blob: 39f8f15921da7973a2300ce4978f37937a1af288 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/file.c
3 *
4 * Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes
5 *
6 * Manage the dynamic fd arrays in the process files_struct.
7 */
8
Al Virofe17f222012-08-21 11:48:11 -04009#include <linux/syscalls.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050010#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/fs.h>
12#include <linux/mm.h>
Andrew Morton6d4831c2011-04-27 15:26:41 -070013#include <linux/mmzone.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070014#include <linux/time.h>
Alexey Dobriyand43c36d2009-10-07 17:09:06 +040015#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/slab.h>
17#include <linux/vmalloc.h>
18#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040019#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/bitops.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070021#include <linux/interrupt.h>
22#include <linux/spinlock.h>
23#include <linux/rcupdate.h>
24#include <linux/workqueue.h>
25
Eric Dumazet9cfe0152008-02-06 01:37:16 -080026int sysctl_nr_open __read_mostly = 1024*1024;
Al Viroeceea0b2008-05-10 10:08:32 -040027int sysctl_nr_open_min = BITS_PER_LONG;
Al Viro7f4b36f2014-03-14 12:45:29 -040028/* our max() is unusable in constant expressions ;-/ */
29#define __const_max(x, y) ((x) < (y) ? (x) : (y))
30int sysctl_nr_open_max = __const_max(INT_MAX, ~(size_t)0/sizeof(void *)) &
31 -BITS_PER_LONG;
Eric Dumazet9cfe0152008-02-06 01:37:16 -080032
David Howells1fd36ad2012-02-16 17:49:54 +000033static void *alloc_fdmem(size_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -070034{
Andrew Morton6d4831c2011-04-27 15:26:41 -070035 /*
36 * Very large allocations can stress page reclaim, so fall back to
37 * vmalloc() if the allocation size will be considered "large" by the VM.
38 */
39 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
Eric W. Biederman96c7a2f2014-02-10 14:25:41 -080040 void *data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN|__GFP_NORETRY);
Andrew Morton6d4831c2011-04-27 15:26:41 -070041 if (data != NULL)
42 return data;
43 }
Changli Gaoa892e2d2010-08-10 18:01:35 -070044 return vmalloc(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045}
46
Changli Gaoa892e2d2010-08-10 18:01:35 -070047static void __free_fdtable(struct fdtable *fdt)
Vadim Lobanov5466b452006-12-10 02:21:22 -080048{
Al Virof6c0a192014-04-23 10:18:46 -040049 kvfree(fdt->fd);
50 kvfree(fdt->open_fds);
Changli Gaoa892e2d2010-08-10 18:01:35 -070051 kfree(fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070052}
53
Al Viro7cf4dc32012-08-15 19:56:12 -040054static void free_fdtable_rcu(struct rcu_head *rcu)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070055{
Al Viroac3e3c52013-04-28 21:42:33 -040056 __free_fdtable(container_of(rcu, struct fdtable, rcu));
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070057}
58
Linus Torvaldsf3f86e32015-10-30 16:53:57 -070059#define BITBIT_NR(nr) BITS_TO_LONGS(BITS_TO_LONGS(nr))
60#define BITBIT_SIZE(nr) (BITBIT_NR(nr) * sizeof(long))
61
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070062/*
Eric Biggersea5c58e2015-11-06 00:32:04 -060063 * Copy 'count' fd bits from the old table to the new table and clear the extra
64 * space if any. This does not copy the file pointers. Called with the files
65 * spinlock held for write.
66 */
67static void copy_fd_bitmaps(struct fdtable *nfdt, struct fdtable *ofdt,
68 unsigned int count)
69{
70 unsigned int cpy, set;
71
72 cpy = count / BITS_PER_BYTE;
73 set = (nfdt->max_fds - count) / BITS_PER_BYTE;
74 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
75 memset((char *)nfdt->open_fds + cpy, 0, set);
76 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
77 memset((char *)nfdt->close_on_exec + cpy, 0, set);
78
79 cpy = BITBIT_SIZE(count);
80 set = BITBIT_SIZE(nfdt->max_fds) - cpy;
81 memcpy(nfdt->full_fds_bits, ofdt->full_fds_bits, cpy);
82 memset((char *)nfdt->full_fds_bits + cpy, 0, set);
83}
84
85/*
86 * Copy all file descriptors from the old table to the new, expanded table and
87 * clear the extra space. Called with the files spinlock held for write.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070088 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080089static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070090{
Vadim Lobanov5466b452006-12-10 02:21:22 -080091 unsigned int cpy, set;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070092
Vadim Lobanov5466b452006-12-10 02:21:22 -080093 BUG_ON(nfdt->max_fds < ofdt->max_fds);
Vadim Lobanov5466b452006-12-10 02:21:22 -080094
95 cpy = ofdt->max_fds * sizeof(struct file *);
96 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
97 memcpy(nfdt->fd, ofdt->fd, cpy);
Eric Biggersea5c58e2015-11-06 00:32:04 -060098 memset((char *)nfdt->fd + cpy, 0, set);
Vadim Lobanov5466b452006-12-10 02:21:22 -080099
Eric Biggersea5c58e2015-11-06 00:32:04 -0600100 copy_fd_bitmaps(nfdt, ofdt, ofdt->max_fds);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101}
102
Vadim Lobanov5466b452006-12-10 02:21:22 -0800103static struct fdtable * alloc_fdtable(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104{
Vadim Lobanov5466b452006-12-10 02:21:22 -0800105 struct fdtable *fdt;
David Howells1fd36ad2012-02-16 17:49:54 +0000106 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700108 /*
Vadim Lobanov5466b452006-12-10 02:21:22 -0800109 * Figure out how many fds we actually want to support in this fdtable.
