file.c 10.9 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 *  linux/fs/file.c
 *
 *  Copyright (C) 1998-1999, Stephen Tweedie and Bill Hawes
 *
 *  Manage the dynamic fd arrays in the process files_struct.
 */

#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/time.h>
#include <linux/slab.h>
#include <linux/vmalloc.h>
#include <linux/file.h>
A
Al Viro 已提交
15
#include <linux/fdtable.h>
L
Linus Torvalds 已提交
16
#include <linux/bitops.h>
17 18 19 20 21 22 23 24 25 26 27
#include <linux/interrupt.h>
#include <linux/spinlock.h>
#include <linux/rcupdate.h>
#include <linux/workqueue.h>

struct fdtable_defer {
	spinlock_t lock;
	struct work_struct wq;
	struct fdtable *next;
};

28
int sysctl_nr_open __read_mostly = 1024*1024;
29 30
int sysctl_nr_open_min = BITS_PER_LONG;
int sysctl_nr_open_max = 1024 * 1024; /* raised later */
31

32 33 34 35 36 37 38
/*
 * We use this list to defer free fdtables that have vmalloced
 * sets/arrays. By keeping a per-cpu list, we avoid having to embed
 * the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
 * this per-task structure.
 */
static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
L
Linus Torvalds 已提交
39

40
static inline void * alloc_fdmem(unsigned int size)
L
Linus Torvalds 已提交
41 42
{
	if (size <= PAGE_SIZE)
43 44 45
		return kmalloc(size, GFP_KERNEL);
	else
		return vmalloc(size);
L
Linus Torvalds 已提交
46 47
}

48
static inline void free_fdarr(struct fdtable *fdt)
L
Linus Torvalds 已提交
49
{
50 51
	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
		kfree(fdt->fd);
L
Linus Torvalds 已提交
52
	else
53
		vfree(fdt->fd);
L
Linus Torvalds 已提交
54 55
}

56
static inline void free_fdset(struct fdtable *fdt)
L
Linus Torvalds 已提交
57
{
58 59 60 61
	if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
		kfree(fdt->open_fds);
	else
		vfree(fdt->open_fds);
62
}
L
Linus Torvalds 已提交
63

64
static void free_fdtable_work(struct work_struct *work)
65
{
66 67
	struct fdtable_defer *f =
		container_of(work, struct fdtable_defer, wq);
68
	struct fdtable *fdt;
L
Linus Torvalds 已提交
69

70 71 72 73 74 75
	spin_lock_bh(&f->lock);
	fdt = f->next;
	f->next = NULL;
	spin_unlock_bh(&f->lock);
	while(fdt) {
		struct fdtable *next = fdt->next;
76 77 78
		vfree(fdt->fd);
		free_fdset(fdt);
		kfree(fdt);
79 80 81
		fdt = next;
	}
}
L
Linus Torvalds 已提交
82

83
void free_fdtable_rcu(struct rcu_head *rcu)
84 85 86
{
	struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
	struct fdtable_defer *fddef;
L
Linus Torvalds 已提交
87

88 89
	BUG_ON(!fdt);

90
	if (fdt->max_fds <= NR_OPEN_DEFAULT) {
91
		/*
92 93
		 * This fdtable is embedded in the files structure and that
		 * structure itself is getting destroyed.
94
		 */
95 96
		kmem_cache_free(files_cachep,
				container_of(fdt, struct files_struct, fdtab));
97 98
		return;
	}
99
	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
100
		kfree(fdt->fd);
101
		kfree(fdt->open_fds);
102
		kfree(fdt);
L
Linus Torvalds 已提交
103
	} else {
104 105 106 107
		fddef = &get_cpu_var(fdtable_defer_list);
		spin_lock(&fddef->lock);
		fdt->next = fddef->next;
		fddef->next = fdt;
108 109
		/* vmallocs are handled from the workqueue context */
		schedule_work(&fddef->wq);
110 111
		spin_unlock(&fddef->lock);
		put_cpu_var(fdtable_defer_list);
L
Linus Torvalds 已提交
112
	}
113 114 115 116 117 118
}

