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 EXPORT_SYMBOL(get_empty_filp);
 151
 152 /**
 153  * alloc_file - allocate and initialize a 'struct file'
 154  * @mnt: the vfsmount on which the file will reside
 155  * @dentry: the dentry representing the new file
 156  * @mode: the mode with which the new file will be opened
 157  * @fop: the 'struct file_operations' for the new file
 158  *
 159  * Use this instead of get_empty_filp() to get a new
 160  * 'struct file'.  Do so because of the same initialization
 161  * pitfalls reasons listed for init_file().  This is a
 162  * preferred interface to using init_file().
 163  *
 164  * If all the callers of init_file() are eliminated, its
 165  * code should be moved into this function.
 166  */
 167 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
 168                 fmode_t mode, const struct file_operations *fop)
 169 {
 170         struct file *file;
 171
 172         file = get_empty_filp();
 173         if (!file)
 174                 return NULL;
 175
 176         init_file(file, mnt, dentry, mode, fop);
 177         return file;
 178 }
 179 EXPORT_SYMBOL(alloc_file);
 180
 181 /**
 182  * init_file - initialize a 'struct file'
 183  * @file: the already allocated 'struct file' to initialized
 184  * @mnt: the vfsmount on which the file resides
 185  * @dentry: the dentry representing this file
 186  * @mode: the mode the file is opened with
 187  * @fop: the 'struct file_operations' for this file
 188  *
 189  * Use this instead of setting the members directly.  Doing so
 190  * avoids making mistakes like forgetting the mntget() or
 191  * forgetting to take a write on the mnt.
 192  *
 193  * Note: This is a crappy interface.  It is here to make
 194  * merging with the existing users of get_empty_filp()
 195  * who have complex failure logic easier.  All users
 196  * of this should be moving to alloc_file().
 197  */
 198 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
 199            fmode_t mode, const struct file_operations *fop)
 200 {
 201         int error = 0;
 202         file->f_path.dentry = dentry;
 203         file->f_path.mnt = mntget(mnt);
 204         file->f_mapping = dentry->d_inode->i_mapping;
 205         file->f_mode = mode;
 206         file->f_op = fop;
 207
 208         /*
 209          * These mounts don't really matter in practice
 210          * for r/o bind mounts.  They aren't userspace-
 211          * visible.  We do this for consistency, and so
 212          * that we can do debugging checks at __fput()
 213          */
 214         if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
 215                 file_take_write(file);
 216                 error = mnt_clone_write(mnt);
 217                 WARN_ON(error);
 218         }
 219         return error;
 220 }
 221 EXPORT_SYMBOL(init_file);
 222
 223 void fput(struct file *file)
 224 {
 225         if (atomic_long_dec_and_test(&file->f_count))
 226                 __fput(file);
 227 }
 228
 229 EXPORT_SYMBOL(fput);
 230
 231 /**
 232  * drop_file_write_access - give up ability to write to a file
 233  * @file: the file to which we will stop writing
 234  *
 235  * This is a central place which will give up the ability
 236  * to write to @file, along with access to write through
 237  * its vfsmount.
 238  */
 239 void drop_file_write_access(struct file *file)
 240 {
 241         struct vfsmount *mnt = file->f_path.mnt;
 242         struct dentry *dentry = file->f_path.dentry;
 243         struct inode *inode = dentry->d_inode;
 244
 245         put_write_access(inode);
 246
 247         if (special_file(inode->i_mode))
 248                 return;
 249         if (file_check_writeable(file) != 0)
 250                 return;
 251         mnt_drop_write(mnt);
 252         file_release_write(file);
 253 }
 254 EXPORT_SYMBOL_GPL(drop_file_write_access);
 255
 256 /* __fput is called from task context when aio completion releases the last
 257  * last use of a struct file *.  Do not use otherwise.
 258  */
 259 void __fput(struct file *file)
 260 {
 261         struct dentry *dentry = file->f_path.dentry;
 262         struct vfsmount *mnt = file->f_path.mnt;
 263         struct inode *inode = dentry->d_inode;
 264
 265         might_sleep();
 266
 267         fsnotify_close(file);
 268         /*
 269          * The function eventpoll_release() should be the first called
 270          * in the file cleanup chain.
