Merge tag 'pnp-extra-4.8-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
[cascardo/linux.git] / arch / x86 / kernel / ptrace.c
1 /* By Ross Biro 1/23/92 */
2 /*
3  * Pentium III FXSR, SSE support
4  *      Gareth Hughes <gareth@valinux.com>, May 2000
5  */
6
7 #include <linux/kernel.h>
8 #include <linux/sched.h>
9 #include <linux/mm.h>
10 #include <linux/smp.h>
11 #include <linux/errno.h>
12 #include <linux/slab.h>
13 #include <linux/ptrace.h>
14 #include <linux/tracehook.h>
15 #include <linux/user.h>
16 #include <linux/elf.h>
17 #include <linux/security.h>
18 #include <linux/audit.h>
19 #include <linux/seccomp.h>
20 #include <linux/signal.h>
21 #include <linux/perf_event.h>
22 #include <linux/hw_breakpoint.h>
23 #include <linux/rcupdate.h>
24 #include <linux/export.h>
25 #include <linux/context_tracking.h>
26
27 #include <asm/uaccess.h>
28 #include <asm/pgtable.h>
29 #include <asm/processor.h>
30 #include <asm/fpu/internal.h>
31 #include <asm/fpu/signal.h>
32 #include <asm/fpu/regset.h>
33 #include <asm/debugreg.h>
34 #include <asm/ldt.h>
35 #include <asm/desc.h>
36 #include <asm/prctl.h>
37 #include <asm/proto.h>
38 #include <asm/hw_breakpoint.h>
39 #include <asm/traps.h>
40 #include <asm/syscall.h>
41
42 #include "tls.h"
43
44 enum x86_regset {
45         REGSET_GENERAL,
46         REGSET_FP,
47         REGSET_XFP,
48         REGSET_IOPERM64 = REGSET_XFP,
49         REGSET_XSTATE,
50         REGSET_TLS,
51         REGSET_IOPERM32,
52 };
53
54 struct pt_regs_offset {
55         const char *name;
56         int offset;
57 };
58
59 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
60 #define REG_OFFSET_END {.name = NULL, .offset = 0}
61
62 static const struct pt_regs_offset regoffset_table[] = {
63 #ifdef CONFIG_X86_64
64         REG_OFFSET_NAME(r15),
65         REG_OFFSET_NAME(r14),
66         REG_OFFSET_NAME(r13),
67         REG_OFFSET_NAME(r12),
68         REG_OFFSET_NAME(r11),
69         REG_OFFSET_NAME(r10),
70         REG_OFFSET_NAME(r9),
71         REG_OFFSET_NAME(r8),
72 #endif
73         REG_OFFSET_NAME(bx),
74         REG_OFFSET_NAME(cx),
75         REG_OFFSET_NAME(dx),
76         REG_OFFSET_NAME(si),
77         REG_OFFSET_NAME(di),
78         REG_OFFSET_NAME(bp),
79         REG_OFFSET_NAME(ax),
80 #ifdef CONFIG_X86_32
81         REG_OFFSET_NAME(ds),
82         REG_OFFSET_NAME(es),
83         REG_OFFSET_NAME(fs),
84         REG_OFFSET_NAME(gs),
85 #endif
86         REG_OFFSET_NAME(orig_ax),
87         REG_OFFSET_NAME(ip),
88         REG_OFFSET_NAME(cs),
89         REG_OFFSET_NAME(flags),
90         REG_OFFSET_NAME(sp),
91         REG_OFFSET_NAME(ss),
92         REG_OFFSET_END,
93 };
94
95 /**
96  * regs_query_register_offset() - query register offset from its name
97  * @name:       the name of a register
98  *
99  * regs_query_register_offset() returns the offset of a register in struct
100  * pt_regs from its name. If the name is invalid, this returns -EINVAL;
101  */
102 int regs_query_register_offset(const char *name)
103 {
104         const struct pt_regs_offset *roff;
105         for (roff = regoffset_table; roff->name != NULL; roff++)
106                 if (!strcmp(roff->name, name))
107                         return roff->offset;
108         return -EINVAL;
109 }
110
111 /**
112  * regs_query_register_name() - query register name from its offset
113  * @offset:     the offset of a register in struct pt_regs.
114  *
115  * regs_query_register_name() returns the name of a register from its
116  * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
117  */
118 const char *regs_query_register_name(unsigned int offset)
119 {
120         const struct pt_regs_offset *roff;
121         for (roff = regoffset_table; roff->name != NULL; roff++)
122                 if (roff->offset == offset)
123                         return roff->name;
124         return NULL;
125 }
126
127 /*
128  * does not yet catch signals sent when the child dies.
129  * in exit.c or in signal.c.
130  */
131
132 /*
133  * Determines which flags the user has access to [1 = access, 0 = no access].
134  */
135 #define FLAG_MASK_32            ((unsigned long)                        \
136                                  (X86_EFLAGS_CF | X86_EFLAGS_PF |       \
137                                   X86_EFLAGS_AF | X86_EFLAGS_ZF |       \
138                                   X86_EFLAGS_SF | X86_EFLAGS_TF |       \
139                                   X86_EFLAGS_DF | X86_EFLAGS_OF |       \
140                                   X86_EFLAGS_RF | X86_EFLAGS_AC))
141
142 /*
143  * Determines whether a value may be installed in a segment register.
144  */
145 static inline bool invalid_selector(u16 value)
146 {
147         return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
148 }
149
150 #ifdef CONFIG_X86_32
151
152 #define FLAG_MASK               FLAG_MASK_32
153
154 /*
155  * X86_32 CPUs don't save ss and esp if the CPU is already in kernel mode
156  * when it traps.  The previous stack will be directly underneath the saved
157  * registers, and 'sp/ss' won't even have been saved. Thus the '&regs->sp'.
