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 3027639 : static void __free_fdtable(struct fdtable *fdt)
35 : {
36 3027639 : kvfree(fdt->fd);
37 3027348 : kvfree(fdt->open_fds);
38 3026927 : kfree(fdt);
39 3026969 : }
40 :
41 4645 : static void free_fdtable_rcu(struct rcu_head *rcu)
42 : {
43 4645 : __free_fdtable(container_of(rcu, struct fdtable, rcu));
44 4648 : }
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 20588668 : static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
55 : unsigned int count)
56 : {
57 20588668 : unsigned int cpy, set;
58 :
59 20588668 : cpy = count / BITS_PER_BYTE;
60 20588668 : set = (nfdt->max_fds - count) / BITS_PER_BYTE;
61 41177336 : memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
62 20588668 : memset((char *)nfdt->open_fds + cpy, 0, set);
63 41177336 : memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
64 20588668 : memset((char *)nfdt->close_on_exec + cpy, 0, set);
65 :
66 20588668 : cpy = BITBIT_SIZE(count);
67 20588668 : set = BITBIT_SIZE(nfdt->max_fds) - cpy;
68 41177336 : memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
69 20588668 : memset((char *)nfdt->full_fds_bits + cpy, 0, set);
70 20588668 : }
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 7844 : static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
77 : {
78 7844 : size_t cpy, set;
79 :
80 7844 : BUG_ON(nfdt->max_fds < ofdt->max_fds);
81 :
82 7844 : cpy = ofdt->max_fds * sizeof(struct file *);
83 7844 : set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
84 15688 : memcpy(nfdt->fd, ofdt->fd, cpy);
85 7844 : memset((char *)nfdt->fd + cpy, 0, set);
86 :
87 7844 : copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
88 7844 : }
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 3028144 : static struct fdtable * alloc_fdtable(unsigned int nr)
106 : {
107 3028144 : struct fdtable *fdt;
108 3028144 : 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 3028144 : nr /= (1024 / sizeof(struct file *));
118 6050214 : nr = roundup_pow_of_two(nr + 1);
119 3028144 : nr *= (1024 / sizeof(struct file *));
120 3028144 : 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 3028144 : if (unlikely(nr > sysctl_nr_open))
130 0 : nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
131 :
132 3028144 : fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
133 3028143 : if (!fdt)
134 0 : goto out;
135 3028143 : fdt->max_fds = nr;
136 3028143 : data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
137 3028144 : if (!data)
138 0 : goto out_fdt;
139 3028144 : fdt->fd = data;
140 :
141 3028144 : data = kvmalloc(max_t(size_t,
142 : 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
143 : GFP_KERNEL_ACCOUNT);
144 3028144 : if (!data)
145 0 : goto out_arr;
146 3028144 : fdt->open_fds = data;
147 3028144 : data += nr / BITS_PER_BYTE;
148 3028144 : fdt->close_on_exec = data;
149 3028144 : data += nr / BITS_PER_BYTE;
150 3028144 : fdt->full_fds_bits = data;
151 :
152 3028144 : 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 7844 : static int expand_fdtable(struct files_struct *files, unsigned int nr)
170 : __releases(files->file_lock)
171 : __acquires(files->file_lock)
172 : {
173 7844 : struct fdtable *new_fdt, *cur_fdt;
174 :
175 7844 : spin_unlock(&files->file_lock);
176 7844 : 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 7844 : if (atomic_read(&files->count) > 1)
182 12 : synchronize_rcu();
183 :
184 7844 : spin_lock(&files->file_lock);
185 7844 : 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 7844 : if (unlikely(new_fdt->max_fds <= nr)) {
192 0 : __free_fdtable(new_fdt);
193 0 : return -EMFILE;
194 : }
195 7844 : cur_fdt = files_fdtable(files);
196 7844 : BUG_ON(nr < cur_fdt->max_fds);
197 7844 : copy_fdtable(new_fdt, cur_fdt);
198 7844 : rcu_assign_pointer(files->fdt, new_fdt);
