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>
35 #include <asm/syscall.h> /* for syscall_get_* */
38 * Access another process' address space via ptrace.
39 * Source/target buffer must be kernel space,
40 * Do not walk the page table directly, use get_user_pages
42 int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
43 void *buf, int len, unsigned int gup_flags)
48 mm = get_task_mm(tsk);
53 (current != tsk->parent) ||
54 ((get_dumpable(mm) != SUID_DUMP_USER) &&
55 !ptracer_capable(tsk, mm->user_ns))) {
60 ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
67 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
68 const struct cred *ptracer_cred)
70 BUG_ON(!list_empty(&child->ptrace_entry));
71 list_add(&child->ptrace_entry, &new_parent->ptraced);
72 child->parent = new_parent;
73 child->ptracer_cred = get_cred(ptracer_cred);
77 * ptrace a task: make the debugger its new parent and
78 * move it to the ptrace list.
80 * Must be called with the tasklist lock write-held.
82 static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
84 __ptrace_link(child, new_parent, current_cred());
88 * __ptrace_unlink - unlink ptracee and restore its execution state
89 * @child: ptracee to be unlinked
91 * Remove @child from the ptrace list, move it back to the original parent,
92 * and restore the execution state so that it conforms to the group stop
95 * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
96 * exiting. For PTRACE_DETACH, unless the ptracee has been killed between
97 * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
98 * If the ptracer is exiting, the ptracee can be in any state.
100 * After detach, the ptracee should be in a state which conforms to the
101 * group stop. If the group is stopped or in the process of stopping, the
102 * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
103 * up from TASK_TRACED.
105 * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
106 * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
107 * to but in the opposite direction of what happens while attaching to a
108 * stopped task. However, in this direction, the intermediate RUNNING
109 * state is not hidden even from the current ptracer and if it immediately
110 * re-attaches and performs a WNOHANG wait(2), it may fail.
113 * write_lock_irq(tasklist_lock)
115 void __ptrace_unlink(struct task_struct *child)
117 const struct cred *old_cred;
118 BUG_ON(!child->ptrace);
120 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
121 #ifdef TIF_SYSCALL_EMU
122 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
125 child->parent = child->real_parent;
126 list_del_init(&child->ptrace_entry);
127 old_cred = child->ptracer_cred;
128 child->ptracer_cred = NULL;
131 spin_lock(&child->sighand->siglock);
134 * Clear all pending traps and TRAPPING. TRAPPING should be
135 * cleared regardless of JOBCTL_STOP_PENDING. Do it explicitly.
137 task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
138 task_clear_jobctl_trapping(child);
141 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
144 if (!(child->flags & PF_EXITING) &&
145 (child->signal->flags & SIGNAL_STOP_STOPPED ||
146 child->signal->group_stop_count)) {
147 child->jobctl |= JOBCTL_STOP_PENDING;
150 * This is only possible if this thread was cloned by the
151 * traced task running in the stopped group, set the signal
152 * for the future reports.
153 * FIXME: we should change ptrace_init_task() to handle this
156 if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
157 child->jobctl |= SIGSTOP;
161 * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
162 * @child in the butt. Note that @resume should be used iff @child
163 * is in TASK_TRACED; otherwise, we might unduly disrupt
164 * TASK_KILLABLE sleeps.
166 if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
167 ptrace_signal_wake_up(child, true);
169 spin_unlock(&child->sighand->siglock);
172 static bool looks_like_a_spurious_pid(struct task_struct *task)
174 if (task->exit_code != ((PTRACE_EVENT_EXEC << 8) | SIGTRAP))
177 if (task_pid_vnr(task) == task->ptrace_message)
180 * The tracee changed its pid but the PTRACE_EVENT_EXEC event
181 * was not wait()'ed, most probably debugger targets the old
182 * leader which was destroyed in de_thread().
