SHM_UNLOCK: fix long unpreemptible section
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kernel / ptrace.c
blob82528799c5deece6e4da9758a520b82420763bfc
1 /* By Ross Biro 1/23/92 */
2 /*
3 * Pentium III FXSR, SSE support
4 * Gareth Hughes <gareth@valinux.com>, May 2000
5 */
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>
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>
37 #include "tls.h"
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/syscalls.h>
42 enum 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,
52 struct pt_regs_offset {
53 const char *name;
54 int offset;
57 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
58 #define REG_OFFSET_END {.name = NULL, .offset = 0}
60 static 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,
93 /**
94 * regs_query_register_offset() - query register offset from its name
95 * @name: the name of a register
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;
100 int regs_query_register_offset(const char *name)
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;
110 * regs_query_register_name() - query register name from its offset
111 * @offset: the offset of a register in struct pt_regs.
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;
116 const char *regs_query_register_name(unsigned int offset)
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;
125 static 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
141 * does not yet catch signals sent when the child dies.
142 * in exit.c or in signal.c.
146 * Determines which flags the user has access to [1 = access, 0 = no access].
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))
156 * Determines whether a value may be installed in a segment register.
158 static inline bool invalid_selector(u16 value)
160 return unlikely(value != 0 && (value & SEGMENT_RPL_MASK) != USER_RPL);
163 #ifdef CONFIG_X86_32
165 #define FLAG_MASK FLAG_MASK_32
167 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long regno)
169 BUILD_BUG_ON(offsetof(struct pt_regs, bx) != 0);
170 return &regs->bx + (regno >> 2);
173 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
176 * Returning the value truncates it to 16 bits.
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);
187 return retval;
190 static int set_segment_reg(struct task_struct *task,
191 unsigned long offset, u16 value)
194 * The value argument was already truncated to 16 bits.
196 if (invalid_selector(value))
197 return -EIO;
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.
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;
214 default:
215 *pt_regs_access(task_pt_regs(task), offset) = value;
216 break;
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;
225 return 0;
228 #else /* CONFIG_X86_64 */
230 #define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
232 static unsigned long *pt_regs_access(struct pt_regs *regs, unsigned long offset)
234 BUILD_BUG_ON(offsetof(struct pt_regs, r15) != 0);
235 return &regs->r15 + (offset / sizeof(regs->r15));
238 static u16 get_segment_reg(struct task_struct *task, unsigned long offset)
241 * Returning the value truncates it to 16 bits.
243 unsigned int seg;
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;
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;
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;
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;
270 return task->thread.es;
272 case offsetof(struct user_regs_struct, cs):
273 case offsetof(struct user_regs_struct, ss):
274 break;
276 return *pt_regs_access(task_pt_regs(task), offset);
279 static int set_segment_reg(struct task_struct *task,
280 unsigned long offset, u16 value)
283 * The value argument was already truncated to 16 bits.
285 if (invalid_selector(value))
286 return -EIO;
288 switch (offset) {
289 case offsetof(struct user_regs_struct,fs):
291 * If this is setting fs as for normal 64-bit use but
292 * setting fs_base has implicitly changed it, leave it.
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):
305 * If this is setting gs as for normal 64-bit use but
306 * setting gs_base has implicitly changed it, leave it.
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;
329 * Can't actually change these in 64-bit mode.
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;
349 return 0;
352 #endif /* CONFIG_X86_32 */
354 static unsigned long get_flags(struct task_struct *task)
356 unsigned long retval = task_pt_regs(task)->flags;
359 * If the debugger set TF, hide it from the readout.
361 if (test_tsk_thread_flag(task, TIF_FORCED_TF))
362 retval &= ~X86_EFLAGS_TF;
364 return retval;
367 static int set_flags(struct task_struct *task, unsigned long value)
369 struct pt_regs *regs = task_pt_regs(task);
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.
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;
381 regs->flags = (regs->flags & ~FLAG_MASK) | (value & FLAG_MASK);
383 return 0;
386 static int putreg(struct task_struct *child,
387 unsigned long offset, unsigned long value)
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);
398 case offsetof(struct user_regs_struct, flags):
399 return set_flags(child, value);
401 #ifdef CONFIG_X86_64
402 case offsetof(struct user_regs_struct,fs_base):
403 if (value >= TASK_SIZE_OF(child))
404 return -EIO;
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.
