Merge tag 'devicetree-for-4.19' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-block.git] / fs / file.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
1da177e4
LT
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
fe17f22d 10#include <linux/syscalls.h>
630d9c47 11#include <linux/export.h>
1da177e4
LT
12#include <linux/fs.h>
13#include <linux/mm.h>
3f07c014 14#include <linux/sched/signal.h>
1da177e4 15#include <linux/slab.h>
1da177e4 16#include <linux/file.h>
9f3acc31 17#include <linux/fdtable.h>
1da177e4 18#include <linux/bitops.h>
ab2af1f5
DS
19#include <linux/spinlock.h>
20#include <linux/rcupdate.h>
ab2af1f5 21
9b80a184
AD
22unsigned int sysctl_nr_open __read_mostly = 1024*1024;
23unsigned int sysctl_nr_open_min = BITS_PER_LONG;
752343be
RV
24/* our min() is unusable in constant expressions ;-/ */
25#define __const_min(x, y) ((x) < (y) ? (x) : (y))
9b80a184
AD
26unsigned int sysctl_nr_open_max =
27 __const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG;
9cfe015a 28
a892e2d7 29static void __free_fdtable(struct fdtable *fdt)
1da177e4 30{
f6c0a192
AV
31 kvfree(fdt->fd);
32 kvfree(fdt->open_fds);
a892e2d7 33 kfree(fdt);
ab2af1f5 34}
1da177e4 35
7cf4dc3c 36static void free_fdtable_rcu(struct rcu_head *rcu)
ab2af1f5 37{
ac3e3c5b 38 __free_fdtable(container_of(rcu, struct fdtable, rcu));
ab2af1f5
DS
39}
40
f3f86e33
LT
41#define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr))
42#define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
43
ab2af1f5 44/*
ea5c58e7
EB
45 * Copy 'count' fd bits from the old table to the new table and clear the extra
46 * space if any. This does not copy the file pointers. Called with the files
47 * spinlock held for write.
48 */
49static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
50 unsigned int count)
51{
52 unsigned int cpy, set;
53
54 cpy = count / BITS_PER_BYTE;
55 set = (nfdt->max_fds - count) / BITS_PER_BYTE;
56 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
57 memset((char *)nfdt->open_fds + cpy, 0, set);
58 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
59 memset((char *)nfdt->close_on_exec + cpy, 0, set);
60
61 cpy = BITBIT_SIZE(count);
62 set = BITBIT_SIZE(nfdt->max_fds) - cpy;
63 memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
64 memset((char *)nfdt->full_fds_bits + cpy, 0, set);
65}
66
67/*
68 * Copy all file descriptors from the old table to the new, expanded table and
69 * clear the extra space. Called with the files spinlock held for write.
ab2af1f5 70 */
5466b456 71static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
ab2af1f5 72{
5466b456 73 unsigned int cpy, set;
ab2af1f5 74
5466b456 75 BUG_ON(nfdt->max_fds < ofdt->max_fds);
5466b456
VL
76
77 cpy = ofdt->max_fds * sizeof(struct file *);
78 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
79 memcpy(nfdt->fd, ofdt->fd, cpy);
ea5c58e7 80 memset((char *)nfdt->fd + cpy, 0, set);
5466b456 81
ea5c58e7 82 copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
1da177e4
LT
83}
84
5466b456 85static struct fdtable * alloc_fdtable(unsigned int nr)
1da177e4 86{
5466b456 87 struct fdtable *fdt;
1fd36adc 88 void *data;
1da177e4 89
ab2af1f5 90 /*
5466b456
VL
91 * Figure out how many fds we actually want to support in this fdtable.
92 * Allocation steps are keyed to the size of the fdarray, since it
93 * grows far faster than any of the other dynamic data. We try to fit
94 * the fdarray into comfortable page-tuned chunks: starting at 1024B
95 * and growing in powers of two from there on.
ab2af1f5 96 */
5466b456
VL
97 nr /= (1024 / sizeof(struct file *));
98 nr = roundup_pow_of_two(nr + 1);
99 nr *= (1024 / sizeof(struct file *));
5c598b34
AV
100 /*
101 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
102 * had been set lower between the check in expand_files() and here. Deal
103 * with that in caller, it's cheaper that way.
104 *
105 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
106 * bitmaps handling below becomes unpleasant, to put it mildly...
