g.,
99 Note that the probed function's args may be passed on the stack
100 or in registers. The jprobe will work in either case, so long as the
101 handler's prototype matches that of the probed function.
105 1.3.1 How Does a Return Probe Work?
107 When you call register_kretprobe(), Kprobes establishes a kprobe at
108 the entry to the function. When the probed function is called and this
109 probe is hit, Kprobes saves a copy of the return address, and replaces
110 the return address with the address of a "trampoline." The trampoline
111 is an arbitrary piece of code -- typically just a nop instruction.
112 At boot time, Kprobes registers a kprobe at the trampoline.
114 When the probed function executes its return instruction, control
115 passes to the trampoline and that probe is hit. Kprobes' trampoline
116 handler calls the user-specified return handler associated with the
117 kretprobe, then sets the saved instruction pointer to the saved return
118 address, and that's where execution resumes upon return from the trap.
120 While the probed function is executing, its return address is
121 stored in an object of type kretprobe_instance. Before calling
122 register_kretprobe(), the user sets the maxactive field of the
123 kretprobe struct to specify how many instances of the specified
124 function can be probed simultaneously. register_kretprobe()
125 pre-allocates the indicated number of kretprobe_instance objects.
127 For example, if the function is non-recursive and is called with a
128 spinlock held, maxactive = 1 should be enough. If the function is
129 non-recursive and can never relinquish the CPU (e.g., via a semaphore
130 or preemption), NR_CPUS should be enough. If maxactive <= 0, it is
131 set to a default value. If CONFIG_PREEMPT is enabled, the default
132 is max(10, 2*NR_CPUS). Otherwise, the default is NR_CPUS.
134 It's not a disaster if you set maxactive too low; you'll just miss
135 some probes. In the kretprobe struct, the nmissed field is set to
136 zero when the return probe is registered, and is incremented every
137 time the probed function is entered but there is no kretprobe_instance
138 object available for establishing the return probe.
140 1.3.2 Kretprobe entry-handler
142 Kretprobes also provides an optional user-specified handler which runs
143 on function entry. This handler is specified by setting the entry_handler
144 field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the
145 function entry is hit, the user-defined entry_handler, if any, is invoked.
146 If the entry_handler returns 0 (success) then a corresponding return handler
147 is guaranteed to be called upon function return. If the entry_handler
148 returns a non-zero error then Kprobes leaves the return address as is, and
149 the kretprobe has no further effect for that particular function instance.
151 Multiple entry and return handler invocations are matched using the unique
152 kretprobe_instance object associated with them. Additionally, a user
153 may also specify per return-instance private data to be part of each
154 kretprobe_instance object. This is especially useful when sharing private
155 data between corresponding user entry and return handlers. The size of each
156 private data object can be specified at kretprobe registration time by
157 setting the data_size field of the kretprobe struct. This data can be
158 accessed through the data field of each kretprobe_instance object.
160 In case probed function is entered but there is no kretprobe_instance
161 object available, then in addition to incrementing the nmissed count,
162 the user entry_handler invocation is also skipped.
164 2. Architectures Supported
166 Kprobes, jprobes, and return probes are implemented on the following
170 - x86_64 (AMD-64, EM64T)
172 - ia64 (Does not support probes on instruction slot1.)
173 - sparc64 (Return probes not yet implemented.)
177 3. Configuring Kprobes
179 When configuring the kernel using make menuconfig/xconfig/oldconfig,
180 ensure that CONFIG_KPROBES is set to "y". Under "Instrumentation
181 Support", look for "Kprobes".
183 So that you can load and unload Kprobes-based instrumentation modules,
184 make sure "Loadable module support" (CONFIG_MODULES) and "Module
185 unloading" (CONFIG_MODULE_UNLOAD) are set to "y".
187 Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL
188 are set to "y", since kallsyms_lookup_name() is used by the in-kernel
189 kprobe address resolution code.
191 If you need to insert a probe in the middle of a function, you may find
192 it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO),
193 so you can use "objdump -d -l vmlinux" to see the source-to-object
198 The Kprobes API includes a "register" function and an "unregister"
199 function for each type of probe. The API also includes "register_*probes"
200 and "unregister_*probes" functions for (un)registering arrays of probes.
