blob: 6c672ad329e9a8c6bf3b7925f9e344ea1c5991c4 [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
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070059/*
60 * Expand the fdset in the files_struct. Called with the files spinlock
61 * held for write.
62 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080063static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070064{
Vadim Lobanov5466b452006-12-10 02:21:22 -080065 unsigned int cpy, set;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070066
Vadim Lobanov5466b452006-12-10 02:21:22 -080067 BUG_ON(nfdt->max_fds < ofdt->max_fds);
Vadim Lobanov5466b452006-12-10 02:21:22 -080068
69 cpy = ofdt->max_fds * sizeof(struct file *);
70 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
71 memcpy(nfdt->fd, ofdt->fd, cpy);
72 memset((char *)(nfdt->fd) + cpy, 0, set);
73
74 cpy = ofdt->max_fds / BITS_PER_BYTE;
75 set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
76 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
77 memset((char *)(nfdt->open_fds) + cpy, 0, set);
78 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
79 memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080}
81
Vadim Lobanov5466b452006-12-10 02:21:22 -080082static struct fdtable * alloc_fdtable(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070083{
Vadim Lobanov5466b452006-12-10 02:21:22 -080084 struct fdtable *fdt;
David Howells1fd36ad2012-02-16 17:49:54 +000085 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070087 /*
Vadim Lobanov5466b452006-12-10 02:21:22 -080088 * Figure out how many fds we actually want to support in this fdtable.
89 * Allocation steps are keyed to the size of the fdarray, since it
90 * grows far faster than any of the other dynamic data. We try to fit
91 * the fdarray into comfortable page-tuned chunks: starting at 1024B
92 * and growing in powers of two from there on.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070093 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080094 nr /= (1024 / sizeof(struct file *));
95 nr = roundup_pow_of_two(nr + 1);
96 nr *= (1024 / sizeof(struct file *));
Al Viro5c598b32008-04-27 20:04:15 -040097 /*
98 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
99 * had been set lower between the check in expand_files() and here. Deal
100 * with that in caller, it's cheaper that way.
101 *
102 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
103 * bitmaps handling below becomes unpleasant, to put it mildly...
104 */
105 if (unlikely(nr > sysctl_nr_open))
106 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800107
Vadim Lobanov5466b452006-12-10 02:21:22 -0800108 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
109 if (!fdt)
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800110 goto out;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800111 fdt->max_fds = nr;
112 data = alloc_fdmem(nr * sizeof(struct file *));
113 if (!data)
114 goto out_fdt;
David Howells1fd36ad2012-02-16 17:49:54 +0000115 fdt->fd = data;
116
117 data = alloc_fdmem(max_t(size_t,
Vadim Lobanov5466b452006-12-10 02:21:22 -0800118 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
119 if (!data)
120 goto out_arr;
David Howells1fd36ad2012-02-16 17:49:54 +0000121 fdt->open_fds = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800122 data += nr / BITS_PER_BYTE;
David Howells1fd36ad2012-02-16 17:49:54 +0000123 fdt->close_on_exec = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800124
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700125 return fdt;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800126
127out_arr:
Al Virof6c0a192014-04-23 10:18:46 -0400128 kvfree(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800129out_fdt:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700130 kfree(fdt);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800131out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700132 return NULL;
133}
134
135/*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700136 * Expand the file descriptor table.
137 * This function will allocate a new fdtable and both fd array and fdset, of
138 * the given size.
139 * Return <0 error code on error; 1 on successful completion.
140 * The files->file_lock should be held on entry, and will be held on exit.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700141 */
142static int expand_fdtable(struct files_struct *files, int nr)
143 __releases(files->file_lock)
144 __acquires(files->file_lock)
145{
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700146 struct fdtable *new_fdt, *cur_fdt;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700147
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 spin_unlock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700149 new_fdt = alloc_fdtable(nr);
Eric Dumazet8a812522015-06-30 15:54:08 +0200150
151 /* make sure all __fd_install() have seen resize_in_progress
152 * or have finished their rcu_read_lock_sched() section.
