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 /* Ensure that nothing can wake it up, even SIGKILL */
173 static bool ptrace_freeze_traced(struct task_struct *task)
177 /* Lockless, nobody but us can set this flag */
178 if (task->jobctl & JOBCTL_LISTENING)
181 spin_lock_irq(&task->sighand->siglock);
182 if (task_is_traced(task) && !__fatal_signal_pending(task)) {
183 task->state = __TASK_TRACED;
186 spin_unlock_irq(&task->sighand->siglock);
191 static void ptrace_unfreeze_traced(struct task_struct *task)
193 if (task->state != __TASK_TRACED)
196 WARN_ON(!task->ptrace || task->parent != current);
199 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
200 * Recheck state under the lock to close this race.
202 spin_lock_irq(&task->sighand->siglock);
203 if (task->state == __TASK_TRACED) {
204 if (__fatal_signal_pending(task))
205 wake_up_state(task, __TASK_TRACED);
207 task->state = TASK_TRACED;
209 spin_unlock_irq(&task->sighand->siglock);
213 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
214 * @child: ptracee to check for
215 * @ignore_state: don't check whether @child is currently %TASK_TRACED
217 * Check whether @child is being ptraced by %current and ready for further
218 * ptrace operations. If @ignore_state is %false, @child also should be in
219 * %TASK_TRACED state and on return the child is guaranteed to be traced
220 * and not executing. If @ignore_state is %true, @child can be in any
224 * Grabs and releases tasklist_lock and @child->sighand->siglock.
227 * 0 on success, -ESRCH if %child is not ready.
229 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
234 * We take the read lock around doing both checks to close a
235 * possible race where someone else was tracing our child and
236 * detached between these two checks. After this locked check,
237 * we are sure that this is our traced child and that can only
238 * be changed by us so it's not changing right after this.
240 read_lock(&tasklist_lock);
241 if (child->ptrace && child->parent == current) {
242 WARN_ON(child->state == __TASK_TRACED);
244 * child->sighand can't be NULL, release_task()
245 * does ptrace_unlink() before __exit_signal().
247 if (ignore_state || ptrace_freeze_traced(child))
250 read_unlock(&tasklist_lock);
252 if (!ret && !ignore_state) {
253 if (!wait_task_inactive(child, __TASK_TRACED)) {
255 * This can only happen if may_ptrace_stop() fails and
256 * ptrace_stop() changes ->state back to TASK_RUNNING,
257 * so we should not worry about leaking __TASK_TRACED.
259 WARN_ON(child->state == __TASK_TRACED);
267 static bool ptrace_has_cap(const struct cred *cred, struct user_namespace *ns,
272 if (mode & PTRACE_MODE_NOAUDIT)
273 ret = security_capable(cred, ns, CAP_SYS_PTRACE, CAP_OPT_NOAUDIT);
275 ret = security_capable(cred, ns, CAP_SYS_PTRACE, CAP_OPT_NONE);
280 /* Returns 0 on success, -errno on denial. */
281 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
283 const struct cred *cred = current_cred(), *tcred;
284 struct mm_struct *mm;
288 if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
289 WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
293 /* May we inspect the given task?
294 * This check is used both for attaching with ptrace
295 * and for allowing access to sensitive information in /proc.
297 * ptrace_attach denies several cases that /proc allows
298 * because setting up the necessary parent/child relationship
299 * or halting the specified task is impossible.
302 /* Don't let security modules deny introspection */
303 if (same_thread_group(task, current))
306 if (mode & PTRACE_MODE_FSCREDS) {
307 caller_uid = cred->fsuid;
308 caller_gid = cred->fsgid;
311 * Using the euid would make more sense here, but something
312 * in userland might rely on the old behavior, and this
313 * shouldn't be a security problem since
314 * PTRACE_MODE_REALCREDS implies that the caller explicitly
315 * used a syscall that requests access to another process
316 * (and not a filesystem syscall to procfs).
