1 /* By Ross Biro 1/23/92 */
3 * Pentium III FXSR, SSE support
4 * Gareth Hughes <gareth@valinux.com>, May 2000
7 * Markus Metzger <markus.t.metzger@intel.com>, Dec 2007
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
13 #include <linux/smp.h>
14 #include <linux/errno.h>
15 #include <linux/ptrace.h>
16 #include <linux/regset.h>
17 #include <linux/tracehook.h>
18 #include <linux/user.h>
19 #include <linux/elf.h>
20 #include <linux/security.h>
21 #include <linux/audit.h>
22 #include <linux/seccomp.h>
23 #include <linux/signal.h>
24 #include <linux/workqueue.h>
25 #include <linux/perf_event.h>
26 #include <linux/hw_breakpoint.h>
28 #include <asm/uaccess.h>
29 #include <asm/pgtable.h>
30 #include <asm/system.h>
31 #include <asm/processor.h>
33 #include <asm/debugreg.h>
36 #include <asm/prctl.h>
37 #include <asm/proto.h>
39 #include <asm/hw_breakpoint.h>
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/syscalls.h>
50 REGSET_IOPERM64
= REGSET_XFP
,
55 struct pt_regs_offset
{
60 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
61 #define REG_OFFSET_END {.name = NULL, .offset = 0}
63 static const struct pt_regs_offset regoffset_table
[] = {
87 REG_OFFSET_NAME(orig_ax
),
90 REG_OFFSET_NAME(flags
),
97 * regs_query_register_offset() - query register offset from its name
98 * @name: the name of a register
100 * regs_query_register_offset() returns the offset of a register in struct
101 * pt_regs from its name. If the name is invalid, this returns -EINVAL;
103 int regs_query_register_offset(const char *name
)
105 const struct pt_regs_offset
*roff
;
106 for (roff
= regoffset_table
; roff
->name
!= NULL
; roff
++)
107 if (!strcmp(roff
->name
, name
))
113 * regs_query_register_name() - query register name from its offset
114 * @offset: the offset of a register in struct pt_regs.
116 * regs_query_register_name() returns the name of a register from its
117 * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
119 const char *regs_query_register_name(unsigned int offset
)
121 const struct pt_regs_offset
*roff
;
122 for (roff
= regoffset_table
; roff
->name
!= NULL
; roff
++)
123 if (roff
->offset
== offset
)
128 static const int arg_offs_table
[] = {
130 [0] = offsetof(struct pt_regs
, ax
),
131 [1] = offsetof(struct pt_regs
, dx
),
132 [2] = offsetof(struct pt_regs
, cx
)
133 #else /* CONFIG_X86_64 */
134 [0] = offsetof(struct pt_regs
, di
),
135 [1] = offsetof(struct pt_regs
, si
),
136 [2] = offsetof(struct pt_regs
, dx
),
137 [3] = offsetof(struct pt_regs
, cx
),
138 [4] = offsetof(struct pt_regs
, r8
),
139 [5] = offsetof(struct pt_regs
, r9
)
144 * regs_get_argument_nth() - get Nth argument at function call
145 * @regs: pt_regs which contains registers at function entry.
146 * @n: argument number.
148 * regs_get_argument_nth() returns @n th argument of a function call.
149 * Since usually the kernel stack will be changed right after function entry,
150 * you must use this at function entry. If the @n th entry is NOT in the
151 * kernel stack or pt_regs, this returns 0.
153 unsigned long regs_get_argument_nth(struct pt_regs
*regs
, unsigned int n
)
155 if (n
< ARRAY_SIZE(arg_offs_table
))
156 return *(unsigned long *)((char *)regs
+ arg_offs_table
[n
]);
159 * The typical case: arg n is on the stack.
160 * (Note: stack[0] = return address, so skip it)
162 n
-= ARRAY_SIZE(arg_offs_table
);
163 return regs_get_kernel_stack_nth(regs
, 1 + n
);
168 * does not yet catch signals sent when the child dies.
169 * in exit.c or in signal.c.
173 * Determines which flags the user has access to [1 = access, 0 = no access].
175 #define FLAG_MASK_32 ((unsigned long) \
176 (X86_EFLAGS_CF | X86_EFLAGS_PF | \
177 X86_EFLAGS_AF | X86_EFLAGS_ZF | \
178 X86_EFLAGS_SF | X86_EFLAGS_TF | \
179 X86_EFLAGS_DF | X86_EFLAGS_OF | \
180 X86_EFLAGS_RF | X86_EFLAGS_AC))
183 * Determines whether a value may be installed in a segment register.
185 static inline bool invalid_selector(u16 value
)
187 return unlikely(value
!= 0 && (value
& SEGMENT_RPL_MASK
) != USER_RPL
);
192 #define FLAG_MASK FLAG_MASK_32
194 static unsigned long *pt_regs_access(struct pt_regs
*regs
, unsigned long regno
)
196 BUILD_BUG_ON(offsetof(struct pt_regs
, bx
) != 0);
197 return ®s
->bx
+ (regno
>> 2);
200 static u16
get_segment_reg(struct task_struct
*task
, unsigned long offset
)
203 * Returning the value truncates it to 16 bits.
206 if (offset
!= offsetof(struct user_regs_struct
, gs
))
207 retval
= *pt_regs_access(task_pt_regs(task
), offset
);
210 retval
= get_user_gs(task_pt_regs(task
));
212 retval
= task_user_gs(task
);
217 static int set_segment_reg(struct task_struct
*task
,
218 unsigned long offset
, u16 value
)
221 * The value argument was already truncated to 16 bits.
223 if (invalid_selector(value
))
227 * For %cs and %ss we cannot permit a null selector.
228 * We can permit a bogus selector as long as it has USER_RPL.
229 * Null selectors are fine for other segment registers, but
230 * we will never get back to user mode with invalid %cs or %ss
231 * and will take the trap in iret instead. Much code relies
232 * on user_mode() to distinguish a user trap frame (which can
233 * safely use invalid selectors) from a kernel trap frame.
