Merge branch 'tracing-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[safe/jmp/linux-2.6] / kernel / exit.c
index 25ed2ad..f80dec3 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);
 
@@ -112,9 +118,7 @@ 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, tsk->utime);
-               sig->stime = cputime_add(sig->stime, tsk->stime);
-               sig->gtime = cputime_add(sig->gtime, tsk->gtime);
+               sig->gtime = cputime_add(sig->gtime, task_gtime(tsk));
                sig->min_flt += tsk->min_flt;
                sig->maj_flt += tsk->maj_flt;
                sig->nvcsw += tsk->nvcsw;
@@ -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)
@@ -583,8 +597,6 @@ mm_need_new_owner(struct mm_struct *mm, struct task_struct *p)
         * If there are other users of the mm and the owner (us) is exiting
         * we need to find a new owner to take on the responsibility.
         */
-       if (!mm)
-               return 0;
        if (atomic_read(&mm->mm_users) <= 1)
                return 0;
        if (mm->owner != p)
@@ -627,6 +639,12 @@ retry:
        } while_each_thread(g, c);
 
        read_unlock(&tasklist_lock);
+       /*
+        * We found no owner yet mm_users > 1: this implies that we are
+        * most likely racing with swapoff (try_to_unuse()) or /proc or
+        * ptrace or page migration (get_task_mm()).  Mark owner as NULL.
+        */
+       mm->owner = NULL;
        return;
 
 assign_new_owner:
@@ -647,7 +665,6 @@ assign_new_owner:
                put_task_struct(c);
                goto retry;
        }
-       cgroup_mm_owner_callbacks(mm->owner, c);
        mm->owner = c;
        task_unlock(c);
        put_task_struct(c);
@@ -1016,8 +1033,6 @@ 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();
        }
@@ -1036,24 +1051,13 @@ NORET_TYPE void do_exit(long code)
                                preempt_count());
 
        acct_update_integrals(tsk);
-       if (tsk->mm) {
-               update_hiwater_rss(tsk->mm);
-               update_hiwater_vm(tsk->mm);
-       }
+
        group_dead = atomic_dec_and_test(&tsk->signal->live);
        if (group_dead) {
                hrtimer_cancel(&tsk->signal->real_timer);
                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))
@@ -1066,13 +1070,14 @@ 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);
@@ -1117,7 +1122,6 @@ 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".  */
@@ -1137,7 +1141,7 @@ NORET_TYPE void complete_and_exit(struct completion *comp, long code)
 
 EXPORT_SYMBOL(complete_and_exit);
 
-asmlinkage long sys_exit(int error_code)
+SYSCALL_DEFINE1(exit, int, error_code)
 {
        do_exit((error_code&0xff)<<8);
 }
@@ -1178,9 +1182,11 @@ do_group_exit(int exit_code)
  * wait4()-ing process will get the correct exit code - even if this
  * thread is not the thread group leader.
  */
-asmlinkage void sys_exit_group(int error_code)
+SYSCALL_DEFINE1(exit_group, int, error_code)
 {
        do_group_exit((error_code & 0xff) << 8);
+       /* NOTREACHED */
+       return 0;
 }
 
 static struct pid *task_pid_type(struct task_struct *task, enum pid_type type)
@@ -1257,12 +1263,12 @@ 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;
 
@@ -1294,6 +1300,7 @@ 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
@@ -1309,20 +1316,23 @@ 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,
@@ -1379,7 +1389,7 @@ 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;
 
@@ -1444,7 +1454,8 @@ 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)
@@ -1518,10 +1529,10 @@ 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);
 
@@ -1667,6 +1678,8 @@ 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:
        /*
@@ -1741,9 +1754,8 @@ end:
        return retval;
 }
 
-asmlinkage long sys_waitid(int which, pid_t upid,
-                          struct siginfo __user *infop, int options,
-                          struct rusage __user *ru)
+SYSCALL_DEFINE5(waitid, int, which, pid_t, upid, struct siginfo __user *,
+               infop, int, options, struct rusage __user *, ru)
 {
        struct pid *pid = NULL;
        enum pid_type type;
@@ -1782,8 +1794,8 @@ asmlinkage long sys_waitid(int which, pid_t upid,
        return ret;
 }
 
-asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
-                         int options, struct rusage __user *ru)
+SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
+               int, options, struct rusage __user *, ru)
 {
        struct pid *pid = NULL;
        enum pid_type type;
@@ -1820,7 +1832,7 @@ asmlinkage long sys_wait4(pid_t upid, int __user *stat_addr,
  * sys_waitpid() remains for compatibility. waitpid() should be
  * implemented by calling sys_wait4() from libc.a.
  */
-asmlinkage long sys_waitpid(pid_t pid, int __user *stat_addr, int options)
+SYSCALL_DEFINE3(waitpid, pid_t, pid, int __user *, stat_addr, int, options)
 {
        return sys_wait4(pid, stat_addr, options, NULL);
 }