linux/fs/file.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 *  linux/fs/file.c
   4 *
   5 *  Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes
   6 *
   7 *  Manage the dynamic fd arrays in the process files_struct.
   8 */
   9
  10#include <linux/syscalls.h>
  11#include <linux/export.h>
  12#include <linux/fs.h>
  13#include <linux/kernel.h>
  14#include <linux/mm.h>
  15#include <linux/sched/signal.h>
  16#include <linux/slab.h>
  17#include <linux/file.h>
  18#include <linux/fdtable.h>
  19#include <linux/bitops.h>
  20#include <linux/spinlock.h>
  21#include <linux/rcupdate.h>
  22#include <linux/close_range.h>
  23#include <net/sock.h>
  24
  25#include "internal.h"
  26
  27unsigned int sysctl_nr_open __read_mostly = 1024*1024;
  28unsigned int sysctl_nr_open_min = BITS_PER_LONG;
  29/* our min() is unusable in constant expressions ;-/ */
  30#define __const_min(x, y) ((x) < (y) ? (x) : (y))
  31unsigned int sysctl_nr_open_max =
  32        __const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG;
  33
  34static void __free_fdtable(struct fdtable *fdt)
  35{
  36        kvfree(fdt->fd);
  37        kvfree(fdt->open_fds);
  38        kfree(fdt);
  39}
  40
  41static void free_fdtable_rcu(struct rcu_head *rcu)
  42{
  43        __free_fdtable(container_of(rcu, struct fdtable, rcu));
  44}
  45
  46#define BITBIT_NR(nr)   BITS_TO_LONGS(BITS_TO_LONGS(nr))
  47#define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
  48
  49/*
  50 * Copy 'count' fd bits from the old table to the new table and clear the extra
  51 * space if any.  This does not copy the file pointers.  Called with the files
  52 * spinlock held for write.
  53 */
  54static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
  55                            unsigned int count)
  56{
  57        unsigned int cpy, set;
  58
  59        cpy = count / BITS_PER_BYTE;
  60        set = (nfdt->max_fds - count) / BITS_PER_BYTE;
  61        memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
  62        memset((char *)nfdt->open_fds + cpy, 0, set);
  63        memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
  64        memset((char *)nfdt->close_on_exec + cpy, 0, set);
  65
  66        cpy = BITBIT_SIZE(count);
  67        set = BITBIT_SIZE(nfdt->max_fds) - cpy;
  68        memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
  69        memset((char *)nfdt->full_fds_bits + cpy, 0, set);
  70}
  71
  72/*
  73 * Copy all file descriptors from the old table to the new, expanded table and
  74 * clear the extra space.  Called with the files spinlock held for write.
  75 */
  76static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
  77{
  78        size_t cpy, set;
  79
  80        BUG_ON(nfdt->max_fds < ofdt->max_fds);
  81
  82        cpy = ofdt->max_fds * sizeof(struct file *);
  83        set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
  84        memcpy(nfdt->fd, ofdt->fd, cpy);
  85        memset((char *)nfdt->fd + cpy, 0, set);
  86
  87        copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
  88}
  89
  90static struct fdtable * alloc_fdtable(unsigned int nr)
  91{
  92        struct fdtable *fdt;
  93        void *data;
  94
  95        /*
  96         * Figure out how many fds we actually want to support in this fdtable.
  97         * Allocation steps are keyed to the size of the fdarray, since it
  98         * grows far faster than any of the other dynamic data. We try to fit
  99         * the fdarray into comfortable page-tuned chunks: starting at 1024B
 100         * and growing in powers of two from there on.
 101         */
 102        nr /= (1024 / sizeof(struct file *));
 103        nr = roundup_pow_of_two(nr + 1);
 104        nr *= (1024 / sizeof(struct file *));
 105        /*
 106         * Note that this can drive nr *below* what we had passed if sysctl_nr_open
 107         * had been set lower between the check in expand_files() and here.  Deal
 108         * with that in caller, it's cheaper that way.
 109         *
 110         * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
 111         * bitmaps handling below becomes unpleasant, to put it mildly...
 112         */
 113        if (unlikely(nr > sysctl_nr_open))
 114                nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
 115
 116        fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
 117        if (!fdt)
 118                goto out;
 119        fdt->max_fds = nr;
 120        data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
 121        if (!data)
 122                goto out_fdt;
 123        fdt->fd = data;
 124
 125        data = kvmalloc(max_t(size_t,
 126                                 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
 127                                 GFP_KERNEL_ACCOUNT);
 128        if (!data)
 129                goto out_arr;
 130        fdt->open_fds = data;
 131        data += nr / BITS_PER_BYTE;
 132        fdt->close_on_exec = data;
 133        data += nr / BITS_PER_BYTE;
 134        fdt->full_fds_bits = data;
 135
 136        return fdt;
 137
 138out_arr:
 139        kvfree(fdt->fd);
 140out_fdt:
 141        kfree(fdt);
 142out:
 143        return NULL;
 144}
 145
 146/*
 147 * Expand the file descriptor table.
 148 * This function will allocate a new fdtable and both fd array and fdset, of
 149 * the given size.
 150 * Return <0 error code on error; 1 on successful completion.
 151 * The files->file_lock should be held on entry, and will be held on exit.
