[PATCH] consolidate sys_ptrace()
[linux-2.6/libata-dev.git] / arch / i386 / kernel / ptrace.c
blob5ffbb4b7ad059a038b7273de448b8e5f97ac5c20
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 */
8 #include <linux/kernel.h>
9 #include <linux/sched.h>
10 #include <linux/mm.h>
11 #include <linux/smp.h>
12 #include <linux/smp_lock.h>
13 #include <linux/errno.h>
14 #include <linux/ptrace.h>
15 #include <linux/user.h>
16 #include <linux/security.h>
17 #include <linux/audit.h>
18 #include <linux/seccomp.h>
19 #include <linux/signal.h>
21 #include <asm/uaccess.h>
22 #include <asm/pgtable.h>
23 #include <asm/system.h>
24 #include <asm/processor.h>
25 #include <asm/i387.h>
26 #include <asm/debugreg.h>
27 #include <asm/ldt.h>
28 #include <asm/desc.h>
31 * does not yet catch signals sent when the child dies.
32 * in exit.c or in signal.c.
35 /* determines which flags the user has access to. */
36 /* 1 = access 0 = no access */
37 #define FLAG_MASK 0x00044dd5
39 /* set's the trap flag. */
40 #define TRAP_FLAG 0x100
43 * Offset of eflags on child stack..
45 #define EFL_OFFSET ((EFL-2)*4-sizeof(struct pt_regs))
47 static inline struct pt_regs *get_child_regs(struct task_struct *task)
49 void *stack_top = (void *)task->thread.esp0;
50 return stack_top - sizeof(struct pt_regs);
54 * this routine will get a word off of the processes privileged stack.
55 * the offset is how far from the base addr as stored in the TSS.
56 * this routine assumes that all the privileged stacks are in our
57 * data space.
58 */
59 static inline int get_stack_long(struct task_struct *task, int offset)
61 unsigned char *stack;
63 stack = (unsigned char *)task->thread.esp0;
64 stack += offset;
65 return (*((int *)stack));
69 * this routine will put a word on the processes privileged stack.
70 * the offset is how far from the base addr as stored in the TSS.
71 * this routine assumes that all the privileged stacks are in our
72 * data space.
74 static inline int put_stack_long(struct task_struct *task, int offset,
75 unsigned long data)
77 unsigned char * stack;
79 stack = (unsigned char *) task->thread.esp0;
80 stack += offset;
81 *(unsigned long *) stack = data;
82 return 0;
85 static int putreg(struct task_struct *child,
86 unsigned long regno, unsigned long value)
88 switch (regno >> 2) {
89 case FS:
90 if (value && (value & 3) != 3)
91 return -EIO;
92 child->thread.fs = value;
93 return 0;
94 case GS:
95 if (value && (value & 3) != 3)
96 return -EIO;
97 child->thread.gs = value;
98 return 0;
99 case DS:
100 case ES:
101 if (value && (value & 3) != 3)
102 return -EIO;
103 value &= 0xffff;
104 break;
105 case SS:
106 case CS:
107 if ((value & 3) != 3)
108 return -EIO;
109 value &= 0xffff;
110 break;
111 case EFL:
112 value &= FLAG_MASK;
113 value |= get_stack_long(child, EFL_OFFSET) & ~FLAG_MASK;
114 break;
116 if (regno > GS*4)
117 regno -= 2*4;
118 put_stack_long(child, regno - sizeof(struct pt_regs), value);
119 return 0;
122 static unsigned long getreg(struct task_struct *child,
123 unsigned long regno)
125 unsigned long retval = ~0UL;
127 switch (regno >> 2) {
128 case FS:
129 retval = child->thread.fs;
130 break;
131 case GS:
132 retval = child->thread.gs;
133 break;
134 case DS:
135 case ES:
136 case SS:
137 case CS:
138 retval = 0xffff;
139 /* fall through */
140 default:
141 if (regno > GS*4)
142 regno -= 2*4;
143 regno = regno - sizeof(struct pt_regs);
144 retval &= get_stack_long(child, regno);
146 return retval;
149 #define LDT_SEGMENT 4
151 static unsigned long convert_eip_to_linear(struct task_struct *child, struct pt_regs *regs)
153 unsigned long addr, seg;
155 addr = regs->eip;
156 seg = regs->xcs & 0xffff;
157 if (regs->eflags & VM_MASK) {
158 addr = (addr & 0xffff) + (seg << 4);
159 return addr;
163 * We'll assume that the code segments in the GDT
164 * are all zero-based. That is largely true: the
165 * TLS segments are used for data, and the PNPBIOS
166 * and APM bios ones we just ignore here.