158  *
159  * Now, if the stack is empty, '&regs->sp' is out of range. In this
160  * case we try to take the previous stack. To always return a non-null
161  * stack pointer we fall back to regs as stack if no previous stack
162  * exists.
163  *
164  * This is valid only for kernel mode traps.
165  */
166 unsigned long kernel_stack_pointer(struct pt_regs *regs)
167 {
168         unsigned long context = (unsigned long)regs & ~(THREAD_SIZE - 1);
169         unsigned long sp = (unsigned long)&regs->sp;
170         u32 *prev_esp;
171
172         if (context == (sp & ~(THREAD_SIZE - 1)))
173                 return sp;
174
175         prev_esp = (u32 *)(context);
176         if (prev_esp)
177                 return (unsigned long)prev_esp;
178
179         return (unsigned long)regs;
180 }
181 EXPORT_SYMBOL_GPL(kernel_stack_pointer);
182
183 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
184 {
185         BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
186         return &regs->bx + (regno >> 2);
187 }
188
189 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
190 {
191         /*
192          * Returning the value truncates it to 16 bits.
193          */
194         unsigned int retval;
195         if (offset != offsetof(struct user_regs_struct, gs))
196                 retval = *pt_regs_access(task_pt_regs(task), offset);
197         else {
198                 if (task == current)
199                         retval = get_user_gs(task_pt_regs(task));
200                 else
201                         retval = task_user_gs(task);
202         }
203         return retval;
204 }
205
206 static int set_segment_reg(struct task_struct *task,
207                            unsigned long offset, u16 value)
208 {
209         /*
210          * The value argument was already truncated to 16 bits.
211          */
212         if (invalid_selector(value))
213                 return -EIO;
214
215         /*
216          * For %cs and %ss we cannot permit a null selector.
217          * We can permit a bogus selector as long as it has USER_RPL.
218          * Null selectors are fine for other segment registers, but
219          * we will never get back to user mode with invalid %cs or %ss
220          * and will take the trap in iret instead.  Much code relies
221          * on user_mode() to distinguish a user trap frame (which can
222          * safely use invalid selectors) from a kernel trap frame.
223          */
224         switch (offset) {
225         case offsetof(struct user_regs_struct, cs):
226         case offsetof(struct user_regs_struct, ss):
227                 if (unlikely(value == 0))
228                         return -EIO;
229
230         default:
231                 *pt_regs_access(task_pt_regs(task), offset) = value;
232                 break;
233
234         case offsetof(struct user_regs_struct, gs):
235                 if (task == current)
236                         set_user_gs(task_pt_regs(task), value);
237                 else
238                         task_user_gs(task) = value;
239         }
240
241         return 0;
242 }
243
244 #else  /* CONFIG_X86_64 */
245
246 #define FLAG_MASK               (FLAG_MASK_32 | X86_EFLAGS_NT)
247
248 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
249 {
250         BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
251         return &regs->r15 + (offset / sizeof(regs->r15));
252 }
253
254 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
255 {
256         /*
257          * Returning the value truncates it to 16 bits.
258          */
259         unsigned int seg;
260
261         switch (offset) {
262         case offsetof(struct user_regs_struct, fs):
263                 if (task == current) {
264                         /* Older gas can't assemble movq %?s,%r?? */
265                         asm("movl %%fs,%0" : "=r" (seg));
266                         return seg;
267                 }
268                 return task->thread.fsindex;
269         case offsetof(struct user_regs_struct, gs):
270                 if (task == current) {
271                         asm("movl %%gs,%0" : "=r" (seg));
272                         return seg;
273                 }
274                 return task->thread.gsindex;
275         case offsetof(struct user_regs_struct, ds):
276                 if (task == current) {
277                         asm("movl %%ds,%0" : "=r" (seg));
278                         return seg;
279                 }
280                 return task->thread.ds;
281         case offsetof(struct user_regs_struct, es):
282                 if (task == current) {
283                         asm("movl %%es,%0" : "=r" (seg));
284                         return seg;
285                 }
286                 return task->thread.es;
287
288         case offsetof(struct user_regs_struct, cs):
289         case offsetof(struct user_regs_struct, ss):
290                 break;
291         }
292         return *pt_regs_access(task_pt_regs(task), offset);
293 }
294
295 static int set_segment_reg(struct task_struct *task,
296                            unsigned long offset, u16 value)
297 {
298         /*
299          * The value argument was already truncated to 16 bits.
300          */
301         if (invalid_selector(value))
302                 return -EIO;
303
304         switch (offset) {
305         case offsetof(struct user_regs_struct,fs):
306                 task->thread.fsindex = value;
307                 if (task == current)
308                         loadsegment(fs, task->thread.fsindex);
309                 break;
310         case offsetof(struct user_regs_struct,gs):
311                 task->thread.gsindex = value;
312                 if (task == current)
313                         load_gs_index(task->thread.gsindex);
314                 break;
315         case offsetof(struct user_regs_struct,ds):
316                 task->thread.ds = value;
317                 if (task == current)
318                         loadsegment(ds, task->thread.ds);
319                 break;
320         case offsetof(struct user_regs_struct,es):
321                 task->thread.es = value;
322                 if (task == current)
323                         loadsegment(es, task->thread.es);
324                 break;
325
326                 /*
327                  * Can't actually change these in 64-bit mode.
328                  */
329         case offsetof(struct user_regs_struct,cs):
330                 if (unlikely(value == 0))
331                         return -EIO;
332                 task_pt_regs(task)->cs = value;
333                 break;
334         case offsetof(struct user_regs_struct,ss):
335                 if (unlikely(value == 0))
336                         return -EIO;
337                 task_pt_regs(task)->ss = value;
338                 break;
339         }
340
341         return 0;
342 }
343
344 #endif  /* CONFIG_X86_32 */
345
346 static unsigned long get_flags(struct task_struct *task)
347 {
348         unsigned long retval = task_pt_regs(task)->flags;
349
350         /*
351          * If the debugger set TF, hide it from the readout.
