1 // SPDX-License-Identifier: GPL-2.0-only
3 * linux/kernel/ptrace.c
5 * (C) Copyright 1999 Linus Torvalds
7 * Common interfaces for "ptrace()" which we do not want
8 * to continually duplicate across every architecture.
11 #include <linux/capability.h>
12 #include <linux/export.h>
13 #include <linux/sched.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/coredump.h>
16 #include <linux/sched/task.h>
17 #include <linux/errno.h>
19 #include <linux/highmem.h>
20 #include <linux/pagemap.h>
21 #include <linux/ptrace.h>
22 #include <linux/security.h>
23 #include <linux/signal.h>
24 #include <linux/uio.h>
25 #include <linux/audit.h>
26 #include <linux/pid_namespace.h>
27 #include <linux/syscalls.h>
28 #include <linux/uaccess.h>
29 #include <linux/regset.h>
30 #include <linux/hw_breakpoint.h>
31 #include <linux/cn_proc.h>
32 #include <linux/compat.h>
33 #include <linux/sched/signal.h>
34 #include <linux/minmax.h>
36 #include <asm/syscall.h> /* for syscall_get_* */
39 * Access another process' address space via ptrace.
40 * Source/target buffer must be kernel space,
41 * Do not walk the page table directly, use get_user_pages
43 int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
44 void *buf, int len, unsigned int gup_flags)
49 mm = get_task_mm(tsk);
54 (current != tsk->parent) ||
55 ((get_dumpable(mm) != SUID_DUMP_USER) &&
56 !ptracer_capable(tsk, mm->user_ns))) {
61 ret = __access_remote_vm(mm, addr, buf, len, gup_flags);
68 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
69 const struct cred *ptracer_cred)
71 BUG_ON(!list_empty(&child->ptrace_entry));
72 list_add(&child->ptrace_entry, &new_parent->ptraced);
73 child->parent = new_parent;
74 child->ptracer_cred = get_cred(ptracer_cred);
78 * ptrace a task: make the debugger its new parent and
79 * move it to the ptrace list.
81 * Must be called with the tasklist lock write-held.
83 static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
85 __ptrace_link(child, new_parent, current_cred());
89 * __ptrace_unlink - unlink ptracee and restore its execution state
90 * @child: ptracee to be unlinked
92 * Remove @child from the ptrace list, move it back to the original parent,
93 * and restore the execution state so that it conforms to the group stop
96 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
97 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
98 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
99 * If the ptracer is exiting, the ptracee can be in any state.
101 * After detach, the ptracee should be in a state which conforms to the
102 * group stop. If the group is stopped or in the process of stopping, the
103 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
104 * up from TASK_TRACED.
106 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
107 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
108 * to but in the opposite direction of what happens while attaching to a
109 * stopped task. However, in this direction, the intermediate RUNNING
110 * state is not hidden even from the current ptracer and if it immediately
111 * re-attaches and performs a WNOHANG wait(2), it may fail.
114 * write_lock_irq(tasklist_lock)
116 void __ptrace_unlink(struct task_struct *child)
118 const struct cred *old_cred;
119 BUG_ON(!child->ptrace);
121 clear_task_syscall_work(child, SYSCALL_TRACE);
122 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
123 clear_task_syscall_work(child, SYSCALL_EMU);
126 child->parent = child->real_parent;
127 list_del_init(&child->ptrace_entry);
128 old_cred = child->ptracer_cred;
129 child->ptracer_cred = NULL;
132 spin_lock(&child->sighand->siglock);
135 * Clear all pending traps and TRAPPING. TRAPPING should be
136 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
138 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
139 task_clear_jobctl_trapping(child);
142 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
145 if (!(child->flags & PF_EXITING) &&
146 (child->signal->flags & SIGNAL_STOP_STOPPED ||
147 child->signal->group_stop_count)) {
148 child->jobctl |= JOBCTL_STOP_PENDING;
151 * This is only possible if this thread was cloned by the
152 * traced task running in the stopped group, set the signal
153 * for the future reports.
