Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/kernel/ptrace.c
   4 *
   5 * (C) Copyright 1999 Linus Torvalds
   6 *
   7 * Common interfaces for "ptrace()" which we do not want
   8 * to continually duplicate across every architecture.
   9 */
  10
  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>
  18#include <linux/mm.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>
  35
  36#include <asm/syscall.h>	/* for syscall_get_* */
  37
  38/*
  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
  42 */
  43int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
  44		     void *buf, int len, unsigned int gup_flags)
  45{
  46	struct mm_struct *mm;
  47	int ret;
  48
  49	mm = get_task_mm(tsk);
  50	if (!mm)
  51		return 0;
  52
  53	if (!tsk->ptrace ||
  54	    (current != tsk->parent) ||
  55	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
  56	     !ptracer_capable(tsk, mm->user_ns))) {
  57		mmput(mm);
  58		return 0;
  59	}
  60
  61	ret = __access_remote_vm(mm, addr, buf, len, gup_flags);
  62	mmput(mm);
  63
  64	return ret;
  65}
  66
  67
  68void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
  69		   const struct cred *ptracer_cred)
  70{
  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);
  75}
  76
  77/*
  78 * ptrace a task: make the debugger its new parent and
  79 * move it to the ptrace list.
  80 *
  81 * Must be called with the tasklist lock write-held.
  82 */
  83static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
  84{
  85	__ptrace_link(child, new_parent, current_cred());
  86}
  87
  88/**
  89 * __ptrace_unlink - unlink ptracee and restore its execution state
  90 * @child: ptracee to be unlinked
  91 *
  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
  94 * state.
  95 *
  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.
 100 *
 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.
 105 *
 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.
 112 *
 113 * CONTEXT:
 114 * write_lock_irq(tasklist_lock)
 115 */
 116void __ptrace_unlink(struct task_struct *child)
 117{
 118	const struct cred *old_cred;
 119	BUG_ON(!child->ptrace);
 120
 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);
 124#endif
 125
 126	child->parent = child->real_parent;
 127	list_del_init(&child->ptrace_entry);
 128	old_cred = child->ptracer_cred;
 129	child->ptracer_cred = NULL;
 130	put_cred(old_cred);
 131
 132	spin_lock(&child->sighand->siglock);
 133	child->ptrace = 0;
 134	/*
 135	 * Clear all pending traps and TRAPPING.  TRAPPING should be
 136	 * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
 137	 */
 138	task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
 139	task_clear_jobctl_trapping(child);
 140
 141	/*
 142	 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
 143	 * @child isn't dead.
 144	 */
 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;
 149
 150		/*
 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
 155		 * case.
 156		 */
 157		if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
 158			child->jobctl |= SIGSTOP;
 159	}
 160
 161	/*
 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.
 166	 */
 167	if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
 168		ptrace_signal_wake_up(child, true);
 169
 170	spin_unlock(&child->sighand->siglock);
 171}
 172
 173static bool looks_like_a_spurious_pid(struct task_struct *task)
 174{
 175	if (task->exit_code != ((PTRACE_EVENT_EXEC << 8) | SIGTRAP))
 176		return false;
 177
 178	if (task_pid_vnr(task) == task->ptrace_message)
 179		return false;
 180	/*
 181	 * The tracee changed its pid but the PTRACE_EVENT_EXEC event
 182	 * was not wait()'ed, most probably debugger targets the old
 183	 * leader which was destroyed in de_thread().
 184	 */
 185	return true;
 186}
 187
 188/*
 189 * Ensure that nothing can wake it up, even SIGKILL
 190 *
 191 * A task is switched to this state while a ptrace operation is in progress;
 192 * such that the ptrace operation is uninterruptible.
 193 */
 194static bool ptrace_freeze_traced(struct task_struct *task)
 195{
 196	bool ret = false;
 197
 198	/* Lockless, nobody but us can set this flag */
 199	if (task->jobctl & JOBCTL_LISTENING)
 200		return ret;
 201
 202	spin_lock_irq(&task->sighand->siglock);
 203	if (task_is_traced(task) && !looks_like_a_spurious_pid(task) &&
 204	    !__fatal_signal_pending(task)) {
 205		task->jobctl |= JOBCTL_PTRACE_FROZEN;
 206		ret = true;
 207	}
 208	spin_unlock_irq(&task->sighand->siglock);
 209
 210	return ret;
 211}
 212
 213static void ptrace_unfreeze_traced(struct task_struct *task)
 214{
 215	unsigned long flags;
 
 
 
 216
 217	/*
 218	 * The child may be awake and may have cleared
 219	 * JOBCTL_PTRACE_FROZEN (see ptrace_resume).  The child will
 220	 * not set JOBCTL_PTRACE_FROZEN or enter __TASK_TRACED anew.
 221	 */
 222	if (lock_task_sighand(task, &flags)) {
 223		task->jobctl &= ~JOBCTL_PTRACE_FROZEN;
 224		if (__fatal_signal_pending(task)) {
 225			task->jobctl &= ~JOBCTL_TRACED;
 226			wake_up_state(task, __TASK_TRACED);
 227		}
 228		unlock_task_sighand(task, &flags);
 229	}
 
 230}
 231
 232/**
 233 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
 234 * @child: ptracee to check for
 235 * @ignore_state: don't check whether @child is currently %TASK_TRACED
 236 *
 237 * Check whether @child is being ptraced by %current and ready for further
 238 * ptrace operations.  If @ignore_state is %false, @child also should be in
 239 * %TASK_TRACED state and on return the child is guaranteed to be traced
 240 * and not executing.  If @ignore_state is %true, @child can be in any
 241 * state.
 242 *
 243 * CONTEXT:
 244 * Grabs and releases tasklist_lock and @child->sighand->siglock.
 245 *
 246 * RETURNS:
 247 * 0 on success, -ESRCH if %child is not ready.
 248 */
 249static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
 250{
 251	int ret = -ESRCH;
 252
 253	/*
 254	 * We take the read lock around doing both checks to close a
 255	 * possible race where someone else was tracing our child and
 256	 * detached between these two checks.  After this locked check,
 257	 * we are sure that this is our traced child and that can only
 258	 * be changed by us so it's not changing right after this.
 259	 */
 260	read_lock(&tasklist_lock);
 261	if (child->ptrace && child->parent == current) {
 
 262		/*
 263		 * child->sighand can't be NULL, release_task()
 264		 * does ptrace_unlink() before __exit_signal().
 265		 */
 266		if (ignore_state || ptrace_freeze_traced(child))
 267			ret = 0;
 268	}
 269	read_unlock(&tasklist_lock);
 270
 271	if (!ret && !ignore_state &&
 272	    WARN_ON_ONCE(!wait_task_inactive(child, __TASK_TRACED|TASK_FROZEN)))
 273		ret = -ESRCH;
 
 
 
 
 
 
 
 
 274
 275	return ret;
 276}
 277
 278static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
 279{
 280	if (mode & PTRACE_MODE_NOAUDIT)
 281		return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
 282	return ns_capable(ns, CAP_SYS_PTRACE);
 283}
 284
 285/* Returns 0 on success, -errno on denial. */
 286static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
 287{
 288	const struct cred *cred = current_cred(), *tcred;
 289	struct mm_struct *mm;
 290	kuid_t caller_uid;
 291	kgid_t caller_gid;
 292
 293	if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
 294		WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
 295		return -EPERM;
 296	}
 297
 298	/* May we inspect the given task?
 299	 * This check is used both for attaching with ptrace
 300	 * and for allowing access to sensitive information in /proc.
 301	 *
 302	 * ptrace_attach denies several cases that /proc allows
 303	 * because setting up the necessary parent/child relationship
 304	 * or halting the specified task is impossible.
 305	 */
 306
 307	/* Don't let security modules deny introspection */
 308	if (same_thread_group(task, current))
 309		return 0;
 310	rcu_read_lock();
 311	if (mode & PTRACE_MODE_FSCREDS) {
 312		caller_uid = cred->fsuid;
 313		caller_gid = cred->fsgid;
 314	} else {
 315		/*
 316		 * Using the euid would make more sense here, but something
 317		 * in userland might rely on the old behavior, and this
 318		 * shouldn't be a security problem since
 319		 * PTRACE_MODE_REALCREDS implies that the caller explicitly
 320		 * used a syscall that requests access to another process
 321		 * (and not a filesystem syscall to procfs).
 322		 */
 323		caller_uid = cred->uid;
 324		caller_gid = cred->gid;
 325	}
 326	tcred = __task_cred(task);
 327	if (uid_eq(caller_uid, tcred->euid) &&
 328	    uid_eq(caller_uid, tcred->suid) &&
 329	    uid_eq(caller_uid, tcred->uid)  &&
 330	    gid_eq(caller_gid, tcred->egid) &&
 331	    gid_eq(caller_gid, tcred->sgid) &&
 332	    gid_eq(caller_gid, tcred->gid))
 333		goto ok;
 334	if (ptrace_has_cap(tcred->user_ns, mode))
 335		goto ok;
 336	rcu_read_unlock();
 337	return -EPERM;
 338ok:
 339	rcu_read_unlock();
 340	/*
 341	 * If a task drops privileges and becomes nondumpable (through a syscall
 342	 * like setresuid()) while we are trying to access it, we must ensure
 343	 * that the dumpability is read after the credentials; otherwise,
 344	 * we may be able to attach to a task that we shouldn't be able to
 345	 * attach to (as if the task had dropped privileges without becoming
 346	 * nondumpable).
 347	 * Pairs with a write barrier in commit_creds().
 348	 */
 349	smp_rmb();
 350	mm = task->mm;
 351	if (mm &&
 352	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
 353	     !ptrace_has_cap(mm->user_ns, mode)))
 354	    return -EPERM;
 355
 356	return security_ptrace_access_check(task, mode);
 357}
 358
 359bool ptrace_may_access(struct task_struct *task, unsigned int mode)
 360{
 361	int err;
 362	task_lock(task);
 363	err = __ptrace_may_access(task, mode);
 364	task_unlock(task);
 365	return !err;
 366}
 367
 368static int check_ptrace_options(unsigned long data)
 369{
 370	if (data & ~(unsigned long)PTRACE_O_MASK)
 371		return -EINVAL;
 372
 373	if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
 374		if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
 375		    !IS_ENABLED(CONFIG_SECCOMP))
 376			return -EINVAL;
 377
 378		if (!capable(CAP_SYS_ADMIN))
 379			return -EPERM;
 380
 381		if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
 382		    current->ptrace & PT_SUSPEND_SECCOMP)
 383			return -EPERM;
 384	}
 385	return 0;
 386}
 387
 388static int ptrace_attach(struct task_struct *task, long request,
 389			 unsigned long addr,
 390			 unsigned long flags)
 391{
 392	bool seize = (request == PTRACE_SEIZE);
 393	int retval;
 394
 395	retval = -EIO;
 396	if (seize) {
 397		if (addr != 0)
 398			goto out;
 399		/*
 400		 * This duplicates the check in check_ptrace_options() because
 401		 * ptrace_attach() and ptrace_setoptions() have historically
 402		 * used different error codes for unknown ptrace options.
 403		 */
 404		if (flags & ~(unsigned long)PTRACE_O_MASK)
 405			goto out;
 406		retval = check_ptrace_options(flags);
 407		if (retval)
 408			return retval;
 409		flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
 410	} else {
 411		flags = PT_PTRACED;
 412	}
 413
 414	audit_ptrace(task);
 415
 416	retval = -EPERM;
 417	if (unlikely(task->flags & PF_KTHREAD))
 418		goto out;
 419	if (same_thread_group(task, current))
 420		goto out;
 421
 422	/*
 423	 * Protect exec's credential calculations against our interference;
 424	 * SUID, SGID and LSM creds get determined differently
 425	 * under ptrace.
 426	 */
 427	retval = -ERESTARTNOINTR;
 428	if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
 429		goto out;
 430
 431	task_lock(task);
 432	retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
 433	task_unlock(task);
 434	if (retval)
 435		goto unlock_creds;
 436
 437	write_lock_irq(&tasklist_lock);
 438	retval = -EPERM;
 439	if (unlikely(task->exit_state))
 440		goto unlock_tasklist;
 441	if (task->ptrace)
 442		goto unlock_tasklist;
 443
 
