hung_task: fix false positive during vfork
[linux-2.6.git] / kernel / ptrace.c
index e6a99d2..a70d2a5 100644 (file)
 #include <linux/mm.h>
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
-#include <linux/smp_lock.h>
 #include <linux/ptrace.h>
 #include <linux/security.h>
 #include <linux/signal.h>
 #include <linux/audit.h>
 #include <linux/pid_namespace.h>
+#include <linux/syscalls.h>
+#include <linux/uaccess.h>
+#include <linux/regset.h>
+#include <linux/hw_breakpoint.h>
+#include <linux/cn_proc.h>
 
-#include <asm/pgtable.h>
-#include <asm/uaccess.h>
+
+static int ptrace_trapping_sleep_fn(void *flags)
+{
+       schedule();
+       return 0;
+}
 
 /*
  * ptrace a task: make the debugger its new parent and
  */
 void __ptrace_link(struct task_struct *child, struct task_struct *new_parent)
 {
-       BUG_ON(!list_empty(&child->ptrace_list));
-       if (child->parent == new_parent)
-               return;
-       list_add(&child->ptrace_list, &child->parent->ptrace_children);
-       remove_parent(child);
+       BUG_ON(!list_empty(&child->ptrace_entry));
+       list_add(&child->ptrace_entry, &new_parent->ptraced);
        child->parent = new_parent;
-       add_parent(child);
-}
-/*
- * Turn a tracing stop into a normal stop now, since with no tracer there
- * would be no way to wake it up with SIGCONT or SIGKILL.  If there was a
- * signal sent that would resume the child, but didn't because it was in
- * TASK_TRACED, resume it now.
- * Requires that irqs be disabled.
- */
-void ptrace_untrace(struct task_struct *child)
-{
-       spin_lock(&child->sighand->siglock);
-       if (child->state == TASK_TRACED) {
-               if (child->signal->flags & SIGNAL_STOP_STOPPED) {
-                       child->state = TASK_STOPPED;
-               } else {
-                       signal_wake_up(child, 1);
-               }
-       }
-       spin_unlock(&child->sighand->siglock);
 }
 
-/*
- * unptrace a task: move it back to its original parent and
- * remove it from the ptrace list.
+/**
+ * __ptrace_unlink - unlink ptracee and restore its execution state
+ * @child: ptracee to be unlinked
  *
- * Must be called with the tasklist lock write-held.
+ * Remove @child from the ptrace list, move it back to the original parent,
+ * and restore the execution state so that it conforms to the group stop
+ * state.
+ *
+ * Unlinking can happen via two paths - explicit PTRACE_DETACH or ptracer
+ * exiting.  For PTRACE_DETACH, unless the ptracee has been killed between
+ * ptrace_check_attach() and here, it's guaranteed to be in TASK_TRACED.
+ * If the ptracer is exiting, the ptracee can be in any state.
+ *
+ * After detach, the ptracee should be in a state which conforms to the
+ * group stop.  If the group is stopped or in the process of stopping, the
+ * ptracee should be put into TASK_STOPPED; otherwise, it should be woken
+ * up from TASK_TRACED.
+ *
+ * If the ptracee is in TASK_TRACED and needs to be moved to TASK_STOPPED,
+ * it goes through TRACED -> RUNNING -> STOPPED transition which is similar
+ * to but in the opposite direction of what happens while attaching to a
+ * stopped task.  However, in this direction, the intermediate RUNNING
+ * state is not hidden even from the current ptracer and if it immediately
+ * re-attaches and performs a WNOHANG wait(2), it may fail.
+ *
+ * CONTEXT:
+ * write_lock_irq(tasklist_lock)
  */
 void __ptrace_unlink(struct task_struct *child)
 {
        BUG_ON(!child->ptrace);
 
        child->ptrace = 0;
-       if (!list_empty(&child->ptrace_list)) {
-               list_del_init(&child->ptrace_list);
-               remove_parent(child);
-               child->parent = child->real_parent;
-               add_parent(child);
-       }
+       child->parent = child->real_parent;
+       list_del_init(&child->ptrace_entry);
+
+       spin_lock(&child->sighand->siglock);
+
+       /*
+        * Clear all pending traps and TRAPPING.  TRAPPING should be
+        * cleared regardless of JOBCTL_STOP_PENDING.  Do it explicitly.
+        */
+       task_clear_jobctl_pending(child, JOBCTL_TRAP_MASK);
+       task_clear_jobctl_trapping(child);
+
+       /*
+        * Reinstate JOBCTL_STOP_PENDING if group stop is in effect and
+        * @child isn't dead.
+        */
+       if (!(child->flags & PF_EXITING) &&
+           (child->signal->flags & SIGNAL_STOP_STOPPED ||
+            child->signal->group_stop_count))
+               child->jobctl |= JOBCTL_STOP_PENDING;
 
