]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/x86/kernel/ptrace.c
i387: Split up <asm/i387.h> into exported and internal interfaces
[mirror_ubuntu-artful-kernel.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/regset.h>
15 #include <linux/tracehook.h>
16 #include <linux/user.h>
17 #include <linux/elf.h>
18 #include <linux/security.h>
19 #include <linux/audit.h>
20 #include <linux/seccomp.h>
21 #include <linux/signal.h>
22 #include <linux/perf_event.h>
23 #include <linux/hw_breakpoint.h>
24
25 #include <asm/uaccess.h>
26 #include <asm/pgtable.h>
27 #include <asm/system.h>
28 #include <asm/processor.h>
29 #include <asm/i387.h>
30 #include <asm/fpu-internal.h>
31 #include <asm/debugreg.h>
32 #include <asm/ldt.h>
33 #include <asm/desc.h>
34 #include <asm/prctl.h>
35 #include <asm/proto.h>
36 #include <asm/hw_breakpoint.h>
37
38 #include "tls.h"
39
40 #define CREATE_TRACE_POINTS
41 #include <trace/events/syscalls.h>
42
43 enum x86_regset {
44 REGSET_GENERAL,
45 REGSET_FP,
46 REGSET_XFP,
47 REGSET_IOPERM64 = REGSET_XFP,
48 REGSET_XSTATE,
49 REGSET_TLS,
50 REGSET_IOPERM32,
51 };
52
53 struct pt_regs_offset {
54 const char *name;
55 int offset;
56 };
57
58 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
59 #define REG_OFFSET_END {.name = NULL, .offset = 0}
60
61 static const struct pt_regs_offset regoffset_table[] = {
62 #ifdef CONFIG_X86_64
63 REG_OFFSET_NAME(r15),
64 REG_OFFSET_NAME(r14),
65 REG_OFFSET_NAME(r13),
66 REG_OFFSET_NAME(r12),
67 REG_OFFSET_NAME(r11),
68 REG_OFFSET_NAME(r10),
69 REG_OFFSET_NAME(r9),
70 REG_OFFSET_NAME(r8),
71 #endif
72 REG_OFFSET_NAME(bx),
73 REG_OFFSET_NAME(cx),
74 REG_OFFSET_NAME(dx),
75 REG_OFFSET_NAME(si),
76 REG_OFFSET_NAME(di),
77 REG_OFFSET_NAME(bp),
78 REG_OFFSET_NAME(ax),
79 #ifdef CONFIG_X86_32
80 REG_OFFSET_NAME(ds),
81 REG_OFFSET_NAME(es),
82 REG_OFFSET_NAME(fs),
83 REG_OFFSET_NAME(gs),
84 #endif
85 REG_OFFSET_NAME(orig_ax),
86 REG_OFFSET_NAME(ip),
87 REG_OFFSET_NAME(cs),
88 REG_OFFSET_NAME(flags),
89 REG_OFFSET_NAME(sp),
90 REG_OFFSET_NAME(ss),
91 REG_OFFSET_END,
92 };
93
94 /**
95 * regs_query_register_offset() - query register offset from its name
96 * @name: the name of a register
97 *
98 * regs_query_register_offset() returns the offset of a register in struct
99 * pt_regs from its name. If the name is invalid, this returns -EINVAL;
100 */
101 int regs_query_register_offset(const char *name)
102 {
103 const struct pt_regs_offset *roff;
104 for (roff = regoffset_table; roff->name != NULL; roff++)
105 if (!strcmp(roff->name, name))
106 return roff->offset;
107 return -EINVAL;
108 }
109
110 /**
111 * regs_query_register_name() - query register name from its offset
112 * @offset: the offset of a register in struct pt_regs.
113 *
114 * regs_query_register_name() returns the name of a register from its
115 * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
116 */
117 const char *regs_query_register_name(unsigned int offset)
118 {
119 const struct pt_regs_offset *roff;
120 for (roff = regoffset_table; roff->name != NULL; roff++)
121 if (roff->offset == offset)
122 return roff->name;
123 return NULL;
124 }
125
126 static const int arg_offs_table[] = {
127 #ifdef CONFIG_X86_32
128 [0] = offsetof(struct pt_regs, ax),
129 [1] = offsetof(struct pt_regs, dx),
130 [2] = offsetof(struct pt_regs, cx)
131 #else /* CONFIG_X86_64 */
132 [0] = offsetof(struct pt_regs, di),
133 [1] = offsetof(struct pt_regs, si),
134 [2] = offsetof(struct pt_regs, dx),
135 [3] = offsetof(struct pt_regs, cx),
136 [4] = offsetof(struct pt_regs, r8),
137 [5] = offsetof(struct pt_regs, r9)
138 #endif
139 };
140
141 /*
142 * does not yet catch signals sent when the child dies.
143 * in exit.c or in signal.c.
144 */
145
146 /*
147 * Determines which flags the user has access to [1 = access, 0 = no access].
148 */
149 #define FLAG_MASK_32 ((unsigned long) \
150 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
151 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
152 X86_EFLAGS_SF | X86_EFLAGS_TF | \
153 X86_EFLAGS_DF | X86_EFLAGS_OF | \
154 X86_EFLAGS_RF | X86_EFLAGS_AC))
155
156 /*
157 * Determines whether a value may be installed in a segment register.
158 */
159 static inline bool invalid_selector(u16 value)
160 {
161 return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
162 }
163
164 #ifdef CONFIG_X86_32
165
166 #define FLAG_MASK FLAG_MASK_32
167
168 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
169 {
170 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
171 return &regs->bx + (regno >> 2);
172 }
173
174 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
175 {
176 /*
177 * Returning the value truncates it to 16 bits.
