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) {
107 down(&child->mm->context.sem);
108 if (unlikely((seg >> 3) >= child->mm->context.size))
109 addr = -1L; /* bogus selector, access would fault */
111 desc = child->mm->context.ldt + seg;
112 base = ((desc[0] >> 16) |
113 ((desc[1] & 0xff) << 16) |
114 (desc[1] & 0xff000000));
116 /* 16-bit code segment? */
117 if (!((desc[1] >> 22) & 1))
121 up(&child->mm->context.sem);
127 static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
130 unsigned char opcode[15];
131 unsigned long addr = convert_rip_to_linear(child, regs);
133 copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
134 for (i = 0; i < copied; i++) {
137 case 0x9d: case 0xcf:
142 /* opcode and address size prefixes */
143 case 0x66: case 0x67:
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:
153 if (regs->cs != __USER_CS)
154 /* 32-bit mode: register increment */
156 /* 64-bit mode: REX prefix */
159 /* CHECKME: f2, f3 */
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.
176 static void set_singlestep(struct task_struct *child)
178 struct pt_regs *regs = task_pt_regs(child);
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.
185 set_tsk_thread_flag(child, TIF_SINGLESTEP);
188 * If TF was already set, don't do anything else
190 if (regs->eflags & TRAP_FLAG)
193 /* Set TF on the kernel stack.. */
194 regs->eflags |= TRAP_FLAG;
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.
201 if (is_setting_trap_flag(child, regs))
204 child->ptrace |= PT_DTRACE;
207 static void clear_singlestep(struct task_struct *child)
209 /* Always clear TIF_SINGLESTEP... */
210 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
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;
221 * Called by kernel/ptrace.c when detaching..
223 * Make sure the single step bit is not set.
225 void ptrace_disable(struct task_struct *child)
227 clear_singlestep(child);
230 static int putreg(struct task_struct *child,
231 unsigned long regno, unsigned long value)
235 /* Some code in the 64bit emulation may not be 64bit clean.
236 Don't take any chances. */
237 if (test_tsk_thread_flag(child, TIF_IA32))
240 case offsetof(struct user_regs_struct,fs):
241 if (value && (value & 3) != 3)
243 child->thread.fsindex = value & 0xffff;
245 case offsetof(struct user_regs_struct,gs):
246 if (value && (value & 3) != 3)
248 child->thread.gsindex = value & 0xffff;
250 case offsetof(struct user_regs_struct,ds):
251 if (value && (value & 3) != 3)
253 child->thread.ds = value & 0xffff;
255 case offsetof(struct user_regs_struct,es):
256 if (value && (value & 3) != 3)
258 child->thread.es = value & 0xffff;
260 case offsetof(struct user_regs_struct,ss):
261 if ((value & 3) != 3)
265 case offsetof(struct user_regs_struct,fs_base):
266 if (value >= TASK_SIZE_OF(child))
268 child->thread.fs = value;
270 case offsetof(struct user_regs_struct,gs_base):
271 if (value >= TASK_SIZE_OF(child))
273 child->thread.gs = value;
275 case offsetof(struct user_regs_struct, eflags):
277 tmp = get_stack_long(child, EFL_OFFSET);
281 case offsetof(struct user_regs_struct,cs):
282 if ((value & 3) != 3)
287 put_stack_long(child, regno - sizeof(struct pt_regs), value);
291 static unsigned long getreg(struct task_struct *child, unsigned long regno)
295 case offsetof(struct user_regs_struct, fs):
296 return child->thread.fsindex;
297 case offsetof(struct user_regs_struct, gs):
298 return child->thread.gsindex;
299 case offsetof(struct user_regs_struct, ds):
300 return child->thread.ds;
301 case offsetof(struct user_regs_struct, es):
302 return child->thread.es;
303 case offsetof(struct user_regs_struct, fs_base):
304 return child->thread.fs;
305 case offsetof(struct user_regs_struct, gs_base):
306 return child->thread.gs;
308 regno = regno - sizeof(struct pt_regs);
309 val = get_stack_long(child, regno);
310 if (test_tsk_thread_flag(child, TIF_IA32))