-       if (child->state == TASK_TRACED)
-               ptrace_untrace(child);
+       /*
+        * If transition to TASK_STOPPED is pending or in TASK_TRACED, kick
+        * @child in the butt.  Note that @resume should be used iff @child
+        * is in TASK_TRACED; otherwise, we might unduly disrupt
+        * TASK_KILLABLE sleeps.
+        */
+       if (child->jobctl & JOBCTL_STOP_PENDING || task_is_traced(child))
+               signal_wake_up(child, task_is_traced(child));
+
+       spin_unlock(&child->sighand->siglock);
 }
 
-/*
- * Check that we have indeed attached to the thing..
+/**
+ * ptrace_check_attach - check whether ptracee is ready for ptrace operation
+ * @child: ptracee to check for
+ * @ignore_state: don't check whether @child is currently %TASK_TRACED
+ *
+ * Check whether @child is being ptraced by %current and ready for further
+ * ptrace operations.  If @ignore_state is %false, @child also should be in
+ * %TASK_TRACED state and on return the child is guaranteed to be traced
+ * and not executing.  If @ignore_state is %true, @child can be in any
+ * state.
+ *
+ * CONTEXT:
+ * Grabs and releases tasklist_lock and @child->sighand->siglock.
+ *
+ * RETURNS:
+ * 0 on success, -ESRCH if %child is not ready.
  */
-int ptrace_check_attach(struct task_struct *child, int kill)
+int ptrace_check_attach(struct task_struct *child, bool ignore_state)
 {
        int ret = -ESRCH;
 
@@ -98,30 +140,31 @@ int ptrace_check_attach(struct task_struct *child, int kill)
         * be changed by us so it's not changing right after this.
         */
        read_lock(&tasklist_lock);
-       if ((child->ptrace & PT_PTRACED) && child->parent == current &&
-           (!(child->ptrace & PT_ATTACHED) || child->real_parent != current)
-           && child->signal != NULL) {
-               ret = 0;
+       if ((child->ptrace & PT_PTRACED) && child->parent == current) {
+               /*
+                * child->sighand can't be NULL, release_task()
+                * does ptrace_unlink() before __exit_signal().
+                */
                spin_lock_irq(&child->sighand->siglock);
-               if (child->state == TASK_STOPPED) {
-                       child->state = TASK_TRACED;
-               } else if (child->state != TASK_TRACED && !kill) {
-                       ret = -ESRCH;
-               }
+               WARN_ON_ONCE(task_is_stopped(child));
+               if (ignore_state || (task_is_traced(child) &&
+                                    !(child->jobctl & JOBCTL_LISTENING)))
+                       ret = 0;
                spin_unlock_irq(&child->sighand->siglock);
        }
        read_unlock(&tasklist_lock);
 
-       if (!ret && !kill) {
-               wait_task_inactive(child);
-       }
+       if (!ret && !ignore_state)
+               ret = wait_task_inactive(child, TASK_TRACED) ? 0 : -ESRCH;
 
        /* All systems go.. */
        return ret;
 }
 
-int __ptrace_may_attach(struct task_struct *task)
+int __ptrace_may_access(struct task_struct *task, unsigned int mode)
 {
+       const struct cred *cred = current_cred(), *tcred;
+
        /* May we inspect the given task?
         * This check is used both for attaching with ptrace
         * and for allowing access to sensitive information in /proc.
@@ -134,101 +177,230 @@ int __ptrace_may_attach(struct task_struct *task)
        /* Don't let security modules deny introspection */
        if (task == current)
                return 0;
-       if (((current->uid != task->euid) ||
-            (current->uid != task->suid) ||
-            (current->uid != task->uid) ||
-            (current->gid != task->egid) ||
-            (current->gid != task->sgid) ||
-            (current->gid != task->gid)) && !capable(CAP_SYS_PTRACE))
-               return -EPERM;
+       rcu_read_lock();
+       tcred = __task_cred(task);
+       if (cred->user->user_ns == tcred->user->user_ns &&
+           (cred->uid == tcred->euid &&
+            cred->uid == tcred->suid &&
+            cred->uid == tcred->uid  &&
+            cred->gid == tcred->egid &&
+            cred->gid == tcred->sgid &&
+            cred->gid == tcred->gid))
+               goto ok;
+       if (ns_capable(tcred->user->user_ns, CAP_SYS_PTRACE))
+               goto ok;
+       rcu_read_unlock();
+       return -EPERM;
+ok:
+       rcu_read_unlock();
        smp_rmb();
        if (task->mm)
                dumpable = get_dumpable(task->mm);
-       if (!dumpable && !capable(CAP_SYS_PTRACE))
+       if (!dumpable && !task_ns_capable(task, CAP_SYS_PTRACE))
                return -EPERM;
 
-       return security_ptrace(current, task);
+       return security_ptrace_access_check(task, mode);
 }
 
-int ptrace_may_attach(struct task_struct *task)
+bool ptrace_may_access(struct task_struct *task, unsigned int mode)
 {
        int err;
        task_lock(task);
-       err = __ptrace_may_attach(task);
+       err = __ptrace_may_access(task, mode);
        task_unlock(task);
        return !err;
 }
 