178 */
179 unsigned int retval;
180 if (offset != offsetof(struct user_regs_struct, gs))
181 retval = *pt_regs_access(task_pt_regs(task), offset);
182 else {
183 if (task == current)
184 retval = get_user_gs(task_pt_regs(task));
185 else
186 retval = task_user_gs(task);
187 }
188 return retval;
189 }
190
191 static int set_segment_reg(struct task_struct *task,
192 unsigned long offset, u16 value)
193 {
194 /*
195 * The value argument was already truncated to 16 bits.
196 */
197 if (invalid_selector(value))
198 return -EIO;
199
200 /*
201 * For %cs and %ss we cannot permit a null selector.
202 * We can permit a bogus selector as long as it has USER_RPL.
203 * Null selectors are fine for other segment registers, but
204 * we will never get back to user mode with invalid %cs or %ss
205 * and will take the trap in iret instead. Much code relies
206 * on user_mode() to distinguish a user trap frame (which can
207 * safely use invalid selectors) from a kernel trap frame.
208 */
209 switch (offset) {
210 case offsetof(struct user_regs_struct, cs):
211 case offsetof(struct user_regs_struct, ss):
212 if (unlikely(value == 0))
213 return -EIO;
214
215 default:
216 *pt_regs_access(task_pt_regs(task), offset) = value;
217 break;
218
219 case offsetof(struct user_regs_struct, gs):
220 if (task == current)
221 set_user_gs(task_pt_regs(task), value);
222 else
223 task_user_gs(task) = value;
224 }
225
226 return 0;
227 }
228
229 #else /* CONFIG_X86_64 */
230
231 #define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
232
233 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
234 {
235 BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
236 return &regs->r15 + (offset / sizeof(regs->r15));
237 }
238
239 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
240 {
241 /*
242 * Returning the value truncates it to 16 bits.
243 */
244 unsigned int seg;
245
246 switch (offset) {
247 case offsetof(struct user_regs_struct, fs):
248 if (task == current) {
249 /* Older gas can't assemble movq %?s,%r?? */
250 asm("movl %%fs,%0" : "=r" (seg));
251 return seg;
252 }
253 return task->thread.fsindex;
254 case offsetof(struct user_regs_struct, gs):
255 if (task == current) {
256 asm("movl %%gs,%0" : "=r" (seg));
257 return seg;
258 }
259 return task->thread.gsindex;
260 case offsetof(struct user_regs_struct, ds):
261 if (task == current) {
262 asm("movl %%ds,%0" : "=r" (seg));
263 return seg;
264 }
265 return task->thread.ds;
266 case offsetof(struct user_regs_struct, es):
267 if (task == current) {
268 asm("movl %%es,%0" : "=r" (seg));
269 return seg;
270 }
271 return task->thread.es;
272
273 case offsetof(struct user_regs_struct, cs):
274 case offsetof(struct user_regs_struct, ss):
275 break;
276 }
277 return *pt_regs_access(task_pt_regs(task), offset);
278 }
279
280 static int set_segment_reg(struct task_struct *task,
281 unsigned long offset, u16 value)
282 {
283 /*
284 * The value argument was already truncated to 16 bits.
285 */
286 if (invalid_selector(value))
287 return -EIO;
288
289 switch (offset) {
290 case offsetof(struct user_regs_struct,fs):
291 /*
292 * If this is setting fs as for normal 64-bit use but
293 * setting fs_base has implicitly changed it, leave it.
294 */
295 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
296 task->thread.fs != 0) ||
297 (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
298 task->thread.fs == 0))
299 break;
300 task->thread.fsindex = value;
301 if (task == current)
302 loadsegment(fs, task->thread.fsindex);
303 break;
304 case offsetof(struct user_regs_struct,gs):
305 /*
306 * If this is setting gs as for normal 64-bit use but
307 * setting gs_base has implicitly changed it, leave it.
308 */
309 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
310 task->thread.gs != 0) ||
311 (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
312 task->thread.gs == 0))
313 break;
314 task->thread.gsindex = value;
315 if (task == current)
316 load_gs_index(task->thread.gsindex);
317 break;
318 case offsetof(struct user_regs_struct,ds):
319 task->thread.ds = value;
320 if (task == current)
321 loadsegment(ds, task->thread.ds);
322 break;
323 case offsetof(struct user_regs_struct,es):
324 task->thread.es = value;
325 if (task == current)
326 loadsegment(es, task->thread.es);
327 break;
328
329 /*
330 * Can't actually change these in 64-bit mode.
331 */
332 case offsetof(struct user_regs_struct,cs):
333 if (unlikely(value == 0))
334 return -EIO;
335 #ifdef CONFIG_IA32_EMULATION
336 if (test_tsk_thread_flag(task, TIF_IA32))
337 task_pt_regs(task)->cs = value;
338 #endif
339 break;
340 case offsetof(struct user_regs_struct,ss):
341 if (unlikely(value == 0))
342 return -EIO;
343 #ifdef CONFIG_IA32_EMULATION
344 if (test_tsk_thread_flag(task, TIF_IA32))
345 task_pt_regs(task)->ss = value;
346 #endif
347 break;
348 }
349
350 return 0;
351 }
352
353 #endif /* CONFIG_X86_32 */
354
355 static unsigned long get_flags(struct task_struct *task)
356 {
357 unsigned long retval = task_pt_regs(task)->flags;
358
359 /*
360 * If the debugger set TF, hide it from the readout.
