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