Merge branch 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[safe/jmp/linux-2.6] / kernel / fork.c
index adbea16..bfee931 100644 (file)
@@ -17,7 +17,6 @@
 #include <linux/module.h>
 #include <linux/vmalloc.h>
 #include <linux/completion.h>
-#include <linux/mnt_namespace.h>
 #include <linux/personality.h>
 #include <linux/mempolicy.h>
 #include <linux/sem.h>
@@ -60,8 +59,9 @@
 #include <linux/tty.h>
 #include <linux/proc_fs.h>
 #include <linux/blkdev.h>
-#include <trace/sched.h>
+#include <linux/fs_struct.h>
 #include <linux/magic.h>
+#include <linux/perf_counter.h>
 
 #include <asm/pgtable.h>
 #include <asm/pgalloc.h>
@@ -70,6 +70,8 @@
 #include <asm/cacheflush.h>
 #include <asm/tlbflush.h>
 
+#include <trace/events/sched.h>
+
 /*
  * Protected counters by write_lock_irq(&tasklist_lock)
  */
@@ -82,8 +84,6 @@ DEFINE_PER_CPU(unsigned long, process_counts) = 0;
 
 __cacheline_aligned DEFINE_RWLOCK(tasklist_lock);  /* outer */
 
-DEFINE_TRACE(sched_process_fork);
-
 int nr_processes(void)
 {
        int cpu;
@@ -152,8 +152,7 @@ void __put_task_struct(struct task_struct *tsk)
        WARN_ON(atomic_read(&tsk->usage));
        WARN_ON(tsk == current);
 
-       put_cred(tsk->real_cred);
-       put_cred(tsk->cred);
+       exit_creds(tsk);
        delayacct_tsk_free(tsk);
 
        if (!profile_handoff_task(tsk))
@@ -177,7 +176,7 @@ void __init fork_init(unsigned long mempages)
        /* create a slab on which task_structs can be allocated */
        task_struct_cachep =
                kmem_cache_create("task_struct", sizeof(struct task_struct),
-                       ARCH_MIN_TASKALIGN, SLAB_PANIC, NULL);
+                       ARCH_MIN_TASKALIGN, SLAB_PANIC | SLAB_NOTRACK, NULL);
 #endif
 
        /* do the arch specific task caches init */
@@ -567,18 +566,18 @@ void mm_release(struct task_struct *tsk, struct mm_struct *mm)
         * the value intact in a core dump, and to save the unnecessary
         * trouble otherwise.  Userland only wants this done for a sys_exit.
         */
-       if (tsk->clear_child_tid
-           && !(tsk->flags & PF_SIGNALED)
-           && atomic_read(&mm->mm_users) > 1) {
-               u32 __user * tidptr = tsk->clear_child_tid;
+       if (tsk->clear_child_tid) {
+               if (!(tsk->flags & PF_SIGNALED) &&
+                   atomic_read(&mm->mm_users) > 1) {
+                       /*
+                        * We don't check the error code - if userspace has
+                        * not set up a proper pointer then tough luck.
+                        */
+                       put_user(0, tsk->clear_child_tid);
+                       sys_futex(tsk->clear_child_tid, FUTEX_WAKE,
+                                       1, NULL, NULL, 0);
+               }
                tsk->clear_child_tid = NULL;
-
-               /*
-                * We don't check the error code - if userspace has
-                * not set up a proper pointer then tough luck.
-                */
-               put_user(0, tidptr);
-               sys_futex(tidptr, FUTEX_WAKE, 1, NULL, NULL, 0);
        }
 }
 
@@ -644,6 +643,9 @@ static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
 
        tsk->min_flt = tsk->maj_flt = 0;
        tsk->nvcsw = tsk->nivcsw = 0;
+#ifdef CONFIG_DETECT_HUNG_TASK
+       tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
+#endif
 
        tsk->mm = NULL;
        tsk->active_mm = NULL;
@@ -681,38 +683,21 @@ fail_nomem:
        return retval;
 }
 
-static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
-{
-       struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
-       /* We don't need to lock fs - think why ;-) */
-       if (fs) {
-               atomic_set(&fs->count, 1);
-               rwlock_init(&fs->lock);
-               fs->umask = old->umask;
-               read_lock(&old->lock);
-               fs->root = old->root;
-               path_get(&old->root);
-               fs->pwd = old->pwd;
-               path_get(&old->pwd);
-               read_unlock(&old->lock);
-       }
-       return fs;
-}
-
-struct fs_struct *copy_fs_struct(struct fs_struct *old)
-{
-       return __copy_fs_struct(old);
-}
-
-EXPORT_SYMBOL_GPL(copy_fs_struct);
-
 static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
 {
+       struct fs_struct *fs = current->fs;
        if (clone_flags & CLONE_FS) {
-               atomic_inc(&current->fs->count);
+               /* tsk->fs is already what we want */
+               write_lock(&fs->lock);
+               if (fs->in_exec) {
+                       write_unlock(&fs->lock);
+                       return -EAGAIN;
+               }
+               fs->users++;
+               write_unlock(&fs->lock);
                return 0;
        }
-       tsk->fs = __copy_fs_struct(current->fs);
+       tsk->fs = copy_fs_struct(fs);
        if (!tsk->fs)
                return -ENOMEM;
        return 0;
@@ -813,6 +798,12 @@ static void posix_cpu_timers_init_group(struct signal_struct *sig)
        sig->cputime_expires.virt_exp = cputime_zero;
        sig->cputime_expires.sched_exp = 0;
 
