[net-next-2.6.git] / fs / file.c

/*
 *  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>
#include <linux/fdtable.h>
#include <linux/bitops.h>
#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;
};

int sysctl_nr_open __read_mostly = 1024*1024;

/*
 * 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);

static inline void * alloc_fdmem(unsigned int size)
{
	if (size <= PAGE_SIZE)
		return kmalloc(size, GFP_KERNEL);
	else
		return vmalloc(size);
}

static inline void free_fdarr(struct fdtable *fdt)
{
	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
		kfree(fdt->fd);
	else
		vfree(fdt->fd);
}

static inline void free_fdset(struct fdtable *fdt)
{
	if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
		kfree(fdt->open_fds);
	else
		vfree(fdt->open_fds);
}

static void free_fdtable_work(struct work_struct *work)
{
	struct fdtable_defer *f =
		container_of(work, struct fdtable_defer, wq);
	struct fdtable *fdt;

	spin_lock_bh(&f->lock);
	fdt = f->next;
	f->next = NULL;
	spin_unlock_bh(&f->lock);
	while(fdt) {
		struct fdtable *next = fdt->next;
		vfree(fdt->fd);
		free_fdset(fdt);
		kfree(fdt);
		fdt = next;
	}
}

void free_fdtable_rcu(struct rcu_head *rcu)
{
	struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
	struct fdtable_defer *fddef;

	BUG_ON(!fdt);

	if (fdt->max_fds <= NR_OPEN_DEFAULT) {
		/*
		 * This fdtable is embedded in the files structure and that
		 * structure itself is getting destroyed.
		 */
		kmem_cache_free(files_cachep,
				container_of(fdt, struct files_struct, fdtab));
		return;
	}
	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
		kfree(fdt->fd);
		kfree(fdt->open_fds);
		kfree(fdt);
	} else {
		fddef = &get_cpu_var(fdtable_defer_list);
		spin_lock(&fddef->lock);
		fdt->next = fddef->next;
		fddef->next = fdt;
		/* vmallocs are handled from the workqueue context */
		schedule_work(&fddef->wq);
		spin_unlock(&fddef->lock);
		put_cpu_var(fdtable_defer_list);
	}
}

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

	BUG_ON(nfdt->max_fds < ofdt->max_fds);
	if (ofdt->max_fds == 0)
		return;

	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);
}

static struct fdtable * alloc_fdtable(unsigned int nr)
{
	struct fdtable *fdt;
	char *data;

	/*
	 * 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.
	 */
	nr /= (1024 / sizeof(struct file *));
	nr = roundup_pow_of_two(nr + 1);
	nr *= (1024 / sizeof(struct file *));
	/*
	 * 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;

	fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
	if (!fdt)
		goto out;
	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;

	return fdt;

out_arr:
	free_fdarr(fdt);
out_fdt:
	kfree(fdt);
out:
	return NULL;
}

/*
 * 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.
 */
static int expand_fdtable(struct files_struct *files, int nr)
	__releases(files->file_lock)
	__acquires(files->file_lock)
{
	struct fdtable *new_fdt, *cur_fdt;

	spin_unlock(&files->file_lock);
	new_fdt = alloc_fdtable(nr);
	spin_lock(&files->file_lock);
	if (!new_fdt)
		return -ENOMEM;
	/*
	 * 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;
	}
	/*
	 * Check again since another task may have expanded the fd table while
	 * we dropped the lock
	 */
	cur_fdt = files_fdtable(files);
	if (nr >= cur_fdt->max_fds) {
		/* Continue as planned */
		copy_fdtable(new_fdt, cur_fdt);
		rcu_assign_pointer(files->fdt, new_fdt);
		if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
			free_fdtable(cur_fdt);
	} else {
		/* Somebody else expanded, so undo our attempt */
		free_fdarr(new_fdt);
		free_fdset(new_fdt);
		kfree(new_fdt);
	}
	return 1;
}

/*
 * Expand files.
 * 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.
 */
int expand_files(struct files_struct *files, int nr)
{
	struct fdtable *fdt;

	fdt = files_fdtable(files);
	/* Do we need to expand? */
	if (nr < fdt->max_fds)
		return 0;
	/* Can we expand? */
	if (nr >= sysctl_nr_open)
		return -EMFILE;