107 */
108 if (unlikely(nr > sysctl_nr_open))
109 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
bbea9f69 110
5d097056 111 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
5466b456 112 if (!fdt)
bbea9f69 113 goto out;
5466b456 114 fdt->max_fds = nr;
c823bd92 115 data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
5466b456
VL
116 if (!data)
117 goto out_fdt;
1fd36adc
DH
118 fdt->fd = data;
119
c823bd92
MH
120 data = kvmalloc(max_t(size_t,
121 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
122 GFP_KERNEL_ACCOUNT);
5466b456
VL
123 if (!data)
124 goto out_arr;
1fd36adc 125 fdt->open_fds = data;
5466b456 126 data += nr / BITS_PER_BYTE;
1fd36adc 127 fdt->close_on_exec = data;
f3f86e33
LT
128 data += nr / BITS_PER_BYTE;
129 fdt->full_fds_bits = data;
5466b456 130
ab2af1f5 131 return fdt;
5466b456
VL
132
133out_arr:
f6c0a192 134 kvfree(fdt->fd);
5466b456 135out_fdt:
ab2af1f5 136 kfree(fdt);
5466b456 137out:
ab2af1f5
DS
138 return NULL;
139}
1da177e4 140
ab2af1f5 141/*
74d392aa
VL
142 * Expand the file descriptor table.
143 * This function will allocate a new fdtable and both fd array and fdset, of
144 * the given size.
145 * Return <0 error code on error; 1 on successful completion.
146 * The files->file_lock should be held on entry, and will be held on exit.
ab2af1f5 147 */
9b80a184 148static int expand_fdtable(struct files_struct *files, unsigned int nr)
ab2af1f5
DS
149 __releases(files->file_lock)
150 __acquires(files->file_lock)
151{
74d392aa 152 struct fdtable *new_fdt, *cur_fdt;
ab2af1f5
DS
153
154 spin_unlock(&files->file_lock);
74d392aa 155 new_fdt = alloc_fdtable(nr);
8a81252b
ED
156
157 /* make sure all __fd_install() have seen resize_in_progress
158 * or have finished their rcu_read_lock_sched() section.
159 */
160 if (atomic_read(&files->count) > 1)
161 synchronize_sched();
162
ab2af1f5 163 spin_lock(&files->file_lock);
74d392aa
VL
164 if (!new_fdt)
165 return -ENOMEM;
5c598b34
AV
166 /*
167 * extremely unlikely race - sysctl_nr_open decreased between the check in
168 * caller and alloc_fdtable(). Cheaper to catch it here...
169 */
170 if (unlikely(new_fdt->max_fds <= nr)) {
a892e2d7 171 __free_fdtable(new_fdt);
5c598b34
AV
172 return -EMFILE;
173 }
74d392aa 174 cur_fdt = files_fdtable(files);
8a81252b
ED
175 BUG_ON(nr < cur_fdt->max_fds);
176 copy_fdtable(new_fdt, cur_fdt);
177 rcu_assign_pointer(files->fdt, new_fdt);
178 if (cur_fdt != &files->fdtab)
179 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
180 /* coupled with smp_rmb() in __fd_install() */
181 smp_wmb();
74d392aa 182 return 1;
1da177e4
LT
183}
184
185/*
186 * Expand files.
74d392aa
VL
187 * This function will expand the file structures, if the requested size exceeds
188 * the current capacity and there is room for expansion.
189 * Return <0 error code on error; 0 when nothing done; 1 when files were
190 * expanded and execution may have blocked.
191 * The files->file_lock should be held on entry, and will be held on exit.