168 if (seg & LDT_SEGMENT) {
169 u32 *desc;
170 unsigned long base;
172 down(&child->mm->context.sem);
173 desc = child->mm->context.ldt + (seg & ~7);
174 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
176 /* 16-bit code segment? */
177 if (!((desc[1] >> 22) & 1))
178 addr &= 0xffff;
179 addr += base;
180 up(&child->mm->context.sem);
182 return addr;
185 static inline int is_at_popf(struct task_struct *child, struct pt_regs *regs)
187 int i, copied;
188 unsigned char opcode[16];
189 unsigned long addr = convert_eip_to_linear(child, regs);
191 copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
192 for (i = 0; i < copied; i++) {
193 switch (opcode[i]) {
194 /* popf */
195 case 0x9d:
196 return 1;
197 /* opcode and address size prefixes */
198 case 0x66: case 0x67:
199 continue;
200 /* irrelevant prefixes (segment overrides and repeats) */
201 case 0x26: case 0x2e:
202 case 0x36: case 0x3e:
203 case 0x64: case 0x65:
204 case 0xf0: case 0xf2: case 0xf3:
205 continue;
208 * pushf: NOTE! We should probably not let
209 * the user see the TF bit being set. But
210 * it's more pain than it's worth to avoid
211 * it, and a debugger could emulate this
212 * all in user space if it _really_ cares.
214 case 0x9c:
215 default:
216 return 0;
219 return 0;
222 static void set_singlestep(struct task_struct *child)
224 struct pt_regs *regs = get_child_regs(child);
227 * Always set TIF_SINGLESTEP - this guarantees that
228 * we single-step system calls etc.. This will also
229 * cause us to set TF when returning to user mode.
231 set_tsk_thread_flag(child, TIF_SINGLESTEP);
234 * If TF was already set, don't do anything else
236 if (regs->eflags & TRAP_FLAG)
237 return;
239 /* Set TF on the kernel stack.. */
240 regs->eflags |= TRAP_FLAG;
243 * ..but if TF is changed by the instruction we will trace,
244 * don't mark it as being "us" that set it, so that we
245 * won't clear it by hand later.
247 if (is_at_popf(child, regs))
248 return;
250 child->ptrace |= PT_DTRACE;
253 static void clear_singlestep(struct task_struct *child)
255 /* Always clear TIF_SINGLESTEP... */
256 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
258 /* But touch TF only if it was set by us.. */
259 if (child->ptrace & PT_DTRACE) {
260 struct pt_regs *regs = get_child_regs(child);
261 regs->eflags &= ~TRAP_FLAG;
262 child->ptrace &= ~PT_DTRACE;
267 * Called by kernel/ptrace.c when detaching..
269 * Make sure the single step bit is not set.
271 void ptrace_disable(struct task_struct *child)
273 clear_singlestep(child);
274 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
275 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
279 * Perform get_thread_area on behalf of the traced child.
