1 /* By Ross Biro 1/23/92 */
3 * Pentium III FXSR, SSE support
4 * Gareth Hughes <gareth@valinux.com>, May 2000
6 * x86-64 port 2000-2002 Andi Kleen
9 #include <linux/kernel.h>
10 #include <linux/sched.h>
12 #include <linux/smp.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/prctl.h>
27 #include <asm/debugreg.h>
30 #include <asm/proto.h>
34 * does not yet catch signals sent when the child dies.
35 * in exit.c or in signal.c.
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).
43 #define FLAG_MASK 0x54dd5UL
46 * eflags and offset of eflags on child stack..
48 #define EFLAGS offsetof(struct pt_regs, eflags)
49 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
52 * this routine will get a word off of the processes privileged stack.
53 * the offset is how far from the base addr as stored in the TSS.
54 * this routine assumes that all the privileged stacks are in our
57 static inline unsigned long get_stack_long(struct task_struct
*task
, int offset
)
61 stack
= (unsigned char *)task
->thread
.rsp0
;
63 return (*((unsigned long *)stack
));
67 * this routine will put a word on the processes privileged stack.
68 * the offset is how far from the base addr as stored in the TSS.
69 * this routine assumes that all the privileged stacks are in our
72 static inline long put_stack_long(struct task_struct
*task
, int offset
,
75 unsigned char * stack
;
77 stack
= (unsigned char *) task
->thread
.rsp0
;
79 *(unsigned long *) stack
= data
;
85 unsigned long convert_rip_to_linear(struct task_struct
*child
, struct pt_regs
*regs
)
87 unsigned long addr
, seg
;
90 seg
= regs
->cs
& 0xffff;
93 * We'll assume that the code segments in the GDT
94 * are all zero-based. That is largely true: the
95 * TLS segments are used for data, and the PNPBIOS
96 * and APM bios ones we just ignore here.
98 if (seg
& LDT_SEGMENT
) {
104 mutex_lock(&child
->mm
->context
.lock
);
105 if (unlikely((seg
>> 3) >= child
->mm
->context
.size
))
106 addr
= -1L; /* bogus selector, access would fault */
108 desc
= child
->mm
->context
.ldt
+ seg
;
109 base
= ((desc
[0] >> 16) |
110 ((desc
[1] & 0xff) << 16) |
111 (desc
[1] & 0xff000000));
113 /* 16-bit code segment? */
114 if (!((desc
[1] >> 22) & 1))
118 mutex_unlock(&child
->mm
->context
.lock
);
124 static int is_setting_trap_flag(struct task_struct
*child
, struct pt_regs
*regs
)
127 unsigned char opcode
[15];
128 unsigned long addr
= convert_rip_to_linear(child
, regs
);
130 copied
= access_process_vm(child
, addr
, opcode
, sizeof(opcode
), 0);
131 for (i
= 0; i
< copied
; i
++) {
134 case 0x9d: case 0xcf:
139 /* opcode and address size prefixes */
140 case 0x66: case 0x67:
142 /* irrelevant prefixes (segment overrides and repeats) */
143 case 0x26: case 0x2e:
144 case 0x36: case 0x3e:
145 case 0x64: case 0x65:
146 case 0xf2: case 0xf3:
150 if (regs
->cs
!= __USER_CS
)
151 /* 32-bit mode: register increment */
153 /* 64-bit mode: REX prefix */
156 /* CHECKME: f2, f3 */
159 * pushf: NOTE! We should probably not let
160 * the user see the TF bit being set. But
161 * it's more pain than it's worth to avoid
162 * it, and a debugger could emulate this
163 * all in user space if it _really_ cares.
173 static void set_singlestep(struct task_struct
*child
)
175 struct pt_regs
*regs
= task_pt_regs(child
);
178 * Always set TIF_SINGLESTEP - this guarantees that
179 * we single-step system calls etc.. This will also
180 * cause us to set TF when returning to user mode.
182 set_tsk_thread_flag(child
, TIF_SINGLESTEP
);
185 * If TF was already set, don't do anything else
187 if (regs
->eflags
& X86_EFLAGS_TF
)
190 /* Set TF on the kernel stack.. */
191 regs
->eflags
|= X86_EFLAGS_TF
;
194 * ..but if TF is changed by the instruction we will trace,
195 * don't mark it as being "us" that set it, so that we
196 * won't clear it by hand later.