318 caller_uid = cred->uid;
319 caller_gid = cred->gid;
321 tcred = __task_cred(task);
322 if (uid_eq(caller_uid, tcred->euid) &&
323 uid_eq(caller_uid, tcred->suid) &&
324 uid_eq(caller_uid, tcred->uid) &&
325 gid_eq(caller_gid, tcred->egid) &&
326 gid_eq(caller_gid, tcred->sgid) &&
327 gid_eq(caller_gid, tcred->gid))
329 if (ptrace_has_cap(cred, tcred->user_ns, mode))
336 * If a task drops privileges and becomes nondumpable (through a syscall
337 * like setresuid()) while we are trying to access it, we must ensure
338 * that the dumpability is read after the credentials; otherwise,
339 * we may be able to attach to a task that we shouldn't be able to
340 * attach to (as if the task had dropped privileges without becoming
342 * Pairs with a write barrier in commit_creds().
347 ((get_dumpable(mm) != SUID_DUMP_USER) &&
348 !ptrace_has_cap(cred, mm->user_ns, mode)))
351 return security_ptrace_access_check(task, mode);
354 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
358 err = __ptrace_may_access(task, mode);
363 static int ptrace_attach(struct task_struct *task, long request,
367 bool seize = (request == PTRACE_SEIZE);
374 if (flags & ~(unsigned long)PTRACE_O_MASK)
376 flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
384 if (unlikely(task->flags & PF_KTHREAD))
386 if (same_thread_group(task, current))
390 * Protect exec's credential calculations against our interference;
391 * SUID, SGID and LSM creds get determined differently
394 retval = -ERESTARTNOINTR;
395 if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
399 retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
404 write_lock_irq(&tasklist_lock);
406 if (unlikely(task->exit_state))
407 goto unlock_tasklist;
409 goto unlock_tasklist;
413 task->ptrace = flags;
415 ptrace_link(task, current);
417 /* SEIZE doesn't trap tracee on attach */
419 send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
421 spin_lock(&task->sighand->siglock);
424 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
425 * TRAPPING, and kick it so that it transits to TRACED. TRAPPING
426 * will be cleared if the child completes the transition or any
427 * event which clears the group stop states happens. We'll wait
428 * for the transition to complete before returning from this
431 * This hides STOPPED -> RUNNING -> TRACED transition from the
432 * attaching thread but a different thread in the same group can
433 * still observe the transient RUNNING state. IOW, if another
434 * thread's WNOHANG wait(2) on the stopped tracee races against
435 * ATTACH, the wait(2) may fail due to the transient RUNNING.
437 * The following task_is_stopped() test is safe as both transitions
438 * in and out of STOPPED are protected by siglock.
440 if (task_is_stopped(task) &&
441 task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
442 signal_wake_up_state(task, __TASK_STOPPED);
444 spin_unlock(&task->sighand->siglock);
448 write_unlock_irq(&tasklist_lock);
450 mutex_unlock(&task->signal->cred_guard_mutex);
454 * We do not bother to change retval or clear JOBCTL_TRAPPING
455 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
456 * not return to user-mode, it will exit and clear this bit in
457 * __ptrace_unlink() if it wasn't already cleared by the tracee;
458 * and until then nobody can ptrace this task.
460 wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
461 proc_ptrace_connector(task, PTRACE_ATTACH);
468 * ptrace_traceme -- helper for PTRACE_TRACEME
470 * Performs checks and sets PT_PTRACED.
471 * Should be used by all ptrace implementations for PTRACE_TRACEME.
473 static int ptrace_traceme(void)
477 write_lock_irq(&tasklist_lock);
478 /* Are we already being traced? */
479 if (!current->ptrace) {
480 ret = security_ptrace_traceme(current->parent);
482 * Check PF_EXITING to ensure ->real_parent has not passed
483 * exit_ptrace(). Otherwise we don't report the error but
484 * pretend ->real_parent untraces us right after return.
486 if (!ret && !(current->real_parent->flags & PF_EXITING)) {
487 current->ptrace = PT_PTRACED;
488 ptrace_link(current, current->real_parent);
491 write_unlock_irq(&tasklist_lock);
497 * Called with irqs disabled, returns true if childs should reap themselves.
