blob: 50c1208f64502d460407f46b1618ae786b71f6c8 [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;
28int sysctl_nr_open_max = 1024 * 1024; /* raised later */
Eric Dumazet9cfe0152008-02-06 01:37:16 -080029
David Howells1fd36ad2012-02-16 17:49:54 +000030static void *alloc_fdmem(size_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -070031{
Andrew Morton6d4831c2011-04-27 15:26:41 -070032 /*
33 * Very large allocations can stress page reclaim, so fall back to
34 * vmalloc() if the allocation size will be considered "large" by the VM.
35 */
36 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
37 void *data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN);
38 if (data != NULL)
39 return data;
40 }
Changli Gaoa892e2d2010-08-10 18:01:35 -070041 return vmalloc(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042}
43
Changli Gaoa892e2d2010-08-10 18:01:35 -070044static void free_fdmem(void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070045{
Changli Gaoa892e2d2010-08-10 18:01:35 -070046 is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr);
Vadim Lobanov5466b452006-12-10 02:21:22 -080047}
48
Changli Gaoa892e2d2010-08-10 18:01:35 -070049static void __free_fdtable(struct fdtable *fdt)
Vadim Lobanov5466b452006-12-10 02:21:22 -080050{
Changli Gaoa892e2d2010-08-10 18:01:35 -070051 free_fdmem(fdt->fd);
52 free_fdmem(fdt->open_fds);
53 kfree(fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070054}
55
Al Viro7cf4dc32012-08-15 19:56:12 -040056static void free_fdtable_rcu(struct rcu_head *rcu)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070057{
Al Viroac3e3c52013-04-28 21:42:33 -040058 __free_fdtable(container_of(rcu, struct fdtable, rcu));
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070059}
60
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070061/*
62 * Expand the fdset in the files_struct. Called with the files spinlock
63 * held for write.
64 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080065static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070066{
Vadim Lobanov5466b452006-12-10 02:21:22 -080067 unsigned int cpy, set;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070068
Vadim Lobanov5466b452006-12-10 02:21:22 -080069 BUG_ON(nfdt->max_fds < ofdt->max_fds);
Vadim Lobanov5466b452006-12-10 02:21:22 -080070
71 cpy = ofdt->max_fds * sizeof(struct file *);
72 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
73 memcpy(nfdt->fd, ofdt->fd, cpy);
74 memset((char *)(nfdt->fd) + cpy, 0, set);
75
76 cpy = ofdt->max_fds / BITS_PER_BYTE;
77 set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
78 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
79 memset((char *)(nfdt->open_fds) + cpy, 0, set);
80 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
81 memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082}
83
Vadim Lobanov5466b452006-12-10 02:21:22 -080084static struct fdtable * alloc_fdtable(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085{
Vadim Lobanov5466b452006-12-10 02:21:22 -080086 struct fdtable *fdt;
David Howells1fd36ad2012-02-16 17:49:54 +000087 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070089 /*
Vadim Lobanov5466b452006-12-10 02:21:22 -080090 * Figure out how many fds we actually want to support in this fdtable.
91 * Allocation steps are keyed to the size of the fdarray, since it
92 * grows far faster than any of the other dynamic data. We try to fit
93 * the fdarray into comfortable page-tuned chunks: starting at 1024B
94 * and growing in powers of two from there on.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070095 */
Vadim Lobanov5466b452006-12-10 02:21:22 -080096 nr /= (1024 / sizeof(struct file *));
97 nr = roundup_pow_of_two(nr + 1);
98 nr *= (1024 / sizeof(struct file *));
Al Viro5c598b32008-04-27 20:04:15 -040099 /*
100 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
101 * had been set lower between the check in expand_files() and here. Deal
102 * with that in caller, it's cheaper that way.
103 *
104 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
105 * bitmaps handling below becomes unpleasant, to put it mildly...
106 */
107 if (unlikely(nr > sysctl_nr_open))
108 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800109
Vadim Lobanov5466b452006-12-10 02:21:22 -0800110 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
111 if (!fdt)
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800112 goto out;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800113 fdt->max_fds = nr;
114 data = alloc_fdmem(nr * sizeof(struct file *));
115 if (!data)
116 goto out_fdt;
David Howells1fd36ad2012-02-16 17:49:54 +0000117 fdt->fd = data;
118
119 data = alloc_fdmem(max_t(size_t,
Vadim Lobanov5466b452006-12-10 02:21:22 -0800120 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
121 if (!data)
122 goto out_arr;
David Howells1fd36ad2012-02-16 17:49:54 +0000123 fdt->open_fds = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800124 data += nr / BITS_PER_BYTE;
David Howells1fd36ad2012-02-16 17:49:54 +0000125 fdt->close_on_exec = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800126
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700127 return fdt;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800128
129out_arr:
Changli Gaoa892e2d2010-08-10 18:01:35 -0700130 free_fdmem(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800131out_fdt:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700132 kfree(fdt);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800133out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700134 return NULL;
135}
136
137/*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700138 * Expand the file descriptor table.
