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 15786080 : static void __free_fdtable(struct fdtable *fdt)
35 : {
36 15786080 : kvfree(fdt->fd);
37 15787538 : kvfree(fdt->open_fds);
38 15787535 : kfree(fdt);
39 15787523 : }
40 :
41 15561 : static void free_fdtable_rcu(struct rcu_head *rcu)
42 : {
43 15561 : __free_fdtable(container_of(rcu, struct fdtable, rcu));
44 15561 : }
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 51874869 : static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
55 : unsigned int count)
56 : {
57 51874869 : unsigned int cpy, set;
58 :
59 51874869 : cpy = count / BITS_PER_BYTE;
60 51874869 : set = (nfdt->max_fds - count) / BITS_PER_BYTE;
61 103749738 : memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
62 51874869 : memset((char *)nfdt->open_fds + cpy, 0, set);
63 103749738 : memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
64 51874869 : memset((char *)nfdt->close_on_exec + cpy, 0, set);
65 :
66 51874869 : cpy = BITBIT_SIZE(count);
67 51874869 : set = BITBIT_SIZE(nfdt->max_fds) - cpy;
68 103749738 : memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
69 51874869 : memset((char *)nfdt->full_fds_bits + cpy, 0, set);
70 51874869 : }
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 25672 : static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
77 : {
78 25672 : size_t cpy, set;
79 :
80 25672 : BUG_ON(nfdt->max_fds < ofdt->max_fds);
81 :
82 25672 : cpy = ofdt->max_fds * sizeof(struct file *);
83 25672 : set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
84 51344 : memcpy(nfdt->fd, ofdt->fd, cpy);
85 25672 : memset((char *)nfdt->fd + cpy, 0, set);
86 :
87 25672 : copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
88 25672 : }
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 15787803 : static struct fdtable * alloc_fdtable(unsigned int nr)
106 : {
107 15787803 : struct fdtable *fdt;
108 15787803 : 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 15787803 : nr /= (1024 / sizeof(struct file *));
118 15787803 : nr = roundup_pow_of_two(nr + 1);
119 15787803 : nr *= (1024 / sizeof(struct file *));
120 15787803 : 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 15787803 : if (unlikely(nr > sysctl_nr_open))
130 0 : nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
131 :
132 15787803 : fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
133 15787803 : if (!fdt)
134 0 : goto out;
135 15787803 : fdt->max_fds = nr;
136 15787803 : data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
137 15787804 : if (!data)
138 0 : goto out_fdt;
139 15787804 : fdt->fd = data;
140 :
141 15787804 : data = kvmalloc(max_t(size_t,
142 : 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
143 : GFP_KERNEL_ACCOUNT);
144 15787803 : if (!data)
145 0 : goto out_arr;
146 15787803 : fdt->open_fds = data;
147 15787803 : data += nr / BITS_PER_BYTE;
148 15787803 : fdt->close_on_exec = data;
149 15787803 : data += nr / BITS_PER_BYTE;
150 15787803 : fdt->full_fds_bits = data;
151 :
152 15787803 : 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 25672 : static int expand_fdtable(struct files_struct *files, unsigned int nr)
170 : __releases(files->file_lock)
171 : __acquires(files->file_lock)
172 : {
173 25672 : struct fdtable *new_fdt, *cur_fdt;
174 :
175 25672 : spin_unlock(&files->file_lock);
176 25672 : 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 25672 : if (atomic_read(&files->count) > 1)
182 72 : synchronize_rcu();
183 :
184 25672 : spin_lock(&files->file_lock);
185 25672 : 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 25672 : if (unlikely(new_fdt->max_fds <= nr)) {
192 0 : __free_fdtable(new_fdt);
193 0 : return -EMFILE;
194 : }
195 25672 : cur_fdt = files_fdtable(files);
196 25672 : BUG_ON(nr < cur_fdt->max_fds);
197 25672 : copy_fdtable(new_fdt, cur_fdt);
