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/fdtable.h>
  12 #include <linux/init.h>
  13 #include <linux/module.h>
  14 #include <linux/fs.h>
  15 #include <linux/security.h>
  16 #include <linux/ima.h>
  17 #include <linux/eventpoll.h>
  18 #include <linux/rcupdate.h>
  19 #include <linux/mount.h>
  20 #include <linux/capability.h>
  21 #include <linux/cdev.h>
  22 #include <linux/fsnotify.h>
  23 #include <linux/sysctl.h>
  24 #include <linux/percpu_counter.h>
  25
  26 #include <asm/atomic.h>
  27
  28 /* sysctl tunables... */
  29 struct files_stat_struct files_stat = {
  30         .max_files = NR_FILE
  31 };
  32
  33 /* public. Not pretty! */
  34 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
  35
  36 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  37
  38 static inline void file_free_rcu(struct rcu_head *head)
  39 {
  40         struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  41
  42         put_cred(f->f_cred);
  43         kmem_cache_free(filp_cachep, f);
  44 }
  45
  46 static inline void file_free(struct file *f)
  47 {
  48         percpu_counter_dec(&nr_files);
  49         file_check_state(f);
  50         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  51 }
  52
  53 /*
  54  * Return the total number of open files in the system
  55  */
  56 static int get_nr_files(void)
  57 {
  58         return percpu_counter_read_positive(&nr_files);
  59 }
  60
  61 /*
  62  * Return the maximum number of open files in the system
  63  */
  64 int get_max_files(void)
  65 {
  66         return files_stat.max_files;
  67 }
  68 EXPORT_SYMBOL_GPL(get_max_files);
  69
  70 /*
  71  * Handle nr_files sysctl
  72  */
  73 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  74 int proc_nr_files(ctl_table *table, int write, struct file *filp,
  75                      void __user *buffer, size_t *lenp, loff_t *ppos)
  76 {
  77         files_stat.nr_files = get_nr_files();
  78         return proc_dointvec(table, write, filp, buffer, lenp, ppos);
  79 }
  80 #else
  81 int proc_nr_files(ctl_table *table, int write, struct file *filp,
  82                      void __user *buffer, size_t *lenp, loff_t *ppos)
  83 {
  84         return -ENOSYS;
  85 }
  86 #endif
  87
  88 /* Find an unused file structure and return a pointer to it.
  89  * Returns NULL, if there are no more free file structures or
  90  * we run out of memory.
  91  *
  92  * Be very careful using this.  You are responsible for
  93  * getting write access to any mount that you might assign
  94  * to this filp, if it is opened for write.  If this is not
  95  * done, you will imbalance int the mount's writer count
  96  * and a warning at __fput() time.
  97  */
  98 struct file *get_empty_filp(void)
  99 {
 100         const struct cred *cred = current_cred();
 101         static int old_max;
 102         struct file * f;
 103
 104         /*
 105          * Privileged users can go above max_files
 106          */
 107         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 108                 /*
 109                  * percpu_counters are inaccurate.  Do an expensive check before
 110                  * we go and fail.
 111                  */
 112                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 113                         goto over;
 114         }
 115
 116         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 117         if (f == NULL)
 118                 goto fail;
 119
 120         percpu_counter_inc(&nr_files);
 121         if (security_file_alloc(f))
 122                 goto fail_sec;
 123
 124         INIT_LIST_HEAD(&f->f_u.fu_list);
 125         atomic_long_set(&f->f_count, 1);
 126         rwlock_init(&f->f_owner.lock);
 127         f->f_cred = get_cred(cred);
 128         eventpoll_init_file(f);
 129         /* f->f_version: 0 */
 130         return f;
 131
 132 over:
 133         /* Ran out of filps - report that */
 134         if (get_nr_files() > old_max) {
 135                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
 136                                         get_max_files());
 137                 old_max = get_nr_files();
 138         }
 139         goto fail;
 140
 141 fail_sec:
 142         file_free(f);
 143 fail:
 144         return NULL;
 145 }
 146
 147 EXPORT_SYMBOL(get_empty_filp);
 148
 149 /**
 150  * alloc_file - allocate and initialize a 'struct file'
 151  * @mnt: the vfsmount on which the file will reside
 152  * @dentry: the dentry representing the new file
 153  * @mode: the mode with which the new file will be opened
 154  * @fop: the 'struct file_operations' for the new file
 155  *
 156  * Use this instead of get_empty_filp() to get a new
 157  * 'struct file'.  Do so because of the same initialization
 158  * pitfalls reasons listed for init_file().  This is a
 159  * preferred interface to using init_file().