 152 */
 153static int expand_fdtable(struct files_struct *files, unsigned int nr)
 154        __releases(files->file_lock)
 155        __acquires(files->file_lock)
 156{
 157        struct fdtable *new_fdt, *cur_fdt;
 158
 159        spin_unlock(&files->file_lock);
 160        new_fdt = alloc_fdtable(nr);
 161
 162        /* make sure all fd_install() have seen resize_in_progress
 163         * or have finished their rcu_read_lock_sched() section.
 164         */
 165        if (atomic_read(&files->count) > 1)
 166                synchronize_rcu();
 167
 168        spin_lock(&files->file_lock);
 169        if (!new_fdt)
 170                return -ENOMEM;
 171        /*
 172         * extremely unlikely race - sysctl_nr_open decreased between the check in
 173         * caller and alloc_fdtable().  Cheaper to catch it here...
 174         */
 175        if (unlikely(new_fdt->max_fds <= nr)) {
 176                __free_fdtable(new_fdt);
 177                return -EMFILE;
 178        }
 179        cur_fdt = files_fdtable(files);
 180        BUG_ON(nr < cur_fdt->max_fds);
 181        copy_fdtable(new_fdt, cur_fdt);
 182        rcu_assign_pointer(files->fdt, new_fdt);
 183        if (cur_fdt != &files->fdtab)
 184                call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
 185        /* coupled with smp_rmb() in fd_install() */
 186        smp_wmb();
 187        return 1;
 188}
 189
 190/*
 191 * Expand files.
 192 * This function will expand the file structures, if the requested size exceeds
 193 * the current capacity and there is room for expansion.
 194 * Return <0 error code on error; 0 when nothing done; 1 when files were
 195 * expanded and execution may have blocked.
 196 * The files->file_lock should be held on entry, and will be held on exit.
 197 */
 198static int expand_files(struct files_struct *files, unsigned int nr)
 199        __releases(files->file_lock)
 200        __acquires(files->file_lock)
 201{
 202        struct fdtable *fdt;
 203        int expanded = 0;
 204
 205repeat:
 206        fdt = files_fdtable(files);
 207
 208        /* Do we need to expand? */
 209        if (nr < fdt->max_fds)
 210                return expanded;
 211
 212        /* Can we expand? */
 213        if (nr >= sysctl_nr_open)
 214                return -EMFILE;
 215
 216        if (unlikely(files->resize_in_progress)) {
 217                spin_unlock(&files->file_lock);
 218                expanded = 1;
 219                wait_event(files->resize_wait, !files->resize_in_progress);
 220                spin_lock(&files->file_lock);
 221                goto repeat;
 222        }
 223
 224        /* All good, so we try */
 225        files->resize_in_progress = true;
 226        expanded = expand_fdtable(files, nr);
 227        files->resize_in_progress = false;
 228
 229        wake_up_all(&files->resize_wait);
 230        return expanded;
 231}
 232
 233static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
 234{
 235        __set_bit(fd, fdt->close_on_exec);
 236}
 237
 238static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
 239{
 240        if (test_bit(fd, fdt->close_on_exec))
 241                __clear_bit(fd, fdt->close_on_exec);
 242}
 243
 244static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
 245{
 246        __set_bit(fd, fdt->open_fds);
 247        fd /= BITS_PER_LONG;
 248        if (!~fdt->open_fds[fd])
 249                __set_bit(fd, fdt->full_fds_bits);
 250}
 251
 252static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
 253{
 254        __clear_bit(fd, fdt->open_fds);
 255        __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
 256}
 257
 258static unsigned int count_open_files(struct fdtable *fdt)
 259{
 260        unsigned int size = fdt->max_fds;
 261        unsigned int i;
 262
 263        /* Find the last open fd */
 264        for (i = size / BITS_PER_LONG; i > 0; ) {
 265                if (fdt->open_fds[--i])
 266                        break;
 267        }
 268        i = (i + 1) * BITS_PER_LONG;
 269        return i;
 270}
 271
 272static unsigned int sane_fdtable_size(struct fdtable *fdt, unsigned int max_fds)
 273{
 274        unsigned int count;
 275
 276        count = count_open_files(fdt);
 277        if (max_fds < NR_OPEN_DEFAULT)
 278                max_fds = NR_OPEN_DEFAULT;
 279        return min(count, max_fds);
 280}
 281
 282/*
 283 * Allocate a new files structure and copy contents from the
 284 * passed in files structure.
 285 * errorp will be valid only when the returned files_struct is NULL.
 286 */
 287struct files_struct *dup_fd(struct files_struct *oldf, unsigned int max_fds, int *errorp)
 288{
 289        struct files_struct *newf;
 290        struct file **old_fds, **new_fds;
 291        unsigned int open_files, i;
 292        struct fdtable *old_fdt, *new_fdt;
 293
 294        *errorp = -ENOMEM;
 295        newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
 296        if (!newf)
 297                goto out;
 298
 299        atomic_set(&newf->count, 1);
 300
 301        spin_lock_init(&newf->file_lock);
 302        newf->resize_in_progress = false;
 303        init_waitqueue_head(&newf->resize_wait);
 304        newf->next_fd = 0;
 305        new_fdt = &newf->fdtab;
 306        new_fdt->max_fds = NR_OPEN_DEFAULT;
 307        new_fdt->close_on_exec = newf->close_on_exec_init;
 308        new_fdt->open_fds = newf->open_fds_init;
 309        new_fdt->full_fds_bits = newf->full_fds_bits_init;
 310        new_fdt->fd = &newf->fd_array[0];
 311
 312        spin_lock(&oldf->file_lock);
 313        old_fdt = files_fdtable(oldf);
 314        open_files = sane_fdtable_size(old_fdt, max_fds);
 315
 316        /*
 317         * Check whether we need to allocate a larger fd array and fd set.
