linux/fs/file_table.c
<<
>>
Prefs
   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/ima.h>
  27
  28#include <linux/atomic.h>
  29
  30#include "internal.h"
  31
  32/* sysctl tunables... */
  33struct files_stat_struct files_stat = {
  34        .max_files = NR_FILE
  35};
  36
  37DECLARE_LGLOCK(files_lglock);
  38DEFINE_LGLOCK(files_lglock);
  39
  40/* SLAB cache for file structures */
  41static struct kmem_cache *filp_cachep __read_mostly;
  42
  43static struct percpu_counter nr_files __cacheline_aligned_in_smp;
  44
  45static inline void file_free_rcu(struct rcu_head *head)
  46{
  47        struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
  48
  49        put_cred(f->f_cred);
  50        kmem_cache_free(filp_cachep, f);
  51}
  52
  53static inline void file_free(struct file *f)
  54{
  55        percpu_counter_dec(&nr_files);
  56        file_check_state(f);
  57        call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
  58}
  59
  60/*
  61 * Return the total number of open files in the system
  62 */
  63static long get_nr_files(void)
  64{
  65        return percpu_counter_read_positive(&nr_files);
  66}
  67
  68/*
  69 * Return the maximum number of open files in the system
  70 */
  71unsigned long get_max_files(void)
  72{
  73        return files_stat.max_files;
  74}
  75EXPORT_SYMBOL_GPL(get_max_files);
  76
  77/*
  78 * Handle nr_files sysctl
  79 */
  80#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
  81int proc_nr_files(ctl_table *table, int write,
  82                     void __user *buffer, size_t *lenp, loff_t *ppos)
  83{
  84        files_stat.nr_files = get_nr_files();
  85        return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
  86}
  87#else
  88int proc_nr_files(ctl_table *table, int write,
  89                     void __user *buffer, size_t *lenp, loff_t *ppos)
  90{
  91        return -ENOSYS;
  92}
  93#endif
  94
  95/* Find an unused file structure and return a pointer to it.
  96 * Returns NULL, if there are no more free file structures or
  97 * we run out of memory.
  98 *
  99 * Be very careful using this.  You are responsible for
 100 * getting write access to any mount that you might assign
 101 * to this filp, if it is opened for write.  If this is not
 102 * done, you will imbalance int the mount's writer count
 103 * and a warning at __fput() time.
 104 */
 105struct file *get_empty_filp(void)
 106{
 107        const struct cred *cred = current_cred();
 108        static long old_max;
 109        struct file * f;
 110
 111        /*
 112         * Privileged users can go above max_files
 113         */
 114        if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
 115                /*
 116                 * percpu_counters are inaccurate.  Do an expensive check before
 117                 * we go and fail.
 118                 */
 119                if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
 120                        goto over;
 121        }
 122
 123        f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
 124        if (f == NULL)
 125                goto fail;
 126
 127        percpu_counter_inc(&nr_files);
 128        f->f_cred = get_cred(cred);
 129        if (security_file_alloc(f))
 130                goto fail_sec;
 131
 132        INIT_LIST_HEAD(&f->f_u.fu_list);
 133        atomic_long_set(&f->f_count, 1);
 134        rwlock_init(&f->f_owner.lock);
 135        spin_lock_init(&f->f_lock);
 136        eventpoll_init_file(f);
 137        /* f->f_version: 0 */
 138        return f;
 139
 140over:
 141        /* Ran out of filps - report that */
 142        if (get_nr_files() > old_max) {
 143                pr_info("VFS: file-max limit %lu reached\n", get_max_files());
 144                old_max = get_nr_files();
 145        }
 146        goto fail;
 147
 148fail_sec:
 149        file_free(f);
 150fail:
 151        return NULL;
 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 */
 169struct 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 (!file)
 176                return NULL;
 177
 178        file->f_path = *path;
 179        file->f_mapping = path->dentry->d_inode->i_mapping;
 180        file->f_mode = mode;
 181        file->f_op = fop;
 182
 183        /*
 184         * These mounts don't really matter in practice
 185         * for r/o bind mounts.  They aren't userspace-
 186         * visible.  We do this for consistency, and so
 187         * that we can do debugging checks at __fput()
 188         */
 189        if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
 190                file_take_write(file);
 191                WARN_ON(mnt_clone_write(path->mnt));
 192        }
 193        if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 194                i_readcount_inc(path->dentry->d_inode);
 195        return file;
 196}
 197EXPORT_SYMBOL(alloc_file);
 198
 199/**
 200 * drop_file_write_access - give up ability to write to a file
 201 * @file: the file to which we will stop writing
 202 *
 203 * This is a central place which will give up the ability
 204 * to write to @file, along with access to write through
 205 * its vfsmount.