198 25672 : rcu_assign_pointer(files->fdt, new_fdt);
199 25672 : if (cur_fdt != &files->fdtab)
200 15565 : call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
201 : /* coupled with smp_rmb() in fd_install() */
202 25672 : smp_wmb();
203 25672 : 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 1402644997 : static int expand_files(struct files_struct *files, unsigned int nr)
215 : __releases(files->file_lock)
216 : __acquires(files->file_lock)
217 : {
218 1402644997 : struct fdtable *fdt;
219 1402644997 : int expanded = 0;
220 :
221 1402649357 : repeat:
222 1402649357 : fdt = files_fdtable(files);
223 :
224 : /* Do we need to expand? */
225 1402649357 : if (nr < fdt->max_fds)
226 1402619325 : return expanded;
227 :
228 : /* Can we expand? */
229 30032 : if (nr >= sysctl_nr_open)
230 : return -EMFILE;
231 :
232 30032 : if (unlikely(files->resize_in_progress)) {
233 4360 : spin_unlock(&files->file_lock);
234 4359 : expanded = 1;
235 8712 : wait_event(files->resize_wait, !files->resize_in_progress);
236 4360 : spin_lock(&files->file_lock);
237 4360 : goto repeat;
238 : }
239 :
240 : /* All good, so we try */
241 25672 : files->resize_in_progress = true;
242 25672 : expanded = expand_fdtable(files, nr);
243 25672 : files->resize_in_progress = false;
244 :
245 25672 : wake_up_all(&files->resize_wait);
246 25672 : return expanded;
247 : }
248 :
249 449395674 : static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
250 : {
251 449395674 : __set_bit(fd, fdt->close_on_exec);
252 449751784 : }
253 :
254 992620907 : static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
255 : {
256 1984450794 : if (test_bit(fd, fdt->close_on_exec))
257 58489404 : __clear_bit(fd, fdt->close_on_exec);
258 991846607 : }
259 :
260 1401645939 : static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
261 : {
262 1401645939 : __set_bit(fd, fdt->open_fds);
263 1401765127 : fd /= BITS_PER_LONG;
264 1401765127 : if (!~fdt->open_fds[fd])
265 4017163 : __set_bit(fd, fdt->full_fds_bits);
266 1401765057 : }
267 :
268 7483074510 : static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
269 : {
270 7483074510 : __clear_bit(fd, fdt->open_fds);
271 7483145108 : __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
272 7483403713 : }
273 :
274 67621735 : static unsigned int count_open_files(struct fdtable *fdt)
275 : {
276 67621735 : unsigned int size = fdt->max_fds;
277 67621735 : unsigned int i;
278 :
279 : /* Find the last open fd */
280 136556214 : for (i = size / BITS_PER_LONG; i > 0; ) {
281 136287883 : if (fdt->open_fds[--i])
282 : break;
283 : }
284 67621735 : i = (i + 1) * BITS_PER_LONG;
285 67621735 : 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 67615201 : unsigned int count;
304 :
305 67615201 : count = count_open_files(fdt);
306 67615201 : if (max_fds < NR_OPEN_DEFAULT)
307 : max_fds = NR_OPEN_DEFAULT;
308 67615201 : 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 51839802 : struct files_struct *dup_fd(struct files_struct *oldf, unsigned int max_fds, int *errorp)
317 : {
318 51839802 : struct files_struct *newf;
319 51839802 : struct file **old_fds, **new_fds;
320 51839802 : unsigned int open_files, i;
321 51839802 : struct fdtable *old_fdt, *new_fdt;
322 :
323 51839802 : *errorp = -ENOMEM;
324 51839802 : newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
325 51853067 : if (!newf)
326 0 : goto out;
327 :
328 51853067 : atomic_set(&newf->count, 1);
329 :
330 51853067 : spin_lock_init(&newf->file_lock);
331 51850438 : newf->resize_in_progress = false;
332 51850438 : init_waitqueue_head(&newf->resize_wait);
333 51846090 : newf->next_fd = 0;
334 51846090 : new_fdt = &newf->fdtab;
335 51846090 : new_fdt->max_fds = NR_OPEN_DEFAULT;
336 51846090 : new_fdt->close_on_exec = newf->close_on_exec_init;
337 51846090 : new_fdt->open_fds = newf->open_fds_init;