139 * This function will allocate a new fdtable and both fd array and fdset, of
140 * the given size.
141 * Return <0 error code on error; 1 on successful completion.
142 * The files->file_lock should be held on entry, and will be held on exit.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700143 */
144static int expand_fdtable(struct files_struct *files, int nr)
145 __releases(files->file_lock)
146 __acquires(files->file_lock)
147{
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700148 struct fdtable *new_fdt, *cur_fdt;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 spin_unlock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700151 new_fdt = alloc_fdtable(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700153 if (!new_fdt)
154 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700155 /*
Al Viro5c598b32008-04-27 20:04:15 -0400156 * extremely unlikely race - sysctl_nr_open decreased between the check in
157 * caller and alloc_fdtable(). Cheaper to catch it here...
158 */
159 if (unlikely(new_fdt->max_fds <= nr)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700160 __free_fdtable(new_fdt);
Al Viro5c598b32008-04-27 20:04:15 -0400161 return -EMFILE;
162 }
163 /*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700164 * Check again since another task may have expanded the fd table while
165 * we dropped the lock
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700166 */
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700167 cur_fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800168 if (nr >= cur_fdt->max_fds) {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700169 /* Continue as planned */
170 copy_fdtable(new_fdt, cur_fdt);
171 rcu_assign_pointer(files->fdt, new_fdt);
Al Viroac3e3c52013-04-28 21:42:33 -0400172 if (cur_fdt != &files->fdtab)
Al Viro1983e782012-08-15 20:06:36 -0400173 call_rcu(&cur_fdt->rcu, free_fdtable_rcu);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700174 } else {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700175 /* Somebody else expanded, so undo our attempt */
Changli Gaoa892e2d2010-08-10 18:01:35 -0700176 __free_fdtable(new_fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700177 }
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700178 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179}
180
181/*
182 * Expand files.
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700183 * This function will expand the file structures, if the requested size exceeds
184 * the current capacity and there is room for expansion.
185 * Return <0 error code on error; 0 when nothing done; 1 when files were
186 * expanded and execution may have blocked.
187 * The files->file_lock should be held on entry, and will be held on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 */
Al Viroad47bd72012-08-21 20:11:34 -0400189static int expand_files(struct files_struct *files, int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700191 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700193 fdt = files_fdtable(files);
Al Viro4e1e0182008-07-26 16:01:20 -0400194
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700195 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800196 if (nr < fdt->max_fds)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700197 return 0;
Al Viro4e1e0182008-07-26 16:01:20 -0400198
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700199 /* Can we expand? */
Eric Dumazet9cfe0152008-02-06 01:37:16 -0800200 if (nr >= sysctl_nr_open)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700201 return -EMFILE;
202
203 /* All good, so we try */
204 return expand_fdtable(files, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700206
Al Virob8318b02012-08-21 20:09:42 -0400207static inline void __set_close_on_exec(int fd, struct fdtable *fdt)
208{
209 __set_bit(fd, fdt->close_on_exec);
210}
211
212static inline void __clear_close_on_exec(int fd, struct fdtable *fdt)
213{
214 __clear_bit(fd, fdt->close_on_exec);
215}
216
217static inline void __set_open_fd(int fd, struct fdtable *fdt)
218{
219 __set_bit(fd, fdt->open_fds);
220}
221
222static inline void __clear_open_fd(int fd, struct fdtable *fdt)
223{
224 __clear_bit(fd, fdt->open_fds);
225}
226
Al Viro02afc6262008-05-08 19:42:56 -0400227static int count_open_files(struct fdtable *fdt)
228{
229 int size = fdt->max_fds;
230 int i;
231
232 /* Find the last open fd */
David Howells1fd36ad2012-02-16 17:49:54 +0000233 for (i = size / BITS_PER_LONG; i > 0; ) {
234 if (fdt->open_fds[--i])
Al Viro02afc6262008-05-08 19:42:56 -0400235 break;
236 }
David Howells1fd36ad2012-02-16 17:49:54 +0000237 i = (i + 1) * BITS_PER_LONG;
Al Viro02afc6262008-05-08 19:42:56 -0400238 return i;
239}
240
Al Viro02afc6262008-05-08 19:42:56 -0400241/*
242 * Allocate a new files structure and copy contents from the
243 * passed in files structure.
