Back to home page

LXR

 
 

    


0001 /*
0002  * linux/kernel/ptrace.c
0003  *
0004  * (C) Copyright 1999 Linus Torvalds
0005  *
0006  * Common interfaces for "ptrace()" which we do not want
0007  * to continually duplicate across every architecture.
0008  */
0009 
0010 #include <linux/capability.h>
0011 #include <linux/export.h>
0012 #include <linux/sched.h>
0013 #include <linux/errno.h>
0014 #include <linux/mm.h>
0015 #include <linux/highmem.h>
0016 #include <linux/pagemap.h>
0017 #include <linux/ptrace.h>
0018 #include <linux/security.h>
0019 #include <linux/signal.h>
0020 #include <linux/uio.h>
0021 #include <linux/audit.h>
0022 #include <linux/pid_namespace.h>
0023 #include <linux/syscalls.h>
0024 #include <linux/uaccess.h>
0025 #include <linux/regset.h>
0026 #include <linux/hw_breakpoint.h>
0027 #include <linux/cn_proc.h>
0028 #include <linux/compat.h>
0029 
0030 /*
0031  * Access another process' address space via ptrace.
0032  * Source/target buffer must be kernel space,
0033  * Do not walk the page table directly, use get_user_pages
0034  */
0035 int ptrace_access_vm(struct task_struct *tsk, unsigned long addr,
0036              void *buf, int len, unsigned int gup_flags)
0037 {
0038     struct mm_struct *mm;
0039     int ret;
0040 
0041     mm = get_task_mm(tsk);
0042     if (!mm)
0043         return 0;
0044 
0045     if (!tsk->ptrace ||
0046         (current != tsk->parent) ||
0047         ((get_dumpable(mm) != SUID_DUMP_USER) &&
0048          !ptracer_capable(tsk, mm->user_ns))) {
0049         mmput(mm);
0050         return 0;
0051     }
0052 
0053     ret = __access_remote_vm(tsk, mm, addr, buf, len, gup_flags);
0054     mmput(mm);
0055 
0056     return ret;
0057 }
0058 
0059 
0060 /*
0061  * ptrace a task: make the debugger its new parent and
0062  * move it to the ptrace list.
0063  *
0064  * Must be called with the tasklist lock write-held.
0065  */
0066 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
0067 {
0068     BUG_ON(!list_empty(&child->ptrace_entry));
0069     list_add(&child->ptrace_entry, &new_parent->ptraced);
0070     child->parent = new_parent;
0071     rcu_read_lock();
0072     child->ptracer_cred = get_cred(__task_cred(new_parent));
0073     rcu_read_unlock();
0074 }
0075 
0076 /**
0077  * __ptrace_unlink - unlink ptracee and restore its execution state
0078  * @child: ptracee to be unlinked
0079  *
0080  * Remove @child from the ptrace list, move it back to the original parent,
0081  * and restore the execution state so that it conforms to the group stop
0082  * state.
0083  *
0084  * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
0085  * exiting.  For PTRACE_DETACH, unless the ptracee has been killed between
0086  * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
0087  * If the ptracer is exiting, the ptracee can be in any state.
0088  *
0089  * After detach, the ptracee should be in a state which conforms to the
0090  * group stop.  If the group is stopped or in the process of stopping, the
0091  * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
0092  * up from TASK_TRACED.
0093  *
0094  * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
0095  * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
0096  * to but in the opposite direction of what happens while attaching to a
0097  * stopped task.  However, in this direction, the intermediate RUNNING
0098  * state is not hidden even from the current ptracer and if it immediately
0099  * re-attaches and performs a WNOHANG wait(2), it may fail.
0100  *
0101  * CONTEXT:
0102  * write_lock_irq(tasklist_lock)
0103  */
0104 void __ptrace_unlink(struct task_struct *child)
0105 {
0106     const struct cred *old_cred;
0107     BUG_ON(!child->ptrace);
0108 
0109     clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
0110 
0111     child->parent = child->real_parent;
0112     list_del_init(&child->ptrace_entry);
0113     old_cred = child->ptracer_cred;
0114     child->ptracer_cred = NULL;
0115     put_cred(old_cred);
0116 
0117     spin_lock(&child->sighand->siglock);
0118     child->ptrace = 0;
0119     /*
0120      * Clear all pending traps and TRAPPING.  TRAPPING should be
0121      * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
0122      */
0123     task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
0124     task_clear_jobctl_trapping(child);
0125 
0126     /*
0127      * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
0128      * @child isn't dead.
0129      */
0130     if (!(child->flags & PF_EXITING) &&
0131         (child->signal->flags & SIGNAL_STOP_STOPPED ||
0132          child->signal->group_stop_count)) {
0133         child->jobctl |= JOBCTL_STOP_PENDING;
0134 
0135         /*
0136          * This is only possible if this thread was cloned by the
0137          * traced task running in the stopped group, set the signal
0138          * for the future reports.
0139          * FIXME: we should change ptrace_init_task() to handle this
0140          * case.
0141          */
0142         if (!(child->jobctl & JOBCTL_STOP_SIGMASK))
0143             child->jobctl |= SIGSTOP;
0144     }
0145 
0146     /*
0147      * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
0148      * @child in the butt.  Note that @resume should be used iff @child
0149      * is in TASK_TRACED; otherwise, we might unduly disrupt
0150      * TASK_KILLABLE sleeps.