110 * Allocation steps are keyed to the size of the fdarray, since it
111 * grows far faster than any of the other dynamic data. We try to fit
112 * the fdarray into comfortable page-tuned chunks: starting at 1024B
113 * and growing in powers of two from there on.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700114 */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800115 nr /= (1024 / sizeof(struct file *));
116 nr = roundup_pow_of_two(nr + 1);
117 nr *= (1024 / sizeof(struct file *));
Al Viro5c598b32008-04-27 20:04:15 -0400118 /*
119 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
120 * had been set lower between the check in expand_files() and here. Deal
121 * with that in caller, it's cheaper that way.
122 *
123 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
124 * bitmaps handling below becomes unpleasant, to put it mildly...
125 */
126 if (unlikely(nr > sysctl_nr_open))
127 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800128
Vadim Lobanov5466b452006-12-10 02:21:22 -0800129 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
130 if (!fdt)
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800131 goto out;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800132 fdt->max_fds = nr;
133 data = alloc_fdmem(nr * sizeof(struct file *));
134 if (!data)
135 goto out_fdt;
David Howells1fd36ad2012-02-16 17:49:54 +0000136 fdt->fd = data;
137
138 data = alloc_fdmem(max_t(size_t,
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700139 2 * nr / BITS_PER_BYTE + BITBIT_SIZE(nr), L1_CACHE_BYTES));
Vadim Lobanov5466b452006-12-10 02:21:22 -0800140 if (!data)
141 goto out_arr;
David Howells1fd36ad2012-02-16 17:49:54 +0000142 fdt->open_fds = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800143 data += nr / BITS_PER_BYTE;
David Howells1fd36ad2012-02-16 17:49:54 +0000144 fdt->close_on_exec = data;
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700145 data += nr / BITS_PER_BYTE;
146 fdt->full_fds_bits = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800147
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700148 return fdt;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800149
150out_arr:
Al Virof6c0a192014-04-23 10:18:46 -0400151 kvfree(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800152out_fdt:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700153 kfree(fdt);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800154out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700155 return NULL;
156}
157
158/*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700159 * Expand the file descriptor table.
160 * This function will allocate a new fdtable and both fd array and fdset, of
161 * the given size.
162 * Return <0 error code on error; 1 on successful completion.
163 * The files->file_lock should be held on entry, and will be held on exit.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700164 */
165static int expand_fdtable(struct files_struct *files, int nr)
166 __releases(files->file_lock)
167 __acquires(files->file_lock)
168{
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700169 struct fdtable *new_fdt, *cur_fdt;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 spin_unlock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700172 new_fdt = alloc_fdtable(nr);
Eric Dumazet8a812522015-06-30 15:54:08 +0200173
174 /* make sure all __fd_install() have seen resize_in_progress
175 * or have finished their rcu_read_lock_sched() section.
176 */
177 if (atomic_read(&files->count) > 1)
178 synchronize_sched();
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700181 if (!new_fdt)
182 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700183 /*
Al Viro5c598b32008-04-27 20:04:15 -0400184 * extremely unlikely race - sysctl_nr_open decreased between the check in
185 * caller and alloc_fdtable(). Cheaper to catch it here...
186 */
187 if (unlikely(new_fdt->max_fds <= nr)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700188 __free_fdtable(new_fdt);
Al Viro5c598b32008-04-27 20:04:15 -0400189 return -EMFILE;
190 }
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700191 cur_fdt = files_fdtable(files);
Eric Dumazet8a812522015-06-30 15:54:08 +0200192 BUG_ON(nr < cur_fdt->max_fds);
193 copy_fdtable(new_fdt, cur_fdt);
194 rcu_assign_pointer(files->fdt, new_fdt);
195 if (cur_fdt != &files->fdtab)
196 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
197 /* coupled with smp_rmb() in __fd_install() */
198 smp_wmb();
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700199 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200}
201
202/*
203 * Expand files.
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700204 * This function will expand the file structures, if the requested size exceeds
205 * the current capacity and there is room for expansion.
206 * Return <0 error code on error; 0 when nothing done; 1 when files were
207 * expanded and execution may have blocked.
