aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2010-03-05 13:50:22 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2010-03-05 13:50:22 -0500
commit660f6a360be399f4ebdd6572a3d24afe54e9bb1c (patch)
tree9c16463c495a656e34577d59c97b58997b61d242 /arch/x86/kernel
parent586fac13f8685bf9dfb32e1ee98bfb14f0dd0061 (diff)
parente5a11016643d1ab7172193591506d33a844734cc (diff)
Merge branch 'perf-probes-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
* 'perf-probes-for-linus-2' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: x86: Issue at least one memory barrier in stop_machine_text_poke() perf probe: Correct probe syntax on command line help perf probe: Add lazy line matching support perf probe: Show more lines after last line perf probe: Check function address range strictly in line finder perf probe: Use libdw callback routines perf probe: Use elfutils-libdw for analyzing debuginfo perf probe: Rename probe finder functions perf probe: Fix bugs in line range finder perf probe: Update perf probe document perf probe: Do not show --line option without dwarf support kprobes: Add documents of jump optimization kprobes/x86: Support kprobes jump optimization on x86 x86: Add text_poke_smp for SMP cross modifying code kprobes/x86: Cleanup save/restore registers kprobes/x86: Boost probes when reentering kprobes: Jump optimization sysctl interface kprobes: Introduce kprobes jump optimization kprobes: Introduce generic insn_slot framework kprobes/x86: Cleanup RELATIVEJUMP_INSTRUCTION to RELATIVEJUMP_OPCODE
Diffstat (limited to 'arch/x86/kernel')
-rw-r--r--arch/x86/kernel/alternative.c60
-rw-r--r--arch/x86/kernel/kprobes.c609
2 files changed, 564 insertions, 105 deletions
diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
index e6ea0342c8f8..3a4bf35c179b 100644
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -7,6 +7,7 @@
7#include <linux/mm.h> 7#include <linux/mm.h>
8#include <linux/vmalloc.h> 8#include <linux/vmalloc.h>
9#include <linux/memory.h> 9#include <linux/memory.h>
10#include <linux/stop_machine.h>
10#include <asm/alternative.h> 11#include <asm/alternative.h>
11#include <asm/sections.h> 12#include <asm/sections.h>
12#include <asm/pgtable.h> 13#include <asm/pgtable.h>
@@ -572,3 +573,62 @@ void *__kprobes text_poke(void *addr, const void *opcode, size_t len)
572 local_irq_restore(flags); 573 local_irq_restore(flags);
573 return addr; 574 return addr;
574} 575}
576
577/*
578 * Cross-modifying kernel text with stop_machine().
579 * This code originally comes from immediate value.
580 */
581static atomic_t stop_machine_first;
582static int wrote_text;
583
584struct text_poke_params {
585 void *addr;
586 const void *opcode;
587 size_t len;
588};
589
590static int __kprobes stop_machine_text_poke(void *data)
591{
592 struct text_poke_params *tpp = data;
593
594 if (atomic_dec_and_test(&stop_machine_first)) {
595 text_poke(tpp->addr, tpp->opcode, tpp->len);
596 smp_wmb(); /* Make sure other cpus see that this has run */
597 wrote_text = 1;
598 } else {
599 while (!wrote_text)
600 cpu_relax();
601 smp_mb(); /* Load wrote_text before following execution */
602 }
603
604 flush_icache_range((unsigned long)tpp->addr,
605 (unsigned long)tpp->addr + tpp->len);
606 return 0;
607}
608
609/**
610 * text_poke_smp - Update instructions on a live kernel on SMP
611 * @addr: address to modify
612 * @opcode: source of the copy
613 * @len: length to copy
614 *
615 * Modify multi-byte instruction by using stop_machine() on SMP. This allows
616 * user to poke/set multi-byte text on SMP. Only non-NMI/MCE code modifying
617 * should be allowed, since stop_machine() does _not_ protect code against
618 * NMI and MCE.
619 *
620 * Note: Must be called under get_online_cpus() and text_mutex.
