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/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#include <linux/rcupdate.h>
25#include <linux/export.h>
26#include <linux/context_tracking.h>
27#include <linux/nospec.h>
28
29#include <linux/uaccess.h>
30#include <asm/processor.h>
31#include <asm/fpu/signal.h>
32#include <asm/fpu/regset.h>
33#include <asm/fpu/xstate.h>
34#include <asm/debugreg.h>
35#include <asm/ldt.h>
36#include <asm/desc.h>
37#include <asm/prctl.h>
38#include <asm/proto.h>
39#include <asm/hw_breakpoint.h>
40#include <asm/traps.h>
41#include <asm/syscall.h>
42#include <asm/fsgsbase.h>
43#include <asm/io_bitmap.h>
44
45#include "tls.h"
46
47enum x86_regset_32 {
48 REGSET32_GENERAL,
49 REGSET32_FP,
50 REGSET32_XFP,
51 REGSET32_XSTATE,
52 REGSET32_TLS,
53 REGSET32_IOPERM,
54};
55
56enum x86_regset_64 {
57 REGSET64_GENERAL,
58 REGSET64_FP,
59 REGSET64_IOPERM,
60 REGSET64_XSTATE,
61 REGSET64_SSP,
62};
63
64#define REGSET_GENERAL \
65({ \
66 BUILD_BUG_ON((int)REGSET32_GENERAL != (int)REGSET64_GENERAL); \
67 REGSET32_GENERAL; \
68})
69
70#define REGSET_FP \
71({ \
72 BUILD_BUG_ON((int)REGSET32_FP != (int)REGSET64_FP); \
73 REGSET32_FP; \
74})
75
76
77struct pt_regs_offset {
78 const char *name;
79 int offset;
80};
81
82#define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
83#define REG_OFFSET_END {.name = NULL, .offset = 0}
84
85static const struct pt_regs_offset regoffset_table[] = {
86#ifdef CONFIG_X86_64
87 REG_OFFSET_NAME(r15),
88 REG_OFFSET_NAME(r14),
89 REG_OFFSET_NAME(r13),
90 REG_OFFSET_NAME(r12),
91 REG_OFFSET_NAME(r11),
92 REG_OFFSET_NAME(r10),
93 REG_OFFSET_NAME(r9),
94 REG_OFFSET_NAME(r8),
95#endif
96 REG_OFFSET_NAME(bx),
97 REG_OFFSET_NAME(cx),
98 REG_OFFSET_NAME(dx),
99 REG_OFFSET_NAME(si),
100 REG_OFFSET_NAME(di),
101 REG_OFFSET_NAME(bp),
102 REG_OFFSET_NAME(ax),
103#ifdef CONFIG_X86_32
104 REG_OFFSET_NAME(ds),
105 REG_OFFSET_NAME(es),
106 REG_OFFSET_NAME(fs),
107 REG_OFFSET_NAME(gs),
108#endif
109 REG_OFFSET_NAME(orig_ax),
110 REG_OFFSET_NAME(ip),
111 REG_OFFSET_NAME(cs),
112 REG_OFFSET_NAME(flags),
113 REG_OFFSET_NAME(sp),
114 REG_OFFSET_NAME(ss),
115 REG_OFFSET_END,
116};
117
118/**
119 * regs_query_register_offset() - query register offset from its name
120 * @name: the name of a register
121 *
122 * regs_query_register_offset() returns the offset of a register in struct
123 * pt_regs from its name. If the name is invalid, this returns -EINVAL;
124 */
125int regs_query_register_offset(const char *name)
126{
127 const struct pt_regs_offset *roff;
128 for (roff = regoffset_table; roff->name != NULL; roff++)
129 if (!strcmp(roff->name, name))
130 return roff->offset;
131 return -EINVAL;
132}
133
134/**
135 * regs_query_register_name() - query register name from its offset
136 * @offset: the offset of a register in struct pt_regs.
137 *
138 * regs_query_register_name() returns the name of a register from its
139 * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
140 */
141const char *regs_query_register_name(unsigned int offset)
142{
143 const struct pt_regs_offset *roff;
144 for (roff = regoffset_table; roff->name != NULL; roff++)
145 if (roff->offset == offset)
146 return roff->name;
147 return NULL;
148}
149
150/*
151 * does not yet catch signals sent when the child dies.
152 * in exit.c or in signal.c.
153 */
154
155/*
156 * Determines which flags the user has access to [1 = access, 0 = no access].
157 */
158#define FLAG_MASK_32 ((unsigned long) \
159 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
160 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
161 X86_EFLAGS_SF | X86_EFLAGS_TF | \
162 X86_EFLAGS_DF | X86_EFLAGS_OF | \
163 X86_EFLAGS_RF | X86_EFLAGS_AC))
164
165/*
166 * Determines whether a value may be installed in a segment register.
167 */
168static inline bool invalid_selector(u16 value)
169{
170 return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
171}
172
173#ifdef CONFIG_X86_32
174
175#define FLAG_MASK FLAG_MASK_32
176
177static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
178{
179 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
180 return &regs->bx + (regno >> 2);
181}
182
183static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
184{
185 /*
186 * Returning the value truncates it to 16 bits.
187 */
188 unsigned int retval;
189 if (offset != offsetof(struct user_regs_struct, gs))
190 retval = *pt_regs_access(task_pt_regs(task), offset);
191 else {
192 if (task == current)
193 savesegment(gs, retval);
194 else
195 retval = task->thread.gs;
196 }
197 return retval;
198}
199
200static int set_segment_reg(struct task_struct *task,
201 unsigned long offset, u16 value)
202{
203 if (WARN_ON_ONCE(task == current))
204 return -EIO;
205
206 /*
207 * The value argument was already truncated to 16 bits.
208 */
209 if (invalid_selector(value))
210 return -EIO;
211
212 /*
213 * For %cs and %ss we cannot permit a null selector.