-int ptrace_attach(struct task_struct *task)
+static int ptrace_attach(struct task_struct *task, long request,
+                        unsigned long flags)
 {
+       bool seize = (request == PTRACE_SEIZE);
        int retval;
-       unsigned long flags;
+
+       /*
+        * SEIZE will enable new ptrace behaviors which will be implemented
+        * gradually.  SEIZE_DEVEL is used to prevent applications
+        * expecting full SEIZE behaviors trapping on kernel commits which
+        * are still in the process of implementing them.
+        *
+        * Only test programs for new ptrace behaviors being implemented
+        * should set SEIZE_DEVEL.  If unset, SEIZE will fail with -EIO.
+        *
+        * Once SEIZE behaviors are completely implemented, this flag and
+        * the following test will be removed.
+        */
+       retval = -EIO;
+       if (seize && !(flags & PTRACE_SEIZE_DEVEL))
+               goto out;
 
        audit_ptrace(task);
 
        retval = -EPERM;
-       if (task->pid <= 1)
+       if (unlikely(task->flags & PF_KTHREAD))
                goto out;
        if (same_thread_group(task, current))
                goto out;
 
-repeat:
        /*
-        * Nasty, nasty.
-        *
-        * We want to hold both the task-lock and the
-        * tasklist_lock for writing at the same time.
-        * But that's against the rules (tasklist_lock
-        * is taken for reading by interrupts on other
-        * cpu's that may have task_lock).
+        * Protect exec's credential calculations against our interference;
+        * interference; SUID, SGID and LSM creds get determined differently
+        * under ptrace.
         */
-       task_lock(task);
-       if (!write_trylock_irqsave(&tasklist_lock, flags)) {
-               task_unlock(task);
-               do {
-                       cpu_relax();
-               } while (!write_can_lock(&tasklist_lock));
-               goto repeat;
-       }
+       retval = -ERESTARTNOINTR;
+       if (mutex_lock_interruptible(&task->signal->cred_guard_mutex))
+               goto out;
 
-       if (!task->mm)
-               goto bad;
-       /* the same process cannot be attached many times */
-       if (task->ptrace & PT_PTRACED)
-               goto bad;
-       retval = __ptrace_may_attach(task);
+       task_lock(task);
+       retval = __ptrace_may_access(task, PTRACE_MODE_ATTACH);
+       task_unlock(task);
        if (retval)
-               goto bad;
+               goto unlock_creds;
 
-       /* Go */
-       task->ptrace |= PT_PTRACED | ((task->real_parent != current)
-                                     ? PT_ATTACHED : 0);
-       if (capable(CAP_SYS_PTRACE))
+       write_lock_irq(&tasklist_lock);
+       retval = -EPERM;
+       if (unlikely(task->exit_state))
+               goto unlock_tasklist;
+       if (task->ptrace)
+               goto unlock_tasklist;
+
+       task->ptrace = PT_PTRACED;
+       if (seize)
+               task->ptrace |= PT_SEIZED;
+       if (task_ns_capable(task, CAP_SYS_PTRACE))
                task->ptrace |= PT_PTRACE_CAP;
 
        __ptrace_link(task, current);
 
-       force_sig_specific(SIGSTOP, task);
+       /* SEIZE doesn't trap tracee on attach */
+       if (!seize)
+               send_sig_info(SIGSTOP, SEND_SIG_FORCED, task);
 
-bad:
-       write_unlock_irqrestore(&tasklist_lock, flags);
-       task_unlock(task);
+       spin_lock(&task->sighand->siglock);
+
+       /*
+        * If the task is already STOPPED, set JOBCTL_TRAP_STOP and
+        * TRAPPING, and kick it so that it transits to TRACED.  TRAPPING
+        * will be cleared if the child completes the transition or any
+        * event which clears the group stop states happens.  We'll wait
+        * for the transition to complete before returning from this
+        * function.
+        *
+        * This hides STOPPED -> RUNNING -> TRACED transition from the
+        * attaching thread but a different thread in the same group can
+        * still observe the transient RUNNING state.  IOW, if another
+        * thread's WNOHANG wait(2) on the stopped tracee races against
+        * ATTACH, the wait(2) may fail due to the transient RUNNING.
+        *
+        * The following task_is_stopped() test is safe as both transitions
+        * in and out of STOPPED are protected by siglock.
+        */
+       if (task_is_stopped(task) &&
+           task_set_jobctl_pending(task, JOBCTL_TRAP_STOP | JOBCTL_TRAPPING))
+               signal_wake_up(task, 1);
+
+       spin_unlock(&task->sighand->siglock);
+
+       retval = 0;
+unlock_tasklist:
+       write_unlock_irq(&tasklist_lock);
+unlock_creds:
+       mutex_unlock(&task->signal->cred_guard_mutex);
 out:
+       if (!retval) {
+               wait_on_bit(&task->jobctl, JOBCTL_TRAPPING_BIT,
+                           ptrace_trapping_sleep_fn, TASK_UNINTERRUPTIBLE);
+               proc_ptrace_connector(task, PTRACE_ATTACH);
+       }
+
        return retval;
 }
 
