Loading...
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/debugreg.h>
31#include <asm/ldt.h>
32#include <asm/desc.h>
33#include <asm/prctl.h>
34#include <asm/proto.h>
35#include <asm/hw_breakpoint.h>
36
37#include "tls.h"
38
39#define CREATE_TRACE_POINTS
40#include <trace/events/syscalls.h>
41
42enum x86_regset {
43 REGSET_GENERAL,
44 REGSET_FP,
45 REGSET_XFP,
46 REGSET_IOPERM64 = REGSET_XFP,
47 REGSET_XSTATE,
48 REGSET_TLS,
49 REGSET_IOPERM32,
50};
51
52struct pt_regs_offset {
53 const char *name;
54 int offset;
55};
56
57#define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
58#define REG_OFFSET_END {.name = NULL, .offset = 0}
59
60static const struct pt_regs_offset regoffset_table[] = {
61#ifdef CONFIG_X86_64
62 REG_OFFSET_NAME(r15),
63 REG_OFFSET_NAME(r14),
64 REG_OFFSET_NAME(r13),
65 REG_OFFSET_NAME(r12),
66 REG_OFFSET_NAME(r11),
67 REG_OFFSET_NAME(r10),
68 REG_OFFSET_NAME(r9),
69 REG_OFFSET_NAME(r8),
70#endif
71 REG_OFFSET_NAME(bx),
72 REG_OFFSET_NAME(cx),
73 REG_OFFSET_NAME(dx),
74 REG_OFFSET_NAME(si),
75 REG_OFFSET_NAME(di),
76 REG_OFFSET_NAME(bp),
77 REG_OFFSET_NAME(ax),
78#ifdef CONFIG_X86_32
79 REG_OFFSET_NAME(ds),
80 REG_OFFSET_NAME(es),
81 REG_OFFSET_NAME(fs),
82 REG_OFFSET_NAME(gs),
83#endif
84 REG_OFFSET_NAME(orig_ax),
85 REG_OFFSET_NAME(ip),
86 REG_OFFSET_NAME(cs),
87 REG_OFFSET_NAME(flags),
88 REG_OFFSET_NAME(sp),
89 REG_OFFSET_NAME(ss),
90 REG_OFFSET_END,
91};
92
93/**
94 * regs_query_register_offset() - query register offset from its name
95 * @name: the name of a register
96 *
97 * regs_query_register_offset() returns the offset of a register in struct
98 * pt_regs from its name. If the name is invalid, this returns -EINVAL;
99 */
100int regs_query_register_offset(const char *name)
101{
102 const struct pt_regs_offset *roff;
103 for (roff = regoffset_table; roff->name != NULL; roff++)
104 if (!strcmp(roff->name, name))
105 return roff->offset;
106 return -EINVAL;
107}
108
109/**
110 * regs_query_register_name() - query register name from its offset
111 * @offset: the offset of a register in struct pt_regs.
112 *
113 * regs_query_register_name() returns the name of a register from its
114 * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
115 */
116const char *regs_query_register_name(unsigned int offset)
117{
118 const struct pt_regs_offset *roff;
119 for (roff = regoffset_table; roff->name != NULL; roff++)
120 if (roff->offset == offset)
121 return roff->name;
122 return NULL;
123}
124
125static const int arg_offs_table[] = {
126#ifdef CONFIG_X86_32
127 [0] = offsetof(struct pt_regs, ax),
128 [1] = offsetof(struct pt_regs, dx),
129 [2] = offsetof(struct pt_regs, cx)
130#else /* CONFIG_X86_64 */
131 [0] = offsetof(struct pt_regs, di),
132 [1] = offsetof(struct pt_regs, si),
133 [2] = offsetof(struct pt_regs, dx),
134 [3] = offsetof(struct pt_regs, cx),
135 [4] = offsetof(struct pt_regs, r8),
136 [5] = offsetof(struct pt_regs, r9)
137#endif
138};
139
140/*
141 * does not yet catch signals sent when the child dies.
142 * in exit.c or in signal.c.
143 */
144
145/*
146 * Determines which flags the user has access to [1 = access, 0 = no access].
147 */
148#define FLAG_MASK_32 ((unsigned long) \
149 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
150 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
151 X86_EFLAGS_SF | X86_EFLAGS_TF | \
152 X86_EFLAGS_DF | X86_EFLAGS_OF | \
153 X86_EFLAGS_RF | X86_EFLAGS_AC))
154
155/*
156 * Determines whether a value may be installed in a segment register.
157 */
158static inline bool invalid_selector(u16 value)
159{
160 return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
161}
162
163#ifdef CONFIG_X86_32
164
165#define FLAG_MASK FLAG_MASK_32
166
167static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
168{
169 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
170 return ®s->bx + (regno >> 2);
171}
172
173static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
174{
175 /*
176 * Returning the value truncates it to 16 bits.
177 */
178 unsigned int retval;
179 if (offset != offsetof(struct user_regs_struct, gs))
180 retval = *pt_regs_access(task_pt_regs(task), offset);
181 else {
182 if (task == current)
183 retval = get_user_gs(task_pt_regs(task));
184 else
185 retval = task_user_gs(task);
186 }
187 return retval;
188}
189
190static int set_segment_reg(struct task_struct *task,
191 unsigned long offset, u16 value)
192{
193 /*
194 * The value argument was already truncated to 16 bits.
195 */
196 if (invalid_selector(value))
197 return -EIO;
198
199 /*
200 * For %cs and %ss we cannot permit a null selector.
201 * We can permit a bogus selector as long as it has USER_RPL.
202 * Null selectors are fine for other segment registers, but
203 * we will never get back to user mode with invalid %cs or %ss
204 * and will take the trap in iret instead. Much code relies
205 * on user_mode() to distinguish a user trap frame (which can
206 * safely use invalid selectors) from a kernel trap frame.
207 */
208 switch (offset) {
209 case offsetof(struct user_regs_struct, cs):
210 case offsetof(struct user_regs_struct, ss):
211 if (unlikely(value == 0))
212 return -EIO;
213
214 default:
215 *pt_regs_access(task_pt_regs(task), offset) = value;
216 break;
217
218 case offsetof(struct user_regs_struct, gs):
219 if (task == current)
220 set_user_gs(task_pt_regs(task), value);
221 else
222 task_user_gs(task) = value;
223 }
224
225 return 0;
226}
227
228#else /* CONFIG_X86_64 */
229
230#define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
231
232static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
233{
234 BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
235 return ®s->r15 + (offset / sizeof(regs->r15));
236}
237
238static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
239{
240 /*
241 * Returning the value truncates it to 16 bits.
242 */
243 unsigned int seg;
244
245 switch (offset) {
246 case offsetof(struct user_regs_struct, fs):
247 if (task == current) {
248 /* Older gas can't assemble movq %?s,%r?? */
249 asm("movl %%fs,%0" : "=r" (seg));
250 return seg;
251 }
252 return task->thread.fsindex;
253 case offsetof(struct user_regs_struct, gs):
254 if (task == current) {
255 asm("movl %%gs,%0" : "=r" (seg));
256 return seg;
257 }
258 return task->thread.gsindex;
259 case offsetof(struct user_regs_struct, ds):
260 if (task == current) {
261 asm("movl %%ds,%0" : "=r" (seg));
262 return seg;
263 }
264 return task->thread.ds;
265 case offsetof(struct user_regs_struct, es):
266 if (task == current) {
267 asm("movl %%es,%0" : "=r" (seg));
268 return seg;
269 }
270 return task->thread.es;
271
272 case offsetof(struct user_regs_struct, cs):
273 case offsetof(struct user_regs_struct, ss):
274 break;
275 }
276 return *pt_regs_access(task_pt_regs(task), offset);
277}
278
279static int set_segment_reg(struct task_struct *task,
280 unsigned long offset, u16 value)
281{
282 /*
283 * The value argument was already truncated to 16 bits.
284 */
285 if (invalid_selector(value))
286 return -EIO;
287
288 switch (offset) {
289 case offsetof(struct user_regs_struct,fs):
290 /*
291 * If this is setting fs as for normal 64-bit use but
292 * setting fs_base has implicitly changed it, leave it.
293 */
294 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
295 task->thread.fs != 0) ||
296 (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
297 task->thread.fs == 0))
298 break;
299 task->thread.fsindex = value;
300 if (task == current)
301 loadsegment(fs, task->thread.fsindex);
302 break;
303 case offsetof(struct user_regs_struct,gs):
304 /*
305 * If this is setting gs as for normal 64-bit use but
306 * setting gs_base has implicitly changed it, leave it.
307 */
308 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
309 task->thread.gs != 0) ||
310 (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
311 task->thread.gs == 0))
312 break;
313 task->thread.gsindex = value;
314 if (task == current)
315 load_gs_index(task->thread.gsindex);
316 break;
317 case offsetof(struct user_regs_struct,ds):
318 task->thread.ds = value;
319 if (task == current)
320 loadsegment(ds, task->thread.ds);
321 break;
322 case offsetof(struct user_regs_struct,es):
323 task->thread.es = value;
324 if (task == current)
325 loadsegment(es, task->thread.es);
326 break;
327
328 /*
329 * Can't actually change these in 64-bit mode.
330 */
331 case offsetof(struct user_regs_struct,cs):
332 if (unlikely(value == 0))
333 return -EIO;
334#ifdef CONFIG_IA32_EMULATION
335 if (test_tsk_thread_flag(task, TIF_IA32))
336 task_pt_regs(task)->cs = value;
337#endif
338 break;
339 case offsetof(struct user_regs_struct,ss):
340 if (unlikely(value == 0))
341 return -EIO;
342#ifdef CONFIG_IA32_EMULATION
343 if (test_tsk_thread_flag(task, TIF_IA32))
344 task_pt_regs(task)->ss = value;
345#endif
346 break;
347 }
348
349 return 0;
350}
351
352#endif /* CONFIG_X86_32 */
353
354static unsigned long get_flags(struct task_struct *task)
355{
356 unsigned long retval = task_pt_regs(task)->flags;
357
358 /*
359 * If the debugger set TF, hide it from the readout.
