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 DEFINE_STATIC_LGLOCK(files_lglock);
  40
  41 /* SLAB cache for file structures */
  42 static struct kmem_cache *filp_cachep __read_mostly;
  43
  44 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  45
  46 static void file_free_rcu(struct rcu_head *head)
  47 {
  48         struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  49
  50         put_cred(f->f_cred);
  51         kmem_cache_free(filp_cachep, f);
  52 }
  53
  54 static inline void file_free(struct file *f)
  55 {
  56         percpu_counter_dec(&nr_files);
  57         file_check_state(f);
  58         call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  59 }
  60
  61 /*
  62  * Return the total number of open files in the system
  63  */
  64 static long get_nr_files(void)
  65 {
  66         return percpu_counter_read_positive(&nr_files);
  67 }
  68
  69 /*
  70  * Return the maximum number of open files in the system
  71  */
  72 unsigned long get_max_files(void)
  73 {
  74         return files_stat.max_files;
  75 }
  76 EXPORT_SYMBOL_GPL(get_max_files);
  77
  78 /*
  79  * Handle nr_files sysctl
  80  */
  81 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  82 int proc_nr_files(ctl_table *table, int write,
  83                      void __user *buffer, size_t *lenp, loff_t *ppos)
  84 {
  85         files_stat.nr_files = get_nr_files();
  86         return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
  87 }
  88 #else
  89 int proc_nr_files(ctl_table *table, int write,
  90                      void __user *buffer, size_t *lenp, loff_t *ppos)
  91 {
  92         return -ENOSYS;
  93 }
  94 #endif
  95
  96 /* Find an unused file structure and return a pointer to it.
  97  * Returns an error pointer if some error happend e.g. we over file
  98  * structures limit, run out of memory or operation is not permitted.
  99  *
 100  * Be very careful using this.  You are responsible for
 101  * getting write access to any mount that you might assign
 102  * to this filp, if it is opened for write.  If this is not
 103  * done, you will imbalance int the mount's writer count
 104  * and a warning at __fput() time.
 105  */
 106 struct file *get_empty_filp(void)
 107 {
 108         const struct cred *cred = current_cred();
 109         static long old_max;
 110         struct file *f;
 111         int error;
 112
 113         /*
 114          * Privileged users can go above max_files
 115          */
 116         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 117                 /*
 118                  * percpu_counters are inaccurate.  Do an expensive check before
 119                  * we go and fail.
 120                  */
 121                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 122                         goto over;
 123         }
 124
 125         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 126         if (unlikely(!f))
 127                 return ERR_PTR(-ENOMEM);
 128
 129         percpu_counter_inc(&nr_files);
 130         f->f_cred = get_cred(cred);
 131         error = security_file_alloc(f);
 132         if (unlikely(error)) {
 133                 file_free(f);
 134                 return ERR_PTR(error);
 135         }
 136
 137         INIT_LIST_HEAD(&f->f_u.fu_list);
 138         atomic_long_set(&f->f_count, 1);
 139         rwlock_init(&f->f_owner.lock);
 140         spin_lock_init(&f->f_lock);
 141         eventpoll_init_file(f);
 142         /* f->f_version: 0 */
 143         return f;
 144
 145 over:
 146         /* Ran out of filps - report that */
 147         if (get_nr_files() > old_max) {
 148                 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
 149                 old_max = get_nr_files();
 150         }
 151         return ERR_PTR(-ENFILE);
 152 }
 153
 154 /**
 155  * alloc_file - allocate and initialize a 'struct file'
 156  * @mnt: the vfsmount on which the file will reside
 157  * @dentry: the dentry representing the new file
 158  * @mode: the mode with which the new file will be opened
 159  * @fop: the 'struct file_operations' for the new file
 160  *
 161  * Use this instead of get_empty_filp() to get a new
 162  * 'struct file'.  Do so because of the same initialization
 163  * pitfalls reasons listed for init_file().  This is a
 164  * preferred interface to using init_file().
 165  *
 166  * If all the callers of init_file() are eliminated, its
 167  * code should be moved into this function.
 168  */
 169 struct file *alloc_file(struct path *path, fmode_t mode,
 170                 const struct file_operations *fop)
 171 {
 172         struct file *file;
 173
 174         file = get_empty_filp();
 175         if (IS_ERR(file))
 176                 return file;
 177
 178         file->f_path = *path;
 179         file->f_inode = path->dentry->d_inode;
 180         file->f_mapping = path->dentry->d_inode->i_mapping;
 181         file->f_mode = mode;
 182         file->f_op = fop;
 183
 184         /*
 185          * These mounts don't really matter in practice
 186          * for r/o bind mounts.  They aren't userspace-
 187          * visible.  We do this for consistency, and so
 188          * that we can do debugging checks at __fput()
 189          */
 190         if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
 191                 file_take_write(file);
 192                 WARN_ON(mnt_clone_write(path->mnt));
 193         }
 194         if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 195                 i_readcount_inc(path->dentry->d_inode);
 196         return file;
 197 }
 198 EXPORT_SYMBOL(alloc_file);
 199
 200 /**
 201  * drop_file_write_access - give up ability to write to a file
 202  * @file: the file to which we will stop writing
 203  *
 204  * This is a central place which will give up the ability
 205  * to write to @file, along with access to write through
 206  * its vfsmount.