187 /* Ensure that nothing can wake it up, even SIGKILL */
188 static bool ptrace_freeze_traced(struct task_struct *task)
192 /* Lockless, nobody but us can set this flag */
193 if (task->jobctl & JOBCTL_LISTENING)
196 spin_lock_irq(&task->sighand->siglock);
197 if (task_is_traced(task) && !looks_like_a_spurious_pid(task) &&
198 !__fatal_signal_pending(task)) {
199 task->state = __TASK_TRACED;
202 spin_unlock_irq(&task->sighand->siglock);
207 static void ptrace_unfreeze_traced(struct task_struct *task)
209 if (task->state != __TASK_TRACED)
212 WARN_ON(!task->ptrace || task->parent != current);
215 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
216 * Recheck state under the lock to close this race.
218 spin_lock_irq(&task->sighand->siglock);
219 if (task->state == __TASK_TRACED) {
220 if (__fatal_signal_pending(task))
221 wake_up_state(task, __TASK_TRACED);
223 task->state = TASK_TRACED;
225 spin_unlock_irq(&task->sighand->siglock);
229 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
230 * @child: ptracee to check for
231 * @ignore_state: don't check whether @child is currently %TASK_TRACED
233 * Check whether @child is being ptraced by %current and ready for further
234 * ptrace operations. If @ignore_state is %false, @child also should be in
235 * %TASK_TRACED state and on return the child is guaranteed to be traced
236 * and not executing. If @ignore_state is %true, @child can be in any
240 * Grabs and releases tasklist_lock and @child->sighand->siglock.
243 * 0 on success, -ESRCH if %child is not ready.
245 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
250 * We take the read lock around doing both checks to close a
251 * possible race where someone else was tracing our child and
252 * detached between these two checks. After this locked check,
253 * we are sure that this is our traced child and that can only
254 * be changed by us so it's not changing right after this.
256 read_lock(&tasklist_lock);
257 if (child->ptrace && child->parent == current) {
258 WARN_ON(child->state == __TASK_TRACED);
260 * child->sighand can't be NULL, release_task()
261 * does ptrace_unlink() before __exit_signal().
263 if (ignore_state || ptrace_freeze_traced(child))
266 read_unlock(&tasklist_lock);
268 if (!ret && !ignore_state) {
269 if (!wait_task_inactive(child, __TASK_TRACED)) {
271 * This can only happen if may_ptrace_stop() fails and
272 * ptrace_stop() changes ->state back to TASK_RUNNING,
273 * so we should not worry about leaking __TASK_TRACED.
275 WARN_ON(child->state == __TASK_TRACED);
283 static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
285 if (mode & PTRACE_MODE_NOAUDIT)
286 return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
287 return ns_capable(ns, CAP_SYS_PTRACE);
290 /* Returns 0 on success, -errno on denial. */
291 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
293 const struct cred *cred = current_cred(), *tcred;
294 struct mm_struct *mm;
298 if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
299 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
303 /* May we inspect the given task?
304 * This check is used both for attaching with ptrace
305 * and for allowing access to sensitive information in /proc.
307 * ptrace_attach denies several cases that /proc allows
308 * because setting up the necessary parent/child relationship
309 * or halting the specified task is impossible.
312 /* Don't let security modules deny introspection */
313 if (same_thread_group(task, current))
316 if (mode & PTRACE_MODE_FSCREDS) {
317 caller_uid = cred->fsuid;
318 caller_gid = cred->fsgid;
321 * Using the euid would make more sense here, but something
322 * in userland might rely on the old behavior, and this
323 * shouldn't be a security problem since
324 * PTRACE_MODE_REALCREDS implies that the caller explicitly
325 * used a syscall that requests access to another process
326 * (and not a filesystem syscall to procfs).