360 */
361 if (test_tsk_thread_flag(task, TIF_FORCED_TF))
362 retval &= ~X86_EFLAGS_TF;
363
364 return retval;
365}
366
367static int set_flags(struct task_struct *task, unsigned long value)
368{
369 struct pt_regs *regs = task_pt_regs(task);
370
371 /*
372 * If the user value contains TF, mark that
373 * it was not "us" (the debugger) that set it.
374 * If not, make sure it stays set if we had.
375 */
376 if (value & X86_EFLAGS_TF)
377 clear_tsk_thread_flag(task, TIF_FORCED_TF);
378 else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
379 value |= X86_EFLAGS_TF;
380
381 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
382
383 return 0;
384}
385
386static int putreg(struct task_struct *child,
387 unsigned long offset, unsigned long value)
388{
389 switch (offset) {
390 case offsetof(struct user_regs_struct, cs):
391 case offsetof(struct user_regs_struct, ds):
392 case offsetof(struct user_regs_struct, es):
393 case offsetof(struct user_regs_struct, fs):
394 case offsetof(struct user_regs_struct, gs):
395 case offsetof(struct user_regs_struct, ss):
396 return set_segment_reg(child, offset, value);
397
398 case offsetof(struct user_regs_struct, flags):
399 return set_flags(child, value);
400
401#ifdef CONFIG_X86_64
402 case offsetof(struct user_regs_struct,fs_base):
403 if (value >= TASK_SIZE_OF(child))
404 return -EIO;
405 /*
406 * When changing the segment base, use do_arch_prctl
407 * to set either thread.fs or thread.fsindex and the
408 * corresponding GDT slot.
409 */
410 if (child->thread.fs != value)
411 return do_arch_prctl(child, ARCH_SET_FS, value);
412 return 0;
413 case offsetof(struct user_regs_struct,gs_base):
414 /*
415 * Exactly the same here as the %fs handling above.
416 */
417 if (value >= TASK_SIZE_OF(child))
418 return -EIO;
419 if (child->thread.gs != value)
420 return do_arch_prctl(child, ARCH_SET_GS, value);
421 return 0;
422#endif
423 }
424
425 *pt_regs_access(task_pt_regs(child), offset) = value;
426 return 0;
427}
428
429static unsigned long getreg(struct task_struct *task, unsigned long offset)
430{
431 switch (offset) {
432 case offsetof(struct user_regs_struct, cs):
433 case offsetof(struct user_regs_struct, ds):
434 case offsetof(struct user_regs_struct, es):
435 case offsetof(struct user_regs_struct, fs):
436 case offsetof(struct user_regs_struct, gs):
437 case offsetof(struct user_regs_struct, ss):
438 return get_segment_reg(task, offset);
439
440 case offsetof(struct user_regs_struct, flags):
441 return get_flags(task);
442
443#ifdef CONFIG_X86_64
444 case offsetof(struct user_regs_struct, fs_base): {
445 /*
446 * do_arch_prctl may have used a GDT slot instead of
447 * the MSR. To userland, it appears the same either
448 * way, except the %fs segment selector might not be 0.
449 */
450 unsigned int seg = task->thread.fsindex;
451 if (task->thread.fs != 0)
452 return task->thread.fs;
453 if (task == current)
454 asm("movl %%fs,%0" : "=r" (seg));
455 if (seg != FS_TLS_SEL)
456 return 0;
457 return get_desc_base(&task->thread.tls_array[FS_TLS]);
458 }
459 case offsetof(struct user_regs_struct, gs_base): {
460 /*
461 * Exactly the same here as the %fs handling above.
462 */
463 unsigned int seg = task->thread.gsindex;
464 if (task->thread.gs != 0)
465 return task->thread.gs;
466 if (task == current)
467 asm("movl %%gs,%0" : "=r" (seg));
468 if (seg != GS_TLS_SEL)
469 return 0;
470 return get_desc_base(&task->thread.tls_array[GS_TLS]);
471 }
472#endif
473 }
474
475 return *pt_regs_access(task_pt_regs(task), offset);
476}
477
478static int genregs_get(struct task_struct *target,
479 const struct user_regset *regset,
480 unsigned int pos, unsigned int count,
481 void *kbuf, void __user *ubuf)
482{
483 if (kbuf) {
484 unsigned long *k = kbuf;
485 while (count >= sizeof(*k)) {
486 *k++ = getreg(target, pos);
487 count -= sizeof(*k);
488 pos += sizeof(*k);
489 }
490 } else {
491 unsigned long __user *u = ubuf;
492 while (count >= sizeof(*u)) {
493 if (__put_user(getreg(target, pos), u++))
494 return -EFAULT;
495 count -= sizeof(*u);
496 pos += sizeof(*u);
497 }
498 }
499
500 return 0;
501}
502
503static int genregs_set(struct task_struct *target,
504 const struct user_regset *regset,
505 unsigned int pos, unsigned int count,
506 const void *kbuf, const void __user *ubuf)
507{
508 int ret = 0;
509 if (kbuf) {
510 const unsigned long *k = kbuf;
511 while (count >= sizeof(*k) && !ret) {
512 ret = putreg(target, pos, *k++);
513 count -= sizeof(*k);
514 pos += sizeof(*k);
515 }
516 } else {
517 const unsigned long __user *u = ubuf;
518 while (count >= sizeof(*u) && !ret) {
519 unsigned long word;
520 ret = __get_user(word, u++);
521 if (ret)
522 break;
523 ret = putreg(target, pos, word);
524 count -= sizeof(*u);
525 pos += sizeof(*u);
526 }
527 }
528 return ret;
529}
530
531static void ptrace_triggered(struct perf_event *bp,
532 struct perf_sample_data *data,
533 struct pt_regs *regs)
534{
535 int i;
536 struct thread_struct *thread = &(current->thread);
537
538 /*
539 * Store in the virtual DR6 register the fact that the breakpoint
540 * was hit so the thread's debugger will see it.
541 */
542 for (i = 0; i < HBP_NUM; i++) {
543 if (thread->ptrace_bps[i] == bp)
544 break;
545 }
546
547 thread->debugreg6 |= (DR_TRAP0 << i);
548}
549
550/*
551 * Walk through every ptrace breakpoints for this thread and
552 * build the dr7 value on top of their attributes.
553 *
554 */
555static unsigned long ptrace_get_dr7(struct perf_event *bp[])
556{
557 int i;
558 int dr7 = 0;
559 struct arch_hw_breakpoint *info;
560
561 for (i = 0; i < HBP_NUM; i++) {
562 if (bp[i] && !bp[i]->attr.disabled) {
563 info = counter_arch_bp(bp[i]);
564 dr7 |= encode_dr7(i, info->len, info->type);
565 }
566 }
567
568 return dr7;
569}
570
571static int
572ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
573 struct task_struct *tsk, int disabled)
574{
575 int err;
576 int gen_len, gen_type;
577 struct perf_event_attr attr;
578
579 /*
580 * We should have at least an inactive breakpoint at this
581 * slot. It means the user is writing dr7 without having
582 * written the address register first
583 */
584 if (!bp)
585 return -EINVAL;
586
587 err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
588 if (err)
589 return err;
590
591 attr = bp->attr;
592 attr.bp_len = gen_len;
593 attr.bp_type = gen_type;
594 attr.disabled = disabled;
595
596 return modify_user_hw_breakpoint(bp, &attr);
597}
598
599/*
600 * Handle ptrace writes to debug register 7.
601 */
602static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
603{
604 struct thread_struct *thread = &(tsk->thread);
605 unsigned long old_dr7;
606 int i, orig_ret = 0, rc = 0;
607 int enabled, second_pass = 0;
608 unsigned len, type;
609 struct perf_event *bp;
610
611 if (ptrace_get_breakpoints(tsk) < 0)
612 return -ESRCH;
613
614 data &= ~DR_CONTROL_RESERVED;
615 old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
616restore:
617 /*
618 * Loop through all the hardware breakpoints, making the
619 * appropriate changes to each.
620 */
621 for (i = 0; i < HBP_NUM; i++) {
622 enabled = decode_dr7(data, i, &len, &type);
623 bp = thread->ptrace_bps[i];
624
625 if (!enabled) {
626 if (bp) {
627 /*
628 * Don't unregister the breakpoints right-away,
629 * unless all register_user_hw_breakpoint()
630 * requests have succeeded. This prevents
631 * any window of opportunity for debug
632 * register grabbing by other users.
633 */
634 if (!second_pass)
635 continue;
636
637 rc = ptrace_modify_breakpoint(bp, len, type,
638 tsk, 1);
639 if (rc)
640 break;
641 }
642 continue;
643 }
644
645 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
646 if (rc)
647 break;
648 }
649 /*
650 * Make a second pass to free the remaining unused breakpoints
651 * or to restore the original breakpoints if an error occurred.
652 */
653 if (!second_pass) {
654 second_pass = 1;
655 if (rc < 0) {
656 orig_ret = rc;
657 data = old_dr7;
658 }
659 goto restore;
660 }
661
662 ptrace_put_breakpoints(tsk);
663
664 return ((orig_ret < 0) ? orig_ret : rc);
665}
666
667/*
668 * Handle PTRACE_PEEKUSR calls for the debug register area.