 271          */
 272         eventpoll_release(file);
 273         locks_remove_flock(file);
 274
 275         if (unlikely(file->f_flags & FASYNC)) {
 276                 if (file->f_op && file->f_op->fasync)
 277                         file->f_op->fasync(-1, file, 0);
 278         }
 279         if (file->f_op && file->f_op->release)
 280                 file->f_op->release(inode, file);
 281         security_file_free(file);
 282         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 283                 cdev_put(inode->i_cdev);
 284         fops_put(file->f_op);
 285         put_pid(file->f_owner.pid);
 286         file_kill(file);
 287         if (file->f_mode & FMODE_WRITE)
 288                 drop_file_write_access(file);
 289         file->f_path.dentry = NULL;
 290         file->f_path.mnt = NULL;
 291         file_free(file);
 292         dput(dentry);
 293         mntput(mnt);
 294 }
 295
 296 struct file *fget(unsigned int fd)
 297 {
 298         struct file *file;
 299         struct files_struct *files = current->files;
 300
 301         rcu_read_lock();
 302         file = fcheck_files(files, fd);
 303         if (file) {
 304                 if (!atomic_long_inc_not_zero(&file->f_count)) {
 305                         /* File object ref couldn't be taken */
 306                         rcu_read_unlock();
 307                         return NULL;
 308                 }
 309         }
 310         rcu_read_unlock();
 311
 312         return file;
 313 }
 314
 315 EXPORT_SYMBOL(fget);
 316
 317 /*
 318  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 319  * You can use this only if it is guranteed that the current task already
 320  * holds a refcnt to that file. That check has to be done at fget() only
 321  * and a flag is returned to be passed to the corresponding fput_light().
 322  * There must not be a cloning between an fget_light/fput_light pair.
 323  */
 324 struct file *fget_light(unsigned int fd, int *fput_needed)
 325 {
 326         struct file *file;
 327         struct files_struct *files = current->files;
 328
 329         *fput_needed = 0;
 330         if (likely((atomic_read(&files->count) == 1))) {
 331                 file = fcheck_files(files, fd);
 332         } else {
 333                 rcu_read_lock();
 334                 file = fcheck_files(files, fd);
 335                 if (file) {
 336                         if (atomic_long_inc_not_zero(&file->f_count))
 337                                 *fput_needed = 1;
 338                         else
 339                                 /* Didn't get the reference, someone's freed */
 340                                 file = NULL;
 341                 }
 342                 rcu_read_unlock();
 343         }
 344
 345         return file;
 346 }
 347
 348
 349 void put_filp(struct file *file)
 350 {
 351         if (atomic_long_dec_and_test(&file->f_count)) {
 352                 security_file_free(file);
 353                 file_kill(file);
 354                 file_free(file);
 355         }
 356 }
 357
 358 void file_move(struct file *file, struct list_head *list)
 359 {
 360         if (!list)
 361                 return;
 362         file_list_lock();
 363         list_move(&file->f_u.fu_list, list);
 364         file_list_unlock();
 365 }
 366
 367 void file_kill(struct file *file)
 368 {
 369         if (!list_empty(&file->f_u.fu_list)) {
 370                 file_list_lock();
 371                 list_del_init(&file->f_u.fu_list);
 372                 file_list_unlock();
 373         }
 374 }
 375
 376 int fs_may_remount_ro(struct super_block *sb)
 377 {
 378         struct file *file;
 379
 380         /* Check that no files are currently opened for writing. */
 381         file_list_lock();
 382         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 383                 struct inode *inode = file->f_path.dentry->d_inode;
 384
 385                 /* File with pending delete? */
 386                 if (inode->i_nlink == 0)
 387                         goto too_bad;
 388
 389                 /* Writeable file? */
 390                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 391                         goto too_bad;
 392         }
 393         file_list_unlock();
 394         return 1; /* Tis' cool bro. */
 395 too_bad:
 396         file_list_unlock();
 397         return 0;
 398 }
 399
 400 /**
 401  *      mark_files_ro - mark all files read-only
 402  *      @sb: superblock in question
 403  *
 404  *      All files are marked read-only.  We don't care about pending
 405  *      delete files so this should be used in 'force' mode only.
 406  */
 407 void mark_files_ro(struct super_block *sb)
 408 {
 409         struct file *f;
 410
 411 retry:
 412         file_list_lock();
 413         list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
 414                 struct vfsmount *mnt;
 415                 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
 416                        continue;
 417                 if (!file_count(f))
 418                         continue;
 419                 if (!(f->f_mode & FMODE_WRITE))
 420                         continue;
 421                 f->f_mode &= ~FMODE_WRITE;
 422                 if (file_check_writeable(f) != 0)
 423                         continue;
 424                 file_release_write(f);
 425                 mnt = mntget(f->f_path.mnt);
 426                 file_list_unlock();
 427                 /*
 428                  * This can sleep, so we can't hold
 429                  * the file_list_lock() spinlock.
 430                  */
 431                 mnt_drop_write(mnt);
 432                 mntput(mnt);
 433                 goto retry;
 434         }
 435         file_list_unlock();
 436 }
 437
 438 void __init files_init(unsigned long mempages)
 439 {
 440         int n;
 441
 442         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 443                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 444
 445         /*
 446          * One file with associated inode and dcache is very roughly 1K.
 447          * Per default don't use more than 10% of our memory for files.
 448          */
 449
 450         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 451         files_stat.max_files = n;
 452         if (files_stat.max_files < NR_FILE)
 453                 files_stat.max_files = NR_FILE;
 454         files_defer_init();
 455         percpu_counter_init(&nr_files, 0);
 456 }