214 * We can permit a bogus selector as long as it has USER_RPL.
215 * Null selectors are fine for other segment registers, but
216 * we will never get back to user mode with invalid %cs or %ss
217 * and will take the trap in iret instead. Much code relies
218 * on user_mode() to distinguish a user trap frame (which can
219 * safely use invalid selectors) from a kernel trap frame.
220 */
221 switch (offset) {
222 case offsetof(struct user_regs_struct, cs):
223 case offsetof(struct user_regs_struct, ss):
224 if (unlikely(value == 0))
225 return -EIO;
226 fallthrough;
227
228 default:
229 *pt_regs_access(task_pt_regs(task), offset) = value;
230 break;
231
232 case offsetof(struct user_regs_struct, gs):
233 task->thread.gs = value;
234 }
235
236 return 0;
237}
238
239#else /* CONFIG_X86_64 */
240
241#define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
242
243static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
244{
245 BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
246 return &regs->r15 + (offset / sizeof(regs->r15));
247}
248
249static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
250{
251 /*
252 * Returning the value truncates it to 16 bits.
253 */
254 unsigned int seg;
255
256 switch (offset) {
257 case offsetof(struct user_regs_struct, fs):
258 if (task == current) {
259 /* Older gas can't assemble movq %?s,%r?? */
260 asm("movl %%fs,%0" : "=r" (seg));
261 return seg;
262 }
263 return task->thread.fsindex;
264 case offsetof(struct user_regs_struct, gs):
265 if (task == current) {
266 asm("movl %%gs,%0" : "=r" (seg));
267 return seg;
268 }
269 return task->thread.gsindex;
270 case offsetof(struct user_regs_struct, ds):
271 if (task == current) {
272 asm("movl %%ds,%0" : "=r" (seg));
273 return seg;
274 }
275 return task->thread.ds;
276 case offsetof(struct user_regs_struct, es):
277 if (task == current) {
278 asm("movl %%es,%0" : "=r" (seg));
279 return seg;
280 }
281 return task->thread.es;
282
283 case offsetof(struct user_regs_struct, cs):
284 case offsetof(struct user_regs_struct, ss):
285 break;
286 }
287 return *pt_regs_access(task_pt_regs(task), offset);
288}
289
290static int set_segment_reg(struct task_struct *task,
291 unsigned long offset, u16 value)
292{
293 if (WARN_ON_ONCE(task == current))
294 return -EIO;
295
296 /*
297 * The value argument was already truncated to 16 bits.
298 */
299 if (invalid_selector(value))
300 return -EIO;
301
302 /*
303 * Writes to FS and GS will change the stored selector. Whether
304 * this changes the segment base as well depends on whether
305 * FSGSBASE is enabled.
306 */
307
308 switch (offset) {
309 case offsetof(struct user_regs_struct,fs):
310 task->thread.fsindex = value;
311 break;
312 case offsetof(struct user_regs_struct,gs):
313 task->thread.gsindex = value;
314 break;
315 case offsetof(struct user_regs_struct,ds):
316 task->thread.ds = value;
317 break;
318 case offsetof(struct user_regs_struct,es):
319 task->thread.es = value;
320 break;
321
322 /*
323 * Can't actually change these in 64-bit mode.
324 */
325 case offsetof(struct user_regs_struct,cs):
326 if (unlikely(value == 0))
327 return -EIO;
328 task_pt_regs(task)->cs = value;
329 break;
330 case offsetof(struct user_regs_struct,ss):
331 if (unlikely(value == 0))
332 return -EIO;
333 task_pt_regs(task)->ss = value;
334 break;
335 }
336
337 return 0;
338}
339
340#endif /* CONFIG_X86_32 */
341
342static unsigned long get_flags(struct task_struct *task)
343{
344 unsigned long retval = task_pt_regs(task)->flags;
345
346 /*
347 * If the debugger set TF, hide it from the readout.
348 */
349 if (test_tsk_thread_flag(tsk: task, TIF_FORCED_TF))
350 retval &= ~X86_EFLAGS_TF;
351
352 return retval;
353}
354
355static int set_flags(struct task_struct *task, unsigned long value)
356{
357 struct pt_regs *regs = task_pt_regs(task);
358
359 /*
360 * If the user value contains TF, mark that
361 * it was not "us" (the debugger) that set it.
362 * If not, make sure it stays set if we had.