 
 444	task->ptrace = flags;
 445
 446	ptrace_link(task, current);
 447
 448	/* SEIZE doesn't trap tracee on attach */
 449	if (!seize)
 450		send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
 451
 452	spin_lock(&task->sighand->siglock);
 453
 454	/*
 455	 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
 456	 * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
 457	 * will be cleared if the child completes the transition or any
 458	 * event which clears the group stop states happens.  We'll wait
 459	 * for the transition to complete before returning from this
 460	 * function.
 461	 *
 462	 * This hides STOPPED -> RUNNING -> TRACED transition from the
 463	 * attaching thread but a different thread in the same group can
 464	 * still observe the transient RUNNING state.  IOW, if another
 465	 * thread's WNOHANG wait(2) on the stopped tracee races against
 466	 * ATTACH, the wait(2) may fail due to the transient RUNNING.
 467	 *
 468	 * The following task_is_stopped() test is safe as both transitions
 469	 * in and out of STOPPED are protected by siglock.
 470	 */
 471	if (task_is_stopped(task) &&
 472	    task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING)) {
 473		task->jobctl &= ~JOBCTL_STOPPED;
 474		signal_wake_up_state(task, __TASK_STOPPED);
 475	}
 476
 477	spin_unlock(&task->sighand->siglock);
 478
 479	retval = 0;
 480unlock_tasklist:
 481	write_unlock_irq(&tasklist_lock);
 482unlock_creds:
 483	mutex_unlock(&task->signal->cred_guard_mutex);
 484out:
 485	if (!retval) {
 486		/*
 487		 * We do not bother to change retval or clear JOBCTL_TRAPPING
 488		 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
 489		 * not return to user-mode, it will exit and clear this bit in
 490		 * __ptrace_unlink() if it wasn't already cleared by the tracee;
 491		 * and until then nobody can ptrace this task.
 492		 */
 493		wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
 494		proc_ptrace_connector(task, PTRACE_ATTACH);
 495	}
 496
 497	return retval;
 498}
 499
 500/**
 501 * ptrace_traceme  --  helper for PTRACE_TRACEME
 502 *
 503 * Performs checks and sets PT_PTRACED.
 504 * Should be used by all ptrace implementations for PTRACE_TRACEME.
 505 */
 506static int ptrace_traceme(void)
 507{
 508	int ret = -EPERM;
 509
 510	write_lock_irq(&tasklist_lock);
 511	/* Are we already being traced? */
 512	if (!current->ptrace) {
 513		ret = security_ptrace_traceme(current->parent);
 514		/*
 515		 * Check PF_EXITING to ensure ->real_parent has not passed
 516		 * exit_ptrace(). Otherwise we don't report the error but
 517		 * pretend ->real_parent untraces us right after return.
 518		 */
 519		if (!ret && !(current->real_parent->flags & PF_EXITING)) {
 520			current->ptrace = PT_PTRACED;
 521			ptrace_link(current, current->real_parent);
 522		}
 523	}
 524	write_unlock_irq(&tasklist_lock);
 525
 526	return ret;
 527}
 528
 529/*
 530 * Called with irqs disabled, returns true if childs should reap themselves.
 531 */
 532static int ignoring_children(struct sighand_struct *sigh)
 533{
 534	int ret;
 535	spin_lock(&sigh->siglock);
 536	ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
 537	      (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
 538	spin_unlock(&sigh->siglock);
 539	return ret;
 540}
 541
 542/*
 543 * Called with tasklist_lock held for writing.
 544 * Unlink a traced task, and clean it up if it was a traced zombie.
 545 * Return true if it needs to be reaped with release_task().
 546 * (We can't call release_task() here because we already hold tasklist_lock.)
 547 *
 548 * If it's a zombie, our attachedness prevented normal parent notification
 549 * or self-reaping.  Do notification now if it would have happened earlier.
 550 * If it should reap itself, return true.
 551 *
 552 * If it's our own child, there is no notification to do. But if our normal
 553 * children self-reap, then this child was prevented by ptrace and we must
 554 * reap it now, in that case we must also wake up sub-threads sleeping in
 555 * do_wait().
 556 */
 557static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
 558{
 559	bool dead;
 560
 561	__ptrace_unlink(p);
 562
 563	if (p->exit_state != EXIT_ZOMBIE)
 564		return false;
 565
 566	dead = !thread_group_leader(p);
 567
 568	if (!dead && thread_group_empty(p)) {
 569		if (!same_thread_group(p->real_parent, tracer))
 570			dead = do_notify_parent(p, p->exit_signal);
 571		else if (ignoring_children(tracer->sighand)) {
 572			__wake_up_parent(p, tracer);
 573			dead = true;
 574		}
 575	}
 576	/* Mark it as in the process of being reaped. */
 577	if (dead)
 578		p->exit_state = EXIT_DEAD;
 579	return dead;
 580}
 581
 582static int ptrace_detach(struct task_struct *child, unsigned int data)
 583{
 584	if (!valid_signal(data))
 585		return -EIO;
 586
 587	/* Architecture-specific hardware disable .. */
 588	ptrace_disable(child);
 589
 590	write_lock_irq(&tasklist_lock);
 591	/*
 592	 * We rely on ptrace_freeze_traced(). It can't be killed and
 593	 * untraced by another thread, it can't be a zombie.
 594	 */
 595	WARN_ON(!child->ptrace || child->exit_state);
 596	/*
 597	 * tasklist_lock avoids the race with wait_task_stopped(), see
 598	 * the comment in ptrace_resume().
 599	 */
 600	child->exit_code = data;
 601	__ptrace_detach(current, child);
 602	write_unlock_irq(&tasklist_lock);
 603
 604	proc_ptrace_connector(child, PTRACE_DETACH);
 605
 606	return 0;
 607}
 608
 609/*
 610 * Detach all tasks we were using ptrace on. Called with tasklist held
 611 * for writing.
 612 */
 613void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
 614{
 615	struct task_struct *p, *n;
 616
 617	list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
 618		if (unlikely(p->ptrace & PT_EXITKILL))
 619			send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
 620
 621		if (__ptrace_detach(tracer, p))
 622			list_add(&p->ptrace_entry, dead);
 623	}
 624}
 625
 626int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 627{
 628	int copied = 0;
 629
 630	while (len > 0) {
 631		char buf[128];
 632		int this_len, retval;
 633
 634		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 635		retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
 636
 637		if (!retval) {
 638			if (copied)
 639				break;
 640			return -EIO;
 641		}
 642		if (copy_to_user(dst, buf, retval))
 643			return -EFAULT;
 644		copied += retval;
 645		src += retval;
 646		dst += retval;
 647		len -= retval;
 648	}
 649	return copied;
 650}
 651
 652int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
 653{
 654	int copied = 0;
 655
 656	while (len > 0) {
 657		char buf[128];
 658		int this_len, retval;
 659
 660		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 661		if (copy_from_user(buf, src, this_len))
 662			return -EFAULT;
 663		retval = ptrace_access_vm(tsk, dst, buf, this_len,
 664				FOLL_FORCE | FOLL_WRITE);
 665		if (!retval) {
 666			if (copied)
 667				break;
 668			return -EIO;
 669		}
 670		copied += retval;
 671		src += retval;
 672		dst += retval;
 673		len -= retval;
 674	}
 675	return copied;
 676}
 677
 678static int ptrace_setoptions(struct task_struct *child, unsigned long data)
 679{
 680	unsigned flags;
 681	int ret;
 682
 683	ret = check_ptrace_options(data);
 684	if (ret)
 685		return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 686
 687	/* Avoid intermediate state when all opts are cleared */
 688	flags = child->ptrace;
 689	flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
 690	flags |= (data << PT_OPT_FLAG_SHIFT);
 691	child->ptrace = flags;
 692
 693	return 0;
 694}
 695
 696static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
 697{
 698	unsigned long flags;
 699	int error = -ESRCH;
 700
 701	if (lock_task_sighand(child, &flags)) {
 702		error = -EINVAL;
 703		if (likely(child->last_siginfo != NULL)) {
 704			copy_siginfo(info, child->last_siginfo);
 705			error = 0;
 706		}
 707		unlock_task_sighand(child, &flags);
 708	}
 709	return error;
 710}
 711
 712static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
 713{
 714	unsigned long flags;
 715	int error = -ESRCH;
 716
 717	if (lock_task_sighand(child, &flags)) {
 718		error = -EINVAL;
 719		if (likely(child->last_siginfo != NULL)) {
 720			copy_siginfo(child->last_siginfo, info);
 721			error = 0;
 722		}
 723		unlock_task_sighand(child, &flags);
 724	}
 725	return error;
 726}
 727
 728static int ptrace_peek_siginfo(struct task_struct *child,
 729				unsigned long addr,
 730				unsigned long data)
 731{
 732	struct ptrace_peeksiginfo_args arg;
 733	struct sigpending *pending;
 734	struct sigqueue *q;
 735	int ret, i;
 736
 737	ret = copy_from_user(&arg, (void __user *) addr,
 738				sizeof(struct ptrace_peeksiginfo_args));
 739	if (ret)
 740		return -EFAULT;
 741
 742	if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
 743		return -EINVAL; /* unknown flags */
 744
 745	if (arg.nr < 0)
 746		return -EINVAL;
 747
 748	/* Ensure arg.off fits in an unsigned long */
 749	if (arg.off > ULONG_MAX)
 750		return 0;
 751
 752	if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
 753		pending = &child->signal->shared_pending;
 754	else
 755		pending = &child->pending;
 756
 757	for (i = 0; i < arg.nr; ) {
 758		kernel_siginfo_t info;
 759		unsigned long off = arg.off + i;
 760		bool found = false;
 761
 762		spin_lock_irq(&child->sighand->siglock);
 763		list_for_each_entry(q, &pending->list, list) {
 764			if (!off--) {
 765				found = true;
 766				copy_siginfo(&info, &q->info);
 767				break;
 768			}
 769		}
 770		spin_unlock_irq(&child->sighand->siglock);
 771
 772		if (!found) /* beyond the end of the list */
 773			break;
 774
 775#ifdef CONFIG_COMPAT
 776		if (unlikely(in_compat_syscall())) {
 777			compat_siginfo_t __user *uinfo = compat_ptr(data);
 778
 779			if (copy_siginfo_to_user32(uinfo, &info)) {
 780				ret = -EFAULT;
 781				break;
 782			}
 783
 784		} else
 785#endif
 786		{
 787			siginfo_t __user *uinfo = (siginfo_t __user *) data;
 788
 789			if (copy_siginfo_to_user(uinfo, &info)) {
 790				ret = -EFAULT;
 791				break;
 792			}
 793		}
 794
 795		data += sizeof(siginfo_t);
 796		i++;
 797
 798		if (signal_pending(current))
 799			break;
 800
 801		cond_resched();
 802	}
 803
 804	if (i > 0)
 805		return i;
 806
 807	return ret;
 808}
 809
 810#ifdef CONFIG_RSEQ
 811static long ptrace_get_rseq_configuration(struct task_struct *task,
 812					  unsigned long size, void __user *data)
 813{
 814	struct ptrace_rseq_configuration conf = {
 815		.rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
 816		.rseq_abi_size = sizeof(*task->rseq),
 817		.signature = task->rseq_sig,
 818		.flags = 0,
 819	};
 820
 821	size = min_t(unsigned long, size, sizeof(conf));
 822	if (copy_to_user(data, &conf, size))
 823		return -EFAULT;
 824	return sizeof(conf);
 825}
 826#endif
 827
 