338 51846090 : new_fdt->full_fds_bits = newf->full_fds_bits_init;
339 51846090 : new_fdt->fd = &newf->fd_array[0];
340 :
341 51846090 : spin_lock(&oldf->file_lock);
342 51853069 : old_fdt = files_fdtable(oldf);
343 51853069 : 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 67615201 : while (unlikely(open_files > new_fdt->max_fds)) {
349 15762132 : spin_unlock(&oldf->file_lock);
350 :
351 15762132 : if (new_fdt != &newf->fdtab)
352 0 : __free_fdtable(new_fdt);
353 :
354 15762132 : new_fdt = alloc_fdtable(open_files - 1);
355 15762131 : if (!new_fdt) {
356 0 : *errorp = -ENOMEM;
357 0 : goto out_release;
358 : }
359 :
360 : /* beyond sysctl_nr_open; nothing to do */
361 15762131 : 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 15762131 : spin_lock(&oldf->file_lock);
373 15762132 : old_fdt = files_fdtable(oldf);
374 15762132 : open_files = sane_fdtable_size(old_fdt, max_fds);
375 : }
376 :
377 51853069 : copy_fd_bitmaps(new_fdt, old_fdt, open_files);
378 :
379 51843007 : old_fds = old_fdt->fd;
380 51843007 : new_fds = new_fdt->fd;
381 :
382 6393041905 : for (i = open_files; i != 0; i--) {
383 6341178824 : struct file *f = *old_fds++;
384 6341178824 : if (f) {
385 298110065 : 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 6043068759 : __clear_open_fd(open_files - i, new_fdt);
394 : }
395 6341146051 : rcu_assign_pointer(*new_fds++, f);
396 : }
397 51863081 : spin_unlock(&oldf->file_lock);
398 :
399 : /* clear the remainder */
400 51863620 : memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
401 :
402 51863620 : rcu_assign_pointer(newf->fdt, new_fdt);
403 :
404 51863319 : return newf;
405 :
406 0 : out_release:
407 0 : kmem_cache_free(files_cachep, newf);
408 : out:
409 : return NULL;
410 : }
411 :
412 51860867 : 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 51860867 : struct fdtable *fdt = rcu_dereference_raw(files->fdt);
420 51860867 : unsigned int i, j = 0;
421 :
422 151208609 : for (;;) {
423 151208609 : unsigned long set;
424 151208609 : i = j * BITS_PER_LONG;
425 151208609 : if (i >= fdt->max_fds)
426 : break;
427 99347964 : set = fdt->open_fds[j++];
428 352458273 : while (set) {
429 253110531 : if (set & 1) {
430 173336617 : struct file * file = xchg(&fdt->fd[i], NULL);
431 173346465 : if (file) {
432 173346465 : filp_close(file, files);
433 173332630 : cond_resched();
434 : }
435 : }
436 253110309 : i++;
437 253110309 : set >>= 1;
438 : }
439 : }
440 :
441 51860645 : return fdt;
442 : }
443 :
444 93559185 : void put_files_struct(struct files_struct *files)
445 : {
446 93559185 : if (atomic_dec_and_test(&files->count)) {
447 51862596 : struct fdtable *fdt = close_files(files);
448 :
449 : /* free the arrays if they are not embedded */
450 51860938 : if (fdt != &files->fdtab)
451 15771630 : __free_fdtable(fdt);
452 51861287 : kmem_cache_free(files_cachep, files);
453 : }
454 93558477 : }
455 :
456 93572308 : void exit_files(struct task_struct *tsk)
457 : {
458 93572308 : struct files_struct * files = tsk->files;
459 :
460 93572308 : if (files) {
461 93572308 : task_lock(tsk);
462 93554307 : tsk->files = NULL;
463 93554307 : task_unlock(tsk);
464 93595061 : put_files_struct(files);
465 : }
466 93551788 : }
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 1314783710 : static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
483 : {
484 1314783710 : unsigned int maxfd = fdt->max_fds;
485 1314783710 : unsigned int maxbit = maxfd / BITS_PER_LONG;
486 1314783710 : unsigned int bitbit = start / BITS_PER_LONG;
487 :
488 1314783710 : bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
489 1314072626 : if (bitbit > maxfd)
490 : return maxfd;
491 1314204011 : if (bitbit > start)
492 : start = bitbit;
493 1314204011 : return find_next_zero_bit(fdt->open_fds, maxfd, start);
494 : }
495 :
496 : /*
497 : * allocate a file descriptor, mark it busy.