236 case offsetof(struct user_regs_struct
, cs
):
237 case offsetof(struct user_regs_struct
, ss
):
238 if (unlikely(value
== 0))
242 *pt_regs_access(task_pt_regs(task
), offset
) = value
;
245 case offsetof(struct user_regs_struct
, gs
):
247 set_user_gs(task_pt_regs(task
), value
);
249 task_user_gs(task
) = value
;
255 #else /* CONFIG_X86_64 */
257 #define FLAG_MASK (FLAG_MASK_32 | X86_EFLAGS_NT)
259 static unsigned long *pt_regs_access(struct pt_regs
*regs
, unsigned long offset
)
261 BUILD_BUG_ON(offsetof(struct pt_regs
, r15
) != 0);
262 return ®s
->r15
+ (offset
/ sizeof(regs
->r15
));
265 static u16
get_segment_reg(struct task_struct
*task
, unsigned long offset
)
268 * Returning the value truncates it to 16 bits.
273 case offsetof(struct user_regs_struct
, fs
):
274 if (task
== current
) {
275 /* Older gas can't assemble movq %?s,%r?? */
276 asm("movl %%fs,%0" : "=r" (seg
));
279 return task
->thread
.fsindex
;
280 case offsetof(struct user_regs_struct
, gs
):
281 if (task
== current
) {
282 asm("movl %%gs,%0" : "=r" (seg
));
285 return task
->thread
.gsindex
;
286 case offsetof(struct user_regs_struct
, ds
):
287 if (task
== current
) {
288 asm("movl %%ds,%0" : "=r" (seg
));
291 return task
->thread
.ds
;
292 case offsetof(struct user_regs_struct
, es
):
293 if (task
== current
) {
294 asm("movl %%es,%0" : "=r" (seg
));
297 return task
->thread
.es
;
299 case offsetof(struct user_regs_struct
, cs
):
300 case offsetof(struct user_regs_struct
, ss
):
303 return *pt_regs_access(task_pt_regs(task
), offset
);
306 static int set_segment_reg(struct task_struct
*task
,
307 unsigned long offset
, u16 value
)
310 * The value argument was already truncated to 16 bits.
312 if (invalid_selector(value
))
316 case offsetof(struct user_regs_struct
,fs
):
318 * If this is setting fs as for normal 64-bit use but
319 * setting fs_base has implicitly changed it, leave it.
321 if ((value
== FS_TLS_SEL
&& task
->thread
.fsindex
== 0 &&
322 task
->thread
.fs
!= 0) ||
323 (value
== 0 && task
->thread
.fsindex
== FS_TLS_SEL
&&
324 task
->thread
.fs
== 0))
326 task
->thread
.fsindex
= value
;
328 loadsegment(fs
, task
->thread
.fsindex
);
330 case offsetof(struct user_regs_struct
,gs
):
332 * If this is setting gs as for normal 64-bit use but
333 * setting gs_base has implicitly changed it, leave it.
335 if ((value
== GS_TLS_SEL
&& task
->thread
.gsindex
== 0 &&
336 task
->thread
.gs
!= 0) ||
337 (value
== 0 && task
->thread
.gsindex
== GS_TLS_SEL
&&
338 task
->thread
.gs
== 0))
340 task
->thread
.gsindex
= value
;
342 load_gs_index(task
->thread
.gsindex
);
344 case offsetof(struct user_regs_struct
,ds
):
345 task
->thread
.ds
= value
;
347 loadsegment(ds
, task
->thread
.ds
);
349 case offsetof(struct user_regs_struct
,es
):
350 task
->thread
.es
= value
;
352 loadsegment(es
, task
->thread
.es
);
356 * Can't actually change these in 64-bit mode.
358 case offsetof(struct user_regs_struct
,cs
):
359 if (unlikely(value
== 0))
361 #ifdef CONFIG_IA32_EMULATION
362 if (test_tsk_thread_flag(task
, TIF_IA32
))
363 task_pt_regs(task
)->cs
= value
;
366 case offsetof(struct user_regs_struct
,ss
):
367 if (unlikely(value
== 0))
369 #ifdef CONFIG_IA32_EMULATION
370 if (test_tsk_thread_flag(task
, TIF_IA32
))
371 task_pt_regs(task
)->ss
= value
;
379 #endif /* CONFIG_X86_32 */
381 static unsigned long get_flags(struct task_struct
*task
)
383 unsigned long retval
= task_pt_regs(task
)->flags
;
386 * If the debugger set TF, hide it from the readout.
388 if (test_tsk_thread_flag(task
, TIF_FORCED_TF
))
389 retval
&= ~X86_EFLAGS_TF
;
394 static int set_flags(struct task_struct
*task
, unsigned long value
)
396 struct pt_regs
*regs
= task_pt_regs(task
);
399 * If the user value contains TF, mark that
400 * it was not "us" (the debugger) that set it.
401 * If not, make sure it stays set if we had.
403 if (value
& X86_EFLAGS_TF
)
404 clear_tsk_thread_flag(task
, TIF_FORCED_TF
);
405 else if (test_tsk_thread_flag(task
, TIF_FORCED_TF
))
406 value
|= X86_EFLAGS_TF
;
408 regs
->flags
= (regs
->flags
& ~FLAG_MASK
) | (value
& FLAG_MASK
);
413 static int putreg(struct task_struct
*child
,
414 unsigned long offset
, unsigned long value
)
417 case offsetof(struct user_regs_struct
, cs
):
418 case offsetof(struct user_regs_struct
, ds
):
419 case offsetof(struct user_regs_struct
, es
):
420 case offsetof(struct user_regs_struct
, fs
):
421 case offsetof(struct user_regs_struct
, gs
):
422 case offsetof(struct user_regs_struct
, ss
):
423 return set_segment_reg(child
, offset
, value
);
425 case offsetof(struct user_regs_struct
, flags
):
426 return set_flags(child
, value
);
429 case offsetof(struct user_regs_struct
,fs_base
):
430 if (value
>= TASK_SIZE_OF(child
))
433 * When changing the segment base, use do_arch_prctl
434 * to set either thread.fs or thread.fsindex and the
435 * corresponding GDT slot.
437 if (child
->thread
.fs
!= value
)
438 return do_arch_prctl(child
, ARCH_SET_FS
, value
);
440 case offsetof(struct user_regs_struct
,gs_base
):
442 * Exactly the same here as the %fs handling above.