499 static int ignoring_children(struct sighand_struct *sigh)
502 spin_lock(&sigh->siglock);
503 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
504 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
505 spin_unlock(&sigh->siglock);
510 * Called with tasklist_lock held for writing.
511 * Unlink a traced task, and clean it up if it was a traced zombie.
512 * Return true if it needs to be reaped with release_task().
513 * (We can't call release_task() here because we already hold tasklist_lock.)
515 * If it's a zombie, our attachedness prevented normal parent notification
516 * or self-reaping. Do notification now if it would have happened earlier.
517 * If it should reap itself, return true.
519 * If it's our own child, there is no notification to do. But if our normal
520 * children self-reap, then this child was prevented by ptrace and we must
521 * reap it now, in that case we must also wake up sub-threads sleeping in
524 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
530 if (p->exit_state != EXIT_ZOMBIE)
533 dead = !thread_group_leader(p);
535 if (!dead && thread_group_empty(p)) {
536 if (!same_thread_group(p->real_parent, tracer))
537 dead = do_notify_parent(p, p->exit_signal);
538 else if (ignoring_children(tracer->sighand)) {
539 __wake_up_parent(p, tracer);
543 /* Mark it as in the process of being reaped. */
545 p->exit_state = EXIT_DEAD;
549 static int ptrace_detach(struct task_struct *child, unsigned int data)
551 if (!valid_signal(data))
554 /* Architecture-specific hardware disable .. */
555 ptrace_disable(child);
557 write_lock_irq(&tasklist_lock);
559 * We rely on ptrace_freeze_traced(). It can't be killed and
560 * untraced by another thread, it can't be a zombie.
562 WARN_ON(!child->ptrace || child->exit_state);
564 * tasklist_lock avoids the race with wait_task_stopped(), see
565 * the comment in ptrace_resume().
567 child->exit_code = data;
568 __ptrace_detach(current, child);
569 write_unlock_irq(&tasklist_lock);
571 proc_ptrace_connector(child, PTRACE_DETACH);
577 * Detach all tasks we were using ptrace on. Called with tasklist held
580 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
582 struct task_struct *p, *n;
584 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
585 if (unlikely(p->ptrace & PT_EXITKILL))
586 send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
588 if (__ptrace_detach(tracer, p))
589 list_add(&p->ptrace_entry, dead);
593 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
599 int this_len, retval;
601 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
602 retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
609 if (copy_to_user(dst, buf, retval))
619 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
625 int this_len, retval;
627 this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
628 if (copy_from_user(buf, src, this_len))
630 retval = ptrace_access_vm(tsk, dst, buf, this_len,
631 FOLL_FORCE | FOLL_WRITE);
645 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
649 if (data & ~(unsigned long)PTRACE_O_MASK)
652 if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
653 if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
654 !IS_ENABLED(CONFIG_SECCOMP))
657 if (!capable(CAP_SYS_ADMIN))
660 if (seccomp_mode(¤t->seccomp) != SECCOMP_MODE_DISABLED ||
661 current->ptrace & PT_SUSPEND_SECCOMP)
665 /* Avoid intermediate state when all opts are cleared */
666 flags = child->ptrace;
667 flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
668 flags |= (data << PT_OPT_FLAG_SHIFT);
669 child->ptrace = flags;
674 static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
679 if (lock_task_sighand(child, &flags)) {
681 if (likely(child->last_siginfo != NULL)) {
682 copy_siginfo(info, child->last_siginfo);
685 unlock_task_sighand(child, &flags);
690 static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
695 if (lock_task_sighand(child, &flags)) {
697 if (likely(child->last_siginfo != NULL)) {
698 copy_siginfo(child->last_siginfo, info);
701 unlock_task_sighand(child, &flags);
706 static int ptrace_peek_siginfo(struct task_struct *child,
710 struct ptrace_peeksiginfo_args arg;