498 : */
499 1312643719 : static int alloc_fd(unsigned start, unsigned end, unsigned flags)
500 : {
501 1312643719 : struct files_struct *files = current->files;
502 1312643719 : unsigned int fd;
503 1312643719 : int error;
504 1312643719 : struct fdtable *fdt;
505 :
506 1312643719 : spin_lock(&files->file_lock);
507 1314654507 : repeat:
508 1314654507 : fdt = files_fdtable(files);
509 1314654507 : fd = start;
510 1314654507 : if (fd < files->next_fd)
511 : fd = files->next_fd;
512 :
513 1314654507 : if (fd < fdt->max_fds)
514 1314794595 : 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 1314035882 : error = -EMFILE;
521 1314035882 : if (fd >= end)
522 215 : goto out;
523 :
524 1314035667 : error = expand_files(files, fd);
525 1313040916 : 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 1313040916 : if (error)
533 13168 : goto repeat;
534 :
535 1313027748 : if (start <= files->next_fd)
536 1297157912 : files->next_fd = fd + 1;
537 :
538 1313027748 : __set_open_fd(fd, fdt);
539 1312679301 : if (flags & O_CLOEXEC)
540 410144476 : __set_close_on_exec(fd, fdt);
541 : else
542 902534825 : __clear_close_on_exec(fd, fdt);
543 1313352585 : error = fd;
544 : #if 1
545 : /* Sanity check */
546 1313352585 : 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 1313352585 : out:
553 1313352800 : spin_unlock(&files->file_lock);
554 1315230410 : return error;
555 : }
556 :
557 0 : int __get_unused_fd_flags(unsigned flags, unsigned long nofile)
558 : {
559 1294889627 : return alloc_fd(0, nofile, flags);
560 : }
561 :
562 1294889627 : int get_unused_fd_flags(unsigned flags)
563 : {
564 1294889627 : return __get_unused_fd_flags(flags, rlimit(RLIMIT_NOFILE));
565 : }
566 : EXPORT_SYMBOL(get_unused_fd_flags);
567 :
568 1440716291 : static void __put_unused_fd(struct files_struct *files, unsigned int fd)
569 : {
570 1440716291 : struct fdtable *fdt = files_fdtable(files);
571 1440716291 : __clear_open_fd(fd, fdt);
572 1440578076 : if (fd < files->next_fd)
573 808632783 : files->next_fd = fd;
574 1440578076 : }
575 :
576 28007647 : void put_unused_fd(unsigned int fd)
577 : {
578 28007647 : struct files_struct *files = current->files;
579 28007647 : spin_lock(&files->file_lock);
580 28021992 : __put_unused_fd(files, fd);
581 28006598 : spin_unlock(&files->file_lock);
582 28031510 : }
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 1284919135 : void fd_install(unsigned int fd, struct file *file)
603 : {
604 1284919135 : struct files_struct *files = current->files;
605 1284919135 : struct fdtable *fdt;
606 :
607 1284919135 : rcu_read_lock_sched();
608 :
609 1285344994 : if (unlikely(files->resize_in_progress)) {
610 6918 : rcu_read_unlock_sched();
611 6918 : spin_lock(&files->file_lock);
612 6918 : fdt = files_fdtable(files);
613 6918 : BUG_ON(fdt->fd[fd] != NULL);
614 6918 : rcu_assign_pointer(fdt->fd[fd], file);
615 6918 : spin_unlock(&files->file_lock);
616 6918 : return;
617 : }
618 : /* coupled with smp_wmb() in expand_fdtable() */
619 1285338076 : smp_rmb();
620 1285806265 : fdt = rcu_dereference_sched(files->fdt);
621 1285806265 : BUG_ON(fdt->fd[fd] != NULL);
622 1285806265 : rcu_assign_pointer(fdt->fd[fd], file);
