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/init.h>
  12 #include <linux/module.h>
  13 #include <linux/fs.h>
  14 #include <linux/security.h>
  15 #include <linux/eventpoll.h>
  16 #include <linux/rcupdate.h>
  17 #include <linux/mount.h>
  18 #include <linux/capability.h>
  19 #include <linux/cdev.h>
  20 #include <linux/fsnotify.h>
  21 #include <linux/sysctl.h>
  22 #include <linux/percpu_counter.h>
  23
  24 #include <asm/atomic.h>
  25
  26 /* sysctl tunables... */
  27 struct files_stat_struct files_stat = {
  28         .max_files = NR_FILE
  29 };
  30
  31 /* public. Not pretty! */
  32 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
  33
  34 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  35
  36 static inline void file_free_rcu(struct rcu_head *head)
  37 {
  38         struct file *f =  container_of(head, struct file, f_u.fu_rcuhead);
  39         kmem_cache_free(filp_cachep, f);
  40 }
  41
  42 static inline void file_free(struct file *f)
  43 {
  44         percpu_counter_dec(&nr_files);
  45         file_check_state(f);
  46         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  47 }
  48
  49 /*
  50  * Return the total number of open files in the system
  51  */
  52 static int get_nr_files(void)
  53 {
  54         return percpu_counter_read_positive(&nr_files);
  55 }
  56
  57 /*
  58  * Return the maximum number of open files in the system
  59  */
  60 int get_max_files(void)
  61 {
  62         return files_stat.max_files;
  63 }
  64 EXPORT_SYMBOL_GPL(get_max_files);
  65
  66 /*
  67  * Handle nr_files sysctl
  68  */
  69 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  70 int proc_nr_files(ctl_table *table, int write, struct file *filp,
  71                      void __user *buffer, size_t *lenp, loff_t *ppos)
  72 {
  73         files_stat.nr_files = get_nr_files();
  74         return proc_dointvec(table, write, filp, buffer, lenp, ppos);
  75 }
  76 #else
  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         return -ENOSYS;
  81 }
  82 #endif
  83
  84 /* Find an unused file structure and return a pointer to it.
  85  * Returns NULL, if there are no more free file structures or
  86  * we run out of memory.
  87  *
  88  * Be very careful using this.  You are responsible for
  89  * getting write access to any mount that you might assign
  90  * to this filp, if it is opened for write.  If this is not
  91  * done, you will imbalance int the mount's writer count
  92  * and a warning at __fput() time.
  93  */
  94 struct file *get_empty_filp(void)
  95 {
  96         struct task_struct *tsk;
  97         static int old_max;
  98         struct file * f;
  99
 100         /*
 101          * Privileged users can go above max_files
 102          */
 103         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 104                 /*
 105                  * percpu_counters are inaccurate.  Do an expensive check before
 106                  * we go and fail.
 107                  */
 108                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 109                         goto over;
 110         }
 111
 112         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 113         if (f == NULL)
 114                 goto fail;
 115
 116         percpu_counter_inc(&nr_files);
 117         if (security_file_alloc(f))
 118                 goto fail_sec;
 119
 120         tsk = current;
 121         INIT_LIST_HEAD(&f->f_u.fu_list);
 122         atomic_set(&f->f_count, 1);
 123         rwlock_init(&f->f_owner.lock);
 124         f->f_uid = tsk->fsuid;
 125         f->f_gid = tsk->fsgid;
 126         eventpoll_init_file(f);
 127         /* f->f_version: 0 */
 128         return f;
 129
 130 over:
 131         /* Ran out of filps - report that */
 132         if (get_nr_files() > old_max) {
 133                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
 134                                         get_max_files());
 135                 old_max = get_nr_files();
 136         }
 137         goto fail;
 138
 139 fail_sec:
 140         file_free(f);
 141 fail:
 142         return NULL;
 143 }
 144
 145 EXPORT_SYMBOL(get_empty_filp);
 146
 147 /**
 148  * alloc_file - allocate and initialize a 'struct file'
 149  * @mnt: the vfsmount on which the file will reside
 150  * @dentry: the dentry representing the new file
 151  * @mode: the mode with which the new file will be opened
 152  * @fop: the 'struct file_operations' for the new file
 153  *
 154  * Use this instead of get_empty_filp() to get a new
 155  * 'struct file'.  Do so because of the same initialization
 156  * pitfalls reasons listed for init_file().  This is a
 157  * preferred interface to using init_file().