 318         */
 319        while (unlikely(open_files > new_fdt->max_fds)) {
 320                spin_unlock(&oldf->file_lock);
 321
 322                if (new_fdt != &newf->fdtab)
 323                        __free_fdtable(new_fdt);
 324
 325                new_fdt = alloc_fdtable(open_files - 1);
 326                if (!new_fdt) {
 327                        *errorp = -ENOMEM;
 328                        goto out_release;
 329                }
 330
 331                /* beyond sysctl_nr_open; nothing to do */
 332                if (unlikely(new_fdt->max_fds < open_files)) {
 333                        __free_fdtable(new_fdt);
 334                        *errorp = -EMFILE;
 335                        goto out_release;
 336                }
 337
 338                /*
 339                 * Reacquire the oldf lock and a pointer to its fd table
 340                 * who knows it may have a new bigger fd table. We need
 341                 * the latest pointer.
 342                 */
 343                spin_lock(&oldf->file_lock);
 344                old_fdt = files_fdtable(oldf);
 345                open_files = sane_fdtable_size(old_fdt, max_fds);
 346        }
 347
 348        copy_fd_bitmaps(new_fdt, old_fdt, open_files);
 349
 350        old_fds = old_fdt->fd;
 351        new_fds = new_fdt->fd;
 352
 353        for (i = open_files; i != 0; i--) {
 354                struct file *f = *old_fds++;
 355                if (f) {
 356                        get_file(f);
 357                } else {
 358                        /*
 359                         * The fd may be claimed in the fd bitmap but not yet
 360                         * instantiated in the files array if a sibling thread
 361                         * is partway through open().  So make sure that this
 362                         * fd is available to the new process.
 363                         */
 364                        __clear_open_fd(open_files - i, new_fdt);
 365                }
 366                rcu_assign_pointer(*new_fds++, f);
 367        }
 368        spin_unlock(&oldf->file_lock);
 369
 370        /* clear the remainder */
 371        memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
 372
 373        rcu_assign_pointer(newf->fdt, new_fdt);
 374
 375        return newf;
 376
 377out_release:
 378        kmem_cache_free(files_cachep, newf);
 379out:
 380        return NULL;
 381}
 382
 383static struct fdtable *close_files(struct files_struct * files)
 384{
 385        /*
 386         * It is safe to dereference the fd table without RCU or
 387         * ->file_lock because this is the last reference to the
 388         * files structure.
 389         */
 390        struct fdtable *fdt = rcu_dereference_raw(files->fdt);
 391        unsigned int i, j = 0;
 392
 393        for (;;) {
 394                unsigned long set;
 395                i = j * BITS_PER_LONG;
 396                if (i >= fdt->max_fds)
 397                        break;
 398                set = fdt->open_fds[j++];
 399                while (set) {
 400                        if (set & 1) {
 401                                struct file * file = xchg(&fdt->fd[i], NULL);
 402                                if (file) {
 403                                        filp_close(file, files);
 404                                        cond_resched();
 405                                }
 406                        }
 407                        i++;
 408                        set >>= 1;
 409                }
 410        }
 411
 412        return fdt;
 413}
 414
 415void put_files_struct(struct files_struct *files)
 416{
 417        if (atomic_dec_and_test(&files->count)) {
 418                struct fdtable *fdt = close_files(files);
 419
 420                /* free the arrays if they are not embedded */
 421                if (fdt != &files->fdtab)
 422                        __free_fdtable(fdt);
 423                kmem_cache_free(files_cachep, files);
 424        }
 425}
 426
 427void exit_files(struct task_struct *tsk)
 428{
 429        struct files_struct * files = tsk->files;
 430
 431        if (files) {
 432                task_lock(tsk);
 433                tsk->files = NULL;
 434                task_unlock(tsk);
 435                put_files_struct(files);
 436        }
 437}
 438
 439struct files_struct init_files = {
 440        .count          = ATOMIC_INIT(1),
 441        .fdt            = &init_files.fdtab,
 442        .fdtab          = {
 443                .max_fds        = NR_OPEN_DEFAULT,
 444                .fd             = &init_files.fd_array[0],
 445                .close_on_exec  = init_files.close_on_exec_init,
 446                .open_fds       = init_files.open_fds_init,
 447                .full_fds_bits  = init_files.full_fds_bits_init,
 448        },
 449        .file_lock      = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
 450        .resize_wait    = __WAIT_QUEUE_HEAD_INITIALIZER(init_files.resize_wait),
 451};
 452
 453static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
 454{
 455        unsigned int maxfd = fdt->max_fds;
 456        unsigned int maxbit = maxfd / BITS_PER_LONG;
 457        unsigned int bitbit = start / BITS_PER_LONG;
 458
 459        bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
 460        if (bitbit > maxfd)
 461                return maxfd;
 462        if (bitbit > start)
 463                start = bitbit;
 464        return find_next_zero_bit(fdt->open_fds, maxfd, start);
 465}
 466
 467/*
 468 * allocate a file descriptor, mark it busy.
