Back to home page

LXR

 
 

    


0001 Title   : Kernel Probes (Kprobes)
0002 Authors : Jim Keniston <jkenisto@us.ibm.com>
0003         : Prasanna S Panchamukhi <prasanna.panchamukhi@gmail.com>
0004         : Masami Hiramatsu <mhiramat@redhat.com>
0005 
0006 CONTENTS
0007 
0008 1. Concepts: Kprobes, Jprobes, Return Probes
0009 2. Architectures Supported
0010 3. Configuring Kprobes
0011 4. API Reference
0012 5. Kprobes Features and Limitations
0013 6. Probe Overhead
0014 7. TODO
0015 8. Kprobes Example
0016 9. Jprobes Example
0017 10. Kretprobes Example
0018 Appendix A: The kprobes debugfs interface
0019 Appendix B: The kprobes sysctl interface
0020 
0021 1. Concepts: Kprobes, Jprobes, Return Probes
0022 
0023 Kprobes enables you to dynamically break into any kernel routine and
0024 collect debugging and performance information non-disruptively. You
0025 can trap at almost any kernel code address(*), specifying a handler
0026 routine to be invoked when the breakpoint is hit.
0027 (*: some parts of the kernel code can not be trapped, see 1.5 Blacklist)
0028 
0029 There are currently three types of probes: kprobes, jprobes, and
0030 kretprobes (also called return probes).  A kprobe can be inserted
0031 on virtually any instruction in the kernel.  A jprobe is inserted at
0032 the entry to a kernel function, and provides convenient access to the
0033 function's arguments.  A return probe fires when a specified function
0034 returns.
0035 
0036 In the typical case, Kprobes-based instrumentation is packaged as
0037 a kernel module.  The module's init function installs ("registers")
0038 one or more probes, and the exit function unregisters them.  A
0039 registration function such as register_kprobe() specifies where
0040 the probe is to be inserted and what handler is to be called when
0041 the probe is hit.
0042 
0043 There are also register_/unregister_*probes() functions for batch
0044 registration/unregistration of a group of *probes. These functions
0045 can speed up unregistration process when you have to unregister
0046 a lot of probes at once.
0047 
0048 The next four subsections explain how the different types of
0049 probes work and how jump optimization works.  They explain certain
0050 things that you'll need to know in order to make the best use of
0051 Kprobes -- e.g., the difference between a pre_handler and
0052 a post_handler, and how to use the maxactive and nmissed fields of
0053 a kretprobe.  But if you're in a hurry to start using Kprobes, you
0054 can skip ahead to section 2.
0055 
0056 1.1 How Does a Kprobe Work?
0057 
0058 When a kprobe is registered, Kprobes makes a copy of the probed
0059 instruction and replaces the first byte(s) of the probed instruction
0060 with a breakpoint instruction (e.g., int3 on i386 and x86_64).
0061 
0062 When a CPU hits the breakpoint instruction, a trap occurs, the CPU's
0063 registers are saved, and control passes to Kprobes via the
0064 notifier_call_chain mechanism.  Kprobes executes the "pre_handler"
0065 associated with the kprobe, passing the handler the addresses of the
0066 kprobe struct and the saved registers.
0067 
0068 Next, Kprobes single-steps its copy of the probed instruction.
0069 (It would be simpler to single-step the actual instruction in place,
0070 but then Kprobes would have to temporarily remove the breakpoint
0071 instruction.  This would open a small time window when another CPU
0072 could sail right past the probepoint.)
0073 
0074 After the instruction is single-stepped, Kprobes executes the
0075 "post_handler," if any, that is associated with the kprobe.
0076 Execution then continues with the instruction following the probepoint.
0077 
0078 1.2 How Does a Jprobe Work?
0079 
0080 A jprobe is implemented using a kprobe that is placed on a function's
0081 entry point.  It employs a simple mirroring principle to allow
0082 seamless access to the probed function's arguments.  The jprobe
0083 handler routine should have the same signature (arg list and return
0084 type) as the function being probed, and must always end by calling
0085 the Kprobes function jprobe_return().
0086 
0087 Here's how it works.  When the probe is hit, Kprobes makes a copy of
0088 the saved registers and a generous portion of the stack (see below).