199 7844 : if (cur_fdt != &files->fdtab)
200 4649 : call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
201 : /* coupled with smp_rmb() in fd_install() */
202 7844 : smp_wmb();
203 7844 : 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 645780402 : static int expand_files(struct files_struct *files, unsigned int nr)
215 : __releases(files->file_lock)
216 : __acquires(files->file_lock)
217 : {
218 645780402 : struct fdtable *fdt;
219 645780402 : int expanded = 0;
220 :
221 645780922 : repeat:
222 645780922 : fdt = files_fdtable(files);
223 :
224 : /* Do we need to expand? */
225 645780922 : if (nr < fdt->max_fds)
226 645772558 : return expanded;
227 :
228 : /* Can we expand? */
229 8364 : if (nr >= sysctl_nr_open)
230 : return -EMFILE;
231 :
232 8364 : if (unlikely(files->resize_in_progress)) {
233 520 : spin_unlock(&files->file_lock);
234 520 : expanded = 1;
235 1038 : wait_event(files->resize_wait, !files->resize_in_progress);
236 520 : spin_lock(&files->file_lock);
237 520 : goto repeat;
238 : }
239 :
240 : /* All good, so we try */
241 7844 : files->resize_in_progress = true;
242 7844 : expanded = expand_fdtable(files, nr);
243 7844 : files->resize_in_progress = false;
244 :
245 7844 : wake_up_all(&files->resize_wait);
246 7844 : return expanded;
247 : }
248 :
249 200775088 : static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
250 : {
251 200775088 : __set_bit(fd, fdt->close_on_exec);
252 200775088 : }
253 :
254 463408059 : static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
255 : {
256 463408059 : if (test_bit(fd, fdt->close_on_exec))
257 23729144 : __clear_bit(fd, fdt->close_on_exec);
258 463408059 : }
259 :
260 645822787 : static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
261 : {
262 645822787 : __set_bit(fd, fdt->open_fds);
263 645822787 : fd /= BITS_PER_LONG;
264 645822787 : if (!~fdt->open_fds[fd])
265 1976370 : __set_bit(fd, fdt->full_fds_bits);
266 645822787 : }
267 :
268 2453438530 : static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
269 : {
270 2453438530 : __clear_bit(fd, fdt->open_fds);
271 2453438530 : __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
272 2453438530 : }
273 :
274 23596418 : static unsigned int count_open_files(struct fdtable *fdt)
275 : {
276 23596418 : unsigned int size = fdt->max_fds;
277 23596418 : unsigned int i;
278 :
279 : /* Find the last open fd */
280 59534348 : for (i = size / BITS_PER_LONG; i > 0; ) {
281 59501321 : if (fdt->open_fds[--i])
282 : break;
283 : }
284 23596418 : i = (i + 1) * BITS_PER_LONG;
285 23596418 : 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 23602867 : unsigned int count;
304 :
305 23602867 : count = count_open_files(fdt);
306 23602867 : if (max_fds < NR_OPEN_DEFAULT)
307 : max_fds = NR_OPEN_DEFAULT;
308 23602867 : 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 20580956 : struct files_struct *dup_fd(struct files_struct *oldf, unsigned int max_fds, int *errorp)
317 : {
318 20580956 : struct files_struct *newf;
319 20580956 : struct file **old_fds, **new_fds;
320 20580956 : unsigned int open_files, i;
321 20580956 : struct fdtable *old_fdt, *new_fdt;
322 :
323 20580956 : *errorp = -ENOMEM;
324 20580956 : newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
325 20598246 : if (!newf)
326 0 : goto out;
327 :
328 20598246 : atomic_set(&newf->count, 1);
329 :
330 20598246 : spin_lock_init(&newf->file_lock);
331 20588682 : newf->resize_in_progress = false;
332 20588682 : init_waitqueue_head(&newf->resize_wait);
333 20584040 : newf->next_fd = 0;
334 20584040 : new_fdt = &newf->fdtab;
335 20584040 : new_fdt->max_fds = NR_OPEN_DEFAULT;
336 20584040 : new_fdt->close_on_exec = newf->close_on_exec_init;
337 20584040 : new_fdt->open_fds = newf->open_fds_init;