198 if (is_setting_trap_flag(child
, regs
))
201 child
->ptrace
|= PT_DTRACE
;
204 static void clear_singlestep(struct task_struct
*child
)
206 /* Always clear TIF_SINGLESTEP... */
207 clear_tsk_thread_flag(child
, TIF_SINGLESTEP
);
209 /* But touch TF only if it was set by us.. */
210 if (child
->ptrace
& PT_DTRACE
) {
211 struct pt_regs
*regs
= task_pt_regs(child
);
212 regs
->eflags
&= ~X86_EFLAGS_TF
;
213 child
->ptrace
&= ~PT_DTRACE
;
218 * Called by kernel/ptrace.c when detaching..
220 * Make sure the single step bit is not set.
222 void ptrace_disable(struct task_struct
*child
)
224 clear_singlestep(child
);
227 static int putreg(struct task_struct
*child
,
228 unsigned long regno
, unsigned long value
)
233 case offsetof(struct user_regs_struct
,fs
):
234 if (value
&& (value
& 3) != 3)
236 child
->thread
.fsindex
= value
& 0xffff;
238 case offsetof(struct user_regs_struct
,gs
):
239 if (value
&& (value
& 3) != 3)
241 child
->thread
.gsindex
= value
& 0xffff;
243 case offsetof(struct user_regs_struct
,ds
):
244 if (value
&& (value
& 3) != 3)
246 child
->thread
.ds
= value
& 0xffff;
248 case offsetof(struct user_regs_struct
,es
):
249 if (value
&& (value
& 3) != 3)
251 child
->thread
.es
= value
& 0xffff;
253 case offsetof(struct user_regs_struct
,ss
):
254 if ((value
& 3) != 3)
258 case offsetof(struct user_regs_struct
,fs_base
):
259 if (value
>= TASK_SIZE_OF(child
))
262 * When changing the segment base, use do_arch_prctl
263 * to set either thread.fs or thread.fsindex and the
264 * corresponding GDT slot.
266 if (child
->thread
.fs
!= value
)
267 return do_arch_prctl(child
, ARCH_SET_FS
, value
);
269 case offsetof(struct user_regs_struct
,gs_base
):
271 * Exactly the same here as the %fs handling above.
273 if (value
>= TASK_SIZE_OF(child
))
275 if (child
->thread
.gs
!= value
)
276 return do_arch_prctl(child
, ARCH_SET_GS
, value
);
278 case offsetof(struct user_regs_struct
, eflags
):
280 tmp
= get_stack_long(child
, EFL_OFFSET
);
284 case offsetof(struct user_regs_struct
,cs
):
285 if ((value
& 3) != 3)
290 put_stack_long(child
, regno
- sizeof(struct pt_regs
), value
);
294 static unsigned long getreg(struct task_struct
*child
, unsigned long regno
)
298 case offsetof(struct user_regs_struct
, fs
):
299 return child
->thread
.fsindex
;
300 case offsetof(struct user_regs_struct
, gs
):
301 return child
->thread
.gsindex
;
302 case offsetof(struct user_regs_struct
, ds
):
303 return child
->thread
.ds
;
304 case offsetof(struct user_regs_struct
, es
):
305 return child
->thread
.es
;
306 case offsetof(struct user_regs_struct
, fs_base
):
308 * do_arch_prctl may have used a GDT slot instead of
309 * the MSR. To userland, it appears the same either
310 * way, except the %fs segment selector might not be 0.
312 if (child
->thread
.fs
!= 0)
313 return child
->thread
.fs
;
314 if (child
->thread
.fsindex
!= FS_TLS_SEL
)
316 return get_desc_base(&child
->thread
.tls_array
[FS_TLS
]);
317 case offsetof(struct user_regs_struct
, gs_base
):
319 * Exactly the same here as the %fs handling above.