153 */
154 if (atomic_read(&files->count) > 1)
155 synchronize_sched();
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700158 if (!new_fdt)
159 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700160 /*
Al Viro5c598b32008-04-27 20:04:15 -0400161 * extremely unlikely race - sysctl_nr_open decreased between the check in
162 * caller and alloc_fdtable(). Cheaper to catch it here...
163 */
164 if (unlikely(new_fdt->max_fds <= nr)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700165 __free_fdtable(new_fdt);
Al Viro5c598b32008-04-27 20:04:15 -0400166 return -EMFILE;
167 }
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700168 cur_fdt = files_fdtable(files);
Eric Dumazet8a812522015-06-30 15:54:08 +0200169 BUG_ON(nr < cur_fdt->max_fds);
170 copy_fdtable(new_fdt, cur_fdt);
171 rcu_assign_pointer(files->fdt, new_fdt);
172 if (cur_fdt != &files->fdtab)
173 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
174 /* coupled with smp_rmb() in __fd_install() */
175 smp_wmb();
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700176 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177}
178
179/*
180 * Expand files.
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700181 * This function will expand the file structures, if the requested size exceeds
182 * the current capacity and there is room for expansion.
183 * Return <0 error code on error; 0 when nothing done; 1 when files were
184 * expanded and execution may have blocked.
185 * The files->file_lock should be held on entry, and will be held on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 */
Al Viroad47bd72012-08-21 20:11:34 -0400187static int expand_files(struct files_struct *files, int nr)
Eric Dumazet8a812522015-06-30 15:54:08 +0200188 __releases(files->file_lock)
189 __acquires(files->file_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700191 struct fdtable *fdt;
Eric Dumazet8a812522015-06-30 15:54:08 +0200192 int expanded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Eric Dumazet8a812522015-06-30 15:54:08 +0200194repeat:
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700195 fdt = files_fdtable(files);
Al Viro4e1e0182008-07-26 16:01:20 -0400196
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700197 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800198 if (nr < fdt->max_fds)
Eric Dumazet8a812522015-06-30 15:54:08 +0200199 return expanded;
Al Viro4e1e0182008-07-26 16:01:20 -0400200
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700201 /* Can we expand? */
Eric Dumazet9cfe0152008-02-06 01:37:16 -0800202 if (nr >= sysctl_nr_open)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700203 return -EMFILE;
204
Eric Dumazet8a812522015-06-30 15:54:08 +0200205 if (unlikely(files->resize_in_progress)) {
206 spin_unlock(&files->file_lock);
207 expanded = 1;
208 wait_event(files->resize_wait, !files->resize_in_progress);
209 spin_lock(&files->file_lock);
210 goto repeat;
211 }
212
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700213 /* All good, so we try */
Eric Dumazet8a812522015-06-30 15:54:08 +0200214 files->resize_in_progress = true;
215 expanded = expand_fdtable(files, nr);
216 files->resize_in_progress = false;
217
218 wake_up_all(&files->resize_wait);
219 return expanded;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700221
Al Virob8318b02012-08-21 20:09:42 -0400222static inline void __set_close_on_exec(int fd, struct fdtable *fdt)
223{
224 __set_bit(fd, fdt->close_on_exec);
225}
226
227static inline void __clear_close_on_exec(int fd, struct fdtable *fdt)
228{
229 __clear_bit(fd, fdt->close_on_exec);
230}
231
232static inline void __set_open_fd(int fd, struct fdtable *fdt)
233{
234 __set_bit(fd, fdt->open_fds);
235}
236
237static inline void __clear_open_fd(int fd, struct fdtable *fdt)
238{
239 __clear_bit(fd, fdt->open_fds);
240}
241
Al Viro02afc6262008-05-08 19:42:56 -0400242static int count_open_files(struct fdtable *fdt)
243{
244 int size = fdt->max_fds;
245 int i;
246
247 /* Find the last open fd */
David Howells1fd36ad2012-02-16 17:49:54 +0000248 for (i = size / BITS_PER_LONG; i > 0; ) {
249 if (fdt->open_fds[--i])
Al Viro02afc6262008-05-08 19:42:56 -0400250 break;
251 }
David Howells1fd36ad2012-02-16 17:49:54 +0000252 i = (i + 1) * BITS_PER_LONG;
Al Viro02afc6262008-05-08 19:42:56 -0400253 return i;
254}
255
Al Viro02afc6262008-05-08 19:42:56 -0400256/*
257 * Allocate a new files structure and copy contents from the
258 * passed in files structure.
