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_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 inline 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 NULL, if there are no more free file structures or
  98  * we run out of memory.
  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
 112         /*
 113          * Privileged users can go above max_files
 114          */
 115         if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 116                 /*
 117                  * percpu_counters are inaccurate.  Do an expensive check before
 118                  * we go and fail.
 119                  */
 120                 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 121                         goto over;
 122         }
 123
 124         f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 125         if (f == NULL)
 126                 goto fail;
 127
 128         percpu_counter_inc(&nr_files);
 129         f->f_cred = get_cred(cred);
 130         if (security_file_alloc(f))
 131                 goto fail_sec;
 132
 133         INIT_LIST_HEAD(&f->f_u.fu_list);
 134         atomic_long_set(&f->f_count, 1);
 135         rwlock_init(&f->f_owner.lock);
 136         spin_lock_init(&f->f_lock);
 137         eventpoll_init_file(f);
 138         /* f->f_version: 0 */
 139         return f;
 140
 141 over:
 142         /* Ran out of filps - report that */
 143         if (get_nr_files() > old_max) {
 144                 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
 145                 old_max = get_nr_files();
 146         }
 147         goto fail;
 148
 149 fail_sec:
 150         file_free(f);
 151 fail:
 152         return NULL;
 153 }
 154
 155 /**
 156  * alloc_file - allocate and initialize a 'struct file'
 157  * @mnt: the vfsmount on which the file will reside
 158  * @dentry: the dentry representing the new file
 159  * @mode: the mode with which the new file will be opened
 160  * @fop: the 'struct file_operations' for the new file
 161  *
 162  * Use this instead of get_empty_filp() to get a new
 163  * 'struct file'.  Do so because of the same initialization
 164  * pitfalls reasons listed for init_file().  This is a
 165  * preferred interface to using init_file().
 166  *
 167  * If all the callers of init_file() are eliminated, its
 168  * code should be moved into this function.
 169  */
 170 struct file *alloc_file(struct path *path, fmode_t mode,
 171                 const struct file_operations *fop)
 172 {
 173         struct file *file;
 174
 175         file = get_empty_filp();
 176         if (!file)
 177                 return NULL;
 178
 179         file->f_path = *path;
 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 = dentry->d_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         if (file->f_op && file->f_op->release)
 247                 file->f_op->release(inode, file);
 248         security_file_free(file);
 249         ima_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_free(file);
 263         dput(dentry);
 264         mntput(mnt);
 265 }
 266
 267 static DEFINE_SPINLOCK(delayed_fput_lock);
 268 static LIST_HEAD(delayed_fput_list);
 269 static void delayed_fput(struct work_struct *unused)
 270 {
 271         LIST_HEAD(head);
 272         spin_lock_irq(&delayed_fput_lock);
 273         list_splice_init(&delayed_fput_list, &head);
 274         spin_unlock_irq(&delayed_fput_lock);
 275         while (!list_empty(&head)) {
 276                 struct file *f = list_first_entry(&head, struct file, f_u.fu_list);
 277                 list_del_init(&f->f_u.fu_list);
 278                 __fput(f);
 279         }
 280 }
 281
 282 static void ____fput(struct callback_head *work)
 283 {
 284         __fput(container_of(work, struct file, f_u.fu_rcuhead));
 285 }
 286
 287 /*
 288  * If kernel thread really needs to have the final fput() it has done
 289  * to complete, call this.  The only user right now is the boot - we
 290  * *do* need to make sure our writes to binaries on initramfs has
 291  * not left us with opened struct file waiting for __fput() - execve()
 292  * won't work without that.  Please, don't add more callers without
 293  * very good reasons; in particular, never call that with locks
 294  * held and never call that from a thread that might need to do
 295  * some work on any kind of umount.
 296  */
 297 void flush_delayed_fput(void)
 298 {
 299         delayed_fput(NULL);
 300 }
 301
 302 static DECLARE_WORK(delayed_fput_work, delayed_fput);
 303
 304 void fput(struct file *file)
 305 {
 306         if (atomic_long_dec_and_test(&file->f_count)) {
 307                 struct task_struct *task = current;
 308                 file_sb_list_del(file);
 309                 if (unlikely(in_interrupt() || task->flags & PF_KTHREAD)) {
 310                         unsigned long flags;
 311                         spin_lock_irqsave(&delayed_fput_lock, flags);
 312                         list_add(&file->f_u.fu_list, &delayed_fput_list);
 313                         schedule_work(&delayed_fput_work);
 314                         spin_unlock_irqrestore(&delayed_fput_lock, flags);
 315                         return;
 316                 }
 317                 init_task_work(&file->f_u.fu_rcuhead, ____fput);
 318                 task_work_add(task, &file->f_u.fu_rcuhead, true);
 319         }
 320 }
 321
 322 /*
 323  * synchronous analog of fput(); for kernel threads that might be needed
 324  * in some umount() (and thus can't use flush_delayed_fput() without
 325  * risking deadlocks), need to wait for completion of __fput() and know
 326  * for this specific struct file it won't involve anything that would
 327  * need them.  Use only if you really need it - at the very least,
 328  * don't blindly convert fput() by kernel thread to that.