 469 */
 470static int alloc_fd(unsigned start, unsigned end, unsigned flags)
 471{
 472        struct files_struct *files = current->files;
 473        unsigned int fd;
 474        int error;
 475        struct fdtable *fdt;
 476
 477        spin_lock(&files->file_lock);
 478repeat:
 479        fdt = files_fdtable(files);
 480        fd = start;
 481        if (fd < files->next_fd)
 482                fd = files->next_fd;
 483
 484        if (fd < fdt->max_fds)
 485                fd = find_next_fd(fdt, fd);
 486
 487        /*
 488         * N.B. For clone tasks sharing a files structure, this test
 489         * will limit the total number of files that can be opened.
 490         */
 491        error = -EMFILE;
 492        if (fd >= end)
 493                goto out;
 494
 495        error = expand_files(files, fd);
 496        if (error < 0)
 497                goto out;
 498
 499        /*
 500         * If we needed to expand the fs array we
 501         * might have blocked - try again.
 502         */
 503        if (error)
 504                goto repeat;
 505
 506        if (start <= files->next_fd)
 507                files->next_fd = fd + 1;
 508
 509        __set_open_fd(fd, fdt);
 510        if (flags & O_CLOEXEC)
 511                __set_close_on_exec(fd, fdt);
 512        else
 513                __clear_close_on_exec(fd, fdt);
 514        error = fd;
 515#if 1
 516        /* Sanity check */
 517        if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
 518                printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
 519                rcu_assign_pointer(fdt->fd[fd], NULL);
 520        }
 521#endif
 522
 523out:
 524        spin_unlock(&files->file_lock);
 525        return error;
 526}
 527
 528int __get_unused_fd_flags(unsigned flags, unsigned long nofile)
 529{
 530        return alloc_fd(0, nofile, flags);
 531}
 532
 533int get_unused_fd_flags(unsigned flags)
 534{
 535        return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE));
 536}
 537EXPORT_SYMBOL(get_unused_fd_flags);
 538
 539static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 540{
 541        struct fdtable *fdt = files_fdtable(files);
 542        __clear_open_fd(fd, fdt);
 543        if (fd < files->next_fd)
 544                files->next_fd = fd;
 545}
 546
 547void put_unused_fd(unsigned int fd)
 548{
 549        struct files_struct *files = current->files;
 550        spin_lock(&files->file_lock);
 551        __put_unused_fd(files, fd);
 552        spin_unlock(&files->file_lock);
 553}
 554
 555EXPORT_SYMBOL(put_unused_fd);
 556
 557/*
 558 * Install a file pointer in the fd array.
 559 *
 560 * The VFS is full of places where we drop the files lock between
 561 * setting the open_fds bitmap and installing the file in the file
 562 * array.  At any such point, we are vulnerable to a dup2() race
 563 * installing a file in the array before us.  We need to detect this and
 564 * fput() the struct file we are about to overwrite in this case.
 565 *
 566 * It should never happen - if we allow dup2() do it, _really_ bad things
 567 * will follow.
 568 *
 569 * This consumes the "file" refcount, so callers should treat it
 570 * as if they had called fput(file).
 571 */
 572
 573void fd_install(unsigned int fd, struct file *file)
 574{
 575        struct files_struct *files = current->files;
 576        struct fdtable *fdt;
 577
 578        rcu_read_lock_sched();
 579
 580        if (unlikely(files->resize_in_progress)) {
 581                rcu_read_unlock_sched();
 582                spin_lock(&files->file_lock);
 583                fdt = files_fdtable(files);
 584                BUG_ON(fdt->fd[fd] != NULL);
 585                rcu_assign_pointer(fdt->fd[fd], file);
 586                spin_unlock(&files->file_lock);
 587                return;
 588        }
 589        /* coupled with smp_wmb() in expand_fdtable() */
 590        smp_rmb();
 591        fdt = rcu_dereference_sched(files->fdt);
 592        BUG_ON(fdt->fd[fd] != NULL);
 593        rcu_assign_pointer(fdt->fd[fd], file);
 594        rcu_read_unlock_sched();
 595}
 596
 597EXPORT_SYMBOL(fd_install);
 598
 599/**
 600 * pick_file - return file associatd with fd
 601 * @files: file struct to retrieve file from
 602 * @fd: file descriptor to retrieve file for
 603 *
 604 * If this functions returns an EINVAL error pointer the fd was beyond the
 605 * current maximum number of file descriptors for that fdtable.
 606 *
 607 * Returns: The file associated with @fd, on error returns an error pointer.
 608 */
 609static struct file *pick_file(struct files_struct *files, unsigned fd)
 610{
 611        struct file *file;
 612        struct fdtable *fdt;
 613
 614        spin_lock(&files->file_lock);
 615        fdt = files_fdtable(files);
 616        if (fd >= fdt->max_fds) {
 617                file = ERR_PTR(-EINVAL);
 618                goto out_unlock;
 619        }
 620        file = fdt->fd[fd];
 621        if (!file) {
 622                file = ERR_PTR(-EBADF);
 623                goto out_unlock;
 624        }
 625        rcu_assign_pointer(fdt->fd[fd], NULL);
 626        __put_unused_fd(files, fd);
 627
 628out_unlock:
 629        spin_unlock(&files->file_lock);
 630        return file;
 631}
 632
 633int close_fd(unsigned fd)
 634{
 635        struct files_struct *files = current->files;
 636        struct file *file;
 637
 638        file = pick_file(files, fd);
 639        if (IS_ERR(file))
 640                return -EBADF;
 641
 642        return filp_close(file, files);
 643}
 644EXPORT_SYMBOL(close_fd); /* for ksys_close() */
 645
 646/**
 647 * last_fd - return last valid index into fd table
 648 * @cur_fds: files struct
 649 *
 650 * Context: Either rcu read lock or files_lock must be held.