352          */
353         if (test_tsk_thread_flag(task, TIF_FORCED_TF))
354                 retval &= ~X86_EFLAGS_TF;
355
356         return retval;
357 }
358
359 static int set_flags(struct task_struct *task, unsigned long value)
360 {
361         struct pt_regs *regs = task_pt_regs(task);
362
363         /*
364          * If the user value contains TF, mark that
365          * it was not "us" (the debugger) that set it.
366          * If not, make sure it stays set if we had.
367          */
368         if (value & X86_EFLAGS_TF)
369                 clear_tsk_thread_flag(task, TIF_FORCED_TF);
370         else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
371                 value |= X86_EFLAGS_TF;
372
373         regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
374
375         return 0;
376 }
377
378 static int putreg(struct task_struct *child,
379                   unsigned long offset, unsigned long value)
380 {
381         switch (offset) {
382         case offsetof(struct user_regs_struct, cs):
383         case offsetof(struct user_regs_struct, ds):
384         case offsetof(struct user_regs_struct, es):
385         case offsetof(struct user_regs_struct, fs):
386         case offsetof(struct user_regs_struct, gs):
387         case offsetof(struct user_regs_struct, ss):
388                 return set_segment_reg(child, offset, value);
389
390         case offsetof(struct user_regs_struct, flags):
391                 return set_flags(child, value);
392
393 #ifdef CONFIG_X86_64
394         case offsetof(struct user_regs_struct,fs_base):
395                 if (value >= TASK_SIZE_MAX)
396                         return -EIO;
397                 /*
398                  * When changing the segment base, use do_arch_prctl
399                  * to set either thread.fs or thread.fsindex and the
400                  * corresponding GDT slot.
401                  */
402                 if (child->thread.fsbase != value)
403                         return do_arch_prctl(child, ARCH_SET_FS, value);
404                 return 0;
405         case offsetof(struct user_regs_struct,gs_base):
406                 /*
407                  * Exactly the same here as the %fs handling above.
408                  */
409                 if (value >= TASK_SIZE_MAX)
410                         return -EIO;
411                 if (child->thread.gsbase != value)
412                         return do_arch_prctl(child, ARCH_SET_GS, value);
413                 return 0;
414 #endif
415         }
416
417         *pt_regs_access(task_pt_regs(child), offset) = value;
418         return 0;
419 }
420
421 static unsigned long getreg(struct task_struct *task, unsigned long offset)
422 {
423         switch (offset) {
424         case offsetof(struct user_regs_struct, cs):
425         case offsetof(struct user_regs_struct, ds):
426         case offsetof(struct user_regs_struct, es):
427         case offsetof(struct user_regs_struct, fs):
428         case offsetof(struct user_regs_struct, gs):
429         case offsetof(struct user_regs_struct, ss):
430                 return get_segment_reg(task, offset);
431
432         case offsetof(struct user_regs_struct, flags):
433                 return get_flags(task);
434
435 #ifdef CONFIG_X86_64
436         case offsetof(struct user_regs_struct, fs_base): {
437                 /*
438                  * XXX: This will not behave as expected if called on
439                  * current or if fsindex != 0.
440                  */
441                 return task->thread.fsbase;
442         }
443         case offsetof(struct user_regs_struct, gs_base): {
444                 /*
445                  * XXX: This will not behave as expected if called on
446                  * current or if fsindex != 0.
447                  */
448                 return task->thread.gsbase;
449         }
450 #endif
451         }
452
453         return *pt_regs_access(task_pt_regs(task), offset);
454 }
455
456 static int genregs_get(struct task_struct *target,
457                        const struct user_regset *regset,
458                        unsigned int pos, unsigned int count,
459                        void *kbuf, void __user *ubuf)
460 {
461         if (kbuf) {
462                 unsigned long *k = kbuf;
463                 while (count >= sizeof(*k)) {
464                         *k++ = getreg(target, pos);
465                         count -= sizeof(*k);
466                         pos += sizeof(*k);
467                 }
468         } else {
469                 unsigned long __user *u = ubuf;
470                 while (count >= sizeof(*u)) {
471                         if (__put_user(getreg(target, pos), u++))
472                                 return -EFAULT;
473                         count -= sizeof(*u);
474                         pos += sizeof(*u);
475                 }
476         }
477
478         return 0;
479 }
480
481 static int genregs_set(struct task_struct *target,
482                        const struct user_regset *regset,
483                        unsigned int pos, unsigned int count,
484                        const void *kbuf, const void __user *ubuf)
485 {
486         int ret = 0;
487         if (kbuf) {
488                 const unsigned long *k = kbuf;
489                 while (count >= sizeof(*k) && !ret) {
490                         ret = putreg(target, pos, *k++);
491                         count -= sizeof(*k);
492                         pos += sizeof(*k);
493                 }
494         } else {
495                 const unsigned long  __user *u = ubuf;
496                 while (count >= sizeof(*u) && !ret) {
497                         unsigned long word;
498                         ret = __get_user(word, u++);
499                         if (ret)
500                                 break;
501                         ret = putreg(target, pos, word);
502                         count -= sizeof(*u);
503                         pos += sizeof(*u);
504                 }
505         }
506         return ret;
507 }
508
509 static void ptrace_triggered(struct perf_event *bp,
510                              struct perf_sample_data *data,
511                              struct pt_regs *regs)
512 {
513         int i;
514         struct thread_struct *thread = &(current->thread);
515
516         /*
517          * Store in the virtual DR6 register the fact that the breakpoint
518          * was hit so the thread's debugger will see it.