0089 Kprobes then points the saved instruction pointer at the jprobe's
0090 handler routine, and returns from the trap.  As a result, control
0091 passes to the handler, which is presented with the same register and
0092 stack contents as the probed function.  When it is done, the handler
0093 calls jprobe_return(), which traps again to restore the original stack
0094 contents and processor state and switch to the probed function.
0095 
0096 By convention, the callee owns its arguments, so gcc may produce code
0097 that unexpectedly modifies that portion of the stack.  This is why
0098 Kprobes saves a copy of the stack and restores it after the jprobe
0099 handler has run.  Up to MAX_STACK_SIZE bytes are copied -- e.g.,
0100 64 bytes on i386.
0101 
0102 Note that the probed function's args may be passed on the stack
0103 or in registers.  The jprobe will work in either case, so long as the
0104 handler's prototype matches that of the probed function.
0105 
0106 Note that in some architectures (e.g.: arm64 and sparc64) the stack
0107 copy is not done, as the actual location of stacked parameters may be
0108 outside of a reasonable MAX_STACK_SIZE value and because that location
0109 cannot be determined by the jprobes code. In this case the jprobes
0110 user must be careful to make certain the calling signature of the
0111 function does not cause parameters to be passed on the stack (e.g.:
0112 more than eight function arguments, an argument of more than sixteen
0113 bytes, or more than 64 bytes of argument data, depending on
0114 architecture).
0115 
0116 1.3 Return Probes
0117 
0118 1.3.1 How Does a Return Probe Work?
0119 
0120 When you call register_kretprobe(), Kprobes establishes a kprobe at
0121 the entry to the function.  When the probed function is called and this
0122 probe is hit, Kprobes saves a copy of the return address, and replaces
0123 the return address with the address of a "trampoline."  The trampoline
0124 is an arbitrary piece of code -- typically just a nop instruction.
0125 At boot time, Kprobes registers a kprobe at the trampoline.
0126 
0127 When the probed function executes its return instruction, control
0128 passes to the trampoline and that probe is hit.  Kprobes' trampoline
0129 handler calls the user-specified return handler associated with the
0130 kretprobe, then sets the saved instruction pointer to the saved return
0131 address, and that's where execution resumes upon return from the trap.
0132 
0133 While the probed function is executing, its return address is
0134 stored in an object of type kretprobe_instance.  Before calling
0135 register_kretprobe(), the user sets the maxactive field of the
0136 kretprobe struct to specify how many instances of the specified
0137 function can be probed simultaneously.  register_kretprobe()
0138 pre-allocates the indicated number of kretprobe_instance objects.
0139 
0140 For example, if the function is non-recursive and is called with a
0141 spinlock held, maxactive = 1 should be enough.  If the function is
0142 non-recursive and can never relinquish the CPU (e.g., via a semaphore
0143 or preemption), NR_CPUS should be enough.  If maxactive <= 0, it is
0144 set to a default value.  If CONFIG_PREEMPT is enabled, the default
0145 is max(10, 2*NR_CPUS).  Otherwise, the default is NR_CPUS.
0146 
0147 It's not a disaster if you set maxactive too low; you'll just miss
0148 some probes.  In the kretprobe struct, the nmissed field is set to
0149 zero when the return probe is registered, and is incremented every
0150 time the probed function is entered but there is no kretprobe_instance
0151 object available for establishing the return probe.
0152 
0153 1.3.2 Kretprobe entry-handler
0154 
0155 Kretprobes also provides an optional user-specified handler which runs
0156 on function entry. This handler is specified by setting the entry_handler
0157 field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the
0158 function entry is hit, the user-defined entry_handler, if any, is invoked.
0159 If the entry_handler returns 0 (success) then a corresponding return handler
0160 is guaranteed to be called upon function return. If the entry_handler
0161 returns a non-zero error then Kprobes leaves the return address as is, and
0162 the kretprobe has no further effect for that particular function instance.
0163 
0164 Multiple entry and return handler invocations are matched using the unique
0165 kretprobe_instance object associated with them. Additionally, a user
0166 may also specify per return-instance private data to be part of each
0167 kretprobe_instance object. This is especially useful when sharing private
0168 data between corresponding user entry and return handlers. The size of each
0169 private data object can be specified at kretprobe registration time by
0170 setting the data_size field of the kretprobe struct. This data can be
0171 accessed through the data field of each kretprobe_instance object.
0172 
0173 In case probed function is entered but there is no kretprobe_instance
0174 object available, then in addition to incrementing the nmissed count,
0175 the user entry_handler invocation is also skipped.
