blob: 7ffd6e9d103d6470cbc2ea0b1618ea2f2cb028c2 [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
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
Al Virofe17f222012-08-21 11:48:11 -040010#include <linux/syscalls.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050011#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/fs.h>
13#include <linux/mm.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010014#include <linux/sched/signal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040017#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/bitops.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070019#include <linux/spinlock.h>
20#include <linux/rcupdate.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070021
Alexey Dobriyan9b80a182016-09-02 00:38:52 +030022unsigned int sysctl_nr_open __read_mostly = 1024*1024;
23unsigned int sysctl_nr_open_min = BITS_PER_LONG;
Rasmus Villemoes752343b2015-10-29 12:01:41 +010024/* our min() is unusable in constant expressions ;-/ */
25#define __const_min(x, y) ((x) < (y) ? (x) : (y))
Alexey Dobriyan9b80a182016-09-02 00:38:52 +030026unsigned int sysctl_nr_open_max =
27 __const_min(INT_MAX, ~(size_t)0/sizeof(void *)) & -BITS_PER_LONG;
Eric Dumazet9cfe0152008-02-06 01:37:16 -080028
Changli Gaoa892e2d2010-08-10 18:01:35 -070029static void __free_fdtable(struct fdtable *fdt)
Vadim Lobanov5466b452006-12-10 02:21:22 -080030{
Al Virof6c0a192014-04-23 10:18:46 -040031 kvfree(fdt->fd);
32 kvfree(fdt->open_fds);
Changli Gaoa892e2d2010-08-10 18:01:35 -070033 kfree(fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070034}
35
Al Viro7cf4dc32012-08-15 19:56:12 -040036static void free_fdtable_rcu(struct rcu_head *rcu)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070037{
Al Viroac3e3c52013-04-28 21:42:33 -040038 __free_fdtable(container_of(rcu, struct fdtable, rcu));
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070039}
40
Linus Torvaldsf3f86e32015-10-30 16:53:57 -070041#define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr))
42#define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
43
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070044/*
Eric Biggersea5c58e2015-11-06 00:32:04 -060045 * Copy 'count' fd bits from the old table to the new table and clear the extra
46 * space if any. This does not copy the file pointers. Called with the files
47 * spinlock held for write.
48 */
49static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
50 unsigned int count)
51{
52 unsigned int cpy, set;
53
54 cpy = count / BITS_PER_BYTE;
55 set = (nfdt->max_fds - count) / BITS_PER_BYTE;
56 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
57 memset((char *)nfdt->open_fds + cpy, 0, set);
58 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
59 memset((char *)nfdt->close_on_exec + cpy, 0, set);
60
61 cpy = BITBIT_SIZE(count);
62 set = BITBIT_SIZE(nfdt->max_fds) - cpy;
63 memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
64 memset((char *)nfdt->full_fds_bits + cpy, 0, set);
65}
66
67/*
68 * Copy all file descriptors from the old table to the new, expanded table and
69 * clear the extra space. Called with the files spinlock held for write.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070070 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080071static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070072{
Vadim Lobanov5466b452006-12-10 02:21:22 -080073 unsigned int cpy, set;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070074
Vadim Lobanov5466b452006-12-10 02:21:22 -080075 BUG_ON(nfdt->max_fds < ofdt->max_fds);
Vadim Lobanov5466b452006-12-10 02:21:22 -080076
77 cpy = ofdt->max_fds * sizeof(struct file *);
78 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
79 memcpy(nfdt->fd, ofdt->fd, cpy);
Eric Biggersea5c58e2015-11-06 00:32:04 -060080 memset((char *)nfdt->fd + cpy, 0, set);
Vadim Lobanov5466b452006-12-10 02:21:22 -080081
Eric Biggersea5c58e2015-11-06 00:32:04 -060082 copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083}
84
Vadim Lobanov5466b452006-12-10 02:21:22 -080085static struct fdtable * alloc_fdtable(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
Vadim Lobanov5466b452006-12-10 02:21:22 -080087 struct fdtable *fdt;
David Howells1fd36ad2012-02-16 17:49:54 +000088 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -070089
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070090 /*
Vadim Lobanov5466b452006-12-10 02:21:22 -080091 * Figure out how many fds we actually want to support in this fdtable.
92 * Allocation steps are keyed to the size of the fdarray, since it
93 * grows far faster than any of the other dynamic data. We try to fit
94 * the fdarray into comfortable page-tuned chunks: starting at 1024B
95 * and growing in powers of two from there on.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070096 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080097 nr /= (1024 / sizeof(struct file *));
98 nr = roundup_pow_of_two(nr + 1);
99 nr *= (1024 / sizeof(struct file *));
Al Viro5c598b32008-04-27 20:04:15 -0400100 /*
101 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
102 * had been set lower between the check in expand_files() and here. Deal
103 * with that in caller, it's cheaper that way.
104 *
105 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
106 * bitmaps handling below becomes unpleasant, to put it mildly...
