1 /* 2 * linux/fs/file.c 3 * 4 * Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes 5 * 6 * Manage the dynamic fd arrays in the process files_struct. 7 */ 8 9 #include <linux/syscalls.h> 10 #include <linux/export.h> 11 #include <linux/fs.h> 12 #include <linux/mm.h> 13 #include <linux/mmzone.h> 14 #include <linux/time.h> 15 #include <linux/sched.h> 16 #include <linux/slab.h> 17 #include <linux/vmalloc.h> 18 #include <linux/file.h> 19 #include <linux/fdtable.h> 20 #include <linux/bitops.h> 21 #include <linux/interrupt.h> 22 #include <linux/spinlock.h> 23 #include <linux/rcupdate.h> 24 #include <linux/workqueue.h> 25 26 struct fdtable_defer { 27 spinlock_t lock; 28 struct work_struct wq; 29 struct fdtable *next; 30 }; 31 32 int sysctl_nr_open __read_mostly = 1024*1024; 33 int sysctl_nr_open_min = BITS_PER_LONG; 34 int sysctl_nr_open_max = 1024 * 1024; /* raised later */ 35 36 /* 37 * We use this list to defer free fdtables that have vmalloced 38 * sets/arrays. By keeping a per-cpu list, we avoid having to embed 39 * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in 40 * this per-task structure. 41 */ 42 static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list); 43 44 static void *alloc_fdmem(size_t size) 45 { 46 /* 47 * Very large allocations can stress page reclaim, so fall back to 48 * vmalloc() if the allocation size will be considered "large" by the VM. 49 */ 50 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { 51 void *data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN); 52 if (data != NULL) 53 return data; 54 } 55 return vmalloc(size); 56 } 57 58 static void free_fdmem(void *ptr) 59 { 60 is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr); 61 } 62 63 static void __free_fdtable(struct fdtable *fdt) 64 { 65 free_fdmem(fdt->fd); 66 free_fdmem(fdt->open_fds); 67 kfree(fdt); 68 } 69 70 static void free_fdtable_work(struct work_struct *work) 71 { 72 struct fdtable_defer *f = 73 container_of(work, struct fdtable_defer, wq); 74 struct fdtable *fdt; 75 76 spin_lock_bh(&f->lock); 77 fdt = f->next; 78 f->next = NULL; 79 spin_unlock_bh(&f->lock); 80 while(fdt) { 81 struct fdtable *next = fdt->next; 82 83 __free_fdtable(fdt); 84 fdt = next; 85 } 86 } 87 88 static void free_fdtable_rcu(struct rcu_head *rcu) 89 { 90 struct fdtable *fdt = container_of(rcu, struct fdtable, rcu); 91 struct fdtable_defer *fddef; 92 93 BUG_ON(!fdt); 94 BUG_ON(fdt->max_fds <= NR_OPEN_DEFAULT); 95 96 if (!is_vmalloc_addr(fdt->fd) && !is_vmalloc_addr(fdt->open_fds)) { 97 kfree(fdt->fd); 98 kfree(fdt->open_fds); 99 kfree(fdt); 100 } else { 101 fddef = &get_cpu_var(fdtable_defer_list); 102 spin_lock(&fddef->lock); 103 fdt->next = fddef->next; 104 fddef->next = fdt; 105 /* vmallocs are handled from the workqueue context */ 106 schedule_work(&fddef->wq); 107 spin_unlock(&fddef->lock); 108 put_cpu_var(fdtable_defer_list); 109 } 110 } 111 112 /* 113 * Expand the fdset in the files_struct. Called with the files spinlock 114 * held for write. 