0151      */
0152     if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
0153         ptrace_signal_wake_up(child, true);
0154 
0155     spin_unlock(&child->sighand->siglock);
0156 }
0157 
0158 /* Ensure that nothing can wake it up, even SIGKILL */
0159 static bool ptrace_freeze_traced(struct task_struct *task)
0160 {
0161     bool ret = false;
0162 
0163     /* Lockless, nobody but us can set this flag */
0164     if (task->jobctl & JOBCTL_LISTENING)
0165         return ret;
0166 
0167     spin_lock_irq(&task->sighand->siglock);
0168     if (task_is_traced(task) && !__fatal_signal_pending(task)) {
0169         task->state = __TASK_TRACED;
0170         ret = true;
0171     }
0172     spin_unlock_irq(&task->sighand->siglock);
0173 
0174     return ret;
0175 }
0176 
0177 static void ptrace_unfreeze_traced(struct task_struct *task)
0178 {
0179     if (task->state != __TASK_TRACED)
0180         return;
0181 
0182     WARN_ON(!task->ptrace || task->parent != current);
0183 
0184     spin_lock_irq(&task->sighand->siglock);
0185     if (__fatal_signal_pending(task))
0186         wake_up_state(task, __TASK_TRACED);
0187     else
0188         task->state = TASK_TRACED;
0189     spin_unlock_irq(&task->sighand->siglock);
0190 }
0191 
0192 /**
0193  * ptrace_check_attach - check whether ptracee is ready for ptrace operation
0194  * @child: ptracee to check for
0195  * @ignore_state: don't check whether @child is currently %TASK_TRACED
0196  *
0197  * Check whether @child is being ptraced by %current and ready for further
0198  * ptrace operations.  If @ignore_state is %false, @child also should be in
0199  * %TASK_TRACED state and on return the child is guaranteed to be traced
0200  * and not executing.  If @ignore_state is %true, @child can be in any
0201  * state.
0202  *
0203  * CONTEXT:
0204  * Grabs and releases tasklist_lock and @child->sighand->siglock.
0205  *
0206  * RETURNS:
0207  * 0 on success, -ESRCH if %child is not ready.
0208  */
0209 static int ptrace_check_attach(struct task_struct *child, bool ignore_state)
0210 {
0211     int ret = -ESRCH;
0212 
0213     /*
0214      * We take the read lock around doing both checks to close a
0215      * possible race where someone else was tracing our child and
0216      * detached between these two checks.  After this locked check,
0217      * we are sure that this is our traced child and that can only
0218      * be changed by us so it's not changing right after this.
0219      */
0220     read_lock(&tasklist_lock);
0221     if (child->ptrace && child->parent == current) {
0222         WARN_ON(child->state == __TASK_TRACED);
0223         /*
0224          * child->sighand can't be NULL, release_task()
0225          * does ptrace_unlink() before __exit_signal().
0226          */
0227         if (ignore_state || ptrace_freeze_traced(child))
0228             ret = 0;
0229     }
0230     read_unlock(&tasklist_lock);
0231 
0232     if (!ret && !ignore_state) {
0233         if (!wait_task_inactive(child, __TASK_TRACED)) {
0234             /*
0235              * This can only happen if may_ptrace_stop() fails and
0236              * ptrace_stop() changes ->state back to TASK_RUNNING,
0237              * so we should not worry about leaking __TASK_TRACED.
0238              */
0239             WARN_ON(child->state == __TASK_TRACED);
0240             ret = -ESRCH;
0241         }
0242     }
0243 
0244     return ret;
0245 }
0246 
0247 static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
0248 {
0249     if (mode & PTRACE_MODE_NOAUDIT)
0250         return has_ns_capability_noaudit(current, ns, CAP_SYS_PTRACE);
0251     else
0252         return has_ns_capability(current, ns, CAP_SYS_PTRACE);
0253 }
0254 
0255 /* Returns 0 on success, -errno on denial. */
0256 static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
0257 {
0258     const struct cred *cred = current_cred(), *tcred;
0259     struct mm_struct *mm;
0260     kuid_t caller_uid;
0261     kgid_t caller_gid;
0262 
0263     if (!(mode & PTRACE_MODE_FSCREDS) == !(mode & PTRACE_MODE_REALCREDS)) {
0264         WARN(1, "denying ptrace access check without PTRACE_MODE_*CREDS\n");
0265         return -EPERM;
0266     }
0267 
0268     /* May we inspect the given task?
0269      * This check is used both for attaching with ptrace
0270      * and for allowing access to sensitive information in /proc.
0271      *
0272      * ptrace_attach denies several cases that /proc allows
0273      * because setting up the necessary parent/child relationship
0274      * or halting the specified task is impossible.
0275      */
0276 
0277     /* Don't let security modules deny introspection */
0278     if (same_thread_group(task, current))
0279         return 0;
0280     rcu_read_lock();
0281     if (mode & PTRACE_MODE_FSCREDS) {
0282         caller_uid = cred->fsuid;
0283         caller_gid = cred->fsgid;
0284     } else {
0285         /*
0286          * Using the euid would make more sense here, but something
0287          * in userland might rely on the old behavior, and this
0288          * shouldn't be a security problem since
0289          * PTRACE_MODE_REALCREDS implies that the caller explicitly
0290          * used a syscall that requests access to another process
0291          * (and not a filesystem syscall to procfs).
