2 * Kernel Probes (KProbes)
3 * arch/i386/kernel/kprobes.c
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 * Copyright (C) IBM Corporation, 2002, 2004
21 * 2002-Oct Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
22 * Probes initial implementation ( includes contributions from
24 * 2004-July Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
25 * interface to access function arguments.
26 * 2005-May Hien Nguyen <hien@us.ibm.com>, Jim Keniston
27 * <jkenisto@us.ibm.com> and Prasanna S Panchamukhi
28 * <prasanna@in.ibm.com> added function-return probes.
31 #include <linux/config.h>
32 #include <linux/kprobes.h>
33 #include <linux/ptrace.h>
34 #include <linux/preempt.h>
35 #include <asm/cacheflush.h>
36 #include <asm/kdebug.h>
39 void jprobe_return_end(void);
41 DEFINE_PER_CPU(struct kprobe *, current_kprobe) = NULL;
42 DEFINE_PER_CPU(struct kprobe_ctlblk, kprobe_ctlblk);
45 * returns non-zero if opcode modifies the interrupt flag.
47 static inline int is_IF_modifier(kprobe_opcode_t opcode)
52 case 0xcf: /* iret/iretd */
53 case 0x9d: /* popf/popfd */
59 int __kprobes arch_prepare_kprobe(struct kprobe *p)
64 void __kprobes arch_copy_kprobe(struct kprobe *p)
66 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
70 void __kprobes arch_arm_kprobe(struct kprobe *p)
72 *p->addr = BREAKPOINT_INSTRUCTION;
73 flush_icache_range((unsigned long) p->addr,
74 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
77 void __kprobes arch_disarm_kprobe(struct kprobe *p)
80 flush_icache_range((unsigned long) p->addr,
81 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
84 void __kprobes arch_remove_kprobe(struct kprobe *p)
88 static inline void save_previous_kprobe(struct kprobe_ctlblk *kcb)
90 kcb->prev_kprobe.kp = kprobe_running();
91 kcb->prev_kprobe.status = kcb->kprobe_status;
92 kcb->prev_kprobe.old_eflags = kcb->kprobe_old_eflags;
93 kcb->prev_kprobe.saved_eflags = kcb->kprobe_saved_eflags;
96 static inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb)
98 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
99 kcb->kprobe_status = kcb->prev_kprobe.status;
100 kcb->kprobe_old_eflags = kcb->prev_kprobe.old_eflags;
101 kcb->kprobe_saved_eflags = kcb->prev_kprobe.saved_eflags;
104 static inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
105 struct kprobe_ctlblk *kcb)
107 __get_cpu_var(current_kprobe) = p;
108 kcb->kprobe_saved_eflags = kcb->kprobe_old_eflags
109 = (regs->eflags & (TF_MASK | IF_MASK));
110 if (is_IF_modifier(p->opcode))
111 kcb->kprobe_saved_eflags &= ~IF_MASK;
114 static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
116 regs->eflags |= TF_MASK;
117 regs->eflags &= ~IF_MASK;
118 /*single step inline if the instruction is an int3*/
119 if (p->opcode == BREAKPOINT_INSTRUCTION)
120 regs->eip = (unsigned long)p->addr;
122 regs->eip = (unsigned long)&p->ainsn.insn;
125 /* Called with kretprobe_lock held */
126 void __kprobes arch_prepare_kretprobe(struct kretprobe *rp,
127 struct pt_regs *regs)
129 unsigned long *sara = (unsigned long *)®s->esp;
130 struct kretprobe_instance *ri;
132 if ((ri = get_free_rp_inst(rp)) != NULL) {
135 ri->ret_addr = (kprobe_opcode_t *) *sara;
137 /* Replace the return addr with trampoline addr */
138 *sara = (unsigned long) &kretprobe_trampoline;
147 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
148 * remain disabled thorough out this function.
150 static int __kprobes kprobe_handler(struct pt_regs *regs)
154 kprobe_opcode_t *addr = NULL;
156 struct kprobe_ctlblk *kcb;
159 * We don't want to be preempted for the entire
160 * duration of kprobe processing
163 kcb = get_kprobe_ctlblk();
165 /* Check if the application is using LDT entry for its code segment and
166 * calculate the address by reading the base address from the LDT entry.
168 if ((regs->xcs & 4) && (current->mm)) {
169 lp = (unsigned long *) ((unsigned long)((regs->xcs >> 3) * 8)
170 + (char *) current->mm->context.ldt);
171 addr = (kprobe_opcode_t *) (get_desc_base(lp) + regs->eip -
172 sizeof(kprobe_opcode_t));
174 addr = (kprobe_opcode_t *)(regs->eip - sizeof(kprobe_opcode_t));
176 /* Check we're not actually recursing */
177 if (kprobe_running()) {
178 p = get_kprobe(addr);
180 if (kcb->kprobe_status == KPROBE_HIT_SS &&
181 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
182 regs->eflags &= ~TF_MASK;
183 regs->eflags |= kcb->kprobe_saved_eflags;
186 /* We have reentered the kprobe_handler(), since
187 * another probe was hit while within the handler.