244 * errorp will be valid only when the returned files_struct is NULL.
245 */
246struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
247{
248 struct files_struct *newf;
249 struct file **old_fds, **new_fds;
250 int open_files, size, i;
251 struct fdtable *old_fdt, *new_fdt;
252
253 *errorp = -ENOMEM;
Al Viroafbec7f2008-05-08 21:11:17 -0400254 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
Al Viro02afc6262008-05-08 19:42:56 -0400255 if (!newf)
256 goto out;
257
Al Viroafbec7f2008-05-08 21:11:17 -0400258 atomic_set(&newf->count, 1);
259
260 spin_lock_init(&newf->file_lock);
261 newf->next_fd = 0;
262 new_fdt = &newf->fdtab;
263 new_fdt->max_fds = NR_OPEN_DEFAULT;
David Howells1fd36ad2012-02-16 17:49:54 +0000264 new_fdt->close_on_exec = newf->close_on_exec_init;
265 new_fdt->open_fds = newf->open_fds_init;
Al Viroafbec7f2008-05-08 21:11:17 -0400266 new_fdt->fd = &newf->fd_array[0];
Al Viroafbec7f2008-05-08 21:11:17 -0400267
Al Viro02afc6262008-05-08 19:42:56 -0400268 spin_lock(&oldf->file_lock);
269 old_fdt = files_fdtable(oldf);
Al Viro02afc6262008-05-08 19:42:56 -0400270 open_files = count_open_files(old_fdt);
271
272 /*
273 * Check whether we need to allocate a larger fd array and fd set.
Al Viro02afc6262008-05-08 19:42:56 -0400274 */
Al Viroadbecb12008-05-08 21:19:42 -0400275 while (unlikely(open_files > new_fdt->max_fds)) {
Al Viro02afc6262008-05-08 19:42:56 -0400276 spin_unlock(&oldf->file_lock);
Al Viro9dec3c42008-05-08 21:02:45 -0400277
Changli Gaoa892e2d2010-08-10 18:01:35 -0700278 if (new_fdt != &newf->fdtab)
279 __free_fdtable(new_fdt);
Al Viroadbecb12008-05-08 21:19:42 -0400280
Al Viro9dec3c42008-05-08 21:02:45 -0400281 new_fdt = alloc_fdtable(open_files - 1);
282 if (!new_fdt) {
283 *errorp = -ENOMEM;
Al Viro02afc6262008-05-08 19:42:56 -0400284 goto out_release;
Al Viro9dec3c42008-05-08 21:02:45 -0400285 }
286
287 /* beyond sysctl_nr_open; nothing to do */
288 if (unlikely(new_fdt->max_fds < open_files)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700289 __free_fdtable(new_fdt);
Al Viro9dec3c42008-05-08 21:02:45 -0400290 *errorp = -EMFILE;
291 goto out_release;
292 }
Al Viro9dec3c42008-05-08 21:02:45 -0400293
Al Viro02afc6262008-05-08 19:42:56 -0400294 /*
295 * Reacquire the oldf lock and a pointer to its fd table
296 * who knows it may have a new bigger fd table. We need
297 * the latest pointer.
298 */
299 spin_lock(&oldf->file_lock);
300 old_fdt = files_fdtable(oldf);
Al Viroadbecb12008-05-08 21:19:42 -0400301 open_files = count_open_files(old_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400302 }
303
304 old_fds = old_fdt->fd;
305 new_fds = new_fdt->fd;
306
David Howells1fd36ad2012-02-16 17:49:54 +0000307 memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8);
308 memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8);
Al Viro02afc6262008-05-08 19:42:56 -0400309
310 for (i = open_files; i != 0; i--) {
311 struct file *f = *old_fds++;
312 if (f) {
313 get_file(f);
314 } else {
315 /*
316 * The fd may be claimed in the fd bitmap but not yet
317 * instantiated in the files array if a sibling thread
318 * is partway through open(). So make sure that this
319 * fd is available to the new process.
