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/config.h>
   9 #include <linux/string.h>
  10 #include <linux/slab.h>
  11 #include <linux/file.h>
  12 #include <linux/init.h>
  13 #include <linux/module.h>
  14 #include <linux/smp_lock.h>
  15 #include <linux/fs.h>
  16 #include <linux/security.h>
  17 #include <linux/eventpoll.h>
  18 #include <linux/rcupdate.h>
  19 #include <linux/mount.h>
  20 #include <linux/capability.h>
  21 #include <linux/cdev.h>
  22 #include <linux/fsnotify.h>
  23 #include <linux/sysctl.h>
  24 #include <linux/percpu_counter.h>
  25
  26 #include <asm/atomic.h>
  27
  28 /* sysctl tunables... */
  29 struct files_stat_struct files_stat = {
  30         .max_files = NR_FILE
  31 };
  32
  33 /* public. Not pretty! */
  34 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
  35
  36 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  37
  38 static inline void file_free_rcu(struct rcu_head *head)
  39 {
  40         struct file *f =  container_of(head, struct file, f_u.fu_rcuhead);
  41         kmem_cache_free(filp_cachep, f);
  42 }
  43
  44 static inline void file_free(struct file *f)
  45 {
  46         percpu_counter_dec(&nr_files);
  47         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  48 }
  49
  50 /*
  51  * Return the total number of open files in the system
  52  */
  53 static int get_nr_files(void)
  54 {
  55         return percpu_counter_read_positive(&nr_files);
  56 }
  57
  58 /*
  59  * Return the maximum number of open files in the system
  60  */
  61 int get_max_files(void)
  62 {
  63         return files_stat.max_files;
  64 }
  65 EXPORT_SYMBOL_GPL(get_max_files);
  66
  67 /*
  68  * Handle nr_files sysctl
  69  */
  70 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  71 int proc_nr_files(ctl_table *table, int write, struct file *filp,
  72                      void __user *buffer, size_t *lenp, loff_t *ppos)
  73 {
  74         files_stat.nr_files = get_nr_files();
  75         return proc_dointvec(table, write, filp, buffer, lenp, ppos);
  76 }
  77 #else
  78 int proc_nr_files(ctl_table *table, int write, struct file *filp,
  79                      void __user *buffer, size_t *lenp, loff_t *ppos)
  80 {
  81         return -ENOSYS;
  82 }
  83 #endif
  84
  85 /* Find an unused file structure and return a pointer to it.
  86  * Returns NULL, if there are no more free file structures or
  87  * we run out of memory.
  88  */
  89 struct file *get_empty_filp(void)
  90 {
  91         static int old_max;
  92         struct file * f;
  93
  94         /*
  95          * Privileged users can go above max_files
  96          */
  97         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
  98                 /*
  99                  * percpu_counters are inaccurate.  Do an expensive check before
 100                  * we go and fail.
 101                  */
 102                 if (percpu_counter_sum(&nr_files) >= files_stat.max_files)
 103                         goto over;
 104         }
 105
 106         f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
 107         if (f == NULL)
 108                 goto fail;
 109
 110         percpu_counter_inc(&nr_files);
 111         memset(f, 0, sizeof(*f));
 112         if (security_file_alloc(f))
 113                 goto fail_sec;
 114
 115         eventpoll_init_file(f);
 116         atomic_set(&f->f_count, 1);
 117         f->f_uid = current->fsuid;
 118         f->f_gid = current->fsgid;
 119         rwlock_init(&f->f_owner.lock);
 120         /* f->f_version: 0 */
 121         INIT_LIST_HEAD(&f->f_u.fu_list);
 122         return f;
 123
 124 over:
 125         /* Ran out of filps - report that */
 126         if (get_nr_files() > old_max) {
 127                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
 128                                         get_max_files());
 129                 old_max = get_nr_files();
 130         }
 131         goto fail;
 132
 133 fail_sec:
 134         file_free(f);
 135 fail:
 136         return NULL;
 137 }
 138
 139 EXPORT_SYMBOL(get_empty_filp);
 140
 141 void fastcall fput(struct file *file)
 142 {
 143         if (atomic_dec_and_test(&file->f_count))
 144                 __fput(file);
 145 }
 146
 147 EXPORT_SYMBOL(fput);
 148
 149 /* __fput is called from task context when aio completion releases the last
 150  * last use of a struct file *.  Do not use otherwise.
