blob: 55f4e70225631b275f85215ee543b104507caacc [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
9#include <linux/fs.h>
10#include <linux/mm.h>
11#include <linux/time.h>
12#include <linux/slab.h>
13#include <linux/vmalloc.h>
14#include <linux/file.h>
15#include <linux/bitops.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070016#include <linux/interrupt.h>
17#include <linux/spinlock.h>
18#include <linux/rcupdate.h>
19#include <linux/workqueue.h>
20
21struct fdtable_defer {
22 spinlock_t lock;
23 struct work_struct wq;
24 struct timer_list timer;
25 struct fdtable *next;
26};
27
28/*
29 * We use this list to defer free fdtables that have vmalloced
30 * sets/arrays. By keeping a per-cpu list, we avoid having to embed
31 * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
32 * this per-task structure.
33 */
34static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
36
37/*
38 * Allocate an fd array, using kmalloc or vmalloc.
39 * Note: the array isn't cleared at allocation time.
40 */
41struct file ** alloc_fd_array(int num)
42{
43 struct file **new_fds;
44 int size = num * sizeof(struct file *);
45
46 if (size <= PAGE_SIZE)
47 new_fds = (struct file **) kmalloc(size, GFP_KERNEL);
48 else
49 new_fds = (struct file **) vmalloc(size);
50 return new_fds;
51}
52
53void free_fd_array(struct file **array, int num)
54{
55 int size = num * sizeof(struct file *);
56
57 if (!array) {
58 printk (KERN_ERR "free_fd_array: array = 0 (num = %d)\n", num);
59 return;
60 }
61
62 if (num <= NR_OPEN_DEFAULT) /* Don't free the embedded fd array! */
63 return;
64 else if (size <= PAGE_SIZE)
65 kfree(array);
66 else
67 vfree(array);
68}
69
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070070static void __free_fdtable(struct fdtable *fdt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
Dipankar Sarma0b175a72005-09-15 00:48:42 +053072 free_fdset(fdt->open_fds, fdt->max_fdset);
73 free_fdset(fdt->close_on_exec, fdt->max_fdset);
74 free_fd_array(fdt->fd, fdt->max_fds);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070075 kfree(fdt);
76}
77
78static void fdtable_timer(unsigned long data)
79{
80 struct fdtable_defer *fddef = (struct fdtable_defer *)data;
81
82 spin_lock(&fddef->lock);
83 /*
84 * If someone already emptied the queue return.
85 */
86 if (!fddef->next)
87 goto out;
88 if (!schedule_work(&fddef->wq))
89 mod_timer(&fddef->timer, 5);
90out:
91 spin_unlock(&fddef->lock);
92}
93
94static void free_fdtable_work(struct fdtable_defer *f)
95{
Dipankar Sarmabadf1662005-09-09 13:04:10 -070096 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070098 spin_lock_bh(&f->lock);
99 fdt = f->next;
100 f->next = NULL;
101 spin_unlock_bh(&f->lock);
102 while(fdt) {
103 struct fdtable *next = fdt->next;
104 __free_fdtable(fdt);
105 fdt = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700107}
108
109static void free_fdtable_rcu(struct rcu_head *rcu)
110{
111 struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
112 int fdset_size, fdarray_size;
113 struct fdtable_defer *fddef;
114
115 BUG_ON(!fdt);
116 fdset_size = fdt->max_fdset / 8;
117 fdarray_size = fdt->max_fds * sizeof(struct file *);
118
119 if (fdt->free_files) {
120 /*
121 * The this fdtable was embedded in the files structure
122 * and the files structure itself was getting destroyed.
123 * It is now safe to free the files structure.
124 */
125 kmem_cache_free(files_cachep, fdt->free_files);
126 return;
127 }
Eric Dumazet0c9e63f2006-03-23 03:00:12 -0800128 if (fdt->max_fdset <= EMBEDDED_FD_SET_SIZE &&
129 fdt->max_fds <= NR_OPEN_DEFAULT) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700130 /*
131 * The fdtable was embedded
132 */
133 return;
134 }
135 if (fdset_size <= PAGE_SIZE && fdarray_size <= PAGE_SIZE) {
136 kfree(fdt->open_fds);
137 kfree(fdt->close_on_exec);
138 kfree(fdt->fd);
139 kfree(fdt);
140 } else {
141 fddef = &get_cpu_var(fdtable_defer_list);
142 spin_lock(&fddef->lock);
143 fdt->next = fddef->next;
144 fddef->next = fdt;
145 /*
146 * vmallocs are handled from the workqueue context.
