/arch/i386/kernel/kprobes.c

  1/*
  2 *  Kernel Probes (KProbes)
  3 *  arch/i386/kernel/kprobes.c
  4 *
  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.
  9 *
 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.
 14 *
 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.
 18 *
 19 * Copyright (C) IBM Corporation, 2002, 2004
 20 *
 21 * 2002-Oct	Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
 22 *		Probes initial implementation ( includes contributions from
 23 *		Rusty Russell).
 24 * 2004-July	Suparna Bhattacharya <suparna@in.ibm.com> added jumper probes
 25 *		interface to access function arguments.
 26 */
 27
 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>
 34#include <asm/desc.h>
 35
 36/* kprobe_status settings */
 37#define KPROBE_HIT_ACTIVE	0x00000001
 38#define KPROBE_HIT_SS		0x00000002
 39
 40static struct kprobe *current_kprobe;
 41static unsigned long kprobe_status, kprobe_old_eflags, kprobe_saved_eflags;
 42static struct pt_regs jprobe_saved_regs;
 43static long *jprobe_saved_esp;
 44/* copy of the kernel stack at the probe fire time */
 45static kprobe_opcode_t jprobes_stack[MAX_STACK_SIZE];
 46void jprobe_return_end(void);
 47
 48/*
 49 * returns non-zero if opcode modifies the interrupt flag.
 50 */
 51static inline int is_IF_modifier(kprobe_opcode_t opcode)
 52{
 53	switch (opcode) {
 54	case 0xfa:		/* cli */
 55	case 0xfb:		/* sti */
 56	case 0xcf:		/* iret/iretd */
 57	case 0x9d:		/* popf/popfd */
 58		return 1;
 59	}
 60	return 0;
 61}
 62
 63int arch_prepare_kprobe(struct kprobe *p)
 64{
 65	return 0;
 66}
 67
 68void arch_copy_kprobe(struct kprobe *p)
 69{
 70	memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t));
 71}
 72
 73void arch_remove_kprobe(struct kprobe *p)
 74{
 75}
 76
 77static inline void disarm_kprobe(struct kprobe *p, struct pt_regs *regs)
 78{
 79	*p->addr = p->opcode;
 80	regs->eip = (unsigned long)p->addr;
 81}
 82
 83static inline void prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
 84{
 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;
 90	else
 91		regs->eip = (unsigned long)&p->ainsn.insn;
 92}
 93
 94/*
 95 * Interrupts are disabled on entry as trap3 is an interrupt gate and they
 96 * remain disabled thorough out this function.
 97 */
 98static int kprobe_handler(struct pt_regs *regs)
 99{
100	struct kprobe *p;
101	int ret = 0;
102	kprobe_opcode_t *addr = NULL;
103	unsigned long *lp;
104
105	/* We're in an interrupt, but this is clear and BUG()-safe. */
106	preempt_disable();
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.
109	 */
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));
115	} else {
116		addr = (kprobe_opcode_t *)(regs->eip - sizeof(kprobe_opcode_t));
117	}
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);
123		if (p) {
124			if (kprobe_status == KPROBE_HIT_SS) {
125				regs->eflags &= ~TF_MASK;
126				regs->eflags |= kprobe_saved_eflags;
127				unlock_kprobes();
128				goto no_kprobe;
129			}
130			disarm_kprobe(p, regs);
131			ret = 1;
132		} else {
133			p = current_kprobe;
134			if (p->break_handler && p->break_handler(p, regs)) {
135				goto ss_probe;
136			}
137		}
138		/* If it's not ours, can't be delete race, (we hold lock). */
139		goto no_kprobe;
140	}
141
142	lock_kprobes();
143	p = get_kprobe(addr);
144	if (!p) {
145		unlock_kprobes();
146		if (regs->eflags & VM_MASK) {
147			/* We are in virtual-8086 mode. Return 0 */
148			goto no_kprobe;
149		}
150
151		if (*addr != BREAKPOINT_INSTRUCTION) {
152			/*
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.
158			 */
159			ret = 1;
160		}
161		/* Not one of ours: let kernel handle it */
162		goto no_kprobe;
163	}
164
165	kprobe_status = KPROBE_HIT_ACTIVE;
166	current_kprobe = p;
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;
171
172	if (p->pre_handler && p->pre_handler(p, regs))
173		/* handler has already set things up, so skip ss setup */
174		return 1;
175
176ss_probe:
177	prepare_singlestep(p, regs);
178	kprobe_status = KPROBE_HIT_SS;
179	return 1;
180
181no_kprobe:
182	preempt_enable_no_resched();
183	return ret;
184}
185
186/*
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.
193 *
194 * This function prepares to return from the post-single-step
195 * interrupt.  We have to fix up the stack as follows:
196 *
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.
200 *
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.
203 *
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.
