blob: ba3f6053025cf44915ebd5ca42605d3281429d33 [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
Paul Gortmaker630d9c42011-11-16 23:57:37 -05009#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/fs.h>
11#include <linux/mm.h>
Andrew Morton6d4831c2011-04-27 15:26:41 -070012#include <linux/mmzone.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/time.h>
Alexey Dobriyand43c36d2009-10-07 17:09:06 +040014#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/slab.h>
16#include <linux/vmalloc.h>
17#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040018#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/bitops.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070020#include <linux/interrupt.h>
21#include <linux/spinlock.h>
22#include <linux/rcupdate.h>
23#include <linux/workqueue.h>
24
25struct fdtable_defer {
26 spinlock_t lock;
27 struct work_struct wq;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070028 struct fdtable *next;
29};
30
Eric Dumazet9cfe0152008-02-06 01:37:16 -080031int sysctl_nr_open __read_mostly = 1024*1024;
Al Viroeceea0b2008-05-10 10:08:32 -040032int sysctl_nr_open_min = BITS_PER_LONG;
33int sysctl_nr_open_max = 1024 * 1024; /* raised later */
Eric Dumazet9cfe0152008-02-06 01:37:16 -080034
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070035/*
36 * We use this list to defer free fdtables that have vmalloced
37 * sets/arrays. By keeping a per-cpu list, we avoid having to embed
38 * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
39 * this per-task structure.
40 */
41static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
David Howells1fd36ad2012-02-16 17:49:54 +000043static void *alloc_fdmem(size_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -070044{
Andrew Morton6d4831c2011-04-27 15:26:41 -070045 /*
46 * Very large allocations can stress page reclaim, so fall back to
47 * vmalloc() if the allocation size will be considered "large" by the VM.
48 */
49 if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
50 void *data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN);
51 if (data != NULL)
52 return data;
53 }
Changli Gaoa892e2d2010-08-10 18:01:35 -070054 return vmalloc(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055}
56
Changli Gaoa892e2d2010-08-10 18:01:35 -070057static void free_fdmem(void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058{
Changli Gaoa892e2d2010-08-10 18:01:35 -070059 is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr);
Vadim Lobanov5466b452006-12-10 02:21:22 -080060}
61
Changli Gaoa892e2d2010-08-10 18:01:35 -070062static void __free_fdtable(struct fdtable *fdt)
Vadim Lobanov5466b452006-12-10 02:21:22 -080063{
Changli Gaoa892e2d2010-08-10 18:01:35 -070064 free_fdmem(fdt->fd);
65 free_fdmem(fdt->open_fds);
66 kfree(fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070067}
68
David Howells65f27f32006-11-22 14:55:48 +000069static void free_fdtable_work(struct work_struct *work)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070070{
David Howells65f27f32006-11-22 14:55:48 +000071 struct fdtable_defer *f =
72 container_of(work, struct fdtable_defer, wq);
Dipankar Sarmabadf1662005-09-09 13:04:10 -070073 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070075 spin_lock_bh(&f->lock);
76 fdt = f->next;
77 f->next = NULL;
78 spin_unlock_bh(&f->lock);
79 while(fdt) {
80 struct fdtable *next = fdt->next;
Changli Gaoa892e2d2010-08-10 18:01:35 -070081
82 __free_fdtable(fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070083 fdt = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070085}
86
Vadim Lobanov4fd45812006-12-10 02:21:17 -080087void free_fdtable_rcu(struct rcu_head *rcu)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070088{
89 struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070090 struct fdtable_defer *fddef;
91
92 BUG_ON(!fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070093
Vadim Lobanov4fd45812006-12-10 02:21:17 -080094 if (fdt->max_fds <= NR_OPEN_DEFAULT) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070095 /*
Vadim Lobanov4fd45812006-12-10 02:21:17 -080096 * This fdtable is embedded in the files structure and that
97 * structure itself is getting destroyed.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070098 */
Vadim Lobanov4fd45812006-12-10 02:21:17 -080099 kmem_cache_free(files_cachep,
100 container_of(fdt, struct files_struct, fdtab));
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700101 return;
102 }
Changli Gaoa892e2d2010-08-10 18:01:35 -0700103 if (!is_vmalloc_addr(fdt->fd) && !is_vmalloc_addr(fdt->open_fds)) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700104 kfree(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800105 kfree(fdt->open_fds);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700106 kfree(fdt);
107 } else {
108 fddef = &get_cpu_var(fdtable_defer_list);
109 spin_lock(&fddef->lock);
110 fdt->next = fddef->next;
111 fddef->next = fdt;
Tejun Heo593be072006-12-06 20:36:01 -0800112 /* vmallocs are handled from the workqueue context */
113 schedule_work(&fddef->wq);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700114 spin_unlock(&fddef->lock);
115 put_cpu_var(fdtable_defer_list);
116 }
117}
118
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700119/*
120 * Expand the fdset in the files_struct. Called with the files spinlock
121 * held for write.
