Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * linux/fs/file.c
4 : *
5 : * Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes
6 : *
7 : * Manage the dynamic fd arrays in the process files_struct.
8 : */
9 :
10 : #include <linux/syscalls.h>
11 : #include <linux/export.h>
12 : #include <linux/fs.h>
13 : #include <linux/kernel.h>
14 : #include <linux/mm.h>
15 : #include <linux/sched/signal.h>
16 : #include <linux/slab.h>
17 : #include <linux/file.h>
18 : #include <linux/fdtable.h>
19 : #include <linux/bitops.h>
20 : #include <linux/spinlock.h>
21 : #include <linux/rcupdate.h>
22 : #include <linux/close_range.h>
23 : #include <net/sock.h>
24 :
25 : #include "internal.h"
26 :
27 : unsigned int sysctl_nr_open __read_mostly = 1024*1024;
28 : unsigned int sysctl_nr_open_min = BITS_PER_LONG;
29 : /* our min() is unusable in constant expressions ;-/ */
30 : #define __const_min(x, y) ((x) < (y) ? (x) : (y))
31 : unsigned int sysctl_nr_open_max =
32 : __const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG;
33 :
34 3102758 : static void __free_fdtable(struct fdtable *fdt)
35 : {
36 3102758 : kvfree(fdt->fd);
37 3102624 : kvfree(fdt->open_fds);
38 3102312 : kfree(fdt);
39 3102117 : }
40 :
41 5346 : static void free_fdtable_rcu(struct rcu_head *rcu)
42 : {
43 5346 : __free_fdtable(container_of(rcu, struct fdtable, rcu));
44 5346 : }
45 :
46 : #define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr))
47 : #define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
48 :
49 : /*
50 : * Copy 'count' fd bits from the old table to the new table and clear the extra
51 : * space if any. This does not copy the file pointers. Called with the files
52 : * spinlock held for write.
53 : */
54 20815035 : static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
55 : unsigned int count)
56 : {
57 20815035 : unsigned int cpy, set;
58 :
59 20815035 : cpy = count / BITS_PER_BYTE;
60 20815035 : set = (nfdt->max_fds - count) / BITS_PER_BYTE;
61 41630070 : memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
62 20815035 : memset((char *)nfdt->open_fds + cpy, 0, set);
63 41630070 : memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
64 20815035 : memset((char *)nfdt->close_on_exec + cpy, 0, set);
65 :
66 20815035 : cpy = BITBIT_SIZE(count);
67 20815035 : set = BITBIT_SIZE(nfdt->max_fds) - cpy;
68 41630070 : memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
69 20815035 : memset((char *)nfdt->full_fds_bits + cpy, 0, set);
70 20815035 : }
71 :
72 : /*
73 : * Copy all file descriptors from the old table to the new, expanded table and
74 : * clear the extra space. Called with the files spinlock held for write.
75 : */
76 8885 : static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
77 : {
78 8885 : size_t cpy, set;
79 :
80 8885 : BUG_ON(nfdt->max_fds < ofdt->max_fds);
81 :
82 8885 : cpy = ofdt->max_fds * sizeof(struct file *);
83 8885 : set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
84 17770 : memcpy(nfdt->fd, ofdt->fd, cpy);
85 8885 : memset((char *)nfdt->fd + cpy, 0, set);
86 :
87 8885 : copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
88 8885 : }
89 :
90 : /*
91 : * Note how the fdtable bitmap allocations very much have to be a multiple of
92 : * BITS_PER_LONG. This is not only because we walk those things in chunks of
93 : * 'unsigned long' in some places, but simply because that is how the Linux
94 : * kernel bitmaps are defined to work: they are not "bits in an array of bytes",
95 : * they are very much "bits in an array of unsigned long".
96 : *
97 : * The ALIGN(nr, BITS_PER_LONG) here is for clarity: since we just multiplied
98 : * by that "1024/sizeof(ptr)" before, we already know there are sufficient
99 : * clear low bits. Clang seems to realize that, gcc ends up being confused.
100 : *
101 : * On a 128-bit machine, the ALIGN() would actually matter. In the meantime,
102 : * let's consider it documentation (and maybe a test-case for gcc to improve
103 : * its code generation ;)
104 : */
105 3103073 : static struct fdtable * alloc_fdtable(unsigned int nr)
106 : {
107 3103073 : struct fdtable *fdt;
108 3103073 : void *data;
109 :
110 : /*
111 : * Figure out how many fds we actually want to support in this fdtable.
112 : * Allocation steps are keyed to the size of the fdarray, since it
113 : * grows far faster than any of the other dynamic data. We try to fit
114 : * the fdarray into comfortable page-tuned chunks: starting at 1024B
115 : * and growing in powers of two from there on.
116 : */
117 3103073 : nr /= (1024 / sizeof(struct file *));
118 6199398 : nr = roundup_pow_of_two(nr + 1);
119 3103073 : nr *= (1024 / sizeof(struct file *));
120 3103073 : nr = ALIGN(nr, BITS_PER_LONG);
121 : /*
122 : * Note that this can drive nr *below* what we had passed if sysctl_nr_open
123 : * had been set lower between the check in expand_files() and here. Deal
124 : * with that in caller, it's cheaper that way.
125 : *
126 : * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
127 : * bitmaps handling below becomes unpleasant, to put it mildly...
128 : */
129 3103073 : if (unlikely(nr > sysctl_nr_open))
130 0 : nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
131 :
132 3103073 : fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
133 3103075 : if (!fdt)
134 0 : goto out;
135 3103075 : fdt->max_fds = nr;
136 3103075 : data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
137 3103075 : if (!data)
138 0 : goto out_fdt;
139 3103075 : fdt->fd = data;
140 :
141 3103075 : data = kvmalloc(max_t(size_t,
142 : 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
143 : GFP_KERNEL_ACCOUNT);
144 3103075 : if (!data)
145 0 : goto out_arr;
146 3103075 : fdt->open_fds = data;
147 3103075 : data += nr / BITS_PER_BYTE;
148 3103075 : fdt->close_on_exec = data;
149 3103075 : data += nr / BITS_PER_BYTE;
150 3103075 : fdt->full_fds_bits = data;
151 :
152 3103075 : return fdt;
153 :
154 : out_arr:
155 0 : kvfree(fdt->fd);
156 0 : out_fdt:
157 0 : kfree(fdt);
158 : out:
159 : return NULL;
160 : }
161 :
162 : /*
163 : * Expand the file descriptor table.