107 */
108 if (unlikely(nr > sysctl_nr_open))
109 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800110
Vladimir Davydov5d097052016-01-14 15:18:21 -0800111 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL_ACCOUNT);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800112 if (!fdt)
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800113 goto out;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800114 fdt->max_fds = nr;
Michal Hockoc823bd92017-07-06 15:36:19 -0700115 data = kvmalloc_array(nr, sizeof(struct file *), GFP_KERNEL_ACCOUNT);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800116 if (!data)
117 goto out_fdt;
David Howells1fd36ad2012-02-16 17:49:54 +0000118 fdt->fd = data;
119
Michal Hockoc823bd92017-07-06 15:36:19 -0700120 data = kvmalloc(max_t(size_t,
121 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES),
122 GFP_KERNEL_ACCOUNT);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800123 if (!data)
124 goto out_arr;
David Howells1fd36ad2012-02-16 17:49:54 +0000125 fdt->open_fds = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800126 data += nr / BITS_PER_BYTE;
David Howells1fd36ad2012-02-16 17:49:54 +0000127 fdt->close_on_exec = data;
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700128 data += nr / BITS_PER_BYTE;
129 fdt->full_fds_bits = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800130
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700131 return fdt;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800132
133out_arr:
Al Virof6c0a192014-04-23 10:18:46 -0400134 kvfree(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800135out_fdt:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700136 kfree(fdt);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800137out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700138 return NULL;
139}
140
141/*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700142 * Expand the file descriptor table.
143 * This function will allocate a new fdtable and both fd array and fdset, of
144 * the given size.
145 * Return <0 error code on error; 1 on successful completion.
146 * The files->file_lock should be held on entry, and will be held on exit.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700147 */
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300148static int expand_fdtable(struct files_struct *files, unsigned int nr)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700149 __releases(files->file_lock)
150 __acquires(files->file_lock)
151{
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700152 struct fdtable *new_fdt, *cur_fdt;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 spin_unlock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700155 new_fdt = alloc_fdtable(nr);
Eric Dumazet8a812522015-06-30 15:54:08 +0200156
157 /* make sure all __fd_install() have seen resize_in_progress
158 * or have finished their rcu_read_lock_sched() section.
159 */
160 if (atomic_read(&files->count) > 1)
161 synchronize_sched();
162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700164 if (!new_fdt)
165 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700166 /*
Al Viro5c598b32008-04-27 20:04:15 -0400167 * extremely unlikely race - sysctl_nr_open decreased between the check in
168 * caller and alloc_fdtable(). Cheaper to catch it here...
169 */
170 if (unlikely(new_fdt->max_fds <= nr)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700171 __free_fdtable(new_fdt);
Al Viro5c598b32008-04-27 20:04:15 -0400172 return -EMFILE;
173 }
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700174 cur_fdt = files_fdtable(files);
Eric Dumazet8a812522015-06-30 15:54:08 +0200175 BUG_ON(nr < cur_fdt->max_fds);
176 copy_fdtable(new_fdt, cur_fdt);
177 rcu_assign_pointer(files->fdt, new_fdt);
178 if (cur_fdt != &files->fdtab)
179 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
180 /* coupled with smp_rmb() in __fd_install() */
181 smp_wmb();
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700182 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185/*
186 * Expand files.
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700187 * This function will expand the file structures, if the requested size exceeds
188 * the current capacity and there is room for expansion.
189 * Return <0 error code on error; 0 when nothing done; 1 when files were
190 * expanded and execution may have blocked.
191 * The files->file_lock should be held on entry, and will be held on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 */
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300193static int expand_files(struct files_struct *files, unsigned int nr)
Eric Dumazet8a812522015-06-30 15:54:08 +0200194 __releases(files->file_lock)
195 __acquires(files->file_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700197 struct fdtable *fdt;
Eric Dumazet8a812522015-06-30 15:54:08 +0200198 int expanded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
Eric Dumazet8a812522015-06-30 15:54:08 +0200200repeat:
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700201 fdt = files_fdtable(files);
Al Viro4e1e0182008-07-26 16:01:20 -0400202
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700203 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800204 if (nr < fdt->max_fds)
Eric Dumazet8a812522015-06-30 15:54:08 +0200205 return expanded;
Al Viro4e1e0182008-07-26 16:01:20 -0400206
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700207 /* Can we expand? */
Eric Dumazet9cfe0152008-02-06 01:37:16 -0800208 if (nr >= sysctl_nr_open)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700209 return -EMFILE;