328 caller_uid = cred->uid;
329 caller_gid = cred->gid;
331 tcred = __task_cred(task);
332 if (uid_eq(caller_uid, tcred->euid) &&
333 uid_eq(caller_uid, tcred->suid) &&
334 uid_eq(caller_uid, tcred->uid) &&
335 gid_eq(caller_gid, tcred->egid) &&
336 gid_eq(caller_gid, tcred->sgid) &&
337 gid_eq(caller_gid, tcred->gid))
339 if (ptrace_has_cap(tcred->user_ns, mode))
346 * If a task drops privileges and becomes nondumpable (through a syscall
347 * like setresuid()) while we are trying to access it, we must ensure
348 * that the dumpability is read after the credentials; otherwise,
349 * we may be able to attach to a task that we shouldn't be able to
350 * attach to (as if the task had dropped privileges without becoming
352 * Pairs with a write barrier in commit_creds().
357 ((get_dumpable(mm) != SUID_DUMP_USER) &&
358 !ptrace_has_cap(mm->user_ns, mode)))
361 return security_ptrace_access_check(task, mode);
364 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
368 err = __ptrace_may_access(task, mode);
373 static int ptrace_attach(struct task_struct *task, long request,
377 bool seize = (request == PTRACE_SEIZE);
384 if (flags & ~(unsigned long)PTRACE_O_MASK)
386 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
394 if (unlikely(task->flags & PF_KTHREAD))
396 if (same_thread_group(task, current))
400 * Protect exec's credential calculations against our interference;
401 * SUID, SGID and LSM creds get determined differently
404 retval = -ERESTARTNOINTR;
405 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
409 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
414 write_lock_irq(&tasklist_lock);
416 if (unlikely(task->exit_state))
417 goto unlock_tasklist;
419 goto unlock_tasklist;
423 task->ptrace = flags;
425 ptrace_link(task, current);
427 /* SEIZE doesn't trap tracee on attach */
429 send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
431 spin_lock(&task->sighand->siglock);
434 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
435 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
436 * will be cleared if the child completes the transition or any
437 * event which clears the group stop states happens. We'll wait
438 * for the transition to complete before returning from this
441 * This hides STOPPED -> RUNNING -> TRACED transition from the
442 * attaching thread but a different thread in the same group can
443 * still observe the transient RUNNING state. IOW, if another
444 * thread's WNOHANG wait(2) on the stopped tracee races against
445 * ATTACH, the wait(2) may fail due to the transient RUNNING.
447 * The following task_is_stopped() test is safe as both transitions
448 * in and out of STOPPED are protected by siglock.
450 if (task_is_stopped(task) &&
451 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
452 signal_wake_up_state(task, __TASK_STOPPED);
454 spin_unlock(&task->sighand->siglock);
458 write_unlock_irq(&tasklist_lock);
460 mutex_unlock(&task->signal->cred_guard_mutex);
464 * We do not bother to change retval or clear JOBCTL_TRAPPING
465 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
466 * not return to user-mode, it will exit and clear this bit in
467 * __ptrace_unlink() if it wasn't already cleared by the tracee;
468 * and until then nobody can ptrace this task.
470 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
471 proc_ptrace_connector(task, PTRACE_ATTACH);
478 * ptrace_traceme -- helper for PTRACE_TRACEME
480 * Performs checks and sets PT_PTRACED.
481 * Should be used by all ptrace implementations for PTRACE_TRACEME.
483 static int ptrace_traceme(void)
487 write_lock_irq(&tasklist_lock);
488 /* Are we already being traced? */
489 if (!current->ptrace) {
490 ret = security_ptrace_traceme(current->parent);
492 * Check PF_EXITING to ensure ->real_parent has not passed
493 * exit_ptrace(). Otherwise we don't report the error but
494 * pretend ->real_parent untraces us right after return.
496 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
497 current->ptrace = PT_PTRACED;
498 ptrace_link(current, current->real_parent);
501 write_unlock_irq(&tasklist_lock);
507 * Called with irqs disabled, returns true if childs should reap themselves.
509 static int ignoring_children(struct sighand_struct *sigh)
512 spin_lock(&sigh->siglock);
513 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
514 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
515 spin_unlock(&sigh->siglock);
520 * Called with tasklist_lock held for writing.