338 20584040 : new_fdt->full_fds_bits = newf->full_fds_bits_init;
339 20584040 : new_fdt->fd = &newf->fd_array[0];
340 :
341 20584040 : spin_lock(&oldf->file_lock);
342 20582568 : old_fdt = files_fdtable(oldf);
343 20582568 : 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 23602867 : while (unlikely(open_files > new_fdt->max_fds)) {
349 3020299 : spin_unlock(&oldf->file_lock);
350 :
351 3020297 : if (new_fdt != &newf->fdtab)
352 0 : __free_fdtable(new_fdt);
353 :
354 3020297 : new_fdt = alloc_fdtable(open_files - 1);
355 3020299 : if (!new_fdt) {
356 0 : *errorp = -ENOMEM;
357 0 : goto out_release;
358 : }
359 :
360 : /* beyond sysctl_nr_open; nothing to do */
361 3020299 : 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 3020299 : spin_lock(&oldf->file_lock);
373 3020299 : old_fdt = files_fdtable(oldf);
374 3020299 : open_files = sane_fdtable_size(old_fdt, max_fds);
375 : }
376 :
377 20582568 : copy_fd_bitmaps(new_fdt, old_fdt, open_files);
378 :
379 20575714 : old_fds = old_fdt->fd;
380 20575714 : new_fds = new_fdt->fd;
381 :
382 1916252505 : for (i = open_files; i != 0; i--) {
383 1895657828 : struct file *f = *old_fds++;
384 1895657828 : if (f) {
385 108580537 : 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 1787077291 : __clear_open_fd(open_files - i, new_fdt);
394 : }
395 1895667855 : rcu_assign_pointer(*new_fds++, f);
396 : }
397 20594677 : spin_unlock(&oldf->file_lock);
398 :
399 : /* clear the remainder */
400 20598090 : memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
401 :
402 20598090 : rcu_assign_pointer(newf->fdt, new_fdt);
403 :
404 20598090 : return newf;
405 :
406 0 : out_release:
407 0 : kmem_cache_free(files_cachep, newf);
408 : out:
409 : return NULL;
410 : }
411 :
412 20596886 : 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 20596886 : struct fdtable *fdt = rcu_dereference_raw(files->fdt);
420 20596886 : unsigned int i, j = 0;
421 :
422 50271109 : for (;;) {
423 50271109 : unsigned long set;
424 50271109 : i = j * BITS_PER_LONG;
425 50271109 : if (i >= fdt->max_fds)
426 : break;
427 29679086 : set = fdt->open_fds[j++];
428 120336358 : while (set) {
429 90662135 : if (set & 1) {
430 61032395 : struct file * file = xchg(&fdt->fd[i], NULL);
431 61032870 : if (file) {
432 61033770 : filp_close(file, files);
433 61027791 : cond_resched();
434 : }
435 : }
436 90657272 : i++;
437 90657272 : set >>= 1;
438 : }
439 : }
440 :
441 20592023 : return fdt;
442 : }
443 :
444 33108422 : void put_files_struct(struct files_struct *files)
445 : {
446 66216003 : if (atomic_dec_and_test(&files->count)) {
447 20591002 : struct fdtable *fdt = close_files(files);
448 :
449 : /* free the arrays if they are not embedded */
450 20592174 : if (fdt != &files->fdtab)
451 3022989 : __free_fdtable(fdt);
452 20591229 : kmem_cache_free(files_cachep, files);
453 : }
454 33105759 : }
455 :
456 33106482 : void exit_files(struct task_struct *tsk)
457 : {
458 33106482 : struct files_struct * files = tsk->files;
459 :
460 33106482 : if (files) {
461 33106482 : task_lock(tsk);
462 33108037 : tsk->files = NULL;
463 33108037 : task_unlock(tsk);
464 33095298 : put_files_struct(files);
465 : }
466 33107464 : }
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 618246969 : static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
483 : {
484 618246969 : unsigned int maxfd = fdt->max_fds;
485 618246969 : unsigned int maxbit = maxfd / BITS_PER_LONG;
486 618246969 : unsigned int bitbit = start / BITS_PER_LONG;
487 :
488 618246969 : bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
489 618213827 : if (bitbit > maxfd)
490 : return maxfd;
491 618211054 : if (bitbit > start)
492 : start = bitbit;
493 618211054 : return find_next_zero_bit(fdt->open_fds, maxfd, start);
494 : }
495 :
496 : /*
497 : * allocate a file descriptor, mark it busy.
