blob: 93c5f89c248b07b6fba50fd31f4bd36ba9520fef [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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700151 if (!new_fdt)
152 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700153 /*
Al Viro5c598b32008-04-27 20:04:15 -0400154 * extremely unlikely race - sysctl_nr_open decreased between the check in
155 * caller and alloc_fdtable(). Cheaper to catch it here...
156 */
157 if (unlikely(new_fdt->max_fds <= nr)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700158 __free_fdtable(new_fdt);
Al Viro5c598b32008-04-27 20:04:15 -0400159 return -EMFILE;
160 }
161 /*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700162 * Check again since another task may have expanded the fd table while
163 * we dropped the lock
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700164 */
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700165 cur_fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800166 if (nr >= cur_fdt->max_fds) {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700167 /* Continue as planned */
168 copy_fdtable(new_fdt, cur_fdt);
169 rcu_assign_pointer(files->fdt, new_fdt);
Al Viroac3e3c52013-04-28 21:42:33 -0400170 if (cur_fdt != &files->fdtab)
Al Viro1983e782012-08-15 20:06:36 -0400171 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700172 } else {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700173 /* Somebody else expanded, so undo our attempt */
Changli Gaoa892e2d2010-08-10 18:01:35 -0700174 __free_fdtable(new_fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700175 }
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)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700189 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700191 fdt = files_fdtable(files);
Al Viro4e1e0182008-07-26 16:01:20 -0400192
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700193 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800194 if (nr < fdt->max_fds)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700195 return 0;
Al Viro4e1e0182008-07-26 16:01:20 -0400196
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700197 /* Can we expand? */
Eric Dumazet9cfe0152008-02-06 01:37:16 -0800198 if (nr >= sysctl_nr_open)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700199 return -EMFILE;
200
201 /* All good, so we try */
202 return expand_fdtable(files, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700204
Al Virob8318b02012-08-21 20:09:42 -0400205static inline void __set_close_on_exec(int fd, struct fdtable *fdt)
206{
207 __set_bit(fd, fdt->close_on_exec);
208}
209
210static inline void __clear_close_on_exec(int fd, struct fdtable *fdt)
211{
212 __clear_bit(fd, fdt->close_on_exec);
213}
214
215static inline void __set_open_fd(int fd, struct fdtable *fdt)
216{
217 __set_bit(fd, fdt->open_fds);
218}
219
220static inline void __clear_open_fd(int fd, struct fdtable *fdt)
221{
222 __clear_bit(fd, fdt->open_fds);
223}
224
Al Viro02afc6262008-05-08 19:42:56 -0400225static int count_open_files(struct fdtable *fdt)
226{
227 int size = fdt->max_fds;
228 int i;
229
230 /* Find the last open fd */
David Howells1fd36ad2012-02-16 17:49:54 +0000231 for (i = size / BITS_PER_LONG; i > 0; ) {
232 if (fdt->open_fds[--i])
Al Viro02afc6262008-05-08 19:42:56 -0400233 break;
234 }
David Howells1fd36ad2012-02-16 17:49:54 +0000235 i = (i + 1) * BITS_PER_LONG;
Al Viro02afc6262008-05-08 19:42:56 -0400236 return i;
237}
238
Al Viro02afc6262008-05-08 19:42:56 -0400239/*
240 * Allocate a new files structure and copy contents from the
241 * passed in files structure.
242 * errorp will be valid only when the returned files_struct is NULL.
243 */
244struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
245{
246 struct files_struct *newf;
247 struct file **old_fds, **new_fds;
248 int open_files, size, i;
249 struct fdtable *old_fdt, *new_fdt;
250
251 *errorp = -ENOMEM;
Al Viroafbec7f2008-05-08 21:11:17 -0400252 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
Al Viro02afc6262008-05-08 19:42:56 -0400253 if (!newf)
254 goto out;
255
Al Viroafbec7f2008-05-08 21:11:17 -0400256 atomic_set(&newf->count, 1);
257
258 spin_lock_init(&newf->file_lock);
259 newf->next_fd = 0;
260 new_fdt = &newf->fdtab;
261 new_fdt->max_fds = NR_OPEN_DEFAULT;
David Howells1fd36ad2012-02-16 17:49:54 +0000262 new_fdt->close_on_exec = newf->close_on_exec_init;
263 new_fdt->open_fds = newf->open_fds_init;
Al Viroafbec7f2008-05-08 21:11:17 -0400264 new_fdt->fd = &newf->fd_array[0];
Al Viroafbec7f2008-05-08 21:11:17 -0400265
Al Viro02afc6262008-05-08 19:42:56 -0400266 spin_lock(&oldf->file_lock);
267 old_fdt = files_fdtable(oldf);
Al Viro02afc6262008-05-08 19:42:56 -0400268 open_files = count_open_files(old_fdt);
269
270 /*
271 * Check whether we need to allocate a larger fd array and fd set.