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):
415 * Exactly the same here as the %fs handling above.
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
425 *pt_regs_access(task_pt_regs(child), offset) = value;
426 return 0;
429 static unsigned long getreg(struct task_struct *task, unsigned long offset)
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);
440 case offsetof(struct user_regs_struct, flags):
441 return get_flags(task);
443 #ifdef CONFIG_X86_64
444 case offsetof(struct user_regs_struct, fs_base): {
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.
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]);
459 case offsetof(struct user_regs_struct, gs_base): {
461 * Exactly the same here as the %fs handling above.
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]);
472 #endif
475 return *pt_regs_access(task_pt_regs(task), offset);
478 static 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)
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);
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);
500 return 0;
503 static 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)
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);
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);
528 return ret;
531 static void ptrace_triggered(struct perf_event *bp,
532 struct perf_sample_data *data,
533 struct pt_regs *regs)
535 int i;
536 struct thread_struct *thread = &(current->thread);
539 * Store in the virtual DR6 register the fact that the breakpoint
540 * was hit so the thread's debugger will see it.
542 for (i = 0; i < HBP_NUM; i++) {
543 if (thread->ptrace_bps[i] == bp)
544 break;
547 thread->debugreg6 |= (DR_TRAP0 << i);
551 * Walk through every ptrace breakpoints for this thread and
552 * build the dr7 value on top of their attributes.
555 static unsigned long ptrace_get_dr7(struct perf_event *bp[])
557 int i;
558 int dr7 = 0;
559 struct arch_hw_breakpoint *info;
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);
568 return dr7;
571 static int
572 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
573 struct task_struct *tsk, int disabled)
575 int err;
576 int gen_len, gen_type;
577 struct perf_event_attr attr;
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
584 if (!bp)
585 return -EINVAL;
587 err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
588 if (err)
589 return err;
591 attr = bp->attr;
592 attr.bp_len = gen_len;
593 attr.bp_type = gen_type;
594 attr.disabled = disabled;
596 return modify_user_hw_breakpoint(bp, &attr);
600 * Handle ptrace writes to debug register 7.
602 static int ptrace_write_dr7(struct task_struct *tsk, unsigned long data)
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;
611 if (ptrace_get_breakpoints(tsk) < 0)
612 return -ESRCH;
614 data &= ~DR_CONTROL_RESERVED;
615 old_dr7 = ptrace_get_dr7(thread->ptrace_bps);
616 restore:
618 * Loop through all the hardware breakpoints, making the
619 * appropriate changes to each.
621 for (i = 0; i < HBP_NUM; i++) {
622 enabled = decode_dr7(data, i, &len, &type);
623 bp = thread->ptrace_bps[i];
625 if (!enabled) {
626 if (bp) {
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.
634 if (!second_pass)
635 continue;
637 rc = ptrace_modify_breakpoint(bp, len, type,
638 tsk, 1);
639 if (rc)
640 break;
642 continue;
645 rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
646 if (rc)
647 break;
650 * Make a second pass to free the remaining unused breakpoints
651 * or to restore the original breakpoints if an error occurred.
653 if (!second_pass) {
654 second_pass = 1;
655 if (rc < 0) {
656 orig_ret = rc;
657 data = old_dr7;
659 goto restore;
662 ptrace_put_breakpoints(tsk);
664 return ((orig_ret < 0) ? orig_ret : rc);
668 * Handle PTRACE_PEEKUSR calls for the debug register area.
670 static unsigned long ptrace_get_debugreg(struct task_struct *tsk, int n)
672 struct thread_struct *thread = &(tsk->thread);
673 unsigned long val = 0;
675 if (n < HBP_NUM) {
676 struct perf_event *bp;
678 if (ptrace_get_breakpoints(tsk) < 0)
679 return -ESRCH;
681 bp = thread->ptrace_bps[n];
682 if (!bp)
683 val = 0;
684 else
685 val = bp->hw.info.address;
687 ptrace_put_breakpoints(tsk);
688 } else if (n == 6) {
689 val = thread->debugreg6;
690 } else if (n == 7) {
691 val = thread->ptrace_dr7;
693 return val;
696 static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
697 unsigned long addr)
699 struct perf_event *bp;
700 struct thread_struct *t = &tsk->thread;
701 struct perf_event_attr attr;
702 int err = 0;
704 if (ptrace_get_breakpoints(tsk) < 0)
705 return -ESRCH;
707 if (!t->ptrace_bps[nr]) {
708 ptrace_breakpoint_init(&attr);
710 * Put stub len and type to register (reserve) an inactive but
711 * correct bp
713 attr.bp_addr = addr;
714 attr.bp_len = HW_BREAKPOINT_LEN_1;
715 attr.bp_type = HW_BREAKPOINT_W;
716 attr.disabled = 1;
718 bp = register_user_hw_breakpoint(&attr, ptrace_triggered,
719 NULL, tsk);
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.