154 * FIXME: we should change ptrace_init_task() to handle this
157 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
158 child->jobctl |= SIGSTOP;
162 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
163 * @child in the butt. Note that @resume should be used iff @child
164 * is in TASK_TRACED; otherwise, we might unduly disrupt
165 * TASK_KILLABLE sleeps.
167 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
168 ptrace_signal_wake_up(child, true);
170 spin_unlock(&child->sighand->siglock);
173 /* Ensure that nothing can wake it up, even SIGKILL */
174 static bool ptrace_freeze_traced(struct task_struct *task)
178 /* Lockless, nobody but us can set this flag */
179 if (task->jobctl & JOBCTL_LISTENING)
182 spin_lock_irq(&task->sighand->siglock);
183 if (task_is_traced(task) && !__fatal_signal_pending(task)) {
184 task->state = __TASK_TRACED;
187 spin_unlock_irq(&task->sighand->siglock);
192 static void ptrace_unfreeze_traced(struct task_struct *task)
194 if (task->state != __TASK_TRACED)
197 WARN_ON(!task->ptrace || task->parent != current);
200 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
201 * Recheck state under the lock to close this race.
203 spin_lock_irq(&task->sighand->siglock);
204 if (task->state == __TASK_TRACED) {
205 if (__fatal_signal_pending(task))
206 wake_up_state(task, __TASK_TRACED);
208 task->state = TASK_TRACED;
210 spin_unlock_irq(&task->sighand->siglock);
214 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
215 * @child: ptracee to check for
216 * @ignore_state: don't check whether @child is currently %TASK_TRACED
218 * Check whether @child is being ptraced by %current and ready for further
219 * ptrace operations. If @ignore_state is %false, @child also should be in
220 * %TASK_TRACED state and on return the child is guaranteed to be traced
221 * and not executing. If @ignore_state is %true, @child can be in any
225 * Grabs and releases tasklist_lock and @child->sighand->siglock.
228 * 0 on success, -ESRCH if %child is not ready.
230 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
235 * We take the read lock around doing both checks to close a
236 * possible race where someone else was tracing our child and
237 * detached between these two checks. After this locked check,
238 * we are sure that this is our traced child and that can only
239 * be changed by us so it's not changing right after this.
241 read_lock(&tasklist_lock);
242 if (child->ptrace && child->parent == current) {
243 WARN_ON(child->state == __TASK_TRACED);
245 * child->sighand can't be NULL, release_task()
246 * does ptrace_unlink() before __exit_signal().
248 if (ignore_state || ptrace_freeze_traced(child))
251 read_unlock(&tasklist_lock);
253 if (!ret && !ignore_state) {
254 if (!wait_task_inactive(child, __TASK_TRACED)) {
256 * This can only happen if may_ptrace_stop() fails and
257 * ptrace_stop() changes ->state back to TASK_RUNNING,
258 * so we should not worry about leaking __TASK_TRACED.
260 WARN_ON(child->state == __TASK_TRACED);
268 static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
270 if (mode & PTRACE_MODE_NOAUDIT)
271 return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
272 return ns_capable(ns, CAP_SYS_PTRACE);
275 /* Returns 0 on success, -errno on denial. */
276 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
278 const struct cred *cred = current_cred(), *tcred;
279 struct mm_struct *mm;
283 if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
284 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
288 /* May we inspect the given task?
289 * This check is used both for attaching with ptrace
290 * and for allowing access to sensitive information in /proc.
292 * ptrace_attach denies several cases that /proc allows
293 * because setting up the necessary parent/child relationship
294 * or halting the specified task is impossible.
297 /* Don't let security modules deny introspection */
298 if (same_thread_group(task, current))
301 if (mode & PTRACE_MODE_FSCREDS) {
302 caller_uid = cred->fsuid;
303 caller_gid = cred->fsgid;
306 * Using the euid would make more sense here, but something
307 * in userland might rely on the old behavior, and this
308 * shouldn't be a security problem since
309 * PTRACE_MODE_REALCREDS implies that the caller explicitly
310 * used a syscall that requests access to another process
311 * (and not a filesystem syscall to procfs).
