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/spinlock.h>
35 #include <linux/preempt.h>
36 #include <asm/kdebug.h>
39 /* kprobe_status settings */
40 #define KPROBE_HIT_ACTIVE 0x00000001
41 #define KPROBE_HIT_SS 0x00000002
43 static struct kprobe
*current_kprobe
;
44 static unsigned long kprobe_status
, kprobe_old_eflags
, kprobe_saved_eflags
;
45 static struct pt_regs jprobe_saved_regs
;
46 static long *jprobe_saved_esp
;
47 /* copy of the kernel stack at the probe fire time */
48 static kprobe_opcode_t jprobes_stack
[MAX_STACK_SIZE
];
49 void jprobe_return_end(void);
52 * returns non-zero if opcode modifies the interrupt flag.
54 static inline int is_IF_modifier(kprobe_opcode_t opcode
)
59 case 0xcf: /* iret/iretd */
60 case 0x9d: /* popf/popfd */
66 int arch_prepare_kprobe(struct kprobe
*p
)
71 void arch_copy_kprobe(struct kprobe
*p
)
73 memcpy(p
->ainsn
.insn
, p
->addr
, MAX_INSN_SIZE
* sizeof(kprobe_opcode_t
));
76 void arch_remove_kprobe(struct kprobe
*p
)
80 static inline void disarm_kprobe(struct kprobe
*p
, struct pt_regs
*regs
)
83 regs
->eip
= (unsigned long)p
->addr
;
86 static inline void prepare_singlestep(struct kprobe
*p
, struct pt_regs
*regs
)
88 regs
->eflags
|= TF_MASK
;
89 regs
->eflags
&= ~IF_MASK
;
90 /*single step inline if the instruction is an int3*/
91 if (p
->opcode
== BREAKPOINT_INSTRUCTION
)
92 regs
->eip
= (unsigned long)p
->addr
;
94 regs
->eip
= (unsigned long)&p
->ainsn
.insn
;
97 struct task_struct
*arch_get_kprobe_task(void *ptr
)
99 return ((struct thread_info
*) (((unsigned long) ptr
) &
100 (~(THREAD_SIZE
-1))))->task
;
103 void arch_prepare_kretprobe(struct kretprobe
*rp
, struct pt_regs
*regs
)
105 unsigned long *sara
= (unsigned long *)®s
->esp
;
106 struct kretprobe_instance
*ri
;
107 static void *orig_ret_addr
;
110 * Save the return address when the return probe hits
111 * the first time, and use it to populate the (krprobe
112 * instance)->ret_addr for subsequent return probes at
113 * the same addrress since stack address would have
114 * the kretprobe_trampoline by then.
116 if (((void*) *sara
) != kretprobe_trampoline
)
117 orig_ret_addr
= (void*) *sara
;
119 if ((ri
= get_free_rp_inst(rp
)) != NULL
) {
121 ri
->stack_addr
= sara
;
122 ri
->ret_addr
= orig_ret_addr
;
124 /* Replace the return addr with trampoline addr */
125 *sara
= (unsigned long) &kretprobe_trampoline
;
131 void arch_kprobe_flush_task(struct task_struct
*tk
, spinlock_t
*kp_lock
)
133 unsigned long flags
= 0;
134 struct kretprobe_instance
*ri
;
135 spin_lock_irqsave(kp_lock
, flags
);
136 while ((ri
= get_rp_inst_tsk(tk
)) != NULL
) {
137 *((unsigned long *)(ri
->stack_addr
)) =
138 (unsigned long) ri
->ret_addr
;
141 spin_unlock_irqrestore(kp_lock
, flags
);
145 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
146 * remain disabled thorough out this function.
148 static int kprobe_handler(struct pt_regs
*regs
)
152 kprobe_opcode_t
*addr
= NULL
;
155 /* We're in an interrupt, but this is clear and BUG()-safe. */
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 /* We *are* holding lock here, so this is safe.
171 Disarm the probe we just hit, and ignore it. */
172 p
= get_kprobe(addr
);
174 if (kprobe_status
== KPROBE_HIT_SS
) {
175 regs
->eflags
&= ~TF_MASK
;
176 regs
->eflags
|= kprobe_saved_eflags
;
180 disarm_kprobe(p
, regs
);
184 if (p
->break_handler
&& p
->break_handler(p
, regs
)) {
188 /* If it's not ours, can't be delete race, (we hold lock). */
193 p
= get_kprobe(addr
);
196 if (regs
->eflags
& VM_MASK
) {
197 /* We are in virtual-8086 mode. Return 0 */
201 if (*addr
!= BREAKPOINT_INSTRUCTION
) {
203 * The breakpoint instruction was removed right
204 * after we hit it. Another cpu has removed
205 * either a probepoint or a debugger breakpoint
206 * at this address. In either case, no further
207 * handling of this interrupt is appropriate.