0292          */
0293         caller_uid = cred->uid;
0294         caller_gid = cred->gid;
0295     }
0296     tcred = __task_cred(task);
0297     if (uid_eq(caller_uid, tcred->euid) &&
0298         uid_eq(caller_uid, tcred->suid) &&
0299         uid_eq(caller_uid, tcred->uid)  &&
0300         gid_eq(caller_gid, tcred->egid) &&
0301         gid_eq(caller_gid, tcred->sgid) &&
0302         gid_eq(caller_gid, tcred->gid))
0303         goto ok;
0304     if (ptrace_has_cap(tcred->user_ns, mode))
0305         goto ok;
0306     rcu_read_unlock();
0307     return -EPERM;
0308 ok:
0309     rcu_read_unlock();
0310     mm = task->mm;
0311     if (mm &&
0312         ((get_dumpable(mm) != SUID_DUMP_USER) &&
0313          !ptrace_has_cap(mm->user_ns, mode)))
0314         return -EPERM;
0315 
0316     return security_ptrace_access_check(task, mode);
0317 }
0318 
0319 bool ptrace_may_access(struct task_struct *task, unsigned int mode)
0320 {
0321     int err;
0322     task_lock(task);
0323     err = __ptrace_may_access(task, mode);
0324     task_unlock(task);
0325     return !err;
0326 }
0327 
0328 static int ptrace_attach(struct task_struct *task, long request,
0329              unsigned long addr,
0330              unsigned long flags)
0331 {
0332     bool seize = (request == PTRACE_SEIZE);
0333     int retval;
0334 
0335     retval = -EIO;
0336     if (seize) {
0337         if (addr != 0)
0338             goto out;
0339         if (flags & ~(unsigned long)PTRACE_O_MASK)
0340             goto out;
0341         flags = PT_PTRACED | PT_SEIZED | (flags << PT_OPT_FLAG_SHIFT);
0342     } else {
0343         flags = PT_PTRACED;
0344     }
0345 
0346     audit_ptrace(task);
0347 
0348     retval = -EPERM;
0349     if (unlikely(task->flags & PF_KTHREAD))
0350         goto out;
0351     if (same_thread_group(task, current))
0352         goto out;
0353 
0354     /*
0355      * Protect exec's credential calculations against our interference;
0356      * SUID, SGID and LSM creds get determined differently
0357      * under ptrace.
0358      */
0359     retval = -ERESTARTNOINTR;
0360     if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
0361         goto out;
0362 
0363     task_lock(task);
0364     retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH_REALCREDS);
0365     task_unlock(task);
0366     if (retval)
0367         goto unlock_creds;
0368 
0369     write_lock_irq(&tasklist_lock);
0370     retval = -EPERM;
0371     if (unlikely(task->exit_state))
0372         goto unlock_tasklist;
0373     if (task->ptrace)
0374         goto unlock_tasklist;
0375 
0376     if (seize)
0377         flags |= PT_SEIZED;
0378     task->ptrace = flags;
0379 
0380     __ptrace_link(task, current);
0381 
0382     /* SEIZE doesn't trap tracee on attach */
0383     if (!seize)
0384         send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
0385 
0386     spin_lock(&task->sighand->siglock);
0387 
0388     /*
0389      * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
0390      * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
0391      * will be cleared if the child completes the transition or any
0392      * event which clears the group stop states happens.  We'll wait
0393      * for the transition to complete before returning from this
0394      * function.
0395      *
0396      * This hides STOPPED -> RUNNING -> TRACED transition from the
0397      * attaching thread but a different thread in the same group can
0398      * still observe the transient RUNNING state.  IOW, if another
0399      * thread's WNOHANG wait(2) on the stopped tracee races against
0400      * ATTACH, the wait(2) may fail due to the transient RUNNING.
0401      *
0402      * The following task_is_stopped() test is safe as both transitions
0403      * in and out of STOPPED are protected by siglock.
0404      */
0405     if (task_is_stopped(task) &&
0406         task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
0407         signal_wake_up_state(task, __TASK_STOPPED);
0408 
0409     spin_unlock(&task->sighand->siglock);
0410 
0411     retval = 0;
0412 unlock_tasklist:
0413     write_unlock_irq(&tasklist_lock);
0414 unlock_creds:
0415     mutex_unlock(&task->signal->cred_guard_mutex);
0416 out:
0417     if (!retval) {
0418         /*
0419          * We do not bother to change retval or clear JOBCTL_TRAPPING
0420          * if wait_on_bit() was interrupted by SIGKILL. The tracer will
0421          * not return to user-mode, it will exit and clear this bit in
0422          * __ptrace_unlink() if it wasn't already cleared by the tracee;
0423          * and until then nobody can ptrace this task.
0424          */
0425         wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT, TASK_KILLABLE);
0426         proc_ptrace_connector(task, PTRACE_ATTACH);
0427     }
0428 
0429     return retval;
0430 }
0431 
0432 /**
0433  * ptrace_traceme  --  helper for PTRACE_TRACEME
0434  *
0435  * Performs checks and sets PT_PTRACED.
0436  * Should be used by all ptrace implementations for PTRACE_TRACEME.
0437  */
0438 static int ptrace_traceme(void)
0439 {
0440     int ret = -EPERM;
0441 
0442     write_lock_irq(&tasklist_lock);
0443     /* Are we already being traced? */
0444     if (!current->ptrace) {
0445         ret = security_ptrace_traceme(current->parent);
0446         /*
0447          * Check PF_EXITING to ensure ->real_parent has not passed
0448          * exit_ptrace(). Otherwise we don't report the error but
0449          * pretend ->real_parent untraces us right after return.
0450          */
0451         if (!ret && !(current->real_parent->flags & PF_EXITING)) {
0452             current->ptrace = PT_PTRACED;
0453             __ptrace_link(current, current->real_parent);
0454         }
0455     }
0456     write_unlock_irq(&tasklist_lock);
0457 
0458     return ret;
0459 }
0460 
0461 /*
0462  * Called with irqs disabled, returns true if childs should reap themselves.