Al Viro02afc6262008-05-08 19:42:56 -0400272 */
Al Viroadbecb12008-05-08 21:19:42 -0400273 while (unlikely(open_files > new_fdt->max_fds)) {
Al Viro02afc6262008-05-08 19:42:56 -0400274 spin_unlock(&oldf->file_lock);
Al Viro9dec3c42008-05-08 21:02:45 -0400275
Changli Gaoa892e2d2010-08-10 18:01:35 -0700276 if (new_fdt != &newf->fdtab)
277 __free_fdtable(new_fdt);
Al Viroadbecb12008-05-08 21:19:42 -0400278
Al Viro9dec3c42008-05-08 21:02:45 -0400279 new_fdt = alloc_fdtable(open_files - 1);
280 if (!new_fdt) {
281 *errorp = -ENOMEM;
Al Viro02afc6262008-05-08 19:42:56 -0400282 goto out_release;
Al Viro9dec3c42008-05-08 21:02:45 -0400283 }
284
285 /* beyond sysctl_nr_open; nothing to do */
286 if (unlikely(new_fdt->max_fds < open_files)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700287 __free_fdtable(new_fdt);
Al Viro9dec3c42008-05-08 21:02:45 -0400288 *errorp = -EMFILE;
289 goto out_release;
290 }
Al Viro9dec3c42008-05-08 21:02:45 -0400291
Al Viro02afc6262008-05-08 19:42:56 -0400292 /*
293 * Reacquire the oldf lock and a pointer to its fd table
294 * who knows it may have a new bigger fd table. We need
295 * the latest pointer.
296 */
297 spin_lock(&oldf->file_lock);
298 old_fdt = files_fdtable(oldf);
Al Viroadbecb12008-05-08 21:19:42 -0400299 open_files = count_open_files(old_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400300 }
301
302 old_fds = old_fdt->fd;
303 new_fds = new_fdt->fd;
304
David Howells1fd36ad2012-02-16 17:49:54 +0000305 memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8);
306 memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8);
Al Viro02afc6262008-05-08 19:42:56 -0400307
308 for (i = open_files; i != 0; i--) {
309 struct file *f = *old_fds++;
310 if (f) {
311 get_file(f);
312 } else {
313 /*
314 * The fd may be claimed in the fd bitmap but not yet
315 * instantiated in the files array if a sibling thread
316 * is partway through open(). So make sure that this
317 * fd is available to the new process.
318 */
David Howells1dce27c2012-02-16 17:49:42 +0000319 __clear_open_fd(open_files - i, new_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400320 }
321 rcu_assign_pointer(*new_fds++, f);
322 }
323 spin_unlock(&oldf->file_lock);
324
325 /* compute the remainder to be cleared */
326 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
327
328 /* This is long word aligned thus could use a optimized version */
329 memset(new_fds, 0, size);
330
331 if (new_fdt->max_fds > open_files) {
David Howells1fd36ad2012-02-16 17:49:54 +0000332 int left = (new_fdt->max_fds - open_files) / 8;
333 int start = open_files / BITS_PER_LONG;
Al Viro02afc6262008-05-08 19:42:56 -0400334
David Howells1fd36ad2012-02-16 17:49:54 +0000335 memset(&new_fdt->open_fds[start], 0, left);
336 memset(&new_fdt->close_on_exec[start], 0, left);
Al Viro02afc6262008-05-08 19:42:56 -0400337 }
338
Al Viroafbec7f2008-05-08 21:11:17 -0400339 rcu_assign_pointer(newf->fdt, new_fdt);
340
Al Viro02afc6262008-05-08 19:42:56 -0400341 return newf;
342
343out_release:
344 kmem_cache_free(files_cachep, newf);
345out:
346 return NULL;
347}
348
Oleg Nesterovce08b622014-01-11 19:19:53 +0100349static struct fdtable *close_files(struct files_struct * files)
Al Viro7cf4dc32012-08-15 19:56:12 -0400350{
Al Viro7cf4dc32012-08-15 19:56:12 -0400351 /*
352 * It is safe to dereference the fd table without RCU or
353 * ->file_lock because this is the last reference to the
Oleg Nesterovce08b622014-01-11 19:19:53 +0100354 * files structure.
