[IA64] pvops: add to hooks, pv_time_ops, for steal time accounting.
[pandora-kernel.git] / kernel / fork.c
index 6067e42..19908b2 100644 (file)
@@ -22,6 +22,7 @@
 #include <linux/mempolicy.h>
 #include <linux/sem.h>
 #include <linux/file.h>
+#include <linux/fdtable.h>
 #include <linux/key.h>
 #include <linux/binfmts.h>
 #include <linux/mman.h>
@@ -381,14 +382,13 @@ static struct mm_struct * mm_init(struct mm_struct * mm, struct task_struct *p)
        mm->ioctx_list = NULL;
        mm->free_area_cache = TASK_UNMAPPED_BASE;
        mm->cached_hole_size = ~0UL;
-       mm_init_cgroup(mm, p);
+       mm_init_owner(mm, p);
 
        if (likely(!mm_alloc_pgd(mm))) {
                mm->def_flags = 0;
                return mm;
        }
 
-       mm_free_cgroup(mm);
        free_mm(mm);
        return NULL;
 }
@@ -432,13 +432,13 @@ void mmput(struct mm_struct *mm)
        if (atomic_dec_and_test(&mm->mm_users)) {
                exit_aio(mm);
                exit_mmap(mm);
+               set_mm_exe_file(mm, NULL);
                if (!list_empty(&mm->mmlist)) {
                        spin_lock(&mmlist_lock);
                        list_del(&mm->mmlist);
                        spin_unlock(&mmlist_lock);
                }
                put_swap_token(mm);
-               mm_free_cgroup(mm);
                mmdrop(mm);
        }
 }
@@ -545,6 +545,8 @@ struct mm_struct *dup_mm(struct task_struct *tsk)
        if (init_new_context(tsk, mm))
                goto fail_nocontext;
 
+       dup_mm_exe_file(oldmm, mm);
+
        err = dup_mmap(mm, oldmm);
        if (err)
                goto free_pt;
@@ -658,136 +660,6 @@ static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
        return 0;
 }
 
-static int count_open_files(struct fdtable *fdt)
-{
-       int size = fdt->max_fds;
-       int i;
-
-       /* Find the last open fd */
-       for (i = size/(8*sizeof(long)); i > 0; ) {
-               if (fdt->open_fds->fds_bits[--i])
-                       break;
-       }
-       i = (i+1) * 8 * sizeof(long);
-       return i;
-}
-
-static struct files_struct *alloc_files(void)
-{
-       struct files_struct *newf;
-       struct fdtable *fdt;
-
-       newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
-       if (!newf)
-               goto out;
-
-       atomic_set(&newf->count, 1);
-
-       spin_lock_init(&newf->file_lock);
-       newf->next_fd = 0;
-       fdt = &newf->fdtab;
-       fdt->max_fds = NR_OPEN_DEFAULT;
-       fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
-       fdt->open_fds = (fd_set *)&newf->open_fds_init;
-       fdt->fd = &newf->fd_array[0];
-       INIT_RCU_HEAD(&fdt->rcu);
-       fdt->next = NULL;
-       rcu_assign_pointer(newf->fdt, fdt);
-out:
-       return newf;
-}
-
-/*
- * Allocate a new files structure and copy contents from the
- * passed in files structure.
- * errorp will be valid only when the returned files_struct is NULL.
- */
-static struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
-{
-       struct files_struct *newf;
-       struct file **old_fds, **new_fds;
-       int open_files, size, i;
-       struct fdtable *old_fdt, *new_fdt;
-
-       *errorp = -ENOMEM;
-       newf = alloc_files();
-       if (!newf)
-               goto out;
-
-       spin_lock(&oldf->file_lock);
-       old_fdt = files_fdtable(oldf);
-       new_fdt = files_fdtable(newf);
-       open_files = count_open_files(old_fdt);
-
-       /*
-        * Check whether we need to allocate a larger fd array and fd set.
-        * Note: we're not a clone task, so the open count won't change.
-        */
-       if (open_files > new_fdt->max_fds) {
-               new_fdt->max_fds = 0;
-               spin_unlock(&oldf->file_lock);
-               spin_lock(&newf->file_lock);
-               *errorp = expand_files(newf, open_files-1);
-               spin_unlock(&newf->file_lock);
-               if (*errorp < 0)
-                       goto out_release;
-               new_fdt = files_fdtable(newf);
-               /*
-                * Reacquire the oldf lock and a pointer to its fd table
-                * who knows it may have a new bigger fd table. We need
-                * the latest pointer.
-                */
-               spin_lock(&oldf->file_lock);
-               old_fdt = files_fdtable(oldf);
-       }
-
-       old_fds = old_fdt->fd;
-       new_fds = new_fdt->fd;
-
-       memcpy(new_fdt->open_fds->fds_bits,
-               old_fdt->open_fds->fds_bits, open_files/8);
-       memcpy(new_fdt->close_on_exec->fds_bits,
-               old_fdt->close_on_exec->fds_bits, open_files/8);
-
-       for (i = open_files; i != 0; i--) {
-               struct file *f = *old_fds++;
-               if (f) {
-                       get_file(f);
-               } else {
-                       /*
-                        * The fd may be claimed in the fd bitmap but not yet
-                        * instantiated in the files array if a sibling thread
-                        * is partway through open().  So make sure that this
-                        * fd is available to the new process.
-                        */
-                       FD_CLR(open_files - i, new_fdt->open_fds);
-               }
-               rcu_assign_pointer(*new_fds++, f);
-       }
-       spin_unlock(&oldf->file_lock);
-
-       /* compute the remainder to be cleared */
-       size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
-
-       /* This is long word aligned thus could use a optimized version */
-       memset(new_fds, 0, size);
-
-       if (new_fdt->max_fds > open_files) {
-               int left = (new_fdt->max_fds-open_files)/8;
-               int start = open_files / (8 * sizeof(unsigned long));
-
-               memset(&new_fdt->open_fds->fds_bits[start], 0, left);
-               memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
-       }
-
-       return newf;
-
-out_release:
-       kmem_cache_free(files_cachep, newf);
-out:
-       return NULL;
-}
-
 static int copy_files(unsigned long clone_flags, struct task_struct * tsk)
 {
        struct files_struct *oldf, *newf;
@@ -891,7 +763,7 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
        sig->group_exit_code = 0;
        sig->group_exit_task = NULL;
        sig->group_stop_count = 0;
-       sig->curr_target = NULL;
+       sig->curr_target = tsk;
        init_sigpending(&sig->shared_pending);
        INIT_LIST_HEAD(&sig->posix_timers);
 
