vfs: rename 'do_follow_link' to 'should_follow_link'
[linux-2.6.git] / fs / file.c
index 0f705c7..4c6992d 100644 (file)
--- a/fs/file.c
+++ b/fs/file.c
@@ -6,9 +6,12 @@
  *  Manage the dynamic fd arrays in the process files_struct.
  */
 
+#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>
 #include <linux/vmalloc.h>
 #include <linux/file.h>
@@ -26,6 +29,8 @@ struct fdtable_defer {
 };
 
 int sysctl_nr_open __read_mostly = 1024*1024;
+int sysctl_nr_open_min = BITS_PER_LONG;
+int sysctl_nr_open_max = 1024 * 1024; /* raised later */
 
 /*
  * We use this list to defer free fdtables that have vmalloced
@@ -35,28 +40,30 @@ int sysctl_nr_open __read_mostly = 1024*1024;
  */
 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)
@@ -71,9 +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;
        }
 }
@@ -94,7 +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);
@@ -174,13 +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:
@@ -210,9 +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;
        }
        /*
@@ -228,9 +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;
 }
@@ -248,9 +249,18 @@ int expand_files(struct files_struct *files, int nr)
        struct fdtable *fdt;
 
        fdt = files_fdtable(files);
+
+       /*
+        * N.B. For clone tasks sharing a files structure, this test
+        * will limit the total number of files that can be opened.
+        */
+       if (nr >= rlimit(RLIMIT_NOFILE))
+               return -EMFILE;
+
        /* Do we need to expand? */
        if (nr < fdt->max_fds)
                return 0;
+
        /* Can we expand? */
        if (nr >= sysctl_nr_open)
                return -EMFILE;
@@ -299,7 +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);
@@ -312,11 +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) {
@@ -326,9 +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;
                }
@@ -405,6 +409,8 @@ void __init files_defer_init(void)
        int i;
        for_each_possible_cpu(i)
                fdtable_defer_list_init(i);
+       sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
+                            -BITS_PER_LONG;
 }
 
 struct files_struct init_files = {
@@ -415,7 +421,66 @@ 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),
 };
+
+/*
+ * allocate a file descriptor, mark it busy.
+ */
+int alloc_fd(unsigned start, unsigned flags)
+{
+       struct files_struct *files = current->files;
+       unsigned int fd;
+       int error;
+       struct fdtable *fdt;
+
+       spin_lock(&files->file_lock);
+repeat:
+       fdt = files_fdtable(files);
+       fd = start;
+       if (fd < files->next_fd)
+               fd = files->next_fd;
+
+       if (fd < fdt->max_fds)
+               fd = find_next_zero_bit(fdt->open_fds->fds_bits,
+                                          fdt->max_fds, fd);
+
+       error = expand_files(files, fd);
+       if (error < 0)
+               goto out;
+
+       /*
+        * If we needed to expand the fs array we
+        * might have blocked - try again.
+        */
+       if (error)
+               goto repeat;
+
+       if (start <= files->next_fd)
+               files->next_fd = fd + 1;
+
+       FD_SET(fd, fdt->open_fds);
+       if (flags & O_CLOEXEC)
+               FD_SET(fd, fdt->close_on_exec);
+       else
+               FD_CLR(fd, fdt->close_on_exec);
+       error = fd;
+#if 1
+       /* Sanity check */
+       if (rcu_dereference_raw(fdt->fd[fd]) != NULL) {
+               printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
+               rcu_assign_pointer(fdt->fd[fd], NULL);
+       }
+#endif
+
+out:
+       spin_unlock(&files->file_lock);
+       return error;
+}
+
+int get_unused_fd(void)
+{
+       return alloc_fd(0, 0);
+}
+EXPORT_SYMBOL(get_unused_fd);