fs/file_table.c

   1 /*
   2  *  linux/fs/file_table.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *  Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
   6  */
   7
   8 #include <linux/string.h>
   9 #include <linux/slab.h>
  10 #include <linux/file.h>
  11 #include <linux/init.h>
  12 #include <linux/module.h>
  13 #include <linux/smp_lock.h>
  14 #include <linux/fs.h>
  15 #include <linux/security.h>
  16 #include <linux/eventpoll.h>
  17 #include <linux/mount.h>
  18 #include <linux/cdev.h>
  19 #include <linux/fsnotify.h>
  20
  21 /* sysctl tunables... */
  22 struct files_stat_struct files_stat = {
  23         .max_files = NR_FILE
  24 };
  25
  26 EXPORT_SYMBOL(files_stat); /* Needed by unix.o */
  27
  28 /* public. Not pretty! */
  29  __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
  30
  31 static DEFINE_SPINLOCK(filp_count_lock);
  32
  33 /* slab constructors and destructors are called from arbitrary
  34  * context and must be fully threaded - use a local spinlock
  35  * to protect files_stat.nr_files
  36  */
  37 void filp_ctor(void * objp, struct kmem_cache_s *cachep, unsigned long cflags)
  38 {
  39         if ((cflags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
  40             SLAB_CTOR_CONSTRUCTOR) {
  41                 unsigned long flags;
  42                 spin_lock_irqsave(&filp_count_lock, flags);
  43                 files_stat.nr_files++;
  44                 spin_unlock_irqrestore(&filp_count_lock, flags);
  45         }
  46 }
  47
  48 void filp_dtor(void * objp, struct kmem_cache_s *cachep, unsigned long dflags)
  49 {
  50         unsigned long flags;
  51         spin_lock_irqsave(&filp_count_lock, flags);
  52         files_stat.nr_files--;
  53         spin_unlock_irqrestore(&filp_count_lock, flags);
  54 }
  55
  56 static inline void file_free(struct file *f)
  57 {
  58         kmem_cache_free(filp_cachep, f);
  59 }
  60
  61 /* Find an unused file structure and return a pointer to it.
  62  * Returns NULL, if there are no more free file structures or
  63  * we run out of memory.
  64  */
  65 struct file *get_empty_filp(void)
  66 {
  67         static int old_max;
  68         struct file * f;
  69
  70         /*
  71          * Privileged users can go above max_files
  72          */
  73         if (files_stat.nr_files >= files_stat.max_files &&
  74                                 !capable(CAP_SYS_ADMIN))
  75                 goto over;
  76
  77         f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
  78         if (f == NULL)
  79                 goto fail;
  80
  81         memset(f, 0, sizeof(*f));
  82         if (security_file_alloc(f))
  83                 goto fail_sec;
  84
  85         eventpoll_init_file(f);
  86         atomic_set(&f->f_count, 1);
  87         f->f_uid = current->fsuid;
  88         f->f_gid = current->fsgid;
  89         rwlock_init(&f->f_owner.lock);
  90         /* f->f_version: 0 */
  91         INIT_LIST_HEAD(&f->f_list);
  92         return f;
  93
  94 over:
  95         /* Ran out of filps - report that */
  96         if (files_stat.nr_files > old_max) {
  97                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
  98                                         files_stat.max_files);
  99                 old_max = files_stat.nr_files;
 100         }
 101         goto fail;
 102
 103 fail_sec:
 104         file_free(f);
 105 fail:
 106         return NULL;
 107 }
 108
 109 EXPORT_SYMBOL(get_empty_filp);
 110
 111 void fastcall fput(struct file *file)
 112 {
 113         if (atomic_dec_and_test(&file->f_count))
 114                 __fput(file);
 115 }
 116
 117 EXPORT_SYMBOL(fput);
 118
 119 /* __fput is called from task context when aio completion releases the last
 120  * last use of a struct file *.  Do not use otherwise.