669 */
670static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
671{
672 struct thread_struct *thread = &(tsk->thread);
673 unsigned long val = 0;
674
675 if (n < HBP_NUM) {
676 struct perf_event *bp;
677
678 if (ptrace_get_breakpoints(tsk) < 0)
679 return -ESRCH;
680
681 bp = thread->ptrace_bps[n];
682 if (!bp)
683 val = 0;
684 else
685 val = bp->hw.info.address;
686
687 ptrace_put_breakpoints(tsk);
688 } else if (n == 6) {
689 val = thread->debugreg6;
690 } else if (n == 7) {
691 val = thread->ptrace_dr7;
692 }
693 return val;
694}
695
696static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
697 unsigned long addr)
698{
699 struct perf_event *bp;
700 struct thread_struct *t = &tsk->thread;
701 struct perf_event_attr attr;
702 int err = 0;
703
704 if (ptrace_get_breakpoints(tsk) < 0)
705 return -ESRCH;
706
707 if (!t->ptrace_bps[nr]) {
708 ptrace_breakpoint_init(&attr);
709 /*
710 * Put stub len and type to register (reserve) an inactive but
711 * correct bp
712 */
713 attr.bp_addr = addr;
714 attr.bp_len = HW_BREAKPOINT_LEN_1;
715 attr.bp_type = HW_BREAKPOINT_W;
716 attr.disabled = 1;
717
718 bp = register_user_hw_breakpoint(&attr, ptrace_triggered,
719 NULL, tsk);
720
721 /*
722 * CHECKME: the previous code returned -EIO if the addr wasn't
723 * a valid task virtual addr. The new one will return -EINVAL in
724 * this case.
725 * -EINVAL may be what we want for in-kernel breakpoints users,
726 * but -EIO looks better for ptrace, since we refuse a register
727 * writing for the user. And anyway this is the previous
728 * behaviour.
729 */
730 if (IS_ERR(bp)) {
731 err = PTR_ERR(bp);
732 goto put;
733 }
734
735 t->ptrace_bps[nr] = bp;
736 } else {
737 bp = t->ptrace_bps[nr];
738
739 attr = bp->attr;
740 attr.bp_addr = addr;
741 err = modify_user_hw_breakpoint(bp, &attr);
742 }
743
744put:
745 ptrace_put_breakpoints(tsk);
746 return err;
747}
748
749/*
750 * Handle PTRACE_POKEUSR calls for the debug register area.
751 */
752int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val)
753{
754 struct thread_struct *thread = &(tsk->thread);
755 int rc = 0;
756
757 /* There are no DR4 or DR5 registers */
758 if (n == 4 || n == 5)
759 return -EIO;
760
761 if (n == 6) {
762 thread->debugreg6 = val;
763 goto ret_path;
764 }
765 if (n < HBP_NUM) {
766 rc = ptrace_set_breakpoint_addr(tsk, n, val);
767 if (rc)
768 return rc;
769 }
770 /* All that's left is DR7 */
771 if (n == 7) {
772 rc = ptrace_write_dr7(tsk, val);
773 if (!rc)
774 thread->ptrace_dr7 = val;
775 }
776
777ret_path:
778 return rc;
779}
780
781/*
782 * These access the current or another (stopped) task's io permission
783 * bitmap for debugging or core dump.
784 */
785static int ioperm_active(struct task_struct *target,
786 const struct user_regset *regset)
787{
788 return target->thread.io_bitmap_max / regset->size;
789}
790
791static int ioperm_get(struct task_struct *target,
792 const struct user_regset *regset,
793 unsigned int pos, unsigned int count,
794 void *kbuf, void __user *ubuf)
795{
796 if (!target->thread.io_bitmap_ptr)
797 return -ENXIO;
798
799 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
800 target->thread.io_bitmap_ptr,
801 0, IO_BITMAP_BYTES);
802}
803
804/*
805 * Called by kernel/ptrace.c when detaching..
806 *
807 * Make sure the single step bit is not set.
808 */
809void ptrace_disable(struct task_struct *child)
810{
811 user_disable_single_step(child);
812#ifdef TIF_SYSCALL_EMU
813 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
814#endif
815}
816
817#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
818static const struct user_regset_view user_x86_32_view; /* Initialized below. */
819#endif
820
821long arch_ptrace(struct task_struct *child, long request,
822 unsigned long addr, unsigned long data)
823{
824 int ret;
825 unsigned long __user *datap = (unsigned long __user *)data;
826
827 switch (request) {
828 /* read the word at location addr in the USER area. */
829 case PTRACE_PEEKUSR: {
830 unsigned long tmp;
831
832 ret = -EIO;
833 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
834 break;
835
836 tmp = 0; /* Default return condition */
837 if (addr < sizeof(struct user_regs_struct))
838 tmp = getreg(child, addr);
839 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
840 addr <= offsetof(struct user, u_debugreg[7])) {
841 addr -= offsetof(struct user, u_debugreg[0]);
842 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
843 }
844 ret = put_user(tmp, datap);
845 break;
846 }
847
848 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
849 ret = -EIO;
850 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
851 break;
852
853 if (addr < sizeof(struct user_regs_struct))
854 ret = putreg(child, addr, data);
855 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
856 addr <= offsetof(struct user, u_debugreg[7])) {
857 addr -= offsetof(struct user, u_debugreg[0]);
858 ret = ptrace_set_debugreg(child,
859 addr / sizeof(data), data);
860 }
861 break;
862
863 case PTRACE_GETREGS: /* Get all gp regs from the child. */
864 return copy_regset_to_user(child,
865 task_user_regset_view(current),
866 REGSET_GENERAL,
867 0, sizeof(struct user_regs_struct),
868 datap);
869
870 case PTRACE_SETREGS: /* Set all gp regs in the child. */
871 return copy_regset_from_user(child,
872 task_user_regset_view(current),
873 REGSET_GENERAL,
874 0, sizeof(struct user_regs_struct),
875 datap);
876
877 case PTRACE_GETFPREGS: /* Get the child FPU state. */
878 return copy_regset_to_user(child,
879 task_user_regset_view(current),
880 REGSET_FP,
881 0, sizeof(struct user_i387_struct),
882 datap);
883
884 case PTRACE_SETFPREGS: /* Set the child FPU state. */
885 return copy_regset_from_user(child,
886 task_user_regset_view(current),
887 REGSET_FP,
888 0, sizeof(struct user_i387_struct),
889 datap);
890
891#ifdef CONFIG_X86_32
892 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
893 return copy_regset_to_user(child, &user_x86_32_view,
894 REGSET_XFP,
895 0, sizeof(struct user_fxsr_struct),
896 datap) ? -EIO : 0;
897
898 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
899 return copy_regset_from_user(child, &user_x86_32_view,
900 REGSET_XFP,
901 0, sizeof(struct user_fxsr_struct),
902 datap) ? -EIO : 0;
903#endif
904
905#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
906 case PTRACE_GET_THREAD_AREA:
907 if ((int) addr < 0)
908 return -EIO;
909 ret = do_get_thread_area(child, addr,
910 (struct user_desc __user *)data);
911 break;
912
913 case PTRACE_SET_THREAD_AREA:
914 if ((int) addr < 0)
915 return -EIO;
916 ret = do_set_thread_area(child, addr,
917 (struct user_desc __user *)data, 0);
918 break;
919#endif
920
921#ifdef CONFIG_X86_64
922 /* normal 64bit interface to access TLS data.
923 Works just like arch_prctl, except that the arguments
924 are reversed. */
925 case PTRACE_ARCH_PRCTL:
926 ret = do_arch_prctl(child, data, addr);
927 break;
928#endif
929
930 default:
931 ret = ptrace_request(child, request, addr, data);
932 break;
933 }
934
935 return ret;
936}
937
938#ifdef CONFIG_IA32_EMULATION
939
940#include <linux/compat.h>
941#include <linux/syscalls.h>
942#include <asm/ia32.h>
943#include <asm/user32.h>
944
945#define R32(l,q) \
946 case offsetof(struct user32, regs.l): \
947 regs->q = value; break
948
949#define SEG32(rs) \
950 case offsetof(struct user32, regs.rs): \
951 return set_segment_reg(child, \
952 offsetof(struct user_regs_struct, rs), \
953 value); \
954 break
955
956static int putreg32(struct task_struct *child, unsigned regno, u32 value)
957{
958 struct pt_regs *regs = task_pt_regs(child);
959
960 switch (regno) {
961
962 SEG32(cs);
963 SEG32(ds);
964 SEG32(es);
965 SEG32(fs);
966 SEG32(gs);
967 SEG32(ss);
968
969 R32(ebx, bx);
970 R32(ecx, cx);
971 R32(edx, dx);
972 R32(edi, di);
973 R32(esi, si);
974 R32(ebp, bp);
975 R32(eax, ax);
976 R32(eip, ip);
977 R32(esp, sp);
978
979 case offsetof(struct user32, regs.orig_eax):
980 /*
981 * A 32-bit debugger setting orig_eax means to restore
982 * the state of the task restarting a 32-bit syscall.
983 * Make sure we interpret the -ERESTART* codes correctly
984 * in case the task is not actually still sitting at the
985 * exit from a 32-bit syscall with TS_COMPAT still set.
986 */
987 regs->orig_ax = value;
988 if (syscall_get_nr(child, regs) >= 0)
989 task_thread_info(child)->status |= TS_COMPAT;
990 break;
991
992 case offsetof(struct user32, regs.eflags):
993 return set_flags(child, value);
994
995 case offsetof(struct user32, u_debugreg[0]) ...
