1
2
3
4
5
6
7
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
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
51
52
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
74
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
97
98
99
100
101
102 nr /= (1024 / sizeof(struct file *));
103 nr = roundup_pow_of_two(nr + 1);
104 nr *= (1024 / sizeof(struct file *));
105
106
107
108
109
110
111
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
148
149
150
151
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
163
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
173
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
186 smp_wmb();
187 return 1;
188}
189
190
191
192
193
194
195
196
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
209 if (nr < fdt->max_fds)
210 return expanded;
211
212
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
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
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
284
285
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
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
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
340
341
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
360
361
362
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
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
387
388
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
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
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
489
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
501
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
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
559
560
561
562
563
564
565
566
567
568
569
570
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
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
601
602
603
604
605
606
607
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);
645
646
647
648
649
650
651
652
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
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
682 filp_close(file, cur_fds);
683 cond_resched();
684 continue;
685 }
686
687
688 if (PTR_ERR(file) == -EINVAL)
689 return;
690 }
691}
692
693
694
695
696
697
698
699
700
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
719
720
721
722 if (!(flags & CLOSE_RANGE_CLOEXEC)) {
723
724
725
726
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
740
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
754
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
767
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
793
794
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
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 struct file *__fget_files(struct files_struct *files, unsigned int fd,
845 fmode_t mask, unsigned int refs)
846{
847 struct file *file;
848
849 rcu_read_lock();
850loop:
851 file = files_lookup_fd_rcu(files, fd);
852 if (file) {
853
854
855
856
857 if (file->f_mode & mask)
858 file = NULL;
859 else if (!get_file_rcu_many(file, refs))
860 goto loop;
861 }
862 rcu_read_unlock();
863
864 return file;
865}
866
867static inline struct file *__fget(unsigned int fd, fmode_t mask,
868 unsigned int refs)
869{
870 return __fget_files(current->files, fd, mask, refs);
871}
872
873struct file *fget_many(unsigned int fd, unsigned int refs)
874{
875 return __fget(fd, FMODE_PATH, refs);
876}
877
878struct file *fget(unsigned int fd)
879{
880 return __fget(fd, FMODE_PATH, 1);
881}
882EXPORT_SYMBOL(fget);
883
884struct file *fget_raw(unsigned int fd)
885{
886 return __fget(fd, 0, 1);
887}
888EXPORT_SYMBOL(fget_raw);
889
890struct file *fget_task(struct task_struct *task, unsigned int fd)
891{
892 struct file *file = NULL;
893
894 task_lock(task);