361 */
362 if (test_tsk_thread_flag(task, TIF_FORCED_TF))
363 retval &= ~X86_EFLAGS_TF;
364
365 return retval;
366 }
367
368 static int set_flags(struct task_struct *task, unsigned long value)
369 {
370 struct pt_regs *regs = task_pt_regs(task);
371
372 /*
373 * If the user value contains TF, mark that
374 * it was not "us" (the debugger) that set it.
375 * If not, make sure it stays set if we had.
376 */
377 if (value & X86_EFLAGS_TF)
378 clear_tsk_thread_flag(task, TIF_FORCED_TF);
379 else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
380 value |= X86_EFLAGS_TF;
381
382 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
383
384 return 0;
385 }
386
387 static int putreg(struct task_struct *child,
388 unsigned long offset, unsigned long value)
389 {
390 switch (offset) {
391 case offsetof(struct user_regs_struct, cs):
392 case offsetof(struct user_regs_struct, ds):
393 case offsetof(struct user_regs_struct, es):
394 case offsetof(struct user_regs_struct, fs):
395 case offsetof(struct user_regs_struct, gs):
396 case offsetof(struct user_regs_struct, ss):
397 return set_segment_reg(child, offset, value);
398
399 case offsetof(struct user_regs_struct, flags):
400 return set_flags(child, value);
401
402 #ifdef CONFIG_X86_64
403 case offsetof(struct user_regs_struct,fs_base):
404 if (value >= TASK_SIZE_OF(child))
405 return -EIO;
406 /*
407 * When changing the segment base, use do_arch_prctl
408 * to set either thread.fs or thread.fsindex and the
409 * corresponding GDT slot.
410 */
411 if (child->thread.fs != value)
412 return do_arch_prctl(child, ARCH_SET_FS, value);
413 return 0;
414 case offsetof(struct user_regs_struct,gs_base):
415 /*
416 * Exactly the same here as the %fs handling above.
417 */
418 if (value >= TASK_SIZE_OF(child))
419 return -EIO;
420 if (child->thread.gs != value)
421 return do_arch_prctl(child, ARCH_SET_GS, value);
422 return 0;
423 #endif
424 }
425
426 *pt_regs_access(task_pt_regs(child), offset) = value;
427 return 0;
428 }
429
430 static unsigned long getreg(struct task_struct *task, unsigned long offset)
431 {
432 switch (offset) {
433 case offsetof(struct user_regs_struct, cs):
434 case offsetof(struct user_regs_struct, ds):
435 case offsetof(struct user_regs_struct, es):
436 case offsetof(struct user_regs_struct, fs):
437 case offsetof(struct user_regs_struct, gs):
438 case offsetof(struct user_regs_struct, ss):
439 return get_segment_reg(task, offset);
440
441 case offsetof(struct user_regs_struct, flags):
442 return get_flags(task);
443
444 #ifdef CONFIG_X86_64
445 case offsetof(struct user_regs_struct, fs_base): {
446 /*
447 * do_arch_prctl may have used a GDT slot instead of
448 * the MSR. To userland, it appears the same either
449 * way, except the %fs segment selector might not be 0.
450 */
451 unsigned int seg = task->thread.fsindex;
452 if (task->thread.fs != 0)
453 return task->thread.fs;
454 if (task == current)
455 asm("movl %%fs,%0" : "=r" (seg));
456 if (seg != FS_TLS_SEL)
457 return 0;
458 return get_desc_base(&task->thread.tls_array[FS_TLS]);
459 }
460 case offsetof(struct user_regs_struct, gs_base): {
461 /*
462 * Exactly the same here as the %fs handling above.
463 */
464 unsigned int seg = task->thread.gsindex;
465 if (task->thread.gs != 0)
466 return task->thread.gs;
467 if (task == current)
468 asm("movl %%gs,%0" : "=r" (seg));
469 if (seg != GS_TLS_SEL)
470 return 0;
471 return get_desc_base(&task->thread.tls_array[GS_TLS]);
472 }
473 #endif
474 }
475
476 return *pt_regs_access(task_pt_regs(task), offset);
477 }
478
479 static int genregs_get(struct task_struct *target,
480 const struct user_regset *regset,
481 unsigned int pos, unsigned int count,
482 void *kbuf, void __user *ubuf)
483 {
484 if (kbuf) {
485 unsigned long *k = kbuf;
486 while (count >= sizeof(*k)) {
487 *k++ = getreg(target, pos);
488 count -= sizeof(*k);
489 pos += sizeof(*k);
490 }
491 } else {
492 unsigned long __user *u = ubuf;
493 while (count >= sizeof(*u)) {
494 if (__put_user(getreg(target, pos), u++))
495 return -EFAULT;
496 count -= sizeof(*u);
497 pos += sizeof(*u);
498 }
499 }
500
501 return 0;
502 }
503
504 static int genregs_set(struct task_struct *target,
505 const struct user_regset *regset,
506 unsigned int pos, unsigned int count,
507 const void *kbuf, const void __user *ubuf)
508 {
509 int ret = 0;
510 if (kbuf) {
511 const unsigned long *k = kbuf;
512 while (count >= sizeof(*k) && !ret) {
513 ret = putreg(target, pos, *k++);
514 count -= sizeof(*k);
515 pos += sizeof(*k);
516 }
517 } else {
518 const unsigned long __user *u = ubuf;
519 while (count >= sizeof(*u) && !ret) {
520 unsigned long word;
521 ret = __get_user(word, u++);
522 if (ret)
523 break;
524 ret = putreg(target, pos, word);
525 count -= sizeof(*u);
526 pos += sizeof(*u);
527 }
528 }
529 return ret;
530 }
531
532 static void ptrace_triggered(struct perf_event *bp,
533 struct perf_sample_data *data,
534 struct pt_regs *regs)
535 {
536 int i;
537 struct thread_struct *thread = &(current->thread);
538
539 /*
540 * Store in the virtual DR6 register the fact that the breakpoint
541 * was hit so the thread's debugger will see it.