996 offsetof(struct user32, u_debugreg[7]):
997 regno -= offsetof(struct user32, u_debugreg[0]);
998 return ptrace_set_debugreg(child, regno / 4, value);
999
1000 default:
1001 if (regno > sizeof(struct user32) || (regno & 3))
1002 return -EIO;
1003
1004 /*
1005 * Other dummy fields in the virtual user structure
1006 * are ignored
1007 */
1008 break;
1009 }
1010 return 0;
1011}
1012
1013#undef R32
1014#undef SEG32
1015
1016#define R32(l,q) \
1017 case offsetof(struct user32, regs.l): \
1018 *val = regs->q; break
1019
1020#define SEG32(rs) \
1021 case offsetof(struct user32, regs.rs): \
1022 *val = get_segment_reg(child, \
1023 offsetof(struct user_regs_struct, rs)); \
1024 break
1025
1026static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1027{
1028 struct pt_regs *regs = task_pt_regs(child);
1029
1030 switch (regno) {
1031
1032 SEG32(ds);
1033 SEG32(es);
1034 SEG32(fs);
1035 SEG32(gs);
1036
1037 R32(cs, cs);
1038 R32(ss, ss);
1039 R32(ebx, bx);
1040 R32(ecx, cx);
1041 R32(edx, dx);
1042 R32(edi, di);
1043 R32(esi, si);
1044 R32(ebp, bp);
1045 R32(eax, ax);
1046 R32(orig_eax, orig_ax);
1047 R32(eip, ip);
1048 R32(esp, sp);
1049
1050 case offsetof(struct user32, regs.eflags):
1051 *val = get_flags(child);
1052 break;
1053
1054 case offsetof(struct user32, u_debugreg[0]) ...
1055 offsetof(struct user32, u_debugreg[7]):
1056 regno -= offsetof(struct user32, u_debugreg[0]);
1057 *val = ptrace_get_debugreg(child, regno / 4);
1058 break;
1059
1060 default:
1061 if (regno > sizeof(struct user32) || (regno & 3))
1062 return -EIO;
1063
1064 /*
1065 * Other dummy fields in the virtual user structure
1066 * are ignored
1067 */
1068 *val = 0;
1069 break;
1070 }
1071 return 0;
1072}
1073
1074#undef R32
1075#undef SEG32
1076
1077static int genregs32_get(struct task_struct *target,
1078 const struct user_regset *regset,
1079 unsigned int pos, unsigned int count,
1080 void *kbuf, void __user *ubuf)
1081{
1082 if (kbuf) {
1083 compat_ulong_t *k = kbuf;
1084 while (count >= sizeof(*k)) {
1085 getreg32(target, pos, k++);
1086 count -= sizeof(*k);
1087 pos += sizeof(*k);
1088 }
1089 } else {
1090 compat_ulong_t __user *u = ubuf;
1091 while (count >= sizeof(*u)) {
1092 compat_ulong_t word;
1093 getreg32(target, pos, &word);
1094 if (__put_user(word, u++))
1095 return -EFAULT;
1096 count -= sizeof(*u);
1097 pos += sizeof(*u);
1098 }
1099 }
1100
1101 return 0;
1102}
1103
1104static int genregs32_set(struct task_struct *target,
1105 const struct user_regset *regset,
1106 unsigned int pos, unsigned int count,
1107 const void *kbuf, const void __user *ubuf)
1108{
1109 int ret = 0;
1110 if (kbuf) {
1111 const compat_ulong_t *k = kbuf;
1112 while (count >= sizeof(*k) && !ret) {
1113 ret = putreg32(target, pos, *k++);
1114 count -= sizeof(*k);
1115 pos += sizeof(*k);
1116 }
1117 } else {
1118 const compat_ulong_t __user *u = ubuf;
1119 while (count >= sizeof(*u) && !ret) {
1120 compat_ulong_t word;
1121 ret = __get_user(word, u++);
1122 if (ret)
1123 break;
1124 ret = putreg32(target, pos, word);
1125 count -= sizeof(*u);
1126 pos += sizeof(*u);
1127 }
1128 }
1129 return ret;
1130}
1131
1132long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1133 compat_ulong_t caddr, compat_ulong_t cdata)
1134{
1135 unsigned long addr = caddr;
1136 unsigned long data = cdata;
1137 void __user *datap = compat_ptr(data);
1138 int ret;
1139 __u32 val;
1140
1141 switch (request) {
1142 case PTRACE_PEEKUSR:
1143 ret = getreg32(child, addr, &val);
1144 if (ret == 0)
1145 ret = put_user(val, (__u32 __user *)datap);
1146 break;
1147
1148 case PTRACE_POKEUSR:
1149 ret = putreg32(child, addr, data);
1150 break;
1151
1152 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1153 return copy_regset_to_user(child, &user_x86_32_view,
1154 REGSET_GENERAL,
1155 0, sizeof(struct user_regs_struct32),
1156 datap);
1157
1158 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1159 return copy_regset_from_user(child, &user_x86_32_view,
1160 REGSET_GENERAL, 0,
1161 sizeof(struct user_regs_struct32),
1162 datap);
1163
1164 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1165 return copy_regset_to_user(child, &user_x86_32_view,
1166 REGSET_FP, 0,
1167 sizeof(struct user_i387_ia32_struct),
1168 datap);
1169
1170 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1171 return copy_regset_from_user(
1172 child, &user_x86_32_view, REGSET_FP,
1173 0, sizeof(struct user_i387_ia32_struct), datap);
1174
1175 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1176 return copy_regset_to_user(child, &user_x86_32_view,
1177 REGSET_XFP, 0,
1178 sizeof(struct user32_fxsr_struct),
1179 datap);
1180
1181 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1182 return copy_regset_from_user(child, &user_x86_32_view,
1183 REGSET_XFP, 0,
1184 sizeof(struct user32_fxsr_struct),
1185 datap);
1186
1187 case PTRACE_GET_THREAD_AREA:
1188 case PTRACE_SET_THREAD_AREA:
1189 return arch_ptrace(child, request, addr, data);
1190
1191 default:
1192 return compat_ptrace_request(child, request, addr, data);
1193 }
1194
1195 return ret;
1196}
1197
1198#endif /* CONFIG_IA32_EMULATION */
1199
1200#ifdef CONFIG_X86_64
1201
1202static struct user_regset x86_64_regsets[] __read_mostly = {
1203 [REGSET_GENERAL] = {
1204 .core_note_type = NT_PRSTATUS,
1205 .n = sizeof(struct user_regs_struct) / sizeof(long),
1206 .size = sizeof(long), .align = sizeof(long),
1207 .get = genregs_get, .set = genregs_set
1208 },
1209 [REGSET_FP] = {
1210 .core_note_type = NT_PRFPREG,
1211 .n = sizeof(struct user_i387_struct) / sizeof(long),
1212 .size = sizeof(long), .align = sizeof(long),
1213 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1214 },
1215 [REGSET_XSTATE] = {
1216 .core_note_type = NT_X86_XSTATE,
1217 .size = sizeof(u64), .align = sizeof(u64),
1218 .active = xstateregs_active, .get = xstateregs_get,
1219 .set = xstateregs_set
1220 },
1221 [REGSET_IOPERM64] = {
1222 .core_note_type = NT_386_IOPERM,
1223 .n = IO_BITMAP_LONGS,
1224 .size = sizeof(long), .align = sizeof(long),
1225 .active = ioperm_active, .get = ioperm_get
1226 },
1227};
1228
1229static const struct user_regset_view user_x86_64_view = {
1230 .name = "x86_64", .e_machine = EM_X86_64,
1231 .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1232};
1233
1234#else /* CONFIG_X86_32 */
1235
1236#define user_regs_struct32 user_regs_struct
1237#define genregs32_get genregs_get
1238#define genregs32_set genregs_set
1239
1240#define user_i387_ia32_struct user_i387_struct
1241#define user32_fxsr_struct user_fxsr_struct
1242
1243#endif /* CONFIG_X86_64 */
1244
1245#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1246static struct user_regset x86_32_regsets[] __read_mostly = {
1247 [REGSET_GENERAL] = {
1248 .core_note_type = NT_PRSTATUS,
1249 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1250 .size = sizeof(u32), .align = sizeof(u32),
1251 .get = genregs32_get, .set = genregs32_set
1252 },
1253 [REGSET_FP] = {
1254 .core_note_type = NT_PRFPREG,
1255 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1256 .size = sizeof(u32), .align = sizeof(u32),
1257 .active = fpregs_active, .get = fpregs_get, .set = fpregs_set
1258 },
1259 [REGSET_XFP] = {
1260 .core_note_type = NT_PRXFPREG,
1261 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1262 .size = sizeof(u32), .align = sizeof(u32),
1263 .active = xfpregs_active, .get = xfpregs_get, .set = xfpregs_set
1264 },
1265 [REGSET_XSTATE] = {
1266 .core_note_type = NT_X86_XSTATE,
1267 .size = sizeof(u64), .align = sizeof(u64),
1268 .active = xstateregs_active, .get = xstateregs_get,
1269 .set = xstateregs_set
1270 },
1271 [REGSET_TLS] = {
1272 .core_note_type = NT_386_TLS,
1273 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1274 .size = sizeof(struct user_desc),
1275 .align = sizeof(struct user_desc),
1276 .active = regset_tls_active,
1277 .get = regset_tls_get, .set = regset_tls_set
1278 },
1279 [REGSET_IOPERM32] = {
1280 .core_note_type = NT_386_IOPERM,
1281 .n = IO_BITMAP_BYTES / sizeof(u32),
1282 .size = sizeof(u32), .align = sizeof(u32),
1283 .active = ioperm_active, .get = ioperm_get
1284 },
1285};
1286
1287static const struct user_regset_view user_x86_32_view = {
1288 .name = "i386", .e_machine = EM_386,
1289 .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1290};
1291#endif
1292
1293/*
1294 * This represents bytes 464..511 in the memory layout exported through
1295 * the REGSET_XSTATE interface.