281 static int
282 ptrace_get_thread_area(struct task_struct *child,
283 int idx, struct user_desc __user *user_desc)
285 struct user_desc info;
286 struct desc_struct *desc;
289 * Get the current Thread-Local Storage area:
292 #define GET_BASE(desc) ( \
293 (((desc)->a >> 16) & 0x0000ffff) | \
294 (((desc)->b << 16) & 0x00ff0000) | \
295 ( (desc)->b & 0xff000000) )
297 #define GET_LIMIT(desc) ( \
298 ((desc)->a & 0x0ffff) | \
299 ((desc)->b & 0xf0000) )
301 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
302 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
303 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
304 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
305 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
306 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
308 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
309 return -EINVAL;
311 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
313 info.entry_number = idx;
314 info.base_addr = GET_BASE(desc);
315 info.limit = GET_LIMIT(desc);
316 info.seg_32bit = GET_32BIT(desc);
317 info.contents = GET_CONTENTS(desc);
318 info.read_exec_only = !GET_WRITABLE(desc);
319 info.limit_in_pages = GET_LIMIT_PAGES(desc);
320 info.seg_not_present = !GET_PRESENT(desc);
321 info.useable = GET_USEABLE(desc);
323 if (copy_to_user(user_desc, &info, sizeof(info)))
324 return -EFAULT;
326 return 0;
330 * Perform set_thread_area on behalf of the traced child.
332 static int
333 ptrace_set_thread_area(struct task_struct *child,
334 int idx, struct user_desc __user *user_desc)
336 struct user_desc info;
337 struct desc_struct *desc;
339 if (copy_from_user(&info, user_desc, sizeof(info)))
340 return -EFAULT;
342 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
343 return -EINVAL;
345 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
346 if (LDT_empty(&info)) {
347 desc->a = 0;
348 desc->b = 0;
349 } else {
350 desc->a = LDT_entry_a(&info);
351 desc->b = LDT_entry_b(&info);
354 return 0;
357 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
359 struct user * dummy = NULL;
360 int i, ret;
361 unsigned long __user *datap = (unsigned long __user *)data;
363 switch (request) {
364 /* when I and D space are separate, these will need to be fixed. */
365 case PTRACE_PEEKTEXT: /* read word at location addr. */
366 case PTRACE_PEEKDATA: {
367 unsigned long tmp;
368 int copied;
370 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
371 ret = -EIO;
372 if (copied != sizeof(tmp))
373 break;
374 ret = put_user(tmp, datap);
375 break;
378 /* read the word at location addr in the USER area. */
379 case PTRACE_PEEKUSR: {
380 unsigned long tmp;
382 ret = -EIO;
383 if ((addr & 3) || addr < 0 ||
384 addr > sizeof(struct user) - 3)
385 break;
387 tmp = 0; /* Default return condition */
388 if(addr < FRAME_SIZE*sizeof(long))
389 tmp = getreg(child, addr);
390 if(addr >= (long) &dummy->u_debugreg[0] &&
391 addr <= (long) &dummy->u_debugreg[7]){
392 addr -= (long) &dummy->u_debugreg[0];
393 addr = addr >> 2;
394 tmp = child->thread.debugreg[addr];
396 ret = put_user(tmp, datap);
397 break;
400 /* when I and D space are separate, this will have to be fixed. */
401 case PTRACE_POKETEXT: /* write the word at location addr. */
402 case PTRACE_POKEDATA:
403 ret = 0;
404 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
405 break;
406 ret = -EIO;
407 break;
409 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
410 ret = -EIO;
411 if ((addr & 3) || addr < 0 ||
412 addr > sizeof(struct user) - 3)
413 break;
415 if (addr < FRAME_SIZE*sizeof(long)) {
416 ret = putreg(child, addr, data);
417 break;
419 /* We need to be very careful here. We implicitly
420 want to modify a portion of the task_struct, and we
421 have to be selective about what portions we allow someone
422 to modify. */
424 ret = -EIO;
425 if(addr >= (long) &dummy->u_debugreg[0] &&
426 addr <= (long) &dummy->u_debugreg[7]){
428 if(addr == (long) &dummy->u_debugreg[4]) break;
429 if(addr == (long) &dummy->u_debugreg[5]) break;
430 if(addr < (long) &dummy->u_debugreg[4] &&
431 ((unsigned long) data) >= TASK_SIZE-3) break;
433 /* Sanity-check data. Take one half-byte at once with
434 * check = (val >> (16 + 4*i)) & 0xf. It contains the
435 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
436 * 2 and 3 are LENi. Given a list of invalid values,
437 * we do mask |= 1 << invalid_value, so that
438 * (mask >> check) & 1 is a correct test for invalid
439 * values.
