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 /* public *and* exported. Not pretty! */
  26 spinlock_t __cacheline_aligned_in_smp files_lock = SPIN_LOCK_UNLOCKED;
  27
  28 static spinlock_t filp_count_lock = SPIN_LOCK_UNLOCKED;
  29
  30 /* slab constructors and destructors are called from arbitrary
  31  * context and must be fully threaded - use a local spinlock
  32  * to protect files_stat.nr_files
  33  */
  34 void filp_ctor(void * objp, struct kmem_cache_s *cachep, unsigned long cflags)
  35 {
  36         if ((cflags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
  37             SLAB_CTOR_CONSTRUCTOR) {
  38                 unsigned long flags;
  39                 spin_lock_irqsave(&filp_count_lock, flags);
  40                 files_stat.nr_files++;
  41                 spin_unlock_irqrestore(&filp_count_lock, flags);
  42         }
  43 }
  44
  45 void filp_dtor(void * objp, struct kmem_cache_s *cachep, unsigned long dflags)
  46 {
  47         unsigned long flags;
  48         spin_lock_irqsave(&filp_count_lock, flags);
  49         files_stat.nr_files--;
  50         spin_unlock_irqrestore(&filp_count_lock, flags);
  51 }
  52
  53 static inline void file_free(struct file *f)
  54 {
  55         kmem_cache_free(filp_cachep, f);
  56 }
  57
  58 /* Find an unused file structure and return a pointer to it.
  59  * Returns NULL, if there are no more free file structures or
  60  * we run out of memory.
  61  */
  62 struct file *get_empty_filp(void)
  63 {
  64 static int old_max = 0;
  65         struct file * f;
  66
  67         /*
  68          * Privileged users can go above max_files
  69          */
  70         if (files_stat.nr_files < files_stat.max_files ||
  71                                 capable(CAP_SYS_ADMIN)) {
  72                 f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
  73                 if (f) {
  74                         memset(f, 0, sizeof(*f));
  75                         if (security_file_alloc(f)) {
  76                                 file_free(f);
  77                                 goto fail;
  78                         }
  79                         eventpoll_init_file(f);
  80                         atomic_set(&f->f_count, 1);
  81                         f->f_uid = current->fsuid;
  82                         f->f_gid = current->fsgid;
  83                         f->f_owner.lock = RW_LOCK_UNLOCKED;
  84                         /* f->f_version: 0 */
  85                         INIT_LIST_HEAD(&f->f_list);
  86                         return f;
  87                 }
  88         }
  89
  90         /* Ran out of filps - report that */
  91         if (files_stat.max_files >= old_max) {
  92                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
  93                                         files_stat.max_files);
  94                 old_max = files_stat.max_files;
  95         } else {
  96                 /* Big problems... */
  97                 printk(KERN_WARNING "VFS: filp allocation failed\n");
  98         }
  99 fail:
 100         return NULL;
 101 }
 102
 103 /*
 104  * Clear and initialize a (private) struct file for the given dentry,
 105  * allocate the security structure, and call the open function (if any).
 106  * The file should be released using close_private_file.
 107  */
 108 int open_private_file(struct file *filp, struct dentry *dentry, int flags)
 109 {
 110         int error;
 111         memset(filp, 0, sizeof(*filp));
 112         eventpoll_init_file(filp);
 113         filp->f_flags  = flags;
 114         filp->f_mode   = (flags+1) & O_ACCMODE;
 115         atomic_set(&filp->f_count, 1);
 116         filp->f_dentry = dentry;
 117         filp->f_uid    = current->fsuid;
 118         filp->f_gid    = current->fsgid;
 119         filp->f_op     = dentry->d_inode->i_fop;
 120         INIT_LIST_HEAD(&filp->f_list);
 121         error = security_file_alloc(filp);
 122         if (!error)
 123                 if (filp->f_op && filp->f_op->open) {
 124                         error = filp->f_op->open(dentry->d_inode, filp);
 125                         if (error)
 126                                 security_file_free(filp);
 127                 }
 128         return error;
 129 }
 130
 131 /*
 132  * Release a private file by calling the release function (if any) and
 133  * freeing the security structure.
 134  */
 135 void close_private_file(struct file *file)
 136 {
 137         struct inode * inode = file->f_dentry->d_inode;
 138
 139         if (file->f_op && file->f_op->release)
 140                 file->f_op->release(inode, file);
 141         security_file_free(file);
 142 }
 143
 144 void fput(struct file *file)
 145 {
 146         if (atomic_dec_and_test(&file->f_count))
 147                 __fput(file);
 148 }
 149
 150 /* __fput is called from task context when aio completion releases the last
 151  * last use of a struct file *.  Do not use otherwise.