519          */
520         for (i = 0; i < HBP_NUM; i++) {
521                 if (thread->ptrace_bps[i] == bp)
522                         break;
523         }
524
525         thread->debugreg6 |= (DR_TRAP0 << i);
526 }
527
528 /*
529  * Walk through every ptrace breakpoints for this thread and
530  * build the dr7 value on top of their attributes.
531  *
532  */
533 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
534 {
535         int i;
536         int dr7 = 0;
537         struct arch_hw_breakpoint *info;
538
539         for (i = 0; i < HBP_NUM; i++) {
540                 if (bp[i] && !bp[i]->attr.disabled) {
541                         info = counter_arch_bp(bp[i]);
542                         dr7 |= encode_dr7(i, info->len, info->type);
543                 }
544         }
545
546         return dr7;
547 }
548
549 static int ptrace_fill_bp_fields(struct perf_event_attr *attr,
550                                         int len, int type, bool disabled)
551 {
552         int err, bp_len, bp_type;
553
554         err = arch_bp_generic_fields(len, type, &bp_len, &bp_type);
555         if (!err) {
556                 attr->bp_len = bp_len;
557                 attr->bp_type = bp_type;
558                 attr->disabled = disabled;
559         }
560
561         return err;
562 }
563
564 static struct perf_event *
565 ptrace_register_breakpoint(struct task_struct *tsk, int len, int type,
566                                 unsigned long addr, bool disabled)
567 {
568         struct perf_event_attr attr;
569         int err;
570
571         ptrace_breakpoint_init(&attr);
572         attr.bp_addr = addr;
573
574         err = ptrace_fill_bp_fields(&attr, len, type, disabled);
575         if (err)
576                 return ERR_PTR(err);
577
578         return register_user_hw_breakpoint(&attr, ptrace_triggered,
579                                                  NULL, tsk);
580 }
581
582 static int ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
583                                         int disabled)
584 {
585         struct perf_event_attr attr = bp->attr;
586         int err;
587
588         err = ptrace_fill_bp_fields(&attr, len, type, disabled);
589         if (err)
590                 return err;
591
592         return modify_user_hw_breakpoint(bp, &attr);
593 }
594
595 /*
596  * Handle ptrace writes to debug register 7.
597  */
598 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
599 {
600         struct thread_struct *thread = &tsk->thread;
601         unsigned long old_dr7;
602         bool second_pass = false;
603         int i, rc, ret = 0;
604
605         data &= ~DR_CONTROL_RESERVED;
606         old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
607
608 restore:
609         rc = 0;
610         for (i = 0; i < HBP_NUM; i++) {
611                 unsigned len, type;
612                 bool disabled = !decode_dr7(data, i, &len, &type);
613                 struct perf_event *bp = thread->ptrace_bps[i];
614
615                 if (!bp) {
616                         if (disabled)
617                                 continue;
618
619                         bp = ptrace_register_breakpoint(tsk,
620                                         len, type, 0, disabled);
621                         if (IS_ERR(bp)) {
622                                 rc = PTR_ERR(bp);
623                                 break;
624                         }
625
626                         thread->ptrace_bps[i] = bp;
627                         continue;
628                 }
629
630                 rc = ptrace_modify_breakpoint(bp, len, type, disabled);
631                 if (rc)
632                         break;
633         }
634
635         /* Restore if the first pass failed, second_pass shouldn't fail. */
636         if (rc && !WARN_ON(second_pass)) {
637                 ret = rc;
638                 data = old_dr7;
639                 second_pass = true;
640                 goto restore;
641         }
642
643         return ret;
644 }
645
646 /*
647  * Handle PTRACE_PEEKUSR calls for the debug register area.
648  */
649 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
650 {
651         struct thread_struct *thread = &tsk->thread;
652         unsigned long val = 0;
653
654         if (n < HBP_NUM) {
655                 struct perf_event *bp = thread->ptrace_bps[n];
656
657                 if (bp)
658                         val = bp->hw.info.address;
659         } else if (n == 6) {
660                 val = thread->debugreg6;
661         } else if (n == 7) {
662                 val = thread->ptrace_dr7;
663         }
664         return val;
665 }
666
667 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
668                                       unsigned long addr)
669 {
670         struct thread_struct *t = &tsk->thread;
671         struct perf_event *bp = t->ptrace_bps[nr];
672         int err = 0;
673
674         if (!bp) {
675                 /*
676                  * Put stub len and type to create an inactive but correct bp.
677                  *
678                  * CHECKME: the previous code returned -EIO if the addr wasn't
679                  * a valid task virtual addr. The new one will return -EINVAL in
680                  *  this case.
681                  * -EINVAL may be what we want for in-kernel breakpoints users,
682                  * but -EIO looks better for ptrace, since we refuse a register
683                  * writing for the user. And anyway this is the previous
684                  * behaviour.
685                  */
686                 bp = ptrace_register_breakpoint(tsk,
687                                 X86_BREAKPOINT_LEN_1, X86_BREAKPOINT_WRITE,
688                                 addr, true);
689                 if (IS_ERR(bp))
690                         err = PTR_ERR(bp);
691                 else
692                         t->ptrace_bps[nr] = bp;
693         } else {
694                 struct perf_event_attr attr = bp->attr;
695
696                 attr.bp_addr = addr;
697                 err = modify_user_hw_breakpoint(bp, &attr);
698         }
699
700         return err;
701 }
702
703 /*
704  * Handle PTRACE_POKEUSR calls for the debug register area.
