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