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.
28 #include <linux/config.h>
29 #include <linux/kprobes.h>
30 #include <linux/ptrace.h>
31 #include <linux/spinlock.h>
32 #include <linux/preempt.h>
33 #include <asm/kdebug.h>
36 /* kprobe_status settings */
37 #define KPROBE_HIT_ACTIVE 0x00000001
38 #define KPROBE_HIT_SS 0x00000002
40 static struct kprobe *current_kprobe;
41 static unsigned long kprobe_status, kprobe_old_eflags, kprobe_saved_eflags;
42 static struct pt_regs jprobe_saved_regs;
43 static long *jprobe_saved_esp;
44 /* copy of the kernel stack at the probe fire time */
45 static kprobe_opcode_t jprobes_stack[MAX_STACK_SIZE];
46 void jprobe_return_end(void);
49 * returns non-zero if opcode modifies the interrupt flag.
51 static inline int is_IF_modifier(kprobe_opcode_t opcode)
56 case 0xcf: /* iret/iretd */
57 case 0x9d: /* popf/popfd */
63 int arch_prepare_kprobe(struct kprobe *p)
68 void arch_copy_kprobe(struct kprobe *p)
70 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
73 void arch_remove_kprobe(struct kprobe *p)
77 static inline void disarm_kprobe(struct kprobe *p, struct pt_regs *regs)
80 regs->eip = (unsigned long)p->addr;
83 static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
85 regs->eflags |= TF_MASK;
86 regs->eflags &= ~IF_MASK;
87 /*single step inline if the instruction is an int3*/
88 if (p->opcode == BREAKPOINT_INSTRUCTION)
89 regs->eip = (unsigned long)p->addr;
91 regs->eip = (unsigned long)&p->ainsn.insn;
95 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
96 * remain disabled thorough out this function.
98 static int kprobe_handler(struct pt_regs *regs)
102 kprobe_opcode_t *addr = NULL;
105 /* We're in an interrupt, but this is clear and BUG()-safe. */
107 /* Check if the application is using LDT entry for its code segment and
108 * calculate the address by reading the base address from the LDT entry.
110 if ((regs->xcs & 4) && (current->mm)) {
111 lp = (unsigned long *) ((unsigned long)((regs->xcs >> 3) * 8)
112 + (char *) current->mm->context.ldt);
113 addr = (kprobe_opcode_t *) (get_desc_base(lp) + regs->eip -
114 sizeof(kprobe_opcode_t));
116 addr = (kprobe_opcode_t *)(regs->eip - sizeof(kprobe_opcode_t));
118 /* Check we're not actually recursing */
119 if (kprobe_running()) {
120 /* We *are* holding lock here, so this is safe.
121 Disarm the probe we just hit, and ignore it. */
122 p = get_kprobe(addr);
124 if (kprobe_status == KPROBE_HIT_SS) {
125 regs->eflags &= ~TF_MASK;
126 regs->eflags |= kprobe_saved_eflags;
130 disarm_kprobe(p, regs);
134 if (p->break_handler && p->break_handler(p, regs)) {
138 /* If it's not ours, can't be delete race, (we hold lock). */
143 p = get_kprobe(addr);
146 if (regs->eflags & VM_MASK) {
147 /* We are in virtual-8086 mode. Return 0 */
151 if (*addr != BREAKPOINT_INSTRUCTION) {
153 * The breakpoint instruction was removed right
154 * after we hit it. Another cpu has removed
155 * either a probepoint or a debugger breakpoint
156 * at this address. In either case, no further
157 * handling of this interrupt is appropriate.
161 /* Not one of ours: let kernel handle it */
165 kprobe_status = KPROBE_HIT_ACTIVE;
167 kprobe_saved_eflags = kprobe_old_eflags
168 = (regs->eflags & (TF_MASK | IF_MASK));
169 if (is_IF_modifier(p->opcode))
170 kprobe_saved_eflags &= ~IF_MASK;
172 if (p->pre_handler && p->pre_handler(p, regs))
173 /* handler has already set things up, so skip ss setup */
177 prepare_singlestep(p, regs);
178 kprobe_status = KPROBE_HIT_SS;
182 preempt_enable_no_resched();
187 * Called after single-stepping. p->addr is the address of the
188 * instruction whose first byte has been replaced by the "int 3"
189 * instruction. To avoid the SMP problems that can occur when we
190 * temporarily put back the original opcode to single-step, we
191 * single-stepped a copy of the instruction. The address of this
192 * copy is p->ainsn.insn.
194 * This function prepares to return from the post-single-step
195 * interrupt. We have to fix up the stack as follows:
197 * 0) Except in the case of absolute or indirect jump or call instructions,
198 * the new eip is relative to the copied instruction. We need to make
199 * it relative to the original instruction.
201 * 1) If the single-stepped instruction was pushfl, then the TF and IF
202 * flags are set in the just-pushed eflags, and may need to be cleared.
204 * 2) If the single-stepped instruction was a call, the return address
205 * that is atop the stack is the address following the copied instruction.