Al Viro7cf4dc32012-08-15 19:56:12 -0400355 */
Oleg Nesterovce08b622014-01-11 19:19:53 +0100356 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
357 int i, j = 0;
358
Al Viro7cf4dc32012-08-15 19:56:12 -0400359 for (;;) {
360 unsigned long set;
361 i = j * BITS_PER_LONG;
362 if (i >= fdt->max_fds)
363 break;
364 set = fdt->open_fds[j++];
365 while (set) {
366 if (set & 1) {
367 struct file * file = xchg(&fdt->fd[i], NULL);
368 if (file) {
369 filp_close(file, files);
Paul E. McKenneybde6c3a2014-07-01 11:26:57 -0700370 cond_resched_rcu_qs();
Al Viro7cf4dc32012-08-15 19:56:12 -0400371 }
372 }
373 i++;
374 set >>= 1;
375 }
376 }
Oleg Nesterovce08b622014-01-11 19:19:53 +0100377
378 return fdt;
Al Viro7cf4dc32012-08-15 19:56:12 -0400379}
380
381struct files_struct *get_files_struct(struct task_struct *task)
382{
383 struct files_struct *files;
384
385 task_lock(task);
386 files = task->files;
387 if (files)
388 atomic_inc(&files->count);
389 task_unlock(task);
390
391 return files;
392}
393
394void put_files_struct(struct files_struct *files)
395{
Al Viro7cf4dc32012-08-15 19:56:12 -0400396 if (atomic_dec_and_test(&files->count)) {
Oleg Nesterovce08b622014-01-11 19:19:53 +0100397 struct fdtable *fdt = close_files(files);
398
Al Virob9e02af2012-08-15 20:00:58 -0400399 /* free the arrays if they are not embedded */
400 if (fdt != &files->fdtab)
401 __free_fdtable(fdt);
402 kmem_cache_free(files_cachep, files);
Al Viro7cf4dc32012-08-15 19:56:12 -0400403 }
404}
405
406void reset_files_struct(struct files_struct *files)
407{
408 struct task_struct *tsk = current;
409 struct files_struct *old;
410
411 old = tsk->files;
412 task_lock(tsk);
413 tsk->files = files;
414 task_unlock(tsk);
415 put_files_struct(old);
416}
417
418void exit_files(struct task_struct *tsk)
419{
420 struct files_struct * files = tsk->files;
421
422 if (files) {
423 task_lock(tsk);
424 tsk->files = NULL;
425 task_unlock(tsk);
426 put_files_struct(files);
427 }
428}
429
Al Virof52111b2008-05-08 18:19:16 -0400430struct files_struct init_files = {
431 .count = ATOMIC_INIT(1),
432 .fdt = &init_files.fdtab,
433 .fdtab = {
434 .max_fds = NR_OPEN_DEFAULT,
435 .fd = &init_files.fd_array[0],
David Howells1fd36ad2012-02-16 17:49:54 +0000436 .close_on_exec = init_files.close_on_exec_init,
437 .open_fds = init_files.open_fds_init,
Al Virof52111b2008-05-08 18:19:16 -0400438 },
Thomas Gleixnereece09e2011-07-17 21:25:03 +0200439 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
Al Virof52111b2008-05-08 18:19:16 -0400440};
Al Viro1027abe2008-07-30 04:13:04 -0400441
442/*
443 * allocate a file descriptor, mark it busy.
444 */
Al Virodcfadfa2012-08-12 17:27:30 -0400445int __alloc_fd(struct files_struct *files,
446 unsigned start, unsigned end, unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400447{
Al Viro1027abe2008-07-30 04:13:04 -0400448 unsigned int fd;
449 int error;
450 struct fdtable *fdt;
451
452 spin_lock(&files->file_lock);
453repeat:
454 fdt = files_fdtable(files);
455 fd = start;
456 if (fd < files->next_fd)
457 fd = files->next_fd;
458
459 if (fd < fdt->max_fds)
David Howells1fd36ad2012-02-16 17:49:54 +0000460 fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd);
Al Viro1027abe2008-07-30 04:13:04 -0400461
Al Virof33ff992012-08-12 16:17:59 -0400462 /*
463 * N.B. For clone tasks sharing a files structure, this test
464 * will limit the total number of files that can be opened.
