vfs: rename 'do_follow_link' to 'should_follow_link'
[linux-2.6.git] / fs / file.c
index 34bb7f7..4c6992d 100644 (file)
--- a/fs/file.c
+++ b/fs/file.c
@@ -9,6 +9,7 @@
 #include <linux/module.h>
 #include <linux/fs.h>
 #include <linux/mm.h>
+#include <linux/mmzone.h>
 #include <linux/time.h>
 #include <linux/sched.h>
 #include <linux/slab.h>
@@ -39,28 +40,30 @@ int sysctl_nr_open_max = 1024 * 1024; /* raised later */
  */
 static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
 
-static inline void * alloc_fdmem(unsigned int size)
+static void *alloc_fdmem(unsigned int size)
 {
-       if (size <= PAGE_SIZE)
-               return kmalloc(size, GFP_KERNEL);
-       else
-               return vmalloc(size);
+       /*
+        * Very large allocations can stress page reclaim, so fall back to
+        * vmalloc() if the allocation size will be considered "large" by the VM.
+        */
+       if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
+               void *data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN);
+               if (data != NULL)
+                       return data;
+       }
+       return vmalloc(size);
 }
 
-static inline void free_fdarr(struct fdtable *fdt)
+static void free_fdmem(void *ptr)
 {
-       if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
-               kfree(fdt->fd);
-       else
-               vfree(fdt->fd);
+       is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr);
 }
 
-static inline void free_fdset(struct fdtable *fdt)
+static void __free_fdtable(struct fdtable *fdt)
 {
-       if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
-               kfree(fdt->open_fds);
-       else
-               vfree(fdt->open_fds);
+       free_fdmem(fdt->fd);
+       free_fdmem(fdt->open_fds);
+       kfree(fdt);
 }
 
 static void free_fdtable_work(struct work_struct *work)
@@ -75,9 +78,8 @@ static void free_fdtable_work(struct work_struct *work)
        spin_unlock_bh(&f->lock);
        while(fdt) {
                struct fdtable *next = fdt->next;
-               vfree(fdt->fd);
-               free_fdset(fdt);
-               kfree(fdt);
+
+               __free_fdtable(fdt);
                fdt = next;
        }
 }
@@ -98,7 +100,7 @@ void free_fdtable_rcu(struct rcu_head *rcu)
                                container_of(fdt, struct files_struct, fdtab));
                return;
        }
-       if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
+       if (!is_vmalloc_addr(fdt->fd) && !is_vmalloc_addr(fdt->open_fds)) {
                kfree(fdt->fd);
                kfree(fdt->open_fds);
                kfree(fdt);
@@ -178,13 +180,12 @@ static struct fdtable * alloc_fdtable(unsigned int nr)
        fdt->open_fds = (fd_set *)data;
        data += nr / BITS_PER_BYTE;
        fdt->close_on_exec = (fd_set *)data;
-       INIT_RCU_HEAD(&fdt->rcu);
        fdt->next = NULL;
 
        return fdt;
 
 out_arr:
-       free_fdarr(fdt);
+       free_fdmem(fdt->fd);
 out_fdt:
        kfree(fdt);
 out:
@@ -214,9 +215,7 @@ static int expand_fdtable(struct files_struct *files, int nr)
         * caller and alloc_fdtable().  Cheaper to catch it here...
         */
        if (unlikely(new_fdt->max_fds <= nr)) {
-               free_fdarr(new_fdt);
-               free_fdset(new_fdt);
-               kfree(new_fdt);
+               __free_fdtable(new_fdt);
                return -EMFILE;
        }
        /*
@@ -232,9 +231,7 @@ static int expand_fdtable(struct files_struct *files, int nr)
                        free_fdtable(cur_fdt);
        } else {
                /* Somebody else expanded, so undo our attempt */
-               free_fdarr(new_fdt);
-               free_fdset(new_fdt);
-               kfree(new_fdt);
+               __free_fdtable(new_fdt);
        }
        return 1;
 }
@@ -312,7 +309,6 @@ struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
        new_fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
        new_fdt->open_fds = (fd_set *)&newf->open_fds_init;
        new_fdt->fd = &newf->fd_array[0];
-       INIT_RCU_HEAD(&new_fdt->rcu);
        new_fdt->next = NULL;
 
        spin_lock(&oldf->file_lock);
@@ -325,11 +321,8 @@ struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
        while (unlikely(open_files > new_fdt->max_fds)) {
                spin_unlock(&oldf->file_lock);
 
-               if (new_fdt != &newf->fdtab) {
-                       free_fdarr(new_fdt);
-                       free_fdset(new_fdt);
-                       kfree(new_fdt);
-               }
+               if (new_fdt != &newf->fdtab)
+                       __free_fdtable(new_fdt);
 
                new_fdt = alloc_fdtable(open_files - 1);
                if (!new_fdt) {
@@ -339,9 +332,7 @@ struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
 
                /* beyond sysctl_nr_open; nothing to do */
                if (unlikely(new_fdt->max_fds < open_files)) {
-                       free_fdarr(new_fdt);
-                       free_fdset(new_fdt);
-                       kfree(new_fdt);
+                       __free_fdtable(new_fdt);
                        *errorp = -EMFILE;
                        goto out_release;
                }
@@ -430,7 +421,6 @@ struct files_struct init_files = {
                .fd             = &init_files.fd_array[0],
                .close_on_exec  = (fd_set *)&init_files.close_on_exec_init,
                .open_fds       = (fd_set *)&init_files.open_fds_init,
-               .rcu            = RCU_HEAD_INIT,
        },
        .file_lock      = __SPIN_LOCK_UNLOCKED(init_task.file_lock),
 };