@@ -982,6 +854,13 @@ static void rt_mutex_init_task(struct task_struct *p)
 #endif
 }
 
+#ifdef CONFIG_MM_OWNER
+void mm_init_owner(struct mm_struct *mm, struct task_struct *p)
+{
+       mm->owner = p;
+}
+#endif /* CONFIG_MM_OWNER */
+
 /*
  * This creates a new process as a copy of the old one,
  * but does not actually start it yet.
@@ -1663,18 +1542,6 @@ static int unshare_fd(unsigned long unshare_flags, struct files_struct **new_fdp
        return 0;
 }
 
-/*
- * Unsharing of semundo for tasks created with CLONE_SYSVSEM is not
- * supported yet
- */
-static int unshare_semundo(unsigned long unshare_flags, struct sem_undo_list **new_ulistp)
-{
-       if (unshare_flags & CLONE_SYSVSEM)
-               return -EINVAL;
-
-       return 0;
-}
-
 /*
  * unshare allows a process to 'unshare' part of the process
  * context which was originally shared using clone.  copy_*
@@ -1690,8 +1557,8 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
        struct sighand_struct *new_sigh = NULL;
        struct mm_struct *mm, *new_mm = NULL, *active_mm = NULL;
        struct files_struct *fd, *new_fd = NULL;
-       struct sem_undo_list *new_ulist = NULL;
        struct nsproxy *new_nsproxy = NULL;
+       int do_sysvsem = 0;
 
        check_unshare_flags(&unshare_flags);
 
@@ -1703,6 +1570,13 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
                                CLONE_NEWNET))
                goto bad_unshare_out;
 
+       /*
+        * CLONE_NEWIPC must also detach from the undolist: after switching
+        * to a new ipc namespace, the semaphore arrays from the old
+        * namespace are unreachable.
+        */
+       if (unshare_flags & (CLONE_NEWIPC|CLONE_SYSVSEM))
+               do_sysvsem = 1;
        if ((err = unshare_thread(unshare_flags)))
                goto bad_unshare_out;
        if ((err = unshare_fs(unshare_flags, &new_fs)))
@@ -1713,13 +1587,17 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
                goto bad_unshare_cleanup_sigh;
        if ((err = unshare_fd(unshare_flags, &new_fd)))
                goto bad_unshare_cleanup_vm;
-       if ((err = unshare_semundo(unshare_flags, &new_ulist)))
-               goto bad_unshare_cleanup_fd;
        if ((err = unshare_nsproxy_namespaces(unshare_flags, &new_nsproxy,
                        new_fs)))
-               goto bad_unshare_cleanup_semundo;
+               goto bad_unshare_cleanup_fd;
 
-       if (new_fs ||  new_mm || new_fd || new_ulist || new_nsproxy) {
+       if (new_fs ||  new_mm || new_fd || do_sysvsem || new_nsproxy) {
+               if (do_sysvsem) {
+                       /*
+                        * CLONE_SYSVSEM is equivalent to sys_exit().
+                        */
+                       exit_sem(current);
+               }
 
                if (new_nsproxy) {
                        switch_task_namespaces(current, new_nsproxy);
@@ -1755,7 +1633,6 @@ asmlinkage long sys_unshare(unsigned long unshare_flags)
        if (new_nsproxy)
                put_nsproxy(new_nsproxy);
 
-bad_unshare_cleanup_semundo:
 bad_unshare_cleanup_fd:
        if (new_fd)
                put_files_struct(new_fd);