208 * The files->file_lock should be held on entry, and will be held on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 */
Al Viroad47bd72012-08-21 20:11:34 -0400210static int expand_files(struct files_struct *files, int nr)
Eric Dumazet8a812522015-06-30 15:54:08 +0200211 __releases(files->file_lock)
212 __acquires(files->file_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700214 struct fdtable *fdt;
Eric Dumazet8a812522015-06-30 15:54:08 +0200215 int expanded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
Eric Dumazet8a812522015-06-30 15:54:08 +0200217repeat:
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700218 fdt = files_fdtable(files);
Al Viro4e1e0182008-07-26 16:01:20 -0400219
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700220 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800221 if (nr < fdt->max_fds)
Eric Dumazet8a812522015-06-30 15:54:08 +0200222 return expanded;
Al Viro4e1e0182008-07-26 16:01:20 -0400223
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700224 /* Can we expand? */
Eric Dumazet9cfe0152008-02-06 01:37:16 -0800225 if (nr >= sysctl_nr_open)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700226 return -EMFILE;
227
Eric Dumazet8a812522015-06-30 15:54:08 +0200228 if (unlikely(files->resize_in_progress)) {
229 spin_unlock(&files->file_lock);
230 expanded = 1;
231 wait_event(files->resize_wait, !files->resize_in_progress);
232 spin_lock(&files->file_lock);
233 goto repeat;
234 }
235
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700236 /* All good, so we try */
Eric Dumazet8a812522015-06-30 15:54:08 +0200237 files->resize_in_progress = true;
238 expanded = expand_fdtable(files, nr);
239 files->resize_in_progress = false;
240
241 wake_up_all(&files->resize_wait);
242 return expanded;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700244
Al Virob8318b02012-08-21 20:09:42 -0400245static inline void __set_close_on_exec(int fd, struct fdtable *fdt)
246{
247 __set_bit(fd, fdt->close_on_exec);
248}
249
250static inline void __clear_close_on_exec(int fd, struct fdtable *fdt)
251{
Linus Torvaldsfc908882015-10-31 16:06:40 -0700252 if (test_bit(fd, fdt->close_on_exec))
253 __clear_bit(fd, fdt->close_on_exec);
Al Virob8318b02012-08-21 20:09:42 -0400254}
255
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700256static inline void __set_open_fd(unsigned int fd, struct fdtable *fdt)
Al Virob8318b02012-08-21 20:09:42 -0400257{
258 __set_bit(fd, fdt->open_fds);
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700259 fd /= BITS_PER_LONG;
260 if (!~fdt->open_fds[fd])
261 __set_bit(fd, fdt->full_fds_bits);
Al Virob8318b02012-08-21 20:09:42 -0400262}
263
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700264static inline void __clear_open_fd(unsigned int fd, struct fdtable *fdt)
Al Virob8318b02012-08-21 20:09:42 -0400265{
266 __clear_bit(fd, fdt->open_fds);
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700267 __clear_bit(fd / BITS_PER_LONG, fdt->full_fds_bits);
Al Virob8318b02012-08-21 20:09:42 -0400268}
269
Al Viro02afc6262008-05-08 19:42:56 -0400270static int count_open_files(struct fdtable *fdt)
271{
272 int size = fdt->max_fds;
273 int i;
274
275 /* Find the last open fd */
David Howells1fd36ad2012-02-16 17:49:54 +0000276 for (i = size / BITS_PER_LONG; i > 0; ) {
277 if (fdt->open_fds[--i])
Al Viro02afc6262008-05-08 19:42:56 -0400278 break;
279 }
David Howells1fd36ad2012-02-16 17:49:54 +0000280 i = (i + 1) * BITS_PER_LONG;
Al Viro02afc6262008-05-08 19:42:56 -0400281 return i;
282}
283
Al Viro02afc6262008-05-08 19:42:56 -0400284/*
285 * Allocate a new files structure and copy contents from the
286 * passed in files structure.
287 * errorp will be valid only when the returned files_struct is NULL.
288 */
289struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
290{
291 struct files_struct *newf;
292 struct file **old_fds, **new_fds;
Eric Biggersea5c58e2015-11-06 00:32:04 -0600293 int open_files, i;
Al Viro02afc6262008-05-08 19:42:56 -0400294 struct fdtable *old_fdt, *new_fdt;
295
296 *errorp = -ENOMEM;
Al Viroafbec7f2008-05-08 21:11:17 -0400297 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
Al Viro02afc6262008-05-08 19:42:56 -0400298 if (!newf)
299 goto out;
300
Al Viroafbec7f2008-05-08 21:11:17 -0400301 atomic_set(&newf->count, 1);
302
303 spin_lock_init(&newf->file_lock);
Eric Dumazet8a812522015-06-30 15:54:08 +0200304 newf->resize_in_progress = false;
305 init_waitqueue_head(&newf->resize_wait);
Al Viroafbec7f2008-05-08 21:11:17 -0400306 newf->next_fd = 0;
307 new_fdt = &newf->fdtab;
308 new_fdt->max_fds = NR_OPEN_DEFAULT;
David Howells1fd36ad2012-02-16 17:49:54 +0000309 new_fdt->close_on_exec = newf->close_on_exec_init;
310 new_fdt->open_fds = newf->open_fds_init;
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700311 new_fdt->full_fds_bits = newf->full_fds_bits_init;
Al Viroafbec7f2008-05-08 21:11:17 -0400312 new_fdt->fd = &newf->fd_array[0];
Al Viroafbec7f2008-05-08 21:11:17 -0400313
Al Viro02afc6262008-05-08 19:42:56 -0400314 spin_lock(&oldf->file_lock);
315 old_fdt = files_fdtable(oldf);
Al Viro02afc6262008-05-08 19:42:56 -0400316 open_files = count_open_files(old_fdt);
317
318 /*
319 * Check whether we need to allocate a larger fd array and fd set.
