blob: 8f294cfac69749024c2c2e19d4b156755130e9ed [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_fdmem(void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070048{
Changli Gaoa892e2d2010-08-10 18:01:35 -070049 is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr);
Vadim Lobanov5466b452006-12-10 02:21:22 -080050}
51
Changli Gaoa892e2d2010-08-10 18:01:35 -070052static void __free_fdtable(struct fdtable *fdt)
Vadim Lobanov5466b452006-12-10 02:21:22 -080053{
Changli Gaoa892e2d2010-08-10 18:01:35 -070054 free_fdmem(fdt->fd);
55 free_fdmem(fdt->open_fds);
56 kfree(fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070057}
58
Al Viro7cf4dc32012-08-15 19:56:12 -040059static void free_fdtable_rcu(struct rcu_head *rcu)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070060{
Al Viroac3e3c52013-04-28 21:42:33 -040061 __free_fdtable(container_of(rcu, struct fdtable, rcu));
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070062}
63
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070064/*
65 * Expand the fdset in the files_struct. Called with the files spinlock
66 * held for write.
67 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080068static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070069{
Vadim Lobanov5466b452006-12-10 02:21:22 -080070 unsigned int cpy, set;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070071
Vadim Lobanov5466b452006-12-10 02:21:22 -080072 BUG_ON(nfdt->max_fds < ofdt->max_fds);
Vadim Lobanov5466b452006-12-10 02:21:22 -080073
74 cpy = ofdt->max_fds * sizeof(struct file *);
75 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
76 memcpy(nfdt->fd, ofdt->fd, cpy);
77 memset((char *)(nfdt->fd) + cpy, 0, set);
78
79 cpy = ofdt->max_fds / BITS_PER_BYTE;
80 set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
81 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
82 memset((char *)(nfdt->open_fds) + cpy, 0, set);
83 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
84 memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085}
86
Vadim Lobanov5466b452006-12-10 02:21:22 -080087static struct fdtable * alloc_fdtable(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Vadim Lobanov5466b452006-12-10 02:21:22 -080089 struct fdtable *fdt;
David Howells1fd36ad2012-02-16 17:49:54 +000090 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070092 /*
Vadim Lobanov5466b452006-12-10 02:21:22 -080093 * Figure out how many fds we actually want to support in this fdtable.
94 * Allocation steps are keyed to the size of the fdarray, since it
95 * grows far faster than any of the other dynamic data. We try to fit
96 * the fdarray into comfortable page-tuned chunks: starting at 1024B
97 * and growing in powers of two from there on.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070098 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080099 nr /= (1024 / sizeof(struct file *));
100 nr = roundup_pow_of_two(nr + 1);
101 nr *= (1024 / sizeof(struct file *));
Al Viro5c598b32008-04-27 20:04:15 -0400102 /*
103 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
104 * had been set lower between the check in expand_files() and here. Deal
105 * with that in caller, it's cheaper that way.
106 *
107 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
108 * bitmaps handling below becomes unpleasant, to put it mildly...
109 */
110 if (unlikely(nr > sysctl_nr_open))
111 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800112
Vadim Lobanov5466b452006-12-10 02:21:22 -0800113 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
114 if (!fdt)
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800115 goto out;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800116 fdt->max_fds = nr;
117 data = alloc_fdmem(nr * sizeof(struct file *));
118 if (!data)
119 goto out_fdt;
David Howells1fd36ad2012-02-16 17:49:54 +0000120 fdt->fd = data;
121
122 data = alloc_fdmem(max_t(size_t,
Vadim Lobanov5466b452006-12-10 02:21:22 -0800123 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
124 if (!data)
125 goto out_arr;
David Howells1fd36ad2012-02-16 17:49:54 +0000126 fdt->open_fds = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800127 data += nr / BITS_PER_BYTE;
David Howells1fd36ad2012-02-16 17:49:54 +0000128 fdt->close_on_exec = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800129
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700130 return fdt;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800131
132out_arr:
Changli Gaoa892e2d2010-08-10 18:01:35 -0700133 free_fdmem(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800134out_fdt:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700135 kfree(fdt);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800136out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700137 return NULL;
138}
139
140/*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700141 * Expand the file descriptor table.
142 * This function will allocate a new fdtable and both fd array and fdset, of
143 * the given size.
144 * Return <0 error code on error; 1 on successful completion.