705  */
706 static int ptrace_set_debugreg(struct task_struct *tsk, int n,
707                                unsigned long val)
708 {
709         struct thread_struct *thread = &tsk->thread;
710         /* There are no DR4 or DR5 registers */
711         int rc = -EIO;
712
713         if (n < HBP_NUM) {
714                 rc = ptrace_set_breakpoint_addr(tsk, n, val);
715         } else if (n == 6) {
716                 thread->debugreg6 = val;
717                 rc = 0;
718         } else if (n == 7) {
719                 rc = ptrace_write_dr7(tsk, val);
720                 if (!rc)
721                         thread->ptrace_dr7 = val;
722         }
723         return rc;
724 }
725
726 /*
727  * These access the current or another (stopped) task's io permission
728  * bitmap for debugging or core dump.
729  */
730 static int ioperm_active(struct task_struct *target,
731                          const struct user_regset *regset)
732 {
733         return target->thread.io_bitmap_max / regset->size;
734 }
735
736 static int ioperm_get(struct task_struct *target,
737                       const struct user_regset *regset,
738                       unsigned int pos, unsigned int count,
739                       void *kbuf, void __user *ubuf)
740 {
741         if (!target->thread.io_bitmap_ptr)
742                 return -ENXIO;
743
744         return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
745                                    target->thread.io_bitmap_ptr,
746                                    0, IO_BITMAP_BYTES);
747 }
748
749 /*
750  * Called by kernel/ptrace.c when detaching..
751  *
752  * Make sure the single step bit is not set.
753  */
754 void ptrace_disable(struct task_struct *child)
755 {
756         user_disable_single_step(child);
757 #ifdef TIF_SYSCALL_EMU
758         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
759 #endif
760 }
761
762 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
763 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
764 #endif
765
766 long arch_ptrace(struct task_struct *child, long request,
767                  unsigned long addr, unsigned long data)
768 {
769         int ret;
770         unsigned long __user *datap = (unsigned long __user *)data;
771
772         switch (request) {
773         /* read the word at location addr in the USER area. */
774         case PTRACE_PEEKUSR: {
775                 unsigned long tmp;
776
777                 ret = -EIO;
778                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
779                         break;
780
781                 tmp = 0;  /* Default return condition */
782                 if (addr < sizeof(struct user_regs_struct))
783                         tmp = getreg(child, addr);
784                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
785                          addr <= offsetof(struct user, u_debugreg[7])) {
786                         addr -= offsetof(struct user, u_debugreg[0]);
787                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
788                 }
789                 ret = put_user(tmp, datap);
790                 break;
791         }
792
793         case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
794                 ret = -EIO;
795                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
796                         break;
797
798                 if (addr < sizeof(struct user_regs_struct))
799                         ret = putreg(child, addr, data);
800                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
801                          addr <= offsetof(struct user, u_debugreg[7])) {
802                         addr -= offsetof(struct user, u_debugreg[0]);
803                         ret = ptrace_set_debugreg(child,
804                                                   addr / sizeof(data), data);
805                 }
806                 break;
807
808         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
809                 return copy_regset_to_user(child,
810                                            task_user_regset_view(current),
811                                            REGSET_GENERAL,
812                                            0, sizeof(struct user_regs_struct),
813                                            datap);
814
815         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
816                 return copy_regset_from_user(child,
817                                              task_user_regset_view(current),
818                                              REGSET_GENERAL,
819                                              0, sizeof(struct user_regs_struct),
820                                              datap);
821
822         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
823                 return copy_regset_to_user(child,
824                                            task_user_regset_view(current),
825                                            REGSET_FP,
826                                            0, sizeof(struct user_i387_struct),
827                                            datap);
828
829         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
830                 return copy_regset_from_user(child,
831                                              task_user_regset_view(current),
832                                              REGSET_FP,
833                                              0, sizeof(struct user_i387_struct),
834                                              datap);
835
836 #ifdef CONFIG_X86_32
837         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
838                 return copy_regset_to_user(child, &user_x86_32_view,
839                                            REGSET_XFP,
840                                            0, sizeof(struct user_fxsr_struct),
841                                            datap) ? -EIO : 0;
842
843         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
844                 return copy_regset_from_user(child, &user_x86_32_view,
845                                              REGSET_XFP,
846                                              0, sizeof(struct user_fxsr_struct),
847                                              datap) ? -EIO : 0;
848 #endif
849
850 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
851         case PTRACE_GET_THREAD_AREA:
852                 if ((int) addr < 0)
853                         return -EIO;
854                 ret = do_get_thread_area(child, addr,
855                                         (struct user_desc __user *)data);
856                 break;
857
858         case PTRACE_SET_THREAD_AREA:
859                 if ((int) addr < 0)
860                         return -EIO;
861                 ret = do_set_thread_area(child, addr,
862                                         (struct user_desc __user *)data, 0);
863                 break;
864 #endif
865
866 #ifdef CONFIG_X86_64
867                 /* normal 64bit interface to access TLS data.
868                    Works just like arch_prctl, except that the arguments
869                    are reversed. */
870         case PTRACE_ARCH_PRCTL:
871                 ret = do_arch_prctl(child, data, addr);
872                 break;
873 #endif
874
875         default:
876                 ret = ptrace_request(child, request, addr, data);
877                 break;
878         }
879
880         return ret;
881 }
882
883 #ifdef CONFIG_IA32_EMULATION
884
885 #include <linux/compat.h>
886 #include <linux/syscalls.h>
887 #include <asm/ia32.h>
888 #include <asm/user32.h>
889
890 #define R32(l,q)                                                        \
891         case offsetof(struct user32, regs.l):                           \
892                 regs->q = value; break
893
894 #define SEG32(rs)                                                       \
895         case offsetof(struct user32, regs.rs):                          \
896                 return set_segment_reg(child,                           \
897                                        offsetof(struct user_regs_struct, rs), \
898                                        value);                          \
899                 break
900
901 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
902 {
903         struct pt_regs *regs = task_pt_regs(child);
904
905         switch (regno) {
906
907         SEG32(cs);
908         SEG32(ds);
909         SEG32(es);
910         SEG32(fs);
911         SEG32(gs);
912         SEG32(ss);
913
914         R32(ebx, bx);
915         R32(ecx, cx);
916         R32(edx, dx);
917         R32(edi, di);
918         R32(esi, si);
919         R32(ebp, bp);
920         R32(eax, ax);
921         R32(eip, ip);
922         R32(esp, sp);
923
924         case offsetof(struct user32, regs.orig_eax):
925                 /*
926                  * A 32-bit debugger setting orig_eax means to restore
927                  * the state of the task restarting a 32-bit syscall.