0176 
0177 1.4 How Does Jump Optimization Work?
0178 
0179 If your kernel is built with CONFIG_OPTPROBES=y (currently this flag
0180 is automatically set 'y' on x86/x86-64, non-preemptive kernel) and
0181 the "debug.kprobes_optimization" kernel parameter is set to 1 (see
0182 sysctl(8)), Kprobes tries to reduce probe-hit overhead by using a jump
0183 instruction instead of a breakpoint instruction at each probepoint.
0184 
0185 1.4.1 Init a Kprobe
0186 
0187 When a probe is registered, before attempting this optimization,
0188 Kprobes inserts an ordinary, breakpoint-based kprobe at the specified
0189 address. So, even if it's not possible to optimize this particular
0190 probepoint, there'll be a probe there.
0191 
0192 1.4.2 Safety Check
0193 
0194 Before optimizing a probe, Kprobes performs the following safety checks:
0195 
0196 - Kprobes verifies that the region that will be replaced by the jump
0197 instruction (the "optimized region") lies entirely within one function.
0198 (A jump instruction is multiple bytes, and so may overlay multiple
0199 instructions.)
0200 
0201 - Kprobes analyzes the entire function and verifies that there is no
0202 jump into the optimized region.  Specifically:
0203   - the function contains no indirect jump;
0204   - the function contains no instruction that causes an exception (since
0205   the fixup code triggered by the exception could jump back into the
0206   optimized region -- Kprobes checks the exception tables to verify this);
0207   and
0208   - there is no near jump to the optimized region (other than to the first
0209   byte).
0210 
0211 - For each instruction in the optimized region, Kprobes verifies that
0212 the instruction can be executed out of line.
0213 
0214 1.4.3 Preparing Detour Buffer
0215 
0216 Next, Kprobes prepares a "detour" buffer, which contains the following
0217 instruction sequence:
0218 - code to push the CPU's registers (emulating a breakpoint trap)
0219 - a call to the trampoline code which calls user's probe handlers.
0220 - code to restore registers
0221 - the instructions from the optimized region
0222 - a jump back to the original execution path.
0223 
0224 1.4.4 Pre-optimization
0225 
0226 After preparing the detour buffer, Kprobes verifies that none of the
0227 following situations exist:
0228 - The probe has either a break_handler (i.e., it's a jprobe) or a
0229 post_handler.
0230 - Other instructions in the optimized region are probed.
0231 - The probe is disabled.
0232 In any of the above cases, Kprobes won't start optimizing the probe.
0233 Since these are temporary situations, Kprobes tries to start
0234 optimizing it again if the situation is changed.
0235 
0236 If the kprobe can be optimized, Kprobes enqueues the kprobe to an
0237 optimizing list, and kicks the kprobe-optimizer workqueue to optimize
0238 it.  If the to-be-optimized probepoint is hit before being optimized,
0239 Kprobes returns control to the original instruction path by setting
0240 the CPU's instruction pointer to the copied code in the detour buffer
0241 -- thus at least avoiding the single-step.
0242 
0243 1.4.5 Optimization
0244 
0245 The Kprobe-optimizer doesn't insert the jump instruction immediately;
0246 rather, it calls synchronize_sched() for safety first, because it's
0247 possible for a CPU to be interrupted in the middle of executing the
0248 optimized region(*).  As you know, synchronize_sched() can ensure
0249 that all interruptions that were active when synchronize_sched()
0250 was called are done, but only if CONFIG_PREEMPT=n.  So, this version
0251 of kprobe optimization supports only kernels with CONFIG_PREEMPT=n.(**)
0252 
0253 After that, the Kprobe-optimizer calls stop_machine() to replace
0254 the optimized region with a jump instruction to the detour buffer,
0255 using text_poke_smp().
0256 
0257 1.4.6 Unoptimization
0258 
0259 When an optimized kprobe is unregistered, disabled, or blocked by
0260 another kprobe, it will be unoptimized.  If this happens before
0261 the optimization is complete, the kprobe is just dequeued from the
0262 optimized list.  If the optimization has been done, the jump is
0263 replaced with the original code (except for an int3 breakpoint in
0264 the first byte) by using text_poke_smp().