0463  */
0464 static int ignoring_children(struct sighand_struct *sigh)
0465 {
0466     int ret;
0467     spin_lock(&sigh->siglock);
0468     ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
0469           (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
0470     spin_unlock(&sigh->siglock);
0471     return ret;
0472 }
0473 
0474 /*
0475  * Called with tasklist_lock held for writing.
0476  * Unlink a traced task, and clean it up if it was a traced zombie.
0477  * Return true if it needs to be reaped with release_task().
0478  * (We can't call release_task() here because we already hold tasklist_lock.)
0479  *
0480  * If it's a zombie, our attachedness prevented normal parent notification
0481  * or self-reaping.  Do notification now if it would have happened earlier.
0482  * If it should reap itself, return true.
0483  *
0484  * If it's our own child, there is no notification to do. But if our normal
0485  * children self-reap, then this child was prevented by ptrace and we must
0486  * reap it now, in that case we must also wake up sub-threads sleeping in
0487  * do_wait().
0488  */
0489 static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
0490 {
0491     bool dead;
0492 
0493     __ptrace_unlink(p);
0494 
0495     if (p->exit_state != EXIT_ZOMBIE)
0496         return false;
0497 
0498     dead = !thread_group_leader(p);
0499 
0500     if (!dead && thread_group_empty(p)) {
0501         if (!same_thread_group(p->real_parent, tracer))
0502             dead = do_notify_parent(p, p->exit_signal);
0503         else if (ignoring_children(tracer->sighand)) {
0504             __wake_up_parent(p, tracer);
0505             dead = true;
0506         }
0507     }
0508     /* Mark it as in the process of being reaped. */
0509     if (dead)
0510         p->exit_state = EXIT_DEAD;
0511     return dead;
0512 }
0513 
0514 static int ptrace_detach(struct task_struct *child, unsigned int data)
0515 {
0516     if (!valid_signal(data))
0517         return -EIO;
0518 
0519     /* Architecture-specific hardware disable .. */
0520     ptrace_disable(child);
0521 
0522     write_lock_irq(&tasklist_lock);
0523     /*
0524      * We rely on ptrace_freeze_traced(). It can't be killed and
0525      * untraced by another thread, it can't be a zombie.
0526      */
0527     WARN_ON(!child->ptrace || child->exit_state);
0528     /*
0529      * tasklist_lock avoids the race with wait_task_stopped(), see
0530      * the comment in ptrace_resume().
0531      */
0532     child->exit_code = data;
0533     __ptrace_detach(current, child);
0534     write_unlock_irq(&tasklist_lock);
0535 
0536     proc_ptrace_connector(child, PTRACE_DETACH);
0537 
0538     return 0;
0539 }
0540 
0541 /*
0542  * Detach all tasks we were using ptrace on. Called with tasklist held
0543  * for writing.
0544  */
0545 void exit_ptrace(struct task_struct *tracer, struct list_head *dead)
0546 {
0547     struct task_struct *p, *n;
0548 
0549     list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
0550         if (unlikely(p->ptrace & PT_EXITKILL))
0551             send_sig_info(SIGKILL, SEND_SIG_FORCED, p);
0552 
0553         if (__ptrace_detach(tracer, p))
0554             list_add(&p->ptrace_entry, dead);
0555     }
0556 }
0557 
0558 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
0559 {
0560     int copied = 0;
0561 
0562     while (len > 0) {
0563         char buf[128];
0564         int this_len, retval;
0565 
0566         this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
0567         retval = ptrace_access_vm(tsk, src, buf, this_len, FOLL_FORCE);
0568 
0569         if (!retval) {
0570             if (copied)
0571                 break;
0572             return -EIO;
0573         }
0574         if (copy_to_user(dst, buf, retval))
0575             return -EFAULT;
0576         copied += retval;
0577         src += retval;
0578         dst += retval;
0579         len -= retval;
0580     }
0581     return copied;
0582 }
0583 
0584 int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long dst, int len)
0585 {
0586     int copied = 0;
0587 
0588     while (len > 0) {
0589         char buf[128];
0590         int this_len, retval;
0591 
0592         this_len = (len > sizeof(buf)) ? sizeof(buf) : len;
0593         if (copy_from_user(buf, src, this_len))
0594             return -EFAULT;
0595         retval = ptrace_access_vm(tsk, dst, buf, this_len,
0596                 FOLL_FORCE | FOLL_WRITE);
0597         if (!retval) {
0598             if (copied)
0599                 break;
0600             return -EIO;
0601         }
0602         copied += retval;
0603         src += retval;
0604         dst += retval;
0605         len -= retval;
0606     }
0607     return copied;
0608 }
0609 
0610 static int ptrace_setoptions(struct task_struct *child, unsigned long data)
0611 {
0612     unsigned flags;
0613 
0614     if (data & ~(unsigned long)PTRACE_O_MASK)
0615         return -EINVAL;
0616 
0617     if (unlikely(data & PTRACE_O_SUSPEND_SECCOMP)) {
0618         if (!IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) ||
0619             !IS_ENABLED(CONFIG_SECCOMP))
0620             return -EINVAL;
0621 
0622         if (!capable(CAP_SYS_ADMIN))
0623             return -EPERM;
0624 
0625         if (seccomp_mode(&current->seccomp) != SECCOMP_MODE_DISABLED ||
0626             current->ptrace & PT_SUSPEND_SECCOMP)
0627             return -EPERM;
0628     }
0629 
0630     /* Avoid intermediate state when all opts are cleared */