621 */
622void *__kprobes text_poke_smp(void *addr, const void *opcode, size_t len)
623{
624 struct text_poke_params tpp;
625
626 tpp.addr = addr;
627 tpp.opcode = opcode;
628 tpp.len = len;
629 atomic_set(&stop_machine_first, 1);
630 wrote_text = 0;
631 stop_machine(stop_machine_text_poke, (void *)&tpp, NULL);
632 return addr;
633}
634
diff --git a/arch/x86/kernel/kprobes.c b/arch/x86/kernel/kprobes.c
index 5de9f4a9c3fd..b43bbaebe2c0 100644
--- a/arch/x86/kernel/kprobes.c
+++ b/arch/x86/kernel/kprobes.c
@@ -49,6 +49,7 @@
49#include <linux/module.h> 49#include <linux/module.h>
50#include <linux/kdebug.h> 50#include <linux/kdebug.h>
51#include <linux/kallsyms.h> 51#include <linux/kallsyms.h>
52#include <linux/ftrace.h>
52 53
53#include <asm/cacheflush.h> 54#include <asm/cacheflush.h>
54#include <asm/desc.h> 55#include <asm/desc.h>
@@ -106,16 +107,22 @@ struct kretprobe_blackpoint kretprobe_blacklist[] = {
106}; 107};
107const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist); 108const int kretprobe_blacklist_size = ARRAY_SIZE(kretprobe_blacklist);
108 109
109/* Insert a jump instruction at address 'from', which jumps to address 'to'.*/ 110static void __kprobes __synthesize_relative_insn(void *from, void *to, u8 op)
110static void __kprobes set_jmp_op(void *from, void *to)
111{ 111{
112 struct __arch_jmp_op { 112 struct __arch_relative_insn {
113 char op; 113 u8 op;
114 s32 raddr; 114 s32 raddr;
115 } __attribute__((packed)) * jop; 115 } __attribute__((packed)) *insn;
116 jop = (struct __arch_jmp_op *)from; 116
117 jop->raddr = (s32)((long)(to) - ((long)(from) + 5)); 117 insn = (struct __arch_relative_insn *)from;
118 jop->op = RELATIVEJUMP_INSTRUCTION; 118 insn->raddr = (s32)((long)(to) - ((long)(from) + 5));
119 insn->op = op;
120}
121
122/* Insert a jump instruction at address 'from', which jumps to address 'to'.*/
123static void __kprobes synthesize_reljump(void *from, void *to)
124{
125 __synthesize_relative_insn(from, to, RELATIVEJUMP_OPCODE);
119} 126}
120 127
121/* 128/*
@@ -202,7 +209,7 @@ static int recover_probed_instruction(kprobe_opcode_t *buf, unsigned long addr)
202 /* 209 /*
203 * Basically, kp->ainsn.insn has an original instruction. 210 * Basically, kp->ainsn.insn has an original instruction.
204 * However, RIP-relative instruction can not do single-stepping 211 * However, RIP-relative instruction can not do single-stepping
205 * at different place, fix_riprel() tweaks the displacement of 212 * at different place, __copy_instruction() tweaks the displacement of
206 * that instruction. In that case, we can't recover the instruction 213 * that instruction. In that case, we can't recover the instruction
207 * from the kp->ainsn.insn. 214 * from the kp->ainsn.insn.
208 * 215 *
@@ -284,21 +291,37 @@ static int __kprobes is_IF_modifier(kprobe_opcode_t *insn)
284} 291}
285 292
286/* 293/*
287 * Adjust the displacement if the instruction uses the %rip-relative 294 * Copy an instruction and adjust the displacement if the instruction
288 * addressing mode. 295 * uses the %rip-relative addressing mode.
289 * If it does, Return the address of the 32-bit displacement word. 296 * If it does, Return the address of the 32-bit displacement word.
290 * If not, return null. 297 * If not, return null.
291 * Only applicable to 64-bit x86. 298 * Only applicable to 64-bit x86.
292 */ 299 */
293static void __kprobes fix_riprel(struct kprobe *p) 300static int __kprobes __copy_instruction(u8 *dest, u8 *src, int recover)
294{ 301{
295#ifdef CONFIG_X86_64
296 struct insn insn; 302 struct insn insn;
297 kernel_insn_init(&insn, p->ainsn.insn); 303 int ret;
304 kprobe_opcode_t buf[MAX_INSN_SIZE];
298 305
306 kernel_insn_init(&insn, src);
307 if (recover) {
308 insn_get_opcode(&insn);
309 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION) {
310 ret = recover_probed_instruction(buf,
311 (unsigned long)src);
312 if (ret)
313 return 0;
314 kernel_insn_init(&insn, buf);
315 }
316 }
317 insn_get_length(&insn);
318 memcpy(dest, insn.kaddr, insn.length);
319
320#ifdef CONFIG_X86_64
299 if (insn_rip_relative(&insn)) { 321 if (insn_rip_relative(&insn)) {
300 s64 newdisp; 322 s64 newdisp;
301 u8 *disp; 323 u8 *disp;
324 kernel_insn_init(&insn, dest);
302 insn_get_displacement(&insn); 325 insn_get_displacement(&insn);
303 /* 326 /*
304 * The copied instruction uses the %rip-relative addressing 327 * The copied instruction uses the %rip-relative addressing
@@ -312,20 +335,23 @@ static void __kprobes fix_riprel(struct kprobe *p)
312 * extension of the original signed 32-bit displacement would 335 * extension of the original signed 32-bit displacement would
313 * have given. 336 * have given.