 160  *
 161  * If all the callers of init_file() are eliminated, its
 162  * code should be moved into this function.
 163  */
 164 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
 165                 fmode_t mode, const struct file_operations *fop)
 166 {
 167         struct file *file;
 168         struct path;
 169
 170         file = get_empty_filp();
 171         if (!file)
 172                 return NULL;
 173
 174         init_file(file, mnt, dentry, mode, fop);
 175         return file;
 176 }
 177 EXPORT_SYMBOL(alloc_file);
 178
 179 /**
 180  * init_file - initialize a 'struct file'
 181  * @file: the already allocated 'struct file' to initialized
 182  * @mnt: the vfsmount on which the file resides
 183  * @dentry: the dentry representing this file
 184  * @mode: the mode the file is opened with
 185  * @fop: the 'struct file_operations' for this file
 186  *
 187  * Use this instead of setting the members directly.  Doing so
 188  * avoids making mistakes like forgetting the mntget() or
 189  * forgetting to take a write on the mnt.
 190  *
 191  * Note: This is a crappy interface.  It is here to make
 192  * merging with the existing users of get_empty_filp()
 193  * who have complex failure logic easier.  All users
 194  * of this should be moving to alloc_file().
 195  */
 196 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
 197            fmode_t mode, const struct file_operations *fop)
 198 {
 199         int error = 0;
 200         file->f_path.dentry = dentry;
 201         file->f_path.mnt = mntget(mnt);
 202         file->f_mapping = dentry->d_inode->i_mapping;
 203         file->f_mode = mode;
 204         file->f_op = fop;
 205
 206         /*
 207          * These mounts don't really matter in practice
 208          * for r/o bind mounts.  They aren't userspace-
 209          * visible.  We do this for consistency, and so
 210          * that we can do debugging checks at __fput()
 211          */
 212         if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
 213                 file_take_write(file);
 214                 error = mnt_want_write(mnt);
 215                 WARN_ON(error);
 216         }
 217         return error;
 218 }
 219 EXPORT_SYMBOL(init_file);
 220
 221 void fput(struct file *file)
 222 {
 223         if (atomic_long_dec_and_test(&file->f_count))
 224                 __fput(file);
 225 }
 226
 227 EXPORT_SYMBOL(fput);
 228
 229 /**
 230  * drop_file_write_access - give up ability to write to a file
 231  * @file: the file to which we will stop writing
 232  *
 233  * This is a central place which will give up the ability
 234  * to write to @file, along with access to write through
 235  * its vfsmount.
 236  */
 237 void drop_file_write_access(struct file *file)
 238 {
 239         struct vfsmount *mnt = file->f_path.mnt;
 240         struct dentry *dentry = file->f_path.dentry;
 241         struct inode *inode = dentry->d_inode;
 242
 243         put_write_access(inode);
 244
 245         if (special_file(inode->i_mode))
 246                 return;
 247         if (file_check_writeable(file) != 0)
 248                 return;
 249         mnt_drop_write(mnt);
 250         file_release_write(file);
 251 }
 252 EXPORT_SYMBOL_GPL(drop_file_write_access);
 253
 254 /* __fput is called from task context when aio completion releases the last
 255  * last use of a struct file *.  Do not use otherwise.
 256  */
 257 void __fput(struct file *file)
 258 {
 259         struct dentry *dentry = file->f_path.dentry;
 260         struct vfsmount *mnt = file->f_path.mnt;
 261         struct inode *inode = dentry->d_inode;
 262
 263         might_sleep();
 264
 265         fsnotify_close(file);
 266         /*
 267          * The function eventpoll_release() should be the first called
 268          * in the file cleanup chain.
