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