-static inline void __ptrace_detach(struct task_struct *child, unsigned int data)
+/**
+ * ptrace_traceme  --  helper for PTRACE_TRACEME
+ *
+ * Performs checks and sets PT_PTRACED.
+ * Should be used by all ptrace implementations for PTRACE_TRACEME.
+ */
+static int ptrace_traceme(void)
+{
+       int ret = -EPERM;
+
+       write_lock_irq(&tasklist_lock);
+       /* Are we already being traced? */
+       if (!current->ptrace) {
+               ret = security_ptrace_traceme(current->parent);
+               /*
+                * Check PF_EXITING to ensure ->real_parent has not passed
+                * exit_ptrace(). Otherwise we don't report the error but
+                * pretend ->real_parent untraces us right after return.
+                */
+               if (!ret && !(current->real_parent->flags & PF_EXITING)) {
+                       current->ptrace = PT_PTRACED;
+                       __ptrace_link(current, current->real_parent);
+               }
+       }
+       write_unlock_irq(&tasklist_lock);
+
+       return ret;
+}
+
+/*
+ * Called with irqs disabled, returns true if childs should reap themselves.
+ */
+static int ignoring_children(struct sighand_struct *sigh)
 {
-       child->exit_code = data;
-       /* .. re-parent .. */
-       __ptrace_unlink(child);
-       /* .. and wake it up. */
-       if (child->exit_state != EXIT_ZOMBIE)
-               wake_up_process(child);
+       int ret;
+       spin_lock(&sigh->siglock);
+       ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
+             (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
+       spin_unlock(&sigh->siglock);
+       return ret;
+}
+
+/*
+ * Called with tasklist_lock held for writing.
+ * Unlink a traced task, and clean it up if it was a traced zombie.
+ * Return true if it needs to be reaped with release_task().
+ * (We can't call release_task() here because we already hold tasklist_lock.)
+ *
+ * If it's a zombie, our attachedness prevented normal parent notification
+ * or self-reaping.  Do notification now if it would have happened earlier.
+ * If it should reap itself, return true.
+ *
+ * If it's our own child, there is no notification to do. But if our normal
+ * children self-reap, then this child was prevented by ptrace and we must
+ * reap it now, in that case we must also wake up sub-threads sleeping in
+ * do_wait().
+ */
+static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
+{
+       bool dead;
+
+       __ptrace_unlink(p);
+
+       if (p->exit_state != EXIT_ZOMBIE)
+               return false;
+
+       dead = !thread_group_leader(p);
+
+       if (!dead && thread_group_empty(p)) {
+               if (!same_thread_group(p->real_parent, tracer))
+                       dead = do_notify_parent(p, p->exit_signal);
+               else if (ignoring_children(tracer->sighand)) {
+                       __wake_up_parent(p, tracer);
+                       dead = true;
+               }
+       }
+       /* Mark it as in the process of being reaped. */
+       if (dead)
+               p->exit_state = EXIT_DEAD;
+       return dead;
 }
 
-int ptrace_detach(struct task_struct *child, unsigned int data)
+static int ptrace_detach(struct task_struct *child, unsigned int data)
 {
+       bool dead = false;
+
        if (!valid_signal(data))
                return -EIO;
 
@@ -237,14 +409,54 @@ int ptrace_detach(struct task_struct *child, unsigned int data)
        clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
 
        write_lock_irq(&tasklist_lock);
-       /* protect against de_thread()->release_task() */
-       if (child->ptrace)
-               __ptrace_detach(child, data);
+       /*
+        * This child can be already killed. Make sure de_thread() or
+        * our sub-thread doing do_wait() didn't do release_task() yet.
+        */
+       if (child->ptrace) {
+               child->exit_code = data;
+               dead = __ptrace_detach(current, child);
+       }
        write_unlock_irq(&tasklist_lock);
 
+       proc_ptrace_connector(child, PTRACE_DETACH);
+       if (unlikely(dead))
+               release_task(child);
+
        return 0;
 }
 
