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// SPDX-License-Identifier: GPL-2.0-only
2/* By Ross Biro 1/23/92 */
3/*
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
6 */
7
8#include <linux/kernel.h>
9#include <linux/sched.h>
10#include <linux/sched/task_stack.h>
11#include <linux/mm.h>
12#include <linux/smp.h>
13#include <linux/errno.h>
14#include <linux/slab.h>
15#include <linux/ptrace.h>
16#include <linux/tracehook.h>
17#include <linux/user.h>
18#include <linux/elf.h>
19#include <linux/security.h>
20#include <linux/audit.h>
21#include <linux/seccomp.h>
22#include <linux/signal.h>
23#include <linux/perf_event.h>
24#include <linux/hw_breakpoint.h>
25#include <linux/rcupdate.h>
26#include <linux/export.h>
27#include <linux/context_tracking.h>
28#include <linux/nospec.h>
29
30#include <linux/uaccess.h>
31#include <asm/processor.h>
32#include <asm/fpu/internal.h>
33#include <asm/fpu/signal.h>
34#include <asm/fpu/regset.h>
35#include <asm/debugreg.h>
36#include <asm/ldt.h>
37#include <asm/desc.h>
38#include <asm/prctl.h>
39#include <asm/proto.h>
40#include <asm/hw_breakpoint.h>
41#include <asm/traps.h>
42#include <asm/syscall.h>
43#include <asm/fsgsbase.h>
44#include <asm/io_bitmap.h>
45
46#include "tls.h"
47
48enum x86_regset {
49 REGSET_GENERAL,
50 REGSET_FP,
51 REGSET_XFP,
52 REGSET_IOPERM64 = REGSET_XFP,
53 REGSET_XSTATE,
54 REGSET_TLS,
55 REGSET_IOPERM32,
56};
57
58struct pt_regs_offset {
59 const char *name;
60 int offset;
61};
62
63#define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
64#define REG_OFFSET_END {.name = NULL, .offset = 0}
65
66static const struct pt_regs_offset regoffset_table[] = {
67#ifdef CONFIG_X86_64
68 REG_OFFSET_NAME(r15),
69 REG_OFFSET_NAME(r14),
70 REG_OFFSET_NAME(r13),
71 REG_OFFSET_NAME(r12),
72 REG_OFFSET_NAME(r11),
73 REG_OFFSET_NAME(r10),
74 REG_OFFSET_NAME(r9),
75 REG_OFFSET_NAME(r8),
76#endif
77 REG_OFFSET_NAME(bx),
78 REG_OFFSET_NAME(cx),
79 REG_OFFSET_NAME(dx),
80 REG_OFFSET_NAME(si),
81 REG_OFFSET_NAME(di),
82 REG_OFFSET_NAME(bp),
83 REG_OFFSET_NAME(ax),
84#ifdef CONFIG_X86_32
85 REG_OFFSET_NAME(ds),
86 REG_OFFSET_NAME(es),
87 REG_OFFSET_NAME(fs),
88 REG_OFFSET_NAME(gs),
89#endif
90 REG_OFFSET_NAME(orig_ax),
91 REG_OFFSET_NAME(ip),
92 REG_OFFSET_NAME(cs),
93 REG_OFFSET_NAME(flags),
94 REG_OFFSET_NAME(sp),
95 REG_OFFSET_NAME(ss),
96 REG_OFFSET_END,
97};
98
99/**
100 * regs_query_register_offset() - query register offset from its name
101 * @name: the name of a register
102 *
103 * regs_query_register_offset() returns the offset of a register in struct
104 * pt_regs from its name. If the name is invalid, this returns -EINVAL;
105 */
106int regs_query_register_offset(const char *name)
107{
108 const struct pt_regs_offset *roff;
109 for (roff = regoffset_table; roff->name != NULL; roff++)
110 if (!strcmp(roff->name, name))
111 return roff->offset;
112 return -EINVAL;
113}
114
115/**
116 * regs_query_register_name() - query register name from its offset
117 * @offset: the offset of a register in struct pt_regs.
118 *
119 * regs_query_register_name() returns the name of a register from its
120 * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
121 */
122const char *regs_query_register_name(unsigned int offset)
123{
124 const struct pt_regs_offset *roff;
125 for (roff = regoffset_table; roff->name != NULL; roff++)
126 if (roff->offset == offset)
127 return roff->name;
128 return NULL;
129}
130
131/*
132 * does not yet catch signals sent when the child dies.
133 * in exit.c or in signal.c.
134 */
135
136/*
137 * Determines which flags the user has access to [1 = access, 0 = no access].
138 */
139#define FLAG_MASK_32 ((unsigned long) \
140 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
141 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
142 X86_EFLAGS_SF | X86_EFLAGS_TF | \
143 X86_EFLAGS_DF | X86_EFLAGS_OF | \
144 X86_EFLAGS_RF | X86_EFLAGS_AC))
145
146/*
147 * Determines whether a value may be installed in a segment register.
148 */
149static inline bool invalid_selector(u16 value)
150{
151 return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
152}
153
154#ifdef CONFIG_X86_32
155
156#define FLAG_MASK FLAG_MASK_32
157
158static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
159{
160 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
161 return ®s->bx + (regno >> 2);
162}
163
164static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
165{
166 /*
167 * Returning the value truncates it to 16 bits.
