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 /* SLAB cache for file structures */
  37 static struct kmem_cache *filp_cachep __read_mostly;
  38
  39 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  40
  41 static inline void file_free_rcu(struct rcu_head *head)
  42 {
  43         struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  44
  45         put_cred(f->f_cred);
  46         kmem_cache_free(filp_cachep, f);
  47 }
  48
  49 static inline void file_free(struct file *f)
  50 {
  51         percpu_counter_dec(&nr_files);
  52         file_check_state(f);
  53         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  54 }
  55
  56 /*
  57  * Return the total number of open files in the system
  58  */
  59 static int get_nr_files(void)
  60 {
  61         return percpu_counter_read_positive(&nr_files);
  62 }
  63
  64 /*
  65  * Return the maximum number of open files in the system
  66  */
  67 int get_max_files(void)
  68 {
  69         return files_stat.max_files;
  70 }
  71 EXPORT_SYMBOL_GPL(get_max_files);
  72
  73 /*
  74  * Handle nr_files sysctl
  75  */
  76 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  77 int proc_nr_files(ctl_table *table, int write,
  78                      void __user *buffer, size_t *lenp, loff_t *ppos)
  79 {
  80         files_stat.nr_files = get_nr_files();
  81         return proc_dointvec(table, write, buffer, lenp, ppos);
  82 }
  83 #else
  84 int proc_nr_files(ctl_table *table, int write,
  85                      void __user *buffer, size_t *lenp, loff_t *ppos)
  86 {
  87         return -ENOSYS;
  88 }
  89 #endif
  90
  91 /* Find an unused file structure and return a pointer to it.
  92  * Returns NULL, if there are no more free file structures or
  93  * we run out of memory.
  94  *
  95  * Be very careful using this.  You are responsible for
  96  * getting write access to any mount that you might assign
  97  * to this filp, if it is opened for write.  If this is not
  98  * done, you will imbalance int the mount's writer count
  99  * and a warning at __fput() time.
 100  */
 101 struct file *get_empty_filp(void)
 102 {
 103         const struct cred *cred = current_cred();
 104         static int old_max;
 105         struct file * f;
 106
 107         /*
 108          * Privileged users can go above max_files
 109          */
 110         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 111                 /*
 112                  * percpu_counters are inaccurate.  Do an expensive check before
 113                  * we go and fail.
 114                  */
 115                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 116                         goto over;
 117         }
 118
 119         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 120         if (f == NULL)
 121                 goto fail;
 122
 123         percpu_counter_inc(&nr_files);
 124         if (security_file_alloc(f))
 125                 goto fail_sec;
 126
 127         INIT_LIST_HEAD(&f->f_u.fu_list);
 128         atomic_long_set(&f->f_count, 1);
 129         rwlock_init(&f->f_owner.lock);
 130         f->f_cred = get_cred(cred);
 131         spin_lock_init(&f->f_lock);
 132         eventpoll_init_file(f);
 133         /* f->f_version: 0 */
 134         return f;
 135
 136 over:
 137         /* Ran out of filps - report that */
 138         if (get_nr_files() > old_max) {
 139                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
 140                                         get_max_files());
 141                 old_max = get_nr_files();
 142         }
 143         goto fail;
 144
 145 fail_sec:
 146         file_free(f);
 147 fail:
 148         return NULL;
 149 }
 150
 151 EXPORT_SYMBOL(get_empty_filp);
 152
 153 /**
 154  * alloc_file - allocate and initialize a 'struct file'
 155  * @mnt: the vfsmount on which the file will reside
 156  * @dentry: the dentry representing the new file
 157  * @mode: the mode with which the new file will be opened
 158  * @fop: the 'struct file_operations' for the new file
 159  *
 160  * Use this instead of get_empty_filp() to get a new
 161  * 'struct file'.  Do so because of the same initialization
 162  * pitfalls reasons listed for init_file().  This is a
 163  * preferred interface to using init_file().
 164  *
 165  * If all the callers of init_file() are eliminated, its
 166  * code should be moved into this function.
 167  */
 168 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
 169                 fmode_t mode, const struct file_operations *fop)
 170 {
 171         struct file *file;
 172
 173         file = get_empty_filp();
 174         if (!file)
 175                 return NULL;
 176
 177         init_file(file, mnt, dentry, mode, fop);
 178         return file;
 179 }
 180 EXPORT_SYMBOL(alloc_file);
 181
 182 /**
 183  * init_file - initialize a 'struct file'
 184  * @file: the already allocated 'struct file' to initialized
 185  * @mnt: the vfsmount on which the file resides
 186  * @dentry: the dentry representing this file
 187  * @mode: the mode the file is opened with
 188  * @fop: the 'struct file_operations' for this file
 189  *
 190  * Use this instead of setting the members directly.  Doing so
 191  * avoids making mistakes like forgetting the mntget() or
 192  * forgetting to take a write on the mnt.