730 if (IS_ERR(bp)) {
731 err = PTR_ERR(bp);
732 goto put;
735 t->ptrace_bps[nr] = bp;
736 } else {
737 bp = t->ptrace_bps[nr];
739 attr = bp->attr;
740 attr.bp_addr = addr;
741 err = modify_user_hw_breakpoint(bp, &attr);
744 put:
745 ptrace_put_breakpoints(tsk);
746 return err;
750 * Handle PTRACE_POKEUSR calls for the debug register area.
752 int ptrace_set_debugreg(struct task_struct *tsk, int n, unsigned long val)
754 struct thread_struct *thread = &(tsk->thread);
755 int rc = 0;
757 /* There are no DR4 or DR5 registers */
758 if (n == 4 || n == 5)
759 return -EIO;
761 if (n == 6) {
762 thread->debugreg6 = val;
763 goto ret_path;
765 if (n < HBP_NUM) {
766 rc = ptrace_set_breakpoint_addr(tsk, n, val);
767 if (rc)
768 return rc;
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;
777 ret_path:
778 return rc;
782 * These access the current or another (stopped) task's io permission
783 * bitmap for debugging or core dump.
785 static int ioperm_active(struct task_struct *target,
786 const struct user_regset *regset)
788 return target->thread.io_bitmap_max / regset->size;
791 static 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)
796 if (!target->thread.io_bitmap_ptr)
797 return -ENXIO;
799 return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
800 target->thread.io_bitmap_ptr,
801 0, IO_BITMAP_BYTES);
805 * Called by kernel/ptrace.c when detaching..
807 * Make sure the single step bit is not set.
809 void ptrace_disable(struct task_struct *child)
811 user_disable_single_step(child);
812 #ifdef TIF_SYSCALL_EMU
813 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
814 #endif
817 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
818 static const struct user_regset_view user_x86_32_view; /* Initialized below. */
819 #endif
821 long arch_ptrace(struct task_struct *child, long request,
822 unsigned long addr, unsigned long data)
824 int ret;
825 unsigned long __user *datap = (unsigned long __user *)data;
827 switch (request) {
828 /* read the word at location addr in the USER area. */
829 case PTRACE_PEEKUSR: {
830 unsigned long tmp;
832 ret = -EIO;
833 if ((addr & (sizeof(data) - 1)) || addr >= sizeof(struct user))
834 break;
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));
844 ret = put_user(tmp, datap);
845 break;
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;
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);
861 break;
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);
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);
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);
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);
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;
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
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;
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
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
930 default:
931 ret = ptrace_request(child, request, addr, data);
932 break;
935 return ret;
938 #ifdef CONFIG_IA32_EMULATION
940 #include <linux/compat.h>
941 #include <linux/syscalls.h>
942 #include <asm/ia32.h>
943 #include <asm/user32.h>
945 #define R32(l,q) \
946 case offsetof(struct user32, regs.l): \
947 regs->q = value; break
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
956 static int putreg32(struct task_struct *child, unsigned regno, u32 value)
958 struct pt_regs *regs = task_pt_regs(child);
960 switch (regno) {
962 SEG32(cs);
963 SEG32(ds);
964 SEG32(es);
965 SEG32(fs);
966 SEG32(gs);
967 SEG32(ss);
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);
979 case offsetof(struct user32, regs.orig_eax):
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.