259 * errorp will be valid only when the returned files_struct is NULL.
260 */
261struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
262{
263 struct files_struct *newf;
264 struct file **old_fds, **new_fds;
265 int open_files, size, i;
266 struct fdtable *old_fdt, *new_fdt;
267
268 *errorp = -ENOMEM;
Al Viroafbec7f2008-05-08 21:11:17 -0400269 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
Al Viro02afc6262008-05-08 19:42:56 -0400270 if (!newf)
271 goto out;
272
Al Viroafbec7f2008-05-08 21:11:17 -0400273 atomic_set(&newf->count, 1);
274
275 spin_lock_init(&newf->file_lock);
Eric Dumazet8a812522015-06-30 15:54:08 +0200276 newf->resize_in_progress = false;
277 init_waitqueue_head(&newf->resize_wait);
Al Viroafbec7f2008-05-08 21:11:17 -0400278 newf->next_fd = 0;
279 new_fdt = &newf->fdtab;
280 new_fdt->max_fds = NR_OPEN_DEFAULT;
David Howells1fd36ad2012-02-16 17:49:54 +0000281 new_fdt->close_on_exec = newf->close_on_exec_init;
282 new_fdt->open_fds = newf->open_fds_init;
Al Viroafbec7f2008-05-08 21:11:17 -0400283 new_fdt->fd = &newf->fd_array[0];
Al Viroafbec7f2008-05-08 21:11:17 -0400284
Al Viro02afc6262008-05-08 19:42:56 -0400285 spin_lock(&oldf->file_lock);
286 old_fdt = files_fdtable(oldf);
Al Viro02afc6262008-05-08 19:42:56 -0400287 open_files = count_open_files(old_fdt);
288
289 /*
290 * Check whether we need to allocate a larger fd array and fd set.
Al Viro02afc6262008-05-08 19:42:56 -0400291 */
Al Viroadbecb12008-05-08 21:19:42 -0400292 while (unlikely(open_files > new_fdt->max_fds)) {
Al Viro02afc6262008-05-08 19:42:56 -0400293 spin_unlock(&oldf->file_lock);
Al Viro9dec3c42008-05-08 21:02:45 -0400294
Changli Gaoa892e2d2010-08-10 18:01:35 -0700295 if (new_fdt != &newf->fdtab)
296 __free_fdtable(new_fdt);
Al Viroadbecb12008-05-08 21:19:42 -0400297
Al Viro9dec3c42008-05-08 21:02:45 -0400298 new_fdt = alloc_fdtable(open_files - 1);
299 if (!new_fdt) {
300 *errorp = -ENOMEM;
Al Viro02afc6262008-05-08 19:42:56 -0400301 goto out_release;
Al Viro9dec3c42008-05-08 21:02:45 -0400302 }
303
304 /* beyond sysctl_nr_open; nothing to do */
305 if (unlikely(new_fdt->max_fds < open_files)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700306 __free_fdtable(new_fdt);
Al Viro9dec3c42008-05-08 21:02:45 -0400307 *errorp = -EMFILE;
308 goto out_release;
309 }
Al Viro9dec3c42008-05-08 21:02:45 -0400310
Al Viro02afc6262008-05-08 19:42:56 -0400311 /*
312 * Reacquire the oldf lock and a pointer to its fd table
313 * who knows it may have a new bigger fd table. We need
314 * the latest pointer.
315 */
316 spin_lock(&oldf->file_lock);
317 old_fdt = files_fdtable(oldf);
Al Viroadbecb12008-05-08 21:19:42 -0400318 open_files = count_open_files(old_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400319 }
320
321 old_fds = old_fdt->fd;
322 new_fds = new_fdt->fd;
323
David Howells1fd36ad2012-02-16 17:49:54 +0000324 memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8);
325 memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8);
Al Viro02afc6262008-05-08 19:42:56 -0400326
327 for (i = open_files; i != 0; i--) {
328 struct file *f = *old_fds++;
329 if (f) {
330 get_file(f);
331 } else {
332 /*
333 * The fd may be claimed in the fd bitmap but not yet
334 * instantiated in the files array if a sibling thread
335 * is partway through open(). So make sure that this
336 * fd is available to the new process.