 158  *
 159  * If all the callers of init_file() are eliminated, its
 160  * code should be moved into this function.
 161  */
 162 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
 163                 mode_t mode, const struct file_operations *fop)
 164 {
 165         struct file *file;
 166         struct path;
 167
 168         file = get_empty_filp();
 169         if (!file)
 170                 return NULL;
 171
 172         init_file(file, mnt, dentry, mode, fop);
 173         return file;
 174 }
 175 EXPORT_SYMBOL(alloc_file);
 176
 177 /**
 178  * init_file - initialize a 'struct file'
 179  * @file: the already allocated 'struct file' to initialized
 180  * @mnt: the vfsmount on which the file resides
 181  * @dentry: the dentry representing this file
 182  * @mode: the mode the file is opened with
 183  * @fop: the 'struct file_operations' for this file
 184  *
 185  * Use this instead of setting the members directly.  Doing so
 186  * avoids making mistakes like forgetting the mntget() or
 187  * forgetting to take a write on the mnt.
 188  *
 189  * Note: This is a crappy interface.  It is here to make
 190  * merging with the existing users of get_empty_filp()
 191  * who have complex failure logic easier.  All users
 192  * of this should be moving to alloc_file().
 193  */
 194 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
 195            mode_t mode, const struct file_operations *fop)
 196 {
 197         int error = 0;
 198         file->f_path.dentry = dentry;
 199         file->f_path.mnt = mntget(mnt);
 200         file->f_mapping = dentry->d_inode->i_mapping;
 201         file->f_mode = mode;
 202         file->f_op = fop;
 203
 204         /*
 205          * These mounts don't really matter in practice
 206          * for r/o bind mounts.  They aren't userspace-
 207          * visible.  We do this for consistency, and so
 208          * that we can do debugging checks at __fput()
 209          */
 210         if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) {
 211                 file_take_write(file);
 212                 error = mnt_want_write(mnt);
 213                 WARN_ON(error);
 214         }
 215         return error;
 216 }
 217 EXPORT_SYMBOL(init_file);
 218
 219 void fput(struct file *file)
 220 {
 221         if (atomic_dec_and_test(&file->f_count))
 222                 __fput(file);
 223 }
 224
 225 EXPORT_SYMBOL(fput);
 226
 227 /**
 228  * drop_file_write_access - give up ability to write to a file
 229  * @file: the file to which we will stop writing
 230  *
 231  * This is a central place which will give up the ability
 232  * to write to @file, along with access to write through
 233  * its vfsmount.
 234  */
 235 void drop_file_write_access(struct file *file)
 236 {
 237         struct vfsmount *mnt = file->f_path.mnt;
 238         struct dentry *dentry = file->f_path.dentry;
 239         struct inode *inode = dentry->d_inode;
 240
 241         put_write_access(inode);
 242
 243         if (special_file(inode->i_mode))
 244                 return;
 245         if (file_check_writeable(file) != 0)
 246                 return;
 247         mnt_drop_write(mnt);
 248         file_release_write(file);
 249 }
 250 EXPORT_SYMBOL_GPL(drop_file_write_access);
 251
 252 /* __fput is called from task context when aio completion releases the last
 253  * last use of a struct file *.  Do not use otherwise.
 254  */
 255 void __fput(struct file *file)
 256 {
 257         struct dentry *dentry = file->f_path.dentry;
 258         struct vfsmount *mnt = file->f_path.mnt;
 259         struct inode *inode = dentry->d_inode;
 260
 261         might_sleep();
 262
 263         fsnotify_close(file);
 264         /*
 265          * The function eventpoll_release() should be the first called
 266          * in the file cleanup chain.