895 if (task->files)
896 file = __fget_files(task->files, fd, 0, 1);
897 task_unlock(task);
898
899 return file;
900}
901
902struct file *task_lookup_fd_rcu(struct task_struct *task, unsigned int fd)
903{
904
905 struct files_struct *files;
906 struct file *file = NULL;
907
908 task_lock(task);
909 files = task->files;
910 if (files)
911 file = files_lookup_fd_rcu(files, fd);
912 task_unlock(task);
913
914 return file;
915}
916
917struct file *task_lookup_next_fd_rcu(struct task_struct *task, unsigned int *ret_fd)
918{
919
920 struct files_struct *files;
921 unsigned int fd = *ret_fd;
922 struct file *file = NULL;
923
924 task_lock(task);
925 files = task->files;
926 if (files) {
927 for (; fd < files_fdtable(files)->max_fds; fd++) {
928 file = files_lookup_fd_rcu(files, fd);
929 if (file)
930 break;
931 }
932 }
933 task_unlock(task);
934 *ret_fd = fd;
935 return file;
936}
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954static unsigned long __fget_light(unsigned int fd, fmode_t mask)
955{
956 struct files_struct *files = current->files;
957 struct file *file;
958
959 if (atomic_read(&files->count) == 1) {
960 file = files_lookup_fd_raw(files, fd);
961 if (!file || unlikely(file->f_mode & mask))
962 return 0;
963 return (unsigned long)file;
964 } else {
965 file = __fget(fd, mask, 1);
966 if (!file)
967 return 0;
968 return FDPUT_FPUT | (unsigned long)file;
969 }
970}
971unsigned long __fdget(unsigned int fd)
972{
973 return __fget_light(fd, FMODE_PATH);
974}
975EXPORT_SYMBOL(__fdget);
976
977unsigned long __fdget_raw(unsigned int fd)
978{
979 return __fget_light(fd, 0);
980}
981
982unsigned long __fdget_pos(unsigned int fd)
983{
984 unsigned long v = __fdget(fd);
985 struct file *file = (struct file *)(v & ~3);
986
987 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
988 if (file_count(file) > 1) {
989 v |= FDPUT_POS_UNLOCK;
990 mutex_lock(&file->f_pos_lock);
991 }
992 }
993 return v;
994}
995
996void __f_unlock_pos(struct file *f)
997{
998 mutex_unlock(&f->f_pos_lock);
999}
1000
1001
1002
1003
1004
1005
1006
1007void set_close_on_exec(unsigned int fd, int flag)
1008{
1009 struct files_struct *files = current->files;
1010 struct fdtable *fdt;
1011 spin_lock(&files->file_lock);
1012 fdt = files_fdtable(files);
1013 if (flag)
1014 __set_close_on_exec(fd, fdt);
1015 else
1016 __clear_close_on_exec(fd, fdt);
1017 spin_unlock(&files->file_lock);
1018}
1019
1020bool get_close_on_exec(unsigned int fd)
1021{
1022 struct files_struct *files = current->files;
1023 struct fdtable *fdt;
1024 bool res;
1025 rcu_read_lock();
1026 fdt = files_fdtable(files);
1027 res = close_on_exec(fd, fdt);
1028 rcu_read_unlock();
1029 return res;
1030}
1031
1032static int do_dup2(struct files_struct *files,
1033 struct file *file, unsigned fd, unsigned flags)
1034__releases(&files->file_lock)
1035{
1036 struct file *tofree;
1037 struct fdtable *fdt;
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053 fdt = files_fdtable(files);
1054 tofree = fdt->fd[fd];
1055 if (!tofree && fd_is_open(fd, fdt))
1056 goto Ebusy;
1057 get_file(file);
1058 rcu_assign_pointer(fdt->fd[fd], file);
1059 __set_open_fd(fd, fdt);
1060 if (flags & O_CLOEXEC)
1061 __set_close_on_exec(fd, fdt);
1062 else
1063 __clear_close_on_exec(fd, fdt);
1064 spin_unlock(&files->file_lock);
1065
1066 if (tofree)
1067 filp_close(tofree, files);
1068
1069 return fd;
1070
1071Ebusy:
1072 spin_unlock(&files->file_lock);
1073 return -EBUSY;
1074}
1075
1076int replace_fd(unsigned fd, struct file *file, unsigned flags)
1077{
1078 int err;
1079 struct files_struct *files = current->files;
1080
1081 if (!file)
1082 return close_fd(fd);
1083
1084 if (fd >= rlimit(RLIMIT_NOFILE))
1085 return -EBADF;
1086
1087 spin_lock(&files->file_lock);
