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/cacheflush.h>
37 #include <asm/kdebug.h>
40 static struct kprobe *current_kprobe;
41 static unsigned long kprobe_status, kprobe_old_eflags, kprobe_saved_eflags;
42 static struct kprobe *kprobe_prev;
43 static unsigned long kprobe_status_prev, kprobe_old_eflags_prev, kprobe_saved_eflags_prev;
44 static struct pt_regs jprobe_saved_regs;
45 static long *jprobe_saved_esp;
46 /* copy of the kernel stack at the probe fire time */
47 static kprobe_opcode_t jprobes_stack[MAX_STACK_SIZE];
48 void jprobe_return_end(void);
51 * returns non-zero if opcode modifies the interrupt flag.
53 static inline int is_IF_modifier(kprobe_opcode_t opcode)
58 case 0xcf: /* iret/iretd */
59 case 0x9d: /* popf/popfd */
65 int arch_prepare_kprobe(struct kprobe *p)
70 void arch_copy_kprobe(struct kprobe *p)
72 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
76 void arch_arm_kprobe(struct kprobe *p)
78 *p->addr = BREAKPOINT_INSTRUCTION;
79 flush_icache_range((unsigned long) p->addr,
80 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
83 void arch_disarm_kprobe(struct kprobe *p)
86 flush_icache_range((unsigned long) p->addr,
87 (unsigned long) p->addr + sizeof(kprobe_opcode_t));
90 void arch_remove_kprobe(struct kprobe *p)
94 static inline void save_previous_kprobe(void)
96 kprobe_prev = current_kprobe;
97 kprobe_status_prev = kprobe_status;
98 kprobe_old_eflags_prev = kprobe_old_eflags;
99 kprobe_saved_eflags_prev = kprobe_saved_eflags;
102 static inline void restore_previous_kprobe(void)
104 current_kprobe = kprobe_prev;
105 kprobe_status = kprobe_status_prev;
106 kprobe_old_eflags = kprobe_old_eflags_prev;
107 kprobe_saved_eflags = kprobe_saved_eflags_prev;
110 static inline void set_current_kprobe(struct kprobe *p, struct pt_regs *regs)
113 kprobe_saved_eflags = kprobe_old_eflags
114 = (regs->eflags & (TF_MASK | IF_MASK));
115 if (is_IF_modifier(p->opcode))
116 kprobe_saved_eflags &= ~IF_MASK;
119 static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
121 regs->eflags |= TF_MASK;
122 regs->eflags &= ~IF_MASK;
123 /*single step inline if the instruction is an int3*/
124 if (p->opcode == BREAKPOINT_INSTRUCTION)
125 regs->eip = (unsigned long)p->addr;
127 regs->eip = (unsigned long)&p->ainsn.insn;
130 void arch_prepare_kretprobe(struct kretprobe *rp, struct pt_regs *regs)
132 unsigned long *sara = (unsigned long *)®s->esp;
133 struct kretprobe_instance *ri;
135 if ((ri = get_free_rp_inst(rp)) != NULL) {
138 ri->ret_addr = (kprobe_opcode_t *) *sara;
140 /* Replace the return addr with trampoline addr */
141 *sara = (unsigned long) &kretprobe_trampoline;
150 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
151 * remain disabled thorough out this function.
153 static int kprobe_handler(struct pt_regs *regs)
157 kprobe_opcode_t *addr = NULL;
160 /* We're in an interrupt, but this is clear and BUG()-safe. */
162 /* Check if the application is using LDT entry for its code segment and
163 * calculate the address by reading the base address from the LDT entry.
165 if ((regs->xcs & 4) && (current->mm)) {
166 lp = (unsigned long *) ((unsigned long)((regs->xcs >> 3) * 8)
167 + (char *) current->mm->context.ldt);
168 addr = (kprobe_opcode_t *) (get_desc_base(lp) + regs->eip -
169 sizeof(kprobe_opcode_t));
171 addr = (kprobe_opcode_t *)(regs->eip - sizeof(kprobe_opcode_t));
173 /* Check we're not actually recursing */
174 if (kprobe_running()) {
175 /* We *are* holding lock here, so this is safe.
176 Disarm the probe we just hit, and ignore it. */
177 p = get_kprobe(addr);
179 if (kprobe_status == KPROBE_HIT_SS) {
180 regs->eflags &= ~TF_MASK;
181 regs->eflags |= kprobe_saved_eflags;
185 /* We have reentered the kprobe_handler(), since
186 * another probe was hit while within the handler.
187 * We here save the original kprobes variables and
188 * just single step on the instruction of the new probe
189 * without calling any user handlers.