115 */ 116 static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt) 117 { 118 unsigned int cpy, set; 119 120 BUG_ON(nfdt->max_fds < ofdt->max_fds); 121 122 cpy = ofdt->max_fds * sizeof(struct file *); 123 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *); 124 memcpy(nfdt->fd, ofdt->fd, cpy); 125 memset((char *)(nfdt->fd) + cpy, 0, set); 126 127 cpy = ofdt->max_fds / BITS_PER_BYTE; 128 set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE; 129 memcpy(nfdt->open_fds, ofdt->open_fds, cpy); 130 memset((char *)(nfdt->open_fds) + cpy, 0, set); 131 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy); 132 memset((char *)(nfdt->close_on_exec) + cpy, 0, set); 133 } 134 135 static struct fdtable * alloc_fdtable(unsigned int nr) 136 { 137 struct fdtable *fdt; 138 void *data; 139 140 /* 141 * Figure out how many fds we actually want to support in this fdtable. 142 * Allocation steps are keyed to the size of the fdarray, since it 143 * grows far faster than any of the other dynamic data. We try to fit 144 * the fdarray into comfortable page-tuned chunks: starting at 1024B 145 * and growing in powers of two from there on. 146 */ 147 nr /= (1024 / sizeof(struct file *)); 148 nr = roundup_pow_of_two(nr + 1); 149 nr *= (1024 / sizeof(struct file *)); 150 /* 151 * Note that this can drive nr *below* what we had passed if sysctl_nr_open 152 * had been set lower between the check in expand_files() and here. Deal 153 * with that in caller, it's cheaper that way. 154 * 155 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise 156 * bitmaps handling below becomes unpleasant, to put it mildly... 157 */ 158 if (unlikely(nr > sysctl_nr_open)) 159 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1; 160 161 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL); 162 if (!fdt) 163 goto out; 164 fdt->max_fds = nr; 165 data = alloc_fdmem(nr * sizeof(struct file *)); 166 if (!data) 167 goto out_fdt; 168 fdt->fd = data; 169 170 data = alloc_fdmem(max_t(size_t, 171 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES)); 172 if (!data) 173 goto out_arr; 174 fdt->open_fds = data; 175 data += nr / BITS_PER_BYTE; 176 fdt->close_on_exec = data; 177 fdt->next = NULL; 178 179 return fdt; 180 181 out_arr: 182 free_fdmem(fdt->fd); 183 out_fdt: 184 kfree(fdt); 185 out: 186 return NULL; 187 } 188 189 /* 190 * Expand the file descriptor table. 191 * This function will allocate a new fdtable and both fd array and fdset, of 192 * the given size. 193 * Return <0 error code on error; 1 on successful completion. 194 * The files->file_lock should be held on entry, and will be held on exit. 195 */ 196 static int expand_fdtable(struct files_struct *files, int nr) 197 __releases(files->file_lock) 198 __acquires(files->file_lock) 199 { 200 struct fdtable *new_fdt, *cur_fdt; 201 202 spin_unlock(&files->file_lock); 203 new_fdt = alloc_fdtable(nr); 204 spin_lock(&files->file_lock); 205 if (!new_fdt) 206 return -ENOMEM; 207 /* 208 * extremely unlikely race - sysctl_nr_open decreased between the check in 209 * caller and alloc_fdtable(). Cheaper to catch it here... 210 */ 211 if (unlikely(new_fdt->max_fds <= nr)) { 212 __free_fdtable(new_fdt); 213 return -EMFILE; 214 } 215 /* 216 * Check again since another task may have expanded the fd table while 217 * we dropped the lock 218 */ 219 cur_fdt = files_fdtable(files); 220 if (nr >= cur_fdt->max_fds) { 221 /* Continue as planned */ 222 copy_fdtable(new_fdt, cur_fdt); 223 rcu_assign_pointer(files->fdt, new_fdt); 224 if (cur_fdt->max_fds > NR_OPEN_DEFAULT) 225 call_rcu(&cur_fdt->rcu, free_fdtable_rcu); 226 } else { 227 /* Somebody else expanded, so undo our attempt */ 228 __free_fdtable(new_fdt); 229 } 230 return 1; 231 } 232 233 /* 234 * Expand files. 