314 */ 337 */
315 newdisp = (u8 *) p->addr + (s64) insn.displacement.value - 338 newdisp = (u8 *) src + (s64) insn.displacement.value -
316 (u8 *) p->ainsn.insn; 339 (u8 *) dest;
317 BUG_ON((s64) (s32) newdisp != newdisp); /* Sanity check. */ 340 BUG_ON((s64) (s32) newdisp != newdisp); /* Sanity check. */
318 disp = (u8 *) p->ainsn.insn + insn_offset_displacement(&insn); 341 disp = (u8 *) dest + insn_offset_displacement(&insn);
319 *(s32 *) disp = (s32) newdisp; 342 *(s32 *) disp = (s32) newdisp;
320 } 343 }
321#endif 344#endif
345 return insn.length;
322} 346}
323 347
324static void __kprobes arch_copy_kprobe(struct kprobe *p) 348static void __kprobes arch_copy_kprobe(struct kprobe *p)
325{ 349{
326 memcpy(p->ainsn.insn, p->addr, MAX_INSN_SIZE * sizeof(kprobe_opcode_t)); 350 /*
327 351 * Copy an instruction without recovering int3, because it will be
328 fix_riprel(p); 352 * put by another subsystem.
353 */
354 __copy_instruction(p->ainsn.insn, p->addr, 0);
329 355
330 if (can_boost(p->addr)) 356 if (can_boost(p->addr))
331 p->ainsn.boostable = 0; 357 p->ainsn.boostable = 0;
@@ -406,18 +432,6 @@ static void __kprobes restore_btf(void)
406 update_debugctlmsr(current->thread.debugctlmsr); 432 update_debugctlmsr(current->thread.debugctlmsr);
407} 433}
408 434
409static void __kprobes prepare_singlestep(struct kprobe *p, struct pt_regs *regs)
410{
411 clear_btf();
412 regs->flags |= X86_EFLAGS_TF;
413 regs->flags &= ~X86_EFLAGS_IF;
414 /* single step inline if the instruction is an int3 */
415 if (p->opcode == BREAKPOINT_INSTRUCTION)
416 regs->ip = (unsigned long)p->addr;
417 else
418 regs->ip = (unsigned long)p->ainsn.insn;
419}
420
421void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri, 435void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
422 struct pt_regs *regs) 436 struct pt_regs *regs)
423{ 437{
@@ -429,20 +443,50 @@ void __kprobes arch_prepare_kretprobe(struct kretprobe_instance *ri,
429 *sara = (unsigned long) &kretprobe_trampoline; 443 *sara = (unsigned long) &kretprobe_trampoline;
430} 444}
431 445
446#ifdef CONFIG_OPTPROBES
447static int __kprobes setup_detour_execution(struct kprobe *p,
448 struct pt_regs *regs,
449 int reenter);
450#else
451#define setup_detour_execution(p, regs, reenter) (0)
452#endif
453
432static void __kprobes setup_singlestep(struct kprobe *p, struct pt_regs *regs, 454static void __kprobes setup_singlestep(struct kprobe *p, struct pt_regs *regs,
433 struct kprobe_ctlblk *kcb) 455 struct kprobe_ctlblk *kcb, int reenter)
434{ 456{
457 if (setup_detour_execution(p, regs, reenter))
458 return;
459
435#if !defined(CONFIG_PREEMPT) 460#if !defined(CONFIG_PREEMPT)
436 if (p->ainsn.boostable == 1 && !p->post_handler) { 461 if (p->ainsn.boostable == 1 && !p->post_handler) {
437 /* Boost up -- we can execute copied instructions directly */ 462 /* Boost up -- we can execute copied instructions directly */
438 reset_current_kprobe(); 463 if (!reenter)
464 reset_current_kprobe();
465 /*
466 * Reentering boosted probe doesn't reset current_kprobe,
467 * nor set current_kprobe, because it doesn't use single
468 * stepping.