498 : */
499 618229256 : static int alloc_fd(unsigned start, unsigned end, unsigned flags)
500 : {
501 618229256 : struct files_struct *files = current->files;
502 618229256 : unsigned int fd;
503 618229256 : int error;
504 618229256 : struct fdtable *fdt;
505 :
506 618229256 : spin_lock(&files->file_lock);
507 618221912 : repeat:
508 618221912 : fdt = files_fdtable(files);
509 618221912 : fd = start;
510 618221912 : if (fd < files->next_fd)
511 : fd = files->next_fd;
512 :
513 618221912 : if (fd < fdt->max_fds)
514 618184040 : 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 618189548 : error = -EMFILE;
521 618189548 : if (fd >= end)
522 20 : goto out;
523 :
524 618189528 : error = expand_files(files, fd);
525 618141629 : 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 618141629 : if (error)
533 2664 : goto repeat;
534 :
535 618138965 : if (start <= files->next_fd)
536 614729876 : files->next_fd = fd + 1;
537 :
538 618138965 : __set_open_fd(fd, fdt);
539 618249006 : if (flags & O_CLOEXEC)
540 182572156 : __set_close_on_exec(fd, fdt);
541 : else
542 435676850 : __clear_close_on_exec(fd, fdt);
543 618204217 : error = fd;
544 : #if 1
545 : /* Sanity check */
546 618204217 : 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 618204217 : out:
553 618204237 : spin_unlock(&files->file_lock);
554 618138849 : return error;
555 : }
556 :
557 0 : int __get_unused_fd_flags(unsigned flags, unsigned long nofile)
558 : {
559 614405570 : return alloc_fd(0, nofile, flags);
560 : }
561 :
562 614405570 : int get_unused_fd_flags(unsigned flags)
563 : {
564 614405570 : return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE));
565 : }
566 : EXPORT_SYMBOL(get_unused_fd_flags);
567 :
568 666490125 : static void __put_unused_fd(struct files_struct *files, unsigned int fd)
569 : {
570 666490125 : struct fdtable *fdt = files_fdtable(files);
571 666490125 : __clear_open_fd(fd, fdt);
572 666492122 : if (fd < files->next_fd)
573 390750311 : files->next_fd = fd;
574 666492122 : }
575 :
576 51233639 : void put_unused_fd(unsigned int fd)
577 : {
578 51233639 : struct files_struct *files = current->files;
579 51233639 : spin_lock(&files->file_lock);
580 51238147 : __put_unused_fd(files, fd);
581 51234918 : spin_unlock(&files->file_lock);
582 51233997 : }
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 567080129 : void fd_install(unsigned int fd, struct file *file)
603 : {
604 567080129 : struct files_struct *files = current->files;
605 567080129 : struct fdtable *fdt;
606 :
607 567080129 : rcu_read_lock_sched();
608 :
609 567097717 : if (unlikely(files->resize_in_progress)) {
610 73 : rcu_read_unlock_sched();
611 73 : spin_lock(&files->file_lock);
612 73 : fdt = files_fdtable(files);
613 73 : BUG_ON(fdt->fd[fd] != NULL);
614 73 : rcu_assign_pointer(fdt->fd[fd], file);
615 73 : spin_unlock(&files->file_lock);
616 73 : return;
617 : }
618 : /* coupled with smp_wmb() in expand_fdtable() */
619 567097644 : smp_rmb();
620 567120844 : fdt = rcu_dereference_sched(files->fdt);
621 567120844 : BUG_ON(fdt->fd[fd] != NULL);
622 567120844 : rcu_assign_pointer(fdt->fd[fd], file);
623 567107226 : 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 623992579 : static struct file *pick_file(struct files_struct *files, unsigned fd)
638 : {
639 623992579 : struct fdtable *fdt = files_fdtable(files);
640 623992579 : struct file *file;
641 :
642 623992579 : if (fd >= fdt->max_fds)
643 : return NULL;
644 :
645 622732769 : fd = array_index_nospec(fd, fdt->max_fds);
646 622711875 : file = fdt->fd[fd];
647 622711875 : if (file) {
648 612944876 : rcu_assign_pointer(fdt->fd[fd], NULL);
649 612944876 : __put_unused_fd(files, fd);
650 : }
651 : return file;
652 : }
653 :
654 621134514 : int close_fd(unsigned fd)
655 : {
656 621134514 : struct files_struct *files = current->files;