122 */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800123static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700124{
Vadim Lobanov5466b452006-12-10 02:21:22 -0800125 unsigned int cpy, set;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700126
Vadim Lobanov5466b452006-12-10 02:21:22 -0800127 BUG_ON(nfdt->max_fds < ofdt->max_fds);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800128
129 cpy = ofdt->max_fds * sizeof(struct file *);
130 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
131 memcpy(nfdt->fd, ofdt->fd, cpy);
132 memset((char *)(nfdt->fd) + cpy, 0, set);
133
134 cpy = ofdt->max_fds / BITS_PER_BYTE;
135 set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
136 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
137 memset((char *)(nfdt->open_fds) + cpy, 0, set);
138 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
139 memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140}
141
Vadim Lobanov5466b452006-12-10 02:21:22 -0800142static struct fdtable * alloc_fdtable(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
Vadim Lobanov5466b452006-12-10 02:21:22 -0800144 struct fdtable *fdt;
David Howells1fd36ad2012-02-16 17:49:54 +0000145 void *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700147 /*
Vadim Lobanov5466b452006-12-10 02:21:22 -0800148 * Figure out how many fds we actually want to support in this fdtable.
149 * Allocation steps are keyed to the size of the fdarray, since it
150 * grows far faster than any of the other dynamic data. We try to fit
151 * the fdarray into comfortable page-tuned chunks: starting at 1024B
152 * and growing in powers of two from there on.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700153 */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800154 nr /= (1024 / sizeof(struct file *));
155 nr = roundup_pow_of_two(nr + 1);
156 nr *= (1024 / sizeof(struct file *));
Al Viro5c598b32008-04-27 20:04:15 -0400157 /*
158 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
159 * had been set lower between the check in expand_files() and here. Deal
160 * with that in caller, it's cheaper that way.
161 *
162 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
163 * bitmaps handling below becomes unpleasant, to put it mildly...
164 */
165 if (unlikely(nr > sysctl_nr_open))
166 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800167
Vadim Lobanov5466b452006-12-10 02:21:22 -0800168 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
169 if (!fdt)
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800170 goto out;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800171 fdt->max_fds = nr;
172 data = alloc_fdmem(nr * sizeof(struct file *));
173 if (!data)
174 goto out_fdt;
David Howells1fd36ad2012-02-16 17:49:54 +0000175 fdt->fd = data;
176
177 data = alloc_fdmem(max_t(size_t,
Vadim Lobanov5466b452006-12-10 02:21:22 -0800178 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
179 if (!data)
180 goto out_arr;
David Howells1fd36ad2012-02-16 17:49:54 +0000181 fdt->open_fds = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800182 data += nr / BITS_PER_BYTE;
David Howells1fd36ad2012-02-16 17:49:54 +0000183 fdt->close_on_exec = data;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800184 fdt->next = NULL;
185
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700186 return fdt;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800187
188out_arr:
Changli Gaoa892e2d2010-08-10 18:01:35 -0700189 free_fdmem(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800190out_fdt:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700191 kfree(fdt);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800192out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700193 return NULL;
194}
195
196/*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700197 * Expand the file descriptor table.
198 * This function will allocate a new fdtable and both fd array and fdset, of
199 * the given size.
200 * Return <0 error code on error; 1 on successful completion.