313 caller_uid = cred->uid;
314 caller_gid = cred->gid;
316 tcred = __task_cred(task);
317 if (uid_eq(caller_uid, tcred->euid) &&
318 uid_eq(caller_uid, tcred->suid) &&
319 uid_eq(caller_uid, tcred->uid) &&
320 gid_eq(caller_gid, tcred->egid) &&
321 gid_eq(caller_gid, tcred->sgid) &&
322 gid_eq(caller_gid, tcred->gid))
324 if (ptrace_has_cap(tcred->user_ns, mode))
331 * If a task drops privileges and becomes nondumpable (through a syscall
332 * like setresuid()) while we are trying to access it, we must ensure
333 * that the dumpability is read after the credentials; otherwise,
334 * we may be able to attach to a task that we shouldn't be able to
335 * attach to (as if the task had dropped privileges without becoming
337 * Pairs with a write barrier in commit_creds().
342 ((get_dumpable(mm) != SUID_DUMP_USER) &&
343 !ptrace_has_cap(mm->user_ns, mode)))
346 return security_ptrace_access_check(task, mode);
349 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
353 err = __ptrace_may_access(task, mode);
358 static int ptrace_attach(struct task_struct *task, long request,
362 bool seize = (request == PTRACE_SEIZE);
369 if (flags & ~(unsigned long)PTRACE_O_MASK)
371 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
379 if (unlikely(task->flags & PF_KTHREAD))
381 if (same_thread_group(task, current))
385 * Protect exec's credential calculations against our interference;
386 * SUID, SGID and LSM creds get determined differently
389 retval = -ERESTARTNOINTR;
390 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
394 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
399 write_lock_irq(&tasklist_lock);
401 if (unlikely(task->exit_state))
402 goto unlock_tasklist;
404 goto unlock_tasklist;
408 task->ptrace = flags;
410 ptrace_link(task, current);
412 /* SEIZE doesn't trap tracee on attach */
414 send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
416 spin_lock(&task->sighand->siglock);
419 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
420 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
421 * will be cleared if the child completes the transition or any
422 * event which clears the group stop states happens. We'll wait
423 * for the transition to complete before returning from this
426 * This hides STOPPED -> RUNNING -> TRACED transition from the
427 * attaching thread but a different thread in the same group can
428 * still observe the transient RUNNING state. IOW, if another
429 * thread's WNOHANG wait(2) on the stopped tracee races against
430 * ATTACH, the wait(2) may fail due to the transient RUNNING.
432 * The following task_is_stopped() test is safe as both transitions
433 * in and out of STOPPED are protected by siglock.
435 if (task_is_stopped(task) &&
436 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
437 signal_wake_up_state(task, __TASK_STOPPED);
439 spin_unlock(&task->sighand->siglock);
443 write_unlock_irq(&tasklist_lock);
445 mutex_unlock(&task->signal->cred_guard_mutex);
449 * We do not bother to change retval or clear JOBCTL_TRAPPING
450 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
451 * not return to user-mode, it will exit and clear this bit in
452 * __ptrace_unlink() if it wasn't already cleared by the tracee;
453 * and until then nobody can ptrace this task.
455 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
456 proc_ptrace_connector(task, PTRACE_ATTACH);
463 * ptrace_traceme -- helper for PTRACE_TRACEME
465 * Performs checks and sets PT_PTRACED.
466 * Should be used by all ptrace implementations for PTRACE_TRACEME.
468 static int ptrace_traceme(void)
472 write_lock_irq(&tasklist_lock);
473 /* Are we already being traced? */
474 if (!current->ptrace) {
475 ret = security_ptrace_traceme(current->parent);
477 * Check PF_EXITING to ensure ->real_parent has not passed
478 * exit_ptrace(). Otherwise we don't report the error but
479 * pretend ->real_parent untraces us right after return.
481 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
482 current->ptrace = PT_PTRACED;
483 ptrace_link(current, current->real_parent);
486 write_unlock_irq(&tasklist_lock);
492 * Called with irqs disabled, returns true if childs should reap themselves.
