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 void fput(struct file *file)
 198 {
 199         if (atomic_long_dec_and_test(&file->f_count))
 200                 __fput(file);
 201 }
 202
 203 EXPORT_SYMBOL(fput);
 204
 205 /**
 206  * drop_file_write_access - give up ability to write to a file
 207  * @file: the file to which we will stop writing
 208  *
 209  * This is a central place which will give up the ability
 210  * to write to @file, along with access to write through
 211  * its vfsmount.
 212  */
 213 void drop_file_write_access(struct file *file)
 214 {
 215         struct vfsmount *mnt = file->f_path.mnt;
 216         struct dentry *dentry = file->f_path.dentry;
 217         struct inode *inode = dentry->d_inode;
 218
 219         put_write_access(inode);
 220
 221         if (special_file(inode->i_mode))
 222                 return;
 223         if (file_check_writeable(file) != 0)
 224                 return;
 225         mnt_drop_write(mnt);
 226         file_release_write(file);
 227 }
 228 EXPORT_SYMBOL_GPL(drop_file_write_access);
 229
 230 /* __fput is called from task context when aio completion releases the last
 231  * last use of a struct file *.  Do not use otherwise.
 232  */
 233 void __fput(struct file *file)
 234 {
 235         struct dentry *dentry = file->f_path.dentry;
 236         struct vfsmount *mnt = file->f_path.mnt;
 237         struct inode *inode = dentry->d_inode;
 238
 239         might_sleep();
 240
 241         fsnotify_close(file);
 242         /*
 243          * The function eventpoll_release() should be the first called
 244          * in the file cleanup chain.
 245          */
 246         eventpoll_release(file);
 247         locks_remove_flock(file);
 248
 249         if (unlikely(file->f_flags & FASYNC)) {
 250                 if (file->f_op && file->f_op->fasync)
 251                         file->f_op->fasync(-1, file, 0);
 252         }
 253         if (file->f_op && file->f_op->release)
 254                 file->f_op->release(inode, file);
 255         security_file_free(file);
 256         ima_file_free(file);
 257         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 258                 cdev_put(inode->i_cdev);
 259         fops_put(file->f_op);
 260         put_pid(file->f_owner.pid);
 261         file_kill(file);
 262         if (file->f_mode & FMODE_WRITE)
 263                 drop_file_write_access(file);
 264         file->f_path.dentry = NULL;
 265         file->f_path.mnt = NULL;
 266         file_free(file);
 267         dput(dentry);
 268         mntput(mnt);
 269 }
 270
 271 struct file *fget(unsigned int fd)
 272 {
 273         struct file *file;
 274         struct files_struct *files = current->files;
 275
 276         rcu_read_lock();
 277         file = fcheck_files(files, fd);
 278         if (file) {
 279                 if (!atomic_long_inc_not_zero(&file->f_count)) {
 280                         /* File object ref couldn't be taken */
 281                         rcu_read_unlock();
 282                         return NULL;
 283                 }
 284         }
 285         rcu_read_unlock();
 286
 287         return file;
 288 }
 289
 290 EXPORT_SYMBOL(fget);
 291
 292 /*
 293  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 294  * You can use this only if it is guranteed that the current task already
 295  * holds a refcnt to that file. That check has to be done at fget() only
 296  * and a flag is returned to be passed to the corresponding fput_light().
 297  * There must not be a cloning between an fget_light/fput_light pair.
 298  */
 299 struct file *fget_light(unsigned int fd, int *fput_needed)
 300 {
 301         struct file *file;
 302         struct files_struct *files = current->files;
 303
 304         *fput_needed = 0;
 305         if (likely((atomic_read(&files->count) == 1))) {
 306                 file = fcheck_files(files, fd);
 307         } else {
 308                 rcu_read_lock();
 309                 file = fcheck_files(files, fd);
 310                 if (file) {
 311                         if (atomic_long_inc_not_zero(&file->f_count))
 312                                 *fput_needed = 1;
 313                         else
 314                                 /* Didn't get the reference, someone's freed */
 315                                 file = NULL;
 316                 }
 317                 rcu_read_unlock();
 318         }
 319
 320         return file;
 321 }
 322
 323
 324 void put_filp(struct file *file)
 325 {
 326         if (atomic_long_dec_and_test(&file->f_count)) {
 327                 security_file_free(file);
 328                 file_kill(file);
 329                 file_free(file);
 330         }
 331 }
 332
 333 void file_move(struct file *file, struct list_head *list)
 334 {
 335         if (!list)
 336                 return;
 337         file_list_lock();
 338         list_move(&file->f_u.fu_list, list);
 339         file_list_unlock();
 340 }
 341
 342 void file_kill(struct file *file)
 343 {
 344         if (!list_empty(&file->f_u.fu_list)) {
 345                 file_list_lock();
 346                 list_del_init(&file->f_u.fu_list);
 347                 file_list_unlock();
 348         }
 349 }
 350
 351 int fs_may_remount_ro(struct super_block *sb)
 352 {
 353         struct file *file;
 354
 355         /* Check that no files are currently opened for writing. */
 356         file_list_lock();
 357         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 358                 struct inode *inode = file->f_path.dentry->d_inode;
 359
 360                 /* File with pending delete? */
 361                 if (inode->i_nlink == 0)
 362                         goto too_bad;
 363
 364                 /* Writeable file? */
 365                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 366                         goto too_bad;
 367         }
 368         file_list_unlock();
 369         return 1; /* Tis' cool bro. */
 370 too_bad:
 371         file_list_unlock();
 372         return 0;
 373 }
 374
 375 /**
 376  *      mark_files_ro - mark all files read-only
 377  *      @sb: superblock in question
 378  *
 379  *      All files are marked read-only.  We don't care about pending
 380  *      delete files so this should be used in 'force' mode only.
 381  */
 382 void mark_files_ro(struct super_block *sb)
 383 {
 384         struct file *f;
 385
 386 retry:
 387         file_list_lock();
 388         list_for_each_entry(f, &sb->s_files, f_u.fu_list) {
 389                 struct vfsmount *mnt;
 390                 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
 391                        continue;
 392                 if (!file_count(f))
 393                         continue;
 394                 if (!(f->f_mode & FMODE_WRITE))
 395                         continue;
 396                 spin_lock(&f->f_lock);
 397                 f->f_mode &= ~FMODE_WRITE;
 398                 spin_unlock(&f->f_lock);
 399                 if (file_check_writeable(f) != 0)
 400                         continue;
 401                 file_release_write(f);
 402                 mnt = mntget(f->f_path.mnt);
 403                 file_list_unlock();
 404                 /*
 405                  * This can sleep, so we can't hold
 406                  * the file_list_lock() spinlock.
 407                  */
 408                 mnt_drop_write(mnt);
 409                 mntput(mnt);
 410                 goto retry;
 411         }
 412         file_list_unlock();
 413 }
 414
 415 void __init files_init(unsigned long mempages)
 416 {
 417         int n;
 418
 419         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 420                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 421
 422         /*
 423          * One file with associated inode and dcache is very roughly 1K.
 424          * Per default don't use more than 10% of our memory for files.
 425          */
 426
 427         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 428         files_stat.max_files = n;
 429         if (files_stat.max_files < NR_FILE)
 430                 files_stat.max_files = NR_FILE;
 431         files_defer_init();
 432         percpu_counter_init(&nr_files, 0);
 433 }