542 */
543 for (i = 0; i < HBP_NUM; i++) {
544 if (thread->ptrace_bps[i] == bp)
545 break;
546 }
547
548 thread->debugreg6 |= (DR_TRAP0 << i);
549 }
550
551 /*
552 * Walk through every ptrace breakpoints for this thread and
553 * build the dr7 value on top of their attributes.
554 *
555 */
556 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
557 {
558 int i;
559 int dr7 = 0;
560 struct arch_hw_breakpoint *info;
561
562 for (i = 0; i < HBP_NUM; i++) {
563 if (bp[i] && !bp[i]->attr.disabled) {
564 info = counter_arch_bp(bp[i]);
565 dr7 |= encode_dr7(i, info->len, info->type);
566 }
567 }
568
569 return dr7;
570 }
571
572 static int
573 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
574 struct task_struct *tsk, int disabled)
575 {
576 int err;
577 int gen_len, gen_type;
578 struct perf_event_attr attr;
579
580 /*
581 * We should have at least an inactive breakpoint at this
582 * slot. It means the user is writing dr7 without having
583 * written the address register first
584 */
585 if (!bp)
586 return -EINVAL;
587
588 err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
589 if (err)
590 return err;
591
592 attr = bp->attr;
593 attr.bp_len = gen_len;
594 attr.bp_type = gen_type;
595 attr.disabled = disabled;
596
597 return modify_user_hw_breakpoint(bp, &attr);
598 }
599
600 /*
601 * Handle ptrace writes to debug register 7.
602 */
603 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
604 {
605 struct thread_struct *thread = &(tsk->thread);
606 unsigned long old_dr7;
607 int i, orig_ret = 0, rc = 0;
608 int enabled, second_pass = 0;
609 unsigned len, type;
610 struct perf_event *bp;
611
612 if (ptrace_get_breakpoints(tsk) < 0)
613 return -ESRCH;
614
615 data &= ~DR_CONTROL_RESERVED;
616 old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
617 restore:
618 /*
619 * Loop through all the hardware breakpoints, making the
620 * appropriate changes to each.
621 */
622 for (i = 0; i < HBP_NUM; i++) {
623 enabled = decode_dr7(data, i, &len, &type);
624 bp = thread->ptrace_bps[i];
625
626 if (!enabled) {
627 if (bp) {
628 /*
629 * Don't unregister the breakpoints right-away,
630 * unless all register_user_hw_breakpoint()
631 * requests have succeeded. This prevents
632 * any window of opportunity for debug
633 * register grabbing by other users.
634 */
635 if (!second_pass)
636 continue;
637
638 rc = ptrace_modify_breakpoint(bp, len, type,
639 tsk, 1);
640 if (rc)
641 break;
642 }
643 continue;
644 }
645
646 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
647 if (rc)
648 break;
649 }
650 /*
651 * Make a second pass to free the remaining unused breakpoints
652 * or to restore the original breakpoints if an error occurred.
653 */
654 if (!second_pass) {
655 second_pass = 1;
656 if (rc < 0) {
657 orig_ret = rc;
658 data = old_dr7;
659 }
660 goto restore;
661 }
662
663 ptrace_put_breakpoints(tsk);
664
665 return ((orig_ret < 0) ? orig_ret : rc);
666 }
667
668 /*
669 * Handle PTRACE_PEEKUSR calls for the debug register area.
670 */
671 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
672 {
673 struct thread_struct *thread = &(tsk->thread);
674 unsigned long val = 0;
675
676 if (n < HBP_NUM) {
677 struct perf_event *bp;
678
679 if (ptrace_get_breakpoints(tsk) < 0)
680 return -ESRCH;
681
682 bp = thread->ptrace_bps[n];
683 if (!bp)
684 val = 0;
685 else
686 val = bp->hw.info.address;
687
688 ptrace_put_breakpoints(tsk);
689 } else if (n == 6) {
690 val = thread->debugreg6;
691 } else if (n == 7) {
692 val = thread->ptrace_dr7;
693 }
694 return val;
695 }
696
697 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
698 unsigned long addr)
699 {
700 struct perf_event *bp;
701 struct thread_struct *t = &tsk->thread;
702 struct perf_event_attr attr;
703 int err = 0;
704
705 if (ptrace_get_breakpoints(tsk) < 0)
706 return -ESRCH;
707
708 if (!t->ptrace_bps[nr]) {
709 ptrace_breakpoint_init(&attr);
710 /*
711 * Put stub len and type to register (reserve) an inactive but
712 * correct bp
713 */
714 attr.bp_addr = addr;
715 attr.bp_len = HW_BREAKPOINT_LEN_1;
716 attr.bp_type = HW_BREAKPOINT_W;
717 attr.disabled = 1;
718
719 bp = register_user_hw_breakpoint(&attr, ptrace_triggered,
720 NULL, tsk);
721
722 /*
723 * CHECKME: the previous code returned -EIO if the addr wasn't
724 * a valid task virtual addr. The new one will return -EINVAL in
725 * this case.