1da177e4 192 */
9b80a184 193static int expand_files(struct files_struct *files, unsigned int nr)
8a81252b
ED
194 __releases(files->file_lock)
195 __acquires(files->file_lock)
1da177e4 196{
badf1662 197 struct fdtable *fdt;
8a81252b 198 int expanded = 0;
1da177e4 199
8a81252b 200repeat:
badf1662 201 fdt = files_fdtable(files);
4e1e018e 202
74d392aa 203 /* Do we need to expand? */
bbea9f69 204 if (nr < fdt->max_fds)
8a81252b 205 return expanded;
4e1e018e 206
74d392aa 207 /* Can we expand? */
9cfe015a 208 if (nr >= sysctl_nr_open)
74d392aa
VL
209 return -EMFILE;
210
8a81252b
ED
211 if (unlikely(files->resize_in_progress)) {
212 spin_unlock(&files->file_lock);
213 expanded = 1;
214 wait_event(files->resize_wait, !files->resize_in_progress);
215 spin_lock(&files->file_lock);
216 goto repeat;
217 }
218
74d392aa 219 /* All good, so we try */
8a81252b
ED
220 files->resize_in_progress = true;
221 expanded = expand_fdtable(files, nr);
222 files->resize_in_progress = false;
223
224 wake_up_all(&files->resize_wait);
225 return expanded;
1da177e4 226}
ab2af1f5 227
9b80a184 228static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
b8318b01
AV
229{
230 __set_bit(fd, fdt->close_on_exec);
231}
232
9b80a184 233static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
b8318b01 234{
fc90888d
LT
235 if (test_bit(fd, fdt->close_on_exec))
236 __clear_bit(fd, fdt->close_on_exec);
b8318b01
AV
237}
238
f3f86e33 239static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
b8318b01
AV
240{
241 __set_bit(fd, fdt->open_fds);
f3f86e33
LT
242 fd /= BITS_PER_LONG;
243 if (!~fdt->open_fds[fd])
244 __set_bit(fd, fdt->full_fds_bits);
b8318b01
AV
245}
246
f3f86e33 247static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
b8318b01
AV
248{
249 __clear_bit(fd, fdt->open_fds);
f3f86e33 250 __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
b8318b01
AV
251}
252
9b80a184 253static unsigned int count_open_files(struct fdtable *fdt)
02afc626 254{
9b80a184
AD
255 unsigned int size = fdt->max_fds;
256 unsigned int i;
02afc626
AV
257
258 /* Find the last open fd */
1fd36adc
DH
259 for (i = size / BITS_PER_LONG; i > 0; ) {
260 if (fdt->open_fds[--i])
02afc626
AV
261 break;
262 }
1fd36adc 263 i = (i + 1) * BITS_PER_LONG;
02afc626
AV
264 return i;
265}
266
02afc626
AV
267/*
268 * Allocate a new files structure and copy contents from the
269 * passed in files structure.
270 * errorp will be valid only when the returned files_struct is NULL.
271 */
272struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
273{
274 struct files_struct *newf;
275 struct file **old_fds, **new_fds;
9b80a184 276 unsigned int open_files, i;
02afc626
AV
277 struct fdtable *old_fdt, *new_fdt;
278
279 *errorp = -ENOMEM;
afbec7ff 280 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
02afc626
AV
281 if (!newf)
282 goto out;
283
afbec7ff
AV
284 atomic_set(&newf->count, 1);
285
286 spin_lock_init(&newf->file_lock);
8a81252b
ED
287 newf->resize_in_progress = false;
288 init_waitqueue_head(&newf->resize_wait);
afbec7ff
AV
289 newf->next_fd = 0;
290 new_fdt = &newf->fdtab;
291 new_fdt->max_fds = NR_OPEN_DEFAULT;
1fd36adc
DH
292 new_fdt->close_on_exec = newf->close_on_exec_init;
293 new_fdt->open_fds = newf->open_fds_init;
f3f86e33 294 new_fdt->full_fds_bits = newf->full_fds_bits_init;
afbec7ff 295 new_fdt->fd = &newf->fd_array[0];
afbec7ff 296
02afc626
AV
297 spin_lock(&oldf->file_lock);
298 old_fdt = files_fdtable(oldf);
02afc626
AV
299 open_files = count_open_files(old_fdt);
300
301 /*
302 * Check whether we need to allocate a larger fd array and fd set.
02afc626 303 */
adbecb12 304 while (unlikely(open_files > new_fdt->max_fds)) {
02afc626 305 spin_unlock(&oldf->file_lock);
9dec3c4d 306
a892e2d7
CG
307 if (new_fdt != &newf->fdtab)
308 __free_fdtable(new_fdt);
adbecb12 309
9dec3c4d
AV
310 new_fdt = alloc_fdtable(open_files - 1);
311 if (!new_fdt) {
312 *errorp = -ENOMEM;
313 goto out_release;
314 }
315
316 /* beyond sysctl_nr_open; nothing to do */
317 if (unlikely(new_fdt->max_fds < open_files)) {
a892e2d7 318 __free_fdtable(new_fdt);
9dec3c4d 319 *errorp = -EMFILE;
02afc626 320 goto out_release;
9dec3c4d 321 }
9dec3c4d 322
02afc626
AV
323 /*
324 * Reacquire the oldf lock and a pointer to its fd table
325 * who knows it may have a new bigger fd table. We need
326 * the latest pointer.