444 if (value
>= TASK_SIZE_OF(child
))
446 if (child
->thread
.gs
!= value
)
447 return do_arch_prctl(child
, ARCH_SET_GS
, value
);
452 *pt_regs_access(task_pt_regs(child
), offset
) = value
;
456 static unsigned long getreg(struct task_struct
*task
, unsigned long offset
)
459 case offsetof(struct user_regs_struct
, cs
):
460 case offsetof(struct user_regs_struct
, ds
):
461 case offsetof(struct user_regs_struct
, es
):
462 case offsetof(struct user_regs_struct
, fs
):
463 case offsetof(struct user_regs_struct
, gs
):
464 case offsetof(struct user_regs_struct
, ss
):
465 return get_segment_reg(task
, offset
);
467 case offsetof(struct user_regs_struct
, flags
):
468 return get_flags(task
);
471 case offsetof(struct user_regs_struct
, fs_base
): {
473 * do_arch_prctl may have used a GDT slot instead of
474 * the MSR. To userland, it appears the same either
475 * way, except the %fs segment selector might not be 0.
477 unsigned int seg
= task
->thread
.fsindex
;
478 if (task
->thread
.fs
!= 0)
479 return task
->thread
.fs
;
481 asm("movl %%fs,%0" : "=r" (seg
));
482 if (seg
!= FS_TLS_SEL
)
484 return get_desc_base(&task
->thread
.tls_array
[FS_TLS
]);
486 case offsetof(struct user_regs_struct
, gs_base
): {
488 * Exactly the same here as the %fs handling above.
490 unsigned int seg
= task
->thread
.gsindex
;
491 if (task
->thread
.gs
!= 0)
492 return task
->thread
.gs
;
494 asm("movl %%gs,%0" : "=r" (seg
));
495 if (seg
!= GS_TLS_SEL
)
497 return get_desc_base(&task
->thread
.tls_array
[GS_TLS
]);
502 return *pt_regs_access(task_pt_regs(task
), offset
);
505 static int genregs_get(struct task_struct
*target
,
506 const struct user_regset
*regset
,
507 unsigned int pos
, unsigned int count
,
508 void *kbuf
, void __user
*ubuf
)
511 unsigned long *k
= kbuf
;
512 while (count
>= sizeof(*k
)) {
513 *k
++ = getreg(target
, pos
);
518 unsigned long __user
*u
= ubuf
;
519 while (count
>= sizeof(*u
)) {
520 if (__put_user(getreg(target
, pos
), u
++))
530 static int genregs_set(struct task_struct
*target
,
531 const struct user_regset
*regset
,
532 unsigned int pos
, unsigned int count
,
533 const void *kbuf
, const void __user
*ubuf
)
537 const unsigned long *k
= kbuf
;
538 while (count
>= sizeof(*k
) && !ret
) {
539 ret
= putreg(target
, pos
, *k
++);
544 const unsigned long __user
*u
= ubuf
;
545 while (count
>= sizeof(*u
) && !ret
) {
547 ret
= __get_user(word
, u
++);
550 ret
= putreg(target
, pos
, word
);
558 static void ptrace_triggered(struct perf_event
*bp
, int nmi
,
559 struct perf_sample_data
*data
,
560 struct pt_regs
*regs
)
563 struct thread_struct
*thread
= &(current
->thread
);
566 * Store in the virtual DR6 register the fact that the breakpoint
567 * was hit so the thread's debugger will see it.
569 for (i
= 0; i
< HBP_NUM
; i
++) {
570 if (thread
->ptrace_bps
[i
] == bp
)
574 thread
->debugreg6
|= (DR_TRAP0
<< i
);
578 * Walk through every ptrace breakpoints for this thread and
579 * build the dr7 value on top of their attributes.
582 static unsigned long ptrace_get_dr7(struct perf_event
*bp
[])
586 struct arch_hw_breakpoint
*info
;
588 for (i
= 0; i
< HBP_NUM
; i
++) {
589 if (bp
[i
] && !bp
[i
]->attr
.disabled
) {
590 info
= counter_arch_bp(bp
[i
]);
591 dr7
|= encode_dr7(i
, info
->len
, info
->type
);
599 ptrace_modify_breakpoint(struct perf_event
*bp
, int len
, int type
,
600 struct task_struct
*tsk
, int disabled
)
603 int gen_len
, gen_type
;
604 struct perf_event_attr attr
;
607 * We shoud have at least an inactive breakpoint at this
608 * slot. It means the user is writing dr7 without having
609 * written the address register first
614 err
= arch_bp_generic_fields(len
, type
, &gen_len
, &gen_type
);
619 attr
.bp_len
= gen_len
;
620 attr
.bp_type
= gen_type
;
621 attr
.disabled
= disabled
;
623 return modify_user_hw_breakpoint(bp
, &attr
);
627 * Handle ptrace writes to debug register 7.
629 static int ptrace_write_dr7(struct task_struct
*tsk
, unsigned long data
)
631 struct thread_struct
*thread
= &(tsk
->thread
);
632 unsigned long old_dr7
;
633 int i
, orig_ret
= 0, rc
= 0;
634 int enabled
, second_pass
= 0;
636 struct perf_event
*bp
;
638 data
&= ~DR_CONTROL_RESERVED
;
639 old_dr7
= ptrace_get_dr7(thread
->ptrace_bps
);
642 * Loop through all the hardware breakpoints, making the
643 * appropriate changes to each.
645 for (i
= 0; i
< HBP_NUM
; i
++) {
646 enabled
= decode_dr7(data
, i
, &len
, &type
);
647 bp
= thread
->ptrace_bps
[i
];
652 * Don't unregister the breakpoints right-away,
653 * unless all register_user_hw_breakpoint()
654 * requests have succeeded. This prevents
655 * any window of opportunity for debug
656 * register grabbing by other users.
661 rc
= ptrace_modify_breakpoint(bp
, len
, type
,
669 rc
= ptrace_modify_breakpoint(bp
, len
, type
, tsk
, 0);
674 * Make a second pass to free the remaining unused breakpoints
675 * or to restore the original breakpoints if an error occurred.