0265 
0266 (*)Please imagine that the 2nd instruction is interrupted and then
0267 the optimizer replaces the 2nd instruction with the jump *address*
0268 while the interrupt handler is running. When the interrupt
0269 returns to original address, there is no valid instruction,
0270 and it causes an unexpected result.
0271 
0272 (**)This optimization-safety checking may be replaced with the
0273 stop-machine method that ksplice uses for supporting a CONFIG_PREEMPT=y
0274 kernel.
0275 
0276 NOTE for geeks:
0277 The jump optimization changes the kprobe's pre_handler behavior.
0278 Without optimization, the pre_handler can change the kernel's execution
0279 path by changing regs->ip and returning 1.  However, when the probe
0280 is optimized, that modification is ignored.  Thus, if you want to
0281 tweak the kernel's execution path, you need to suppress optimization,
0282 using one of the following techniques:
0283 - Specify an empty function for the kprobe's post_handler or break_handler.
0284  or
0285 - Execute 'sysctl -w debug.kprobes_optimization=n'
0286 
0287 1.5 Blacklist
0288 
0289 Kprobes can probe most of the kernel except itself. This means
0290 that there are some functions where kprobes cannot probe. Probing
0291 (trapping) such functions can cause a recursive trap (e.g. double
0292 fault) or the nested probe handler may never be called.
0293 Kprobes manages such functions as a blacklist.
0294 If you want to add a function into the blacklist, you just need
0295 to (1) include linux/kprobes.h and (2) use NOKPROBE_SYMBOL() macro
0296 to specify a blacklisted function.
0297 Kprobes checks the given probe address against the blacklist and
0298 rejects registering it, if the given address is in the blacklist.
0299 
0300 2. Architectures Supported
0301 
0302 Kprobes, jprobes, and return probes are implemented on the following
0303 architectures:
0304 
0305 - i386 (Supports jump optimization)
0306 - x86_64 (AMD-64, EM64T) (Supports jump optimization)
0307 - ppc64
0308 - ia64 (Does not support probes on instruction slot1.)
0309 - sparc64 (Return probes not yet implemented.)
0310 - arm
0311 - ppc
0312 - mips
0313 - s390
0314 
0315 3. Configuring Kprobes
0316 
0317 When configuring the kernel using make menuconfig/xconfig/oldconfig,
0318 ensure that CONFIG_KPROBES is set to "y". Under "General setup", look
0319 for "Kprobes".
0320 
0321 So that you can load and unload Kprobes-based instrumentation modules,
0322 make sure "Loadable module support" (CONFIG_MODULES) and "Module
0323 unloading" (CONFIG_MODULE_UNLOAD) are set to "y".
0324 
0325 Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL
0326 are set to "y", since kallsyms_lookup_name() is used by the in-kernel
0327 kprobe address resolution code.
0328 
0329 If you need to insert a probe in the middle of a function, you may find
0330 it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO),
0331 so you can use "objdump -d -l vmlinux" to see the source-to-object
0332 code mapping.
0333 
0334 4. API Reference
0335 
0336 The Kprobes API includes a "register" function and an "unregister"
0337 function for each type of probe. The API also includes "register_*probes"
0338 and "unregister_*probes" functions for (un)registering arrays of probes.
0339 Here are terse, mini-man-page specifications for these functions and
0340 the associated probe handlers that you'll write. See the files in the
0341 samples/kprobes/ sub-directory for examples.
0342 
0343 4.1 register_kprobe
0344 
0345 #include <linux/kprobes.h>
0346 int register_kprobe(struct kprobe *kp);
0347 
0348 Sets a breakpoint at the address kp->addr.  When the breakpoint is
0349 hit, Kprobes calls kp->pre_handler.  After the probed instruction
0350 is single-stepped, Kprobe calls kp->post_handler.  If a fault
0351 occurs during execution of kp->pre_handler or kp->post_handler,
0352 or during single-stepping of the probed instruction, Kprobes calls
0353 kp->fault_handler.  Any or all handlers can be NULL. If kp->flags
0354 is set KPROBE_FLAG_DISABLED, that kp will be registered but disabled,
0355 so, its handlers aren't hit until calling enable_kprobe(kp).
0356 
0357 NOTE:
0358 1. With the introduction of the "symbol_name" field to struct kprobe,
0359 the probepoint address resolution will now be taken care of by the kernel.