320 */
David Howells1dce27c2012-02-16 17:49:42 +0000321 __clear_open_fd(open_files - i, new_fdt);
Al Viro02afc6262008-05-08 19:42:56 -0400322 }
323 rcu_assign_pointer(*new_fds++, f);
324 }
325 spin_unlock(&oldf->file_lock);
326
327 /* compute the remainder to be cleared */
328 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
329
330 /* This is long word aligned thus could use a optimized version */
331 memset(new_fds, 0, size);
332
333 if (new_fdt->max_fds > open_files) {
David Howells1fd36ad2012-02-16 17:49:54 +0000334 int left = (new_fdt->max_fds - open_files) / 8;
335 int start = open_files / BITS_PER_LONG;
Al Viro02afc6262008-05-08 19:42:56 -0400336
David Howells1fd36ad2012-02-16 17:49:54 +0000337 memset(&new_fdt->open_fds[start], 0, left);
338 memset(&new_fdt->close_on_exec[start], 0, left);
Al Viro02afc6262008-05-08 19:42:56 -0400339 }
340
Al Viroafbec7f2008-05-08 21:11:17 -0400341 rcu_assign_pointer(newf->fdt, new_fdt);
342
Al Viro02afc6262008-05-08 19:42:56 -0400343 return newf;
344
345out_release:
346 kmem_cache_free(files_cachep, newf);
347out:
348 return NULL;
349}
350
Oleg Nesterovce08b622014-01-11 19:19:53 +0100351static struct fdtable *close_files(struct files_struct * files)
Al Viro7cf4dc32012-08-15 19:56:12 -0400352{
Al Viro7cf4dc32012-08-15 19:56:12 -0400353 /*
354 * It is safe to dereference the fd table without RCU or
355 * ->file_lock because this is the last reference to the
Oleg Nesterovce08b622014-01-11 19:19:53 +0100356 * files structure.
Al Viro7cf4dc32012-08-15 19:56:12 -0400357 */
Oleg Nesterovce08b622014-01-11 19:19:53 +0100358 struct fdtable *fdt = rcu_dereference_raw(files->fdt);
359 int i, j = 0;
360
Al Viro7cf4dc32012-08-15 19:56:12 -0400361 for (;;) {
362 unsigned long set;
363 i = j * BITS_PER_LONG;
364 if (i >= fdt->max_fds)
365 break;
366 set = fdt->open_fds[j++];
367 while (set) {
368 if (set & 1) {
369 struct file * file = xchg(&fdt->fd[i], NULL);
370 if (file) {
371 filp_close(file, files);
372 cond_resched();
373 }
374 }
375 i++;
376 set >>= 1;
377 }
378 }
Oleg Nesterovce08b622014-01-11 19:19:53 +0100379
380 return fdt;
Al Viro7cf4dc32012-08-15 19:56:12 -0400381}
382
383struct files_struct *get_files_struct(struct task_struct *task)
384{
385 struct files_struct *files;
386
387 task_lock(task);
388 files = task->files;
389 if (files)
390 atomic_inc(&files->count);
391 task_unlock(task);
392
393 return files;
394}
395
396void put_files_struct(struct files_struct *files)
397{
Al Viro7cf4dc32012-08-15 19:56:12 -0400398 if (atomic_dec_and_test(&files->count)) {
Oleg Nesterovce08b622014-01-11 19:19:53 +0100399 struct fdtable *fdt = close_files(files);
400
Al Virob9e02af2012-08-15 20:00:58 -0400401 /* free the arrays if they are not embedded */
402 if (fdt != &files->fdtab)
403 __free_fdtable(fdt);
404 kmem_cache_free(files_cachep, files);
Al Viro7cf4dc32012-08-15 19:56:12 -0400405 }
406}
407
408void reset_files_struct(struct files_struct *files)
409{
410 struct task_struct *tsk = current;
411 struct files_struct *old;
412
413 old = tsk->files;
414 task_lock(tsk);
415 tsk->files = files;
416 task_unlock(tsk);
417 put_files_struct(old);
418}
419
420void exit_files(struct task_struct *tsk)
421{
422 struct files_struct * files = tsk->files;
423
424 if (files) {
425 task_lock(tsk);
426 tsk->files = NULL;
427 task_unlock(tsk);
428 put_files_struct(files);
429 }
430}
431
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700432void __init files_defer_init(void)
433{
Al Viroeceea0b2008-05-10 10:08:32 -0400434 sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
435 -BITS_PER_LONG;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700436}
Al Virof52111b2008-05-08 18:19:16 -0400437
438struct files_struct init_files = {
439 .count = ATOMIC_INIT(1),
440 .fdt = &init_files.fdtab,
441 .fdtab = {
442 .max_fds = NR_OPEN_DEFAULT,
443 .fd = &init_files.fd_array[0],
David Howells1fd36ad2012-02-16 17:49:54 +0000444 .close_on_exec = init_files.close_on_exec_init,
445 .open_fds = init_files.open_fds_init,
Al Virof52111b2008-05-08 18:19:16 -0400446 },
Thomas Gleixnereece09e2011-07-17 21:25:03 +0200447 .file_lock = __SPIN_LOCK_UNLOCKED(init_files.file_lock),
Al Virof52111b2008-05-08 18:19:16 -0400448};
Al Viro1027abe2008-07-30 04:13:04 -0400449
450/*
451 * allocate a file descriptor, mark it busy.