363 */
364 if (value & X86_EFLAGS_TF)
365 clear_tsk_thread_flag(tsk: task, TIF_FORCED_TF);
366 else if (test_tsk_thread_flag(tsk: task, TIF_FORCED_TF))
367 value |= X86_EFLAGS_TF;
368
369 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
370
371 return 0;
372}
373
374static int putreg(struct task_struct *child,
375 unsigned long offset, unsigned long value)
376{
377 switch (offset) {
378 case offsetof(struct user_regs_struct, cs):
379 case offsetof(struct user_regs_struct, ds):
380 case offsetof(struct user_regs_struct, es):
381 case offsetof(struct user_regs_struct, fs):
382 case offsetof(struct user_regs_struct, gs):
383 case offsetof(struct user_regs_struct, ss):
384 return set_segment_reg(task: child, offset, value);
385
386 case offsetof(struct user_regs_struct, flags):
387 return set_flags(task: child, value);
388
389#ifdef CONFIG_X86_64
390 case offsetof(struct user_regs_struct,fs_base):
391 if (value >= TASK_SIZE_MAX)
392 return -EIO;
393 x86_fsbase_write_task(task: child, fsbase: value);
394 return 0;
395 case offsetof(struct user_regs_struct,gs_base):
396 if (value >= TASK_SIZE_MAX)
397 return -EIO;
398 x86_gsbase_write_task(task: child, gsbase: value);
399 return 0;
400#endif
401 }
402
403 *pt_regs_access(task_pt_regs(child), offset) = value;
404 return 0;
405}
406
407static unsigned long getreg(struct task_struct *task, unsigned long offset)
408{
409 switch (offset) {
410 case offsetof(struct user_regs_struct, cs):
411 case offsetof(struct user_regs_struct, ds):
412 case offsetof(struct user_regs_struct, es):
413 case offsetof(struct user_regs_struct, fs):
414 case offsetof(struct user_regs_struct, gs):
415 case offsetof(struct user_regs_struct, ss):
416 return get_segment_reg(task, offset);
417
418 case offsetof(struct user_regs_struct, flags):
419 return get_flags(task);
420
421#ifdef CONFIG_X86_64
422 case offsetof(struct user_regs_struct, fs_base):
423 return x86_fsbase_read_task(task);
424 case offsetof(struct user_regs_struct, gs_base):
425 return x86_gsbase_read_task(task);
426#endif
427 }
428
429 return *pt_regs_access(task_pt_regs(task), offset);
430}
431
432static int genregs_get(struct task_struct *target,
433 const struct user_regset *regset,
434 struct membuf to)
435{
436 int reg;
437
438 for (reg = 0; to.left; reg++)
439 membuf_store(&to, getreg(target, reg * sizeof(unsigned long)));
440 return 0;
441}
442
443static int genregs_set(struct task_struct *target,
444 const struct user_regset *regset,
445 unsigned int pos, unsigned int count,
446 const void *kbuf, const void __user *ubuf)
447{
448 int ret = 0;
449 if (kbuf) {
450 const unsigned long *k = kbuf;
451 while (count >= sizeof(*k) && !ret) {
452 ret = putreg(child: target, offset: pos, value: *k++);
453 count -= sizeof(*k);
454 pos += sizeof(*k);
455 }
456 } else {
457 const unsigned long __user *u = ubuf;
458 while (count >= sizeof(*u) && !ret) {
459 unsigned long word;
460 ret = __get_user(word, u++);
461 if (ret)
462 break;
463 ret = putreg(child: target, offset: pos, value: word);
464 count -= sizeof(*u);
465 pos += sizeof(*u);
466 }
467 }
468 return ret;
469}
470
471static void ptrace_triggered(struct perf_event *bp,
472 struct perf_sample_data *data,
473 struct pt_regs *regs)
474{
475 int i;
476 struct thread_struct *thread = &(current->thread);
477
478 /*
479 * Store in the virtual DR6 register the fact that the breakpoint
480 * was hit so the thread's debugger will see it.
481 */
482 for (i = 0; i < HBP_NUM; i++) {
483 if (thread->ptrace_bps[i] == bp)
484 break;
485 }
486
487 thread->virtual_dr6 |= (DR_TRAP0 << i);
488}
489
490/*
491 * Walk through every ptrace breakpoints for this thread and
492 * build the dr7 value on top of their attributes.
493 *
494 */
495static unsigned long ptrace_get_dr7(struct perf_event *bp[])
496{
497 int i;
498 int dr7 = 0;
499 struct arch_hw_breakpoint *info;
500
501 for (i = 0; i < HBP_NUM; i++) {
502 if (bp[i] && !bp[i]->attr.disabled) {
503 info = counter_arch_bp(bp: bp[i]);
504 dr7 |= encode_dr7(drnum: i, len: info->len, type: info->type);
505 }
506 }
507
508 return dr7;
509}
510
511static int ptrace_fill_bp_fields(struct perf_event_attr *attr,
512 int len, int type, bool disabled)
513{
514 int err, bp_len, bp_type;
515
516 err = arch_bp_generic_fields(x86_len: len, x86_type: type, gen_len: &bp_len, gen_type: &bp_type);
517 if (!err) {
518 attr->bp_len = bp_len;
519 attr->bp_type = bp_type;
520 attr->disabled = disabled;
521 }
522
523 return err;
524}
525
526static struct perf_event *
527ptrace_register_breakpoint(struct task_struct *tsk, int len, int type,
528 unsigned long addr, bool disabled)
529{
530 struct perf_event_attr attr;
531 int err;
532
533 ptrace_breakpoint_init(attr: &attr);
534 attr.bp_addr = addr;
535
536 err = ptrace_fill_bp_fields(attr: &attr, len, type, disabled);
537 if (err)
538 return ERR_PTR(error: err);
539
540 return register_user_hw_breakpoint(attr: &attr, triggered: ptrace_triggered,
541 NULL, tsk);
542}
543
544static int ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
545 int disabled)
546{
547 struct perf_event_attr attr = bp->attr;
548 int err;
549
550 err = ptrace_fill_bp_fields(attr: &attr, len, type, disabled);
551 if (err)
552 return err;
553
554 return modify_user_hw_breakpoint(bp, attr: &attr);
555}
556
557/*
558 * Handle ptrace writes to debug register 7.
559 */
560static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
561{
562 struct thread_struct *thread = &tsk->thread;
563 unsigned long old_dr7;
564 bool second_pass = false;
565 int i, rc, ret = 0;
566
567 data &= ~DR_CONTROL_RESERVED;
568 old_dr7 = ptrace_get_dr7(bp: thread->ptrace_bps);
569
570restore:
571 rc = 0;
572 for (i = 0; i < HBP_NUM; i++) {
573 unsigned len, type;
574 bool disabled = !decode_dr7(dr7: data, bpnum: i, len: &len, type: &type);
575 struct perf_event *bp = thread->ptrace_bps[i];
576
577 if (!bp) {
578 if (disabled)
579 continue;
580
581 bp = ptrace_register_breakpoint(tsk,
582 len, type, addr: 0, disabled);
583 if (IS_ERR(ptr: bp)) {
584 rc = PTR_ERR(ptr: bp);
585 break;
586 }
587
588 thread->ptrace_bps[i] = bp;
589 continue;
590 }
591
592 rc = ptrace_modify_breakpoint(bp, len, type, disabled);
593 if (rc)
594 break;
595 }
596
597 /* Restore if the first pass failed, second_pass shouldn't fail. */
598 if (rc && !WARN_ON(second_pass)) {
599 ret = rc;
600 data = old_dr7;
601 second_pass = true;
602 goto restore;
603 }
604
605 return ret;
606}
607
608/*
609 * Handle PTRACE_PEEKUSR calls for the debug register area.
