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