1296 */
1297u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1298
1299void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1300{
1301#ifdef CONFIG_X86_64
1302 x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1303#endif
1304#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1305 x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1306#endif
1307 xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1308}
1309
1310const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1311{
1312#ifdef CONFIG_IA32_EMULATION
1313 if (test_tsk_thread_flag(task, TIF_IA32))
1314#endif
1315#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1316 return &user_x86_32_view;
1317#endif
1318#ifdef CONFIG_X86_64
1319 return &user_x86_64_view;
1320#endif
1321}
1322
1323static void fill_sigtrap_info(struct task_struct *tsk,
1324 struct pt_regs *regs,
1325 int error_code, int si_code,
1326 struct siginfo *info)
1327{
1328 tsk->thread.trap_no = 1;
1329 tsk->thread.error_code = error_code;
1330
1331 memset(info, 0, sizeof(*info));
1332 info->si_signo = SIGTRAP;
1333 info->si_code = si_code;
1334 info->si_addr = user_mode_vm(regs) ? (void __user *)regs->ip : NULL;
1335}
1336
1337void user_single_step_siginfo(struct task_struct *tsk,
1338 struct pt_regs *regs,
1339 struct siginfo *info)
1340{
1341 fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1342}
1343
1344void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1345 int error_code, int si_code)
1346{
1347 struct siginfo info;
1348
1349 fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1350 /* Send us the fake SIGTRAP */
1351 force_sig_info(SIGTRAP, &info, tsk);
1352}
1353
1354
1355#ifdef CONFIG_X86_32
1356# define IS_IA32 1
1357#elif defined CONFIG_IA32_EMULATION
1358# define IS_IA32 is_compat_task()
1359#else
1360# define IS_IA32 0
1361#endif
1362
1363/*
1364 * We must return the syscall number to actually look up in the table.
1365 * This can be -1L to skip running any syscall at all.
1366 */
1367long syscall_trace_enter(struct pt_regs *regs)
1368{
1369 long ret = 0;
1370
1371 /*
1372 * If we stepped into a sysenter/syscall insn, it trapped in
1373 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1374 * If user-mode had set TF itself, then it's still clear from
1375 * do_debug() and we need to set it again to restore the user
1376 * state. If we entered on the slow path, TF was already set.
1377 */
1378 if (test_thread_flag(TIF_SINGLESTEP))
1379 regs->flags |= X86_EFLAGS_TF;
1380
1381 /* do the secure computing check first */
1382 secure_computing(regs->orig_ax);
1383
1384 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1385 ret = -1L;
1386
1387 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1388 tracehook_report_syscall_entry(regs))
1389 ret = -1L;
1390
1391 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1392 trace_sys_enter(regs, regs->orig_ax);
1393
1394 if (unlikely(current->audit_context)) {
1395 if (IS_IA32)
1396 audit_syscall_entry(AUDIT_ARCH_I386,
1397 regs->orig_ax,
1398 regs->bx, regs->cx,
1399 regs->dx, regs->si);
1400#ifdef CONFIG_X86_64
1401 else
1402 audit_syscall_entry(AUDIT_ARCH_X86_64,
1403 regs->orig_ax,
1404 regs->di, regs->si,
1405 regs->dx, regs->r10);
1406#endif
1407 }
1408
1409 return ret ?: regs->orig_ax;
1410}
1411
1412void syscall_trace_leave(struct pt_regs *regs)
1413{
1414 bool step;
1415
1416 if (unlikely(current->audit_context))
1417 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1418
1419 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1420 trace_sys_exit(regs, regs->ax);
1421
1422 /*
1423 * If TIF_SYSCALL_EMU is set, we only get here because of
1424 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1425 * We already reported this syscall instruction in
1426 * syscall_trace_enter().
1427 */
1428 step = unlikely(test_thread_flag(TIF_SINGLESTEP)) &&
1429 !test_thread_flag(TIF_SYSCALL_EMU);
1430 if (step || test_thread_flag(TIF_SYSCALL_TRACE))
1431 tracehook_report_syscall_exit(regs, step);
1432}
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
44enum 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
54struct 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
62static 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 */
102int 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 */
118const 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 */
145static 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 '®s->sp'.
158 *
159 * Now, if the stack is empty, '®s->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 */
166unsigned 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)®s->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}
181EXPORT_SYMBOL_GPL(kernel_stack_pointer);
182
183static 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 ®s->bx + (regno >> 2);
187}
188
189static 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
206static 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
248static 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 ®s->r15 + (offset / sizeof(regs->r15));
252}
253
254static 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
295static 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 /*
307 * If this is setting fs as for normal 64-bit use but
308 * setting fs_base has implicitly changed it, leave it.
309 */
310 if ((value == FS_TLS_SEL && task->thread.fsindex == 0 &&
311 task->thread.fs != 0) ||
312 (value == 0 && task->thread.fsindex == FS_TLS_SEL &&
313 task->thread.fs == 0))
314 break;
315 task->thread.fsindex = value;
316 if (task == current)
317 loadsegment(fs, task->thread.fsindex);
318 break;
319 case offsetof(struct user_regs_struct,gs):
320 /*
321 * If this is setting gs as for normal 64-bit use but
322 * setting gs_base has implicitly changed it, leave it.
323 */
324 if ((value == GS_TLS_SEL && task->thread.gsindex == 0 &&
325 task->thread.gs != 0) ||
326 (value == 0 && task->thread.gsindex == GS_TLS_SEL &&
327 task->thread.gs == 0))
328 break;
329 task->thread.gsindex = value;
330 if (task == current)
331 load_gs_index(task->thread.gsindex);
332 break;
333 case offsetof(struct user_regs_struct,ds):
334 task->thread.ds = value;
335 if (task == current)
336 loadsegment(ds, task->thread.ds);
337 break;
338 case offsetof(struct user_regs_struct,es):
339 task->thread.es = value;
340 if (task == current)
341 loadsegment(es, task->thread.es);
342 break;
343
344 /*
345 * Can't actually change these in 64-bit mode.
346 */
347 case offsetof(struct user_regs_struct,cs):
348 if (unlikely(value == 0))
349 return -EIO;
350 task_pt_regs(task)->cs = value;
351 break;
352 case offsetof(struct user_regs_struct,ss):
353 if (unlikely(value == 0))
354 return -EIO;
355 task_pt_regs(task)->ss = value;
356 break;
357 }
358
359 return 0;
360}
361
362#endif /* CONFIG_X86_32 */
363
364static unsigned long get_flags(struct task_struct *task)
365{
366 unsigned long retval = task_pt_regs(task)->flags;
367
368 /*
369 * If the debugger set TF, hide it from the readout.
370 */
371 if (test_tsk_thread_flag(task, TIF_FORCED_TF))
372 retval &= ~X86_EFLAGS_TF;
373
374 return retval;
375}
376
377static int set_flags(struct task_struct *task, unsigned long value)
378{
379 struct pt_regs *regs = task_pt_regs(task);
380
381 /*
382 * If the user value contains TF, mark that
383 * it was not "us" (the debugger) that set it.
384 * If not, make sure it stays set if we had.
385 */
386 if (value & X86_EFLAGS_TF)
387 clear_tsk_thread_flag(task, TIF_FORCED_TF);
388 else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
389 value |= X86_EFLAGS_TF;
390
391 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
392
393 return 0;
394}
395
396static int putreg(struct task_struct *child,
397 unsigned long offset, unsigned long value)
398{
399 switch (offset) {
400 case offsetof(struct user_regs_struct, cs):
401 case offsetof(struct user_regs_struct, ds):
402 case offsetof(struct user_regs_struct, es):
403 case offsetof(struct user_regs_struct, fs):
404 case offsetof(struct user_regs_struct, gs):
405 case offsetof(struct user_regs_struct, ss):
406 return set_segment_reg(child, offset, value);
407
408 case offsetof(struct user_regs_struct, flags):
409 return set_flags(child, value);
410
411#ifdef CONFIG_X86_64
412 case offsetof(struct user_regs_struct,fs_base):
413 if (value >= TASK_SIZE_OF(child))
414 return -EIO;
415 /*
416 * When changing the segment base, use do_arch_prctl
417 * to set either thread.fs or thread.fsindex and the
418 * corresponding GDT slot.
419 */
420 if (child->thread.fs != value)
421 return do_arch_prctl(child, ARCH_SET_FS, value);
422 return 0;
423 case offsetof(struct user_regs_struct,gs_base):
424 /*
425 * Exactly the same here as the %fs handling above.
426 */
427 if (value >= TASK_SIZE_OF(child))
428 return -EIO;
429 if (child->thread.gs != value)
430 return do_arch_prctl(child, ARCH_SET_GS, value);
431 return 0;
432#endif
433 }
434
435 *pt_regs_access(task_pt_regs(child), offset) = value;
436 return 0;
437}
438
439static unsigned long getreg(struct task_struct *task, unsigned long offset)
440{
441 switch (offset) {
442 case offsetof(struct user_regs_struct, cs):
443 case offsetof(struct user_regs_struct, ds):
444 case offsetof(struct user_regs_struct, es):
445 case offsetof(struct user_regs_struct, fs):
446 case offsetof(struct user_regs_struct, gs):
447 case offsetof(struct user_regs_struct, ss):
448 return get_segment_reg(task, offset);
449
450 case offsetof(struct user_regs_struct, flags):
451 return get_flags(task);
452
453#ifdef CONFIG_X86_64
454 case offsetof(struct user_regs_struct, fs_base): {
455 /*
456 * do_arch_prctl may have used a GDT slot instead of
457 * the MSR. To userland, it appears the same either
458 * way, except the %fs segment selector might not be 0.
459 */
460 unsigned int seg = task->thread.fsindex;
461 if (task->thread.fs != 0)
462 return task->thread.fs;
463 if (task == current)
464 asm("movl %%fs,%0" : "=r" (seg));
465 if (seg != FS_TLS_SEL)
466 return 0;
467 return get_desc_base(&task->thread.tls_array[FS_TLS]);
468 }
469 case offsetof(struct user_regs_struct, gs_base): {
470 /*
471 * Exactly the same here as the %fs handling above.