465 */
466 error = -EMFILE;
467 if (fd >= end)
468 goto out;
469
Al Viro1027abe2008-07-30 04:13:04 -0400470 error = expand_files(files, fd);
471 if (error < 0)
472 goto out;
473
474 /*
475 * If we needed to expand the fs array we
476 * might have blocked - try again.
477 */
478 if (error)
479 goto repeat;
480
481 if (start <= files->next_fd)
482 files->next_fd = fd + 1;
483
David Howells1dce27c2012-02-16 17:49:42 +0000484 __set_open_fd(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400485 if (flags & O_CLOEXEC)
David Howells1dce27c2012-02-16 17:49:42 +0000486 __set_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400487 else
David Howells1dce27c2012-02-16 17:49:42 +0000488 __clear_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400489 error = fd;
490#if 1
491 /* Sanity check */
Paul E. McKenneyadd1f092014-02-12 12:51:09 -0800492 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
Al Viro1027abe2008-07-30 04:13:04 -0400493 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
494 rcu_assign_pointer(fdt->fd[fd], NULL);
495 }
496#endif
497
498out:
499 spin_unlock(&files->file_lock);
500 return error;
501}
502
Al Viroad47bd72012-08-21 20:11:34 -0400503static int alloc_fd(unsigned start, unsigned flags)
Al Virodcfadfa2012-08-12 17:27:30 -0400504{
505 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
506}
507
Al Viro1a7bd222012-08-12 17:18:05 -0400508int get_unused_fd_flags(unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400509{
Al Virodcfadfa2012-08-12 17:27:30 -0400510 return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags);
Al Viro1027abe2008-07-30 04:13:04 -0400511}
Al Viro1a7bd222012-08-12 17:18:05 -0400512EXPORT_SYMBOL(get_unused_fd_flags);
Al Viro56007ca2012-08-15 21:03:26 -0400513
514static void __put_unused_fd(struct files_struct *files, unsigned int fd)
515{
516 struct fdtable *fdt = files_fdtable(files);
517 __clear_open_fd(fd, fdt);
518 if (fd < files->next_fd)
519 files->next_fd = fd;
520}
521
522void put_unused_fd(unsigned int fd)
523{
524 struct files_struct *files = current->files;
525 spin_lock(&files->file_lock);
526 __put_unused_fd(files, fd);
527 spin_unlock(&files->file_lock);
528}
529
530EXPORT_SYMBOL(put_unused_fd);
531
532/*
533 * Install a file pointer in the fd array.
534 *
535 * The VFS is full of places where we drop the files lock between
536 * setting the open_fds bitmap and installing the file in the file
537 * array. At any such point, we are vulnerable to a dup2() race
538 * installing a file in the array before us. We need to detect this and
539 * fput() the struct file we are about to overwrite in this case.
540 *
541 * It should never happen - if we allow dup2() do it, _really_ bad things
542 * will follow.
Al Virof869e8a2012-08-15 21:06:33 -0400543 *
544 * NOTE: __fd_install() variant is really, really low-level; don't
545 * use it unless you are forced to by truly lousy API shoved down
546 * your throat. 'files' *MUST* be either current->files or obtained
547 * by get_files_struct(current) done by whoever had given it to you,
548 * or really bad things will happen. Normally you want to use
549 * fd_install() instead.
Al Viro56007ca2012-08-15 21:03:26 -0400550 */
551
Al Virof869e8a2012-08-15 21:06:33 -0400552void __fd_install(struct files_struct *files, unsigned int fd,
553 struct file *file)
Al Viro56007ca2012-08-15 21:03:26 -0400554{
Al Viro56007ca2012-08-15 21:03:26 -0400555 struct fdtable *fdt;
556 spin_lock(&files->file_lock);
557 fdt = files_fdtable(files);
558 BUG_ON(fdt->fd[fd] != NULL);
559 rcu_assign_pointer(fdt->fd[fd], file);
560 spin_unlock(&files->file_lock);
561}
562
Al Virof869e8a2012-08-15 21:06:33 -0400563void fd_install(unsigned int fd, struct file *file)
564{
565 __fd_install(current->files, fd, file);
566}
567
Al Viro56007ca2012-08-15 21:03:26 -0400568EXPORT_SYMBOL(fd_install);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400569
Al Viro483ce1d2012-08-19 12:04:24 -0400570/*
571 * The same warnings as for __alloc_fd()/__fd_install() apply here...