0631     flags = child->ptrace;
0632     flags &= ~(PTRACE_O_MASK << PT_OPT_FLAG_SHIFT);
0633     flags |= (data << PT_OPT_FLAG_SHIFT);
0634     child->ptrace = flags;
0635 
0636     return 0;
0637 }
0638 
0639 static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
0640 {
0641     unsigned long flags;
0642     int error = -ESRCH;
0643 
0644     if (lock_task_sighand(child, &flags)) {
0645         error = -EINVAL;
0646         if (likely(child->last_siginfo != NULL)) {
0647             *info = *child->last_siginfo;
0648             error = 0;
0649         }
0650         unlock_task_sighand(child, &flags);
0651     }
0652     return error;
0653 }
0654 
0655 static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
0656 {
0657     unsigned long flags;
0658     int error = -ESRCH;
0659 
0660     if (lock_task_sighand(child, &flags)) {
0661         error = -EINVAL;
0662         if (likely(child->last_siginfo != NULL)) {
0663             *child->last_siginfo = *info;
0664             error = 0;
0665         }
0666         unlock_task_sighand(child, &flags);
0667     }
0668     return error;
0669 }
0670 
0671 static int ptrace_peek_siginfo(struct task_struct *child,
0672                 unsigned long addr,
0673                 unsigned long data)
0674 {
0675     struct ptrace_peeksiginfo_args arg;
0676     struct sigpending *pending;
0677     struct sigqueue *q;
0678     int ret, i;
0679 
0680     ret = copy_from_user(&arg, (void __user *) addr,
0681                 sizeof(struct ptrace_peeksiginfo_args));
0682     if (ret)
0683         return -EFAULT;
0684 
0685     if (arg.flags & ~PTRACE_PEEKSIGINFO_SHARED)
0686         return -EINVAL; /* unknown flags */
0687 
0688     if (arg.nr < 0)
0689         return -EINVAL;
0690 
0691     if (arg.flags & PTRACE_PEEKSIGINFO_SHARED)
0692         pending = &child->signal->shared_pending;
0693     else
0694         pending = &child->pending;
0695 
0696     for (i = 0; i < arg.nr; ) {
0697         siginfo_t info;
0698         s32 off = arg.off + i;
0699 
0700         spin_lock_irq(&child->sighand->siglock);
0701         list_for_each_entry(q, &pending->list, list) {
0702             if (!off--) {
0703                 copy_siginfo(&info, &q->info);
0704                 break;
0705             }
0706         }
0707         spin_unlock_irq(&child->sighand->siglock);
0708 
0709         if (off >= 0) /* beyond the end of the list */
0710             break;
0711 
0712 #ifdef CONFIG_COMPAT
0713         if (unlikely(in_compat_syscall())) {
0714             compat_siginfo_t __user *uinfo = compat_ptr(data);
0715 
0716             if (copy_siginfo_to_user32(uinfo, &info) ||
0717                 __put_user(info.si_code, &uinfo->si_code)) {
0718                 ret = -EFAULT;
0719                 break;
0720             }
0721 
0722         } else
0723 #endif
0724         {
0725             siginfo_t __user *uinfo = (siginfo_t __user *) data;
0726 
0727             if (copy_siginfo_to_user(uinfo, &info) ||
0728                 __put_user(info.si_code, &uinfo->si_code)) {
0729                 ret = -EFAULT;
0730                 break;
0731             }
0732         }
0733 
0734         data += sizeof(siginfo_t);
0735         i++;
0736 
0737         if (signal_pending(current))
0738             break;
0739 
0740         cond_resched();
0741     }
0742 
0743     if (i > 0)
0744         return i;
0745 
0746     return ret;
0747 }
0748 
0749 #ifdef PTRACE_SINGLESTEP
0750 #define is_singlestep(request)      ((request) == PTRACE_SINGLESTEP)
0751 #else
0752 #define is_singlestep(request)      0
0753 #endif
0754 
0755 #ifdef PTRACE_SINGLEBLOCK
0756 #define is_singleblock(request)     ((request) == PTRACE_SINGLEBLOCK)
0757 #else
0758 #define is_singleblock(request)     0
0759 #endif
0760 
0761 #ifdef PTRACE_SYSEMU
0762 #define is_sysemu_singlestep(request)   ((request) == PTRACE_SYSEMU_SINGLESTEP)
0763 #else
0764 #define is_sysemu_singlestep(request)   0
0765 #endif
0766 
0767 static int ptrace_resume(struct task_struct *child, long request,
0768              unsigned long data)
0769 {
0770     bool need_siglock;
0771 
0772     if (!valid_signal(data))
0773         return -EIO;
0774 
0775     if (request == PTRACE_SYSCALL)
0776         set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
0777     else
0778         clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
0779 
0780 #ifdef TIF_SYSCALL_EMU
0781     if (request == PTRACE_SYSEMU || request == PTRACE_SYSEMU_SINGLESTEP)
0782         set_tsk_thread_flag(child, TIF_SYSCALL_EMU);
0783     else
0784         clear_tsk_thread_flag(child, TIF_SYSCALL_EMU);
0785 #endif
0786 
0787     if (is_singleblock(request)) {
0788         if (unlikely(!arch_has_block_step()))
0789             return -EIO;
0790         user_enable_block_step(child);
0791     } else if (is_singlestep(request) || is_sysemu_singlestep(request)) {
0792         if (unlikely(!arch_has_single_step()))
0793             return -EIO;
0794         user_enable_single_step(child);
0795     } else {
0796         user_disable_single_step(child);
0797     }
0798 
0799     /*
0800      * Change ->exit_code and ->state under siglock to avoid the race
0801      * with wait_task_stopped() in between; a non-zero ->exit_code will
0802      * wrongly look like another report from tracee.
0803      *
0804      * Note that we need siglock even if ->exit_code == data and/or this
0805      * status was not reported yet, the new status must not be cleared by
0806      * wait_task_stopped() after resume.