Al Viro02afc6262008-05-08 19:42:56 -0400320 */
Al Viroadbecb12008-05-08 21:19:42 -0400321 while (unlikely(open_files > new_fdt->max_fds)) {
Al Viro02afc6262008-05-08 19:42:56 -0400322 spin_unlock(&oldf->file_lock);
Al Viro9dec3c42008-05-08 21:02:45 -0400323
Changli Gaoa892e2d2010-08-10 18:01:35 -0700324 if (new_fdt != &newf->fdtab)
325 __free_fdtable(new_fdt);
Al Viroadbecb12008-05-08 21:19:42 -0400326
Al Viro9dec3c42008-05-08 21:02:45 -0400327 new_fdt = alloc_fdtable(open_files - 1);
328 if (!new_fdt) {
329 *errorp = -ENOMEM;
Al Viro02afc6262008-05-08 19:42:56 -0400330 goto out_release;
Al Viro9dec3c42008-05-08 21:02:45 -0400331 }
332
333 /* beyond sysctl_nr_open; nothing to do */
334 if (unlikely(new_fdt->max_fds < open_files)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700335 __free_fdtable(new_fdt);
Al Viro9dec3c42008-05-08 21:02:45 -0400336 *errorp = -EMFILE;
337 goto out_release;
338 }
Al Viro9dec3c42008-05-08 21:02:45 -0400339
Al Viro02afc6262008-05-08 19:42:56 -0400340 /*
341 * Reacquire the oldf lock and a pointer to its fd table
342 * who knows it may have a new bigger fd table. We need
343 * the latest pointer.
344 */
345 spin_lock(&oldf->file_lock);
346 old_fdt = files_fdtable(oldf);
Al Viroadbecb12008-05-08 21:19:42 -0400347 open_files = count_open_files(old_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400348 }
349
Eric Biggersea5c58e2015-11-06 00:32:04 -0600350 copy_fd_bitmaps(new_fdt, old_fdt, open_files);
351
Al Viro02afc6262008-05-08 19:42:56 -0400352 old_fds = old_fdt->fd;
353 new_fds = new_fdt->fd;
354
Al Viro02afc6262008-05-08 19:42:56 -0400355 for (i = open_files; i != 0; i--) {
356 struct file *f = *old_fds++;
357 if (f) {
358 get_file(f);
359 } else {
360 /*
361 * The fd may be claimed in the fd bitmap but not yet
362 * instantiated in the files array if a sibling thread
363 * is partway through open(). So make sure that this
364 * fd is available to the new process.
365 */
David Howells1dce27c2012-02-16 17:49:42 +0000366 __clear_open_fd(open_files - i, new_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400367 }
368 rcu_assign_pointer(*new_fds++, f);
369 }
370 spin_unlock(&oldf->file_lock);
371
Eric Biggersea5c58e2015-11-06 00:32:04 -0600372 /* clear the remainder */
373 memset(new_fds, 0, (new_fdt->max_fds - open_files) * sizeof(struct file *));
Al Viro02afc6262008-05-08 19:42:56 -0400374
Al Viroafbec7f2008-05-08 21:11:17 -0400375 rcu_assign_pointer(newf->fdt, new_fdt);
376
Al Viro02afc6262008-05-08 19:42:56 -0400377 return newf;
378
379out_release:
380 kmem_cache_free(files_cachep, newf);
381out:
382 return NULL;
383}
384
Oleg Nesterovce08b622014-01-11 19:19:53 +0100385static struct fdtable *close_files(struct files_struct * files)
Al Viro7cf4dc32012-08-15 19:56:12 -0400386{
Al Viro7cf4dc32012-08-15 19:56:12 -0400387 /*
388 * It is safe to dereference the fd table without RCU or
389 * ->file_lock because this is the last reference to the
Oleg Nesterovce08b622014-01-11 19:19:53 +0100390 * files structure.
Al Viro7cf4dc32012-08-15 19:56:12 -0400391 */
Oleg Nesterovce08b622014-01-11 19:19:53 +0100392 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
393 int i, j = 0;
394
Al Viro7cf4dc32012-08-15 19:56:12 -0400395 for (;;) {
396 unsigned long set;
397 i = j * BITS_PER_LONG;
398 if (i >= fdt->max_fds)
399 break;
400 set = fdt->open_fds[j++];
401 while (set) {
402 if (set & 1) {
403 struct file * file = xchg(&fdt->fd[i], NULL);
404 if (file) {
405 filp_close(file, files);
Paul E. McKenneybde6c3a2014-07-01 11:26:57 -0700406 cond_resched_rcu_qs();
Al Viro7cf4dc32012-08-15 19:56:12 -0400407 }
408 }
409 i++;
410 set >>= 1;
411 }
412 }
Oleg Nesterovce08b622014-01-11 19:19:53 +0100413
414 return fdt;
Al Viro7cf4dc32012-08-15 19:56:12 -0400415}
416
417struct files_struct *get_files_struct(struct task_struct *task)
418{
419 struct files_struct *files;
420
421 task_lock(task);
422 files = task->files;
423 if (files)
424 atomic_inc(&files->count);
425 task_unlock(task);
426
427 return files;
428}
429
430void put_files_struct(struct files_struct *files)
431{
Al Viro7cf4dc32012-08-15 19:56:12 -0400432 if (atomic_dec_and_test(&files->count)) {
Oleg Nesterovce08b622014-01-11 19:19:53 +0100433 struct fdtable *fdt = close_files(files);
434
Al Virob9e02af2012-08-15 20:00:58 -0400435 /* free the arrays if they are not embedded */
436 if (fdt != &files->fdtab)
437 __free_fdtable(fdt);
438 kmem_cache_free(files_cachep, files);
Al Viro7cf4dc32012-08-15 19:56:12 -0400439 }
440}
441
442void reset_files_struct(struct files_struct *files)
443{
444 struct task_struct *tsk = current;
445 struct files_struct *old;
446
447 old = tsk->files;
448 task_lock(tsk);