711 struct sigpending *pending;
715 ret = copy_from_user(&arg, (void __user *) addr,
716 sizeof(struct ptrace_peeksiginfo_args));
720 if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
721 return -EINVAL; /* unknown flags */
726 /* Ensure arg.off fits in an unsigned long */
727 if (arg.off > ULONG_MAX)
730 if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
731 pending = &child->signal->shared_pending;
733 pending = &child->pending;
735 for (i = 0; i < arg.nr; ) {
736 kernel_siginfo_t info;
737 unsigned long off = arg.off + i;
740 spin_lock_irq(&child->sighand->siglock);
741 list_for_each_entry(q, &pending->list, list) {
744 copy_siginfo(&info, &q->info);
748 spin_unlock_irq(&child->sighand->siglock);
750 if (!found) /* beyond the end of the list */
754 if (unlikely(in_compat_syscall())) {
755 compat_siginfo_t __user *uinfo = compat_ptr(data);
757 if (copy_siginfo_to_user32(uinfo, &info)) {
765 siginfo_t __user *uinfo = (siginfo_t __user *) data;
767 if (copy_siginfo_to_user(uinfo, &info)) {
773 data += sizeof(siginfo_t);
776 if (signal_pending(current))
788 #ifdef PTRACE_SINGLESTEP
789 #define is_singlestep(request) ((request) == PTRACE_SINGLESTEP)
791 #define is_singlestep(request) 0
794 #ifdef PTRACE_SINGLEBLOCK
795 #define is_singleblock(request) ((request) == PTRACE_SINGLEBLOCK)
797 #define is_singleblock(request) 0
801 #define is_sysemu_singlestep(request) ((request) == PTRACE_SYSEMU_SINGLESTEP)
803 #define is_sysemu_singlestep(request) 0
806 static int ptrace_resume(struct task_struct *child, long request,
811 if (!valid_signal(data))
814 if (request == PTRACE_SYSCALL)
815 set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
817 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
819 #ifdef TIF_SYSCALL_EMU
820 if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
821 set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
823 clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
826 if (is_singleblock(request)) {
827 if (unlikely(!arch_has_block_step()))
829 user_enable_block_step(child);
830 } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
831 if (unlikely(!arch_has_single_step()))
833 user_enable_single_step(child);
835 user_disable_single_step(child);
839 * Change ->exit_code and ->state under siglock to avoid the race
840 * with wait_task_stopped() in between; a non-zero ->exit_code will
841 * wrongly look like another report from tracee.
843 * Note that we need siglock even if ->exit_code == data and/or this
844 * status was not reported yet, the new status must not be cleared by
845 * wait_task_stopped() after resume.
847 * If data == 0 we do not care if wait_task_stopped() reports the old
848 * status and clears the code too; this can't race with the tracee, it
849 * takes siglock after resume.
851 need_siglock = data && !thread_group_empty(current);
853 spin_lock_irq(&child->sighand->siglock);
854 child->exit_code = data;
855 wake_up_state(child, __TASK_TRACED);
857 spin_unlock_irq(&child->sighand->siglock);
862 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
864 static const struct user_regset *
865 find_regset(const struct user_regset_view *view, unsigned int type)
867 const struct user_regset *regset;
870 for (n = 0; n < view->n; ++n) {
871 regset = view->regsets + n;
872 if (regset->core_note_type == type)
879 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
882 const struct user_regset_view *view = task_user_regset_view(task);
883 const struct user_regset *regset = find_regset(view, type);
886 if (!regset || (kiov->iov_len % regset->size) != 0)
889 regset_no = regset - view->regsets;
890 kiov->iov_len = min(kiov->iov_len,
891 (__kernel_size_t) (regset->n * regset->size));
893 if (req == PTRACE_GETREGSET)
894 return copy_regset_to_user(task, view, regset_no, 0,
895 kiov->iov_len, kiov->iov_base);
897 return copy_regset_from_user(task, view, regset_no, 0,
898 kiov->iov_len, kiov->iov_base);
902 * This is declared in linux/regset.h and defined in machine-dependent
903 * code. We put the export here, near the primary machine-neutral use,
904 * to ensure no machine forgets it.