201 Here are terse, mini-man-page specifications for these functions and
202 the associated probe handlers that you'll write. See the files in the
203 samples/kprobes/ sub-directory for examples.
207 #include <linux/kprobes.h>
208 int register_kprobe(struct kprobe *kp);
210 Sets a breakpoint at the address kp->addr. When the breakpoint is
211 hit, Kprobes calls kp->pre_handler. After the probed instruction
212 is single-stepped, Kprobe calls kp->post_handler. If a fault
213 occurs during execution of kp->pre_handler or kp->post_handler,
214 or during single-stepping of the probed instruction, Kprobes calls
215 kp->fault_handler. Any or all handlers can be NULL. If kp->flags
216 is set KPROBE_FLAG_DISABLED, that kp will be registered but disabled,
217 so, it's handlers aren't hit until calling enable_kprobe(kp).
220 1. With the introduction of the "symbol_name" field to struct kprobe,
221 the probepoint address resolution will now be taken care of by the kernel.
222 The following will now work:
224 kp.symbol_name = "symbol_name";
226 (64-bit powerpc intricacies such as function descriptors are handled
229 2. Use the "offset" field of struct kprobe if the offset into the symbol
230 to install a probepoint is known. This field is used to calculate the
233 3. Specify either the kprobe "symbol_name" OR the "addr". If both are
234 specified, kprobe registration will fail with -EINVAL.
236 4. With CISC architectures (such as i386 and x86_64), the kprobes code
237 does not validate if the kprobe.addr is at an instruction boundary.
238 Use "offset" with caution.
240 register_kprobe() returns 0 on success, or a negative errno otherwise.
242 User's pre-handler (kp->pre_handler):
243 #include <linux/kprobes.h>
244 #include <linux/ptrace.h>
245 int pre_handler(struct kprobe *p, struct pt_regs *regs);
247 Called with p pointing to the kprobe associated with the breakpoint,
248 and regs pointing to the struct containing the registers saved when
249 the breakpoint was hit. Return 0 here unless you're a Kprobes geek.
251 User's post-handler (kp->post_handler):
252 #include <linux/kprobes.h>
253 #include <linux/ptrace.h>
254 void post_handler(struct kprobe *p, struct pt_regs *regs,
255 unsigned long flags);
257 p and regs are as described for the pre_handler. flags always seems
260 User's fault-handler (kp->fault_handler):
261 #include <linux/kprobes.h>
262 #include <linux/ptrace.h>
263 int fault_handler(struct kprobe *p, struct pt_regs *regs, int trapnr);
265 p and regs are as described for the pre_handler. trapnr is the
266 architecture-specific trap number associated with the fault (e.g.,
267 on i386, 13 for a general protection fault or 14 for a page fault).
268 Returns 1 if it successfully handled the exception.
272 #include <linux/kprobes.h>
273 int register_jprobe(struct jprobe *jp)
275 Sets a breakpoint at the address jp->kp.addr, which must be the address
276 of the first instruction of a function. When the breakpoint is hit,
277 Kprobes runs the handler whose address is jp->entry.
279 The handler should have the same arg list and return type as the probed
280 function; and just before it returns, it must call jprobe_return().
281 (The handler never actually returns, since jprobe_return() returns
282 control to Kprobes.) If the probed function is declared asmlinkage
283 or anything else that affects how args are passed, the handler's
284 declaration must match.
286 register_jprobe() returns 0 on success, or a negative errno otherwise.
288 4.3 register_kretprobe
290 #include <linux/kprobes.h>
291 int register_kretprobe(struct kretprobe *rp);
293 Establishes a return probe for the function whose address is
294 rp->kp.addr. When that function returns, Kprobes calls rp->handler.
295 You must set rp->maxactive appropriately before you call
296 register_kretprobe(); see "How Does a Return Probe Work?" for details.
298 register_kretprobe() returns 0 on success, or a negative errno
301 User's return-probe handler (rp->handler):
302 #include <linux/kprobes.h>
303 #include <linux/ptrace.h>
304 int kretprobe_handler(struct kretprobe_instance *ri, struct pt_regs *regs);
306 regs is as described for kprobe.pre_handler. ri points to the
307 kretprobe_instance object, of which the following fields may be
309 - ret_addr: the return address
310 - rp: points to the corresponding kretprobe object
311 - task: points to the corresponding task struct
312 - data: points to per return-instance private data; see "Kretprobe
313 entry-handler" for details.