0360 The following will now work:
0361 
0362         kp.symbol_name = "symbol_name";
0363 
0364 (64-bit powerpc intricacies such as function descriptors are handled
0365 transparently)
0366 
0367 2. Use the "offset" field of struct kprobe if the offset into the symbol
0368 to install a probepoint is known. This field is used to calculate the
0369 probepoint.
0370 
0371 3. Specify either the kprobe "symbol_name" OR the "addr". If both are
0372 specified, kprobe registration will fail with -EINVAL.
0373 
0374 4. With CISC architectures (such as i386 and x86_64), the kprobes code
0375 does not validate if the kprobe.addr is at an instruction boundary.
0376 Use "offset" with caution.
0377 
0378 register_kprobe() returns 0 on success, or a negative errno otherwise.
0379 
0380 User's pre-handler (kp->pre_handler):
0381 #include <linux/kprobes.h>
0382 #include <linux/ptrace.h>
0383 int pre_handler(struct kprobe *p, struct pt_regs *regs);
0384 
0385 Called with p pointing to the kprobe associated with the breakpoint,
0386 and regs pointing to the struct containing the registers saved when
0387 the breakpoint was hit.  Return 0 here unless you're a Kprobes geek.
0388 
0389 User's post-handler (kp->post_handler):
0390 #include <linux/kprobes.h>
0391 #include <linux/ptrace.h>
0392 void post_handler(struct kprobe *p, struct pt_regs *regs,
0393         unsigned long flags);
0394 
0395 p and regs are as described for the pre_handler.  flags always seems
0396 to be zero.
0397 
0398 User's fault-handler (kp->fault_handler):
0399 #include <linux/kprobes.h>
0400 #include <linux/ptrace.h>
0401 int fault_handler(struct kprobe *p, struct pt_regs *regs, int trapnr);
0402 
0403 p and regs are as described for the pre_handler.  trapnr is the
0404 architecture-specific trap number associated with the fault (e.g.,
0405 on i386, 13 for a general protection fault or 14 for a page fault).
0406 Returns 1 if it successfully handled the exception.
0407 
0408 4.2 register_jprobe
0409 
0410 #include <linux/kprobes.h>
0411 int register_jprobe(struct jprobe *jp)
0412 
0413 Sets a breakpoint at the address jp->kp.addr, which must be the address
0414 of the first instruction of a function.  When the breakpoint is hit,
0415 Kprobes runs the handler whose address is jp->entry.
0416 
0417 The handler should have the same arg list and return type as the probed
0418 function; and just before it returns, it must call jprobe_return().
0419 (The handler never actually returns, since jprobe_return() returns
0420 control to Kprobes.)  If the probed function is declared asmlinkage
0421 or anything else that affects how args are passed, the handler's
0422 declaration must match.
0423 
0424 register_jprobe() returns 0 on success, or a negative errno otherwise.
0425 
0426 4.3 register_kretprobe
0427 
0428 #include <linux/kprobes.h>
0429 int register_kretprobe(struct kretprobe *rp);
0430 
0431 Establishes a return probe for the function whose address is
0432 rp->kp.addr.  When that function returns, Kprobes calls rp->handler.
0433 You must set rp->maxactive appropriately before you call
0434 register_kretprobe(); see "How Does a Return Probe Work?" for details.
0435 
0436 register_kretprobe() returns 0 on success, or a negative errno
0437 otherwise.
0438 
0439 User's return-probe handler (rp->handler):
0440 #include <linux/kprobes.h>
0441 #include <linux/ptrace.h>
0442 int kretprobe_handler(struct kretprobe_instance *ri, struct pt_regs *regs);
0443 
0444 regs is as described for kprobe.pre_handler.  ri points to the
0445 kretprobe_instance object, of which the following fields may be
0446 of interest:
0447 - ret_addr: the return address
0448 - rp: points to the corresponding kretprobe object
0449 - task: points to the corresponding task struct
0450 - data: points to per return-instance private data; see "Kretprobe
0451         entry-handler" for details.
0452 
0453 The regs_return_value(regs) macro provides a simple abstraction to
0454 extract the return value from the appropriate register as defined by
0455 the architecture's ABI.
0456 
0457 The handler's return value is currently ignored.
0458 
0459 4.4 unregister_*probe
0460 
0461 #include <linux/kprobes.h>
0462 void unregister_kprobe(struct kprobe *kp);
0463 void unregister_jprobe(struct jprobe *jp);
0464 void unregister_kretprobe(struct kretprobe *rp);
0465 
0466 Removes the specified probe.  The unregister function can be called
0467 at any time after the probe has been registered.