 828#define is_singlestep(request)		((request) == PTRACE_SINGLESTEP)
 
 
 
 829
 830#ifdef PTRACE_SINGLEBLOCK
 831#define is_singleblock(request)		((request) == PTRACE_SINGLEBLOCK)
 832#else
 833#define is_singleblock(request)		0
 834#endif
 835
 836#ifdef PTRACE_SYSEMU
 837#define is_sysemu_singlestep(request)	((request) == PTRACE_SYSEMU_SINGLESTEP)
 838#else
 839#define is_sysemu_singlestep(request)	0
 840#endif
 841
 842static int ptrace_resume(struct task_struct *child, long request,
 843			 unsigned long data)
 844{
 
 
 845	if (!valid_signal(data))
 846		return -EIO;
 847
 848	if (request == PTRACE_SYSCALL)
 849		set_task_syscall_work(child, SYSCALL_TRACE);
 850	else
 851		clear_task_syscall_work(child, SYSCALL_TRACE);
 852
 853#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
 854	if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 855		set_task_syscall_work(child, SYSCALL_EMU);
 856	else
 857		clear_task_syscall_work(child, SYSCALL_EMU);
 858#endif
 859
 860	if (is_singleblock(request)) {
 861		if (unlikely(!arch_has_block_step()))
 862			return -EIO;
 863		user_enable_block_step(child);
 864	} else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 865		if (unlikely(!arch_has_single_step()))
 866			return -EIO;
 867		user_enable_single_step(child);
 868	} else {
 869		user_disable_single_step(child);
 870	}
 871
 872	/*
 873	 * Change ->exit_code and ->state under siglock to avoid the race
 874	 * with wait_task_stopped() in between; a non-zero ->exit_code will
 875	 * wrongly look like another report from tracee.
 876	 *
 877	 * Note that we need siglock even if ->exit_code == data and/or this
 878	 * status was not reported yet, the new status must not be cleared by
 879	 * wait_task_stopped() after resume.
 
 
 
 
 880	 */
 881	spin_lock_irq(&child->sighand->siglock);
 
 
 882	child->exit_code = data;
 883	child->jobctl &= ~JOBCTL_TRACED;
 884	wake_up_state(child, __TASK_TRACED);
 885	spin_unlock_irq(&child->sighand->siglock);
 
 886
 887	return 0;
 888}
 889
 890#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 891
 892static const struct user_regset *
 893find_regset(const struct user_regset_view *view, unsigned int type)
 894{
 895	const struct user_regset *regset;
 896	int n;
 897
 898	for (n = 0; n < view->n; ++n) {
 899		regset = view->regsets + n;
 900		if (regset->core_note_type == type)
 901			return regset;
 902	}
 903
 904	return NULL;
 905}
 906
 907static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
 908			 struct iovec *kiov)
 909{
 910	const struct user_regset_view *view = task_user_regset_view(task);
 911	const struct user_regset *regset = find_regset(view, type);
 912	int regset_no;
 913
 914	if (!regset || (kiov->iov_len % regset->size) != 0)
 915		return -EINVAL;
 916
 917	regset_no = regset - view->regsets;
 918	kiov->iov_len = min(kiov->iov_len,
 919			    (__kernel_size_t) (regset->n * regset->size));
 920
 921	if (req == PTRACE_GETREGSET)
 922		return copy_regset_to_user(task, view, regset_no, 0,
 923					   kiov->iov_len, kiov->iov_base);
 924	else
 925		return copy_regset_from_user(task, view, regset_no, 0,
 926					     kiov->iov_len, kiov->iov_base);
 927}
 928
 929/*
 930 * This is declared in linux/regset.h and defined in machine-dependent
 931 * code.  We put the export here, near the primary machine-neutral use,
 932 * to ensure no machine forgets it.
 933 */
 934EXPORT_SYMBOL_GPL(task_user_regset_view);
 935
 936static unsigned long
 937ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
 938			      struct ptrace_syscall_info *info)
 939{
 940	unsigned long args[ARRAY_SIZE(info->entry.args)];
 941	int i;
 942
 943	info->op = PTRACE_SYSCALL_INFO_ENTRY;
 944	info->entry.nr = syscall_get_nr(child, regs);
 945	syscall_get_arguments(child, regs, args);
 946	for (i = 0; i < ARRAY_SIZE(args); i++)
 947		info->entry.args[i] = args[i];
 948
 949	/* args is the last field in struct ptrace_syscall_info.entry */
 950	return offsetofend(struct ptrace_syscall_info, entry.args);
 951}
 952
 953static unsigned long
 954ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
 955				struct ptrace_syscall_info *info)
 956{
 957	/*
 958	 * As struct ptrace_syscall_info.entry is currently a subset
 959	 * of struct ptrace_syscall_info.seccomp, it makes sense to
 960	 * initialize that subset using ptrace_get_syscall_info_entry().
 961	 * This can be reconsidered in the future if these structures
 962	 * diverge significantly enough.
 963	 */
 964	ptrace_get_syscall_info_entry(child, regs, info);
 965	info->op = PTRACE_SYSCALL_INFO_SECCOMP;
 966	info->seccomp.ret_data = child->ptrace_message;
 967
 968	/* ret_data is the last field in struct ptrace_syscall_info.seccomp */
 969	return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
 970}
 971
 972static unsigned long
 973ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
 974			     struct ptrace_syscall_info *info)
 975{
 976	info->op = PTRACE_SYSCALL_INFO_EXIT;
 977	info->exit.rval = syscall_get_error(child, regs);
 978	info->exit.is_error = !!info->exit.rval;
 979	if (!info->exit.is_error)
 980		info->exit.rval = syscall_get_return_value(child, regs);
 981
 982	/* is_error is the last field in struct ptrace_syscall_info.exit */
 983	return offsetofend(struct ptrace_syscall_info, exit.is_error);
 984}
 985
 986static int
 987ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
 988			void __user *datavp)
 989{
 990	struct pt_regs *regs = task_pt_regs(child);
 991	struct ptrace_syscall_info info = {
 992		.op = PTRACE_SYSCALL_INFO_NONE,
 993		.arch = syscall_get_arch(child),
 994		.instruction_pointer = instruction_pointer(regs),
 995		.stack_pointer = user_stack_pointer(regs),
 996	};
 997	unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
 998	unsigned long write_size;
 999
1000	/*
1001	 * This does not need lock_task_sighand() to access
1002	 * child->last_siginfo because ptrace_freeze_traced()
1003	 * called earlier by ptrace_check_attach() ensures that
1004	 * the tracee cannot go away and clear its last_siginfo.
1005	 */
1006	switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
1007	case SIGTRAP | 0x80:
1008		switch (child->ptrace_message) {
1009		case PTRACE_EVENTMSG_SYSCALL_ENTRY:
1010			actual_size = ptrace_get_syscall_info_entry(child, regs,
1011								    &info);
1012			break;
1013		case PTRACE_EVENTMSG_SYSCALL_EXIT:
1014			actual_size = ptrace_get_syscall_info_exit(child, regs,
1015								   &info);
1016			break;
1017		}
1018		break;
1019	case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1020		actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1021							      &info);
1022		break;
1023	}
1024
1025	write_size = min(actual_size, user_size);
1026	return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1027}
1028#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1029
1030int ptrace_request(struct task_struct *child, long request,
1031		   unsigned long addr, unsigned long data)
1032{
1033	bool seized = child->ptrace & PT_SEIZED;
1034	int ret = -EIO;
1035	kernel_siginfo_t siginfo, *si;
1036	void __user *datavp = (void __user *) data;
1037	unsigned long __user *datalp = datavp;
1038	unsigned long flags;
1039
1040	switch (request) {
1041	case PTRACE_PEEKTEXT:
1042	case PTRACE_PEEKDATA:
1043		return generic_ptrace_peekdata(child, addr, data);
1044	case PTRACE_POKETEXT:
1045	case PTRACE_POKEDATA:
1046		return generic_ptrace_pokedata(child, addr, data);
1047
1048#ifdef PTRACE_OLDSETOPTIONS
1049	case PTRACE_OLDSETOPTIONS:
1050#endif
1051	case PTRACE_SETOPTIONS:
1052		ret = ptrace_setoptions(child, data);
1053		break;
1054	case PTRACE_GETEVENTMSG:
1055		ret = put_user(child->ptrace_message, datalp);
1056		break;
1057
1058	case PTRACE_PEEKSIGINFO:
1059		ret = ptrace_peek_siginfo(child, addr, data);
1060		break;
1061
1062	case PTRACE_GETSIGINFO:
1063		ret = ptrace_getsiginfo(child, &siginfo);
1064		if (!ret)
1065			ret = copy_siginfo_to_user(datavp, &siginfo);
1066		break;
1067
1068	case PTRACE_SETSIGINFO:
1069		ret = copy_siginfo_from_user(&siginfo, datavp);
1070		if (!ret)
1071			ret = ptrace_setsiginfo(child, &siginfo);
1072		break;
1073
1074	case PTRACE_GETSIGMASK: {
1075		sigset_t *mask;
1076
1077		if (addr != sizeof(sigset_t)) {
1078			ret = -EINVAL;
1079			break;
1080		}
1081
1082		if (test_tsk_restore_sigmask(child))
1083			mask = &child->saved_sigmask;
1084		else
1085			mask = &child->blocked;
1086
1087		if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1088			ret = -EFAULT;
1089		else
1090			ret = 0;
1091
1092		break;
1093	}
1094
1095	case PTRACE_SETSIGMASK: {
1096		sigset_t new_set;
1097
1098		if (addr != sizeof(sigset_t)) {
1099			ret = -EINVAL;
1100			break;
1101		}
1102
1103		if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1104			ret = -EFAULT;
1105			break;
1106		}
1107
1108		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1109
1110		/*
1111		 * Every thread does recalc_sigpending() after resume, so
1112		 * retarget_shared_pending() and recalc_sigpending() are not
1113		 * called here.
1114		 */
1115		spin_lock_irq(&child->sighand->siglock);
1116		child->blocked = new_set;
1117		spin_unlock_irq(&child->sighand->siglock);
1118
1119		clear_tsk_restore_sigmask(child);
1120
1121		ret = 0;
1122		break;
1123	}
1124
1125	case PTRACE_INTERRUPT:
1126		/*
1127		 * Stop tracee without any side-effect on signal or job
1128		 * control.  At least one trap is guaranteed to happen
1129		 * after this request.  If @child is already trapped, the
1130		 * current trap is not disturbed and another trap will
1131		 * happen after the current trap is ended with PTRACE_CONT.
1132		 *
1133		 * The actual trap might not be PTRACE_EVENT_STOP trap but
1134		 * the pending condition is cleared regardless.
1135		 */
1136		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1137			break;
1138
1139		/*
1140		 * INTERRUPT doesn't disturb existing trap sans one
1141		 * exception.  If ptracer issued LISTEN for the current
1142		 * STOP, this INTERRUPT should clear LISTEN and re-trap
1143		 * tracee into STOP.
1144		 */
1145		if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1146			ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1147
1148		unlock_task_sighand(child, &flags);
1149		ret = 0;
1150		break;
1151
1152	case PTRACE_LISTEN:
1153		/*
1154		 * Listen for events.  Tracee must be in STOP.  It's not
1155		 * resumed per-se but is not considered to be in TRACED by
1156		 * wait(2) or ptrace(2).  If an async event (e.g. group
1157		 * stop state change) happens, tracee will enter STOP trap
1158		 * again.  Alternatively, ptracer can issue INTERRUPT to
1159		 * finish listening and re-trap tracee into STOP.
1160		 */
1161		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1162			break;
1163
1164		si = child->last_siginfo;
1165		if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1166			child->jobctl |= JOBCTL_LISTENING;
1167			/*
1168			 * If NOTIFY is set, it means event happened between
1169			 * start of this trap and now.  Trigger re-trap.
1170			 */
1171			if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1172				ptrace_signal_wake_up(child, true);
1173			ret = 0;
1174		}
1175		unlock_task_sighand(child, &flags);
1176		break;
1177
1178	case PTRACE_DETACH:	 /* detach a process that was attached. */
1179		ret = ptrace_detach(child, data);
1180		break;
1181
1182#ifdef CONFIG_BINFMT_ELF_FDPIC
1183	case PTRACE_GETFDPIC: {
1184		struct mm_struct *mm = get_task_mm(child);
1185		unsigned long tmp = 0;
1186
1187		ret = -ESRCH;
1188		if (!mm)
1189			break;
1190
1191		switch (addr) {
1192		case PTRACE_GETFDPIC_EXEC:
1193			tmp = mm->context.exec_fdpic_loadmap;
1194			break;
1195		case PTRACE_GETFDPIC_INTERP:
1196			tmp = mm->context.interp_fdpic_loadmap;
1197			break;
1198		default:
1199			break;
1200		}
1201		mmput(mm);
1202
1203		ret = put_user(tmp, datalp);
1204		break;
1205	}
1206#endif
1207
 
