1 // SPDX-License-Identifier: GPL-2.0
3 * Kernel support for the ptrace() and syscall tracing interfaces.
5 * Copyright (C) 2000 Hewlett-Packard Co, Linuxcare Inc.
6 * Copyright (C) 2000 Matthew Wilcox <matthew@wil.cx>
7 * Copyright (C) 2000 David Huggins-Daines <dhd@debian.org>
8 * Copyright (C) 2008-2016 Helge Deller <deller@gmx.de>
11 #include <linux/kernel.h>
12 #include <linux/sched.h>
14 #include <linux/smp.h>
15 #include <linux/elf.h>
16 #include <linux/errno.h>
17 #include <linux/ptrace.h>
18 #include <linux/user.h>
19 #include <linux/personality.h>
20 #include <linux/regset.h>
21 #include <linux/security.h>
22 #include <linux/seccomp.h>
23 #include <linux/compat.h>
24 #include <linux/signal.h>
25 #include <linux/audit.h>
27 #include <linux/uaccess.h>
28 #include <asm/processor.h>
29 #include <asm/asm-offsets.h>
31 /* PSW bits we allow the debugger to modify */
32 #define USER_PSW_BITS (PSW_N | PSW_B | PSW_V | PSW_CB)
34 #define CREATE_TRACE_POINTS
35 #include <trace/events/syscalls.h>
38 * These are our native regset flavors.
46 * Called by kernel/ptrace.c when detaching..
48 * Make sure single step bits etc are not set.
50 void ptrace_disable(struct task_struct *task)
52 clear_tsk_thread_flag(task, TIF_SINGLESTEP);
53 clear_tsk_thread_flag(task, TIF_BLOCKSTEP);
55 /* make sure the trap bits are not set */
63 * The following functions are called by ptrace_resume() when
64 * enabling or disabling single/block tracing.
66 void user_disable_single_step(struct task_struct *task)
71 void user_enable_single_step(struct task_struct *task)
73 clear_tsk_thread_flag(task, TIF_BLOCKSTEP);
74 set_tsk_thread_flag(task, TIF_SINGLESTEP);
76 if (pa_psw(task)->n) {
77 /* Nullified, just crank over the queue. */
78 task_regs(task)->iaoq[0] = task_regs(task)->iaoq[1];
79 task_regs(task)->iasq[0] = task_regs(task)->iasq[1];
80 task_regs(task)->iaoq[1] = task_regs(task)->iaoq[0] + 4;
87 /* Don't wake up the task, but let the
88 parent know something happened. */
89 force_sig_fault_to_task(SIGTRAP, TRAP_TRACE,
90 (void __user *) (task_regs(task)->iaoq[0] & ~3),
92 /* notify_parent(task, SIGCHLD); */
96 /* Enable recovery counter traps. The recovery counter
97 * itself will be set to zero on a task switch. If the
98 * task is suspended on a syscall then the syscall return
99 * path will overwrite the recovery counter with a suitable
100 * value such that it traps once back in user space. We
101 * disable interrupts in the tasks PSW here also, to avoid
102 * interrupts while the recovery counter is decrementing.
110 void user_enable_block_step(struct task_struct *task)
112 clear_tsk_thread_flag(task, TIF_SINGLESTEP);
113 set_tsk_thread_flag(task, TIF_BLOCKSTEP);
115 /* Enable taken branch trap. */
122 long arch_ptrace(struct task_struct *child, long request,
123 unsigned long addr, unsigned long data)
125 unsigned long __user *datap = (unsigned long __user *)data;
131 /* Read the word at location addr in the USER area. For ptraced
132 processes, the kernel saves all regs on a syscall. */
134 if ((addr & (sizeof(unsigned long)-1)) ||
135 addr >= sizeof(struct pt_regs))
137 tmp = *(unsigned long *) ((char *) task_regs(child) + addr);
138 ret = put_user(tmp, datap);
141 /* Write the word at location addr in the USER area. This will need
142 to change when the kernel no longer saves all regs on a syscall.
143 FIXME. There is a problem at the moment in that r3-r18 are only
144 saved if the process is ptraced on syscall entry, and even then
145 those values are overwritten by actual register values on syscall
148 /* Some register values written here may be ignored in
149 * entry.S:syscall_restore_rfi; e.g. iaoq is written with
150 * r31/r31+4, and not with the values in pt_regs.