211 /* Not one of ours: let kernel handle it */
215 kprobe_status
= KPROBE_HIT_ACTIVE
;
217 kprobe_saved_eflags
= kprobe_old_eflags
218 = (regs
->eflags
& (TF_MASK
| IF_MASK
));
219 if (is_IF_modifier(p
->opcode
))
220 kprobe_saved_eflags
&= ~IF_MASK
;
222 if (p
->pre_handler
&& p
->pre_handler(p
, regs
))
223 /* handler has already set things up, so skip ss setup */
227 prepare_singlestep(p
, regs
);
228 kprobe_status
= KPROBE_HIT_SS
;
232 preempt_enable_no_resched();
237 * For function-return probes, init_kprobes() establishes a probepoint
238 * here. When a retprobed function returns, this probe is hit and
239 * trampoline_probe_handler() runs, calling the kretprobe's handler.
241 void kretprobe_trampoline_holder(void)
243 asm volatile ( ".global kretprobe_trampoline\n"
244 "kretprobe_trampoline: \n"
249 * Called when we hit the probe point at kretprobe_trampoline
251 int trampoline_probe_handler(struct kprobe
*p
, struct pt_regs
*regs
)
253 struct task_struct
*tsk
;
254 struct kretprobe_instance
*ri
;
255 struct hlist_head
*head
;
256 struct hlist_node
*node
;
257 unsigned long *sara
= ((unsigned long *) ®s
->esp
) - 1;
259 tsk
= arch_get_kprobe_task(sara
);
260 head
= kretprobe_inst_table_head(tsk
);
262 hlist_for_each_entry(ri
, node
, head
, hlist
) {
263 if (ri
->stack_addr
== sara
&& ri
->rp
) {
265 ri
->rp
->handler(ri
, regs
);
271 void trampoline_post_handler(struct kprobe
*p
, struct pt_regs
*regs
,
274 struct kretprobe_instance
*ri
;
275 /* RA already popped */
276 unsigned long *sara
= ((unsigned long *)®s
->esp
) - 1;
278 while ((ri
= get_rp_inst(sara
))) {
279 regs
->eip
= (unsigned long)ri
->ret_addr
;
282 regs
->eflags
&= ~TF_MASK
;
286 * Called after single-stepping. p->addr is the address of the
287 * instruction whose first byte has been replaced by the "int 3"
288 * instruction. To avoid the SMP problems that can occur when we
289 * temporarily put back the original opcode to single-step, we
290 * single-stepped a copy of the instruction. The address of this
291 * copy is p->ainsn.insn.
293 * This function prepares to return from the post-single-step
294 * interrupt. We have to fix up the stack as follows:
296 * 0) Except in the case of absolute or indirect jump or call instructions,
297 * the new eip is relative to the copied instruction. We need to make
298 * it relative to the original instruction.
300 * 1) If the single-stepped instruction was pushfl, then the TF and IF
301 * flags are set in the just-pushed eflags, and may need to be cleared.
303 * 2) If the single-stepped instruction was a call, the return address
304 * that is atop the stack is the address following the copied instruction.
305 * We need to make it the address following the original instruction.
307 static void resume_execution(struct kprobe
*p
, struct pt_regs
*regs
)
309 unsigned long *tos
= (unsigned long *)®s
->esp
;
310 unsigned long next_eip
= 0;
311 unsigned long copy_eip
= (unsigned long)&p
->ainsn
.insn
;
312 unsigned long orig_eip
= (unsigned long)p
->addr
;
314 switch (p
->ainsn
.insn
[0]) {
315 case 0x9c: /* pushfl */
316 *tos
&= ~(TF_MASK
| IF_MASK
);
317 *tos
|= kprobe_old_eflags
;
319 case 0xc3: /* ret/lret */
323 regs
->eflags
&= ~TF_MASK
;
324 /* eip is already adjusted, no more changes required*/
326 case 0xe8: /* call relative - Fix return addr */
327 *tos
= orig_eip
+ (*tos
- copy_eip
);
330 if ((p
->ainsn
.insn
[1] & 0x30) == 0x10) {
331 /* call absolute, indirect */
332 /* Fix return addr; eip is correct. */
333 next_eip
= regs
->eip
;
334 *tos
= orig_eip
+ (*tos
- copy_eip
);
335 } else if (((p
->ainsn
.insn
[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
336 ((p
->ainsn
.insn
[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
337 /* eip is correct. */
338 next_eip
= regs
->eip
;
341 case 0xea: /* jmp absolute -- eip is correct */
342 next_eip
= regs
->eip
;
348 regs
->eflags
&= ~TF_MASK
;
350 regs
->eip
= next_eip
;
352 regs
->eip
= orig_eip
+ (regs
->eip
- copy_eip
);
357 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
358 * remain disabled thoroughout this function. And we hold kprobe lock.