188 * We here save the original kprobes variables and
189 * just single step on the instruction of the new probe
190 * without calling any user handlers.
192 save_previous_kprobe(kcb);
193 set_current_kprobe(p, regs, kcb);
195 prepare_singlestep(p, regs);
196 kcb->kprobe_status = KPROBE_REENTER;
199 p = __get_cpu_var(current_kprobe);
200 if (p->break_handler && p->break_handler(p, regs)) {
207 p = get_kprobe(addr);
209 if (regs->eflags & VM_MASK) {
210 /* We are in virtual-8086 mode. Return 0 */
214 if (*addr != BREAKPOINT_INSTRUCTION) {
216 * The breakpoint instruction was removed right
217 * after we hit it. Another cpu has removed
218 * either a probepoint or a debugger breakpoint
219 * at this address. In either case, no further
220 * handling of this interrupt is appropriate.
221 * Back up over the (now missing) int3 and run
222 * the original instruction.
224 regs->eip -= sizeof(kprobe_opcode_t);
227 /* Not one of ours: let kernel handle it */
231 set_current_kprobe(p, regs, kcb);
232 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
234 if (p->pre_handler && p->pre_handler(p, regs))
235 /* handler has already set things up, so skip ss setup */
239 prepare_singlestep(p, regs);
240 kcb->kprobe_status = KPROBE_HIT_SS;
244 preempt_enable_no_resched();
249 * For function-return probes, init_kprobes() establishes a probepoint
250 * here. When a retprobed function returns, this probe is hit and
251 * trampoline_probe_handler() runs, calling the kretprobe's handler.
253 void kretprobe_trampoline_holder(void)
255 asm volatile ( ".global kretprobe_trampoline\n"
256 "kretprobe_trampoline: \n"
261 * Called when we hit the probe point at kretprobe_trampoline
263 int __kprobes trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
265 struct kretprobe_instance *ri = NULL;
266 struct hlist_head *head;
267 struct hlist_node *node, *tmp;
268 unsigned long flags, orig_ret_address = 0;
269 unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
271 spin_lock_irqsave(&kretprobe_lock, flags);
272 head = kretprobe_inst_table_head(current);
275 * It is possible to have multiple instances associated with a given
276 * task either because an multiple functions in the call path
277 * have a return probe installed on them, and/or more then one return
278 * return probe was registered for a target function.
280 * We can handle this because:
281 * - instances are always inserted at the head of the list
282 * - when multiple return probes are registered for the same
283 * function, the first instance's ret_addr will point to the
284 * real return address, and all the rest will point to
285 * kretprobe_trampoline
287 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
288 if (ri->task != current)
289 /* another task is sharing our hash bucket */
292 if (ri->rp && ri->rp->handler)
293 ri->rp->handler(ri, regs);
295 orig_ret_address = (unsigned long)ri->ret_addr;
298 if (orig_ret_address != trampoline_address)
300 * This is the real return address. Any other
301 * instances associated with this task are for
302 * other calls deeper on the call stack
307 BUG_ON(!orig_ret_address || (orig_ret_address == trampoline_address));
308 regs->eip = orig_ret_address;
310 reset_current_kprobe();
311 spin_unlock_irqrestore(&kretprobe_lock, flags);
312 preempt_enable_no_resched();
315 * By returning a non-zero value, we are telling
316 * kprobe_handler() that we don't want the post_handler
317 * to run (and have re-enabled preemption)
323 * Called after single-stepping. p->addr is the address of the
324 * instruction whose first byte has been replaced by the "int 3"
325 * instruction. To avoid the SMP problems that can occur when we
326 * temporarily put back the original opcode to single-step, we
327 * single-stepped a copy of the instruction. The address of this
328 * copy is p->ainsn.insn.
330 * This function prepares to return from the post-single-step
331 * interrupt. We have to fix up the stack as follows:
333 * 0) Except in the case of absolute or indirect jump or call instructions,
334 * the new eip is relative to the copied instruction. We need to make
335 * it relative to the original instruction.
337 * 1) If the single-stepped instruction was pushfl, then the TF and IF
338 * flags are set in the just-pushed eflags, and may need to be cleared.
340 * 2) If the single-stepped instruction was a call, the return address
341 * that is atop the stack is the address following the copied instruction.
342 * We need to make it the address following the original instruction.