145 * The files->file_lock should be held on entry, and will be held on exit.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700146 */
147static int expand_fdtable(struct files_struct *files, int nr)
148 __releases(files->file_lock)
149 __acquires(files->file_lock)
150{
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700151 struct fdtable *new_fdt, *cur_fdt;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 spin_unlock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700154 new_fdt = alloc_fdtable(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700156 if (!new_fdt)
157 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700158 /*
Al Viro5c598b32008-04-27 20:04:15 -0400159 * extremely unlikely race - sysctl_nr_open decreased between the check in
160 * caller and alloc_fdtable(). Cheaper to catch it here...
161 */
162 if (unlikely(new_fdt->max_fds <= nr)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700163 __free_fdtable(new_fdt);
Al Viro5c598b32008-04-27 20:04:15 -0400164 return -EMFILE;
165 }
166 /*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700167 * Check again since another task may have expanded the fd table while
168 * we dropped the lock
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700169 */
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700170 cur_fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800171 if (nr >= cur_fdt->max_fds) {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700172 /* Continue as planned */
173 copy_fdtable(new_fdt, cur_fdt);
174 rcu_assign_pointer(files->fdt, new_fdt);
Al Viroac3e3c52013-04-28 21:42:33 -0400175 if (cur_fdt != &files->fdtab)
Al Viro1983e782012-08-15 20:06:36 -0400176 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700177 } else {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700178 /* Somebody else expanded, so undo our attempt */
Changli Gaoa892e2d2010-08-10 18:01:35 -0700179 __free_fdtable(new_fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700180 }
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700181 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
184/*
185 * Expand files.
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700186 * This function will expand the file structures, if the requested size exceeds
187 * the current capacity and there is room for expansion.
188 * Return <0 error code on error; 0 when nothing done; 1 when files were
189 * expanded and execution may have blocked.
190 * The files->file_lock should be held on entry, and will be held on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 */
Al Viroad47bd72012-08-21 20:11:34 -0400192static int expand_files(struct files_struct *files, int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700194 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700196 fdt = files_fdtable(files);
Al Viro4e1e0182008-07-26 16:01:20 -0400197
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700198 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800199 if (nr < fdt->max_fds)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700200 return 0;
Al Viro4e1e0182008-07-26 16:01:20 -0400201
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700202 /* Can we expand? */
Eric Dumazet9cfe0152008-02-06 01:37:16 -0800203 if (nr >= sysctl_nr_open)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700204 return -EMFILE;
205
206 /* All good, so we try */
207 return expand_fdtable(files, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700209
Al Virob8318b02012-08-21 20:09:42 -0400210static inline void __set_close_on_exec(int fd, struct fdtable *fdt)
211{
212 __set_bit(fd, fdt->close_on_exec);
213}
214
215static inline void __clear_close_on_exec(int fd, struct fdtable *fdt)
216{
217 __clear_bit(fd, fdt->close_on_exec);
218}
219
220static inline void __set_open_fd(int fd, struct fdtable *fdt)
221{
222 __set_bit(fd, fdt->open_fds);
223}
224
225static inline void __clear_open_fd(int fd, struct fdtable *fdt)
226{
227 __clear_bit(fd, fdt->open_fds);
228}
229
Al Viro02afc622008-05-08 19:42:56 -0400230static int count_open_files(struct fdtable *fdt)
231{
232 int size = fdt->max_fds;
233 int i;
234
235 /* Find the last open fd */
David Howells1fd36ad2012-02-16 17:49:54 +0000236 for (i = size / BITS_PER_LONG; i > 0; ) {
237 if (fdt->open_fds[--i])
Al Viro02afc622008-05-08 19:42:56 -0400238 break;
239 }
David Howells1fd36ad2012-02-16 17:49:54 +0000240 i = (i + 1) * BITS_PER_LONG;
Al Viro02afc622008-05-08 19:42:56 -0400241 return i;
242}
243
Al Viro02afc622008-05-08 19:42:56 -0400244/*
245 * Allocate a new files structure and copy contents from the
246 * passed in files structure.
247 * errorp will be valid only when the returned files_struct is NULL.
