aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kernel/ptrace_64.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kernel/ptrace_64.c')
-rw-r--r--arch/x86/kernel/ptrace_64.c627
1 files changed, 627 insertions, 0 deletions
diff --git a/arch/x86/kernel/ptrace_64.c b/arch/x86/kernel/ptrace_64.c
new file mode 100644
index 000000000000..eea3702427b4
--- /dev/null
+++ b/arch/x86/kernel/ptrace_64.c
@@ -0,0 +1,627 @@
1/* ptrace.c */
2/* By Ross Biro 1/23/92 */
3/*
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
6 *
7 * x86-64 port 2000-2002 Andi Kleen
8 */
9
10#include <linux/kernel.h>
11#include <linux/sched.h>
12#include <linux/mm.h>
13#include <linux/smp.h>
14#include <linux/errno.h>
15#include <linux/ptrace.h>
16#include <linux/user.h>
17#include <linux/security.h>
18#include <linux/audit.h>
19#include <linux/seccomp.h>
20#include <linux/signal.h>
21
22#include <asm/uaccess.h>
23#include <asm/pgtable.h>
24#include <asm/system.h>
25#include <asm/processor.h>
26#include <asm/i387.h>
27#include <asm/debugreg.h>
28#include <asm/ldt.h>
29#include <asm/desc.h>
30#include <asm/proto.h>
31#include <asm/ia32.h>
32
33/*
34 * does not yet catch signals sent when the child dies.
35 * in exit.c or in signal.c.
36 */
37
38/*
39 * Determines which flags the user has access to [1 = access, 0 = no access].
40 * Prohibits changing ID(21), VIP(20), VIF(19), VM(17), IOPL(12-13), IF(9).
41 * Also masks reserved bits (63-22, 15, 5, 3, 1).
42 */
43#define FLAG_MASK 0x54dd5UL
44
45/* set's the trap flag. */
46#define TRAP_FLAG 0x100UL
47
48/*
49 * eflags and offset of eflags on child stack..
50 */
51#define EFLAGS offsetof(struct pt_regs, eflags)
52#define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
53
54/*
55 * this routine will get a word off of the processes privileged stack.
56 * the offset is how far from the base addr as stored in the TSS.
57 * this routine assumes that all the privileged stacks are in our
58 * data space.
59 */
60static inline unsigned long get_stack_long(struct task_struct *task, int offset)
61{
62 unsigned char *stack;
63
64 stack = (unsigned char *)task->thread.rsp0;
65 stack += offset;
66 return (*((unsigned long *)stack));
67}
68
69/*
70 * this routine will put a word on the processes privileged stack.
71 * the offset is how far from the base addr as stored in the TSS.
72 * this routine assumes that all the privileged stacks are in our
73 * data space.
74 */
75static inline long put_stack_long(struct task_struct *task, int offset,
76 unsigned long data)
77{
78 unsigned char * stack;
79
80 stack = (unsigned char *) task->thread.rsp0;
81 stack += offset;
82 *(unsigned long *) stack = data;
83 return 0;
84}
85
86#define LDT_SEGMENT 4
87
88unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
89{
90 unsigned long addr, seg;
91
92 addr = regs->rip;
93 seg = regs->cs & 0xffff;
94
95 /*
96 * We'll assume that the code segments in the GDT
97 * are all zero-based. That is largely true: the
98 * TLS segments are used for data, and the PNPBIOS
99 * and APM bios ones we just ignore here.
100 */
101 if (seg & LDT_SEGMENT) {
102 u32 *desc;
103 unsigned long base;
104
105 seg &= ~7UL;
106
107 down(&child->mm->context.sem);
108 if (unlikely((seg >> 3) >= child->mm->context.size))
109 addr = -1L; /* bogus selector, access would fault */
110 else {
111 desc = child->mm->context.ldt + seg;
112 base = ((desc[0] >> 16) |
113 ((desc[1] & 0xff) << 16) |
114 (desc[1] & 0xff000000));
115
116 /* 16-bit code segment? */
117 if (!((desc[1] >> 22) & 1))
118 addr &= 0xffff;
119 addr += base;
120 }
121 up(&child->mm->context.sem);
122 }
123
124 return addr;
125}
126
127static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
128{
129 int i, copied;
130 unsigned char opcode[15];
131 unsigned long addr = convert_rip_to_linear(child, regs);
132
133 copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
134 for (i = 0; i < copied; i++) {
135 switch (opcode[i]) {
136 /* popf and iret */
137 case 0x9d: case 0xcf:
138 return 1;
139
140 /* CHECKME: 64 65 */
141
142 /* opcode and address size prefixes */
143 case 0x66: case 0x67:
144 continue;
145 /* irrelevant prefixes (segment overrides and repeats) */
146 case 0x26: case 0x2e:
147 case 0x36: case 0x3e:
148 case 0x64: case 0x65:
149 case 0xf2: case 0xf3:
150 continue;
151
152 case 0x40 ... 0x4f:
153 if (regs->cs != __USER_CS)
154 /* 32-bit mode: register increment */
155 return 0;
156 /* 64-bit mode: REX prefix */
157 continue;
158
159 /* CHECKME: f2, f3 */
160
161 /*
162 * pushf: NOTE! We should probably not let
163 * the user see the TF bit being set. But
164 * it's more pain than it's worth to avoid
165 * it, and a debugger could emulate this
166 * all in user space if it _really_ cares.