152 if (addr == PT_PSW) {
153 /* Allow writing to Nullify, Divide-step-correction,
154 * and carry/borrow bits.
155 * BEWARE, if you set N, and then single step, it won't
156 * stop on the nullified instruction.
158 data &= USER_PSW_BITS;
159 task_regs(child)->gr[0] &= ~USER_PSW_BITS;
160 task_regs(child)->gr[0] |= data;
165 if ((addr & (sizeof(unsigned long)-1)) ||
166 addr >= sizeof(struct pt_regs))
168 if (addr == PT_IAOQ0 || addr == PT_IAOQ1) {
169 data |= 3; /* ensure userspace privilege */
171 if ((addr >= PT_GR1 && addr <= PT_GR31) ||
172 addr == PT_IAOQ0 || addr == PT_IAOQ1 ||
173 (addr >= PT_FR0 && addr <= PT_FR31 + 4) ||
175 *(unsigned long *) ((char *) task_regs(child) + addr) = data;
180 case PTRACE_GETREGS: /* Get all gp regs from the child. */
181 return copy_regset_to_user(child,
182 task_user_regset_view(current),
184 0, sizeof(struct user_regs_struct),
187 case PTRACE_SETREGS: /* Set all gp regs in the child. */
188 return copy_regset_from_user(child,
189 task_user_regset_view(current),
191 0, sizeof(struct user_regs_struct),
194 case PTRACE_GETFPREGS: /* Get the child FPU state. */
195 return copy_regset_to_user(child,
196 task_user_regset_view(current),
198 0, sizeof(struct user_fp_struct),
201 case PTRACE_SETFPREGS: /* Set the child FPU state. */
202 return copy_regset_from_user(child,
203 task_user_regset_view(current),
205 0, sizeof(struct user_fp_struct),
209 ret = ptrace_request(child, request, addr, data);
219 /* This function is needed to translate 32 bit pt_regs offsets in to
220 * 64 bit pt_regs offsets. For example, a 32 bit gdb under a 64 bit kernel
221 * will request offset 12 if it wants gr3, but the lower 32 bits of
222 * the 64 bit kernels view of gr3 will be at offset 28 (3*8 + 4).
223 * This code relies on a 32 bit pt_regs being comprised of 32 bit values
224 * except for the fp registers which (a) are 64 bits, and (b) follow
225 * the gr registers at the start of pt_regs. The 32 bit pt_regs should
226 * be half the size of the 64 bit pt_regs, plus 32*4 to allow for fr[]
227 * being 64 bit in both cases.
230 static compat_ulong_t translate_usr_offset(compat_ulong_t offset)
234 if (offset < 32*4) /* gr[0..31] */
235 pos = offset * 2 + 4;
236 else if (offset < 32*4+32*8) /* fr[0] ... fr[31] */
237 pos = (offset - 32*4) + PT_FR0;
238 else if (offset < sizeof(struct pt_regs)/2 + 32*4) /* sr[0] ... ipsw */
239 pos = (offset - 32*4 - 32*8) * 2 + PT_SR0 + 4;
241 pos = sizeof(struct pt_regs);
246 long compat_arch_ptrace(struct task_struct *child, compat_long_t request,
247 compat_ulong_t addr, compat_ulong_t data)
255 if (addr & (sizeof(compat_uint_t)-1))
257 addr = translate_usr_offset(addr);
258 if (addr >= sizeof(struct pt_regs))
261 tmp = *(compat_uint_t *) ((char *) task_regs(child) + addr);
262 ret = put_user(tmp, (compat_uint_t *) (unsigned long) data);
265 /* Write the word at location addr in the USER area. This will need
266 to change when the kernel no longer saves all regs on a syscall.
267 FIXME. There is a problem at the moment in that r3-r18 are only
268 saved if the process is ptraced on syscall entry, and even then
269 those values are overwritten by actual register values on syscall
272 /* Some register values written here may be ignored in
273 * entry.S:syscall_restore_rfi; e.g. iaoq is written with
274 * r31/r31+4, and not with the values in pt_regs.