201 * The files->file_lock should be held on entry, and will be held on exit.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700202 */
203static int expand_fdtable(struct files_struct *files, int nr)
204 __releases(files->file_lock)
205 __acquires(files->file_lock)
206{
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700207 struct fdtable *new_fdt, *cur_fdt;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700208
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 spin_unlock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700210 new_fdt = alloc_fdtable(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700212 if (!new_fdt)
213 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700214 /*
Al Viro5c598b32008-04-27 20:04:15 -0400215 * extremely unlikely race - sysctl_nr_open decreased between the check in
216 * caller and alloc_fdtable(). Cheaper to catch it here...
217 */
218 if (unlikely(new_fdt->max_fds <= nr)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700219 __free_fdtable(new_fdt);
Al Viro5c598b32008-04-27 20:04:15 -0400220 return -EMFILE;
221 }
222 /*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700223 * Check again since another task may have expanded the fd table while
224 * we dropped the lock
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700225 */
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700226 cur_fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800227 if (nr >= cur_fdt->max_fds) {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700228 /* Continue as planned */
229 copy_fdtable(new_fdt, cur_fdt);
230 rcu_assign_pointer(files->fdt, new_fdt);
Vadim Lobanov4fd45812006-12-10 02:21:17 -0800231 if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
Vadim Lobanov01b2d932006-12-22 01:10:43 -0800232 free_fdtable(cur_fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700233 } else {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700234 /* Somebody else expanded, so undo our attempt */
Changli Gaoa892e2d2010-08-10 18:01:35 -0700235 __free_fdtable(new_fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700236 }
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700237 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
240/*
241 * Expand files.
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700242 * This function will expand the file structures, if the requested size exceeds
243 * the current capacity and there is room for expansion.
244 * Return <0 error code on error; 0 when nothing done; 1 when files were
245 * expanded and execution may have blocked.
246 * The files->file_lock should be held on entry, and will be held on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 */
248int expand_files(struct files_struct *files, int nr)
249{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700250 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700252 fdt = files_fdtable(files);
Al Viro4e1e0182008-07-26 16:01:20 -0400253
254 /*
255 * N.B. For clone tasks sharing a files structure, this test
256 * will limit the total number of files that can be opened.
257 */
Jiri Slabyd554ed892010-03-05 13:42:42 -0800258 if (nr >= rlimit(RLIMIT_NOFILE))
Al Viro4e1e0182008-07-26 16:01:20 -0400259 return -EMFILE;
260
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700261 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800262 if (nr < fdt->max_fds)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700263 return 0;
Al Viro4e1e0182008-07-26 16:01:20 -0400264
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700265 /* Can we expand? */
Eric Dumazet9cfe0152008-02-06 01:37:16 -0800266 if (nr >= sysctl_nr_open)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700267 return -EMFILE;
268
269 /* All good, so we try */
270 return expand_fdtable(files, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700272
Al Viro02afc622008-05-08 19:42:56 -0400273static int count_open_files(struct fdtable *fdt)
274{
275 int size = fdt->max_fds;
276 int i;
277
278 /* Find the last open fd */
David Howells1fd36ad2012-02-16 17:49:54 +0000279 for (i = size / BITS_PER_LONG; i > 0; ) {
280 if (fdt->open_fds[--i])
Al Viro02afc622008-05-08 19:42:56 -0400281 break;
282 }
David Howells1fd36ad2012-02-16 17:49:54 +0000283 i = (i + 1) * BITS_PER_LONG;
Al Viro02afc622008-05-08 19:42:56 -0400284 return i;
285}
286
Al Viro02afc622008-05-08 19:42:56 -0400287/*
288 * Allocate a new files structure and copy contents from the
289 * passed in files structure.
290 * errorp will be valid only when the returned files_struct is NULL.