 207  */
 208 static void drop_file_write_access(struct file *file)
 209 {
 210         struct vfsmount *mnt = file->f_path.mnt;
 211         struct dentry *dentry = file->f_path.dentry;
 212         struct inode *inode = dentry->d_inode;
 213
 214         put_write_access(inode);
 215
 216         if (special_file(inode->i_mode))
 217                 return;
 218         if (file_check_writeable(file) != 0)
 219                 return;
 220         __mnt_drop_write(mnt);
 221         file_release_write(file);
 222 }
 223
 224 /* the real guts of fput() - releasing the last reference to file
 225  */
 226 static void __fput(struct file *file)
 227 {
 228         struct dentry *dentry = file->f_path.dentry;
 229         struct vfsmount *mnt = file->f_path.mnt;
 230         struct inode *inode = file->f_inode;
 231
 232         might_sleep();
 233
 234         fsnotify_close(file);
 235         /*
 236          * The function eventpoll_release() should be the first called
 237          * in the file cleanup chain.
 238          */
 239         eventpoll_release(file);
 240         locks_remove_flock(file);
 241
 242         if (unlikely(file->f_flags & FASYNC)) {
 243                 if (file->f_op && file->f_op->fasync)
 244                         file->f_op->fasync(-1, file, 0);
 245         }
 246         ima_file_free(file);
 247         if (file->f_op && file->f_op->release)
 248                 file->f_op->release(inode, file);
 249         security_file_free(file);
 250         if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
 251                      !(file->f_mode & FMODE_PATH))) {
 252                 cdev_put(inode->i_cdev);
 253         }
 254         fops_put(file->f_op);
 255         put_pid(file->f_owner.pid);
 256         if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 257                 i_readcount_dec(inode);
 258         if (file->f_mode & FMODE_WRITE)
 259                 drop_file_write_access(file);
 260         file->f_path.dentry = NULL;
 261         file->f_path.mnt = NULL;
 262         file->f_inode = NULL;
 263         file_free(file);
 264         dput(dentry);
 265         mntput(mnt);
 266 }
 267
 268 static LLIST_HEAD(delayed_fput_list);
 269 static void delayed_fput(struct work_struct *unused)
 270 {
 271         struct llist_node *node = llist_del_all(&delayed_fput_list);
 272         struct llist_node *next;
 273
 274         for (; node; node = next) {
 275                 next = llist_next(node);
 276                 __fput(llist_entry(node, struct file, f_u.fu_llist));
 277         }
 278 }
 279
 280 static void ____fput(struct callback_head *work)
 281 {
 282         __fput(container_of(work, struct file, f_u.fu_rcuhead));
 283 }
 284
 285 /*
 286  * If kernel thread really needs to have the final fput() it has done
 287  * to complete, call this.  The only user right now is the boot - we
 288  * *do* need to make sure our writes to binaries on initramfs has
 289  * not left us with opened struct file waiting for __fput() - execve()
 290  * won't work without that.  Please, don't add more callers without
 291  * very good reasons; in particular, never call that with locks
 292  * held and never call that from a thread that might need to do
 293  * some work on any kind of umount.
 294  */
 295 void flush_delayed_fput(void)
 296 {
 297         delayed_fput(NULL);
 298 }
 299
 300 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
 301
 302 void fput(struct file *file)
 303 {
 304         if (atomic_long_dec_and_test(&file->f_count)) {
 305                 struct task_struct *task = current;
 306
 307                 file_sb_list_del(file);
 308                 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
 309                         init_task_work(&file->f_u.fu_rcuhead, ____fput);
 310                         if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
 311                                 return;
 312                         /*
 313                          * After this task has run exit_task_work(),
 314                          * task_work_add() will fail.  Fall through to delayed
 315                          * fput to avoid leaking *file.
 316                          */
 317                 }
 318
 319                 if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
 320                         schedule_delayed_work(&delayed_fput_work, 1);
 321         }
 322 }
 323
 324 /*
 325  * synchronous analog of fput(); for kernel threads that might be needed
 326  * in some umount() (and thus can't use flush_delayed_fput() without
 327  * risking deadlocks), need to wait for completion of __fput() and know
 328  * for this specific struct file it won't involve anything that would
 329  * need them.  Use only if you really need it - at the very least,
 330  * don't blindly convert fput() by kernel thread to that.