276 if (addr == PT_PSW) {
277 /* Since PT_PSW==0, it is valid for 32 bit processes
278 * under 64 bit kernels as well.
280 ret = arch_ptrace(child, request, addr, data);
282 if (addr & (sizeof(compat_uint_t)-1))
284 addr = translate_usr_offset(addr);
285 if (addr >= sizeof(struct pt_regs))
287 if (addr == PT_IAOQ0+4 || addr == PT_IAOQ1+4) {
288 data |= 3; /* ensure userspace privilege */
290 if (addr >= PT_FR0 && addr <= PT_FR31 + 4) {
291 /* Special case, fp regs are 64 bits anyway */
292 *(__u32 *) ((char *) task_regs(child) + addr) = data;
295 else if ((addr >= PT_GR1+4 && addr <= PT_GR31+4) ||
296 addr == PT_IAOQ0+4 || addr == PT_IAOQ1+4 ||
298 /* Zero the top 32 bits */
299 *(__u32 *) ((char *) task_regs(child) + addr - 4) = 0;
300 *(__u32 *) ((char *) task_regs(child) + addr) = data;
307 ret = compat_ptrace_request(child, request, addr, data);
315 long do_syscall_trace_enter(struct pt_regs *regs)
317 if (test_thread_flag(TIF_SYSCALL_TRACE)) {
318 int rc = ptrace_report_syscall_entry(regs);
321 * As tracesys_next does not set %r28 to -ENOSYS
322 * when %r20 is set to -1, initialize it here.
324 regs->gr[28] = -ENOSYS;
328 * A nonzero return code from
329 * ptrace_report_syscall_entry() tells us
330 * to prevent the syscall execution. Skip
331 * the syscall call and the syscall restart handling.
333 * Note that the tracer may also just change
334 * regs->gr[20] to an invalid syscall number,
335 * that is handled by tracesys_next.
342 /* Do the secure computing check after ptrace. */
343 if (secure_computing() == -1)
346 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
347 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
348 trace_sys_enter(regs, regs->gr[20]);
352 if (!is_compat_task())
353 audit_syscall_entry(regs->gr[20], regs->gr[26], regs->gr[25],
354 regs->gr[24], regs->gr[23]);
357 audit_syscall_entry(regs->gr[20] & 0xffffffff,
358 regs->gr[26] & 0xffffffff,
359 regs->gr[25] & 0xffffffff,
360 regs->gr[24] & 0xffffffff,
361 regs->gr[23] & 0xffffffff);
364 * Sign extend the syscall number to 64bit since it may have been
365 * modified by a compat ptrace call
367 return (int) ((u32) regs->gr[20]);
370 void do_syscall_trace_exit(struct pt_regs *regs)
372 int stepping = test_thread_flag(TIF_SINGLESTEP) ||
373 test_thread_flag(TIF_BLOCKSTEP);
375 audit_syscall_exit(regs);
377 #ifdef CONFIG_HAVE_SYSCALL_TRACEPOINTS
378 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
379 trace_sys_exit(regs, regs->gr[20]);
382 if (stepping || test_thread_flag(TIF_SYSCALL_TRACE))
383 ptrace_report_syscall_exit(regs, stepping);
391 static int fpr_get(struct task_struct *target,
392 const struct user_regset *regset,
395 struct pt_regs *regs = task_regs(target);
397 return membuf_write(&to, regs->fr, ELF_NFPREG * sizeof(__u64));
400 static int fpr_set(struct task_struct *target,
401 const struct user_regset *regset,
402 unsigned int pos, unsigned int count,
403 const void *kbuf, const void __user *ubuf)
405 struct pt_regs *regs = task_regs(target);
406 const __u64 *k = kbuf;
407 const __u64 __user *u = ubuf;
411 count /= sizeof(reg);
414 for (; count > 0 && pos < ELF_NFPREG; --count)
415 regs->fr[pos++] = *k++;