248 */
249struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
250{
251 struct files_struct *newf;
252 struct file **old_fds, **new_fds;
253 int open_files, size, i;
254 struct fdtable *old_fdt, *new_fdt;
255
256 *errorp = -ENOMEM;
Al Viroafbec7f2008-05-08 21:11:17 -0400257 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
Al Viro02afc622008-05-08 19:42:56 -0400258 if (!newf)
259 goto out;
260
Al Viroafbec7f2008-05-08 21:11:17 -0400261 atomic_set(&newf->count, 1);
262
263 spin_lock_init(&newf->file_lock);
264 newf->next_fd = 0;
265 new_fdt = &newf->fdtab;
266 new_fdt->max_fds = NR_OPEN_DEFAULT;
David Howells1fd36ad2012-02-16 17:49:54 +0000267 new_fdt->close_on_exec = newf->close_on_exec_init;
268 new_fdt->open_fds = newf->open_fds_init;
Al Viroafbec7f2008-05-08 21:11:17 -0400269 new_fdt->fd = &newf->fd_array[0];
Al Viroafbec7f2008-05-08 21:11:17 -0400270
Al Viro02afc622008-05-08 19:42:56 -0400271 spin_lock(&oldf->file_lock);
272 old_fdt = files_fdtable(oldf);
Al Viro02afc622008-05-08 19:42:56 -0400273 open_files = count_open_files(old_fdt);
274
275 /*
276 * Check whether we need to allocate a larger fd array and fd set.
Al Viro02afc622008-05-08 19:42:56 -0400277 */
Al Viroadbecb12008-05-08 21:19:42 -0400278 while (unlikely(open_files > new_fdt->max_fds)) {
Al Viro02afc622008-05-08 19:42:56 -0400279 spin_unlock(&oldf->file_lock);
Al Viro9dec3c42008-05-08 21:02:45 -0400280
Changli Gaoa892e2d2010-08-10 18:01:35 -0700281 if (new_fdt != &newf->fdtab)
282 __free_fdtable(new_fdt);
Al Viroadbecb12008-05-08 21:19:42 -0400283
Al Viro9dec3c42008-05-08 21:02:45 -0400284 new_fdt = alloc_fdtable(open_files - 1);
285 if (!new_fdt) {
286 *errorp = -ENOMEM;
Al Viro02afc622008-05-08 19:42:56 -0400287 goto out_release;
Al Viro9dec3c42008-05-08 21:02:45 -0400288 }
289
290 /* beyond sysctl_nr_open; nothing to do */
291 if (unlikely(new_fdt->max_fds < open_files)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700292 __free_fdtable(new_fdt);
Al Viro9dec3c42008-05-08 21:02:45 -0400293 *errorp = -EMFILE;
294 goto out_release;
295 }
Al Viro9dec3c42008-05-08 21:02:45 -0400296
Al Viro02afc622008-05-08 19:42:56 -0400297 /*
298 * Reacquire the oldf lock and a pointer to its fd table
299 * who knows it may have a new bigger fd table. We need
300 * the latest pointer.
301 */
302 spin_lock(&oldf->file_lock);
303 old_fdt = files_fdtable(oldf);
Al Viroadbecb12008-05-08 21:19:42 -0400304 open_files = count_open_files(old_fdt);
Al Viro02afc622008-05-08 19:42:56 -0400305 }
306
307 old_fds = old_fdt->fd;
308 new_fds = new_fdt->fd;
309
David Howells1fd36ad2012-02-16 17:49:54 +0000310 memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8);
311 memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8);
Al Viro02afc622008-05-08 19:42:56 -0400312
313 for (i = open_files; i != 0; i--) {
314 struct file *f = *old_fds++;
315 if (f) {
316 get_file(f);
317 } else {
318 /*
319 * The fd may be claimed in the fd bitmap but not yet
320 * instantiated in the files array if a sibling thread
321 * is partway through open(). So make sure that this
322 * fd is available to the new process.