167 */
168 case 0x9c:
169 default:
170 return 0;
171 }
172 }
173 return 0;
174}
175
176static void set_singlestep(struct task_struct *child)
177{
178 struct pt_regs *regs = task_pt_regs(child);
179
180 /*
181 * Always set TIF_SINGLESTEP - this guarantees that
182 * we single-step system calls etc.. This will also
183 * cause us to set TF when returning to user mode.
184 */
185 set_tsk_thread_flag(child, TIF_SINGLESTEP);
186
187 /*
188 * If TF was already set, don't do anything else
189 */
190 if (regs->eflags & TRAP_FLAG)
191 return;
192
193 /* Set TF on the kernel stack.. */
194 regs->eflags |= TRAP_FLAG;
195
196 /*
197 * ..but if TF is changed by the instruction we will trace,
198 * don't mark it as being "us" that set it, so that we
199 * won't clear it by hand later.
200 */
201 if (is_setting_trap_flag(child, regs))
202 return;
203
204 child->ptrace |= PT_DTRACE;
205}
206
207static void clear_singlestep(struct task_struct *child)
208{
209 /* Always clear TIF_SINGLESTEP... */
210 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
211
212 /* But touch TF only if it was set by us.. */
213 if (child->ptrace & PT_DTRACE) {
214 struct pt_regs *regs = task_pt_regs(child);
215 regs->eflags &= ~TRAP_FLAG;
216 child->ptrace &= ~PT_DTRACE;
217 }
218}
219
220/*
221 * Called by kernel/ptrace.c when detaching..
222 *
223 * Make sure the single step bit is not set.
224 */
225void ptrace_disable(struct task_struct *child)
226{
227 clear_singlestep(child);
228}
229
230static int putreg(struct task_struct *child,
231 unsigned long regno, unsigned long value)
232{
233 unsigned long tmp;
234
235 switch (regno) {
236 case offsetof(struct user_regs_struct,fs):
237 if (value && (value & 3) != 3)
238 return -EIO;
239 child->thread.fsindex = value & 0xffff;
240 return 0;
241 case offsetof(struct user_regs_struct,gs):
242 if (value && (value & 3) != 3)
243 return -EIO;
244 child->thread.gsindex = value & 0xffff;
245 return 0;
246 case offsetof(struct user_regs_struct,ds):
247 if (value && (value & 3) != 3)
248 return -EIO;
249 child->thread.ds = value & 0xffff;
250 return 0;
251 case offsetof(struct user_regs_struct,es):
252 if (value && (value & 3) != 3)
253 return -EIO;
254 child->thread.es = value & 0xffff;
255 return 0;
256 case offsetof(struct user_regs_struct,ss):
257 if ((value & 3) != 3)
258 return -EIO;
259 value &= 0xffff;
260 return 0;
261 case offsetof(struct user_regs_struct,fs_base):
262 if (value >= TASK_SIZE_OF(child))
263 return -EIO;
264 child->thread.fs = value;
265 return 0;
266 case offsetof(struct user_regs_struct,gs_base):
267 if (value >= TASK_SIZE_OF(child))
268 return -EIO;
269 child->thread.gs = value;
270 return 0;
271 case offsetof(struct user_regs_struct, eflags):
272 value &= FLAG_MASK;
273 tmp = get_stack_long(child, EFL_OFFSET);
274 tmp &= ~FLAG_MASK;
275 value |= tmp;
276 break;
277 case offsetof(struct user_regs_struct,cs):
278 if ((value & 3) != 3)
279 return -EIO;
280 value &= 0xffff;
281 break;
282 }
283 put_stack_long(child, regno - sizeof(struct pt_regs), value);
284 return 0;
285}
286
287static unsigned long getreg(struct task_struct *child, unsigned long regno)
288{
289 unsigned long val;
290 switch (regno) {
291 case offsetof(struct user_regs_struct, fs):
292 return child->thread.fsindex;
293 case offsetof(struct user_regs_struct, gs):
294 return child->thread.gsindex;
295 case offsetof(struct user_regs_struct, ds):
296 return child->thread.ds;
297 case offsetof(struct user_regs_struct, es):
298 return child->thread.es;
299 case offsetof(struct user_regs_struct, fs_base):
300 return child->thread.fs;
301 case offsetof(struct user_regs_struct, gs_base):
302 return child->thread.gs;
303 default:
304 regno = regno - sizeof(struct pt_regs);
305 val = get_stack_long(child, regno);
306 if (test_tsk_thread_flag(child, TIF_IA32))
307 val &= 0xffffffff;
308 return val;
309 }
310
311}
312
313long arch_ptrace(struct task_struct *child, long request, long addr, long data)
314{
315 long i, ret;
316 unsigned ui;
317
318 switch (request) {
319 /* when I and D space are separate, these will need to be fixed. */
320 case PTRACE_PEEKTEXT: /* read word at location addr. */
321 case PTRACE_PEEKDATA:
322 ret = generic_ptrace_peekdata(child, addr, data);
323 break;
324