 329  */
 330 void __fput_sync(struct file *file)
 331 {
 332         if (atomic_long_dec_and_test(&file->f_count)) {
 333                 struct task_struct *task = current;
 334                 file_sb_list_del(file);
 335                 BUG_ON(!(task->flags & PF_KTHREAD));
 336                 __fput(file);
 337         }
 338 }
 339
 340 EXPORT_SYMBOL(fput);
 341
 342 struct file *fget(unsigned int fd)
 343 {
 344         struct file *file;
 345         struct files_struct *files = current->files;
 346
 347         rcu_read_lock();
 348         file = fcheck_files(files, fd);
 349         if (file) {
 350                 /* File object ref couldn't be taken */
 351                 if (file->f_mode & FMODE_PATH ||
 352                     !atomic_long_inc_not_zero(&file->f_count))
 353                         file = NULL;
 354         }
 355         rcu_read_unlock();
 356
 357         return file;
 358 }
 359
 360 EXPORT_SYMBOL(fget);
 361
 362 struct file *fget_raw(unsigned int fd)
 363 {
 364         struct file *file;
 365         struct files_struct *files = current->files;
 366
 367         rcu_read_lock();
 368         file = fcheck_files(files, fd);
 369         if (file) {
 370                 /* File object ref couldn't be taken */
 371                 if (!atomic_long_inc_not_zero(&file->f_count))
 372                         file = NULL;
 373         }
 374         rcu_read_unlock();
 375
 376         return file;
 377 }
 378
 379 EXPORT_SYMBOL(fget_raw);
 380
 381 /*
 382  * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 383  *
 384  * You can use this instead of fget if you satisfy all of the following
 385  * conditions:
 386  * 1) You must call fput_light before exiting the syscall and returning control
 387  *    to userspace (i.e. you cannot remember the returned struct file * after
 388  *    returning to userspace).
 389  * 2) You must not call filp_close on the returned struct file * in between
 390  *    calls to fget_light and fput_light.
 391  * 3) You must not clone the current task in between the calls to fget_light
 392  *    and fput_light.
 393  *
 394  * The fput_needed flag returned by fget_light should be passed to the
 395  * corresponding fput_light.
 396  */
 397 struct file *fget_light(unsigned int fd, int *fput_needed)
 398 {
 399         struct file *file;
 400         struct files_struct *files = current->files;
 401
 402         *fput_needed = 0;
 403         if (atomic_read(&files->count) == 1) {
 404                 file = fcheck_files(files, fd);
 405                 if (file && (file->f_mode & FMODE_PATH))
 406                         file = NULL;
 407         } else {
 408                 rcu_read_lock();
 409                 file = fcheck_files(files, fd);
 410                 if (file) {
 411                         if (!(file->f_mode & FMODE_PATH) &&
 412                             atomic_long_inc_not_zero(&file->f_count))
 413                                 *fput_needed = 1;
 414                         else
 415                                 /* Didn't get the reference, someone's freed */
 416                                 file = NULL;
 417                 }
 418                 rcu_read_unlock();
 419         }
 420
 421         return file;
 422 }
 423
 424 struct file *fget_raw_light(unsigned int fd, int *fput_needed)
 425 {
 426         struct file *file;
 427         struct files_struct *files = current->files;
 428
 429         *fput_needed = 0;
 430         if (atomic_read(&files->count) == 1) {
 431                 file = fcheck_files(files, fd);
 432         } else {
 433                 rcu_read_lock();
 434                 file = fcheck_files(files, fd);
 435                 if (file) {
 436                         if (atomic_long_inc_not_zero(&file->f_count))
 437                                 *fput_needed = 1;
 438                         else
 439                                 /* Didn't get the reference, someone's freed */
 440                                 file = NULL;
 441                 }
 442                 rcu_read_unlock();
 443         }
 444
 445         return file;
 446 }
 447
 448 void put_filp(struct file *file)
 449 {
 450         if (atomic_long_dec_and_test(&file->f_count)) {
 451                 security_file_free(file);
 452                 file_sb_list_del(file);
 453                 file_free(file);
 454         }
 455 }
 456
 457 static inline int file_list_cpu(struct file *file)
 458 {
 459 #ifdef CONFIG_SMP
 460         return file->f_sb_list_cpu;
 461 #else
 462         return smp_processor_id();
 463 #endif
 464 }
 465
 466 /* helper for file_sb_list_add to reduce ifdefs */
 467 static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
 468 {
 469         struct list_head *list;
 470 #ifdef CONFIG_SMP
 471         int cpu;
 472         cpu = smp_processor_id();
 473         file->f_sb_list_cpu = cpu;
 474         list = per_cpu_ptr(sb->s_files, cpu);
 475 #else
 476         list = &sb->s_files;
 477 #endif
 478         list_add(&file->f_u.fu_list, list);
 479 }
 480
 481 /**
 482  * file_sb_list_add - add a file to the sb's file list
 483  * @file: file to add
 484  * @sb: sb to add it to
 485  *
 486  * Use this function to associate a file with the superblock of the inode it
 487  * refers to.
