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, struct file *filp,
  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, filp, buffer, lenp, ppos);
  82 }
  83 #else
  84 int proc_nr_files(ctl_table *table, int write, struct file *filp,
  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         struct path;
 173
 174         file = get_empty_filp();
 175         if (!file)
 176                 return NULL;
 177
 178         init_file(file, mnt, dentry, mode, fop);
 179         return file;
 180 }
 181 EXPORT_SYMBOL(alloc_file);
 182
 183 /**
 184  * init_file - initialize a 'struct file'
 185  * @file: the already allocated 'struct file' to initialized
 186  * @mnt: the vfsmount on which the file resides
 187  * @dentry: the dentry representing this file
 188  * @mode: the mode the file is opened with
 189  * @fop: the 'struct file_operations' for this file
 190  *
 191  * Use this instead of setting the members directly.  Doing so
 192  * avoids making mistakes like forgetting the mntget() or
 193  * forgetting to take a write on the mnt.
 194  *
 195  * Note: This is a crappy interface.  It is here to make
 196  * merging with the existing users of get_empty_filp()
 197  * who have complex failure logic easier.  All users
 198  * of this should be moving to alloc_file().
 199  */
 200 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
 201            fmode_t mode, const struct file_operations *fop)
 202 {
 203         int error = 0;
 204         file->f_path.dentry = dentry;
 205         file->f_path.mnt = mntget(mnt);
 206         file->f_mapping = dentry->d_inode->i_mapping;
 207         file->f_mode = mode;
 208         file->f_op = fop;
 209
 210         /*
 211          * These mounts don't really matter in practice
 212          * for r/o bind mounts.  They aren't userspace-
 213          * visible.  We do this for consistency, and so
 214          * that we can do debugging checks at __fput()
 215          */
 216         if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
 217                 file_take_write(file);
 218                 error = mnt_want_write(mnt);
 219                 WARN_ON(error);
 220         }
 221         return error;
 222 }
 223 EXPORT_SYMBOL(init_file);
 224
 225 void fput(struct file *file)
 226 {
 227         if (atomic_long_dec_and_test(&file->f_count))
 228                 __fput(file);
 229 }
 230
 231 EXPORT_SYMBOL(fput);
 232
 233 /**
 234  * drop_file_write_access - give up ability to write to a file
 235  * @file: the file to which we will stop writing
 236  *
 237  * This is a central place which will give up the ability
 238  * to write to @file, along with access to write through
 239  * its vfsmount.
 240  */
 241 void drop_file_write_access(struct file *file)
 242 {
 243         struct vfsmount *mnt = file->f_path.mnt;
 244         struct dentry *dentry = file->f_path.dentry;
 245         struct inode *inode = dentry->d_inode;
 246
 247         put_write_access(inode);
 248
 249         if (special_file(inode->i_mode))
 250                 return;
 251         if (file_check_writeable(file) != 0)
 252                 return;
 253         mnt_drop_write(mnt);
 254         file_release_write(file);
 255 }
 256 EXPORT_SYMBOL_GPL(drop_file_write_access);
 257
 258 /* __fput is called from task context when aio completion releases the last
 259  * last use of a struct file *.  Do not use otherwise.
 260  */
 261 void __fput(struct file *file)
 262 {
 263         struct dentry *dentry = file->f_path.dentry;
 264         struct vfsmount *mnt = file->f_path.mnt;
 265         struct inode *inode = dentry->d_inode;
 266
 267         might_sleep();
 268
 269         fsnotify_close(file);
 270         /*
 271          * The function eventpoll_release() should be the first called
 272          * in the file cleanup chain.