325 /* read the word at location addr in the USER area. */
326 case PTRACE_PEEKUSR: {
327 unsigned long tmp;
328
329 ret = -EIO;
330 if ((addr & 7) ||
331 addr > sizeof(struct user) - 7)
332 break;
333
334 switch (addr) {
335 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
336 tmp = getreg(child, addr);
337 break;
338 case offsetof(struct user, u_debugreg[0]):
339 tmp = child->thread.debugreg0;
340 break;
341 case offsetof(struct user, u_debugreg[1]):
342 tmp = child->thread.debugreg1;
343 break;
344 case offsetof(struct user, u_debugreg[2]):
345 tmp = child->thread.debugreg2;
346 break;
347 case offsetof(struct user, u_debugreg[3]):
348 tmp = child->thread.debugreg3;
349 break;
350 case offsetof(struct user, u_debugreg[6]):
351 tmp = child->thread.debugreg6;
352 break;
353 case offsetof(struct user, u_debugreg[7]):
354 tmp = child->thread.debugreg7;
355 break;
356 default:
357 tmp = 0;
358 break;
359 }
360 ret = put_user(tmp,(unsigned long __user *) data);
361 break;
362 }
363
364 /* when I and D space are separate, this will have to be fixed. */
365 case PTRACE_POKETEXT: /* write the word at location addr. */
366 case PTRACE_POKEDATA:
367 ret = generic_ptrace_pokedata(child, addr, data);
368 break;
369
370 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
371 {
372 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
373 ret = -EIO;
374 if ((addr & 7) ||
375 addr > sizeof(struct user) - 7)
376 break;
377
378 switch (addr) {
379 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
380 ret = putreg(child, addr, data);
381 break;
382 /* Disallows to set a breakpoint into the vsyscall */
383 case offsetof(struct user, u_debugreg[0]):
384 if (data >= TASK_SIZE_OF(child) - dsize) break;
385 child->thread.debugreg0 = data;
386 ret = 0;
387 break;
388 case offsetof(struct user, u_debugreg[1]):
389 if (data >= TASK_SIZE_OF(child) - dsize) break;
390 child->thread.debugreg1 = data;
391 ret = 0;
392 break;
393 case offsetof(struct user, u_debugreg[2]):
394 if (data >= TASK_SIZE_OF(child) - dsize) break;
395 child->thread.debugreg2 = data;
396 ret = 0;
397 break;
398 case offsetof(struct user, u_debugreg[3]):
399 if (data >= TASK_SIZE_OF(child) - dsize) break;
400 child->thread.debugreg3 = data;
401 ret = 0;
402 break;
403 case offsetof(struct user, u_debugreg[6]):
404 if (data >> 32)
405 break;
406 child->thread.debugreg6 = data;
407 ret = 0;
408 break;
409 case offsetof(struct user, u_debugreg[7]):
410 /* See arch/i386/kernel/ptrace.c for an explanation of
411 * this awkward check.*/
412 data &= ~DR_CONTROL_RESERVED;
413 for(i=0; i<4; i++)
414 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
415 break;
416 if (i == 4) {
417 child->thread.debugreg7 = data;
418 if (data)
419 set_tsk_thread_flag(child, TIF_DEBUG);
420 else
421 clear_tsk_thread_flag(child, TIF_DEBUG);
422 ret = 0;
423 }
424 break;
425 }
426 break;
427 }
428 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
429 case PTRACE_CONT: /* restart after signal. */
430
431 ret = -EIO;
432 if (!valid_signal(data))
433 break;
434 if (request == PTRACE_SYSCALL)
435 set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
436 else
437 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
438 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
439 child->exit_code = data;
440 /* make sure the single step bit is not set. */
441 clear_singlestep(child);
442 wake_up_process(child);
443 ret = 0;
444 break;
445
446#ifdef CONFIG_IA32_EMULATION
447 /* This makes only sense with 32bit programs. Allow a
448 64bit debugger to fully examine them too. Better
449 don't use it against 64bit processes, use
450 PTRACE_ARCH_PRCTL instead. */
451 case PTRACE_SET_THREAD_AREA: {
452 struct user_desc __user *p;
453 int old;
454 p = (struct user_desc __user *)data;
455 get_user(old, &p->entry_number);
456 put_user(addr, &p->entry_number);
457 ret = do_set_thread_area(&child->thread, p);
458 put_user(old, &p->entry_number);
459 break;
460 case PTRACE_GET_THREAD_AREA:
461 p = (struct user_desc __user *)data;
462 get_user(old, &p->entry_number);
463 put_user(addr, &p->entry_number);
464 ret = do_get_thread_area(&child->thread, p);
465 put_user(old, &p->entry_number);
466 break;
467 }
468#endif
469 /* normal 64bit interface to access TLS data.