449 tsk->files = files;
450 task_unlock(tsk);
451 put_files_struct(old);
452}
453
454void exit_files(struct task_struct *tsk)
455{
456 struct files_struct * files = tsk->files;
457
458 if (files) {
459 task_lock(tsk);
460 tsk->files = NULL;
461 task_unlock(tsk);
462 put_files_struct(files);
463 }
464}
465
Al Virof52111b2008-05-08 18:19:16 -0400466struct files_struct init_files = {
467 .count = ATOMIC_INIT(1),
468 .fdt = &init_files.fdtab,
469 .fdtab = {
470 .max_fds = NR_OPEN_DEFAULT,
471 .fd = &init_files.fd_array[0],
David Howells1fd36ad2012-02-16 17:49:54 +0000472 .close_on_exec = init_files.close_on_exec_init,
473 .open_fds = init_files.open_fds_init,
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700474 .full_fds_bits = init_files.full_fds_bits_init,
Al Virof52111b2008-05-08 18:19:16 -0400475 },
Thomas Gleixnereece09e2011-07-17 21:25:03 +0200476 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
Al Virof52111b2008-05-08 18:19:16 -0400477};
Al Viro1027abe2008-07-30 04:13:04 -0400478
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700479static unsigned long find_next_fd(struct fdtable *fdt, unsigned long start)
480{
481 unsigned long maxfd = fdt->max_fds;
482 unsigned long maxbit = maxfd / BITS_PER_LONG;
483 unsigned long bitbit = start / BITS_PER_LONG;
484
485 bitbit = find_next_zero_bit(fdt->full_fds_bits, maxbit, bitbit) * BITS_PER_LONG;
486 if (bitbit > maxfd)
487 return maxfd;
488 if (bitbit > start)
489 start = bitbit;
490 return find_next_zero_bit(fdt->open_fds, maxfd, start);
491}
492
Al Viro1027abe2008-07-30 04:13:04 -0400493/*
494 * allocate a file descriptor, mark it busy.
495 */
Al Virodcfadfa2012-08-12 17:27:30 -0400496int __alloc_fd(struct files_struct *files,
497 unsigned start, unsigned end, unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400498{
Al Viro1027abe2008-07-30 04:13:04 -0400499 unsigned int fd;
500 int error;
501 struct fdtable *fdt;
502
503 spin_lock(&files->file_lock);
504repeat:
505 fdt = files_fdtable(files);
506 fd = start;
507 if (fd < files->next_fd)
508 fd = files->next_fd;
509
510 if (fd < fdt->max_fds)
Linus Torvaldsf3f86e32015-10-30 16:53:57 -0700511 fd = find_next_fd(fdt, fd);
Al Viro1027abe2008-07-30 04:13:04 -0400512
Al Virof33ff992012-08-12 16:17:59 -0400513 /*
514 * N.B. For clone tasks sharing a files structure, this test
515 * will limit the total number of files that can be opened.
516 */
517 error = -EMFILE;
518 if (fd >= end)
519 goto out;
520
Al Viro1027abe2008-07-30 04:13:04 -0400521 error = expand_files(files, fd);
522 if (error < 0)
523 goto out;
524
525 /*
526 * If we needed to expand the fs array we
527 * might have blocked - try again.
528 */
529 if (error)
530 goto repeat;
531
532 if (start <= files->next_fd)
533 files->next_fd = fd + 1;
534
David Howells1dce27c2012-02-16 17:49:42 +0000535 __set_open_fd(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400536 if (flags & O_CLOEXEC)
David Howells1dce27c2012-02-16 17:49:42 +0000537 __set_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400538 else
David Howells1dce27c2012-02-16 17:49:42 +0000539 __clear_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400540 error = fd;
541#if 1
542 /* Sanity check */
Paul E. McKenneyadd1f092014-02-12 12:51:09 -0800543 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
Al Viro1027abe2008-07-30 04:13:04 -0400544 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
545 rcu_assign_pointer(fdt->fd[fd], NULL);
546 }
547#endif
548
549out:
550 spin_unlock(&files->file_lock);
551 return error;
552}
553
Al Viroad47bd72012-08-21 20:11:34 -0400554static int alloc_fd(unsigned start, unsigned flags)
Al Virodcfadfa2012-08-12 17:27:30 -0400555{
556 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
557}
558
Al Viro1a7bd222012-08-12 17:18:05 -0400559int get_unused_fd_flags(unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400560{
Al Virodcfadfa2012-08-12 17:27:30 -0400561 return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags);
Al Viro1027abe2008-07-30 04:13:04 -0400562}
Al Viro1a7bd222012-08-12 17:18:05 -0400563EXPORT_SYMBOL(get_unused_fd_flags);
Al Viro56007ca2012-08-15 21:03:26 -0400564
565static void __put_unused_fd(struct files_struct *files, unsigned int fd)
566{
567 struct fdtable *fdt = files_fdtable(files);
568 __clear_open_fd(fd, fdt);
569 if (fd < files->next_fd)
570 files->next_fd = fd;
571}
572
573void put_unused_fd(unsigned int fd)
574{
575 struct files_struct *files = current->files;
576 spin_lock(&files->file_lock);
577 __put_unused_fd(files, fd);
578 spin_unlock(&files->file_lock);
579}
580
581EXPORT_SYMBOL(put_unused_fd);
582
583/*
584 * Install a file pointer in the fd array.