168 */
169 unsigned int retval;
170 if (offset != offsetof(struct user_regs_struct, gs))
171 retval = *pt_regs_access(task_pt_regs(task), offset);
172 else {
173 if (task == current)
174 retval = get_user_gs(task_pt_regs(task));
175 else
176 retval = task_user_gs(task);
177 }
178 return retval;
179}
180
181static int set_segment_reg(struct task_struct *task,
182 unsigned long offset, u16 value)
183{
184 if (WARN_ON_ONCE(task == current))
185 return -EIO;
186
187 /*
188 * The value argument was already truncated to 16 bits.
189 */
190 if (invalid_selector(value))
191 return -EIO;
192
193 /*
194 * For %cs and %ss we cannot permit a null selector.
195 * We can permit a bogus selector as long as it has USER_RPL.
196 * Null selectors are fine for other segment registers, but
197 * we will never get back to user mode with invalid %cs or %ss
198 * and will take the trap in iret instead. Much code relies
199 * on user_mode() to distinguish a user trap frame (which can
200 * safely use invalid selectors) from a kernel trap frame.
201 */
202 switch (offset) {
203 case offsetof(struct user_regs_struct, cs):
204 case offsetof(struct user_regs_struct, ss):
205 if (unlikely(value == 0))
206 return -EIO;
207 fallthrough;
208
209 default:
210 *pt_regs_access(task_pt_regs(task), offset) = value;
211 break;
212
213 case offsetof(struct user_regs_struct, gs):
214 task_user_gs(task) = value;
215 }
216
217 return 0;
218}
219
220#else /* CONFIG_X86_64 */
221
222#define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
223
224static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
225{
226 BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
227 return ®s->r15 + (offset / sizeof(regs->r15));
228}
229
230static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
231{
232 /*
233 * Returning the value truncates it to 16 bits.
234 */
235 unsigned int seg;
236
237 switch (offset) {
238 case offsetof(struct user_regs_struct, fs):
239 if (task == current) {
240 /* Older gas can't assemble movq %?s,%r?? */
241 asm("movl %%fs,%0" : "=r" (seg));
242 return seg;
243 }
244 return task->thread.fsindex;
245 case offsetof(struct user_regs_struct, gs):
246 if (task == current) {
247 asm("movl %%gs,%0" : "=r" (seg));
248 return seg;
249 }
250 return task->thread.gsindex;
251 case offsetof(struct user_regs_struct, ds):
252 if (task == current) {
253 asm("movl %%ds,%0" : "=r" (seg));
254 return seg;
255 }
256 return task->thread.ds;
257 case offsetof(struct user_regs_struct, es):
258 if (task == current) {
259 asm("movl %%es,%0" : "=r" (seg));
260 return seg;
261 }
262 return task->thread.es;
263
264 case offsetof(struct user_regs_struct, cs):
265 case offsetof(struct user_regs_struct, ss):
266 break;
267 }
268 return *pt_regs_access(task_pt_regs(task), offset);
269}
270
271static int set_segment_reg(struct task_struct *task,
272 unsigned long offset, u16 value)
273{
274 if (WARN_ON_ONCE(task == current))
275 return -EIO;
276
277 /*
278 * The value argument was already truncated to 16 bits.
279 */
280 if (invalid_selector(value))
281 return -EIO;
282
283 /*
284 * Writes to FS and GS will change the stored selector. Whether
285 * this changes the segment base as well depends on whether
286 * FSGSBASE is enabled.
287 */
288
289 switch (offset) {
290 case offsetof(struct user_regs_struct,fs):
291 task->thread.fsindex = value;
292 break;
293 case offsetof(struct user_regs_struct,gs):
294 task->thread.gsindex = value;
295 break;
296 case offsetof(struct user_regs_struct,ds):
297 task->thread.ds = value;
298 break;
299 case offsetof(struct user_regs_struct,es):
300 task->thread.es = value;
301 break;
302
303 /*
304 * Can't actually change these in 64-bit mode.
305 */
306 case offsetof(struct user_regs_struct,cs):
307 if (unlikely(value == 0))
308 return -EIO;
309 task_pt_regs(task)->cs = value;
310 break;
311 case offsetof(struct user_regs_struct,ss):
312 if (unlikely(value == 0))
313 return -EIO;
314 task_pt_regs(task)->ss = value;
315 break;
316 }
317
318 return 0;
319}
320
321#endif /* CONFIG_X86_32 */
322
323static unsigned long get_flags(struct task_struct *task)
324{
325 unsigned long retval = task_pt_regs(task)->flags;
326
327 /*
328 * If the debugger set TF, hide it from the readout.
329 */
330 if (test_tsk_thread_flag(task, TIF_FORCED_TF))
331 retval &= ~X86_EFLAGS_TF;
332
333 return retval;
334}
335
336static int set_flags(struct task_struct *task, unsigned long value)
337{
338 struct pt_regs *regs = task_pt_regs(task);
339
340 /*
341 * If the user value contains TF, mark that
342 * it was not "us" (the debugger) that set it.
343 * If not, make sure it stays set if we had.