 206 */
 207void drop_file_write_access(struct file *file)
 208{
 209        struct vfsmount *mnt = file->f_path.mnt;
 210        struct dentry *dentry = file->f_path.dentry;
 211        struct inode *inode = dentry->d_inode;
 212
 213        put_write_access(inode);
 214
 215        if (special_file(inode->i_mode))
 216                return;
 217        if (file_check_writeable(file) != 0)
 218                return;
 219        mnt_drop_write(mnt);
 220        file_release_write(file);
 221}
 222EXPORT_SYMBOL_GPL(drop_file_write_access);
 223
 224/* the real guts of fput() - releasing the last reference to file
 225 */
 226static 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        file_sb_list_del(file);
 257        if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
 258                i_readcount_dec(inode);
 259        if (file->f_mode & FMODE_WRITE)
 260                drop_file_write_access(file);
 261        file->f_path.dentry = NULL;
 262        file->f_path.mnt = NULL;
 263        file_free(file);
 264        dput(dentry);
 265        mntput(mnt);
 266}
 267
 268void fput(struct file *file)
 269{
 270        if (atomic_long_dec_and_test(&file->f_count))
 271                __fput(file);
 272}
 273
 274EXPORT_SYMBOL(fput);
 275
 276struct file *fget(unsigned int fd)
 277{
 278        struct file *file;
 279        struct files_struct *files = current->files;
 280
 281        rcu_read_lock();
 282        file = fcheck_files(files, fd);
 283        if (file) {
 284                /* File object ref couldn't be taken */
 285                if (file->f_mode & FMODE_PATH ||
 286                    !atomic_long_inc_not_zero(&file->f_count))
 287                        file = NULL;
 288        }
 289        rcu_read_unlock();
 290
 291        return file;
 292}
 293
 294EXPORT_SYMBOL(fget);
 295
 296struct file *fget_raw(unsigned int fd)
 297{
 298        struct file *file;
 299        struct files_struct *files = current->files;
 300
 301        rcu_read_lock();
 302        file = fcheck_files(files, fd);
 303        if (file) {
 304                /* File object ref couldn't be taken */
 305                if (!atomic_long_inc_not_zero(&file->f_count))
 306                        file = NULL;
 307        }
 308        rcu_read_unlock();
 309
 310        return file;
 311}
 312
 313EXPORT_SYMBOL(fget_raw);
 314
 315/*
 316 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 317 *
 318 * You can use this instead of fget if you satisfy all of the following
 319 * conditions:
 320 * 1) You must call fput_light before exiting the syscall and returning control
 321 *    to userspace (i.e. you cannot remember the returned struct file * after
 322 *    returning to userspace).
 323 * 2) You must not call filp_close on the returned struct file * in between
 324 *    calls to fget_light and fput_light.
 325 * 3) You must not clone the current task in between the calls to fget_light
 326 *    and fput_light.
 327 *
 328 * The fput_needed flag returned by fget_light should be passed to the
 329 * corresponding fput_light.
 330 */
 331struct file *fget_light(unsigned int fd, int *fput_needed)
 332{
 333        struct file *file;
 334        struct files_struct *files = current->files;
 335
 336        *fput_needed = 0;
 337        if (atomic_read(&files->count) == 1) {
 338                file = fcheck_files(files, fd);
 339                if (file && (file->f_mode & FMODE_PATH))
 340                        file = NULL;
 341        } else {
 342                rcu_read_lock();
 343                file = fcheck_files(files, fd);
 344                if (file) {
 345                        if (!(file->f_mode & FMODE_PATH) &&
 346                            atomic_long_inc_not_zero(&file->f_count))
 347                                *fput_needed = 1;
 348                        else
 349                                /* Didn't get the reference, someone's freed */
 350                                file = NULL;
 351                }
 352                rcu_read_unlock();
 353        }
 354
 355        return file;
 356}
 357
 358struct file *fget_raw_light(unsigned int fd, int *fput_needed)
 359{
 360        struct file *file;
 361        struct files_struct *files = current->files;
 362
 363        *fput_needed = 0;
 364        if (atomic_read(&files->count) == 1) {
 365                file = fcheck_files(files, fd);
 366        } else {
 367                rcu_read_lock();
 368                file = fcheck_files(files, fd);
 369                if (file) {
 370                        if (atomic_long_inc_not_zero(&file->f_count))
 371                                *fput_needed = 1;
 372                        else
 373                                /* Didn't get the reference, someone's freed */
 374                                file = NULL;
 375                }
 376                rcu_read_unlock();
 377        }
 378
 379        return file;
 380}
 381
 382void put_filp(struct file *file)
 383{
 384        if (atomic_long_dec_and_test(&file->f_count)) {
 385                security_file_free(file);
 386                file_sb_list_del(file);
 387                file_free(file);
 388        }
 389}
 390
 391static inline int file_list_cpu(struct file *file)
 392{
 393#ifdef CONFIG_SMP
 394        return file->f_sb_list_cpu;
 395#else
 396        return smp_processor_id();
 397#endif
 398}
 399
 400/* helper for file_sb_list_add to reduce ifdefs */
 401static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
 402{
 403        struct list_head *list;
 404#ifdef CONFIG_SMP
 405        int cpu;
 406        cpu = smp_processor_id();
 407        file->f_sb_list_cpu = cpu;
 408        list = per_cpu_ptr(sb->s_files, cpu);
 409#else
 410        list = &sb->s_files;
 411#endif
 412        list_add(&file->f_u.fu_list, list);
 413}
 414
 415/**
 416 * file_sb_list_add - add a file to the sb's file list
 417 * @file: file to add
 418 * @sb: sb to add it to
 419 *
 420 * Use this function to associate a file with the superblock of the inode it
 421 * refers to.