441 * R/Wi contains the type of the breakpoint /
442 * watchpoint, LENi contains the length of the watched
443 * data in the watchpoint case.
445 * The invalid values are:
446 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
447 * - R/Wi == 0x10 (break on I/O reads or writes), so
448 * mask |= 0x4444.
449 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
450 * 0x1110.
452 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
454 * See the Intel Manual "System Programming Guide",
455 * 15.2.4
457 * Note that LENi == 0x10 is defined on x86_64 in long
458 * mode (i.e. even for 32-bit userspace software, but
459 * 64-bit kernel), so the x86_64 mask value is 0x5454.
460 * See the AMD manual no. 24593 (AMD64 System
461 * Programming)*/
463 if(addr == (long) &dummy->u_debugreg[7]) {
464 data &= ~DR_CONTROL_RESERVED;
465 for(i=0; i<4; i++)
466 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
467 goto out_tsk;
470 addr -= (long) &dummy->u_debugreg;
471 addr = addr >> 2;
472 child->thread.debugreg[addr] = data;
473 ret = 0;
475 break;
477 case PTRACE_SYSEMU: /* continue and stop at next syscall, which will not be executed */
478 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
479 case PTRACE_CONT: /* restart after signal. */
480 ret = -EIO;
481 if (!valid_signal(data))
482 break;
483 if (request == PTRACE_SYSEMU) {
484 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
485 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
486 } else if (request == PTRACE_SYSCALL) {
487 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
488 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
489 } else {
490 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
491 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
493 child->exit_code = data;
494 /* make sure the single step bit is not set. */
495 clear_singlestep(child);
496 wake_up_process(child);
497 ret = 0;
498 break;
501 * make the child exit. Best I can do is send it a sigkill.
502 * perhaps it should be put in the status that it wants to
503 * exit.
505 case PTRACE_KILL:
506 ret = 0;
507 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
508 break;
509 child->exit_code = SIGKILL;
510 /* make sure the single step bit is not set. */
511 clear_singlestep(child);
512 wake_up_process(child);
513 break;
515 case PTRACE_SYSEMU_SINGLESTEP: /* Same as SYSEMU, but singlestep if not syscall */
516 case PTRACE_SINGLESTEP: /* set the trap flag. */
517 ret = -EIO;
518 if (!valid_signal(data))
519 break;
521 if (request == PTRACE_SYSEMU_SINGLESTEP)
522 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
523 else
524 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
526 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
527 set_singlestep(child);
528 child->exit_code = data;
529 /* give it a chance to run. */
530 wake_up_process(child);
531 ret = 0;
532 break;
534 case PTRACE_DETACH:
535 /* detach a process that was attached. */
536 ret = ptrace_detach(child, data);
537 break;
539 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
540 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
541 ret = -EIO;
542 break;
544 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
545 __put_user(getreg(child, i), datap);
546 datap++;
548 ret = 0;
549 break;
552 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
553 unsigned long tmp;
554 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
555 ret = -EIO;
556 break;
558 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
559 __get_user(tmp, datap);
560 putreg(child, i, tmp);
561 datap++;
563 ret = 0;
564 break;
567 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
568 if (!access_ok(VERIFY_WRITE, datap,
569 sizeof(struct user_i387_struct))) {
570 ret = -EIO;
571 break;
573 ret = 0;
574 if (!tsk_used_math(child))
575 init_fpu(child);
576 get_fpregs((struct user_i387_struct __user *)data, child);
577 break;
580 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
581 if (!access_ok(VERIFY_READ, datap,
582 sizeof(struct user_i387_struct))) {
583 ret = -EIO;
584 break;