 267          */
 268         eventpoll_release(file);
 269         locks_remove_flock(file);
 270
 271         if (file->f_op && file->f_op->release)
 272                 file->f_op->release(inode, file);
 273         security_file_free(file);
 274         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 275                 cdev_put(inode->i_cdev);
 276         fops_put(file->f_op);
 277         put_pid(file->f_owner.pid);
 278         file_kill(file);
 279         if (file->f_mode & FMODE_WRITE)
 280                 drop_file_write_access(file);
 281         file->f_path.dentry = NULL;
 282         file->f_path.mnt = NULL;
 283         file_free(file);
 284         dput(dentry);
 285         mntput(mnt);
 286 }
 287
 288 struct file *fget(unsigned int fd)
 289 {
 290         struct file *file;
 291         struct files_struct *files = current->files;
 292
 293         rcu_read_lock();
 294         file = fcheck_files(files, fd);
 295         if (file) {
 296                 if (!atomic_inc_not_zero(&file->f_count)) {
 297                         /* File object ref couldn't be taken */
 298                         rcu_read_unlock();
 299                         return NULL;
 300                 }
 301         }
 302         rcu_read_unlock();
 303
 304         return file;
 305 }
 306
 307 EXPORT_SYMBOL(fget);
 308
 309 /*
 310  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 311  * You can use this only if it is guranteed that the current task already
 312  * holds a refcnt to that file. That check has to be done at fget() only
 313  * and a flag is returned to be passed to the corresponding fput_light().
 314  * There must not be a cloning between an fget_light/fput_light pair.
 315  */
 316 struct file *fget_light(unsigned int fd, int *fput_needed)
 317 {
 318         struct file *file;
 319         struct files_struct *files = current->files;
 320
 321         *fput_needed = 0;
 322         if (likely((atomic_read(&files->count) == 1))) {
 323                 file = fcheck_files(files, fd);
 324         } else {
 325                 rcu_read_lock();
 326                 file = fcheck_files(files, fd);
 327                 if (file) {
 328                         if (atomic_inc_not_zero(&file->f_count))
 329                                 *fput_needed = 1;
 330                         else
 331                                 /* Didn't get the reference, someone's freed */
 332                                 file = NULL;
 333                 }
 334                 rcu_read_unlock();
 335         }
 336
 337         return file;
 338 }
 339
 340
 341 void put_filp(struct file *file)
 342 {
 343         if (atomic_dec_and_test(&file->f_count)) {
 344                 security_file_free(file);
 345                 file_kill(file);
 346                 file_free(file);
 347         }
 348 }
 349
 350 void file_move(struct file *file, struct list_head *list)
 351 {
 352         if (!list)
 353                 return;
 354         file_list_lock();
 355         list_move(&file->f_u.fu_list, list);
 356         file_list_unlock();
 357 }
 358
 359 void file_kill(struct file *file)
 360 {
 361         if (!list_empty(&file->f_u.fu_list)) {
 362                 file_list_lock();
 363                 list_del_init(&file->f_u.fu_list);
 364                 file_list_unlock();
 365         }
 366 }
 367
 368 int fs_may_remount_ro(struct super_block *sb)
 369 {
 370         struct file *file;
 371
 372         /* Check that no files are currently opened for writing. */
 373         file_list_lock();
 374         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 375                 struct inode *inode = file->f_path.dentry->d_inode;
 376
 377                 /* File with pending delete? */
 378                 if (inode->i_nlink == 0)
 379                         goto too_bad;
 380
 381                 /* Writeable file? */
 382                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 383                         goto too_bad;
 384         }
 385         file_list_unlock();
 386         return 1; /* Tis' cool bro. */
 387 too_bad:
 388         file_list_unlock();
 389         return 0;
 390 }
 391
 392 void __init files_init(unsigned long mempages)
 393 {
 394         int n;
 395         /* One file with associated inode and dcache is very roughly 1K.
 396          * Per default don't use more than 10% of our memory for files.
 397          */
 398
 399         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 400         files_stat.max_files = n;
 401         if (files_stat.max_files < NR_FILE)
 402                 files_stat.max_files = NR_FILE;
 403         files_defer_init();
 404         percpu_counter_init(&nr_files, 0);
 405 }