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 p = __get_cpu_var(current_kprobe);
192 if (p->break_handler && p->break_handler(p, regs)) {
199 p = get_kprobe(addr);
201 if (regs->eflags & VM_MASK) {
202 /* We are in virtual-8086 mode. Return 0 */
206 if (*addr != BREAKPOINT_INSTRUCTION) {
208 * The breakpoint instruction was removed right
209 * after we hit it. Another cpu has removed
210 * either a probepoint or a debugger breakpoint
211 * at this address. In either case, no further
212 * handling of this interrupt is appropriate.
213 * Back up over the (now missing) int3 and run
214 * the original instruction.
216 regs->eip -= sizeof(kprobe_opcode_t);
219 /* Not one of ours: let kernel handle it */
223 set_current_kprobe(p, regs, kcb);
224 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
226 if (p->pre_handler && p->pre_handler(p, regs))
227 /* handler has already set things up, so skip ss setup */
231 prepare_singlestep(p, regs);
232 kcb->kprobe_status = KPROBE_HIT_SS;
236 preempt_enable_no_resched();
241 * For function-return probes, init_kprobes() establishes a probepoint
242 * here. When a retprobed function returns, this probe is hit and
243 * trampoline_probe_handler() runs, calling the kretprobe's handler.
245 void kretprobe_trampoline_holder(void)
247 asm volatile ( ".global kretprobe_trampoline\n"
248 "kretprobe_trampoline: \n"
253 * Called when we hit the probe point at kretprobe_trampoline
255 int __kprobes trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
257 struct kretprobe_instance *ri = NULL;
258 struct hlist_head *head;
259 struct hlist_node *node, *tmp;
260 unsigned long flags, orig_ret_address = 0;
261 unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
263 spin_lock_irqsave(&kretprobe_lock, flags);
264 head = kretprobe_inst_table_head(current);
267 * It is possible to have multiple instances associated with a given
268 * task either because an multiple functions in the call path
269 * have a return probe installed on them, and/or more then one return
270 * return probe was registered for a target function.
272 * We can handle this because:
273 * - instances are always inserted at the head of the list
274 * - when multiple return probes are registered for the same
275 * function, the first instance's ret_addr will point to the
276 * real return address, and all the rest will point to
277 * kretprobe_trampoline
279 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
280 if (ri->task != current)
281 /* another task is sharing our hash bucket */
284 if (ri->rp && ri->rp->handler)
285 ri->rp->handler(ri, regs);
287 orig_ret_address = (unsigned long)ri->ret_addr;
290 if (orig_ret_address != trampoline_address)
292 * This is the real return address. Any other
293 * instances associated with this task are for
294 * other calls deeper on the call stack
299 BUG_ON(!orig_ret_address || (orig_ret_address == trampoline_address));
300 regs->eip = orig_ret_address;
302 reset_current_kprobe();
303 spin_unlock_irqrestore(&kretprobe_lock, flags);
304 preempt_enable_no_resched();
307 * By returning a non-zero value, we are telling
308 * kprobe_handler() that we don't want the post_handler
309 * to run (and have re-enabled preemption)
315 * Called after single-stepping. p->addr is the address of the
316 * instruction whose first byte has been replaced by the "int 3"
317 * instruction. To avoid the SMP problems that can occur when we
318 * temporarily put back the original opcode to single-step, we
319 * single-stepped a copy of the instruction. The address of this
320 * copy is p->ainsn.insn.
322 * This function prepares to return from the post-single-step
323 * interrupt. We have to fix up the stack as follows:
325 * 0) Except in the case of absolute or indirect jump or call instructions,
326 * the new eip is relative to the copied instruction. We need to make
327 * it relative to the original instruction.
329 * 1) If the single-stepped instruction was pushfl, then the TF and IF
330 * flags are set in the just-pushed eflags, and may need to be cleared.
332 * 2) If the single-stepped instruction was a call, the return address
333 * that is atop the stack is the address following the copied instruction.
334 * We need to make it the address following the original instruction.
336 static void __kprobes resume_execution(struct kprobe *p,
337 struct pt_regs *regs, struct kprobe_ctlblk *kcb)
339 unsigned long *tos = (unsigned long *)®s->esp;
340 unsigned long next_eip = 0;
341 unsigned long copy_eip = (unsigned long)&p->ainsn.insn;
342 unsigned long orig_eip = (unsigned long)p->addr;
344 switch (p->ainsn.insn[0]) {
345 case 0x9c: /* pushfl */
346 *tos &= ~(TF_MASK | IF_MASK);
347 *tos |= kcb->kprobe_old_eflags;
349 case 0xc3: /* ret/lret */
353 regs->eflags &= ~TF_MASK;
354 /* eip is already adjusted, no more changes required*/
356 case 0xe8: /* call relative - Fix return addr */
357 *tos = orig_eip + (*tos - copy_eip);
360 if ((p->ainsn.insn[1] & 0x30) == 0x10) {
361 /* call absolute, indirect */
362 /* Fix return addr; eip is correct. */
363 next_eip = regs->eip;
364 *tos = orig_eip + (*tos - copy_eip);
365 } else if (((p->ainsn.insn[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
366 ((p->ainsn.insn[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
367 /* eip is correct. */
368 next_eip = regs->eip;
371 case 0xea: /* jmp absolute -- eip is correct */
372 next_eip = regs->eip;
378 regs->eflags &= ~TF_MASK;
380 regs->eip = next_eip;
382 regs->eip = orig_eip + (regs->eip - copy_eip);
387 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
388 * remain disabled thoroughout this function.