344 */
345 if (value & X86_EFLAGS_TF)
346 clear_tsk_thread_flag(task, TIF_FORCED_TF);
347 else if (test_tsk_thread_flag(task, TIF_FORCED_TF))
348 value |= X86_EFLAGS_TF;
349
350 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
351
352 return 0;
353}
354
355static int putreg(struct task_struct *child,
356 unsigned long offset, unsigned long value)
357{
358 switch (offset) {
359 case offsetof(struct user_regs_struct, cs):
360 case offsetof(struct user_regs_struct, ds):
361 case offsetof(struct user_regs_struct, es):
362 case offsetof(struct user_regs_struct, fs):
363 case offsetof(struct user_regs_struct, gs):
364 case offsetof(struct user_regs_struct, ss):
365 return set_segment_reg(child, offset, value);
366
367 case offsetof(struct user_regs_struct, flags):
368 return set_flags(child, value);
369
370#ifdef CONFIG_X86_64
371 case offsetof(struct user_regs_struct,fs_base):
372 if (value >= TASK_SIZE_MAX)
373 return -EIO;
374 x86_fsbase_write_task(child, value);
375 return 0;
376 case offsetof(struct user_regs_struct,gs_base):
377 if (value >= TASK_SIZE_MAX)
378 return -EIO;
379 x86_gsbase_write_task(child, value);
380 return 0;
381#endif
382 }
383
384 *pt_regs_access(task_pt_regs(child), offset) = value;
385 return 0;
386}
387
388static unsigned long getreg(struct task_struct *task, unsigned long offset)
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 get_segment_reg(task, offset);
398
399 case offsetof(struct user_regs_struct, flags):
400 return get_flags(task);
401
402#ifdef CONFIG_X86_64
403 case offsetof(struct user_regs_struct, fs_base):
404 return x86_fsbase_read_task(task);
405 case offsetof(struct user_regs_struct, gs_base):
406 return x86_gsbase_read_task(task);
407#endif
408 }
409
410 return *pt_regs_access(task_pt_regs(task), offset);
411}
412
413static int genregs_get(struct task_struct *target,
414 const struct user_regset *regset,
415 struct membuf to)
416{
417 int reg;
418
419 for (reg = 0; to.left; reg++)
420 membuf_store(&to, getreg(target, reg * sizeof(unsigned long)));
421 return 0;
422}
423
424static int genregs_set(struct task_struct *target,
425 const struct user_regset *regset,
426 unsigned int pos, unsigned int count,
427 const void *kbuf, const void __user *ubuf)
428{
429 int ret = 0;
430 if (kbuf) {
431 const unsigned long *k = kbuf;
432 while (count >= sizeof(*k) && !ret) {
433 ret = putreg(target, pos, *k++);
434 count -= sizeof(*k);
435 pos += sizeof(*k);
436 }
437 } else {
438 const unsigned long __user *u = ubuf;
439 while (count >= sizeof(*u) && !ret) {
440 unsigned long word;
441 ret = __get_user(word, u++);
442 if (ret)
443 break;
444 ret = putreg(target, pos, word);
445 count -= sizeof(*u);
446 pos += sizeof(*u);
447 }
448 }
449 return ret;
450}
451
452static void ptrace_triggered(struct perf_event *bp,
453 struct perf_sample_data *data,
454 struct pt_regs *regs)
455{
456 int i;
457 struct thread_struct *thread = &(current->thread);
458
459 /*
460 * Store in the virtual DR6 register the fact that the breakpoint
461 * was hit so the thread's debugger will see it.
462 */
463 for (i = 0; i < HBP_NUM; i++) {
464 if (thread->ptrace_bps[i] == bp)
465 break;
466 }
467
468 thread->virtual_dr6 |= (DR_TRAP0 << i);
469}
470
471/*
472 * Walk through every ptrace breakpoints for this thread and
473 * build the dr7 value on top of their attributes.
474 *
475 */
476static unsigned long ptrace_get_dr7(struct perf_event *bp[])
477{
478 int i;
479 int dr7 = 0;
480 struct arch_hw_breakpoint *info;
481
482 for (i = 0; i < HBP_NUM; i++) {
483 if (bp[i] && !bp[i]->attr.disabled) {
484 info = counter_arch_bp(bp[i]);
485 dr7 |= encode_dr7(i, info->len, info->type);
486 }
487 }
488
489 return dr7;
490}
491
492static int ptrace_fill_bp_fields(struct perf_event_attr *attr,
493 int len, int type, bool disabled)
494{
495 int err, bp_len, bp_type;
496
497 err = arch_bp_generic_fields(len, type, &bp_len, &bp_type);
498 if (!err) {
499 attr->bp_len = bp_len;
500 attr->bp_type = bp_type;
501 attr->disabled = disabled;
502 }
503
504 return err;
505}
506
507static struct perf_event *
508ptrace_register_breakpoint(struct task_struct *tsk, int len, int type,
509 unsigned long addr, bool disabled)
510{
511 struct perf_event_attr attr;
512 int err;
513
514 ptrace_breakpoint_init(&attr);
515 attr.bp_addr = addr;
516
517 err = ptrace_fill_bp_fields(&attr, len, type, disabled);
518 if (err)
519 return ERR_PTR(err);
520
521 return register_user_hw_breakpoint(&attr, ptrace_triggered,
522 NULL, tsk);
523}
524
525static int ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
526 int disabled)
527{
528 struct perf_event_attr attr = bp->attr;
529 int err;
530
531 err = ptrace_fill_bp_fields(&attr, len, type, disabled);
532 if (err)
533 return err;
534
535 return modify_user_hw_breakpoint(bp, &attr);
536}
537
538/*
539 * Handle ptrace writes to debug register 7.