470 Works just like arch_prctl, except that the arguments
471 are reversed. */
472 case PTRACE_ARCH_PRCTL:
473 ret = do_arch_prctl(child, data, addr);
474 break;
475
476/*
477 * make the child exit. Best I can do is send it a sigkill.
478 * perhaps it should be put in the status that it wants to
479 * exit.
480 */
481 case PTRACE_KILL:
482 ret = 0;
483 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
484 break;
485 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
486 child->exit_code = SIGKILL;
487 /* make sure the single step bit is not set. */
488 clear_singlestep(child);
489 wake_up_process(child);
490 break;
491
492 case PTRACE_SINGLESTEP: /* set the trap flag. */
493 ret = -EIO;
494 if (!valid_signal(data))
495 break;
496 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
497 set_singlestep(child);
498 child->exit_code = data;
499 /* give it a chance to run. */
500 wake_up_process(child);
501 ret = 0;
502 break;
503
504 case PTRACE_DETACH:
505 /* detach a process that was attached. */
506 ret = ptrace_detach(child, data);
507 break;
508
509 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
510 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
511 sizeof(struct user_regs_struct))) {
512 ret = -EIO;
513 break;
514 }
515 ret = 0;
516 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
517 ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
518 data += sizeof(long);
519 }
520 break;
521 }
522
523 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
524 unsigned long tmp;
525 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
526 sizeof(struct user_regs_struct))) {
527 ret = -EIO;
528 break;
529 }
530 ret = 0;
531 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
532 ret = __get_user(tmp, (unsigned long __user *) data);
533 if (ret)
534 break;
535 ret = putreg(child, ui, tmp);
536 if (ret)
537 break;
538 data += sizeof(long);
539 }
540 break;
541 }
542
543 case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
544 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
545 sizeof(struct user_i387_struct))) {
546 ret = -EIO;
547 break;
548 }
549 ret = get_fpregs((struct user_i387_struct __user *)data, child);
550 break;
551 }
552
553 case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
554 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
555 sizeof(struct user_i387_struct))) {
556 ret = -EIO;
557 break;
558 }
559 set_stopped_child_used_math(child);
560 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
561 break;
562 }
563
564 default:
565 ret = ptrace_request(child, request, addr, data);
566 break;
567 }
568 return ret;
569}
570
571static void syscall_trace(struct pt_regs *regs)
572{
573
574#if 0
575 printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
576 current->comm,
577 regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
578 current_thread_info()->flags, current->ptrace);
579#endif
580
581 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
582 ? 0x80 : 0));
583 /*
584 * this isn't the same as continuing with a signal, but it will do
585 * for normal use. strace only continues with a signal if the
586 * stopping signal is not SIGTRAP. -brl
587 */
588 if (current->exit_code) {
589 send_sig(current->exit_code, current, 1);
590 current->exit_code = 0;
591 }
592}
593
594asmlinkage void syscall_trace_enter(struct pt_regs *regs)
595{
596 /* do the secure computing check first */
597 secure_computing(regs->orig_rax);
598
599 if (test_thread_flag(TIF_SYSCALL_TRACE)
600 && (current->ptrace & PT_PTRACED))
601 syscall_trace(regs);
602
603 if (unlikely(current->audit_context)) {
604 if (test_thread_flag(TIF_IA32)) {
605 audit_syscall_entry(AUDIT_ARCH_I386,
606 regs->orig_rax,
607 regs->rbx, regs->rcx,
608 regs->rdx, regs->rsi);
609 } else {
610 audit_syscall_entry(AUDIT_ARCH_X86_64,
611 regs->orig_rax,
612 regs->rdi, regs->rsi,
613 regs->rdx, regs->r10);
614 }
615 }
616}
617
618asmlinkage void syscall_trace_leave(struct pt_regs *regs)
619{
620 if (unlikely(current->audit_context))
621 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
622
623 if ((test_thread_flag(TIF_SYSCALL_TRACE)
624 || test_thread_flag(TIF_SINGLESTEP))
625 && (current->ptrace & PT_PTRACED))
626 syscall_trace(regs);
627}