494 static int ignoring_children(struct sighand_struct *sigh)
497 spin_lock(&sigh->siglock);
498 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
499 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
500 spin_unlock(&sigh->siglock);
505 * Called with tasklist_lock held for writing.
506 * Unlink a traced task, and clean it up if it was a traced zombie.
507 * Return true if it needs to be reaped with release_task().
508 * (We can't call release_task() here because we already hold tasklist_lock.)
510 * If it's a zombie, our attachedness prevented normal parent notification
511 * or self-reaping. Do notification now if it would have happened earlier.
512 * If it should reap itself, return true.
514 * If it's our own child, there is no notification to do. But if our normal
515 * children self-reap, then this child was prevented by ptrace and we must
516 * reap it now, in that case we must also wake up sub-threads sleeping in
519 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
525 if (p->exit_state != EXIT_ZOMBIE)
528 dead = !thread_group_leader(p);
530 if (!dead && thread_group_empty(p)) {
531 if (!same_thread_group(p->real_parent, tracer))
532 dead = do_notify_parent(p, p->exit_signal);
533 else if (ignoring_children(tracer->sighand)) {
534 __wake_up_parent(p, tracer);
538 /* Mark it as in the process of being reaped. */
540 p->exit_state = EXIT_DEAD;
544 static int ptrace_detach(struct task_struct *child, unsigned int data)
546 if (!valid_signal(data))
549 /* Architecture-specific hardware disable .. */
550 ptrace_disable(child);
552 write_lock_irq(&tasklist_lock);
554 * We rely on ptrace_freeze_traced(). It can't be killed and
555 * untraced by another thread, it can't be a zombie.
557 WARN_ON(!child->ptrace || child->exit_state);
559 * tasklist_lock avoids the race with wait_task_stopped(), see
560 * the comment in ptrace_resume().
562 child->exit_code = data;
563 __ptrace_detach(current, child);
564 write_unlock_irq(&tasklist_lock);
566 proc_ptrace_connector(child, PTRACE_DETACH);
572 * Detach all tasks we were using ptrace on. Called with tasklist held
575 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
577 struct task_struct *p, *n;
579 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
580 if (unlikely(p->ptrace & PT_EXITKILL))
581 send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
583 if (__ptrace_detach(tracer, p))
584 list_add(&p->ptrace_entry, dead);
588 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
594 int this_len, retval;
596 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
597 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
604 if (copy_to_user(dst, buf, retval))
614 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
620 int this_len, retval;
622 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
623 if (copy_from_user(buf, src, this_len))
625 retval = ptrace_access_vm(tsk, dst, buf, this_len,
626 FOLL_FORCE | FOLL_WRITE);
640 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
644 if (data & ~(unsigned long)PTRACE_O_MASK)
647 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
648 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
649 !IS_ENABLED(CONFIG_SECCOMP))
652 if (!capable(CAP_SYS_ADMIN))
655 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
656 current->ptrace & PT_SUSPEND_SECCOMP)
660 /* Avoid intermediate state when all opts are cleared */
661 flags = child->ptrace;
662 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
663 flags |= (data << PT_OPT_FLAG_SHIFT);
664 child->ptrace = flags;
669 static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
674 if (lock_task_sighand(child, &flags)) {
676 if (likely(child->last_siginfo != NULL)) {
677 copy_siginfo(info, child->last_siginfo);
680 unlock_task_sighand(child, &flags);
685 static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
690 if (lock_task_sighand(child, &flags)) {
692 if (likely(child->last_siginfo != NULL)) {
693 copy_siginfo(child->last_siginfo, info);
696 unlock_task_sighand(child, &flags);
701 static int ptrace_peek_siginfo(struct task_struct *child,
705 struct ptrace_peeksiginfo_args arg;
706 struct sigpending *pending;
710 ret = copy_from_user(&arg, (void __user *) addr,
711 sizeof(struct ptrace_peeksiginfo_args));
715 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