1208	case PTRACE_SINGLESTEP:
 
1209#ifdef PTRACE_SINGLEBLOCK
1210	case PTRACE_SINGLEBLOCK:
1211#endif
1212#ifdef PTRACE_SYSEMU
1213	case PTRACE_SYSEMU:
1214	case PTRACE_SYSEMU_SINGLESTEP:
1215#endif
1216	case PTRACE_SYSCALL:
1217	case PTRACE_CONT:
1218		return ptrace_resume(child, request, data);
1219
1220	case PTRACE_KILL:
1221		send_sig_info(SIGKILL, SEND_SIG_NOINFO, child);
1222		return 0;
 
1223
1224#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1225	case PTRACE_GETREGSET:
1226	case PTRACE_SETREGSET: {
1227		struct iovec kiov;
1228		struct iovec __user *uiov = datavp;
1229
1230		if (!access_ok(uiov, sizeof(*uiov)))
1231			return -EFAULT;
1232
1233		if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1234		    __get_user(kiov.iov_len, &uiov->iov_len))
1235			return -EFAULT;
1236
1237		ret = ptrace_regset(child, request, addr, &kiov);
1238		if (!ret)
1239			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1240		break;
1241	}
1242
1243	case PTRACE_GET_SYSCALL_INFO:
1244		ret = ptrace_get_syscall_info(child, addr, datavp);
1245		break;
1246#endif
1247
1248	case PTRACE_SECCOMP_GET_FILTER:
1249		ret = seccomp_get_filter(child, addr, datavp);
1250		break;
1251
1252	case PTRACE_SECCOMP_GET_METADATA:
1253		ret = seccomp_get_metadata(child, addr, datavp);
1254		break;
1255
1256#ifdef CONFIG_RSEQ
1257	case PTRACE_GET_RSEQ_CONFIGURATION:
1258		ret = ptrace_get_rseq_configuration(child, addr, datavp);
1259		break;
1260#endif
1261
1262	default:
1263		break;
1264	}
1265
1266	return ret;
1267}
1268
 
 
 
 
1269SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1270		unsigned long, data)
1271{
1272	struct task_struct *child;
1273	long ret;
1274
1275	if (request == PTRACE_TRACEME) {
1276		ret = ptrace_traceme();
 
 
1277		goto out;
1278	}
1279
1280	child = find_get_task_by_vpid(pid);
1281	if (!child) {
1282		ret = -ESRCH;
1283		goto out;
1284	}
1285
1286	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1287		ret = ptrace_attach(child, request, addr, data);
 
 
 
 
 
 
1288		goto out_put_task_struct;
1289	}
1290
1291	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1292				  request == PTRACE_INTERRUPT);
1293	if (ret < 0)
1294		goto out_put_task_struct;
1295
1296	ret = arch_ptrace(child, request, addr, data);
1297	if (ret || request != PTRACE_DETACH)
1298		ptrace_unfreeze_traced(child);
1299
1300 out_put_task_struct:
1301	put_task_struct(child);
1302 out:
1303	return ret;
1304}
1305
1306int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1307			    unsigned long data)
1308{
1309	unsigned long tmp;
1310	int copied;
1311
1312	copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1313	if (copied != sizeof(tmp))
1314		return -EIO;
1315	return put_user(tmp, (unsigned long __user *)data);
1316}
1317
1318int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1319			    unsigned long data)
1320{
1321	int copied;
1322
1323	copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1324			FOLL_FORCE | FOLL_WRITE);
1325	return (copied == sizeof(data)) ? 0 : -EIO;
1326}
1327
1328#if defined CONFIG_COMPAT
1329
1330int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1331			  compat_ulong_t addr, compat_ulong_t data)
1332{
1333	compat_ulong_t __user *datap = compat_ptr(data);
1334	compat_ulong_t word;
1335	kernel_siginfo_t siginfo;
1336	int ret;
1337
1338	switch (request) {
1339	case PTRACE_PEEKTEXT:
1340	case PTRACE_PEEKDATA:
1341		ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1342				FOLL_FORCE);
1343		if (ret != sizeof(word))
1344			ret = -EIO;
1345		else
1346			ret = put_user(word, datap);
1347		break;
1348
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);
1354		break;
1355
1356	case PTRACE_GETEVENTMSG:
1357		ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1358		break;
1359
1360	case PTRACE_GETSIGINFO:
1361		ret = ptrace_getsiginfo(child, &siginfo);
1362		if (!ret)
1363			ret = copy_siginfo_to_user32(
1364				(struct compat_siginfo __user *) datap,
1365				&siginfo);
1366		break;
1367
1368	case PTRACE_SETSIGINFO:
1369		ret = copy_siginfo_from_user32(
1370			&siginfo, (struct compat_siginfo __user *) datap);
1371		if (!ret)
1372			ret = ptrace_setsiginfo(child, &siginfo);
1373		break;
1374#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1375	case PTRACE_GETREGSET:
1376	case PTRACE_SETREGSET:
1377	{
1378		struct iovec kiov;
1379		struct compat_iovec __user *uiov =
1380			(struct compat_iovec __user *) datap;
1381		compat_uptr_t ptr;
1382		compat_size_t len;
1383
1384		if (!access_ok(uiov, sizeof(*uiov)))
1385			return -EFAULT;
1386
1387		if (__get_user(ptr, &uiov->iov_base) ||
1388		    __get_user(len, &uiov->iov_len))
1389			return -EFAULT;
1390
1391		kiov.iov_base = compat_ptr(ptr);
1392		kiov.iov_len = len;
1393
1394		ret = ptrace_regset(child, request, addr, &kiov);
1395		if (!ret)
1396			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1397		break;
1398	}
1399#endif
1400
1401	default:
1402		ret = ptrace_request(child, request, addr, data);
1403	}
1404
1405	return ret;
1406}
1407
1408COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1409		       compat_long_t, addr, compat_long_t, data)
1410{
1411	struct task_struct *child;
1412	long ret;
1413
1414	if (request == PTRACE_TRACEME) {
1415		ret = ptrace_traceme();
1416		goto out;
1417	}
1418
1419	child = find_get_task_by_vpid(pid);
1420	if (!child) {
1421		ret = -ESRCH;
1422		goto out;
1423	}
1424
1425	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1426		ret = ptrace_attach(child, request, addr, data);
 
 
 
 
 
