[TG3]: Set minimal hw interrupt mitigation.
[linux-2.6/verdex.git] / arch / i386 / kernel / ptrace.c
blobe34f651fa13c001de7b84967c0882d71bda35c1e
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);
277 * Perform get_thread_area on behalf of the traced child.
279 static int
280 ptrace_get_thread_area(struct task_struct *child,
281 int idx, struct user_desc __user *user_desc)
283 struct user_desc info;
284 struct desc_struct *desc;
287 * Get the current Thread-Local Storage area:
290 #define GET_BASE(desc) ( \
291 (((desc)->a >> 16) & 0x0000ffff) | \
292 (((desc)->b << 16) & 0x00ff0000) | \
293 ( (desc)->b & 0xff000000) )
295 #define GET_LIMIT(desc) ( \
296 ((desc)->a & 0x0ffff) | \
297 ((desc)->b & 0xf0000) )
299 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
300 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
301 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
302 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
303 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
304 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
306 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
307 return -EINVAL;
309 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
311 info.entry_number = idx;
312 info.base_addr = GET_BASE(desc);
313 info.limit = GET_LIMIT(desc);
314 info.seg_32bit = GET_32BIT(desc);
315 info.contents = GET_CONTENTS(desc);
316 info.read_exec_only = !GET_WRITABLE(desc);
317 info.limit_in_pages = GET_LIMIT_PAGES(desc);
318 info.seg_not_present = !GET_PRESENT(desc);
319 info.useable = GET_USEABLE(desc);
321 if (copy_to_user(user_desc, &info, sizeof(info)))
322 return -EFAULT;
324 return 0;
328 * Perform set_thread_area on behalf of the traced child.
330 static int
331 ptrace_set_thread_area(struct task_struct *child,
332 int idx, struct user_desc __user *user_desc)
334 struct user_desc info;
335 struct desc_struct *desc;
337 if (copy_from_user(&info, user_desc, sizeof(info)))
338 return -EFAULT;
340 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
341 return -EINVAL;
343 desc = child->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
344 if (LDT_empty(&info)) {
345 desc->a = 0;
346 desc->b = 0;
347 } else {
348 desc->a = LDT_entry_a(&info);
349 desc->b = LDT_entry_b(&info);
352 return 0;
355 asmlinkage int sys_ptrace(long request, long pid, long addr, long data)
357 struct task_struct *child;
358 struct user * dummy = NULL;
359 int i, ret;
360 unsigned long __user *datap = (unsigned long __user *)data;
362 lock_kernel();
363 ret = -EPERM;
364 if (request == PTRACE_TRACEME) {
365 /* are we already being traced? */
366 if (current->ptrace & PT_PTRACED)
367 goto out;
368 ret = security_ptrace(current->parent, current);
369 if (ret)
370 goto out;
371 /* set the ptrace bit in the process flags. */
372 current->ptrace |= PT_PTRACED;
373 ret = 0;
374 goto out;
376 ret = -ESRCH;
377 read_lock(&tasklist_lock);
378 child = find_task_by_pid(pid);
379 if (child)
380 get_task_struct(child);
381 read_unlock(&tasklist_lock);
382 if (!child)
383 goto out;
385 ret = -EPERM;
386 if (pid == 1) /* you may not mess with init */
387 goto out_tsk;
389 if (request == PTRACE_ATTACH) {
390 ret = ptrace_attach(child);
391 goto out_tsk;
394 ret = ptrace_check_attach(child, request == PTRACE_KILL);
395 if (ret < 0)
396 goto out_tsk;
398 switch (request) {
399 /* when I and D space are separate, these will need to be fixed. */
400 case PTRACE_PEEKTEXT: /* read word at location addr. */
401 case PTRACE_PEEKDATA: {
402 unsigned long tmp;
403 int copied;
405 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
406 ret = -EIO;
407 if (copied != sizeof(tmp))
408 break;
