blob: cea7cbea11d0d5fca944ab11a331568fe0949fb7 [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 }
128 if (fdt->max_fdset <= __FD_SETSIZE && fdt->max_fds <= NR_OPEN_DEFAULT) {
129 /*
130 * The fdtable was embedded
131 */
132 return;
133 }
134 if (fdset_size <= PAGE_SIZE && fdarray_size <= PAGE_SIZE) {
135 kfree(fdt->open_fds);
136 kfree(fdt->close_on_exec);
137 kfree(fdt->fd);
138 kfree(fdt);
139 } else {
140 fddef = &get_cpu_var(fdtable_defer_list);
141 spin_lock(&fddef->lock);
142 fdt->next = fddef->next;
143 fddef->next = fdt;
144 /*
145 * vmallocs are handled from the workqueue context.
146 * If the per-cpu workqueue is running, then we
147 * defer work scheduling through a timer.
148 */
149 if (!schedule_work(&fddef->wq))
150 mod_timer(&fddef->timer, 5);
151 spin_unlock(&fddef->lock);
152 put_cpu_var(fdtable_defer_list);
153 }
154}
155
156void free_fdtable(struct fdtable *fdt)
157{
158 if (fdt->free_files || fdt->max_fdset > __FD_SETSIZE ||
159 fdt->max_fds > NR_OPEN_DEFAULT)
160 call_rcu(&fdt->rcu, free_fdtable_rcu);
161}
162
163/*
164 * Expand the fdset in the files_struct. Called with the files spinlock
165 * held for write.
166 */
167static void copy_fdtable(struct fdtable *nfdt, struct fdtable *fdt)
168{
169 int i;
170 int count;
171
172 BUG_ON(nfdt->max_fdset < fdt->max_fdset);
173 BUG_ON(nfdt->max_fds < fdt->max_fds);
174 /* Copy the existing tables and install the new pointers */
175
176 i = fdt->max_fdset / (sizeof(unsigned long) * 8);
177 count = (nfdt->max_fdset - fdt->max_fdset) / 8;
178
179 /*
180 * Don't copy the entire array if the current fdset is
181 * not yet initialised.
182 */
183 if (i) {
184 memcpy (nfdt->open_fds, fdt->open_fds,
185 fdt->max_fdset/8);
186 memcpy (nfdt->close_on_exec, fdt->close_on_exec,
187 fdt->max_fdset/8);
188 memset (&nfdt->open_fds->fds_bits[i], 0, count);
189 memset (&nfdt->close_on_exec->fds_bits[i], 0, count);
190 }
191
192 /* Don't copy/clear the array if we are creating a new
193 fd array for fork() */
194 if (fdt->max_fds) {
195 memcpy(nfdt->fd, fdt->fd,
196 fdt->max_fds * sizeof(struct file *));
197 /* clear the remainder of the array */
198 memset(&nfdt->fd[fdt->max_fds], 0,
199 (nfdt->max_fds - fdt->max_fds) *
200 sizeof(struct file *));
201 }
202 nfdt->next_fd = fdt->next_fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
205/*
206 * Allocate an fdset array, using kmalloc or vmalloc.
207 * Note: the array isn't cleared at allocation time.
208 */
209fd_set * alloc_fdset(int num)
210{
211 fd_set *new_fdset;
212 int size = num / 8;
213
214 if (size <= PAGE_SIZE)
215 new_fdset = (fd_set *) kmalloc(size, GFP_KERNEL);
216 else
217 new_fdset = (fd_set *) vmalloc(size);
218 return new_fdset;
219}
220
221void free_fdset(fd_set *array, int num)
222{
223 int size = num / 8;
224
225 if (num <= __FD_SETSIZE) /* Don't free an embedded fdset */
226 return;
227 else if (size <= PAGE_SIZE)
228 kfree(array);
229 else
230 vfree(array);
231}
232
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700233static struct fdtable *alloc_fdtable(int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234{
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700235 struct fdtable *fdt = NULL;
236 int nfds = 0;
237 fd_set *new_openset = NULL, *new_execset = NULL;
238 struct file **new_fds;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700240 fdt = kmalloc(sizeof(*fdt), GFP_KERNEL);
241 if (!fdt)
242 goto out;
243 memset(fdt, 0, sizeof(*fdt));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700245 nfds = __FD_SETSIZE;
246 /* Expand to the max in easy steps */
247 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 if (nfds < (PAGE_SIZE * 8))
249 nfds = PAGE_SIZE * 8;
250 else {
251 nfds = nfds * 2;
252 if (nfds > NR_OPEN)
253 nfds = NR_OPEN;
254 }
255 } while (nfds <= nr);
256
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700257 new_openset = alloc_fdset(nfds);
258 new_execset = alloc_fdset(nfds);
259 if (!new_openset || !new_execset)
260 goto out;
261 fdt->open_fds = new_openset;
262 fdt->close_on_exec = new_execset;
263 fdt->max_fdset = nfds;
264
265 nfds = NR_OPEN_DEFAULT;
266 /*
267 * Expand to the max in easy steps, and keep expanding it until
268 * we have enough for the requested fd array size.