685 return ((orig_ret
< 0) ? orig_ret
: rc
);
689 * Handle PTRACE_PEEKUSR calls for the debug register area.
691 static unsigned long ptrace_get_debugreg(struct task_struct
*tsk
, int n
)
693 struct thread_struct
*thread
= &(tsk
->thread
);
694 unsigned long val
= 0;
697 struct perf_event
*bp
;
698 bp
= thread
->ptrace_bps
[n
];
701 val
= bp
->hw
.info
.address
;
703 val
= thread
->debugreg6
;
705 val
= ptrace_get_dr7(thread
->ptrace_bps
);
710 static int ptrace_set_breakpoint_addr(struct task_struct
*tsk
, int nr
,
713 struct perf_event
*bp
;
714 struct thread_struct
*t
= &tsk
->thread
;
715 struct perf_event_attr attr
;
717 if (!t
->ptrace_bps
[nr
]) {
718 hw_breakpoint_init(&attr
);
720 * Put stub len and type to register (reserve) an inactive but
724 attr
.bp_len
= HW_BREAKPOINT_LEN_1
;
725 attr
.bp_type
= HW_BREAKPOINT_W
;
728 bp
= register_user_hw_breakpoint(&attr
, ptrace_triggered
, tsk
);
731 * CHECKME: the previous code returned -EIO if the addr wasn't
732 * a valid task virtual addr. The new one will return -EINVAL in
734 * -EINVAL may be what we want for in-kernel breakpoints users,
735 * but -EIO looks better for ptrace, since we refuse a register
736 * writing for the user. And anyway this is the previous
742 t
->ptrace_bps
[nr
] = bp
;
746 bp
= t
->ptrace_bps
[nr
];
750 err
= modify_user_hw_breakpoint(bp
, &attr
);
760 * Handle PTRACE_POKEUSR calls for the debug register area.
762 int ptrace_set_debugreg(struct task_struct
*tsk
, int n
, unsigned long val
)
764 struct thread_struct
*thread
= &(tsk
->thread
);
767 /* There are no DR4 or DR5 registers */
768 if (n
== 4 || n
== 5)
772 thread
->debugreg6
= val
;
776 rc
= ptrace_set_breakpoint_addr(tsk
, n
, val
);
780 /* All that's left is DR7 */
782 rc
= ptrace_write_dr7(tsk
, val
);
789 * These access the current or another (stopped) task's io permission
790 * bitmap for debugging or core dump.
792 static int ioperm_active(struct task_struct
*target
,
793 const struct user_regset
*regset
)
795 return target
->thread
.io_bitmap_max
/ regset
->size
;
798 static int ioperm_get(struct task_struct
*target
,
799 const struct user_regset
*regset
,
800 unsigned int pos
, unsigned int count
,
801 void *kbuf
, void __user
*ubuf
)
803 if (!target
->thread
.io_bitmap_ptr
)
806 return user_regset_copyout(&pos
, &count
, &kbuf
, &ubuf
,
807 target
->thread
.io_bitmap_ptr
,
811 #ifdef CONFIG_X86_PTRACE_BTS
813 * A branch trace store context.
815 * Contexts may only be installed by ptrace_bts_config() and only for
818 * Contexts are destroyed when the tracee is detached from the tracer.
819 * The actual destruction work requires interrupts enabled, so the
820 * work is deferred and will be scheduled during __ptrace_unlink().
822 * Contexts hold an additional task_struct reference on the traced
823 * task, as well as a reference on the tracer's mm.
825 * Ptrace already holds a task_struct for the duration of ptrace operations,
826 * but since destruction is deferred, it may be executed after both
827 * tracer and tracee exited.
830 /* The branch trace handle. */
831 struct bts_tracer
*tracer
;
833 /* The buffer used to store the branch trace and its size. */
837 /* The mm that paid for the above buffer. */
838 struct mm_struct
*mm
;
840 /* The task this context belongs to. */
841 struct task_struct
*task
;
843 /* The signal to send on a bts buffer overflow. */
844 unsigned int bts_ovfl_signal
;
846 /* The work struct to destroy a context. */
847 struct work_struct work
;
850 static int alloc_bts_buffer(struct bts_context
*context
, unsigned int size
)
855 err
= account_locked_memory(current
->mm
, current
->signal
->rlim
, size
);
859 buffer
= kzalloc(size
, GFP_KERNEL
);
863 context
->buffer
= buffer
;
864 context
->size
= size
;
865 context
->mm
= get_task_mm(current
);
870 refund_locked_memory(current
->mm
, size
);
874 static inline void free_bts_buffer(struct bts_context
*context
)
876 if (!context
->buffer
)
879 kfree(context
->buffer
);
880 context
->buffer
= NULL
;
882 refund_locked_memory(context
->mm
, context
->size
);
889 static void free_bts_context_work(struct work_struct
*w
)
891 struct bts_context
*context
;
893 context
= container_of(w
, struct bts_context
, work
);
895 ds_release_bts(context
->tracer
);
896 put_task_struct(context
->task
);
897 free_bts_buffer(context
);
901 static inline void free_bts_context(struct bts_context
*context
)
903 INIT_WORK(&context