521 * Unlink a traced task, and clean it up if it was a traced zombie.
522 * Return true if it needs to be reaped with release_task().
523 * (We can't call release_task() here because we already hold tasklist_lock.)
525 * If it's a zombie, our attachedness prevented normal parent notification
526 * or self-reaping. Do notification now if it would have happened earlier.
527 * If it should reap itself, return true.
529 * If it's our own child, there is no notification to do. But if our normal
530 * children self-reap, then this child was prevented by ptrace and we must
531 * reap it now, in that case we must also wake up sub-threads sleeping in
534 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
540 if (p->exit_state != EXIT_ZOMBIE)
543 dead = !thread_group_leader(p);
545 if (!dead && thread_group_empty(p)) {
546 if (!same_thread_group(p->real_parent, tracer))
547 dead = do_notify_parent(p, p->exit_signal);
548 else if (ignoring_children(tracer->sighand)) {
549 __wake_up_parent(p, tracer);
553 /* Mark it as in the process of being reaped. */
555 p->exit_state = EXIT_DEAD;
559 static int ptrace_detach(struct task_struct *child, unsigned int data)
561 if (!valid_signal(data))
564 /* Architecture-specific hardware disable .. */
565 ptrace_disable(child);
567 write_lock_irq(&tasklist_lock);
569 * We rely on ptrace_freeze_traced(). It can't be killed and
570 * untraced by another thread, it can't be a zombie.
572 WARN_ON(!child->ptrace || child->exit_state);
574 * tasklist_lock avoids the race with wait_task_stopped(), see
575 * the comment in ptrace_resume().
577 child->exit_code = data;
578 __ptrace_detach(current, child);
579 write_unlock_irq(&tasklist_lock);
581 proc_ptrace_connector(child, PTRACE_DETACH);
587 * Detach all tasks we were using ptrace on. Called with tasklist held
590 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
592 struct task_struct *p, *n;
594 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
595 if (unlikely(p->ptrace & PT_EXITKILL))
596 send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
598 if (__ptrace_detach(tracer, p))
599 list_add(&p->ptrace_entry, dead);
603 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
609 int this_len, retval;
611 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
612 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
619 if (copy_to_user(dst, buf, retval))
629 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
635 int this_len, retval;
637 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
638 if (copy_from_user(buf, src, this_len))
640 retval = ptrace_access_vm(tsk, dst, buf, this_len,
641 FOLL_FORCE | FOLL_WRITE);
655 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
659 if (data & ~(unsigned long)PTRACE_O_MASK)
662 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
663 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
664 !IS_ENABLED(CONFIG_SECCOMP))
667 if (!capable(CAP_SYS_ADMIN))
670 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
671 current->ptrace & PT_SUSPEND_SECCOMP)
675 /* Avoid intermediate state when all opts are cleared */
676 flags = child->ptrace;
677 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
678 flags |= (data << PT_OPT_FLAG_SHIFT);
679 child->ptrace = flags;
684 static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
689 if (lock_task_sighand(child, &flags)) {
691 if (likely(child->last_siginfo != NULL)) {
692 copy_siginfo(info, child->last_siginfo);
695 unlock_task_sighand(child, &flags);
700 static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
705 if (lock_task_sighand(child, &flags)) {
707 if (likely(child->last_siginfo != NULL)) {
708 copy_siginfo(child->last_siginfo, info);
711 unlock_task_sighand(child, &flags);
716 static int ptrace_peek_siginfo(struct task_struct *child,
720 struct ptrace_peeksiginfo_args arg;
721 struct sigpending *pending;
725 ret = copy_from_user(&arg, (void __user *) addr,
726 sizeof(struct ptrace_peeksiginfo_args));
