Title : Kernel Probes (Kprobes) Authors : Jim Keniston <jkenisto@us.ibm.com> : Prasanna S Panchamukhi <prasanna@in.ibm.com> CONTENTS 1. Concepts: Kprobes, Jprobes, Return Probes 2. Architectures Supported 3. Configuring Kprobes 4. API Reference 5. Kprobes Features and Limitations 6. Probe Overhead 7. TODO 8. Kprobes Example 9. Jprobes Example 10. Kretprobes Example Appendix A: The kprobes debugfs interface 1. Concepts: Kprobes, Jprobes, Return Probes Kprobes enables you to dynamically break into any kernel routine and collect debugging and performance information non-disruptively. You can trap at almost any kernel code address, specifying a handler routine to be invoked when the breakpoint is hit. There are currently three types of probes: kprobes, jprobes, and kretprobes (also called return probes). A kprobe can be inserted on virtually any instruction in the kernel. A jprobe is inserted at the entry to a kernel function, and provides convenient access to the function's arguments. A return probe fires when a specified function returns. In the typical case, Kprobes-based instrumentation is packaged as a kernel module. The module's init function installs ("registers") one or more probes, and the exit function unregisters them. A registration function such as register_kprobe() specifies where the probe is to be inserted and what handler is to be called when the probe is hit. The next three subsections explain how the different types of probes work. They explain certain things that you'll need to know in order to make the best use of Kprobes -- e.g., the difference between a pre_handler and a post_handler, and how to use the maxactive and nmissed fields of a kretprobe. But if you're in a hurry to start using Kprobes, you can skip ahead to section 2. 1.1 How Does a Kprobe Work? When a kprobe is registered, Kprobes makes a copy of the probed instruction and replaces the first byte(s) of the probed instruction with a breakpoint instruction (e.g., int3 on i386 and x86_64). When a CPU hits the breakpoint instruction, a trap occurs, the CPU's registers are saved, and control passes to Kprobes via the notifier_call_chain mechanism. Kprobes executes the "pre_handler" associated with the kprobe, passing the handler the addresses of the kprobe struct and the saved registers. Next, Kprobes single-steps its copy of the probed instruction. (It would be simpler to single-step the actual instruction in place, but then Kprobes would have to temporarily remove the breakpoint instruction. This would open a small time window when another CPU could sail right past the probepoint.) After the instruction is single-stepped, Kprobes executes the "post_handler," if any, that is associated with the kprobe. Execution then continues with the instruction following the probepoint. 1.2 How Does a Jprobe Work? A jprobe is implemented using a kprobe that is placed on a function's entry point. It employs a simple mirroring principle to allow seamless access to the probed function's arguments. The jprobe handler routine should have the same signature (arg list and return type) as the function being probed, and must always end by calling the Kprobes function jprobe_return(). Here's how it works. When the probe is hit, Kprobes makes a copy of the saved registers and a generous portion of the stack (see below). Kprobes then points the saved instruction pointer at the jprobe's handler routine, and returns from the trap. As a result, control passes to the handler, which is presented with the same register and stack contents as the probed function. When it is done, the handler calls jprobe_return(), which traps again to restore the original stack contents and processor state and switch to the probed function. By convention, the callee owns its arguments, so gcc may produce code that unexpectedly modifies that portion of the stack. This is why Kprobes saves a copy of the stack and restores it after the jprobe handler has run. Up to MAX_STACK_SIZE bytes are copied -- e.g., 64 bytes on i386. Note that the probed function's args may be passed on the stack or in registers. The jprobe will work in either case, so long as the handler's prototype matches that of the probed function. 