 422 */
 423void file_sb_list_add(struct file *file, struct super_block *sb)
 424{
 425        lg_local_lock(files_lglock);
 426        __file_sb_list_add(file, sb);
 427        lg_local_unlock(files_lglock);
 428}
 429
 430/**
 431 * file_sb_list_del - remove a file from the sb's file list
 432 * @file: file to remove
 433 * @sb: sb to remove it from
 434 *
 435 * Use this function to remove a file from its superblock.
 436 */
 437void file_sb_list_del(struct file *file)
 438{
 439        if (!list_empty(&file->f_u.fu_list)) {
 440                lg_local_lock_cpu(files_lglock, file_list_cpu(file));
 441                list_del_init(&file->f_u.fu_list);
 442                lg_local_unlock_cpu(files_lglock, file_list_cpu(file));
 443        }
 444}
 445
 446#ifdef CONFIG_SMP
 447
 448/*
 449 * These macros iterate all files on all CPUs for a given superblock.
 450 * files_lglock must be held globally.
 451 */
 452#define do_file_list_for_each_entry(__sb, __file)               \
 453{                                                               \
 454        int i;                                                  \
 455        for_each_possible_cpu(i) {                              \
 456                struct list_head *list;                         \
 457                list = per_cpu_ptr((__sb)->s_files, i);         \
 458                list_for_each_entry((__file), list, f_u.fu_list)
 459
 460#define while_file_list_for_each_entry                          \
 461        }                                                       \
 462}
 463
 464#else
 465
 466#define do_file_list_for_each_entry(__sb, __file)               \
 467{                                                               \
 468        struct list_head *list;                                 \
 469        list = &(sb)->s_files;                                  \
 470        list_for_each_entry((__file), list, f_u.fu_list)
 471
 472#define while_file_list_for_each_entry                          \
 473}
 474
 475#endif
 476
 477int fs_may_remount_ro(struct super_block *sb)
 478{
 479        struct file *file;
 480        /* Check that no files are currently opened for writing. */
 481        lg_global_lock(files_lglock);
 482        do_file_list_for_each_entry(sb, file) {
 483                struct inode *inode = file->f_path.dentry->d_inode;
 484
 485                /* File with pending delete? */
 486                if (inode->i_nlink == 0)
 487                        goto too_bad;
 488
 489                /* Writeable file? */
 490                if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
 491                        goto too_bad;
 492        } while_file_list_for_each_entry;
 493        lg_global_unlock(files_lglock);
 494        return 1; /* Tis' cool bro. */
 495too_bad:
 496        lg_global_unlock(files_lglock);
 497        return 0;
 498}
 499
 500/**
 501 *      mark_files_ro - mark all files read-only
 502 *      @sb: superblock in question
 503 *
 504 *      All files are marked read-only.  We don't care about pending
 505 *      delete files so this should be used in 'force' mode only.
 506 */
 507void mark_files_ro(struct super_block *sb)
 508{
 509        struct file *f;
 510
 511retry:
 512        lg_global_lock(files_lglock);
 513        do_file_list_for_each_entry(sb, f) {
 514                struct vfsmount *mnt;
 515                if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
 516                       continue;
 517                if (!file_count(f))
 518                        continue;
 519                if (!(f->f_mode & FMODE_WRITE))
 520                        continue;
 521                spin_lock(&f->f_lock);
 522                f->f_mode &= ~FMODE_WRITE;
 523                spin_unlock(&f->f_lock);
 524                if (file_check_writeable(f) != 0)
 525                        continue;
 526                file_release_write(f);
 527                mnt = mntget(f->f_path.mnt);
 528                /* This can sleep, so we can't hold the spinlock. */
 529                lg_global_unlock(files_lglock);
 530                mnt_drop_write(mnt);
 531                mntput(mnt);
 532                goto retry;
 533        } while_file_list_for_each_entry;
 534        lg_global_unlock(files_lglock);
 535}
 536
 537void __init files_init(unsigned long mempages)
 538{ 
 539        unsigned long n;
 540
 541        filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
 542                        SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
 543
 544        /*
 545         * One file with associated inode and dcache is very roughly 1K.
 546         * Per default don't use more than 10% of our memory for files. 
 547         */ 
 548
 549        n = (mempages * (PAGE_SIZE / 1024)) / 10;
 550        files_stat.max_files = max_t(unsigned long, n, NR_FILE);
 551        files_defer_init();
 552        lg_lock_init(files_lglock);
 553        percpu_counter_init(&nr_files, 0);
 554} 
 555
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.