726 * -EINVAL may be what we want for in-kernel breakpoints users,
727 * but -EIO looks better for ptrace, since we refuse a register
728 * writing for the user. And anyway this is the previous
729 * behaviour.
730 */
731 if (IS_ERR(bp)) {
732 err = PTR_ERR(bp);
733 goto put;
734 }
735
736 t->ptrace_bps[nr] = bp;
737 } else {
738 bp = t->ptrace_bps[nr];
739
740 attr = bp->attr;
741 attr.bp_addr = addr;
742 err = modify_user_hw_breakpoint(bp, &attr);
743 }
744
745 put:
746 ptrace_put_breakpoints(tsk);
747 return err;
748 }
749
750 /*
751 * Handle PTRACE_POKEUSR calls for the debug register area.
752 */
753 static int ptrace_set_debugreg(struct task_struct *tsk, int n,
754 unsigned long val)
755 {
756 struct thread_struct *thread = &(tsk->thread);
757 int rc = 0;
758
759 /* There are no DR4 or DR5 registers */
760 if (n == 4 || n == 5)
761 return -EIO;
762
763 if (n == 6) {
764 thread->debugreg6 = val;
765 goto ret_path;
766 }
767 if (n < HBP_NUM) {
768 rc = ptrace_set_breakpoint_addr(tsk, n, val);
769 if (rc)
770 return rc;
771 }
772 /* All that's left is DR7 */
773 if (n == 7) {
774 rc = ptrace_write_dr7(tsk, val);
775 if (!rc)
776 thread->ptrace_dr7 = val;
777 }
778
779 ret_path:
780 return rc;
781 }
782
783 /*
784 * These access the current or another (stopped) task's io permission
785 * bitmap for debugging or core dump.
786 */
787 static int ioperm_active(struct task_struct *target,
788 const struct user_regset *regset)
789 {
790 return target->thread.io_bitmap_max / regset->size;
791 }
792
793 static int ioperm_get(struct task_struct *target,
794 const struct user_regset *regset,
795 unsigned int pos, unsigned int count,
796 void *kbuf, void __user *ubuf)
797 {
798 if (!target->thread.io_bitmap_ptr)
799 return -ENXIO;
800
801 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
802 target->thread.io_bitmap_ptr,
803 0, IO_BITMAP_BYTES);
804 }
805
806 /*
807 * Called by kernel/ptrace.c when detaching..
808 *
809 * Make sure the single step bit is not set.
810 */
811 void ptrace_disable(struct task_struct *child)
812 {
813 user_disable_single_step(child);
814 #ifdef TIF_SYSCALL_EMU
815 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
816 #endif
817 }
818
819 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
820 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
821 #endif
822
823 long arch_ptrace(struct task_struct *child, long request,
824 unsigned long addr, unsigned long data)
825 {
826 int ret;
827 unsigned long __user *datap = (unsigned long __user *)data;
828
829 switch (request) {
830 /* read the word at location addr in the USER area. */
831 case PTRACE_PEEKUSR: {
832 unsigned long tmp;
833
834 ret = -EIO;
835 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
836 break;
837
838 tmp = 0; /* Default return condition */
839 if (addr < sizeof(struct user_regs_struct))
840 tmp = getreg(child, addr);
841 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
842 addr <= offsetof(struct user, u_debugreg[7])) {
843 addr -= offsetof(struct user, u_debugreg[0]);
844 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
845 }
846 ret = put_user(tmp, datap);
847 break;
848 }
849
850 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
851 ret = -EIO;
852 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
853 break;
854
855 if (addr < sizeof(struct user_regs_struct))
856 ret = putreg(child, addr, data);
857 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
858 addr <= offsetof(struct user, u_debugreg[7])) {
859 addr -= offsetof(struct user, u_debugreg[0]);
860 ret = ptrace_set_debugreg(child,
861 addr / sizeof(data), data);
862 }
863 break;
864
865 case PTRACE_GETREGS: /* Get all gp regs from the child. */
866 return copy_regset_to_user(child,
867 task_user_regset_view(current),
868 REGSET_GENERAL,
869 0, sizeof(struct user_regs_struct),
870 datap);
871
872 case PTRACE_SETREGS: /* Set all gp regs in the child. */
873 return copy_regset_from_user(child,
874 task_user_regset_view(current),
875 REGSET_GENERAL,
876 0, sizeof(struct user_regs_struct),
877 datap);
878
879 case PTRACE_GETFPREGS: /* Get the child FPU state. */
880 return copy_regset_to_user(child,
881 task_user_regset_view(current),
882 REGSET_FP,
883 0, sizeof(struct user_i387_struct),
884 datap);
885
886 case PTRACE_SETFPREGS: /* Set the child FPU state. */
887 return copy_regset_from_user(child,
888 task_user_regset_view(current),
889 REGSET_FP,
890 0, sizeof(struct user_i387_struct),
891 datap);
892
893 #ifdef CONFIG_X86_32
894 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
895 return copy_regset_to_user(child, &user_x86_32_view,
896 REGSET_XFP,
897 0, sizeof(struct user_fxsr_struct),
898 datap) ? -EIO : 0;
899
900 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
901 return copy_regset_from_user(child, &user_x86_32_view,
902 REGSET_XFP,
903 0, sizeof(struct user_fxsr_struct),
904 datap) ? -EIO : 0;
905 #endif
906
907 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
908 case PTRACE_GET_THREAD_AREA:
909 if ((int) addr < 0)
910 return -EIO;
911 ret = do_get_thread_area(child, addr,
912 (struct user_desc __user *)data);
913 break;
914
915 case PTRACE_SET_THREAD_AREA:
916 if ((int) addr < 0)
917 return -EIO;
918 ret = do_set_thread_area(child, addr,
919 (struct user_desc __user *)data, 0);
920 break;
921 #endif
922
923 #ifdef CONFIG_X86_64
924 /* normal 64bit interface to access TLS data.