417 for (; count > 0 && pos < ELF_NFPREG; --count) {
418 if (__get_user(reg, u++))
420 regs->fr[pos++] = reg;
426 count *= sizeof(reg);
427 user_regset_copyin_ignore(&pos, &count, &kbuf, &ubuf,
428 ELF_NFPREG * sizeof(reg), -1);
432 #define RI(reg) (offsetof(struct user_regs_struct,reg) / sizeof(long))
434 static unsigned long get_reg(struct pt_regs *regs, int num)
437 case RI(gr[0]) ... RI(gr[31]): return regs->gr[num - RI(gr[0])];
438 case RI(sr[0]) ... RI(sr[7]): return regs->sr[num - RI(sr[0])];
439 case RI(iasq[0]): return regs->iasq[0];
440 case RI(iasq[1]): return regs->iasq[1];
441 case RI(iaoq[0]): return regs->iaoq[0];
442 case RI(iaoq[1]): return regs->iaoq[1];
443 case RI(sar): return regs->sar;
444 case RI(iir): return regs->iir;
445 case RI(isr): return regs->isr;
446 case RI(ior): return regs->ior;
447 case RI(ipsw): return regs->ipsw;
448 case RI(cr27): return regs->cr27;
449 case RI(cr0): return mfctl(0);
450 case RI(cr24): return mfctl(24);
451 case RI(cr25): return mfctl(25);
452 case RI(cr26): return mfctl(26);
453 case RI(cr28): return mfctl(28);
454 case RI(cr29): return mfctl(29);
455 case RI(cr30): return mfctl(30);
456 case RI(cr31): return mfctl(31);
457 case RI(cr8): return mfctl(8);
458 case RI(cr9): return mfctl(9);
459 case RI(cr12): return mfctl(12);
460 case RI(cr13): return mfctl(13);
461 case RI(cr10): return mfctl(10);
462 case RI(cr15): return mfctl(15);
467 static void set_reg(struct pt_regs *regs, int num, unsigned long val)
472 * Allow writing to Nullify, Divide-step-correction,
473 * and carry/borrow bits.
474 * BEWARE, if you set N, and then single step, it won't
475 * stop on the nullified instruction.
477 val &= USER_PSW_BITS;
478 regs->gr[0] &= ~USER_PSW_BITS;
481 case RI(gr[1]) ... RI(gr[31]):
482 regs->gr[num - RI(gr[0])] = val;
486 /* set 2 lowest bits to ensure userspace privilege: */
487 regs->iaoq[num - RI(iaoq[0])] = val | 3;
489 case RI(sar): regs->sar = val;
493 /* do not allow to change any of the following registers (yet) */
494 case RI(sr[0]) ... RI(sr[7]): return regs->sr[num - RI(sr[0])];
495 case RI(iasq[0]): return regs->iasq[0];
496 case RI(iasq[1]): return regs->iasq[1];
497 case RI(iir): return regs->iir;
498 case RI(isr): return regs->isr;
499 case RI(ior): return regs->ior;
500 case RI(ipsw): return regs->ipsw;
501 case RI(cr27): return regs->cr27;
502 case cr0, cr24, cr25, cr26, cr27, cr28, cr29, cr30, cr31;
503 case cr8, cr9, cr12, cr13, cr10, cr15;
508 static int gpr_get(struct task_struct *target,
509 const struct user_regset *regset,
512 struct pt_regs *regs = task_regs(target);
515 for (pos = 0; pos < ELF_NGREG; pos++)
516 membuf_store(&to, get_reg(regs, pos));
520 static int gpr_set(struct task_struct *target,
521 const struct user_regset *regset,
522 unsigned int pos, unsigned int count,
523 const void *kbuf, const void __user *ubuf)
525 struct pt_regs *regs = task_regs(target);
526 const unsigned long *k = kbuf;
527 const unsigned long __user *u = ubuf;
531 count /= sizeof(reg);
534 for (; count > 0 && pos < ELF_NGREG; --count)
535 set_reg(regs, pos++, *k++);
537 for (; count > 0 && pos < ELF_NGREG; --count) {
538 if (__get_user(reg, u++))
540 set_reg(regs, pos++, reg);