623 1285806581 : 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 1424957866 : static struct file *pick_file(struct files_struct *files, unsigned fd)
638 : {
639 1424957866 : struct fdtable *fdt = files_fdtable(files);
640 1424957866 : struct file *file;
641 :
642 1424957866 : if (fd >= fdt->max_fds)
643 : return NULL;
644 :
645 1422297097 : fd = array_index_nospec(fd, fdt->max_fds);
646 1422298776 : file = fdt->fd[fd];
647 1422298776 : if (file) {
648 1397638738 : rcu_assign_pointer(fdt->fd[fd], NULL);
649 1397638738 : __put_unused_fd(files, fd);
650 : }
651 : return file;
652 : }
653 :
654 1415264022 : int close_fd(unsigned fd)
655 : {
656 1415264022 : struct files_struct *files = current->files;
657 1415264022 : struct file *file;
658 :
659 1415264022 : spin_lock(&files->file_lock);
660 1417207529 : file = pick_file(files, fd);
661 1416434936 : spin_unlock(&files->file_lock);
662 1417325657 : if (!file)
663 : return -EBADF;
664 :
665 1396317812 : 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 114028 : 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 114028 : static inline void __range_close(struct files_struct *cur_fds, unsigned int fd,
697 : unsigned int max_fd)
698 : {
699 114028 : unsigned n;
700 :
701 114028 : rcu_read_lock();
702 114028 : n = last_fd(files_fdtable(cur_fds));
703 114028 : rcu_read_unlock();
704 114029 : max_fd = min(max_fd, n);
705 :
706 8026995 : while (fd <= max_fd) {
707 7912965 : struct file *file;
708 :
709 7912965 : spin_lock(&cur_fds->file_lock);
710 7912944 : file = pick_file(cur_fds, fd++);
711 7912910 : spin_unlock(&cur_fds->file_lock);
712 :
713 7912982 : if (file) {
714 : /* found a valid file to close */
715 1600225 : filp_close(file, cur_fds);
716 1600227 : cond_resched();
717 : }
718 : }
719 114030 : }
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 114029 : int __close_range(unsigned fd, unsigned max_fd, unsigned int flags)
731 : {
732 114029 : struct task_struct *me = current;
733 114029 : struct files_struct *cur_fds = me->files, *fds = NULL;
734 :
735 114029 : if (flags & ~(CLOSE_RANGE_UNSHARE | CLOSE_RANGE_CLOEXEC))
736 : return -EINVAL;
737 :
738 114029 : if (fd > max_fd)
739 : return -EINVAL;
740 :
741 114029 : 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 114029 : if (flags & CLOSE_RANGE_CLOEXEC)
775 0 : __range_cloexec(cur_fds, fd, max_fd);
776 : else
777 114029 : __range_close(cur_fds, fd, max_fd);
778 :
779 114030 : 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 36949407 : void do_close_on_exec(struct files_struct *files)
819 : {
820 36949407 : unsigned i;
821 36949407 : struct fdtable *fdt;
822 :
823 : /* exec unshares first */
824 36949407 : spin_lock(&files->file_lock);
825 107113690 : for (i = 0; ; i++) {
826 107122785 : unsigned long set;
827 107122785 : unsigned fd = i * BITS_PER_LONG;
828 107122785 : fdt = files_fdtable(files);
829 107122785 : if (fd >= fdt->max_fds)
830 : break;
831 70164432 : set = fdt->close_on_exec[i];
832 70164432 : if (!set)
833 23929160 : continue;
834 46235272 : fdt->close_on_exec[i] = 0;
835 1302983392 : for ( ; set ; fd++, set >>= 1) {
836 1256748269 : struct file *file;