 
1427		goto out_put_task_struct;
1428	}
1429
1430	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1431				  request == PTRACE_INTERRUPT);
1432	if (!ret) {
1433		ret = compat_arch_ptrace(child, request, addr, data);
1434		if (ret || request != PTRACE_DETACH)
1435			ptrace_unfreeze_traced(child);
1436	}
1437
1438 out_put_task_struct:
1439	put_task_struct(child);
1440 out:
1441	return ret;
1442}
1443#endif	/* CONFIG_COMPAT */
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * linux/kernel/ptrace.c
   4 *
   5 * (C) Copyright 1999 Linus Torvalds
   6 *
   7 * Common interfaces for "ptrace()" which we do not want
   8 * to continually duplicate across every architecture.
   9 */
  10
  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>
  18#include <linux/mm.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>
  35
  36#include <asm/syscall.h>	/* for syscall_get_* */
  37
  38/*
  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
  42 */
  43int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
  44		     void *buf, int len, unsigned int gup_flags)
  45{
  46	struct mm_struct *mm;
  47	int ret;
  48
  49	mm = get_task_mm(tsk);
  50	if (!mm)
  51		return 0;
  52
  53	if (!tsk->ptrace ||
  54	    (current != tsk->parent) ||
  55	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
  56	     !ptracer_capable(tsk, mm->user_ns))) {
  57		mmput(mm);
  58		return 0;
  59	}
  60
  61	ret = __access_remote_vm(mm, addr, buf, len, gup_flags);
  62	mmput(mm);
  63
  64	return ret;
  65}
  66
  67
  68void __ptrace_link(struct task_struct *child, struct task_struct *new_parent,
  69		   const struct cred *ptracer_cred)
  70{
  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);
  75}
  76
  77/*
  78 * ptrace a task: make the debugger its new parent and
  79 * move it to the ptrace list.
  80 *
  81 * Must be called with the tasklist lock write-held.
  82 */
  83static void ptrace_link(struct task_struct *child, struct task_struct *new_parent)
  84{
  85	__ptrace_link(child, new_parent, current_cred());
  86}
  87
  88/**
  89 * __ptrace_unlink - unlink ptracee and restore its execution state
  90 * @child: ptracee to be unlinked
  91 *
  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
  94 * state.
  95 *
  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.
 100 *
 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.
 105 *
 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.
 112 *
 113 * CONTEXT:
 114 * write_lock_irq(tasklist_lock)
 115 */
 116void __ptrace_unlink(struct task_struct *child)
 117{
 118	const struct cred *old_cred;
 119	BUG_ON(!child->ptrace);
 120
 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);
 124#endif
 125
 126	child->parent = child->real_parent;
 127	list_del_init(&child->ptrace_entry);
 128	old_cred = child->ptracer_cred;
 129	child->ptracer_cred = NULL;
 130	put_cred(old_cred);
 131
 132	spin_lock(&child->sighand->siglock);
 133	child->ptrace = 0;
 134	/*
 135	 * Clear all pending traps and TRAPPING.  TRAPPING should be
 136	 * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
 137	 */
 138	task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
 139	task_clear_jobctl_trapping(child);
 140
 141	/*
 142	 * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
 143	 * @child isn't dead.
 144	 */
 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;
 149
 150		/*
 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
 155		 * case.
 156		 */
 157		if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
 158			child->jobctl |= SIGSTOP;
 159	}
 160
 161	/*
 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.
 166	 */
 167	if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
 168		ptrace_signal_wake_up(child, true);
 169
 170	spin_unlock(&child->sighand->siglock);
 171}
 172
 173static bool looks_like_a_spurious_pid(struct task_struct *task)
 174{
 175	if (task->exit_code != ((PTRACE_EVENT_EXEC << 8) | SIGTRAP))
 176		return false;
 177
 178	if (task_pid_vnr(task) == task->ptrace_message)
 179		return false;
 180	/*
 181	 * The tracee changed its pid but the PTRACE_EVENT_EXEC event
 182	 * was not wait()'ed, most probably debugger targets the old
 183	 * leader which was destroyed in de_thread().
 184	 */
 185	return true;
 186}
 187
 188/* Ensure that nothing can wake it up, even SIGKILL */
 
 
 
 
 
 189static bool ptrace_freeze_traced(struct task_struct *task)
 190{
 191	bool ret = false;
 192
 193	/* Lockless, nobody but us can set this flag */
 194	if (task->jobctl & JOBCTL_LISTENING)
 195		return ret;
 196
 197	spin_lock_irq(&task->sighand->siglock);
 198	if (task_is_traced(task) && !looks_like_a_spurious_pid(task) &&
 199	    !__fatal_signal_pending(task)) {
 200		WRITE_ONCE(task->__state, __TASK_TRACED);
 201		ret = true;
 202	}
 203	spin_unlock_irq(&task->sighand->siglock);
 204
 205	return ret;
 206}
 207
 208static void ptrace_unfreeze_traced(struct task_struct *task)
 209{
 210	if (READ_ONCE(task->__state) != __TASK_TRACED)
 211		return;
 212
 213	WARN_ON(!task->ptrace || task->parent != current);
 214
 215	/*
 216	 * PTRACE_LISTEN can allow ptrace_trap_notify to wake us up remotely.
 217	 * Recheck state under the lock to close this race.
 
 218	 */
 219	spin_lock_irq(&task->sighand->siglock);
 220	if (READ_ONCE(task->__state) == __TASK_TRACED) {
 221		if (__fatal_signal_pending(task))
 
 222			wake_up_state(task, __TASK_TRACED);
 223		else
 224			WRITE_ONCE(task->__state, TASK_TRACED);
 225	}
 226	spin_unlock_irq(&task->sighand->siglock);
 227}
 228
 229/**
 230 * ptrace_check_attach - check whether ptracee is ready for ptrace operation
 231 * @child: ptracee to check for
 232 * @ignore_state: don't check whether @child is currently %TASK_TRACED
 233 *
 234 * Check whether @child is being ptraced by %current and ready for further
 235 * ptrace operations.  If @ignore_state is %false, @child also should be in
 236 * %TASK_TRACED state and on return the child is guaranteed to be traced
 237 * and not executing.  If @ignore_state is %true, @child can be in any
 238 * state.
 239 *
 240 * CONTEXT:
 241 * Grabs and releases tasklist_lock and @child->sighand->siglock.
 242 *
 243 * RETURNS:
 244 * 0 on success, -ESRCH if %child is not ready.
 245 */
 246static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
 247{
 248	int ret = -ESRCH;
 249
 250	/*
 251	 * We take the read lock around doing both checks to close a
 252	 * possible race where someone else was tracing our child and
 253	 * detached between these two checks.  After this locked check,
 254	 * we are sure that this is our traced child and that can only
 255	 * be changed by us so it's not changing right after this.
 256	 */
 257	read_lock(&tasklist_lock);
 258	if (child->ptrace && child->parent == current) {
 259		WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
 260		/*
 261		 * child->sighand can't be NULL, release_task()
 262		 * does ptrace_unlink() before __exit_signal().
 263		 */
 264		if (ignore_state || ptrace_freeze_traced(child))
 265			ret = 0;
 266	}
 267	read_unlock(&tasklist_lock);
 268
 269	if (!ret && !ignore_state) {
 270		if (!wait_task_inactive(child, __TASK_TRACED)) {
 271			/*
 272			 * This can only happen if may_ptrace_stop() fails and
 273			 * ptrace_stop() changes ->state back to TASK_RUNNING,
 274			 * so we should not worry about leaking __TASK_TRACED.
 275			 */
 276			WARN_ON(READ_ONCE(child->__state) == __TASK_TRACED);
 277			ret = -ESRCH;
 278		}
 279	}
 280
 281	return ret;
 282}
 283
 284static bool ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
 285{
 286	if (mode & PTRACE_MODE_NOAUDIT)
 287		return ns_capable_noaudit(ns, CAP_SYS_PTRACE);
 288	return ns_capable(ns, CAP_SYS_PTRACE);
 289}
 290
 291/* Returns 0 on success, -errno on denial. */
 292static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
 293{
 294	const struct cred *cred = current_cred(), *tcred;
 295	struct mm_struct *mm;
 296	kuid_t caller_uid;
 297	kgid_t caller_gid;
 298
 299	if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
 300		WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
 301		return -EPERM;
 302	}
 303
 304	/* May we inspect the given task?
 305	 * This check is used both for attaching with ptrace
 306	 * and for allowing access to sensitive information in /proc.
 307	 *
 308	 * ptrace_attach denies several cases that /proc allows
 309	 * because setting up the necessary parent/child relationship
 310	 * or halting the specified task is impossible.
 311	 */
 312
 313	/* Don't let security modules deny introspection */
 314	if (same_thread_group(task, current))
 315		return 0;
 316	rcu_read_lock();
 317	if (mode & PTRACE_MODE_FSCREDS) {
 318		caller_uid = cred->fsuid;
 319		caller_gid = cred->fsgid;
 320	} else {
 321		/*
 322		 * Using the euid would make more sense here, but something
 323		 * in userland might rely on the old behavior, and this
 324		 * shouldn't be a security problem since
 325		 * PTRACE_MODE_REALCREDS implies that the caller explicitly
 326		 * used a syscall that requests access to another process
 327		 * (and not a filesystem syscall to procfs).
 328		 */
 329		caller_uid = cred->uid;
 330		caller_gid = cred->gid;
 331	}
 332	tcred = __task_cred(task);
 333	if (uid_eq(caller_uid, tcred->euid) &&
 334	    uid_eq(caller_uid, tcred->suid) &&
 335	    uid_eq(caller_uid, tcred->uid)  &&
 336	    gid_eq(caller_gid, tcred->egid) &&
 337	    gid_eq(caller_gid, tcred->sgid) &&
 338	    gid_eq(caller_gid, tcred->gid))
 339		goto ok;
 340	if (ptrace_has_cap(tcred->user_ns, mode))
 341		goto ok;
 342	rcu_read_unlock();
 343	return -EPERM;
 344ok:
 345	rcu_read_unlock();
 346	/*
 347	 * If a task drops privileges and becomes nondumpable (through a syscall
 348	 * like setresuid()) while we are trying to access it, we must ensure
 349	 * that the dumpability is read after the credentials; otherwise,
 350	 * we may be able to attach to a task that we shouldn't be able to
 351	 * attach to (as if the task had dropped privileges without becoming
 352	 * nondumpable).
 353	 * Pairs with a write barrier in commit_creds().
 354	 */
 355	smp_rmb();
 356	mm = task->mm;
 357	if (mm &&
 358	    ((get_dumpable(mm) != SUID_DUMP_USER) &&
 359	     !ptrace_has_cap(mm->user_ns, mode)))
 360	    return -EPERM;
 361
 362	return security_ptrace_access_check(task, mode);
 363}
 364
 365bool ptrace_may_access(struct task_struct *task, unsigned int mode)
 366{
 367	int err;
 368	task_lock(task);
 369	err = __ptrace_may_access(task, mode);
 370	task_unlock(task);
 371	return !err;
 372}
 373
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 374static int ptrace_attach(struct task_struct *task, long request,
 375			 unsigned long addr,
 376			 unsigned long flags)
 377{
 378	bool seize = (request == PTRACE_SEIZE);
 379	int retval;
 380
 381	retval = -EIO;
 382	if (seize) {
 383		if (addr != 0)
 384			goto out;
 
 
 
 
 
 385		if (flags & ~(unsigned long)PTRACE_O_MASK)
 386			goto out;
 
 
 
 387		flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
 388	} else {
 389		flags = PT_PTRACED;
 390	}
 391
 392	audit_ptrace(task);
 393
 394	retval = -EPERM;
 395	if (unlikely(task->flags & PF_KTHREAD))
 396		goto out;
 397	if (same_thread_group(task, current))
 398		goto out;
 399
 400	/*
 401	 * Protect exec's credential calculations against our interference;
 402	 * SUID, SGID and LSM creds get determined differently
 403	 * under ptrace.
 404	 */
 405	retval = -ERESTARTNOINTR;
 406	if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
 407		goto out;
 408
 409	task_lock(task);
 410	retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
 411	task_unlock(task);
 412	if (retval)
 413		goto unlock_creds;
 414
 415	write_lock_irq(&tasklist_lock);
 416	retval = -EPERM;
 417	if (unlikely(task->exit_state))
 418		goto unlock_tasklist;
 419	if (task->ptrace)
 420		goto unlock_tasklist;
 421
 422	if (seize)
 423		flags |= PT_SEIZED;
 424	task->ptrace = flags;
 425
 426	ptrace_link(task, current);
 427
 428	/* SEIZE doesn't trap tracee on attach */
 429	if (!seize)
 430		send_sig_info(SIGSTOP, SEND_SIG_PRIV, task);
 431
 432	spin_lock(&task->sighand->siglock);
 433
 434	/*
 435	 * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
 436	 * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
 437	 * will be cleared if the child completes the transition or any
 438	 * event which clears the group stop states happens.  We'll wait
 439	 * for the transition to complete before returning from this
 440	 * function.
 441	 *
 442	 * This hides STOPPED -> RUNNING -> TRACED transition from the
 443	 * attaching thread but a different thread in the same group can
 444	 * still observe the transient RUNNING state.  IOW, if another
 445	 * thread's WNOHANG wait(2) on the stopped tracee races against
 446	 * ATTACH, the wait(2) may fail due to the transient RUNNING.
 447	 *
 448	 * The following task_is_stopped() test is safe as both transitions
 449	 * in and out of STOPPED are protected by siglock.
 450	 */
 451	if (task_is_stopped(task) &&
 452	    task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
 