610 */
611static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
612{
613 struct thread_struct *thread = &tsk->thread;
614 unsigned long val = 0;
615
616 if (n < HBP_NUM) {
617 int index = array_index_nospec(n, HBP_NUM);
618 struct perf_event *bp = thread->ptrace_bps[index];
619
620 if (bp)
621 val = bp->hw.info.address;
622 } else if (n == 6) {
623 val = thread->virtual_dr6 ^ DR6_RESERVED; /* Flip back to arch polarity */
624 } else if (n == 7) {
625 val = thread->ptrace_dr7;
626 }
627 return val;
628}
629
630static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
631 unsigned long addr)
632{
633 struct thread_struct *t = &tsk->thread;
634 struct perf_event *bp = t->ptrace_bps[nr];
635 int err = 0;
636
637 if (!bp) {
638 /*
639 * Put stub len and type to create an inactive but correct bp.
640 *
641 * CHECKME: the previous code returned -EIO if the addr wasn't
642 * a valid task virtual addr. The new one will return -EINVAL in
643 * this case.
644 * -EINVAL may be what we want for in-kernel breakpoints users,
645 * but -EIO looks better for ptrace, since we refuse a register
646 * writing for the user. And anyway this is the previous
647 * behaviour.
648 */
649 bp = ptrace_register_breakpoint(tsk,
650 X86_BREAKPOINT_LEN_1, X86_BREAKPOINT_WRITE,
651 addr, disabled: true);
652 if (IS_ERR(ptr: bp))
653 err = PTR_ERR(ptr: bp);
654 else
655 t->ptrace_bps[nr] = bp;
656 } else {
657 struct perf_event_attr attr = bp->attr;
658
659 attr.bp_addr = addr;
660 err = modify_user_hw_breakpoint(bp, attr: &attr);
661 }
662
663 return err;
664}
665
666/*
667 * Handle PTRACE_POKEUSR calls for the debug register area.
668 */
669static int ptrace_set_debugreg(struct task_struct *tsk, int n,
670 unsigned long val)
671{
672 struct thread_struct *thread = &tsk->thread;
673 /* There are no DR4 or DR5 registers */
674 int rc = -EIO;
675
676 if (n < HBP_NUM) {
677 rc = ptrace_set_breakpoint_addr(tsk, nr: n, addr: val);
678 } else if (n == 6) {
679 thread->virtual_dr6 = val ^ DR6_RESERVED; /* Flip to positive polarity */
680 rc = 0;
681 } else if (n == 7) {
682 rc = ptrace_write_dr7(tsk, data: val);
683 if (!rc)
684 thread->ptrace_dr7 = val;
685 }
686 return rc;
687}
688
689/*
690 * These access the current or another (stopped) task's io permission
691 * bitmap for debugging or core dump.
692 */
693static int ioperm_active(struct task_struct *target,
694 const struct user_regset *regset)
695{
696 struct io_bitmap *iobm = target->thread.io_bitmap;
697
698 return iobm ? DIV_ROUND_UP(iobm->max, regset->size) : 0;
699}
700
701static int ioperm_get(struct task_struct *target,
702 const struct user_regset *regset,
703 struct membuf to)
704{
705 struct io_bitmap *iobm = target->thread.io_bitmap;
706
707 if (!iobm)
708 return -ENXIO;
709
710 return membuf_write(s: &to, v: iobm->bitmap, IO_BITMAP_BYTES);
711}
712
713/*
714 * Called by kernel/ptrace.c when detaching..
715 *
716 * Make sure the single step bit is not set.
717 */
718void ptrace_disable(struct task_struct *child)
719{
720 user_disable_single_step(child);
721}
722
723#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
724static const struct user_regset_view user_x86_32_view; /* Initialized below. */
725#endif
726#ifdef CONFIG_X86_64
727static const struct user_regset_view user_x86_64_view; /* Initialized below. */
728#endif
729
730long arch_ptrace(struct task_struct *child, long request,
731 unsigned long addr, unsigned long data)
732{
733 int ret;
734 unsigned long __user *datap = (unsigned long __user *)data;
735
736#ifdef CONFIG_X86_64
737 /* This is native 64-bit ptrace() */
738 const struct user_regset_view *regset_view = &user_x86_64_view;
739#else
740 /* This is native 32-bit ptrace() */
741 const struct user_regset_view *regset_view = &user_x86_32_view;
742#endif
743
744 switch (request) {
745 /* read the word at location addr in the USER area. */
746 case PTRACE_PEEKUSR: {
747 unsigned long tmp;
748
749 ret = -EIO;
750 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
751 break;
752
753 tmp = 0; /* Default return condition */
754 if (addr < sizeof(struct user_regs_struct))
755 tmp = getreg(task: child, offset: addr);
756 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
757 addr <= offsetof(struct user, u_debugreg[7])) {
758 addr -= offsetof(struct user, u_debugreg[0]);
759 tmp = ptrace_get_debugreg(tsk: child, n: addr / sizeof(data));
760 }
761 ret = put_user(tmp, datap);
762 break;
763 }
764
765 case PTRACE_POKEUSR: /* write the word at location addr in the USER area */
766 ret = -EIO;
767 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
768 break;
769
770 if (addr < sizeof(struct user_regs_struct))
771 ret = putreg(child, offset: addr, value: data);
772 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
