blob: 4c6f0ea12c419c52154558c271586b7887dabf13 [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>
Al Viro9f3acc32008-04-24 07:44:08 -040015#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/bitops.h>
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070017#include <linux/interrupt.h>
18#include <linux/spinlock.h>
19#include <linux/rcupdate.h>
20#include <linux/workqueue.h>
21
22struct fdtable_defer {
23 spinlock_t lock;
24 struct work_struct wq;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070025 struct fdtable *next;
26};
27
Eric Dumazet9cfe0152008-02-06 01:37:16 -080028int sysctl_nr_open __read_mostly = 1024*1024;
29
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070030/*
31 * We use this list to defer free fdtables that have vmalloced
32 * sets/arrays. By keeping a per-cpu list, we avoid having to embed
33 * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
34 * this per-task structure.
35 */
36static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Vadim Lobanov5466b452006-12-10 02:21:22 -080038static inline void * alloc_fdmem(unsigned int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -070039{
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 if (size <= PAGE_SIZE)
Vadim Lobanov5466b452006-12-10 02:21:22 -080041 return kmalloc(size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 else
Vadim Lobanov5466b452006-12-10 02:21:22 -080043 return vmalloc(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -070044}
45
Vadim Lobanov5466b452006-12-10 02:21:22 -080046static inline void free_fdarr(struct fdtable *fdt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070047{
Vadim Lobanov5466b452006-12-10 02:21:22 -080048 if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
49 kfree(fdt->fd);
50 else
51 vfree(fdt->fd);
52}
53
54static inline void free_fdset(struct fdtable *fdt)
55{
56 if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
57 kfree(fdt->open_fds);
58 else
59 vfree(fdt->open_fds);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070060}
61
David Howells65f27f32006-11-22 14:55:48 +000062static void free_fdtable_work(struct work_struct *work)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070063{
David Howells65f27f32006-11-22 14:55:48 +000064 struct fdtable_defer *f =
65 container_of(work, struct fdtable_defer, wq);
Dipankar Sarmabadf1662005-09-09 13:04:10 -070066 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070068 spin_lock_bh(&f->lock);
69 fdt = f->next;
70 f->next = NULL;
71 spin_unlock_bh(&f->lock);
72 while(fdt) {
73 struct fdtable *next = fdt->next;
Vadim Lobanov5466b452006-12-10 02:21:22 -080074 vfree(fdt->fd);
75 free_fdset(fdt);
76 kfree(fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070077 fdt = next;
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 }
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070079}
80
Vadim Lobanov4fd45812006-12-10 02:21:17 -080081void free_fdtable_rcu(struct rcu_head *rcu)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070082{
83 struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070084 struct fdtable_defer *fddef;
85
86 BUG_ON(!fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070087
Vadim Lobanov4fd45812006-12-10 02:21:17 -080088 if (fdt->max_fds <= NR_OPEN_DEFAULT) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070089 /*
Vadim Lobanov4fd45812006-12-10 02:21:17 -080090 * This fdtable is embedded in the files structure and that
91 * structure itself is getting destroyed.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070092 */
Vadim Lobanov4fd45812006-12-10 02:21:17 -080093 kmem_cache_free(files_cachep,
94 container_of(fdt, struct files_struct, fdtab));
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070095 return;
96 }
Vadim Lobanov5466b452006-12-10 02:21:22 -080097 if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -070098 kfree(fdt->fd);
Vadim Lobanov5466b452006-12-10 02:21:22 -080099 kfree(fdt->open_fds);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700100 kfree(fdt);
101 } else {
102 fddef = &get_cpu_var(fdtable_defer_list);
103 spin_lock(&fddef->lock);
104 fdt->next = fddef->next;
105 fddef->next = fdt;
Tejun Heo593be072006-12-06 20:36:01 -0800106 /* vmallocs are handled from the workqueue context */
107 schedule_work(&fddef->wq);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700108 spin_unlock(&fddef->lock);
109 put_cpu_var(fdtable_defer_list);
110 }
111}
112
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700113/*
114 * Expand the fdset in the files_struct. Called with the files spinlock
115 * held for write.
116 */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800117static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700118{
Vadim Lobanov5466b452006-12-10 02:21:22 -0800119 unsigned int cpy, set;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700120
Vadim Lobanov5466b452006-12-10 02:21:22 -0800121 BUG_ON(nfdt->max_fds < ofdt->max_fds);
122 if (ofdt->max_fds == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123 return;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800124
125 cpy = ofdt->max_fds * sizeof(struct file *);
126 set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
127 memcpy(nfdt->fd, ofdt->fd, cpy);
128 memset((char *)(nfdt->fd) + cpy, 0, set);
129
130 cpy = ofdt->max_fds / BITS_PER_BYTE;
131 set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
132 memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
133 memset((char *)(nfdt->open_fds) + cpy, 0, set);
134 memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
135 memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136}
137
Vadim Lobanov5466b452006-12-10 02:21:22 -0800138static struct fdtable * alloc_fdtable(unsigned int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
Vadim Lobanov5466b452006-12-10 02:21:22 -0800140 struct fdtable *fdt;
141 char *data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700143 /*
Vadim Lobanov5466b452006-12-10 02:21:22 -0800144 * Figure out how many fds we actually want to support in this fdtable.