323 */
David Howells1dce27c2012-02-16 17:49:42 +0000324 __clear_open_fd(open_files - i, new_fdt);
Al Viro02afc622008-05-08 19:42:56 -0400325 }
326 rcu_assign_pointer(*new_fds++, f);
327 }
328 spin_unlock(&oldf->file_lock);
329
330 /* compute the remainder to be cleared */
331 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
332
333 /* This is long word aligned thus could use a optimized version */
334 memset(new_fds, 0, size);
335
336 if (new_fdt->max_fds > open_files) {
David Howells1fd36ad2012-02-16 17:49:54 +0000337 int left = (new_fdt->max_fds - open_files) / 8;
338 int start = open_files / BITS_PER_LONG;
Al Viro02afc622008-05-08 19:42:56 -0400339
David Howells1fd36ad2012-02-16 17:49:54 +0000340 memset(&new_fdt->open_fds[start], 0, left);
341 memset(&new_fdt->close_on_exec[start], 0, left);
Al Viro02afc622008-05-08 19:42:56 -0400342 }
343
Al Viroafbec7f2008-05-08 21:11:17 -0400344 rcu_assign_pointer(newf->fdt, new_fdt);
345
Al Viro02afc622008-05-08 19:42:56 -0400346 return newf;
347
348out_release:
349 kmem_cache_free(files_cachep, newf);
350out:
351 return NULL;
352}
353
Oleg Nesterovce08b622014-01-11 19:19:53 +0100354static struct fdtable *close_files(struct files_struct * files)
Al Viro7cf4dc32012-08-15 19:56:12 -0400355{
Al Viro7cf4dc32012-08-15 19:56:12 -0400356 /*
357 * It is safe to dereference the fd table without RCU or
358 * ->file_lock because this is the last reference to the
Oleg Nesterovce08b622014-01-11 19:19:53 +0100359 * files structure.
Al Viro7cf4dc32012-08-15 19:56:12 -0400360 */
Oleg Nesterovce08b622014-01-11 19:19:53 +0100361 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
362 int i, j = 0;
363
Al Viro7cf4dc32012-08-15 19:56:12 -0400364 for (;;) {
365 unsigned long set;
366 i = j * BITS_PER_LONG;
367 if (i >= fdt->max_fds)
368 break;
369 set = fdt->open_fds[j++];
370 while (set) {
371 if (set & 1) {
372 struct file * file = xchg(&fdt->fd[i], NULL);
373 if (file) {
374 filp_close(file, files);
375 cond_resched();
376 }
377 }
378 i++;
379 set >>= 1;
380 }
381 }
Oleg Nesterovce08b622014-01-11 19:19:53 +0100382
383 return fdt;
Al Viro7cf4dc32012-08-15 19:56:12 -0400384}
385
386struct files_struct *get_files_struct(struct task_struct *task)
387{
388 struct files_struct *files;
389
390 task_lock(task);
391 files = task->files;
392 if (files)
393 atomic_inc(&files->count);
394 task_unlock(task);
395
396 return files;
397}
398
399void put_files_struct(struct files_struct *files)
400{
Al Viro7cf4dc32012-08-15 19:56:12 -0400401 if (atomic_dec_and_test(&files->count)) {
Oleg Nesterovce08b622014-01-11 19:19:53 +0100402 struct fdtable *fdt = close_files(files);
403
Al Virob9e02af2012-08-15 20:00:58 -0400404 /* free the arrays if they are not embedded */
405 if (fdt != &files->fdtab)
406 __free_fdtable(fdt);
407 kmem_cache_free(files_cachep, files);
Al Viro7cf4dc32012-08-15 19:56:12 -0400408 }
409}
410
411void reset_files_struct(struct files_struct *files)
412{
413 struct task_struct *tsk = current;
414 struct files_struct *old;
415
416 old = tsk->files;
417 task_lock(tsk);
418 tsk->files = files;
419 task_unlock(tsk);
420 put_files_struct(old);
421}
422
423void exit_files(struct task_struct *tsk)
424{
425 struct files_struct * files = tsk->files;
426
427 if (files) {
428 task_lock(tsk);
429 tsk->files = NULL;
430 task_unlock(tsk);
431 put_files_struct(files);
432 }
433}
434
Al Virof52111b2008-05-08 18:19:16 -0400435struct files_struct init_files = {
436 .count = ATOMIC_INIT(1),
437 .fdt = &init_files.fdtab,
438 .fdtab = {
439 .max_fds = NR_OPEN_DEFAULT,
440 .fd = &init_files.fd_array[0],
David Howells1fd36ad2012-02-16 17:49:54 +0000441 .close_on_exec = init_files.close_on_exec_init,
442 .open_fds = init_files.open_fds_init,
Al Virof52111b2008-05-08 18:19:16 -0400443 },
Thomas Gleixnereece09e2011-07-17 21:25:03 +0200444 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
Al Virof52111b2008-05-08 18:19:16 -0400445};
Al Viro1027abe2008-07-30 04:13:04 -0400446
447/*
448 * allocate a file descriptor, mark it busy.