->work
, free_bts_context_work
);
904 schedule_work(&context
->work
);
907 static inline struct bts_context
*alloc_bts_context(struct task_struct
*task
)
909 struct bts_context
*context
= kzalloc(sizeof(*context
), GFP_KERNEL
);
911 context
->task
= task
;
914 get_task_struct(task
);
920 static int ptrace_bts_read_record(struct task_struct
*child
, size_t index
,
921 struct bts_struct __user
*out
)
923 struct bts_context
*context
;
924 const struct bts_trace
*trace
;
925 struct bts_struct bts
;
926 const unsigned char *at
;
929 context
= child
->bts
;
933 trace
= ds_read_bts(context
->tracer
);
937 at
= trace
->ds
.top
- ((index
+ 1) * trace
->ds
.size
);
938 if ((void *)at
< trace
->ds
.begin
)
939 at
+= (trace
->ds
.n
* trace
->ds
.size
);
944 error
= trace
->read(context
->tracer
, at
, &bts
);
948 if (copy_to_user(out
, &bts
, sizeof(bts
)))
954 static int ptrace_bts_drain(struct task_struct
*child
,
956 struct bts_struct __user
*out
)
958 struct bts_context
*context
;
959 const struct bts_trace
*trace
;
960 const unsigned char *at
;
961 int error
, drained
= 0;
963 context
= child
->bts
;
967 trace
= ds_read_bts(context
->tracer
);
974 if (size
< (trace
->ds
.top
- trace
->ds
.begin
))
977 for (at
= trace
->ds
.begin
; (void *)at
< trace
->ds
.top
;
978 out
++, drained
++, at
+= trace
->ds
.size
) {
979 struct bts_struct bts
;
981 error
= trace
->read(context
->tracer
, at
, &bts
);
985 if (copy_to_user(out
, &bts
, sizeof(bts
)))
989 memset(trace
->ds
.begin
, 0, trace
->ds
.n
* trace
->ds
.size
);
991 error
= ds_reset_bts(context
->tracer
);
998 static int ptrace_bts_config(struct task_struct
*child
,
1000 const struct ptrace_bts_config __user
*ucfg
)
1002 struct bts_context
*context
;
1003 struct ptrace_bts_config cfg
;
1004 unsigned int flags
= 0;
1006 if (cfg_size
< sizeof(cfg
))
1009 if (copy_from_user(&cfg
, ucfg
, sizeof(cfg
)))
1012 context
= child
->bts
;
1014 context
= alloc_bts_context(child
);
1018 if (cfg
.flags
& PTRACE_BTS_O_SIGNAL
) {
1023 context
->bts_ovfl_signal
= cfg
.signal
;
1026 ds_release_bts(context
->tracer
);
1027 context
->tracer
= NULL
;
1029 if ((cfg
.flags
& PTRACE_BTS_O_ALLOC
) && (cfg
.size
!= context
->size
)) {
1032 free_bts_buffer(context
);
1036 err
= alloc_bts_buffer(context
, cfg
.size
);
1041 if (cfg
.flags
& PTRACE_BTS_O_TRACE
)
1044 if (cfg
.flags
& PTRACE_BTS_O_SCHED
)
1045 flags
|= BTS_TIMESTAMPS
;
1048 ds_request_bts_task(child
, context
->buffer
, context
->size
,
1049 NULL
, (size_t)-1, flags
);
1050 if (unlikely(IS_ERR(context
->tracer
))) {
1051 int error
= PTR_ERR(context
->tracer
);
1053 free_bts_buffer(context
);
1054 context
->tracer
= NULL
;
1061 static int ptrace_bts_status(struct task_struct
*child
,
1063 struct ptrace_bts_config __user
*ucfg
)
1065 struct bts_context
*context
;
1066 const struct bts_trace
*trace
;
1067 struct ptrace_bts_config cfg
;
1069 context
= child
->bts
;
1073 if (cfg_size
< sizeof(cfg
))
1076 trace
= ds_read_bts(context
->tracer
);
1080 memset(&cfg
, 0, sizeof(cfg
));
1081 cfg
.size
= trace
->ds
.end
- trace
->ds
.begin
;
1082 cfg
.signal
= context
->bts_ovfl_signal
;
1083 cfg
.bts_size
= sizeof(struct bts_struct
);
1086 cfg
.flags
|= PTRACE_BTS_O_SIGNAL
;
1088 if (trace
->ds
.flags
& BTS_USER
)
1089 cfg
.flags
|= PTRACE_BTS_O_TRACE
;
1091 if (trace
->ds
.flags
& BTS_TIMESTAMPS
)
1092 cfg
.flags
|= PTRACE_BTS_O_SCHED
;
1094 if (copy_to_user(ucfg
, &cfg
, sizeof(cfg
)))
1100 static int ptrace_bts_clear(struct task_struct
*child
)
1102 struct bts_context
*context
;
1103 const struct bts_trace
*trace
;
1105 context
= child
->bts
;
1109 trace
= ds_read_bts(context
->tracer
);
1113 memset(trace
->ds
.begin
, 0, trace
->ds
.n
* trace
->ds
.size
);
1115 return ds_reset_bts(context
->tracer
);
1118 static int ptrace_bts_size(struct task_struct
*child
)
1120 struct bts_context
*context
;
1121 const struct bts_trace
*trace
;
1123 context
= child
->bts
;
1127 trace
= ds_read_bts(context
->tracer
);
1131 return (trace
->ds
.top
- trace
->ds
.begin
) / trace
->ds
.size
;
1135 * Called from __ptrace_unlink() after the child has been moved back
1136 * to its original parent.
1138 void ptrace_bts_untrace(struct task_struct
*child
)
1140 if (unlikely(child
->bts
)) {
1141 free_bts_context(child
->bts
);
1145 #endif /* CONFIG_X86_PTRACE_BTS */
1148 * Called by kernel/ptrace.c when detaching..
1150 * Make sure the single step bit is not set.