 651 *
 652 * Returns: Last valid index into fdtable.
 653 */
 654static inline unsigned last_fd(struct fdtable *fdt)
 655{
 656        return fdt->max_fds - 1;
 657}
 658
 659static inline void __range_cloexec(struct files_struct *cur_fds,
 660                                   unsigned int fd, unsigned int max_fd)
 661{
 662        struct fdtable *fdt;
 663
 664        /* make sure we're using the correct maximum value */
 665        spin_lock(&cur_fds->file_lock);
 666        fdt = files_fdtable(cur_fds);
 667        max_fd = min(last_fd(fdt), max_fd);
 668        if (fd <= max_fd)
 669                bitmap_set(fdt->close_on_exec, fd, max_fd - fd + 1);
 670        spin_unlock(&cur_fds->file_lock);
 671}
 672
 673static inline void __range_close(struct files_struct *cur_fds, unsigned int fd,
 674                                 unsigned int max_fd)
 675{
 676        while (fd <= max_fd) {
 677                struct file *file;
 678
 679                file = pick_file(cur_fds, fd++);
 680                if (!IS_ERR(file)) {
 681                        /* found a valid file to close */
 682                        filp_close(file, cur_fds);
 683                        cond_resched();
 684                        continue;
 685                }
 686
 687                /* beyond the last fd in that table */
 688                if (PTR_ERR(file) == -EINVAL)
 689                        return;
 690        }
 691}
 692
 693/**
 694 * __close_range() - Close all file descriptors in a given range.
 695 *
 696 * @fd:     starting file descriptor to close
 697 * @max_fd: last file descriptor to close
 698 *
 699 * This closes a range of file descriptors. All file descriptors
 700 * from @fd up to and including @max_fd are closed.
 701 */
 702int __close_range(unsigned fd, unsigned max_fd, unsigned int flags)
 703{
 704        struct task_struct *me = current;
 705        struct files_struct *cur_fds = me->files, *fds = NULL;
 706
 707        if (flags & ~(CLOSE_RANGE_UNSHARE | CLOSE_RANGE_CLOEXEC))
 708                return -EINVAL;
 709
 710        if (fd > max_fd)
 711                return -EINVAL;
 712
 713        if (flags & CLOSE_RANGE_UNSHARE) {
 714                int ret;
 715                unsigned int max_unshare_fds = NR_OPEN_MAX;
 716
 717                /*
 718                 * If the caller requested all fds to be made cloexec we always
 719                 * copy all of the file descriptors since they still want to
 720                 * use them.
 721                 */
 722                if (!(flags & CLOSE_RANGE_CLOEXEC)) {
 723                        /*
 724                         * If the requested range is greater than the current
 725                         * maximum, we're closing everything so only copy all
 726                         * file descriptors beneath the lowest file descriptor.
 727                         */
 728                        rcu_read_lock();
 729                        if (max_fd >= last_fd(files_fdtable(cur_fds)))
 730                                max_unshare_fds = fd;
 731                        rcu_read_unlock();
 732                }
 733
 734                ret = unshare_fd(CLONE_FILES, max_unshare_fds, &fds);
 735                if (ret)
 736                        return ret;
 737
 738                /*
 739                 * We used to share our file descriptor table, and have now
 740                 * created a private one, make sure we're using it below.
 741                 */
 742                if (fds)
 743                        swap(cur_fds, fds);
 744        }
 745
 746        if (flags & CLOSE_RANGE_CLOEXEC)
 747                __range_cloexec(cur_fds, fd, max_fd);
 748        else
 749                __range_close(cur_fds, fd, max_fd);
 750
 751        if (fds) {
 752                /*
 753                 * We're done closing the files we were supposed to. Time to install
 754                 * the new file descriptor table and drop the old one.
 755                 */
 756                task_lock(me);
 757                me->files = cur_fds;
 758                task_unlock(me);
 759                put_files_struct(fds);
 760        }
 761
 762        return 0;
 763}
 764
 765/*
 766 * See close_fd_get_file() below, this variant assumes current->files->file_lock
 767 * is held.
 768 */
 769int __close_fd_get_file(unsigned int fd, struct file **res)
 770{
 771        struct files_struct *files = current->files;
 772        struct file *file;
 773        struct fdtable *fdt;
 774
 775        fdt = files_fdtable(files);
 776        if (fd >= fdt->max_fds)
 777                goto out_err;
 778        file = fdt->fd[fd];
 779        if (!file)
 780                goto out_err;
 781        rcu_assign_pointer(fdt->fd[fd], NULL);
 782        __put_unused_fd(files, fd);
 783        get_file(file);
 784        *res = file;
 785        return 0;
 786out_err:
 787        *res = NULL;
 788        return -ENOENT;
 789}
 790
 791/*
 792 * variant of close_fd that gets a ref on the file for later fput.
 793 * The caller must ensure that filp_close() called on the file, and then
 794 * an fput().