585 *
586 * The VFS is full of places where we drop the files lock between
587 * setting the open_fds bitmap and installing the file in the file
588 * array. At any such point, we are vulnerable to a dup2() race
589 * installing a file in the array before us. We need to detect this and
590 * fput() the struct file we are about to overwrite in this case.
591 *
592 * It should never happen - if we allow dup2() do it, _really_ bad things
593 * will follow.
Al Virof869e8a2012-08-15 21:06:33 -0400594 *
595 * NOTE: __fd_install() variant is really, really low-level; don't
596 * use it unless you are forced to by truly lousy API shoved down
597 * your throat. 'files' *MUST* be either current->files or obtained
598 * by get_files_struct(current) done by whoever had given it to you,
599 * or really bad things will happen. Normally you want to use
600 * fd_install() instead.
Al Viro56007ca2012-08-15 21:03:26 -0400601 */
602
Al Virof869e8a2012-08-15 21:06:33 -0400603void __fd_install(struct files_struct *files, unsigned int fd,
604 struct file *file)
Al Viro56007ca2012-08-15 21:03:26 -0400605{
Al Viro56007ca2012-08-15 21:03:26 -0400606 struct fdtable *fdt;
Eric Dumazet8a812522015-06-30 15:54:08 +0200607
608 might_sleep();
609 rcu_read_lock_sched();
610
611 while (unlikely(files->resize_in_progress)) {
612 rcu_read_unlock_sched();
613 wait_event(files->resize_wait, !files->resize_in_progress);
614 rcu_read_lock_sched();
615 }
616 /* coupled with smp_wmb() in expand_fdtable() */
617 smp_rmb();
618 fdt = rcu_dereference_sched(files->fdt);
Al Viro56007ca2012-08-15 21:03:26 -0400619 BUG_ON(fdt->fd[fd] != NULL);
620 rcu_assign_pointer(fdt->fd[fd], file);
Eric Dumazet8a812522015-06-30 15:54:08 +0200621 rcu_read_unlock_sched();
Al Viro56007ca2012-08-15 21:03:26 -0400622}
623
Al Virof869e8a2012-08-15 21:06:33 -0400624void fd_install(unsigned int fd, struct file *file)
625{
626 __fd_install(current->files, fd, file);
627}
628
Al Viro56007ca2012-08-15 21:03:26 -0400629EXPORT_SYMBOL(fd_install);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400630
Al Viro483ce1d2012-08-19 12:04:24 -0400631/*
632 * The same warnings as for __alloc_fd()/__fd_install() apply here...
633 */
634int __close_fd(struct files_struct *files, unsigned fd)
635{
636 struct file *file;
637 struct fdtable *fdt;
638
639 spin_lock(&files->file_lock);
640 fdt = files_fdtable(files);
641 if (fd >= fdt->max_fds)
642 goto out_unlock;
643 file = fdt->fd[fd];
644 if (!file)
645 goto out_unlock;
646 rcu_assign_pointer(fdt->fd[fd], NULL);
647 __clear_close_on_exec(fd, fdt);
648 __put_unused_fd(files, fd);
649 spin_unlock(&files->file_lock);
650 return filp_close(file, files);
651
652out_unlock:
653 spin_unlock(&files->file_lock);
654 return -EBADF;
655}
656
Al Viro6a6d27d2012-08-21 09:56:33 -0400657void do_close_on_exec(struct files_struct *files)
658{
659 unsigned i;
660 struct fdtable *fdt;
661
662 /* exec unshares first */
Al Viro6a6d27d2012-08-21 09:56:33 -0400663 spin_lock(&files->file_lock);
664 for (i = 0; ; i++) {
665 unsigned long set;
666 unsigned fd = i * BITS_PER_LONG;
667 fdt = files_fdtable(files);
668 if (fd >= fdt->max_fds)
669 break;
670 set = fdt->close_on_exec[i];
671 if (!set)
672 continue;
673 fdt->close_on_exec[i] = 0;
674 for ( ; set ; fd++, set >>= 1) {
675 struct file *file;
676 if (!(set & 1))
677 continue;
678 file = fdt->fd[fd];
679 if (!file)
680 continue;
681 rcu_assign_pointer(fdt->fd[fd], NULL);
682 __put_unused_fd(files, fd);
683 spin_unlock(&files->file_lock);
684 filp_close(file, files);
685 cond_resched();
686 spin_lock(&files->file_lock);
687 }
688
689 }
690 spin_unlock(&files->file_lock);
691}
692
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100693static struct file *__fget(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400694{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400695 struct files_struct *files = current->files;
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100696 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400697
698 rcu_read_lock();
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200699loop:
Al Viro0ee8cdf2012-08-15 21:12:10 -0400700 file = fcheck_files(files, fd);
701 if (file) {
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200702 /* File object ref couldn't be taken.