925 Works just like arch_prctl, except that the arguments
926 are reversed. */
927 case PTRACE_ARCH_PRCTL:
928 ret = do_arch_prctl(child, data, addr);
929 break;
930 #endif
931
932 default:
933 ret = ptrace_request(child, request, addr, data);
934 break;
935 }
936
937 return ret;
938 }
939
940 #ifdef CONFIG_IA32_EMULATION
941
942 #include <linux/compat.h>
943 #include <linux/syscalls.h>
944 #include <asm/ia32.h>
945 #include <asm/user32.h>
946
947 #define R32(l,q) \
948 case offsetof(struct user32, regs.l): \
949 regs->q = value; break
950
951 #define SEG32(rs) \
952 case offsetof(struct user32, regs.rs): \
953 return set_segment_reg(child, \
954 offsetof(struct user_regs_struct, rs), \
955 value); \
956 break
957
958 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
959 {
960 struct pt_regs *regs = task_pt_regs(child);
961
962 switch (regno) {
963
964 SEG32(cs);
965 SEG32(ds);
966 SEG32(es);
967 SEG32(fs);
968 SEG32(gs);
969 SEG32(ss);
970
971 R32(ebx, bx);
972 R32(ecx, cx);
973 R32(edx, dx);
974 R32(edi, di);
975 R32(esi, si);
976 R32(ebp, bp);
977 R32(eax, ax);
978 R32(eip, ip);
979 R32(esp, sp);
980
981 case offsetof(struct user32, regs.orig_eax):
982 /*
983 * A 32-bit debugger setting orig_eax means to restore
984 * the state of the task restarting a 32-bit syscall.
985 * Make sure we interpret the -ERESTART* codes correctly
986 * in case the task is not actually still sitting at the
987 * exit from a 32-bit syscall with TS_COMPAT still set.
988 */
989 regs->orig_ax = value;
990 if (syscall_get_nr(child, regs) >= 0)
991 task_thread_info(child)->status |= TS_COMPAT;
992 break;
993
994 case offsetof(struct user32, regs.eflags):
995 return set_flags(child, value);
996
997 case offsetof(struct user32, u_debugreg[0]) ...
998 offsetof(struct user32, u_debugreg[7]):
999 regno -= offsetof(struct user32, u_debugreg[0]);
1000 return ptrace_set_debugreg(child, regno / 4, value);
1001
1002 default:
1003 if (regno > sizeof(struct user32) || (regno & 3))
1004 return -EIO;
1005
1006 /*
1007 * Other dummy fields in the virtual user structure
1008 * are ignored
1009 */
1010 break;
1011 }
1012 return 0;
1013 }
1014
1015 #undef R32
1016 #undef SEG32
1017
1018 #define R32(l,q) \
1019 case offsetof(struct user32, regs.l): \
1020 *val = regs->q; break
1021
1022 #define SEG32(rs) \
1023 case offsetof(struct user32, regs.rs): \
1024 *val = get_segment_reg(child, \
1025 offsetof(struct user_regs_struct, rs)); \
1026 break
1027
1028 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1029 {
1030 struct pt_regs *regs = task_pt_regs(child);
1031
1032 switch (regno) {
1033
1034 SEG32(ds);
1035 SEG32(es);
1036 SEG32(fs);
1037 SEG32(gs);
1038
1039 R32(cs, cs);
1040 R32(ss, ss);
1041 R32(ebx, bx);
1042 R32(ecx, cx);
1043 R32(edx, dx);
1044 R32(edi, di);
1045 R32(esi, si);
1046 R32(ebp, bp);
1047 R32(eax, ax);
1048 R32(orig_eax, orig_ax);
1049 R32(eip, ip);
1050 R32(esp, sp);
1051
1052 case offsetof(struct user32, regs.eflags):
1053 *val = get_flags(child);
1054 break;
1055
1056 case offsetof(struct user32, u_debugreg[0]) ...