 795 */
 796int close_fd_get_file(unsigned int fd, struct file **res)
 797{
 798        struct files_struct *files = current->files;
 799        int ret;
 800
 801        spin_lock(&files->file_lock);
 802        ret = __close_fd_get_file(fd, res);
 803        spin_unlock(&files->file_lock);
 804
 805        return ret;
 806}
 807
 808void do_close_on_exec(struct files_struct *files)
 809{
 810        unsigned i;
 811        struct fdtable *fdt;
 812
 813        /* exec unshares first */
 814        spin_lock(&files->file_lock);
 815        for (i = 0; ; i++) {
 816                unsigned long set;
 817                unsigned fd = i * BITS_PER_LONG;
 818                fdt = files_fdtable(files);
 819                if (fd >= fdt->max_fds)
 820                        break;
 821                set = fdt->close_on_exec[i];
 822                if (!set)
 823                        continue;
 824                fdt->close_on_exec[i] = 0;
 825                for ( ; set ; fd++, set >>= 1) {
 826                        struct file *file;
 827                        if (!(set & 1))
 828                                continue;
 829                        file = fdt->fd[fd];
 830                        if (!file)
 831                                continue;
 832                        rcu_assign_pointer(fdt->fd[fd], NULL);
 833                        __put_unused_fd(files, fd);
 834                        spin_unlock(&files->file_lock);
 835                        filp_close(file, files);
 836                        cond_resched();
 837                        spin_lock(&files->file_lock);
 838                }
 839
 840        }
 841        spin_unlock(&files->file_lock);
 842}
 843
 844static inline struct file *__fget_files_rcu(struct files_struct *files,
 845        unsigned int fd, fmode_t mask, unsigned int refs)
 846{
 847        for (;;) {
 848                struct file *file;
 849                struct fdtable *fdt = rcu_dereference_raw(files->fdt);
 850                struct file __rcu **fdentry;
 851
 852                if (unlikely(fd >= fdt->max_fds))
 853                        return NULL;
 854
 855                fdentry = fdt->fd + array_index_nospec(fd, fdt->max_fds);
 856                file = rcu_dereference_raw(*fdentry);
 857                if (unlikely(!file))
 858                        return NULL;
 859
 860                if (unlikely(file->f_mode & mask))
 861                        return NULL;
 862
 863                /*
 864                 * Ok, we have a file pointer. However, because we do
 865                 * this all locklessly under RCU, we may be racing with
 866                 * that file being closed.
 867                 *
 868                 * Such a race can take two forms:
 869                 *
 870                 *  (a) the file ref already went down to zero,
 871                 *      and get_file_rcu_many() fails. Just try
 872                 *      again:
 873                 */
 874                if (unlikely(!get_file_rcu_many(file, refs)))
 875                        continue;
 876
 877                /*
 878                 *  (b) the file table entry has changed under us.
 879                 *       Note that we don't need to re-check the 'fdt->fd'
 880                 *       pointer having changed, because it always goes
 881                 *       hand-in-hand with 'fdt'.
 882                 *
 883                 * If so, we need to put our refs and try again.
 884                 */
 885                if (unlikely(rcu_dereference_raw(files->fdt) != fdt) ||
 886                    unlikely(rcu_dereference_raw(*fdentry) != file)) {
 887                        fput_many(file, refs);
 888                        continue;
 889                }
 890
 891                /*
 892                 * Ok, we have a ref to the file, and checked that it
 893                 * still exists.
 894                 */
 895                return file;
 896        }
 897}
 898
 899static struct file *__fget_files(struct files_struct *files, unsigned int fd,
 900                                 fmode_t mask, unsigned int refs)
 901{
 902        struct file *file;
 903
 904        rcu_read_lock();
 905        file = __fget_files_rcu(files, fd, mask, refs);
 906        rcu_read_unlock();
 907
 908        return file;
 909}
 910
 911static inline struct file *__fget(unsigned int fd, fmode_t mask,
 912                                  unsigned int refs)
 913{
 914        return __fget_files(current->files, fd, mask, refs);
 915}
 916
 917struct file *fget_many(unsigned int fd, unsigned int refs)
 918{
 919        return __fget(fd, FMODE_PATH, refs);
 920}
 921
 922struct file *fget(unsigned int fd)
 923{
 924        return __fget(fd, FMODE_PATH, 1);
 925}
 926EXPORT_SYMBOL(fget);
 927
 928struct file *fget_raw(unsigned int fd)
 929{
 930        return __fget(fd, 0, 1);
 931}
 932EXPORT_SYMBOL(fget_raw);
 933
 934struct file *fget_task(struct task_struct *task, unsigned int fd)
 935{
 936        struct file *file = NULL;
 937
 938        task_lock(task);
 939        if (task->files)
 940                file = __fget_files(task->files, fd, 0, 1);
 941        task_unlock(task);
 942
 943        return file;
 944}
 945
 946struct file *task_lookup_fd_rcu(struct task_struct *task, unsigned int fd)
 947{
 948        /* Must be called with rcu_read_lock held */
 949        struct files_struct *files;
 950        struct file *file = NULL;
 951
 952        task_lock(task);
 953        files = task->files;
 954        if (files)
 955                file = files_lookup_fd_rcu(files, fd);
 956        task_unlock(task);
 957
 958        return file;
 959}
 960
 961struct file *task_lookup_next_fd_rcu(struct task_struct *task, unsigned int *ret_fd)
 962{
 963        /* Must be called with rcu_read_lock held */
 964        struct files_struct *files;
 965        unsigned int fd = *ret_fd;
 966        struct file *file = NULL;
 967
 968        task_lock(task);
 969        files = task->files;
 970        if (files) {
 971                for (; fd < files_fdtable(files)->max_fds; fd++) {
 972                        file = files_lookup_fd_rcu(files, fd);
 973                        if (file)
 974                                break;
 975                }
 976        }
 977        task_unlock(task);
 978        *ret_fd = fd;
 979        return file;
 980}
 981
 982/*
 983 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
 984 *
 985 * You can use this instead of fget if you satisfy all of the following
 986 * conditions:
 987 * 1) You must call fput_light before exiting the syscall and returning control
 988 *    to userspace (i.e. you cannot remember the returned struct file * after
 989 *    returning to userspace).