191 save_previous_kprobe();
192 set_current_kprobe(p, regs);
194 prepare_singlestep(p, regs);
195 kprobe_status = KPROBE_REENTER;
199 if (p->break_handler && p->break_handler(p, regs)) {
203 /* If it's not ours, can't be delete race, (we hold lock). */
208 p = get_kprobe(addr);
211 if (regs->eflags & VM_MASK) {
212 /* We are in virtual-8086 mode. Return 0 */
216 if (*addr != BREAKPOINT_INSTRUCTION) {
218 * The breakpoint instruction was removed right
219 * after we hit it. Another cpu has removed
220 * either a probepoint or a debugger breakpoint
221 * at this address. In either case, no further
222 * handling of this interrupt is appropriate.
226 /* Not one of ours: let kernel handle it */
230 kprobe_status = KPROBE_HIT_ACTIVE;
231 set_current_kprobe(p, regs);
233 if (p->pre_handler && p->pre_handler(p, regs))
234 /* handler has already set things up, so skip ss setup */
238 prepare_singlestep(p, regs);
239 kprobe_status = KPROBE_HIT_SS;
243 preempt_enable_no_resched();
248 * For function-return probes, init_kprobes() establishes a probepoint
249 * here. When a retprobed function returns, this probe is hit and
250 * trampoline_probe_handler() runs, calling the kretprobe's handler.
252 void kretprobe_trampoline_holder(void)
254 asm volatile ( ".global kretprobe_trampoline\n"
255 "kretprobe_trampoline: \n"
260 * Called when we hit the probe point at kretprobe_trampoline
262 int trampoline_probe_handler(struct kprobe *p, struct pt_regs *regs)
264 struct kretprobe_instance *ri = NULL;
265 struct hlist_head *head;
266 struct hlist_node *node, *tmp;
267 unsigned long orig_ret_address = 0;
268 unsigned long trampoline_address =(unsigned long)&kretprobe_trampoline;
270 head = kretprobe_inst_table_head(current);
273 * It is possible to have multiple instances associated with a given
274 * task either because an multiple functions in the call path
275 * have a return probe installed on them, and/or more then one return
276 * return probe was registered for a target function.
278 * We can handle this because:
279 * - instances are always inserted at the head of the list
280 * - when multiple return probes are registered for the same
281 * function, the first instance's ret_addr will point to the
282 * real return address, and all the rest will point to
283 * kretprobe_trampoline
285 hlist_for_each_entry_safe(ri, node, tmp, head, hlist) {
286 if (ri->task != current)
287 /* another task is sharing our hash bucket */
290 if (ri->rp && ri->rp->handler)
291 ri->rp->handler(ri, regs);
293 orig_ret_address = (unsigned long)ri->ret_addr;
296 if (orig_ret_address != trampoline_address)
298 * This is the real return address. Any other
299 * instances associated with this task are for
300 * other calls deeper on the call stack
305 BUG_ON(!orig_ret_address || (orig_ret_address == trampoline_address));
306 regs->eip = orig_ret_address;
309 preempt_enable_no_resched();
312 * By returning a non-zero value, we are telling
313 * kprobe_handler() that we have handled unlocking
314 * and re-enabling preemption.
320 * Called after single-stepping. p->addr is the address of the
321 * instruction whose first byte has been replaced by the "int 3"
322 * instruction. To avoid the SMP problems that can occur when we
323 * temporarily put back the original opcode to single-step, we
324 * single-stepped a copy of the instruction. The address of this
325 * copy is p->ainsn.insn.
327 * This function prepares to return from the post-single-step
328 * interrupt. We have to fix up the stack as follows:
330 * 0) Except in the case of absolute or indirect jump or call instructions,
331 * the new eip is relative to the copied instruction. We need to make
332 * it relative to the original instruction.
334 * 1) If the single-stepped instruction was pushfl, then the TF and IF
335 * flags are set in the just-pushed eflags, and may need to be cleared.
337 * 2) If the single-stepped instruction was a call, the return address
338 * that is atop the stack is the address following the copied instruction.
339 * We need to make it the address following the original instruction.