1057 offsetof(struct user32, u_debugreg[7]):
1058 regno -= offsetof(struct user32, u_debugreg[0]);
1059 *val = ptrace_get_debugreg(child, regno / 4);
1060 break;
1061
1062 default:
1063 if (regno > sizeof(struct user32) || (regno & 3))
1064 return -EIO;
1065
1066 /*
1067 * Other dummy fields in the virtual user structure
1068 * are ignored
1069 */
1070 *val = 0;
1071 break;
1072 }
1073 return 0;
1074 }
1075
1076 #undef R32
1077 #undef SEG32
1078
1079 static int genregs32_get(struct task_struct *target,
1080 const struct user_regset *regset,
1081 unsigned int pos, unsigned int count,
1082 void *kbuf, void __user *ubuf)
1083 {
1084 if (kbuf) {
1085 compat_ulong_t *k = kbuf;
1086 while (count >= sizeof(*k)) {
1087 getreg32(target, pos, k++);
1088 count -= sizeof(*k);
1089 pos += sizeof(*k);
1090 }
1091 } else {
1092 compat_ulong_t __user *u = ubuf;
1093 while (count >= sizeof(*u)) {
1094 compat_ulong_t word;
1095 getreg32(target, pos, &word);
1096 if (__put_user(word, u++))
1097 return -EFAULT;
1098 count -= sizeof(*u);
1099 pos += sizeof(*u);
1100 }
1101 }
1102
1103 return 0;
1104 }
1105
1106 static int genregs32_set(struct task_struct *target,
1107 const struct user_regset *regset,
1108 unsigned int pos, unsigned int count,
1109 const void *kbuf, const void __user *ubuf)
1110 {
1111 int ret = 0;
1112 if (kbuf) {
1113 const compat_ulong_t *k = kbuf;
1114 while (count >= sizeof(*k) && !ret) {
1115 ret = putreg32(target, pos, *k++);
1116 count -= sizeof(*k);
1117 pos += sizeof(*k);
1118 }
1119 } else {
1120 const compat_ulong_t __user *u = ubuf;
1121 while (count >= sizeof(*u) && !ret) {
1122 compat_ulong_t word;
1123 ret = __get_user(word, u++);
1124 if (ret)
1125 break;
1126 ret = putreg32(target, pos, word);
1127 count -= sizeof(*u);
1128 pos += sizeof(*u);
1129 }
1130 }
1131 return ret;
1132 }
1133
1134 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1135 compat_ulong_t caddr, compat_ulong_t cdata)
1136 {
1137 unsigned long addr = caddr;
1138 unsigned long data = cdata;
1139 void __user *datap = compat_ptr(data);
1140 int ret;
1141 __u32 val;
1142
1143 switch (request) {
1144 case PTRACE_PEEKUSR:
1145 ret = getreg32(child, addr, &val);
1146 if (ret == 0)
1147 ret = put_user(val, (__u32 __user *)datap);
1148 break;
1149
1150 case PTRACE_POKEUSR:
1151 ret = putreg32(child, addr, data);
1152 break;
1153
1154 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1155 return copy_regset_to_user(child, &user_x86_32_view,
1156 REGSET_GENERAL,
1157 0, sizeof(struct user_regs_struct32),
1158 datap);
1159
1160 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1161 return copy_regset_from_user(child, &user_x86_32_view,
1162 REGSET_GENERAL, 0,
1163 sizeof(struct user_regs_struct32),
1164 datap);
1165
1166 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1167 return copy_regset_to_user(child, &user_x86_32_view,
1168 REGSET_FP, 0,
1169 sizeof(struct user_i387_ia32_struct),
1170 datap);
1171
1172 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1173 return copy_regset_from_user(
1174 child, &user_x86_32_view, REGSET_FP,
1175 0, sizeof(struct user_i387_ia32_struct), datap);
1176
1177 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1178 return copy_regset_to_user(child, &user_x86_32_view,
1179 REGSET_XFP, 0,
1180 sizeof(struct user32_fxsr_struct),
1181 datap);
1182
1183 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1184 return copy_regset_from_user(child, &user_x86_32_view,
1185 REGSET_XFP, 0,
1186 sizeof(struct user32_fxsr_struct),
1187 datap);
1188
1189 case PTRACE_GET_THREAD_AREA:
1190 case PTRACE_SET_THREAD_AREA:
1191 return arch_ptrace(child, request, addr, data);
1192
1193 default:
1194 return compat_ptrace_request(child, request, addr, data);
1195 }
1196
1197 return ret;
1198 }
1199
1200 #endif /* CONFIG_IA32_EMULATION */
1201
1202 #ifdef CONFIG_X86_64
1203
1204 static struct user_regset x86_64_regsets[] __read_mostly = {
1205 [REGSET_GENERAL] = {
1206 .core_note_type = NT_PRSTATUS,
1207 .n = sizeof(struct user_regs_struct) / sizeof(long),
1208 .size = sizeof(long), .align = sizeof(long),
1209 .get = genregs_get, .set = genregs_set
1210 },
1211 [REGSET_FP] = {
1212 .core_note_type = NT_PRFPREG,
1213 .n = sizeof(struct user_i387_struct) / sizeof(long),
1214 .size = sizeof(long), .align = sizeof(long),
1215 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1216 },
1217 [REGSET_XSTATE] = {
1218 .core_note_type = NT_X86_XSTATE,
1219 .size = sizeof(u64), .align = sizeof(u64),
1220 .active = xstateregs_active, .get = xstateregs_get,
1221 .set = xstateregs_set
1222 },
1223 [REGSET_IOPERM64] = {
1224 .core_note_type = NT_386_IOPERM,
1225 .n = IO_BITMAP_LONGS,
1226 .size = sizeof(long), .align = sizeof(long),
1227 .active = ioperm_active, .get = ioperm_get
1228 },
1229 };
1230
1231 static const struct user_regset_view user_x86_64_view = {
1232 .name = "x86_64", .e_machine = EM_X86_64,
1233 .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1234 };
1235
1236 #else /* CONFIG_X86_32 */
1237
1238 #define user_regs_struct32 user_regs_struct
1239 #define genregs32_get genregs_get
1240 #define genregs32_set genregs_set
1241
1242 #define user_i387_ia32_struct user_i387_struct
1243 #define user32_fxsr_struct user_fxsr_struct
1244
1245 #endif /* CONFIG_X86_64 */
1246
1247 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1248 static struct user_regset x86_32_regsets[] __read_mostly = {
1249 [REGSET_GENERAL] = {
1250 .core_note_type = NT_PRSTATUS,