337 */
David Howells1dce27c2012-02-16 17:49:42 +0000338 __clear_open_fd(open_files - i, new_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400339 }
340 rcu_assign_pointer(*new_fds++, f);
341 }
342 spin_unlock(&oldf->file_lock);
343
344 /* compute the remainder to be cleared */
345 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
346
347 /* This is long word aligned thus could use a optimized version */
348 memset(new_fds, 0, size);
349
350 if (new_fdt->max_fds > open_files) {
David Howells1fd36ad2012-02-16 17:49:54 +0000351 int left = (new_fdt->max_fds - open_files) / 8;
352 int start = open_files / BITS_PER_LONG;
Al Viro02afc6262008-05-08 19:42:56 -0400353
David Howells1fd36ad2012-02-16 17:49:54 +0000354 memset(&new_fdt->open_fds[start], 0, left);
355 memset(&new_fdt->close_on_exec[start], 0, left);
Al Viro02afc6262008-05-08 19:42:56 -0400356 }
357
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);
376 int i, j = 0;
377
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. McKenneybde6c3a2014-07-01 11:26:57 -0700389 cond_resched_rcu_qs();
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,
Al Virof52111b2008-05-08 18:19:16 -0400457 },
Thomas Gleixnereece09e2011-07-17 21:25:03 +0200458 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
Al Virof52111b2008-05-08 18:19:16 -0400459};
Al Viro1027abe2008-07-30 04:13:04 -0400460
461/*
462 * allocate a file descriptor, mark it busy.
463 */
Al Virodcfadfa2012-08-12 17:27:30 -0400464int __alloc_fd(struct files_struct *files,
465 unsigned start, unsigned end, unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400466{
Al Viro1027abe2008-07-30 04:13:04 -0400467 unsigned int fd;
468 int error;
469 struct fdtable *fdt;
470
471 spin_lock(&files->file_lock);
472repeat:
473 fdt = files_fdtable(files);
474 fd = start;
475 if (fd < files->next_fd)
476 fd = files->next_fd;
477
478 if (fd < fdt->max_fds)
David Howells1fd36ad2012-02-16 17:49:54 +0000479 fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd);
Al Viro1027abe2008-07-30 04:13:04 -0400480
Al Virof33ff992012-08-12 16:17:59 -0400481 /*
482 * N.B. For clone tasks sharing a files structure, this test
483 * will limit the total number of files that can be opened.
484 */
485 error = -EMFILE;
486 if (fd >= end)
487 goto out;
488
Al Viro1027abe2008-07-30 04:13:04 -0400489 error = expand_files(files, fd);
490 if (error < 0)
491 goto out;
492
493 /*
494 * If we needed to expand the fs array we
495 * might have blocked - try again.
496 */
497 if (error)
498 goto repeat;
499
500 if (start <= files->next_fd)
501 files->next_fd = fd + 1;
502
David Howells1dce27c2012-02-16 17:49:42 +0000503 __set_open_fd(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400504 if (flags & O_CLOEXEC)
David Howells1dce27c2012-02-16 17:49:42 +0000505 __set_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400506 else
David Howells1dce27c2012-02-16 17:49:42 +0000507 __clear_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400508 error = fd;
509#if 1
510 /* Sanity check */
Paul E. McKenneyadd1f092014-02-12 12:51:09 -0800511 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
Al Viro1027abe2008-07-30 04:13:04 -0400512 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
513 rcu_assign_pointer(fdt->fd[fd], NULL);
514 }
515#endif
516
517out:
518 spin_unlock(&files->file_lock);
519 return error;
520}
521
Al Viroad47bd72012-08-21 20:11:34 -0400522static int alloc_fd(unsigned start, unsigned flags)
Al Virodcfadfa2012-08-12 17:27:30 -0400523{
524 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
525}
526
Al Viro1a7bd222012-08-12 17:18:05 -0400527int get_unused_fd_flags(unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400528{
Al Virodcfadfa2012-08-12 17:27:30 -0400529 return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags);
Al Viro1027abe2008-07-30 04:13:04 -0400530}
Al Viro1a7bd222012-08-12 17:18:05 -0400531EXPORT_SYMBOL(get_unused_fd_flags);
Al Viro56007ca2012-08-15 21:03:26 -0400532
533static void __put_unused_fd(struct files_struct *files, unsigned int fd)
534{
535 struct fdtable *fdt = files_fdtable(files);
536 __clear_open_fd(fd, fdt);
537 if (fd < files->next_fd)
538 files->next_fd = fd;
539}
540
541void put_unused_fd(unsigned int fd)
542{
543 struct files_struct *files = current->files;
544 spin_lock(&files->file_lock);
545 __put_unused_fd(files, fd);
546 spin_unlock(&files->file_lock);
547}
548
549EXPORT_SYMBOL(put_unused_fd);
550
551/*
552 * Install a file pointer in the fd array.