341 static void resume_execution(struct kprobe *p, struct pt_regs *regs)
343 unsigned long *tos = (unsigned long *)®s->esp;
344 unsigned long next_eip = 0;
345 unsigned long copy_eip = (unsigned long)&p->ainsn.insn;
346 unsigned long orig_eip = (unsigned long)p->addr;
348 switch (p->ainsn.insn[0]) {
349 case 0x9c: /* pushfl */
350 *tos &= ~(TF_MASK | IF_MASK);
351 *tos |= kprobe_old_eflags;
353 case 0xc3: /* ret/lret */
357 regs->eflags &= ~TF_MASK;
358 /* eip is already adjusted, no more changes required*/
360 case 0xe8: /* call relative - Fix return addr */
361 *tos = orig_eip + (*tos - copy_eip);
364 if ((p->ainsn.insn[1] & 0x30) == 0x10) {
365 /* call absolute, indirect */
366 /* Fix return addr; eip is correct. */
367 next_eip = regs->eip;
368 *tos = orig_eip + (*tos - copy_eip);
369 } else if (((p->ainsn.insn[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
370 ((p->ainsn.insn[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
371 /* eip is correct. */
372 next_eip = regs->eip;
375 case 0xea: /* jmp absolute -- eip is correct */
376 next_eip = regs->eip;
382 regs->eflags &= ~TF_MASK;
384 regs->eip = next_eip;
386 regs->eip = orig_eip + (regs->eip - copy_eip);
391 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
392 * remain disabled thoroughout this function. And we hold kprobe lock.
394 static inline int post_kprobe_handler(struct pt_regs *regs)
396 if (!kprobe_running())
399 if ((kprobe_status != KPROBE_REENTER) && current_kprobe->post_handler) {
400 kprobe_status = KPROBE_HIT_SSDONE;
401 current_kprobe->post_handler(current_kprobe, regs, 0);
404 resume_execution(current_kprobe, regs);
405 regs->eflags |= kprobe_saved_eflags;
407 /*Restore back the original saved kprobes variables and continue. */
408 if (kprobe_status == KPROBE_REENTER) {
409 restore_previous_kprobe();
414 preempt_enable_no_resched();
417 * if somebody else is singlestepping across a probe point, eflags
418 * will have TF set, in which case, continue the remaining processing
419 * of do_debug, as if this is not a probe hit.
421 if (regs->eflags & TF_MASK)
427 /* Interrupts disabled, kprobe_lock held. */
428 static inline int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
430 if (current_kprobe->fault_handler
431 && current_kprobe->fault_handler(current_kprobe, regs, trapnr))
434 if (kprobe_status & KPROBE_HIT_SS) {
435 resume_execution(current_kprobe, regs);
436 regs->eflags |= kprobe_old_eflags;
439 preempt_enable_no_resched();
445 * Wrapper routine to for handling exceptions.
447 int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val,
450 struct die_args *args = (struct die_args *)data;
453 if (kprobe_handler(args->regs))
457 if (post_kprobe_handler(args->regs))
461 if (kprobe_running() &&
462 kprobe_fault_handler(args->regs, args->trapnr))
466 if (kprobe_running() &&
467 kprobe_fault_handler(args->regs, args->trapnr))
476 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
478 struct jprobe *jp = container_of(p, struct jprobe, kp);
481 jprobe_saved_regs = *regs;
482 jprobe_saved_esp = ®s->esp;
483 addr = (unsigned long)jprobe_saved_esp;
486 * TBD: As Linus pointed out, gcc assumes that the callee
487 * owns the argument space and could overwrite it, e.g.
488 * tailcall optimization. So, to be absolutely safe
489 * we also save and restore enough stack bytes to cover
492 memcpy(jprobes_stack, (kprobe_opcode_t *) addr, MIN_STACK_SIZE(addr));
493 regs->eflags &= ~IF_MASK;
494 regs->eip = (unsigned long)(jp->entry);
498 void jprobe_return(void)
500 preempt_enable_no_resched();
501 asm volatile (" xchgl %%ebx,%%esp \n"
503 " .globl jprobe_return_end \n"
504 " jprobe_return_end: \n"
506 (jprobe_saved_esp):"memory");
509 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
511 u8 *addr = (u8 *) (regs->eip - 1);
512 unsigned long stack_addr = (unsigned long)jprobe_saved_esp;
513 struct jprobe *jp = container_of(p, struct jprobe, kp);
515 if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
516 if (®s->esp != jprobe_saved_esp) {
517 struct pt_regs *saved_regs =
518 container_of(jprobe_saved_esp, struct pt_regs, esp);
519 printk("current esp %p does not match saved esp %p\n",
520 ®s->esp, jprobe_saved_esp);
521 printk("Saved registers for jprobe %p\n", jp);
522 show_registers(saved_regs);
523 printk("Current registers\n");
524 show_registers(regs);
527 *regs = jprobe_saved_regs;
528 memcpy((kprobe_opcode_t *) stack_addr, jprobes_stack,
529 MIN_STACK_SIZE(stack_addr));
535 static struct kprobe trampoline_p = {
536 .addr = (kprobe_opcode_t *) &kretprobe_trampoline,
537 .pre_handler = trampoline_probe_handler
540 int __init arch_init(void)
542 return register_kprobe(&trampoline_p);