 453		signal_wake_up_state(task, __TASK_STOPPED);
 
 454
 455	spin_unlock(&task->sighand->siglock);
 456
 457	retval = 0;
 458unlock_tasklist:
 459	write_unlock_irq(&tasklist_lock);
 460unlock_creds:
 461	mutex_unlock(&task->signal->cred_guard_mutex);
 462out:
 463	if (!retval) {
 464		/*
 465		 * We do not bother to change retval or clear JOBCTL_TRAPPING
 466		 * if wait_on_bit() was interrupted by SIGKILL. The tracer will
 467		 * not return to user-mode, it will exit and clear this bit in
 468		 * __ptrace_unlink() if it wasn't already cleared by the tracee;
 469		 * and until then nobody can ptrace this task.
 470		 */
 471		wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
 472		proc_ptrace_connector(task, PTRACE_ATTACH);
 473	}
 474
 475	return retval;
 476}
 477
 478/**
 479 * ptrace_traceme  --  helper for PTRACE_TRACEME
 480 *
 481 * Performs checks and sets PT_PTRACED.
 482 * Should be used by all ptrace implementations for PTRACE_TRACEME.
 483 */
 484static int ptrace_traceme(void)
 485{
 486	int ret = -EPERM;
 487
 488	write_lock_irq(&tasklist_lock);
 489	/* Are we already being traced? */
 490	if (!current->ptrace) {
 491		ret = security_ptrace_traceme(current->parent);
 492		/*
 493		 * Check PF_EXITING to ensure ->real_parent has not passed
 494		 * exit_ptrace(). Otherwise we don't report the error but
 495		 * pretend ->real_parent untraces us right after return.
 496		 */
 497		if (!ret && !(current->real_parent->flags & PF_EXITING)) {
 498			current->ptrace = PT_PTRACED;
 499			ptrace_link(current, current->real_parent);
 500		}
 501	}
 502	write_unlock_irq(&tasklist_lock);
 503
 504	return ret;
 505}
 506
 507/*
 508 * Called with irqs disabled, returns true if childs should reap themselves.
 509 */
 510static int ignoring_children(struct sighand_struct *sigh)
 511{
 512	int ret;
 513	spin_lock(&sigh->siglock);
 514	ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
 515	      (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
 516	spin_unlock(&sigh->siglock);
 517	return ret;
 518}
 519
 520/*
 521 * Called with tasklist_lock held for writing.
 522 * Unlink a traced task, and clean it up if it was a traced zombie.
 523 * Return true if it needs to be reaped with release_task().
 524 * (We can't call release_task() here because we already hold tasklist_lock.)
 525 *
 526 * If it's a zombie, our attachedness prevented normal parent notification
 527 * or self-reaping.  Do notification now if it would have happened earlier.
 528 * If it should reap itself, return true.
 529 *
 530 * If it's our own child, there is no notification to do. But if our normal
 531 * children self-reap, then this child was prevented by ptrace and we must
 532 * reap it now, in that case we must also wake up sub-threads sleeping in
 533 * do_wait().
 534 */
 535static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
 536{
 537	bool dead;
 538
 539	__ptrace_unlink(p);
 540
 541	if (p->exit_state != EXIT_ZOMBIE)
 542		return false;
 543
 544	dead = !thread_group_leader(p);
 545
 546	if (!dead && thread_group_empty(p)) {
 547		if (!same_thread_group(p->real_parent, tracer))
 548			dead = do_notify_parent(p, p->exit_signal);
 549		else if (ignoring_children(tracer->sighand)) {
 550			__wake_up_parent(p, tracer);
 551			dead = true;
 552		}
 553	}
 554	/* Mark it as in the process of being reaped. */
 555	if (dead)
 556		p->exit_state = EXIT_DEAD;
 557	return dead;
 558}
 559
 560static int ptrace_detach(struct task_struct *child, unsigned int data)
 561{
 562	if (!valid_signal(data))
 563		return -EIO;
 564
 565	/* Architecture-specific hardware disable .. */
 566	ptrace_disable(child);
 567
 568	write_lock_irq(&tasklist_lock);
 569	/*
 570	 * We rely on ptrace_freeze_traced(). It can't be killed and
 571	 * untraced by another thread, it can't be a zombie.
 572	 */
 573	WARN_ON(!child->ptrace || child->exit_state);
 574	/*
 575	 * tasklist_lock avoids the race with wait_task_stopped(), see
 576	 * the comment in ptrace_resume().
 577	 */
 578	child->exit_code = data;
 579	__ptrace_detach(current, child);
 580	write_unlock_irq(&tasklist_lock);
 581
 582	proc_ptrace_connector(child, PTRACE_DETACH);
 583
 584	return 0;
 585}
 586
 587/*
 588 * Detach all tasks we were using ptrace on. Called with tasklist held
 589 * for writing.
 590 */
 591void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
 592{
 593	struct task_struct *p, *n;
 594
 595	list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
 596		if (unlikely(p->ptrace & PT_EXITKILL))
 597			send_sig_info(SIGKILL, SEND_SIG_PRIV, p);
 598
 599		if (__ptrace_detach(tracer, p))
 600			list_add(&p->ptrace_entry, dead);
 601	}
 602}
 603
 604int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 605{
 606	int copied = 0;
 607
 608	while (len > 0) {
 609		char buf[128];
 610		int this_len, retval;
 611
 612		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 613		retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
 614
 615		if (!retval) {
 616			if (copied)
 617				break;
 618			return -EIO;
 619		}
 620		if (copy_to_user(dst, buf, retval))
 621			return -EFAULT;
 622		copied += retval;
 623		src += retval;
 624		dst += retval;
 625		len -= retval;
 626	}
 627	return copied;
 628}
 629
 630int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
 631{
 632	int copied = 0;
 633
 634	while (len > 0) {
 635		char buf[128];
 636		int this_len, retval;
 637
 638		this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
 639		if (copy_from_user(buf, src, this_len))
 640			return -EFAULT;
 641		retval = ptrace_access_vm(tsk, dst, buf, this_len,
 642				FOLL_FORCE | FOLL_WRITE);
 643		if (!retval) {
 644			if (copied)
 645				break;
 646			return -EIO;
 647		}
 648		copied += retval;
 649		src += retval;
 650		dst += retval;
 651		len -= retval;
 652	}
 653	return copied;
 654}
 655
 656static int ptrace_setoptions(struct task_struct *child, unsigned long data)
 657{
 658	unsigned flags;
 
 659
 660	if (data & ~(unsigned long)PTRACE_O_MASK)
 661		return -EINVAL;
 662
 663	if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
 664		if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
 665		    !IS_ENABLED(CONFIG_SECCOMP))
 666			return -EINVAL;
 667
 668		if (!capable(CAP_SYS_ADMIN))
 669			return -EPERM;
 670
 671		if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
 672		    current->ptrace & PT_SUSPEND_SECCOMP)
 673			return -EPERM;
 674	}
 675
 676	/* Avoid intermediate state when all opts are cleared */
 677	flags = child->ptrace;
 678	flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
 679	flags |= (data << PT_OPT_FLAG_SHIFT);
 680	child->ptrace = flags;
 681
 682	return 0;
 683}
 684
 685static int ptrace_getsiginfo(struct task_struct *child, kernel_siginfo_t *info)
 686{
 687	unsigned long flags;
 688	int error = -ESRCH;
 689
 690	if (lock_task_sighand(child, &flags)) {
 691		error = -EINVAL;
 692		if (likely(child->last_siginfo != NULL)) {
 693			copy_siginfo(info, child->last_siginfo);
 694			error = 0;
 695		}
 696		unlock_task_sighand(child, &flags);
 697	}
 698	return error;
 699}
 700
 701static int ptrace_setsiginfo(struct task_struct *child, const kernel_siginfo_t *info)
 702{
 703	unsigned long flags;
 704	int error = -ESRCH;
 705
 706	if (lock_task_sighand(child, &flags)) {
 707		error = -EINVAL;
 708		if (likely(child->last_siginfo != NULL)) {
 709			copy_siginfo(child->last_siginfo, info);
 710			error = 0;
 711		}
 712		unlock_task_sighand(child, &flags);
 713	}
 714	return error;
 715}
 716
 717static int ptrace_peek_siginfo(struct task_struct *child,
 718				unsigned long addr,
 719				unsigned long data)
 720{
 721	struct ptrace_peeksiginfo_args arg;
 722	struct sigpending *pending;
 723	struct sigqueue *q;
 724	int ret, i;
 725
 726	ret = copy_from_user(&arg, (void __user *) addr,
 727				sizeof(struct ptrace_peeksiginfo_args));
 728	if (ret)
 729		return -EFAULT;
 730
 731	if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
 732		return -EINVAL; /* unknown flags */
 733
 734	if (arg.nr < 0)
 735		return -EINVAL;
 736
 737	/* Ensure arg.off fits in an unsigned long */
 738	if (arg.off > ULONG_MAX)
 739		return 0;
 740
 741	if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
 742		pending = &child->signal->shared_pending;
 743	else
 744		pending = &child->pending;
 745
 746	for (i = 0; i < arg.nr; ) {
 747		kernel_siginfo_t info;
 748		unsigned long off = arg.off + i;
 749		bool found = false;
 750
 751		spin_lock_irq(&child->sighand->siglock);
 752		list_for_each_entry(q, &pending->list, list) {
 753			if (!off--) {
 754				found = true;
 755				copy_siginfo(&info, &q->info);
 756				break;
 757			}
 758		}
 759		spin_unlock_irq(&child->sighand->siglock);
 760
 761		if (!found) /* beyond the end of the list */
 762			break;
 763
 764#ifdef CONFIG_COMPAT
 765		if (unlikely(in_compat_syscall())) {
 766			compat_siginfo_t __user *uinfo = compat_ptr(data);
 767
 768			if (copy_siginfo_to_user32(uinfo, &info)) {
 769				ret = -EFAULT;
 770				break;
 771			}
 772
 773		} else
 774#endif
 775		{
 776			siginfo_t __user *uinfo = (siginfo_t __user *) data;
 777
 778			if (copy_siginfo_to_user(uinfo, &info)) {
 779				ret = -EFAULT;
 780				break;
 781			}
 782		}
 783
 784		data += sizeof(siginfo_t);
 785		i++;
 786
 787		if (signal_pending(current))
 788			break;
 789
 790		cond_resched();
 791	}
 792
 793	if (i > 0)
 794		return i;
 795
 796	return ret;
 797}
 798
 799#ifdef CONFIG_RSEQ
 800static long ptrace_get_rseq_configuration(struct task_struct *task,
 801					  unsigned long size, void __user *data)
 802{
 803	struct ptrace_rseq_configuration conf = {
 804		.rseq_abi_pointer = (u64)(uintptr_t)task->rseq,
 805		.rseq_abi_size = sizeof(*task->rseq),
 806		.signature = task->rseq_sig,
 807		.flags = 0,
 808	};
 809
 810	size = min_t(unsigned long, size, sizeof(conf));
 811	if (copy_to_user(data, &conf, size))
 812		return -EFAULT;
 813	return sizeof(conf);
 814}
 815#endif
 816
 817#ifdef PTRACE_SINGLESTEP
 818#define is_singlestep(request)		((request) == PTRACE_SINGLESTEP)
 819#else
 820#define is_singlestep(request)		0
 821#endif
 822
 823#ifdef PTRACE_SINGLEBLOCK
 824#define is_singleblock(request)		((request) == PTRACE_SINGLEBLOCK)
 825#else
 826#define is_singleblock(request)		0
 827#endif
 828
 829#ifdef PTRACE_SYSEMU
 830#define is_sysemu_singlestep(request)	((request) == PTRACE_SYSEMU_SINGLESTEP)
 831#else
 832#define is_sysemu_singlestep(request)	0
 833#endif
 834
 835static int ptrace_resume(struct task_struct *child, long request,
 836			 unsigned long data)
 837{
 838	bool need_siglock;
 839
 840	if (!valid_signal(data))
 841		return -EIO;
 842
 843	if (request == PTRACE_SYSCALL)
 844		set_task_syscall_work(child, SYSCALL_TRACE);
 845	else
 846		clear_task_syscall_work(child, SYSCALL_TRACE);
 847
 848#if defined(CONFIG_GENERIC_ENTRY) || defined(TIF_SYSCALL_EMU)
 849	if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
 850		set_task_syscall_work(child, SYSCALL_EMU);
 851	else
 852		clear_task_syscall_work(child, SYSCALL_EMU);
 853#endif
 854
 855	if (is_singleblock(request)) {
 856		if (unlikely(!arch_has_block_step()))
 857			return -EIO;
 858		user_enable_block_step(child);
 859	} else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
 860		if (unlikely(!arch_has_single_step()))
 861			return -EIO;
 862		user_enable_single_step(child);
 863	} else {
 864		user_disable_single_step(child);
 865	}
 866
 867	/*
 868	 * Change ->exit_code and ->state under siglock to avoid the race
 869	 * with wait_task_stopped() in between; a non-zero ->exit_code will
 870	 * wrongly look like another report from tracee.
 871	 *
 872	 * Note that we need siglock even if ->exit_code == data and/or this
 873	 * status was not reported yet, the new status must not be cleared by
 874	 * wait_task_stopped() after resume.
 875	 *
 876	 * If data == 0 we do not care if wait_task_stopped() reports the old
 877	 * status and clears the code too; this can't race with the tracee, it
 878	 * takes siglock after resume.
 879	 */
 880	need_siglock = data && !thread_group_empty(current);
 881	if (need_siglock)
 882		spin_lock_irq(&child->sighand->siglock);
 883	child->exit_code = data;
 