837 1256748269 : if (!(set & 1))
838 1043424131 : continue;
839 213324138 : file = fdt->fd[fd];
840 213324138 : if (!file)
841 197864823 : continue;
842 15459315 : rcu_assign_pointer(fdt->fd[fd], NULL);
843 15459315 : __put_unused_fd(files, fd);
844 15455552 : spin_unlock(&files->file_lock);
845 15459650 : filp_close(file, files);
846 15453208 : cond_resched();
847 15453411 : spin_lock(&files->file_lock);
848 : }
849 :
850 : }
851 36958353 : spin_unlock(&files->file_lock);
852 36953318 : }
853 :
854 2898790901 : static inline struct file *__fget_files_rcu(struct files_struct *files,
855 : unsigned int fd, fmode_t mask)
856 : {
857 2898790901 : for (;;) {
858 2898790901 : struct file *file;
859 2898790901 : struct fdtable *fdt = rcu_dereference_raw(files->fdt);
860 2898790901 : struct file __rcu **fdentry;
861 :
862 2898790901 : if (unlikely(fd >= fdt->max_fds))
863 : return NULL;
864 :
865 2898790901 : fdentry = fdt->fd + array_index_nospec(fd, fdt->max_fds);
866 2902853774 : file = rcu_dereference_raw(*fdentry);
867 2902853774 : if (unlikely(!file))
868 : return NULL;
869 :
870 2902853774 : 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 5822920450 : 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 2920066676 : if (unlikely(rcu_dereference_raw(files->fdt) != fdt) ||
895 2920066676 : 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 2892379063 : static struct file *__fget_files(struct files_struct *files, unsigned int fd,
909 : fmode_t mask)
910 : {
911 2892379063 : struct file *file;
912 :
913 2892379063 : rcu_read_lock();
914 2892981698 : file = __fget_files_rcu(files, fd, mask);
915 2916200202 : rcu_read_unlock();
916 :
917 2917020289 : return file;
918 : }
919 :
920 : static inline struct file *__fget(unsigned int fd, fmode_t mask)
921 : {
922 2889198829 : return __fget_files(current->files, fd, mask);
923 : }
924 :
925 797134498 : struct file *fget(unsigned int fd)
926 : {
927 797134498 : return __fget(fd, FMODE_PATH);
928 : }
929 : EXPORT_SYMBOL(fget);
930 :
931 2209591 : struct file *fget_raw(unsigned int fd)
932 : {
933 2209591 : return __fget(fd, 0);
934 : }
935 : EXPORT_SYMBOL(fget_raw);
936 :
937 2363935 : struct file *fget_task(struct task_struct *task, unsigned int fd)
938 : {
939 2363935 : struct file *file = NULL;
940 :
941 2363935 : task_lock(task);
942 2364544 : if (task->files)
943 2364544 : file = __fget_files(task->files, fd, 0);
944 2365102 : task_unlock(task);
945 :
946 2365234 : return file;
947 : }
948 :
949 2364243 : struct file *task_lookup_fd_rcu(struct task_struct *task, unsigned int fd)
950 : {
951 : /* Must be called with rcu_read_lock held */
952 2364243 : struct files_struct *files;
953 2364243 : struct file *file = NULL;
954 :
955 2364243 : task_lock(task);
956 2364452 : files = task->files;
957 2364452 : if (files)
958 2364452 : file = files_lookup_fd_rcu(files, fd);
959 2364526 : task_unlock(task);
960 :
961 2365087 : return file;
962 : }
963 :
964 429575 : 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 429575 : struct files_struct *files;
968 429575 : unsigned int fd = *ret_fd;
969 429575 : struct file *file = NULL;
970 :
971 429575 : task_lock(task);
972 429575 : files = task->files;