+       if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
+               sig->cputime_expires.prof_exp =
+                       secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
+               sig->cputimer.running = 1;
+       }
+
        /* The timer lists. */
        INIT_LIST_HEAD(&sig->cpu_timers[0]);
        INIT_LIST_HEAD(&sig->cpu_timers[1]);
@@ -823,16 +814,10 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
 {
        struct signal_struct *sig;
 
-       if (clone_flags & CLONE_THREAD) {
-               atomic_inc(&current->signal->count);
-               atomic_inc(&current->signal->live);
+       if (clone_flags & CLONE_THREAD)
                return 0;
-       }
-       sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
-
-       if (sig)
-               posix_cpu_timers_init_group(sig);
 
+       sig = kmem_cache_alloc(signal_cachep, GFP_KERNEL);
        tsk->signal = sig;
        if (!sig)
                return -ENOMEM;
@@ -872,6 +857,8 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
        memcpy(sig->rlim, current->signal->rlim, sizeof sig->rlim);
        task_unlock(current->group_leader);
 
+       posix_cpu_timers_init_group(sig);
+
        acct_init_pacct(&sig->pacct);
 
        tty_audit_fork(sig);
@@ -886,16 +873,6 @@ void __cleanup_signal(struct signal_struct *sig)
        kmem_cache_free(signal_cachep, sig);
 }
 
-static void cleanup_signal(struct task_struct *tsk)
-{
-       struct signal_struct *sig = tsk->signal;
-
-       atomic_dec(&sig->live);
-
-       if (atomic_dec_and_test(&sig->count))
-               __cleanup_signal(sig);
-}
-
 static void copy_flags(unsigned long clone_flags, struct task_struct *p)
 {
        unsigned long new_flags = p->flags;
@@ -990,6 +967,8 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        if (!p)
                goto fork_out;
 
+       ftrace_graph_init_task(p);
+
        rt_mutex_init_task(p);
 
 #ifdef CONFIG_PROVE_LOCKING
@@ -1028,14 +1007,10 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        copy_flags(clone_flags, p);
        INIT_LIST_HEAD(&p->children);
        INIT_LIST_HEAD(&p->sibling);
-#ifdef CONFIG_PREEMPT_RCU
-       p->rcu_read_lock_nesting = 0;
-       p->rcu_flipctr_idx = 0;
-#endif /* #ifdef CONFIG_PREEMPT_RCU */
+       rcu_copy_process(p);
        p->vfork_done = NULL;
        spin_lock_init(&p->alloc_lock);
 
-       clear_tsk_thread_flag(p, TIF_SIGPENDING);
        init_sigpending(&p->pending);
 
        p->utime = cputime_zero;
@@ -1048,11 +1023,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 
        p->default_timer_slack_ns = current->timer_slack_ns;
 
-#ifdef CONFIG_DETECT_SOFTLOCKUP
-       p->last_switch_count = 0;
-       p->last_switch_timestamp = 0;
-#endif
-
        task_io_accounting_init(&p->ioac);
        acct_clear_integrals(p);
 
@@ -1102,12 +1072,16 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 #ifdef CONFIG_DEBUG_MUTEXES
        p->blocked_on = NULL; /* not blocked yet */
 #endif
-       if (unlikely(current->ptrace))
-               ptrace_fork(p, clone_flags);
+
+       p->bts = NULL;
 
        /* Perform scheduler related setup. Assign this task to a CPU. */
        sched_fork(p, clone_flags);
 
+       retval = perf_counter_init_task(p);
+       if (retval)
+               goto bad_fork_cleanup_policy;
+
        if ((retval = audit_alloc(p)))
                goto bad_fork_cleanup_policy;
        /* copy all the process information */
@@ -1144,8 +1118,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
                }
        }
 
-       ftrace_graph_init_task(p);
-
        p->pid = pid_nr(pid);
        p->tgid = p->pid;
        if (clone_flags & CLONE_THREAD)
@@ -1154,7 +1126,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        if (current->nsproxy != p->nsproxy) {
                retval = ns_cgroup_clone(p, pid);
                if (retval)
-                       goto bad_fork_free_graph;
+                       goto bad_fork_free_pid;
        }
 
        p->set_child_tid = (clone_flags & CLONE_CHILD_SETTID) ? child_tidptr : NULL;
@@ -1246,10 +1218,12 @@ static struct task_struct *copy_process(unsigned long clone_flags,
                spin_unlock(&current->sighand->siglock);
                write_unlock_irq(&tasklist_lock);
                retval = -ERESTARTNOINTR;
-               goto bad_fork_free_graph;
+               goto bad_fork_free_pid;
        }
 