317 long arch_ptrace(struct task_struct *child, long request, long addr, long data)
323 /* when I and D space are separate, these will need to be fixed. */
324 case PTRACE_PEEKTEXT: /* read word at location addr. */
325 case PTRACE_PEEKDATA:
326 ret = generic_ptrace_peekdata(child, addr, data);
329 /* read the word at location addr in the USER area. */
330 case PTRACE_PEEKUSR: {
335 addr > sizeof(struct user) - 7)
339 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
340 tmp = getreg(child, addr);
342 case offsetof(struct user, u_debugreg[0]):
343 tmp = child->thread.debugreg0;
345 case offsetof(struct user, u_debugreg[1]):
346 tmp = child->thread.debugreg1;
348 case offsetof(struct user, u_debugreg[2]):
349 tmp = child->thread.debugreg2;
351 case offsetof(struct user, u_debugreg[3]):
352 tmp = child->thread.debugreg3;
354 case offsetof(struct user, u_debugreg[6]):
355 tmp = child->thread.debugreg6;
357 case offsetof(struct user, u_debugreg[7]):
358 tmp = child->thread.debugreg7;
364 ret = put_user(tmp,(unsigned long __user *) data);
368 /* when I and D space are separate, this will have to be fixed. */
369 case PTRACE_POKETEXT: /* write the word at location addr. */
370 case PTRACE_POKEDATA:
371 ret = generic_ptrace_pokedata(child, addr, data);
374 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
376 int dsize = test_tsk_thread_flag(child, TIF_IA32) ? 3 : 7;
379 addr > sizeof(struct user) - 7)
383 case 0 ... sizeof(struct user_regs_struct) - sizeof(long):
384 ret = putreg(child, addr, data);
386 /* Disallows to set a breakpoint into the vsyscall */
387 case offsetof(struct user, u_debugreg[0]):
388 if (data >= TASK_SIZE_OF(child) - dsize) break;
389 child->thread.debugreg0 = data;
392 case offsetof(struct user, u_debugreg[1]):
393 if (data >= TASK_SIZE_OF(child) - dsize) break;
394 child->thread.debugreg1 = data;
397 case offsetof(struct user, u_debugreg[2]):
398 if (data >= TASK_SIZE_OF(child) - dsize) break;
399 child->thread.debugreg2 = data;
402 case offsetof(struct user, u_debugreg[3]):
403 if (data >= TASK_SIZE_OF(child) - dsize) break;
404 child->thread.debugreg3 = data;
407 case offsetof(struct user, u_debugreg[6]):
410 child->thread.debugreg6 = data;
413 case offsetof(struct user, u_debugreg[7]):
414 /* See arch/i386/kernel/ptrace.c for an explanation of
415 * this awkward check.*/
416 data &= ~DR_CONTROL_RESERVED;
418 if ((0x5554 >> ((data >> (16 + 4*i)) & 0xf)) & 1)
421 child->thread.debugreg7 = data;
423 set_tsk_thread_flag(child, TIF_DEBUG);
425 clear_tsk_thread_flag(child, TIF_DEBUG);
432 case PTRACE_SYSCALL: /* continue and stop at next (return from) syscall */
433 case PTRACE_CONT: /* restart after signal. */
436 if (!valid_signal(data))
438 if (request == PTRACE_SYSCALL)
439 set_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
441 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
442 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
443 child->exit_code = data;
444 /* make sure the single step bit is not set. */
445 clear_singlestep(child);
446 wake_up_process(child);
450 #ifdef CONFIG_IA32_EMULATION
451 /* This makes only sense with 32bit programs. Allow a
452 64bit debugger to fully examine them too. Better
453 don't use it against 64bit processes, use
454 PTRACE_ARCH_PRCTL instead. */
455 case PTRACE_SET_THREAD_AREA: {
456 struct user_desc __user *p;
458 p = (struct user_desc __user *)data;
459 get_user(old, &p->entry_number);
460 put_user(addr, &p->entry_number);
461 ret = do_set_thread_area(&child->thread, p);
462 put_user(old, &p->entry_number);
464 case PTRACE_GET_THREAD_AREA:
465 p = (struct user_desc __user *)data;
466 get_user(old, &p->entry_number);
467 put_user(addr, &p->entry_number);
468 ret = do_get_thread_area(&child->thread, p);
469 put_user(old, &p->entry_number);
473 /* normal 64bit interface to access TLS data.