973 429575 : if (files) {
974 2991106 : for (; fd < files_fdtable(files)->max_fds; fd++) {
975 2909248 : file = files_lookup_fd_rcu(files, fd);
976 2909248 : if (file)
977 : break;
978 : }
979 : }
980 429575 : task_unlock(task);
981 429575 : *ret_fd = fd;
982 429575 : 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 13213130631 : static unsigned long __fget_light(unsigned int fd, fmode_t mask)
1003 : {
1004 13213130631 : struct files_struct *files = current->files;
1005 13213130631 : 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 13213130631 : if (atomic_read_acquire(&files->count) == 1) {
1017 11126065927 : file = files_lookup_fd_raw(files, fd);
1018 11128107981 : if (!file || unlikely(file->f_mode & mask))
1019 : return 0;
1020 11124279438 : return (unsigned long)file;
1021 : } else {
1022 2089854740 : file = __fget(fd, mask);
1023 2114875250 : if (!file)
1024 : return 0;
1025 2114875250 : return FDPUT_FPUT | (unsigned long)file;
1026 : }
1027 : }
1028 6939229862 : unsigned long __fdget(unsigned int fd)
1029 : {
1030 6939229862 : return __fget_light(fd, FMODE_PATH);
1031 : }
1032 : EXPORT_SYMBOL(__fdget);
1033 :
1034 2122274335 : unsigned long __fdget_raw(unsigned int fd)
1035 : {
1036 2122274335 : return __fget_light(fd, 0);
1037 : }
1038 :
1039 4161696267 : unsigned long __fdget_pos(unsigned int fd)
1040 : {
1041 4161696267 : unsigned long v = __fdget(fd);
1042 4162480233 : struct file *file = (struct file *)(v & ~3);
1043 :
1044 4162480233 : if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
1045 1721776557 : v |= FDPUT_POS_UNLOCK;
1046 1721776557 : mutex_lock(&file->f_pos_lock);
1047 : }
1048 4162780887 : return v;
1049 : }
1050 :
1051 1721302007 : void __f_unlock_pos(struct file *f)
1052 : {
1053 1721302007 : mutex_unlock(&f->f_pos_lock);
1054 1721935641 : }
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 39970529 : void set_close_on_exec(unsigned int fd, int flag)
1063 : {
1064 39970529 : struct files_struct *files = current->files;
1065 39970529 : struct fdtable *fdt;
1066 39970529 : spin_lock(&files->file_lock);
1067 40081753 : fdt = files_fdtable(files);
1068 40081753 : if (flag)
1069 39596250 : __set_close_on_exec(fd, fdt);
1070 : else
1071 485503 : __clear_close_on_exec(fd, fdt);
1072 40087544 : spin_unlock(&files->file_lock);
1073 40191158 : }
1074 :
1075 55036849 : bool get_close_on_exec(unsigned int fd)
1076 : {
1077 55036849 : struct files_struct *files = current->files;
1078 55036849 : struct fdtable *fdt;
1079 55036849 : bool res;
1080 55036849 : rcu_read_lock();
1081 54989959 : fdt = files_fdtable(files);
1082 54989959 : res = close_on_exec(fd, fdt);
1083 55102454 : rcu_read_unlock();
1084 55145784 : return res;
1085 : }
1086 :
1087 88308272 : static int do_dup2(struct files_struct *files,
1088 : struct file *file, unsigned fd, unsigned flags)
1089 : __releases(&files->file_lock)
1090 : {
1091 88308272 : struct file *tofree;
1092 88308272 : 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 88308272 : fdt = files_fdtable(files);
1109 88308272 : tofree = fdt->fd[fd];
1110 88308272 : if (!tofree && fd_is_open(fd, fdt))
1111 0 : goto Ebusy;