/*
 * Expand the fdset in the files_struct.  Called with the files spinlock
 * held for write.
 */
119
static void copy_fdtable(struct fdtable *nfdt, struct fdtable *ofdt)
120
{
121
	unsigned int cpy, set;
122

123 124 125 126 127 128 129 130 131 132 133 134 135
	BUG_ON(nfdt->max_fds < ofdt->max_fds);

	cpy = ofdt->max_fds * sizeof(struct file *);
	set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
	memcpy(nfdt->fd, ofdt->fd, cpy);
	memset((char *)(nfdt->fd) + cpy, 0, set);

	cpy = ofdt->max_fds / BITS_PER_BYTE;
	set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
	memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
	memset((char *)(nfdt->open_fds) + cpy, 0, set);
	memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
	memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
L
Linus Torvalds 已提交
136 137
}

138
static struct fdtable * alloc_fdtable(unsigned int nr)
L
Linus Torvalds 已提交
139
{
140 141
	struct fdtable *fdt;
	char *data;
L
Linus Torvalds 已提交
142

143
	/*
144 145 146 147 148
	 * Figure out how many fds we actually want to support in this fdtable.
	 * Allocation steps are keyed to the size of the fdarray, since it
	 * grows far faster than any of the other dynamic data. We try to fit
	 * the fdarray into comfortable page-tuned chunks: starting at 1024B
	 * and growing in powers of two from there on.
149
	 */
150 151 152
	nr /= (1024 / sizeof(struct file *));
	nr = roundup_pow_of_two(nr + 1);
	nr *= (1024 / sizeof(struct file *));
A
Al Viro 已提交
153 154 155 156 157 158 159 160 161 162
	/*
	 * Note that this can drive nr *below* what we had passed if sysctl_nr_open
	 * had been set lower between the check in expand_files() and here.  Deal
	 * with that in caller, it's cheaper that way.
	 *
	 * We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
	 * bitmaps handling below becomes unpleasant, to put it mildly...
	 */
	if (unlikely(nr > sysctl_nr_open))
		nr = ((sysctl_nr_open - 1) | (BITS_PER_LONG - 1)) + 1;
163

164 165
	fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
	if (!fdt)
166
		goto out;
167 168 169 170 171 172 173 174 175 176 177 178 179 180 181
	fdt->max_fds = nr;
	data = alloc_fdmem(nr * sizeof(struct file *));
	if (!data)
		goto out_fdt;
	fdt->fd = (struct file **)data;
	data = alloc_fdmem(max_t(unsigned int,
				 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
	if (!data)
		goto out_arr;
	fdt->open_fds = (fd_set *)data;
	data += nr / BITS_PER_BYTE;
	fdt->close_on_exec = (fd_set *)data;
	INIT_RCU_HEAD(&fdt->rcu);
	fdt->next = NULL;

182
	return fdt;
183 184 185 186

out_arr:
	free_fdarr(fdt);
out_fdt:
187
	kfree(fdt);
188
out:
189 190
	return NULL;
}
L
Linus Torvalds 已提交
191

192
/*
193 194 195 196 197
 * Expand the file descriptor table.
 * This function will allocate a new fdtable and both fd array and fdset, of
 * the given size.
 * Return <0 error code on error; 1 on successful completion.
 * The files->file_lock should be held on entry, and will be held on exit.
198 199 200 201 202
 */
static int expand_fdtable(struct files_struct *files, int nr)
	__releases(files->file_lock)
	__acquires(files->file_lock)
{
203
	struct fdtable *new_fdt, *cur_fdt;
204 205