469 */
439 regs->ip = (unsigned long)p->ainsn.insn; 470 regs->ip = (unsigned long)p->ainsn.insn;
440 preempt_enable_no_resched(); 471 preempt_enable_no_resched();
441 return; 472 return;
442 } 473 }
443#endif 474#endif
444 prepare_singlestep(p, regs); 475 if (reenter) {
445 kcb->kprobe_status = KPROBE_HIT_SS; 476 save_previous_kprobe(kcb);
477 set_current_kprobe(p, regs, kcb);
478 kcb->kprobe_status = KPROBE_REENTER;
479 } else
480 kcb->kprobe_status = KPROBE_HIT_SS;
481 /* Prepare real single stepping */
482 clear_btf();
483 regs->flags |= X86_EFLAGS_TF;
484 regs->flags &= ~X86_EFLAGS_IF;
485 /* single step inline if the instruction is an int3 */
486 if (p->opcode == BREAKPOINT_INSTRUCTION)
487 regs->ip = (unsigned long)p->addr;
488 else
489 regs->ip = (unsigned long)p->ainsn.insn;
446} 490}
447 491
448/* 492/*
@@ -456,11 +500,8 @@ static int __kprobes reenter_kprobe(struct kprobe *p, struct pt_regs *regs,
456 switch (kcb->kprobe_status) { 500 switch (kcb->kprobe_status) {
457 case KPROBE_HIT_SSDONE: 501 case KPROBE_HIT_SSDONE:
458 case KPROBE_HIT_ACTIVE: 502 case KPROBE_HIT_ACTIVE:
459 save_previous_kprobe(kcb);
460 set_current_kprobe(p, regs, kcb);
461 kprobes_inc_nmissed_count(p); 503 kprobes_inc_nmissed_count(p);
462 prepare_singlestep(p, regs); 504 setup_singlestep(p, regs, kcb, 1);
463 kcb->kprobe_status = KPROBE_REENTER;
464 break; 505 break;
465 case KPROBE_HIT_SS: 506 case KPROBE_HIT_SS:
466 /* A probe has been hit in the codepath leading up to, or just 507 /* A probe has been hit in the codepath leading up to, or just
@@ -535,13 +576,13 @@ static int __kprobes kprobe_handler(struct pt_regs *regs)
535 * more here. 576 * more here.
536 */ 577 */
537 if (!p->pre_handler || !p->pre_handler(p, regs)) 578 if (!p->pre_handler || !p->pre_handler(p, regs))
538 setup_singlestep(p, regs, kcb); 579 setup_singlestep(p, regs, kcb, 0);
539 return 1; 580 return 1;
540 } 581 }
541 } else if (kprobe_running()) { 582 } else if (kprobe_running()) {
542 p = __get_cpu_var(current_kprobe); 583 p = __get_cpu_var(current_kprobe);
543 if (p->break_handler && p->break_handler(p, regs)) { 584 if (p->break_handler && p->break_handler(p, regs)) {
544 setup_singlestep(p, regs, kcb); 585 setup_singlestep(p, regs, kcb, 0);
545 return 1; 586 return 1;
546 } 587 }
547 } /* else: not a kprobe fault; let the kernel handle it */ 588 } /* else: not a kprobe fault; let the kernel handle it */
@@ -550,6 +591,69 @@ static int __kprobes kprobe_handler(struct pt_regs *regs)
550 return 0; 591 return 0;
551} 592}
552 593
594#ifdef CONFIG_X86_64
595#define SAVE_REGS_STRING \
596 /* Skip cs, ip, orig_ax. */ \
597 " subq $24, %rsp\n" \
598 " pushq %rdi\n" \
599 " pushq %rsi\n" \
600 " pushq %rdx\n" \
601 " pushq %rcx\n" \
602 " pushq %rax\n" \
603 " pushq %r8\n" \
604 " pushq %r9\n" \
605 " pushq %r10\n" \
606 " pushq %r11\n" \
607 " pushq %rbx\n" \
608 " pushq %rbp\n" \
609 " pushq %r12\n" \
610 " pushq %r13\n" \
611 " pushq %r14\n" \
612 " pushq %r15\n"
613#define RESTORE_REGS_STRING \
614 " popq %r15\n" \
615 " popq %r14\n" \
616 " popq %r13\n" \
617 " popq %r12\n" \
618 " popq %rbp\n" \
619 " popq %rbx\n" \
620 " popq %r11\n" \
621 " popq %r10\n" \
622 " popq %r9\n" \
623 " popq %r8\n" \
624 " popq %rax\n" \
625 " popq %rcx\n" \
626 " popq %rdx\n" \
627 " popq %rsi\n" \
628 " popq %rdi\n" \
629 /* Skip orig_ax, ip, cs */ \
630 " addq $24, %rsp\n"
631#else
632#define SAVE_REGS_STRING \
633 /* Skip cs, ip, orig_ax and gs. */ \
634 " subl $16, %esp\n" \
635 " pushl %fs\n" \
636 " pushl %ds\n" \
637 " pushl %es\n" \
638 " pushl %eax\n" \
639 " pushl %ebp\n" \
640 " pushl %edi\n" \
641 " pushl %esi\n" \
642 " pushl %edx\n" \
643 " pushl %ecx\n" \
644 " pushl %ebx\n"
645#define RESTORE_REGS_STRING \
646 " popl %ebx\n" \
647 " popl %ecx\n" \
648 " popl %edx\n" \
649 " popl %esi\n" \
650 " popl %edi\n" \
651 " popl %ebp\n" \
652 " popl %eax\n" \
653 /* Skip ds, es, fs, gs, orig_ax, and ip. Note: don't pop cs here*/\
654 " addl $24, %esp\n"
655#endif
656
553/* 657/*
554 * When a retprobed function returns, this code saves registers and 658 * When a retprobed function returns, this code saves registers and
555 * calls trampoline_handler() runs, which calls the kretprobe's handler. 659 * calls trampoline_handler() runs, which calls the kretprobe's handler.