540 */
541static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
542{
543 struct thread_struct *thread = &tsk->thread;
544 unsigned long old_dr7;
545 bool second_pass = false;
546 int i, rc, ret = 0;
547
548 data &= ~DR_CONTROL_RESERVED;
549 old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
550
551restore:
552 rc = 0;
553 for (i = 0; i < HBP_NUM; i++) {
554 unsigned len, type;
555 bool disabled = !decode_dr7(data, i, &len, &type);
556 struct perf_event *bp = thread->ptrace_bps[i];
557
558 if (!bp) {
559 if (disabled)
560 continue;
561
562 bp = ptrace_register_breakpoint(tsk,
563 len, type, 0, disabled);
564 if (IS_ERR(bp)) {
565 rc = PTR_ERR(bp);
566 break;
567 }
568
569 thread->ptrace_bps[i] = bp;
570 continue;
571 }
572
573 rc = ptrace_modify_breakpoint(bp, len, type, disabled);
574 if (rc)
575 break;
576 }
577
578 /* Restore if the first pass failed, second_pass shouldn't fail. */
579 if (rc && !WARN_ON(second_pass)) {
580 ret = rc;
581 data = old_dr7;
582 second_pass = true;
583 goto restore;
584 }
585
586 return ret;
587}
588
589/*
590 * Handle PTRACE_PEEKUSR calls for the debug register area.
591 */
592static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
593{
594 struct thread_struct *thread = &tsk->thread;
595 unsigned long val = 0;
596
597 if (n < HBP_NUM) {
598 int index = array_index_nospec(n, HBP_NUM);
599 struct perf_event *bp = thread->ptrace_bps[index];
600
601 if (bp)
602 val = bp->hw.info.address;
603 } else if (n == 6) {
604 val = thread->virtual_dr6 ^ DR6_RESERVED; /* Flip back to arch polarity */
605 } else if (n == 7) {
606 val = thread->ptrace_dr7;
607 }
608 return val;
609}
610
611static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
612 unsigned long addr)
613{
614 struct thread_struct *t = &tsk->thread;
615 struct perf_event *bp = t->ptrace_bps[nr];
616 int err = 0;
617
618 if (!bp) {
619 /*
620 * Put stub len and type to create an inactive but correct bp.
621 *
622 * CHECKME: the previous code returned -EIO if the addr wasn't
623 * a valid task virtual addr. The new one will return -EINVAL in
624 * this case.
625 * -EINVAL may be what we want for in-kernel breakpoints users,
626 * but -EIO looks better for ptrace, since we refuse a register
627 * writing for the user. And anyway this is the previous
628 * behaviour.
629 */
630 bp = ptrace_register_breakpoint(tsk,
631 X86_BREAKPOINT_LEN_1, X86_BREAKPOINT_WRITE,
632 addr, true);
633 if (IS_ERR(bp))
634 err = PTR_ERR(bp);
635 else
636 t->ptrace_bps[nr] = bp;
637 } else {
638 struct perf_event_attr attr = bp->attr;
639
640 attr.bp_addr = addr;
641 err = modify_user_hw_breakpoint(bp, &attr);
642 }
643
644 return err;
645}
646
647/*
648 * Handle PTRACE_POKEUSR calls for the debug register area.
649 */
650static int ptrace_set_debugreg(struct task_struct *tsk, int n,
651 unsigned long val)
652{
653 struct thread_struct *thread = &tsk->thread;
654 /* There are no DR4 or DR5 registers */
655 int rc = -EIO;
656
657 if (n < HBP_NUM) {
658 rc = ptrace_set_breakpoint_addr(tsk, n, val);
659 } else if (n == 6) {
660 thread->virtual_dr6 = val ^ DR6_RESERVED; /* Flip to positive polarity */
661 rc = 0;
662 } else if (n == 7) {
663 rc = ptrace_write_dr7(tsk, val);
664 if (!rc)
665 thread->ptrace_dr7 = val;
666 }
667 return rc;
668}
669
670/*
671 * These access the current or another (stopped) task's io permission
672 * bitmap for debugging or core dump.
673 */
674static int ioperm_active(struct task_struct *target,
675 const struct user_regset *regset)
676{
677 struct io_bitmap *iobm = target->thread.io_bitmap;
678
679 return iobm ? DIV_ROUND_UP(iobm->max, regset->size) : 0;
680}
681
682static int ioperm_get(struct task_struct *target,
683 const struct user_regset *regset,
684 struct membuf to)
685{
686 struct io_bitmap *iobm = target->thread.io_bitmap;
687
688 if (!iobm)
689 return -ENXIO;
690
691 return membuf_write(&to, iobm->bitmap, IO_BITMAP_BYTES);
692}
693
694/*
695 * Called by kernel/ptrace.c when detaching..
696 *
697 * Make sure the single step bit is not set.