449 */
Al Virodcfadfa2012-08-12 17:27:30 -0400450int __alloc_fd(struct files_struct *files,
451 unsigned start, unsigned end, unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400452{
Al Viro1027abe2008-07-30 04:13:04 -0400453 unsigned int fd;
454 int error;
455 struct fdtable *fdt;
456
457 spin_lock(&files->file_lock);
458repeat:
459 fdt = files_fdtable(files);
460 fd = start;
461 if (fd < files->next_fd)
462 fd = files->next_fd;
463
464 if (fd < fdt->max_fds)
David Howells1fd36ad2012-02-16 17:49:54 +0000465 fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd);
Al Viro1027abe2008-07-30 04:13:04 -0400466
Al Virof33ff992012-08-12 16:17:59 -0400467 /*
468 * N.B. For clone tasks sharing a files structure, this test
469 * will limit the total number of files that can be opened.
470 */
471 error = -EMFILE;
472 if (fd >= end)
473 goto out;
474
Al Viro1027abe2008-07-30 04:13:04 -0400475 error = expand_files(files, fd);
476 if (error < 0)
477 goto out;
478
479 /*
480 * If we needed to expand the fs array we
481 * might have blocked - try again.
482 */
483 if (error)
484 goto repeat;
485
486 if (start <= files->next_fd)
487 files->next_fd = fd + 1;
488
David Howells1dce27c2012-02-16 17:49:42 +0000489 __set_open_fd(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400490 if (flags & O_CLOEXEC)
David Howells1dce27c2012-02-16 17:49:42 +0000491 __set_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400492 else
David Howells1dce27c2012-02-16 17:49:42 +0000493 __clear_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400494 error = fd;
495#if 1
496 /* Sanity check */
Paul E. McKenneyadd1f092014-02-12 12:51:09 -0800497 if (rcu_access_pointer(fdt->fd[fd]) != NULL) {
Al Viro1027abe2008-07-30 04:13:04 -0400498 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
499 rcu_assign_pointer(fdt->fd[fd], NULL);
500 }
501#endif
502
503out:
504 spin_unlock(&files->file_lock);
505 return error;
506}
507
Al Viroad47bd72012-08-21 20:11:34 -0400508static int alloc_fd(unsigned start, unsigned flags)
Al Virodcfadfa2012-08-12 17:27:30 -0400509{
510 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
511}
512
Al Viro1a7bd222012-08-12 17:18:05 -0400513int get_unused_fd_flags(unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400514{
Al Virodcfadfa2012-08-12 17:27:30 -0400515 return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags);
Al Viro1027abe2008-07-30 04:13:04 -0400516}
Al Viro1a7bd222012-08-12 17:18:05 -0400517EXPORT_SYMBOL(get_unused_fd_flags);
Al Viro56007ca2012-08-15 21:03:26 -0400518
519static void __put_unused_fd(struct files_struct *files, unsigned int fd)
520{
521 struct fdtable *fdt = files_fdtable(files);
522 __clear_open_fd(fd, fdt);
523 if (fd < files->next_fd)
524 files->next_fd = fd;
525}
526
527void put_unused_fd(unsigned int fd)
528{
529 struct files_struct *files = current->files;
530 spin_lock(&files->file_lock);
531 __put_unused_fd(files, fd);
532 spin_unlock(&files->file_lock);
533}
534
535EXPORT_SYMBOL(put_unused_fd);
536
537/*
538 * Install a file pointer in the fd array.
539 *
540 * The VFS is full of places where we drop the files lock between
541 * setting the open_fds bitmap and installing the file in the file
542 * array. At any such point, we are vulnerable to a dup2() race
543 * installing a file in the array before us. We need to detect this and
544 * fput() the struct file we are about to overwrite in this case.
545 *
546 * It should never happen - if we allow dup2() do it, _really_ bad things
547 * will follow.
Al Virof869e8a2012-08-15 21:06:33 -0400548 *
549 * NOTE: __fd_install() variant is really, really low-level; don't
550 * use it unless you are forced to by truly lousy API shoved down
551 * your throat. 'files' *MUST* be either current->files or obtained
552 * by get_files_struct(current) done by whoever had given it to you,
553 * or really bad things will happen. Normally you want to use
554 * fd_install() instead.