657 621134514 : struct file *file;
658 :
659 621134514 : spin_lock(&files->file_lock);
660 621134710 : file = pick_file(files, fd);
661 621098043 : spin_unlock(&files->file_lock);
662 621142430 : if (!file)
663 : return -EBADF;
664 :
665 612375049 : 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 41996 : 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 41996 : static inline void __range_close(struct files_struct *cur_fds, unsigned int fd,
697 : unsigned int max_fd)
698 : {
699 41996 : unsigned n;
700 :
701 41996 : rcu_read_lock();
702 41996 : n = last_fd(files_fdtable(cur_fds));
703 41996 : rcu_read_unlock();
704 41996 : max_fd = min(max_fd, n);
705 :
706 2906086 : while (fd <= max_fd) {
707 2864090 : struct file *file;
708 :
709 2864090 : spin_lock(&cur_fds->file_lock);
710 2864118 : file = pick_file(cur_fds, fd++);
711 2864101 : spin_unlock(&cur_fds->file_lock);
712 :
713 2864098 : if (file) {
714 : /* found a valid file to close */
715 604678 : filp_close(file, cur_fds);
716 604679 : cond_resched();
717 : }
718 : }
719 41996 : }
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 41996 : int __close_range(unsigned fd, unsigned max_fd, unsigned int flags)
731 : {
732 41996 : struct task_struct *me = current;
733 41996 : struct files_struct *cur_fds = me->files, *fds = NULL;
734 :
735 41996 : if (flags & ~(CLOSE_RANGE_UNSHARE | CLOSE_RANGE_CLOEXEC))
736 : return -EINVAL;
737 :
738 41996 : if (fd > max_fd)
739 : return -EINVAL;
740 :
741 41996 : 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 41996 : if (flags & CLOSE_RANGE_CLOEXEC)
775 0 : __range_cloexec(cur_fds, fd, max_fd);
776 : else
777 41996 : __range_close(cur_fds, fd, max_fd);
778 :
779 41996 : 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 15460454 : void do_close_on_exec(struct files_struct *files)
819 : {
820 15460454 : unsigned i;
821 15460454 : struct fdtable *fdt;
822 :
823 : /* exec unshares first */
824 15460454 : spin_lock(&files->file_lock);
825 37555862 : for (i = 0; ; i++) {
826 37536172 : unsigned long set;
827 37536172 : unsigned fd = i * BITS_PER_LONG;
828 37536172 : fdt = files_fdtable(files);
829 37536172 : if (fd >= fdt->max_fds)
830 : break;
831 22089965 : set = fdt->close_on_exec[i];
832 22089965 : if (!set)
833 4730578 : continue;
834 17359387 : fdt->close_on_exec[i] = 0;
835 399811832 : for ( ; set ; fd++, set >>= 1) {
836 382447002 : struct file *file;
837 382447002 : if (!(set & 1))
838 290031502 : continue;
839 92415500 : file = fdt->fd[fd];
840 92415500 : if (!file)
841 90112913 : continue;
842 2302587 : rcu_assign_pointer(fdt->fd[fd], NULL);
843 2302587 : __put_unused_fd(files, fd);
844 2302547 : spin_unlock(&files->file_lock);
845 2288918 : filp_close(file, files);
846 2287070 : cond_resched();
847 2287177 : spin_lock(&files->file_lock);
848 : }
849 :
850 : }
851 15446207 : spin_unlock(&files->file_lock);
852 15429671 : }
853 :
854 1692632848 : static inline struct file *__fget_files_rcu(struct files_struct *files,
855 : unsigned int fd, fmode_t mask)
856 : {
857 1692632848 : for (;;) {
858 1692632848 : struct file *file;
859 1692632848 : struct fdtable *fdt = rcu_dereference_raw(files->fdt);
860 1692632848 : struct file __rcu **fdentry;
861 :
862 1692632848 : if (unlikely(fd >= fdt->max_fds))
863 : return NULL;
864 :
865 1692632848 : fdentry = fdt->fd + array_index_nospec(fd, fdt->max_fds);
866 1692456201 : file = rcu_dereference_raw(*fdentry);
867 1692456201 : if (unlikely(!file))
868 : return NULL;
869 :
870 1692456201 : 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 3385710872 : 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 1693254671 : if (unlikely(rcu_dereference_raw(files->fdt) != fdt) ||
895 1693254671 : 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 1692600894 : static struct file *__fget_files(struct files_struct *files, unsigned int fd,