147 * If the per-cpu workqueue is running, then we
148 * defer work scheduling through a timer.
149 */
150 if (!schedule_work(&fddef->wq))
151 mod_timer(&fddef->timer, 5);
152 spin_unlock(&fddef->lock);
153 put_cpu_var(fdtable_defer_list);
154 }
155}
156
157void free_fdtable(struct fdtable *fdt)
158{
Eric Dumazet0c9e63f2006-03-23 03:00:12 -0800159 if (fdt->free_files ||
160 fdt->max_fdset > EMBEDDED_FD_SET_SIZE ||
161 fdt->max_fds > NR_OPEN_DEFAULT)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700162 call_rcu(&fdt->rcu, free_fdtable_rcu);
163}
164
165/*
166 * Expand the fdset in the files_struct. Called with the files spinlock
167 * held for write.
168 */
169static void copy_fdtable(struct fdtable *nfdt, struct fdtable *fdt)
170{
171 int i;
172 int count;
173
174 BUG_ON(nfdt->max_fdset < fdt->max_fdset);
175 BUG_ON(nfdt->max_fds < fdt->max_fds);
176 /* Copy the existing tables and install the new pointers */
177
178 i = fdt->max_fdset / (sizeof(unsigned long) * 8);
179 count = (nfdt->max_fdset - fdt->max_fdset) / 8;
180
181 /*
182 * Don't copy the entire array if the current fdset is
183 * not yet initialised.
184 */
185 if (i) {
186 memcpy (nfdt->open_fds, fdt->open_fds,
187 fdt->max_fdset/8);
188 memcpy (nfdt->close_on_exec, fdt->close_on_exec,
189 fdt->max_fdset/8);
190 memset (&nfdt->open_fds->fds_bits[i], 0, count);
191 memset (&nfdt->close_on_exec->fds_bits[i], 0, count);
192 }
193
194 /* Don't copy/clear the array if we are creating a new
195 fd array for fork() */
196 if (fdt->max_fds) {
197 memcpy(nfdt->fd, fdt->fd,
198 fdt->max_fds * sizeof(struct file *));
199 /* clear the remainder of the array */
200 memset(&nfdt->fd[fdt->max_fds], 0,
201 (nfdt->max_fds - fdt->max_fds) *
202 sizeof(struct file *));
203 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204}
205
206/*
207 * Allocate an fdset array, using kmalloc or vmalloc.
208 * Note: the array isn't cleared at allocation time.
209 */
210fd_set * alloc_fdset(int num)
211{
212 fd_set *new_fdset;
213 int size = num / 8;
214
215 if (size <= PAGE_SIZE)
216 new_fdset = (fd_set *) kmalloc(size, GFP_KERNEL);
217 else
218 new_fdset = (fd_set *) vmalloc(size);
219 return new_fdset;
220}
221
222void free_fdset(fd_set *array, int num)
223{
Eric Dumazet0c9e63f2006-03-23 03:00:12 -0800224 if (num <= EMBEDDED_FD_SET_SIZE) /* Don't free an embedded fdset */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 return;
Eric Dumazet0c9e63f2006-03-23 03:00:12 -0800226 else if (num <= 8 * PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 kfree(array);
228 else
229 vfree(array);
230}
231
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700232static struct fdtable *alloc_fdtable(int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700234 struct fdtable *fdt = NULL;
235 int nfds = 0;
236 fd_set *new_openset = NULL, *new_execset = NULL;
237 struct file **new_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
Eric Dumazet0c9e63f2006-03-23 03:00:12 -0800239 fdt = kzalloc(sizeof(*fdt), GFP_KERNEL);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700240 if (!fdt)
241 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
Eric Dumazet0c9e63f2006-03-23 03:00:12 -0800243 nfds = 8 * L1_CACHE_BYTES;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700244 /* Expand to the max in easy steps */
Eric Dumazet0c9e63f2006-03-23 03:00:12 -0800245 while (nfds <= nr) {
246 nfds = nfds * 2;
247 if (nfds > NR_OPEN)
248 nfds = NR_OPEN;
249 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700251 new_openset = alloc_fdset(nfds);
252 new_execset = alloc_fdset(nfds);
253 if (!new_openset || !new_execset)
254 goto out;
255 fdt->open_fds = new_openset;
256 fdt->close_on_exec = new_execset;
257 fdt->max_fdset = nfds;
258
259 nfds = NR_OPEN_DEFAULT;
260 /*
261 * Expand to the max in easy steps, and keep expanding it until
262 * we have enough for the requested fd array size.