773 addr <= offsetof(struct user, u_debugreg[7])) {
774 addr -= offsetof(struct user, u_debugreg[0]);
775 ret = ptrace_set_debugreg(tsk: child,
776 n: addr / sizeof(data), val: data);
777 }
778 break;
779
780 case PTRACE_GETREGS: /* Get all gp regs from the child. */
781 return copy_regset_to_user(target: child,
782 view: regset_view,
783 REGSET_GENERAL,
784 offset: 0, size: sizeof(struct user_regs_struct),
785 data: datap);
786
787 case PTRACE_SETREGS: /* Set all gp regs in the child. */
788 return copy_regset_from_user(target: child,
789 view: regset_view,
790 REGSET_GENERAL,
791 offset: 0, size: sizeof(struct user_regs_struct),
792 data: datap);
793
794 case PTRACE_GETFPREGS: /* Get the child FPU state. */
795 return copy_regset_to_user(target: child,
796 view: regset_view,
797 REGSET_FP,
798 offset: 0, size: sizeof(struct user_i387_struct),
799 data: datap);
800
801 case PTRACE_SETFPREGS: /* Set the child FPU state. */
802 return copy_regset_from_user(target: child,
803 view: regset_view,
804 REGSET_FP,
805 offset: 0, size: sizeof(struct user_i387_struct),
806 data: datap);
807
808#ifdef CONFIG_X86_32
809 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
810 return copy_regset_to_user(child, &user_x86_32_view,
811 REGSET32_XFP,
812 0, sizeof(struct user_fxsr_struct),
813 datap) ? -EIO : 0;
814
815 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
816 return copy_regset_from_user(child, &user_x86_32_view,
817 REGSET32_XFP,
818 0, sizeof(struct user_fxsr_struct),
819 datap) ? -EIO : 0;
820#endif
821
822#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
823 case PTRACE_GET_THREAD_AREA:
824 if ((int) addr < 0)
825 return -EIO;
826 ret = do_get_thread_area(p: child, idx: addr,
827 info: (struct user_desc __user *)data);
828 break;
829
830 case PTRACE_SET_THREAD_AREA:
831 if ((int) addr < 0)
832 return -EIO;
833 ret = do_set_thread_area(p: child, idx: addr,
834 info: (struct user_desc __user *)data, can_allocate: 0);
835 break;
836#endif
837
838#ifdef CONFIG_X86_64
839 /* normal 64bit interface to access TLS data.
840 Works just like arch_prctl, except that the arguments
841 are reversed. */
842 case PTRACE_ARCH_PRCTL:
843 ret = do_arch_prctl_64(task: child, option: data, arg2: addr);
844 break;
845#endif
846
847 default:
848 ret = ptrace_request(child, request, addr, data);
849 break;
850 }
851
852 return ret;
853}
854
855#ifdef CONFIG_IA32_EMULATION
856
857#include <linux/compat.h>
858#include <linux/syscalls.h>
859#include <asm/ia32.h>
860#include <asm/user32.h>
861
862#define R32(l,q) \
863 case offsetof(struct user32, regs.l): \
864 regs->q = value; break
865
866#define SEG32(rs) \
867 case offsetof(struct user32, regs.rs): \
868 return set_segment_reg(child, \
869 offsetof(struct user_regs_struct, rs), \
870 value); \
871 break
872
873static int putreg32(struct task_struct *child, unsigned regno, u32 value)
874{
875 struct pt_regs *regs = task_pt_regs(child);
876 int ret;
877
878 switch (regno) {
879
880 SEG32(cs);
881 SEG32(ds);
882 SEG32(es);
883
884 /*
885 * A 32-bit ptracer on a 64-bit kernel expects that writing
886 * FS or GS will also update the base. This is needed for
887 * operations like PTRACE_SETREGS to fully restore a saved
888 * CPU state.
889 */
890
891 case offsetof(struct user32, regs.fs):
892 ret = set_segment_reg(task: child,
893 offsetof(struct user_regs_struct, fs),
894 value);
895 if (ret == 0)
896 child->thread.fsbase =
897 x86_fsgsbase_read_task(task: child, selector: value);
898 return ret;
899
900 case offsetof(struct user32, regs.gs):
901 ret = set_segment_reg(task: child,
902 offsetof(struct user_regs_struct, gs),
903 value);
904 if (ret == 0)
905 child->thread.gsbase =
906 x86_fsgsbase_read_task(task: child, selector: value);
907 return ret;
908
909 SEG32(ss);
910
911 R32(ebx, bx);
912 R32(ecx, cx);
913 R32(edx, dx);
914 R32(edi, di);
915 R32(esi, si);
916 R32(ebp, bp);
917 R32(eax, ax);
918 R32(eip, ip);
919 R32(esp, sp);
920
921 case offsetof(struct user32, regs.orig_eax):
922 /*
923 * Warning: bizarre corner case fixup here. A 32-bit
924 * debugger setting orig_eax to -1 wants to disable
925 * syscall restart. Make sure that the syscall
926 * restart code sign-extends orig_ax. Also make sure
927 * we interpret the -ERESTART* codes correctly if
928 * loaded into regs->ax in case the task is not
929 * actually still sitting at the exit from a 32-bit
930 * syscall with TS_COMPAT still set.
931 */
932 regs->orig_ax = value;
933 if (syscall_get_nr(task: child, regs) != -1)
934 child->thread_info.status |= TS_I386_REGS_POKED;
935 break;
936
937 case offsetof(struct user32, regs.eflags):
938 return set_flags(task: child, value);
939
940 case offsetof(struct user32, u_debugreg[0]) ...