291 */
292struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
293{
294 struct files_struct *newf;
295 struct file **old_fds, **new_fds;
296 int open_files, size, i;
297 struct fdtable *old_fdt, *new_fdt;
298
299 *errorp = -ENOMEM;
Al Viroafbec7f2008-05-08 21:11:17 -0400300 newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
Al Viro02afc622008-05-08 19:42:56 -0400301 if (!newf)
302 goto out;
303
Al Viroafbec7f2008-05-08 21:11:17 -0400304 atomic_set(&newf->count, 1);
305
306 spin_lock_init(&newf->file_lock);
307 newf->next_fd = 0;
308 new_fdt = &newf->fdtab;
309 new_fdt->max_fds = NR_OPEN_DEFAULT;
David Howells1fd36ad2012-02-16 17:49:54 +0000310 new_fdt->close_on_exec = newf->close_on_exec_init;
311 new_fdt->open_fds = newf->open_fds_init;
Al Viroafbec7f2008-05-08 21:11:17 -0400312 new_fdt->fd = &newf->fd_array[0];
Al Viroafbec7f2008-05-08 21:11:17 -0400313 new_fdt->next = NULL;
314
Al Viro02afc622008-05-08 19:42:56 -0400315 spin_lock(&oldf->file_lock);
316 old_fdt = files_fdtable(oldf);
Al Viro02afc622008-05-08 19:42:56 -0400317 open_files = count_open_files(old_fdt);
318
319 /*
320 * Check whether we need to allocate a larger fd array and fd set.
Al Viro02afc622008-05-08 19:42:56 -0400321 */
Al Viroadbecb12008-05-08 21:19:42 -0400322 while (unlikely(open_files > new_fdt->max_fds)) {
Al Viro02afc622008-05-08 19:42:56 -0400323 spin_unlock(&oldf->file_lock);
Al Viro9dec3c42008-05-08 21:02:45 -0400324
Changli Gaoa892e2d2010-08-10 18:01:35 -0700325 if (new_fdt != &newf->fdtab)
326 __free_fdtable(new_fdt);
Al Viroadbecb12008-05-08 21:19:42 -0400327
Al Viro9dec3c42008-05-08 21:02:45 -0400328 new_fdt = alloc_fdtable(open_files - 1);
329 if (!new_fdt) {
330 *errorp = -ENOMEM;
Al Viro02afc622008-05-08 19:42:56 -0400331 goto out_release;
Al Viro9dec3c42008-05-08 21:02:45 -0400332 }
333
334 /* beyond sysctl_nr_open; nothing to do */
335 if (unlikely(new_fdt->max_fds < open_files)) {
Changli Gaoa892e2d2010-08-10 18:01:35 -0700336 __free_fdtable(new_fdt);
Al Viro9dec3c42008-05-08 21:02:45 -0400337 *errorp = -EMFILE;
338 goto out_release;
339 }
Al Viro9dec3c42008-05-08 21:02:45 -0400340
Al Viro02afc622008-05-08 19:42:56 -0400341 /*
342 * Reacquire the oldf lock and a pointer to its fd table
343 * who knows it may have a new bigger fd table. We need
344 * the latest pointer.
345 */
346 spin_lock(&oldf->file_lock);
347 old_fdt = files_fdtable(oldf);
Al Viroadbecb12008-05-08 21:19:42 -0400348 open_files = count_open_files(old_fdt);
Al Viro02afc622008-05-08 19:42:56 -0400349 }
350
351 old_fds = old_fdt->fd;
352 new_fds = new_fdt->fd;
353
David Howells1fd36ad2012-02-16 17:49:54 +0000354 memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8);
355 memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8);
Al Viro02afc622008-05-08 19:42:56 -0400356
357 for (i = open_files; i != 0; i--) {
358 struct file *f = *old_fds++;
359 if (f) {
360 get_file(f);
361 } else {
362 /*
363 * The fd may be claimed in the fd bitmap but not yet
364 * instantiated in the files array if a sibling thread
365 * is partway through open(). So make sure that this
366 * fd is available to the new process.
