2 /* By Ross Biro 1/23/92 */
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
7 * x86-64 port 2000-2002 Andi Kleen
10 #include <linux/kernel.h>
11 #include <linux/sched.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>
22 #include <asm/uaccess.h>
23 #include <asm/pgtable.h>
24 #include <asm/system.h>
25 #include <asm/processor.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
45 /* set's the trap flag. */
46 #define TRAP_FLAG 0x100UL
49 * eflags and offset of eflags on child stack..
51 #define EFLAGS offsetof(struct pt_regs, eflags)
52 #define EFL_OFFSET ((int)(EFLAGS-sizeof(struct pt_regs)))
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
60 static inline unsigned long get_stack_long(struct task_struct *task, int offset)
64 stack = (unsigned char *)task->thread.rsp0;
66 return (*((unsigned long *)stack));
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
75 static inline long put_stack_long(struct task_struct *task, int offset,
78 unsigned char * stack;
80 stack = (unsigned char *) task->thread.rsp0;
82 *(unsigned long *) stack = data;
88 unsigned long convert_rip_to_linear(struct task_struct *child, struct pt_regs *regs)
90 unsigned long addr, seg;
93 seg = regs->cs & 0xffff;
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.
101 if (seg & LDT_SEGMENT) {
105 down(&child->mm->context.sem);
106 desc = child->mm->context.ldt + (seg & ~7);
107 base = (desc[0] >> 16) | ((desc[1] & 0xff) << 16) | (desc[1] & 0xff000000);
109 /* 16-bit code segment? */
110 if (!((desc[1] >> 22) & 1))
113 up(&child->mm->context.sem);
118 static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
121 unsigned char opcode[15];
122 unsigned long addr = convert_rip_to_linear(child, regs);
124 copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
125 for (i = 0; i < copied; i++) {
128 case 0x9d: case 0xcf:
133 /* opcode and address size prefixes */
134 case 0x66: case 0x67:
136 /* irrelevant prefixes (segment overrides and repeats) */
137 case 0x26: case 0x2e:
138 case 0x36: case 0x3e:
139 case 0x64: case 0x65:
140 case 0xf2: case 0xf3:
144 if (regs->cs != __USER_CS)
145 /* 32-bit mode: register increment */
147 /* 64-bit mode: REX prefix */
150 /* CHECKME: f2, f3 */
153 * pushf: NOTE! We should probably not let
154 * the user see the TF bit being set. But
155 * it's more pain than it's worth to avoid
156 * it, and a debugger could emulate this
157 * all in user space if it _really_ cares.
167 static void set_singlestep(struct task_struct *child)
169 struct pt_regs *regs = task_pt_regs(child);
172 * Always set TIF_SINGLESTEP - this guarantees that
173 * we single-step system calls etc.. This will also
174 * cause us to set TF when returning to user mode.
176 set_tsk_thread_flag(child, TIF_SINGLESTEP);
179 * If TF was already set, don't do anything else
181 if (regs->eflags & TRAP_FLAG)
184 /* Set TF on the kernel stack.. */
185 regs->eflags |= TRAP_FLAG;
188 * ..but if TF is changed by the instruction we will trace,
189 * don't mark it as being "us" that set it, so that we
190 * won't clear it by hand later.
192 if (is_setting_trap_flag(child, regs))
195 child->ptrace |= PT_DTRACE;
198 static void clear_singlestep(struct task_struct *child)
200 /* Always clear TIF_SINGLESTEP... */
201 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
203 /* But touch TF only if it was set by us.. */
204 if (child->ptrace & PT_DTRACE) {
205 struct pt_regs *regs = task_pt_regs(child);
206 regs->eflags &= ~TRAP_FLAG;
207 child->ptrace &= ~PT_DTRACE;
212 * Called by kernel/ptrace.c when detaching..
214 * Make sure the single step bit is not set.
216 void ptrace_disable(struct task_struct *child)
218 clear_singlestep(child);
221 static int putreg(struct task_struct *child,
222 unsigned long regno, unsigned long value)
226 /* Some code in the 64bit emulation may not be 64bit clean.