360 static inline int post_kprobe_handler(struct pt_regs
*regs
)
362 if (!kprobe_running())
365 if (current_kprobe
->post_handler
)
366 current_kprobe
->post_handler(current_kprobe
, regs
, 0);
368 if (current_kprobe
->post_handler
!= trampoline_post_handler
)
369 resume_execution(current_kprobe
, regs
);
370 regs
->eflags
|= kprobe_saved_eflags
;
373 preempt_enable_no_resched();
376 * if somebody else is singlestepping across a probe point, eflags
377 * will have TF set, in which case, continue the remaining processing
378 * of do_debug, as if this is not a probe hit.
380 if (regs
->eflags
& TF_MASK
)
386 /* Interrupts disabled, kprobe_lock held. */
387 static inline int kprobe_fault_handler(struct pt_regs
*regs
, int trapnr
)
389 if (current_kprobe
->fault_handler
390 && current_kprobe
->fault_handler(current_kprobe
, regs
, trapnr
))
393 if (kprobe_status
& KPROBE_HIT_SS
) {
394 resume_execution(current_kprobe
, regs
);
395 regs
->eflags
|= kprobe_old_eflags
;
398 preempt_enable_no_resched();
404 * Wrapper routine to for handling exceptions.
406 int kprobe_exceptions_notify(struct notifier_block
*self
, unsigned long val
,
409 struct die_args
*args
= (struct die_args
*)data
;
412 if (kprobe_handler(args
->regs
))
416 if (post_kprobe_handler(args
->regs
))
420 if (kprobe_running() &&
421 kprobe_fault_handler(args
->regs
, args
->trapnr
))
425 if (kprobe_running() &&
426 kprobe_fault_handler(args
->regs
, args
->trapnr
))
435 int setjmp_pre_handler(struct kprobe
*p
, struct pt_regs
*regs
)
437 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
440 jprobe_saved_regs
= *regs
;
441 jprobe_saved_esp
= ®s
->esp
;
442 addr
= (unsigned long)jprobe_saved_esp
;
445 * TBD: As Linus pointed out, gcc assumes that the callee
446 * owns the argument space and could overwrite it, e.g.
447 * tailcall optimization. So, to be absolutely safe
448 * we also save and restore enough stack bytes to cover
451 memcpy(jprobes_stack
, (kprobe_opcode_t
*) addr
, MIN_STACK_SIZE(addr
));
452 regs
->eflags
&= ~IF_MASK
;
453 regs
->eip
= (unsigned long)(jp
->entry
);
457 void jprobe_return(void)
459 preempt_enable_no_resched();
460 asm volatile (" xchgl %%ebx,%%esp \n"
462 " .globl jprobe_return_end \n"
463 " jprobe_return_end: \n"
465 (jprobe_saved_esp
):"memory");
468 int longjmp_break_handler(struct kprobe
*p
, struct pt_regs
*regs
)
470 u8
*addr
= (u8
*) (regs
->eip
- 1);
471 unsigned long stack_addr
= (unsigned long)jprobe_saved_esp
;
472 struct jprobe
*jp
= container_of(p
, struct jprobe
, kp
);
474 if ((addr
> (u8
*) jprobe_return
) && (addr
< (u8
*) jprobe_return_end
)) {
475 if (®s
->esp
!= jprobe_saved_esp
) {
476 struct pt_regs
*saved_regs
=
477 container_of(jprobe_saved_esp
, struct pt_regs
, esp
);
478 printk("current esp %p does not match saved esp %p\n",
479 ®s
->esp
, jprobe_saved_esp
);
480 printk("Saved registers for jprobe %p\n", jp
);
481 show_registers(saved_regs
);
482 printk("Current registers\n");
483 show_registers(regs
);
486 *regs
= jprobe_saved_regs
;
487 memcpy((kprobe_opcode_t
*) stack_addr
, jprobes_stack
,
488 MIN_STACK_SIZE(stack_addr
));