210
Eric Dumazet8a812522015-06-30 15:54:08 +0200211 if (unlikely(files->resize_in_progress)) {
212 spin_unlock(&files->file_lock);
213 expanded = 1;
214 wait_event(files->resize_wait, !files->resize_in_progress);
215 spin_lock(&files->file_lock);
216 goto repeat;
217 }
218
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700219 /* All good, so we try */
Eric Dumazet8a812522015-06-30 15:54:08 +0200220 files->resize_in_progress = true;
221 expanded = expand_fdtable(files, nr);
222 files->resize_in_progress = false;
223
224 wake_up_all(&files->resize_wait);
225 return expanded;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700227
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300228static inline void __set_close_on_exec(unsigned int fd, struct fdtable *fdt)
Al Virob8318b02012-08-21 20:09:42 -0400229{
230 __set_bit(fd, fdt->close_on_exec);
231}
232
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300233static inline void __clear_close_on_exec(unsigned int fd, struct fdtable *fdt)
Al Virob8318b02012-08-21 20:09:42 -0400234{
Linus Torvaldsfc908882015-10-31 16:06:40 -0700235 if (test_bit(fd, fdt->close_on_exec))
236 __clear_bit(fd, fdt->close_on_exec);
Al Virob8318b02012-08-21 20:09:42 -0400237}
238
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700239static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
Al Virob8318b02012-08-21 20:09:42 -0400240{
241 __set_bit(fd, fdt->open_fds);
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700242 fd /= BITS_PER_LONG;
243 if (!~fdt->open_fds[fd])
244 __set_bit(fd, fdt->full_fds_bits);
Al Virob8318b02012-08-21 20:09:42 -0400245}
246
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700247static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
Al Virob8318b02012-08-21 20:09:42 -0400248{
249 __clear_bit(fd, fdt->open_fds);
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700250 __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
Al Virob8318b02012-08-21 20:09:42 -0400251}
252
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300253static unsigned int count_open_files(struct fdtable *fdt)
Al Viro02afc6262008-05-08 19:42:56 -0400254{
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300255 unsigned int size = fdt->max_fds;
256 unsigned int i;
Al Viro02afc6262008-05-08 19:42:56 -0400257
258 /* Find the last open fd */
David Howells1fd36ad2012-02-16 17:49:54 +0000259 for (i = size / BITS_PER_LONG; i > 0; ) {
260 if (fdt->open_fds[--i])
Al Viro02afc6262008-05-08 19:42:56 -0400261 break;
262 }
David Howells1fd36ad2012-02-16 17:49:54 +0000263 i = (i + 1) * BITS_PER_LONG;
Al Viro02afc6262008-05-08 19:42:56 -0400264 return i;
265}
266
Al Viro02afc6262008-05-08 19:42:56 -0400267/*
268 * Allocate a new files structure and copy contents from the
269 * passed in files structure.
270 * errorp will be valid only when the returned files_struct is NULL.
271 */
272struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
273{
274 struct files_struct *newf;
275 struct file **old_fds, **new_fds;
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300276 unsigned int open_files, i;
Al Viro02afc6262008-05-08 19:42:56 -0400277 struct fdtable *old_fdt, *new_fdt;
278
279 *errorp = -ENOMEM;
Al Viroafbec7f2008-05-08 21:11:17 -0400280 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
Al Viro02afc6262008-05-08 19:42:56 -0400281 if (!newf)
282 goto out;
283
Al Viroafbec7f2008-05-08 21:11:17 -0400284 atomic_set(&newf->count, 1);
285
286 spin_lock_init(&newf->file_lock);
Eric Dumazet8a812522015-06-30 15:54:08 +0200287 newf->resize_in_progress = false;
288 init_waitqueue_head(&newf->resize_wait);
Al Viroafbec7f2008-05-08 21:11:17 -0400289 newf->next_fd = 0;
290 new_fdt = &newf->fdtab;
291 new_fdt->max_fds = NR_OPEN_DEFAULT;
David Howells1fd36ad2012-02-16 17:49:54 +0000292 new_fdt->close_on_exec = newf->close_on_exec_init;
293 new_fdt->open_fds = newf->open_fds_init;
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700294 new_fdt->full_fds_bits = newf->full_fds_bits_init;
Al Viroafbec7f2008-05-08 21:11:17 -0400295 new_fdt->fd = &newf->fd_array[0];
Al Viroafbec7f2008-05-08 21:11:17 -0400296
Al Viro02afc6262008-05-08 19:42:56 -0400297 spin_lock(&oldf->file_lock);
298 old_fdt = files_fdtable(oldf);
Al Viro02afc6262008-05-08 19:42:56 -0400299 open_files = count_open_files(old_fdt);
300
301 /*
302 * Check whether we need to allocate a larger fd array and fd set.
Al Viro02afc6262008-05-08 19:42:56 -0400303 */
Al Viroadbecb12008-05-08 21:19:42 -0400304 while (unlikely(open_files > new_fdt->max_fds)) {
Al Viro02afc6262008-05-08 19:42:56 -0400305 spin_unlock(&oldf->file_lock);
Al Viro9dec3c42008-05-08 21:02:45 -0400306
Changli Gaoa892e2d2010-08-10 18:01:35 -0700307 if (new_fdt != &newf->fdtab)
308 __free_fdtable(new_fdt);
Al Viroadbecb12008-05-08 21:19:42 -0400309
Al Viro9dec3c42008-05-08 21:02:45 -0400310 new_fdt = alloc_fdtable(open_files - 1);
311 if (!new_fdt) {
312 *errorp = -ENOMEM;
Al Viro02afc6262008-05-08 19:42:56 -0400313 goto out_release;
Al Viro9dec3c42008-05-08 21:02:45 -0400314 }
315
316 /* beyond sysctl_nr_open; nothing to do */
317 if (unlikely(new_fdt->max_fds < open_files)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700318 __free_fdtable(new_fdt);
Al Viro9dec3c42008-05-08 21:02:45 -0400319 *errorp = -EMFILE;
320 goto out_release;
321 }
Al Viro9dec3c42008-05-08 21:02:45 -0400322
Al Viro02afc6262008-05-08 19:42:56 -0400323 /*
324 * Reacquire the oldf lock and a pointer to its fd table
325 * who knows it may have a new bigger fd table. We need
326 * the latest pointer.
