Merge branch 'linus' into stackprotector
[linux-2.6.git] / kernel / exit.c
index c8d0485..e69edc7 100644 (file)
 #include <linux/cn_proc.h>
 #include <linux/mutex.h>
 #include <linux/futex.h>
-#include <linux/compat.h>
 #include <linux/pipe_fs_i.h>
 #include <linux/audit.h> /* for audit_free() */
 #include <linux/resource.h>
 #include <linux/blkdev.h>
 #include <linux/task_io_accounting_ops.h>
 #include <linux/tracehook.h>
+#include <linux/init_task.h>
+#include <trace/sched.h>
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
 #include <asm/pgtable.h>
 #include <asm/mmu_context.h>
+#include "cred-internals.h"
+
+DEFINE_TRACE(sched_process_free);
+DEFINE_TRACE(sched_process_exit);
+DEFINE_TRACE(sched_process_wait);
 
 static void exit_mm(struct task_struct * tsk);
 
@@ -112,8 +118,6 @@ static void __exit_signal(struct task_struct *tsk)
                 * We won't ever get here for the group leader, since it
                 * will have been the last reference on the signal_struct.
                 */
-               sig->utime = cputime_add(sig->utime, task_utime(tsk));
-               sig->stime = cputime_add(sig->stime, task_stime(tsk));
                sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
                sig->min_flt += tsk->min_flt;
                sig->maj_flt += tsk->maj_flt;
@@ -122,7 +126,6 @@ static void __exit_signal(struct task_struct *tsk)
                sig->inblock += task_io_get_inblock(tsk);
                sig->oublock += task_io_get_oublock(tsk);
                task_io_accounting_add(&sig->ioac, &tsk->ioac);
-               sig->sum_sched_runtime += tsk->se.sum_exec_runtime;
                sig = NULL; /* Marker for below. */
        }
 
@@ -143,13 +146,21 @@ static void __exit_signal(struct task_struct *tsk)
        if (sig) {
                flush_sigqueue(&sig->shared_pending);
                taskstats_tgid_free(sig);
+               /*
+                * Make sure ->signal can't go away under rq->lock,
+                * see account_group_exec_runtime().
+                */
+               task_rq_unlock_wait(tsk);
                __cleanup_signal(sig);
        }
 }
 
 static void delayed_put_task_struct(struct rcu_head *rhp)
 {
-       put_task_struct(container_of(rhp, struct task_struct, rcu));
+       struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
+
+       trace_sched_process_free(tsk);
+       put_task_struct(tsk);
 }
 
 
@@ -159,7 +170,10 @@ void release_task(struct task_struct * p)
        int zap_leader;
 repeat:
        tracehook_prepare_release_task(p);
-       atomic_dec(&p->user->processes);
+       /* don't need to get the RCU readlock here - the process is dead and
+        * can't be modifying its own credentials */
+       atomic_dec(&__task_cred(p)->user->processes);
+
        proc_flush_task(p);
        write_lock_irq(&tasklist_lock);
        tracehook_finish_release_task(p);
@@ -334,12 +348,12 @@ static void reparent_to_kthreadd(void)
        /* cpus_allowed? */
        /* rt_priority? */
        /* signals? */
-       security_task_reparent_to_init(current);
        memcpy(current->signal->rlim, init_task.signal->rlim,
               sizeof(current->signal->rlim));
-       atomic_inc(&(INIT_USER->__count));
+
+       atomic_inc(&init_cred.usage);
+       commit_creds(&init_cred);
        write_unlock_irq(&tasklist_lock);
-       switch_uid(INIT_USER);
 }
 
 void __set_special_pids(struct pid *pid)
@@ -640,24 +654,23 @@ retry:
 assign_new_owner:
        BUG_ON(c == p);
        get_task_struct(c);
+       read_unlock(&tasklist_lock);
+       down_write(&mm->mmap_sem);
        /*
         * The task_lock protects c->mm from changing.
         * We always want mm->owner->mm == mm
         */
        task_lock(c);
-       /*
-        * Delay read_unlock() till we have the task_lock()
-        * to ensure that c does not slip away underneath us
-        */
-       read_unlock(&tasklist_lock);
        if (c->mm != mm) {
                task_unlock(c);
+               up_write(&mm->mmap_sem);
                put_task_struct(c);
                goto retry;
        }
        cgroup_mm_owner_callbacks(mm->owner, c);
        mm->owner = c;
        task_unlock(c);
+       up_write(&mm->mmap_sem);
        put_task_struct(c);
 }
 #endif /* CONFIG_MM_OWNER */
@@ -1021,8 +1034,6 @@ NORET_TYPE void do_exit(long code)
                 * task into the wait for ever nirwana as well.
                 */
                tsk->flags |= PF_EXITPIDONE;
-               if (tsk->io_context)
-                       exit_io_context();
                set_current_state(TASK_UNINTERRUPTIBLE);
                schedule();
        }
@@ -1051,14 +1062,6 @@ NORET_TYPE void do_exit(long code)
                exit_itimers(tsk->signal);
        }
        acct_collect(code, group_dead);
-#ifdef CONFIG_FUTEX
-       if (unlikely(tsk->robust_list))
-               exit_robust_list(tsk);
-#ifdef CONFIG_COMPAT
-       if (unlikely(tsk->compat_robust_list))
-               compat_exit_robust_list(tsk);
-#endif
-#endif
        if (group_dead)
                tty_audit_exit();
        if (unlikely(tsk->audit_context))
@@ -1071,13 +1074,14 @@ NORET_TYPE void do_exit(long code)
 