0807      *
0808      * If data == 0 we do not care if wait_task_stopped() reports the old
0809      * status and clears the code too; this can't race with the tracee, it
0810      * takes siglock after resume.
0811      */
0812     need_siglock = data && !thread_group_empty(current);
0813     if (need_siglock)
0814         spin_lock_irq(&child->sighand->siglock);
0815     child->exit_code = data;
0816     wake_up_state(child, __TASK_TRACED);
0817     if (need_siglock)
0818         spin_unlock_irq(&child->sighand->siglock);
0819 
0820     return 0;
0821 }
0822 
0823 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
0824 
0825 static const struct user_regset *
0826 find_regset(const struct user_regset_view *view, unsigned int type)
0827 {
0828     const struct user_regset *regset;
0829     int n;
0830 
0831     for (n = 0; n < view->n; ++n) {
0832         regset = view->regsets + n;
0833         if (regset->core_note_type == type)
0834             return regset;
0835     }
0836 
0837     return NULL;
0838 }
0839 
0840 static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
0841              struct iovec *kiov)
0842 {
0843     const struct user_regset_view *view = task_user_regset_view(task);
0844     const struct user_regset *regset = find_regset(view, type);
0845     int regset_no;
0846 
0847     if (!regset || (kiov->iov_len % regset->size) != 0)
0848         return -EINVAL;
0849 
0850     regset_no = regset - view->regsets;
0851     kiov->iov_len = min(kiov->iov_len,
0852                 (__kernel_size_t) (regset->n * regset->size));
0853 
0854     if (req == PTRACE_GETREGSET)
0855         return copy_regset_to_user(task, view, regset_no, 0,
0856                        kiov->iov_len, kiov->iov_base);
0857     else
0858         return copy_regset_from_user(task, view, regset_no, 0,
0859                          kiov->iov_len, kiov->iov_base);
0860 }
0861 
0862 /*
0863  * This is declared in linux/regset.h and defined in machine-dependent
0864  * code.  We put the export here, near the primary machine-neutral use,
0865  * to ensure no machine forgets it.
0866  */
0867 EXPORT_SYMBOL_GPL(task_user_regset_view);
0868 #endif
0869 
0870 int ptrace_request(struct task_struct *child, long request,
0871            unsigned long addr, unsigned long data)
0872 {
0873     bool seized = child->ptrace & PT_SEIZED;
0874     int ret = -EIO;
0875     siginfo_t siginfo, *si;
0876     void __user *datavp = (void __user *) data;
0877     unsigned long __user *datalp = datavp;
0878     unsigned long flags;
0879 
0880     switch (request) {
0881     case PTRACE_PEEKTEXT:
0882     case PTRACE_PEEKDATA:
0883         return generic_ptrace_peekdata(child, addr, data);
0884     case PTRACE_POKETEXT:
0885     case PTRACE_POKEDATA:
0886         return generic_ptrace_pokedata(child, addr, data);
0887 
0888 #ifdef PTRACE_OLDSETOPTIONS
0889     case PTRACE_OLDSETOPTIONS:
0890 #endif
0891     case PTRACE_SETOPTIONS:
0892         ret = ptrace_setoptions(child, data);
0893         break;
0894     case PTRACE_GETEVENTMSG:
0895         ret = put_user(child->ptrace_message, datalp);
0896         break;
0897 
0898     case PTRACE_PEEKSIGINFO:
0899         ret = ptrace_peek_siginfo(child, addr, data);
0900         break;
0901 
0902     case PTRACE_GETSIGINFO:
0903         ret = ptrace_getsiginfo(child, &siginfo);
0904         if (!ret)
0905             ret = copy_siginfo_to_user(datavp, &siginfo);
0906         break;
0907 
0908     case PTRACE_SETSIGINFO:
0909         if (copy_from_user(&siginfo, datavp, sizeof siginfo))
0910             ret = -EFAULT;
0911         else
0912             ret = ptrace_setsiginfo(child, &siginfo);
0913         break;
0914 
0915     case PTRACE_GETSIGMASK:
0916         if (addr != sizeof(sigset_t)) {
0917             ret = -EINVAL;
0918             break;
0919         }
0920 
0921         if (copy_to_user(datavp, &child->blocked, sizeof(sigset_t)))
0922             ret = -EFAULT;
0923         else
0924             ret = 0;
0925 
0926         break;
0927 
0928     case PTRACE_SETSIGMASK: {
0929         sigset_t new_set;
0930 
0931         if (addr != sizeof(sigset_t)) {
0932             ret = -EINVAL;
0933             break;
0934         }
0935 
0936         if (copy_from_user(&new_set, datavp, sizeof(sigset_t))) {
0937             ret = -EFAULT;
0938             break;
0939         }
0940 
0941         sigdelsetmask(&new_set, sigmask(SIGKILL)|sigmask(SIGSTOP));
0942 
0943         /*
0944          * Every thread does recalc_sigpending() after resume, so
0945          * retarget_shared_pending() and recalc_sigpending() are not
0946          * called here.
0947          */
0948         spin_lock_irq(&child->sighand->siglock);
0949         child->blocked = new_set;
0950         spin_unlock_irq(&child->sighand->siglock);
0951 
0952         ret = 0;
0953         break;
0954     }
0955 
0956     case PTRACE_INTERRUPT:
0957         /*
0958          * Stop tracee without any side-effect on signal or job
0959          * control.  At least one trap is guaranteed to happen
0960          * after this request.  If @child is already trapped, the
0961          * current trap is not disturbed and another trap will
0962          * happen after the current trap is ended with PTRACE_CONT.