 273          */
 274         eventpoll_release(file);
 275         locks_remove_flock(file);
 276
 277         if (unlikely(file->f_flags & FASYNC)) {
 278                 if (file->f_op && file->f_op->fasync)
 279                         file->f_op->fasync(-1, file, 0);
 280         }
 281         if (file->f_op && file->f_op->release)
 282                 file->f_op->release(inode, file);
 283         security_file_free(file);
 284         ima_file_free(file);
 285         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 286                 cdev_put(inode->i_cdev);
 287         fops_put(file->f_op);
 288         put_pid(file->f_owner.pid);
 289         file_kill(file);
 290         if (file->f_mode & FMODE_WRITE)
 291                 drop_file_write_access(file);
 292         file->f_path.dentry = NULL;
 293         file->f_path.mnt = NULL;
 294         file_free(file);
 295         dput(dentry);
 296         mntput(mnt);
 297 }
 298
 299 struct file *fget(unsigned int fd)
 300 {
 301         struct file *file;
 302         struct files_struct *files = current->files;
 303
 304         rcu_read_lock();
 305         file = fcheck_files(files, fd);
 306         if (file) {
 307                 if (!atomic_long_inc_not_zero(&file->f_count)) {
 308                         /* File object ref couldn't be taken */
 309                         rcu_read_unlock();
 310                         return NULL;
 311                 }
 312         }
 313         rcu_read_unlock();
 314
 315         return file;
 316 }
 317
 318 EXPORT_SYMBOL(fget);
 319
 320 /*
 321  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 322  * You can use this only if it is guranteed that the current task already
 323  * holds a refcnt to that file. That check has to be done at fget() only
 324  * and a flag is returned to be passed to the corresponding fput_light().
 325  * There must not be a cloning between an fget_light/fput_light pair.
 326  */
 327 struct file *fget_light(unsigned int fd, int *fput_needed)
 328 {
 329         struct file *file;
 330         struct files_struct *files = current->files;
 331
 332         *fput_needed = 0;
 333         if (likely((atomic_read(&files->count) == 1))) {
 334                 file = fcheck_files(files, fd);
 335         } else {
 336                 rcu_read_lock();
 337                 file = fcheck_files(files, fd);
 338                 if (file) {
 339                         if (atomic_long_inc_not_zero(&file->f_count))
 340                                 *fput_needed = 1;
 341                         else
 342                                 /* Didn't get the reference, someone's freed */
 343                                 file = NULL;
 344                 }
 345                 rcu_read_unlock();
 346         }
 347
 348         return file;
 349 }
 350
 351
 352 void put_filp(struct file *file)
 353 {
 354         if (atomic_long_dec_and_test(&file->f_count)) {
 355                 security_file_free(file);
 356                 file_kill(file);
 357                 file_free(file);
 358         }
 359 }
 360
 361 void file_move(struct file *file, struct list_head *list)
 362 {
 363         if (!list)
 364                 return;
 365         file_list_lock();
 366         list_move(&file->f_u.fu_list, list);
 367         file_list_unlock();
 368 }
 369
 370 void file_kill(struct file *file)
 371 {
 372         if (!list_empty(&file->f_u.fu_list)) {
 373                 file_list_lock();
 374                 list_del_init(&file->f_u.fu_list);
 375                 file_list_unlock();
 376         }
 377 }
 378
 379 int fs_may_remount_ro(struct super_block *sb)
 380 {
 381         struct file *file;
 382
 383         /* Check that no files are currently opened for writing. */
 384         file_list_lock();
 385         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 386                 struct inode *inode = file->f_path.dentry->d_inode;
 387
 388                 /* File with pending delete? */
 389                 if (inode->i_nlink == 0)
 390                         goto too_bad;
 391
 392                 /* Writeable file? */
 393                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 394                         goto too_bad;
 395         }
 396         file_list_unlock();
 397         return 1; /* Tis' cool bro. */
 398 too_bad:
 399         file_list_unlock();
 400         return 0;
 401 }
 402
 403 void __init files_init(unsigned long mempages)
 404 {
 405         int n;
 406
 407         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 408                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 409
 410         /*
 411          * One file with associated inode and dcache is very roughly 1K.
 412          * Per default don't use more than 10% of our memory for files.
 413          */
 414
 415         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 416         files_stat.max_files = n;
 417         if (files_stat.max_files < NR_FILE)
 418                 files_stat.max_files = NR_FILE;
 419         files_defer_init();
 420         percpu_counter_init(&nr_files, 0);
 421 }