327 */
328 spin_lock(&oldf->file_lock);
329 old_fdt = files_fdtable(oldf);
adbecb12 330 open_files = count_open_files(old_fdt);
02afc626
AV
331 }
332
ea5c58e7
EB
333 copy_fd_bitmaps(new_fdt, old_fdt, open_files);
334
02afc626
AV
335 old_fds = old_fdt->fd;
336 new_fds = new_fdt->fd;
337
02afc626
AV
338 for (i = open_files; i != 0; i--) {
339 struct file *f = *old_fds++;
340 if (f) {
341 get_file(f);
342 } else {
343 /*
344 * The fd may be claimed in the fd bitmap but not yet
345 * instantiated in the files array if a sibling thread
346 * is partway through open(). So make sure that this
347 * fd is available to the new process.
348 */
1dce27c5 349 __clear_open_fd(open_files - i, new_fdt);
02afc626
AV
350 }
351 rcu_assign_pointer(*new_fds++, f);
352 }
353 spin_unlock(&oldf->file_lock);
354
ea5c58e7
EB
355 /* clear the remainder */
356 memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
02afc626 357
afbec7ff
AV
358 rcu_assign_pointer(newf->fdt, new_fdt);
359
02afc626
AV
360 return newf;
361
362out_release:
363 kmem_cache_free(files_cachep, newf);
364out:
365 return NULL;
366}
367
ce08b62d 368static struct fdtable *close_files(struct files_struct * files)
7cf4dc3c 369{
7cf4dc3c
AV
370 /*
371 * It is safe to dereference the fd table without RCU or
372 * ->file_lock because this is the last reference to the
ce08b62d 373 * files structure.
7cf4dc3c 374 */
ce08b62d 375 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
9b80a184 376 unsigned int i, j = 0;
ce08b62d 377
7cf4dc3c
AV
378 for (;;) {
379 unsigned long set;
380 i = j * BITS_PER_LONG;
381 if (i >= fdt->max_fds)
382 break;
383 set = fdt->open_fds[j++];
384 while (set) {
385 if (set & 1) {
386 struct file * file = xchg(&fdt->fd[i], NULL);
387 if (file) {
388 filp_close(file, files);
388a4c88 389 cond_resched();
7cf4dc3c
AV
390 }
391 }
392 i++;
393 set >>= 1;
394 }
395 }
ce08b62d
ON
396
397 return fdt;
7cf4dc3c
AV
398}
399
400struct files_struct *get_files_struct(struct task_struct *task)
401{
402 struct files_struct *files;
403
404 task_lock(task);
405 files = task->files;
406 if (files)
407 atomic_inc(&files->count);
408 task_unlock(task);
409
410 return files;
411}
412
413void put_files_struct(struct files_struct *files)
414{
7cf4dc3c 415 if (atomic_dec_and_test(&files->count)) {
ce08b62d
ON
416 struct fdtable *fdt = close_files(files);
417
b9e02af0
AV
418 /* free the arrays if they are not embedded */
419 if (fdt != &files->fdtab)
420 __free_fdtable(fdt);
421 kmem_cache_free(files_cachep, files);
7cf4dc3c
AV
422 }
423}
424
425void reset_files_struct(struct files_struct *files)
426{
427 struct task_struct *tsk = current;
428 struct files_struct *old;
429
430 old = tsk->files;
431 task_lock(tsk);
432 tsk->files = files;
433 task_unlock(tsk);
434 put_files_struct(old);
435}
436
437void exit_files(struct task_struct *tsk)
438{
439 struct files_struct * files = tsk->files;
440
441 if (files) {
442 task_lock(tsk);
443 tsk->files = NULL;
444 task_unlock(tsk);
445 put_files_struct(files);
446 }
447}
448
f52111b1
AV
449struct files_struct init_files = {
450 .count = ATOMIC_INIT(1),
451 .fdt = &init_files.fdtab,
452 .fdtab = {
453 .max_fds = NR_OPEN_DEFAULT,
454 .fd = &init_files.fd_array[0],
1fd36adc
DH
455 .close_on_exec = init_files.close_on_exec_init,
456 .open_fds = init_files.open_fds_init,
f3f86e33 457 .full_fds_bits = init_files.full_fds_bits_init,
f52111b1 458 },
eece09ec 459 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
f52111b1 460};
1027abe8 461
9b80a184 462static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
f3f86e33 463{
9b80a184
AD
464 unsigned int maxfd = fdt->max_fds;
465 unsigned int maxbit = maxfd / BITS_PER_LONG;
466 unsigned int bitbit = start / BITS_PER_LONG;
f3f86e33
LT
467
468 bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
469 if (bitbit > maxfd)
470 return maxfd;
471 if (bitbit > start)
472 start = bitbit;
473 return find_next_zero_bit(fdt->open_fds, maxfd, start);
474}
475
1027abe8
AV
476/*
477 * allocate a file descriptor, mark it busy.