327 */
328 spin_lock(&oldf->file_lock);
329 old_fdt = files_fdtable(oldf);
Al Viroadbecb12008-05-08 21:19:42 -0400330 open_files = count_open_files(old_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400331 }
332
Eric Biggersea5c58e2015-11-06 00:32:04 -0600333 copy_fd_bitmaps(new_fdt, old_fdt, open_files);
334
Al Viro02afc6262008-05-08 19:42:56 -0400335 old_fds = old_fdt->fd;
336 new_fds = new_fdt->fd;
337
Al Viro02afc6262008-05-08 19:42:56 -0400338 for (i = open_files; i != 0; i--) {
339 struct file *f = *old_fds++;
340 if (f) {
341 get_file(f);
342 } else {
343 /*
344 * The fd may be claimed in the fd bitmap but not yet
345 * instantiated in the files array if a sibling thread
346 * is partway through open(). So make sure that this
347 * fd is available to the new process.
348 */
David Howells1dce27c2012-02-16 17:49:42 +0000349 __clear_open_fd(open_files - i, new_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400350 }
351 rcu_assign_pointer(*new_fds++, f);
352 }
353 spin_unlock(&oldf->file_lock);
354
Eric Biggersea5c58e2015-11-06 00:32:04 -0600355 /* clear the remainder */
356 memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
Al Viro02afc6262008-05-08 19:42:56 -0400357
Al Viroafbec7f2008-05-08 21:11:17 -0400358 rcu_assign_pointer(newf->fdt, new_fdt);
359
Al Viro02afc6262008-05-08 19:42:56 -0400360 return newf;
361
362out_release:
363 kmem_cache_free(files_cachep, newf);
364out:
365 return NULL;
366}
367
Oleg Nesterovce08b622014-01-11 19:19:53 +0100368static struct fdtable *close_files(struct files_struct * files)
Al Viro7cf4dc32012-08-15 19:56:12 -0400369{
Al Viro7cf4dc32012-08-15 19:56:12 -0400370 /*
371 * It is safe to dereference the fd table without RCU or
372 * ->file_lock because this is the last reference to the
Oleg Nesterovce08b622014-01-11 19:19:53 +0100373 * files structure.
Al Viro7cf4dc32012-08-15 19:56:12 -0400374 */
Oleg Nesterovce08b622014-01-11 19:19:53 +0100375 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300376 unsigned int i, j = 0;
Oleg Nesterovce08b622014-01-11 19:19:53 +0100377
Al Viro7cf4dc32012-08-15 19:56:12 -0400378 for (;;) {
379 unsigned long set;
380 i = j * BITS_PER_LONG;
381 if (i >= fdt->max_fds)
382 break;
383 set = fdt->open_fds[j++];
384 while (set) {
385 if (set & 1) {
386 struct file * file = xchg(&fdt->fd[i], NULL);
387 if (file) {
388 filp_close(file, files);
Paul E. McKenney388a4c82017-10-24 08:39:34 -0700389 cond_resched();
Al Viro7cf4dc32012-08-15 19:56:12 -0400390 }
391 }
392 i++;
393 set >>= 1;
394 }
395 }
Oleg Nesterovce08b622014-01-11 19:19:53 +0100396
397 return fdt;
Al Viro7cf4dc32012-08-15 19:56:12 -0400398}
399
400struct files_struct *get_files_struct(struct task_struct *task)
401{
402 struct files_struct *files;
403
404 task_lock(task);
405 files = task->files;
406 if (files)
407 atomic_inc(&files->count);
408 task_unlock(task);
409
410 return files;
411}
412
413void put_files_struct(struct files_struct *files)
414{
Al Viro7cf4dc32012-08-15 19:56:12 -0400415 if (atomic_dec_and_test(&files->count)) {
Oleg Nesterovce08b622014-01-11 19:19:53 +0100416 struct fdtable *fdt = close_files(files);
417
Al Virob9e02af2012-08-15 20:00:58 -0400418 /* free the arrays if they are not embedded */
419 if (fdt != &files->fdtab)
420 __free_fdtable(fdt);
421 kmem_cache_free(files_cachep, files);
Al Viro7cf4dc32012-08-15 19:56:12 -0400422 }
423}
424
425void reset_files_struct(struct files_struct *files)
426{
427 struct task_struct *tsk = current;
428 struct files_struct *old;
429
430 old = tsk->files;
431 task_lock(tsk);
432 tsk->files = files;
433 task_unlock(tsk);
434 put_files_struct(old);
435}
436
437void exit_files(struct task_struct *tsk)
438{
439 struct files_struct * files = tsk->files;
440
441 if (files) {
442 task_lock(tsk);
443 tsk->files = NULL;
444 task_unlock(tsk);
445 put_files_struct(files);
446 }
447}
448
Al Virof52111b2008-05-08 18:19:16 -0400449struct files_struct init_files = {
450 .count = ATOMIC_INIT(1),
451 .fdt = &init_files.fdtab,
452 .fdtab = {
453 .max_fds = NR_OPEN_DEFAULT,
454 .fd = &init_files.fd_array[0],
David Howells1fd36ad2012-02-16 17:49:54 +0000455 .close_on_exec = init_files.close_on_exec_init,
456 .open_fds = init_files.open_fds_init,
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700457 .full_fds_bits = init_files.full_fds_bits_init,
Al Virof52111b2008-05-08 18:19:16 -0400458 },
Thomas Gleixnereece09e2011-07-17 21:25:03 +0200459 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
Al Virof52111b2008-05-08 18:19:16 -0400460};
Al Viro1027abe2008-07-30 04:13:04 -0400461
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300462static unsigned int find_next_fd(struct fdtable *fdt, unsigned int start)
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700463{
Alexey Dobriyan9b80a182016-09-02 00:38:52 +0300464 unsigned int maxfd = fdt->max_fds;
465 unsigned int maxbit = maxfd / BITS_PER_LONG;
466 unsigned int bitbit = start / BITS_PER_LONG;
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700467
468 bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
469 if (bitbit > maxfd)
470 return maxfd;
471 if (bitbit > start)
472 start = bitbit;
473 return find_next_zero_bit(fdt->open_fds, maxfd, start);
474}
475
Al Viro1027abe2008-07-30 04:13:04 -0400476/*
477 * allocate a file descriptor, mark it busy.