553 *
554 * The VFS is full of places where we drop the files lock between
555 * setting the open_fds bitmap and installing the file in the file
556 * array. At any such point, we are vulnerable to a dup2() race
557 * installing a file in the array before us. We need to detect this and
558 * fput() the struct file we are about to overwrite in this case.
559 *
560 * It should never happen - if we allow dup2() do it, _really_ bad things
561 * will follow.
Al Virof869e8a2012-08-15 21:06:33 -0400562 *
563 * NOTE: __fd_install() variant is really, really low-level; don't
564 * use it unless you are forced to by truly lousy API shoved down
565 * your throat. 'files' *MUST* be either current->files or obtained
566 * by get_files_struct(current) done by whoever had given it to you,
567 * or really bad things will happen. Normally you want to use
568 * fd_install() instead.
Al Viro56007ca2012-08-15 21:03:26 -0400569 */
570
Al Virof869e8a2012-08-15 21:06:33 -0400571void __fd_install(struct files_struct *files, unsigned int fd,
572 struct file *file)
Al Viro56007ca2012-08-15 21:03:26 -0400573{
Al Viro56007ca2012-08-15 21:03:26 -0400574 struct fdtable *fdt;
Eric Dumazet8a812522015-06-30 15:54:08 +0200575
576 might_sleep();
577 rcu_read_lock_sched();
578
579 while (unlikely(files->resize_in_progress)) {
580 rcu_read_unlock_sched();
581 wait_event(files->resize_wait, !files->resize_in_progress);
582 rcu_read_lock_sched();
583 }
584 /* coupled with smp_wmb() in expand_fdtable() */
585 smp_rmb();
586 fdt = rcu_dereference_sched(files->fdt);
Al Viro56007ca2012-08-15 21:03:26 -0400587 BUG_ON(fdt->fd[fd] != NULL);
588 rcu_assign_pointer(fdt->fd[fd], file);
Eric Dumazet8a812522015-06-30 15:54:08 +0200589 rcu_read_unlock_sched();
Al Viro56007ca2012-08-15 21:03:26 -0400590}
591
Al Virof869e8a2012-08-15 21:06:33 -0400592void fd_install(unsigned int fd, struct file *file)
593{
594 __fd_install(current->files, fd, file);
595}
596
Al Viro56007ca2012-08-15 21:03:26 -0400597EXPORT_SYMBOL(fd_install);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400598
Al Viro483ce1d2012-08-19 12:04:24 -0400599/*
600 * The same warnings as for __alloc_fd()/__fd_install() apply here...