@@ -563,65 +667,16 @@ static void __used __kprobes kretprobe_trampoline_holder(void)
563 /* We don't bother saving the ss register */ 667 /* We don't bother saving the ss register */
564 " pushq %rsp\n" 668 " pushq %rsp\n"
565 " pushfq\n" 669 " pushfq\n"
566 /* 670 SAVE_REGS_STRING
567 * Skip cs, ip, orig_ax.
568 * trampoline_handler() will plug in these values
569 */
570 " subq $24, %rsp\n"
571 " pushq %rdi\n"
572 " pushq %rsi\n"
573 " pushq %rdx\n"
574 " pushq %rcx\n"
575 " pushq %rax\n"
576 " pushq %r8\n"
577 " pushq %r9\n"
578 " pushq %r10\n"
579 " pushq %r11\n"
580 " pushq %rbx\n"
581 " pushq %rbp\n"
582 " pushq %r12\n"
583 " pushq %r13\n"
584 " pushq %r14\n"
585 " pushq %r15\n"
586 " movq %rsp, %rdi\n" 671 " movq %rsp, %rdi\n"
587 " call trampoline_handler\n" 672 " call trampoline_handler\n"
588 /* Replace saved sp with true return address. */ 673 /* Replace saved sp with true return address. */
589 " movq %rax, 152(%rsp)\n" 674 " movq %rax, 152(%rsp)\n"
590 " popq %r15\n" 675 RESTORE_REGS_STRING
591 " popq %r14\n"
592 " popq %r13\n"
593 " popq %r12\n"
594 " popq %rbp\n"
595 " popq %rbx\n"
596 " popq %r11\n"
597 " popq %r10\n"
598 " popq %r9\n"
599 " popq %r8\n"
600 " popq %rax\n"
601 " popq %rcx\n"
602 " popq %rdx\n"
603 " popq %rsi\n"
604 " popq %rdi\n"
605 /* Skip orig_ax, ip, cs */
606 " addq $24, %rsp\n"
607 " popfq\n" 676 " popfq\n"
608#else 677#else
609 " pushf\n" 678 " pushf\n"
610 /* 679 SAVE_REGS_STRING
611 * Skip cs, ip, orig_ax and gs.
612 * trampoline_handler() will plug in these values
613 */
614 " subl $16, %esp\n"
615 " pushl %fs\n"
616 " pushl %es\n"
617 " pushl %ds\n"
618 " pushl %eax\n"
619 " pushl %ebp\n"
620 " pushl %edi\n"
621 " pushl %esi\n"
622 " pushl %edx\n"
623 " pushl %ecx\n"
624 " pushl %ebx\n"
625 " movl %esp, %eax\n" 680 " movl %esp, %eax\n"
626 " call trampoline_handler\n" 681 " call trampoline_handler\n"
627 /* Move flags to cs */ 682 /* Move flags to cs */
@@ -629,15 +684,7 @@ static void __used __kprobes kretprobe_trampoline_holder(void)
629 " movl %edx, 52(%esp)\n" 684 " movl %edx, 52(%esp)\n"
630 /* Replace saved flags with true return address. */ 685 /* Replace saved flags with true return address. */
631 " movl %eax, 56(%esp)\n" 686 " movl %eax, 56(%esp)\n"
632 " popl %ebx\n" 687 RESTORE_REGS_STRING
633 " popl %ecx\n"
634 " popl %edx\n"
635 " popl %esi\n"
636 " popl %edi\n"
637 " popl %ebp\n"
638 " popl %eax\n"
639 /* Skip ds, es, fs, gs, orig_ax and ip */
640 " addl $24, %esp\n"
641 " popf\n" 688 " popf\n"
642#endif 689#endif
643 " ret\n"); 690 " ret\n");
@@ -805,8 +852,8 @@ static void __kprobes resume_execution(struct kprobe *p,
805 * These instructions can be executed directly if it 852 * These instructions can be executed directly if it
806 * jumps back to correct address. 853 * jumps back to correct address.