928                  * Make sure we interpret the -ERESTART* codes correctly
929                  * in case the task is not actually still sitting at the
930                  * exit from a 32-bit syscall with TS_COMPAT still set.
931                  */
932                 regs->orig_ax = value;
933                 if (syscall_get_nr(child, regs) >= 0)
934                         task_thread_info(child)->status |= TS_COMPAT;
935                 break;
936
937         case offsetof(struct user32, regs.eflags):
938                 return set_flags(child, value);
939
940         case offsetof(struct user32, u_debugreg[0]) ...
941                 offsetof(struct user32, u_debugreg[7]):
942                 regno -= offsetof(struct user32, u_debugreg[0]);
943                 return ptrace_set_debugreg(child, regno / 4, value);
944
945         default:
946                 if (regno > sizeof(struct user32) || (regno & 3))
947                         return -EIO;
948
949                 /*
950                  * Other dummy fields in the virtual user structure
951                  * are ignored
952                  */
953                 break;
954         }
955         return 0;
956 }
957
958 #undef R32
959 #undef SEG32
960
961 #define R32(l,q)                                                        \
962         case offsetof(struct user32, regs.l):                           \
963                 *val = regs->q; break
964
965 #define SEG32(rs)                                                       \
966         case offsetof(struct user32, regs.rs):                          \
967                 *val = get_segment_reg(child,                           \
968                                        offsetof(struct user_regs_struct, rs)); \
969                 break
970
971 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
972 {
973         struct pt_regs *regs = task_pt_regs(child);
974
975         switch (regno) {
976
977         SEG32(ds);
978         SEG32(es);
979         SEG32(fs);
980         SEG32(gs);
981
982         R32(cs, cs);
983         R32(ss, ss);
984         R32(ebx, bx);
985         R32(ecx, cx);
986         R32(edx, dx);
987         R32(edi, di);
988         R32(esi, si);
989         R32(ebp, bp);
990         R32(eax, ax);
991         R32(orig_eax, orig_ax);
992         R32(eip, ip);
993         R32(esp, sp);
994
995         case offsetof(struct user32, regs.eflags):
996                 *val = get_flags(child);
997                 break;
998
999         case offsetof(struct user32, u_debugreg[0]) ...
1000                 offsetof(struct user32, u_debugreg[7]):
1001                 regno -= offsetof(struct user32, u_debugreg[0]);
1002                 *val = ptrace_get_debugreg(child, regno / 4);
1003                 break;
1004
1005         default:
1006                 if (regno > sizeof(struct user32) || (regno & 3))
1007                         return -EIO;
1008
1009                 /*
1010                  * Other dummy fields in the virtual user structure
1011                  * are ignored
1012                  */
1013                 *val = 0;
1014                 break;
1015         }
1016         return 0;
1017 }
1018
1019 #undef R32
1020 #undef SEG32
1021
1022 static int genregs32_get(struct task_struct *target,
1023                          const struct user_regset *regset,
1024                          unsigned int pos, unsigned int count,
1025                          void *kbuf, void __user *ubuf)
1026 {
1027         if (kbuf) {
1028                 compat_ulong_t *k = kbuf;
1029                 while (count >= sizeof(*k)) {
1030                         getreg32(target, pos, k++);
1031                         count -= sizeof(*k);
1032                         pos += sizeof(*k);
1033                 }
1034         } else {
1035                 compat_ulong_t __user *u = ubuf;
1036                 while (count >= sizeof(*u)) {
1037                         compat_ulong_t word;
1038                         getreg32(target, pos, &word);
1039                         if (__put_user(word, u++))
1040                                 return -EFAULT;
1041                         count -= sizeof(*u);
1042                         pos += sizeof(*u);
1043                 }
1044         }
1045
1046         return 0;
1047 }
1048
1049 static int genregs32_set(struct task_struct *target,
1050                          const struct user_regset *regset,
1051                          unsigned int pos, unsigned int count,
1052                          const void *kbuf, const void __user *ubuf)
1053 {
1054         int ret = 0;
1055         if (kbuf) {
1056                 const compat_ulong_t *k = kbuf;
1057                 while (count >= sizeof(*k) && !ret) {
1058                         ret = putreg32(target, pos, *k++);
1059                         count -= sizeof(*k);
1060                         pos += sizeof(*k);
1061                 }
1062         } else {
1063                 const compat_ulong_t __user *u = ubuf;
1064                 while (count >= sizeof(*u) && !ret) {
1065                         compat_ulong_t word;
1066                         ret = __get_user(word, u++);
1067                         if (ret)
1068                                 break;
1069                         ret = putreg32(target, pos, word);
1070                         count -= sizeof(*u);
1071                         pos += sizeof(*u);
1072                 }
1073         }
1074         return ret;
1075 }
1076
1077 static long ia32_arch_ptrace(struct task_struct *child, compat_long_t request,
1078                              compat_ulong_t caddr, compat_ulong_t cdata)
1079 {
1080         unsigned long addr = caddr;
1081         unsigned long data = cdata;
1082         void __user *datap = compat_ptr(data);
1083         int ret;
1084         __u32 val;
1085
1086         switch (request) {
1087         case PTRACE_PEEKUSR:
1088                 ret = getreg32(child, addr, &val);
1089                 if (ret == 0)
1090                         ret = put_user(val, (__u32 __user *)datap);
1091                 break;
1092
1093         case PTRACE_POKEUSR:
1094                 ret = putreg32(child, addr, data);
1095                 break;
1096
1097         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1098                 return copy_regset_to_user(child, &user_x86_32_view,
1099                                            REGSET_GENERAL,
1100                                            0, sizeof(struct user_regs_struct32),