0963          *
0964          * The actual trap might not be PTRACE_EVENT_STOP trap but
0965          * the pending condition is cleared regardless.
0966          */
0967         if (unlikely(!seized || !lock_task_sighand(child, &flags)))
0968             break;
0969 
0970         /*
0971          * INTERRUPT doesn't disturb existing trap sans one
0972          * exception.  If ptracer issued LISTEN for the current
0973          * STOP, this INTERRUPT should clear LISTEN and re-trap
0974          * tracee into STOP.
0975          */
0976         if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
0977             ptrace_signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
0978 
0979         unlock_task_sighand(child, &flags);
0980         ret = 0;
0981         break;
0982 
0983     case PTRACE_LISTEN:
0984         /*
0985          * Listen for events.  Tracee must be in STOP.  It's not
0986          * resumed per-se but is not considered to be in TRACED by
0987          * wait(2) or ptrace(2).  If an async event (e.g. group
0988          * stop state change) happens, tracee will enter STOP trap
0989          * again.  Alternatively, ptracer can issue INTERRUPT to
0990          * finish listening and re-trap tracee into STOP.
0991          */
0992         if (unlikely(!seized || !lock_task_sighand(child, &flags)))
0993             break;
0994 
0995         si = child->last_siginfo;
0996         if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
0997             child->jobctl |= JOBCTL_LISTENING;
0998             /*
0999              * If NOTIFY is set, it means event happened between
1000              * start of this trap and now.  Trigger re-trap.
1001              */
1002             if (child->jobctl & JOBCTL_TRAP_NOTIFY)
1003                 ptrace_signal_wake_up(child, true);
1004             ret = 0;
1005         }
1006         unlock_task_sighand(child, &flags);
1007         break;
1008 
1009     case PTRACE_DETACH:  /* detach a process that was attached. */
1010         ret = ptrace_detach(child, data);
1011         break;
1012 
1013 #ifdef CONFIG_BINFMT_ELF_FDPIC
1014     case PTRACE_GETFDPIC: {
1015         struct mm_struct *mm = get_task_mm(child);
1016         unsigned long tmp = 0;
1017 
1018         ret = -ESRCH;
1019         if (!mm)
1020             break;
1021 
1022         switch (addr) {
1023         case PTRACE_GETFDPIC_EXEC:
1024             tmp = mm->context.exec_fdpic_loadmap;
1025             break;
1026         case PTRACE_GETFDPIC_INTERP:
1027             tmp = mm->context.interp_fdpic_loadmap;
1028             break;
1029         default:
1030             break;
1031         }
1032         mmput(mm);
1033 
1034         ret = put_user(tmp, datalp);
1035         break;
1036     }
1037 #endif
1038 
1039 #ifdef PTRACE_SINGLESTEP
1040     case PTRACE_SINGLESTEP:
1041 #endif
1042 #ifdef PTRACE_SINGLEBLOCK
1043     case PTRACE_SINGLEBLOCK:
1044 #endif
1045 #ifdef PTRACE_SYSEMU
1046     case PTRACE_SYSEMU:
1047     case PTRACE_SYSEMU_SINGLESTEP:
1048 #endif
1049     case PTRACE_SYSCALL:
1050     case PTRACE_CONT:
1051         return ptrace_resume(child, request, data);
1052 
1053     case PTRACE_KILL:
1054         if (child->exit_state)  /* already dead */
1055             return 0;
1056         return ptrace_resume(child, request, SIGKILL);
1057 
1058 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1059     case PTRACE_GETREGSET:
1060     case PTRACE_SETREGSET: {
1061         struct iovec kiov;
1062         struct iovec __user *uiov = datavp;
1063 
1064         if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1065             return -EFAULT;
1066 
1067         if (__get_user(kiov.iov_base, &uiov->iov_base) ||
1068             __get_user(kiov.iov_len, &uiov->iov_len))
1069             return -EFAULT;
1070 
1071         ret = ptrace_regset(child, request, addr, &kiov);
1072         if (!ret)
1073             ret = __put_user(kiov.iov_len, &uiov->iov_len);
1074         break;
1075     }
1076 #endif
1077 
1078     case PTRACE_SECCOMP_GET_FILTER:
1079         ret = seccomp_get_filter(child, addr, datavp);
1080         break;
1081 
1082     default:
1083         break;
1084     }
1085 
1086     return ret;
1087 }
1088 
1089 static struct task_struct *ptrace_get_task_struct(pid_t pid)
1090 {
1091     struct task_struct *child;
1092 
1093     rcu_read_lock();
1094     child = find_task_by_vpid(pid);
1095     if (child)
1096         get_task_struct(child);
1097     rcu_read_unlock();
1098 
1099     if (!child)
1100         return ERR_PTR(-ESRCH);
1101     return child;
1102 }
1103 
1104 #ifndef arch_ptrace_attach
1105 #define arch_ptrace_attach(child)   do { } while (0)
1106 #endif
1107 
1108 SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
1109         unsigned long, data)
1110 {
1111     struct task_struct *child;
1112     long ret;
1113 
1114     if (request == PTRACE_TRACEME) {
1115         ret = ptrace_traceme();
1116         if (!ret)
1117             arch_ptrace_attach(current);
1118         goto out;
1119     }
1120 
1121     child = ptrace_get_task_struct(pid);
1122     if (IS_ERR(child)) {
1123         ret = PTR_ERR(child);
1124         goto out;
1125     }
1126 
1127     if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1128         ret = ptrace_attach(child, request, addr, data);
1129         /*
1130          * Some architectures need to do book-keeping after
1131          * a ptrace attach.