698 */
699void ptrace_disable(struct task_struct *child)
700{
701 user_disable_single_step(child);
702}
703
704#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
705static const struct user_regset_view user_x86_32_view; /* Initialized below. */
706#endif
707#ifdef CONFIG_X86_64
708static const struct user_regset_view user_x86_64_view; /* Initialized below. */
709#endif
710
711long arch_ptrace(struct task_struct *child, long request,
712 unsigned long addr, unsigned long data)
713{
714 int ret;
715 unsigned long __user *datap = (unsigned long __user *)data;
716
717#ifdef CONFIG_X86_64
718 /* This is native 64-bit ptrace() */
719 const struct user_regset_view *regset_view = &user_x86_64_view;
720#else
721 /* This is native 32-bit ptrace() */
722 const struct user_regset_view *regset_view = &user_x86_32_view;
723#endif
724
725 switch (request) {
726 /* read the word at location addr in the USER area. */
727 case PTRACE_PEEKUSR: {
728 unsigned long tmp;
729
730 ret = -EIO;
731 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
732 break;
733
734 tmp = 0; /* Default return condition */
735 if (addr < sizeof(struct user_regs_struct))
736 tmp = getreg(child, addr);
737 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
738 addr <= offsetof(struct user, u_debugreg[7])) {
739 addr -= offsetof(struct user, u_debugreg[0]);
740 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
741 }
742 ret = put_user(tmp, datap);
743 break;
744 }
745
746 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
747 ret = -EIO;
748 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
749 break;
750
751 if (addr < sizeof(struct user_regs_struct))
752 ret = putreg(child, addr, data);
753 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
754 addr <= offsetof(struct user, u_debugreg[7])) {
755 addr -= offsetof(struct user, u_debugreg[0]);
756 ret = ptrace_set_debugreg(child,
757 addr / sizeof(data), data);
758 }
759 break;
760
761 case PTRACE_GETREGS: /* Get all gp regs from the child. */
762 return copy_regset_to_user(child,
763 regset_view,
764 REGSET_GENERAL,
765 0, sizeof(struct user_regs_struct),
766 datap);
767
768 case PTRACE_SETREGS: /* Set all gp regs in the child. */
769 return copy_regset_from_user(child,
770 regset_view,
771 REGSET_GENERAL,
772 0, sizeof(struct user_regs_struct),
773 datap);
774
775 case PTRACE_GETFPREGS: /* Get the child FPU state. */
776 return copy_regset_to_user(child,
777 regset_view,
778 REGSET_FP,
779 0, sizeof(struct user_i387_struct),
780 datap);
781
782 case PTRACE_SETFPREGS: /* Set the child FPU state. */
783 return copy_regset_from_user(child,
784 regset_view,
785 REGSET_FP,
786 0, sizeof(struct user_i387_struct),
787 datap);
788
789#ifdef CONFIG_X86_32
790 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
791 return copy_regset_to_user(child, &user_x86_32_view,
792 REGSET_XFP,
793 0, sizeof(struct user_fxsr_struct),
794 datap) ? -EIO : 0;
795
796 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
797 return copy_regset_from_user(child, &user_x86_32_view,
798 REGSET_XFP,
799 0, sizeof(struct user_fxsr_struct),
800 datap) ? -EIO : 0;
801#endif
802
803#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
804 case PTRACE_GET_THREAD_AREA:
805 if ((int) addr < 0)
806 return -EIO;
807 ret = do_get_thread_area(child, addr,
808 (struct user_desc __user *)data);
809 break;
810
811 case PTRACE_SET_THREAD_AREA:
812 if ((int) addr < 0)
813 return -EIO;
814 ret = do_set_thread_area(child, addr,
815 (struct user_desc __user *)data, 0);
816 break;
817#endif
818
819#ifdef CONFIG_X86_64
820 /* normal 64bit interface to access TLS data.
821 Works just like arch_prctl, except that the arguments
822 are reversed. */
823 case PTRACE_ARCH_PRCTL:
824 ret = do_arch_prctl_64(child, data, addr);
825 break;
826#endif
827
828 default:
829 ret = ptrace_request(child, request, addr, data);
830 break;
831 }
832
833 return ret;
834}
835
836#ifdef CONFIG_IA32_EMULATION
837
838#include <linux/compat.h>
839#include <linux/syscalls.h>
840#include <asm/ia32.h>
841#include <asm/user32.h>
842
843#define R32(l,q) \
844 case offsetof(struct user32, regs.l): \
845 regs->q = value; break
846
847#define SEG32(rs) \
848 case offsetof(struct user32, regs.rs): \
849 return set_segment_reg(child, \
850 offsetof(struct user_regs_struct, rs), \
851 value); \
852 break
853
854static int putreg32(struct task_struct *child, unsigned regno, u32 value)
855{
856 struct pt_regs *regs = task_pt_regs(child);
857 int ret;
858
859 switch (regno) {
860
861 SEG32(cs);
862 SEG32(ds);
863 SEG32(es);
864
865 /*
866 * A 32-bit ptracer on a 64-bit kernel expects that writing
867 * FS or GS will also update the base. This is needed for
868 * operations like PTRACE_SETREGS to fully restore a saved
869 * CPU state.
870 */
871
872 case offsetof(struct user32, regs.fs):
873 ret = set_segment_reg(child,
874 offsetof(struct user_regs_struct, fs),
875 value);
876 if (ret == 0)
877 child->thread.fsbase =
878 x86_fsgsbase_read_task(child, value);
879 return ret;
880
881 case offsetof(struct user32, regs.gs):
882 ret = set_segment_reg(child,
883 offsetof(struct user_regs_struct, gs),
884 value);
885 if (ret == 0)
886 child->thread.gsbase =
887 x86_fsgsbase_read_task(child, value);
888 return ret;
889
890 SEG32(ss);
891
892 R32(ebx, bx);
893 R32(ecx, cx);
894 R32(edx, dx);
895 R32(edi, di);
896 R32(esi, si);
897 R32(ebp, bp);
898 R32(eax, ax);
899 R32(eip, ip);
900 R32(esp, sp);
901
902 case offsetof(struct user32, regs.orig_eax):
903 /*
904 * Warning: bizarre corner case fixup here. A 32-bit
905 * debugger setting orig_eax to -1 wants to disable
906 * syscall restart. Make sure that the syscall
907 * restart code sign-extends orig_ax. Also make sure
908 * we interpret the -ERESTART* codes correctly if
909 * loaded into regs->ax in case the task is not
910 * actually still sitting at the exit from a 32-bit
911 * syscall with TS_COMPAT still set.