941 offsetof(struct user32, u_debugreg[7]):
942 regno -= offsetof(struct user32, u_debugreg[0]);
943 return ptrace_set_debugreg(tsk: child, n: regno / 4, val: value);
944
945 default:
946 if (regno > sizeof(struct user32) || (regno & 3))
947 return -EIO;
948
949 /*
950 * Other dummy fields in the virtual user structure
951 * are ignored
952 */
953 break;
954 }
955 return 0;
956}
957
958#undef R32
959#undef SEG32
960
961#define R32(l,q) \
962 case offsetof(struct user32, regs.l): \
963 *val = regs->q; break
964
965#define SEG32(rs) \
966 case offsetof(struct user32, regs.rs): \
967 *val = get_segment_reg(child, \
968 offsetof(struct user_regs_struct, rs)); \
969 break
970
971static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
972{
973 struct pt_regs *regs = task_pt_regs(child);
974
975 switch (regno) {
976
977 SEG32(ds);
978 SEG32(es);
979 SEG32(fs);
980 SEG32(gs);
981
982 R32(cs, cs);
983 R32(ss, ss);
984 R32(ebx, bx);
985 R32(ecx, cx);
986 R32(edx, dx);
987 R32(edi, di);
988 R32(esi, si);
989 R32(ebp, bp);
990 R32(eax, ax);
991 R32(orig_eax, orig_ax);
992 R32(eip, ip);
993 R32(esp, sp);
994
995 case offsetof(struct user32, regs.eflags):
996 *val = get_flags(task: child);
997 break;
998
999 case offsetof(struct user32, u_debugreg[0]) ...
1000 offsetof(struct user32, u_debugreg[7]):
1001 regno -= offsetof(struct user32, u_debugreg[0]);
1002 *val = ptrace_get_debugreg(tsk: child, n: regno / 4);
1003 break;
1004
1005 default:
1006 if (regno > sizeof(struct user32) || (regno & 3))
1007 return -EIO;
1008
1009 /*
1010 * Other dummy fields in the virtual user structure
1011 * are ignored
1012 */
1013 *val = 0;
1014 break;
1015 }
1016 return 0;
1017}
1018
1019#undef R32
1020#undef SEG32
1021
1022static int genregs32_get(struct task_struct *target,
1023 const struct user_regset *regset,
1024 struct membuf to)
1025{
1026 int reg;
1027
1028 for (reg = 0; to.left; reg++) {
1029 u32 val;
1030 getreg32(child: target, regno: reg * 4, val: &val);
1031 membuf_store(&to, val);
1032 }
1033 return 0;
1034}
1035
1036static int genregs32_set(struct task_struct *target,
1037 const struct user_regset *regset,
1038 unsigned int pos, unsigned int count,
1039 const void *kbuf, const void __user *ubuf)
1040{
1041 int ret = 0;
1042 if (kbuf) {
1043 const compat_ulong_t *k = kbuf;
1044 while (count >= sizeof(*k) && !ret) {
1045 ret = putreg32(child: target, regno: pos, value: *k++);
1046 count -= sizeof(*k);
1047 pos += sizeof(*k);
1048 }
1049 } else {
1050 const compat_ulong_t __user *u = ubuf;
1051 while (count >= sizeof(*u) && !ret) {
1052 compat_ulong_t word;
1053 ret = __get_user(word, u++);
1054 if (ret)
1055 break;
1056 ret = putreg32(child: target, regno: pos, value: word);
1057 count -= sizeof(*u);
1058 pos += sizeof(*u);
1059 }
1060 }
1061 return ret;
1062}
1063
1064static long ia32_arch_ptrace(struct task_struct *child, compat_long_t request,
1065 compat_ulong_t caddr, compat_ulong_t cdata)
1066{
1067 unsigned long addr = caddr;
1068 unsigned long data = cdata;
1069 void __user *datap = compat_ptr(uptr: data);
1070 int ret;
1071 __u32 val;
1072
1073 switch (request) {
1074 case PTRACE_PEEKUSR:
1075 ret = getreg32(child, regno: addr, val: &val);
1076 if (ret == 0)
1077 ret = put_user(val, (__u32 __user *)datap);
1078 break;
1079
1080 case PTRACE_POKEUSR:
1081 ret = putreg32(child, regno: addr, value: data);
1082 break;
1083
1084 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1085 return copy_regset_to_user(target: child, view: &user_x86_32_view,
1086 REGSET_GENERAL,
1087 offset: 0, size: sizeof(struct user_regs_struct32),
1088 data: datap);
1089
1090 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1091 return copy_regset_from_user(target: child, view: &user_x86_32_view,
1092 REGSET_GENERAL, offset: 0,
1093 size: sizeof(struct user_regs_struct32),
1094 data: datap);
1095
1096 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1097 return copy_regset_to_user(target: child, view: &user_x86_32_view,
1098 REGSET_FP, offset: 0,
1099 size: sizeof(struct user_i387_ia32_struct),
1100 data: datap);
1101
1102 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1103 return copy_regset_from_user(
1104 target: child, view: &user_x86_32_view, REGSET_FP,
1105 offset: 0, size: sizeof(struct user_i387_ia32_struct), data: datap);
1106
1107 case PTRACE_GETFPXREGS: /* Get the child extended FPU state. */
1108 return copy_regset_to_user(target: child, view: &user_x86_32_view,
1109 setno: REGSET32_XFP, offset: 0,
1110 size: sizeof(struct user32_fxsr_struct),
1111 data: datap);
1112
1113 case PTRACE_SETFPXREGS: /* Set the child extended FPU state. */
1114 return copy_regset_from_user(target: child, view: &user_x86_32_view,
1115 setno: REGSET32_XFP, offset: 0,
1116 size: sizeof(struct user32_fxsr_struct),
1117 data: datap);
1118
1119 case PTRACE_GET_THREAD_AREA:
1120 case PTRACE_SET_THREAD_AREA:
1121 return arch_ptrace(child, request, addr, data);
1122
1123 default:
1124 return compat_ptrace_request(child, request, addr, data);
1125 }
1126
1127 return ret;
1128}
1129#endif /* CONFIG_IA32_EMULATION */
1130
1131#ifdef CONFIG_X86_X32_ABI
1132static long x32_arch_ptrace(struct task_struct *child,
1133 compat_long_t request, compat_ulong_t caddr,
1134 compat_ulong_t cdata)
1135{
1136 unsigned long addr = caddr;
1137 unsigned long data = cdata;
1138 void __user *datap = compat_ptr(uptr: data);
1139 int ret;
1140
1141 switch (request) {
1142 /* Read 32bits at location addr in the USER area. Only allow
1143 to return the lower 32bits of segment and debug registers. */
1144 case PTRACE_PEEKUSR: {
1145 u32 tmp;
1146
1147 ret = -EIO;
1148 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1149 addr < offsetof(struct user_regs_struct, cs))
1150 break;
1151
1152 tmp = 0; /* Default return condition */
1153 if (addr < sizeof(struct user_regs_struct))
1154 tmp = getreg(task: child, offset: addr);
1155 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1156 addr <= offsetof(struct user, u_debugreg[7])) {
1157 addr -= offsetof(struct user, u_debugreg[0]);
1158 tmp = ptrace_get_debugreg(tsk: child, n: addr / sizeof(data));
1159 }
1160 ret = put_user(tmp, (__u32 __user *)datap);
1161 break;
1162 }
1163
1164 /* Write the word at location addr in the USER area. Only allow
1165 to update segment and debug registers with the upper 32bits
1166 zero-extended. */
1167 case PTRACE_POKEUSR:
1168 ret = -EIO;
1169 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user) ||
1170 addr < offsetof(struct user_regs_struct, cs))
1171 break;
1172
1173 if (addr < sizeof(struct user_regs_struct))
1174 ret = putreg(child, offset: addr, value: data);
1175 else if (addr >= offsetof(struct user, u_debugreg[0]) &&
1176 addr <= offsetof(struct user, u_debugreg[7])) {
1177 addr -= offsetof(struct user, u_debugreg[0]);
1178 ret = ptrace_set_debugreg(tsk: child,
1179 n: addr / sizeof(data), val: data);
1180 }
1181 break;
1182
1183 case PTRACE_GETREGS: /* Get all gp regs from the child. */
1184 return copy_regset_to_user(target: child,
1185 view: &user_x86_64_view,
1186 REGSET_GENERAL,
1187 offset: 0, size: sizeof(struct user_regs_struct),
1188 data: datap);
1189
1190 case PTRACE_SETREGS: /* Set all gp regs in the child. */
1191 return copy_regset_from_user(target: child,
1192 view: &user_x86_64_view,
1193 REGSET_GENERAL,
1194 offset: 0, size: sizeof(struct user_regs_struct),
1195 data: datap);
1196
1197 case PTRACE_GETFPREGS: /* Get the child FPU state. */
1198 return copy_regset_to_user(target: child,
1199 view: &user_x86_64_view,
1200 REGSET_FP,
1201 offset: 0, size: sizeof(struct user_i387_struct),
1202 data: datap);
1203
1204 case PTRACE_SETFPREGS: /* Set the child FPU state. */
1205 return copy_regset_from_user(target: child,
1206 view: &user_x86_64_view,
1207 REGSET_FP,
1208 offset: 0, size: sizeof(struct user_i387_struct),
1209 data: datap);
1210
1211 default:
1212 return compat_ptrace_request(child, request, addr, data);
1213 }
1214
1215 return ret;
1216}
1217#endif
1218
1219#ifdef CONFIG_COMPAT
1220long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1221 compat_ulong_t caddr, compat_ulong_t cdata)
1222{
1223#ifdef CONFIG_X86_X32_ABI
1224 if (!in_ia32_syscall())
1225 return x32_arch_ptrace(child, request, caddr, cdata);
1226#endif
1227#ifdef CONFIG_IA32_EMULATION
1228 return ia32_arch_ptrace(child, request, caddr, cdata);
1229#else
1230 return 0;
1231#endif
1232}
1233#endif /* CONFIG_COMPAT */
1234
1235#ifdef CONFIG_X86_64
1236
1237static struct user_regset x86_64_regsets[] __ro_after_init = {
1238 [REGSET64_GENERAL] = {
1239 .core_note_type = NT_PRSTATUS,
1240 .n = sizeof(struct user_regs_struct) / sizeof(long),
1241 .size = sizeof(long),
1242 .align = sizeof(long),
1243 .regset_get = genregs_get,
1244 .set = genregs_set
1245 },
1246 [REGSET64_FP] = {
1247 .core_note_type = NT_PRFPREG,
1248 .n = sizeof(struct fxregs_state) / sizeof(long),
1249 .size = sizeof(long),
1250 .align = sizeof(long),
1251 .active = regset_xregset_fpregs_active,
1252 .regset_get = xfpregs_get,
1253 .set = xfpregs_set
1254 },
1255 [REGSET64_XSTATE] = {
1256 .core_note_type = NT_X86_XSTATE,
1257 .size = sizeof(u64),
1258 .align = sizeof(u64),
1259 .active = xstateregs_active,
1260 .regset_get = xstateregs_get,
1261 .set = xstateregs_set
1262 },
1263 [REGSET64_IOPERM] = {
1264 .core_note_type = NT_386_IOPERM,
1265 .n = IO_BITMAP_LONGS,
1266 .size = sizeof(long),
1267 .align = sizeof(long),
1268 .active = ioperm_active,
1269 .regset_get = ioperm_get
1270 },
1271#ifdef CONFIG_X86_USER_SHADOW_STACK
1272 [REGSET64_SSP] = {
1273 .core_note_type = NT_X86_SHSTK,
1274 .n = 1,
1275 .size = sizeof(u64),
1276 .align = sizeof(u64),
1277 .active = ssp_active,
1278 .regset_get = ssp_get,
1279 .set = ssp_set
1280 },
1281#endif
1282};
1283