 151  */
 152 void fastcall __fput(struct file *file)
 153 {
 154         struct dentry *dentry = file->f_dentry;
 155         struct vfsmount *mnt = file->f_vfsmnt;
 156         struct inode *inode = dentry->d_inode;
 157
 158         might_sleep();
 159
 160         fsnotify_close(file);
 161         /*
 162          * The function eventpoll_release() should be the first called
 163          * in the file cleanup chain.
 164          */
 165         eventpoll_release(file);
 166         locks_remove_flock(file);
 167
 168         if (file->f_op && file->f_op->release)
 169                 file->f_op->release(inode, file);
 170         security_file_free(file);
 171         if (unlikely(inode->i_cdev != NULL))
 172                 cdev_put(inode->i_cdev);
 173         fops_put(file->f_op);
 174         if (file->f_mode & FMODE_WRITE)
 175                 put_write_access(inode);
 176         file_kill(file);
 177         file->f_dentry = NULL;
 178         file->f_vfsmnt = NULL;
 179         file_free(file);
 180         dput(dentry);
 181         mntput(mnt);
 182 }
 183
 184 struct file fastcall *fget(unsigned int fd)
 185 {
 186         struct file *file;
 187         struct files_struct *files = current->files;
 188
 189         rcu_read_lock();
 190         file = fcheck_files(files, fd);
 191         if (file) {
 192                 if (!atomic_inc_not_zero(&file->f_count)) {
 193                         /* File object ref couldn't be taken */
 194                         rcu_read_unlock();
 195                         return NULL;
 196                 }
 197         }
 198         rcu_read_unlock();
 199
 200         return file;
 201 }
 202
 203 EXPORT_SYMBOL(fget);
 204
 205 /*
 206  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 207  * You can use this only if it is guranteed that the current task already
 208  * holds a refcnt to that file. That check has to be done at fget() only
 209  * and a flag is returned to be passed to the corresponding fput_light().
 210  * There must not be a cloning between an fget_light/fput_light pair.
 211  */
 212 struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
 213 {
 214         struct file *file;
 215         struct files_struct *files = current->files;
 216
 217         *fput_needed = 0;
 218         if (likely((atomic_read(&files->count) == 1))) {
 219                 file = fcheck_files(files, fd);
 220         } else {
 221                 rcu_read_lock();
 222                 file = fcheck_files(files, fd);
 223                 if (file) {
 224                         if (atomic_inc_not_zero(&file->f_count))
 225                                 *fput_needed = 1;
 226                         else
 227                                 /* Didn't get the reference, someone's freed */
 228                                 file = NULL;
 229                 }
 230                 rcu_read_unlock();
 231         }
 232
 233         return file;
 234 }
 235
 236
 237 void put_filp(struct file *file)
 238 {
 239         if (atomic_dec_and_test(&file->f_count)) {
 240                 security_file_free(file);
 241                 file_kill(file);
 242                 file_free(file);
 243         }
 244 }
 245
 246 void file_move(struct file *file, struct list_head *list)
 247 {
 248         if (!list)
 249                 return;
 250         file_list_lock();
 251         list_move(&file->f_u.fu_list, list);
 252         file_list_unlock();
 253 }
 254
 255 void file_kill(struct file *file)
 256 {
 257         if (!list_empty(&file->f_u.fu_list)) {
 258                 file_list_lock();
 259                 list_del_init(&file->f_u.fu_list);
 260                 file_list_unlock();
 261         }
 262 }
 263
 264 int fs_may_remount_ro(struct super_block *sb)
 265 {
 266         struct list_head *p;
 267
 268         /* Check that no files are currently opened for writing. */
 269         file_list_lock();
 270         list_for_each(p, &sb->s_files) {
 271                 struct file *file = list_entry(p, struct file, f_u.fu_list);
 272                 struct inode *inode = file->f_dentry->d_inode;
 273
 274                 /* File with pending delete? */
 275                 if (inode->i_nlink == 0)
 276                         goto too_bad;
 277
 278                 /* Writeable file? */
 279                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 280                         goto too_bad;
 281         }
 282         file_list_unlock();
 283         return 1; /* Tis' cool bro. */
 284 too_bad:
 285         file_list_unlock();
 286         return 0;
 287 }
 288
 289 void __init files_init(unsigned long mempages)
 290 {
 291         int n;
 292         /* One file with associated inode and dcache is very roughly 1K.
 293          * Per default don't use more than 10% of our memory for files.
 294          */
 295
 296         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 297         files_stat.max_files = n;
 298         if (files_stat.max_files < NR_FILE)
 299                 files_stat.max_files = NR_FILE;
 300         files_defer_init();
 301         percpu_counter_init(&nr_files);
 302 }