390 static inline int post_kprobe_handler(struct pt_regs *regs)
392 struct kprobe *cur = kprobe_running();
393 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
398 if ((kcb->kprobe_status != KPROBE_REENTER) && cur->post_handler) {
399 kcb->kprobe_status = KPROBE_HIT_SSDONE;
400 cur->post_handler(cur, regs, 0);
403 resume_execution(cur, regs, kcb);
404 regs->eflags |= kcb->kprobe_saved_eflags;
406 /*Restore back the original saved kprobes variables and continue. */
407 if (kcb->kprobe_status == KPROBE_REENTER) {
408 restore_previous_kprobe(kcb);
411 reset_current_kprobe();
413 preempt_enable_no_resched();
416 * if somebody else is singlestepping across a probe point, eflags
417 * will have TF set, in which case, continue the remaining processing
418 * of do_debug, as if this is not a probe hit.
420 if (regs->eflags & TF_MASK)
426 static inline int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
428 struct kprobe *cur = kprobe_running();
429 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
431 if (cur->fault_handler && cur->fault_handler(cur, regs, trapnr))
434 if (kcb->kprobe_status & KPROBE_HIT_SS) {
435 resume_execution(cur, regs, kcb);
436 regs->eflags |= kcb->kprobe_old_eflags;
438 reset_current_kprobe();
439 preempt_enable_no_resched();
445 * Wrapper routine to for handling exceptions.
447 int __kprobes kprobe_exceptions_notify(struct notifier_block *self,
448 unsigned long val, void *data)
450 struct die_args *args = (struct die_args *)data;
451 int ret = NOTIFY_DONE;
455 if (kprobe_handler(args->regs))
459 if (post_kprobe_handler(args->regs))
464 /* kprobe_running() needs smp_processor_id() */
466 if (kprobe_running() &&
467 kprobe_fault_handler(args->regs, args->trapnr))
477 int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
479 struct jprobe *jp = container_of(p, struct jprobe, kp);
481 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
483 kcb->jprobe_saved_regs = *regs;
484 kcb->jprobe_saved_esp = ®s->esp;
485 addr = (unsigned long)(kcb->jprobe_saved_esp);
488 * TBD: As Linus pointed out, gcc assumes that the callee
489 * owns the argument space and could overwrite it, e.g.
490 * tailcall optimization. So, to be absolutely safe
491 * we also save and restore enough stack bytes to cover
494 memcpy(kcb->jprobes_stack, (kprobe_opcode_t *)addr,
495 MIN_STACK_SIZE(addr));
496 regs->eflags &= ~IF_MASK;
497 regs->eip = (unsigned long)(jp->entry);
501 void __kprobes jprobe_return(void)
503 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
505 asm volatile (" xchgl %%ebx,%%esp \n"
507 " .globl jprobe_return_end \n"
508 " jprobe_return_end: \n"
510 (kcb->jprobe_saved_esp):"memory");
513 int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
515 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
516 u8 *addr = (u8 *) (regs->eip - 1);
517 unsigned long stack_addr = (unsigned long)(kcb->jprobe_saved_esp);
518 struct jprobe *jp = container_of(p, struct jprobe, kp);
520 if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
521 if (®s->esp != kcb->jprobe_saved_esp) {
522 struct pt_regs *saved_regs =
523 container_of(kcb->jprobe_saved_esp,
524 struct pt_regs, esp);
525 printk("current esp %p does not match saved esp %p\n",
526 ®s->esp, kcb->jprobe_saved_esp);
527 printk("Saved registers for jprobe %p\n", jp);
528 show_registers(saved_regs);
529 printk("Current registers\n");
530 show_registers(regs);
533 *regs = kcb->jprobe_saved_regs;
534 memcpy((kprobe_opcode_t *) stack_addr, kcb->jprobes_stack,
535 MIN_STACK_SIZE(stack_addr));
536 preempt_enable_no_resched();
542 static struct kprobe trampoline_p = {
543 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
544 .pre_handler = trampoline_probe_handler
547 int __init arch_init_kprobes(void)
549 return register_kprobe(&trampoline_p);