1152 void ptrace_disable(struct task_struct
*child
)
1154 user_disable_single_step(child
);
1155 #ifdef TIF_SYSCALL_EMU
1156 clear_tsk_thread_flag(child
, TIF_SYSCALL_EMU
);
1160 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1161 static const struct user_regset_view user_x86_32_view
; /* Initialized below. */
1164 long arch_ptrace(struct task_struct
*child
, long request
, long addr
, long data
)
1167 unsigned long __user
*datap
= (unsigned long __user
*)data
;
1170 /* read the word at location addr in the USER area. */
1171 case PTRACE_PEEKUSR
: {
1175 if ((addr
& (sizeof(data
) - 1)) || addr
< 0 ||
1176 addr
>= sizeof(struct user
))
1179 tmp
= 0; /* Default return condition */
1180 if (addr
< sizeof(struct user_regs_struct
))
1181 tmp
= getreg(child
, addr
);
1182 else if (addr
>= offsetof(struct user
, u_debugreg
[0]) &&
1183 addr
<= offsetof(struct user
, u_debugreg
[7])) {
1184 addr
-= offsetof(struct user
, u_debugreg
[0]);
1185 tmp
= ptrace_get_debugreg(child
, addr
/ sizeof(data
));
1187 ret
= put_user(tmp
, datap
);
1191 case PTRACE_POKEUSR
: /* write the word at location addr in the USER area */
1193 if ((addr
& (sizeof(data
) - 1)) || addr
< 0 ||
1194 addr
>= sizeof(struct user
))
1197 if (addr
< sizeof(struct user_regs_struct
))
1198 ret
= putreg(child
, addr
, data
);
1199 else if (addr
>= offsetof(struct user
, u_debugreg
[0]) &&
1200 addr
<= offsetof(struct user
, u_debugreg
[7])) {
1201 addr
-= offsetof(struct user
, u_debugreg
[0]);
1202 ret
= ptrace_set_debugreg(child
,
1203 addr
/ sizeof(data
), data
);
1207 case PTRACE_GETREGS
: /* Get all gp regs from the child. */
1208 return copy_regset_to_user(child
,
1209 task_user_regset_view(current
),
1211 0, sizeof(struct user_regs_struct
),
1214 case PTRACE_SETREGS
: /* Set all gp regs in the child. */
1215 return copy_regset_from_user(child
,
1216 task_user_regset_view(current
),
1218 0, sizeof(struct user_regs_struct
),
1221 case PTRACE_GETFPREGS
: /* Get the child FPU state. */
1222 return copy_regset_to_user(child
,
1223 task_user_regset_view(current
),
1225 0, sizeof(struct user_i387_struct
),
1228 case PTRACE_SETFPREGS
: /* Set the child FPU state. */
1229 return copy_regset_from_user(child
,
1230 task_user_regset_view(current
),
1232 0, sizeof(struct user_i387_struct
),
1235 #ifdef CONFIG_X86_32
1236 case PTRACE_GETFPXREGS
: /* Get the child extended FPU state. */
1237 return copy_regset_to_user(child
, &user_x86_32_view
,
1239 0, sizeof(struct user_fxsr_struct
),
1242 case PTRACE_SETFPXREGS
: /* Set the child extended FPU state. */
1243 return copy_regset_from_user(child
, &user_x86_32_view
,
1245 0, sizeof(struct user_fxsr_struct
),
1249 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1250 case PTRACE_GET_THREAD_AREA
:
1253 ret
= do_get_thread_area(child
, addr
,
1254 (struct user_desc __user
*) data
);
1257 case PTRACE_SET_THREAD_AREA
:
1260 ret
= do_set_thread_area(child
, addr
,
1261 (struct user_desc __user
*) data
, 0);
1265 #ifdef CONFIG_X86_64
1266 /* normal 64bit interface to access TLS data.
1267 Works just like arch_prctl, except that the arguments
1269 case PTRACE_ARCH_PRCTL
:
1270 ret
= do_arch_prctl(child
, data
, addr
);
1275 * These bits need more cooking - not enabled yet:
1277 #ifdef CONFIG_X86_PTRACE_BTS
1278 case PTRACE_BTS_CONFIG
:
1279 ret
= ptrace_bts_config
1280 (child
, data
, (struct ptrace_bts_config __user
*)addr
);
1283 case PTRACE_BTS_STATUS
:
1284 ret
= ptrace_bts_status
1285 (child
, data
, (struct ptrace_bts_config __user
*)addr
);
1288 case PTRACE_BTS_SIZE
:
1289 ret
= ptrace_bts_size(child
);
1292 case PTRACE_BTS_GET
:
1293 ret
= ptrace_bts_read_record
1294 (child
, data
, (struct bts_struct __user
*) addr
);
1297 case PTRACE_BTS_CLEAR
:
1298 ret
= ptrace_bts_clear(child
);
1301 case PTRACE_BTS_DRAIN
:
1302 ret
= ptrace_bts_drain
1303 (child
, data
, (struct bts_struct __user
*) addr
);
1305 #endif /* CONFIG_X86_PTRACE_BTS */
1308 ret
= ptrace_request(child
, request
, addr
, data
);
1315 #ifdef CONFIG_IA32_EMULATION
1317 #include <linux/compat.h>
1318 #include <linux/syscalls.h>
1319 #include <asm/ia32.h>
1320 #include <asm/user32.h>
1323 case offsetof(struct user32, regs.l): \
1324 regs->q = value; break
1327 case offsetof(struct user32, regs.rs): \
1328 return set_segment_reg(child, \
1329 offsetof(struct user_regs_struct, rs), \
1333 static int putreg32(struct task_struct
*child
, unsigned regno
, u32 value
)
1335 struct pt_regs
*regs
= task_pt_regs(child
);
1356 case offsetof(struct user32
, regs
.orig_eax
):
1358 * A 32-bit debugger setting orig_eax means to restore
1359 * the state of the task restarting a 32-bit syscall.
1360 * Make sure we interpret the -ERESTART* codes correctly
1361 * in case the task is not actually still sitting at the
1362 * exit from a 32-bit syscall with TS_COMPAT still set.
1364 regs
->orig_ax
= value
;
1365 if (syscall_get_nr(child
, regs
) >= 0)
1366 task_thread_info(child
)->status
|= TS_COMPAT
;
1369 case offsetof(struct user32
, regs
.eflags
):
1370 return set_flags(child
, value
);
1372 case offsetof(struct user32
, u_debugreg
[0]) ...
1373 offsetof(struct user32
, u_debugreg
[7]):
1374 regno
-= offsetof(struct user32
, u_debugreg
[0]);
1375 return ptrace_set_debugreg(child
, regno
/ 4, value
);
1378 if (regno
> sizeof(struct user32
) || (regno
& 3))
1382 * Other dummy fields in the virtual user structure
1394 case offsetof(struct user32, regs.l): \
1395 *val = regs->q; break
1398 case offsetof(struct user32, regs.rs): \
1399 *val = get_segment_reg(child, \
1400 offsetof(struct user_regs_struct, rs)); \
1403 static int getreg32(struct task_struct
*child
, unsigned regno
, u32
*val
)
1405 struct pt_regs
*regs
= task_pt_regs(child
);
1423 R32(orig_eax
, orig_ax
);
1427 case offsetof(struct user32
, regs
.eflags
):
1428 *val
= get_flags(child
);
1431 case offsetof(struct user32
, u_debugreg
[0]) ...