572 */
573int __close_fd(struct files_struct *files, unsigned fd)
574{
575 struct file *file;
576 struct fdtable *fdt;
577
578 spin_lock(&files->file_lock);
579 fdt = files_fdtable(files);
580 if (fd >= fdt->max_fds)
581 goto out_unlock;
582 file = fdt->fd[fd];
583 if (!file)
584 goto out_unlock;
585 rcu_assign_pointer(fdt->fd[fd], NULL);
586 __clear_close_on_exec(fd, fdt);
587 __put_unused_fd(files, fd);
588 spin_unlock(&files->file_lock);
589 return filp_close(file, files);
590
591out_unlock:
592 spin_unlock(&files->file_lock);
593 return -EBADF;
594}
595
Al Viro6a6d27d2012-08-21 09:56:33 -0400596void do_close_on_exec(struct files_struct *files)
597{
598 unsigned i;
599 struct fdtable *fdt;
600
601 /* exec unshares first */
Al Viro6a6d27d2012-08-21 09:56:33 -0400602 spin_lock(&files->file_lock);
603 for (i = 0; ; i++) {
604 unsigned long set;
605 unsigned fd = i * BITS_PER_LONG;
606 fdt = files_fdtable(files);
607 if (fd >= fdt->max_fds)
608 break;
609 set = fdt->close_on_exec[i];
610 if (!set)
611 continue;
612 fdt->close_on_exec[i] = 0;
613 for ( ; set ; fd++, set >>= 1) {
614 struct file *file;
615 if (!(set & 1))
616 continue;
617 file = fdt->fd[fd];
618 if (!file)
619 continue;
620 rcu_assign_pointer(fdt->fd[fd], NULL);
621 __put_unused_fd(files, fd);
622 spin_unlock(&files->file_lock);
623 filp_close(file, files);
624 cond_resched();
625 spin_lock(&files->file_lock);
626 }
627
628 }
629 spin_unlock(&files->file_lock);
630}
631
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100632static struct file *__fget(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400633{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400634 struct files_struct *files = current->files;
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100635 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400636
637 rcu_read_lock();
638 file = fcheck_files(files, fd);
639 if (file) {
640 /* File object ref couldn't be taken */
Konstantin Khlebnikov90f31d02015-04-16 12:47:56 -0700641 if ((file->f_mode & mask) || !get_file_rcu(file))
Al Viro0ee8cdf2012-08-15 21:12:10 -0400642 file = NULL;
643 }
644 rcu_read_unlock();
645
646 return file;
647}
648
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100649struct file *fget(unsigned int fd)
650{
651 return __fget(fd, FMODE_PATH);
652}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400653EXPORT_SYMBOL(fget);
654
655struct file *fget_raw(unsigned int fd)
656{
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100657 return __fget(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400658}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400659EXPORT_SYMBOL(fget_raw);
660
661/*
662 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
663 *
664 * You can use this instead of fget if you satisfy all of the following
665 * conditions:
666 * 1) You must call fput_light before exiting the syscall and returning control
667 * to userspace (i.e. you cannot remember the returned struct file * after
668 * returning to userspace).
669 * 2) You must not call filp_close on the returned struct file * in between
670 * calls to fget_light and fput_light.
671 * 3) You must not clone the current task in between the calls to fget_light
672 * and fput_light.
673 *
674 * The fput_needed flag returned by fget_light should be passed to the
675 * corresponding fput_light.