315 The regs_return_value(regs) macro provides a simple abstraction to
316 extract the return value from the appropriate register as defined by
317 the architecture's ABI.
319 The handler's return value is currently ignored.
321 4.4 unregister_*probe
323 #include <linux/kprobes.h>
324 void unregister_kprobe(struct kprobe *kp);
325 void unregister_jprobe(struct jprobe *jp);
326 void unregister_kretprobe(struct kretprobe *rp);
328 Removes the specified probe. The unregister function can be called
329 at any time after the probe has been registered.
332 If the functions find an incorrect probe (ex. an unregistered probe),
333 they clear the addr field of the probe.
337 #include <linux/kprobes.h>
338 int register_kprobes(struct kprobe **kps, int num);
339 int register_kretprobes(struct kretprobe **rps, int num);
340 int register_jprobes(struct jprobe **jps, int num);
342 Registers each of the num probes in the specified array. If any
343 error occurs during registration, all probes in the array, up to
344 the bad probe, are safely unregistered before the register_*probes
346 - kps/rps/jps: an array of pointers to *probe data structures
347 - num: the number of the array entries.
350 You have to allocate(or define) an array of pointers and set all
351 of the array entries before using these functions.
353 4.6 unregister_*probes
355 #include <linux/kprobes.h>
356 void unregister_kprobes(struct kprobe **kps, int num);
357 void unregister_kretprobes(struct kretprobe **rps, int num);
358 void unregister_jprobes(struct jprobe **jps, int num);
360 Removes each of the num probes in the specified array at once.
363 If the functions find some incorrect probes (ex. unregistered
364 probes) in the specified array, they clear the addr field of those
365 incorrect probes. However, other probes in the array are
366 unregistered correctly.
370 #include <linux/kprobes.h>
371 int disable_kprobe(struct kprobe *kp);
373 Temporarily disables the specified kprobe. You can enable it again by using
374 enable_kprobe(). You must specify the kprobe which has been registered.
378 #include <linux/kprobes.h>
379 int enable_kprobe(struct kprobe *kp);
381 Enables kprobe which has been disabled by disable_kprobe(). You must specify
382 the kprobe which has been registered.
384 5. Kprobes Features and Limitations
386 Kprobes allows multiple probes at the same address. Currently,
387 however, there cannot be multiple jprobes on the same function at
390 In general, you can install a probe anywhere in the kernel.
391 In particular, you can probe interrupt handlers. Known exceptions
392 are discussed in this section.
394 The register_*probe functions will return -EINVAL if you attempt
395 to install a probe in the code that implements Kprobes (mostly
396 kernel/kprobes.c and arch/*/kernel/kprobes.c, but also functions such
397 as do_page_fault and notifier_call_chain).
399 If you install a probe in an inline-able function, Kprobes makes
400 no attempt to chase down all inline instances of the function and
401 install probes there. gcc may inline a function without being asked,
402 so keep this in mind if you're not seeing the probe hits you expect.
404 A probe handler can modify the environment of the probed function
405 -- e.g., by modifying kernel data structures, or by modifying the
406 contents of the pt_regs struct (which are restored to the registers
407 upon return from the breakpoint). So Kprobes can be used, for example,
408 to install a bug fix or to inject faults for testing. Kprobes, of
409 course, has no way to distinguish the deliberately injected faults
410 from the accidental ones. Don't drink and probe.
412 Kprobes makes no attempt to prevent probe handlers from stepping on
413 each other -- e.g., probing printk() and then calling printk() from a
414 probe handler. If a probe handler hits a probe, that second probe's
415 handlers won't be run in that instance, and the kprobe.nmissed member
416 of the second probe will be incremented.
418 As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of
419 the same handler) may run concurrently on different CPUs.
421 Kprobes does not use mutexes or allocate memory except during
422 registration and unregistration.
424 Probe handlers are run with preemption disabled. Depending on the
425 architecture, handlers may also run with interrupts disabled. In any
426 case, your handler should not yield the CPU (e.g., by attempting to
427 acquire a semaphore).