164 : * This function will allocate a new fdtable and both fd array and fdset, of
165 : * the given size.
166 : * Return <0 error code on error; 1 on successful completion.
167 : * The files->file_lock should be held on entry, and will be held on exit.
168 : */
169 8885 : static int expand_fdtable(struct files_struct *files, unsigned int nr)
170 : __releases(files->file_lock)
171 : __acquires(files->file_lock)
172 : {
173 8885 : struct fdtable *new_fdt, *cur_fdt;
174 :
175 8885 : spin_unlock(&files->file_lock);
176 8885 : new_fdt = alloc_fdtable(nr);
177 :
178 : /* make sure all fd_install() have seen resize_in_progress
179 : * or have finished their rcu_read_lock_sched() section.
180 : */
181 8885 : if (atomic_read(&files->count) > 1)
182 6 : synchronize_rcu();
183 :
184 8885 : spin_lock(&files->file_lock);
185 8885 : if (!new_fdt)
186 : return -ENOMEM;
187 : /*
188 : * extremely unlikely race - sysctl_nr_open decreased between the check in
189 : * caller and alloc_fdtable(). Cheaper to catch it here...
190 : */
191 8885 : if (unlikely(new_fdt->max_fds <= nr)) {
192 0 : __free_fdtable(new_fdt);
193 0 : return -EMFILE;
194 : }
195 8885 : cur_fdt = files_fdtable(files);
196 8885 : BUG_ON(nr < cur_fdt->max_fds);
197 8885 : copy_fdtable(new_fdt, cur_fdt);
198 8885 : rcu_assign_pointer(files->fdt, new_fdt);
199 8885 : if (cur_fdt != &files->fdtab)
200 5346 : call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
201 : /* coupled with smp_rmb() in fd_install() */
202 8885 : smp_wmb();
203 8885 : return 1;
204 : }
205 :
206 : /*
207 : * Expand files.
208 : * This function will expand the file structures, if the requested size exceeds
209 : * the current capacity and there is room for expansion.
210 : * Return <0 error code on error; 0 when nothing done; 1 when files were
211 : * expanded and execution may have blocked.
212 : * The files->file_lock should be held on entry, and will be held on exit.
213 : */
214 760373963 : static int expand_files(struct files_struct *files, unsigned int nr)
215 : __releases(files->file_lock)
216 : __acquires(files->file_lock)
217 : {
218 760373963 : struct fdtable *fdt;
219 760373963 : int expanded = 0;
220 :
221 760374145 : repeat:
222 760374145 : fdt = files_fdtable(files);
223 :
224 : /* Do we need to expand? */
225 760374145 : if (nr < fdt->max_fds)
226 760365078 : return expanded;
227 :
228 : /* Can we expand? */
229 9067 : if (nr >= sysctl_nr_open)
230 : return -EMFILE;
231 :
232 9067 : if (unlikely(files->resize_in_progress)) {
233 182 : spin_unlock(&files->file_lock);
234 182 : expanded = 1;
235 364 : wait_event(files->resize_wait, !files->resize_in_progress);
236 182 : spin_lock(&files->file_lock);
237 182 : goto repeat;
238 : }
239 :
240 : /* All good, so we try */
241 8885 : files->resize_in_progress = true;
242 8885 : expanded = expand_fdtable(files, nr);
243 8885 : files->resize_in_progress = false;
244 :
245 8885 : wake_up_all(&files->resize_wait);
246 8885 : return expanded;
247 : }
248 :
249 200341473 : static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
250 : {
251 200341473 : __set_bit(fd, fdt->close_on_exec);
252 200341473 : }
253 :
254 577060295 : static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
255 : {
256 577060295 : if (test_bit(fd, fdt->close_on_exec))
257 22781506 : __clear_bit(fd, fdt->close_on_exec);
258 577060295 : }
259 :
260 760378112 : static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
261 : {
262 760378112 : __set_bit(fd, fdt->open_fds);
263 760378112 : fd /= BITS_PER_LONG;
264 760378112 : if (!~fdt->open_fds[fd])
265 2703521 : __set_bit(fd, fdt->full_fds_bits);
266 760378112 : }
267 :
268 2594175764 : static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
269 : {
270 2594175764 : __clear_bit(fd, fdt->open_fds);
271 2594175764 : __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
272 2594175764 : }
273 :
274 23899784 : static unsigned int count_open_files(struct fdtable *fdt)
275 : {
276 23899784 : unsigned int size = fdt->max_fds;
277 23899784 : unsigned int i;
278 :
279 : /* Find the last open fd */
280 60155688 : for (i = size / BITS_PER_LONG; i > 0; ) {
281 60115006 : if (fdt->open_fds[--i])
282 : break;
283 : }
284 23899784 : i = (i + 1) * BITS_PER_LONG;
285 23899784 : return i;
286 : }
287 :
288 : /*
289 : * Note that a sane fdtable size always has to be a multiple of
290 : * BITS_PER_LONG, since we have bitmaps that are sized by this.
291 : *
292 : * 'max_fds' will normally already be properly aligned, but it
293 : * turns out that in the close_range() -> __close_range() ->
294 : * unshare_fd() -> dup_fd() -> sane_fdtable_size() we can end
295 : * up having a 'max_fds' value that isn't already aligned.
296 : *
297 : * Rather than make close_range() have to worry about this,
298 : * just make that BITS_PER_LONG alignment be part of a sane
299 : * fdtable size. Becuase that's really what it is.
300 : */
301 : static unsigned int sane_fdtable_size(struct fdtable *fdt, unsigned int max_fds)
302 : {
303 23899130 : unsigned int count;
304 :
305 23899130 : count = count_open_files(fdt);
306 23899130 : if (max_fds < NR_OPEN_DEFAULT)
307 : max_fds = NR_OPEN_DEFAULT;
308 23899130 : return ALIGN(min(count, max_fds), BITS_PER_LONG);
309 : }
310 :
311 : /*
312 : * Allocate a new files structure and copy contents from the
313 : * passed in files structure.
314 : * errorp will be valid only when the returned files_struct is NULL.