912 */
913 regs->orig_ax = value;
914 if (syscall_get_nr(child, regs) != -1)
915 child->thread_info.status |= TS_I386_REGS_POKED;
916 break;
917
918 case offsetof(struct user32, regs.eflags):
919 return set_flags(child, value);
920
921 case offsetof(struct user32, u_debugreg[0]) ...
922 offsetof(struct user32, u_debugreg[7]):
923 regno -= offsetof(struct user32, u_debugreg[0]);
924 return ptrace_set_debugreg(child, regno / 4, value);
925
926 default:
927 if (regno > sizeof(struct user32) || (regno & 3))
928 return -EIO;
929
930 /*
931 * Other dummy fields in the virtual user structure
932 * are ignored
933 */
934 break;
935 }
936 return 0;
937}
938
939#undef R32
940#undef SEG32
941
942#define R32(l,q) \
943 case offsetof(struct user32, regs.l): \
944 *val = regs->q; break
945
946#define SEG32(rs) \
947 case offsetof(struct user32, regs.rs): \
948 *val = get_segment_reg(child, \
949 offsetof(struct user_regs_struct, rs)); \
950 break
951
952static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
953{
954 struct pt_regs *regs = task_pt_regs(child);
955
956 switch (regno) {
957
958 SEG32(ds);
959 SEG32(es);
960 SEG32(fs);
961 SEG32(gs);
962
963 R32(cs, cs);
964 R32(ss, ss);
965 R32(ebx, bx);
966 R32(ecx, cx);
967 R32(edx, dx);
968 R32(edi, di);
969 R32(esi, si);
970 R32(ebp, bp);
971 R32(eax, ax);
972 R32(orig_eax, orig_ax);
973 R32(eip, ip);
974 R32(esp, sp);
975
976 case offsetof(struct user32, regs.eflags):
977 *val = get_flags(child);
978 break;
979
980 case offsetof(struct user32, u_debugreg[0]) ...
981 offsetof(struct user32, u_debugreg[7]):
982 regno -= offsetof(struct user32, u_debugreg[0]);
983 *val = ptrace_get_debugreg(child, regno / 4);
984 break;
985
986 default:
987 if (regno > sizeof(struct user32) || (regno & 3))
988 return -EIO;
989
990 /*
991 * Other dummy fields in the virtual user structure
992 * are ignored
993 */
994 *val = 0;
995 break;
996 }
997 return 0;
998}
999
1000#undef R32
1001#undef SEG32
1002
1003static int genregs32_get(struct task_struct *target,
1004 const struct user_regset *regset,
1005 struct membuf to)
1006{
1007 int reg;
1008
1009 for (reg = 0; to.left; reg++) {
1010 u32 val;
1011 getreg32(target, reg * 4, &val);
1012 membuf_store(&to, val);
1013 }
1014 return 0;
1015}
1016
1017static int genregs32_set(struct task_struct *target,
1018 const struct user_regset *regset,
1019 unsigned int pos, unsigned int count,
1020 const void *kbuf, const void __user *ubuf)
1021{
1022 int ret = 0;
1023 if (kbuf) {
1024 const compat_ulong_t *k = kbuf;
1025 while (count >= sizeof(*k) && !ret) {
1026 ret = putreg32(target, pos, *k++);
1027 count -= sizeof(*k);
1028 pos += sizeof(*k);
1029 }
1030 } else {
1031 const compat_ulong_t __user *u = ubuf;
1032 while (count >= sizeof(*u) && !ret) {
1033 compat_ulong_t word;
1034 ret = __get_user(word, u++);
1035 if (ret)
1036 break;
1037 ret = putreg32(target, pos, word);
1038 count -= sizeof(*u);
1039 pos += sizeof(*u);
1040 }
1041 }
1042 return ret;
1043}
1044
1045static long ia32_arch_ptrace(struct task_struct *child, compat_long_t request,
1046 compat_ulong_t caddr, compat_ulong_t cdata)
1047{
1048 unsigned long addr = caddr;
1049 unsigned long data = cdata;
1050 void __user *datap = compat_ptr(data);
1051 int ret;
1052 __u32 val;
1053
1054 switch (request) {
1055 case PTRACE_PEEKUSR:
1056 ret = getreg32(child, addr, &val);
1057 if (ret == 0)
1058 ret = put_user(val, (__u32 __user *)datap);
1059 break;
1060
1061 case PTRACE_POKEUSR:
1062 ret = putreg32(child, addr, data);
1063 break;
1064
1065 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1066 return copy_regset_to_user(child, &user_x86_32_view,
1067 REGSET_GENERAL,
1068 0, sizeof(struct user_regs_struct32),
1069 datap);
1070
1071 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1072 return copy_regset_from_user(child, &user_x86_32_view,
1073 REGSET_GENERAL, 0,
1074 sizeof(struct user_regs_struct32),
1075 datap);
1076
1077 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1078 return copy_regset_to_user(child, &user_x86_32_view,
1079 REGSET_FP, 0,
1080 sizeof(struct user_i387_ia32_struct),
1081 datap);
1082
1083 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1084 return copy_regset_from_user(
1085 child, &user_x86_32_view, REGSET_FP,
1086 0, sizeof(struct user_i387_ia32_struct), datap);
1087
1088 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1089 return copy_regset_to_user(child, &user_x86_32_view,
1090 REGSET_XFP, 0,
1091 sizeof(struct user32_fxsr_struct),
1092 datap);
1093
1094 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1095 return copy_regset_from_user(child, &user_x86_32_view,