807 */ 854 */
808 set_jmp_op((void *)regs->ip, 855 synthesize_reljump((void *)regs->ip,
809 (void *)orig_ip + (regs->ip - copy_ip)); 856 (void *)orig_ip + (regs->ip - copy_ip));
810 p->ainsn.boostable = 1; 857 p->ainsn.boostable = 1;
811 } else { 858 } else {
812 p->ainsn.boostable = -1; 859 p->ainsn.boostable = -1;
@@ -1033,6 +1080,358 @@ int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs)
1033 return 0; 1080 return 0;
1034} 1081}
1035 1082
1083
1084#ifdef CONFIG_OPTPROBES
1085
1086/* Insert a call instruction at address 'from', which calls address 'to'.*/
1087static void __kprobes synthesize_relcall(void *from, void *to)
1088{
1089 __synthesize_relative_insn(from, to, RELATIVECALL_OPCODE);
1090}
1091
1092/* Insert a move instruction which sets a pointer to eax/rdi (1st arg). */
1093static void __kprobes synthesize_set_arg1(kprobe_opcode_t *addr,
1094 unsigned long val)
1095{
1096#ifdef CONFIG_X86_64
1097 *addr++ = 0x48;
1098 *addr++ = 0xbf;
1099#else
1100 *addr++ = 0xb8;
1101#endif
1102 *(unsigned long *)addr = val;
1103}
1104
1105void __kprobes kprobes_optinsn_template_holder(void)
1106{
1107 asm volatile (
1108 ".global optprobe_template_entry\n"
1109 "optprobe_template_entry: \n"
1110#ifdef CONFIG_X86_64
1111 /* We don't bother saving the ss register */
1112 " pushq %rsp\n"
1113 " pushfq\n"
1114 SAVE_REGS_STRING
1115 " movq %rsp, %rsi\n"
1116 ".global optprobe_template_val\n"
1117 "optprobe_template_val: \n"
1118 ASM_NOP5
1119 ASM_NOP5
1120 ".global optprobe_template_call\n"
1121 "optprobe_template_call: \n"
1122 ASM_NOP5
1123 /* Move flags to rsp */
1124 " movq 144(%rsp), %rdx\n"
1125 " movq %rdx, 152(%rsp)\n"
1126 RESTORE_REGS_STRING
1127 /* Skip flags entry */
1128 " addq $8, %rsp\n"
1129 " popfq\n"
1130#else /* CONFIG_X86_32 */
1131 " pushf\n"
1132 SAVE_REGS_STRING
1133 " movl %esp, %edx\n"
1134 ".global optprobe_template_val\n"
1135 "optprobe_template_val: \n"
1136 ASM_NOP5
1137 ".global optprobe_template_call\n"
1138 "optprobe_template_call: \n"
1139 ASM_NOP5
1140 RESTORE_REGS_STRING
1141 " addl $4, %esp\n" /* skip cs */
1142 " popf\n"
1143#endif
1144 ".global optprobe_template_end\n"
1145 "optprobe_template_end: \n");
1146}
1147
1148#define TMPL_MOVE_IDX \
1149 ((long)&optprobe_template_val - (long)&optprobe_template_entry)
1150#define TMPL_CALL_IDX \
1151 ((long)&optprobe_template_call - (long)&optprobe_template_entry)
1152#define TMPL_END_IDX \
1153 ((long)&optprobe_template_end - (long)&optprobe_template_entry)
1154
1155#define INT3_SIZE sizeof(kprobe_opcode_t)
1156
1157/* Optimized kprobe call back function: called from optinsn */
1158static void __kprobes optimized_callback(struct optimized_kprobe *op,
1159 struct pt_regs *regs)
1160{
1161 struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
1162
1163 preempt_disable();
1164 if (kprobe_running()) {
1165 kprobes_inc_nmissed_count(&op->kp);
1166 } else {
1167 /* Save skipped registers */
1168#ifdef CONFIG_X86_64
1169 regs->cs = __KERNEL_CS;
1170#else
1171 regs->cs = __KERNEL_CS | get_kernel_rpl();
1172 regs->gs = 0;
1173#endif
1174 regs->ip = (unsigned long)op->kp.addr + INT3_SIZE;
1175 regs->orig_ax = ~0UL;
1176
1177 __get_cpu_var(current_kprobe) = &op->kp;
1178 kcb->kprobe_status = KPROBE_HIT_ACTIVE;
1179 opt_pre_handler(&op->kp, regs);
1180 __get_cpu_var(current_kprobe) = NULL;
1181 }
1182 preempt_enable_no_resched();
1183}
1184
1185static int __kprobes copy_optimized_instructions(u8 *dest, u8 *src)
1186{
1187 int len = 0, ret;
1188
1189 while (len < RELATIVEJUMP_SIZE) {
1190 ret = __copy_instruction(dest + len, src + len, 1);
1191 if (!ret || !can_boost(dest + len))
1192 return -EINVAL;
1193 len += ret;
1194 }
1195 /* Check whether the address range is reserved */
1196 if (ftrace_text_reserved(src, src + len - 1) ||
1197 alternatives_text_reserved(src, src + len - 1))
1198 return -EBUSY;
1199
1200 return len;
1201}
1202
1203/* Check whether insn is indirect jump */
1204static int __kprobes insn_is_indirect_jump(struct insn *insn)
1205{
1206 return ((insn->opcode.bytes[0] == 0xff &&
1207 (X86_MODRM_REG(insn->modrm.value) & 6) == 4) || /* Jump */