315 : */
316 20805994 : struct files_struct *dup_fd(struct files_struct *oldf, unsigned int max_fds, int *errorp)
317 : {
318 20805994 : struct files_struct *newf;
319 20805994 : struct file **old_fds, **new_fds;
320 20805994 : unsigned int open_files, i;
321 20805994 : struct fdtable *old_fdt, *new_fdt;
322 :
323 20805994 : *errorp = -ENOMEM;
324 20805994 : newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
325 20809947 : if (!newf)
326 0 : goto out;
327 :
328 20809947 : atomic_set(&newf->count, 1);
329 :
330 20809947 : spin_lock_init(&newf->file_lock);
331 20806884 : newf->resize_in_progress = false;
332 20806884 : init_waitqueue_head(&newf->resize_wait);
333 20807068 : newf->next_fd = 0;
334 20807068 : new_fdt = &newf->fdtab;
335 20807068 : new_fdt->max_fds = NR_OPEN_DEFAULT;
336 20807068 : new_fdt->close_on_exec = newf->close_on_exec_init;
337 20807068 : new_fdt->open_fds = newf->open_fds_init;
338 20807068 : new_fdt->full_fds_bits = newf->full_fds_bits_init;
339 20807068 : new_fdt->fd = &newf->fd_array[0];
340 :
341 20807068 : spin_lock(&oldf->file_lock);
342 20804940 : old_fdt = files_fdtable(oldf);
343 20804940 : open_files = sane_fdtable_size(old_fdt, max_fds);
344 :
345 : /*
346 : * Check whether we need to allocate a larger fd array and fd set.
347 : */
348 23899130 : while (unlikely(open_files > new_fdt->max_fds)) {
349 3094188 : spin_unlock(&oldf->file_lock);
350 :
351 3094189 : if (new_fdt != &newf->fdtab)
352 0 : __free_fdtable(new_fdt);
353 :
354 3094189 : new_fdt = alloc_fdtable(open_files - 1);
355 3094189 : if (!new_fdt) {
356 0 : *errorp = -ENOMEM;
357 0 : goto out_release;
358 : }
359 :
360 : /* beyond sysctl_nr_open; nothing to do */
361 3094189 : if (unlikely(new_fdt->max_fds < open_files)) {
362 0 : __free_fdtable(new_fdt);
363 0 : *errorp = -EMFILE;
364 0 : goto out_release;
365 : }
366 :
367 : /*
368 : * Reacquire the oldf lock and a pointer to its fd table
369 : * who knows it may have a new bigger fd table. We need
370 : * the latest pointer.
371 : */
372 3094189 : spin_lock(&oldf->file_lock);
373 3094190 : old_fdt = files_fdtable(oldf);
374 3094190 : open_files = sane_fdtable_size(old_fdt, max_fds);
375 : }
376 :
377 20804942 : copy_fd_bitmaps(new_fdt, old_fdt, open_files);
378 :
379 20802355 : old_fds = old_fdt->fd;
380 20802355 : new_fds = new_fdt->fd;
381 :
382 1945248795 : for (i = open_files; i != 0; i--) {
383 1924435213 : struct file *f = *old_fds++;
384 1924435213 : if (f) {
385 110305111 : get_file(f);
386 : } else {
387 : /*
388 : * The fd may be claimed in the fd bitmap but not yet
389 : * instantiated in the files array if a sibling thread
390 : * is partway through open(). So make sure that this
391 : * fd is available to the new process.
392 : */
393 1814130102 : __clear_open_fd(open_files - i, new_fdt);
394 : }
395 1924437124 : rcu_assign_pointer(*new_fds++, f);
396 : }
397 20813582 : spin_unlock(&oldf->file_lock);
398 :
399 : /* clear the remainder */
400 20815152 : memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
401 :
402 20815152 : rcu_assign_pointer(newf->fdt, new_fdt);
403 :
404 20815152 : return newf;
405 :
406 0 : out_release:
407 0 : kmem_cache_free(files_cachep, newf);
408 : out:
409 : return NULL;
410 : }
411 :
412 20815549 : static struct fdtable *close_files(struct files_struct * files)
413 : {
414 : /*
415 : * It is safe to dereference the fd table without RCU or
416 : * ->file_lock because this is the last reference to the
417 : * files structure.
418 : */
419 20815549 : struct fdtable *fdt = rcu_dereference_raw(files->fdt);
420 20815549 : unsigned int i, j = 0;
421 :
422 50937953 : for (;;) {
423 50937953 : unsigned long set;
424 50937953 : i = j * BITS_PER_LONG;
425 50937953 : if (i >= fdt->max_fds)
426 : break;
427 30123051 : set = fdt->open_fds[j++];
428 121855124 : while (set) {
429 91732720 : if (set & 1) {
430 62443486 : struct file * file = xchg(&fdt->fd[i], NULL);
431 62443554 : if (file) {
432 62442586 : filp_close(file, files);
433 62441724 : cond_resched();
434 : }
435 : }
436 91732073 : i++;
437 91732073 : set >>= 1;
438 : }
439 : }
440 :
441 20814902 : return fdt;
442 : }
443 :
444 30162294 : void put_files_struct(struct files_struct *files)
445 : {
446 60329764 : if (atomic_dec_and_test(&files->count)) {
447 20815561 : struct fdtable *fdt = close_files(files);
448 :
449 : /* free the arrays if they are not embedded */
450 20814840 : if (fdt != &files->fdtab)
451 3097426 : __free_fdtable(fdt);
452 20814026 : kmem_cache_free(files_cachep, files);
453 : }
454 30164818 : }
455 :
456 30162160 : void exit_files(struct task_struct *tsk)
457 : {
458 30162160 : struct files_struct * files = tsk->files;
459 :
460 30162160 : if (files) {
461 30162160 : task_lock(tsk);
462 30166174 : tsk->files = NULL;
463 30166174 : task_unlock(tsk);
464 30161120 : put_files_struct(files);
465 : }
466 30161204 : }
467 :
468 : struct files_struct init_files = {
469 : .count = ATOMIC_INIT(1),
470 : .fdt = &init_files.fdtab,
471 : .fdtab = {
472 : .max_fds = NR_OPEN_DEFAULT,
473 : .fd = &init_files.fd_array[0],
474 : .close_on_exec = init_files.close_on_exec_init,
475 : .open_fds = init_files.open_fds_init,
476 : .full_fds_bits = init_files.full_fds_bits_init,
477 : },
478 : .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
479 : .resize_wait = __WAIT_QUEUE_HEAD_INITIALIZER(init_files.resize_wait),
480 : };
481 :
482 731518426 : static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
483 : {
484 731518426 : unsigned int maxfd = fdt->max_fds;
485 731518426 : unsigned int maxbit = maxfd / BITS_PER_LONG;
486 731518426 : unsigned int bitbit = start / BITS_PER_LONG;
487 :
488 731518426 : bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
489 731526320 : if (bitbit > maxfd)
490 : return maxfd;
491 731521241 : if (bitbit > start)
492 : start = bitbit;
493 731521241 : return find_next_zero_bit(fdt->open_fds, maxfd, start);
494 : }
495 :
496 : /*
497 : * allocate a file descriptor, mark it busy.