 193  *
 194  * Note: This is a crappy interface.  It is here to make
 195  * merging with the existing users of get_empty_filp()
 196  * who have complex failure logic easier.  All users
 197  * of this should be moving to alloc_file().
 198  */
 199 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
 200            fmode_t mode, const struct file_operations *fop)
 201 {
 202         int error = 0;
 203         file->f_path.dentry = dentry;
 204         file->f_path.mnt = mntget(mnt);
 205         file->f_mapping = dentry->d_inode->i_mapping;
 206         file->f_mode = mode;
 207         file->f_op = fop;
 208
 209         /*
 210          * These mounts don't really matter in practice
 211          * for r/o bind mounts.  They aren't userspace-
 212          * visible.  We do this for consistency, and so
 213          * that we can do debugging checks at __fput()
 214          */
 215         if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
 216                 file_take_write(file);
 217                 error = mnt_clone_write(mnt);
 218                 WARN_ON(error);
 219         }
 220         return error;
 221 }
 222 EXPORT_SYMBOL(init_file);
 223
 224 void fput(struct file *file)
 225 {
 226         if (atomic_long_dec_and_test(&file->f_count))
 227                 __fput(file);
 228 }
 229
 230 EXPORT_SYMBOL(fput);
 231
 232 /**
 233  * drop_file_write_access - give up ability to write to a file
 234  * @file: the file to which we will stop writing
 235  *
 236  * This is a central place which will give up the ability
 237  * to write to @file, along with access to write through
 238  * its vfsmount.
 239  */
 240 void drop_file_write_access(struct file *file)
 241 {
 242         struct vfsmount *mnt = file->f_path.mnt;
 243         struct dentry *dentry = file->f_path.dentry;
 244         struct inode *inode = dentry->d_inode;
 245
 246         put_write_access(inode);
 247
 248         if (special_file(inode->i_mode))
 249                 return;
 250         if (file_check_writeable(file) != 0)
 251                 return;
 252         mnt_drop_write(mnt);
 253         file_release_write(file);
 254 }
 255 EXPORT_SYMBOL_GPL(drop_file_write_access);
 256
 257 /* __fput is called from task context when aio completion releases the last
 258  * last use of a struct file *.  Do not use otherwise.
 259  */
 260 void __fput(struct file *file)
 261 {
 262         struct dentry *dentry = file->f_path.dentry;
 263         struct vfsmount *mnt = file->f_path.mnt;
 264         struct inode *inode = dentry->d_inode;
 265
 266         might_sleep();
 267
 268         fsnotify_close(file);
 269         /*
 270          * The function eventpoll_release() should be the first called
 271          * in the file cleanup chain.
 272          */
 273         eventpoll_release(file);
 274         locks_remove_flock(file);
 275
 276         if (unlikely(file->f_flags & FASYNC)) {
 277                 if (file->f_op && file->f_op->fasync)
 278                         file->f_op->fasync(-1, file, 0);
 279         }
 280         if (file->f_op && file->f_op->release)
 281                 file->f_op->release(inode, file);
 282         security_file_free(file);
 283         ima_file_free(file);
 284         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 285                 cdev_put(inode->i_cdev);
 286         fops_put(file->f_op);
 287         put_pid(file->f_owner.pid);
 288         file_kill(file);
 289         if (file->f_mode & FMODE_WRITE)
 290                 drop_file_write_access(file);
 291         file->f_path.dentry = NULL;
 292         file->f_path.mnt = NULL;
 293         file_free(file);
 294         dput(dentry);
 295         mntput(mnt);
 296 }
 297
 298 struct file *fget(unsigned int fd)
 299 {
 300         struct file *file;
 301         struct files_struct *files = current->files;
 302
 303         rcu_read_lock();
 304         file = fcheck_files(files, fd);
 305         if (file) {
 306                 if (!atomic_long_inc_not_zero(&file->f_count)) {
 307                         /* File object ref couldn't be taken */
 308                         rcu_read_unlock();
 309                         return NULL;
 310                 }
 311         }
 312         rcu_read_unlock();
 313
 314         return file;
 315 }
 316
 317 EXPORT_SYMBOL(fget);
 318
 319 /*
 320  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 321  *
 322  * You can use this instead of fget if you satisfy all of the following
 323  * conditions:
 324  * 1) You must call fput_light before exiting the syscall and returning control
 325  *    to userspace (i.e. you cannot remember the returned struct file * after
 326  *    returning to userspace).