472 */
473 unsigned int seg = task->thread.gsindex;
474 if (task->thread.gs != 0)
475 return task->thread.gs;
476 if (task == current)
477 asm("movl %%gs,%0" : "=r" (seg));
478 if (seg != GS_TLS_SEL)
479 return 0;
480 return get_desc_base(&task->thread.tls_array[GS_TLS]);
481 }
482#endif
483 }
484
485 return *pt_regs_access(task_pt_regs(task), offset);
486}
487
488static int genregs_get(struct task_struct *target,
489 const struct user_regset *regset,
490 unsigned int pos, unsigned int count,
491 void *kbuf, void __user *ubuf)
492{
493 if (kbuf) {
494 unsigned long *k = kbuf;
495 while (count >= sizeof(*k)) {
496 *k++ = getreg(target, pos);
497 count -= sizeof(*k);
498 pos += sizeof(*k);
499 }
500 } else {
501 unsigned long __user *u = ubuf;
502 while (count >= sizeof(*u)) {
503 if (__put_user(getreg(target, pos), u++))
504 return -EFAULT;
505 count -= sizeof(*u);
506 pos += sizeof(*u);
507 }
508 }
509
510 return 0;
511}
512
513static int genregs_set(struct task_struct *target,
514 const struct user_regset *regset,
515 unsigned int pos, unsigned int count,
516 const void *kbuf, const void __user *ubuf)
517{
518 int ret = 0;
519 if (kbuf) {
520 const unsigned long *k = kbuf;
521 while (count >= sizeof(*k) && !ret) {
522 ret = putreg(target, pos, *k++);
523 count -= sizeof(*k);
524 pos += sizeof(*k);
525 }
526 } else {
527 const unsigned long __user *u = ubuf;
528 while (count >= sizeof(*u) && !ret) {
529 unsigned long word;
530 ret = __get_user(word, u++);
531 if (ret)
532 break;
533 ret = putreg(target, pos, word);
534 count -= sizeof(*u);
535 pos += sizeof(*u);
536 }
537 }
538 return ret;
539}
540
541static void ptrace_triggered(struct perf_event *bp,
542 struct perf_sample_data *data,
543 struct pt_regs *regs)
544{
545 int i;
546 struct thread_struct *thread = &(current->thread);
547
548 /*
549 * Store in the virtual DR6 register the fact that the breakpoint
550 * was hit so the thread's debugger will see it.
551 */
552 for (i = 0; i < HBP_NUM; i++) {
553 if (thread->ptrace_bps[i] == bp)
554 break;
555 }
556
557 thread->debugreg6 |= (DR_TRAP0 << i);
558}
559
560/*
561 * Walk through every ptrace breakpoints for this thread and
562 * build the dr7 value on top of their attributes.
563 *
564 */
565static unsigned long ptrace_get_dr7(struct perf_event *bp[])
566{
567 int i;
568 int dr7 = 0;
569 struct arch_hw_breakpoint *info;
570
571 for (i = 0; i < HBP_NUM; i++) {
572 if (bp[i] && !bp[i]->attr.disabled) {
573 info = counter_arch_bp(bp[i]);
574 dr7 |= encode_dr7(i, info->len, info->type);
575 }
576 }
577
578 return dr7;
579}
580
581static int ptrace_fill_bp_fields(struct perf_event_attr *attr,
582 int len, int type, bool disabled)
583{
584 int err, bp_len, bp_type;
585
586 err = arch_bp_generic_fields(len, type, &bp_len, &bp_type);
587 if (!err) {
588 attr->bp_len = bp_len;
589 attr->bp_type = bp_type;
590 attr->disabled = disabled;
591 }
592
593 return err;
594}
595
596static struct perf_event *
597ptrace_register_breakpoint(struct task_struct *tsk, int len, int type,
598 unsigned long addr, bool disabled)
599{
600 struct perf_event_attr attr;
601 int err;
602
603 ptrace_breakpoint_init(&attr);
604 attr.bp_addr = addr;
605
606 err = ptrace_fill_bp_fields(&attr, len, type, disabled);
607 if (err)
608 return ERR_PTR(err);
609
610 return register_user_hw_breakpoint(&attr, ptrace_triggered,
611 NULL, tsk);
612}
613
614static int ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
615 int disabled)
616{
617 struct perf_event_attr attr = bp->attr;
618 int err;
619
620 err = ptrace_fill_bp_fields(&attr, len, type, disabled);
621 if (err)
622 return err;
623
624 return modify_user_hw_breakpoint(bp, &attr);
625}
626
627/*
628 * Handle ptrace writes to debug register 7.
629 */
630static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
631{
632 struct thread_struct *thread = &tsk->thread;
633 unsigned long old_dr7;
634 bool second_pass = false;
635 int i, rc, ret = 0;
636
637 data &= ~DR_CONTROL_RESERVED;
638 old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
639
640restore:
641 rc = 0;
642 for (i = 0; i < HBP_NUM; i++) {
643 unsigned len, type;
644 bool disabled = !decode_dr7(data, i, &len, &type);
645 struct perf_event *bp = thread->ptrace_bps[i];
646
647 if (!bp) {
648 if (disabled)
649 continue;
650
651 bp = ptrace_register_breakpoint(tsk,
652 len, type, 0, disabled);
653 if (IS_ERR(bp)) {
654 rc = PTR_ERR(bp);
655 break;
656 }
657
658 thread->ptrace_bps[i] = bp;
659 continue;
660 }
661
662 rc = ptrace_modify_breakpoint(bp, len, type, disabled);
663 if (rc)
664 break;
665 }
666
667 /* Restore if the first pass failed, second_pass shouldn't fail. */
668 if (rc && !WARN_ON(second_pass)) {
669 ret = rc;
670 data = old_dr7;
671 second_pass = true;
672 goto restore;
673 }
674
675 return ret;
676}
677
678/*
679 * Handle PTRACE_PEEKUSR calls for the debug register area.
680 */
681static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
682{
683 struct thread_struct *thread = &tsk->thread;
684 unsigned long val = 0;
685
686 if (n < HBP_NUM) {
687 struct perf_event *bp = thread->ptrace_bps[n];
688
689 if (bp)
690 val = bp->hw.info.address;
691 } else if (n == 6) {
692 val = thread->debugreg6;
693 } else if (n == 7) {
694 val = thread->ptrace_dr7;
695 }
696 return val;
697}
698
699static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
700 unsigned long addr)
701{
702 struct thread_struct *t = &tsk->thread;
703 struct perf_event *bp = t->ptrace_bps[nr];
704 int err = 0;
705
706 if (!bp) {
707 /*
708 * Put stub len and type to create an inactive but correct bp.
709 *
710 * CHECKME: the previous code returned -EIO if the addr wasn't
711 * a valid task virtual addr. The new one will return -EINVAL in
712 * this case.
713 * -EINVAL may be what we want for in-kernel breakpoints users,
714 * but -EIO looks better for ptrace, since we refuse a register
715 * writing for the user. And anyway this is the previous
716 * behaviour.
717 */
718 bp = ptrace_register_breakpoint(tsk,
719 X86_BREAKPOINT_LEN_1, X86_BREAKPOINT_WRITE,
720 addr, true);
721 if (IS_ERR(bp))
722 err = PTR_ERR(bp);
723 else
724 t->ptrace_bps[nr] = bp;
725 } else {
726 struct perf_event_attr attr = bp->attr;
727
728 attr.bp_addr = addr;
729 err = modify_user_hw_breakpoint(bp, &attr);
730 }
731
732 return err;
733}
734
735/*
736 * Handle PTRACE_POKEUSR calls for the debug register area.
737 */
738static int ptrace_set_debugreg(struct task_struct *tsk, int n,
739 unsigned long val)
740{
741 struct thread_struct *thread = &tsk->thread;
742 /* There are no DR4 or DR5 registers */
743 int rc = -EIO;
744
745 if (n < HBP_NUM) {
746 rc = ptrace_set_breakpoint_addr(tsk, n, val);
747 } else if (n == 6) {
748 thread->debugreg6 = val;
749 rc = 0;
750 } else if (n == 7) {
751 rc = ptrace_write_dr7(tsk, val);
752 if (!rc)
753 thread->ptrace_dr7 = val;
754 }
755 return rc;
756}
757
758/*
759 * These access the current or another (stopped) task's io permission
760 * bitmap for debugging or core dump.
761 */
762static int ioperm_active(struct task_struct *target,
763 const struct user_regset *regset)
764{
765 return target->thread.io_bitmap_max / regset->size;
766}
767
768static int ioperm_get(struct task_struct *target,
769 const struct user_regset *regset,
770 unsigned int pos, unsigned int count,
771 void *kbuf, void __user *ubuf)
772{
773 if (!target->thread.io_bitmap_ptr)
774 return -ENXIO;
775
776 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
777 target->thread.io_bitmap_ptr,
778 0, IO_BITMAP_BYTES);
779}
780
781/*
782 * Called by kernel/ptrace.c when detaching..
783 *
784 * Make sure the single step bit is not set.