478 */
Al Virodcfadfa2012-08-12 17:27:30 -0400479int __alloc_fd(struct files_struct *files,
480 unsigned start, unsigned end, unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400481{
Al Viro1027abe2008-07-30 04:13:04 -0400482 unsigned int fd;
483 int error;
484 struct fdtable *fdt;
485
486 spin_lock(&files->file_lock);
487repeat:
488 fdt = files_fdtable(files);
489 fd = start;
490 if (fd < files->next_fd)
491 fd = files->next_fd;
492
493 if (fd < fdt->max_fds)
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700494 fd = find_next_fd(fdt, fd);
Al Viro1027abe2008-07-30 04:13:04 -0400495
Al Virof33ff992012-08-12 16:17:59 -0400496 /*
497 * N.B. For clone tasks sharing a files structure, this test
498 * will limit the total number of files that can be opened.
499 */
500 error = -EMFILE;
501 if (fd >= end)
502 goto out;
503
Al Viro1027abe2008-07-30 04:13:04 -0400504 error = expand_files(files, fd);
505 if (error < 0)
506 goto out;
507
508 /*
509 * If we needed to expand the fs array we
510 * might have blocked - try again.
511 */
512 if (error)
513 goto repeat;
514
515 if (start <= files->next_fd)
516 files->next_fd = fd + 1;
517
David Howells1dce27c2012-02-16 17:49:42 +0000518 __set_open_fd(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400519 if (flags & O_CLOEXEC)
David Howells1dce27c2012-02-16 17:49:42 +0000520 __set_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400521 else
David Howells1dce27c2012-02-16 17:49:42 +0000522 __clear_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400523 error = fd;
524#if 1
525 /* Sanity check */
Paul E. McKenneyadd1f092014-02-12 12:51:09 -0800526 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
Al Viro1027abe2008-07-30 04:13:04 -0400527 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
528 rcu_assign_pointer(fdt->fd[fd], NULL);
529 }
530#endif
531
532out:
533 spin_unlock(&files->file_lock);
534 return error;
535}
536
Al Viroad47bd72012-08-21 20:11:34 -0400537static int alloc_fd(unsigned start, unsigned flags)
Al Virodcfadfa2012-08-12 17:27:30 -0400538{
539 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
540}
541
Al Viro1a7bd222012-08-12 17:18:05 -0400542int get_unused_fd_flags(unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400543{
Al Virodcfadfa2012-08-12 17:27:30 -0400544 return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags);
Al Viro1027abe2008-07-30 04:13:04 -0400545}
Al Viro1a7bd222012-08-12 17:18:05 -0400546EXPORT_SYMBOL(get_unused_fd_flags);
Al Viro56007ca2012-08-15 21:03:26 -0400547
548static void __put_unused_fd(struct files_struct *files, unsigned int fd)
549{
550 struct fdtable *fdt = files_fdtable(files);
551 __clear_open_fd(fd, fdt);
552 if (fd < files->next_fd)
553 files->next_fd = fd;
554}
555
556void put_unused_fd(unsigned int fd)
557{
558 struct files_struct *files = current->files;
559 spin_lock(&files->file_lock);
560 __put_unused_fd(files, fd);
561 spin_unlock(&files->file_lock);
562}
563
564EXPORT_SYMBOL(put_unused_fd);
565
566/*
567 * Install a file pointer in the fd array.
568 *
569 * The VFS is full of places where we drop the files lock between
570 * setting the open_fds bitmap and installing the file in the file
571 * array. At any such point, we are vulnerable to a dup2() race
572 * installing a file in the array before us. We need to detect this and
573 * fput() the struct file we are about to overwrite in this case.
574 *
575 * It should never happen - if we allow dup2() do it, _really_ bad things
576 * will follow.
Al Virof869e8a2012-08-15 21:06:33 -0400577 *
578 * NOTE: __fd_install() variant is really, really low-level; don't
579 * use it unless you are forced to by truly lousy API shoved down
580 * your throat. 'files' *MUST* be either current->files or obtained
581 * by get_files_struct(current) done by whoever had given it to you,
582 * or really bad things will happen. Normally you want to use
583 * fd_install() instead.