906 EXPORT_SYMBOL_GPL(task_user_regset_view);
909 ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
910 struct ptrace_syscall_info *info)
912 unsigned long args[ARRAY_SIZE(info->entry.args)];
915 info->op = PTRACE_SYSCALL_INFO_ENTRY;
916 info->entry.nr = syscall_get_nr(child, regs);
917 syscall_get_arguments(child, regs, args);
918 for (i = 0; i < ARRAY_SIZE(args); i++)
919 info->entry.args[i] = args[i];
921 /* args is the last field in struct ptrace_syscall_info.entry */
922 return offsetofend(struct ptrace_syscall_info, entry.args);
926 ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
927 struct ptrace_syscall_info *info)
930 * As struct ptrace_syscall_info.entry is currently a subset
931 * of struct ptrace_syscall_info.seccomp, it makes sense to
932 * initialize that subset using ptrace_get_syscall_info_entry().
933 * This can be reconsidered in the future if these structures
934 * diverge significantly enough.
936 ptrace_get_syscall_info_entry(child, regs, info);
937 info->op = PTRACE_SYSCALL_INFO_SECCOMP;
938 info->seccomp.ret_data = child->ptrace_message;
940 /* ret_data is the last field in struct ptrace_syscall_info.seccomp */
941 return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
945 ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
946 struct ptrace_syscall_info *info)
948 info->op = PTRACE_SYSCALL_INFO_EXIT;
949 info->exit.rval = syscall_get_error(child, regs);
950 info->exit.is_error = !!info->exit.rval;
951 if (!info->exit.is_error)
952 info->exit.rval = syscall_get_return_value(child, regs);
954 /* is_error is the last field in struct ptrace_syscall_info.exit */
955 return offsetofend(struct ptrace_syscall_info, exit.is_error);
959 ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
962 struct pt_regs *regs = task_pt_regs(child);
963 struct ptrace_syscall_info info = {
964 .op = PTRACE_SYSCALL_INFO_NONE,
965 .arch = syscall_get_arch(child),
966 .instruction_pointer = instruction_pointer(regs),
967 .stack_pointer = user_stack_pointer(regs),
969 unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
970 unsigned long write_size;
973 * This does not need lock_task_sighand() to access
974 * child->last_siginfo because ptrace_freeze_traced()
975 * called earlier by ptrace_check_attach() ensures that
976 * the tracee cannot go away and clear its last_siginfo.
978 switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
980 switch (child->ptrace_message) {
981 case PTRACE_EVENTMSG_SYSCALL_ENTRY:
982 actual_size = ptrace_get_syscall_info_entry(child, regs,
985 case PTRACE_EVENTMSG_SYSCALL_EXIT:
986 actual_size = ptrace_get_syscall_info_exit(child, regs,
991 case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
992 actual_size = ptrace_get_syscall_info_seccomp(child, regs,
997 write_size = min(actual_size, user_size);
998 return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1000 #endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1002 int ptrace_request(struct task_struct *child, long request,
1003 unsigned long addr, unsigned long data)
1005 bool seized = child->ptrace & PT_SEIZED;
1007 kernel_siginfo_t siginfo, *si;
1008 void __user *datavp = (void __user *) data;
1009 unsigned long __user *datalp = datavp;
1010 unsigned long flags;
1013 case PTRACE_PEEKTEXT:
1014 case PTRACE_PEEKDATA:
1015 return generic_ptrace_peekdata(child, addr, data);
1016 case PTRACE_POKETEXT:
1017 case PTRACE_POKEDATA:
1018 return generic_ptrace_pokedata(child, addr, data);
1020 #ifdef PTRACE_OLDSETOPTIONS
1021 case PTRACE_OLDSETOPTIONS:
1023 case PTRACE_SETOPTIONS:
1024 ret = ptrace_setoptions(child, data);
1026 case PTRACE_GETEVENTMSG:
1027 ret = put_user(child->ptrace_message, datalp);
1030 case PTRACE_PEEKSIGINFO:
1031 ret = ptrace_peek_siginfo(child, addr, data);
1034 case PTRACE_GETSIGINFO:
1035 ret = ptrace_getsiginfo(child, &siginfo);
1037 ret = copy_siginfo_to_user(datavp, &siginfo);
1040 case PTRACE_SETSIGINFO:
1041 ret = copy_siginfo_from_user(&siginfo, datavp);
1043 ret = ptrace_setsiginfo(child, &siginfo);
1046 case PTRACE_GETSIGMASK: {
1049 if (addr != sizeof(sigset_t)) {
1054 if (test_tsk_restore_sigmask(child))
1055 mask = &child->saved_sigmask;
1057 mask = &child->blocked;
1059 if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1067 case PTRACE_SETSIGMASK: {
1070 if (addr != sizeof(sigset_t)) {
1075 if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1080 sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1083 * Every thread does recalc_sigpending() after resume, so
1084 * retarget_shared_pending() and recalc_sigpending() are not
1087 spin_lock_irq(&child->sighand->siglock);
1088 child->blocked = new_set;
1089 spin_unlock_irq(&child->sighand->siglock);
1091 clear_tsk_restore_sigmask(child);
1097 case PTRACE_INTERRUPT:
1099 * Stop tracee without any side-effect on signal or job
1100 * control. At least one trap is guaranteed to happen
1101 * after this request. If @child is already trapped, the
1102 * current trap is not disturbed and another trap will
1103 * happen after the current trap is ended with PTRACE_CONT.
