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)
61 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
66 void __kprobes arch_arm_kprobe(struct kprobe *p)
68 *p->addr = BREAKPOINT_INSTRUCTION;
69 flush_icache_range((unsigned long) p->addr,
70 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
73 void __kprobes arch_disarm_kprobe(struct kprobe *p)
76 flush_icache_range((unsigned long) p->addr,
77 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
80 static inline void save_previous_kprobe(struct kprobe_ctlblk *kcb)
82 kcb->prev_kprobe.kp = kprobe_running();
83 kcb->prev_kprobe.status = kcb->kprobe_status;
84 kcb->prev_kprobe.old_eflags = kcb->kprobe_old_eflags;
85 kcb->prev_kprobe.saved_eflags = kcb->kprobe_saved_eflags;
88 static inline void restore_previous_kprobe(struct kprobe_ctlblk *kcb)
90 __get_cpu_var(current_kprobe) = kcb->prev_kprobe.kp;
91 kcb->kprobe_status = kcb->prev_kprobe.status;
92 kcb->kprobe_old_eflags = kcb->prev_kprobe.old_eflags;
93 kcb->kprobe_saved_eflags = kcb->prev_kprobe.saved_eflags;
96 static inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs,
97 struct kprobe_ctlblk *kcb)
99 __get_cpu_var(current_kprobe) = p;
100 kcb->kprobe_saved_eflags = kcb->kprobe_old_eflags
101 = (regs->eflags & (TF_MASK | IF_MASK));
102 if (is_IF_modifier(p->opcode))
103 kcb->kprobe_saved_eflags &= ~IF_MASK;
106 static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
108 regs->eflags |= TF_MASK;
109 regs->eflags &= ~IF_MASK;
110 /*single step inline if the instruction is an int3*/
111 if (p->opcode == BREAKPOINT_INSTRUCTION)
112 regs->eip = (unsigned long)p->addr;
114 regs->eip = (unsigned long)&p->ainsn.insn;
117 /* Called with kretprobe_lock held */
118 void __kprobes arch_prepare_kretprobe(struct kretprobe *rp,
119 struct pt_regs *regs)
121 unsigned long *sara = (unsigned long *)®s->esp;
122 struct kretprobe_instance *ri;
124 if ((ri = get_free_rp_inst(rp)) != NULL) {
127 ri->ret_addr = (kprobe_opcode_t *) *sara;
129 /* Replace the return addr with trampoline addr */
130 *sara = (unsigned long) &kretprobe_trampoline;
139 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
140 * remain disabled thorough out this function.
142 static int __kprobes kprobe_handler(struct pt_regs *regs)
146 kprobe_opcode_t *addr = NULL;
148 struct kprobe_ctlblk *kcb;
151 * We don't want to be preempted for the entire
152 * duration of kprobe processing
155 kcb = get_kprobe_ctlblk();
157 /* Check if the application is using LDT entry for its code segment and
158 * calculate the address by reading the base address from the LDT entry.
160 if ((regs->xcs & 4) && (current->mm)) {
161 lp = (unsigned long *) ((unsigned long)((regs->xcs >> 3) * 8)
162 + (char *) current->mm->context.ldt);
163 addr = (kprobe_opcode_t *) (get_desc_base(lp) + regs->eip -
164 sizeof(kprobe_opcode_t));
166 addr = (kprobe_opcode_t *)(regs->eip - sizeof(kprobe_opcode_t));
168 /* Check we're not actually recursing */
169 if (kprobe_running()) {
170 p = get_kprobe(addr);
172 if (kcb->kprobe_status == KPROBE_HIT_SS &&
173 *p->ainsn.insn == BREAKPOINT_INSTRUCTION) {
174 regs->eflags &= ~TF_MASK;
175 regs->eflags |= kcb->kprobe_saved_eflags;
178 /* We have reentered the kprobe_handler(), since
179 * another probe was hit while within the handler.
180 * We here save the original kprobes variables and
181 * just single step on the instruction of the new probe
182 * without calling any user handlers.