586 set_stopped_child_used_math(child);
587 set_fpregs(child, (struct user_i387_struct __user *)data);
588 ret = 0;
589 break;
592 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
593 if (!access_ok(VERIFY_WRITE, datap,
594 sizeof(struct user_fxsr_struct))) {
595 ret = -EIO;
596 break;
598 if (!tsk_used_math(child))
599 init_fpu(child);
600 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
601 break;
604 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
605 if (!access_ok(VERIFY_READ, datap,
606 sizeof(struct user_fxsr_struct))) {
607 ret = -EIO;
608 break;
610 set_stopped_child_used_math(child);
611 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
612 break;
615 case PTRACE_GET_THREAD_AREA:
616 ret = ptrace_get_thread_area(child, addr,
617 (struct user_desc __user *) data);
618 break;
620 case PTRACE_SET_THREAD_AREA:
621 ret = ptrace_set_thread_area(child, addr,
622 (struct user_desc __user *) data);
623 break;
625 default:
626 ret = ptrace_request(child, request, addr, data);
627 break;
629 out_tsk:
630 return ret;
633 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
635 struct siginfo info;
637 tsk->thread.trap_no = 1;
638 tsk->thread.error_code = error_code;
640 memset(&info, 0, sizeof(info));
641 info.si_signo = SIGTRAP;
642 info.si_code = TRAP_BRKPT;
644 /* User-mode eip? */
645 info.si_addr = user_mode_vm(regs) ? (void __user *) regs->eip : NULL;
647 /* Send us the fakey SIGTRAP */
648 force_sig_info(SIGTRAP, &info, tsk);
651 /* notification of system call entry/exit
652 * - triggered by current->work.syscall_trace
654 __attribute__((regparm(3)))
655 int do_syscall_trace(struct pt_regs *regs, int entryexit)
657 int is_sysemu = test_thread_flag(TIF_SYSCALL_EMU);
659 * With TIF_SYSCALL_EMU set we want to ignore TIF_SINGLESTEP for syscall
660 * interception
662 int is_singlestep = !is_sysemu && test_thread_flag(TIF_SINGLESTEP);
663 int ret = 0;
665 /* do the secure computing check first */
666 if (!entryexit)
667 secure_computing(regs->orig_eax);
669 if (unlikely(current->audit_context)) {
670 if (entryexit)
671 audit_syscall_exit(current, AUDITSC_RESULT(regs->eax),
672 regs->eax);
673 /* Debug traps, when using PTRACE_SINGLESTEP, must be sent only
674 * on the syscall exit path. Normally, when TIF_SYSCALL_AUDIT is
675 * not used, entry.S will call us only on syscall exit, not
676 * entry; so when TIF_SYSCALL_AUDIT is used we must avoid
677 * calling send_sigtrap() on syscall entry.
679 * Note that when PTRACE_SYSEMU_SINGLESTEP is used,
680 * is_singlestep is false, despite his name, so we will still do
681 * the correct thing.
683 else if (is_singlestep)
684 goto out;
687 if (!(current->ptrace & PT_PTRACED))
688 goto out;
690 /* If a process stops on the 1st tracepoint with SYSCALL_TRACE
691 * and then is resumed with SYSEMU_SINGLESTEP, it will come in
692 * here. We have to check this and return */
693 if (is_sysemu && entryexit)
694 return 0;
696 /* Fake a debug trap */
697 if (is_singlestep)
698 send_sigtrap(current, regs, 0);
700 if (!test_thread_flag(TIF_SYSCALL_TRACE) && !is_sysemu)
701 goto out;
703 /* the 0x80 provides a way for the tracing parent to distinguish
704 between a syscall stop and SIGTRAP delivery */
705 /* Note that the debugger could change the result of test_thread_flag!*/
706 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80:0));
709 * this isn't the same as continuing with a signal, but it will do
710 * for normal use. strace only continues with a signal if the
711 * stopping signal is not SIGTRAP. -brl
713 if (current->exit_code) {
714 send_sig(current->exit_code, current, 1);
715 current->exit_code = 0;
717 ret = is_sysemu;
718 out:
719 if (unlikely(current->audit_context) && !entryexit)
720 audit_syscall_entry(current, AUDIT_ARCH_I386, regs->orig_eax,
721 regs->ebx, regs->ecx, regs->edx, regs->esi);
722 if (ret == 0)
723 return 0;
725 regs->orig_eax = -1; /* force skip of syscall restarting */
726 if (unlikely(current->audit_context))
727 audit_syscall_exit(current, AUDITSC_RESULT(regs->eax),
728 regs->eax);
729 return 1;