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/eventpoll.h>
  17 #include <linux/rcupdate.h>
  18 #include <linux/mount.h>
  19 #include <linux/capability.h>
  20 #include <linux/cdev.h>
  21 #include <linux/fsnotify.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/percpu_counter.h>
  24
  25 #include <asm/atomic.h>
  26
  27 /* sysctl tunables... */
  28 struct files_stat_struct files_stat = {
  29         .max_files = NR_FILE
  30 };
  31
  32 /* public. Not pretty! */
  33 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
  34
  35 /* SLAB cache for file structures */
  36 static struct kmem_cache *filp_cachep __read_mostly;
  37
  38 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  39
  40 static inline void file_free_rcu(struct rcu_head *head)
  41 {
  42         struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  43
  44         put_cred(f->f_cred);
  45         kmem_cache_free(filp_cachep, f);
  46 }
  47
  48 static inline void file_free(struct file *f)
  49 {
  50         percpu_counter_dec(&nr_files);
  51         file_check_state(f);
  52         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  53 }
  54
  55 /*
  56  * Return the total number of open files in the system
  57  */
  58 static int get_nr_files(void)
  59 {
  60         return percpu_counter_read_positive(&nr_files);
  61 }
  62
  63 /*
  64  * Return the maximum number of open files in the system
  65  */
  66 int get_max_files(void)
  67 {
  68         return files_stat.max_files;
  69 }
  70 EXPORT_SYMBOL_GPL(get_max_files);
  71
  72 /*
  73  * Handle nr_files sysctl
  74  */
  75 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  76 int proc_nr_files(ctl_table *table, int write,
  77                      void __user *buffer, size_t *lenp, loff_t *ppos)
  78 {
  79         files_stat.nr_files = get_nr_files();
  80         return proc_dointvec(table, write, buffer, lenp, ppos);
  81 }
  82 #else
  83 int proc_nr_files(ctl_table *table, int write,
  84                      void __user *buffer, size_t *lenp, loff_t *ppos)
  85 {
  86         return -ENOSYS;
  87 }
  88 #endif
  89
  90 /* Find an unused file structure and return a pointer to it.
  91  * Returns NULL, if there are no more free file structures or
  92  * we run out of memory.
  93  *
  94  * Be very careful using this.  You are responsible for
  95  * getting write access to any mount that you might assign
  96  * to this filp, if it is opened for write.  If this is not
  97  * done, you will imbalance int the mount's writer count
  98  * and a warning at __fput() time.
  99  */
 100 struct file *get_empty_filp(void)
 101 {
 102         const struct cred *cred = current_cred();
 103         static int old_max;
 104         struct file * f;
 105
 106         /*
 107          * Privileged users can go above max_files
 108          */
 109         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 110                 /*
 111                  * percpu_counters are inaccurate.  Do an expensive check before
 112                  * we go and fail.
 113                  */
 114                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 115                         goto over;
 116         }
 117
 118         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 119         if (f == NULL)
 120                 goto fail;
 121
 122         percpu_counter_inc(&nr_files);
 123         if (security_file_alloc(f))
 124                 goto fail_sec;
 125
 126         INIT_LIST_HEAD(&f->f_u.fu_list);
 127         atomic_long_set(&f->f_count, 1);
 128         rwlock_init(&f->f_owner.lock);
 129         f->f_cred = get_cred(cred);
 130         spin_lock_init(&f->f_lock);
 131         eventpoll_init_file(f);
 132         /* f->f_version: 0 */
 133         return f;
 134
 135 over:
 136         /* Ran out of filps - report that */
 137         if (get_nr_files() > old_max) {
 138                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
 139                                         get_max_files());
 140                 old_max = get_nr_files();
 141         }
 142         goto fail;
 143
 144 fail_sec:
 145         file_free(f);
 146 fail:
 147         return NULL;
 148 }
 149
 150 /**
 151  * alloc_file - allocate and initialize a 'struct file'
 152  * @mnt: the vfsmount on which the file will reside
 153  * @dentry: the dentry representing the new file
 154  * @mode: the mode with which the new file will be opened
 155  * @fop: the 'struct file_operations' for the new file
 156  *
 157  * Use this instead of get_empty_filp() to get a new
 158  * 'struct file'.  Do so because of the same initialization
 159  * pitfalls reasons listed for init_file().  This is a
 160  * preferred interface to using init_file().
 161  *
 162  * If all the callers of init_file() are eliminated, its
 163  * code should be moved into this function.