478 */
dcfadfa4
AV
479int __alloc_fd(struct files_struct *files,
480 unsigned start, unsigned end, unsigned flags)
1027abe8 481{
1027abe8
AV
482 unsigned int fd;
483 int error;
484 struct fdtable *fdt;
485
486 spin_lock(&files->file_lock);
487repeat:
488 fdt = files_fdtable(files);
489 fd = start;
490 if (fd < files->next_fd)
491 fd = files->next_fd;
492
493 if (fd < fdt->max_fds)
f3f86e33 494 fd = find_next_fd(fdt, fd);
1027abe8 495
f33ff992
AV
496 /*
497 * N.B. For clone tasks sharing a files structure, this test
498 * will limit the total number of files that can be opened.
499 */
500 error = -EMFILE;
501 if (fd >= end)
502 goto out;
503
1027abe8
AV
504 error = expand_files(files, fd);
505 if (error < 0)
506 goto out;
507
508 /*
509 * If we needed to expand the fs array we
510 * might have blocked - try again.
511 */
512 if (error)
513 goto repeat;
514
515 if (start <= files->next_fd)
516 files->next_fd = fd + 1;
517
1dce27c5 518 __set_open_fd(fd, fdt);
1027abe8 519 if (flags & O_CLOEXEC)
1dce27c5 520 __set_close_on_exec(fd, fdt);
1027abe8 521 else
1dce27c5 522 __clear_close_on_exec(fd, fdt);
1027abe8
AV
523 error = fd;
524#if 1
525 /* Sanity check */
add1f099 526 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
1027abe8
AV
527 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
528 rcu_assign_pointer(fdt->fd[fd], NULL);
529 }
530#endif
531
532out:
533 spin_unlock(&files->file_lock);
534 return error;
535}
536
ad47bd72 537static int alloc_fd(unsigned start, unsigned flags)
dcfadfa4
AV
538{
539 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
540}
541
1a7bd226 542int get_unused_fd_flags(unsigned flags)
1027abe8 543{
dcfadfa4 544 return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags);
1027abe8 545}
1a7bd226 546EXPORT_SYMBOL(get_unused_fd_flags);
56007cae
AV
547
548static void __put_unused_fd(struct files_struct *files, unsigned int fd)
549{
550 struct fdtable *fdt = files_fdtable(files);
551 __clear_open_fd(fd, fdt);
552 if (fd < files->next_fd)
553 files->next_fd = fd;
554}
555
556void put_unused_fd(unsigned int fd)
557{
558 struct files_struct *files = current->files;
559 spin_lock(&files->file_lock);
560 __put_unused_fd(files, fd);
561 spin_unlock(&files->file_lock);
562}
563
564EXPORT_SYMBOL(put_unused_fd);
565
566/*
567 * Install a file pointer in the fd array.
568 *
569 * The VFS is full of places where we drop the files lock between
570 * setting the open_fds bitmap and installing the file in the file
571 * array. At any such point, we are vulnerable to a dup2() race
572 * installing a file in the array before us. We need to detect this and
573 * fput() the struct file we are about to overwrite in this case.
574 *
575 * It should never happen - if we allow dup2() do it, _really_ bad things
576 * will follow.
f869e8a7
AV
577 *
578 * NOTE: __fd_install() variant is really, really low-level; don't
579 * use it unless you are forced to by truly lousy API shoved down
580 * your throat. 'files' *MUST* be either current->files or obtained
581 * by get_files_struct(current) done by whoever had given it to you,
582 * or really bad things will happen. Normally you want to use
583 * fd_install() instead.