235 * This function will expand the file structures, if the requested size exceeds 236 * the current capacity and there is room for expansion. 237 * Return <0 error code on error; 0 when nothing done; 1 when files were 238 * expanded and execution may have blocked. 239 * The files->file_lock should be held on entry, and will be held on exit. 240 */ 241 static int expand_files(struct files_struct *files, int nr) 242 { 243 struct fdtable *fdt; 244 245 fdt = files_fdtable(files); 246 247 /* Do we need to expand? */ 248 if (nr < fdt->max_fds) 249 return 0; 250 251 /* Can we expand? */ 252 if (nr >= sysctl_nr_open) 253 return -EMFILE; 254 255 /* All good, so we try */ 256 return expand_fdtable(files, nr); 257 } 258 259 static inline void __set_close_on_exec(int fd, struct fdtable *fdt) 260 { 261 __set_bit(fd, fdt->close_on_exec); 262 } 263 264 static inline void __clear_close_on_exec(int fd, struct fdtable *fdt) 265 { 266 __clear_bit(fd, fdt->close_on_exec); 267 } 268 269 static inline void __set_open_fd(int fd, struct fdtable *fdt) 270 { 271 __set_bit(fd, fdt->open_fds); 272 } 273 274 static inline void __clear_open_fd(int fd, struct fdtable *fdt) 275 { 276 __clear_bit(fd, fdt->open_fds); 277 } 278 279 static int count_open_files(struct fdtable *fdt) 280 { 281 int size = fdt->max_fds; 282 int i; 283 284 /* Find the last open fd */ 285 for (i = size / BITS_PER_LONG; i > 0; ) { 286 if (fdt->open_fds[--i]) 287 break; 288 } 289 i = (i + 1) * BITS_PER_LONG; 290 return i; 291 } 292 293 /* 294 * Allocate a new files structure and copy contents from the 295 * passed in files structure. 296 * errorp will be valid only when the returned files_struct is NULL. 297 */ 298 struct files_struct *dup_fd(struct files_struct *oldf, int *errorp) 299 { 300 struct files_struct *newf; 301 struct file **old_fds, **new_fds; 302 int open_files, size, i; 303 struct fdtable *old_fdt, *new_fdt; 304 305 *errorp = -ENOMEM; 306 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL); 307 if (!newf) 308 goto out; 309 310 atomic_set(&newf->count, 1); 311 312 spin_lock_init(&newf->file_lock); 313 newf->next_fd = 0; 314 new_fdt = &newf->fdtab; 315 new_fdt->max_fds = NR_OPEN_DEFAULT; 316 new_fdt->close_on_exec = newf->close_on_exec_init; 317 new_fdt->open_fds = newf->open_fds_init; 318 new_fdt->fd = &newf->fd_array[0]; 319 new_fdt->next = NULL; 320 321 spin_lock(&oldf->file_lock); 322 old_fdt = files_fdtable(oldf); 323 open_files = count_open_files(old_fdt); 324 325 /* 326 * Check whether we need to allocate a larger fd array and fd set. 327 */ 328 while (unlikely(open_files > new_fdt->max_fds)) { 329 spin_unlock(&oldf->file_lock); 330 331 if (new_fdt != &newf->fdtab) 332 __free_fdtable(new_fdt); 333 334 new_fdt = alloc_fdtable(open_files - 1); 335 if (!new_fdt) { 336 *errorp = -ENOMEM; 337 goto out_release; 338 } 339 340 /* beyond sysctl_nr_open; nothing to do */ 341 if (unlikely(new_fdt->max_fds < open_files)) { 342 __free_fdtable(new_fdt); 343 *errorp = -EMFILE; 344 goto out_release; 345 } 346 347 /* 348 * Reacquire the oldf lock and a pointer to its fd table 349 * who knows it may have a new bigger fd table. We need 350 * the latest pointer. 