        if (clone_flags & CLONE_THREAD) {
+               atomic_inc(&current->signal->count);
+               atomic_inc(&current->signal->live);
                p->group_leader = current->group_leader;
                list_add_tail_rcu(&p->thread_group, &p->group_leader->thread_group);
        }
@@ -1265,8 +1239,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
                        p->signal->leader_pid = pid;
                        tty_kref_put(p->signal->tty);
                        p->signal->tty = tty_kref_get(current->signal->tty);
-                       set_task_pgrp(p, task_pgrp_nr(current));
-                       set_task_session(p, task_session_nr(current));
                        attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
                        attach_pid(p, PIDTYPE_SID, task_session(current));
                        list_add_tail_rcu(&p->tasks, &init_task.tasks);
@@ -1281,10 +1253,9 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        write_unlock_irq(&tasklist_lock);
        proc_fork_connector(p);
        cgroup_post_fork(p);
+       perf_counter_fork(p);
        return p;
 
-bad_fork_free_graph:
-       ftrace_graph_exit_task(p);
 bad_fork_free_pid:
        if (pid != &init_struct_pid)
                free_pid(pid);
@@ -1296,7 +1267,8 @@ bad_fork_cleanup_mm:
        if (p->mm)
                mmput(p->mm);
 bad_fork_cleanup_signal:
-       cleanup_signal(p);
+       if (!(clone_flags & CLONE_THREAD))
+               __cleanup_signal(p->signal);
 bad_fork_cleanup_sighand:
        __cleanup_sighand(p->sighand);
 bad_fork_cleanup_fs:
@@ -1308,6 +1280,7 @@ bad_fork_cleanup_semundo:
 bad_fork_cleanup_audit:
        audit_free(p);
 bad_fork_cleanup_policy:
+       perf_counter_free_task(p);
 #ifdef CONFIG_NUMA
        mpol_put(p->mempolicy);
 bad_fork_cleanup_cgroup:
@@ -1320,8 +1293,7 @@ bad_fork_cleanup_put_domain:
        module_put(task_thread_info(p)->exec_domain->module);
 bad_fork_cleanup_count:
        atomic_dec(&p->cred->user->processes);
-       put_cred(p->real_cred);
-       put_cred(p->cred);
+       exit_creds(p);
 bad_fork_free:
        free_task(p);
 fork_out:
@@ -1424,7 +1396,7 @@ long do_fork(unsigned long clone_flags,
                }
 
                audit_finish_fork(p);
-               tracehook_report_clone(trace, regs, clone_flags, nr, p);
+               tracehook_report_clone(regs, clone_flags, nr, p);
 
                /*
                 * We set PF_STARTING at creation in case tracing wants to
@@ -1476,20 +1448,20 @@ void __init proc_caches_init(void)
 {
        sighand_cachep = kmem_cache_create("sighand_cache",
                        sizeof(struct sighand_struct), 0,
-                       SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU,
-                       sighand_ctor);
+                       SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_DESTROY_BY_RCU|
+                       SLAB_NOTRACK, sighand_ctor);
        signal_cachep = kmem_cache_create("signal_cache",
                        sizeof(struct signal_struct), 0,
-                       SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
+                       SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
        files_cachep = kmem_cache_create("files_cache",
                        sizeof(struct files_struct), 0,
-                       SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
+                       SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
        fs_cachep = kmem_cache_create("fs_cache",
                        sizeof(struct fs_struct), 0,
-                       SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
+                       SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
        mm_cachep = kmem_cache_create("mm_struct",
                        sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
-                       SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
+                       SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_NOTRACK, NULL);
        vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
        mmap_init();
 }
@@ -1546,12 +1518,16 @@ static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
 {
        struct fs_struct *fs = current->fs;
 
-       if ((unshare_flags & CLONE_FS) &&
-           (fs && atomic_read(&fs->count) > 1)) {
-               *new_fsp = __copy_fs_struct(current->fs);
-               if (!*new_fsp)
-                       return -ENOMEM;
-       }
+       if (!(unshare_flags & CLONE_FS) || !fs)
+               return 0;
+
+       /* don't need lock here; in the worst case we'll do useless copy */
+       if (fs->users == 1)
+               return 0;
+
+       *new_fsp = copy_fs_struct(fs);
+       if (!*new_fsp)
+               return -ENOMEM;
 
        return 0;
 }
@@ -1667,8 +1643,13 @@ SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
 
                if (new_fs) {
                        fs = current->fs;
+                       write_lock(&fs->lock);
                        current->fs = new_fs;
-                       new_fs = fs;
+                       if (--fs->users)
+                               new_fs = NULL;
+                       else
+                               new_fs = fs;
+                       write_unlock(&fs->lock);
                }
 
                if (new_mm) {
@@ -1707,7 +1688,7 @@ bad_unshare_cleanup_sigh:
 
 bad_unshare_cleanup_fs:
        if (new_fs)
-               put_fs_struct(new_fs);
+               free_fs_struct(new_fs);
 
 bad_unshare_cleanup_thread:
 bad_unshare_out: