4 * Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes
6 * Manage the dynamic fd arrays in the process files_struct.
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>
16 #include <linux/interrupt.h>
17 #include <linux/spinlock.h>
18 #include <linux/rcupdate.h>
19 #include <linux/workqueue.h>
21 struct fdtable_defer {
23 struct work_struct wq;
28 * We use this list to defer free fdtables that have vmalloced
29 * sets/arrays. By keeping a per-cpu list, we avoid having to embed
30 * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
31 * this per-task structure.
33 static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
37 * Allocate an fd array, using kmalloc or vmalloc.
38 * Note: the array isn't cleared at allocation time.
40 struct file ** alloc_fd_array(int num)
42 struct file **new_fds;
43 int size = num * sizeof(struct file *);
45 if (size <= PAGE_SIZE)
46 new_fds = (struct file **) kmalloc(size, GFP_KERNEL);
48 new_fds = (struct file **) vmalloc(size);
52 void free_fd_array(struct file **array, int num)
54 int size = num * sizeof(struct file *);
57 printk (KERN_ERR "free_fd_array: array = 0 (num = %d)\n", num);
61 if (num <= NR_OPEN_DEFAULT) /* Don't free the embedded fd array! */
63 else if (size <= PAGE_SIZE)
69 static void __free_fdtable(struct fdtable *fdt)
71 free_fdset(fdt->open_fds, fdt->max_fds);
72 free_fdset(fdt->close_on_exec, fdt->max_fds);
73 free_fd_array(fdt->fd, fdt->max_fds);
77 static void free_fdtable_work(struct work_struct *work)
79 struct fdtable_defer *f =
80 container_of(work, struct fdtable_defer, wq);
83 spin_lock_bh(&f->lock);
86 spin_unlock_bh(&f->lock);
88 struct fdtable *next = fdt->next;
94 static void free_fdtable_rcu(struct rcu_head *rcu)
96 struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
97 int fdset_size, fdarray_size;
98 struct fdtable_defer *fddef;
101 fdset_size = fdt->max_fds / 8;
102 fdarray_size = fdt->max_fds * sizeof(struct file *);
104 if (fdt->free_files) {
106 * The this fdtable was embedded in the files structure
107 * and the files structure itself was getting destroyed.
108 * It is now safe to free the files structure.
110 kmem_cache_free(files_cachep, fdt->free_files);
113 if (fdt->max_fds <= NR_OPEN_DEFAULT)
115 * The fdtable was embedded
118 if (fdset_size <= PAGE_SIZE && fdarray_size <= PAGE_SIZE) {
119 kfree(fdt->open_fds);
120 kfree(fdt->close_on_exec);
124 fddef = &get_cpu_var(fdtable_defer_list);
125 spin_lock(&fddef->lock);
126 fdt->next = fddef->next;
128 /* vmallocs are handled from the workqueue context */
129 schedule_work(&fddef->wq);
130 spin_unlock(&fddef->lock);
131 put_cpu_var(fdtable_defer_list);
135 void free_fdtable(struct fdtable *fdt)
137 if (fdt->free_files || fdt->max_fds > NR_OPEN_DEFAULT)
138 call_rcu(&fdt->rcu, free_fdtable_rcu);
142 * Expand the fdset in the files_struct. Called with the files spinlock
145 static void copy_fdtable(struct fdtable *nfdt, struct fdtable *fdt)
150 BUG_ON(nfdt->max_fds < fdt->max_fds);
151 /* Copy the existing tables and install the new pointers */
153 i = fdt->max_fds / (sizeof(unsigned long) * 8);
154 count = (nfdt->max_fds - fdt->max_fds) / 8;
157 * Don't copy the entire array if the current fdset is
158 * not yet initialised.
161 memcpy (nfdt->open_fds, fdt->open_fds,
163 memcpy (nfdt->close_on_exec, fdt->close_on_exec,
165 memset (&nfdt->open_fds->fds_bits[i], 0, count);
166 memset (&nfdt->close_on_exec->fds_bits[i], 0, count);
169 /* Don't copy/clear the array if we are creating a new
170 fd array for fork() */
172 memcpy(nfdt->fd, fdt->fd,
173 fdt->max_fds * sizeof(struct file *));
174 /* clear the remainder of the array */
175 memset(&nfdt->fd[fdt->max_fds], 0,
176 (nfdt->max_fds - fdt->max_fds) *
177 sizeof(struct file *));
182 * Allocate an fdset array, using kmalloc or vmalloc.
183 * Note: the array isn't cleared at allocation time.
185 fd_set * alloc_fdset(int num)
190 if (size <= PAGE_SIZE)
191 new_fdset = (fd_set *) kmalloc(size, GFP_KERNEL);
193 new_fdset = (fd_set *) vmalloc(size);
197 void free_fdset(fd_set *array, int num)
199 if (num <= NR_OPEN_DEFAULT) /* Don't free an embedded fdset */
201 else if (num <= 8 * PAGE_SIZE)
207 static struct fdtable *alloc_fdtable(int nr)
209 struct fdtable *fdt = NULL;
211 fd_set *new_openset = NULL, *new_execset = NULL;
212 struct file **new_fds;
214 fdt = kzalloc(sizeof(*fdt), GFP_KERNEL);
218 nfds = NR_OPEN_DEFAULT;
220 * Expand to the max in easy steps, and keep expanding it until
221 * we have enough for the requested fd array size.
224 #if NR_OPEN_DEFAULT < 256
229 if (nfds < (PAGE_SIZE / sizeof(struct file *)))
230 nfds = PAGE_SIZE / sizeof(struct file *);
236 } while (nfds <= nr);
238 new_openset = alloc_fdset(nfds);
239 new_execset = alloc_fdset(nfds);
240 if (!new_openset || !new_execset)
242 fdt->open_fds = new_openset;
243 fdt->close_on_exec = new_execset;
245 new_fds = alloc_fd_array(nfds);
250 fdt->free_files = NULL;
253 free_fdset(new_openset, nfds);
254 free_fdset(new_execset, nfds);
260 * Expand the file descriptor table.
261 * This function will allocate a new fdtable and both fd array and fdset, of
263 * Return <0 error code on error; 1 on successful completion.
264 * The files->file_lock should be held on entry, and will be held on exit.
266 static int expand_fdtable(struct files_struct *files, int nr)
267 __releases(files->file_lock)
268 __acquires(files->file_lock)
270 struct fdtable *new_fdt, *cur_fdt;
272 spin_unlock(&files->file_lock);
273 new_fdt = alloc_fdtable(nr);
274 spin_lock(&files->file_lock);
278 * Check again since another task may have expanded the fd table while
279 * we dropped the lock
281 cur_fdt = files_fdtable(files);
282 if (nr >= cur_fdt->max_fds) {
283 /* Continue as planned */
284 copy_fdtable(new_fdt, cur_fdt);
285 rcu_assign_pointer(files->fdt, new_fdt);
286 free_fdtable(cur_fdt);
288 /* Somebody else expanded, so undo our attempt */
289 __free_fdtable(new_fdt);
296 * This function will expand the file structures, if the requested size exceeds
297 * the current capacity and there is room for expansion.
298 * Return <0 error code on error; 0 when nothing done; 1 when files were
299 * expanded and execution may have blocked.
300 * The files->file_lock should be held on entry, and will be held on exit.
302 int expand_files(struct files_struct *files, int nr)
306 fdt = files_fdtable(files);
307 /* Do we need to expand? */
308 if (nr < fdt->max_fds)
314 /* All good, so we try */
315 return expand_fdtable(files, nr);
318 static void __devinit fdtable_defer_list_init(int cpu)
320 struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
321 spin_lock_init(&fddef->lock);
322 INIT_WORK(&fddef->wq, free_fdtable_work);
326 void __init files_defer_init(void)
329 for_each_possible_cpu(i)
330 fdtable_defer_list_init(i);