227 Don't take any chances. */
228 if (test_tsk_thread_flag(child, TIF_IA32))
231 case offsetof(struct user_regs_struct,fs):
232 if (value && (value & 3) != 3)
234 child->thread.fsindex = value & 0xffff;
236 case offsetof(struct user_regs_struct,gs):
237 if (value && (value & 3) != 3)
239 child->thread.gsindex = value & 0xffff;
241 case offsetof(struct user_regs_struct,ds):
242 if (value && (value & 3) != 3)
244 child->thread.ds = value & 0xffff;
246 case offsetof(struct user_regs_struct,es):
247 if (value && (value & 3) != 3)
249 child->thread.es = value & 0xffff;
251 case offsetof(struct user_regs_struct,ss):
252 if ((value & 3) != 3)
256 case offsetof(struct user_regs_struct,fs_base):
257 if (value >= TASK_SIZE_OF(child))
259 child->thread.fs = value;
261 case offsetof(struct user_regs_struct,gs_base):
262 if (value >= TASK_SIZE_OF(child))
264 child->thread.gs = value;
266 case offsetof(struct user_regs_struct, eflags):
268 tmp = get_stack_long(child, EFL_OFFSET);
272 case offsetof(struct user_regs_struct,cs):
273 if ((value & 3) != 3)
278 put_stack_long(child, regno - sizeof(struct pt_regs), value);
282 static unsigned long getreg(struct task_struct *child, unsigned long regno)
286 case offsetof(struct user_regs_struct, fs):
287 return child->thread.fsindex;
288 case offsetof(struct user_regs_struct, gs):
289 return child->thread.gsindex;
290 case offsetof(struct user_regs_struct, ds):
291 return child->thread.ds;
292 case offsetof(struct user_regs_struct, es):
293 return child->thread.es;
294 case offsetof(struct user_regs_struct, fs_base):
295 return child->thread.fs;
296 case offsetof(struct user_regs_struct, gs_base):
297 return child->thread.gs;
299 regno = regno - sizeof(struct pt_regs);
300 val = get_stack_long(child, regno);
301 if (test_tsk_thread_flag(child, TIF_IA32))
308 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
314 /* when I and D space are separate, these will need to be fixed. */
315 case PTRACE_PEEKTEXT: /* read word at location addr. */
316 case PTRACE_PEEKDATA: {
320 copied = access_process_vm(child, addr, &tmp, sizeof(tmp), 0);
322 if (copied != sizeof(tmp))
324 ret = put_user(tmp,(unsigned long __user *) data);
328 /* read the word at location addr in the USER area. */
329 case PTRACE_PEEKUSR: {
334 addr > sizeof(struct user) - 7)
338 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
339 tmp = getreg(child, addr);
341 case offsetof(struct user, u_debugreg[0]):
342 tmp = child->thread.debugreg0;
344 case offsetof(struct user, u_debugreg[1]):
345 tmp = child->thread.debugreg1;
347 case offsetof(struct user, u_debugreg[2]):
348 tmp = child->thread.debugreg2;
350 case offsetof(struct user, u_debugreg[3]):
351 tmp = child->thread.debugreg3;
353 case offsetof(struct user, u_debugreg[6]):
354 tmp = child->thread.debugreg6;
356 case offsetof(struct user, u_debugreg[7]):
357 tmp = child->thread.debugreg7;
363 ret = put_user(tmp,(unsigned long __user *) data);
367 /* when I and D space are separate, this will have to be fixed. */
368 case PTRACE_POKETEXT: /* write the word at location addr. */
369 case PTRACE_POKEDATA:
371 if (access_process_vm(child, addr, &data, sizeof(data), 1) == sizeof(data))
376 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
378 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
381 addr > sizeof(struct user) - 7)
385 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
386 ret = putreg(child, addr, data);
388 /* Disallows to set a breakpoint into the vsyscall */
389 case offsetof(struct user, u_debugreg[0]):
390 if (data >= TASK_SIZE_OF(child) - dsize) break;
391 child->thread.debugreg0 = data;
394 case offsetof(struct user, u_debugreg[1]):
395 if (data >= TASK_SIZE_OF(child) - dsize) break;
396 child->thread.debugreg1 = data;
399 case offsetof(struct user, u_debugreg[2]):
400 if (data >= TASK_SIZE_OF(child) - dsize) break;
401 child->thread.debugreg2 = data;
404 case offsetof(struct user, u_debugreg[3]):
405 if (data >= TASK_SIZE_OF(child) - dsize) break;
406 child->thread.debugreg3 = data;
409 case offsetof(struct user, u_debugreg[6]):
412 child->thread.debugreg6 = data;
415 case offsetof(struct user, u_debugreg[7]):
416 /* See arch/i386/kernel/ptrace.c for an explanation of
417 * this awkward check.*/
418 data &= ~DR_CONTROL_RESERVED;
420 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
423 child->thread.debugreg7 = data;
425 set_tsk_thread_flag(child, TIF_DEBUG);
427 clear_tsk_thread_flag(child, TIF_DEBUG);
434 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
435 case PTRACE_CONT: /* restart after signal. */