730 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
731 return -EINVAL; /* unknown flags */
736 /* Ensure arg.off fits in an unsigned long */
737 if (arg.off > ULONG_MAX)
740 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
741 pending = &child->signal->shared_pending;
743 pending = &child->pending;
745 for (i = 0; i < arg.nr; ) {
746 kernel_siginfo_t info;
747 unsigned long off = arg.off + i;
750 spin_lock_irq(&child->sighand->siglock);
751 list_for_each_entry(q, &pending->list, list) {
754 copy_siginfo(&info, &q->info);
758 spin_unlock_irq(&child->sighand->siglock);
760 if (!found) /* beyond the end of the list */
764 if (unlikely(in_compat_syscall())) {
765 compat_siginfo_t __user *uinfo = compat_ptr(data);
767 if (copy_siginfo_to_user32(uinfo, &info)) {
775 siginfo_t __user *uinfo = (siginfo_t __user *) data;
777 if (copy_siginfo_to_user(uinfo, &info)) {
783 data += sizeof(siginfo_t);
786 if (signal_pending(current))
798 #ifdef PTRACE_SINGLESTEP
799 #define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
801 #define is_singlestep(request) 0
804 #ifdef PTRACE_SINGLEBLOCK
805 #define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
807 #define is_singleblock(request) 0
811 #define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
813 #define is_sysemu_singlestep(request) 0
816 static int ptrace_resume(struct task_struct *child, long request,
821 if (!valid_signal(data))
824 if (request == PTRACE_SYSCALL)
825 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
827 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
829 #ifdef TIF_SYSCALL_EMU
830 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
831 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
833 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
836 if (is_singleblock(request)) {
837 if (unlikely(!arch_has_block_step()))
839 user_enable_block_step(child);
840 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
841 if (unlikely(!arch_has_single_step()))
843 user_enable_single_step(child);
845 user_disable_single_step(child);
849 * Change ->exit_code and ->state under siglock to avoid the race
850 * with wait_task_stopped() in between; a non-zero ->exit_code will
851 * wrongly look like another report from tracee.
853 * Note that we need siglock even if ->exit_code == data and/or this
854 * status was not reported yet, the new status must not be cleared by
855 * wait_task_stopped() after resume.
857 * If data == 0 we do not care if wait_task_stopped() reports the old
858 * status and clears the code too; this can't race with the tracee, it
859 * takes siglock after resume.
861 need_siglock = data && !thread_group_empty(current);
863 spin_lock_irq(&child->sighand->siglock);
864 child->exit_code = data;
865 wake_up_state(child, __TASK_TRACED);
867 spin_unlock_irq(&child->sighand->siglock);
872 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
874 static const struct user_regset *
875 find_regset(const struct user_regset_view *view, unsigned int type)
877 const struct user_regset *regset;
880 for (n = 0; n < view->n; ++n) {
881 regset = view->regsets + n;
882 if (regset->core_note_type == type)
889 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
892 const struct user_regset_view *view = task_user_regset_view(task);
893 const struct user_regset *regset = find_regset(view, type);
896 if (!regset || (kiov->iov_len % regset->size) != 0)
899 regset_no = regset - view->regsets;
900 kiov->iov_len = min(kiov->iov_len,
901 (__kernel_size_t) (regset->n * regset->size));
903 if (req == PTRACE_GETREGSET)
904 return copy_regset_to_user(task, view, regset_no, 0,
905 kiov->iov_len, kiov->iov_base);
907 return copy_regset_from_user(task, view, regset_no, 0,
908 kiov->iov_len, kiov->iov_base);
912 * This is declared in linux/regset.h and defined in machine-dependent
913 * code. We put the export here, near the primary machine-neutral use,
914 * to ensure no machine forgets it.