Al Viro56007ca2012-08-15 21:03:26 -0400584 */
585
Al Virof869e8a2012-08-15 21:06:33 -0400586void __fd_install(struct files_struct *files, unsigned int fd,
587 struct file *file)
Al Viro56007ca2012-08-15 21:03:26 -0400588{
Al Viro56007ca2012-08-15 21:03:26 -0400589 struct fdtable *fdt;
Eric Dumazet8a812522015-06-30 15:54:08 +0200590
Eric Dumazet8a812522015-06-30 15:54:08 +0200591 rcu_read_lock_sched();
592
Mateusz Guzikc02b1a92017-10-03 12:58:15 +0200593 if (unlikely(files->resize_in_progress)) {
Eric Dumazet8a812522015-06-30 15:54:08 +0200594 rcu_read_unlock_sched();
Mateusz Guzikc02b1a92017-10-03 12:58:15 +0200595 spin_lock(&files->file_lock);
596 fdt = files_fdtable(files);
597 BUG_ON(fdt->fd[fd] != NULL);
598 rcu_assign_pointer(fdt->fd[fd], file);
599 spin_unlock(&files->file_lock);
600 return;
Eric Dumazet8a812522015-06-30 15:54:08 +0200601 }
602 /* coupled with smp_wmb() in expand_fdtable() */
603 smp_rmb();
604 fdt = rcu_dereference_sched(files->fdt);
Al Viro56007ca2012-08-15 21:03:26 -0400605 BUG_ON(fdt->fd[fd] != NULL);
606 rcu_assign_pointer(fdt->fd[fd], file);
Eric Dumazet8a812522015-06-30 15:54:08 +0200607 rcu_read_unlock_sched();
Al Viro56007ca2012-08-15 21:03:26 -0400608}
609
Al Virof869e8a2012-08-15 21:06:33 -0400610void fd_install(unsigned int fd, struct file *file)
611{
612 __fd_install(current->files, fd, file);
613}
614
Al Viro56007ca2012-08-15 21:03:26 -0400615EXPORT_SYMBOL(fd_install);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400616
Al Viro483ce1d2012-08-19 12:04:24 -0400617/*
618 * The same warnings as for __alloc_fd()/__fd_install() apply here...
619 */
620int __close_fd(struct files_struct *files, unsigned fd)
621{
622 struct file *file;
623 struct fdtable *fdt;
624
625 spin_lock(&files->file_lock);
626 fdt = files_fdtable(files);
627 if (fd >= fdt->max_fds)
628 goto out_unlock;
629 file = fdt->fd[fd];
630 if (!file)
631 goto out_unlock;
632 rcu_assign_pointer(fdt->fd[fd], NULL);
Al Viro483ce1d2012-08-19 12:04:24 -0400633 __put_unused_fd(files, fd);
634 spin_unlock(&files->file_lock);
635 return filp_close(file, files);
636
637out_unlock:
638 spin_unlock(&files->file_lock);
639 return -EBADF;
640}
Dominik Brodowski2ca2a092018-03-11 11:34:55 +0100641EXPORT_SYMBOL(__close_fd); /* for ksys_close() */
Al Viro483ce1d2012-08-19 12:04:24 -0400642
Al Viro6a6d27d2012-08-21 09:56:33 -0400643void do_close_on_exec(struct files_struct *files)
644{
645 unsigned i;
646 struct fdtable *fdt;
647
648 /* exec unshares first */
Al Viro6a6d27d2012-08-21 09:56:33 -0400649 spin_lock(&files->file_lock);
650 for (i = 0; ; i++) {
651 unsigned long set;
652 unsigned fd = i * BITS_PER_LONG;
653 fdt = files_fdtable(files);
654 if (fd >= fdt->max_fds)
655 break;
656 set = fdt->close_on_exec[i];
657 if (!set)
658 continue;
659 fdt->close_on_exec[i] = 0;
660 for ( ; set ; fd++, set >>= 1) {
661 struct file *file;
662 if (!(set & 1))
663 continue;
664 file = fdt->fd[fd];
665 if (!file)
666 continue;
667 rcu_assign_pointer(fdt->fd[fd], NULL);
668 __put_unused_fd(files, fd);
669 spin_unlock(&files->file_lock);
670 filp_close(file, files);
671 cond_resched();
672 spin_lock(&files->file_lock);
673 }
674
675 }
676 spin_unlock(&files->file_lock);
677}
678
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100679static struct file *__fget(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400680{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400681 struct files_struct *files = current->files;
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100682 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400683
684 rcu_read_lock();
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200685loop:
Al Viro0ee8cdf2012-08-15 21:12:10 -0400686 file = fcheck_files(files, fd);
687 if (file) {
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200688 /* File object ref couldn't be taken.
689 * dup2() atomicity guarantee is the reason
690 * we loop to catch the new file (or NULL pointer)
691 */
692 if (file->f_mode & mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400693 file = NULL;
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200694 else if (!get_file_rcu(file))
695 goto loop;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400696 }
697 rcu_read_unlock();
698
699 return file;
700}
701
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100702struct file *fget(unsigned int fd)
703{
704 return __fget(fd, FMODE_PATH);
705}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400706EXPORT_SYMBOL(fget);
707
708struct file *fget_raw(unsigned int fd)
709{
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100710 return __fget(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400711}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400712EXPORT_SYMBOL(fget_raw);
713
714/*
715 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
716 *
717 * You can use this instead of fget if you satisfy all of the following
718 * conditions:
719 * 1) You must call fput_light before exiting the syscall and returning control
720 * to userspace (i.e. you cannot remember the returned struct file * after
721 * returning to userspace).