1208 insn->opcode.bytes[0] == 0xea); /* Segment based jump */
1209}
1210
1211/* Check whether insn jumps into specified address range */
1212static int insn_jump_into_range(struct insn *insn, unsigned long start, int len)
1213{
1214 unsigned long target = 0;
1215
1216 switch (insn->opcode.bytes[0]) {
1217 case 0xe0: /* loopne */
1218 case 0xe1: /* loope */
1219 case 0xe2: /* loop */
1220 case 0xe3: /* jcxz */
1221 case 0xe9: /* near relative jump */
1222 case 0xeb: /* short relative jump */
1223 break;
1224 case 0x0f:
1225 if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
1226 break;
1227 return 0;
1228 default:
1229 if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
1230 break;
1231 return 0;
1232 }
1233 target = (unsigned long)insn->next_byte + insn->immediate.value;
1234
1235 return (start <= target && target <= start + len);
1236}
1237
1238/* Decode whole function to ensure any instructions don't jump into target */
1239static int __kprobes can_optimize(unsigned long paddr)
1240{
1241 int ret;
1242 unsigned long addr, size = 0, offset = 0;
1243 struct insn insn;
1244 kprobe_opcode_t buf[MAX_INSN_SIZE];
1245 /* Dummy buffers for lookup_symbol_attrs */
1246 static char __dummy_buf[KSYM_NAME_LEN];
1247
1248 /* Lookup symbol including addr */
1249 if (!kallsyms_lookup(paddr, &size, &offset, NULL, __dummy_buf))
1250 return 0;
1251
1252 /* Check there is enough space for a relative jump. */
1253 if (size - offset < RELATIVEJUMP_SIZE)
1254 return 0;
1255
1256 /* Decode instructions */
1257 addr = paddr - offset;
1258 while (addr < paddr - offset + size) { /* Decode until function end */
1259 if (search_exception_tables(addr))
1260 /*
1261 * Since some fixup code will jumps into this function,
1262 * we can't optimize kprobe in this function.
1263 */
1264 return 0;
1265 kernel_insn_init(&insn, (void *)addr);
1266 insn_get_opcode(&insn);
1267 if (insn.opcode.bytes[0] == BREAKPOINT_INSTRUCTION) {
1268 ret = recover_probed_instruction(buf, addr);
1269 if (ret)
1270 return 0;
1271 kernel_insn_init(&insn, buf);
1272 }
1273 insn_get_length(&insn);
1274 /* Recover address */
1275 insn.kaddr = (void *)addr;
1276 insn.next_byte = (void *)(addr + insn.length);
1277 /* Check any instructions don't jump into target */
1278 if (insn_is_indirect_jump(&insn) ||
1279 insn_jump_into_range(&insn, paddr + INT3_SIZE,
1280 RELATIVE_ADDR_SIZE))
1281 return 0;
1282 addr += insn.length;
1283 }
1284
1285 return 1;
1286}
1287
1288/* Check optimized_kprobe can actually be optimized. */
1289int __kprobes arch_check_optimized_kprobe(struct optimized_kprobe *op)
1290{
1291 int i;
1292 struct kprobe *p;
1293
1294 for (i = 1; i < op->optinsn.size; i++) {
1295 p = get_kprobe(op->kp.addr + i);
1296 if (p && !kprobe_disabled(p))
1297 return -EEXIST;
1298 }
1299
1300 return 0;
1301}
1302
1303/* Check the addr is within the optimized instructions. */
1304int __kprobes arch_within_optimized_kprobe(struct optimized_kprobe *op,
1305 unsigned long addr)
1306{
1307 return ((unsigned long)op->kp.addr <= addr &&
1308 (unsigned long)op->kp.addr + op->optinsn.size > addr);
1309}
1310
1311/* Free optimized instruction slot */
1312static __kprobes
1313void __arch_remove_optimized_kprobe(struct optimized_kprobe *op, int dirty)
1314{
1315 if (op->optinsn.insn) {
1316 free_optinsn_slot(op->optinsn.insn, dirty);
1317 op->optinsn.insn = NULL;
1318 op->optinsn.size = 0;
1319 }
1320}
1321
1322void __kprobes arch_remove_optimized_kprobe(struct optimized_kprobe *op)
1323{
1324 __arch_remove_optimized_kprobe(op, 1);
1325}
1326
1327/*
1328 * Copy replacing target instructions
1329 * Target instructions MUST be relocatable (checked inside)
1330 */
1331int __kprobes arch_prepare_optimized_kprobe(struct optimized_kprobe *op)
1332{
1333 u8 *buf;
1334 int ret;
1335 long rel;
1336
1337 if (!can_optimize((unsigned long)op->kp.addr))
1338 return -EILSEQ;
1339
1340 op->optinsn.insn = get_optinsn_slot();
1341 if (!op->optinsn.insn)
1342 return -ENOMEM;
1343
1344 /*
1345 * Verify if the address gap is in 2GB range, because this uses
1346 * a relative jump.