601 */
602int __close_fd(struct files_struct *files, unsigned fd)
603{
604 struct file *file;
605 struct fdtable *fdt;
606
607 spin_lock(&files->file_lock);
608 fdt = files_fdtable(files);
609 if (fd >= fdt->max_fds)
610 goto out_unlock;
611 file = fdt->fd[fd];
612 if (!file)
613 goto out_unlock;
614 rcu_assign_pointer(fdt->fd[fd], NULL);
615 __clear_close_on_exec(fd, fdt);
616 __put_unused_fd(files, fd);
617 spin_unlock(&files->file_lock);
618 return filp_close(file, files);
619
620out_unlock:
621 spin_unlock(&files->file_lock);
622 return -EBADF;
623}
624
Al Viro6a6d27d2012-08-21 09:56:33 -0400625void do_close_on_exec(struct files_struct *files)
626{
627 unsigned i;
628 struct fdtable *fdt;
629
630 /* exec unshares first */
Al Viro6a6d27d2012-08-21 09:56:33 -0400631 spin_lock(&files->file_lock);
632 for (i = 0; ; i++) {
633 unsigned long set;
634 unsigned fd = i * BITS_PER_LONG;
635 fdt = files_fdtable(files);
636 if (fd >= fdt->max_fds)
637 break;
638 set = fdt->close_on_exec[i];
639 if (!set)
640 continue;
641 fdt->close_on_exec[i] = 0;
642 for ( ; set ; fd++, set >>= 1) {
643 struct file *file;
644 if (!(set & 1))
645 continue;
646 file = fdt->fd[fd];
647 if (!file)
648 continue;
649 rcu_assign_pointer(fdt->fd[fd], NULL);
650 __put_unused_fd(files, fd);
651 spin_unlock(&files->file_lock);
652 filp_close(file, files);
653 cond_resched();
654 spin_lock(&files->file_lock);
655 }
656
657 }
658 spin_unlock(&files->file_lock);
659}
660
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100661static struct file *__fget(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400662{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400663 struct files_struct *files = current->files;
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100664 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400665
666 rcu_read_lock();
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200667loop:
Al Viro0ee8cdf2012-08-15 21:12:10 -0400668 file = fcheck_files(files, fd);
669 if (file) {
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200670 /* File object ref couldn't be taken.
671 * dup2() atomicity guarantee is the reason
672 * we loop to catch the new file (or NULL pointer)
673 */
674 if (file->f_mode & mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400675 file = NULL;
Eric Dumazet5ba97d22015-06-29 17:10:30 +0200676 else if (!get_file_rcu(file))
677 goto loop;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400678 }
679 rcu_read_unlock();
680
681 return file;
682}
683
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100684struct file *fget(unsigned int fd)
685{
686 return __fget(fd, FMODE_PATH);
687}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400688EXPORT_SYMBOL(fget);
689
690struct file *fget_raw(unsigned int fd)
691{
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100692 return __fget(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400693}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400694EXPORT_SYMBOL(fget_raw);
695
696/*
697 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
698 *
699 * You can use this instead of fget if you satisfy all of the following
700 * conditions:
701 * 1) You must call fput_light before exiting the syscall and returning control
702 * to userspace (i.e. you cannot remember the returned struct file * after
703 * returning to userspace).
704 * 2) You must not call filp_close on the returned struct file * in between
705 * calls to fget_light and fput_light.
706 * 3) You must not clone the current task in between the calls to fget_light
707 * and fput_light.
708 *
709 * The fput_needed flag returned by fget_light should be passed to the
710 * corresponding fput_light.