 331  */
 332 void __fput_sync(struct file *file)
 333 {
 334         if (atomic_long_dec_and_test(&file->f_count)) {
 335                 struct task_struct *task = current;
 336                 file_sb_list_del(file);
 337                 BUG_ON(!(task->flags & PF_KTHREAD));
 338                 __fput(file);
 339         }
 340 }
 341
 342 EXPORT_SYMBOL(fput);
 343
 344 void put_filp(struct file *file)
 345 {
 346         if (atomic_long_dec_and_test(&file->f_count)) {
 347                 security_file_free(file);
 348                 file_sb_list_del(file);
 349                 file_free(file);
 350         }
 351 }
 352
 353 static inline int file_list_cpu(struct file *file)
 354 {
 355 #ifdef CONFIG_SMP
 356         return file->f_sb_list_cpu;
 357 #else
 358         return smp_processor_id();
 359 #endif
 360 }
 361
 362 /* helper for file_sb_list_add to reduce ifdefs */
 363 static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
 364 {
 365         struct list_head *list;
 366 #ifdef CONFIG_SMP
 367         int cpu;
 368         cpu = smp_processor_id();
 369         file->f_sb_list_cpu = cpu;
 370         list = per_cpu_ptr(sb->s_files, cpu);
 371 #else
 372         list = &sb->s_files;
 373 #endif
 374         list_add(&file->f_u.fu_list, list);
 375 }
 376
 377 /**
 378  * file_sb_list_add - add a file to the sb's file list
 379  * @file: file to add
 380  * @sb: sb to add it to
 381  *
 382  * Use this function to associate a file with the superblock of the inode it
 383  * refers to.
 384  */
 385 void file_sb_list_add(struct file *file, struct super_block *sb)
 386 {
 387         if (likely(!(file->f_mode & FMODE_WRITE)))
 388                 return;
 389         if (!S_ISREG(file_inode(file)->i_mode))
 390                 return;
 391         lg_local_lock(&files_lglock);
 392         __file_sb_list_add(file, sb);
 393         lg_local_unlock(&files_lglock);
 394 }
 395
 396 /**
 397  * file_sb_list_del - remove a file from the sb's file list
 398  * @file: file to remove
 399  * @sb: sb to remove it from
 400  *
 401  * Use this function to remove a file from its superblock.
 402  */
 403 void file_sb_list_del(struct file *file)
 404 {
 405         if (!list_empty(&file->f_u.fu_list)) {
 406                 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
 407                 list_del_init(&file->f_u.fu_list);
 408                 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
 409         }
 410 }
 411
 412 #ifdef CONFIG_SMP
 413
 414 /*
 415  * These macros iterate all files on all CPUs for a given superblock.
 416  * files_lglock must be held globally.
 417  */
 418 #define do_file_list_for_each_entry(__sb, __file)               \
 419 {                                                               \
 420         int i;                                                  \
 421         for_each_possible_cpu(i) {                              \
 422                 struct list_head *list;                         \
 423                 list = per_cpu_ptr((__sb)->s_files, i);         \
 424                 list_for_each_entry((__file), list, f_u.fu_list)
 425
 426 #define while_file_list_for_each_entry                          \
 427         }                                                       \
 428 }
 429
 430 #else
 431
 432 #define do_file_list_for_each_entry(__sb, __file)               \
 433 {                                                               \
 434         struct list_head *list;                                 \
 435         list = &(sb)->s_files;                                  \
 436         list_for_each_entry((__file), list, f_u.fu_list)
 437
 438 #define while_file_list_for_each_entry                          \
 439 }
 440
 441 #endif
 442
 443 /**
 444  *      mark_files_ro - mark all files read-only
 445  *      @sb: superblock in question
 446  *
 447  *      All files are marked read-only.  We don't care about pending
 448  *      delete files so this should be used in 'force' mode only.
 449  */
 450 void mark_files_ro(struct super_block *sb)
 451 {
 452         struct file *f;
 453
 454         lg_global_lock(&files_lglock);
 455         do_file_list_for_each_entry(sb, f) {
 456                 if (!file_count(f))
 457                         continue;
 458                 if (!(f->f_mode & FMODE_WRITE))
 459                         continue;
 460                 spin_lock(&f->f_lock);
 461                 f->f_mode &= ~FMODE_WRITE;
 462                 spin_unlock(&f->f_lock);
 463                 if (file_check_writeable(f) != 0)
 464                         continue;
 465                 __mnt_drop_write(f->f_path.mnt);
 466                 file_release_write(f);
 467         } while_file_list_for_each_entry;
 468         lg_global_unlock(&files_lglock);
 469 }
 470
 471 void __init files_init(unsigned long mempages)
 472 {
 473         unsigned long n;
 474
 475         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 476                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 477
 478         /*
 479          * One file with associated inode and dcache is very roughly 1K.
 480          * Per default don't use more than 10% of our memory for files.
 481          */
 482
 483         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 484         files_stat.max_files = max_t(unsigned long, n, NR_FILE);
 485         files_defer_init();
 486         lg_lock_init(&files_lglock, "files_lglock");
 487         percpu_counter_init(&nr_files, 0);
 488 }