56007cae
AV
584 */
585
f869e8a7
AV
586void __fd_install(struct files_struct *files, unsigned int fd,
587 struct file *file)
56007cae 588{
56007cae 589 struct fdtable *fdt;
8a81252b 590
8a81252b
ED
591 rcu_read_lock_sched();
592
c02b1a9b 593 if (unlikely(files->resize_in_progress)) {
8a81252b 594 rcu_read_unlock_sched();
c02b1a9b
MG
595 spin_lock(&files->file_lock);
596 fdt = files_fdtable(files);
597 BUG_ON(fdt->fd[fd] != NULL);
598 rcu_assign_pointer(fdt->fd[fd], file);
599 spin_unlock(&files->file_lock);
600 return;
8a81252b
ED
601 }
602 /* coupled with smp_wmb() in expand_fdtable() */
603 smp_rmb();
604 fdt = rcu_dereference_sched(files->fdt);
56007cae
AV
605 BUG_ON(fdt->fd[fd] != NULL);
606 rcu_assign_pointer(fdt->fd[fd], file);
8a81252b 607 rcu_read_unlock_sched();
56007cae
AV
608}
609
f869e8a7
AV
610void fd_install(unsigned int fd, struct file *file)
611{
612 __fd_install(current->files, fd, file);
613}
614
56007cae 615EXPORT_SYMBOL(fd_install);
0ee8cdfe 616
483ce1d4
AV
617/*
618 * The same warnings as for __alloc_fd()/__fd_install() apply here...
619 */
620int __close_fd(struct files_struct *files, unsigned fd)
621{
622 struct file *file;
623 struct fdtable *fdt;
624
625 spin_lock(&files->file_lock);
626 fdt = files_fdtable(files);
627 if (fd >= fdt->max_fds)
628 goto out_unlock;
629 file = fdt->fd[fd];
630 if (!file)
631 goto out_unlock;
632 rcu_assign_pointer(fdt->fd[fd], NULL);
483ce1d4
AV
633 __put_unused_fd(files, fd);
634 spin_unlock(&files->file_lock);
635 return filp_close(file, files);
636
637out_unlock:
638 spin_unlock(&files->file_lock);
639 return -EBADF;
640}
2ca2a09d 641EXPORT_SYMBOL(__close_fd); /* for ksys_close() */
483ce1d4 642
6a6d27de
AV
643void do_close_on_exec(struct files_struct *files)
644{
645 unsigned i;
646 struct fdtable *fdt;
647
648 /* exec unshares first */
6a6d27de
AV
649 spin_lock(&files->file_lock);
650 for (i = 0; ; i++) {
651 unsigned long set;
652 unsigned fd = i * BITS_PER_LONG;
653 fdt = files_fdtable(files);
654 if (fd >= fdt->max_fds)
655 break;
656 set = fdt->close_on_exec[i];
657 if (!set)
658 continue;
659 fdt->close_on_exec[i] = 0;
660 for ( ; set ; fd++, set >>= 1) {
661 struct file *file;
662 if (!(set & 1))
663 continue;
664 file = fdt->fd[fd];
665 if (!file)
666 continue;
667 rcu_assign_pointer(fdt->fd[fd], NULL);
668 __put_unused_fd(files, fd);
669 spin_unlock(&files->file_lock);
670 filp_close(file, files);
671 cond_resched();
672 spin_lock(&files->file_lock);
673 }
674
675 }
676 spin_unlock(&files->file_lock);
677}
678
1deb46e2 679static struct file *__fget(unsigned int fd, fmode_t mask)
0ee8cdfe 680{
0ee8cdfe 681 struct files_struct *files = current->files;
1deb46e2 682 struct file *file;
0ee8cdfe
AV
683
684 rcu_read_lock();
5ba97d28 685loop:
0ee8cdfe
AV
686 file = fcheck_files(files, fd);
687 if (file) {
5ba97d28
ED
688 /* File object ref couldn't be taken.
689 * dup2() atomicity guarantee is the reason
690 * we loop to catch the new file (or NULL pointer)
691 */
692 if (file->f_mode & mask)
0ee8cdfe 693 file = NULL;
5ba97d28
ED
694 else if (!get_file_rcu(file))
695 goto loop;
0ee8cdfe
AV
696 }
697 rcu_read_unlock();
698
699 return file;
700}
701
1deb46e2
ON
702struct file *fget(unsigned int fd)
703{
704 return __fget(fd, FMODE_PATH);
705}
0ee8cdfe
AV
706EXPORT_SYMBOL(fget);
707
708struct file *fget_raw(unsigned int fd)
709{
1deb46e2 710 return __fget(fd, 0);
0ee8cdfe 711}
0ee8cdfe
AV
712EXPORT_SYMBOL(fget_raw);
713
714/*
715 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
716 *
717 * You can use this instead of fget if you satisfy all of the following
718 * conditions:
719 * 1) You must call fput_light before exiting the syscall and returning control
720 * to userspace (i.e. you cannot remember the returned struct file * after
721 * returning to userspace).
722 * 2) You must not call filp_close on the returned struct file * in between
723 * calls to fget_light and fput_light.