785 */
786void ptrace_disable(struct task_struct *child)
787{
788 user_disable_single_step(child);
789#ifdef TIF_SYSCALL_EMU
790 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
791#endif
792}
793
794#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
795static const struct user_regset_view user_x86_32_view; /* Initialized below. */
796#endif
797
798long arch_ptrace(struct task_struct *child, long request,
799 unsigned long addr, unsigned long data)
800{
801 int ret;
802 unsigned long __user *datap = (unsigned long __user *)data;
803
804 switch (request) {
805 /* read the word at location addr in the USER area. */
806 case PTRACE_PEEKUSR: {
807 unsigned long tmp;
808
809 ret = -EIO;
810 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
811 break;
812
813 tmp = 0; /* Default return condition */
814 if (addr < sizeof(struct user_regs_struct))
815 tmp = getreg(child, addr);
816 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
817 addr <= offsetof(struct user, u_debugreg[7])) {
818 addr -= offsetof(struct user, u_debugreg[0]);
819 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
820 }
821 ret = put_user(tmp, datap);
822 break;
823 }
824
825 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
826 ret = -EIO;
827 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
828 break;
829
830 if (addr < sizeof(struct user_regs_struct))
831 ret = putreg(child, addr, data);
832 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
833 addr <= offsetof(struct user, u_debugreg[7])) {
834 addr -= offsetof(struct user, u_debugreg[0]);
835 ret = ptrace_set_debugreg(child,
836 addr / sizeof(data), data);
837 }
838 break;
839
840 case PTRACE_GETREGS: /* Get all gp regs from the child. */
841 return copy_regset_to_user(child,
842 task_user_regset_view(current),
843 REGSET_GENERAL,
844 0, sizeof(struct user_regs_struct),
845 datap);
846
847 case PTRACE_SETREGS: /* Set all gp regs in the child. */
848 return copy_regset_from_user(child,
849 task_user_regset_view(current),
850 REGSET_GENERAL,
851 0, sizeof(struct user_regs_struct),
852 datap);
853
854 case PTRACE_GETFPREGS: /* Get the child FPU state. */
855 return copy_regset_to_user(child,
856 task_user_regset_view(current),
857 REGSET_FP,
858 0, sizeof(struct user_i387_struct),
859 datap);
860
861 case PTRACE_SETFPREGS: /* Set the child FPU state. */
862 return copy_regset_from_user(child,
863 task_user_regset_view(current),
864 REGSET_FP,
865 0, sizeof(struct user_i387_struct),
866 datap);
867
868#ifdef CONFIG_X86_32
869 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
870 return copy_regset_to_user(child, &user_x86_32_view,
871 REGSET_XFP,
872 0, sizeof(struct user_fxsr_struct),
873 datap) ? -EIO : 0;
874
875 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
876 return copy_regset_from_user(child, &user_x86_32_view,
877 REGSET_XFP,
878 0, sizeof(struct user_fxsr_struct),
879 datap) ? -EIO : 0;
880#endif
881
882#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
883 case PTRACE_GET_THREAD_AREA:
884 if ((int) addr < 0)
885 return -EIO;
886 ret = do_get_thread_area(child, addr,
887 (struct user_desc __user *)data);
888 break;
889
890 case PTRACE_SET_THREAD_AREA:
891 if ((int) addr < 0)
892 return -EIO;
893 ret = do_set_thread_area(child, addr,
894 (struct user_desc __user *)data, 0);
895 break;
896#endif
897
898#ifdef CONFIG_X86_64
899 /* normal 64bit interface to access TLS data.
900 Works just like arch_prctl, except that the arguments
901 are reversed. */
902 case PTRACE_ARCH_PRCTL:
903 ret = do_arch_prctl(child, data, addr);
904 break;
905#endif
906
907 default:
908 ret = ptrace_request(child, request, addr, data);
909 break;
910 }
911
912 return ret;
913}
914
915#ifdef CONFIG_IA32_EMULATION
916
917#include <linux/compat.h>
918#include <linux/syscalls.h>
919#include <asm/ia32.h>
920#include <asm/user32.h>
921
922#define R32(l,q) \
923 case offsetof(struct user32, regs.l): \
924 regs->q = value; break
925
926#define SEG32(rs) \
927 case offsetof(struct user32, regs.rs): \
928 return set_segment_reg(child, \
929 offsetof(struct user_regs_struct, rs), \
930 value); \
931 break
932
933static int putreg32(struct task_struct *child, unsigned regno, u32 value)
934{
935 struct pt_regs *regs = task_pt_regs(child);
936
937 switch (regno) {
938
939 SEG32(cs);
940 SEG32(ds);
941 SEG32(es);
942 SEG32(fs);
943 SEG32(gs);
944 SEG32(ss);
945
946 R32(ebx, bx);
947 R32(ecx, cx);
948 R32(edx, dx);
949 R32(edi, di);
950 R32(esi, si);
951 R32(ebp, bp);
952 R32(eax, ax);
953 R32(eip, ip);
954 R32(esp, sp);
955
956 case offsetof(struct user32, regs.orig_eax):
957 /*
958 * A 32-bit debugger setting orig_eax means to restore
959 * the state of the task restarting a 32-bit syscall.
960 * Make sure we interpret the -ERESTART* codes correctly
961 * in case the task is not actually still sitting at the
962 * exit from a 32-bit syscall with TS_COMPAT still set.
963 */
964 regs->orig_ax = value;
965 if (syscall_get_nr(child, regs) >= 0)
966 task_thread_info(child)->status |= TS_COMPAT;
967 break;
968
969 case offsetof(struct user32, regs.eflags):
970 return set_flags(child, value);
971
972 case offsetof(struct user32, u_debugreg[0]) ...
973 offsetof(struct user32, u_debugreg[7]):
974 regno -= offsetof(struct user32, u_debugreg[0]);
975 return ptrace_set_debugreg(child, regno / 4, value);
976
977 default:
978 if (regno > sizeof(struct user32) || (regno & 3))
979 return -EIO;
980
981 /*
982 * Other dummy fields in the virtual user structure
983 * are ignored
984 */
985 break;
986 }
987 return 0;
988}
989
990#undef R32
991#undef SEG32
992
993#define R32(l,q) \
994 case offsetof(struct user32, regs.l): \
995 *val = regs->q; break
996
997#define SEG32(rs) \
998 case offsetof(struct user32, regs.rs): \
999 *val = get_segment_reg(child, \
1000 offsetof(struct user_regs_struct, rs)); \
1001 break
1002
1003static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1004{
1005 struct pt_regs *regs = task_pt_regs(child);
1006
1007 switch (regno) {
1008
1009 SEG32(ds);
1010 SEG32(es);
1011 SEG32(fs);
1012 SEG32(gs);
1013
1014 R32(cs, cs);
1015 R32(ss, ss);
1016 R32(ebx, bx);
1017 R32(ecx, cx);
1018 R32(edx, dx);
1019 R32(edi, di);
1020 R32(esi, si);
1021 R32(ebp, bp);
1022 R32(eax, ax);
1023 R32(orig_eax, orig_ax);
1024 R32(eip, ip);
1025 R32(esp, sp);
1026
1027 case offsetof(struct user32, regs.eflags):
1028 *val = get_flags(child);
1029 break;
1030
1031 case offsetof(struct user32, u_debugreg[0]) ...
1032 offsetof(struct user32, u_debugreg[7]):
1033 regno -= offsetof(struct user32, u_debugreg[0]);
1034 *val = ptrace_get_debugreg(child, regno / 4);
1035 break;
1036
1037 default:
1038 if (regno > sizeof(struct user32) || (regno & 3))
1039 return -EIO;
1040
1041 /*
1042 * Other dummy fields in the virtual user structure
1043 * are ignored
1044 */
1045 *val = 0;
1046 break;
1047 }
1048 return 0;
1049}
1050
1051#undef R32
1052#undef SEG32
1053
1054static int genregs32_get(struct task_struct *target,
1055 const struct user_regset *regset,
1056 unsigned int pos, unsigned int count,
1057 void *kbuf, void __user *ubuf)
1058{
1059 if (kbuf) {
1060 compat_ulong_t *k = kbuf;
1061 while (count >= sizeof(*k)) {
1062 getreg32(target, pos, k++);
1063 count -= sizeof(*k);
1064 pos += sizeof(*k);
1065 }
1066 } else {
1067 compat_ulong_t __user *u = ubuf;
1068 while (count >= sizeof(*u)) {
1069 compat_ulong_t word;
1070 getreg32(target, pos, &word);
1071 if (__put_user(word, u++))
1072 return -EFAULT;
1073 count -= sizeof(*u);
1074 pos += sizeof(*u);
1075 }
1076 }
1077
1078 return 0;
1079}
1080
1081static int genregs32_set(struct task_struct *target,
1082 const struct user_regset *regset,
1083 unsigned int pos, unsigned int count,
1084 const void *kbuf, const void __user *ubuf)
1085{
1086 int ret = 0;
1087 if (kbuf) {
1088 const compat_ulong_t *k = kbuf;
1089 while (count >= sizeof(*k) && !ret) {
1090 ret = putreg32(target, pos, *k++);
1091 count -= sizeof(*k);
1092 pos += sizeof(*k);
1093 }
1094 } else {
1095 const compat_ulong_t __user *u = ubuf;
1096 while (count >= sizeof(*u) && !ret) {
1097 compat_ulong_t word;
1098 ret = __get_user(word, u++);
1099 if (ret)
1100 break;
1101 ret = putreg32(target, pos, word);
1102 count -= sizeof(*u);
1103 pos += sizeof(*u);
1104 }
1105 }
1106 return ret;
1107}
1108
1109static long ia32_arch_ptrace(struct task_struct *child, compat_long_t request,
1110 compat_ulong_t caddr, compat_ulong_t cdata)
1111{
1112 unsigned long addr = caddr;
1113 unsigned long data = cdata;
1114 void __user *datap = compat_ptr(data);
1115 int ret;
1116 __u32 val;
1117
1118 switch (request) {
1119 case PTRACE_PEEKUSR:
1120 ret = getreg32(child, addr, &val);
1121 if (ret == 0)
1122 ret = put_user(val, (__u32 __user *)datap);
1123 break;
1124
1125 case PTRACE_POKEUSR:
1126 ret = putreg32(child, addr, data);
1127 break;
1128
1129 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1130 return copy_regset_to_user(child, &user_x86_32_view,
1131 REGSET_GENERAL,
1132 0, sizeof(struct user_regs_struct32),
1133 datap);
1134
1135 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1136 return copy_regset_from_user(child, &user_x86_32_view,
1137 REGSET_GENERAL, 0,