676 */
Al Virobd2a31d2014-03-04 14:54:22 -0500677static unsigned long __fget_light(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400678{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400679 struct files_struct *files = current->files;
Oleg Nesterovad461832014-01-13 16:48:40 +0100680 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400681
Al Viro0ee8cdf2012-08-15 21:12:10 -0400682 if (atomic_read(&files->count) == 1) {
Oleg Nesterova8d4b832014-01-11 19:19:32 +0100683 file = __fcheck_files(files, fd);
Al Virobd2a31d2014-03-04 14:54:22 -0500684 if (!file || unlikely(file->f_mode & mask))
685 return 0;
686 return (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400687 } else {
Oleg Nesterove6ff9a92014-01-13 16:49:06 +0100688 file = __fget(fd, mask);
Al Virobd2a31d2014-03-04 14:54:22 -0500689 if (!file)
690 return 0;
691 return FDPUT_FPUT | (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400692 }
Al Viro0ee8cdf2012-08-15 21:12:10 -0400693}
Al Virobd2a31d2014-03-04 14:54:22 -0500694unsigned long __fdget(unsigned int fd)
Oleg Nesterovad461832014-01-13 16:48:40 +0100695{
Al Virobd2a31d2014-03-04 14:54:22 -0500696 return __fget_light(fd, FMODE_PATH);
Oleg Nesterovad461832014-01-13 16:48:40 +0100697}
Al Virobd2a31d2014-03-04 14:54:22 -0500698EXPORT_SYMBOL(__fdget);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400699
Al Virobd2a31d2014-03-04 14:54:22 -0500700unsigned long __fdget_raw(unsigned int fd)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400701{
Al Virobd2a31d2014-03-04 14:54:22 -0500702 return __fget_light(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400703}
Al Virofe17f222012-08-21 11:48:11 -0400704
Al Virobd2a31d2014-03-04 14:54:22 -0500705unsigned long __fdget_pos(unsigned int fd)
706{
Eric Biggers99aea682014-03-16 15:47:48 -0500707 unsigned long v = __fdget(fd);
708 struct file *file = (struct file *)(v & ~3);
Al Virobd2a31d2014-03-04 14:54:22 -0500709
Eric Biggers99aea682014-03-16 15:47:48 -0500710 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
Al Virobd2a31d2014-03-04 14:54:22 -0500711 if (file_count(file) > 1) {
712 v |= FDPUT_POS_UNLOCK;
713 mutex_lock(&file->f_pos_lock);
714 }
715 }
Eric Biggers99aea682014-03-16 15:47:48 -0500716 return v;
Al Virobd2a31d2014-03-04 14:54:22 -0500717}
718
719/*
720 * We only lock f_pos if we have threads or if the file might be
721 * shared with another process. In both cases we'll have an elevated
722 * file count (done either by fdget() or by fork()).
723 */
724
Al Virofe17f222012-08-21 11:48:11 -0400725void set_close_on_exec(unsigned int fd, int flag)
726{
727 struct files_struct *files = current->files;
728 struct fdtable *fdt;
729 spin_lock(&files->file_lock);
730 fdt = files_fdtable(files);
731 if (flag)
732 __set_close_on_exec(fd, fdt);
733 else
734 __clear_close_on_exec(fd, fdt);
735 spin_unlock(&files->file_lock);
736}
737
738bool get_close_on_exec(unsigned int fd)
739{
740 struct files_struct *files = current->files;
741 struct fdtable *fdt;
742 bool res;
743 rcu_read_lock();
744 fdt = files_fdtable(files);
745 res = close_on_exec(fd, fdt);
746 rcu_read_unlock();
747 return res;
748}
749
Al Viro8280d162012-08-21 12:11:46 -0400750static int do_dup2(struct files_struct *files,
751 struct file *file, unsigned fd, unsigned flags)
Al Viroe9830942014-08-31 14:12:09 -0400752__releases(&files->file_lock)
Al Viro8280d162012-08-21 12:11:46 -0400753{
754 struct file *tofree;
755 struct fdtable *fdt;
756
757 /*
758 * We need to detect attempts to do dup2() over allocated but still
759 * not finished descriptor. NB: OpenBSD avoids that at the price of
760 * extra work in their equivalent of fget() - they insert struct
761 * file immediately after grabbing descriptor, mark it larval if
762 * more work (e.g. actual opening) is needed and make sure that
763 * fget() treats larval files as absent. Potentially interesting,
764 * but while extra work in fget() is trivial, locking implications
765 * and amount of surgery on open()-related paths in VFS are not.
766 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
767 * deadlocks in rather amusing ways, AFAICS. All of that is out of
768 * scope of POSIX or SUS, since neither considers shared descriptor
769 * tables and this condition does not arise without those.