0468 
0469 NOTE:
0470 If the functions find an incorrect probe (ex. an unregistered probe),
0471 they clear the addr field of the probe.
0472 
0473 4.5 register_*probes
0474 
0475 #include <linux/kprobes.h>
0476 int register_kprobes(struct kprobe **kps, int num);
0477 int register_kretprobes(struct kretprobe **rps, int num);
0478 int register_jprobes(struct jprobe **jps, int num);
0479 
0480 Registers each of the num probes in the specified array.  If any
0481 error occurs during registration, all probes in the array, up to
0482 the bad probe, are safely unregistered before the register_*probes
0483 function returns.
0484 - kps/rps/jps: an array of pointers to *probe data structures
0485 - num: the number of the array entries.
0486 
0487 NOTE:
0488 You have to allocate(or define) an array of pointers and set all
0489 of the array entries before using these functions.
0490 
0491 4.6 unregister_*probes
0492 
0493 #include <linux/kprobes.h>
0494 void unregister_kprobes(struct kprobe **kps, int num);
0495 void unregister_kretprobes(struct kretprobe **rps, int num);
0496 void unregister_jprobes(struct jprobe **jps, int num);
0497 
0498 Removes each of the num probes in the specified array at once.
0499 
0500 NOTE:
0501 If the functions find some incorrect probes (ex. unregistered
0502 probes) in the specified array, they clear the addr field of those
0503 incorrect probes. However, other probes in the array are
0504 unregistered correctly.
0505 
0506 4.7 disable_*probe
0507 
0508 #include <linux/kprobes.h>
0509 int disable_kprobe(struct kprobe *kp);
0510 int disable_kretprobe(struct kretprobe *rp);
0511 int disable_jprobe(struct jprobe *jp);
0512 
0513 Temporarily disables the specified *probe. You can enable it again by using
0514 enable_*probe(). You must specify the probe which has been registered.
0515 
0516 4.8 enable_*probe
0517 
0518 #include <linux/kprobes.h>
0519 int enable_kprobe(struct kprobe *kp);
0520 int enable_kretprobe(struct kretprobe *rp);
0521 int enable_jprobe(struct jprobe *jp);
0522 
0523 Enables *probe which has been disabled by disable_*probe(). You must specify
0524 the probe which has been registered.
0525 
0526 5. Kprobes Features and Limitations
0527 
0528 Kprobes allows multiple probes at the same address.  Currently,
0529 however, there cannot be multiple jprobes on the same function at
0530 the same time.  Also, a probepoint for which there is a jprobe or
0531 a post_handler cannot be optimized.  So if you install a jprobe,
0532 or a kprobe with a post_handler, at an optimized probepoint, the
0533 probepoint will be unoptimized automatically.
0534 
0535 In general, you can install a probe anywhere in the kernel.
0536 In particular, you can probe interrupt handlers.  Known exceptions
0537 are discussed in this section.
0538 
0539 The register_*probe functions will return -EINVAL if you attempt
0540 to install a probe in the code that implements Kprobes (mostly
0541 kernel/kprobes.c and arch/*/kernel/kprobes.c, but also functions such
0542 as do_page_fault and notifier_call_chain).
0543 
0544 If you install a probe in an inline-able function, Kprobes makes
0545 no attempt to chase down all inline instances of the function and
0546 install probes there.  gcc may inline a function without being asked,
0547 so keep this in mind if you're not seeing the probe hits you expect.
0548 
0549 A probe handler can modify the environment of the probed function
0550 -- e.g., by modifying kernel data structures, or by modifying the
0551 contents of the pt_regs struct (which are restored to the registers
0552 upon return from the breakpoint).  So Kprobes can be used, for example,
0553 to install a bug fix or to inject faults for testing.  Kprobes, of
0554 course, has no way to distinguish the deliberately injected faults
0555 from the accidental ones.  Don't drink and probe.
0556 
0557 Kprobes makes no attempt to prevent probe handlers from stepping on
0558 each other -- e.g., probing printk() and then calling printk() from a
0559 probe handler.  If a probe handler hits a probe, that second probe's
0560 handlers won't be run in that instance, and the kprobe.nmissed member
0561 of the second probe will be incremented.
0562 
0563 As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of
0564 the same handler) may run concurrently on different CPUs.