703 * dup2() atomicity guarantee is the reason
704 * we loop to catch the new file (or NULL pointer)
705 */
706 if (file->f_mode & mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400707 file = NULL;
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200708 else if (!get_file_rcu(file))
709 goto loop;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400710 }
711 rcu_read_unlock();
712
713 return file;
714}
715
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100716struct file *fget(unsigned int fd)
717{
718 return __fget(fd, FMODE_PATH);
719}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400720EXPORT_SYMBOL(fget);
721
722struct file *fget_raw(unsigned int fd)
723{
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100724 return __fget(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400725}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400726EXPORT_SYMBOL(fget_raw);
727
728/*
729 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
730 *
731 * You can use this instead of fget if you satisfy all of the following
732 * conditions:
733 * 1) You must call fput_light before exiting the syscall and returning control
734 * to userspace (i.e. you cannot remember the returned struct file * after
735 * returning to userspace).
736 * 2) You must not call filp_close on the returned struct file * in between
737 * calls to fget_light and fput_light.
738 * 3) You must not clone the current task in between the calls to fget_light
739 * and fput_light.
740 *
741 * The fput_needed flag returned by fget_light should be passed to the
742 * corresponding fput_light.
743 */
Al Virobd2a31d2014-03-04 14:54:22 -0500744static unsigned long __fget_light(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400745{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400746 struct files_struct *files = current->files;
Oleg Nesterovad461832014-01-13 16:48:40 +0100747 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400748
Al Viro0ee8cdf2012-08-15 21:12:10 -0400749 if (atomic_read(&files->count) == 1) {
Oleg Nesterova8d4b832014-01-11 19:19:32 +0100750 file = __fcheck_files(files, fd);
Al Virobd2a31d2014-03-04 14:54:22 -0500751 if (!file || unlikely(file->f_mode & mask))
752 return 0;
753 return (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400754 } else {
Oleg Nesterove6ff9a92014-01-13 16:49:06 +0100755 file = __fget(fd, mask);
Al Virobd2a31d2014-03-04 14:54:22 -0500756 if (!file)
757 return 0;
758 return FDPUT_FPUT | (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400759 }
Al Viro0ee8cdf2012-08-15 21:12:10 -0400760}
Al Virobd2a31d2014-03-04 14:54:22 -0500761unsigned long __fdget(unsigned int fd)
Oleg Nesterovad461832014-01-13 16:48:40 +0100762{
Al Virobd2a31d2014-03-04 14:54:22 -0500763 return __fget_light(fd, FMODE_PATH);
Oleg Nesterovad461832014-01-13 16:48:40 +0100764}
Al Virobd2a31d2014-03-04 14:54:22 -0500765EXPORT_SYMBOL(__fdget);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400766
Al Virobd2a31d2014-03-04 14:54:22 -0500767unsigned long __fdget_raw(unsigned int fd)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400768{
Al Virobd2a31d2014-03-04 14:54:22 -0500769 return __fget_light(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400770}
Al Virofe17f222012-08-21 11:48:11 -0400771
Al Virobd2a31d2014-03-04 14:54:22 -0500772unsigned long __fdget_pos(unsigned int fd)
773{
Eric Biggers99aea682014-03-16 15:47:48 -0500774 unsigned long v = __fdget(fd);
775 struct file *file = (struct file *)(v & ~3);
Al Virobd2a31d2014-03-04 14:54:22 -0500776
Eric Biggers99aea682014-03-16 15:47:48 -0500777 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
Al Virobd2a31d2014-03-04 14:54:22 -0500778 if (file_count(file) > 1) {
779 v |= FDPUT_POS_UNLOCK;
780 mutex_lock(&file->f_pos_lock);
781 }
782 }
Eric Biggers99aea682014-03-16 15:47:48 -0500783 return v;
Al Virobd2a31d2014-03-04 14:54:22 -0500784}
785
786/*
787 * We only lock f_pos if we have threads or if the file might be
788 * shared with another process. In both cases we'll have an elevated
789 * file count (done either by fdget() or by fork()).
790 */
791
Al Virofe17f222012-08-21 11:48:11 -0400792void set_close_on_exec(unsigned int fd, int flag)
793{
794 struct files_struct *files = current->files;
795 struct fdtable *fdt;
796 spin_lock(&files->file_lock);
797 fdt = files_fdtable(files);
798 if (flag)
799 __set_close_on_exec(fd, fdt);
800 else
801 __clear_close_on_exec(fd, fdt);
802 spin_unlock(&files->file_lock);
803}
804
805bool get_close_on_exec(unsigned int fd)
806{
807 struct files_struct *files = current->files;
808 struct fdtable *fdt;
809 bool res;
810 rcu_read_lock();
811 fdt = files_fdtable(files);
812 res = close_on_exec(fd, fdt);
813 rcu_read_unlock();
814 return res;
815}
816
Al Viro8280d162012-08-21 12:11:46 -0400817static int do_dup2(struct files_struct *files,
818 struct file *file, unsigned fd, unsigned flags)
Al Viroe9830942014-08-31 14:12:09 -0400819__releases(&files->file_lock)
Al Viro8280d162012-08-21 12:11:46 -0400820{
821 struct file *tofree;
822 struct fdtable *fdt;
823
824 /*
825 * We need to detect attempts to do dup2() over allocated but still
826 * not finished descriptor. NB: OpenBSD avoids that at the price of
827 * extra work in their equivalent of fget() - they insert struct
828 * file immediately after grabbing descriptor, mark it larval if
829 * more work (e.g. actual opening) is needed and make sure that
830 * fget() treats larval files as absent. Potentially interesting,
831 * but while extra work in fget() is trivial, locking implications
832 * and amount of surgery on open()-related paths in VFS are not.