269 */
270 do {
271#if NR_OPEN_DEFAULT < 256
272 if (nfds < 256)
273 nfds = 256;
274 else
275#endif
276 if (nfds < (PAGE_SIZE / sizeof(struct file *)))
277 nfds = PAGE_SIZE / sizeof(struct file *);
278 else {
279 nfds = nfds * 2;
280 if (nfds > NR_OPEN)
281 nfds = NR_OPEN;
282 }
283 } while (nfds <= nr);
284 new_fds = alloc_fd_array(nfds);
285 if (!new_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 goto out;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700287 fdt->fd = new_fds;
288 fdt->max_fds = nfds;
289 fdt->free_files = NULL;
290 return fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700292 if (new_openset)
293 free_fdset(new_openset, nfds);
294 if (new_execset)
295 free_fdset(new_execset, nfds);
296 kfree(fdt);
297 return NULL;
298}
299
300/*
301 * Expands the file descriptor table - it will allocate a new fdtable and
302 * both fd array and fdset. It is expected to be called with the
303 * files_lock held.
304 */
305static int expand_fdtable(struct files_struct *files, int nr)
306 __releases(files->file_lock)
307 __acquires(files->file_lock)
308{
309 int error = 0;
310 struct fdtable *fdt;
311 struct fdtable *nfdt = NULL;
312
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 spin_unlock(&files->file_lock);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700314 nfdt = alloc_fdtable(nr);
315 if (!nfdt) {
316 error = -ENOMEM;
317 spin_lock(&files->file_lock);
318 goto out;
319 }
320
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 spin_lock(&files->file_lock);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700322 fdt = files_fdtable(files);
323 /*
324 * Check again since another task may have expanded the
325 * fd table while we dropped the lock
326 */
327 if (nr >= fdt->max_fds || nr >= fdt->max_fdset) {
328 copy_fdtable(nfdt, fdt);
329 } else {
330 /* Somebody expanded while we dropped file_lock */
331 spin_unlock(&files->file_lock);
332 __free_fdtable(nfdt);
333 spin_lock(&files->file_lock);
334 goto out;
335 }
336 rcu_assign_pointer(files->fdt, nfdt);
337 free_fdtable(fdt);
338out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 return error;
340}
341
342/*
343 * Expand files.
344 * Return <0 on error; 0 nothing done; 1 files expanded, we may have blocked.
345 * Should be called with the files->file_lock spinlock held for write.
346 */
347int expand_files(struct files_struct *files, int nr)
348{
349 int err, expand = 0;
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700350 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700352 fdt = files_fdtable(files);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700353 if (nr >= fdt->max_fdset || nr >= fdt->max_fds) {
354 if (fdt->max_fdset >= NR_OPEN ||
355 fdt->max_fds >= NR_OPEN || nr >= NR_OPEN) {
356 err = -EMFILE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 goto out;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 expand = 1;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700360 if ((err = expand_fdtable(files, nr)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 goto out;
362 }
363 err = expand;
364out:
365 return err;
366}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700367
368static void __devinit fdtable_defer_list_init(int cpu)
369{
370 struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
371 spin_lock_init(&fddef->lock);
372 INIT_WORK(&fddef->wq, (void (*)(void *))free_fdtable_work, fddef);
373 init_timer(&fddef->timer);
374 fddef->timer.data = (unsigned long)fddef;
375 fddef->timer.function = fdtable_timer;
376 fddef->next = NULL;
377}
378
379void __init files_defer_init(void)
380{
381 int i;
Eric Dumazet88a2a4ac2006-02-04 23:27:36 -0800382 for_each_cpu(i)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700383 fdtable_defer_list_init(i);
384}