344 static void __kprobes resume_execution(struct kprobe *p,
345 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
347 unsigned long *tos = (unsigned long *)®s->esp;
348 unsigned long next_eip = 0;
349 unsigned long copy_eip = (unsigned long)&p->ainsn.insn;
350 unsigned long orig_eip = (unsigned long)p->addr;
352 switch (p->ainsn.insn[0]) {
353 case 0x9c: /* pushfl */
354 *tos &= ~(TF_MASK | IF_MASK);
355 *tos |= kcb->kprobe_old_eflags;
357 case 0xc3: /* ret/lret */
361 regs->eflags &= ~TF_MASK;
362 /* eip is already adjusted, no more changes required*/
364 case 0xe8: /* call relative - Fix return addr */
365 *tos = orig_eip + (*tos - copy_eip);
368 if ((p->ainsn.insn[1] & 0x30) == 0x10) {
369 /* call absolute, indirect */
370 /* Fix return addr; eip is correct. */
371 next_eip = regs->eip;
372 *tos = orig_eip + (*tos - copy_eip);
373 } else if (((p->ainsn.insn[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
374 ((p->ainsn.insn[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
375 /* eip is correct. */
376 next_eip = regs->eip;
379 case 0xea: /* jmp absolute -- eip is correct */
380 next_eip = regs->eip;
386 regs->eflags &= ~TF_MASK;
388 regs->eip = next_eip;
390 regs->eip = orig_eip + (regs->eip - copy_eip);
395 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
396 * remain disabled thoroughout this function.
398 static inline int post_kprobe_handler(struct pt_regs *regs)
400 struct kprobe *cur = kprobe_running();
401 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
406 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
407 kcb->kprobe_status = KPROBE_HIT_SSDONE;
408 cur->post_handler(cur, regs, 0);
411 resume_execution(cur, regs, kcb);
412 regs->eflags |= kcb->kprobe_saved_eflags;
414 /*Restore back the original saved kprobes variables and continue. */
415 if (kcb->kprobe_status == KPROBE_REENTER) {
416 restore_previous_kprobe(kcb);
419 reset_current_kprobe();
421 preempt_enable_no_resched();
424 * if somebody else is singlestepping across a probe point, eflags
425 * will have TF set, in which case, continue the remaining processing
426 * of do_debug, as if this is not a probe hit.
428 if (regs->eflags & TF_MASK)
434 static inline int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
436 struct kprobe *cur = kprobe_running();
437 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
439 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
442 if (kcb->kprobe_status & KPROBE_HIT_SS) {
443 resume_execution(cur, regs, kcb);
444 regs->eflags |= kcb->kprobe_old_eflags;
446 reset_current_kprobe();
447 preempt_enable_no_resched();
453 * Wrapper routine to for handling exceptions.
455 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
456 unsigned long val, void *data)
458 struct die_args *args = (struct die_args *)data;
459 int ret = NOTIFY_DONE;
463 if (kprobe_handler(args->regs))
467 if (post_kprobe_handler(args->regs))
472 /* kprobe_running() needs smp_processor_id() */
474 if (kprobe_running() &&
475 kprobe_fault_handler(args->regs, args->trapnr))
485 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
487 struct jprobe *jp = container_of(p, struct jprobe, kp);
489 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
491 kcb->jprobe_saved_regs = *regs;
492 kcb->jprobe_saved_esp = ®s->esp;
493 addr = (unsigned long)(kcb->jprobe_saved_esp);
496 * TBD: As Linus pointed out, gcc assumes that the callee
497 * owns the argument space and could overwrite it, e.g.
498 * tailcall optimization. So, to be absolutely safe
499 * we also save and restore enough stack bytes to cover
502 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
503 MIN_STACK_SIZE(addr));
504 regs->eflags &= ~IF_MASK;
505 regs->eip = (unsigned long)(jp->entry);
509 void __kprobes jprobe_return(void)
511 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
513 asm volatile (" xchgl %%ebx,%%esp \n"
515 " .globl jprobe_return_end \n"
516 " jprobe_return_end: \n"
518 (kcb->jprobe_saved_esp):"memory");
521 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
523 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
524 u8 *addr = (u8 *) (regs->eip - 1);
525 unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_esp);
526 struct jprobe *jp = container_of(p, struct jprobe, kp);
528 if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
529 if (®s->esp != kcb->jprobe_saved_esp) {
530 struct pt_regs *saved_regs =
531 container_of(kcb->jprobe_saved_esp,
532 struct pt_regs, esp);
533 printk("current esp %p does not match saved esp %p\n",
534 ®s->esp, kcb->jprobe_saved_esp);
535 printk("Saved registers for jprobe %p\n", jp);
536 show_registers(saved_regs);
537 printk("Current registers\n");
538 show_registers(regs);
541 *regs = kcb->jprobe_saved_regs;
542 memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
543 MIN_STACK_SIZE(stack_addr));
544 preempt_enable_no_resched();
550 static struct kprobe trampoline_p = {
551 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
552 .pre_handler = trampoline_probe_handler
555 int __init arch_init_kprobes(void)
557 return register_kprobe(&trampoline_p);