 488  */
 489 void file_sb_list_add(struct file *file, struct super_block *sb)
 490 {
 491         lg_local_lock(&files_lglock);
 492         __file_sb_list_add(file, sb);
 493         lg_local_unlock(&files_lglock);
 494 }
 495
 496 /**
 497  * file_sb_list_del - remove a file from the sb's file list
 498  * @file: file to remove
 499  * @sb: sb to remove it from
 500  *
 501  * Use this function to remove a file from its superblock.
 502  */
 503 void file_sb_list_del(struct file *file)
 504 {
 505         if (!list_empty(&file->f_u.fu_list)) {
 506                 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
 507                 list_del_init(&file->f_u.fu_list);
 508                 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
 509         }
 510 }
 511
 512 #ifdef CONFIG_SMP
 513
 514 /*
 515  * These macros iterate all files on all CPUs for a given superblock.
 516  * files_lglock must be held globally.
 517  */
 518 #define do_file_list_for_each_entry(__sb, __file)               \
 519 {                                                               \
 520         int i;                                                  \
 521         for_each_possible_cpu(i) {                              \
 522                 struct list_head *list;                         \
 523                 list = per_cpu_ptr((__sb)->s_files, i);         \
 524                 list_for_each_entry((__file), list, f_u.fu_list)
 525
 526 #define while_file_list_for_each_entry                          \
 527         }                                                       \
 528 }
 529
 530 #else
 531
 532 #define do_file_list_for_each_entry(__sb, __file)               \
 533 {                                                               \
 534         struct list_head *list;                                 \
 535         list = &(sb)->s_files;                                  \
 536         list_for_each_entry((__file), list, f_u.fu_list)
 537
 538 #define while_file_list_for_each_entry                          \
 539 }
 540
 541 #endif
 542
 543 /**
 544  *      mark_files_ro - mark all files read-only
 545  *      @sb: superblock in question
 546  *
 547  *      All files are marked read-only.  We don't care about pending
 548  *      delete files so this should be used in 'force' mode only.
 549  */
 550 void mark_files_ro(struct super_block *sb)
 551 {
 552         struct file *f;
 553
 554         lg_global_lock(&files_lglock);
 555         do_file_list_for_each_entry(sb, f) {
 556                 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
 557                        continue;
 558                 if (!file_count(f))
 559                         continue;
 560                 if (!(f->f_mode & FMODE_WRITE))
 561                         continue;
 562                 spin_lock(&f->f_lock);
 563                 f->f_mode &= ~FMODE_WRITE;
 564                 spin_unlock(&f->f_lock);
 565                 if (file_check_writeable(f) != 0)
 566                         continue;
 567                 file_release_write(f);
 568                 mnt_drop_write_file(f);
 569         } while_file_list_for_each_entry;
 570         lg_global_unlock(&files_lglock);
 571 }
 572
 573 void __init files_init(unsigned long mempages)
 574 {
 575         unsigned long n;
 576
 577         filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 578                         SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 579
 580         /*
 581          * One file with associated inode and dcache is very roughly 1K.
 582          * Per default don't use more than 10% of our memory for files.
 583          */
 584
 585         n = (mempages * (PAGE_SIZE / 1024)) / 10;
 586         files_stat.max_files = max_t(unsigned long, n, NR_FILE);
 587         files_defer_init();
 588         lg_lock_init(&files_lglock, "files_lglock");
 589         percpu_counter_init(&nr_files, 0);
 590 }