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/smp_lock.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
  25 #include <asm/atomic.h>
  26
  27 /* sysctl tunables... */
  28 struct files_stat_struct files_stat = {
  29         .max_files = NR_FILE
  30 };
  31
  32 /* public. Not pretty! */
  33 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
  34
  35 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  36
  37 static inline void file_free_rcu(struct rcu_head *head)
  38 {
  39         struct file *f =  container_of(head, struct file, f_u.fu_rcuhead);
  40         kmem_cache_free(filp_cachep, f);
  41 }
  42
  43 static inline void file_free(struct file *f)
  44 {
  45         percpu_counter_dec(&nr_files);
  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 struct file *get_empty_filp(void)
  89 {
  90         struct task_struct *tsk;
  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         tsk = current;
 116         INIT_LIST_HEAD(&f->f_u.fu_list);
 117         atomic_set(&f->f_count, 1);
 118         rwlock_init(&f->f_owner.lock);
 119         f->f_uid = tsk->fsuid;
 120         f->f_gid = tsk->fsgid;
 121         eventpoll_init_file(f);
 122         /* f->f_version: 0 */
 123         return f;
 124
 125 over:
 126         /* Ran out of filps - report that */
 127         if (get_nr_files() > old_max) {
 128                 printk(KERN_INFO "VFS: file-max limit %d reached\n",
 129                                         get_max_files());
 130                 old_max = get_nr_files();
 131         }
 132         goto fail;
 133
 134 fail_sec:
 135         file_free(f);
 136 fail:
 137         return NULL;
 138 }
 139
 140 EXPORT_SYMBOL(get_empty_filp);
 141
 142 void fastcall fput(struct file *file)
 143 {
 144         if (atomic_dec_and_test(&file->f_count))
 145                 __fput(file);
 146 }
 147
 148 EXPORT_SYMBOL(fput);
 149
 150 /* __fput is called from task context when aio completion releases the last
 151  * last use of a struct file *.  Do not use otherwise.
 152  */
 153 void fastcall __fput(struct file *file)
 154 {
 155         struct dentry *dentry = file->f_dentry;
 156         struct vfsmount *mnt = file->f_vfsmnt;
 157         struct inode *inode = dentry->d_inode;
 158
 159         might_sleep();
 160
 161         fsnotify_close(file);
 162         /*
 163          * The function eventpoll_release() should be the first called
 164          * in the file cleanup chain.
 165          */
 166         eventpoll_release(file);
 167         locks_remove_flock(file);
 168
 169         if (file->f_op && file->f_op->release)
 170                 file->f_op->release(inode, file);
 171         security_file_free(file);
 172 #if 0   // mask by Victor Yu. 02-12-2007
 173         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL))
 174                 cdev_put(inode->i_cdev);
 175 #else
 176         if (unlikely(S_ISCHR(inode->i_mode) && inode->u.i_cdev != NULL))
 177                 cdev_put(inode->u.i_cdev);
 178 #endif
 179         fops_put(file->f_op);
 180         if (file->f_mode & FMODE_WRITE)
 181                 put_write_access(inode);
 182         put_pid(file->f_owner.pid);
 183         file_kill(file);
 184         file->f_dentry = NULL;
 185         file->f_vfsmnt = NULL;
 186         file_free(file);
 187         dput(dentry);
 188         mntput(mnt);
 189 }
 190
 191 struct file fastcall *fget(unsigned int fd)
 192 {
 193         struct file *file;
 194         struct files_struct *files = current->files;
 195
 196         rcu_read_lock();
 197         file = fcheck_files(files, fd);
 198         if (file) {
 199                 if (!atomic_inc_not_zero(&file->f_count)) {
 200                         /* File object ref couldn't be taken */
 201                         rcu_read_unlock();
 202                         return NULL;
 203                 }
 204         }
 205         rcu_read_unlock();
 206
 207         return file;
 208 }
 209
 210 EXPORT_SYMBOL(fget);
 211
 212 /*
 213  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 214  * You can use this only if it is guranteed that the current task already
 215  * holds a refcnt to that file. That check has to be done at fget() only
 216  * and a flag is returned to be passed to the corresponding fput_light().
 217  * There must not be a cloning between an fget_light/fput_light pair.
 218  */
 219 struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
 220 {
 221         struct file *file;
 222         struct files_struct *files = current->files;
 223
 224         *fput_needed = 0;
 225         if (likely((atomic_read(&files->count) == 1))) {
 226                 file = fcheck_files(files, fd);
 227         } else {
 228                 rcu_read_lock();
 229                 file = fcheck_files(files, fd);
 230                 if (file) {
 231                         if (atomic_inc_not_zero(&file->f_count))
 232                                 *fput_needed = 1;
 233                         else
 234                                 /* Didn't get the reference, someone's freed */
 235                                 file = NULL;
 236                 }
 237                 rcu_read_unlock();
 238         }
 239
 240         return file;
 241 }
 242
 243
 244 void put_filp(struct file *file)
 245 {
 246         if (atomic_dec_and_test(&file->f_count)) {
 247                 security_file_free(file);
 248                 file_kill(file);
 249                 file_free(file);
 250         }
 251 }
 252
 253 void file_move(struct file *file, struct list_head *list)
 254 {
 255         if (!list)
 256                 return;
 257         file_list_lock();
 258         list_move(&file->f_u.fu_list, list);
 259         file_list_unlock();
 260 }
 261
 262 void file_kill(struct file *file)
 263 {
 264         if (!list_empty(&file->f_u.fu_list)) {
 265                 file_list_lock();
 266                 list_del_init(&file->f_u.fu_list);
 267                 file_list_unlock();
 268         }
 269 }
 270
 271 int fs_may_remount_ro(struct super_block *sb)
 272 {
 273         struct list_head *p;
 274
 275         /* Check that no files are currently opened for writing. */
 276         file_list_lock();
 277         list_for_each(p, &sb->s_files) {
 278                 struct file *file = list_entry(p, struct file, f_u.fu_list);
 279                 struct inode *inode = file->f_dentry->d_inode;
 280
 281                 /* File with pending delete? */
 282                 if (inode->i_nlink == 0)
 283                         goto too_bad;
 284
 285                 /* Writeable file? */
 286                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 287                         goto too_bad;
 288         }
 289         file_list_unlock();
 290         return 1; /* Tis' cool bro. */
 291 too_bad:
 292         file_list_unlock();
 293         return 0;
 294 }
 295
 296 void __init files_init(unsigned long mempages)
 297 {
 298         int n;
 299         /* One file with associated inode and dcache is very roughly 1K.
 300          * Per default don't use more than 10% of our memory for files.
 301          */
 302
 303         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 304         files_stat.max_files = n;
 305         if (files_stat.max_files < NR_FILE)
 306                 files_stat.max_files = NR_FILE;
 307         files_defer_init();
 308         percpu_counter_init(&nr_files, 0);
 309 }