 269          */
 270         eventpoll_release(file);
 271         locks_remove_flock(file);
 272
 273         if (unlikely(file->f_flags & FASYNC)) {
 274                 if (file->f_op && file->f_op->fasync)
 275                         file->f_op->fasync(-1, file, 0);
 276         }
 277         if (file->f_op && file->f_op->release)
 278                 file->f_op->release(inode, file);
 279         security_file_free(file);
 280         ima_file_free(file);
 281         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 282                 cdev_put(inode->i_cdev);
 283         fops_put(file->f_op);
 284         put_pid(file->f_owner.pid);
 285         file_kill(file);
 286         if (file->f_mode & FMODE_WRITE)
 287                 drop_file_write_access(file);
 288         file->f_path.dentry = NULL;
 289         file->f_path.mnt = NULL;
 290         file_free(file);
 291         dput(dentry);
 292         mntput(mnt);
 293 }
 294
 295 struct file *fget(unsigned int fd)
 296 {
 297         struct file *file;
 298         struct files_struct *files = current->files;
 299
 300         rcu_read_lock();
 301         file = fcheck_files(files, fd);
 302         if (file) {
 303                 if (!atomic_long_inc_not_zero(&file->f_count)) {
 304                         /* File object ref couldn't be taken */
 305                         rcu_read_unlock();
 306                         return NULL;
 307                 }
 308         }
 309         rcu_read_unlock();
 310
 311         return file;
 312 }
 313
 314 EXPORT_SYMBOL(fget);
 315
 316 /*
 317  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 318  * You can use this only if it is guranteed that the current task already
 319  * holds a refcnt to that file. That check has to be done at fget() only
 320  * and a flag is returned to be passed to the corresponding fput_light().
 321  * There must not be a cloning between an fget_light/fput_light pair.
 322  */
 323 struct file *fget_light(unsigned int fd, int *fput_needed)
 324 {
 325         struct file *file;
 326         struct files_struct *files = current->files;
 327
 328         *fput_needed = 0;
 329         if (likely((atomic_read(&files->count) == 1))) {
 330                 file = fcheck_files(files, fd);
 331         } else {
 332                 rcu_read_lock();
 333                 file = fcheck_files(files, fd);
 334                 if (file) {
 335                         if (atomic_long_inc_not_zero(&file->f_count))
 336                                 *fput_needed = 1;
 337                         else
 338                                 /* Didn't get the reference, someone's freed */
 339                                 file = NULL;
 340                 }
 341                 rcu_read_unlock();
 342         }
 343
 344         return file;
 345 }
 346
 347
 348 void put_filp(struct file *file)
 349 {
 350         if (atomic_long_dec_and_test(&file->f_count)) {
 351                 security_file_free(file);
 352                 file_kill(file);
 353                 file_free(file);
 354         }
 355 }
 356
 357 void file_move(struct file *file, struct list_head *list)
 358 {
 359         if (!list)
 360                 return;
 361         file_list_lock();
 362         list_move(&file->f_u.fu_list, list);
 363         file_list_unlock();
 364 }
 365
 366 void file_kill(struct file *file)
 367 {
 368         if (!list_empty(&file->f_u.fu_list)) {
 369                 file_list_lock();
 370                 list_del_init(&file->f_u.fu_list);
 371                 file_list_unlock();
 372         }
 373 }
 374
 375 int fs_may_remount_ro(struct super_block *sb)
 376 {
 377         struct file *file;
 378
 379         /* Check that no files are currently opened for writing. */
 380         file_list_lock();
 381         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 382                 struct inode *inode = file->f_path.dentry->d_inode;
 383
 384                 /* File with pending delete? */
 385                 if (inode->i_nlink == 0)
 386                         goto too_bad;
 387
 388                 /* Writeable file? */
 389                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 390                         goto too_bad;
 391         }
 392         file_list_unlock();
 393         return 1; /* Tis' cool bro. */
 394 too_bad:
 395         file_list_unlock();
 396         return 0;
 397 }
 398
 399 void __init files_init(unsigned long mempages)
 400 {
 401         int n;
 402         /* One file with associated inode and dcache is very roughly 1K.
 403          * Per default don't use more than 10% of our memory for files.
 404          */
 405
 406         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 407         files_stat.max_files = n;
 408         if (files_stat.max_files < NR_FILE)
 409                 files_stat.max_files = NR_FILE;
 410         files_defer_init();
 411         percpu_counter_init(&nr_files, 0);
 412 }