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         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  46 }
  47
  48 /*
  49  * Return the total number of open files in the system
  50  */
  51 static int get_nr_files(void)
  52 {
  53         return percpu_counter_read_positive(&nr_files);
  54 }
  55
  56 /*
  57  * Return the maximum number of open files in the system
  58  */
  59 int get_max_files(void)
  60 {
  61         return files_stat.max_files;
  62 }
  63 EXPORT_SYMBOL_GPL(get_max_files);
  64
  65 /*
  66  * Handle nr_files sysctl
  67  */
  68 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  69 int proc_nr_files(ctl_table *table, int write, struct file *filp,
  70                      void __user *buffer, size_t *lenp, loff_t *ppos)
  71 {
  72         files_stat.nr_files = get_nr_files();
  73         return proc_dointvec(table, write, filp, buffer, lenp, ppos);
  74 }
  75 #else
  76 int proc_nr_files(ctl_table *table, int write, struct file *filp,
  77                      void __user *buffer, size_t *lenp, loff_t *ppos)
  78 {
  79         return -ENOSYS;
  80 }
  81 #endif
  82
  83 /* Find an unused file structure and return a pointer to it.
  84  * Returns NULL, if there are no more free file structures or
  85  * we run out of memory.
  86  *
  87  * Be very careful using this.  You are responsible for
  88  * getting write access to any mount that you might assign
  89  * to this filp, if it is opened for write.  If this is not
  90  * done, you will imbalance int the mount's writer count
  91  * and a warning at __fput() time.
  92  */
  93 struct file *get_empty_filp(void)
  94 {
  95         struct task_struct *tsk;
  96         static int old_max;
  97         struct file * f;
  98
  99         /*
 100          * Privileged users can go above max_files
 101          */
 102         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 103                 /*
 104                  * percpu_counters are inaccurate.  Do an expensive check before
 105                  * we go and fail.
 106                  */
 107                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 108                         goto over;
 109         }
 110
 111         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 112         if (f == NULL)
 113                 goto fail;
 114
 115         percpu_counter_inc(&nr_files);
 116         if (security_file_alloc(f))
 117                 goto fail_sec;
 118
 119         tsk = current;
 120         INIT_LIST_HEAD(&f->f_u.fu_list);
 121         atomic_set(&f->f_count, 1);
 122         rwlock_init(&f->f_owner.lock);
 123         f->f_uid = tsk->fsuid;
 124         f->f_gid = tsk->fsgid;
 125         eventpoll_init_file(f);
 126         /* f->f_version: 0 */
 127         return f;
 128
 129 over:
 130         /* Ran out of filps - report that */
 131         if (get_nr_files() > old_max) {
 132                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
 133                                         get_max_files());
 134                 old_max = get_nr_files();
 135         }
 136         goto fail;
 137
 138 fail_sec:
 139         file_free(f);
 140 fail:
 141         return NULL;
 142 }
 143
 144 EXPORT_SYMBOL(get_empty_filp);
 145
 146 /**
 147  * alloc_file - allocate and initialize a 'struct file'
 148  * @mnt: the vfsmount on which the file will reside
 149  * @dentry: the dentry representing the new file
 150  * @mode: the mode with which the new file will be opened
 151  * @fop: the 'struct file_operations' for the new file
 152  *
 153  * Use this instead of get_empty_filp() to get a new
 154  * 'struct file'.  Do so because of the same initialization
 155  * pitfalls reasons listed for init_file().  This is a
 156  * preferred interface to using init_file().
 157  *
 158  * If all the callers of init_file() are eliminated, its
 159  * code should be moved into this function.
 160  */
 161 struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry,
 162                 mode_t mode, const struct file_operations *fop)
 163 {
 164         struct file *file;
 165         struct path;
 166
 167         file = get_empty_filp();
 168         if (!file)
 169                 return NULL;
 170
 171         init_file(file, mnt, dentry, mode, fop);
 172         return file;
 173 }
 174 EXPORT_SYMBOL(alloc_file);
 175
 176 /**
 177  * init_file - initialize a 'struct file'
 178  * @file: the already allocated 'struct file' to initialized
 179  * @mnt: the vfsmount on which the file resides
 180  * @dentry: the dentry representing this file
 181  * @mode: the mode the file is opened with
 182  * @fop: the 'struct file_operations' for this file
 183  *
 184  * Use this instead of setting the members directly.  Doing so
 185  * avoids making mistakes like forgetting the mntget() or
 186  * forgetting to take a write on the mnt.
 187  *
 188  * Note: This is a crappy interface.  It is here to make
 189  * merging with the existing users of get_empty_filp()
 190  * who have complex failure logic easier.  All users
 191  * of this should be moving to alloc_file().