716 return -EINVAL; /* unknown flags */
721 /* Ensure arg.off fits in an unsigned long */
722 if (arg.off > ULONG_MAX)
725 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
726 pending = &child->signal->shared_pending;
728 pending = &child->pending;
730 for (i = 0; i < arg.nr; ) {
731 kernel_siginfo_t info;
732 unsigned long off = arg.off + i;
735 spin_lock_irq(&child->sighand->siglock);
736 list_for_each_entry(q, &pending->list, list) {
739 copy_siginfo(&info, &q->info);
743 spin_unlock_irq(&child->sighand->siglock);
745 if (!found) /* beyond the end of the list */
749 if (unlikely(in_compat_syscall())) {
750 compat_siginfo_t __user *uinfo = compat_ptr(data);
752 if (copy_siginfo_to_user32(uinfo, &info)) {
760 siginfo_t __user *uinfo = (siginfo_t __user *) data;
762 if (copy_siginfo_to_user(uinfo, &info)) {
768 data += sizeof(siginfo_t);
771 if (signal_pending(current))
784 static long ptrace_get_rseq_configuration(struct task_struct *task,
785 unsigned long size, void __user *data)
787 struct ptrace_rseq_configuration conf = {
788 .rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
789 .rseq_abi_size = sizeof(*task->rseq),
790 .signature = task->rseq_sig,
794 size = min_t(unsigned long, size, sizeof(conf));
795 if (copy_to_user(data, &conf, size))
801 #ifdef PTRACE_SINGLESTEP
802 #define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
804 #define is_singlestep(request) 0
807 #ifdef PTRACE_SINGLEBLOCK
808 #define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
810 #define is_singleblock(request) 0
814 #define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
816 #define is_sysemu_singlestep(request) 0
819 static int ptrace_resume(struct task_struct *child, long request,
824 if (!valid_signal(data))
827 if (request == PTRACE_SYSCALL)
828 set_task_syscall_work(child, SYSCALL_TRACE);
830 clear_task_syscall_work(child, SYSCALL_TRACE);
832 #if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
833 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
834 set_task_syscall_work(child, SYSCALL_EMU);
836 clear_task_syscall_work(child, SYSCALL_EMU);
839 if (is_singleblock(request)) {
840 if (unlikely(!arch_has_block_step()))
842 user_enable_block_step(child);
843 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
844 if (unlikely(!arch_has_single_step()))
846 user_enable_single_step(child);
848 user_disable_single_step(child);
852 * Change ->exit_code and ->state under siglock to avoid the race
853 * with wait_task_stopped() in between; a non-zero ->exit_code will
854 * wrongly look like another report from tracee.
856 * Note that we need siglock even if ->exit_code == data and/or this
857 * status was not reported yet, the new status must not be cleared by
858 * wait_task_stopped() after resume.
860 * If data == 0 we do not care if wait_task_stopped() reports the old
861 * status and clears the code too; this can't race with the tracee, it
862 * takes siglock after resume.
864 need_siglock = data && !thread_group_empty(current);
866 spin_lock_irq(&child->sighand->siglock);
867 child->exit_code = data;
868 wake_up_state(child, __TASK_TRACED);
870 spin_unlock_irq(&child->sighand->siglock);
875 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
877 static const struct user_regset *
878 find_regset(const struct user_regset_view *view, unsigned int type)
880 const struct user_regset *regset;
883 for (n = 0; n < view->n; ++n) {
884 regset = view->regsets + n;
885 if (regset->core_note_type == type)
892 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
895 const struct user_regset_view *view = task_user_regset_view(task);
896 const struct user_regset *regset = find_regset(view, type);
899 if (!regset || (kiov->iov_len % regset->size) != 0)
902 regset_no = regset - view->regsets;
903 kiov->iov_len = min(kiov->iov_len,
904 (__kernel_size_t) (regset->n * regset->size));
906 if (req == PTRACE_GETREGSET)
907 return copy_regset_to_user(task, view, regset_no, 0,
908 kiov->iov_len, kiov->iov_base);
910 return copy_regset_from_user(task, view, regset_no, 0,
911 kiov->iov_len, kiov->iov_base);
915 * This is declared in linux/regset.h and defined in machine-dependent
916 * code. We put the export here, near the primary machine-neutral use,
917 * to ensure no machine forgets it.