1088 err = expand_files(files, fd);
1089 if (unlikely(err < 0))
1090 goto out_unlock;
1091 return do_dup2(files, file, fd, flags);
1092
1093out_unlock:
1094 spin_unlock(&files->file_lock);
1095 return err;
1096}
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113int __receive_fd(struct file *file, int __user *ufd, unsigned int o_flags)
1114{
1115 int new_fd;
1116 int error;
1117
1118 error = security_file_receive(file);
1119 if (error)
1120 return error;
1121
1122 new_fd = get_unused_fd_flags(o_flags);
1123 if (new_fd < 0)
1124 return new_fd;
1125
1126 if (ufd) {
1127 error = put_user(new_fd, ufd);
1128 if (error) {
1129 put_unused_fd(new_fd);
1130 return error;
1131 }
1132 }
1133
1134 fd_install(new_fd, get_file(file));
1135 __receive_sock(file);
1136 return new_fd;
1137}
1138
1139int receive_fd_replace(int new_fd, struct file *file, unsigned int o_flags)
1140{
1141 int error;
1142
1143 error = security_file_receive(file);
1144 if (error)
1145 return error;
1146 error = replace_fd(new_fd, file, o_flags);
1147 if (error)
1148 return error;
1149 __receive_sock(file);
1150 return new_fd;
1151}
1152
1153int receive_fd(struct file *file, unsigned int o_flags)
1154{
1155 return __receive_fd(file, NULL, o_flags);
1156}
1157EXPORT_SYMBOL_GPL(receive_fd);
1158
1159static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1160{
1161 int err = -EBADF;
1162 struct file *file;
1163 struct files_struct *files = current->files;
1164
1165 if ((flags & ~O_CLOEXEC) != 0)
1166 return -EINVAL;
1167
1168 if (unlikely(oldfd == newfd))
1169 return -EINVAL;
1170
1171 if (newfd >= rlimit(RLIMIT_NOFILE))
1172 return -EBADF;
1173
1174 spin_lock(&files->file_lock);
1175 err = expand_files(files, newfd);
1176 file = files_lookup_fd_locked(files, oldfd);
1177 if (unlikely(!file))
1178 goto Ebadf;
1179 if (unlikely(err < 0)) {
1180 if (err == -EMFILE)
1181 goto Ebadf;
1182 goto out_unlock;
1183 }
1184 return do_dup2(files, file, newfd, flags);
1185
1186Ebadf:
1187 err = -EBADF;
1188out_unlock:
1189 spin_unlock(&files->file_lock);
1190 return err;
1191}
1192
1193SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1194{
1195 return ksys_dup3(oldfd, newfd, flags);
1196}
1197
1198SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1199{
1200 if (unlikely(newfd == oldfd)) {
1201 struct files_struct *files = current->files;
1202 int retval = oldfd;
1203
1204 rcu_read_lock();
1205 if (!files_lookup_fd_rcu(files, oldfd))
1206 retval = -EBADF;
1207 rcu_read_unlock();
1208 return retval;
1209 }
1210 return ksys_dup3(oldfd, newfd, 0);
1211}
1212
1213SYSCALL_DEFINE1(dup, unsigned int, fildes)
1214{
1215 int ret = -EBADF;
1216 struct file *file = fget_raw(fildes);
1217
1218 if (file) {
1219 ret = get_unused_fd_flags(0);
1220 if (ret >= 0)
1221 fd_install(ret, file);
1222 else
1223 fput(file);
1224 }
1225 return ret;
1226}
1227
1228int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1229{
1230 unsigned long nofile = rlimit(RLIMIT_NOFILE);
1231 int err;
1232 if (from >= nofile)
1233 return -EINVAL;
1234 err = alloc_fd(from, nofile, flags);
1235 if (err >= 0) {
1236 get_file(file);
1237 fd_install(err, file);
1238 }
1239 return err;
1240}
1241
1242int iterate_fd(struct files_struct *files, unsigned n,
1243 int (*f)(const void *, struct file *, unsigned),
1244 const void *p)
1245{
1246 struct fdtable *fdt;
1247 int res = 0;
1248 if (!files)
1249 return 0;
1250 spin_lock(&files->file_lock);
1251 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1252 struct file *file;
1253 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1254 if (!file)
1255 continue;
1256 res = f(p, file, n);
1257 if (res)
1258 break;
1259 }
1260 spin_unlock(&files->file_lock);
1261 return res;
1262}
1263EXPORT_SYMBOL(iterate_fd);
1264