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/lglock.h>
  24 #include <linux/percpu_counter.h>
  25 #include <linux/percpu.h>
  26 #include <linux/hardirq.h>
  27 #include <linux/task_work.h>
  28 #include <linux/ima.h>
  29
  30 #include <linux/atomic.h>
  31
  32 #include "internal.h"
  33
  34 /* sysctl tunables... */
  35 struct files_stat_struct files_stat = {
  36         .max_files = NR_FILE
  37 };
  38
  39 /* SLAB cache for file structures */
  40 static struct kmem_cache *filp_cachep __read_mostly;
  41
  42 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  43
  44 static void file_free_rcu(struct rcu_head *head)
  45 {
  46         struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  47
  48         put_cred(f->f_cred);
  49         kmem_cache_free(filp_cachep, f);
  50 }
  51
  52 static inline void file_free(struct file *f)
  53 {
  54         percpu_counter_dec(&nr_files);
  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 long 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 unsigned long 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_doulongvec_minmax(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 an error pointer if some error happend e.g. we over file
  95  * structures limit, run out of memory or operation is not permitted.
  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 long old_max;
 107         struct file *f;
 108         int error;
 109
 110         /*
 111          * Privileged users can go above max_files
 112          */
 113         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 114                 /*
 115                  * percpu_counters are inaccurate.  Do an expensive check before
 116                  * we go and fail.
 117                  */
 118                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 119                         goto over;
 120         }
 121
 122         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 123         if (unlikely(!f))
 124                 return ERR_PTR(-ENOMEM);
 125
 126         percpu_counter_inc(&nr_files);
 127         f->f_cred = get_cred(cred);
 128         error = security_file_alloc(f);
 129         if (unlikely(error)) {
 130                 file_free(f);
 131                 return ERR_PTR(error);
 132         }
 133
 134         atomic_long_set(&f->f_count, 1);
 135         rwlock_init(&f->f_owner.lock);
 136         spin_lock_init(&f->f_lock);
 137         mutex_init(&f->f_pos_lock);
 138         eventpoll_init_file(f);
 139         /* f->f_version: 0 */
 140         return f;
 141
 142 over:
 143         /* Ran out of filps - report that */
 144         if (get_nr_files() > old_max) {
 145                 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
 146                 old_max = get_nr_files();
 147         }
 148         return ERR_PTR(-ENFILE);
 149 }
 150
 151 /**
 152  * alloc_file - allocate and initialize a 'struct file'
 153  * @mnt: the vfsmount on which the file will reside
 154  * @dentry: the dentry representing the new file
 155  * @mode: the mode with which the new file will be opened
 156  * @fop: the 'struct file_operations' for the new file
 157  *
 158  * Use this instead of get_empty_filp() to get a new
 159  * 'struct file'.  Do so because of the same initialization
 160  * pitfalls reasons listed for init_file().  This is a
 161  * preferred interface to using init_file().
 162  *
 163  * If all the callers of init_file() are eliminated, its
 164  * code should be moved into this function.
 165  */
 166 struct file *alloc_file(struct path *path, fmode_t mode,
 167                 const struct file_operations *fop)
 168 {
 169         struct file *file;
 170
 171         file = get_empty_filp();
 172         if (IS_ERR(file))
 173                 return file;
 174
 175         file->f_path = *path;
 176         file->f_inode = path->dentry->d_inode;
 177         file->f_mapping = path->dentry->d_inode->i_mapping;
 178         file->f_mode = mode;
 179         file->f_op = fop;
 180
 181         /*
 182          * These mounts don't really matter in practice
 183          * for r/o bind mounts.  They aren't userspace-
 184          * visible.  We do this for consistency, and so
 185          * that we can do debugging checks at __fput()
 186          */
 187         if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
 188                 WARN_ON(mnt_clone_write(path->mnt));
 189         }
 190         if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 191                 i_readcount_inc(path->dentry->d_inode);
 192         return file;
 193 }
 194 EXPORT_SYMBOL(alloc_file);
 195
 196 /**
 197  * drop_file_write_access - give up ability to write to a file
 198  * @file: the file to which we will stop writing
 199  *
 200  * This is a central place which will give up the ability
 201  * to write to @file, along with access to write through
 202  * its vfsmount.
 203  */
 204 static void drop_file_write_access(struct file *file)
 205 {
 206         struct vfsmount *mnt = file->f_path.mnt;
 207         struct dentry *dentry = file->f_path.dentry;
 208         struct inode *inode = dentry->d_inode;
 209
 210         if (special_file(inode->i_mode))
 211                 return;
 212
 213         put_write_access(inode);
 214         __mnt_drop_write(mnt);
 215 }
 216
 217 /* the real guts of fput() - releasing the last reference to file
 218  */
 219 static void __fput(struct file *file)
 220 {
 221         struct dentry *dentry = file->f_path.dentry;
 222         struct vfsmount *mnt = file->f_path.mnt;
 223         struct inode *inode = file->f_inode;
 224
 225         might_sleep();
 226
 227         fsnotify_close(file);
 228         /*
 229          * The function eventpoll_release() should be the first called
 230          * in the file cleanup chain.