438 if (!valid_signal(data))
440 if (request == PTRACE_SYSCALL)
441 set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
443 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
444 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
445 child->exit_code = data;
446 /* make sure the single step bit is not set. */
447 clear_singlestep(child);
448 wake_up_process(child);
452 #ifdef CONFIG_IA32_EMULATION
453 /* This makes only sense with 32bit programs. Allow a
454 64bit debugger to fully examine them too. Better
455 don't use it against 64bit processes, use
456 PTRACE_ARCH_PRCTL instead. */
457 case PTRACE_SET_THREAD_AREA: {
458 struct user_desc __user *p;
460 p = (struct user_desc __user *)data;
461 get_user(old, &p->entry_number);
462 put_user(addr, &p->entry_number);
463 ret = do_set_thread_area(&child->thread, p);
464 put_user(old, &p->entry_number);
466 case PTRACE_GET_THREAD_AREA:
467 p = (struct user_desc __user *)data;
468 get_user(old, &p->entry_number);
469 put_user(addr, &p->entry_number);
470 ret = do_get_thread_area(&child->thread, p);
471 put_user(old, &p->entry_number);
475 /* normal 64bit interface to access TLS data.
476 Works just like arch_prctl, except that the arguments
478 case PTRACE_ARCH_PRCTL:
479 ret = do_arch_prctl(child, data, addr);
483 * make the child exit. Best I can do is send it a sigkill.
484 * perhaps it should be put in the status that it wants to
489 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
491 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
492 child->exit_code = SIGKILL;
493 /* make sure the single step bit is not set. */
494 clear_singlestep(child);
495 wake_up_process(child);
498 case PTRACE_SINGLESTEP: /* set the trap flag. */
500 if (!valid_signal(data))
502 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
503 set_singlestep(child);
504 child->exit_code = data;
505 /* give it a chance to run. */
506 wake_up_process(child);
511 /* detach a process that was attached. */
512 ret = ptrace_detach(child, data);
515 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
516 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
517 sizeof(struct user_regs_struct))) {
522 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
523 ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
524 data += sizeof(long);
529 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
531 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
532 sizeof(struct user_regs_struct))) {
537 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
538 ret = __get_user(tmp, (unsigned long __user *) data);
541 ret = putreg(child, ui, tmp);
544 data += sizeof(long);
549 case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
550 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
551 sizeof(struct user_i387_struct))) {
555 ret = get_fpregs((struct user_i387_struct __user *)data, child);
559 case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
560 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
561 sizeof(struct user_i387_struct))) {
565 set_stopped_child_used_math(child);
566 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
571 ret = ptrace_request(child, request, addr, data);
577 static void syscall_trace(struct pt_regs *regs)
581 printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
583 regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
584 current_thread_info()->flags, current->ptrace);
587 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
590 * this isn't the same as continuing with a signal, but it will do
591 * for normal use. strace only continues with a signal if the
592 * stopping signal is not SIGTRAP. -brl
594 if (current->exit_code) {
595 send_sig(current->exit_code, current, 1);
596 current->exit_code = 0;
600 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
602 /* do the secure computing check first */
603 secure_computing(regs->orig_rax);
605 if (test_thread_flag(TIF_SYSCALL_TRACE)
606 && (current->ptrace & PT_PTRACED))
609 if (unlikely(current->audit_context)) {
610 if (test_thread_flag(TIF_IA32)) {
611 audit_syscall_entry(AUDIT_ARCH_I386,
613 regs->rbx, regs->rcx,
614 regs->rdx, regs->rsi);
616 audit_syscall_entry(AUDIT_ARCH_X86_64,
618 regs->rdi, regs->rsi,
619 regs->rdx, regs->r10);
624 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
626 if (unlikely(current->audit_context))
627 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
629 if ((test_thread_flag(TIF_SYSCALL_TRACE)
630 || test_thread_flag(TIF_SINGLESTEP))
631 && (current->ptrace & PT_PTRACED))