+/*
+ * Detach all tasks we were using ptrace on. Called with tasklist held
+ * for writing, and returns with it held too. But note it can release
+ * and reacquire the lock.
+ */
+void exit_ptrace(struct task_struct *tracer)
+       __releases(&tasklist_lock)
+       __acquires(&tasklist_lock)
+{
+       struct task_struct *p, *n;
+       LIST_HEAD(ptrace_dead);
+
+       if (likely(list_empty(&tracer->ptraced)))
+               return;
+
+       list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
+               if (__ptrace_detach(tracer, p))
+                       list_add(&p->ptrace_entry, &ptrace_dead);
+       }
+
+       write_unlock_irq(&tasklist_lock);
+       BUG_ON(!list_empty(&tracer->ptraced));
+
+       list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_entry) {
+               list_del_init(&p->ptrace_entry);
+               release_task(p);
+       }
+
+       write_lock_irq(&tasklist_lock);
+}
+
 int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
 {
        int copied = 0;
@@ -265,7 +477,7 @@ int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst
                copied += retval;
                src += retval;
                dst += retval;
-               len -= retval;                  
+               len -= retval;
        }
        return copied;
 }
@@ -290,12 +502,12 @@ int ptrace_writedata(struct task_struct *tsk, char __user *src, unsigned long ds
                copied += retval;
                src += retval;
                dst += retval;
-               len -= retval;                  
+               len -= retval;
        }
        return copied;
 }
 
-static int ptrace_setoptions(struct task_struct *child, long data)
+static int ptrace_setoptions(struct task_struct *child, unsigned long data)
 {
        child->ptrace &= ~PT_TRACE_MASK;
 
@@ -323,46 +535,35 @@ static int ptrace_setoptions(struct task_struct *child, long data)
        return (data & ~PTRACE_O_MASK) ? -EINVAL : 0;
 }
 
-static int ptrace_getsiginfo(struct task_struct *child, siginfo_t __user * data)
+static int ptrace_getsiginfo(struct task_struct *child, siginfo_t *info)
 {
-       siginfo_t lastinfo;
+       unsigned long flags;
        int error = -ESRCH;
 
-       read_lock(&tasklist_lock);
-       if (likely(child->sighand != NULL)) {
+       if (lock_task_sighand(child, &flags)) {
                error = -EINVAL;
-               spin_lock_irq(&child->sighand->siglock);
                if (likely(child->last_siginfo != NULL)) {
-                       lastinfo = *child->last_siginfo;
+                       *info = *child->last_siginfo;
                        error = 0;
                }
-               spin_unlock_irq(&child->sighand->siglock);
+               unlock_task_sighand(child, &flags);
        }
-       read_unlock(&tasklist_lock);
-       if (!error)
-               return copy_siginfo_to_user(data, &lastinfo);
        return error;
 }
 
-static int ptrace_setsiginfo(struct task_struct *child, siginfo_t __user * data)
+static int ptrace_setsiginfo(struct task_struct *child, const siginfo_t *info)
 {
-       siginfo_t newinfo;
+       unsigned long flags;
        int error = -ESRCH;
 
-       if (copy_from_user(&newinfo, data, sizeof (siginfo_t)))
-               return -EFAULT;
-
-       read_lock(&tasklist_lock);
-       if (likely(child->sighand != NULL)) {
+       if (lock_task_sighand(child, &flags)) {
                error = -EINVAL;
-               spin_lock_irq(&child->sighand->siglock);
                if (likely(child->last_siginfo != NULL)) {
-                       *child->last_siginfo = newinfo;
+                       *child->last_siginfo = *info;
                        error = 0;
                }
-               spin_unlock_irq(&child->sighand->siglock);
+               unlock_task_sighand(child, &flags);
        }
-       read_unlock(&tasklist_lock);
        return error;
 }
 
@@ -385,7 +586,8 @@ static int ptrace_setsiginfo(struct task_struct *child, siginfo_t __user * data)
 #define is_sysemu_singlestep(request)  0
 #endif
 
-static int ptrace_resume(struct task_struct *child, long request, long data)
+static int ptrace_resume(struct task_struct *child, long request,
+                        unsigned long data)
 {
        if (!valid_signal(data))
                return -EIO;
@@ -410,20 +612,66 @@ static int ptrace_resume(struct task_struct *child, long request, long data)
                if (unlikely(!arch_has_single_step()))
                        return -EIO;
                user_enable_single_step(child);
-       }
-       else
+       } else {
                user_disable_single_step(child);
+       }
 
        child->exit_code = data;
-       wake_up_process(child);
+       wake_up_state(child, __TASK_TRACED);
 
        return 0;
 }
 
+#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
+
+static const struct user_regset *
+find_regset(const struct user_regset_view *view, unsigned int type)
+{
+       const struct user_regset *regset;
+       int n;
+
+       for (n = 0; n < view->n; ++n) {
+               regset = view->regsets + n;
+               if (regset->core_note_type == type)
+                       return regset;
+       }
+
+       return NULL;
+}
+
+static int ptrace_regset(struct task_struct *task, int req, unsigned int type,
+                        struct iovec *kiov)
+{
+       const struct user_regset_view *view = task_user_regset_view(task);
+       const struct user_regset *regset = find_regset(view, type);
+       int regset_no;
+
+       if (!regset || (kiov->iov_len % regset->size) != 0)
+               return -EINVAL;
+
+       regset_no = regset - view->regsets;
+       kiov->iov_len = min(kiov->iov_len,
+                           (__kernel_size_t) (regset->n * regset->size));
+
+       if (req == PTRACE_GETREGSET)
+               return copy_regset_to_user(task, view, regset_no, 0,
+                                          kiov->iov_len, kiov->iov_base);
+       else
+               return copy_regset_from_user(task, view, regset_no, 0,
+                                            kiov->iov_len, kiov->iov_base);
+}
+
+#endif
+
 int ptrace_request(struct task_struct *child, long request,
-                  long addr, long data)
+                  unsigned long addr, unsigned long data)
 {
+       bool seized = child->ptrace & PT_SEIZED;
        int ret = -EIO;
+       siginfo_t siginfo, *si;
+       void __user *datavp = (void __user *) data;
+       unsigned long __user *datalp = datavp;
+       unsigned long flags;
 