367 */
David Howells1dce27c2012-02-16 17:49:42 +0000368 __clear_open_fd(open_files - i, new_fdt);
Al Viro02afc622008-05-08 19:42:56 -0400369 }
370 rcu_assign_pointer(*new_fds++, f);
371 }
372 spin_unlock(&oldf->file_lock);
373
374 /* compute the remainder to be cleared */
375 size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
376
377 /* This is long word aligned thus could use a optimized version */
378 memset(new_fds, 0, size);
379
380 if (new_fdt->max_fds > open_files) {
David Howells1fd36ad2012-02-16 17:49:54 +0000381 int left = (new_fdt->max_fds - open_files) / 8;
382 int start = open_files / BITS_PER_LONG;
Al Viro02afc622008-05-08 19:42:56 -0400383
David Howells1fd36ad2012-02-16 17:49:54 +0000384 memset(&new_fdt->open_fds[start], 0, left);
385 memset(&new_fdt->close_on_exec[start], 0, left);
Al Viro02afc622008-05-08 19:42:56 -0400386 }
387
Al Viroafbec7f2008-05-08 21:11:17 -0400388 rcu_assign_pointer(newf->fdt, new_fdt);
389
Al Viro02afc622008-05-08 19:42:56 -0400390 return newf;
391
392out_release:
393 kmem_cache_free(files_cachep, newf);
394out:
395 return NULL;
396}
397
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700398static void __devinit fdtable_defer_list_init(int cpu)
399{
400 struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
401 spin_lock_init(&fddef->lock);
David Howells65f27f32006-11-22 14:55:48 +0000402 INIT_WORK(&fddef->wq, free_fdtable_work);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700403 fddef->next = NULL;
404}
405
406void __init files_defer_init(void)
407{
408 int i;
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800409 for_each_possible_cpu(i)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700410 fdtable_defer_list_init(i);
Al Viroeceea0b2008-05-10 10:08:32 -0400411 sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
412 -BITS_PER_LONG;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700413}
Al Virof52111b2008-05-08 18:19:16 -0400414
415struct files_struct init_files = {
416 .count = ATOMIC_INIT(1),
417 .fdt = &init_files.fdtab,
418 .fdtab = {
419 .max_fds = NR_OPEN_DEFAULT,
420 .fd = &init_files.fd_array[0],
David Howells1fd36ad2012-02-16 17:49:54 +0000421 .close_on_exec = init_files.close_on_exec_init,
422 .open_fds = init_files.open_fds_init,
Al Virof52111b2008-05-08 18:19:16 -0400423 },
424 .file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock),
425};
Al Viro1027abe2008-07-30 04:13:04 -0400426
427/*
428 * allocate a file descriptor, mark it busy.
429 */
430int alloc_fd(unsigned start, unsigned flags)
431{
432 struct files_struct *files = current->files;
433 unsigned int fd;
434 int error;
435 struct fdtable *fdt;
436
437 spin_lock(&files->file_lock);
438repeat:
439 fdt = files_fdtable(files);
440 fd = start;
441 if (fd < files->next_fd)
442 fd = files->next_fd;
443
444 if (fd < fdt->max_fds)
David Howells1fd36ad2012-02-16 17:49:54 +0000445 fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd);
Al Viro1027abe2008-07-30 04:13:04 -0400446
447 error = expand_files(files, fd);
448 if (error < 0)
449 goto out;
450
451 /*
452 * If we needed to expand the fs array we
453 * might have blocked - try again.
454 */
455 if (error)
456 goto repeat;
457
458 if (start <= files->next_fd)
459 files->next_fd = fd + 1;
460
David Howells1dce27c2012-02-16 17:49:42 +0000461 __set_open_fd(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400462 if (flags & O_CLOEXEC)
David Howells1dce27c2012-02-16 17:49:42 +0000463 __set_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400464 else
David Howells1dce27c2012-02-16 17:49:42 +0000465 __clear_close_on_exec(fd, fdt);
Al Viro1027abe2008-07-30 04:13:04 -0400466 error = fd;
467#if 1
468 /* Sanity check */
Paul E. McKenney7dc52152010-02-22 17:04:52 -0800469 if (rcu_dereference_raw(fdt->fd[fd]) != NULL) {
Al Viro1027abe2008-07-30 04:13:04 -0400470 printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
471 rcu_assign_pointer(fdt->fd[fd], NULL);
472 }
473#endif
474
475out:
476 spin_unlock(&files->file_lock);
477 return error;
478}
479
480int get_unused_fd(void)
481{
482 return alloc_fd(0, 0);
483}
484EXPORT_SYMBOL(get_unused_fd);