452 */
Al Virodcfadfa2012-08-12 17:27:30 -0400453int __alloc_fd(struct files_struct *files,
454 unsigned start, unsigned end, unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400455{
Al Viro1027abe2008-07-30 04:13:04 -0400456 unsigned int fd;
457 int error;
458 struct fdtable *fdt;
459
460 spin_lock(&files->file_lock);
461repeat:
462 fdt = files_fdtable(files);
463 fd = start;
464 if (fd < files->next_fd)
465 fd = files->next_fd;
466
467 if (fd < fdt->max_fds)
David Howells1fd36ad2012-02-16 17:49:54 +0000468 fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd);
Al Viro1027abe2008-07-30 04:13:04 -0400469
Al Virof33ff992012-08-12 16:17:59 -0400470 /*
471 * N.B. For clone tasks sharing a files structure, this test
472 * will limit the total number of files that can be opened.
473 */
474 error = -EMFILE;
475 if (fd >= end)
476 goto out;
477
Al Viro1027abe2008-07-30 04:13:04 -0400478 error = expand_files(files, fd);
479 if (error < 0)
480 goto out;
481
482 /*
483 * If we needed to expand the fs array we
484 * might have blocked - try again.
485 */
486 if (error)
487 goto repeat;
488
489 if (start <= files->next_fd)
490 files->next_fd = fd + 1;
491
David Howells1dce27c2012-02-16 17:49:42 +0000492 __set_open_fd(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400493 if (flags & O_CLOEXEC)
David Howells1dce27c2012-02-16 17:49:42 +0000494 __set_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400495 else
David Howells1dce27c2012-02-16 17:49:42 +0000496 __clear_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400497 error = fd;
498#if 1
499 /* Sanity check */
Paul E. McKenney7dc52152010-02-22 17:04:52 -0800500 if (rcu_dereference_raw(fdt->fd[fd]) != NULL) {
Al Viro1027abe2008-07-30 04:13:04 -0400501 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
502 rcu_assign_pointer(fdt->fd[fd], NULL);
503 }
504#endif
505
506out:
507 spin_unlock(&files->file_lock);
508 return error;
509}
510
Al Viroad47bd72012-08-21 20:11:34 -0400511static int alloc_fd(unsigned start, unsigned flags)
Al Virodcfadfa2012-08-12 17:27:30 -0400512{
513 return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
514}
515
Al Viro1a7bd222012-08-12 17:18:05 -0400516int get_unused_fd_flags(unsigned flags)
Al Viro1027abe2008-07-30 04:13:04 -0400517{
Al Virodcfadfa2012-08-12 17:27:30 -0400518 return __alloc_fd(current->files, 0, rlimit(RLIMIT_NOFILE), flags);
Al Viro1027abe2008-07-30 04:13:04 -0400519}
Al Viro1a7bd222012-08-12 17:18:05 -0400520EXPORT_SYMBOL(get_unused_fd_flags);
Al Viro56007ca2012-08-15 21:03:26 -0400521
522static void __put_unused_fd(struct files_struct *files, unsigned int fd)
523{
524 struct fdtable *fdt = files_fdtable(files);
525 __clear_open_fd(fd, fdt);
526 if (fd < files->next_fd)
527 files->next_fd = fd;
528}
529
530void put_unused_fd(unsigned int fd)
531{
532 struct files_struct *files = current->files;
533 spin_lock(&files->file_lock);
534 __put_unused_fd(files, fd);
535 spin_unlock(&files->file_lock);
536}
537
538EXPORT_SYMBOL(put_unused_fd);
539
540/*
541 * Install a file pointer in the fd array.
542 *
543 * The VFS is full of places where we drop the files lock between
544 * setting the open_fds bitmap and installing the file in the file
545 * array. At any such point, we are vulnerable to a dup2() race
546 * installing a file in the array before us. We need to detect this and
547 * fput() the struct file we are about to overwrite in this case.
548 *
549 * It should never happen - if we allow dup2() do it, _really_ bad things
550 * will follow.