1.3 Return Probes 1.3.1 How Does a Return Probe Work? When you call register_kretprobe(), Kprobes establishes a kprobe at the entry to the function. When the probed function is called and this probe is hit, Kprobes saves a copy of the return address, and replaces the return address with the address of a "trampoline." The trampoline is an arbitrary piece of code -- typically just a nop instruction. At boot time, Kprobes registers a kprobe at the trampoline. When the probed function executes its return instruction, control passes to the trampoline and that probe is hit. Kprobes' trampoline handler calls the user-specified return handler associated with the kretprobe, then sets the saved instruction pointer to the saved return address, and that's where execution resumes upon return from the trap. While the probed function is executing, its return address is stored in an object of type kretprobe_instance. Before calling register_kretprobe(), the user sets the maxactive field of the kretprobe struct to specify how many instances of the specified function can be probed simultaneously. register_kretprobe() pre-allocates the indicated number of kretprobe_instance objects. For example, if the function is non-recursive and is called with a spinlock held, maxactive = 1 should be enough. If the function is non-recursive and can never relinquish the CPU (e.g., via a semaphore or preemption), NR_CPUS should be enough. If maxactive <= 0, it is set to a default value. If CONFIG_PREEMPT is enabled, the default is max(10, 2*NR_CPUS). Otherwise, the default is NR_CPUS. It's not a disaster if you set maxactive too low; you'll just miss some probes. In the kretprobe struct, the nmissed field is set to zero when the return probe is registered, and is incremented every time the probed function is entered but there is no kretprobe_instance object available for establishing the return probe. 1.3.2 Kretprobe entry-handler Kretprobes also provides an optional user-specified handler which runs on function entry. This handler is specified by setting the entry_handler field of the kretprobe struct. Whenever the kprobe placed by kretprobe at the function entry is hit, the user-defined entry_handler, if any, is invoked. If the entry_handler returns 0 (success) then a corresponding return handler is guaranteed to be called upon function return. If the entry_handler returns a non-zero error then Kprobes leaves the return address as is, and the kretprobe has no further effect for that particular function instance. Multiple entry and return handler invocations are matched using the unique kretprobe_instance object associated with them. Additionally, a user may also specify per return-instance private data to be part of each kretprobe_instance object. This is especially useful when sharing private data between corresponding user entry and return handlers. The size of each private data object can be specified at kretprobe registration time by setting the data_size field of the kretprobe struct. This data can be accessed through the data field of each kretprobe_instance object. In case probed function is entered but there is no kretprobe_instance object available, then in addition to incrementing the nmissed count, the user entry_handler invocation is also skipped. 2. Architectures Supported Kprobes, jprobes, and return probes are implemented on the following architectures: - i386 - x86_64 (AMD-64, EM64T) - ppc64 - ia64 (Does not support probes on instruction slot1.) - sparc64 (Return probes not yet implemented.) - arm 3. Configuring Kprobes When configuring the kernel using make menuconfig/xconfig/oldconfig, ensure that CONFIG_KPROBES is set to "y". Under "Instrumentation Support", look for "Kprobes". So that you can load and unload Kprobes-based instrumentation modules, make sure "Loadable module support" (CONFIG_MODULES) and "Module unloading" (CONFIG_MODULE_UNLOAD) are set to "y". Also make sure that CONFIG_KALLSYMS and perhaps even CONFIG_KALLSYMS_ALL are set to "y", since kallsyms_lookup_name() is used by the in-kernel kprobe address resolution code. If you need to insert a probe in the middle of a function, you may find it useful to "Compile the kernel with debug info" (CONFIG_DEBUG_INFO), so you can use "objdump -d -l vmlinux" to see the source-to-object code mapping. 4. API Reference The Kprobes API includes a "register" function and an "unregister" function for each type of probe. Here are terse, mini-man-page specifications for these functions and the associated probe handlers that you'll write. See the files in the samples/kprobes/ sub-directory for examples. 4.1 register_kprobe #include <linux/kprobes.h> int register_kprobe(struct kprobe *kp); Sets a breakpoint at the address kp->addr. When the breakpoint is hit, Kprobes calls kp->pre_handler. After the probed instruction is single-stepped, Kprobe calls kp->post_handler. If a fault occurs during execution of kp->pre_handler or kp->post_handler, or during single-stepping of the probed instruction, Kprobes calls kp->fault_handler. Any or all handlers can be NULL. NOTE: 1. With the introduction of the "symbol_name" field to struct kprobe, the probepoint address resolution will now be taken care of by the kernel. The following will now work: kp.symbol_name = "symbol_name"; (64-bit powerpc intricacies such as function descriptors are handled transparently) 2. Use the "offset" field of struct kprobe if the offset into the symbol to install a probepoint is known. This field is used to calculate the probepoint. 