1251 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1252 .size = sizeof(u32), .align = sizeof(u32),
1253 .get = genregs32_get, .set = genregs32_set
1254 },
1255 [REGSET_FP] = {
1256 .core_note_type = NT_PRFPREG,
1257 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1258 .size = sizeof(u32), .align = sizeof(u32),
1259 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1260 },
1261 [REGSET_XFP] = {
1262 .core_note_type = NT_PRXFPREG,
1263 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1264 .size = sizeof(u32), .align = sizeof(u32),
1265 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1266 },
1267 [REGSET_XSTATE] = {
1268 .core_note_type = NT_X86_XSTATE,
1269 .size = sizeof(u64), .align = sizeof(u64),
1270 .active = xstateregs_active, .get = xstateregs_get,
1271 .set = xstateregs_set
1272 },
1273 [REGSET_TLS] = {
1274 .core_note_type = NT_386_TLS,
1275 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1276 .size = sizeof(struct user_desc),
1277 .align = sizeof(struct user_desc),
1278 .active = regset_tls_active,
1279 .get = regset_tls_get, .set = regset_tls_set
1280 },
1281 [REGSET_IOPERM32] = {
1282 .core_note_type = NT_386_IOPERM,
1283 .n = IO_BITMAP_BYTES / sizeof(u32),
1284 .size = sizeof(u32), .align = sizeof(u32),
1285 .active = ioperm_active, .get = ioperm_get
1286 },
1287 };
1288
1289 static const struct user_regset_view user_x86_32_view = {
1290 .name = "i386", .e_machine = EM_386,
1291 .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1292 };
1293 #endif
1294
1295 /*
1296 * This represents bytes 464..511 in the memory layout exported through
1297 * the REGSET_XSTATE interface.
1298 */
1299 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1300
1301 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1302 {
1303 #ifdef CONFIG_X86_64
1304 x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1305 #endif
1306 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1307 x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1308 #endif
1309 xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1310 }
1311
1312 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1313 {
1314 #ifdef CONFIG_IA32_EMULATION
1315 if (test_tsk_thread_flag(task, TIF_IA32))
1316 #endif
1317 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1318 return &user_x86_32_view;
1319 #endif
1320 #ifdef CONFIG_X86_64
1321 return &user_x86_64_view;
1322 #endif
1323 }
1324
1325 static void fill_sigtrap_info(struct task_struct *tsk,
1326 struct pt_regs *regs,
1327 int error_code, int si_code,
1328 struct siginfo *info)
1329 {
1330 tsk->thread.trap_no = 1;
1331 tsk->thread.error_code = error_code;
1332
1333 memset(info, 0, sizeof(*info));
1334 info->si_signo = SIGTRAP;
1335 info->si_code = si_code;
1336 info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1337 }
1338
1339 void user_single_step_siginfo(struct task_struct *tsk,
1340 struct pt_regs *regs,
1341 struct siginfo *info)
1342 {
1343 fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1344 }
1345
1346 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1347 int error_code, int si_code)
1348 {
1349 struct siginfo info;
1350
1351 fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1352 /* Send us the fake SIGTRAP */
1353 force_sig_info(SIGTRAP, &info, tsk);
1354 }
1355
1356
1357 #ifdef CONFIG_X86_32
1358 # define IS_IA32 1
1359 #elif defined CONFIG_IA32_EMULATION
1360 # define IS_IA32 is_compat_task()
1361 #else
1362 # define IS_IA32 0
1363 #endif
1364
1365 /*
1366 * We must return the syscall number to actually look up in the table.
1367 * This can be -1L to skip running any syscall at all.
1368 */
1369 long syscall_trace_enter(struct pt_regs *regs)
1370 {
1371 long ret = 0;
1372
1373 /*
1374 * If we stepped into a sysenter/syscall insn, it trapped in
1375 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1376 * If user-mode had set TF itself, then it's still clear from
1377 * do_debug() and we need to set it again to restore the user
1378 * state. If we entered on the slow path, TF was already set.
1379 */
1380 if (test_thread_flag(TIF_SINGLESTEP))
1381 regs->flags |= X86_EFLAGS_TF;
1382
1383 /* do the secure computing check first */
1384 secure_computing(regs->orig_ax);
1385
1386 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1387 ret = -1L;
1388
1389 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1390 tracehook_report_syscall_entry(regs))
1391 ret = -1L;
1392
1393 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1394 trace_sys_enter(regs, regs->orig_ax);
1395
1396 if (IS_IA32)
1397 audit_syscall_entry(AUDIT_ARCH_I386,
1398 regs->orig_ax,
1399 regs->bx, regs->cx,
1400 regs->dx, regs->si);
1401 #ifdef CONFIG_X86_64
1402 else
1403 audit_syscall_entry(AUDIT_ARCH_X86_64,
1404 regs->orig_ax,
1405 regs->di, regs->si,
1406 regs->dx, regs->r10);
1407 #endif
1408
1409 return ret ?: regs->orig_ax;
1410 }
1411
1412 void syscall_trace_leave(struct pt_regs *regs)
1413 {
1414 bool step;
1415
1416 audit_syscall_exit(regs);
1417
1418 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1419 trace_sys_exit(regs, regs->ax);
1420
1421 /*
1422 * If TIF_SYSCALL_EMU is set, we only get here because of
1423 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1424 * We already reported this syscall instruction in
1425 * syscall_trace_enter().
1426 */
1427 step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1428 !test_thread_flag(TIF_SYSCALL_EMU);
1429 if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1430 tracehook_report_syscall_exit(regs, step);
1431 }