	spin_unlock(&files->file_lock);
206
	new_fdt = alloc_fdtable(nr);
207
	spin_lock(&files->file_lock);
208 209
	if (!new_fdt)
		return -ENOMEM;
A
Al Viro 已提交
210 211 212 213 214 215 216 217 218 219
	/*
	 * extremely unlikely race - sysctl_nr_open decreased between the check in
	 * caller and alloc_fdtable().  Cheaper to catch it here...
	 */
	if (unlikely(new_fdt->max_fds <= nr)) {
		free_fdarr(new_fdt);
		free_fdset(new_fdt);
		kfree(new_fdt);
		return -EMFILE;
	}
220
	/*
221 222
	 * Check again since another task may have expanded the fd table while
	 * we dropped the lock
223
	 */
224
	cur_fdt = files_fdtable(files);
225
	if (nr >= cur_fdt->max_fds) {
226 227 228
		/* Continue as planned */
		copy_fdtable(new_fdt, cur_fdt);
		rcu_assign_pointer(files->fdt, new_fdt);
229
		if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
230
			free_fdtable(cur_fdt);
231
	} else {
232
		/* Somebody else expanded, so undo our attempt */
233 234 235
		free_fdarr(new_fdt);
		free_fdset(new_fdt);
		kfree(new_fdt);
236
	}
237
	return 1;
L
Linus Torvalds 已提交
238 239 240 241
}

/*
 * Expand files.
242 243 244 245 246
 * This function will expand the file structures, if the requested size exceeds
 * the current capacity and there is room for expansion.
 * Return <0 error code on error; 0 when nothing done; 1 when files were
 * expanded and execution may have blocked.
 * The files->file_lock should be held on entry, and will be held on exit.
L
Linus Torvalds 已提交
247 248 249
 */
int expand_files(struct files_struct *files, int nr)
{
250
	struct fdtable *fdt;
L
Linus Torvalds 已提交
251

252
	fdt = files_fdtable(files);
A
Al Viro 已提交
253 254 255 256 257 258 259 260

	/*
	 * N.B. For clone tasks sharing a files structure, this test
	 * will limit the total number of files that can be opened.
	 */
	if (nr >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
		return -EMFILE;

261
	/* Do we need to expand? */
262
	if (nr < fdt->max_fds)
263
		return 0;
A
Al Viro 已提交
264

265
	/* Can we expand? */
266
	if (nr >= sysctl_nr_open)
267 268 269 270
		return -EMFILE;

	/* All good, so we try */
	return expand_fdtable(files, nr);
L
Linus Torvalds 已提交
271
}
272

A
Al Viro 已提交
273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299
static int count_open_files(struct fdtable *fdt)
{
	int size = fdt->max_fds;
	int i;

	/* Find the last open fd */
	for (i = size/(8*sizeof(long)); i > 0; ) {
		if (fdt->open_fds->fds_bits[--i])
			break;
	}
	i = (i+1) * 8 * sizeof(long);
	return i;
}

/*
 * Allocate a new files structure and copy contents from the
 * passed in files structure.
 * errorp will be valid only when the returned files_struct is NULL.
 */
struct files_struct *dup_fd(struct files_struct *oldf, int *errorp)
{
	struct files_struct *newf;
	struct file **old_fds, **new_fds;
	int open_files, size, i;
	struct fdtable *old_fdt, *new_fdt;

	*errorp = -ENOMEM;
A
Al Viro 已提交
300
	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
A
Al Viro 已提交
301 302 303
	if (!newf)
		goto out;

A
Al Viro 已提交
304 305 306 307 308 309 310 311 312 313 314 315
	atomic_set(&newf->count, 1);

	spin_lock_init(&newf->file_lock);
	newf->next_fd = 0;
	new_fdt = &newf->fdtab;
	new_fdt->max_fds = NR_OPEN_DEFAULT;
	new_fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
	new_fdt->open_fds = (fd_set *)&newf->open_fds_init;
	new_fdt->fd = &newf->fd_array[0];
	INIT_RCU_HEAD(&new_fdt->rcu);
	new_fdt->next = NULL;

A
Al Viro 已提交
316 317 318 319 320 321 322
	spin_lock(&oldf->file_lock);
	old_fdt = files_fdtable(oldf);
	open_files = count_open_files(old_fdt);

	/*
	 * Check whether we need to allocate a larger fd array and fd set.
	 */
A
Al Viro 已提交
323
	while (unlikely(open_files > new_fdt->max_fds)) {
A
Al Viro 已提交
324
		spin_unlock(&oldf->file_lock);
A
Al Viro 已提交
325