722 * 2) You must not call filp_close on the returned struct file * in between
723 * calls to fget_light and fput_light.
724 * 3) You must not clone the current task in between the calls to fget_light
725 * and fput_light.
726 *
727 * The fput_needed flag returned by fget_light should be passed to the
728 * corresponding fput_light.
729 */
Al Virobd2a31d2014-03-04 14:54:22 -0500730static unsigned long __fget_light(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400731{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400732 struct files_struct *files = current->files;
Oleg Nesterovad461832014-01-13 16:48:40 +0100733 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400734
Al Viro0ee8cdf2012-08-15 21:12:10 -0400735 if (atomic_read(&files->count) == 1) {
Oleg Nesterova8d4b832014-01-11 19:19:32 +0100736 file = __fcheck_files(files, fd);
Al Virobd2a31d2014-03-04 14:54:22 -0500737 if (!file || unlikely(file->f_mode & mask))
738 return 0;
739 return (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400740 } else {
Oleg Nesterove6ff9a92014-01-13 16:49:06 +0100741 file = __fget(fd, mask);
Al Virobd2a31d2014-03-04 14:54:22 -0500742 if (!file)
743 return 0;
744 return FDPUT_FPUT | (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400745 }
Al Viro0ee8cdf2012-08-15 21:12:10 -0400746}
Al Virobd2a31d2014-03-04 14:54:22 -0500747unsigned long __fdget(unsigned int fd)
Oleg Nesterovad461832014-01-13 16:48:40 +0100748{
Al Virobd2a31d2014-03-04 14:54:22 -0500749 return __fget_light(fd, FMODE_PATH);
Oleg Nesterovad461832014-01-13 16:48:40 +0100750}
Al Virobd2a31d2014-03-04 14:54:22 -0500751EXPORT_SYMBOL(__fdget);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400752
Al Virobd2a31d2014-03-04 14:54:22 -0500753unsigned long __fdget_raw(unsigned int fd)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400754{
Al Virobd2a31d2014-03-04 14:54:22 -0500755 return __fget_light(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400756}
Al Virofe17f222012-08-21 11:48:11 -0400757
Al Virobd2a31d2014-03-04 14:54:22 -0500758unsigned long __fdget_pos(unsigned int fd)
759{
Eric Biggers99aea682014-03-16 15:47:48 -0500760 unsigned long v = __fdget(fd);
761 struct file *file = (struct file *)(v & ~3);
Al Virobd2a31d2014-03-04 14:54:22 -0500762
Eric Biggers99aea682014-03-16 15:47:48 -0500763 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
Al Virobd2a31d2014-03-04 14:54:22 -0500764 if (file_count(file) > 1) {
765 v |= FDPUT_POS_UNLOCK;
766 mutex_lock(&file->f_pos_lock);
767 }
768 }
Eric Biggers99aea682014-03-16 15:47:48 -0500769 return v;
Al Virobd2a31d2014-03-04 14:54:22 -0500770}
771
Al Viro63b6df12016-04-20 17:08:21 -0400772void __f_unlock_pos(struct file *f)
773{
774 mutex_unlock(&f->f_pos_lock);
775}
776
Al Virobd2a31d2014-03-04 14:54:22 -0500777/*
778 * We only lock f_pos if we have threads or if the file might be
779 * shared with another process. In both cases we'll have an elevated
780 * file count (done either by fdget() or by fork()).
781 */
782
Al Virofe17f222012-08-21 11:48:11 -0400783void set_close_on_exec(unsigned int fd, int flag)
784{
785 struct files_struct *files = current->files;
786 struct fdtable *fdt;
787 spin_lock(&files->file_lock);
788 fdt = files_fdtable(files);
789 if (flag)
790 __set_close_on_exec(fd, fdt);
791 else
792 __clear_close_on_exec(fd, fdt);
793 spin_unlock(&files->file_lock);
794}
795
796bool get_close_on_exec(unsigned int fd)
797{
798 struct files_struct *files = current->files;
799 struct fdtable *fdt;
800 bool res;
801 rcu_read_lock();
802 fdt = files_fdtable(files);
803 res = close_on_exec(fd, fdt);
804 rcu_read_unlock();
805 return res;
806}
807
Al Viro8280d162012-08-21 12:11:46 -0400808static int do_dup2(struct files_struct *files,
809 struct file *file, unsigned fd, unsigned flags)
Al Viroe9830942014-08-31 14:12:09 -0400810__releases(&files->file_lock)
Al Viro8280d162012-08-21 12:11:46 -0400811{
812 struct file *tofree;
813 struct fdtable *fdt;
814
815 /*
816 * We need to detect attempts to do dup2() over allocated but still
817 * not finished descriptor. NB: OpenBSD avoids that at the price of
818 * extra work in their equivalent of fget() - they insert struct
819 * file immediately after grabbing descriptor, mark it larval if
820 * more work (e.g. actual opening) is needed and make sure that
821 * fget() treats larval files as absent. Potentially interesting,
822 * but while extra work in fget() is trivial, locking implications
823 * and amount of surgery on open()-related paths in VFS are not.
824 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
825 * deadlocks in rather amusing ways, AFAICS. All of that is out of
826 * scope of POSIX or SUS, since neither considers shared descriptor
827 * tables and this condition does not arise without those.