987 regs->orig_ax = value;
988 if (syscall_get_nr(child, regs) >= 0)
989 task_thread_info(child)->status |= TS_COMPAT;
990 break;
992 case offsetof(struct user32, regs.eflags):
993 return set_flags(child, value);
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);
1000 default:
1001 if (regno > sizeof(struct user32) || (regno & 3))
1002 return -EIO;
1005 * Other dummy fields in the virtual user structure
1006 * are ignored
1008 break;
1010 return 0;
1013 #undef R32
1014 #undef SEG32
1016 #define R32(l,q) \
1017 case offsetof(struct user32, regs.l): \
1018 *val = regs->q; break
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
1026 static int getreg32(struct task_struct *child, unsigned regno, u32 *val)
1028 struct pt_regs *regs = task_pt_regs(child);
1030 switch (regno) {
1032 SEG32(ds);
1033 SEG32(es);
1034 SEG32(fs);
1035 SEG32(gs);
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);
1050 case offsetof(struct user32, regs.eflags):
1051 *val = get_flags(child);
1052 break;
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;
1060 default:
1061 if (regno > sizeof(struct user32) || (regno & 3))
1062 return -EIO;
1065 * Other dummy fields in the virtual user structure
1066 * are ignored
1068 *val = 0;
1069 break;
1071 return 0;
1074 #undef R32
1075 #undef SEG32
1077 static 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)
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);
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);
1101 return 0;
1104 static 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)
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);
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);
1129 return ret;
1132 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
1133 compat_ulong_t caddr, compat_ulong_t cdata)
1135 unsigned long addr = caddr;
1136 unsigned long data = cdata;
1137 void __user *datap = compat_ptr(data);
1138 int ret;
1139 __u32 val;
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;
1148 case PTRACE_POKEUSR:
1149 ret = putreg32(child, addr, data);
1150 break;
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);
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);
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);
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);
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);
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);
1187 case PTRACE_GET_THREAD_AREA:
1188 case PTRACE_SET_THREAD_AREA:
1189 return arch_ptrace(child, request, addr, data);
1191 default:
1192 return compat_ptrace_request(child, request, addr, data);
1195 return ret;
1198 #endif /* CONFIG_IA32_EMULATION */
1200 #ifdef CONFIG_X86_64
1202 static 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
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
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
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
1229 static 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)
1234 #else /* CONFIG_X86_32 */
1236 #define user_regs_struct32 user_regs_struct
1237 #define genregs32_get genregs_get
1238 #define genregs32_set genregs_set
1240 #define user_i387_ia32_struct user_i387_struct
1241 #define user32_fxsr_struct user_fxsr_struct
1243 #endif /* CONFIG_X86_64 */
1245 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1246 static 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
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
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
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
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
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
1287 static 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)
1291 #endif
1294 * This represents bytes 464..511 in the memory layout exported through
1295 * the REGSET_XSTATE interface.
1297 u64 xstate_fx_sw_bytes[USER_XSTATE_FX_SW_WORDS];
1299 void update_regset_xstate_info(unsigned int size, u64 xstate_mask)
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;
1310 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
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
1323 static void fill_sigtrap_info(struct task_struct *tsk,
1324 struct pt_regs *regs,
1325 int error_code, int si_code,
1326 struct siginfo *info)
1328 tsk->thread.trap_no = 1;
1329 tsk->thread.error_code = error_code;
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;
1337 void user_single_step_siginfo(struct task_struct *tsk,
1338 struct pt_regs *regs,
1339 struct siginfo *info)
1341 fill_sigtrap_info(tsk, regs, 0, TRAP_BRKPT, info);
1344 void send_sigtrap(struct task_struct *tsk, struct pt_regs *regs,
1345 int error_code, int si_code)
1347 struct siginfo info;
1349 fill_sigtrap_info(tsk, regs, error_code, si_code, &info);
1350 /* Send us the fake SIGTRAP */
1351 force_sig_info(SIGTRAP, &info, tsk);
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
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.
1367 long syscall_trace_enter(struct pt_regs *regs)
1369 long ret = 0;
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.
1378 if (test_thread_flag(TIF_SINGLESTEP))
1379 regs->flags |= X86_EFLAGS_TF;
1381 /* do the secure computing check first */
1382 secure_computing(regs->orig_ax);
1384 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU)))
1385 ret = -1L;
1387 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
1388 tracehook_report_syscall_entry(regs))
1389 ret = -1L;
1391 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1392 trace_sys_enter(regs, regs->orig_ax);
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
1409 return ret ?: regs->orig_ax;
1412 void syscall_trace_leave(struct pt_regs *regs)
1414 bool step;
1416 if (unlikely(current->audit_context))
1417 audit_syscall_exit(AUDITSC_RESULT(regs->ax), regs->ax);
1419 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
1420 trace_sys_exit(regs, regs->ax);
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().
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);