Al Virof869e8a2012-08-15 21:06:33 -0400551 *
552 * NOTE: __fd_install() variant is really, really low-level; don't
553 * use it unless you are forced to by truly lousy API shoved down
554 * your throat. 'files' *MUST* be either current->files or obtained
555 * by get_files_struct(current) done by whoever had given it to you,
556 * or really bad things will happen. Normally you want to use
557 * fd_install() instead.
Al Viro56007ca2012-08-15 21:03:26 -0400558 */
559
Al Virof869e8a2012-08-15 21:06:33 -0400560void __fd_install(struct files_struct *files, unsigned int fd,
561 struct file *file)
Al Viro56007ca2012-08-15 21:03:26 -0400562{
Al Viro56007ca2012-08-15 21:03:26 -0400563 struct fdtable *fdt;
564 spin_lock(&files->file_lock);
565 fdt = files_fdtable(files);
566 BUG_ON(fdt->fd[fd] != NULL);
567 rcu_assign_pointer(fdt->fd[fd], file);
568 spin_unlock(&files->file_lock);
569}
570
Al Virof869e8a2012-08-15 21:06:33 -0400571void fd_install(unsigned int fd, struct file *file)
572{
573 __fd_install(current->files, fd, file);
574}
575
Al Viro56007ca2012-08-15 21:03:26 -0400576EXPORT_SYMBOL(fd_install);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400577
Al Viro483ce1d2012-08-19 12:04:24 -0400578/*
579 * The same warnings as for __alloc_fd()/__fd_install() apply here...
580 */
581int __close_fd(struct files_struct *files, unsigned fd)
582{
583 struct file *file;
584 struct fdtable *fdt;
585
586 spin_lock(&files->file_lock);
587 fdt = files_fdtable(files);
588 if (fd >= fdt->max_fds)
589 goto out_unlock;
590 file = fdt->fd[fd];
591 if (!file)
592 goto out_unlock;
593 rcu_assign_pointer(fdt->fd[fd], NULL);
594 __clear_close_on_exec(fd, fdt);
595 __put_unused_fd(files, fd);
596 spin_unlock(&files->file_lock);
597 return filp_close(file, files);
598
599out_unlock:
600 spin_unlock(&files->file_lock);
601 return -EBADF;
602}
603
Al Viro6a6d27d2012-08-21 09:56:33 -0400604void do_close_on_exec(struct files_struct *files)
605{
606 unsigned i;
607 struct fdtable *fdt;
608
609 /* exec unshares first */
Al Viro6a6d27d2012-08-21 09:56:33 -0400610 spin_lock(&files->file_lock);
611 for (i = 0; ; i++) {
612 unsigned long set;
613 unsigned fd = i * BITS_PER_LONG;
614 fdt = files_fdtable(files);
615 if (fd >= fdt->max_fds)
616 break;
617 set = fdt->close_on_exec[i];
618 if (!set)
619 continue;
620 fdt->close_on_exec[i] = 0;
621 for ( ; set ; fd++, set >>= 1) {
622 struct file *file;
623 if (!(set & 1))
624 continue;
625 file = fdt->fd[fd];
626 if (!file)
627 continue;
628 rcu_assign_pointer(fdt->fd[fd], NULL);
629 __put_unused_fd(files, fd);
630 spin_unlock(&files->file_lock);
631 filp_close(file, files);
632 cond_resched();
633 spin_lock(&files->file_lock);
634 }
635
636 }
637 spin_unlock(&files->file_lock);
638}
639
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100640static struct file *__fget(unsigned int fd, fmode_t mask)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400641{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400642 struct files_struct *files = current->files;
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100643 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400644
645 rcu_read_lock();
646 file = fcheck_files(files, fd);
647 if (file) {
648 /* File object ref couldn't be taken */
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100649 if ((file->f_mode & mask) ||
Al Viro0ee8cdf2012-08-15 21:12:10 -0400650 !atomic_long_inc_not_zero(&file->f_count))
651 file = NULL;
652 }
653 rcu_read_unlock();
654
655 return file;
656}
657
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100658struct file *fget(unsigned int fd)
659{
660 return __fget(fd, FMODE_PATH);
661}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400662EXPORT_SYMBOL(fget);
663
664struct file *fget_raw(unsigned int fd)
665{
Oleg Nesterov1deb46e2014-01-13 16:48:19 +0100666 return __fget(fd, 0);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400667}
Al Viro0ee8cdf2012-08-15 21:12:10 -0400668EXPORT_SYMBOL(fget_raw);
669
670/*
671 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
672 *
673 * You can use this instead of fget if you satisfy all of the following
674 * conditions:
675 * 1) You must call fput_light before exiting the syscall and returning control
676 * to userspace (i.e. you cannot remember the returned struct file * after
677 * returning to userspace).