 327  * 2) You must not call filp_close on the returned struct file * in between
 328  *    calls to fget_light and fput_light.
 329  * 3) You must not clone the current task in between the calls to fget_light
 330  *    and fput_light.
 331  *
 332  * The fput_needed flag returned by fget_light should be passed to the
 333  * corresponding fput_light.
 334  */
 335 struct file *fget_light(unsigned int fd, int *fput_needed)
 336 {
 337         struct file *file;
 338         struct files_struct *files = current->files;
 339
 340         *fput_needed = 0;
 341         if (likely((atomic_read(&files->count) == 1))) {
 342                 file = fcheck_files(files, fd);
 343         } else {
 344                 rcu_read_lock();
 345                 file = fcheck_files(files, fd);
 346                 if (file) {
 347                         if (atomic_long_inc_not_zero(&file->f_count))
 348                                 *fput_needed = 1;
 349                         else
 350                                 /* Didn't get the reference, someone's freed */
 351                                 file = NULL;
 352                 }
 353                 rcu_read_unlock();
 354         }
 355
 356         return file;
 357 }
 358
 359
 360 void put_filp(struct file *file)
 361 {
 362         if (atomic_long_dec_and_test(&file->f_count)) {
 363                 security_file_free(file);
 364                 file_kill(file);
 365                 file_free(file);
 366         }
 367 }
 368
 369 void file_move(struct file *file, struct list_head *list)
 370 {
 371         if (!list)
 372                 return;
 373         file_list_lock();
 374         list_move(&file->f_u.fu_list, list);
 375         file_list_unlock();
 376 }
 377
 378 void file_kill(struct file *file)
 379 {
 380         if (!list_empty(&file->f_u.fu_list)) {
 381                 file_list_lock();
 382                 list_del_init(&file->f_u.fu_list);
 383                 file_list_unlock();
 384         }
 385 }
 386
 387 int fs_may_remount_ro(struct super_block *sb)
 388 {
 389         struct file *file;
 390
 391         /* Check that no files are currently opened for writing. */
 392         file_list_lock();
 393         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 394                 struct inode *inode = file->f_path.dentry->d_inode;
 395
 396                 /* File with pending delete? */
 397                 if (inode->i_nlink == 0)
 398                         goto too_bad;
 399
 400                 /* Writeable file? */
 401                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 402                         goto too_bad;
 403         }
 404         file_list_unlock();
 405         return 1; /* Tis' cool bro. */
 406 too_bad:
 407         file_list_unlock();
 408         return 0;
 409 }
 410
 411 /**
 412  *      mark_files_ro - mark all files read-only
 413  *      @sb: superblock in question
 414  *
 415  *      All files are marked read-only.  We don't care about pending
 416  *      delete files so this should be used in 'force' mode only.
 417  */
 418 void mark_files_ro(struct super_block *sb)
 419 {
 420         struct file *f;
 421
 422 retry:
 423         file_list_lock();
 424         list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
 425                 struct vfsmount *mnt;
 426                 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
 427                        continue;
 428                 if (!file_count(f))
 429                         continue;
 430                 if (!(f->f_mode & FMODE_WRITE))
 431                         continue;
 432                 f->f_mode &= ~FMODE_WRITE;
 433                 if (file_check_writeable(f) != 0)
 434                         continue;
 435                 file_release_write(f);
 436                 mnt = mntget(f->f_path.mnt);
 437                 file_list_unlock();
 438                 /*
 439                  * This can sleep, so we can't hold
 440                  * the file_list_lock() spinlock.
 441                  */
 442                 mnt_drop_write(mnt);
 443                 mntput(mnt);
 444                 goto retry;
 445         }
 446         file_list_unlock();
 447 }
 448
 449 void __init files_init(unsigned long mempages)
 450 {
 451         int n;
 452
 453         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 454                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 455
 456         /*
 457          * One file with associated inode and dcache is very roughly 1K.
 458          * Per default don't use more than 10% of our memory for files.
 459          */
 460
 461         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 462         files_stat.max_files = n;
 463         if (files_stat.max_files < NR_FILE)
 464                 files_stat.max_files = NR_FILE;
 465         files_defer_init();
 466         percpu_counter_init(&nr_files, 0);
 467 }