blob: c5575de01113eb24aa182c77b35d46d0f0fced48 [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;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070024 struct fdtable *next;
25};
26
27/*
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.
32 */
33static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
Vadim Lobanov5466b452006-12-10 02:21:22 -080035static inline void * alloc_fdmem(unsigned int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -070036{
Linus Torvalds1da177e2005-04-16 15:20:36 -070037 if (size <= PAGE_SIZE)
Vadim Lobanov5466b452006-12-10 02:21:22 -080038 return kmalloc(size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 else
Vadim Lobanov5466b452006-12-10 02:21:22 -080040 return vmalloc(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041}
42
Vadim Lobanov5466b452006-12-10 02:21:22 -080043static inline void free_fdarr(struct fdtable *fdt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070044{
Vadim Lobanov5466b452006-12-10 02:21:22 -080045 if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
46 kfree(fdt->fd);
47 else
48 vfree(fdt->fd);
49}
50
51static inline void free_fdset(struct fdtable *fdt)
52{
53 if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
54 kfree(fdt->open_fds);
55 else
56 vfree(fdt->open_fds);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070057}
58
David Howells65f27f32006-11-22 14:55:48 +000059static void free_fdtable_work(struct work_struct *work)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070060{
David Howells65f27f32006-11-22 14:55:48 +000061 struct fdtable_defer *f =
62 container_of(work, struct fdtable_defer, wq);
Dipankar Sarmabadf1662005-09-09 13:04:10 -070063 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070065 spin_lock_bh(&f->lock);
66 fdt = f->next;
67 f->next = NULL;
68 spin_unlock_bh(&f->lock);
69 while(fdt) {
70 struct fdtable *next = fdt->next;
Vadim Lobanov5466b452006-12-10 02:21:22 -080071 vfree(fdt->fd);
72 free_fdset(fdt);
73 kfree(fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070074 fdt = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070075 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070076}
77
Vadim Lobanov4fd45812006-12-10 02:21:17 -080078void free_fdtable_rcu(struct rcu_head *rcu)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070079{
80 struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070081 struct fdtable_defer *fddef;
82
83 BUG_ON(!fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070084
Vadim Lobanov4fd45812006-12-10 02:21:17 -080085 if (fdt->max_fds <= NR_OPEN_DEFAULT) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070086 /*
Vadim Lobanov4fd45812006-12-10 02:21:17 -080087 * This fdtable is embedded in the files structure and that
88 * structure itself is getting destroyed.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070089 */
Vadim Lobanov4fd45812006-12-10 02:21:17 -080090 kmem_cache_free(files_cachep,
91 container_of(fdt, struct files_struct, fdtab));
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070092 return;
93 }
Vadim Lobanov5466b452006-12-10 02:21:22 -080094 if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070095 kfree(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -080096 kfree(fdt->open_fds);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070097 kfree(fdt);
98 } else {
99 fddef = &get_cpu_var(fdtable_defer_list);
100 spin_lock(&fddef->lock);
101 fdt->next = fddef->next;
102 fddef->next = fdt;
Tejun Heo593be072006-12-06 20:36:01 -0800103 /* vmallocs are handled from the workqueue context */
104 schedule_work(&fddef->wq);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700105 spin_unlock(&fddef->lock);
106 put_cpu_var(fdtable_defer_list);
107 }
108}
109
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700110/*
111 * Expand the fdset in the files_struct. Called with the files spinlock
112 * held for write.
113 */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800114static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700115{
Vadim Lobanov5466b452006-12-10 02:21:22 -0800116 unsigned int cpy, set;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700117
Vadim Lobanov5466b452006-12-10 02:21:22 -0800118 BUG_ON(nfdt->max_fds < ofdt->max_fds);
119 if (ofdt->max_fds == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 return;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800121
122 cpy = ofdt->max_fds * sizeof(struct file *);
123 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
124 memcpy(nfdt->fd, ofdt->fd, cpy);
125 memset((char *)(nfdt->fd) + cpy, 0, set);
126
127 cpy = ofdt->max_fds / BITS_PER_BYTE;
128 set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
129 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
130 memset((char *)(nfdt->open_fds) + cpy, 0, set);
131 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
132 memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
Vadim Lobanov5466b452006-12-10 02:21:22 -0800135static struct fdtable * alloc_fdtable(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
Vadim Lobanov5466b452006-12-10 02:21:22 -0800137 struct fdtable *fdt;
138 char *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700140 /*
Vadim Lobanov5466b452006-12-10 02:21:22 -0800141 * Figure out how many fds we actually want to support in this fdtable.
142 * Allocation steps are keyed to the size of the fdarray, since it
143 * grows far faster than any of the other dynamic data. We try to fit
144 * the fdarray into comfortable page-tuned chunks: starting at 1024B
145 * and growing in powers of two from there on.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700146 */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800147 nr /= (1024 / sizeof(struct file *));
148 nr = roundup_pow_of_two(nr + 1);
149 nr *= (1024 / sizeof(struct file *));
150 if (nr > NR_OPEN)
151 nr = NR_OPEN;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800152
Vadim Lobanov5466b452006-12-10 02:21:22 -0800153 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
154 if (!fdt)
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800155 goto out;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800156 fdt->max_fds = nr;
157 data = alloc_fdmem(nr * sizeof(struct file *));
158 if (!data)
159 goto out_fdt;
160 fdt->fd = (struct file **)data;
161 data = alloc_fdmem(max_t(unsigned int,
162 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
163 if (!data)
164 goto out_arr;
165 fdt->open_fds = (fd_set *)data;
166 data += nr / BITS_PER_BYTE;
167 fdt->close_on_exec = (fd_set *)data;
168 INIT_RCU_HEAD(&fdt->rcu);
169 fdt->next = NULL;
170
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700171 return fdt;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800172
173out_arr:
174 free_fdarr(fdt);
175out_fdt:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700176 kfree(fdt);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800177out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700178 return NULL;
179}
180
181/*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700182 * Expand the file descriptor table.
183 * This function will allocate a new fdtable and both fd array and fdset, of
184 * the given size.
185 * Return <0 error code on error; 1 on successful completion.
186 * The files->file_lock should be held on entry, and will be held on exit.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700187 */
188static int expand_fdtable(struct files_struct *files, int nr)
189 __releases(files->file_lock)
190 __acquires(files->file_lock)
191{
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700192 struct fdtable *new_fdt, *cur_fdt;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 spin_unlock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700195 new_fdt = alloc_fdtable(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700197 if (!new_fdt)
198 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700199 /*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700200 * Check again since another task may have expanded the fd table while
201 * we dropped the lock
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700202 */
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700203 cur_fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800204 if (nr >= cur_fdt->max_fds) {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700205 /* Continue as planned */
206 copy_fdtable(new_fdt, cur_fdt);
207 rcu_assign_pointer(files->fdt, new_fdt);
Vadim Lobanov4fd45812006-12-10 02:21:17 -0800208 if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
Vadim Lobanov01b2d932006-12-22 01:10:43 -0800209 free_fdtable(cur_fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700210 } else {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700211 /* Somebody else expanded, so undo our attempt */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800212 free_fdarr(new_fdt);
213 free_fdset(new_fdt);
214 kfree(new_fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700215 }
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700216 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
219/*
220 * Expand files.
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700221 * This function will expand the file structures, if the requested size exceeds
222 * the current capacity and there is room for expansion.
223 * Return <0 error code on error; 0 when nothing done; 1 when files were
224 * expanded and execution may have blocked.
225 * The files->file_lock should be held on entry, and will be held on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 */
227int expand_files(struct files_struct *files, int nr)
228{
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700229 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
Dipankar Sarmabadf1662005-09-09 13:04:10 -0700231 fdt = files_fdtable(files);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700232 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800233 if (nr < fdt->max_fds)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700234 return 0;
235 /* Can we expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800236 if (nr >= NR_OPEN)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700237 return -EMFILE;
238
239 /* All good, so we try */
240 return expand_fdtable(files, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700242
243static void __devinit fdtable_defer_list_init(int cpu)
244{
245 struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
246 spin_lock_init(&fddef->lock);
David Howells65f27f32006-11-22 14:55:48 +0000247 INIT_WORK(&fddef->wq, free_fdtable_work);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700248 fddef->next = NULL;
249}
250
251void __init files_defer_init(void)
252{
253 int i;
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800254 for_each_possible_cpu(i)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700255 fdtable_defer_list_init(i);
256}