916 EXPORT_SYMBOL_GPL(task_user_regset_view);
919 ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
920 struct ptrace_syscall_info *info)
922 unsigned long args[ARRAY_SIZE(info->entry.args)];
925 info->op = PTRACE_SYSCALL_INFO_ENTRY;
926 info->entry.nr = syscall_get_nr(child, regs);
927 syscall_get_arguments(child, regs, args);
928 for (i = 0; i < ARRAY_SIZE(args); i++)
929 info->entry.args[i] = args[i];
931 /* args is the last field in struct ptrace_syscall_info.entry */
932 return offsetofend(struct ptrace_syscall_info, entry.args);
936 ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
937 struct ptrace_syscall_info *info)
940 * As struct ptrace_syscall_info.entry is currently a subset
941 * of struct ptrace_syscall_info.seccomp, it makes sense to
942 * initialize that subset using ptrace_get_syscall_info_entry().
943 * This can be reconsidered in the future if these structures
944 * diverge significantly enough.
946 ptrace_get_syscall_info_entry(child, regs, info);
947 info->op = PTRACE_SYSCALL_INFO_SECCOMP;
948 info->seccomp.ret_data = child->ptrace_message;
950 /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
951 return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
955 ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
956 struct ptrace_syscall_info *info)
958 info->op = PTRACE_SYSCALL_INFO_EXIT;
959 info->exit.rval = syscall_get_error(child, regs);
960 info->exit.is_error = !!info->exit.rval;
961 if (!info->exit.is_error)
962 info->exit.rval = syscall_get_return_value(child, regs);
964 /* is_error is the last field in struct ptrace_syscall_info.exit */
965 return offsetofend(struct ptrace_syscall_info, exit.is_error);
969 ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
972 struct pt_regs *regs = task_pt_regs(child);
973 struct ptrace_syscall_info info = {
974 .op = PTRACE_SYSCALL_INFO_NONE,
975 .arch = syscall_get_arch(child),
976 .instruction_pointer = instruction_pointer(regs),
977 .stack_pointer = user_stack_pointer(regs),
979 unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
980 unsigned long write_size;
983 * This does not need lock_task_sighand() to access
984 * child->last_siginfo because ptrace_freeze_traced()
985 * called earlier by ptrace_check_attach() ensures that
986 * the tracee cannot go away and clear its last_siginfo.
988 switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
990 switch (child->ptrace_message) {
991 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
992 actual_size = ptrace_get_syscall_info_entry(child, regs,
995 case PTRACE_EVENTMSG_SYSCALL_EXIT:
996 actual_size = ptrace_get_syscall_info_exit(child, regs,
1001 case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1002 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1007 write_size = min(actual_size, user_size);
1008 return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1010 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1012 int ptrace_request(struct task_struct *child, long request,
1013 unsigned long addr, unsigned long data)
1015 bool seized = child->ptrace & PT_SEIZED;
1017 kernel_siginfo_t siginfo, *si;
1018 void __user *datavp = (void __user *) data;
1019 unsigned long __user *datalp = datavp;
1020 unsigned long flags;
1023 case PTRACE_PEEKTEXT:
1024 case PTRACE_PEEKDATA:
1025 return generic_ptrace_peekdata(child, addr, data);
1026 case PTRACE_POKETEXT:
1027 case PTRACE_POKEDATA:
1028 return generic_ptrace_pokedata(child, addr, data);
1030 #ifdef PTRACE_OLDSETOPTIONS
1031 case PTRACE_OLDSETOPTIONS:
1033 case PTRACE_SETOPTIONS:
1034 ret = ptrace_setoptions(child, data);
1036 case PTRACE_GETEVENTMSG:
1037 ret = put_user(child->ptrace_message, datalp);
1040 case PTRACE_PEEKSIGINFO:
1041 ret = ptrace_peek_siginfo(child, addr, data);
1044 case PTRACE_GETSIGINFO:
1045 ret = ptrace_getsiginfo(child, &siginfo);
1047 ret = copy_siginfo_to_user(datavp, &siginfo);
1050 case PTRACE_SETSIGINFO:
1051 ret = copy_siginfo_from_user(&siginfo, datavp);
1053 ret = ptrace_setsiginfo(child, &siginfo);
1056 case PTRACE_GETSIGMASK: {
1059 if (addr != sizeof(sigset_t)) {
1064 if (test_tsk_restore_sigmask(child))
1065 mask = &child->saved_sigmask;
1067 mask = &child->blocked;
1069 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1077 case PTRACE_SETSIGMASK: {
1080 if (addr != sizeof(sigset_t)) {
1085 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1090 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1093 * Every thread does recalc_sigpending() after resume, so
1094 * retarget_shared_pending() and recalc_sigpending() are not
1097 spin_lock_irq(&child->sighand->siglock);
1098 child->blocked = new_set;
1099 spin_unlock_irq(&child->sighand->siglock);
1101 clear_tsk_restore_sigmask(child);
1107 case PTRACE_INTERRUPT:
1109 * Stop tracee without any side-effect on signal or job
1110 * control. At least one trap is guaranteed to happen
1111 * after this request. If @child is already trapped, the
1112 * current trap is not disturbed and another trap will
1113 * happen after the current trap is ended with PTRACE_CONT.