1096 REGSET_XFP, 0,
1097 sizeof(struct user32_fxsr_struct),
1098 datap);
1099
1100 case PTRACE_GET_THREAD_AREA:
1101 case PTRACE_SET_THREAD_AREA:
1102 return arch_ptrace(child, request, addr, data);
1103
1104 default:
1105 return compat_ptrace_request(child, request, addr, data);
1106 }
1107
1108 return ret;
1109}
1110#endif /* CONFIG_IA32_EMULATION */
1111
1112#ifdef CONFIG_X86_X32_ABI
1113static long x32_arch_ptrace(struct task_struct *child,
1114 compat_long_t request, compat_ulong_t caddr,
1115 compat_ulong_t cdata)
1116{
1117 unsigned long addr = caddr;
1118 unsigned long data = cdata;
1119 void __user *datap = compat_ptr(data);
1120 int ret;
1121
1122 switch (request) {
1123 /* Read 32bits at location addr in the USER area. Only allow
1124 to return the lower 32bits of segment and debug registers. */
1125 case PTRACE_PEEKUSR: {
1126 u32 tmp;
1127
1128 ret = -EIO;
1129 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1130 addr < offsetof(struct user_regs_struct, cs))
1131 break;
1132
1133 tmp = 0; /* Default return condition */
1134 if (addr < sizeof(struct user_regs_struct))
1135 tmp = getreg(child, addr);
1136 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1137 addr <= offsetof(struct user, u_debugreg[7])) {
1138 addr -= offsetof(struct user, u_debugreg[0]);
1139 tmp = ptrace_get_debugreg(child, addr / sizeof(data));
1140 }
1141 ret = put_user(tmp, (__u32 __user *)datap);
1142 break;
1143 }
1144
1145 /* Write the word at location addr in the USER area. Only allow
1146 to update segment and debug registers with the upper 32bits
1147 zero-extended. */
1148 case PTRACE_POKEUSR:
1149 ret = -EIO;
1150 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1151 addr < offsetof(struct user_regs_struct, cs))
1152 break;
1153
1154 if (addr < sizeof(struct user_regs_struct))
1155 ret = putreg(child, addr, data);
1156 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1157 addr <= offsetof(struct user, u_debugreg[7])) {
1158 addr -= offsetof(struct user, u_debugreg[0]);
1159 ret = ptrace_set_debugreg(child,
1160 addr / sizeof(data), data);
1161 }
1162 break;
1163
1164 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1165 return copy_regset_to_user(child,
1166 &user_x86_64_view,
1167 REGSET_GENERAL,
1168 0, sizeof(struct user_regs_struct),
1169 datap);
1170
1171 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1172 return copy_regset_from_user(child,
1173 &user_x86_64_view,
1174 REGSET_GENERAL,
1175 0, sizeof(struct user_regs_struct),
1176 datap);
1177
1178 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1179 return copy_regset_to_user(child,
1180 &user_x86_64_view,
1181 REGSET_FP,
1182 0, sizeof(struct user_i387_struct),
1183 datap);
1184
1185 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1186 return copy_regset_from_user(child,
1187 &user_x86_64_view,
1188 REGSET_FP,
1189 0, sizeof(struct user_i387_struct),
1190 datap);
1191
1192 default:
1193 return compat_ptrace_request(child, request, addr, data);
1194 }
1195
1196 return ret;
1197}
1198#endif
1199
1200#ifdef CONFIG_COMPAT
1201long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1202 compat_ulong_t caddr, compat_ulong_t cdata)
1203{
1204#ifdef CONFIG_X86_X32_ABI
1205 if (!in_ia32_syscall())
1206 return x32_arch_ptrace(child, request, caddr, cdata);
1207#endif
1208#ifdef CONFIG_IA32_EMULATION
1209 return ia32_arch_ptrace(child, request, caddr, cdata);
1210#else
1211 return 0;
1212#endif
1213}
1214#endif /* CONFIG_COMPAT */
1215
1216#ifdef CONFIG_X86_64
1217
1218static struct user_regset x86_64_regsets[] __ro_after_init = {
1219 [REGSET_GENERAL] = {
1220 .core_note_type = NT_PRSTATUS,
1221 .n = sizeof(struct user_regs_struct) / sizeof(long),
1222 .size = sizeof(long), .align = sizeof(long),
1223 .regset_get = genregs_get, .set = genregs_set
1224 },
1225 [REGSET_FP] = {
1226 .core_note_type = NT_PRFPREG,
1227 .n = sizeof(struct user_i387_struct) / sizeof(long),
1228 .size = sizeof(long), .align = sizeof(long),
1229 .active = regset_xregset_fpregs_active, .regset_get = xfpregs_get, .set = xfpregs_set
1230 },
1231 [REGSET_XSTATE] = {
1232 .core_note_type = NT_X86_XSTATE,
1233 .size = sizeof(u64), .align = sizeof(u64),
1234 .active = xstateregs_active, .regset_get = xstateregs_get,
1235 .set = xstateregs_set
1236 },
1237 [REGSET_IOPERM64] = {
1238 .core_note_type = NT_386_IOPERM,
1239 .n = IO_BITMAP_LONGS,
1240 .size = sizeof(long), .align = sizeof(long),
1241 .active = ioperm_active, .regset_get = ioperm_get
1242 },
1243};
1244
1245static const struct user_regset_view user_x86_64_view = {
1246 .name = "x86_64", .e_machine = EM_X86_64,