263 */
264 do {
265#if NR_OPEN_DEFAULT < 256
266 if (nfds < 256)
267 nfds = 256;
268 else
269#endif
270 if (nfds < (PAGE_SIZE / sizeof(struct file *)))
271 nfds = PAGE_SIZE / sizeof(struct file *);
272 else {
273 nfds = nfds * 2;
274 if (nfds > NR_OPEN)
275 nfds = NR_OPEN;
276 }
277 } while (nfds <= nr);
278 new_fds = alloc_fd_array(nfds);
279 if (!new_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 goto out;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700281 fdt->fd = new_fds;
282 fdt->max_fds = nfds;
283 fdt->free_files = NULL;
284 return fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700286 if (new_openset)
287 free_fdset(new_openset, nfds);
288 if (new_execset)
289 free_fdset(new_execset, nfds);
290 kfree(fdt);
291 return NULL;
292}
293
294/*
295 * Expands the file descriptor table - it will allocate a new fdtable and
296 * both fd array and fdset. It is expected to be called with the
297 * files_lock held.
298 */
299static int expand_fdtable(struct files_struct *files, int nr)
300 __releases(files->file_lock)
301 __acquires(files->file_lock)
302{
303 int error = 0;
304 struct fdtable *fdt;
305 struct fdtable *nfdt = NULL;
306
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 spin_unlock(&files->file_lock);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700308 nfdt = alloc_fdtable(nr);
309 if (!nfdt) {
310 error = -ENOMEM;
311 spin_lock(&files->file_lock);
312 goto out;
313 }
314
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 spin_lock(&files->file_lock);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700316 fdt = files_fdtable(files);
317 /*
318 * Check again since another task may have expanded the
319 * fd table while we dropped the lock
320 */
321 if (nr >= fdt->max_fds || nr >= fdt->max_fdset) {
322 copy_fdtable(nfdt, fdt);
323 } else {
324 /* Somebody expanded while we dropped file_lock */
325 spin_unlock(&files->file_lock);
326 __free_fdtable(nfdt);
327 spin_lock(&files->file_lock);
328 goto out;
329 }
330 rcu_assign_pointer(files->fdt, nfdt);
331 free_fdtable(fdt);
332out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 return error;
334}
335
336/*
337 * Expand files.
338 * Return <0 on error; 0 nothing done; 1 files expanded, we may have blocked.
339 * Should be called with the files->file_lock spinlock held for write.
340 */
341int expand_files(struct files_struct *files, int nr)
342{
343 int err, expand = 0;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700344 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700346 fdt = files_fdtable(files);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700347 if (nr >= fdt->max_fdset || nr >= fdt->max_fds) {
348 if (fdt->max_fdset >= NR_OPEN ||
349 fdt->max_fds >= NR_OPEN || nr >= NR_OPEN) {
350 err = -EMFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 goto out;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700352 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 expand = 1;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700354 if ((err = expand_fdtable(files, nr)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 goto out;
356 }
357 err = expand;
358out:
359 return err;
360}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700361
362static void __devinit fdtable_defer_list_init(int cpu)
363{
364 struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
365 spin_lock_init(&fddef->lock);
366 INIT_WORK(&fddef->wq, (void (*)(void *))free_fdtable_work, fddef);
367 init_timer(&fddef->timer);
368 fddef->timer.data = (unsigned long)fddef;
369 fddef->timer.function = fdtable_timer;
370 fddef->next = NULL;
371}
372
373void __init files_defer_init(void)
374{
375 int i;
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800376 for_each_possible_cpu(i)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700377 fdtable_defer_list_init(i);
378}