        switch (request) {
        case PTRACE_PEEKTEXT:
@@ -440,18 +688,105 @@ int ptrace_request(struct task_struct *child, long request,
                ret = ptrace_setoptions(child, data);
                break;
        case PTRACE_GETEVENTMSG:
-               ret = put_user(child->ptrace_message, (unsigned long __user *) data);
+               ret = put_user(child->ptrace_message, datalp);
                break;
+
        case PTRACE_GETSIGINFO:
-               ret = ptrace_getsiginfo(child, (siginfo_t __user *) data);
+               ret = ptrace_getsiginfo(child, &siginfo);
+               if (!ret)
+                       ret = copy_siginfo_to_user(datavp, &siginfo);
                break;
+
        case PTRACE_SETSIGINFO:
-               ret = ptrace_setsiginfo(child, (siginfo_t __user *) data);
+               if (copy_from_user(&siginfo, datavp, sizeof siginfo))
+                       ret = -EFAULT;
+               else
+                       ret = ptrace_setsiginfo(child, &siginfo);
                break;
+
+       case PTRACE_INTERRUPT:
+               /*
+                * Stop tracee without any side-effect on signal or job
+                * control.  At least one trap is guaranteed to happen
+                * after this request.  If @child is already trapped, the
+                * current trap is not disturbed and another trap will
+                * happen after the current trap is ended with PTRACE_CONT.
+                *
+                * The actual trap might not be PTRACE_EVENT_STOP trap but
+                * the pending condition is cleared regardless.
+                */
+               if (unlikely(!seized || !lock_task_sighand(child, &flags)))
+                       break;
+
+               /*
+                * INTERRUPT doesn't disturb existing trap sans one
+                * exception.  If ptracer issued LISTEN for the current
+                * STOP, this INTERRUPT should clear LISTEN and re-trap
+                * tracee into STOP.
+                */
+               if (likely(task_set_jobctl_pending(child, JOBCTL_TRAP_STOP)))
+                       signal_wake_up(child, child->jobctl & JOBCTL_LISTENING);
+
+               unlock_task_sighand(child, &flags);
+               ret = 0;
+               break;
+
+       case PTRACE_LISTEN:
+               /*
+                * Listen for events.  Tracee must be in STOP.  It's not
+                * resumed per-se but is not considered to be in TRACED by
+                * wait(2) or ptrace(2).  If an async event (e.g. group
+                * stop state change) happens, tracee will enter STOP trap
+                * again.  Alternatively, ptracer can issue INTERRUPT to
+                * finish listening and re-trap tracee into STOP.
+                */
+               if (unlikely(!seized || !lock_task_sighand(child, &flags)))
+                       break;
+
+               si = child->last_siginfo;
+               if (likely(si && (si->si_code >> 8) == PTRACE_EVENT_STOP)) {
+                       child->jobctl |= JOBCTL_LISTENING;
+                       /*
+                        * If NOTIFY is set, it means event happened between
+                        * start of this trap and now.  Trigger re-trap.
+                        */
+                       if (child->jobctl & JOBCTL_TRAP_NOTIFY)
+                               signal_wake_up(child, true);
+                       ret = 0;
+               }
+               unlock_task_sighand(child, &flags);
+               break;
+
        case PTRACE_DETACH:      /* detach a process that was attached. */
                ret = ptrace_detach(child, data);
                break;
 
+#ifdef CONFIG_BINFMT_ELF_FDPIC
+       case PTRACE_GETFDPIC: {
+               struct mm_struct *mm = get_task_mm(child);
+               unsigned long tmp = 0;
+
+               ret = -ESRCH;
+               if (!mm)
+                       break;
+
+               switch (addr) {
+               case PTRACE_GETFDPIC_EXEC:
+                       tmp = mm->context.exec_fdpic_loadmap;
+                       break;
+               case PTRACE_GETFDPIC_INTERP:
+                       tmp = mm->context.interp_fdpic_loadmap;
+                       break;
+               default:
+                       break;
+               }
+               mmput(mm);
+
+               ret = put_user(tmp, datalp);
+               break;
+       }
+#endif
+
 #ifdef PTRACE_SINGLESTEP
        case PTRACE_SINGLESTEP:
 #endif
@@ -471,65 +806,43 @@ int ptrace_request(struct task_struct *child, long request,
                        return 0;
                return ptrace_resume(child, request, SIGKILL);
 