1115 * The actual trap might not be PTRACE_EVENT_STOP trap but
1116 * the pending condition is cleared regardless.
1118 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1122 * INTERRUPT doesn't disturb existing trap sans one
1123 * exception. If ptracer issued LISTEN for the current
1124 * STOP, this INTERRUPT should clear LISTEN and re-trap
1127 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1128 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1130 unlock_task_sighand(child, &flags);
1136 * Listen for events. Tracee must be in STOP. It's not
1137 * resumed per-se but is not considered to be in TRACED by
1138 * wait(2) or ptrace(2). If an async event (e.g. group
1139 * stop state change) happens, tracee will enter STOP trap
1140 * again. Alternatively, ptracer can issue INTERRUPT to
1141 * finish listening and re-trap tracee into STOP.
1143 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1146 si = child->last_siginfo;
1147 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1148 child->jobctl |= JOBCTL_LISTENING;
1150 * If NOTIFY is set, it means event happened between
1151 * start of this trap and now. Trigger re-trap.
1153 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1154 ptrace_signal_wake_up(child, true);
1157 unlock_task_sighand(child, &flags);
1160 case PTRACE_DETACH: /* detach a process that was attached. */
1161 ret = ptrace_detach(child, data);
1164 #ifdef CONFIG_BINFMT_ELF_FDPIC
1165 case PTRACE_GETFDPIC: {
1166 struct mm_struct *mm = get_task_mm(child);
1167 unsigned long tmp = 0;
1174 case PTRACE_GETFDPIC_EXEC:
1175 tmp = mm->context.exec_fdpic_loadmap;
1177 case PTRACE_GETFDPIC_INTERP:
1178 tmp = mm->context.interp_fdpic_loadmap;
1185 ret = put_user(tmp, datalp);
1190 #ifdef PTRACE_SINGLESTEP
1191 case PTRACE_SINGLESTEP:
1193 #ifdef PTRACE_SINGLEBLOCK
1194 case PTRACE_SINGLEBLOCK:
1196 #ifdef PTRACE_SYSEMU
1198 case PTRACE_SYSEMU_SINGLESTEP:
1200 case PTRACE_SYSCALL:
1202 return ptrace_resume(child, request, data);
1205 if (child->exit_state) /* already dead */
1207 return ptrace_resume(child, request, SIGKILL);
1209 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1210 case PTRACE_GETREGSET:
1211 case PTRACE_SETREGSET: {
1213 struct iovec __user *uiov = datavp;
1215 if (!access_ok(uiov, sizeof(*uiov)))
1218 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1219 __get_user(kiov.iov_len, &uiov->iov_len))
1222 ret = ptrace_regset(child, request, addr, &kiov);
1224 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1228 case PTRACE_GET_SYSCALL_INFO:
1229 ret = ptrace_get_syscall_info(child, addr, datavp);
1233 case PTRACE_SECCOMP_GET_FILTER:
1234 ret = seccomp_get_filter(child, addr, datavp);
1237 case PTRACE_SECCOMP_GET_METADATA:
1238 ret = seccomp_get_metadata(child, addr, datavp);
1248 #ifndef arch_ptrace_attach
1249 #define arch_ptrace_attach(child) do { } while (0)
1252 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1253 unsigned long, data)
1255 struct task_struct *child;
1258 if (request == PTRACE_TRACEME) {
1259 ret = ptrace_traceme();
1261 arch_ptrace_attach(current);
1265 child = find_get_task_by_vpid(pid);
1271 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1272 ret = ptrace_attach(child, request, addr, data);