1112 88308272 : get_file(file);
1113 88435338 : rcu_assign_pointer(fdt->fd[fd], file);
1114 88440309 : __set_open_fd(fd, fdt);
1115 88382717 : if (flags & O_CLOEXEC)
1116 0 : __set_close_on_exec(fd, fdt);
1117 : else
1118 88382717 : __clear_close_on_exec(fd, fdt);
1119 88391669 : spin_unlock(&files->file_lock);
1120 :
1121 88362525 : if (tofree)
1122 87256314 : filp_close(tofree, files);
1123 :
1124 88065064 : 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 87382 : int __receive_fd(struct file *file, int __user *ufd, unsigned int o_flags)
1169 : {
1170 87382 : int new_fd;
1171 87382 : int error;
1172 :
1173 87382 : error = security_file_receive(file);
1174 87382 : if (error)
1175 : return error;
1176 :
1177 87382 : new_fd = get_unused_fd_flags(o_flags);
1178 87382 : if (new_fd < 0)
1179 : return new_fd;
1180 :
1181 87382 : if (ufd) {
1182 87382 : error = put_user(new_fd, ufd);
1183 87382 : if (error) {
1184 0 : put_unused_fd(new_fd);
1185 0 : return error;
1186 : }
1187 : }
1188 :
1189 87382 : fd_install(new_fd, get_file(file));
1190 87382 : __receive_sock(file);
1191 87382 : 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 88083333 : static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
1215 : {
1216 88083333 : int err = -EBADF;
1217 88083333 : struct file *file;
1218 88083333 : struct files_struct *files = current->files;
1219 :
1220 88083333 : if ((flags & ~O_CLOEXEC) != 0)
1221 : return -EINVAL;
1222 :
1223 88083333 : if (unlikely(oldfd == newfd))
1224 : return -EINVAL;
1225 :
1226 88083333 : if (newfd >= rlimit(RLIMIT_NOFILE))
1227 : return -EBADF;
1228 :
1229 88083333 : spin_lock(&files->file_lock);
1230 88338754 : err = expand_files(files, newfd);
1231 88306615 : file = files_lookup_fd_locked(files, oldfd);
1232 88341802 : if (unlikely(!file))
1233 0 : goto Ebadf;
1234 88341802 : if (unlikely(err < 0)) {
1235 0 : if (err == -EMFILE)
1236 0 : goto Ebadf;
1237 0 : goto out_unlock;
1238 : }
1239 88341802 : 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 61316 : SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
1249 : {
1250 30658 : return ksys_dup3(oldfd, newfd, flags);
1251 : }
1252 :
1253 176258537 : SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
1254 : {
1255 88116645 : 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 88116645 : return ksys_dup3(oldfd, newfd, 0);
1266 : }
1267 :
1268 4244412 : SYSCALL_DEFINE1(dup, unsigned int, fildes)
1269 : {
1270 2122180 : int ret = -EBADF;
1271 2122180 : struct file *file = fget_raw(fildes);
1272 :
1273 2122410 : if (file) {
1274 2122410 : ret = get_unused_fd_flags(0);
1275 2122243 : if (ret >= 0)
1276 2122243 : fd_install(ret, file);
1277 : else
1278 0 : fput(file);
1279 : }
1280 2122251 : return ret;
1281 : }
1282 :
1283 17992182 : int f_dupfd(unsigned int from, struct file *file, unsigned flags)
1284 : {
1285 17992182 : unsigned long nofile = rlimit(RLIMIT_NOFILE);
1286 17992182 : int err;
1287 17992182 : if (from >= nofile)
1288 : return -EINVAL;
1289 17992182 : err = alloc_fd(from, nofile, flags);
1290 18019411 : if (err >= 0) {
1291 18021873 : get_file(file);
1292 18039258 : 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);
|