498 : */
499 731488383 : static int alloc_fd(unsigned start, unsigned end, unsigned flags)
500 : {
501 731488383 : struct files_struct *files = current->files;
502 731488383 : unsigned int fd;
503 731488383 : int error;
504 731488383 : struct fdtable *fdt;
505 :
506 731488383 : spin_lock(&files->file_lock);
507 731524784 : repeat:
508 731524784 : fdt = files_fdtable(files);
509 731524784 : fd = start;
510 731524784 : if (fd < files->next_fd)
511 : fd = files->next_fd;
512 :
513 731524784 : if (fd < fdt->max_fds)
514 731501085 : fd = find_next_fd(fdt, fd);
515 :
516 : /*
517 : * N.B. For clone tasks sharing a files structure, this test
518 : * will limit the total number of files that can be opened.
519 : */
520 731517499 : error = -EMFILE;
521 731517499 : if (fd >= end)
522 20 : goto out;
523 :
524 731517479 : error = expand_files(files, fd);
525 731516821 : if (error < 0)
526 0 : goto out;
527 :
528 : /*
529 : * If we needed to expand the fs array we
530 : * might have blocked - try again.
531 : */
532 731516821 : if (error)
533 2578 : goto repeat;
534 :
535 731514243 : if (start <= files->next_fd)
536 727760458 : files->next_fd = fd + 1;
537 :
538 731514243 : __set_open_fd(fd, fdt);
539 731525592 : if (flags & O_CLOEXEC)
540 183400030 : __set_close_on_exec(fd, fdt);
541 : else
542 548125562 : __clear_close_on_exec(fd, fdt);
543 731511606 : error = fd;
544 : #if 1
545 : /* Sanity check */
546 731511606 : if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
547 0 : printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
548 0 : rcu_assign_pointer(fdt->fd[fd], NULL);
549 : }
550 : #endif
551 :
552 731511606 : out:
553 731511626 : spin_unlock(&files->file_lock);
554 731495123 : return error;
555 : }
556 :
557 0 : int __get_unused_fd_flags(unsigned flags, unsigned long nofile)
558 : {
559 727432165 : return alloc_fd(0, nofile, flags);
560 : }
561 :
562 727432165 : int get_unused_fd_flags(unsigned flags)
563 : {
564 727432165 : return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE));
565 : }
566 : EXPORT_SYMBOL(get_unused_fd_flags);
567 :
568 780148786 : static void __put_unused_fd(struct files_struct *files, unsigned int fd)
569 : {
570 780148786 : struct fdtable *fdt = files_fdtable(files);
571 780148786 : __clear_open_fd(fd, fdt);
572 780152954 : if (fd < files->next_fd)
573 424791407 : files->next_fd = fd;
574 780152954 : }
575 :
576 53011678 : void put_unused_fd(unsigned int fd)
577 : {
578 53011678 : struct files_struct *files = current->files;
579 53011678 : spin_lock(&files->file_lock);
580 53012295 : __put_unused_fd(files, fd);
581 53010205 : spin_unlock(&files->file_lock);
582 53010132 : }
583 :
584 : EXPORT_SYMBOL(put_unused_fd);
585 :
586 : /*
587 : * Install a file pointer in the fd array.
588 : *
589 : * The VFS is full of places where we drop the files lock between
590 : * setting the open_fds bitmap and installing the file in the file
591 : * array. At any such point, we are vulnerable to a dup2() race
592 : * installing a file in the array before us. We need to detect this and
593 : * fput() the struct file we are about to overwrite in this case.
594 : *
595 : * It should never happen - if we allow dup2() do it, _really_ bad things
596 : * will follow.
597 : *
598 : * This consumes the "file" refcount, so callers should treat it
599 : * as if they had called fput(file).
600 : */
601 :
602 678545308 : void fd_install(unsigned int fd, struct file *file)
603 : {
604 678545308 : struct files_struct *files = current->files;
605 678545308 : struct fdtable *fdt;
606 :
607 678545308 : rcu_read_lock_sched();
608 :
609 678574373 : if (unlikely(files->resize_in_progress)) {
610 586 : rcu_read_unlock_sched();
611 586 : spin_lock(&files->file_lock);
612 586 : fdt = files_fdtable(files);
613 586 : BUG_ON(fdt->fd[fd] != NULL);
614 586 : rcu_assign_pointer(fdt->fd[fd], file);
615 586 : spin_unlock(&files->file_lock);
616 586 : return;
617 : }
618 : /* coupled with smp_wmb() in expand_fdtable() */
619 678573787 : smp_rmb();
620 678587529 : fdt = rcu_dereference_sched(files->fdt);
621 678587529 : BUG_ON(fdt->fd[fd] != NULL);
622 678587529 : rcu_assign_pointer(fdt->fd[fd], file);
623 678586067 : rcu_read_unlock_sched();
624 : }
625 :
626 : EXPORT_SYMBOL(fd_install);
627 :
628 : /**
629 : * pick_file - return file associatd with fd
630 : * @files: file struct to retrieve file from
631 : * @fd: file descriptor to retrieve file for
632 : *
633 : * Context: files_lock must be held.