-       default:
-               break;
-       }
+#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
+       case PTRACE_GETREGSET:
+       case PTRACE_SETREGSET:
+       {
+               struct iovec kiov;
+               struct iovec __user *uiov = datavp;
 
-       return ret;
-}
+               if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
+                       return -EFAULT;
 
-/**
- * ptrace_traceme  --  helper for PTRACE_TRACEME
- *
- * Performs checks and sets PT_PTRACED.
- * Should be used by all ptrace implementations for PTRACE_TRACEME.
- */
-int ptrace_traceme(void)
-{
-       int ret = -EPERM;
+               if (__get_user(kiov.iov_base, &uiov->iov_base) ||
+                   __get_user(kiov.iov_len, &uiov->iov_len))
+                       return -EFAULT;
 
-       /*
-        * Are we already being traced?
-        */
-       task_lock(current);
-       if (!(current->ptrace & PT_PTRACED)) {
-               ret = security_ptrace(current->parent, current);
-               /*
-                * Set the ptrace bit in the process ptrace flags.
-                */
+               ret = ptrace_regset(child, request, addr, &kiov);
                if (!ret)
-                       current->ptrace |= PT_PTRACED;
+                       ret = __put_user(kiov.iov_len, &uiov->iov_len);
+               break;
        }
-       task_unlock(current);
+#endif
+       default:
+               break;
+       }
+
        return ret;
 }
 
-/**
- * ptrace_get_task_struct  --  grab a task struct reference for ptrace
- * @pid:       process id to grab a task_struct reference of
- *
- * This function is a helper for ptrace implementations.  It checks
- * permissions and then grabs a task struct for use of the actual
- * ptrace implementation.
- *
- * Returns the task_struct for @pid or an ERR_PTR() on failure.
- */
-struct task_struct *ptrace_get_task_struct(pid_t pid)
+static struct task_struct *ptrace_get_task_struct(pid_t pid)
 {
        struct task_struct *child;
 
-       /*
-        * Tracing init is not allowed.
-        */
-       if (pid == 1)
-               return ERR_PTR(-EPERM);
-
-       read_lock(&tasklist_lock);
+       rcu_read_lock();
        child = find_task_by_vpid(pid);
        if (child)
                get_task_struct(child);
+       rcu_read_unlock();
 
-       read_unlock(&tasklist_lock);
        if (!child)
                return ERR_PTR(-ESRCH);
        return child;
@@ -539,16 +852,12 @@ struct task_struct *ptrace_get_task_struct(pid_t pid)
 #define arch_ptrace_attach(child)      do { } while (0)
 #endif
 
-#ifndef __ARCH_SYS_PTRACE
-asmlinkage long sys_ptrace(long request, long pid, long addr, long data)
+SYSCALL_DEFINE4(ptrace, long, request, long, pid, unsigned long, addr,
+               unsigned long, data)
 {
        struct task_struct *child;
        long ret;
 
-       /*
-        * This lock_kernel fixes a subtle race with suid exec
-        */
-       lock_kernel();
        if (request == PTRACE_TRACEME) {
                ret = ptrace_traceme();
                if (!ret)
@@ -562,8 +871,8 @@ asmlinkage long sys_ptrace(long request, long pid, long addr, long data)
                goto out;
        }
 
-       if (request == PTRACE_ATTACH) {
-               ret = ptrace_attach(child);
+       if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
+               ret = ptrace_attach(child, request, data);
                /*
                 * Some architectures need to do book-keeping after
                 * a ptrace attach.
@@ -573,23 +882,21 @@ asmlinkage long sys_ptrace(long request, long pid, long addr, long data)
                goto out_put_task_struct;
        }
 
-       ret = ptrace_check_attach(child, request == PTRACE_KILL);
+       ret = ptrace_check_attach(child, request == PTRACE_KILL ||
+                                 request == PTRACE_INTERRUPT);
        if (ret < 0)
                goto out_put_task_struct;
 
        ret = arch_ptrace(child, request, addr, data);
-       if (ret < 0)
-               goto out_put_task_struct;
 
  out_put_task_struct:
        put_task_struct(child);
  out:
-       unlock_kernel();
        return ret;
 }
-#endif /* __ARCH_SYS_PTRACE */
 
-int generic_ptrace_peekdata(struct task_struct *tsk, long addr, long data)
+int generic_ptrace_peekdata(struct task_struct *tsk, unsigned long addr,
+                           unsigned long data)
 {
        unsigned long tmp;
        int copied;
@@ -600,10 +907,148 @@ int generic_ptrace_peekdata(struct task_struct *tsk, long addr, long data)
        return put_user(tmp, (unsigned long __user *)data);
 }
 