Al Viro56007ca2012-08-15 21:03:26 -0400555 */
556
Al Virof869e8a2012-08-15 21:06:33 -0400557void __fd_install(struct files_struct *files, unsigned int fd,
558 struct file *file)
Al Viro56007ca2012-08-15 21:03:26 -0400559{
Al Viro56007ca2012-08-15 21:03:26 -0400560 struct fdtable *fdt;
561 spin_lock(&files->file_lock);
562 fdt = files_fdtable(files);
563 BUG_ON(fdt->fd[fd] != NULL);
564 rcu_assign_pointer(fdt->fd[fd], file);
565 spin_unlock(&files->file_lock);
566}
567
Al Virof869e8a2012-08-15 21:06:33 -0400568void fd_install(unsigned int fd, struct file *file)
569{
570 __fd_install(current->files, fd, file);
571}
572
Al Viro56007ca2012-08-15 21:03:26 -0400573EXPORT_SYMBOL(fd_install);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400574
Al Viro483ce1d2012-08-19 12:04:24 -0400575/*
576 * The same warnings as for __alloc_fd()/__fd_install() apply here...
577 */
578int __close_fd(struct files_struct *files, unsigned fd)
579{
580 struct file *file;
581 struct fdtable *fdt;
582
583 spin_lock(&files->file_lock);
584 fdt = files_fdtable(files);
585 if (fd >= fdt->max_fds)
586 goto out_unlock;
587 file = fdt->fd[fd];
588 if (!file)
589 goto out_unlock;
590 rcu_assign_pointer(fdt->fd[fd], NULL);
591 __clear_close_on_exec(fd, fdt);
592 __put_unused_fd(files, fd);
593 spin_unlock(&files->file_lock);
594 return filp_close(file, files);
595
596out_unlock:
597 spin_unlock(&files->file_lock);
598 return -EBADF;
599}
600
Al Viro6a6d27d2012-08-21 09:56:33 -0400601void do_close_on_exec(struct files_struct *files)
602{
603 unsigned i;
604 struct fdtable *fdt;
605
606 /* exec unshares first */
Al Viro6a6d27d2012-08-21 09:56:33 -0400607 spin_lock(&files->file_lock);
608 for (i = 0; ; i++) {
609 unsigned long set;
610 unsigned fd = i * BITS_PER_LONG;
611 fdt = files_fdtable(files);
612 if (fd >= fdt->max_fds)
613 break;
614 set = fdt->close_on_exec[i];
615 if (!set)
616 continue;
617 fdt->close_on_exec[i] = 0;
618 for ( ; set ; fd++, set >>= 1) {
619 struct file *file;
620 if (!(set & 1))
621 continue;
622 file = fdt->fd[fd];
623 if (!file)
624 continue;
625 rcu_assign_pointer(fdt->fd[fd], NULL);
626 __put_unused_fd(files, fd);
627 spin_unlock(&files->file_lock);
628 filp_close(file, files);
629 cond_resched();
630 spin_lock(&files->file_lock);
631 }
632
633 }
634 spin_unlock(&files->file_lock);
635}
636
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100637static struct file *__fget(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400638{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400639 struct files_struct *files = current->files;
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100640 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400641
642 rcu_read_lock();
643 file = fcheck_files(files, fd);
644 if (file) {
645 /* File object ref couldn't be taken */
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100646 if ((file->f_mode & mask) ||
Al Viro0ee8cdf2012-08-15 21:12:10 -0400647 !atomic_long_inc_not_zero(&file->f_count))
648 file = NULL;
649 }
650 rcu_read_unlock();
651
652 return file;
653}
654
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100655struct file *fget(unsigned int fd)
656{
657 return __fget(fd, FMODE_PATH);
658}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400659EXPORT_SYMBOL(fget);
660
661struct file *fget_raw(unsigned int fd)
662{
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100663 return __fget(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400664}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400665EXPORT_SYMBOL(fget_raw);
666
667/*
668 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
669 *
670 * You can use this instead of fget if you satisfy all of the following
671 * conditions:
672 * 1) You must call fput_light before exiting the syscall and returning control
673 * to userspace (i.e. you cannot remember the returned struct file * after
674 * returning to userspace).
675 * 2) You must not call filp_close on the returned struct file * in between
676 * calls to fget_light and fput_light.
677 * 3) You must not clone the current task in between the calls to fget_light
678 * and fput_light.
679 *
680 * The fput_needed flag returned by fget_light should be passed to the
681 * corresponding fput_light.