1101                                            datap);
1102
1103         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1104                 return copy_regset_from_user(child, &user_x86_32_view,
1105                                              REGSET_GENERAL, 0,
1106                                              sizeof(struct user_regs_struct32),
1107                                              datap);
1108
1109         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1110                 return copy_regset_to_user(child, &user_x86_32_view,
1111                                            REGSET_FP, 0,
1112                                            sizeof(struct user_i387_ia32_struct),
1113                                            datap);
1114
1115         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1116                 return copy_regset_from_user(
1117                         child, &user_x86_32_view, REGSET_FP,
1118                         0, sizeof(struct user_i387_ia32_struct), datap);
1119
1120         case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1121                 return copy_regset_to_user(child, &user_x86_32_view,
1122                                            REGSET_XFP, 0,
1123                                            sizeof(struct user32_fxsr_struct),
1124                                            datap);
1125
1126         case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1127                 return copy_regset_from_user(child, &user_x86_32_view,
1128                                              REGSET_XFP, 0,
1129                                              sizeof(struct user32_fxsr_struct),
1130                                              datap);
1131
1132         case PTRACE_GET_THREAD_AREA:
1133         case PTRACE_SET_THREAD_AREA:
1134                 return arch_ptrace(child, request, addr, data);
1135
1136         default:
1137                 return compat_ptrace_request(child, request, addr, data);
1138         }
1139
1140         return ret;
1141 }
1142 #endif /* CONFIG_IA32_EMULATION */
1143
1144 #ifdef CONFIG_X86_X32_ABI
1145 static long x32_arch_ptrace(struct task_struct *child,
1146                             compat_long_t request, compat_ulong_t caddr,
1147                             compat_ulong_t cdata)
1148 {
1149         unsigned long addr = caddr;
1150         unsigned long data = cdata;
1151         void __user *datap = compat_ptr(data);
1152         int ret;
1153
1154         switch (request) {
1155         /* Read 32bits at location addr in the USER area.  Only allow
1156            to return the lower 32bits of segment and debug registers.  */
1157         case PTRACE_PEEKUSR: {
1158                 u32 tmp;
1159
1160                 ret = -EIO;
1161                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1162                     addr < offsetof(struct user_regs_struct, cs))
1163                         break;
1164
1165                 tmp = 0;  /* Default return condition */
1166                 if (addr < sizeof(struct user_regs_struct))
1167                         tmp = getreg(child, addr);
1168                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1169                          addr <= offsetof(struct user, u_debugreg[7])) {
1170                         addr -= offsetof(struct user, u_debugreg[0]);
1171                         tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1172                 }
1173                 ret = put_user(tmp, (__u32 __user *)datap);
1174                 break;
1175         }
1176
1177         /* Write the word at location addr in the USER area.  Only allow
1178            to update segment and debug registers with the upper 32bits
1179            zero-extended. */
1180         case PTRACE_POKEUSR:
1181                 ret = -EIO;
1182                 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1183                     addr < offsetof(struct user_regs_struct, cs))
1184                         break;
1185
1186                 if (addr < sizeof(struct user_regs_struct))
1187                         ret = putreg(child, addr, data);
1188                 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1189                          addr <= offsetof(struct user, u_debugreg[7])) {
1190                         addr -= offsetof(struct user, u_debugreg[0]);
1191                         ret = ptrace_set_debugreg(child,
1192                                                   addr / sizeof(data), data);
1193                 }
1194                 break;
1195
1196         case PTRACE_GETREGS:    /* Get all gp regs from the child. */
1197                 return copy_regset_to_user(child,
1198                                            task_user_regset_view(current),
1199                                            REGSET_GENERAL,
1200                                            0, sizeof(struct user_regs_struct),
1201                                            datap);
1202
1203         case PTRACE_SETREGS:    /* Set all gp regs in the child. */
1204                 return copy_regset_from_user(child,
1205                                              task_user_regset_view(current),
1206                                              REGSET_GENERAL,
1207                                              0, sizeof(struct user_regs_struct),
1208                                              datap);
1209
1210         case PTRACE_GETFPREGS:  /* Get the child FPU state. */
1211                 return copy_regset_to_user(child,
1212                                            task_user_regset_view(current),
1213                                            REGSET_FP,
1214                                            0, sizeof(struct user_i387_struct),
1215                                            datap);
1216
1217         case PTRACE_SETFPREGS:  /* Set the child FPU state. */
1218                 return copy_regset_from_user(child,
1219                                              task_user_regset_view(current),
1220                                              REGSET_FP,
1221                                              0, sizeof(struct user_i387_struct),
1222                                              datap);
1223
1224         default:
1225                 return compat_ptrace_request(child, request, addr, data);
1226         }
1227
1228         return ret;
1229 }
1230 #endif
1231
1232 #ifdef CONFIG_COMPAT
1233 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1234                         compat_ulong_t caddr, compat_ulong_t cdata)