409 ret = put_user(tmp, datap);
410 break;
413 /* read the word at location addr in the USER area. */
414 case PTRACE_PEEKUSR: {
415 unsigned long tmp;
417 ret = -EIO;
418 if ((addr & 3) || addr < 0 ||
419 addr > sizeof(struct user) - 3)
420 break;
422 tmp = 0; /* Default return condition */
423 if(addr < FRAME_SIZE*sizeof(long))
424 tmp = getreg(child, addr);
425 if(addr >= (long) &dummy->u_debugreg[0] &&
426 addr <= (long) &dummy->u_debugreg[7]){
427 addr -= (long) &dummy->u_debugreg[0];
428 addr = addr >> 2;
429 tmp = child->thread.debugreg[addr];
431 ret = put_user(tmp, datap);
432 break;
435 /* when I and D space are separate, this will have to be fixed. */
436 case PTRACE_POKETEXT: /* write the word at location addr. */
437 case PTRACE_POKEDATA:
438 ret = 0;
439 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
440 break;
441 ret = -EIO;
442 break;
444 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
445 ret = -EIO;
446 if ((addr & 3) || addr < 0 ||
447 addr > sizeof(struct user) - 3)
448 break;
450 if (addr < FRAME_SIZE*sizeof(long)) {
451 ret = putreg(child, addr, data);
452 break;
454 /* We need to be very careful here. We implicitly
455 want to modify a portion of the task_struct, and we
456 have to be selective about what portions we allow someone
457 to modify. */
459 ret = -EIO;
460 if(addr >= (long) &dummy->u_debugreg[0] &&
461 addr <= (long) &dummy->u_debugreg[7]){
463 if(addr == (long) &dummy->u_debugreg[4]) break;
464 if(addr == (long) &dummy->u_debugreg[5]) break;
465 if(addr < (long) &dummy->u_debugreg[4] &&
466 ((unsigned long) data) >= TASK_SIZE-3) break;
468 /* Sanity-check data. Take one half-byte at once with
469 * check = (val >> (16 + 4*i)) & 0xf. It contains the
470 * R/Wi and LENi bits; bits 0 and 1 are R/Wi, and bits
471 * 2 and 3 are LENi. Given a list of invalid values,
472 * we do mask |= 1 << invalid_value, so that
473 * (mask >> check) & 1 is a correct test for invalid
474 * values.
476 * R/Wi contains the type of the breakpoint /
477 * watchpoint, LENi contains the length of the watched
478 * data in the watchpoint case.
480 * The invalid values are:
481 * - LENi == 0x10 (undefined), so mask |= 0x0f00.
482 * - R/Wi == 0x10 (break on I/O reads or writes), so
483 * mask |= 0x4444.
484 * - R/Wi == 0x00 && LENi != 0x00, so we have mask |=
485 * 0x1110.
487 * Finally, mask = 0x0f00 | 0x4444 | 0x1110 == 0x5f54.
489 * See the Intel Manual "System Programming Guide",
490 * 15.2.4
492 * Note that LENi == 0x10 is defined on x86_64 in long
493 * mode (i.e. even for 32-bit userspace software, but
494 * 64-bit kernel), so the x86_64 mask value is 0x5454.
495 * See the AMD manual no. 24593 (AMD64 System
496 * Programming)*/
498 if(addr == (long) &dummy->u_debugreg[7]) {
499 data &= ~DR_CONTROL_RESERVED;
500 for(i=0; i<4; i++)
501 if ((0x5f54 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
502 goto out_tsk;
505 addr -= (long) &dummy->u_debugreg;
506 addr = addr >> 2;
507 child->thread.debugreg[addr] = data;
508 ret = 0;
510 break;
512 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
513 case PTRACE_CONT: /* restart after signal. */
514 ret = -EIO;
515 if (!valid_signal(data))
516 break;
517 if (request == PTRACE_SYSCALL) {
518 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
520 else {
521 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
523 child->exit_code = data;
524 /* make sure the single step bit is not set. */
525 clear_singlestep(child);
526 wake_up_process(child);
527 ret = 0;
528 break;
531 * make the child exit. Best I can do is send it a sigkill.
532 * perhaps it should be put in the status that it wants to
533 * exit.