3. Specify either the kprobe "symbol_name" OR the "addr". If both are specified, kprobe registration will fail with -EINVAL. 4. With CISC architectures (such as i386 and x86_64), the kprobes code does not validate if the kprobe.addr is at an instruction boundary. Use "offset" with caution. register_kprobe() returns 0 on success, or a negative errno otherwise. User's pre-handler (kp->pre_handler): #include <linux/kprobes.h> #include <linux/ptrace.h> int pre_handler(struct kprobe *p, struct pt_regs *regs); Called with p pointing to the kprobe associated with the breakpoint, and regs pointing to the struct containing the registers saved when the breakpoint was hit. Return 0 here unless you're a Kprobes geek. User's post-handler (kp->post_handler): #include <linux/kprobes.h> #include <linux/ptrace.h> void post_handler(struct kprobe *p, struct pt_regs *regs, unsigned long flags); p and regs are as described for the pre_handler. flags always seems to be zero. User's fault-handler (kp->fault_handler): #include <linux/kprobes.h> #include <linux/ptrace.h> int fault_handler(struct kprobe *p, struct pt_regs *regs, int trapnr); p and regs are as described for the pre_handler. trapnr is the architecture-specific trap number associated with the fault (e.g., on i386, 13 for a general protection fault or 14 for a page fault). Returns 1 if it successfully handled the exception. 4.2 register_jprobe #include <linux/kprobes.h> int register_jprobe(struct jprobe *jp) Sets a breakpoint at the address jp->kp.addr, which must be the address of the first instruction of a function. When the breakpoint is hit, Kprobes runs the handler whose address is jp->entry. The handler should have the same arg list and return type as the probed function; and just before it returns, it must call jprobe_return(). (The handler never actually returns, since jprobe_return() returns control to Kprobes.) If the probed function is declared asmlinkage or anything else that affects how args are passed, the handler's declaration must match. register_jprobe() returns 0 on success, or a negative errno otherwise. 4.3 register_kretprobe #include <linux/kprobes.h> int register_kretprobe(struct kretprobe *rp); Establishes a return probe for the function whose address is rp->kp.addr. When that function returns, Kprobes calls rp->handler. You must set rp->maxactive appropriately before you call register_kretprobe(); see "How Does a Return Probe Work?" for details. register_kretprobe() returns 0 on success, or a negative errno otherwise. User's return-probe handler (rp->handler): #include <linux/kprobes.h> #include <linux/ptrace.h> int kretprobe_handler(struct kretprobe_instance *ri, struct pt_regs *regs); regs is as described for kprobe.pre_handler. ri points to the kretprobe_instance object, of which the following fields may be of interest: - ret_addr: the return address - rp: points to the corresponding kretprobe object - task: points to the corresponding task struct - data: points to per return-instance private data; see "Kretprobe entry-handler" for details. The regs_return_value(regs) macro provides a simple abstraction to extract the return value from the appropriate register as defined by the architecture's ABI. The handler's return value is currently ignored. 4.4 unregister_*probe #include <linux/kprobes.h> void unregister_kprobe(struct kprobe *kp); void unregister_jprobe(struct jprobe *jp); void unregister_kretprobe(struct kretprobe *rp); Removes the specified probe. The unregister function can be called at any time after the probe has been registered. 5. Kprobes Features and Limitations Kprobes allows multiple probes at the same address. Currently, however, there cannot be multiple jprobes on the same function at the same time. In general, you can install a probe anywhere in the kernel. In particular, you can probe interrupt handlers. Known exceptions are discussed in this section. The register_*probe functions will return -EINVAL if you attempt to install a probe in the code that implements Kprobes (mostly kernel/kprobes.c and arch/*/kernel/kprobes.c, but also functions such as do_page_fault and notifier_call_chain). If you install a probe in an inline-able function, Kprobes makes no attempt to chase down all inline instances of the function and install probes there. gcc may inline a function without being asked, so keep this in mind if you're not seeing the probe hits you expect. A probe handler can modify the environment of the probed function -- e.g., by modifying kernel data structures, or by modifying the contents of the pt_regs struct (which are restored to the registers upon return from the breakpoint). So Kprobes can be used, for example, to install a bug fix or to inject faults for testing. Kprobes, of course, has no way to