919 EXPORT_SYMBOL_GPL(task_user_regset_view);
922 ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
923 struct ptrace_syscall_info *info)
925 unsigned long args[ARRAY_SIZE(info->entry.args)];
928 info->op = PTRACE_SYSCALL_INFO_ENTRY;
929 info->entry.nr = syscall_get_nr(child, regs);
930 syscall_get_arguments(child, regs, args);
931 for (i = 0; i < ARRAY_SIZE(args); i++)
932 info->entry.args[i] = args[i];
934 /* args is the last field in struct ptrace_syscall_info.entry */
935 return offsetofend(struct ptrace_syscall_info, entry.args);
939 ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
940 struct ptrace_syscall_info *info)
943 * As struct ptrace_syscall_info.entry is currently a subset
944 * of struct ptrace_syscall_info.seccomp, it makes sense to
945 * initialize that subset using ptrace_get_syscall_info_entry().
946 * This can be reconsidered in the future if these structures
947 * diverge significantly enough.
949 ptrace_get_syscall_info_entry(child, regs, info);
950 info->op = PTRACE_SYSCALL_INFO_SECCOMP;
951 info->seccomp.ret_data = child->ptrace_message;
953 /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
954 return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
958 ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
959 struct ptrace_syscall_info *info)
961 info->op = PTRACE_SYSCALL_INFO_EXIT;
962 info->exit.rval = syscall_get_error(child, regs);
963 info->exit.is_error = !!info->exit.rval;
964 if (!info->exit.is_error)
965 info->exit.rval = syscall_get_return_value(child, regs);
967 /* is_error is the last field in struct ptrace_syscall_info.exit */
968 return offsetofend(struct ptrace_syscall_info, exit.is_error);
972 ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
975 struct pt_regs *regs = task_pt_regs(child);
976 struct ptrace_syscall_info info = {
977 .op = PTRACE_SYSCALL_INFO_NONE,
978 .arch = syscall_get_arch(child),
979 .instruction_pointer = instruction_pointer(regs),
980 .stack_pointer = user_stack_pointer(regs),
982 unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
983 unsigned long write_size;
986 * This does not need lock_task_sighand() to access
987 * child->last_siginfo because ptrace_freeze_traced()
988 * called earlier by ptrace_check_attach() ensures that
989 * the tracee cannot go away and clear its last_siginfo.
991 switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
993 switch (child->ptrace_message) {
994 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
995 actual_size = ptrace_get_syscall_info_entry(child, regs,
998 case PTRACE_EVENTMSG_SYSCALL_EXIT:
999 actual_size = ptrace_get_syscall_info_exit(child, regs,
1004 case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1005 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1010 write_size = min(actual_size, user_size);
1011 return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1013 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1015 int ptrace_request(struct task_struct *child, long request,
1016 unsigned long addr, unsigned long data)
1018 bool seized = child->ptrace & PT_SEIZED;
1020 kernel_siginfo_t siginfo, *si;
1021 void __user *datavp = (void __user *) data;
1022 unsigned long __user *datalp = datavp;
1023 unsigned long flags;
1026 case PTRACE_PEEKTEXT:
1027 case PTRACE_PEEKDATA:
1028 return generic_ptrace_peekdata(child, addr, data);
1029 case PTRACE_POKETEXT:
1030 case PTRACE_POKEDATA:
1031 return generic_ptrace_pokedata(child, addr, data);
1033 #ifdef PTRACE_OLDSETOPTIONS
1034 case PTRACE_OLDSETOPTIONS:
1036 case PTRACE_SETOPTIONS:
1037 ret = ptrace_setoptions(child, data);
1039 case PTRACE_GETEVENTMSG:
1040 ret = put_user(child->ptrace_message, datalp);
1043 case PTRACE_PEEKSIGINFO:
1044 ret = ptrace_peek_siginfo(child, addr, data);
1047 case PTRACE_GETSIGINFO:
1048 ret = ptrace_getsiginfo(child, &siginfo);
1050 ret = copy_siginfo_to_user(datavp, &siginfo);
1053 case PTRACE_SETSIGINFO:
1054 ret = copy_siginfo_from_user(&siginfo, datavp);
1056 ret = ptrace_setsiginfo(child, &siginfo);
1059 case PTRACE_GETSIGMASK: {
1062 if (addr != sizeof(sigset_t)) {
1067 if (test_tsk_restore_sigmask(child))
1068 mask = &child->saved_sigmask;
1070 mask = &child->blocked;
1072 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1080 case PTRACE_SETSIGMASK: {
1083 if (addr != sizeof(sigset_t)) {
1088 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1093 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1096 * Every thread does recalc_sigpending() after resume, so
1097 * retarget_shared_pending() and recalc_sigpending() are not
1100 spin_lock_irq(&child->sighand->siglock);
1101 child->blocked = new_set;
1102 spin_unlock_irq(&child->sighand->siglock);
1104 clear_tsk_restore_sigmask(child);
1110 case PTRACE_INTERRUPT:
1112 * Stop tracee without any side-effect on signal or job
1113 * control. At least one trap is guaranteed to happen
1114 * after this request. If @child is already trapped, the
1115 * current trap is not disturbed and another trap will
1116 * happen after the current trap is ended with PTRACE_CONT.