1432 offsetof(struct user32
, u_debugreg
[7]):
1433 regno
-= offsetof(struct user32
, u_debugreg
[0]);
1434 *val
= ptrace_get_debugreg(child
, regno
/ 4);
1438 if (regno
> sizeof(struct user32
) || (regno
& 3))
1442 * Other dummy fields in the virtual user structure
1454 static int genregs32_get(struct task_struct
*target
,
1455 const struct user_regset
*regset
,
1456 unsigned int pos
, unsigned int count
,
1457 void *kbuf
, void __user
*ubuf
)
1460 compat_ulong_t
*k
= kbuf
;
1461 while (count
>= sizeof(*k
)) {
1462 getreg32(target
, pos
, k
++);
1463 count
-= sizeof(*k
);
1467 compat_ulong_t __user
*u
= ubuf
;
1468 while (count
>= sizeof(*u
)) {
1469 compat_ulong_t word
;
1470 getreg32(target
, pos
, &word
);
1471 if (__put_user(word
, u
++))
1473 count
-= sizeof(*u
);
1481 static int genregs32_set(struct task_struct
*target
,
1482 const struct user_regset
*regset
,
1483 unsigned int pos
, unsigned int count
,
1484 const void *kbuf
, const void __user
*ubuf
)
1488 const compat_ulong_t
*k
= kbuf
;
1489 while (count
>= sizeof(*k
) && !ret
) {
1490 ret
= putreg32(target
, pos
, *k
++);
1491 count
-= sizeof(*k
);
1495 const compat_ulong_t __user
*u
= ubuf
;
1496 while (count
>= sizeof(*u
) && !ret
) {
1497 compat_ulong_t word
;
1498 ret
= __get_user(word
, u
++);
1501 ret
= putreg32(target
, pos
, word
);
1502 count
-= sizeof(*u
);
1509 long compat_arch_ptrace(struct task_struct
*child
, compat_long_t request
,
1510 compat_ulong_t caddr
, compat_ulong_t cdata
)
1512 unsigned long addr
= caddr
;
1513 unsigned long data
= cdata
;
1514 void __user
*datap
= compat_ptr(data
);
1519 case PTRACE_PEEKUSR
:
1520 ret
= getreg32(child
, addr
, &val
);
1522 ret
= put_user(val
, (__u32 __user
*)datap
);
1525 case PTRACE_POKEUSR
:
1526 ret
= putreg32(child
, addr
, data
);
1529 case PTRACE_GETREGS
: /* Get all gp regs from the child. */
1530 return copy_regset_to_user(child
, &user_x86_32_view
,
1532 0, sizeof(struct user_regs_struct32
),
1535 case PTRACE_SETREGS
: /* Set all gp regs in the child. */
1536 return copy_regset_from_user(child
, &user_x86_32_view
,
1538 sizeof(struct user_regs_struct32
),
1541 case PTRACE_GETFPREGS
: /* Get the child FPU state. */
1542 return copy_regset_to_user(child
, &user_x86_32_view
,
1544 sizeof(struct user_i387_ia32_struct
),
1547 case PTRACE_SETFPREGS
: /* Set the child FPU state. */
1548 return copy_regset_from_user(
1549 child
, &user_x86_32_view
, REGSET_FP
,
1550 0, sizeof(struct user_i387_ia32_struct
), datap
);
1552 case PTRACE_GETFPXREGS
: /* Get the child extended FPU state. */
1553 return copy_regset_to_user(child
, &user_x86_32_view
,
1555 sizeof(struct user32_fxsr_struct
),
1558 case PTRACE_SETFPXREGS
: /* Set the child extended FPU state. */
1559 return copy_regset_from_user(child
, &user_x86_32_view
,
1561 sizeof(struct user32_fxsr_struct
),
1564 case PTRACE_GET_THREAD_AREA
:
1565 case PTRACE_SET_THREAD_AREA
:
1566 #ifdef CONFIG_X86_PTRACE_BTS
1567 case PTRACE_BTS_CONFIG
:
1568 case PTRACE_BTS_STATUS
:
1569 case PTRACE_BTS_SIZE
:
1570 case PTRACE_BTS_GET
:
1571 case PTRACE_BTS_CLEAR
:
1572 case PTRACE_BTS_DRAIN
:
1573 #endif /* CONFIG_X86_PTRACE_BTS */
1574 return arch_ptrace(child
, request
, addr
, data
);
1577 return compat_ptrace_request(child
, request
, addr
, data
);
1583 #endif /* CONFIG_IA32_EMULATION */
1585 #ifdef CONFIG_X86_64
1587 static const struct user_regset x86_64_regsets
[] = {
1588 [REGSET_GENERAL
] = {
1589 .core_note_type
= NT_PRSTATUS
,
1590 .n
= sizeof(struct user_regs_struct
) / sizeof(long),
1591 .size
= sizeof(long), .align
= sizeof(long),
1592 .get
= genregs_get
, .set
= genregs_set
1595 .core_note_type
= NT_PRFPREG
,
1596 .n
= sizeof(struct user_i387_struct
) / sizeof(long),
1597 .size
= sizeof(long), .align
= sizeof(long),
1598 .active
= xfpregs_active
, .get
= xfpregs_get
, .set
= xfpregs_set
1600 [REGSET_IOPERM64
] = {
1601 .core_note_type
= NT_386_IOPERM
,
1602 .n
= IO_BITMAP_LONGS
,
1603 .size
= sizeof(long), .align
= sizeof(long),
1604 .active
= ioperm_active
, .get
= ioperm_get
1608 static const struct user_regset_view user_x86_64_view
= {
1609 .name
= "x86_64", .e_machine
= EM_X86_64
,
1610 .regsets
= x86_64_regsets
, .n
= ARRAY_SIZE(x86_64_regsets
)
1613 #else /* CONFIG_X86_32 */
1615 #define user_regs_struct32 user_regs_struct
1616 #define genregs32_get genregs_get
1617 #define genregs32_set genregs_set
1619 #define user_i387_ia32_struct user_i387_struct
1620 #define user32_fxsr_struct user_fxsr_struct
1622 #endif /* CONFIG_X86_64 */
1624 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1625 static const struct user_regset x86_32_regsets