-int generic_ptrace_pokedata(struct task_struct *tsk, long addr, long data)
+int generic_ptrace_pokedata(struct task_struct *tsk, unsigned long addr,
+                           unsigned long data)
 {
        int copied;
 
        copied = access_process_vm(tsk, addr, &data, sizeof(data), 1);
        return (copied == sizeof(data)) ? 0 : -EIO;
 }
+
+#if defined CONFIG_COMPAT
+#include <linux/compat.h>
+
+int compat_ptrace_request(struct task_struct *child, compat_long_t request,
+                         compat_ulong_t addr, compat_ulong_t data)
+{
+       compat_ulong_t __user *datap = compat_ptr(data);
+       compat_ulong_t word;
+       siginfo_t siginfo;
+       int ret;
+
+       switch (request) {
+       case PTRACE_PEEKTEXT:
+       case PTRACE_PEEKDATA:
+               ret = access_process_vm(child, addr, &word, sizeof(word), 0);
+               if (ret != sizeof(word))
+                       ret = -EIO;
+               else
+                       ret = put_user(word, datap);
+               break;
+
+       case PTRACE_POKETEXT:
+       case PTRACE_POKEDATA:
+               ret = access_process_vm(child, addr, &data, sizeof(data), 1);
+               ret = (ret != sizeof(data) ? -EIO : 0);
+               break;
+
+       case PTRACE_GETEVENTMSG:
+               ret = put_user((compat_ulong_t) child->ptrace_message, datap);
+               break;
+
+       case PTRACE_GETSIGINFO:
+               ret = ptrace_getsiginfo(child, &siginfo);
+               if (!ret)
+                       ret = copy_siginfo_to_user32(
+                               (struct compat_siginfo __user *) datap,
+                               &siginfo);
+               break;
+
+       case PTRACE_SETSIGINFO:
+               memset(&siginfo, 0, sizeof siginfo);
+               if (copy_siginfo_from_user32(
+                           &siginfo, (struct compat_siginfo __user *) datap))
+                       ret = -EFAULT;
+               else
+                       ret = ptrace_setsiginfo(child, &siginfo);
+               break;
+#ifdef CONFIG_HAVE_ARCH_TRACEHOOK
+       case PTRACE_GETREGSET:
+       case PTRACE_SETREGSET:
+       {
+               struct iovec kiov;
+               struct compat_iovec __user *uiov =
+                       (struct compat_iovec __user *) datap;
+               compat_uptr_t ptr;
+               compat_size_t len;
+
+               if (!access_ok(VERIFY_WRITE, uiov, sizeof(*uiov)))
+                       return -EFAULT;
+
+               if (__get_user(ptr, &uiov->iov_base) ||
+                   __get_user(len, &uiov->iov_len))
+                       return -EFAULT;
+
+               kiov.iov_base = compat_ptr(ptr);
+               kiov.iov_len = len;
+
+               ret = ptrace_regset(child, request, addr, &kiov);
+               if (!ret)
+                       ret = __put_user(kiov.iov_len, &uiov->iov_len);
+               break;
+       }
+#endif
+
+       default:
+               ret = ptrace_request(child, request, addr, data);
+       }
+
+       return ret;
+}
+
+asmlinkage long compat_sys_ptrace(compat_long_t request, compat_long_t pid,
+                                 compat_long_t addr, compat_long_t data)
+{
+       struct task_struct *child;
+       long ret;
+
+       if (request == PTRACE_TRACEME) {
+               ret = ptrace_traceme();
+               goto out;
+       }
+
+       child = ptrace_get_task_struct(pid);
+       if (IS_ERR(child)) {
+               ret = PTR_ERR(child);
+               goto out;
+       }
+
+       if (request == PTRACE_ATTACH || request == PTRACE_SEIZE) {
+               ret = ptrace_attach(child, request, data);
+               /*
+                * Some architectures need to do book-keeping after
+                * a ptrace attach.
+                */
+               if (!ret)
+                       arch_ptrace_attach(child);
+               goto out_put_task_struct;
+       }
+
+       ret = ptrace_check_attach(child, request == PTRACE_KILL ||
+                                 request == PTRACE_INTERRUPT);
+       if (!ret)
+               ret = compat_arch_ptrace(child, request, addr, data);
+
+ out_put_task_struct:
+       put_task_struct(child);
+ out:
+       return ret;
+}
+#endif /* CONFIG_COMPAT */
+
+#ifdef CONFIG_HAVE_HW_BREAKPOINT
+int ptrace_get_breakpoints(struct task_struct *tsk)
+{
+       if (atomic_inc_not_zero(&tsk->ptrace_bp_refcnt))
+               return 0;
+
+       return -1;
+}
+
+void ptrace_put_breakpoints(struct task_struct *tsk)
+{
+       if (atomic_dec_and_test(&tsk->ptrace_bp_refcnt))
+               flush_ptrace_hw_breakpoint(tsk);
+}
+#endif /* CONFIG_HAVE_HW_BREAKPOINT */