1105 * The actual trap might not be PTRACE_EVENT_STOP trap but
1106 * the pending condition is cleared regardless.
1108 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1112 * INTERRUPT doesn't disturb existing trap sans one
1113 * exception. If ptracer issued LISTEN for the current
1114 * STOP, this INTERRUPT should clear LISTEN and re-trap
1117 if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1118 ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1120 unlock_task_sighand(child, &flags);
1126 * Listen for events. Tracee must be in STOP. It's not
1127 * resumed per-se but is not considered to be in TRACED by
1128 * wait(2) or ptrace(2). If an async event (e.g. group
1129 * stop state change) happens, tracee will enter STOP trap
1130 * again. Alternatively, ptracer can issue INTERRUPT to
1131 * finish listening and re-trap tracee into STOP.
1133 if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1136 si = child->last_siginfo;
1137 if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1138 child->jobctl |= JOBCTL_LISTENING;
1140 * If NOTIFY is set, it means event happened between
1141 * start of this trap and now. Trigger re-trap.
1143 if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1144 ptrace_signal_wake_up(child, true);
1147 unlock_task_sighand(child, &flags);
1150 case PTRACE_DETACH: /* detach a process that was attached. */
1151 ret = ptrace_detach(child, data);
1154 #ifdef CONFIG_BINFMT_ELF_FDPIC
1155 case PTRACE_GETFDPIC: {
1156 struct mm_struct *mm = get_task_mm(child);
1157 unsigned long tmp = 0;
1164 case PTRACE_GETFDPIC_EXEC:
1165 tmp = mm->context.exec_fdpic_loadmap;
1167 case PTRACE_GETFDPIC_INTERP:
1168 tmp = mm->context.interp_fdpic_loadmap;
1175 ret = put_user(tmp, datalp);
1180 #ifdef PTRACE_SINGLESTEP
1181 case PTRACE_SINGLESTEP:
1183 #ifdef PTRACE_SINGLEBLOCK
1184 case PTRACE_SINGLEBLOCK:
1186 #ifdef PTRACE_SYSEMU
1188 case PTRACE_SYSEMU_SINGLESTEP:
1190 case PTRACE_SYSCALL:
1192 return ptrace_resume(child, request, data);
1195 if (child->exit_state) /* already dead */
1197 return ptrace_resume(child, request, SIGKILL);
1199 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1200 case PTRACE_GETREGSET:
1201 case PTRACE_SETREGSET: {
1203 struct iovec __user *uiov = datavp;
1205 if (!access_ok(uiov, sizeof(*uiov)))
1208 if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1209 __get_user(kiov.iov_len, &uiov->iov_len))
1212 ret = ptrace_regset(child, request, addr, &kiov);
1214 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1218 case PTRACE_GET_SYSCALL_INFO:
1219 ret = ptrace_get_syscall_info(child, addr, datavp);
1223 case PTRACE_SECCOMP_GET_FILTER:
1224 ret = seccomp_get_filter(child, addr, datavp);
1227 case PTRACE_SECCOMP_GET_METADATA:
1228 ret = seccomp_get_metadata(child, addr, datavp);
1238 #ifndef arch_ptrace_attach
1239 #define arch_ptrace_attach(child) do { } while (0)
1242 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1243 unsigned long, data)
1245 struct task_struct *child;
1248 if (request == PTRACE_TRACEME) {
1249 ret = ptrace_traceme();
1251 arch_ptrace_attach(current);
1255 child = find_get_task_by_vpid(pid);
1261 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1262 ret = ptrace_attach(child, request, addr, data);