351 */ 352 spin_lock(&oldf->file_lock); 353 old_fdt = files_fdtable(oldf); 354 open_files = count_open_files(old_fdt); 355 } 356 357 old_fds = old_fdt->fd; 358 new_fds = new_fdt->fd; 359 360 memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8); 361 memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8); 362 363 for (i = open_files; i != 0; i--) { 364 struct file *f = *old_fds++; 365 if (f) { 366 get_file(f); 367 } else { 368 /* 369 * The fd may be claimed in the fd bitmap but not yet 370 * instantiated in the files array if a sibling thread 371 * is partway through open(). So make sure that this 372 * fd is available to the new process. 373 */ 374 __clear_open_fd(open_files - i, new_fdt); 375 } 376 rcu_assign_pointer(*new_fds++, f); 377 } 378 spin_unlock(&oldf->file_lock); 379 380 /* compute the remainder to be cleared */ 381 size = (new_fdt->max_fds - open_files) * sizeof(struct file *); 382 383 /* This is long word aligned thus could use a optimized version */ 384 memset(new_fds, 0, size); 385 386 if (new_fdt->max_fds > open_files) { 387 int left = (new_fdt->max_fds - open_files) / 8; 388 int start = open_files / BITS_PER_LONG; 389 390 memset(&new_fdt->open_fds[start], 0, left); 391 memset(&new_fdt->close_on_exec[start], 0, left); 392 } 393 394 rcu_assign_pointer(newf->fdt, new_fdt); 395 396 return newf; 397 398 out_release: 399 kmem_cache_free(files_cachep, newf); 400 out: 401 return NULL; 402 } 403 404 static void close_files(struct files_struct * files) 405 { 406 int i, j; 407 struct fdtable *fdt; 408 409 j = 0; 410 411 /* 412 * It is safe to dereference the fd table without RCU or 413 * ->file_lock because this is the last reference to the 414 * files structure. But use RCU to shut RCU-lockdep up. 415 */ 416 rcu_read_lock(); 417 fdt = files_fdtable(files); 418 rcu_read_unlock(); 419 for (;;) { 420 unsigned long set; 421 i = j * BITS_PER_LONG; 422 if (i >= fdt->max_fds) 423 break; 424 set = fdt->open_fds[j++]; 425 while (set) { 426 if (set & 1) { 427 struct file * file = xchg(&fdt->fd[i], NULL); 428 if (file) { 429 filp_close(file, files); 430 cond_resched(); 431 } 432 } 433 i++; 434 set >>= 1; 435 } 436 } 437 } 438 439 struct files_struct *get_files_struct(struct task_struct *task) 440 { 441 struct files_struct *files; 442 443 task_lock(task); 444 files = task->files; 445 if (files) 446 atomic_inc(&files->count); 447 task_unlock(task); 448 449 return files; 450 } 451 452 void put_files_struct(struct files_struct *files) 453 { 454 struct fdtable *fdt; 455 456 if (atomic_dec_and_test(&files->count)) { 457 close_files(files); 458 /* not really needed, since nobody can see us */ 459 rcu_read_lock(); 460 fdt = files_fdtable(files); 461 rcu_read_unlock(); 462 /* free the arrays if they are not embedded */ 463 if (fdt != &files->fdtab) 464 __free_fdtable(fdt); 465 kmem_cache_free(files_cachep, files); 466 } 467 } 468 469 void reset_files_struct(struct files_struct *files) 470 { 471 struct task_struct *tsk = current; 472 struct files_struct *old; 473 474 old = tsk->files; 475 task_lock(tsk); 476 tsk->files = files; 477 task_unlock(tsk); 478 put_files_struct(old); 479 } 480 481 void exit_files(struct task_struct *tsk) 482 { 483 struct files_struct * files = tsk->files; 484 485 if (files) { 486 task_lock(tsk); 487 tsk->files = NULL; 488 task_unlock(tsk); 489 put_files_struct(files); 490 } 491 } 492 493 static void __devinit fdtable_defer_list_init(int cpu) 494 { 495 struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu); 496 spin_lock_init(&fddef->lock); 