546 count *= sizeof(reg);
547 user_regset_copyin_ignore(&pos, &count, &kbuf, &ubuf,
548 ELF_NGREG * sizeof(reg), -1);
552 static const struct user_regset native_regsets[] = {
554 .core_note_type = NT_PRSTATUS, .n = ELF_NGREG,
555 .size = sizeof(long), .align = sizeof(long),
556 .regset_get = gpr_get, .set = gpr_set
559 .core_note_type = NT_PRFPREG, .n = ELF_NFPREG,
560 .size = sizeof(__u64), .align = sizeof(__u64),
561 .regset_get = fpr_get, .set = fpr_set
565 static const struct user_regset_view user_parisc_native_view = {
566 .name = "parisc", .e_machine = ELF_ARCH, .ei_osabi = ELFOSABI_LINUX,
567 .regsets = native_regsets, .n = ARRAY_SIZE(native_regsets)
571 static int gpr32_get(struct task_struct *target,
572 const struct user_regset *regset,
575 struct pt_regs *regs = task_regs(target);
578 for (pos = 0; pos < ELF_NGREG; pos++)
579 membuf_store(&to, (compat_ulong_t)get_reg(regs, pos));
584 static int gpr32_set(struct task_struct *target,
585 const struct user_regset *regset,
586 unsigned int pos, unsigned int count,
587 const void *kbuf, const void __user *ubuf)
589 struct pt_regs *regs = task_regs(target);
590 const compat_ulong_t *k = kbuf;
591 const compat_ulong_t __user *u = ubuf;
595 count /= sizeof(reg);
598 for (; count > 0 && pos < ELF_NGREG; --count)
599 set_reg(regs, pos++, *k++);
601 for (; count > 0 && pos < ELF_NGREG; --count) {
602 if (__get_user(reg, u++))
604 set_reg(regs, pos++, reg);
610 count *= sizeof(reg);
611 user_regset_copyin_ignore(&pos, &count, &kbuf, &ubuf,
612 ELF_NGREG * sizeof(reg), -1);
617 * These are the regset flavors matching the 32bit native set.
619 static const struct user_regset compat_regsets[] = {
621 .core_note_type = NT_PRSTATUS, .n = ELF_NGREG,
622 .size = sizeof(compat_long_t), .align = sizeof(compat_long_t),
623 .regset_get = gpr32_get, .set = gpr32_set
626 .core_note_type = NT_PRFPREG, .n = ELF_NFPREG,
627 .size = sizeof(__u64), .align = sizeof(__u64),
628 .regset_get = fpr_get, .set = fpr_set
632 static const struct user_regset_view user_parisc_compat_view = {
633 .name = "parisc", .e_machine = EM_PARISC, .ei_osabi = ELFOSABI_LINUX,
634 .regsets = compat_regsets, .n = ARRAY_SIZE(compat_regsets)
636 #endif /* CONFIG_64BIT */
638 const struct user_regset_view *task_user_regset_view(struct task_struct *task)
640 BUILD_BUG_ON(sizeof(struct user_regs_struct)/sizeof(long) != ELF_NGREG);
641 BUILD_BUG_ON(sizeof(struct user_fp_struct)/sizeof(__u64) != ELF_NFPREG);
643 if (is_compat_task())
644 return &user_parisc_compat_view;
646 return &user_parisc_native_view;
650 /* HAVE_REGS_AND_STACK_ACCESS_API feature */
652 struct pt_regs_offset {
657 #define REG_OFFSET_NAME(r) {.name = #r, .offset = offsetof(struct pt_regs, r)}
658 #define REG_OFFSET_INDEX(r,i) {.name = #r#i, .offset = offsetof(struct pt_regs, r[i])}
659 #define REG_OFFSET_END {.name = NULL, .offset = 0}
661 static const struct pt_regs_offset regoffset_table[] = {
662 REG_OFFSET_INDEX(gr,0),
663 REG_OFFSET_INDEX(gr,1),
664 REG_OFFSET_INDEX(gr,2),
665 REG_OFFSET_INDEX(gr,3),
666 REG_OFFSET_INDEX(gr,4),
667 REG_OFFSET_INDEX(gr,5),
668 REG_OFFSET_INDEX(gr,6),
669 REG_OFFSET_INDEX(gr,7),
670 REG_OFFSET_INDEX(gr,8),