770 */
771 fdt = files_fdtable(files);
772 tofree = fdt->fd[fd];
773 if (!tofree && fd_is_open(fd, fdt))
774 goto Ebusy;
775 get_file(file);
776 rcu_assign_pointer(fdt->fd[fd], file);
777 __set_open_fd(fd, fdt);
778 if (flags & O_CLOEXEC)
779 __set_close_on_exec(fd, fdt);
780 else
781 __clear_close_on_exec(fd, fdt);
782 spin_unlock(&files->file_lock);
783
784 if (tofree)
785 filp_close(tofree, files);
786
787 return fd;
788
789Ebusy:
790 spin_unlock(&files->file_lock);
791 return -EBUSY;
792}
793
794int replace_fd(unsigned fd, struct file *file, unsigned flags)
795{
796 int err;
797 struct files_struct *files = current->files;
798
799 if (!file)
800 return __close_fd(files, fd);
801
802 if (fd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000803 return -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400804
805 spin_lock(&files->file_lock);
806 err = expand_files(files, fd);
807 if (unlikely(err < 0))
808 goto out_unlock;
809 return do_dup2(files, file, fd, flags);
810
811out_unlock:
812 spin_unlock(&files->file_lock);
813 return err;
814}
815
Al Virofe17f222012-08-21 11:48:11 -0400816SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
817{
818 int err = -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400819 struct file *file;
820 struct files_struct *files = current->files;
Al Virofe17f222012-08-21 11:48:11 -0400821
822 if ((flags & ~O_CLOEXEC) != 0)
823 return -EINVAL;
824
Richard W.M. Jonesaed97642012-10-09 15:27:43 +0100825 if (unlikely(oldfd == newfd))
826 return -EINVAL;
827
Al Virofe17f222012-08-21 11:48:11 -0400828 if (newfd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000829 return -EBADF;
Al Virofe17f222012-08-21 11:48:11 -0400830
831 spin_lock(&files->file_lock);
832 err = expand_files(files, newfd);
833 file = fcheck(oldfd);
834 if (unlikely(!file))
835 goto Ebadf;
836 if (unlikely(err < 0)) {
837 if (err == -EMFILE)
838 goto Ebadf;
839 goto out_unlock;
840 }
Al Viro8280d162012-08-21 12:11:46 -0400841 return do_dup2(files, file, newfd, flags);
Al Virofe17f222012-08-21 11:48:11 -0400842
843Ebadf:
844 err = -EBADF;
845out_unlock:
846 spin_unlock(&files->file_lock);
847 return err;
848}
849
850SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
851{
852 if (unlikely(newfd == oldfd)) { /* corner case */
853 struct files_struct *files = current->files;
854 int retval = oldfd;
855
856 rcu_read_lock();
857 if (!fcheck_files(files, oldfd))
858 retval = -EBADF;
859 rcu_read_unlock();
860 return retval;
861 }
862 return sys_dup3(oldfd, newfd, 0);
863}
864
865SYSCALL_DEFINE1(dup, unsigned int, fildes)
866{
867 int ret = -EBADF;
868 struct file *file = fget_raw(fildes);
869
870 if (file) {
Yann Droneaud8d10a032014-12-10 15:45:44 -0800871 ret = get_unused_fd_flags(0);
Al Virofe17f222012-08-21 11:48:11 -0400872 if (ret >= 0)
873 fd_install(ret, file);
874 else
875 fput(file);
876 }
877 return ret;
878}
879
880int f_dupfd(unsigned int from, struct file *file, unsigned flags)
881{
882 int err;
883 if (from >= rlimit(RLIMIT_NOFILE))
884 return -EINVAL;
885 err = alloc_fd(from, flags);
886 if (err >= 0) {
887 get_file(file);
888 fd_install(err, file);
889 }
890 return err;
891}
Al Viroc3c073f2012-08-21 22:32:06 -0400892
893int iterate_fd(struct files_struct *files, unsigned n,
894 int (*f)(const void *, struct file *, unsigned),
895 const void *p)
896{
897 struct fdtable *fdt;
Al Viroc3c073f2012-08-21 22:32:06 -0400898 int res = 0;
899 if (!files)
900 return 0;
901 spin_lock(&files->file_lock);
Al Viroa77cfcb2012-11-29 22:57:33 -0500902 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
903 struct file *file;
904 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
905 if (!file)
906 continue;
907 res = f(p, file, n);
908 if (res)
909 break;
Al Viroc3c073f2012-08-21 22:32:06 -0400910 }
911 spin_unlock(&files->file_lock);
912 return res;
913}
914EXPORT_SYMBOL(iterate_fd);