497 INIT_WORK(&fddef->wq, free_fdtable_work); 498 fddef->next = NULL; 499 } 500 501 void __init files_defer_init(void) 502 { 503 int i; 504 for_each_possible_cpu(i) 505 fdtable_defer_list_init(i); 506 sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) & 507 -BITS_PER_LONG; 508 } 509 510 struct files_struct init_files = { 511 .count = ATOMIC_INIT(1), 512 .fdt = &init_files.fdtab, 513 .fdtab = { 514 .max_fds = NR_OPEN_DEFAULT, 515 .fd = &init_files.fd_array[0], 516 .close_on_exec = init_files.close_on_exec_init, 517 .open_fds = init_files.open_fds_init, 518 }, 519 .file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock), 520 }; 521 522 void daemonize_descriptors(void) 523 { 524 atomic_inc(&init_files.count); 525 reset_files_struct(&init_files); 526 } 527 528 /* 529 * allocate a file descriptor, mark it busy. 530 */ 531 int __alloc_fd(struct files_struct *files, 532 unsigned start, unsigned end, unsigned flags) 533 { 534 unsigned int fd; 535 int error; 536 struct fdtable *fdt; 537 538 spin_lock(&files->file_lock); 539 repeat: 540 fdt = files_fdtable(files); 541 fd = start; 542 if (fd < files->next_fd) 543 fd = files->next_fd; 544 545 if (fd < fdt->max_fds) 546 fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd); 547 548 /* 549 * N.B. For clone tasks sharing a files structure, this test 550 * will limit the total number of files that can be opened. 551 */ 552 error = -EMFILE; 553 if (fd >= end) 554 goto out; 555 556 error = expand_files(files, fd); 557 if (error < 0) 558 goto out; 559 560 /* 561 * If we needed to expand the fs array we 562 * might have blocked - try again. 563 */ 564 if (error) 565 goto repeat; 566 567 if (start <= files->next_fd) 568 files->next_fd = fd + 1; 569 570 __set_open_fd(fd, fdt); 571 if (flags & O_CLOEXEC) 572 __set_close_on_exec(fd, fdt); 573 else 574 __clear_close_on_exec(fd, fdt); 575 error = fd; 576 #if 1 577 /* Sanity check */ 578 if (rcu_dereference_raw(fdt->fd[fd]) != NULL) { 579 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd); 580 rcu_assign_pointer(fdt->fd[fd], NULL); 581 } 582 #endif 583 584 out: 585 spin_unlock(&files->file_lock); 586 return error; 587 } 588 589 static int alloc_fd(unsigned start, unsigned flags) 590 { 591 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags); 592 } 593 594 int get_unused_fd_flags(unsigned flags) 595 { 596 return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags); 597 } 598 EXPORT_SYMBOL(get_unused_fd_flags); 599 600 static void __put_unused_fd(struct files_struct *files, unsigned int fd) 601 { 602 struct fdtable *fdt = files_fdtable(files); 603 __clear_open_fd(fd, fdt); 604 if (fd < files->next_fd) 605 files->next_fd = fd; 606 } 607 608 void put_unused_fd(unsigned int fd) 609 { 610 struct files_struct *files = current->files; 611 spin_lock(&files->file_lock); 612 __put_unused_fd(files, fd); 613 spin_unlock(&files->file_lock); 614 } 615 616 EXPORT_SYMBOL(put_unused_fd); 617 618 /* 619 * Install a file pointer in the fd array. 620 * 621 * The VFS is full of places where we drop the files lock between 622 * setting the open_fds bitmap and installing the file in the file 623 * array. At any such point, we are vulnerable to a dup2() race 624 * installing a file in the array before us. We need to detect this and 625 * fput() the struct file we are about to overwrite in this case. 626 * 627 * It should never happen - if we allow dup2() do it, _really_ bad things 628 * will follow. 