535 case PTRACE_KILL:
536 ret = 0;
537 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
538 break;
539 child->exit_code = SIGKILL;
540 /* make sure the single step bit is not set. */
541 clear_singlestep(child);
542 wake_up_process(child);
543 break;
545 case PTRACE_SINGLESTEP: /* set the trap flag. */
546 ret = -EIO;
547 if (!valid_signal(data))
548 break;
549 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
550 set_singlestep(child);
551 child->exit_code = data;
552 /* give it a chance to run. */
553 wake_up_process(child);
554 ret = 0;
555 break;
557 case PTRACE_DETACH:
558 /* detach a process that was attached. */
559 ret = ptrace_detach(child, data);
560 break;
562 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
563 if (!access_ok(VERIFY_WRITE, datap, FRAME_SIZE*sizeof(long))) {
564 ret = -EIO;
565 break;
567 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
568 __put_user(getreg(child, i), datap);
569 datap++;
571 ret = 0;
572 break;
575 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
576 unsigned long tmp;
577 if (!access_ok(VERIFY_READ, datap, FRAME_SIZE*sizeof(long))) {
578 ret = -EIO;
579 break;
581 for ( i = 0; i < FRAME_SIZE*sizeof(long); i += sizeof(long) ) {
582 __get_user(tmp, datap);
583 putreg(child, i, tmp);
584 datap++;
586 ret = 0;
587 break;
590 case PTRACE_GETFPREGS: { /* Get the child FPU state. */
591 if (!access_ok(VERIFY_WRITE, datap,
592 sizeof(struct user_i387_struct))) {
593 ret = -EIO;
594 break;
596 ret = 0;
597 if (!tsk_used_math(child))
598 init_fpu(child);
599 get_fpregs((struct user_i387_struct __user *)data, child);
600 break;
603 case PTRACE_SETFPREGS: { /* Set the child FPU state. */
604 if (!access_ok(VERIFY_READ, datap,
605 sizeof(struct user_i387_struct))) {
606 ret = -EIO;
607 break;
609 set_stopped_child_used_math(child);
610 set_fpregs(child, (struct user_i387_struct __user *)data);
611 ret = 0;
612 break;
615 case PTRACE_GETFPXREGS: { /* Get the child extended FPU state. */
616 if (!access_ok(VERIFY_WRITE, datap,
617 sizeof(struct user_fxsr_struct))) {
618 ret = -EIO;
619 break;
621 if (!tsk_used_math(child))
622 init_fpu(child);
623 ret = get_fpxregs((struct user_fxsr_struct __user *)data, child);
624 break;
627 case PTRACE_SETFPXREGS: { /* Set the child extended FPU state. */
628 if (!access_ok(VERIFY_READ, datap,
629 sizeof(struct user_fxsr_struct))) {
630 ret = -EIO;
631 break;
633 set_stopped_child_used_math(child);
634 ret = set_fpxregs(child, (struct user_fxsr_struct __user *)data);
635 break;
638 case PTRACE_GET_THREAD_AREA:
639 ret = ptrace_get_thread_area(child, addr,
640 (struct user_desc __user *) data);
641 break;
643 case PTRACE_SET_THREAD_AREA:
644 ret = ptrace_set_thread_area(child, addr,
645 (struct user_desc __user *) data);
646 break;
648 default:
649 ret = ptrace_request(child, request, addr, data);
650 break;
652 out_tsk:
653 put_task_struct(child);
654 out:
655 unlock_kernel();
656 return ret;
659 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs, int error_code)
661 struct siginfo info;
663 tsk->thread.trap_no = 1;
664 tsk->thread.error_code = error_code;
666 memset(&info, 0, sizeof(info));
667 info.si_signo = SIGTRAP;
668 info.si_code = TRAP_BRKPT;
670 /* User-mode eip? */
671 info.si_addr = user_mode(regs) ? (void __user *) regs->eip : NULL;
673 /* Send us the fakey SIGTRAP */
674 force_sig_info(SIGTRAP, &info, tsk);
677 /* notification of system call entry/exit
678 * - triggered by current->work.syscall_trace
680 __attribute__((regparm(3)))
681 void do_syscall_trace(struct pt_regs *regs, int entryexit)
683 /* do the secure computing check first */
684 secure_computing(regs->orig_eax);
686 if (unlikely(current->audit_context) && entryexit)
687 audit_syscall_exit(current, AUDITSC_RESULT(regs->eax), regs->eax);
689 if (!(current->ptrace & PT_PTRACED))
690 goto out;
692 /* Fake a debug trap */
693 if (test_thread_flag(TIF_SINGLESTEP))
694 send_sigtrap(current, regs, 0);
696 if (!test_thread_flag(TIF_SYSCALL_TRACE))
697 goto out;
699 /* the 0x80 provides a way for the tracing parent to distinguish
700 between a syscall stop and SIGTRAP delivery */
701 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD) ? 0x80 : 0));
704 * this isn't the same as continuing with a signal, but it will do
705 * for normal use. strace only continues with a signal if the
706 * stopping signal is not SIGTRAP. -brl
708 if (current->exit_code) {
709 send_sig(current->exit_code, current, 1);
710 current->exit_code = 0;
712 out:
713 if (unlikely(current->audit_context) && !entryexit)
714 audit_syscall_entry(current, AUDIT_ARCH_I386, regs->orig_eax,
715 regs->ebx, regs->ecx, regs->edx, regs->esi);