184 save_previous_kprobe(kcb);
185 set_current_kprobe(p, regs, kcb);
186 kprobes_inc_nmissed_count(p);
187 prepare_singlestep(p, regs);
188 kcb->kprobe_status = KPROBE_REENTER;
191 if (regs->eflags & VM_MASK) {
192 /* We are in virtual-8086 mode. Return 0 */
195 if (*addr != BREAKPOINT_INSTRUCTION) {
196 /* The breakpoint instruction was removed by
197 * another cpu right after we hit, no further
198 * handling of this interrupt is appropriate
200 regs->eip -= sizeof(kprobe_opcode_t);
204 p = __get_cpu_var(current_kprobe);
205 if (p->break_handler && p->break_handler(p, regs)) {
212 p = get_kprobe(addr);
214 if (regs->eflags & VM_MASK) {
215 /* We are in virtual-8086 mode. Return 0 */
219 if (*addr != BREAKPOINT_INSTRUCTION) {
221 * The breakpoint instruction was removed right
222 * after we hit it. Another cpu has removed
223 * either a probepoint or a debugger breakpoint
224 * at this address. In either case, no further
225 * handling of this interrupt is appropriate.
226 * Back up over the (now missing) int3 and run
227 * the original instruction.
229 regs->eip -= sizeof(kprobe_opcode_t);
232 /* Not one of ours: let kernel handle it */
236 set_current_kprobe(p, regs, kcb);
237 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
239 if (p->pre_handler && p->pre_handler(p, regs))
240 /* handler has already set things up, so skip ss setup */
244 prepare_singlestep(p, regs);
245 kcb->kprobe_status = KPROBE_HIT_SS;
249 preempt_enable_no_resched();
254 * For function-return probes, init_kprobes() establishes a probepoint
255 * here. When a retprobed function returns, this probe is hit and
256 * trampoline_probe_handler() runs, calling the kretprobe's handler.
258 void kretprobe_trampoline_holder(void)
260 asm volatile ( ".global kretprobe_trampoline\n"
261 "kretprobe_trampoline: \n"
266 * Called when we hit the probe point at kretprobe_trampoline
268 int __kprobes trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
270 struct kretprobe_instance *ri = NULL;
271 struct hlist_head *head;
272 struct hlist_node *node, *tmp;
273 unsigned long flags, orig_ret_address = 0;
274 unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
276 spin_lock_irqsave(&kretprobe_lock, flags);
277 head = kretprobe_inst_table_head(current);
280 * It is possible to have multiple instances associated with a given
281 * task either because an multiple functions in the call path
282 * have a return probe installed on them, and/or more then one return
283 * return probe was registered for a target function.
285 * We can handle this because:
286 * - instances are always inserted at the head of the list
287 * - when multiple return probes are registered for the same
288 * function, the first instance's ret_addr will point to the
289 * real return address, and all the rest will point to
290 * kretprobe_trampoline
292 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
293 if (ri->task != current)
294 /* another task is sharing our hash bucket */
297 if (ri->rp && ri->rp->handler)
298 ri->rp->handler(ri, regs);
300 orig_ret_address = (unsigned long)ri->ret_addr;
303 if (orig_ret_address != trampoline_address)
305 * This is the real return address. Any other
306 * instances associated with this task are for
307 * other calls deeper on the call stack
312 BUG_ON(!orig_ret_address || (orig_ret_address == trampoline_address));
313 regs->eip = orig_ret_address;
315 reset_current_kprobe();
316 spin_unlock_irqrestore(&kretprobe_lock, flags);
317 preempt_enable_no_resched();
320 * By returning a non-zero value, we are telling
321 * kprobe_handler() that we don't want the post_handler
322 * to run (and have re-enabled preemption)
328 * Called after single-stepping. p->addr is the address of the
329 * instruction whose first byte has been replaced by the "int 3"
330 * instruction. To avoid the SMP problems that can occur when we
331 * temporarily put back the original opcode to single-step, we
332 * single-stepped a copy of the instruction. The address of this
333 * copy is p->ainsn.insn.
335 * This function prepares to return from the post-single-step
336 * interrupt. We have to fix up the stack as follows:
338 * 0) Except in the case of absolute or indirect jump or call instructions,
339 * the new eip is relative to the copied instruction. We need to make
340 * it relative to the original instruction.
342 * 1) If the single-stepped instruction was pushfl, then the TF and IF
343 * flags are set in the just-pushed eflags, and may need to be cleared.
345 * 2) If the single-stepped instruction was a call, the return address
346 * that is atop the stack is the address following the copied instruction.
347 * We need to make it the address following the original instruction.