629 * 630 * NOTE: __fd_install() variant is really, really low-level; don't 631 * use it unless you are forced to by truly lousy API shoved down 632 * your throat. 'files' *MUST* be either current->files or obtained 633 * by get_files_struct(current) done by whoever had given it to you, 634 * or really bad things will happen. Normally you want to use 635 * fd_install() instead. 636 */ 637 638 void __fd_install(struct files_struct *files, unsigned int fd, 639 struct file *file) 640 { 641 struct fdtable *fdt; 642 spin_lock(&files->file_lock); 643 fdt = files_fdtable(files); 644 BUG_ON(fdt->fd[fd] != NULL); 645 rcu_assign_pointer(fdt->fd[fd], file); 646 spin_unlock(&files->file_lock); 647 } 648 649 void fd_install(unsigned int fd, struct file *file) 650 { 651 __fd_install(current->files, fd, file); 652 } 653 654 EXPORT_SYMBOL(fd_install); 655 656 /* 657 * The same warnings as for __alloc_fd()/__fd_install() apply here... 658 */ 659 int __close_fd(struct files_struct *files, unsigned fd) 660 { 661 struct file *file; 662 struct fdtable *fdt; 663 664 spin_lock(&files->file_lock); 665 fdt = files_fdtable(files); 666 if (fd >= fdt->max_fds) 667 goto out_unlock; 668 file = fdt->fd[fd]; 669 if (!file) 670 goto out_unlock; 671 rcu_assign_pointer(fdt->fd[fd], NULL); 672 __clear_close_on_exec(fd, fdt); 673 __put_unused_fd(files, fd); 674 spin_unlock(&files->file_lock); 675 return filp_close(file, files); 676 677 out_unlock: 678 spin_unlock(&files->file_lock); 679 return -EBADF; 680 } 681 682 void do_close_on_exec(struct files_struct *files) 683 { 684 unsigned i; 685 struct fdtable *fdt; 686 687 /* exec unshares first */ 688 BUG_ON(atomic_read(&files->count) != 1); 689 spin_lock(&files->file_lock); 690 for (i = 0; ; i++) { 691 unsigned long set; 692 unsigned fd = i * BITS_PER_LONG; 693 fdt = files_fdtable(files); 694 if (fd >= fdt->max_fds) 695 break; 696 set = fdt->close_on_exec[i]; 697 if (!set) 698 continue; 699 fdt->close_on_exec[i] = 0; 700 for ( ; set ; fd++, set >>= 1) { 701 struct file *file; 702 if (!(set & 1)) 703 continue; 704 file = fdt->fd[fd]; 705 if (!file) 706 continue; 707 rcu_assign_pointer(fdt->fd[fd], NULL); 708 __put_unused_fd(files, fd); 709 spin_unlock(&files->file_lock); 710 filp_close(file, files); 711 cond_resched(); 712 spin_lock(&files->file_lock); 713 } 714 715 } 716 spin_unlock(&files->file_lock); 717 } 718 719 struct file *fget(unsigned int fd) 720 { 721 struct file *file; 722 struct files_struct *files = current->files; 723 724 rcu_read_lock(); 725 file = fcheck_files(files, fd); 726 if (file) { 727 /* File object ref couldn't be taken */ 728 if (file->f_mode & FMODE_PATH || 729 !atomic_long_inc_not_zero(&file->f_count)) 730 file = NULL; 731 } 732 rcu_read_unlock(); 733 734 return file; 735 } 736 737 EXPORT_SYMBOL(fget); 738 739 struct file *fget_raw(unsigned int fd) 740 { 741 struct file *file; 742 struct files_struct *files = current->files; 743 744 rcu_read_lock(); 745 file = fcheck_files(files, fd); 746 if (file) { 747 /* File object ref couldn't be taken */ 748 if (!atomic_long_inc_not_zero(&file->f_count)) 749 file = NULL; 750 } 751 rcu_read_unlock(); 752 753 return file; 754 } 755 756 EXPORT_SYMBOL(fget_raw); 757 758 /* 759 * Lightweight file lookup - no refcnt increment if fd table isn't shared. 