 152  */
 153 void __fput(struct file *file)
 154 {
 155         struct dentry *dentry = file->f_dentry;
 156         struct vfsmount *mnt = file->f_vfsmnt;
 157         struct inode *inode = dentry->d_inode;
 158
 159         /*
 160          * The function eventpoll_release() should be the first called
 161          * in the file cleanup chain.
 162          */
 163         eventpoll_release(file);
 164         locks_remove_flock(file);
 165
 166         if (file->f_op && file->f_op->release)
 167                 file->f_op->release(inode, file);
 168         security_file_free(file);
 169         if (unlikely(inode->i_cdev != NULL))
 170                 cdev_put(inode->i_cdev);
 171         fops_put(file->f_op);
 172         if (file->f_mode & FMODE_WRITE)
 173                 put_write_access(inode);
 174         file->f_dentry = NULL;
 175         file->f_vfsmnt = NULL;
 176         file_kill(file);
 177         file_free(file);
 178         dput(dentry);
 179         mntput(mnt);
 180 }
 181
 182 struct file *fget(unsigned int fd)
 183 {
 184         struct file *file;
 185         struct files_struct *files = current->files;
 186
 187         spin_lock(&files->file_lock);
 188         file = fcheck(fd);
 189         if (file)
 190                 get_file(file);
 191         spin_unlock(&files->file_lock);
 192         return file;
 193 }
 194
 195 /*
 196  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 197  * You can use this only if it is guranteed that the current task already
 198  * holds a refcnt to that file. That check has to be done at fget() only
 199  * and a flag is returned to be passed to the corresponding fput_light().
 200  * There must not be a cloning between an fget_light/fput_light pair.
 201  */
 202 struct file *fget_light(unsigned int fd, int *fput_needed)
 203 {
 204         struct file *file;
 205         struct files_struct *files = current->files;
 206
 207         *fput_needed = 0;
 208         if (likely((atomic_read(&files->count) == 1))) {
 209                 file = fcheck(fd);
 210         } else {
 211                 spin_lock(&files->file_lock);
 212                 file = fcheck(fd);
 213                 if (file) {
 214                         get_file(file);
 215                         *fput_needed = 1;
 216                 }
 217                 spin_unlock(&files->file_lock);
 218         }
 219         return file;
 220 }
 221
 222
 223 void put_filp(struct file *file)
 224 {
 225         if (atomic_dec_and_test(&file->f_count)) {
 226                 security_file_free(file);
 227                 file_kill(file);
 228                 file_free(file);
 229         }
 230 }
 231
 232 void file_move(struct file *file, struct list_head *list)
 233 {
 234         if (!list)
 235                 return;
 236         file_list_lock();
 237         list_move(&file->f_list, list);
 238         file_list_unlock();
 239 }
 240
 241 void file_kill(struct file *file)
 242 {
 243         if (!list_empty(&file->f_list)) {
 244                 file_list_lock();
 245                 list_del_init(&file->f_list);
 246                 file_list_unlock();
 247         }
 248 }
 249
 250 int fs_may_remount_ro(struct super_block *sb)
 251 {
 252         struct list_head *p;
 253
 254         /* Check that no files are currently opened for writing. */
 255         file_list_lock();
 256         list_for_each(p, &sb->s_files) {
 257                 struct file *file = list_entry(p, struct file, f_list);
 258                 struct inode *inode = file->f_dentry->d_inode;
 259
 260                 /* File with pending delete? */
 261                 if (inode->i_nlink == 0)
 262                         goto too_bad;
 263
 264                 /* Writeable file? */
 265                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 266                         goto too_bad;
 267         }
 268         file_list_unlock();
 269         return 1; /* Tis' cool bro. */
 270 too_bad:
 271         file_list_unlock();
 272         return 0;
 273 }
 274
 275 void __init files_init(unsigned long mempages)
 276 {
 277         int n;
 278         /* One file with associated inode and dcache is very roughly 1K.
 279          * Per default don't use more than 10% of our memory for files.
 280          */
 281
 282         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 283         files_stat.max_files = n;
 284         if (files_stat.max_files < NR_FILE)
 285                 files_stat.max_files = NR_FILE;
 286 }
 287