 884	wake_up_state(child, __TASK_TRACED);
 885	if (need_siglock)
 886		spin_unlock_irq(&child->sighand->siglock);
 887
 888	return 0;
 889}
 890
 891#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
 892
 893static const struct user_regset *
 894find_regset(const struct user_regset_view *view, unsigned int type)
 895{
 896	const struct user_regset *regset;
 897	int n;
 898
 899	for (n = 0; n < view->n; ++n) {
 900		regset = view->regsets + n;
 901		if (regset->core_note_type == type)
 902			return regset;
 903	}
 904
 905	return NULL;
 906}
 907
 908static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
 909			 struct iovec *kiov)
 910{
 911	const struct user_regset_view *view = task_user_regset_view(task);
 912	const struct user_regset *regset = find_regset(view, type);
 913	int regset_no;
 914
 915	if (!regset || (kiov->iov_len % regset->size) != 0)
 916		return -EINVAL;
 917
 918	regset_no = regset - view->regsets;
 919	kiov->iov_len = min(kiov->iov_len,
 920			    (__kernel_size_t) (regset->n * regset->size));
 921
 922	if (req == PTRACE_GETREGSET)
 923		return copy_regset_to_user(task, view, regset_no, 0,
 924					   kiov->iov_len, kiov->iov_base);
 925	else
 926		return copy_regset_from_user(task, view, regset_no, 0,
 927					     kiov->iov_len, kiov->iov_base);
 928}
 929
 930/*
 931 * This is declared in linux/regset.h and defined in machine-dependent
 932 * code.  We put the export here, near the primary machine-neutral use,
 933 * to ensure no machine forgets it.
 934 */
 935EXPORT_SYMBOL_GPL(task_user_regset_view);
 936
 937static unsigned long
 938ptrace_get_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
 939			      struct ptrace_syscall_info *info)
 940{
 941	unsigned long args[ARRAY_SIZE(info->entry.args)];
 942	int i;
 943
 944	info->op = PTRACE_SYSCALL_INFO_ENTRY;
 945	info->entry.nr = syscall_get_nr(child, regs);
 946	syscall_get_arguments(child, regs, args);
 947	for (i = 0; i < ARRAY_SIZE(args); i++)
 948		info->entry.args[i] = args[i];
 949
 950	/* args is the last field in struct ptrace_syscall_info.entry */
 951	return offsetofend(struct ptrace_syscall_info, entry.args);
 952}
 953
 954static unsigned long
 955ptrace_get_syscall_info_seccomp(struct task_struct *child, struct pt_regs *regs,
 956				struct ptrace_syscall_info *info)
 957{
 958	/*
 959	 * As struct ptrace_syscall_info.entry is currently a subset
 960	 * of struct ptrace_syscall_info.seccomp, it makes sense to
 961	 * initialize that subset using ptrace_get_syscall_info_entry().
 962	 * This can be reconsidered in the future if these structures
 963	 * diverge significantly enough.
 964	 */
 965	ptrace_get_syscall_info_entry(child, regs, info);
 966	info->op = PTRACE_SYSCALL_INFO_SECCOMP;
 967	info->seccomp.ret_data = child->ptrace_message;
 968
 969	/* ret_data is the last field in struct ptrace_syscall_info.seccomp */
 970	return offsetofend(struct ptrace_syscall_info, seccomp.ret_data);
 971}
 972
 973static unsigned long
 974ptrace_get_syscall_info_exit(struct task_struct *child, struct pt_regs *regs,
 975			     struct ptrace_syscall_info *info)
 976{
 977	info->op = PTRACE_SYSCALL_INFO_EXIT;
 978	info->exit.rval = syscall_get_error(child, regs);
 979	info->exit.is_error = !!info->exit.rval;
 980	if (!info->exit.is_error)
 981		info->exit.rval = syscall_get_return_value(child, regs);
 982
 983	/* is_error is the last field in struct ptrace_syscall_info.exit */
 984	return offsetofend(struct ptrace_syscall_info, exit.is_error);
 985}
 986
 987static int
 988ptrace_get_syscall_info(struct task_struct *child, unsigned long user_size,
 989			void __user *datavp)
 990{
 991	struct pt_regs *regs = task_pt_regs(child);
 992	struct ptrace_syscall_info info = {
 993		.op = PTRACE_SYSCALL_INFO_NONE,
 994		.arch = syscall_get_arch(child),
 995		.instruction_pointer = instruction_pointer(regs),
 996		.stack_pointer = user_stack_pointer(regs),
 997	};
 998	unsigned long actual_size = offsetof(struct ptrace_syscall_info, entry);
 999	unsigned long write_size;
1000
1001	/*
1002	 * This does not need lock_task_sighand() to access
1003	 * child->last_siginfo because ptrace_freeze_traced()
1004	 * called earlier by ptrace_check_attach() ensures that
1005	 * the tracee cannot go away and clear its last_siginfo.
1006	 */
1007	switch (child->last_siginfo ? child->last_siginfo->si_code : 0) {
1008	case SIGTRAP | 0x80:
1009		switch (child->ptrace_message) {
1010		case PTRACE_EVENTMSG_SYSCALL_ENTRY:
1011			actual_size = ptrace_get_syscall_info_entry(child, regs,
1012								    &info);
1013			break;
1014		case PTRACE_EVENTMSG_SYSCALL_EXIT:
1015			actual_size = ptrace_get_syscall_info_exit(child, regs,
1016								   &info);
1017			break;
1018		}
1019		break;
1020	case SIGTRAP | (PTRACE_EVENT_SECCOMP << 8):
1021		actual_size = ptrace_get_syscall_info_seccomp(child, regs,
1022							      &info);
1023		break;
1024	}
1025
1026	write_size = min(actual_size, user_size);
1027	return copy_to_user(datavp, &info, write_size) ? -EFAULT : actual_size;
1028}
1029#endif /* CONFIG_HAVE_ARCH_TRACEHOOK */
1030
1031int ptrace_request(struct task_struct *child, long request,
1032		   unsigned long addr, unsigned long data)
1033{
1034	bool seized = child->ptrace & PT_SEIZED;
1035	int ret = -EIO;
1036	kernel_siginfo_t siginfo, *si;
1037	void __user *datavp = (void __user *) data;
1038	unsigned long __user *datalp = datavp;
1039	unsigned long flags;
1040
1041	switch (request) {
1042	case PTRACE_PEEKTEXT:
1043	case PTRACE_PEEKDATA:
1044		return generic_ptrace_peekdata(child, addr, data);
1045	case PTRACE_POKETEXT:
1046	case PTRACE_POKEDATA:
1047		return generic_ptrace_pokedata(child, addr, data);
1048
1049#ifdef PTRACE_OLDSETOPTIONS
1050	case PTRACE_OLDSETOPTIONS:
1051#endif
1052	case PTRACE_SETOPTIONS:
1053		ret = ptrace_setoptions(child, data);
1054		break;
1055	case PTRACE_GETEVENTMSG:
1056		ret = put_user(child->ptrace_message, datalp);
1057		break;
1058
1059	case PTRACE_PEEKSIGINFO:
1060		ret = ptrace_peek_siginfo(child, addr, data);
1061		break;
1062
1063	case PTRACE_GETSIGINFO:
1064		ret = ptrace_getsiginfo(child, &siginfo);
1065		if (!ret)
1066			ret = copy_siginfo_to_user(datavp, &siginfo);
1067		break;
1068
1069	case PTRACE_SETSIGINFO:
1070		ret = copy_siginfo_from_user(&siginfo, datavp);
1071		if (!ret)
1072			ret = ptrace_setsiginfo(child, &siginfo);
1073		break;
1074
1075	case PTRACE_GETSIGMASK: {
1076		sigset_t *mask;
1077
1078		if (addr != sizeof(sigset_t)) {
1079			ret = -EINVAL;
1080			break;
1081		}
1082
1083		if (test_tsk_restore_sigmask(child))
1084			mask = &child->saved_sigmask;
1085		else
1086			mask = &child->blocked;
1087
1088		if (copy_to_user(datavp, mask, sizeof(sigset_t)))
1089			ret = -EFAULT;
1090		else
1091			ret = 0;
1092
1093		break;
1094	}
1095
1096	case PTRACE_SETSIGMASK: {
1097		sigset_t new_set;
1098
1099		if (addr != sizeof(sigset_t)) {
1100			ret = -EINVAL;
1101			break;
1102		}
1103
1104		if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
1105			ret = -EFAULT;
1106			break;
1107		}
1108
1109		sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
1110
1111		/*
1112		 * Every thread does recalc_sigpending() after resume, so
1113		 * retarget_shared_pending() and recalc_sigpending() are not
1114		 * called here.
1115		 */
1116		spin_lock_irq(&child->sighand->siglock);
1117		child->blocked = new_set;
1118		spin_unlock_irq(&child->sighand->siglock);
1119
1120		clear_tsk_restore_sigmask(child);
1121
1122		ret = 0;
1123		break;
1124	}
1125
1126	case PTRACE_INTERRUPT:
1127		/*
1128		 * Stop tracee without any side-effect on signal or job
1129		 * control.  At least one trap is guaranteed to happen
1130		 * after this request.  If @child is already trapped, the
1131		 * current trap is not disturbed and another trap will
1132		 * happen after the current trap is ended with PTRACE_CONT.
1133		 *
1134		 * The actual trap might not be PTRACE_EVENT_STOP trap but
1135		 * the pending condition is cleared regardless.
1136		 */
1137		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1138			break;
1139
1140		/*
1141		 * INTERRUPT doesn't disturb existing trap sans one
1142		 * exception.  If ptracer issued LISTEN for the current
1143		 * STOP, this INTERRUPT should clear LISTEN and re-trap
1144		 * tracee into STOP.
1145		 */
1146		if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