 121  */
 122 void fastcall __fput(struct file *file)
 123 {
 124         struct dentry *dentry = file->f_dentry;
 125         struct vfsmount *mnt = file->f_vfsmnt;
 126         struct inode *inode = dentry->d_inode;
 127
 128         might_sleep();
 129
 130         fsnotify_close(file);
 131         /*
 132          * The function eventpoll_release() should be the first called
 133          * in the file cleanup chain.
 134          */
 135         eventpoll_release(file);
 136         locks_remove_flock(file);
 137
 138         if (file->f_op && file->f_op->release)
 139                 file->f_op->release(inode, file);
 140         security_file_free(file);
 141         if (unlikely(inode->i_cdev != NULL))
 142                 cdev_put(inode->i_cdev);
 143         fops_put(file->f_op);
 144         if (file->f_mode & FMODE_WRITE)
 145                 put_write_access(inode);
 146         file_kill(file);
 147         file->f_dentry = NULL;
 148         file->f_vfsmnt = NULL;
 149         file_free(file);
 150         dput(dentry);
 151         mntput(mnt);
 152 }
 153
 154 struct file fastcall *fget(unsigned int fd)
 155 {
 156         struct file *file;
 157         struct files_struct *files = current->files;
 158
 159         spin_lock(&files->file_lock);
 160         file = fcheck_files(files, fd);
 161         if (file)
 162                 get_file(file);
 163         spin_unlock(&files->file_lock);
 164         return file;
 165 }
 166
 167 EXPORT_SYMBOL(fget);
 168
 169 /*
 170  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 171  * You can use this only if it is guranteed that the current task already
 172  * holds a refcnt to that file. That check has to be done at fget() only
 173  * and a flag is returned to be passed to the corresponding fput_light().
 174  * There must not be a cloning between an fget_light/fput_light pair.
 175  */
 176 struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
 177 {
 178         struct file *file;
 179         struct files_struct *files = current->files;
 180
 181         *fput_needed = 0;
 182         if (likely((atomic_read(&files->count) == 1))) {
 183                 file = fcheck_files(files, fd);
 184         } else {
 185                 spin_lock(&files->file_lock);
 186                 file = fcheck_files(files, fd);
 187                 if (file) {
 188                         get_file(file);
 189                         *fput_needed = 1;
 190                 }
 191                 spin_unlock(&files->file_lock);
 192         }
 193         return file;
 194 }
 195
 196
 197 void put_filp(struct file *file)
 198 {
 199         if (atomic_dec_and_test(&file->f_count)) {
 200                 security_file_free(file);
 201                 file_kill(file);
 202                 file_free(file);
 203         }
 204 }
 205
 206 void file_move(struct file *file, struct list_head *list)
 207 {
 208         if (!list)
 209                 return;
 210         file_list_lock();
 211         list_move(&file->f_list, list);
 212         file_list_unlock();
 213 }
 214
 215 void file_kill(struct file *file)
 216 {
 217         if (!list_empty(&file->f_list)) {
 218                 file_list_lock();
 219                 list_del_init(&file->f_list);
 220                 file_list_unlock();
 221         }
 222 }
 223
 224 int fs_may_remount_ro(struct super_block *sb)
 225 {
 226         struct list_head *p;
 227
 228         /* Check that no files are currently opened for writing. */
 229         file_list_lock();
 230         list_for_each(p, &sb->s_files) {
 231                 struct file *file = list_entry(p, struct file, f_list);
 232                 struct inode *inode = file->f_dentry->d_inode;
 233
 234                 /* File with pending delete? */
 235                 if (inode->i_nlink == 0)
 236                         goto too_bad;
 237
 238                 /* Writeable file? */
 239                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 240                         goto too_bad;
 241         }
 242         file_list_unlock();
 243         return 1; /* Tis' cool bro. */
 244 too_bad:
 245         file_list_unlock();
 246         return 0;
 247 }
 248
 249 void __init files_init(unsigned long mempages)
 250 {
 251         int n;
 252         /* One file with associated inode and dcache is very roughly 1K.
 253          * Per default don't use more than 10% of our memory for files.
 254          */
 255
 256         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 257         files_stat.max_files = n;
 258         if (files_stat.max_files < NR_FILE)
 259                 files_stat.max_files = NR_FILE;
 260 }