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 #include <linux/ima.h>
  25
  26 #include <asm/atomic.h>
  27
  28 #include "internal.h"
  29
  30 /* sysctl tunables... */
  31 struct files_stat_struct files_stat = {
  32         .max_files = NR_FILE
  33 };
  34
  35 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
  36
  37 /* SLAB cache for file structures */
  38 static struct kmem_cache *filp_cachep __read_mostly;
  39
  40 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  41
  42 static inline void file_free_rcu(struct rcu_head *head)
  43 {
  44         struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  45
  46         put_cred(f->f_cred);
  47         kmem_cache_free(filp_cachep, f);
  48 }
  49
  50 static inline void file_free(struct file *f)
  51 {
  52         percpu_counter_dec(&nr_files);
  53         file_check_state(f);
  54         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  55 }
  56
  57 /*
  58  * Return the total number of open files in the system
  59  */
  60 static int get_nr_files(void)
  61 {
  62         return percpu_counter_read_positive(&nr_files);
  63 }
  64
  65 /*
  66  * Return the maximum number of open files in the system
  67  */
  68 int get_max_files(void)
  69 {
  70         return files_stat.max_files;
  71 }
  72 EXPORT_SYMBOL_GPL(get_max_files);
  73
  74 /*
  75  * Handle nr_files sysctl
  76  */
  77 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  78 int proc_nr_files(ctl_table *table, int write,
  79                      void __user *buffer, size_t *lenp, loff_t *ppos)
  80 {
  81         files_stat.nr_files = get_nr_files();
  82         return proc_dointvec(table, write, buffer, lenp, ppos);
  83 }
  84 #else
  85 int proc_nr_files(ctl_table *table, int write,
  86                      void __user *buffer, size_t *lenp, loff_t *ppos)
  87 {
  88         return -ENOSYS;
  89 }
  90 #endif
  91
  92 /* Find an unused file structure and return a pointer to it.
  93  * Returns NULL, if there are no more free file structures or
  94  * we run out of memory.
  95  *
  96  * Be very careful using this.  You are responsible for
  97  * getting write access to any mount that you might assign
  98  * to this filp, if it is opened for write.  If this is not
  99  * done, you will imbalance int the mount's writer count
 100  * and a warning at __fput() time.
 101  */
 102 struct file *get_empty_filp(void)
 103 {
 104         const struct cred *cred = current_cred();
 105         static int old_max;
 106         struct file * f;
 107
 108         /*
 109          * Privileged users can go above max_files
 110          */
 111         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 112                 /*
 113                  * percpu_counters are inaccurate.  Do an expensive check before
 114                  * we go and fail.
 115                  */
 116                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 117                         goto over;
 118         }
 119
 120         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 121         if (f == NULL)
 122                 goto fail;
 123
 124         percpu_counter_inc(&nr_files);
 125         if (security_file_alloc(f))
 126                 goto fail_sec;
 127
 128         INIT_LIST_HEAD(&f->f_u.fu_list);
 129         atomic_long_set(&f->f_count, 1);
 130         rwlock_init(&f->f_owner.lock);
 131         f->f_cred = get_cred(cred);
 132         spin_lock_init(&f->f_lock);
 133         eventpoll_init_file(f);
 134         /* f->f_version: 0 */
 135         return f;
 136
 137 over:
 138         /* Ran out of filps - report that */
 139         if (get_nr_files() > old_max) {
 140                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
 141                                         get_max_files());
 142                 old_max = get_nr_files();
 143         }
 144         goto fail;
 145
 146 fail_sec:
 147         file_free(f);
 148 fail:
 149         return NULL;
 150 }
 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 path *path, fmode_t mode,
 168                 const struct file_operations *fop)
 169 {
 170         struct file *file;
 171
 172         file = get_empty_filp();
 173         if (!file)
 174                 return NULL;
 175
 176         file->f_path = *path;
 177         file->f_mapping = path->dentry->d_inode->i_mapping;
 178         file->f_mode = mode;
 179         file->f_op = fop;
 180
 181         /*
 182          * These mounts don't really matter in practice
 183          * for r/o bind mounts.  They aren't userspace-
 184          * visible.  We do this for consistency, and so
 185          * that we can do debugging checks at __fput()
 186          */
 187         if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
 188                 file_take_write(file);
 189                 WARN_ON(mnt_clone_write(path->mnt));
 190         }
 191         ima_counts_get(file);
 192         return file;
 193 }
 194 EXPORT_SYMBOL(alloc_file);
 195
 196 /**
 197  * drop_file_write_access - give up ability to write to a file
 198  * @file: the file to which we will stop writing
 199  *
 200  * This is a central place which will give up the ability
 201  * to write to @file, along with access to write through
 202  * its vfsmount.