760 * 761 * You can use this instead of fget if you satisfy all of the following 762 * conditions: 763 * 1) You must call fput_light before exiting the syscall and returning control 764 * to userspace (i.e. you cannot remember the returned struct file * after 765 * returning to userspace). 766 * 2) You must not call filp_close on the returned struct file * in between 767 * calls to fget_light and fput_light. 768 * 3) You must not clone the current task in between the calls to fget_light 769 * and fput_light. 770 * 771 * The fput_needed flag returned by fget_light should be passed to the 772 * corresponding fput_light. 773 */ 774 struct file *fget_light(unsigned int fd, int *fput_needed) 775 { 776 struct file *file; 777 struct files_struct *files = current->files; 778 779 *fput_needed = 0; 780 if (atomic_read(&files->count) == 1) { 781 file = fcheck_files(files, fd); 782 if (file && (file->f_mode & FMODE_PATH)) 783 file = NULL; 784 } else { 785 rcu_read_lock(); 786 file = fcheck_files(files, fd); 787 if (file) { 788 if (!(file->f_mode & FMODE_PATH) && 789 atomic_long_inc_not_zero(&file->f_count)) 790 *fput_needed = 1; 791 else 792 /* Didn't get the reference, someone's freed */ 793 file = NULL; 794 } 795 rcu_read_unlock(); 796 } 797 798 return file; 799 } 800 EXPORT_SYMBOL(fget_light); 801 802 struct file *fget_raw_light(unsigned int fd, int *fput_needed) 803 { 804 struct file *file; 805 struct files_struct *files = current->files; 806 807 *fput_needed = 0; 808 if (atomic_read(&files->count) == 1) { 809 file = fcheck_files(files, fd); 810 } else { 811 rcu_read_lock(); 812 file = fcheck_files(files, fd); 813 if (file) { 814 if (atomic_long_inc_not_zero(&file->f_count)) 815 *fput_needed = 1; 816 else 817 /* Didn't get the reference, someone's freed */ 818 file = NULL; 819 } 820 rcu_read_unlock(); 821 } 822 823 return file; 824 } 825 826 void set_close_on_exec(unsigned int fd, int flag) 827 { 828 struct files_struct *files = current->files; 829 struct fdtable *fdt; 830 spin_lock(&files->file_lock); 831 fdt = files_fdtable(files); 832 if (flag) 833 __set_close_on_exec(fd, fdt); 834 else 835 __clear_close_on_exec(fd, fdt); 836 spin_unlock(&files->file_lock); 837 } 838 839 bool get_close_on_exec(unsigned int fd) 840 { 841 struct files_struct *files = current->files; 842 struct fdtable *fdt; 843 bool res; 844 rcu_read_lock(); 845 fdt = files_fdtable(files); 846 res = close_on_exec(fd, fdt); 847 rcu_read_unlock(); 848 return res; 849 } 850 851 static int do_dup2(struct files_struct *files, 852 struct file *file, unsigned fd, unsigned flags) 853 { 854 struct file *tofree; 855 struct fdtable *fdt; 856 857 /* 858 * We need to detect attempts to do dup2() over allocated but still 859 * not finished descriptor. NB: OpenBSD avoids that at the price of 860 * extra work in their equivalent of fget() - they insert struct 861 * file immediately after grabbing descriptor, mark it larval if 862 * more work (e.g. actual opening) is needed and make sure that 863 * fget() treats larval files as absent. Potentially interesting, 864 * but while extra work in fget() is trivial, locking implications 865 * and amount of surgery on open()-related paths in VFS are not. 866 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution" 867 * deadlocks in rather amusing ways, AFAICS. All of that is out of 868 * scope of POSIX or SUS, since neither considers shared descriptor 869 * tables and this condition does not arise without those. 