distinguish the deliberately injected faults from the accidental ones. Don't drink and probe. Kprobes makes no attempt to prevent probe handlers from stepping on each other -- e.g., probing printk() and then calling printk() from a probe handler. If a probe handler hits a probe, that second probe's handlers won't be run in that instance, and the kprobe.nmissed member of the second probe will be incremented. As of Linux v2.6.15-rc1, multiple handlers (or multiple instances of the same handler) may run concurrently on different CPUs. Kprobes does not use mutexes or allocate memory except during registration and unregistration. Probe handlers are run with preemption disabled. Depending on the architecture, handlers may also run with interrupts disabled. In any case, your handler should not yield the CPU (e.g., by attempting to acquire a semaphore). Since a return probe is implemented by replacing the return address with the trampoline's address, stack backtraces and calls to __builtin_return_address() will typically yield the trampoline's address instead of the real return address for kretprobed functions. (As far as we can tell, __builtin_return_address() is used only for instrumentation and error reporting.) If the number of times a function is called does not match the number of times it returns, registering a return probe on that function may produce undesirable results. In such a case, a line: kretprobe BUG!: Processing kretprobe d000000000041aa8 @ c00000000004f48c gets printed. With this information, one will be able to correlate the exact instance of the kretprobe that caused the problem. We have the do_exit() case covered. do_execve() and do_fork() are not an issue. We're unaware of other specific cases where this could be a problem. If, upon entry to or exit from a function, the CPU is running on a stack other than that of the current task, registering a return probe on that function may produce undesirable results. For this reason, Kprobes doesn't support return probes (or kprobes or jprobes) on the x86_64 version of __switch_to(); the registration functions return -EINVAL. 6. Probe Overhead On a typical CPU in use in 2005, a kprobe hit takes 0.5 to 1.0 microseconds to process. Specifically, a benchmark that hits the same probepoint repeatedly, firing a simple handler each time, reports 1-2 million hits per second, depending on the architecture. A jprobe or return-probe hit typically takes 50-75% longer than a kprobe hit. When you have a return probe set on a function, adding a kprobe at the entry to that function adds essentially no overhead. Here are sample overhead figures (in usec) for different architectures. k = kprobe; j = jprobe; r = return probe; kr = kprobe + return probe on same function; jr = jprobe + return probe on same function i386: Intel Pentium M, 1495 MHz, 2957.31 bogomips k = 0.57 usec; j = 1.00; r = 0.92; kr = 0.99; jr = 1.40 x86_64: AMD Opteron 246, 1994 MHz, 3971.48 bogomips k = 0.49 usec; j = 0.76; r = 0.80; kr = 0.82; jr = 1.07 ppc64: POWER5 (gr), 1656 MHz (SMT disabled, 1 virtual CPU per physical CPU) k = 0.77 usec; j = 1.31; r = 1.26; kr = 1.45; jr = 1.99 7. TODO a. SystemTap (http://sourceware.org/systemtap): Provides a simplified programming interface for probe-based instrumentation. Try it out. b. Kernel return probes for sparc64. c. Support for other architectures. d. User-space probes. e. Watchpoint probes (which fire on data references). 8. Kprobes Example See samples/kprobes/kprobe_example.c 9. Jprobes Example See samples/kprobes/jprobe_example.c 10. Kretprobes Example See samples/kprobes/kretprobe_example.c For additional information on Kprobes, refer to the following URLs: http://www-106.ibm.com/developerworks/library/l-kprobes.html?ca=dgr-lnxw42Kprobe http://www.redhat.com/magazine/005mar05/features/kprobes/ http://www-users.cs.umn.edu/~boutcher/kprobes/ http://www.linuxsymposium.org/2006/linuxsymposium_procv2.pdf (pages 101-115) Appendix A: The kprobes debugfs interface With recent kernels (> 2.6.20) the list of registered kprobes is visible under the /debug/kprobes/ directory (assuming debugfs is mounted at /debug). /debug/kprobes/list: Lists all registered probes on the system c015d71a k vfs_read+0x0 c011a316 j do_fork+0x0 c03dedc5 r tcp_v4_rcv+0x0 The first column provides the kernel address where the probe is inserted. The second column identifies the type of probe (k - kprobe, r - kretprobe and j - jprobe), while the third column specifies the symbol+offset of the probe. If the probed function belongs to a module, the module name is also specified. /debug/kprobes/enabled: Turn kprobes ON/OFF Provides a knob to globally turn registered kprobes ON or OFF. By default, all kprobes are enabled. By echoing "0" to this file, all registered probes will be disarmed, till such time a "1" is echoed to this file.