321 if (child
->thread
.gs
!= 0)
322 return child
->thread
.gs
;
323 if (child
->thread
.gsindex
!= GS_TLS_SEL
)
325 return get_desc_base(&child
->thread
.tls_array
[GS_TLS
]);
327 regno
= regno
- sizeof(struct pt_regs
);
328 val
= get_stack_long(child
, regno
);
329 if (test_tsk_thread_flag(child
, TIF_IA32
))
336 long arch_ptrace(struct task_struct
*child
, long request
, long addr
, long data
)
342 /* when I and D space are separate, these will need to be fixed. */
343 case PTRACE_PEEKTEXT
: /* read word at location addr. */
344 case PTRACE_PEEKDATA
:
345 ret
= generic_ptrace_peekdata(child
, addr
, data
);
348 /* read the word at location addr in the USER area. */
349 case PTRACE_PEEKUSR
: {
354 addr
> sizeof(struct user
) - 7)
358 case 0 ... sizeof(struct user_regs_struct
) - sizeof(long):
359 tmp
= getreg(child
, addr
);
361 case offsetof(struct user
, u_debugreg
[0]):
362 tmp
= child
->thread
.debugreg0
;
364 case offsetof(struct user
, u_debugreg
[1]):
365 tmp
= child
->thread
.debugreg1
;
367 case offsetof(struct user
, u_debugreg
[2]):
368 tmp
= child
->thread
.debugreg2
;
370 case offsetof(struct user
, u_debugreg
[3]):
371 tmp
= child
->thread
.debugreg3
;
373 case offsetof(struct user
, u_debugreg
[6]):
374 tmp
= child
->thread
.debugreg6
;
376 case offsetof(struct user
, u_debugreg
[7]):
377 tmp
= child
->thread
.debugreg7
;
383 ret
= put_user(tmp
,(unsigned long __user
*) data
);
387 /* when I and D space are separate, this will have to be fixed. */
388 case PTRACE_POKETEXT
: /* write the word at location addr. */
389 case PTRACE_POKEDATA
:
390 ret
= generic_ptrace_pokedata(child
, addr
, data
);
393 case PTRACE_POKEUSR
: /* write the word at location addr in the USER area */
395 int dsize
= test_tsk_thread_flag(child
, TIF_IA32
) ? 3 : 7;
398 addr
> sizeof(struct user
) - 7)
402 case 0 ... sizeof(struct user_regs_struct
) - sizeof(long):
403 ret
= putreg(child
, addr
, data
);
405 /* Disallows to set a breakpoint into the vsyscall */
406 case offsetof(struct user
, u_debugreg
[0]):
407 if (data
>= TASK_SIZE_OF(child
) - dsize
) break;
408 child
->thread
.debugreg0
= data
;
411 case offsetof(struct user
, u_debugreg
[1]):
412 if (data
>= TASK_SIZE_OF(child
) - dsize
) break;
413 child
->thread
.debugreg1
= data
;
416 case offsetof(struct user
, u_debugreg
[2]):
417 if (data
>= TASK_SIZE_OF(child
) - dsize
) break;
418 child
->thread
.debugreg2
= data
;
421 case offsetof(struct user
, u_debugreg
[3]):
422 if (data
>= TASK_SIZE_OF(child
) - dsize
) break;
423 child
->thread
.debugreg3
= data
;
426 case offsetof(struct user
, u_debugreg
[6]):
429 child
->thread
.debugreg6
= data
;
432 case offsetof(struct user
, u_debugreg
[7]):
433 /* See arch/i386/kernel/ptrace.c for an explanation of
434 * this awkward check.*/
435 data
&= ~DR_CONTROL_RESERVED
;
437 if ((0x5554 >> ((data
>> (16 + 4*i
)) & 0xf)) & 1)
440 child
->thread
.debugreg7
= data
;
442 set_tsk_thread_flag(child
, TIF_DEBUG
);
444 clear_tsk_thread_flag(child
, TIF_DEBUG
);
451 case PTRACE_SYSCALL
: /* continue and stop at next (return from) syscall */
452 case PTRACE_CONT
: /* restart after signal. */
455 if (!valid_signal(data
))
457 if (request
== PTRACE_SYSCALL
)
458 set_tsk_thread_flag(child
,TIF_SYSCALL_TRACE
);
460 clear_tsk_thread_flag(child
,TIF_SYSCALL_TRACE
);
461 clear_tsk_thread_flag(child
, TIF_SINGLESTEP
);
462 child
->exit_code
= data
;
463 /* make sure the single step bit is not set. */
464 clear_singlestep(child
);
465 wake_up_process(child
);
469 #ifdef CONFIG_IA32_EMULATION
470 /* This makes only sense with 32bit programs. Allow a
471 64bit debugger to fully examine them too. Better
472 don't use it against 64bit processes, use
473 PTRACE_ARCH_PRCTL instead. */
474 case PTRACE_GET_THREAD_AREA
:
477 ret
= do_get_thread_area(child
, addr
,
478 (struct user_desc __user
*) data
);
481 case PTRACE_SET_THREAD_AREA
:
484 ret
= do_set_thread_area(child
, addr
,
485 (struct user_desc __user
*) data
, 0);
488 /* normal 64bit interface to access TLS data.