1147			ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
1148
1149		unlock_task_sighand(child, &flags);
1150		ret = 0;
1151		break;
1152
1153	case PTRACE_LISTEN:
1154		/*
1155		 * Listen for events.  Tracee must be in STOP.  It's not
1156		 * resumed per-se but is not considered to be in TRACED by
1157		 * wait(2) or ptrace(2).  If an async event (e.g. group
1158		 * stop state change) happens, tracee will enter STOP trap
1159		 * again.  Alternatively, ptracer can issue INTERRUPT to
1160		 * finish listening and re-trap tracee into STOP.
1161		 */
1162		if (unlikely(!seized || !lock_task_sighand(child, &flags)))
1163			break;
1164
1165		si = child->last_siginfo;
1166		if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
1167			child->jobctl |= JOBCTL_LISTENING;
1168			/*
1169			 * If NOTIFY is set, it means event happened between
1170			 * start of this trap and now.  Trigger re-trap.
1171			 */
1172			if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1173				ptrace_signal_wake_up(child, true);
1174			ret = 0;
1175		}
1176		unlock_task_sighand(child, &flags);
1177		break;
1178
1179	case PTRACE_DETACH:	 /* detach a process that was attached. */
1180		ret = ptrace_detach(child, data);
1181		break;
1182
1183#ifdef CONFIG_BINFMT_ELF_FDPIC
1184	case PTRACE_GETFDPIC: {
1185		struct mm_struct *mm = get_task_mm(child);
1186		unsigned long tmp = 0;
1187
1188		ret = -ESRCH;
1189		if (!mm)
1190			break;
1191
1192		switch (addr) {
1193		case PTRACE_GETFDPIC_EXEC:
1194			tmp = mm->context.exec_fdpic_loadmap;
1195			break;
1196		case PTRACE_GETFDPIC_INTERP:
1197			tmp = mm->context.interp_fdpic_loadmap;
1198			break;
1199		default:
1200			break;
1201		}
1202		mmput(mm);
1203
1204		ret = put_user(tmp, datalp);
1205		break;
1206	}
1207#endif
1208
1209#ifdef PTRACE_SINGLESTEP
1210	case PTRACE_SINGLESTEP:
1211#endif
1212#ifdef PTRACE_SINGLEBLOCK
1213	case PTRACE_SINGLEBLOCK:
1214#endif
1215#ifdef PTRACE_SYSEMU
1216	case PTRACE_SYSEMU:
1217	case PTRACE_SYSEMU_SINGLESTEP:
1218#endif
1219	case PTRACE_SYSCALL:
1220	case PTRACE_CONT:
1221		return ptrace_resume(child, request, data);
1222
1223	case PTRACE_KILL:
1224		if (child->exit_state)	/* already dead */
1225			return 0;
1226		return ptrace_resume(child, request, SIGKILL);
1227
1228#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1229	case PTRACE_GETREGSET:
1230	case PTRACE_SETREGSET: {
1231		struct iovec kiov;
1232		struct iovec __user *uiov = datavp;
1233
1234		if (!access_ok(uiov, sizeof(*uiov)))
1235			return -EFAULT;
1236
1237		if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1238		    __get_user(kiov.iov_len, &uiov->iov_len))
1239			return -EFAULT;
1240
1241		ret = ptrace_regset(child, request, addr, &kiov);
1242		if (!ret)
1243			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1244		break;
1245	}
1246
1247	case PTRACE_GET_SYSCALL_INFO:
1248		ret = ptrace_get_syscall_info(child, addr, datavp);
1249		break;
1250#endif
1251
1252	case PTRACE_SECCOMP_GET_FILTER:
1253		ret = seccomp_get_filter(child, addr, datavp);
1254		break;
1255
1256	case PTRACE_SECCOMP_GET_METADATA:
1257		ret = seccomp_get_metadata(child, addr, datavp);
1258		break;
1259
1260#ifdef CONFIG_RSEQ
1261	case PTRACE_GET_RSEQ_CONFIGURATION:
1262		ret = ptrace_get_rseq_configuration(child, addr, datavp);
1263		break;
1264#endif
1265
1266	default:
1267		break;
1268	}
1269
1270	return ret;
1271}
1272
1273#ifndef arch_ptrace_attach
1274#define arch_ptrace_attach(child)	do { } while (0)
1275#endif
1276
1277SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1278		unsigned long, data)
1279{
1280	struct task_struct *child;
1281	long ret;
1282
1283	if (request == PTRACE_TRACEME) {
1284		ret = ptrace_traceme();
1285		if (!ret)
1286			arch_ptrace_attach(current);
1287		goto out;
1288	}
1289
1290	child = find_get_task_by_vpid(pid);
1291	if (!child) {
1292		ret = -ESRCH;
1293		goto out;
1294	}
1295
1296	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1297		ret = ptrace_attach(child, request, addr, data);
1298		/*
1299		 * Some architectures need to do book-keeping after
1300		 * a ptrace attach.
1301		 */
1302		if (!ret)
1303			arch_ptrace_attach(child);
1304		goto out_put_task_struct;
1305	}
1306
1307	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1308				  request == PTRACE_INTERRUPT);
1309	if (ret < 0)
1310		goto out_put_task_struct;
1311
1312	ret = arch_ptrace(child, request, addr, data);
1313	if (ret || request != PTRACE_DETACH)
1314		ptrace_unfreeze_traced(child);
1315
1316 out_put_task_struct:
1317	put_task_struct(child);
1318 out:
1319	return ret;
1320}
1321
1322int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1323			    unsigned long data)
1324{
1325	unsigned long tmp;
1326	int copied;
1327
1328	copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1329	if (copied != sizeof(tmp))
1330		return -EIO;
1331	return put_user(tmp, (unsigned long __user *)data);
1332}
1333
1334int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1335			    unsigned long data)
1336{
1337	int copied;
1338
1339	copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1340			FOLL_FORCE | FOLL_WRITE);
1341	return (copied == sizeof(data)) ? 0 : -EIO;
1342}
1343
1344#if defined CONFIG_COMPAT
1345
1346int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1347			  compat_ulong_t addr, compat_ulong_t data)
1348{
1349	compat_ulong_t __user *datap = compat_ptr(data);
1350	compat_ulong_t word;
1351	kernel_siginfo_t siginfo;
1352	int ret;
1353
1354	switch (request) {
1355	case PTRACE_PEEKTEXT:
1356	case PTRACE_PEEKDATA:
1357		ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1358				FOLL_FORCE);
1359		if (ret != sizeof(word))
1360			ret = -EIO;
1361		else
1362			ret = put_user(word, datap);
1363		break;
1364
1365	case PTRACE_POKETEXT:
1366	case PTRACE_POKEDATA:
1367		ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1368				FOLL_FORCE | FOLL_WRITE);
1369		ret = (ret != sizeof(data) ? -EIO : 0);
1370		break;
1371
1372	case PTRACE_GETEVENTMSG:
1373		ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1374		break;
1375
1376	case PTRACE_GETSIGINFO:
1377		ret = ptrace_getsiginfo(child, &siginfo);
1378		if (!ret)
1379			ret = copy_siginfo_to_user32(
1380				(struct compat_siginfo __user *) datap,
1381				&siginfo);
1382		break;
1383
1384	case PTRACE_SETSIGINFO:
1385		ret = copy_siginfo_from_user32(
1386			&siginfo, (struct compat_siginfo __user *) datap);
1387		if (!ret)
1388			ret = ptrace_setsiginfo(child, &siginfo);
1389		break;
1390#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1391	case PTRACE_GETREGSET:
1392	case PTRACE_SETREGSET:
1393	{
1394		struct iovec kiov;
1395		struct compat_iovec __user *uiov =
1396			(struct compat_iovec __user *) datap;
1397		compat_uptr_t ptr;
1398		compat_size_t len;
1399
1400		if (!access_ok(uiov, sizeof(*uiov)))
1401			return -EFAULT;
1402
1403		if (__get_user(ptr, &uiov->iov_base) ||
1404		    __get_user(len, &uiov->iov_len))
1405			return -EFAULT;
1406
1407		kiov.iov_base = compat_ptr(ptr);
1408		kiov.iov_len = len;
1409
1410		ret = ptrace_regset(child, request, addr, &kiov);
1411		if (!ret)
1412			ret = __put_user(kiov.iov_len, &uiov->iov_len);
1413		break;
1414	}
1415#endif
1416
1417	default:
1418		ret = ptrace_request(child, request, addr, data);
1419	}
1420
1421	return ret;
1422}
1423
1424COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1425		       compat_long_t, addr, compat_long_t, data)
1426{
1427	struct task_struct *child;
1428	long ret;
1429
1430	if (request == PTRACE_TRACEME) {
1431		ret = ptrace_traceme();
1432		goto out;
1433	}
1434
1435	child = find_get_task_by_vpid(pid);
1436	if (!child) {
1437		ret = -ESRCH;
1438		goto out;
1439	}
1440
1441	if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1442		ret = ptrace_attach(child, request, addr, data);
1443		/*
1444		 * Some architectures need to do book-keeping after
1445		 * a ptrace attach.
1446		 */
1447		if (!ret)
1448			arch_ptrace_attach(child);
1449		goto out_put_task_struct;
1450	}
1451
1452	ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1453				  request == PTRACE_INTERRUPT);
1454	if (!ret) {
1455		ret = compat_arch_ptrace(child, request, addr, data);
1456		if (ret || request != PTRACE_DETACH)
1457			ptrace_unfreeze_traced(child);
1458	}
1459
1460 out_put_task_struct:
1461	put_task_struct(child);
1462 out:
1463	return ret;
1464}
1465#endif	/* CONFIG_COMPAT */