678 * 2) You must not call filp_close on the returned struct file * in between
679 * calls to fget_light and fput_light.
680 * 3) You must not clone the current task in between the calls to fget_light
681 * and fput_light.
682 *
683 * The fput_needed flag returned by fget_light should be passed to the
684 * corresponding fput_light.
685 */
Oleg Nesterovad461832014-01-13 16:48:40 +0100686struct file *__fget_light(unsigned int fd, fmode_t mask, int *fput_needed)
Al Viro0ee8cdf2012-08-15 21:12:10 -0400687{
Al Viro0ee8cdf2012-08-15 21:12:10 -0400688 struct files_struct *files = current->files;
Oleg Nesterovad461832014-01-13 16:48:40 +0100689 struct file *file;
Al Viro0ee8cdf2012-08-15 21:12:10 -0400690
691 *fput_needed = 0;
692 if (atomic_read(&files->count) == 1) {
Oleg Nesterova8d4b832014-01-11 19:19:32 +0100693 file = __fcheck_files(files, fd);
Oleg Nesterovad461832014-01-13 16:48:40 +0100694 if (file && (file->f_mode & mask))
Al Viro0ee8cdf2012-08-15 21:12:10 -0400695 file = NULL;
696 } else {
697 rcu_read_lock();
698 file = fcheck_files(files, fd);
699 if (file) {
Oleg Nesterovad461832014-01-13 16:48:40 +0100700 if (!(file->f_mode & mask) &&
Al Viro0ee8cdf2012-08-15 21:12:10 -0400701 atomic_long_inc_not_zero(&file->f_count))
702 *fput_needed = 1;
703 else
704 /* Didn't get the reference, someone's freed */
705 file = NULL;
706 }
707 rcu_read_unlock();
708 }
709
710 return file;
711}
Oleg Nesterovad461832014-01-13 16:48:40 +0100712struct file *fget_light(unsigned int fd, int *fput_needed)
713{
714 return __fget_light(fd, FMODE_PATH, fput_needed);
715}
Al Viro4557c662012-08-28 10:19:41 -0400716EXPORT_SYMBOL(fget_light);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400717
718struct file *fget_raw_light(unsigned int fd, int *fput_needed)
719{
Oleg Nesterovad461832014-01-13 16:48:40 +0100720 return __fget_light(fd, 0, fput_needed);
Al Viro0ee8cdf2012-08-15 21:12:10 -0400721}
Al Virofe17f222012-08-21 11:48:11 -0400722
723void set_close_on_exec(unsigned int fd, int flag)
724{
725 struct files_struct *files = current->files;
726 struct fdtable *fdt;
727 spin_lock(&files->file_lock);
728 fdt = files_fdtable(files);
729 if (flag)
730 __set_close_on_exec(fd, fdt);
731 else
732 __clear_close_on_exec(fd, fdt);
733 spin_unlock(&files->file_lock);
734}
735
736bool get_close_on_exec(unsigned int fd)
737{
738 struct files_struct *files = current->files;
739 struct fdtable *fdt;
740 bool res;
741 rcu_read_lock();
742 fdt = files_fdtable(files);
743 res = close_on_exec(fd, fdt);
744 rcu_read_unlock();
745 return res;
746}
747
Al Viro8280d162012-08-21 12:11:46 -0400748static int do_dup2(struct files_struct *files,
749 struct file *file, unsigned fd, unsigned flags)
750{
751 struct file *tofree;
752 struct fdtable *fdt;
753
754 /*
755 * We need to detect attempts to do dup2() over allocated but still
756 * not finished descriptor. NB: OpenBSD avoids that at the price of
757 * extra work in their equivalent of fget() - they insert struct
758 * file immediately after grabbing descriptor, mark it larval if
759 * more work (e.g. actual opening) is needed and make sure that
760 * fget() treats larval files as absent. Potentially interesting,
761 * but while extra work in fget() is trivial, locking implications
762 * and amount of surgery on open()-related paths in VFS are not.
763 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
764 * deadlocks in rather amusing ways, AFAICS. All of that is out of
765 * scope of POSIX or SUS, since neither considers shared descriptor
766 * tables and this condition does not arise without those.
