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 *and* exported. Not pretty! */
  28 spinlock_t __cacheline_aligned_in_smp files_lock = SPIN_LOCK_UNLOCKED;
  29
  30 EXPORT_SYMBOL(files_lock);
  31
  32 static spinlock_t filp_count_lock = SPIN_LOCK_UNLOCKED;
  33
  34 /* slab constructors and destructors are called from arbitrary
  35  * context and must be fully threaded - use a local spinlock
  36  * to protect files_stat.nr_files
  37  */
  38 void filp_ctor(void * objp, struct kmem_cache_s *cachep, unsigned long cflags)
  39 {
  40         if ((cflags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
  41             SLAB_CTOR_CONSTRUCTOR) {
  42                 unsigned long flags;
  43                 spin_lock_irqsave(&filp_count_lock, flags);
  44                 files_stat.nr_files++;
  45                 spin_unlock_irqrestore(&filp_count_lock, flags);
  46         }
  47 }
  48
  49 void filp_dtor(void * objp, struct kmem_cache_s *cachep, unsigned long dflags)
  50 {
  51         unsigned long flags;
  52         spin_lock_irqsave(&filp_count_lock, flags);
  53         files_stat.nr_files--;
  54         spin_unlock_irqrestore(&filp_count_lock, flags);
  55 }
  56
  57 static inline void file_free(struct file *f)
  58 {
  59         kmem_cache_free(filp_cachep, f);
  60 }
  61
  62 /* Find an unused file structure and return a pointer to it.
  63  * Returns NULL, if there are no more free file structures or
  64  * we run out of memory.
  65  */
  66 struct file *get_empty_filp(void)
  67 {
  68 static int old_max;
  69         struct file * f;
  70
  71         /*
  72          * Privileged users can go above max_files
  73          */
  74         if (files_stat.nr_files < files_stat.max_files ||
  75                                 capable(CAP_SYS_ADMIN)) {
  76                 f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
  77                 if (f) {
  78                         memset(f, 0, sizeof(*f));
  79                         if (security_file_alloc(f)) {
  80                                 file_free(f);
  81                                 goto fail;
  82                         }
  83                         eventpoll_init_file(f);
  84                         atomic_set(&f->f_count, 1);
  85                         f->f_uid = current->fsuid;
  86                         f->f_gid = current->fsgid;
  87                         f->f_owner.lock = RW_LOCK_UNLOCKED;
  88                         /* f->f_version: 0 */
  89                         INIT_LIST_HEAD(&f->f_list);
  90                         return f;
  91                 }
  92         }
  93
  94         /* Ran out of filps - report that */
  95         if (files_stat.max_files >= old_max) {
  96                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
  97                                         files_stat.max_files);
  98                 old_max = files_stat.max_files;
  99         } else {
 100                 /* Big problems... */
 101                 printk(KERN_WARNING "VFS: filp allocation failed\n");
 102         }
 103 fail:
 104         return NULL;
 105 }
 106
 107 EXPORT_SYMBOL(get_empty_filp);
 108
 109 void fastcall fput(struct file *file)
 110 {
 111         if (atomic_dec_and_test(&file->f_count))
 112                 __fput(file);
 113 }
 114
 115 EXPORT_SYMBOL(fput);
 116
 117 /* __fput is called from task context when aio completion releases the last
 118  * last use of a struct file *.  Do not use otherwise.
 119  */
 120 void fastcall __fput(struct file *file)
 121 {
 122         struct dentry *dentry = file->f_dentry;
 123         struct vfsmount *mnt = file->f_vfsmnt;
 124         struct inode *inode = dentry->d_inode;
 125
 126         might_sleep();
 127         /*
 128          * The function eventpoll_release() should be the first called
 129          * in the file cleanup chain.
 130          */
 131         eventpoll_release(file);
 132         locks_remove_flock(file);
 133
 134         if (file->f_op && file->f_op->release)
 135                 file->f_op->release(inode, file);
 136         security_file_free(file);
 137         if (unlikely(inode->i_cdev != NULL))
 138                 cdev_put(inode->i_cdev);
 139         fops_put(file->f_op);
 140         if (file->f_mode & FMODE_WRITE)
 141                 put_write_access(inode);
 142         file_kill(file);
 143         file->f_dentry = NULL;
 144         file->f_vfsmnt = NULL;
 145         file_free(file);
 146         dput(dentry);
 147         mntput(mnt);
 148 }
 149
 150 struct file fastcall *fget(unsigned int fd)
 151 {
 152         struct file *file;
 153         struct files_struct *files = current->files;
 154
 155         spin_lock(&files->file_lock);
 156         file = fcheck_files(files, fd);
 157         if (file)
 158                 get_file(file);
 159         spin_unlock(&files->file_lock);
 160         return file;
 161 }
 162
 163 EXPORT_SYMBOL(fget);
 164
 165 /*
 166  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 167  * You can use this only if it is guranteed that the current task already
 168  * holds a refcnt to that file. That check has to be done at fget() only
 169  * and a flag is returned to be passed to the corresponding fput_light().
 170  * There must not be a cloning between an fget_light/fput_light pair.
 171  */
 172 struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
 173 {
 174         struct file *file;
 175         struct files_struct *files = current->files;
 176
 177         *fput_needed = 0;
 178         if (likely((atomic_read(&files->count) == 1))) {
 179                 file = fcheck_files(files, fd);
 180         } else {
 181                 spin_lock(&files->file_lock);
 182                 file = fcheck_files(files, fd);
 183                 if (file) {
 184                         get_file(file);
 185                         *fput_needed = 1;
 186                 }
 187                 spin_unlock(&files->file_lock);
 188         }
 189         return file;
 190 }
 191
 192
 193 void put_filp(struct file *file)
 194 {
 195         if (atomic_dec_and_test(&file->f_count)) {
 196                 security_file_free(file);
 197                 file_kill(file);
 198                 file_free(file);
 199         }
 200 }
 201
 202 EXPORT_SYMBOL(put_filp);
 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 }