207 */
208static void resume_execution(struct kprobe *p, struct pt_regs *regs)
209{
210	unsigned long *tos = (unsigned long *)&regs->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;
214
215	switch (p->ainsn.insn[0]) {
216	case 0x9c:		/* pushfl */
217		*tos &= ~(TF_MASK | IF_MASK);
218		*tos |= kprobe_old_eflags;
219		break;
220	case 0xe8:		/* call relative - Fix return addr */
221		*tos = orig_eip + (*tos - copy_eip);
222		break;
223	case 0xff:
224		if ((p->ainsn.insn[1] & 0x30) == 0x10) {
225			/* call absolute, indirect */
226			/* Fix return addr; eip is correct. */
227			next_eip = regs->eip;
228			*tos = orig_eip + (*tos - copy_eip);
229		} else if (((p->ainsn.insn[1] & 0x31) == 0x20) ||	/* jmp near, absolute indirect */
230			   ((p->ainsn.insn[1] & 0x31) == 0x21)) {	/* jmp far, absolute indirect */
231			/* eip is correct. */
232			next_eip = regs->eip;
233		}
234		break;
235	case 0xea:		/* jmp absolute -- eip is correct */
236		next_eip = regs->eip;
237		break;
238	default:
239		break;
240	}
241
242	regs->eflags &= ~TF_MASK;
243	if (next_eip) {
244		regs->eip = next_eip;
245	} else {
246		regs->eip = orig_eip + (regs->eip - copy_eip);
247	}
248}
249
250/*
251 * Interrupts are disabled on entry as trap1 is an interrupt gate and they
252 * remain disabled thoroughout this function.  And we hold kprobe lock.
253 */
254static inline int post_kprobe_handler(struct pt_regs *regs)
255{
256	if (!kprobe_running())
257		return 0;
258
259	if (current_kprobe->post_handler)
260		current_kprobe->post_handler(current_kprobe, regs, 0);
261
262	resume_execution(current_kprobe, regs);
263	regs->eflags |= kprobe_saved_eflags;
264
265	unlock_kprobes();
266	preempt_enable_no_resched();
267
268	/*
269	 * if somebody else is singlestepping across a probe point, eflags
270	 * will have TF set, in which case, continue the remaining processing
271	 * of do_debug, as if this is not a probe hit.
272	 */
273	if (regs->eflags & TF_MASK)
274		return 0;
275
276	return 1;
277}
278
279/* Interrupts disabled, kprobe_lock held. */
280static inline int kprobe_fault_handler(struct pt_regs *regs, int trapnr)
281{
282	if (current_kprobe->fault_handler
283	    && current_kprobe->fault_handler(current_kprobe, regs, trapnr))
284		return 1;
285
286	if (kprobe_status & KPROBE_HIT_SS) {
287		resume_execution(current_kprobe, regs);
288		regs->eflags |= kprobe_old_eflags;
289
290		unlock_kprobes();
291		preempt_enable_no_resched();
292	}
293	return 0;
294}
295
296/*
297 * Wrapper routine to for handling exceptions.
298 */
299int kprobe_exceptions_notify(struct notifier_block *self, unsigned long val,
300			     void *data)
301{
302	struct die_args *args = (struct die_args *)data;
303	switch (val) {
304	case DIE_INT3:
305		if (kprobe_handler(args->regs))
306			return NOTIFY_STOP;
307		break;
308	case DIE_DEBUG:
309		if (post_kprobe_handler(args->regs))
310			return NOTIFY_STOP;
311		break;
312	case DIE_GPF:
313		if (kprobe_running() &&
314		    kprobe_fault_handler(args->regs, args->trapnr))
315			return NOTIFY_STOP;
316		break;
317	case DIE_PAGE_FAULT:
318		if (kprobe_running() &&
319		    kprobe_fault_handler(args->regs, args->trapnr))
320			return NOTIFY_STOP;
321		break;
322	default:
323		break;
324	}
325	return NOTIFY_DONE;
326}
327
328int setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
329{
330	struct jprobe *jp = container_of(p, struct jprobe, kp);
331	unsigned long addr;
332
333	jprobe_saved_regs = *regs;
334	jprobe_saved_esp = &regs->esp;
335	addr = (unsigned long)jprobe_saved_esp;
336
337	/*
338	 * TBD: As Linus pointed out, gcc assumes that the callee
339	 * owns the argument space and could overwrite it, e.g.
340	 * tailcall optimization. So, to be absolutely safe
341	 * we also save and restore enough stack bytes to cover
342	 * the argument area.
343	 */
344	memcpy(jprobes_stack, (kprobe_opcode_t *) addr, MIN_STACK_SIZE(addr));
345	regs->eflags &= ~IF_MASK;
346	regs->eip = (unsigned long)(jp->entry);
347	return 1;
348}
349
350void jprobe_return(void)
351{
352	preempt_enable_no_resched();
353	asm volatile ("       xchgl   %%ebx,%%esp     \n"
354		      "       int3			\n"
355		      "       .globl jprobe_return_end	\n"
356		      "       jprobe_return_end:	\n"
357		      "       nop			\n"::"b"
358		      (jprobe_saved_esp):"memory");
359}
360
361int longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
362{
363	u8 *addr = (u8 *) (regs->eip - 1);
364	unsigned long stack_addr = (unsigned long)jprobe_saved_esp;
365	struct jprobe *jp = container_of(p, struct jprobe, kp);
366
367	if ((addr > (u8 *) jprobe_return) && (addr < (u8 *) jprobe_return_end)) {
368		if (&regs->esp != jprobe_saved_esp) {
369			struct pt_regs *saved_regs =
370			    container_of(jprobe_saved_esp, struct pt_regs, esp);
371			printk("current esp %p does not match saved esp %p\n",
372			       &regs->esp, jprobe_saved_esp);
373			printk("Saved registers for jprobe %p\n", jp);
374			show_registers(saved_regs);
375			printk("Current registers\n");
376			show_registers(regs);
377			BUG();
378		}
379		*regs = jprobe_saved_regs;
380		memcpy((kprobe_opcode_t *) stack_addr, jprobes_stack,
381		       MIN_STACK_SIZE(stack_addr));
382		return 1;
383	}
384	return 0;
385}