1274 * Some architectures need to do book-keeping after
1278 arch_ptrace_attach(child);
1279 goto out_put_task_struct;
1282 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1283 request == PTRACE_INTERRUPT);
1285 goto out_put_task_struct;
1287 ret = arch_ptrace(child, request, addr, data);
1288 if (ret || request != PTRACE_DETACH)
1289 ptrace_unfreeze_traced(child);
1291 out_put_task_struct:
1292 put_task_struct(child);
1297 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1303 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1304 if (copied != sizeof(tmp))
1306 return put_user(tmp, (unsigned long __user *)data);
1309 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1314 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1315 FOLL_FORCE | FOLL_WRITE);
1316 return (copied == sizeof(data)) ? 0 : -EIO;
1319 #if defined CONFIG_COMPAT
1321 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1322 compat_ulong_t addr, compat_ulong_t data)
1324 compat_ulong_t __user *datap = compat_ptr(data);
1325 compat_ulong_t word;
1326 kernel_siginfo_t siginfo;
1330 case PTRACE_PEEKTEXT:
1331 case PTRACE_PEEKDATA:
1332 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1334 if (ret != sizeof(word))
1337 ret = put_user(word, datap);
1340 case PTRACE_POKETEXT:
1341 case PTRACE_POKEDATA:
1342 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1343 FOLL_FORCE | FOLL_WRITE);
1344 ret = (ret != sizeof(data) ? -EIO : 0);
1347 case PTRACE_GETEVENTMSG:
1348 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1351 case PTRACE_GETSIGINFO:
1352 ret = ptrace_getsiginfo(child, &siginfo);
1354 ret = copy_siginfo_to_user32(
1355 (struct compat_siginfo __user *) datap,
1359 case PTRACE_SETSIGINFO:
1360 ret = copy_siginfo_from_user32(
1361 &siginfo, (struct compat_siginfo __user *) datap);
1363 ret = ptrace_setsiginfo(child, &siginfo);
1365 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1366 case PTRACE_GETREGSET:
1367 case PTRACE_SETREGSET:
1370 struct compat_iovec __user *uiov =
1371 (struct compat_iovec __user *) datap;
1375 if (!access_ok(uiov, sizeof(*uiov)))
1378 if (__get_user(ptr, &uiov->iov_base) ||
1379 __get_user(len, &uiov->iov_len))
1382 kiov.iov_base = compat_ptr(ptr);
1385 ret = ptrace_regset(child, request, addr, &kiov);
1387 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1393 ret = ptrace_request(child, request, addr, data);
1399 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1400 compat_long_t, addr, compat_long_t, data)
1402 struct task_struct *child;
1405 if (request == PTRACE_TRACEME) {
1406 ret = ptrace_traceme();
1410 child = find_get_task_by_vpid(pid);
1416 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1417 ret = ptrace_attach(child, request, addr, data);
1419 * Some architectures need to do book-keeping after
1423 arch_ptrace_attach(child);
1424 goto out_put_task_struct;
1427 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1428 request == PTRACE_INTERRUPT);
1430 ret = compat_arch_ptrace(child, request, addr, data);
1431 if (ret || request != PTRACE_DETACH)
1432 ptrace_unfreeze_traced(child);
1435 out_put_task_struct:
1436 put_task_struct(child);
1440 #endif /* CONFIG_COMPAT */