        if (group_dead)
                acct_process();
+       trace_sched_process_exit(tsk);
+
        exit_sem(tsk);
        exit_files(tsk);
        exit_fs(tsk);
        check_stack_usage();
        exit_thread();
        cgroup_exit(tsk, 1);
-       exit_keys(tsk);
 
        if (group_dead && tsk->signal->leader)
                disassociate_ctty(1);
@@ -1122,7 +1126,6 @@ NORET_TYPE void do_exit(long code)
        preempt_disable();
        /* causes final put_task_struct in finish_task_switch(). */
        tsk->state = TASK_DEAD;
-
        schedule();
        BUG();
        /* Avoid "noreturn function does return".  */
@@ -1262,12 +1265,12 @@ static int wait_task_zombie(struct task_struct *p, int options,
        unsigned long state;
        int retval, status, traced;
        pid_t pid = task_pid_vnr(p);
+       uid_t uid = __task_cred(p)->uid;
 
        if (!likely(options & WEXITED))
                return 0;
 
        if (unlikely(options & WNOWAIT)) {
-               uid_t uid = p->uid;
                int exit_code = p->exit_code;
                int why, status;
 
@@ -1299,6 +1302,7 @@ static int wait_task_zombie(struct task_struct *p, int options,
        if (likely(!traced)) {
                struct signal_struct *psig;
                struct signal_struct *sig;
+               struct task_cputime cputime;
 
                /*
                 * The resource counters for the group leader are in its
@@ -1314,20 +1318,23 @@ static int wait_task_zombie(struct task_struct *p, int options,
                 * need to protect the access to p->parent->signal fields,
                 * as other threads in the parent group can be right
                 * here reaping other children at the same time.
+                *
+                * We use thread_group_cputime() to get times for the thread
+                * group, which consolidates times for all threads in the
+                * group including the group leader.
                 */
+               thread_group_cputime(p, &cputime);
                spin_lock_irq(&p->parent->sighand->siglock);
                psig = p->parent->signal;
                sig = p->signal;
                psig->cutime =
                        cputime_add(psig->cutime,
-                       cputime_add(p->utime,
-                       cputime_add(sig->utime,
-                                   sig->cutime)));
+                       cputime_add(cputime.utime,
+                                   sig->cutime));
                psig->cstime =
                        cputime_add(psig->cstime,
-                       cputime_add(p->stime,
-                       cputime_add(sig->stime,
-                                   sig->cstime)));
+                       cputime_add(cputime.stime,
+                                   sig->cstime));
                psig->cgtime =
                        cputime_add(psig->cgtime,
                        cputime_add(p->gtime,
@@ -1384,7 +1391,7 @@ static int wait_task_zombie(struct task_struct *p, int options,
        if (!retval && infop)
                retval = put_user(pid, &infop->si_pid);
        if (!retval && infop)
-               retval = put_user(p->uid, &infop->si_uid);
+               retval = put_user(uid, &infop->si_uid);
        if (!retval)
                retval = pid;
 
@@ -1449,7 +1456,8 @@ static int wait_task_stopped(int ptrace, struct task_struct *p,
        if (!unlikely(options & WNOWAIT))
                p->exit_code = 0;
 
-       uid = p->uid;
+       /* don't need the RCU readlock here as we're holding a spinlock */
+       uid = __task_cred(p)->uid;
 unlock_sig:
        spin_unlock_irq(&p->sighand->siglock);
        if (!exit_code)
@@ -1523,10 +1531,10 @@ static int wait_task_continued(struct task_struct *p, int options,
        }
        if (!unlikely(options & WNOWAIT))
                p->signal->flags &= ~SIGNAL_STOP_CONTINUED;
+       uid = __task_cred(p)->uid;
        spin_unlock_irq(&p->sighand->siglock);
 
        pid = task_pid_vnr(p);
-       uid = p->uid;
        get_task_struct(p);
        read_unlock(&tasklist_lock);
 
@@ -1672,6 +1680,8 @@ static long do_wait(enum pid_type type, struct pid *pid, int options,
        struct task_struct *tsk;
        int retval;
 
+       trace_sched_process_wait(pid);
+
        add_wait_queue(&current->signal->wait_chldexit,&wait);
 repeat:
        /*