429 Since a return probe is implemented by replacing the return
430 address with the trampoline's address, stack backtraces and calls
431 to __builtin_return_address() will typically yield the trampoline's
432 address instead of the real return address for kretprobed functions.
433 (As far as we can tell, __builtin_return_address() is used only
434 for instrumentation and error reporting.)
436 If the number of times a function is called does not match the number
437 of times it returns, registering a return probe on that function may
438 produce undesirable results. In such a case, a line:
439 kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c
440 gets printed. With this information, one will be able to correlate the
441 exact instance of the kretprobe that caused the problem. We have the
442 do_exit() case covered. do_execve() and do_fork() are not an issue.
443 We're unaware of other specific cases where this could be a problem.
445 If, upon entry to or exit from a function, the CPU is running on
446 a stack other than that of the current task, registering a return
447 probe on that function may produce undesirable results. For this
448 reason, Kprobes doesn't support return probes (or kprobes or jprobes)
449 on the x86_64 version of __switch_to(); the registration functions
454 On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0
455 microseconds to process. Specifically, a benchmark that hits the same
456 probepoint repeatedly, firing a simple handler each time, reports 1-2
457 million hits per second, depending on the architecture. A jprobe or
458 return-probe hit typically takes 50-75% longer than a kprobe hit.
459 When you have a return probe set on a function, adding a kprobe at
460 the entry to that function adds essentially no overhead.
462 Here are sample overhead figures (in usec) for different architectures.
463 k = kprobe; j = jprobe; r = return probe; kr = kprobe + return probe
464 on same function; jr = jprobe + return probe on same function
466 i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips
467 k = 0.57 usec; j = 1.00; r = 0.92; kr = 0.99; jr = 1.40
469 x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips
470 k = 0.49 usec; j = 0.76; r = 0.80; kr = 0.82; jr = 1.07
472 ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU)
473 k = 0.77 usec; j = 1.31; r = 1.26; kr = 1.45; jr = 1.99
477 a. SystemTap (http://sourceware.org/systemtap): Provides a simplified
478 programming interface for probe-based instrumentation. Try it out.
479 b. Kernel return probes for sparc64.
480 c. Support for other architectures.
481 d. User-space probes.
482 e. Watchpoint probes (which fire on data references).
486 See samples/kprobes/kprobe_example.c
490 See samples/kprobes/jprobe_example.c
492 10. Kretprobes Example
494 See samples/kprobes/kretprobe_example.c
496 For additional information on Kprobes, refer to the following URLs:
497 http://www-106.ibm.com/developerworks/library/l-kprobes.html?ca=dgr-lnxw42Kprobe
498 http://www.redhat.com/magazine/005mar05/features/kprobes/
499 http://www-users.cs.umn.edu/~boutcher/kprobes/
500 http://www.linuxsymposium.org/2006/linuxsymposium_procv2.pdf (pages 101-115)
503 Appendix A: The kprobes debugfs interface
505 With recent kernels (> 2.6.20) the list of registered kprobes is visible
506 under the /debug/kprobes/ directory (assuming debugfs is mounted at /debug).
508 /debug/kprobes/list: Lists all registered probes on the system
510 c015d71a k vfs_read+0x0
511 c011a316 j do_fork+0x0
512 c03dedc5 r tcp_v4_rcv+0x0
514 The first column provides the kernel address where the probe is inserted.
515 The second column identifies the type of probe (k - kprobe, r - kretprobe
516 and j - jprobe), while the third column specifies the symbol+offset of
517 the probe. If the probed function belongs to a module, the module name
518 is also specified. Following columns show probe status. If the probe is on
519 a virtual address that is no longer valid (module init sections, module
520 virtual addresses that correspond to modules that've been unloaded),
521 such probes are marked with [GONE]. If the probe is temporarily disabled,
522 such probes are marked with [DISABLED].
524 /debug/kprobes/enabled: Turn kprobes ON/OFF forcibly.
526 Provides a knob to globally and forcibly turn registered kprobes ON or OFF.
527 By default, all kprobes are enabled. By echoing "0" to this file, all
528 registered probes will be disarmed, till such time a "1" is echoed to this
529 file. Note that this knob just disarms and arms all kprobes and doesn't
530 change each probe's disabling state. This means that disabled kprobes (marked
531 [DISABLED]) will be not enabled if you turn ON all kprobes by this knob.