828 */
829 fdt = files_fdtable(files);
830 tofree = fdt->fd[fd];
831 if (!tofree && fd_is_open(fd, fdt))
832 goto Ebusy;
833 get_file(file);
834 rcu_assign_pointer(fdt->fd[fd], file);
835 __set_open_fd(fd, fdt);
836 if (flags & O_CLOEXEC)
837 __set_close_on_exec(fd, fdt);
838 else
839 __clear_close_on_exec(fd, fdt);
840 spin_unlock(&files->file_lock);
841
842 if (tofree)
843 filp_close(tofree, files);
844
845 return fd;
846
847Ebusy:
848 spin_unlock(&files->file_lock);
849 return -EBUSY;
850}
851
852int replace_fd(unsigned fd, struct file *file, unsigned flags)
853{
854 int err;
855 struct files_struct *files = current->files;
856
857 if (!file)
858 return __close_fd(files, fd);
859
860 if (fd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000861 return -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400862
863 spin_lock(&files->file_lock);
864 err = expand_files(files, fd);
865 if (unlikely(err < 0))
866 goto out_unlock;
867 return do_dup2(files, file, fd, flags);
868
869out_unlock:
870 spin_unlock(&files->file_lock);
871 return err;
872}
873
Dominik Brodowskic7248322018-03-11 11:34:40 +0100874static int ksys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
Al Virofe17f222012-08-21 11:48:11 -0400875{
876 int err = -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400877 struct file *file;
878 struct files_struct *files = current->files;
Al Virofe17f222012-08-21 11:48:11 -0400879
880 if ((flags & ~O_CLOEXEC) != 0)
881 return -EINVAL;
882
Richard W.M. Jonesaed97642012-10-09 15:27:43 +0100883 if (unlikely(oldfd == newfd))
884 return -EINVAL;
885
Al Virofe17f222012-08-21 11:48:11 -0400886 if (newfd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000887 return -EBADF;
Al Virofe17f222012-08-21 11:48:11 -0400888
889 spin_lock(&files->file_lock);
890 err = expand_files(files, newfd);
891 file = fcheck(oldfd);
892 if (unlikely(!file))
893 goto Ebadf;
894 if (unlikely(err < 0)) {
895 if (err == -EMFILE)
896 goto Ebadf;
897 goto out_unlock;
898 }
Al Viro8280d162012-08-21 12:11:46 -0400899 return do_dup2(files, file, newfd, flags);
Al Virofe17f222012-08-21 11:48:11 -0400900
901Ebadf:
902 err = -EBADF;
903out_unlock:
904 spin_unlock(&files->file_lock);
905 return err;
906}
907
Dominik Brodowskic7248322018-03-11 11:34:40 +0100908SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
909{
910 return ksys_dup3(oldfd, newfd, flags);
911}
912
Al Virofe17f222012-08-21 11:48:11 -0400913SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
914{
915 if (unlikely(newfd == oldfd)) { /* corner case */
916 struct files_struct *files = current->files;
917 int retval = oldfd;
918
919 rcu_read_lock();
920 if (!fcheck_files(files, oldfd))
921 retval = -EBADF;
922 rcu_read_unlock();
923 return retval;
924 }
Dominik Brodowskic7248322018-03-11 11:34:40 +0100925 return ksys_dup3(oldfd, newfd, 0);
Al Virofe17f222012-08-21 11:48:11 -0400926}
927
Dominik Brodowskic7248322018-03-11 11:34:40 +0100928int ksys_dup(unsigned int fildes)
Al Virofe17f222012-08-21 11:48:11 -0400929{
930 int ret = -EBADF;
931 struct file *file = fget_raw(fildes);
932
933 if (file) {
Yann Droneaud8d10a032014-12-10 15:45:44 -0800934 ret = get_unused_fd_flags(0);
Al Virofe17f222012-08-21 11:48:11 -0400935 if (ret >= 0)
936 fd_install(ret, file);
937 else
938 fput(file);
939 }
940 return ret;
941}
942
Dominik Brodowskic7248322018-03-11 11:34:40 +0100943SYSCALL_DEFINE1(dup, unsigned int, fildes)
944{
945 return ksys_dup(fildes);
946}
947
Al Virofe17f222012-08-21 11:48:11 -0400948int f_dupfd(unsigned int from, struct file *file, unsigned flags)
949{
950 int err;
951 if (from >= rlimit(RLIMIT_NOFILE))
952 return -EINVAL;
953 err = alloc_fd(from, flags);
954 if (err >= 0) {
955 get_file(file);
956 fd_install(err, file);
957 }
958 return err;
959}
Al Viroc3c073f2012-08-21 22:32:06 -0400960
961int iterate_fd(struct files_struct *files, unsigned n,
962 int (*f)(const void *, struct file *, unsigned),
963 const void *p)
964{
965 struct fdtable *fdt;
Al Viroc3c073f2012-08-21 22:32:06 -0400966 int res = 0;
967 if (!files)
968 return 0;
969 spin_lock(&files->file_lock);
Al Viroa77cfcb2012-11-29 22:57:33 -0500970 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
971 struct file *file;
972 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
973 if (!file)
974 continue;
975 res = f(p, file, n);
976 if (res)
977 break;
Al Viroc3c073f2012-08-21 22:32:06 -0400978 }
979 spin_unlock(&files->file_lock);
980 return res;
981}
982EXPORT_SYMBOL(iterate_fd);