1264 * Some architectures need to do book-keeping after
1268 arch_ptrace_attach(child);
1269 goto out_put_task_struct;
1272 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1273 request == PTRACE_INTERRUPT);
1275 goto out_put_task_struct;
1277 ret = arch_ptrace(child, request, addr, data);
1278 if (ret || request != PTRACE_DETACH)
1279 ptrace_unfreeze_traced(child);
1281 out_put_task_struct:
1282 put_task_struct(child);
1287 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1293 copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1294 if (copied != sizeof(tmp))
1296 return put_user(tmp, (unsigned long __user *)data);
1299 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1304 copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1305 FOLL_FORCE | FOLL_WRITE);
1306 return (copied == sizeof(data)) ? 0 : -EIO;
1309 #if defined CONFIG_COMPAT
1311 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1312 compat_ulong_t addr, compat_ulong_t data)
1314 compat_ulong_t __user *datap = compat_ptr(data);
1315 compat_ulong_t word;
1316 kernel_siginfo_t siginfo;
1320 case PTRACE_PEEKTEXT:
1321 case PTRACE_PEEKDATA:
1322 ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1324 if (ret != sizeof(word))
1327 ret = put_user(word, datap);
1330 case PTRACE_POKETEXT:
1331 case PTRACE_POKEDATA:
1332 ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1333 FOLL_FORCE | FOLL_WRITE);
1334 ret = (ret != sizeof(data) ? -EIO : 0);
1337 case PTRACE_GETEVENTMSG:
1338 ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1341 case PTRACE_GETSIGINFO:
1342 ret = ptrace_getsiginfo(child, &siginfo);
1344 ret = copy_siginfo_to_user32(
1345 (struct compat_siginfo __user *) datap,
1349 case PTRACE_SETSIGINFO:
1350 ret = copy_siginfo_from_user32(
1351 &siginfo, (struct compat_siginfo __user *) datap);
1353 ret = ptrace_setsiginfo(child, &siginfo);
1355 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1356 case PTRACE_GETREGSET:
1357 case PTRACE_SETREGSET:
1360 struct compat_iovec __user *uiov =
1361 (struct compat_iovec __user *) datap;
1365 if (!access_ok(uiov, sizeof(*uiov)))
1368 if (__get_user(ptr, &uiov->iov_base) ||
1369 __get_user(len, &uiov->iov_len))
1372 kiov.iov_base = compat_ptr(ptr);
1375 ret = ptrace_regset(child, request, addr, &kiov);
1377 ret = __put_user(kiov.iov_len, &uiov->iov_len);
1383 ret = ptrace_request(child, request, addr, data);
1389 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1390 compat_long_t, addr, compat_long_t, data)
1392 struct task_struct *child;
1395 if (request == PTRACE_TRACEME) {
1396 ret = ptrace_traceme();
1400 child = find_get_task_by_vpid(pid);
1406 if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1407 ret = ptrace_attach(child, request, addr, data);
1409 * Some architectures need to do book-keeping after
1413 arch_ptrace_attach(child);
1414 goto out_put_task_struct;
1417 ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1418 request == PTRACE_INTERRUPT);
1420 ret = compat_arch_ptrace(child, request, addr, data);
1421 if (ret || request != PTRACE_DETACH)
1422 ptrace_unfreeze_traced(child);
1425 out_put_task_struct:
1426 put_task_struct(child);
1430 #endif /* CONFIG_COMPAT */