909 : fmode_t mask)
910 : {
911 1692600894 : struct file *file;
912 :
913 1692600894 : rcu_read_lock();
914 1692617543 : file = __fget_files_rcu(files, fd, mask);
915 1693267511 : rcu_read_unlock();
916 :
917 1693252283 : return file;
918 : }
919 :
920 : static inline struct file *__fget(unsigned int fd, fmode_t mask)
921 : {
922 1691692393 : return __fget_files(current->files, fd, mask);
923 : }
924 :
925 140257036 : struct file *fget(unsigned int fd)
926 : {
927 140257036 : return __fget(fd, FMODE_PATH);
928 : }
929 : EXPORT_SYMBOL(fget);
930 :
931 838668 : struct file *fget_raw(unsigned int fd)
932 : {
933 838668 : return __fget(fd, 0);
934 : }
935 : EXPORT_SYMBOL(fget_raw);
936 :
937 946405 : struct file *fget_task(struct task_struct *task, unsigned int fd)
938 : {
939 946405 : struct file *file = NULL;
940 :
941 946405 : task_lock(task);
942 946405 : if (task->files)
943 946405 : file = __fget_files(task->files, fd, 0);
944 946405 : task_unlock(task);
945 :
946 946405 : return file;
947 : }
948 :
949 946405 : struct file *task_lookup_fd_rcu(struct task_struct *task, unsigned int fd)
950 : {
951 : /* Must be called with rcu_read_lock held */
952 946405 : struct files_struct *files;
953 946405 : struct file *file = NULL;
954 :
955 946405 : task_lock(task);
956 946405 : files = task->files;
957 946405 : if (files)
958 946405 : file = files_lookup_fd_rcu(files, fd);
959 946405 : task_unlock(task);
960 :
961 946405 : return file;
962 : }
963 :
964 65644 : 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 65644 : struct files_struct *files;
968 65644 : unsigned int fd = *ret_fd;
969 65644 : struct file *file = NULL;
970 :
971 65644 : task_lock(task);
972 65644 : files = task->files;
973 65644 : if (files) {
974 465292 : for (; fd < files_fdtable(files)->max_fds; fd++) {
975 452736 : file = files_lookup_fd_rcu(files, fd);
976 452736 : if (file)
977 : break;
978 : }
979 : }
980 65644 : task_unlock(task);
981 65644 : *ret_fd = fd;
982 65644 : 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 4945996165 : static unsigned long __fget_light(unsigned int fd, fmode_t mask)
1003 : {
1004 4945996165 : struct files_struct *files = current->files;
1005 4945996165 : 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 4945996165 : if (atomic_read_acquire(&files->count) == 1) {
1017 3395399476 : file = files_lookup_fd_raw(files, fd);
1018 3395247614 : if (!file || unlikely(file->f_mode & mask))
1019 : return 0;
1020 3393383389 : return (unsigned long)file;
1021 : } else {
1022 1550596689 : file = __fget(fd, mask);
1023 1551243326 : if (!file)
1024 : return 0;
1025 1551243326 : return FDPUT_FPUT | (unsigned long)file;
1026 : }
1027 : }
1028 3004472852 : unsigned long __fdget(unsigned int fd)
1029 : {
1030 3004472852 : return __fget_light(fd, FMODE_PATH);
1031 : }
1032 : EXPORT_SYMBOL(__fdget);
1033 :
1034 683925425 : unsigned long __fdget_raw(unsigned int fd)
1035 : {
1036 683925425 : return __fget_light(fd, 0);
1037 : }
1038 :
1039 1259301979 : unsigned long __fdget_pos(unsigned int fd)
1040 : {
1041 1259301979 : unsigned long v = __fdget(fd);
1042 1259377730 : struct file *file = (struct file *)(v & ~3);
1043 :
1044 1259377730 : if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
1045 496928081 : if (file_count(file) > 1) {
1046 188908047 : v |= FDPUT_POS_UNLOCK;
1047 188908047 : mutex_lock(&file->f_pos_lock);
1048 : }
1049 : }
1050 1259556140 : return v;
1051 : }
1052 :
1053 189086677 : void __f_unlock_pos(struct file *f)
1054 : {
1055 189086677 : mutex_unlock(&f->f_pos_lock);
1056 189086544 : }
1057 :
1058 : /*
1059 : * We only lock f_pos if we have threads or if the file might be
1060 : * shared with another process. In both cases we'll have an elevated
1061 : * file count (done either by fdget() or by fork()).