1118 * The actual trap might not be PTRACE_EVENT_STOP trap but
1119 * the pending condition is cleared regardless.
1121 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1125 * INTERRUPT doesn't disturb existing trap sans one
1126 * exception. If ptracer issued LISTEN for the current
1127 * STOP, this INTERRUPT should clear LISTEN and re-trap
1130 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1131 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1133 unlock_task_sighand(child, &flags);
1139 * Listen for events. Tracee must be in STOP. It's not
1140 * resumed per-se but is not considered to be in TRACED by
1141 * wait(2) or ptrace(2). If an async event (e.g. group
1142 * stop state change) happens, tracee will enter STOP trap
1143 * again. Alternatively, ptracer can issue INTERRUPT to
1144 * finish listening and re-trap tracee into STOP.
1146 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1149 si = child->last_siginfo;
1150 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1151 child->jobctl |= JOBCTL_LISTENING;
1153 * If NOTIFY is set, it means event happened between
1154 * start of this trap and now. Trigger re-trap.
1156 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1157 ptrace_signal_wake_up(child, true);
1160 unlock_task_sighand(child, &flags);
1163 case PTRACE_DETACH: /* detach a process that was attached. */
1164 ret = ptrace_detach(child, data);
1167 #ifdef CONFIG_BINFMT_ELF_FDPIC
1168 case PTRACE_GETFDPIC: {
1169 struct mm_struct *mm = get_task_mm(child);
1170 unsigned long tmp = 0;
1177 case PTRACE_GETFDPIC_EXEC:
1178 tmp = mm->context.exec_fdpic_loadmap;
1180 case PTRACE_GETFDPIC_INTERP:
1181 tmp = mm->context.interp_fdpic_loadmap;
1188 ret = put_user(tmp, datalp);
1193 #ifdef PTRACE_SINGLESTEP
1194 case PTRACE_SINGLESTEP:
1196 #ifdef PTRACE_SINGLEBLOCK
1197 case PTRACE_SINGLEBLOCK:
1199 #ifdef PTRACE_SYSEMU
1201 case PTRACE_SYSEMU_SINGLESTEP:
1203 case PTRACE_SYSCALL:
1205 return ptrace_resume(child, request, data);
1208 if (child->exit_state) /* already dead */
1210 return ptrace_resume(child, request, SIGKILL);
1212 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1213 case PTRACE_GETREGSET:
1214 case PTRACE_SETREGSET: {
1216 struct iovec __user *uiov = datavp;
1218 if (!access_ok(uiov, sizeof(*uiov)))
1221 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1222 __get_user(kiov.iov_len, &uiov->iov_len))
1225 ret = ptrace_regset(child, request, addr, &kiov);
1227 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1231 case PTRACE_GET_SYSCALL_INFO:
1232 ret = ptrace_get_syscall_info(child, addr, datavp);
1236 case PTRACE_SECCOMP_GET_FILTER:
1237 ret = seccomp_get_filter(child, addr, datavp);
1240 case PTRACE_SECCOMP_GET_METADATA:
1241 ret = seccomp_get_metadata(child, addr, datavp);
1245 case PTRACE_GET_RSEQ_CONFIGURATION:
1246 ret = ptrace_get_rseq_configuration(child, addr, datavp);
1257 #ifndef arch_ptrace_attach
1258 #define arch_ptrace_attach(child) do { } while (0)
1261 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1262 unsigned long, data)
1264 struct task_struct *child;
1267 if (request == PTRACE_TRACEME) {
1268 ret = ptrace_traceme();
1270 arch_ptrace_attach(current);
1274 child = find_get_task_by_vpid(pid);