1138 sizeof(struct user_regs_struct32),
1139 datap);
1140
1141 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1142 return copy_regset_to_user(child, &user_x86_32_view,
1143 REGSET_FP, 0,
1144 sizeof(struct user_i387_ia32_struct),
1145 datap);
1146
1147 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1148 return copy_regset_from_user(
1149 child, &user_x86_32_view, REGSET_FP,
1150 0, sizeof(struct user_i387_ia32_struct), datap);
1151
1152 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1153 return copy_regset_to_user(child, &user_x86_32_view,
1154 REGSET_XFP, 0,
1155 sizeof(struct user32_fxsr_struct),
1156 datap);
1157
1158 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1159 return copy_regset_from_user(child, &user_x86_32_view,
1160 REGSET_XFP, 0,
1161 sizeof(struct user32_fxsr_struct),
1162 datap);
1163
1164 case PTRACE_GET_THREAD_AREA:
1165 case PTRACE_SET_THREAD_AREA:
1166 return arch_ptrace(child, request, addr, data);
1167
1168 default:
1169 return compat_ptrace_request(child, request, addr, data);
1170 }
1171
1172 return ret;
1173}
1174#endif /* CONFIG_IA32_EMULATION */
1175
1176#ifdef CONFIG_X86_X32_ABI
1177static long x32_arch_ptrace(struct task_struct *child,
1178 compat_long_t request, compat_ulong_t caddr,
1179 compat_ulong_t cdata)
1180{
1181 unsigned long addr = caddr;
1182 unsigned long data = cdata;
1183 void __user *datap = compat_ptr(data);
1184 int ret;
1185
1186 switch (request) {
1187 /* Read 32bits at location addr in the USER area. Only allow
1188 to return the lower 32bits of segment and debug registers. */
1189 case PTRACE_PEEKUSR: {
1190 u32 tmp;
1191
1192 ret = -EIO;
1193 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1194 addr < offsetof(struct user_regs_struct, cs))
1195 break;
1196
1197 tmp = 0; /* Default return condition */
1198 if (addr < sizeof(struct user_regs_struct))
1199 tmp = getreg(child, addr);
1200 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1201 addr <= offsetof(struct user, u_debugreg[7])) {
1202 addr -= offsetof(struct user, u_debugreg[0]);
1203 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1204 }
1205 ret = put_user(tmp, (__u32 __user *)datap);
1206 break;
1207 }
1208
1209 /* Write the word at location addr in the USER area. Only allow
1210 to update segment and debug registers with the upper 32bits
1211 zero-extended. */
1212 case PTRACE_POKEUSR:
1213 ret = -EIO;
1214 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1215 addr < offsetof(struct user_regs_struct, cs))
1216 break;
1217
1218 if (addr < sizeof(struct user_regs_struct))
1219 ret = putreg(child, addr, data);
1220 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1221 addr <= offsetof(struct user, u_debugreg[7])) {
1222 addr -= offsetof(struct user, u_debugreg[0]);
1223 ret = ptrace_set_debugreg(child,
1224 addr / sizeof(data), data);
1225 }
1226 break;
1227
1228 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1229 return copy_regset_to_user(child,
1230 task_user_regset_view(current),
1231 REGSET_GENERAL,
1232 0, sizeof(struct user_regs_struct),
1233 datap);
1234
1235 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1236 return copy_regset_from_user(child,
1237 task_user_regset_view(current),
1238 REGSET_GENERAL,
1239 0, sizeof(struct user_regs_struct),
1240 datap);
1241
1242 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1243 return copy_regset_to_user(child,
1244 task_user_regset_view(current),
1245 REGSET_FP,
1246 0, sizeof(struct user_i387_struct),
1247 datap);
1248
1249 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1250 return copy_regset_from_user(child,
1251 task_user_regset_view(current),
1252 REGSET_FP,
1253 0, sizeof(struct user_i387_struct),
1254 datap);
1255
1256 default:
1257 return compat_ptrace_request(child, request, addr, data);
1258 }
1259
1260 return ret;
1261}
1262#endif
1263
1264#ifdef CONFIG_COMPAT
1265long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1266 compat_ulong_t caddr, compat_ulong_t cdata)
1267{
1268#ifdef CONFIG_X86_X32_ABI
1269 if (!is_ia32_task())
1270 return x32_arch_ptrace(child, request, caddr, cdata);
1271#endif
1272#ifdef CONFIG_IA32_EMULATION
1273 return ia32_arch_ptrace(child, request, caddr, cdata);
1274#else
1275 return 0;
1276#endif
1277}
1278#endif /* CONFIG_COMPAT */
1279
1280#ifdef CONFIG_X86_64
1281
1282static struct user_regset x86_64_regsets[] __read_mostly = {
1283 [REGSET_GENERAL] = {
1284 .core_note_type = NT_PRSTATUS,
1285 .n = sizeof(struct user_regs_struct) / sizeof(long),
1286 .size = sizeof(long), .align = sizeof(long),
1287 .get = genregs_get, .set = genregs_set
1288 },
1289 [REGSET_FP] = {
1290 .core_note_type = NT_PRFPREG,
1291 .n = sizeof(struct user_i387_struct) / sizeof(long),
1292 .size = sizeof(long), .align = sizeof(long),
1293 .active = regset_xregset_fpregs_active, .get = xfpregs_get, .set = xfpregs_set
1294 },
1295 [REGSET_XSTATE] = {
1296 .core_note_type = NT_X86_XSTATE,
1297 .size = sizeof(u64), .align = sizeof(u64),
1298 .active = xstateregs_active, .get = xstateregs_get,
1299 .set = xstateregs_set
1300 },
1301 [REGSET_IOPERM64] = {
1302 .core_note_type = NT_386_IOPERM,
1303 .n = IO_BITMAP_LONGS,
1304 .size = sizeof(long), .align = sizeof(long),
1305 .active = ioperm_active, .get = ioperm_get
1306 },
1307};
1308
1309static const struct user_regset_view user_x86_64_view = {
1310 .name = "x86_64", .e_machine = EM_X86_64,
1311 .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1312};
1313
1314#else /* CONFIG_X86_32 */
1315
1316#define user_regs_struct32 user_regs_struct
1317#define genregs32_get genregs_get
1318#define genregs32_set genregs_set
1319
1320#endif /* CONFIG_X86_64 */
1321
1322#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1323static struct user_regset x86_32_regsets[] __read_mostly = {
1324 [REGSET_GENERAL] = {
1325 .core_note_type = NT_PRSTATUS,
1326 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1327 .size = sizeof(u32), .align = sizeof(u32),
1328 .get = genregs32_get, .set = genregs32_set
1329 },
1330 [REGSET_FP] = {
1331 .core_note_type = NT_PRFPREG,
1332 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1333 .size = sizeof(u32), .align = sizeof(u32),
1334 .active = regset_fpregs_active, .get = fpregs_get, .set = fpregs_set
1335 },
1336 [REGSET_XFP] = {
1337 .core_note_type = NT_PRXFPREG,
1338 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1339 .size = sizeof(u32), .align = sizeof(u32),
1340 .active = regset_xregset_fpregs_active, .get = xfpregs_get, .set = xfpregs_set
1341 },
1342 [REGSET_XSTATE] = {
1343 .core_note_type = NT_X86_XSTATE,
1344 .size = sizeof(u64), .align = sizeof(u64),
1345 .active = xstateregs_active, .get = xstateregs_get,
1346 .set = xstateregs_set
1347 },
1348 [REGSET_TLS] = {
1349 .core_note_type = NT_386_TLS,
1350 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1351 .size = sizeof(struct user_desc),
1352 .align = sizeof(struct user_desc),
1353 .active = regset_tls_active,
1354 .get = regset_tls_get, .set = regset_tls_set
1355 },
1356 [REGSET_IOPERM32] = {
1357 .core_note_type = NT_386_IOPERM,
1358 .n = IO_BITMAP_BYTES / sizeof(u32),
1359 .size = sizeof(u32), .align = sizeof(u32),
1360 .active = ioperm_active, .get = ioperm_get
1361 },
1362};
1363
1364static const struct user_regset_view user_x86_32_view = {
1365 .name = "i386", .e_machine = EM_386,
1366 .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1367};
1368#endif
1369
1370/*
1371 * This represents bytes 464..511 in the memory layout exported through
1372 * the REGSET_XSTATE interface.
1373 */
1374u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1375
1376void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1377{
1378#ifdef CONFIG_X86_64
1379 x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1380#endif
1381#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1382 x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1383#endif
1384 xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1385}
1386
1387const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1388{
1389#ifdef CONFIG_IA32_EMULATION
1390 if (test_tsk_thread_flag(task, TIF_IA32))
1391#endif
1392#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1393 return &user_x86_32_view;
1394#endif
1395#ifdef CONFIG_X86_64
1396 return &user_x86_64_view;
1397#endif
1398}
1399
1400static void fill_sigtrap_info(struct task_struct *tsk,
1401 struct pt_regs *regs,
1402 int error_code, int si_code,
1403 struct siginfo *info)
1404{
1405 tsk->thread.trap_nr = X86_TRAP_DB;
1406 tsk->thread.error_code = error_code;
1407
1408 memset(info, 0, sizeof(*info));
1409 info->si_signo = SIGTRAP;
1410 info->si_code = si_code;
1411 info->si_addr = user_mode(regs) ? (void __user *)regs->ip : NULL;
1412}
1413
1414void user_single_step_siginfo(struct task_struct *tsk,
1415 struct pt_regs *regs,
1416 struct siginfo *info)
1417{
1418 fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1419}
1420
1421void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1422 int error_code, int si_code)
1423{
1424 struct siginfo info;
1425
1426 fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1427 /* Send us the fake SIGTRAP */
1428 force_sig_info(SIGTRAP, &info, tsk);
1429}