 990 * 2) You must not call filp_close on the returned struct file * in between
 991 *    calls to fget_light and fput_light.
 992 * 3) You must not clone the current task in between the calls to fget_light
 993 *    and fput_light.
 994 *
 995 * The fput_needed flag returned by fget_light should be passed to the
 996 * corresponding fput_light.
 997 */
 998static unsigned long __fget_light(unsigned int fd, fmode_t mask)
 999{
1000        struct files_struct *files = current->files;
1001        struct file *file;
1002
1003        if (atomic_read(&files->count) == 1) {
1004                file = files_lookup_fd_raw(files, fd);
1005                if (!file || unlikely(file->f_mode & mask))
1006                        return 0;
1007                return (unsigned long)file;
1008        } else {
1009                file = __fget(fd, mask, 1);
1010                if (!file)
1011                        return 0;
1012                return FDPUT_FPUT | (unsigned long)file;
1013        }
1014}
1015unsigned long __fdget(unsigned int fd)
1016{
1017        return __fget_light(fd, FMODE_PATH);
1018}
1019EXPORT_SYMBOL(__fdget);
1020
1021unsigned long __fdget_raw(unsigned int fd)
1022{
1023        return __fget_light(fd, 0);
1024}
1025
1026unsigned long __fdget_pos(unsigned int fd)
1027{
1028        unsigned long v = __fdget(fd);
1029        struct file *file = (struct file *)(v & ~3);
1030
1031        if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
1032                if (file_count(file) > 1) {
1033                        v |= FDPUT_POS_UNLOCK;
1034                        mutex_lock(&file->f_pos_lock);
1035                }
1036        }
1037        return v;
1038}
1039
1040void __f_unlock_pos(struct file *f)
1041{
1042        mutex_unlock(&f->f_pos_lock);
1043}
1044
1045/*
1046 * We only lock f_pos if we have threads or if the file might be
1047 * shared with another process. In both cases we'll have an elevated
1048 * file count (done either by fdget() or by fork()).
1049 */
1050
1051void set_close_on_exec(unsigned int fd, int flag)
1052{
1053        struct files_struct *files = current->files;
1054        struct fdtable *fdt;
1055        spin_lock(&files->file_lock);
1056        fdt = files_fdtable(files);
1057        if (flag)
1058                __set_close_on_exec(fd, fdt);
1059        else
1060                __clear_close_on_exec(fd, fdt);
1061        spin_unlock(&files->file_lock);
1062}
1063
1064bool get_close_on_exec(unsigned int fd)
1065{
1066        struct files_struct *files = current->files;
1067        struct fdtable *fdt;
1068        bool res;
1069        rcu_read_lock();
1070        fdt = files_fdtable(files);
1071        res = close_on_exec(fd, fdt);
1072        rcu_read_unlock();
1073        return res;
1074}
1075
1076static int do_dup2(struct files_struct *files,
1077        struct file *file, unsigned fd, unsigned flags)
1078__releases(&files->file_lock)
1079{
1080        struct file *tofree;
1081        struct fdtable *fdt;
1082
1083        /*
1084         * We need to detect attempts to do dup2() over allocated but still
1085         * not finished descriptor.  NB: OpenBSD avoids that at the price of
1086         * extra work in their equivalent of fget() - they insert struct
1087         * file immediately after grabbing descriptor, mark it larval if
1088         * more work (e.g. actual opening) is needed and make sure that
1089         * fget() treats larval files as absent.  Potentially interesting,
1090         * but while extra work in fget() is trivial, locking implications
1091         * and amount of surgery on open()-related paths in VFS are not.
1092         * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
1093         * deadlocks in rather amusing ways, AFAICS.  All of that is out of
1094         * scope of POSIX or SUS, since neither considers shared descriptor
1095         * tables and this condition does not arise without those.