1280 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1281 ret = ptrace_attach(child, request, addr, data);
1283 * Some architectures need to do book-keeping after
1287 arch_ptrace_attach(child);
1288 goto out_put_task_struct;
1291 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1292 request == PTRACE_INTERRUPT);
1294 goto out_put_task_struct;
1296 ret = arch_ptrace(child, request, addr, data);
1297 if (ret || request != PTRACE_DETACH)
1298 ptrace_unfreeze_traced(child);
1300 out_put_task_struct:
1301 put_task_struct(child);
1306 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1312 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1313 if (copied != sizeof(tmp))
1315 return put_user(tmp, (unsigned long __user *)data);
1318 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1323 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1324 FOLL_FORCE | FOLL_WRITE);
1325 return (copied == sizeof(data)) ? 0 : -EIO;
1328 #if defined CONFIG_COMPAT
1330 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1331 compat_ulong_t addr, compat_ulong_t data)
1333 compat_ulong_t __user *datap = compat_ptr(data);
1334 compat_ulong_t word;
1335 kernel_siginfo_t siginfo;
1339 case PTRACE_PEEKTEXT:
1340 case PTRACE_PEEKDATA:
1341 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1343 if (ret != sizeof(word))
1346 ret = put_user(word, datap);
1349 case PTRACE_POKETEXT:
1350 case PTRACE_POKEDATA:
1351 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1352 FOLL_FORCE | FOLL_WRITE);
1353 ret = (ret != sizeof(data) ? -EIO : 0);
1356 case PTRACE_GETEVENTMSG:
1357 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1360 case PTRACE_GETSIGINFO:
1361 ret = ptrace_getsiginfo(child, &siginfo);
1363 ret = copy_siginfo_to_user32(
1364 (struct compat_siginfo __user *) datap,
1368 case PTRACE_SETSIGINFO:
1369 ret = copy_siginfo_from_user32(
1370 &siginfo, (struct compat_siginfo __user *) datap);
1372 ret = ptrace_setsiginfo(child, &siginfo);
1374 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1375 case PTRACE_GETREGSET:
1376 case PTRACE_SETREGSET:
1379 struct compat_iovec __user *uiov =
1380 (struct compat_iovec __user *) datap;
1384 if (!access_ok(uiov, sizeof(*uiov)))
1387 if (__get_user(ptr, &uiov->iov_base) ||
1388 __get_user(len, &uiov->iov_len))
1391 kiov.iov_base = compat_ptr(ptr);
1394 ret = ptrace_regset(child, request, addr, &kiov);
1396 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1402 ret = ptrace_request(child, request, addr, data);
1408 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1409 compat_long_t, addr, compat_long_t, data)
1411 struct task_struct *child;
1414 if (request == PTRACE_TRACEME) {
1415 ret = ptrace_traceme();
1419 child = find_get_task_by_vpid(pid);
1425 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1426 ret = ptrace_attach(child, request, addr, data);
1428 * Some architectures need to do book-keeping after
1432 arch_ptrace_attach(child);
1433 goto out_put_task_struct;
1436 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1437 request == PTRACE_INTERRUPT);
1439 ret = compat_arch_ptrace(child, request, addr, data);
1440 if (ret || request != PTRACE_DETACH)
1441 ptrace_unfreeze_traced(child);
1444 out_put_task_struct:
1445 put_task_struct(child);
1449 #endif /* CONFIG_COMPAT */