206 * We need to make it the address following the original instruction.
208 static void resume_execution(struct kprobe *p, struct pt_regs *regs)
210 unsigned long *tos = (unsigned long *)®s->esp;
211 unsigned long next_eip = 0;
212 unsigned long copy_eip = (unsigned long)&p->ainsn.insn;
213 unsigned long orig_eip = (unsigned long)p->addr;
215 switch (p->ainsn.insn[0]) {
216 case 0x9c: /* pushfl */
217 *tos &= ~(TF_MASK | IF_MASK);
218 *tos |= kprobe_old_eflags;
220 case 0xc3: /* ret/lret */
224 regs->eflags &= ~TF_MASK;
225 /* eip is already adjusted, no more changes required*/
227 case 0xe8: /* call relative - Fix return addr */
228 *tos = orig_eip + (*tos - copy_eip);
231 if ((p->ainsn.insn[1] & 0x30) == 0x10) {
232 /* call absolute, indirect */
233 /* Fix return addr; eip is correct. */
234 next_eip = regs->eip;
235 *tos = orig_eip + (*tos - copy_eip);
236 } else if (((p->ainsn.insn[1] & 0x31) == 0x20) || /* jmp near, absolute indirect */
237 ((p->ainsn.insn[1] & 0x31) == 0x21)) { /* jmp far, absolute indirect */
238 /* eip is correct. */
239 next_eip = regs->eip;
242 case 0xea: /* jmp absolute -- eip is correct */
243 next_eip = regs->eip;
249 regs->eflags &= ~TF_MASK;
251 regs->eip = next_eip;
253 regs->eip = orig_eip + (regs->eip - copy_eip);
258 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
259 * remain disabled thoroughout this function. And we hold kprobe lock.
261 static inline int post_kprobe_handler(struct pt_regs *regs)
263 if (!kprobe_running())
266 if (current_kprobe->post_handler)
267 current_kprobe->post_handler(current_kprobe, regs, 0);
269 resume_execution(current_kprobe, regs);
270 regs->eflags |= kprobe_saved_eflags;
273 preempt_enable_no_resched();
276 * if somebody else is singlestepping across a probe point, eflags
277 * will have TF set, in which case, continue the remaining processing
278 * of do_debug, as if this is not a probe hit.
280 if (regs->eflags & TF_MASK)
286 /* Interrupts disabled, kprobe_lock held. */
287 static inline int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
289 if (current_kprobe->fault_handler
290 && current_kprobe->fault_handler(current_kprobe, regs, trapnr))
293 if (kprobe_status & KPROBE_HIT_SS) {
294 resume_execution(current_kprobe, regs);
295 regs->eflags |= kprobe_old_eflags;
298 preempt_enable_no_resched();
304 * Wrapper routine to for handling exceptions.
306 int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val,
309 struct die_args *args = (struct die_args *)data;
312 if (kprobe_handler(args->regs))
316 if (post_kprobe_handler(args->regs))
320 if (kprobe_running() &&
321 kprobe_fault_handler(args->regs, args->trapnr))
325 if (kprobe_running() &&
326 kprobe_fault_handler(args->regs, args->trapnr))
335 int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
337 struct jprobe *jp = container_of(p, struct jprobe, kp);
340 jprobe_saved_regs = *regs;
341 jprobe_saved_esp = ®s->esp;
342 addr = (unsigned long)jprobe_saved_esp;
345 * TBD: As Linus pointed out, gcc assumes that the callee
346 * owns the argument space and could overwrite it, e.g.
347 * tailcall optimization. So, to be absolutely safe
348 * we also save and restore enough stack bytes to cover
351 memcpy(jprobes_stack, (kprobe_opcode_t *) addr, MIN_STACK_SIZE(addr));
352 regs->eflags &= ~IF_MASK;
353 regs->eip = (unsigned long)(jp->entry);
357 void jprobe_return(void)
359 preempt_enable_no_resched();
360 asm volatile (" xchgl %%ebx,%%esp \n"
362 " .globl jprobe_return_end \n"
363 " jprobe_return_end: \n"
365 (jprobe_saved_esp):"memory");
368 int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
370 u8 *addr = (u8 *) (regs->eip - 1);
371 unsigned long stack_addr = (unsigned long)jprobe_saved_esp;
372 struct jprobe *jp = container_of(p, struct jprobe, kp);
374 if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
375 if (®s->esp != jprobe_saved_esp) {
376 struct pt_regs *saved_regs =
377 container_of(jprobe_saved_esp, struct pt_regs, esp);
378 printk("current esp %p does not match saved esp %p\n",
379 ®s->esp, jprobe_saved_esp);
380 printk("Saved registers for jprobe %p\n", jp);
381 show_registers(saved_regs);
382 printk("Current registers\n");
383 show_registers(regs);
386 *regs = jprobe_saved_regs;
387 memcpy((kprobe_opcode_t *) stack_addr, jprobes_stack,
388 MIN_STACK_SIZE(stack_addr));