1247 .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1248};
1249
1250#else /* CONFIG_X86_32 */
1251
1252#define user_regs_struct32 user_regs_struct
1253#define genregs32_get genregs_get
1254#define genregs32_set genregs_set
1255
1256#endif /* CONFIG_X86_64 */
1257
1258#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1259static struct user_regset x86_32_regsets[] __ro_after_init = {
1260 [REGSET_GENERAL] = {
1261 .core_note_type = NT_PRSTATUS,
1262 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1263 .size = sizeof(u32), .align = sizeof(u32),
1264 .regset_get = genregs32_get, .set = genregs32_set
1265 },
1266 [REGSET_FP] = {
1267 .core_note_type = NT_PRFPREG,
1268 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1269 .size = sizeof(u32), .align = sizeof(u32),
1270 .active = regset_fpregs_active, .regset_get = fpregs_get, .set = fpregs_set
1271 },
1272 [REGSET_XFP] = {
1273 .core_note_type = NT_PRXFPREG,
1274 .n = sizeof(struct user32_fxsr_struct) / sizeof(u32),
1275 .size = sizeof(u32), .align = sizeof(u32),
1276 .active = regset_xregset_fpregs_active, .regset_get = xfpregs_get, .set = xfpregs_set
1277 },
1278 [REGSET_XSTATE] = {
1279 .core_note_type = NT_X86_XSTATE,
1280 .size = sizeof(u64), .align = sizeof(u64),
1281 .active = xstateregs_active, .regset_get = xstateregs_get,
1282 .set = xstateregs_set
1283 },
1284 [REGSET_TLS] = {
1285 .core_note_type = NT_386_TLS,
1286 .n = GDT_ENTRY_TLS_ENTRIES, .bias = GDT_ENTRY_TLS_MIN,
1287 .size = sizeof(struct user_desc),
1288 .align = sizeof(struct user_desc),
1289 .active = regset_tls_active,
1290 .regset_get = regset_tls_get, .set = regset_tls_set
1291 },
1292 [REGSET_IOPERM32] = {
1293 .core_note_type = NT_386_IOPERM,
1294 .n = IO_BITMAP_BYTES / sizeof(u32),
1295 .size = sizeof(u32), .align = sizeof(u32),
1296 .active = ioperm_active, .regset_get = ioperm_get
1297 },
1298};
1299
1300static const struct user_regset_view user_x86_32_view = {
1301 .name = "i386", .e_machine = EM_386,
1302 .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1303};
1304#endif
1305
1306/*
1307 * This represents bytes 464..511 in the memory layout exported through
1308 * the REGSET_XSTATE interface.
1309 */
1310u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1311
1312void __init update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1313{
1314#ifdef CONFIG_X86_64
1315 x86_64_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1316#endif
1317#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1318 x86_32_regsets[REGSET_XSTATE].n = size / sizeof(u64);
1319#endif
1320 xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1321}
1322
1323/*
1324 * This is used by the core dump code to decide which regset to dump. The
1325 * core dump code writes out the resulting .e_machine and the corresponding
1326 * regsets. This is suboptimal if the task is messing around with its CS.L
1327 * field, but at worst the core dump will end up missing some information.
1328 *
1329 * Unfortunately, it is also used by the broken PTRACE_GETREGSET and
1330 * PTRACE_SETREGSET APIs. These APIs look at the .regsets field but have
1331 * no way to make sure that the e_machine they use matches the caller's
1332 * expectations. The result is that the data format returned by
1333 * PTRACE_GETREGSET depends on the returned CS field (and even the offset
1334 * of the returned CS field depends on its value!) and the data format
1335 * accepted by PTRACE_SETREGSET is determined by the old CS value. The
1336 * upshot is that it is basically impossible to use these APIs correctly.
1337 *
1338 * The best way to fix it in the long run would probably be to add new
1339 * improved ptrace() APIs to read and write registers reliably, possibly by
1340 * allowing userspace to select the ELF e_machine variant that they expect.
1341 */
1342const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1343{
1344#ifdef CONFIG_IA32_EMULATION
1345 if (!user_64bit_mode(task_pt_regs(task)))
1346#endif
1347#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1348 return &user_x86_32_view;
1349#endif
1350#ifdef CONFIG_X86_64
1351 return &user_x86_64_view;
1352#endif
1353}
1354
1355void send_sigtrap(struct pt_regs *regs, int error_code, int si_code)
1356{
1357 struct task_struct *tsk = current;
1358
1359 tsk->thread.trap_nr = X86_TRAP_DB;
1360 tsk->thread.error_code = error_code;
1361
1362 /* Send us the fake SIGTRAP */
1363 force_sig_fault(SIGTRAP, si_code,
1364 user_mode(regs) ? (void __user *)regs->ip : NULL);
1365}
1366
1367void user_single_step_report(struct pt_regs *regs)
1368{
1369 send_sigtrap(regs, 0, TRAP_BRKPT);
1370}