711 */
Al Virobd2a31d2014-03-04 14:54:22 -0500712static unsigned long __fget_light(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400713{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400714 struct files_struct *files = current->files;
Oleg Nesterovad461832014-01-13 16:48:40 +0100715 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400716
Al Viro0ee8cdf2012-08-15 21:12:10 -0400717 if (atomic_read(&files->count) == 1) {
Oleg Nesterova8d4b832014-01-11 19:19:32 +0100718 file = __fcheck_files(files, fd);
Al Virobd2a31d2014-03-04 14:54:22 -0500719 if (!file || unlikely(file->f_mode & mask))
720 return 0;
721 return (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400722 } else {
Oleg Nesterove6ff9a92014-01-13 16:49:06 +0100723 file = __fget(fd, mask);
Al Virobd2a31d2014-03-04 14:54:22 -0500724 if (!file)
725 return 0;
726 return FDPUT_FPUT | (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400727 }
Al Viro0ee8cdf2012-08-15 21:12:10 -0400728}
Al Virobd2a31d2014-03-04 14:54:22 -0500729unsigned long __fdget(unsigned int fd)
Oleg Nesterovad461832014-01-13 16:48:40 +0100730{
Al Virobd2a31d2014-03-04 14:54:22 -0500731 return __fget_light(fd, FMODE_PATH);
Oleg Nesterovad461832014-01-13 16:48:40 +0100732}
Al Virobd2a31d2014-03-04 14:54:22 -0500733EXPORT_SYMBOL(__fdget);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400734
Al Virobd2a31d2014-03-04 14:54:22 -0500735unsigned long __fdget_raw(unsigned int fd)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400736{
Al Virobd2a31d2014-03-04 14:54:22 -0500737 return __fget_light(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400738}
Al Virofe17f222012-08-21 11:48:11 -0400739
Al Virobd2a31d2014-03-04 14:54:22 -0500740unsigned long __fdget_pos(unsigned int fd)
741{
Eric Biggers99aea682014-03-16 15:47:48 -0500742 unsigned long v = __fdget(fd);
743 struct file *file = (struct file *)(v & ~3);
Al Virobd2a31d2014-03-04 14:54:22 -0500744
Eric Biggers99aea682014-03-16 15:47:48 -0500745 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
Al Virobd2a31d2014-03-04 14:54:22 -0500746 if (file_count(file) > 1) {
747 v |= FDPUT_POS_UNLOCK;
748 mutex_lock(&file->f_pos_lock);
749 }
750 }
Eric Biggers99aea682014-03-16 15:47:48 -0500751 return v;
Al Virobd2a31d2014-03-04 14:54:22 -0500752}
753
754/*
755 * We only lock f_pos if we have threads or if the file might be
756 * shared with another process. In both cases we'll have an elevated
757 * file count (done either by fdget() or by fork()).
758 */
759
Al Virofe17f222012-08-21 11:48:11 -0400760void set_close_on_exec(unsigned int fd, int flag)
761{
762 struct files_struct *files = current->files;
763 struct fdtable *fdt;
764 spin_lock(&files->file_lock);
765 fdt = files_fdtable(files);
766 if (flag)
767 __set_close_on_exec(fd, fdt);
768 else
769 __clear_close_on_exec(fd, fdt);
770 spin_unlock(&files->file_lock);
771}
772
773bool get_close_on_exec(unsigned int fd)
774{
775 struct files_struct *files = current->files;
776 struct fdtable *fdt;
777 bool res;
778 rcu_read_lock();
779 fdt = files_fdtable(files);
780 res = close_on_exec(fd, fdt);
781 rcu_read_unlock();
782 return res;
783}
784
Al Viro8280d162012-08-21 12:11:46 -0400785static int do_dup2(struct files_struct *files,
786 struct file *file, unsigned fd, unsigned flags)
Al Viroe9830942014-08-31 14:12:09 -0400787__releases(&files->file_lock)
Al Viro8280d162012-08-21 12:11:46 -0400788{
789 struct file *tofree;
790 struct fdtable *fdt;
791
792 /*
793 * We need to detect attempts to do dup2() over allocated but still
794 * not finished descriptor. NB: OpenBSD avoids that at the price of
795 * extra work in their equivalent of fget() - they insert struct
796 * file immediately after grabbing descriptor, mark it larval if
797 * more work (e.g. actual opening) is needed and make sure that
798 * fget() treats larval files as absent. Potentially interesting,
799 * but while extra work in fget() is trivial, locking implications
800 * and amount of surgery on open()-related paths in VFS are not.
801 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
802 * deadlocks in rather amusing ways, AFAICS. All of that is out of
803 * scope of POSIX or SUS, since neither considers shared descriptor
804 * tables and this condition does not arise without those.