634 : *
635 : * Returns: The file associated with @fd (NULL if @fd is not open)
636 : */
637 736373657 : static struct file *pick_file(struct files_struct *files, unsigned fd)
638 : {
639 736373657 : struct fdtable *fdt = files_fdtable(files);
640 736373657 : struct file *file;
641 :
642 736373657 : if (fd >= fdt->max_fds)
643 : return NULL;
644 :
645 734776181 : fd = array_index_nospec(fd, fdt->max_fds);
646 734767210 : file = fdt->fd[fd];
647 734767210 : if (file) {
648 724854264 : rcu_assign_pointer(fdt->fd[fd], NULL);
649 724854264 : __put_unused_fd(files, fd);
650 : }
651 : return file;
652 : }
653 :
654 733335612 : int close_fd(unsigned fd)
655 : {
656 733335612 : struct files_struct *files = current->files;
657 733335612 : struct file *file;
658 :
659 733335612 : spin_lock(&files->file_lock);
660 733363555 : file = pick_file(files, fd);
661 733335657 : spin_unlock(&files->file_lock);
662 733338033 : if (!file)
663 : return -EBADF;
664 :
665 724205653 : return filp_close(file, files);
666 : }
667 : EXPORT_SYMBOL(close_fd); /* for ksys_close() */
668 :
669 : /**
670 : * last_fd - return last valid index into fd table
671 : * @cur_fds: files struct
672 : *
673 : * Context: Either rcu read lock or files_lock must be held.
674 : *
675 : * Returns: Last valid index into fdtable.
676 : */
677 : static inline unsigned last_fd(struct fdtable *fdt)
678 : {
679 43913 : return fdt->max_fds - 1;
680 : }
681 :
682 0 : static inline void __range_cloexec(struct files_struct *cur_fds,
683 : unsigned int fd, unsigned int max_fd)
684 : {
685 0 : struct fdtable *fdt;
686 :
687 : /* make sure we're using the correct maximum value */
688 0 : spin_lock(&cur_fds->file_lock);
689 0 : fdt = files_fdtable(cur_fds);
690 0 : max_fd = min(last_fd(fdt), max_fd);
691 0 : if (fd <= max_fd)
692 0 : bitmap_set(fdt->close_on_exec, fd, max_fd - fd + 1);
693 0 : spin_unlock(&cur_fds->file_lock);
694 0 : }
695 :
696 43913 : static inline void __range_close(struct files_struct *cur_fds, unsigned int fd,
697 : unsigned int max_fd)
698 : {
699 43913 : unsigned n;
700 :
701 43913 : rcu_read_lock();
702 43913 : n = last_fd(files_fdtable(cur_fds));
703 43913 : rcu_read_unlock();
704 43913 : max_fd = min(max_fd, n);
705 :
706 3062686 : while (fd <= max_fd) {
707 3018773 : struct file *file;
708 :
709 3018773 : spin_lock(&cur_fds->file_lock);
710 3018775 : file = pick_file(cur_fds, fd++);
711 3018775 : spin_unlock(&cur_fds->file_lock);
712 :
713 3018778 : if (file) {
714 : /* found a valid file to close */
715 640752 : filp_close(file, cur_fds);
716 640750 : cond_resched();
717 : }
718 : }
719 43913 : }
720 :
721 : /**
722 : * __close_range() - Close all file descriptors in a given range.
723 : *
724 : * @fd: starting file descriptor to close
725 : * @max_fd: last file descriptor to close
726 : *
727 : * This closes a range of file descriptors. All file descriptors
728 : * from @fd up to and including @max_fd are closed.
729 : */
730 43911 : int __close_range(unsigned fd, unsigned max_fd, unsigned int flags)
731 : {
732 43911 : struct task_struct *me = current;
733 43911 : struct files_struct *cur_fds = me->files, *fds = NULL;
734 :
735 43911 : if (flags & ~(CLOSE_RANGE_UNSHARE | CLOSE_RANGE_CLOEXEC))
736 : return -EINVAL;
737 :
738 43911 : if (fd > max_fd)
739 : return -EINVAL;
740 :
741 43911 : if (flags & CLOSE_RANGE_UNSHARE) {
742 0 : int ret;
743 0 : unsigned int max_unshare_fds = NR_OPEN_MAX;
744 :
745 : /*
746 : * If the caller requested all fds to be made cloexec we always
747 : * copy all of the file descriptors since they still want to
748 : * use them.
749 : */
750 0 : if (!(flags & CLOSE_RANGE_CLOEXEC)) {
751 : /*
752 : * If the requested range is greater than the current
753 : * maximum, we're closing everything so only copy all
754 : * file descriptors beneath the lowest file descriptor.
755 : */
756 0 : rcu_read_lock();
757 0 : if (max_fd >= last_fd(files_fdtable(cur_fds)))
758 0 : max_unshare_fds = fd;
759 0 : rcu_read_unlock();
760 : }
761 :
762 0 : ret = unshare_fd(CLONE_FILES, max_unshare_fds, &fds);
763 0 : if (ret)
764 : return ret;
765 :
766 : /*
767 : * We used to share our file descriptor table, and have now
768 : * created a private one, make sure we're using it below.
769 : */
770 0 : if (fds)
771 0 : swap(cur_fds, fds);
772 : }
773 :
774 43911 : if (flags & CLOSE_RANGE_CLOEXEC)
775 0 : __range_cloexec(cur_fds, fd, max_fd);
776 : else
777 43911 : __range_close(cur_fds, fd, max_fd);
778 :
779 43913 : if (fds) {
780 : /*
781 : * We're done closing the files we were supposed to. Time to install
782 : * the new file descriptor table and drop the old one.
783 : */
784 0 : task_lock(me);
785 0 : me->files = cur_fds;
786 0 : task_unlock(me);
787 0 : put_files_struct(fds);
788 : }
789 :
790 : return 0;
791 : }
792 :
793 : /*
794 : * See close_fd_get_file() below, this variant assumes current->files->file_lock
795 : * is held.
796 : */
797 0 : struct file *__close_fd_get_file(unsigned int fd)
798 : {
799 0 : return pick_file(current->files, fd);
800 : }
801 :
802 : /*
803 : * variant of close_fd that gets a ref on the file for later fput.
804 : * The caller must ensure that filp_close() called on the file.