1235 {
1236 #ifdef CONFIG_X86_X32_ABI
1237         if (!in_ia32_syscall())
1238                 return x32_arch_ptrace(child, request, caddr, cdata);
1239 #endif
1240 #ifdef CONFIG_IA32_EMULATION
1241         return ia32_arch_ptrace(child, request, caddr, cdata);
1242 #else
1243         return 0;
1244 #endif
1245 }
1246 #endif  /* CONFIG_COMPAT */
1247
1248 #ifdef CONFIG_X86_64
1249
1250 static struct user_regset x86_64_regsets[] __read_mostly = {
1251         [REGSET_GENERAL] = {
1252                 .core_note_type = NT_PRSTATUS,
1253                 .n = sizeof(struct user_regs_struct) / sizeof(long),
1254                 .size = sizeof(long), .align = sizeof(long),
1255                 .get = genregs_get, .set = genregs_set
1256         },
1257         [REGSET_FP] = {
1258                 .core_note_type = NT_PRFPREG,
1259                 .n = sizeof(struct user_i387_struct) / sizeof(long),
1260                 .size = sizeof(long), .align = sizeof(long),
1261                 .active = regset_xregset_fpregs_active, .get = xfpregs_get, .set = xfpregs_set
1262         },
1263         [REGSET_XSTATE] = {
1264                 .core_note_type = NT_X86_XSTATE,
1265                 .size = sizeof(u64), .align = sizeof(u64),
1266                 .active = xstateregs_active, .get = xstateregs_get,
1267                 .set = xstateregs_set
1268         },
1269         [REGSET_IOPERM64] = {
1270                 .core_note_type = NT_386_IOPERM,
1271                 .n = IO_BITMAP_LONGS,
1272                 .size = sizeof(long), .align = sizeof(long),
1273                 .active = ioperm_active, .get = ioperm_get
1274         },
1275 };
1276
1277 static const struct user_regset_view user_x86_64_view = {
1278         .name = "x86_64", .e_machine = EM_X86_64,
1279         .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1280 };
1281
1282 #else  /* CONFIG_X86_32 */
1283
1284 #define user_regs_struct32      user_regs_struct
1285 #define genregs32_get           genregs_get
1286 #define genregs32_set           genregs_set
1287
1288 #endif  /* CONFIG_X86_64 */
1289
1290 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1291 static struct user_regset x86_32_regsets[] __read_mostly = {
1292         [REGSET_GENERAL] = {
1293                 .core_note_type = NT_PRSTATUS,
1294                 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1295                 .size = sizeof(u32), .align = sizeof(u32),
1296                 .get = genregs32_get, .set = genregs32_set
1297         },
1298         [REGSET_FP] = {
1299                 .core_note_type = NT_PRFPREG,
1300                 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1301                 .size = sizeof(u32), .align = sizeof(u32),
1302                 .active = regset_fpregs_active, .get = fpregs_get, .set = fpregs_set
1303         },
1304         [REGSET_XFP] = {
1305                 .core_note_type = NT_PRXFPREG,
1306                 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1307                 .size = sizeof(u32), .align = sizeof(u32),
1308                 .active = regset_xregset_fpregs_active, .get = xfpregs_get, .set = xfpregs_set
1309         },
1310         [REGSET_XSTATE] = {
1311                 .core_note_type = NT_X86_XSTATE,
1312                 .size = sizeof(u64), .align = sizeof(u64),
1313                 .active = xstateregs_active, .get = xstateregs_get,
1314                 .set = xstateregs_set
1315         },
1316         [REGSET_TLS] = {
1317                 .core_note_type = NT_386_TLS,
1318                 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1319                 .size = sizeof(struct user_desc),
1320                 .align = sizeof(struct user_desc),
1321                 .active = regset_tls_active,
1322                 .get = regset_tls_get, .set = regset_tls_set
1323         },
1324         [REGSET_IOPERM32] = {
1325                 .core_note_type = NT_386_IOPERM,
1326                 .n = IO_BITMAP_BYTES / sizeof(u32),
1327                 .size = sizeof(u32), .align = sizeof(u32),
1328                 .active = ioperm_active, .get = ioperm_get
1329         },
1330 };
1331
1332 static const struct user_regset_view user_x86_32_view = {
1333         .name = "i386", .e_machine = EM_386,
1334         .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1335 };
1336 #endif
1337
1338 /*
1339  * This represents bytes 464..511 in the memory layout exported through
1340  * the REGSET_XSTATE interface.
1341  */
1342 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1343
1344 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1345 {
1346 #ifdef CONFIG_X86_64
1347         x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1348 #endif
1349 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1350         x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1351 #endif
1352         xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1353 }
1354
1355 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1356 {
1357 #ifdef CONFIG_IA32_EMULATION
1358         if (test_tsk_thread_flag(task, TIF_IA32))
1359 #endif
1360 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1361                 return &user_x86_32_view;
1362 #endif
1363 #ifdef CONFIG_X86_64
1364         return &user_x86_64_view;
1365 #endif
1366 }
1367
1368 static void fill_sigtrap_info(struct task_struct *tsk,
1369                                 struct pt_regs *regs,
1370                                 int error_code, int si_code,
1371                                 struct siginfo *info)
1372 {
1373         tsk->thread.trap_nr = X86_TRAP_DB;
1374         tsk->thread.error_code = error_code;
1375
1376         memset(info, 0, sizeof(*info));
1377         info->si_signo = SIGTRAP;
1378         info->si_code = si_code;
1379         info->si_addr = user_mode(regs) ? (void __user *)regs->ip : NULL;
1380 }
1381
1382 void user_single_step_siginfo(struct task_struct *tsk,
1383                                 struct pt_regs *regs,
1384                                 struct siginfo *info)
1385 {
1386         fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1387 }
1388
1389 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1390                                          int error_code, int si_code)
1391 {
1392         struct siginfo info;
1393
1394         fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1395         /* Send us the fake SIGTRAP */
1396         force_sig_info(SIGTRAP, &info, tsk);
1397 }