671 REG_OFFSET_INDEX(gr,9),
672 REG_OFFSET_INDEX(gr,10),
673 REG_OFFSET_INDEX(gr,11),
674 REG_OFFSET_INDEX(gr,12),
675 REG_OFFSET_INDEX(gr,13),
676 REG_OFFSET_INDEX(gr,14),
677 REG_OFFSET_INDEX(gr,15),
678 REG_OFFSET_INDEX(gr,16),
679 REG_OFFSET_INDEX(gr,17),
680 REG_OFFSET_INDEX(gr,18),
681 REG_OFFSET_INDEX(gr,19),
682 REG_OFFSET_INDEX(gr,20),
683 REG_OFFSET_INDEX(gr,21),
684 REG_OFFSET_INDEX(gr,22),
685 REG_OFFSET_INDEX(gr,23),
686 REG_OFFSET_INDEX(gr,24),
687 REG_OFFSET_INDEX(gr,25),
688 REG_OFFSET_INDEX(gr,26),
689 REG_OFFSET_INDEX(gr,27),
690 REG_OFFSET_INDEX(gr,28),
691 REG_OFFSET_INDEX(gr,29),
692 REG_OFFSET_INDEX(gr,30),
693 REG_OFFSET_INDEX(gr,31),
694 REG_OFFSET_INDEX(sr,0),
695 REG_OFFSET_INDEX(sr,1),
696 REG_OFFSET_INDEX(sr,2),
697 REG_OFFSET_INDEX(sr,3),
698 REG_OFFSET_INDEX(sr,4),
699 REG_OFFSET_INDEX(sr,5),
700 REG_OFFSET_INDEX(sr,6),
701 REG_OFFSET_INDEX(sr,7),
702 REG_OFFSET_INDEX(iasq,0),
703 REG_OFFSET_INDEX(iasq,1),
704 REG_OFFSET_INDEX(iaoq,0),
705 REG_OFFSET_INDEX(iaoq,1),
706 REG_OFFSET_NAME(cr27),
707 REG_OFFSET_NAME(ksp),
708 REG_OFFSET_NAME(kpc),
709 REG_OFFSET_NAME(sar),
710 REG_OFFSET_NAME(iir),
711 REG_OFFSET_NAME(isr),
712 REG_OFFSET_NAME(ior),
713 REG_OFFSET_NAME(ipsw),
718 * regs_query_register_offset() - query register offset from its name
719 * @name: the name of a register
721 * regs_query_register_offset() returns the offset of a register in struct
722 * pt_regs from its name. If the name is invalid, this returns -EINVAL;
724 int regs_query_register_offset(const char *name)
726 const struct pt_regs_offset *roff;
727 for (roff = regoffset_table; roff->name != NULL; roff++)
728 if (!strcmp(roff->name, name))
734 * regs_query_register_name() - query register name from its offset
735 * @offset: the offset of a register in struct pt_regs.
737 * regs_query_register_name() returns the name of a register from its
738 * offset in struct pt_regs. If the @offset is invalid, this returns NULL;
740 const char *regs_query_register_name(unsigned int offset)
742 const struct pt_regs_offset *roff;
743 for (roff = regoffset_table; roff->name != NULL; roff++)
744 if (roff->offset == offset)
750 * regs_within_kernel_stack() - check the address in the stack
751 * @regs: pt_regs which contains kernel stack pointer.
752 * @addr: address which is checked.
754 * regs_within_kernel_stack() checks @addr is within the kernel stack page(s).
755 * If @addr is within the kernel stack, it returns true. If not, returns false.
757 int regs_within_kernel_stack(struct pt_regs *regs, unsigned long addr)
759 return ((addr & ~(THREAD_SIZE - 1)) ==
760 (kernel_stack_pointer(regs) & ~(THREAD_SIZE - 1)));
764 * regs_get_kernel_stack_nth() - get Nth entry of the stack
765 * @regs: pt_regs which contains kernel stack pointer.
766 * @n: stack entry number.
768 * regs_get_kernel_stack_nth() returns @n th entry of the kernel stack which
769 * is specified by @regs. If the @n th entry is NOT in the kernel stack,
772 unsigned long regs_get_kernel_stack_nth(struct pt_regs *regs, unsigned int n)
774 unsigned long *addr = (unsigned long *)kernel_stack_pointer(regs);
778 if (!regs_within_kernel_stack(regs, (unsigned long)addr))