805 : */
806 0 : struct file *close_fd_get_file(unsigned int fd)
807 : {
808 0 : struct files_struct *files = current->files;
809 0 : struct file *file;
810 :
811 0 : spin_lock(&files->file_lock);
812 0 : file = pick_file(files, fd);
813 0 : spin_unlock(&files->file_lock);
814 :
815 0 : return file;
816 : }
817 :
818 15475057 : void do_close_on_exec(struct files_struct *files)
819 : {
820 15475057 : unsigned i;
821 15475057 : struct fdtable *fdt;
822 :
823 : /* exec unshares first */
824 15475057 : spin_lock(&files->file_lock);
825 37807564 : for (i = 0; ; i++) {
826 37798518 : unsigned long set;
827 37798518 : unsigned fd = i * BITS_PER_LONG;
828 37798518 : fdt = files_fdtable(files);
829 37798518 : if (fd >= fdt->max_fds)
830 : break;
831 22327957 : set = fdt->close_on_exec[i];
832 22327957 : if (!set)
833 4883127 : continue;
834 17444830 : fdt->close_on_exec[i] = 0;
835 404564576 : for ( ; set ; fd++, set >>= 1) {
836 387115196 : struct file *file;
837 387115196 : if (!(set & 1))
838 294638560 : continue;
839 92476636 : file = fdt->fd[fd];
840 92476636 : if (!file)
841 90187877 : continue;
842 2288759 : rcu_assign_pointer(fdt->fd[fd], NULL);
843 2288759 : __put_unused_fd(files, fd);
844 2288951 : spin_unlock(&files->file_lock);
845 2288086 : filp_close(file, files);
846 2290773 : cond_resched();
847 2290799 : spin_lock(&files->file_lock);
848 : }
849 :
850 : }
851 15470561 : spin_unlock(&files->file_lock);
852 15464190 : }
853 :
854 682853952 : static inline struct file *__fget_files_rcu(struct files_struct *files,
855 : unsigned int fd, fmode_t mask)
856 : {
857 682853952 : for (;;) {
858 682853952 : struct file *file;
859 682853952 : struct fdtable *fdt = rcu_dereference_raw(files->fdt);
860 682853952 : struct file __rcu **fdentry;
861 :
862 682853952 : if (unlikely(fd >= fdt->max_fds))
863 : return NULL;
864 :
865 682853952 : fdentry = fdt->fd + array_index_nospec(fd, fdt->max_fds);
866 682811295 : file = rcu_dereference_raw(*fdentry);
867 682811295 : if (unlikely(!file))
868 : return NULL;
869 :
870 682811295 : if (unlikely(file->f_mode & mask))
871 : return NULL;
872 :
873 : /*
874 : * Ok, we have a file pointer. However, because we do
875 : * this all locklessly under RCU, we may be racing with
876 : * that file being closed.
877 : *
878 : * Such a race can take two forms:
879 : *
880 : * (a) the file ref already went down to zero,
881 : * and get_file_rcu() fails. Just try again:
882 : */
883 1365774578 : if (unlikely(!get_file_rcu(file)))
884 0 : continue;
885 :
886 : /*
887 : * (b) the file table entry has changed under us.
888 : * Note that we don't need to re-check the 'fdt->fd'
889 : * pointer having changed, because it always goes
890 : * hand-in-hand with 'fdt'.
891 : *
892 : * If so, we need to put our ref and try again.
893 : */
894 682963283 : if (unlikely(rcu_dereference_raw(files->fdt) != fdt) ||
895 682963283 : unlikely(rcu_dereference_raw(*fdentry) != file)) {
896 0 : fput(file);
897 0 : continue;
898 : }
899 :
900 : /*
901 : * Ok, we have a ref to the file, and checked that it
902 : * still exists.
903 : */
904 : return file;
905 : }
906 : }
907 :
908 682837191 : static struct file *__fget_files(struct files_struct *files, unsigned int fd,
909 : fmode_t mask)
910 : {
911 682837191 : struct file *file;
912 :
913 682837191 : rcu_read_lock();
914 682835945 : file = __fget_files_rcu(files, fd, mask);
915 682968284 : rcu_read_unlock();
916 :
917 682967161 : return file;
918 : }
919 :
920 : static inline struct file *__fget(unsigned int fd, fmode_t mask)
921 : {
922 681798320 : return __fget_files(current->files, fd, mask);
923 : }
924 :
925 148856005 : struct file *fget(unsigned int fd)
926 : {
927 148856005 : return __fget(fd, FMODE_PATH);
928 : }
929 : EXPORT_SYMBOL(fget);
930 :
931 864649 : struct file *fget_raw(unsigned int fd)
932 : {
933 864649 : return __fget(fd, 0);
934 : }
935 : EXPORT_SYMBOL(fget_raw);
936 :
937 1047855 : struct file *fget_task(struct task_struct *task, unsigned int fd)
938 : {
939 1047855 : struct file *file = NULL;
940 :
941 1047855 : task_lock(task);
942 1047855 : if (task->files)
943 1047855 : file = __fget_files(task->files, fd, 0);
944 1047855 : task_unlock(task);
945 :
946 1047855 : return file;
947 : }
948 :
949 1047855 : struct file *task_lookup_fd_rcu(struct task_struct *task, unsigned int fd)
950 : {
951 : /* Must be called with rcu_read_lock held */
952 1047855 : struct files_struct *files;
953 1047855 : struct file *file = NULL;
954 :
955 1047855 : task_lock(task);
956 1047855 : files = task->files;
957 1047855 : if (files)
958 1047855 : file = files_lookup_fd_rcu(files, fd);
959 1047855 : task_unlock(task);
960 :
961 1047855 : return file;
962 : }
963 :
964 65654 : struct file *task_lookup_next_fd_rcu(struct task_struct *task, unsigned int *ret_fd)
965 : {
966 : /* Must be called with rcu_read_lock held */
967 65654 : struct files_struct *files;
968 65654 : unsigned int fd = *ret_fd;
969 65654 : struct file *file = NULL;
970 :
971 65654 : task_lock(task);
972 65654 : files = task->files;
973 65654 : if (files) {
974 465938 : for (; fd < files_fdtable(files)->max_fds; fd++) {
975 453376 : file = files_lookup_fd_rcu(files, fd);
976 453376 : if (file)
977 : break;
978 : }
979 : }
980 65654 : task_unlock(task);
981 65654 : *ret_fd = fd;
982 65654 : return file;
983 : }
984 : EXPORT_SYMBOL(task_lookup_next_fd_rcu);
985 :
986 : /*
987 : * Lightweight file lookup - no refcnt increment if fd table isn't shared.
988 : *
989 : * You can use this instead of fget if you satisfy all of the following
990 : * conditions:
991 : * 1) You must call fput_light before exiting the syscall and returning control
992 : * to userspace (i.e. you cannot remember the returned struct file * after
993 : * returning to userspace).