 203  */
 204 void drop_file_write_access(struct file *file)
 205 {
 206         struct vfsmount *mnt = file->f_path.mnt;
 207         struct dentry *dentry = file->f_path.dentry;
 208         struct inode *inode = dentry->d_inode;
 209
 210         put_write_access(inode);
 211
 212         if (special_file(inode->i_mode))
 213                 return;
 214         if (file_check_writeable(file) != 0)
 215                 return;
 216         mnt_drop_write(mnt);
 217         file_release_write(file);
 218 }
 219 EXPORT_SYMBOL_GPL(drop_file_write_access);
 220
 221 /* the real guts of fput() - releasing the last reference to file
 222  */
 223 static void __fput(struct file *file)
 224 {
 225         struct dentry *dentry = file->f_path.dentry;
 226         struct vfsmount *mnt = file->f_path.mnt;
 227         struct inode *inode = dentry->d_inode;
 228
 229         might_sleep();
 230
 231         fsnotify_close(file);
 232         /*
 233          * The function eventpoll_release() should be the first called
 234          * in the file cleanup chain.
 235          */
 236         eventpoll_release(file);
 237         locks_remove_flock(file);
 238
 239         if (unlikely(file->f_flags & FASYNC)) {
 240                 if (file->f_op && file->f_op->fasync)
 241                         file->f_op->fasync(-1, file, 0);
 242         }
 243         if (file->f_op && file->f_op->release)
 244                 file->f_op->release(inode, file);
 245         security_file_free(file);
 246         ima_file_free(file);
 247         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 248                 cdev_put(inode->i_cdev);
 249         fops_put(file->f_op);
 250         put_pid(file->f_owner.pid);
 251         file_sb_list_del(file);
 252         if (file->f_mode & FMODE_WRITE)
 253                 drop_file_write_access(file);
 254         file->f_path.dentry = NULL;
 255         file->f_path.mnt = NULL;
 256         file_free(file);
 257         dput(dentry);
 258         mntput(mnt);
 259 }
 260
 261 void fput(struct file *file)
 262 {
 263         if (atomic_long_dec_and_test(&file->f_count))
 264                 __fput(file);
 265 }
 266
 267 EXPORT_SYMBOL(fput);
 268
 269 struct file *fget(unsigned int fd)
 270 {
 271         struct file *file;
 272         struct files_struct *files = current->files;
 273
 274         rcu_read_lock();
 275         file = fcheck_files(files, fd);
 276         if (file) {
 277                 if (!atomic_long_inc_not_zero(&file->f_count)) {
 278                         /* File object ref couldn't be taken */
 279                         rcu_read_unlock();
 280                         return NULL;
 281                 }
 282         }
 283         rcu_read_unlock();
 284
 285         return file;
 286 }
 287
 288 EXPORT_SYMBOL(fget);
 289
 290 /*
 291  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 292  *
 293  * You can use this instead of fget if you satisfy all of the following
 294  * conditions:
 295  * 1) You must call fput_light before exiting the syscall and returning control
 296  *    to userspace (i.e. you cannot remember the returned struct file * after
 297  *    returning to userspace).
 298  * 2) You must not call filp_close on the returned struct file * in between
 299  *    calls to fget_light and fput_light.
 300  * 3) You must not clone the current task in between the calls to fget_light
 301  *    and fput_light.
 302  *
 303  * The fput_needed flag returned by fget_light should be passed to the
 304  * corresponding fput_light.