1347 */
1348 rel = (long)op->optinsn.insn - (long)op->kp.addr + RELATIVEJUMP_SIZE;
1349 if (abs(rel) > 0x7fffffff)
1350 return -ERANGE;
1351
1352 buf = (u8 *)op->optinsn.insn;
1353
1354 /* Copy instructions into the out-of-line buffer */
1355 ret = copy_optimized_instructions(buf + TMPL_END_IDX, op->kp.addr);
1356 if (ret < 0) {
1357 __arch_remove_optimized_kprobe(op, 0);
1358 return ret;
1359 }
1360 op->optinsn.size = ret;
1361
1362 /* Copy arch-dep-instance from template */
1363 memcpy(buf, &optprobe_template_entry, TMPL_END_IDX);
1364
1365 /* Set probe information */
1366 synthesize_set_arg1(buf + TMPL_MOVE_IDX, (unsigned long)op);
1367
1368 /* Set probe function call */
1369 synthesize_relcall(buf + TMPL_CALL_IDX, optimized_callback);
1370
1371 /* Set returning jmp instruction at the tail of out-of-line buffer */
1372 synthesize_reljump(buf + TMPL_END_IDX + op->optinsn.size,
1373 (u8 *)op->kp.addr + op->optinsn.size);
1374
1375 flush_icache_range((unsigned long) buf,
1376 (unsigned long) buf + TMPL_END_IDX +
1377 op->optinsn.size + RELATIVEJUMP_SIZE);
1378 return 0;
1379}
1380
1381/* Replace a breakpoint (int3) with a relative jump. */
1382int __kprobes arch_optimize_kprobe(struct optimized_kprobe *op)
1383{
1384 unsigned char jmp_code[RELATIVEJUMP_SIZE];
1385 s32 rel = (s32)((long)op->optinsn.insn -
1386 ((long)op->kp.addr + RELATIVEJUMP_SIZE));
1387
1388 /* Backup instructions which will be replaced by jump address */
1389 memcpy(op->optinsn.copied_insn, op->kp.addr + INT3_SIZE,
1390 RELATIVE_ADDR_SIZE);
1391
1392 jmp_code[0] = RELATIVEJUMP_OPCODE;
1393 *(s32 *)(&jmp_code[1]) = rel;
1394
1395 /*
1396 * text_poke_smp doesn't support NMI/MCE code modifying.
1397 * However, since kprobes itself also doesn't support NMI/MCE
1398 * code probing, it's not a problem.
1399 */
1400 text_poke_smp(op->kp.addr, jmp_code, RELATIVEJUMP_SIZE);
1401 return 0;
1402}
1403
1404/* Replace a relative jump with a breakpoint (int3). */
1405void __kprobes arch_unoptimize_kprobe(struct optimized_kprobe *op)
1406{
1407 u8 buf[RELATIVEJUMP_SIZE];
1408
1409 /* Set int3 to first byte for kprobes */
1410 buf[0] = BREAKPOINT_INSTRUCTION;
1411 memcpy(buf + 1, op->optinsn.copied_insn, RELATIVE_ADDR_SIZE);
1412 text_poke_smp(op->kp.addr, buf, RELATIVEJUMP_SIZE);
1413}
1414
1415static int __kprobes setup_detour_execution(struct kprobe *p,
1416 struct pt_regs *regs,
1417 int reenter)
1418{
1419 struct optimized_kprobe *op;
1420
1421 if (p->flags & KPROBE_FLAG_OPTIMIZED) {
1422 /* This kprobe is really able to run optimized path. */
1423 op = container_of(p, struct optimized_kprobe, kp);
1424 /* Detour through copied instructions */
1425 regs->ip = (unsigned long)op->optinsn.insn + TMPL_END_IDX;
1426 if (!reenter)
1427 reset_current_kprobe();
1428 preempt_enable_no_resched();
1429 return 1;
1430 }
1431 return 0;
1432}
1433#endif
1434
1036int __init arch_init_kprobes(void) 1435int __init arch_init_kprobes(void)
1037{ 1436{
1038 return 0; 1437 return 0;