1132          */
1133         if (!ret)
1134             arch_ptrace_attach(child);
1135         goto out_put_task_struct;
1136     }
1137 
1138     ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1139                   request == PTRACE_INTERRUPT);
1140     if (ret < 0)
1141         goto out_put_task_struct;
1142 
1143     ret = arch_ptrace(child, request, addr, data);
1144     if (ret || request != PTRACE_DETACH)
1145         ptrace_unfreeze_traced(child);
1146 
1147  out_put_task_struct:
1148     put_task_struct(child);
1149  out:
1150     return ret;
1151 }
1152 
1153 int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
1154                 unsigned long data)
1155 {
1156     unsigned long tmp;
1157     int copied;
1158 
1159     copied = ptrace_access_vm(tsk, addr, &tmp, sizeof(tmp), FOLL_FORCE);
1160     if (copied != sizeof(tmp))
1161         return -EIO;
1162     return put_user(tmp, (unsigned long __user *)data);
1163 }
1164 
1165 int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
1166                 unsigned long data)
1167 {
1168     int copied;
1169 
1170     copied = ptrace_access_vm(tsk, addr, &data, sizeof(data),
1171             FOLL_FORCE | FOLL_WRITE);
1172     return (copied == sizeof(data)) ? 0 : -EIO;
1173 }
1174 
1175 #if defined CONFIG_COMPAT
1176 
1177 int compat_ptrace_request(struct task_struct *child, compat_long_t request,
1178               compat_ulong_t addr, compat_ulong_t data)
1179 {
1180     compat_ulong_t __user *datap = compat_ptr(data);
1181     compat_ulong_t word;
1182     siginfo_t siginfo;
1183     int ret;
1184 
1185     switch (request) {
1186     case PTRACE_PEEKTEXT:
1187     case PTRACE_PEEKDATA:
1188         ret = ptrace_access_vm(child, addr, &word, sizeof(word),
1189                 FOLL_FORCE);
1190         if (ret != sizeof(word))
1191             ret = -EIO;
1192         else
1193             ret = put_user(word, datap);
1194         break;
1195 
1196     case PTRACE_POKETEXT:
1197     case PTRACE_POKEDATA:
1198         ret = ptrace_access_vm(child, addr, &data, sizeof(data),
1199                 FOLL_FORCE | FOLL_WRITE);
1200         ret = (ret != sizeof(data) ? -EIO : 0);
1201         break;
1202 
1203     case PTRACE_GETEVENTMSG:
1204         ret = put_user((compat_ulong_t) child->ptrace_message, datap);
1205         break;
1206 
1207     case PTRACE_GETSIGINFO:
1208         ret = ptrace_getsiginfo(child, &siginfo);
1209         if (!ret)
1210             ret = copy_siginfo_to_user32(
1211                 (struct compat_siginfo __user *) datap,
1212                 &siginfo);
1213         break;
1214 
1215     case PTRACE_SETSIGINFO:
1216         memset(&siginfo, 0, sizeof siginfo);
1217         if (copy_siginfo_from_user32(
1218                 &siginfo, (struct compat_siginfo __user *) datap))
1219             ret = -EFAULT;
1220         else
1221             ret = ptrace_setsiginfo(child, &siginfo);
1222         break;
1223 #ifdef CONFIG_HAVE_ARCH_TRACEHOOK
1224     case PTRACE_GETREGSET:
1225     case PTRACE_SETREGSET:
1226     {
1227         struct iovec kiov;
1228         struct compat_iovec __user *uiov =
1229             (struct compat_iovec __user *) datap;
1230         compat_uptr_t ptr;
1231         compat_size_t len;
1232 
1233         if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
1234             return -EFAULT;
1235 
1236         if (__get_user(ptr, &uiov->iov_base) ||
1237             __get_user(len, &uiov->iov_len))
1238             return -EFAULT;
1239 
1240         kiov.iov_base = compat_ptr(ptr);
1241         kiov.iov_len = len;
1242 
1243         ret = ptrace_regset(child, request, addr, &kiov);
1244         if (!ret)
1245             ret = __put_user(kiov.iov_len, &uiov->iov_len);
1246         break;
1247     }
1248 #endif
1249 
1250     default:
1251         ret = ptrace_request(child, request, addr, data);
1252     }
1253 
1254     return ret;
1255 }
1256 
1257 COMPAT_SYSCALL_DEFINE4(ptrace, compat_long_t, request, compat_long_t, pid,
1258                compat_long_t, addr, compat_long_t, data)
1259 {
1260     struct task_struct *child;
1261     long ret;
1262 
1263     if (request == PTRACE_TRACEME) {
1264         ret = ptrace_traceme();
1265         goto out;
1266     }
1267 
1268     child = ptrace_get_task_struct(pid);
1269     if (IS_ERR(child)) {
1270         ret = PTR_ERR(child);
1271         goto out;
1272     }
1273 
1274     if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
1275         ret = ptrace_attach(child, request, addr, data);
1276         /*
1277          * Some architectures need to do book-keeping after
1278          * a ptrace attach.
1279          */
1280         if (!ret)
1281             arch_ptrace_attach(child);
1282         goto out_put_task_struct;
1283     }
1284 
1285     ret = ptrace_check_attach(child, request == PTRACE_KILL ||
1286                   request == PTRACE_INTERRUPT);
1287     if (!ret) {
1288         ret = compat_arch_ptrace(child, request, addr, data);
1289         if (ret || request != PTRACE_DETACH)
1290             ptrace_unfreeze_traced(child);
1291     }
1292 
1293  out_put_task_struct:
1294     put_task_struct(child);
1295  out:
1296     return ret;
1297 }
1298 #endif  /* CONFIG_COMPAT */