 305  */
 306 struct file *fget_light(unsigned int fd, int *fput_needed)
 307 {
 308         struct file *file;
 309         struct files_struct *files = current->files;
 310
 311         *fput_needed = 0;
 312         if (likely((atomic_read(&files->count) == 1))) {
 313                 file = fcheck_files(files, fd);
 314         } else {
 315                 rcu_read_lock();
 316                 file = fcheck_files(files, fd);
 317                 if (file) {
 318                         if (atomic_long_inc_not_zero(&file->f_count))
 319                                 *fput_needed = 1;
 320                         else
 321                                 /* Didn't get the reference, someone's freed */
 322                                 file = NULL;
 323                 }
 324                 rcu_read_unlock();
 325         }
 326
 327         return file;
 328 }
 329
 330 void put_filp(struct file *file)
 331 {
 332         if (atomic_long_dec_and_test(&file->f_count)) {
 333                 security_file_free(file);
 334                 file_sb_list_del(file);
 335                 file_free(file);
 336         }
 337 }
 338
 339 void file_sb_list_add(struct file *file, struct super_block *sb)
 340 {
 341         spin_lock(&files_lock);
 342         BUG_ON(!list_empty(&file->f_u.fu_list));
 343         list_add(&file->f_u.fu_list, &sb->s_files);
 344         spin_unlock(&files_lock);
 345 }
 346
 347 void file_sb_list_del(struct file *file)
 348 {
 349         if (!list_empty(&file->f_u.fu_list)) {
 350                 spin_lock(&files_lock);
 351                 list_del_init(&file->f_u.fu_list);
 352                 spin_unlock(&files_lock);
 353         }
 354 }
 355
 356 int fs_may_remount_ro(struct super_block *sb)
 357 {
 358         struct file *file;
 359
 360         /* Check that no files are currently opened for writing. */
 361         spin_lock(&files_lock);
 362         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 363                 struct inode *inode = file->f_path.dentry->d_inode;
 364
 365                 /* File with pending delete? */
 366                 if (inode->i_nlink == 0)
 367                         goto too_bad;
 368
 369                 /* Writeable file? */
 370                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 371                         goto too_bad;
 372         }
 373         spin_unlock(&files_lock);
 374         return 1; /* Tis' cool bro. */
 375 too_bad:
 376         spin_unlock(&files_lock);
 377         return 0;
 378 }
 379
 380 /**
 381  *      mark_files_ro - mark all files read-only
 382  *      @sb: superblock in question
 383  *
 384  *      All files are marked read-only.  We don't care about pending
 385  *      delete files so this should be used in 'force' mode only.
 386  */
 387 void mark_files_ro(struct super_block *sb)
 388 {
 389         struct file *f;
 390
 391 retry:
 392         spin_lock(&files_lock);
 393         list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
 394                 struct vfsmount *mnt;
 395                 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
 396                        continue;
 397                 if (!file_count(f))
 398                         continue;
 399                 if (!(f->f_mode & FMODE_WRITE))
 400                         continue;
 401                 spin_lock(&f->f_lock);
 402                 f->f_mode &= ~FMODE_WRITE;
 403                 spin_unlock(&f->f_lock);
 404                 if (file_check_writeable(f) != 0)
 405                         continue;
 406                 file_release_write(f);
 407                 mnt = mntget(f->f_path.mnt);
 408                 /* This can sleep, so we can't hold the spinlock. */
 409                 spin_unlock(&files_lock);
 410                 mnt_drop_write(mnt);
 411                 mntput(mnt);
 412                 goto retry;
 413         }
 414         spin_unlock(&files_lock);
 415 }
 416
 417 void __init files_init(unsigned long mempages)
 418 {
 419         int n;
 420
 421         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 422                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 423
 424         /*
 425          * One file with associated inode and dcache is very roughly 1K.
 426          * Per default don't use more than 10% of our memory for files.
 427          */
 428
 429         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 430         files_stat.max_files = n;
 431         if (files_stat.max_files < NR_FILE)
 432                 files_stat.max_files = NR_FILE;
 433         files_defer_init();
 434         percpu_counter_init(&nr_files, 0);
 435 }