833 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
834 * deadlocks in rather amusing ways, AFAICS. All of that is out of
835 * scope of POSIX or SUS, since neither considers shared descriptor
836 * tables and this condition does not arise without those.
837 */
838 fdt = files_fdtable(files);
839 tofree = fdt->fd[fd];
840 if (!tofree && fd_is_open(fd, fdt))
841 goto Ebusy;
842 get_file(file);
843 rcu_assign_pointer(fdt->fd[fd], file);
844 __set_open_fd(fd, fdt);
845 if (flags & O_CLOEXEC)
846 __set_close_on_exec(fd, fdt);
847 else
848 __clear_close_on_exec(fd, fdt);
849 spin_unlock(&files->file_lock);
850
851 if (tofree)
852 filp_close(tofree, files);
853
854 return fd;
855
856Ebusy:
857 spin_unlock(&files->file_lock);
858 return -EBUSY;
859}
860
861int replace_fd(unsigned fd, struct file *file, unsigned flags)
862{
863 int err;
864 struct files_struct *files = current->files;
865
866 if (!file)
867 return __close_fd(files, fd);
868
869 if (fd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000870 return -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400871
872 spin_lock(&files->file_lock);
873 err = expand_files(files, fd);
874 if (unlikely(err < 0))
875 goto out_unlock;
876 return do_dup2(files, file, fd, flags);
877
878out_unlock:
879 spin_unlock(&files->file_lock);
880 return err;
881}
882
Al Virofe17f222012-08-21 11:48:11 -0400883SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
884{
885 int err = -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400886 struct file *file;
887 struct files_struct *files = current->files;
Al Virofe17f222012-08-21 11:48:11 -0400888
889 if ((flags & ~O_CLOEXEC) != 0)
890 return -EINVAL;
891
Richard W.M. Jonesaed97642012-10-09 15:27:43 +0100892 if (unlikely(oldfd == newfd))
893 return -EINVAL;
894
Al Virofe17f222012-08-21 11:48:11 -0400895 if (newfd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000896 return -EBADF;
Al Virofe17f222012-08-21 11:48:11 -0400897
898 spin_lock(&files->file_lock);
899 err = expand_files(files, newfd);
900 file = fcheck(oldfd);
901 if (unlikely(!file))
902 goto Ebadf;
903 if (unlikely(err < 0)) {
904 if (err == -EMFILE)
905 goto Ebadf;
906 goto out_unlock;
907 }
Al Viro8280d162012-08-21 12:11:46 -0400908 return do_dup2(files, file, newfd, flags);
Al Virofe17f222012-08-21 11:48:11 -0400909
910Ebadf:
911 err = -EBADF;
912out_unlock:
913 spin_unlock(&files->file_lock);
914 return err;
915}
916
917SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
918{
919 if (unlikely(newfd == oldfd)) { /* corner case */
920 struct files_struct *files = current->files;
921 int retval = oldfd;
922
923 rcu_read_lock();
924 if (!fcheck_files(files, oldfd))
925 retval = -EBADF;
926 rcu_read_unlock();
927 return retval;
928 }
929 return sys_dup3(oldfd, newfd, 0);
930}
931
932SYSCALL_DEFINE1(dup, unsigned int, fildes)
933{
934 int ret = -EBADF;
935 struct file *file = fget_raw(fildes);
936
937 if (file) {
Yann Droneaud8d10a032014-12-10 15:45:44 -0800938 ret = get_unused_fd_flags(0);
Al Virofe17f222012-08-21 11:48:11 -0400939 if (ret >= 0)
940 fd_install(ret, file);
941 else
942 fput(file);
943 }
944 return ret;
945}
946
947int f_dupfd(unsigned int from, struct file *file, unsigned flags)
948{
949 int err;
950 if (from >= rlimit(RLIMIT_NOFILE))
951 return -EINVAL;
952 err = alloc_fd(from, flags);
953 if (err >= 0) {
954 get_file(file);
955 fd_install(err, file);
956 }
957 return err;
958}
Al Viroc3c073f2012-08-21 22:32:06 -0400959
960int iterate_fd(struct files_struct *files, unsigned n,
961 int (*f)(const void *, struct file *, unsigned),
962 const void *p)
963{
964 struct fdtable *fdt;
Al Viroc3c073f2012-08-21 22:32:06 -0400965 int res = 0;
966 if (!files)
967 return 0;
968 spin_lock(&files->file_lock);
Al Viroa77cfcb2012-11-29 22:57:33 -0500969 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
970 struct file *file;
971 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
972 if (!file)
973 continue;
974 res = f(p, file, n);
975 if (res)
976 break;
Al Viroc3c073f2012-08-21 22:32:06 -0400977 }
978 spin_unlock(&files->file_lock);
979 return res;
980}
981EXPORT_SYMBOL(iterate_fd);