1062 : */
1063 :
1064 18299059 : void set_close_on_exec(unsigned int fd, int flag)
1065 : {
1066 18299059 : struct files_struct *files = current->files;
1067 18299059 : struct fdtable *fdt;
1068 18299059 : spin_lock(&files->file_lock);
1069 18299723 : fdt = files_fdtable(files);
1070 18299723 : if (flag)
1071 18222894 : __set_close_on_exec(fd, fdt);
1072 : else
1073 76829 : __clear_close_on_exec(fd, fdt);
1074 18299671 : spin_unlock(&files->file_lock);
1075 18297558 : }
1076 :
1077 12901870 : bool get_close_on_exec(unsigned int fd)
1078 : {
1079 12901870 : struct files_struct *files = current->files;
1080 12901870 : struct fdtable *fdt;
1081 12901870 : bool res;
1082 12901870 : rcu_read_lock();
1083 12907544 : fdt = files_fdtable(files);
1084 12907544 : res = close_on_exec(fd, fdt);
1085 12908224 : rcu_read_unlock();
1086 12895199 : return res;
1087 : }
1088 :
1089 27659933 : static int do_dup2(struct files_struct *files,
1090 : struct file *file, unsigned fd, unsigned flags)
1091 : __releases(&files->file_lock)
1092 : {
1093 27659933 : struct file *tofree;
1094 27659933 : struct fdtable *fdt;
1095 :
1096 : /*
1097 : * We need to detect attempts to do dup2() over allocated but still
1098 : * not finished descriptor. NB: OpenBSD avoids that at the price of
1099 : * extra work in their equivalent of fget() - they insert struct
1100 : * file immediately after grabbing descriptor, mark it larval if
1101 : * more work (e.g. actual opening) is needed and make sure that
1102 : * fget() treats larval files as absent. Potentially interesting,
1103 : * but while extra work in fget() is trivial, locking implications
1104 : * and amount of surgery on open()-related paths in VFS are not.
1105 : * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
1106 : * deadlocks in rather amusing ways, AFAICS. All of that is out of
1107 : * scope of POSIX or SUS, since neither considers shared descriptor
1108 : * tables and this condition does not arise without those.
1109 : */
1110 27659933 : fdt = files_fdtable(files);
1111 27659933 : tofree = fdt->fd[fd];
1112 27659933 : if (!tofree && fd_is_open(fd, fdt))
1113 0 : goto Ebusy;
1114 27659933 : get_file(file);
1115 27663078 : rcu_assign_pointer(fdt->fd[fd], file);
1116 27661844 : __set_open_fd(fd, fdt);
1117 27660424 : if (flags & O_CLOEXEC)
1118 0 : __set_close_on_exec(fd, fdt);
1119 : else
1120 27660424 : __clear_close_on_exec(fd, fdt);
1121 27661859 : spin_unlock(&files->file_lock);
1122 :
1123 27654870 : if (tofree)
1124 27068864 : filp_close(tofree, files);
1125 :
1126 27671965 : return fd;
1127 :
1128 : Ebusy:
1129 0 : spin_unlock(&files->file_lock);
1130 0 : return -EBUSY;
1131 : }
1132 :
1133 0 : int replace_fd(unsigned fd, struct file *file, unsigned flags)
1134 : {
1135 0 : int err;
1136 0 : struct files_struct *files = current->files;
1137 :
1138 0 : if (!file)
1139 0 : return close_fd(fd);
1140 :
1141 0 : if (fd >= rlimit(RLIMIT_NOFILE))
1142 : return -EBADF;
1143 :
1144 0 : spin_lock(&files->file_lock);
1145 0 : err = expand_files(files, fd);
1146 0 : if (unlikely(err < 0))
1147 0 : goto out_unlock;
1148 0 : return do_dup2(files, file, fd, flags);
1149 :
1150 : out_unlock:
1151 0 : spin_unlock(&files->file_lock);
1152 0 : return err;
1153 : }
1154 :
1155 : /**
1156 : * __receive_fd() - Install received file into file descriptor table
1157 : * @file: struct file that was received from another process
1158 : * @ufd: __user pointer to write new fd number to
1159 : * @o_flags: the O_* flags to apply to the new fd entry
1160 : *
1161 : * Installs a received file into the file descriptor table, with appropriate
1162 : * checks and count updates. Optionally writes the fd number to userspace, if
1163 : * @ufd is non-NULL.
1164 : *
1165 : * This helper handles its own reference counting of the incoming
1166 : * struct file.
1167 : *
1168 : * Returns newly install fd or -ve on error.