682 */
Al Virobd2a31d2014-03-04 14:54:22 -0500683static unsigned long __fget_light(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400684{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400685 struct files_struct *files = current->files;
Oleg Nesterovad461832014-01-13 16:48:40 +0100686 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400687
Al Viro0ee8cdf2012-08-15 21:12:10 -0400688 if (atomic_read(&files->count) == 1) {
Oleg Nesterova8d4b832014-01-11 19:19:32 +0100689 file = __fcheck_files(files, fd);
Al Virobd2a31d2014-03-04 14:54:22 -0500690 if (!file || unlikely(file->f_mode & mask))
691 return 0;
692 return (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400693 } else {
Oleg Nesterove6ff9a92014-01-13 16:49:06 +0100694 file = __fget(fd, mask);
Al Virobd2a31d2014-03-04 14:54:22 -0500695 if (!file)
696 return 0;
697 return FDPUT_FPUT | (unsigned long)file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400698 }
Al Viro0ee8cdf2012-08-15 21:12:10 -0400699}
Al Virobd2a31d2014-03-04 14:54:22 -0500700unsigned long __fdget(unsigned int fd)
Oleg Nesterovad461832014-01-13 16:48:40 +0100701{
Al Virobd2a31d2014-03-04 14:54:22 -0500702 return __fget_light(fd, FMODE_PATH);
Oleg Nesterovad461832014-01-13 16:48:40 +0100703}
Al Virobd2a31d2014-03-04 14:54:22 -0500704EXPORT_SYMBOL(__fdget);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400705
Al Virobd2a31d2014-03-04 14:54:22 -0500706unsigned long __fdget_raw(unsigned int fd)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400707{
Al Virobd2a31d2014-03-04 14:54:22 -0500708 return __fget_light(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400709}
Al Virofe17f222012-08-21 11:48:11 -0400710
Al Virobd2a31d2014-03-04 14:54:22 -0500711unsigned long __fdget_pos(unsigned int fd)
712{
Eric Biggers99aea682014-03-16 15:47:48 -0500713 unsigned long v = __fdget(fd);
714 struct file *file = (struct file *)(v & ~3);
Al Virobd2a31d2014-03-04 14:54:22 -0500715
Eric Biggers99aea682014-03-16 15:47:48 -0500716 if (file && (file->f_mode & FMODE_ATOMIC_POS)) {
Al Virobd2a31d2014-03-04 14:54:22 -0500717 if (file_count(file) > 1) {
718 v |= FDPUT_POS_UNLOCK;
719 mutex_lock(&file->f_pos_lock);
720 }
721 }
Eric Biggers99aea682014-03-16 15:47:48 -0500722 return v;
Al Virobd2a31d2014-03-04 14:54:22 -0500723}
724
725/*
726 * We only lock f_pos if we have threads or if the file might be
727 * shared with another process. In both cases we'll have an elevated
728 * file count (done either by fdget() or by fork()).
729 */
730
Al Virofe17f222012-08-21 11:48:11 -0400731void set_close_on_exec(unsigned int fd, int flag)
732{
733 struct files_struct *files = current->files;
734 struct fdtable *fdt;
735 spin_lock(&files->file_lock);
736 fdt = files_fdtable(files);
737 if (flag)
738 __set_close_on_exec(fd, fdt);
739 else
740 __clear_close_on_exec(fd, fdt);
741 spin_unlock(&files->file_lock);
742}
743
744bool get_close_on_exec(unsigned int fd)
745{
746 struct files_struct *files = current->files;
747 struct fdtable *fdt;
748 bool res;
749 rcu_read_lock();
750 fdt = files_fdtable(files);
751 res = close_on_exec(fd, fdt);
752 rcu_read_unlock();
753 return res;
754}
755
Al Viro8280d162012-08-21 12:11:46 -0400756static int do_dup2(struct files_struct *files,
757 struct file *file, unsigned fd, unsigned flags)
758{
759 struct file *tofree;
760 struct fdtable *fdt;
761
762 /*
763 * We need to detect attempts to do dup2() over allocated but still
764 * not finished descriptor. NB: OpenBSD avoids that at the price of
765 * extra work in their equivalent of fget() - they insert struct
766 * file immediately after grabbing descriptor, mark it larval if
767 * more work (e.g. actual opening) is needed and make sure that
768 * fget() treats larval files as absent. Potentially interesting,
769 * but while extra work in fget() is trivial, locking implications
770 * and amount of surgery on open()-related paths in VFS are not.
771 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
772 * deadlocks in rather amusing ways, AFAICS. All of that is out of
773 * scope of POSIX or SUS, since neither considers shared descriptor
774 * tables and this condition does not arise without those.