 231          */
 232         eventpoll_release(file);
 233         locks_remove_flock(file);
 234
 235         if (unlikely(file->f_flags & FASYNC)) {
 236                 if (file->f_op->fasync)
 237                         file->f_op->fasync(-1, file, 0);
 238         }
 239         ima_file_free(file);
 240         if (file->f_op->release)
 241                 file->f_op->release(inode, file);
 242         security_file_free(file);
 243         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
 244                      !(file->f_mode & FMODE_PATH))) {
 245                 cdev_put(inode->i_cdev);
 246         }
 247         fops_put(file->f_op);
 248         put_pid(file->f_owner.pid);
 249         if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 250                 i_readcount_dec(inode);
 251         if (file->f_mode & FMODE_WRITE)
 252                 drop_file_write_access(file);
 253         file->f_path.dentry = NULL;
 254         file->f_path.mnt = NULL;
 255         file->f_inode = NULL;
 256         file_free(file);
 257         dput(dentry);
 258         mntput(mnt);
 259 }
 260
 261 static LLIST_HEAD(delayed_fput_list);
 262 static void delayed_fput(struct work_struct *unused)
 263 {
 264         struct llist_node *node = llist_del_all(&delayed_fput_list);
 265         struct llist_node *next;
 266
 267         for (; node; node = next) {
 268                 next = llist_next(node);
 269                 __fput(llist_entry(node, struct file, f_u.fu_llist));
 270         }
 271 }
 272
 273 static void ____fput(struct callback_head *work)
 274 {
 275         __fput(container_of(work, struct file, f_u.fu_rcuhead));
 276 }
 277
 278 /*
 279  * If kernel thread really needs to have the final fput() it has done
 280  * to complete, call this.  The only user right now is the boot - we
 281  * *do* need to make sure our writes to binaries on initramfs has
 282  * not left us with opened struct file waiting for __fput() - execve()
 283  * won't work without that.  Please, don't add more callers without
 284  * very good reasons; in particular, never call that with locks
 285  * held and never call that from a thread that might need to do
 286  * some work on any kind of umount.
 287  */
 288 void flush_delayed_fput(void)
 289 {
 290         delayed_fput(NULL);
 291 }
 292
 293 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
 294
 295 void fput(struct file *file)
 296 {
 297         if (atomic_long_dec_and_test(&file->f_count)) {
 298                 struct task_struct *task = current;
 299
 300                 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
 301                         init_task_work(&file->f_u.fu_rcuhead, ____fput);
 302                         if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
 303                                 return;
 304                         /*
 305                          * After this task has run exit_task_work(),
 306                          * task_work_add() will fail.  Fall through to delayed
 307                          * fput to avoid leaking *file.
 308                          */
 309                 }
 310
 311                 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
 312                         schedule_delayed_work(&delayed_fput_work, 1);
 313         }
 314 }
 315
 316 /*
 317  * synchronous analog of fput(); for kernel threads that might be needed
 318  * in some umount() (and thus can't use flush_delayed_fput() without
 319  * risking deadlocks), need to wait for completion of __fput() and know
 320  * for this specific struct file it won't involve anything that would
 321  * need them.  Use only if you really need it - at the very least,
 322  * don't blindly convert fput() by kernel thread to that.
 323  */
 324 void __fput_sync(struct file *file)
 325 {
 326         if (atomic_long_dec_and_test(&file->f_count)) {
 327                 struct task_struct *task = current;
 328                 BUG_ON(!(task->flags & PF_KTHREAD));
 329                 __fput(file);
 330         }
 331 }
 332
 333 EXPORT_SYMBOL(fput);
 334
 335 void put_filp(struct file *file)
 336 {
 337         if (atomic_long_dec_and_test(&file->f_count)) {
 338                 security_file_free(file);
 339                 file_free(file);
 340         }
 341 }
 342
 343 void __init files_init(unsigned long mempages)
 344 {
 345         unsigned long n;
 346
 347         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 348                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 349
 350         /*
 351          * One file with associated inode and dcache is very roughly 1K.
 352          * Per default don't use more than 10% of our memory for files.
 353          */
 354
 355         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 356         files_stat.max_files = max_t(unsigned long, n, NR_FILE);
 357         files_defer_init();
 358         percpu_counter_init(&nr_files, 0);
 359 }