724 * 3) You must not clone the current task in between the calls to fget_light
725 * and fput_light.
726 *
727 * The fput_needed flag returned by fget_light should be passed to the
728 * corresponding fput_light.
729 */
bd2a31d5 730static unsigned long __fget_light(unsigned int fd, fmode_t mask)
0ee8cdfe 731{
0ee8cdfe 732 struct files_struct *files = current->files;
ad461834 733 struct file *file;
0ee8cdfe 734
0ee8cdfe 735 if (atomic_read(&files->count) == 1) {
a8d4b834 736 file = __fcheck_files(files, fd);
bd2a31d5
AV
737 if (!file || unlikely(file->f_mode & mask))
738 return 0;
739 return (unsigned long)file;
0ee8cdfe 740 } else {
e6ff9a9f 741 file = __fget(fd, mask);
bd2a31d5
AV
742 if (!file)
743 return 0;
744 return FDPUT_FPUT | (unsigned long)file;
0ee8cdfe 745 }
0ee8cdfe 746}
bd2a31d5 747unsigned long __fdget(unsigned int fd)
ad461834 748{
bd2a31d5 749 return __fget_light(fd, FMODE_PATH);
ad461834 750}
bd2a31d5 751EXPORT_SYMBOL(__fdget);
0ee8cdfe 752
bd2a31d5 753unsigned long __fdget_raw(unsigned int fd)
0ee8cdfe 754{
bd2a31d5 755 return __fget_light(fd, 0);
0ee8cdfe 756}
fe17f22d 757
bd2a31d5
AV
758unsigned long __fdget_pos(unsigned int fd)
759{
99aea681
EB
760 unsigned long v = __fdget(fd);
761 struct file *file = (struct file *)(v & ~3);
bd2a31d5 762
99aea681 763 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
bd2a31d5
AV
764 if (file_count(file) > 1) {
765 v |= FDPUT_POS_UNLOCK;
766 mutex_lock(&file->f_pos_lock);
767 }
768 }
99aea681 769 return v;
bd2a31d5
AV
770}
771
63b6df14
AV
772void __f_unlock_pos(struct file *f)
773{
774 mutex_unlock(&f->f_pos_lock);
775}
776
bd2a31d5
AV
777/*
778 * We only lock f_pos if we have threads or if the file might be
779 * shared with another process. In both cases we'll have an elevated
780 * file count (done either by fdget() or by fork()).
781 */
782
fe17f22d
AV
783void set_close_on_exec(unsigned int fd, int flag)
784{
785 struct files_struct *files = current->files;
786 struct fdtable *fdt;
787 spin_lock(&files->file_lock);
788 fdt = files_fdtable(files);
789 if (flag)
790 __set_close_on_exec(fd, fdt);
791 else
792 __clear_close_on_exec(fd, fdt);
793 spin_unlock(&files->file_lock);
794}
795
796bool get_close_on_exec(unsigned int fd)
797{
798 struct files_struct *files = current->files;
799 struct fdtable *fdt;
800 bool res;
801 rcu_read_lock();
802 fdt = files_fdtable(files);
803 res = close_on_exec(fd, fdt);
804 rcu_read_unlock();
805 return res;
806}
807
8280d161
AV
808static int do_dup2(struct files_struct *files,
809 struct file *file, unsigned fd, unsigned flags)
e983094d 810__releases(&files->file_lock)
fe17f22d 811{
8280d161 812 struct file *tofree;
fe17f22d
AV
813 struct fdtable *fdt;
814
fe17f22d
AV
815 /*
816 * We need to detect attempts to do dup2() over allocated but still
817 * not finished descriptor. NB: OpenBSD avoids that at the price of
818 * extra work in their equivalent of fget() - they insert struct
819 * file immediately after grabbing descriptor, mark it larval if
820 * more work (e.g. actual opening) is needed and make sure that
821 * fget() treats larval files as absent. Potentially interesting,
822 * but while extra work in fget() is trivial, locking implications
823 * and amount of surgery on open()-related paths in VFS are not.
824 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
825 * deadlocks in rather amusing ways, AFAICS. All of that is out of
826 * scope of POSIX or SUS, since neither considers shared descriptor
827 * tables and this condition does not arise without those.