1096         */
1097        fdt = files_fdtable(files);
1098        tofree = fdt->fd[fd];
1099        if (!tofree && fd_is_open(fd, fdt))
1100                goto Ebusy;
1101        get_file(file);
1102        rcu_assign_pointer(fdt->fd[fd], file);
1103        __set_open_fd(fd, fdt);
1104        if (flags & O_CLOEXEC)
1105                __set_close_on_exec(fd, fdt);
1106        else
1107                __clear_close_on_exec(fd, fdt);
1108        spin_unlock(&files->file_lock);
1109
1110        if (tofree)
1111                filp_close(tofree, files);
1112
1113        return fd;
1114
1115Ebusy:
1116        spin_unlock(&files->file_lock);
1117        return -EBUSY;
1118}
1119
1120int replace_fd(unsigned fd, struct file *file, unsigned flags)
1121{
1122        int err;
1123        struct files_struct *files = current->files;
1124
1125        if (!file)
1126                return close_fd(fd);
1127
1128        if (fd >= rlimit(RLIMIT_NOFILE))
1129                return -EBADF;
1130
1131        spin_lock(&files->file_lock);
1132        err = expand_files(files, fd);
1133        if (unlikely(err < 0))
1134                goto out_unlock;
1135        return do_dup2(files, file, fd, flags);
1136
1137out_unlock:
1138        spin_unlock(&files->file_lock);
1139        return err;
1140}
1141
1142/**
1143 * __receive_fd() - Install received file into file descriptor table
1144 * @file: struct file that was received from another process
1145 * @ufd: __user pointer to write new fd number to
1146 * @o_flags: the O_* flags to apply to the new fd entry
1147 *
1148 * Installs a received file into the file descriptor table, with appropriate
1149 * checks and count updates. Optionally writes the fd number to userspace, if
1150 * @ufd is non-NULL.
1151 *
1152 * This helper handles its own reference counting of the incoming
1153 * struct file.
1154 *
1155 * Returns newly install fd or -ve on error.
1156 */
1157int __receive_fd(struct file *file, int __user *ufd, unsigned int o_flags)
1158{
1159        int new_fd;
1160        int error;
1161
1162        error = security_file_receive(file);
1163        if (error)
1164                return error;
1165
1166        new_fd = get_unused_fd_flags(o_flags);
1167        if (new_fd < 0)
1168                return new_fd;
1169
1170        if (ufd) {
1171                error = put_user(new_fd, ufd);
1172                if (error) {
1173                        put_unused_fd(new_fd);
1174                        return error;
1175                }
1176        }
1177
1178        fd_install(new_fd, get_file(file));
1179        __receive_sock(file);
1180        return new_fd;
1181}
1182
1183int receive_fd_replace(int new_fd, struct file *file, unsigned int o_flags)
1184{
1185        int error;
1186
1187        error = security_file_receive(file);
1188        if (error)
1189                return error;
1190        error = replace_fd(new_fd, file, o_flags);
1191        if (error)
1192                return error;
1193        __receive_sock(file);
1194        return new_fd;
1195}
1196
1197int receive_fd(struct file *file, unsigned int o_flags)
1198{
1199        return __receive_fd(file, NULL, o_flags);
1200}
1201EXPORT_SYMBOL_GPL(receive_fd);
1202
1203static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1204{
1205        int err = -EBADF;
1206        struct file *file;
1207        struct files_struct *files = current->files;
1208
1209        if ((flags & ~O_CLOEXEC) != 0)
1210                return -EINVAL;
1211
1212        if (unlikely(oldfd == newfd))
1213                return -EINVAL;
1214
1215        if (newfd >= rlimit(RLIMIT_NOFILE))
1216                return -EBADF;
1217
1218        spin_lock(&files->file_lock);
1219        err = expand_files(files, newfd);
1220        file = files_lookup_fd_locked(files, oldfd);
1221        if (unlikely(!file))
1222                goto Ebadf;
1223        if (unlikely(err < 0)) {
1224                if (err == -EMFILE)
1225                        goto Ebadf;
1226                goto out_unlock;
1227        }
1228        return do_dup2(files, file, newfd, flags);
1229
1230Ebadf:
1231        err = -EBADF;
1232out_unlock:
1233        spin_unlock(&files->file_lock);
1234        return err;
1235}
1236
1237SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1238{
1239        return ksys_dup3(oldfd, newfd, flags);
1240}
1241
1242SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1243{
1244        if (unlikely(newfd == oldfd)) { /* corner case */
1245                struct files_struct *files = current->files;
1246                int retval = oldfd;
1247
1248                rcu_read_lock();
1249                if (!files_lookup_fd_rcu(files, oldfd))
1250                        retval = -EBADF;
1251                rcu_read_unlock();
1252                return retval;
1253        }
1254        return ksys_dup3(oldfd, newfd, 0);
1255}
1256
1257SYSCALL_DEFINE1(dup, unsigned int, fildes)
1258{
1259        int ret = -EBADF;
1260        struct file *file = fget_raw(fildes);
1261
1262        if (file) {
1263                ret = get_unused_fd_flags(0);
1264                if (ret >= 0)
1265                        fd_install(ret, file);
1266                else
1267                        fput(file);
1268        }
1269        return ret;
1270}
1271
1272int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1273{
1274        unsigned long nofile = rlimit(RLIMIT_NOFILE);
1275        int err;
1276        if (from >= nofile)
1277                return -EINVAL;
1278        err = alloc_fd(from, nofile, flags);
1279        if (err >= 0) {
1280                get_file(file);
1281                fd_install(err, file);
1282        }
1283        return err;
1284}
1285
1286int iterate_fd(struct files_struct *files, unsigned n,
1287                int (*f)(const void *, struct file *, unsigned),
1288                const void *p)
1289{
1290        struct fdtable *fdt;
1291        int res = 0;
1292        if (!files)
1293                return 0;
1294        spin_lock(&files->file_lock);
1295        for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1296                struct file *file;
1297                file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1298                if (!file)
1299                        continue;
1300                res = f(p, file, n);
1301                if (res)
1302                        break;
1303        }
1304        spin_unlock(&files->file_lock);
1305        return res;
1306}
1307EXPORT_SYMBOL(iterate_fd);
1308