870 */ 871 fdt = files_fdtable(files); 872 tofree = fdt->fd[fd]; 873 if (!tofree && fd_is_open(fd, fdt)) 874 goto Ebusy; 875 get_file(file); 876 rcu_assign_pointer(fdt->fd[fd], file); 877 __set_open_fd(fd, fdt); 878 if (flags & O_CLOEXEC) 879 __set_close_on_exec(fd, fdt); 880 else 881 __clear_close_on_exec(fd, fdt); 882 spin_unlock(&files->file_lock); 883 884 if (tofree) 885 filp_close(tofree, files); 886 887 return fd; 888 889 Ebusy: 890 spin_unlock(&files->file_lock); 891 return -EBUSY; 892 } 893 894 int replace_fd(unsigned fd, struct file *file, unsigned flags) 895 { 896 int err; 897 struct files_struct *files = current->files; 898 899 if (!file) 900 return __close_fd(files, fd); 901 902 if (fd >= rlimit(RLIMIT_NOFILE)) 903 return -EMFILE; 904 905 spin_lock(&files->file_lock); 906 err = expand_files(files, fd); 907 if (unlikely(err < 0)) 908 goto out_unlock; 909 return do_dup2(files, file, fd, flags); 910 911 out_unlock: 912 spin_unlock(&files->file_lock); 913 return err; 914 } 915 916 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags) 917 { 918 int err = -EBADF; 919 struct file *file; 920 struct files_struct *files = current->files; 921 922 if ((flags & ~O_CLOEXEC) != 0) 923 return -EINVAL; 924 925 if (unlikely(oldfd == newfd)) 926 return -EINVAL; 927 928 if (newfd >= rlimit(RLIMIT_NOFILE)) 929 return -EMFILE; 930 931 spin_lock(&files->file_lock); 932 err = expand_files(files, newfd); 933 file = fcheck(oldfd); 934 if (unlikely(!file)) 935 goto Ebadf; 936 if (unlikely(err < 0)) { 937 if (err == -EMFILE) 938 goto Ebadf; 939 goto out_unlock; 940 } 941 return do_dup2(files, file, newfd, flags); 942 943 Ebadf: 944 err = -EBADF; 945 out_unlock: 946 spin_unlock(&files->file_lock); 947 return err; 948 } 949 950 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd) 951 { 952 if (unlikely(newfd == oldfd)) { /* corner case */ 953 struct files_struct *files = current->files; 954 int retval = oldfd; 955 956 rcu_read_lock(); 957 if (!fcheck_files(files, oldfd)) 958 retval = -EBADF; 959 rcu_read_unlock(); 960 return retval; 961 } 962 return sys_dup3(oldfd, newfd, 0); 963 } 964 965 SYSCALL_DEFINE1(dup, unsigned int, fildes) 966 { 967 int ret = -EBADF; 968 struct file *file = fget_raw(fildes); 969 970 if (file) { 971 ret = get_unused_fd(); 972 if (ret >= 0) 973 fd_install(ret, file); 974 else 975 fput(file); 976 } 977 return ret; 978 } 979 980 int f_dupfd(unsigned int from, struct file *file, unsigned flags) 981 { 982 int err; 983 if (from >= rlimit(RLIMIT_NOFILE)) 984 return -EINVAL; 985 err = alloc_fd(from, flags); 986 if (err >= 0) { 987 get_file(file); 988 fd_install(err, file); 989 } 990 return err; 991 } 992 993 int iterate_fd(struct files_struct *files, unsigned n, 994 int (*f)(const void *, struct file *, unsigned), 995 const void *p) 996 { 997 struct fdtable *fdt; 998 struct file *file; 999 int res = 0; 1000 if (!files) 1001 return 0; 1002 spin_lock(&files->file_lock); 1003 fdt = files_fdtable(files); 1004 while (!res && n < fdt->max_fds) { 1005 file = rcu_dereference_check_fdtable(files, fdt->fd[n++]); 1006 if (file) 1007 res = f(p, file, n); 1008 } 1009 spin_unlock(&files->file_lock); 1010 return res; 1011 } 1012 EXPORT_SYMBOL(iterate_fd); 1013