0565 
0566 Kprobes does not use mutexes or allocate memory except during
0567 registration and unregistration.
0568 
0569 Probe handlers are run with preemption disabled.  Depending on the
0570 architecture and optimization state, handlers may also run with
0571 interrupts disabled (e.g., kretprobe handlers and optimized kprobe
0572 handlers run without interrupt disabled on x86/x86-64).  In any case,
0573 your handler should not yield the CPU (e.g., by attempting to acquire
0574 a semaphore).
0575 
0576 Since a return probe is implemented by replacing the return
0577 address with the trampoline's address, stack backtraces and calls
0578 to __builtin_return_address() will typically yield the trampoline's
0579 address instead of the real return address for kretprobed functions.
0580 (As far as we can tell, __builtin_return_address() is used only
0581 for instrumentation and error reporting.)
0582 
0583 If the number of times a function is called does not match the number
0584 of times it returns, registering a return probe on that function may
0585 produce undesirable results. In such a case, a line:
0586 kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c
0587 gets printed. With this information, one will be able to correlate the
0588 exact instance of the kretprobe that caused the problem. We have the
0589 do_exit() case covered. do_execve() and do_fork() are not an issue.
0590 We're unaware of other specific cases where this could be a problem.
0591 
0592 If, upon entry to or exit from a function, the CPU is running on
0593 a stack other than that of the current task, registering a return
0594 probe on that function may produce undesirable results.  For this
0595 reason, Kprobes doesn't support return probes (or kprobes or jprobes)
0596 on the x86_64 version of __switch_to(); the registration functions
0597 return -EINVAL.
0598 
0599 On x86/x86-64, since the Jump Optimization of Kprobes modifies
0600 instructions widely, there are some limitations to optimization. To
0601 explain it, we introduce some terminology. Imagine a 3-instruction
0602 sequence consisting of a two 2-byte instructions and one 3-byte
0603 instruction.
0604 
0605         IA
0606          |
0607 [-2][-1][0][1][2][3][4][5][6][7]
0608         [ins1][ins2][  ins3 ]
0609         [<-     DCR       ->]
0610            [<- JTPR ->]
0611 
0612 ins1: 1st Instruction
0613 ins2: 2nd Instruction
0614 ins3: 3rd Instruction
0615 IA:  Insertion Address
0616 JTPR: Jump Target Prohibition Region
0617 DCR: Detoured Code Region
0618 
0619 The instructions in DCR are copied to the out-of-line buffer
0620 of the kprobe, because the bytes in DCR are replaced by
0621 a 5-byte jump instruction. So there are several limitations.
0622 
0623 a) The instructions in DCR must be relocatable.
0624 b) The instructions in DCR must not include a call instruction.
0625 c) JTPR must not be targeted by any jump or call instruction.
0626 d) DCR must not straddle the border between functions.
0627 
0628 Anyway, these limitations are checked by the in-kernel instruction
0629 decoder, so you don't need to worry about that.
0630 
0631 6. Probe Overhead
0632 
0633 On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0
0634 microseconds to process.  Specifically, a benchmark that hits the same
0635 probepoint repeatedly, firing a simple handler each time, reports 1-2
0636 million hits per second, depending on the architecture.  A jprobe or
0637 return-probe hit typically takes 50-75% longer than a kprobe hit.
0638 When you have a return probe set on a function, adding a kprobe at
0639 the entry to that function adds essentially no overhead.
0640 
0641 Here are sample overhead figures (in usec) for different architectures.
0642 k = kprobe; j = jprobe; r = return probe; kr = kprobe + return probe
0643 on same function; jr = jprobe + return probe on same function
0644 
0645 i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips
0646 k = 0.57 usec; j = 1.00; r = 0.92; kr = 0.99; jr = 1.40
0647 
0648 x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips
0649 k = 0.49 usec; j = 0.76; r = 0.80; kr = 0.82; jr = 1.07
0650 
0651 ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU)
0652 k = 0.77 usec; j = 1.31; r = 1.26; kr = 1.45; jr = 1.99
0653 
0654 6.1 Optimized Probe Overhead
0655 
0656 Typically, an optimized kprobe hit takes 0.07 to 0.1 microseconds to
0657 process. Here are sample overhead figures (in usec) for x86 architectures.