828 */
fe17f22d 829 fdt = files_fdtable(files);
8280d161
AV
830 tofree = fdt->fd[fd];
831 if (!tofree && fd_is_open(fd, fdt))
832 goto Ebusy;
fe17f22d 833 get_file(file);
8280d161
AV
834 rcu_assign_pointer(fdt->fd[fd], file);
835 __set_open_fd(fd, fdt);
fe17f22d 836 if (flags & O_CLOEXEC)
8280d161 837 __set_close_on_exec(fd, fdt);
fe17f22d 838 else
8280d161 839 __clear_close_on_exec(fd, fdt);
fe17f22d
AV
840 spin_unlock(&files->file_lock);
841
842 if (tofree)
843 filp_close(tofree, files);
844
8280d161
AV
845 return fd;
846
847Ebusy:
848 spin_unlock(&files->file_lock);
849 return -EBUSY;
850}
851
852int replace_fd(unsigned fd, struct file *file, unsigned flags)
853{
854 int err;
855 struct files_struct *files = current->files;
856
857 if (!file)
858 return __close_fd(files, fd);
859
860 if (fd >= rlimit(RLIMIT_NOFILE))
08f05c49 861 return -EBADF;
8280d161
AV
862
863 spin_lock(&files->file_lock);
864 err = expand_files(files, fd);
865 if (unlikely(err < 0))
866 goto out_unlock;
867 return do_dup2(files, file, fd, flags);
868
869out_unlock:
870 spin_unlock(&files->file_lock);
871 return err;
872}
873
c7248321 874static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
8280d161
AV
875{
876 int err = -EBADF;
877 struct file *file;
878 struct files_struct *files = current->files;
879
880 if ((flags & ~O_CLOEXEC) != 0)
881 return -EINVAL;
882
aed97647
RJ
883 if (unlikely(oldfd == newfd))
884 return -EINVAL;
885
8280d161 886 if (newfd >= rlimit(RLIMIT_NOFILE))
08f05c49 887 return -EBADF;
8280d161
AV
888
889 spin_lock(&files->file_lock);
890 err = expand_files(files, newfd);
891 file = fcheck(oldfd);
892 if (unlikely(!file))
893 goto Ebadf;
894 if (unlikely(err < 0)) {
895 if (err == -EMFILE)
896 goto Ebadf;
897 goto out_unlock;
898 }
899 return do_dup2(files, file, newfd, flags);
fe17f22d
AV
900
901Ebadf:
902 err = -EBADF;
903out_unlock:
904 spin_unlock(&files->file_lock);
905 return err;
906}
907
c7248321
DB
908SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
909{
910 return ksys_dup3(oldfd, newfd, flags);
911}
912
fe17f22d
AV
913SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
914{
915 if (unlikely(newfd == oldfd)) { /* corner case */
916 struct files_struct *files = current->files;
917 int retval = oldfd;
918
919 rcu_read_lock();
920 if (!fcheck_files(files, oldfd))
921 retval = -EBADF;
922 rcu_read_unlock();
923 return retval;
924 }
c7248321 925 return ksys_dup3(oldfd, newfd, 0);
fe17f22d
AV
926}
927
c7248321 928int ksys_dup(unsigned int fildes)
fe17f22d
AV
929{
930 int ret = -EBADF;
931 struct file *file = fget_raw(fildes);
932
933 if (file) {
8d10a035 934 ret = get_unused_fd_flags(0);
fe17f22d
AV
935 if (ret >= 0)
936 fd_install(ret, file);
937 else
938 fput(file);
939 }
940 return ret;
941}
942
c7248321
DB
943SYSCALL_DEFINE1(dup, unsigned int, fildes)
944{
945 return ksys_dup(fildes);
946}
947
fe17f22d
AV
948int f_dupfd(unsigned int from, struct file *file, unsigned flags)
949{
950 int err;
951 if (from >= rlimit(RLIMIT_NOFILE))
952 return -EINVAL;
953 err = alloc_fd(from, flags);
954 if (err >= 0) {
955 get_file(file);
956 fd_install(err, file);
957 }
958 return err;
959}
c3c073f8
AV
960
961int iterate_fd(struct files_struct *files, unsigned n,
962 int (*f)(const void *, struct file *, unsigned),
963 const void *p)
964{
965 struct fdtable *fdt;
c3c073f8
AV
966 int res = 0;
967 if (!files)
968 return 0;
969 spin_lock(&files->file_lock);
a77cfcb4
AV
970 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
971 struct file *file;
972 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
973 if (!file)
974 continue;
975 res = f(p, file, n);
976 if (res)
977 break;
c3c073f8
AV
978 }
979 spin_unlock(&files->file_lock);
980 return res;
981}
982EXPORT_SYMBOL(iterate_fd);