145 * Allocation steps are keyed to the size of the fdarray, since it
146 * grows far faster than any of the other dynamic data. We try to fit
147 * the fdarray into comfortable page-tuned chunks: starting at 1024B
148 * and growing in powers of two from there on.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700149 */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800150 nr /= (1024 / sizeof(struct file *));
151 nr = roundup_pow_of_two(nr + 1);
152 nr *= (1024 / sizeof(struct file *));
Al Viro5c598b32008-04-27 20:04:15 -0400153 /*
154 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
155 * had been set lower between the check in expand_files() and here. Deal
156 * with that in caller, it's cheaper that way.
157 *
158 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
159 * bitmaps handling below becomes unpleasant, to put it mildly...
160 */
161 if (unlikely(nr > sysctl_nr_open))
162 nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800163
Vadim Lobanov5466b452006-12-10 02:21:22 -0800164 fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
165 if (!fdt)
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800166 goto out;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800167 fdt->max_fds = nr;
168 data = alloc_fdmem(nr * sizeof(struct file *));
169 if (!data)
170 goto out_fdt;
171 fdt->fd = (struct file **)data;
172 data = alloc_fdmem(max_t(unsigned int,
173 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
174 if (!data)
175 goto out_arr;
176 fdt->open_fds = (fd_set *)data;
177 data += nr / BITS_PER_BYTE;
178 fdt->close_on_exec = (fd_set *)data;
179 INIT_RCU_HEAD(&fdt->rcu);
180 fdt->next = NULL;
181
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700182 return fdt;
Vadim Lobanov5466b452006-12-10 02:21:22 -0800183
184out_arr:
185 free_fdarr(fdt);
186out_fdt:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700187 kfree(fdt);
Vadim Lobanov5466b452006-12-10 02:21:22 -0800188out:
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700189 return NULL;
190}
191
192/*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700193 * Expand the file descriptor table.
194 * This function will allocate a new fdtable and both fd array and fdset, of
195 * the given size.
196 * Return <0 error code on error; 1 on successful completion.
197 * The files->file_lock should be held on entry, and will be held on exit.
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700198 */
199static int expand_fdtable(struct files_struct *files, int nr)
200 __releases(files->file_lock)
201 __acquires(files->file_lock)
202{
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700203 struct fdtable *new_fdt, *cur_fdt;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700204
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 spin_unlock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700206 new_fdt = alloc_fdtable(nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 spin_lock(&files->file_lock);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700208 if (!new_fdt)
209 return -ENOMEM;
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700210 /*
Al Viro5c598b32008-04-27 20:04:15 -0400211 * extremely unlikely race - sysctl_nr_open decreased between the check in
212 * caller and alloc_fdtable(). Cheaper to catch it here...
213 */
214 if (unlikely(new_fdt->max_fds <= nr)) {
215 free_fdarr(new_fdt);
216 free_fdset(new_fdt);
217 kfree(new_fdt);
218 return -EMFILE;
219 }
220 /*
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700221 * Check again since another task may have expanded the fd table while
222 * we dropped the lock
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700223 */
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700224 cur_fdt = files_fdtable(files);
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800225 if (nr >= cur_fdt->max_fds) {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700226 /* Continue as planned */
227 copy_fdtable(new_fdt, cur_fdt);
228 rcu_assign_pointer(files->fdt, new_fdt);
Vadim Lobanov4fd45812006-12-10 02:21:17 -0800229 if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
Vadim Lobanov01b2d932006-12-22 01:10:43 -0800230 free_fdtable(cur_fdt);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700231 } else {
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700232 /* Somebody else expanded, so undo our attempt */
Vadim Lobanov5466b452006-12-10 02:21:22 -0800233 free_fdarr(new_fdt);
234 free_fdset(new_fdt);
235 kfree(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);
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700253 /* Do we need to expand? */
Vadim Lobanovbbea9f62006-12-10 02:21:12 -0800254 if (nr < fdt->max_fds)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700255 return 0;
256 /* Can we expand? */
Eric Dumazet9cfe0152008-02-06 01:37:16 -0800257 if (nr >= sysctl_nr_open)
Vadim Lobanov74d392a2006-09-29 02:01:43 -0700258 return -EMFILE;
259
260 /* All good, so we try */
261 return expand_fdtable(files, nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262}
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700263
264static void __devinit fdtable_defer_list_init(int cpu)
265{
266 struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
267 spin_lock_init(&fddef->lock);
David Howells65f27f32006-11-22 14:55:48 +0000268 INIT_WORK(&fddef->wq, free_fdtable_work);
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700269 fddef->next = NULL;
270}
271
272void __init files_defer_init(void)
273{
274 int i;
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800275 for_each_possible_cpu(i)
Dipankar Sarmaab2af1f2005-09-09 13:04:13 -0700276 fdtable_defer_list_init(i);
277}