349 static void __kprobes resume_execution(struct kprobe *p,
350 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
352 unsigned long *tos = (unsigned long *)®s->esp;
353 unsigned long next_eip = 0;
354 unsigned long copy_eip = (unsigned long)&p->ainsn.insn;
355 unsigned long orig_eip = (unsigned long)p->addr;
357 switch (p->ainsn.insn[0]) {
358 case 0x9c: /* pushfl */
359 *tos &= ~(TF_MASK | IF_MASK);
360 *tos |= kcb->kprobe_old_eflags;
362 case 0xc3: /* ret/lret */
366 regs->eflags &= ~TF_MASK;
367 /* eip is already adjusted, no more changes required*/
369 case 0xe8: /* call relative - Fix return addr */
370 *tos = orig_eip + (*tos - copy_eip);
373 if ((p->ainsn.insn[1] & 0x30) == 0x10) {
374 /* call absolute, indirect */
375 /* Fix return addr; eip is correct. */
376 next_eip = regs->eip;
377 *tos = orig_eip + (*tos - copy_eip);
378 } else if (((p->ainsn.insn[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
379 ((p->ainsn.insn[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
380 /* eip is correct. */
381 next_eip = regs->eip;
384 case 0xea: /* jmp absolute -- eip is correct */
385 next_eip = regs->eip;
391 regs->eflags &= ~TF_MASK;
393 regs->eip = next_eip;
395 regs->eip = orig_eip + (regs->eip - copy_eip);
400 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
401 * remain disabled thoroughout this function.
403 static inline int post_kprobe_handler(struct pt_regs *regs)
405 struct kprobe *cur = kprobe_running();
406 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
411 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
412 kcb->kprobe_status = KPROBE_HIT_SSDONE;
413 cur->post_handler(cur, regs, 0);
416 resume_execution(cur, regs, kcb);
417 regs->eflags |= kcb->kprobe_saved_eflags;
419 /*Restore back the original saved kprobes variables and continue. */
420 if (kcb->kprobe_status == KPROBE_REENTER) {
421 restore_previous_kprobe(kcb);
424 reset_current_kprobe();
426 preempt_enable_no_resched();
429 * if somebody else is singlestepping across a probe point, eflags
430 * will have TF set, in which case, continue the remaining processing
431 * of do_debug, as if this is not a probe hit.
433 if (regs->eflags & TF_MASK)
439 static inline int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
441 struct kprobe *cur = kprobe_running();
442 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
444 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
447 if (kcb->kprobe_status & KPROBE_HIT_SS) {
448 resume_execution(cur, regs, kcb);
449 regs->eflags |= kcb->kprobe_old_eflags;
451 reset_current_kprobe();
452 preempt_enable_no_resched();
458 * Wrapper routine to for handling exceptions.
460 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
461 unsigned long val, void *data)
463 struct die_args *args = (struct die_args *)data;
464 int ret = NOTIFY_DONE;
468 if (kprobe_handler(args->regs))
472 if (post_kprobe_handler(args->regs))
477 /* kprobe_running() needs smp_processor_id() */
479 if (kprobe_running() &&
480 kprobe_fault_handler(args->regs, args->trapnr))
490 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
492 struct jprobe *jp = container_of(p, struct jprobe, kp);
494 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
496 kcb->jprobe_saved_regs = *regs;
497 kcb->jprobe_saved_esp = ®s->esp;
498 addr = (unsigned long)(kcb->jprobe_saved_esp);
501 * TBD: As Linus pointed out, gcc assumes that the callee
502 * owns the argument space and could overwrite it, e.g.
503 * tailcall optimization. So, to be absolutely safe
504 * we also save and restore enough stack bytes to cover
507 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
508 MIN_STACK_SIZE(addr));
509 regs->eflags &= ~IF_MASK;
510 regs->eip = (unsigned long)(jp->entry);
514 void __kprobes jprobe_return(void)
516 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
518 asm volatile (" xchgl %%ebx,%%esp \n"
520 " .globl jprobe_return_end \n"
521 " jprobe_return_end: \n"
523 (kcb->jprobe_saved_esp):"memory");
526 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
528 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
529 u8 *addr = (u8 *) (regs->eip - 1);
530 unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_esp);
531 struct jprobe *jp = container_of(p, struct jprobe, kp);
533 if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
534 if (®s->esp != kcb->jprobe_saved_esp) {
535 struct pt_regs *saved_regs =
536 container_of(kcb->jprobe_saved_esp,
537 struct pt_regs, esp);
538 printk("current esp %p does not match saved esp %p\n",
539 ®s->esp, kcb->jprobe_saved_esp);
540 printk("Saved registers for jprobe %p\n", jp);
541 show_registers(saved_regs);
542 printk("Current registers\n");
543 show_registers(regs);
546 *regs = kcb->jprobe_saved_regs;
547 memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
548 MIN_STACK_SIZE(stack_addr));
549 preempt_enable_no_resched();
555 static struct kprobe trampoline_p = {
556 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
557 .pre_handler = trampoline_probe_handler
560 int __init arch_init_kprobes(void)
562 return register_kprobe(&trampoline_p);