1169 : */
1170 15614 : int __receive_fd(struct file *file, int __user *ufd, unsigned int o_flags)
1171 : {
1172 15614 : int new_fd;
1173 15614 : int error;
1174 :
1175 15614 : error = security_file_receive(file);
1176 15614 : if (error)
1177 : return error;
1178 :
1179 15614 : new_fd = get_unused_fd_flags(o_flags);
1180 15614 : if (new_fd < 0)
1181 : return new_fd;
1182 :
1183 15614 : if (ufd) {
1184 15614 : error = put_user(new_fd, ufd);
1185 15614 : if (error) {
1186 0 : put_unused_fd(new_fd);
1187 0 : return error;
1188 : }
1189 : }
1190 :
1191 15614 : fd_install(new_fd, get_file(file));
1192 15614 : __receive_sock(file);
1193 15614 : return new_fd;
1194 : }
1195 :
1196 0 : int receive_fd_replace(int new_fd, struct file *file, unsigned int o_flags)
1197 : {
1198 0 : int error;
1199 :
1200 0 : error = security_file_receive(file);
1201 0 : if (error)
1202 : return error;
1203 0 : error = replace_fd(new_fd, file, o_flags);
1204 0 : if (error)
1205 : return error;
1206 0 : __receive_sock(file);
1207 0 : return new_fd;
1208 : }
1209 :
1210 0 : int receive_fd(struct file *file, unsigned int o_flags)
1211 : {
1212 0 : return __receive_fd(file, NULL, o_flags);
1213 : }
1214 : EXPORT_SYMBOL_GPL(receive_fd);
1215 :
1216 27662202 : static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1217 : {
1218 27662202 : int err = -EBADF;
1219 27662202 : struct file *file;
1220 27662202 : struct files_struct *files = current->files;
1221 :
1222 27662202 : if ((flags & ~O_CLOEXEC) != 0)
1223 : return -EINVAL;
1224 :
1225 27662202 : if (unlikely(oldfd == newfd))
1226 : return -EINVAL;
1227 :
1228 27662202 : if (newfd >= rlimit(RLIMIT_NOFILE))
1229 : return -EBADF;
1230 :
1231 27662202 : spin_lock(&files->file_lock);
1232 27660976 : err = expand_files(files, newfd);
1233 27661683 : file = files_lookup_fd_locked(files, oldfd);
1234 27659634 : if (unlikely(!file))
1235 0 : goto Ebadf;
1236 27659634 : if (unlikely(err < 0)) {
1237 0 : if (err == -EMFILE)
1238 0 : goto Ebadf;
1239 0 : goto out_unlock;
1240 : }
1241 27659634 : return do_dup2(files, file, newfd, flags);
1242 :
1243 0 : Ebadf:
1244 : err = -EBADF;
1245 0 : out_unlock:
1246 0 : spin_unlock(&files->file_lock);
1247 0 : return err;
1248 : }
1249 :
1250 55331793 : SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1251 : {
1252 27659126 : return ksys_dup3(oldfd, newfd, flags);
1253 : }
1254 :
1255 0 : SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1256 : {
1257 0 : if (unlikely(newfd == oldfd)) { /* corner case */
1258 0 : struct files_struct *files = current->files;
1259 0 : int retval = oldfd;
1260 :
1261 0 : rcu_read_lock();
1262 0 : if (!files_lookup_fd_rcu(files, oldfd))
1263 0 : retval = -EBADF;
1264 0 : rcu_read_unlock();
1265 0 : return retval;
1266 : }
1267 0 : return ksys_dup3(oldfd, newfd, 0);
1268 : }
1269 :
1270 1646104 : SYSCALL_DEFINE1(dup, unsigned int, fildes)
1271 : {
1272 823056 : int ret = -EBADF;
1273 823056 : struct file *file = fget_raw(fildes);
1274 :
1275 823063 : if (file) {
1276 823063 : ret = get_unused_fd_flags(0);
1277 823049 : if (ret >= 0)
1278 823049 : fd_install(ret, file);
1279 : else
1280 0 : fput(file);
1281 : }
1282 823052 : return ret;
1283 : }
1284 :
1285 3770935 : int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1286 : {
1287 3770935 : unsigned long nofile = rlimit(RLIMIT_NOFILE);
1288 3770935 : int err;
1289 3770935 : if (from >= nofile)
1290 : return -EINVAL;
1291 3770935 : err = alloc_fd(from, nofile, flags);
1292 3772280 : if (err >= 0) {
1293 3772242 : get_file(file);
1294 3768825 : fd_install(err, file);
1295 : }
1296 : return err;
1297 : }
1298 :
1299 0 : int iterate_fd(struct files_struct *files, unsigned n,
1300 : int (*f)(const void *, struct file *, unsigned),
1301 : const void *p)
1302 : {
1303 0 : struct fdtable *fdt;
1304 0 : int res = 0;
1305 0 : if (!files)
1306 : return 0;
1307 0 : spin_lock(&files->file_lock);
1308 0 : for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
1309 0 : struct file *file;
1310 0 : file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
1311 0 : if (!file)
1312 0 : continue;
1313 0 : res = f(p, file, n);
1314 0 : if (res)
1315 : break;
1316 : }
1317 0 : spin_unlock(&files->file_lock);
1318 0 : return res;
1319 : }
1320 : EXPORT_SYMBOL(iterate_fd);
|