 164  */
 165 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
 166                 fmode_t mode, const struct file_operations *fop)
 167 {
 168         struct file *file;
 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_clone_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         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 281                 cdev_put(inode->i_cdev);
 282         fops_put(file->f_op);
 283         put_pid(file->f_owner.pid);
 284         file_kill(file);
 285         if (file->f_mode & FMODE_WRITE)
 286                 drop_file_write_access(file);
 287         file->f_path.dentry = NULL;
 288         file->f_path.mnt = NULL;
 289         file_free(file);
 290         dput(dentry);
 291         mntput(mnt);
 292 }
 293
 294 struct file *fget(unsigned int fd)
 295 {
 296         struct file *file;
 297         struct files_struct *files = current->files;
 298
 299         rcu_read_lock();
 300         file = fcheck_files(files, fd);
 301         if (file) {
 302                 if (!atomic_long_inc_not_zero(&file->f_count)) {
 303                         /* File object ref couldn't be taken */
 304                         rcu_read_unlock();
 305                         return NULL;
 306                 }
 307         }
 308         rcu_read_unlock();
 309
 310         return file;
 311 }
 312
 313 EXPORT_SYMBOL(fget);
 314
 315 /*
 316  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 317  * You can use this only if it is guranteed that the current task already
 318  * holds a refcnt to that file. That check has to be done at fget() only
 319  * and a flag is returned to be passed to the corresponding fput_light().
 320  * There must not be a cloning between an fget_light/fput_light pair.
 321  */
 322 struct file *fget_light(unsigned int fd, int *fput_needed)
 323 {
 324         struct file *file;
 325         struct files_struct *files = current->files;
 326
 327         *fput_needed = 0;
 328         if (likely((atomic_read(&files->count) == 1))) {
 329                 file = fcheck_files(files, fd);
 330         } else {
 331                 rcu_read_lock();
 332                 file = fcheck_files(files, fd);
 333                 if (file) {
 334                         if (atomic_long_inc_not_zero(&file->f_count))
 335                                 *fput_needed = 1;
 336                         else
 337                                 /* Didn't get the reference, someone's freed */
 338                                 file = NULL;
 339                 }
 340                 rcu_read_unlock();
 341         }
 342
 343         return file;
 344 }
 345
 346
 347 void put_filp(struct file *file)
 348 {
 349         if (atomic_long_dec_and_test(&file->f_count)) {
 350                 security_file_free(file);
 351                 file_kill(file);
 352                 file_free(file);
 353         }
 354 }
 355
 356 void file_move(struct file *file, struct list_head *list)
 357 {
 358         if (!list)
 359                 return;
 360         file_list_lock();
 361         list_move(&file->f_u.fu_list, list);
 362         file_list_unlock();
 363 }
 364
 365 void file_kill(struct file *file)
 366 {
 367         if (!list_empty(&file->f_u.fu_list)) {
 368                 file_list_lock();
 369                 list_del_init(&file->f_u.fu_list);
 370                 file_list_unlock();
 371         }
 372 }
 373
 374 int fs_may_remount_ro(struct super_block *sb)
 375 {
 376         struct file *file;
 377
 378         /* Check that no files are currently opened for writing. */
 379         file_list_lock();
 380         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 381                 struct inode *inode = file->f_path.dentry->d_inode;
 382
 383                 /* File with pending delete? */
 384                 if (inode->i_nlink == 0)
 385                         goto too_bad;
 386
 387                 /* Writeable file? */
 388                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 389                         goto too_bad;
 390         }
 391         file_list_unlock();
 392         return 1; /* Tis' cool bro. */
 393 too_bad:
 394         file_list_unlock();
 395         return 0;
 396 }
 397
 398 /**
 399  *      mark_files_ro - mark all files read-only
 400  *      @sb: superblock in question
 401  *
 402  *      All files are marked read-only.  We don't care about pending
 403  *      delete files so this should be used in 'force' mode only.
 404  */
 405 void mark_files_ro(struct super_block *sb)
 406 {
 407         struct file *f;
 408
 409 retry:
 410         file_list_lock();
 411         list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
 412                 struct vfsmount *mnt;
 413                 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
 414                        continue;
 415                 if (!file_count(f))
 416                         continue;
 417                 if (!(f->f_mode & FMODE_WRITE))
 418                         continue;
 419                 f->f_mode &= ~FMODE_WRITE;
 420                 if (file_check_writeable(f) != 0)
 421                         continue;
 422                 file_release_write(f);
 423                 mnt = mntget(f->f_path.mnt);
 424                 file_list_unlock();
 425                 /*
 426                  * This can sleep, so we can't hold
 427                  * the file_list_lock() spinlock.
 428                  */
 429                 mnt_drop_write(mnt);
 430                 mntput(mnt);
 431                 goto retry;
 432         }
 433         file_list_unlock();
 434 }
 435
 436 void __init files_init(unsigned long mempages)
 437 {
 438         int n;
 439
 440         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 441                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 442
 443         /*
 444          * One file with associated inode and dcache is very roughly 1K.
 445          * Per default don't use more than 10% of our memory for files.
 446          */
 447
 448         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 449         files_stat.max_files = n;
 450         if (files_stat.max_files < NR_FILE)
 451                 files_stat.max_files = NR_FILE;
 452         files_defer_init();
 453         percpu_counter_init(&nr_files, 0);
 454 }