0658 k = unoptimized kprobe, b = boosted (single-step skipped), o = optimized kprobe,
0659 r = unoptimized kretprobe, rb = boosted kretprobe, ro = optimized kretprobe.
0660 
0661 i386: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips
0662 k = 0.80 usec; b = 0.33; o = 0.05; r = 1.10; rb = 0.61; ro = 0.33
0663 
0664 x86-64: Intel(R) Xeon(R) E5410, 2.33GHz, 4656.90 bogomips
0665 k = 0.99 usec; b = 0.43; o = 0.06; r = 1.24; rb = 0.68; ro = 0.30
0666 
0667 7. TODO
0668 
0669 a. SystemTap (http://sourceware.org/systemtap): Provides a simplified
0670 programming interface for probe-based instrumentation.  Try it out.
0671 b. Kernel return probes for sparc64.
0672 c. Support for other architectures.
0673 d. User-space probes.
0674 e. Watchpoint probes (which fire on data references).
0675 
0676 8. Kprobes Example
0677 
0678 See samples/kprobes/kprobe_example.c
0679 
0680 9. Jprobes Example
0681 
0682 See samples/kprobes/jprobe_example.c
0683 
0684 10. Kretprobes Example
0685 
0686 See samples/kprobes/kretprobe_example.c
0687 
0688 For additional information on Kprobes, refer to the following URLs:
0689 http://www-106.ibm.com/developerworks/library/l-kprobes.html?ca=dgr-lnxw42Kprobe
0690 http://www.redhat.com/magazine/005mar05/features/kprobes/
0691 http://www-users.cs.umn.edu/~boutcher/kprobes/
0692 http://www.linuxsymposium.org/2006/linuxsymposium_procv2.pdf (pages 101-115)
0693 
0694 
0695 Appendix A: The kprobes debugfs interface
0696 
0697 With recent kernels (> 2.6.20) the list of registered kprobes is visible
0698 under the /sys/kernel/debug/kprobes/ directory (assuming debugfs is mounted at //sys/kernel/debug).
0699 
0700 /sys/kernel/debug/kprobes/list: Lists all registered probes on the system
0701 
0702 c015d71a  k  vfs_read+0x0
0703 c011a316  j  do_fork+0x0
0704 c03dedc5  r  tcp_v4_rcv+0x0
0705 
0706 The first column provides the kernel address where the probe is inserted.
0707 The second column identifies the type of probe (k - kprobe, r - kretprobe
0708 and j - jprobe), while the third column specifies the symbol+offset of
0709 the probe. If the probed function belongs to a module, the module name
0710 is also specified. Following columns show probe status. If the probe is on
0711 a virtual address that is no longer valid (module init sections, module
0712 virtual addresses that correspond to modules that've been unloaded),
0713 such probes are marked with [GONE]. If the probe is temporarily disabled,
0714 such probes are marked with [DISABLED]. If the probe is optimized, it is
0715 marked with [OPTIMIZED]. If the probe is ftrace-based, it is marked with
0716 [FTRACE].
0717 
0718 /sys/kernel/debug/kprobes/enabled: Turn kprobes ON/OFF forcibly.
0719 
0720 Provides a knob to globally and forcibly turn registered kprobes ON or OFF.
0721 By default, all kprobes are enabled. By echoing "0" to this file, all
0722 registered probes will be disarmed, till such time a "1" is echoed to this
0723 file. Note that this knob just disarms and arms all kprobes and doesn't
0724 change each probe's disabling state. This means that disabled kprobes (marked
0725 [DISABLED]) will be not enabled if you turn ON all kprobes by this knob.
0726 
0727 
0728 Appendix B: The kprobes sysctl interface
0729 
0730 /proc/sys/debug/kprobes-optimization: Turn kprobes optimization ON/OFF.
0731 
0732 When CONFIG_OPTPROBES=y, this sysctl interface appears and it provides
0733 a knob to globally and forcibly turn jump optimization (see section
0734 1.4) ON or OFF. By default, jump optimization is allowed (ON).
0735 If you echo "0" to this file or set "debug.kprobes_optimization" to
0736 0 via sysctl, all optimized probes will be unoptimized, and any new
0737 probes registered after that will not be optimized.  Note that this
0738 knob *changes* the optimized state. This means that optimized probes
0739 (marked [OPTIMIZED]) will be unoptimized ([OPTIMIZED] tag will be
0740 removed). If the knob is turned on, they will be optimized again.
0741