994 : * 2) You must not call filp_close on the returned struct file * in between
995 : * calls to fget_light and fput_light.
996 : * 3) You must not clone the current task in between the calls to fget_light
997 : * and fput_light.
998 : *
999 : * The fput_needed flag returned by fget_light should be passed to the
1000 : * corresponding fput_light.
1001 : */
1002 4509909445 : static unsigned long __fget_light(unsigned int fd, fmode_t mask)
1003 : {
1004 4509909445 : struct files_struct *files = current->files;
1005 4509909445 : struct file *file;
1006 :
1007 : /*
1008 : * If another thread is concurrently calling close_fd() followed
1009 : * by put_files_struct(), we must not observe the old table
1010 : * entry combined with the new refcount - otherwise we could
1011 : * return a file that is concurrently being freed.
1012 : *
1013 : * atomic_read_acquire() pairs with atomic_dec_and_test() in
1014 : * put_files_struct().
1015 : */
1016 4509909445 : if (atomic_read_acquire(&files->count) == 1) {
1017 3977831779 : file = files_lookup_fd_raw(files, fd);
1018 3977605386 : if (!file || unlikely(file->f_mode & mask))
1019 : return 0;
1020 3975291361 : return (unsigned long)file;
1021 : } else {
1022 532077666 : file = __fget(fd, mask);
1023 532188261 : if (!file)
1024 : return 0;
1025 532188261 : return FDPUT_FPUT | (unsigned long)file;
1026 : }
1027 : }
1028 2227571216 : unsigned long __fdget(unsigned int fd)
1029 : {
1030 2227571216 : return __fget_light(fd, FMODE_PATH);
1031 : }
1032 : EXPORT_SYMBOL(__fdget);
1033 :
1034 773054790 : unsigned long __fdget_raw(unsigned int fd)
1035 : {
1036 773054790 : return __fget_light(fd, 0);
1037 : }
1038 :
1039 1511229021 : unsigned long __fdget_pos(unsigned int fd)
1040 : {
1041 1511229021 : unsigned long v = __fdget(fd);
1042 1511327203 : struct file *file = (struct file *)(v & ~3);
1043 :
1044 1511327203 : if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
1045 482032172 : v |= FDPUT_POS_UNLOCK;
1046 482032172 : mutex_lock(&file->f_pos_lock);
1047 : }
1048 1511575224 : return v;
1049 : }
1050 :
1051 482243812 : void __f_unlock_pos(struct file *f)
1052 : {
1053 482243812 : mutex_unlock(&f->f_pos_lock);
1054 482264326 : }
1055 :
1056 : /*
1057 : * We only lock f_pos if we have threads or if the file might be
1058 : * shared with another process. In both cases we'll have an elevated
1059 : * file count (done either by fdget() or by fork()).
1060 : */
1061 :
1062 17015414 : void set_close_on_exec(unsigned int fd, int flag)
1063 : {
1064 17015414 : struct files_struct *files = current->files;
1065 17015414 : struct fdtable *fdt;
1066 17015414 : spin_lock(&files->file_lock);
1067 17015407 : fdt = files_fdtable(files);
1068 17015407 : if (flag)
1069 16938357 : __set_close_on_exec(fd, fdt);
1070 : else
1071 77050 : __clear_close_on_exec(fd, fdt);
1072 17014385 : spin_unlock(&files->file_lock);
1073 17014206 : }
1074 :
1075 13617469 : bool get_close_on_exec(unsigned int fd)
1076 : {
1077 13617469 : struct files_struct *files = current->files;
1078 13617469 : struct fdtable *fdt;
1079 13617469 : bool res;
1080 13617469 : rcu_read_lock();
1081 13615886 : fdt = files_fdtable(files);
1082 13615886 : res = close_on_exec(fd, fdt);
1083 13615320 : rcu_read_unlock();
1084 13615074 : return res;
1085 : }
1086 :
1087 28895826 : static int do_dup2(struct files_struct *files,
1088 : struct file *file, unsigned fd, unsigned flags)
1089 : __releases(&files->file_lock)
1090 : {
1091 28895826 : struct file *tofree;
1092 28895826 : struct fdtable *fdt;
1093 :
1094 : /*
1095 : * We need to detect attempts to do dup2() over allocated but still
1096 : * not finished descriptor. NB: OpenBSD avoids that at the price of
1097 : * extra work in their equivalent of fget() - they insert struct
1098 : * file immediately after grabbing descriptor, mark it larval if
1099 : * more work (e.g. actual opening) is needed and make sure that
1100 : * fget() treats larval files as absent. Potentially interesting,
1101 : * but while extra work in fget() is trivial, locking implications
1102 : * and amount of surgery on open()-related paths in VFS are not.
1103 : * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
1104 : * deadlocks in rather amusing ways, AFAICS. All of that is out of
1105 : * scope of POSIX or SUS, since neither considers shared descriptor
1106 : * tables and this condition does not arise without those.
1107 : */
1108 28895826 : fdt = files_fdtable(files);
1109 28895826 : tofree = fdt->fd[fd];
1110 28895826 : if (!tofree && fd_is_open(fd, fdt))
1111 0 : goto Ebusy;
1112 28895826 : get_file(file);
1113 28898618 : rcu_assign_pointer(fdt->fd[fd], file);
1114 28897572 : __set_open_fd(fd, fdt);
1115 28897528 : if (flags & O_CLOEXEC)
1116 0 : __set_close_on_exec(fd, fdt);
1117 : else
1118 28897528 : __clear_close_on_exec(fd, fdt);
1119 28898026 : spin_unlock(&files->file_lock);
1120 :
1121 28897041 : if (tofree)
1122 28199887 : filp_close(tofree, files);
1123 :
1124 28901937 : return fd;
1125 :
1126 : Ebusy:
1127 0 : spin_unlock(&files->file_lock);
1128 0 : return -EBUSY;
1129 : }
1130 :
1131 0 : int replace_fd(unsigned fd, struct file *file, unsigned flags)
1132 : {
1133 0 : int err;
1134 0 : struct files_struct *files = current->files;
1135 :
1136 0 : if (!file)
1137 0 : return close_fd(fd);
1138 :
1139 0 : if (fd >= rlimit(RLIMIT_NOFILE))
1140 : return -EBADF;
1141 :
1142 0 : spin_lock(&files->file_lock);
1143 0 : err = expand_files(files, fd);
1144 0 : if (unlikely(err < 0))
1145 0 : goto out_unlock;
1146 0 : return do_dup2(files, file, fd, flags);
1147 :
1148 : out_unlock:
1149 0 : spin_unlock(&files->file_lock);
1150 0 : return err;
1151 : }
1152 :
1153 : /**
1154 : * __receive_fd() - Install received file into file descriptor table
1155 : * @file: struct file that was received from another process
1156 : * @ufd: __user pointer to write new fd number to
1157 : * @o_flags: the O_* flags to apply to the new fd entry
1158 : *
1159 : * Installs a received file into the file descriptor table, with appropriate
1160 : * checks and count updates. Optionally writes the fd number to userspace, if
1161 : * @ufd is non-NULL.