1284static const struct user_regset_view user_x86_64_view = {
1285 .name = "x86_64", .e_machine = EM_X86_64,
1286 .regsets = x86_64_regsets, .n = ARRAY_SIZE(x86_64_regsets)
1287};
1288
1289#else /* CONFIG_X86_32 */
1290
1291#define user_regs_struct32 user_regs_struct
1292#define genregs32_get genregs_get
1293#define genregs32_set genregs_set
1294
1295#endif /* CONFIG_X86_64 */
1296
1297#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1298static struct user_regset x86_32_regsets[] __ro_after_init = {
1299 [REGSET32_GENERAL] = {
1300 .core_note_type = NT_PRSTATUS,
1301 .n = sizeof(struct user_regs_struct32) / sizeof(u32),
1302 .size = sizeof(u32),
1303 .align = sizeof(u32),
1304 .regset_get = genregs32_get,
1305 .set = genregs32_set
1306 },
1307 [REGSET32_FP] = {
1308 .core_note_type = NT_PRFPREG,
1309 .n = sizeof(struct user_i387_ia32_struct) / sizeof(u32),
1310 .size = sizeof(u32),
1311 .align = sizeof(u32),
1312 .active = regset_fpregs_active,
1313 .regset_get = fpregs_get,
1314 .set = fpregs_set
1315 },
1316 [REGSET32_XFP] = {
1317 .core_note_type = NT_PRXFPREG,
1318 .n = sizeof(struct fxregs_state) / sizeof(u32),
1319 .size = sizeof(u32),
1320 .align = sizeof(u32),
1321 .active = regset_xregset_fpregs_active,
1322 .regset_get = xfpregs_get,
1323 .set = xfpregs_set
1324 },
1325 [REGSET32_XSTATE] = {
1326 .core_note_type = NT_X86_XSTATE,
1327 .size = sizeof(u64),
1328 .align = sizeof(u64),
1329 .active = xstateregs_active,
1330 .regset_get = xstateregs_get,
1331 .set = xstateregs_set
1332 },
1333 [REGSET32_TLS] = {
1334 .core_note_type = NT_386_TLS,
1335 .n = GDT_ENTRY_TLS_ENTRIES,
1336 .bias = GDT_ENTRY_TLS_MIN,
1337 .size = sizeof(struct user_desc),
1338 .align = sizeof(struct user_desc),
1339 .active = regset_tls_active,
1340 .regset_get = regset_tls_get,
1341 .set = regset_tls_set
1342 },
1343 [REGSET32_IOPERM] = {
1344 .core_note_type = NT_386_IOPERM,
1345 .n = IO_BITMAP_BYTES / sizeof(u32),
1346 .size = sizeof(u32),
1347 .align = sizeof(u32),
1348 .active = ioperm_active,
1349 .regset_get = ioperm_get
1350 },
1351};
1352
1353static const struct user_regset_view user_x86_32_view = {
1354 .name = "i386", .e_machine = EM_386,
1355 .regsets = x86_32_regsets, .n = ARRAY_SIZE(x86_32_regsets)
1356};
1357#endif
1358
1359/*
1360 * This represents bytes 464..511 in the memory layout exported through
1361 * the REGSET_XSTATE interface.
1362 */
1363u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1364
1365void __init update_regset_xstate_info(unsigned int size, u64 xstate_mask)
1366{
1367#ifdef CONFIG_X86_64
1368 x86_64_regsets[REGSET64_XSTATE].n = size / sizeof(u64);
1369#endif
1370#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1371 x86_32_regsets[REGSET32_XSTATE].n = size / sizeof(u64);
1372#endif
1373 xstate_fx_sw_bytes[USER_XSTATE_XCR0_WORD] = xstate_mask;
1374}
1375
1376/*
1377 * This is used by the core dump code to decide which regset to dump. The
1378 * core dump code writes out the resulting .e_machine and the corresponding
1379 * regsets. This is suboptimal if the task is messing around with its CS.L
1380 * field, but at worst the core dump will end up missing some information.
1381 *
1382 * Unfortunately, it is also used by the broken PTRACE_GETREGSET and
1383 * PTRACE_SETREGSET APIs. These APIs look at the .regsets field but have
1384 * no way to make sure that the e_machine they use matches the caller's
1385 * expectations. The result is that the data format returned by
1386 * PTRACE_GETREGSET depends on the returned CS field (and even the offset
1387 * of the returned CS field depends on its value!) and the data format
1388 * accepted by PTRACE_SETREGSET is determined by the old CS value. The
1389 * upshot is that it is basically impossible to use these APIs correctly.
1390 *
1391 * The best way to fix it in the long run would probably be to add new
1392 * improved ptrace() APIs to read and write registers reliably, possibly by
1393 * allowing userspace to select the ELF e_machine variant that they expect.
1394 */
1395const struct user_regset_view *task_user_regset_view(struct task_struct *task)
1396{
1397#ifdef CONFIG_IA32_EMULATION
1398 if (!user_64bit_mode(task_pt_regs(task)))
1399#endif
1400#if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1401 return &user_x86_32_view;
1402#endif
1403#ifdef CONFIG_X86_64
1404 return &user_x86_64_view;
1405#endif
1406}
1407
1408void send_sigtrap(struct pt_regs *regs, int error_code, int si_code)
1409{
1410 struct task_struct *tsk = current;
1411
1412 tsk->thread.trap_nr = X86_TRAP_DB;
1413 tsk->thread.error_code = error_code;
1414
1415 /* Send us the fake SIGTRAP */
1416 force_sig_fault(SIGTRAP, code: si_code,
1417 addr: user_mode(regs) ? (void __user *)regs->ip : NULL);
1418}
1419
1420void user_single_step_report(struct pt_regs *regs)
1421{
1422 send_sigtrap(regs, error_code: 0, TRAP_BRKPT);
1423}
1424

source code of linux/arch/x86/kernel/ptrace.c