489 Works just like arch_prctl, except that the arguments
491 case PTRACE_ARCH_PRCTL
:
492 ret
= do_arch_prctl(child
, data
, addr
);
496 * make the child exit. Best I can do is send it a sigkill.
497 * perhaps it should be put in the status that it wants to
502 if (child
->exit_state
== EXIT_ZOMBIE
) /* already dead */
504 clear_tsk_thread_flag(child
, TIF_SINGLESTEP
);
505 child
->exit_code
= SIGKILL
;
506 /* make sure the single step bit is not set. */
507 clear_singlestep(child
);
508 wake_up_process(child
);
511 case PTRACE_SINGLESTEP
: /* set the trap flag. */
513 if (!valid_signal(data
))
515 clear_tsk_thread_flag(child
,TIF_SYSCALL_TRACE
);
516 set_singlestep(child
);
517 child
->exit_code
= data
;
518 /* give it a chance to run. */
519 wake_up_process(child
);
523 case PTRACE_GETREGS
: { /* Get all gp regs from the child. */
524 if (!access_ok(VERIFY_WRITE
, (unsigned __user
*)data
,
525 sizeof(struct user_regs_struct
))) {
530 for (ui
= 0; ui
< sizeof(struct user_regs_struct
); ui
+= sizeof(long)) {
531 ret
|= __put_user(getreg(child
, ui
),(unsigned long __user
*) data
);
532 data
+= sizeof(long);
537 case PTRACE_SETREGS
: { /* Set all gp regs in the child. */
539 if (!access_ok(VERIFY_READ
, (unsigned __user
*)data
,
540 sizeof(struct user_regs_struct
))) {
545 for (ui
= 0; ui
< sizeof(struct user_regs_struct
); ui
+= sizeof(long)) {
546 ret
= __get_user(tmp
, (unsigned long __user
*) data
);
549 ret
= putreg(child
, ui
, tmp
);
552 data
+= sizeof(long);
557 case PTRACE_GETFPREGS
: { /* Get the child extended FPU state. */
558 if (!access_ok(VERIFY_WRITE
, (unsigned __user
*)data
,
559 sizeof(struct user_i387_struct
))) {
563 ret
= get_fpregs((struct user_i387_struct __user
*)data
, child
);
567 case PTRACE_SETFPREGS
: { /* Set the child extended FPU state. */
568 if (!access_ok(VERIFY_READ
, (unsigned __user
*)data
,
569 sizeof(struct user_i387_struct
))) {
573 set_stopped_child_used_math(child
);
574 ret
= set_fpregs(child
, (struct user_i387_struct __user
*)data
);
579 ret
= ptrace_request(child
, request
, addr
, data
);
585 static void syscall_trace(struct pt_regs
*regs
)
589 printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
591 regs
->rip
, regs
->rsp
, regs
->rax
, regs
->orig_rax
, __builtin_return_address(0),
592 current_thread_info()->flags
, current
->ptrace
);
595 ptrace_notify(SIGTRAP
| ((current
->ptrace
& PT_TRACESYSGOOD
)
598 * this isn't the same as continuing with a signal, but it will do
599 * for normal use. strace only continues with a signal if the
600 * stopping signal is not SIGTRAP. -brl
602 if (current
->exit_code
) {
603 send_sig(current
->exit_code
, current
, 1);
604 current
->exit_code
= 0;
608 asmlinkage
void syscall_trace_enter(struct pt_regs
*regs
)
610 /* do the secure computing check first */
611 secure_computing(regs
->orig_rax
);
613 if (test_thread_flag(TIF_SYSCALL_TRACE
)
614 && (current
->ptrace
& PT_PTRACED
))
617 if (unlikely(current
->audit_context
)) {
618 if (test_thread_flag(TIF_IA32
)) {
619 audit_syscall_entry(AUDIT_ARCH_I386
,
621 regs
->rbx
, regs
->rcx
,
622 regs
->rdx
, regs
->rsi
);
624 audit_syscall_entry(AUDIT_ARCH_X86_64
,
626 regs
->rdi
, regs
->rsi
,
627 regs
->rdx
, regs
->r10
);
632 asmlinkage
void syscall_trace_leave(struct pt_regs
*regs
)
634 if (unlikely(current
->audit_context
))
635 audit_syscall_exit(AUDITSC_RESULT(regs
->rax
), regs
->rax
);
637 if ((test_thread_flag(TIF_SYSCALL_TRACE
)
638 || test_thread_flag(TIF_SINGLESTEP
))
639 && (current
->ptrace
& PT_PTRACED
))