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