474 Works just like arch_prctl, except that the arguments
476 case PTRACE_ARCH_PRCTL:
477 ret = do_arch_prctl(child, data, addr);
481 * make the child exit. Best I can do is send it a sigkill.
482 * perhaps it should be put in the status that it wants to
487 if (child->exit_state == EXIT_ZOMBIE) /* already dead */
489 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
490 child->exit_code = SIGKILL;
491 /* make sure the single step bit is not set. */
492 clear_singlestep(child);
493 wake_up_process(child);
496 case PTRACE_SINGLESTEP: /* set the trap flag. */
498 if (!valid_signal(data))
500 clear_tsk_thread_flag(child,TIF_SYSCALL_TRACE);
501 set_singlestep(child);
502 child->exit_code = data;
503 /* give it a chance to run. */
504 wake_up_process(child);
509 /* detach a process that was attached. */
510 ret = ptrace_detach(child, data);
513 case PTRACE_GETREGS: { /* Get all gp regs from the child. */
514 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
515 sizeof(struct user_regs_struct))) {
520 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
521 ret |= __put_user(getreg(child, ui),(unsigned long __user *) data);
522 data += sizeof(long);
527 case PTRACE_SETREGS: { /* Set all gp regs in the child. */
529 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
530 sizeof(struct user_regs_struct))) {
535 for (ui = 0; ui < sizeof(struct user_regs_struct); ui += sizeof(long)) {
536 ret = __get_user(tmp, (unsigned long __user *) data);
539 ret = putreg(child, ui, tmp);
542 data += sizeof(long);
547 case PTRACE_GETFPREGS: { /* Get the child extended FPU state. */
548 if (!access_ok(VERIFY_WRITE, (unsigned __user *)data,
549 sizeof(struct user_i387_struct))) {
553 ret = get_fpregs((struct user_i387_struct __user *)data, child);
557 case PTRACE_SETFPREGS: { /* Set the child extended FPU state. */
558 if (!access_ok(VERIFY_READ, (unsigned __user *)data,
559 sizeof(struct user_i387_struct))) {
563 set_stopped_child_used_math(child);
564 ret = set_fpregs(child, (struct user_i387_struct __user *)data);
569 ret = ptrace_request(child, request, addr, data);
575 static void syscall_trace(struct pt_regs *regs)
579 printk("trace %s rip %lx rsp %lx rax %d origrax %d caller %lx tiflags %x ptrace %x\n",
581 regs->rip, regs->rsp, regs->rax, regs->orig_rax, __builtin_return_address(0),
582 current_thread_info()->flags, current->ptrace);
585 ptrace_notify(SIGTRAP | ((current->ptrace & PT_TRACESYSGOOD)
588 * this isn't the same as continuing with a signal, but it will do
589 * for normal use. strace only continues with a signal if the
590 * stopping signal is not SIGTRAP. -brl
592 if (current->exit_code) {
593 send_sig(current->exit_code, current, 1);
594 current->exit_code = 0;
598 asmlinkage void syscall_trace_enter(struct pt_regs *regs)
600 /* do the secure computing check first */
601 secure_computing(regs->orig_rax);
603 if (test_thread_flag(TIF_SYSCALL_TRACE)
604 && (current->ptrace & PT_PTRACED))
607 if (unlikely(current->audit_context)) {
608 if (test_thread_flag(TIF_IA32)) {
609 audit_syscall_entry(AUDIT_ARCH_I386,
611 regs->rbx, regs->rcx,
612 regs->rdx, regs->rsi);
614 audit_syscall_entry(AUDIT_ARCH_X86_64,
616 regs->rdi, regs->rsi,
617 regs->rdx, regs->r10);
622 asmlinkage void syscall_trace_leave(struct pt_regs *regs)
624 if (unlikely(current->audit_context))
625 audit_syscall_exit(AUDITSC_RESULT(regs->rax), regs->rax);
627 if ((test_thread_flag(TIF_SYSCALL_TRACE)
628 || test_thread_flag(TIF_SINGLESTEP))
629 && (current->ptrace & PT_PTRACED))