805 */
806 fdt = files_fdtable(files);
807 tofree = fdt->fd[fd];
808 if (!tofree && fd_is_open(fd, fdt))
809 goto Ebusy;
810 get_file(file);
811 rcu_assign_pointer(fdt->fd[fd], file);
812 __set_open_fd(fd, fdt);
813 if (flags & O_CLOEXEC)
814 __set_close_on_exec(fd, fdt);
815 else
816 __clear_close_on_exec(fd, fdt);
817 spin_unlock(&files->file_lock);
818
819 if (tofree)
820 filp_close(tofree, files);
821
822 return fd;
823
824Ebusy:
825 spin_unlock(&files->file_lock);
826 return -EBUSY;
827}
828
829int replace_fd(unsigned fd, struct file *file, unsigned flags)
830{
831 int err;
832 struct files_struct *files = current->files;
833
834 if (!file)
835 return __close_fd(files, fd);
836
837 if (fd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000838 return -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400839
840 spin_lock(&files->file_lock);
841 err = expand_files(files, fd);
842 if (unlikely(err < 0))
843 goto out_unlock;
844 return do_dup2(files, file, fd, flags);
845
846out_unlock:
847 spin_unlock(&files->file_lock);
848 return err;
849}
850
Al Virofe17f222012-08-21 11:48:11 -0400851SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
852{
853 int err = -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400854 struct file *file;
855 struct files_struct *files = current->files;
Al Virofe17f222012-08-21 11:48:11 -0400856
857 if ((flags & ~O_CLOEXEC) != 0)
858 return -EINVAL;
859
Richard W.M. Jonesaed97642012-10-09 15:27:43 +0100860 if (unlikely(oldfd == newfd))
861 return -EINVAL;
862
Al Virofe17f222012-08-21 11:48:11 -0400863 if (newfd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000864 return -EBADF;
Al Virofe17f222012-08-21 11:48:11 -0400865
866 spin_lock(&files->file_lock);
867 err = expand_files(files, newfd);
868 file = fcheck(oldfd);
869 if (unlikely(!file))
870 goto Ebadf;
871 if (unlikely(err < 0)) {
872 if (err == -EMFILE)
873 goto Ebadf;
874 goto out_unlock;
875 }
Al Viro8280d162012-08-21 12:11:46 -0400876 return do_dup2(files, file, newfd, flags);
Al Virofe17f222012-08-21 11:48:11 -0400877
878Ebadf:
879 err = -EBADF;
880out_unlock:
881 spin_unlock(&files->file_lock);
882 return err;
883}
884
885SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
886{
887 if (unlikely(newfd == oldfd)) { /* corner case */
888 struct files_struct *files = current->files;
889 int retval = oldfd;
890
891 rcu_read_lock();
892 if (!fcheck_files(files, oldfd))
893 retval = -EBADF;
894 rcu_read_unlock();
895 return retval;
896 }
897 return sys_dup3(oldfd, newfd, 0);
898}
899
900SYSCALL_DEFINE1(dup, unsigned int, fildes)
901{
902 int ret = -EBADF;
903 struct file *file = fget_raw(fildes);
904
905 if (file) {
Yann Droneaud8d10a032014-12-10 15:45:44 -0800906 ret = get_unused_fd_flags(0);
Al Virofe17f222012-08-21 11:48:11 -0400907 if (ret >= 0)
908 fd_install(ret, file);
909 else
910 fput(file);
911 }
912 return ret;
913}
914
915int f_dupfd(unsigned int from, struct file *file, unsigned flags)
916{
917 int err;
918 if (from >= rlimit(RLIMIT_NOFILE))
919 return -EINVAL;
920 err = alloc_fd(from, flags);
921 if (err >= 0) {
922 get_file(file);
923 fd_install(err, file);
924 }
925 return err;
926}
Al Viroc3c073f2012-08-21 22:32:06 -0400927
928int iterate_fd(struct files_struct *files, unsigned n,
929 int (*f)(const void *, struct file *, unsigned),
930 const void *p)
931{
932 struct fdtable *fdt;
Al Viroc3c073f2012-08-21 22:32:06 -0400933 int res = 0;
934 if (!files)
935 return 0;
936 spin_lock(&files->file_lock);
Al Viroa77cfcb2012-11-29 22:57:33 -0500937 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
938 struct file *file;
939 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
940 if (!file)
941 continue;
942 res = f(p, file, n);
943 if (res)
944 break;
Al Viroc3c073f2012-08-21 22:32:06 -0400945 }
946 spin_unlock(&files->file_lock);
947 return res;
948}
949EXPORT_SYMBOL(iterate_fd);