1162 : *
1163 : * This helper handles its own reference counting of the incoming
1164 : * struct file.
1165 : *
1166 : * Returns newly install fd or -ve on error.
1167 : */
1168 16203 : int __receive_fd(struct file *file, int __user *ufd, unsigned int o_flags)
1169 : {
1170 16203 : int new_fd;
1171 16203 : int error;
1172 :
1173 16203 : error = security_file_receive(file);
1174 16203 : if (error)
1175 : return error;
1176 :
1177 16203 : new_fd = get_unused_fd_flags(o_flags);
1178 16203 : if (new_fd < 0)
1179 : return new_fd;
1180 :
1181 16203 : if (ufd) {
1182 16203 : error = put_user(new_fd, ufd);
1183 16203 : if (error) {
1184 0 : put_unused_fd(new_fd);
1185 0 : return error;
1186 : }
1187 : }
1188 :
1189 16203 : fd_install(new_fd, get_file(file));
1190 16203 : __receive_sock(file);
1191 16203 : return new_fd;
1192 : }
1193 :
1194 0 : int receive_fd_replace(int new_fd, struct file *file, unsigned int o_flags)
1195 : {
1196 0 : int error;
1197 :
1198 0 : error = security_file_receive(file);
1199 0 : if (error)
1200 : return error;
1201 0 : error = replace_fd(new_fd, file, o_flags);
1202 0 : if (error)
1203 : return error;
1204 0 : __receive_sock(file);
1205 0 : return new_fd;
1206 : }
1207 :
1208 0 : int receive_fd(struct file *file, unsigned int o_flags)
1209 : {
1210 0 : return __receive_fd(file, NULL, o_flags);
1211 : }
1212 : EXPORT_SYMBOL_GPL(receive_fd);
1213 :
1214 28897129 : static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1215 : {
1216 28897129 : int err = -EBADF;
1217 28897129 : struct file *file;
1218 28897129 : struct files_struct *files = current->files;
1219 :
1220 28897129 : if ((flags & ~O_CLOEXEC) != 0)
1221 : return -EINVAL;
1222 :
1223 28897129 : if (unlikely(oldfd == newfd))
1224 : return -EINVAL;
1225 :
1226 28897129 : if (newfd >= rlimit(RLIMIT_NOFILE))
1227 : return -EBADF;
1228 :
1229 28897129 : spin_lock(&files->file_lock);
1230 28899047 : err = expand_files(files, newfd);
1231 28898072 : file = files_lookup_fd_locked(files, oldfd);
1232 28897791 : if (unlikely(!file))
1233 0 : goto Ebadf;
1234 28897791 : if (unlikely(err < 0)) {
1235 0 : if (err == -EMFILE)
1236 0 : goto Ebadf;
1237 0 : goto out_unlock;
1238 : }
1239 28897791 : return do_dup2(files, file, newfd, flags);
1240 :
1241 0 : Ebadf:
1242 : err = -EBADF;
1243 0 : out_unlock:
1244 0 : spin_unlock(&files->file_lock);
1245 0 : return err;
1246 : }
1247 :
1248 57798877 : SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1249 : {
1250 28896356 : return ksys_dup3(oldfd, newfd, flags);
1251 : }
1252 :
1253 0 : SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1254 : {
1255 0 : if (unlikely(newfd == oldfd)) { /* corner case */
1256 0 : struct files_struct *files = current->files;
1257 0 : int retval = oldfd;
1258 :
1259 0 : rcu_read_lock();
1260 0 : if (!files_lookup_fd_rcu(files, oldfd))
1261 0 : retval = -EBADF;
1262 0 : rcu_read_unlock();
1263 0 : return retval;
1264 : }
1265 0 : return ksys_dup3(oldfd, newfd, 0);
1266 : }
1267 :
1268 1696856 : SYSCALL_DEFINE1(dup, unsigned int, fildes)
1269 : {
1270 848429 : int ret = -EBADF;
1271 848429 : struct file *file = fget_raw(fildes);
1272 :
1273 848432 : if (file) {
1274 848432 : ret = get_unused_fd_flags(0);
1275 848426 : if (ret >= 0)
1276 848426 : fd_install(ret, file);
1277 : else
1278 0 : fput(file);
1279 : }
1280 848436 : return ret;
1281 : }
1282 :
1283 4011538 : int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1284 : {
1285 4011538 : unsigned long nofile = rlimit(RLIMIT_NOFILE);
1286 4011538 : int err;
1287 4011538 : if (from >= nofile)
1288 : return -EINVAL;
1289 4011538 : err = alloc_fd(from, nofile, flags);
1290 4005624 : if (err >= 0) {
1291 4005051 : get_file(file);
1292 4005126 : fd_install(err, file);
1293 : }
1294 : return err;
1295 : }
1296 :
1297 0 : int iterate_fd(struct files_struct *files, unsigned n,
1298 : int (*f)(const void *, struct file *, unsigned),
1299 : const void *p)
1300 : {
1301 0 : struct fdtable *fdt;
1302 0 : int res = 0;
1303 0 : if (!files)
1304 : return 0;
1305 0 : spin_lock(&files->file_lock);
1306 0 : for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1307 0 : struct file *file;
1308 0 : file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1309 0 : if (!file)
1310 0 : continue;
1311 0 : res = f(p, file, n);
1312 0 : if (res)
1313 : break;
1314 : }
1315 0 : spin_unlock(&files->file_lock);
1316 0 : return res;
1317 : }
1318 : EXPORT_SYMBOL(iterate_fd);
|