775 */
776 fdt = files_fdtable(files);
777 tofree = fdt->fd[fd];
778 if (!tofree && fd_is_open(fd, fdt))
779 goto Ebusy;
780 get_file(file);
781 rcu_assign_pointer(fdt->fd[fd], file);
782 __set_open_fd(fd, fdt);
783 if (flags & O_CLOEXEC)
784 __set_close_on_exec(fd, fdt);
785 else
786 __clear_close_on_exec(fd, fdt);
787 spin_unlock(&files->file_lock);
788
789 if (tofree)
790 filp_close(tofree, files);
791
792 return fd;
793
794Ebusy:
795 spin_unlock(&files->file_lock);
796 return -EBUSY;
797}
798
799int replace_fd(unsigned fd, struct file *file, unsigned flags)
800{
801 int err;
802 struct files_struct *files = current->files;
803
804 if (!file)
805 return __close_fd(files, fd);
806
807 if (fd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000808 return -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400809
810 spin_lock(&files->file_lock);
811 err = expand_files(files, fd);
812 if (unlikely(err < 0))
813 goto out_unlock;
814 return do_dup2(files, file, fd, flags);
815
816out_unlock:
817 spin_unlock(&files->file_lock);
818 return err;
819}
820
Al Virofe17f222012-08-21 11:48:11 -0400821SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
822{
823 int err = -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400824 struct file *file;
825 struct files_struct *files = current->files;
Al Virofe17f222012-08-21 11:48:11 -0400826
827 if ((flags & ~O_CLOEXEC) != 0)
828 return -EINVAL;
829
Richard W.M. Jonesaed97642012-10-09 15:27:43 +0100830 if (unlikely(oldfd == newfd))
831 return -EINVAL;
832
Al Virofe17f222012-08-21 11:48:11 -0400833 if (newfd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000834 return -EBADF;
Al Virofe17f222012-08-21 11:48:11 -0400835
836 spin_lock(&files->file_lock);
837 err = expand_files(files, newfd);
838 file = fcheck(oldfd);
839 if (unlikely(!file))
840 goto Ebadf;
841 if (unlikely(err < 0)) {
842 if (err == -EMFILE)
843 goto Ebadf;
844 goto out_unlock;
845 }
Al Viro8280d162012-08-21 12:11:46 -0400846 return do_dup2(files, file, newfd, flags);
Al Virofe17f222012-08-21 11:48:11 -0400847
848Ebadf:
849 err = -EBADF;
850out_unlock:
851 spin_unlock(&files->file_lock);
852 return err;
853}
854
855SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
856{
857 if (unlikely(newfd == oldfd)) { /* corner case */
858 struct files_struct *files = current->files;
859 int retval = oldfd;
860
861 rcu_read_lock();
862 if (!fcheck_files(files, oldfd))
863 retval = -EBADF;
864 rcu_read_unlock();
865 return retval;
866 }
867 return sys_dup3(oldfd, newfd, 0);
868}
869
870SYSCALL_DEFINE1(dup, unsigned int, fildes)
871{
872 int ret = -EBADF;
873 struct file *file = fget_raw(fildes);
874
875 if (file) {
876 ret = get_unused_fd();
877 if (ret >= 0)
878 fd_install(ret, file);
879 else
880 fput(file);
881 }
882 return ret;
883}
884
885int f_dupfd(unsigned int from, struct file *file, unsigned flags)
886{
887 int err;
888 if (from >= rlimit(RLIMIT_NOFILE))
889 return -EINVAL;
890 err = alloc_fd(from, flags);
891 if (err >= 0) {
892 get_file(file);
893 fd_install(err, file);
894 }
895 return err;
896}
Al Viroc3c073f2012-08-21 22:32:06 -0400897
898int iterate_fd(struct files_struct *files, unsigned n,
899 int (*f)(const void *, struct file *, unsigned),
900 const void *p)
901{
902 struct fdtable *fdt;
Al Viroc3c073f2012-08-21 22:32:06 -0400903 int res = 0;
904 if (!files)
905 return 0;
906 spin_lock(&files->file_lock);
Al Viroa77cfcb2012-11-29 22:57:33 -0500907 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
908 struct file *file;
909 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
910 if (!file)
911 continue;
912 res = f(p, file, n);
913 if (res)
914 break;
Al Viroc3c073f2012-08-21 22:32:06 -0400915 }
916 spin_unlock(&files->file_lock);
917 return res;
918}
919EXPORT_SYMBOL(iterate_fd);