]> nv-tegra.nvidia Code Review - linux-2.6.git/blobdiff - kernel/fork.c
Merge branch 'linus' into perfcounters/core
[linux-2.6.git] / kernel / fork.c
index a66fbde20715bb2d93180b7b3d03541d31d9c068..4430eb1376f257bd008dab17d3c6ea19cc9c2d3b 100644 (file)
@@ -60,7 +60,9 @@
 #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>
@@ -69,6 +71,8 @@
 #include <asm/cacheflush.h>
 #include <asm/tlbflush.h>
 
+#include <trace/events/sched.h>
+
 /*
  * Protected counters by write_lock_irq(&tasklist_lock)
  */
@@ -81,8 +85,6 @@ 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;
@@ -212,6 +214,8 @@ static struct task_struct *dup_task_struct(struct task_struct *orig)
 {
        struct task_struct *tsk;
        struct thread_info *ti;
+       unsigned long *stackend;
+
        int err;
 
        prepare_to_copy(orig);
@@ -237,6 +241,8 @@ static struct task_struct *dup_task_struct(struct task_struct *orig)
                goto out;
 
        setup_thread_stack(tsk, orig);
+       stackend = end_of_stack(tsk);
+       *stackend = STACK_END_MAGIC;    /* for overflow detection */
 
 #ifdef CONFIG_CC_STACKPROTECTOR
        tsk->stack_canary = get_random_int();
@@ -279,7 +285,7 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
        mm->free_area_cache = oldmm->mmap_base;
        mm->cached_hole_size = ~0UL;
        mm->map_count = 0;
-       cpus_clear(mm->cpu_vm_mask);
+       cpumask_clear(mm_cpumask(mm));
        mm->mm_rb = RB_ROOT;
        rb_link = &mm->mm_rb.rb_node;
        rb_parent = NULL;
@@ -639,6 +645,9 @@ 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;
@@ -676,38 +685,21 @@ 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;
@@ -808,6 +800,12 @@ 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]);
@@ -823,11 +821,8 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
                atomic_inc(&current->signal->live);
                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;
@@ -836,6 +831,8 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
        atomic_set(&sig->live, 1);
        init_waitqueue_head(&sig->wait_chldexit);
        sig->flags = 0;
+       if (clone_flags & CLONE_NEWPID)
+               sig->flags |= SIGNAL_UNKILLABLE;
        sig->group_exit_code = 0;
        sig->group_exit_task = NULL;
        sig->group_stop_count = 0;
@@ -865,6 +862,8 @@ 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);
@@ -983,6 +982,8 @@ 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
@@ -1041,11 +1042,6 @@ 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);
 
@@ -1095,12 +1091,16 @@ 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 */
@@ -1120,7 +1120,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
                goto bad_fork_cleanup_mm;
        if ((retval = copy_io(clone_flags, p)))
                goto bad_fork_cleanup_namespaces;
-       retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs);
+       retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
        if (retval)
                goto bad_fork_cleanup_io;
 
@@ -1137,8 +1137,6 @@ 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)
@@ -1147,7 +1145,7 @@ 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;
@@ -1179,10 +1177,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 #endif
        clear_all_latency_tracing(p);
 
-       /* Our parent execution domain becomes current domain
-          These must match for thread signalling to apply */
-       p->parent_exec_id = p->self_exec_id;
-
        /* ok, now we should be set up.. */
        p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
        p->pdeath_signal = 0;
@@ -1220,10 +1214,13 @@ static struct task_struct *copy_process(unsigned long clone_flags,
                set_task_cpu(p, smp_processor_id());
 
        /* CLONE_PARENT re-uses the old parent */
-       if (clone_flags & (CLONE_PARENT|CLONE_THREAD))
+       if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
                p->real_parent = current->real_parent;
-       else
+               p->parent_exec_id = current->parent_exec_id;
+       } else {
                p->real_parent = current;
+               p->parent_exec_id = current->self_exec_id;
+       }
 
        spin_lock(&current->sighand->siglock);
 
@@ -1240,7 +1237,7 @@ 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) {
@@ -1259,8 +1256,6 @@ 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);
@@ -1277,8 +1272,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
        cgroup_post_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);
@@ -1302,6 +1295,7 @@ 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:
@@ -1415,10 +1409,16 @@ long do_fork(unsigned long clone_flags,
                if (clone_flags & CLONE_VFORK) {
                        p->vfork_done = &vfork;
                        init_completion(&vfork);
+               } else if (!(clone_flags & CLONE_VM)) {
+                       /*
+                        * vfork will do an exec which will call
+                        * set_task_comm()
+                        */
+                       perf_counter_fork(p);
                }
 
                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
@@ -1484,6 +1484,7 @@ void __init proc_caches_init(void)
        mm_cachep = kmem_cache_create("mm_struct",
                        sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
                        SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
+       vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
        mmap_init();
 }
 
@@ -1539,12 +1540,16 @@ 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;
 }
@@ -1660,8 +1665,13 @@ 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) {
@@ -1700,7 +1710,7 @@ 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: