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 struct file *get_empty_filp(void)
  88 {
  89         struct task_struct *tsk;
  90         static int old_max;
  91         struct file * f;
  92
  93         /*
  94          * Privileged users can go above max_files
  95          */
  96         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
  97                 /*
  98                  * percpu_counters are inaccurate.  Do an expensive check before
  99                  * we go and fail.
 100                  */
 101                 if (percpu_counter_sum(&nr_files) >= files_stat.max_files)
 102                         goto over;
 103         }
 104
 105         f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
 106         if (f == NULL)
 107                 goto fail;
 108
 109         percpu_counter_inc(&nr_files);
 110         memset(f, 0, sizeof(*f));
 111         if (security_file_alloc(f))
 112                 goto fail_sec;
 113
 114         tsk = current;
 115         INIT_LIST_HEAD(&f->f_u.fu_list);
 116         atomic_set(&f->f_count, 1);
 117         rwlock_init(&f->f_owner.lock);
 118         f->f_uid = tsk->fsuid;
 119         f->f_gid = tsk->fsgid;
 120         eventpoll_init_file(f);
 121         /* f->f_version: 0 */
 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_path.dentry;
 155         struct vfsmount *mnt = file->f_path.mnt;
 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(S_ISCHR(inode->i_mode) && 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         put_pid(file->f_owner.pid);
 177         file_kill(file);
 178         file->f_path.dentry = NULL;
 179         file->f_path.mnt = NULL;
 180         file_free(file);
 181         dput(dentry);
 182         mntput(mnt);
 183 }
 184
 185 struct file fastcall *fget(unsigned int fd)
 186 {
 187         struct file *file;
 188         struct files_struct *files = current->files;
 189
 190         rcu_read_lock();
 191         file = fcheck_files(files, fd);
 192         if (file) {
 193                 if (!atomic_inc_not_zero(&file->f_count)) {
 194                         /* File object ref couldn't be taken */
 195                         rcu_read_unlock();
 196                         return NULL;
 197                 }
 198         }
 199         rcu_read_unlock();
 200
 201         return file;
 202 }
 203
 204 EXPORT_SYMBOL(fget);
 205
 206 /*
 207  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 208  * You can use this only if it is guranteed that the current task already
 209  * holds a refcnt to that file. That check has to be done at fget() only
 210  * and a flag is returned to be passed to the corresponding fput_light().
 211  * There must not be a cloning between an fget_light/fput_light pair.
 212  */
 213 struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
 214 {
 215         struct file *file;
 216         struct files_struct *files = current->files;
 217
 218         *fput_needed = 0;
 219         if (likely((atomic_read(&files->count) == 1))) {
 220                 file = fcheck_files(files, fd);
 221         } else {
 222                 rcu_read_lock();
 223                 file = fcheck_files(files, fd);
 224                 if (file) {
 225                         if (atomic_inc_not_zero(&file->f_count))
 226                                 *fput_needed = 1;
 227                         else
 228                                 /* Didn't get the reference, someone's freed */
 229                                 file = NULL;
 230                 }
 231                 rcu_read_unlock();
 232         }
 233
 234         return file;
 235 }
 236
 237
 238 void put_filp(struct file *file)
 239 {
 240         if (atomic_dec_and_test(&file->f_count)) {
 241                 security_file_free(file);
 242                 file_kill(file);
 243                 file_free(file);
 244         }
 245 }
 246
 247 void file_move(struct file *file, struct list_head *list)
 248 {
 249         if (!list)
 250                 return;
 251         file_list_lock();
 252         list_move(&file->f_u.fu_list, list);
 253         file_list_unlock();
 254 }
 255
 256 void file_kill(struct file *file)
 257 {
 258         if (!list_empty(&file->f_u.fu_list)) {
 259                 file_list_lock();
 260                 list_del_init(&file->f_u.fu_list);
 261                 file_list_unlock();
 262         }
 263 }
 264
 265 int fs_may_remount_ro(struct super_block *sb)
 266 {
 267         struct list_head *p;
 268
 269         /* Check that no files are currently opened for writing. */
 270         file_list_lock();
 271         list_for_each(p, &sb->s_files) {
 272                 struct file *file = list_entry(p, struct file, f_u.fu_list);
 273                 struct inode *inode = file->f_path.dentry->d_inode;
 274
 275                 /* File with pending delete? */
 276                 if (inode->i_nlink == 0)
 277                         goto too_bad;
 278
 279                 /* Writeable file? */
 280                 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 281                         goto too_bad;
 282         }
 283         file_list_unlock();
 284         return 1; /* Tis' cool bro. */
 285 too_bad:
 286         file_list_unlock();
 287         return 0;
 288 }
 289
 290 void __init files_init(unsigned long mempages)
 291 {
 292         int n;
 293         /* One file with associated inode and dcache is very roughly 1K.
 294          * Per default don't use more than 10% of our memory for files.
 295          */
 296
 297         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 298         files_stat.max_files = n;
 299         if (files_stat.max_files < NR_FILE)
 300                 files_stat.max_files = NR_FILE;
 301         files_defer_init();
 302         percpu_counter_init(&nr_files, 0);
 303 }