 192  */
 193 int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry,
 194            mode_t mode, const struct file_operations *fop)
 195 {
 196         int error = 0;
 197         file->f_path.dentry = dentry;
 198         file->f_path.mnt = mntget(mnt);
 199         file->f_mapping = dentry->d_inode->i_mapping;
 200         file->f_mode = mode;
 201         file->f_op = fop;
 202         return error;
 203 }
 204 EXPORT_SYMBOL(init_file);
 205
 206 void fput(struct file *file)
 207 {
 208         if (atomic_dec_and_test(&file->f_count))
 209                 __fput(file);
 210 }
 211
 212 EXPORT_SYMBOL(fput);
 213
 214 /* __fput is called from task context when aio completion releases the last
 215  * last use of a struct file *.  Do not use otherwise.
 216  */
 217 void __fput(struct file *file)
 218 {
 219         struct dentry *dentry = file->f_path.dentry;
 220         struct vfsmount *mnt = file->f_path.mnt;
 221         struct inode *inode = dentry->d_inode;
 222
 223         might_sleep();
 224
 225         fsnotify_close(file);
 226         /*
 227          * The function eventpoll_release() should be the first called
 228          * in the file cleanup chain.
 229          */
 230         eventpoll_release(file);
 231         locks_remove_flock(file);
 232
 233         if (file->f_op && file->f_op->release)
 234                 file->f_op->release(inode, file);
 235         security_file_free(file);
 236         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 237                 cdev_put(inode->i_cdev);
 238         fops_put(file->f_op);
 239         if (file->f_mode & FMODE_WRITE)
 240                 put_write_access(inode);
 241         put_pid(file->f_owner.pid);
 242         file_kill(file);
 243         file->f_path.dentry = NULL;
 244         file->f_path.mnt = NULL;
 245         file_free(file);
 246         dput(dentry);
 247         mntput(mnt);
 248 }
 249
 250 struct file *fget(unsigned int fd)
 251 {
 252         struct file *file;
 253         struct files_struct *files = current->files;
 254
 255         rcu_read_lock();
 256         file = fcheck_files(files, fd);
 257         if (file) {
 258                 if (!atomic_inc_not_zero(&file->f_count)) {
 259                         /* File object ref couldn't be taken */
 260                         rcu_read_unlock();
 261                         return NULL;
 262                 }
 263         }
 264         rcu_read_unlock();
 265
 266         return file;
 267 }
 268
 269 EXPORT_SYMBOL(fget);
 270
 271 /*
 272  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 273  * You can use this only if it is guranteed that the current task already
 274  * holds a refcnt to that file. That check has to be done at fget() only
 275  * and a flag is returned to be passed to the corresponding fput_light().
 276  * There must not be a cloning between an fget_light/fput_light pair.
 277  */
 278 struct file *fget_light(unsigned int fd, int *fput_needed)
 279 {
 280         struct file *file;
 281         struct files_struct *files = current->files;
 282
 283         *fput_needed = 0;
 284         if (likely((atomic_read(&files->count) == 1))) {
 285                 file = fcheck_files(files, fd);
 286         } else {
 287                 rcu_read_lock();
 288                 file = fcheck_files(files, fd);
 289                 if (file) {
 290                         if (atomic_inc_not_zero(&file->f_count))
 291                                 *fput_needed = 1;
 292                         else
 293                                 /* Didn't get the reference, someone's freed */
 294                                 file = NULL;
 295                 }
 296                 rcu_read_unlock();
 297         }
 298
 299         return file;
 300 }
 301
 302
 303 void put_filp(struct file *file)
 304 {
 305         if (atomic_dec_and_test(&file->f_count)) {
 306                 security_file_free(file);
 307                 file_kill(file);
 308                 file_free(file);
 309         }
 310 }
 311
 312 void file_move(struct file *file, struct list_head *list)
 313 {
 314         if (!list)
 315                 return;
 316         file_list_lock();
 317         list_move(&file->f_u.fu_list, list);
 318         file_list_unlock();
 319 }
 320
 321 void file_kill(struct file *file)
 322 {
 323         if (!list_empty(&file->f_u.fu_list)) {
 324                 file_list_lock();
 325                 list_del_init(&file->f_u.fu_list);
 326                 file_list_unlock();
 327         }
 328 }
 329
 330 int fs_may_remount_ro(struct super_block *sb)
 331 {
 332         struct file *file;
 333
 334         /* Check that no files are currently opened for writing. */
 335         file_list_lock();
 336         list_for_each_entry(file, &sb->s_files, f_u.fu_list) {
 337                 struct inode *inode = file->f_path.dentry->d_inode;
 338
 339                 /* File with pending delete? */
 340                 if (inode->i_nlink == 0)
 341                         goto too_bad;
 342
 343                 /* Writeable file? */
 344                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 345                         goto too_bad;
 346         }
 347         file_list_unlock();
 348         return 1; /* Tis' cool bro. */
 349 too_bad:
 350         file_list_unlock();
 351         return 0;
 352 }
 353
 354 void __init files_init(unsigned long mempages)
 355 {
 356         int n;
 357         /* One file with associated inode and dcache is very roughly 1K.
 358          * Per default don't use more than 10% of our memory for files.
 359          */
 360
 361         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 362         files_stat.max_files = n;
 363         if (files_stat.max_files < NR_FILE)
 364                 files_stat.max_files = NR_FILE;
 365         files_defer_init();
 366         percpu_counter_init(&nr_files, 0);
 367 }