A
Al Viro 已提交
326 327 328 329 330 331
		if (new_fdt != &newf->fdtab) {
			free_fdarr(new_fdt);
			free_fdset(new_fdt);
			kfree(new_fdt);
		}

A
Al Viro 已提交
332 333 334 335 336 337 338 339 340 341 342 343
		new_fdt = alloc_fdtable(open_files - 1);
		if (!new_fdt) {
			*errorp = -ENOMEM;
			goto out_release;
		}

		/* beyond sysctl_nr_open; nothing to do */
		if (unlikely(new_fdt->max_fds < open_files)) {
			free_fdarr(new_fdt);
			free_fdset(new_fdt);
			kfree(new_fdt);
			*errorp = -EMFILE;
A
Al Viro 已提交
344
			goto out_release;
A
Al Viro 已提交
345 346
		}

A
Al Viro 已提交
347 348 349 350 351 352 353
		/*
		 * Reacquire the oldf lock and a pointer to its fd table
		 * who knows it may have a new bigger fd table. We need
		 * the latest pointer.
		 */
		spin_lock(&oldf->file_lock);
		old_fdt = files_fdtable(oldf);
A
Al Viro 已提交
354
		open_files = count_open_files(old_fdt);
A
Al Viro 已提交
355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395
	}

	old_fds = old_fdt->fd;
	new_fds = new_fdt->fd;

	memcpy(new_fdt->open_fds->fds_bits,
		old_fdt->open_fds->fds_bits, open_files/8);
	memcpy(new_fdt->close_on_exec->fds_bits,
		old_fdt->close_on_exec->fds_bits, open_files/8);

	for (i = open_files; i != 0; i--) {
		struct file *f = *old_fds++;
		if (f) {
			get_file(f);
		} else {
			/*
			 * The fd may be claimed in the fd bitmap but not yet
			 * instantiated in the files array if a sibling thread
			 * is partway through open().  So make sure that this
			 * fd is available to the new process.
			 */
			FD_CLR(open_files - i, new_fdt->open_fds);
		}
		rcu_assign_pointer(*new_fds++, f);
	}
	spin_unlock(&oldf->file_lock);

	/* compute the remainder to be cleared */
	size = (new_fdt->max_fds - open_files) * sizeof(struct file *);

	/* This is long word aligned thus could use a optimized version */
	memset(new_fds, 0, size);

	if (new_fdt->max_fds > open_files) {
		int left = (new_fdt->max_fds-open_files)/8;
		int start = open_files / (8 * sizeof(unsigned long));

		memset(&new_fdt->open_fds->fds_bits[start], 0, left);
		memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
	}

A
Al Viro 已提交
396 397
	rcu_assign_pointer(newf->fdt, new_fdt);

A
Al Viro 已提交
398 399 400 401 402 403 404 405
	return newf;

out_release:
	kmem_cache_free(files_cachep, newf);
out:
	return NULL;
}

406 407 408 409
static void __devinit fdtable_defer_list_init(int cpu)
{
	struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
	spin_lock_init(&fddef->lock);
410
	INIT_WORK(&fddef->wq, free_fdtable_work);
411 412 413 414 415 416
	fddef->next = NULL;
}

void __init files_defer_init(void)
{
	int i;
417
	for_each_possible_cpu(i)
418
		fdtable_defer_list_init(i);
419 420
	sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
			     -BITS_PER_LONG;
421
}
A
Al Viro 已提交
422 423 424 425 426 427 428 429 430 431 432 433 434

struct files_struct init_files = {
	.count		= ATOMIC_INIT(1),
	.fdt		= &init_files.fdtab,
	.fdtab		= {
		.max_fds	= NR_OPEN_DEFAULT,
		.fd		= &init_files.fd_array[0],
		.close_on_exec	= (fd_set *)&init_files.close_on_exec_init,
		.open_fds	= (fd_set *)&init_files.open_fds_init,
		.rcu		= RCU_HEAD_INIT,
	},
	.file_lock	= __SPIN_LOCK_UNLOCKED(init_task.file_lock),
};