767 */
768 fdt = files_fdtable(files);
769 tofree = fdt->fd[fd];
770 if (!tofree && fd_is_open(fd, fdt))
771 goto Ebusy;
772 get_file(file);
773 rcu_assign_pointer(fdt->fd[fd], file);
774 __set_open_fd(fd, fdt);
775 if (flags & O_CLOEXEC)
776 __set_close_on_exec(fd, fdt);
777 else
778 __clear_close_on_exec(fd, fdt);
779 spin_unlock(&files->file_lock);
780
781 if (tofree)
782 filp_close(tofree, files);
783
784 return fd;
785
786Ebusy:
787 spin_unlock(&files->file_lock);
788 return -EBUSY;
789}
790
791int replace_fd(unsigned fd, struct file *file, unsigned flags)
792{
793 int err;
794 struct files_struct *files = current->files;
795
796 if (!file)
797 return __close_fd(files, fd);
798
799 if (fd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000800 return -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400801
802 spin_lock(&files->file_lock);
803 err = expand_files(files, fd);
804 if (unlikely(err < 0))
805 goto out_unlock;
806 return do_dup2(files, file, fd, flags);
807
808out_unlock:
809 spin_unlock(&files->file_lock);
810 return err;
811}
812
Al Virofe17f222012-08-21 11:48:11 -0400813SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
814{
815 int err = -EBADF;
Al Viro8280d162012-08-21 12:11:46 -0400816 struct file *file;
817 struct files_struct *files = current->files;
Al Virofe17f222012-08-21 11:48:11 -0400818
819 if ((flags & ~O_CLOEXEC) != 0)
820 return -EINVAL;
821
Richard W.M. Jonesaed97642012-10-09 15:27:43 +0100822 if (unlikely(oldfd == newfd))
823 return -EINVAL;
824
Al Virofe17f222012-08-21 11:48:11 -0400825 if (newfd >= rlimit(RLIMIT_NOFILE))
Al Viro08f05c42012-10-31 03:37:48 +0000826 return -EBADF;
Al Virofe17f222012-08-21 11:48:11 -0400827
828 spin_lock(&files->file_lock);
829 err = expand_files(files, newfd);
830 file = fcheck(oldfd);
831 if (unlikely(!file))
832 goto Ebadf;
833 if (unlikely(err < 0)) {
834 if (err == -EMFILE)
835 goto Ebadf;
836 goto out_unlock;
837 }
Al Viro8280d162012-08-21 12:11:46 -0400838 return do_dup2(files, file, newfd, flags);
Al Virofe17f222012-08-21 11:48:11 -0400839
840Ebadf:
841 err = -EBADF;
842out_unlock:
843 spin_unlock(&files->file_lock);
844 return err;
845}
846
847SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
848{
849 if (unlikely(newfd == oldfd)) { /* corner case */
850 struct files_struct *files = current->files;
851 int retval = oldfd;
852
853 rcu_read_lock();
854 if (!fcheck_files(files, oldfd))
855 retval = -EBADF;
856 rcu_read_unlock();
857 return retval;
858 }
859 return sys_dup3(oldfd, newfd, 0);
860}
861
862SYSCALL_DEFINE1(dup, unsigned int, fildes)
863{
864 int ret = -EBADF;
865 struct file *file = fget_raw(fildes);
866
867 if (file) {
868 ret = get_unused_fd();
869 if (ret >= 0)
870 fd_install(ret, file);
871 else
872 fput(file);
873 }
874 return ret;
875}
876
877int f_dupfd(unsigned int from, struct file *file, unsigned flags)
878{
879 int err;
880 if (from >= rlimit(RLIMIT_NOFILE))
881 return -EINVAL;
882 err = alloc_fd(from, flags);
883 if (err >= 0) {
884 get_file(file);
885 fd_install(err, file);
886 }
887 return err;
888}
Al Viroc3c073f2012-08-21 22:32:06 -0400889
890int iterate_fd(struct files_struct *files, unsigned n,
891 int (*f)(const void *, struct file *, unsigned),
892 const void *p)
893{
894 struct fdtable *fdt;
Al Viroc3c073f2012-08-21 22:32:06 -0400895 int res = 0;
896 if (!files)
897 return 0;
898 spin_lock(&files->file_lock);
Al Viroa77cfcb2012-11-29 22:57:33 -0500899 for (fdt = files_fdtable(files); n < fdt->max_fds; n++) {
900 struct file *file;
901 file = rcu_dereference_check_fdtable(files, fdt->fd[n]);
902 if (!file)
903 continue;
904 res = f(p, file, n);
905 if (res)
906 break;
Al Viroc3c073f2012-08-21 22:32:06 -0400907 }
908 spin_unlock(&files->file_lock);
909 return res;
910}
911EXPORT_SYMBOL(iterate_fd);