[] = {
1626 [REGSET_GENERAL
] = {
1627 .core_note_type
= NT_PRSTATUS
,
1628 .n
= sizeof(struct user_regs_struct32
) / sizeof(u32
),
1629 .size
= sizeof(u32
), .align
= sizeof(u32
),
1630 .get
= genregs32_get
, .set
= genregs32_set
1633 .core_note_type
= NT_PRFPREG
,
1634 .n
= sizeof(struct user_i387_ia32_struct
) / sizeof(u32
),
1635 .size
= sizeof(u32
), .align
= sizeof(u32
),
1636 .active
= fpregs_active
, .get
= fpregs_get
, .set
= fpregs_set
1639 .core_note_type
= NT_PRXFPREG
,
1640 .n
= sizeof(struct user32_fxsr_struct
) / sizeof(u32
),
1641 .size
= sizeof(u32
), .align
= sizeof(u32
),
1642 .active
= xfpregs_active
, .get
= xfpregs_get
, .set
= xfpregs_set
1645 .core_note_type
= NT_386_TLS
,
1646 .n
= GDT_ENTRY_TLS_ENTRIES
, .bias
= GDT_ENTRY_TLS_MIN
,
1647 .size
= sizeof(struct user_desc
),
1648 .align
= sizeof(struct user_desc
),
1649 .active
= regset_tls_active
,
1650 .get
= regset_tls_get
, .set
= regset_tls_set
1652 [REGSET_IOPERM32
] = {
1653 .core_note_type
= NT_386_IOPERM
,
1654 .n
= IO_BITMAP_BYTES
/ sizeof(u32
),
1655 .size
= sizeof(u32
), .align
= sizeof(u32
),
1656 .active
= ioperm_active
, .get
= ioperm_get
1660 static const struct user_regset_view user_x86_32_view
= {
1661 .name
= "i386", .e_machine
= EM_386
,
1662 .regsets
= x86_32_regsets
, .n
= ARRAY_SIZE(x86_32_regsets
)
1666 const struct user_regset_view
*task_user_regset_view(struct task_struct
*task
)
1668 #ifdef CONFIG_IA32_EMULATION
1669 if (test_tsk_thread_flag(task
, TIF_IA32
))
1671 #if defined CONFIG_X86_32 || defined CONFIG_IA32_EMULATION
1672 return &user_x86_32_view
;
1674 #ifdef CONFIG_X86_64
1675 return &user_x86_64_view
;
1679 static void fill_sigtrap_info(struct task_struct
*tsk
,
1680 struct pt_regs
*regs
,
1681 int error_code
, int si_code
,
1682 struct siginfo
*info
)
1684 tsk
->thread
.trap_no
= 1;
1685 tsk
->thread
.error_code
= error_code
;
1687 memset(info
, 0, sizeof(*info
));
1688 info
->si_signo
= SIGTRAP
;
1689 info
->si_code
= si_code
;
1690 info
->si_addr
= user_mode_vm(regs
) ? (void __user
*)regs
->ip
: NULL
;
1693 void user_single_step_siginfo(struct task_struct
*tsk
,
1694 struct pt_regs
*regs
,
1695 struct siginfo
*info
)
1697 fill_sigtrap_info(tsk
, regs
, 0, TRAP_BRKPT
, info
);
1700 void send_sigtrap(struct task_struct
*tsk
, struct pt_regs
*regs
,
1701 int error_code
, int si_code
)
1703 struct siginfo info
;
1705 fill_sigtrap_info(tsk
, regs
, error_code
, si_code
, &info
);
1706 /* Send us the fake SIGTRAP */
1707 force_sig_info(SIGTRAP
, &info
, tsk
);
1711 #ifdef CONFIG_X86_32
1713 #elif defined CONFIG_IA32_EMULATION
1714 # define IS_IA32 is_compat_task()
1720 * We must return the syscall number to actually look up in the table.
1721 * This can be -1L to skip running any syscall at all.
1723 asmregparm
long syscall_trace_enter(struct pt_regs
*regs
)
1728 * If we stepped into a sysenter/syscall insn, it trapped in
1729 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
1730 * If user-mode had set TF itself, then it's still clear from
1731 * do_debug() and we need to set it again to restore the user
1732 * state. If we entered on the slow path, TF was already set.
1734 if (test_thread_flag(TIF_SINGLESTEP
))
1735 regs
->flags
|= X86_EFLAGS_TF
;
1737 /* do the secure computing check first */
1738 secure_computing(regs
->orig_ax
);
1740 if (unlikely(test_thread_flag(TIF_SYSCALL_EMU
)))
1743 if ((ret
|| test_thread_flag(TIF_SYSCALL_TRACE
)) &&
1744 tracehook_report_syscall_entry(regs
))
1747 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT
)))
1748 trace_sys_enter(regs
, regs
->orig_ax
);
1750 if (unlikely(current
->audit_context
)) {
1752 audit_syscall_entry(AUDIT_ARCH_I386
,
1755 regs
->dx
, regs
->si
);
1756 #ifdef CONFIG_X86_64
1758 audit_syscall_entry(AUDIT_ARCH_X86_64
,
1761 regs
->dx
, regs
->r10
);
1765 return ret
?: regs
->orig_ax
;
1768 asmregparm
void syscall_trace_leave(struct pt_regs
*regs
)
1772 if (unlikely(current
->audit_context
))
1773 audit_syscall_exit(AUDITSC_RESULT(regs
->ax
), regs
->ax
);
1775 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT
)))
1776 trace_sys_exit(regs
, regs
->ax
);
1779 * If TIF_SYSCALL_EMU is set, we only get here because of
1780 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
1781 * We already reported this syscall instruction in
1782 * syscall_trace_enter().
1784 step
= unlikely(test_thread_flag(TIF_SINGLESTEP
)) &&
1785 !test_thread_flag(TIF_SYSCALL_EMU
);
1786 if (step
|| test_thread_flag(TIF_SYSCALL_TRACE
))
1787 tracehook_report_syscall_exit(regs
, step
);