	/* All good, so we try */
	return expand_fdtable(files, nr);
}

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;
}

static struct files_struct *alloc_files(void)
{
	struct files_struct *newf;
	struct fdtable *fdt;

	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
	if (!newf)
		goto out;

	atomic_set(&newf->count, 1);

	spin_lock_init(&newf->file_lock);
	newf->next_fd = 0;
	fdt = &newf->fdtab;
	fdt->max_fds = NR_OPEN_DEFAULT;
	fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
	fdt->open_fds = (fd_set *)&newf->open_fds_init;
	fdt->fd = &newf->fd_array[0];
	INIT_RCU_HEAD(&fdt->rcu);
	fdt->next = NULL;
	rcu_assign_pointer(newf->fdt, fdt);
out:
	return newf;
}

/*
 * 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;
	newf = alloc_files();
	if (!newf)
		goto out;

	spin_lock(&oldf->file_lock);
	old_fdt = files_fdtable(oldf);
	new_fdt = files_fdtable(newf);
	open_files = count_open_files(old_fdt);

	/*
	 * Check whether we need to allocate a larger fd array and fd set.
	 * Note: we're not a clone task, so the open count won't change.
	 */
	if (open_files > new_fdt->max_fds) {
		new_fdt->max_fds = 0;
		spin_unlock(&oldf->file_lock);
		spin_lock(&newf->file_lock);
		*errorp = expand_files(newf, open_files-1);
		spin_unlock(&newf->file_lock);
		if (*errorp < 0)
			goto out_release;
		new_fdt = files_fdtable(newf);
		/*
		 * 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);
	}

	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);
	}

	return newf;

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

static void __devinit fdtable_defer_list_init(int cpu)
{
	struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
	spin_lock_init(&fddef->lock);
	INIT_WORK(&fddef->wq, free_fdtable_work);
	fddef->next = NULL;
}

void __init files_defer_init(void)
{
	int i;
	for_each_possible_cpu(i)
		fdtable_defer_list_init(i);
}

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),
};
Commit	Line	Data
1da177e4 LT	1	/*
	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>
9f3acc31	15	#include <linux/fdtable.h>
1da177e4	16	#include <linux/bitops.h>
ab2af1f5 DS	17	#include <linux/interrupt.h>
	18	#include <linux/spinlock.h>
	19	#include <linux/rcupdate.h>
	20	#include <linux/workqueue.h>
	21
	22	struct fdtable_defer {
	23	spinlock_t lock;
	24	struct work_struct wq;
ab2af1f5 DS	25	struct fdtable *next;
	26	};
	27
9cfe015a ED	28	int sysctl_nr_open __read_mostly = 1024*1024;
9cfe015a ED	29
ab2af1f5 DS	30	/*
	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	*/
	36	static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
1da177e4	37
5466b456	38	static inline void * alloc_fdmem(unsigned int size)
1da177e4	39	{
1da177e4	40	if (size <= PAGE_SIZE)
5466b456 VL	41	return kmalloc(size, GFP_KERNEL);
	42	else
	43	return vmalloc(size);
1da177e4 LT	44	}
1da177e4 LT	45
5466b456	46	static inline void free_fdarr(struct fdtable *fdt)
1da177e4	47	{
5466b456 VL	48	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
5466b456 VL	49	kfree(fdt->fd);
1da177e4	50	else
5466b456	51	vfree(fdt->fd);
1da177e4 LT	52	}
1da177e4 LT	53
5466b456	54	static inline void free_fdset(struct fdtable *fdt)
1da177e4	55	{
5466b456 VL	56	if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
	57	kfree(fdt->open_fds);
	58	else
	59	vfree(fdt->open_fds);
ab2af1f5	60	}
1da177e4	61
65f27f38	62	static void free_fdtable_work(struct work_struct *work)
ab2af1f5	63	{
65f27f38 DH	64	struct fdtable_defer *f =
65f27f38 DH	65	container_of(work, struct fdtable_defer, wq);
ab2af1f5	66	struct fdtable *fdt;
1da177e4	67
ab2af1f5 DS	68	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;
5466b456 VL	74	vfree(fdt->fd);
	75	free_fdset(fdt);
	76	kfree(fdt);
ab2af1f5 DS	77	fdt = next;
	78	}
	79	}
1da177e4	80
4fd45812	81	void free_fdtable_rcu(struct rcu_head *rcu)
ab2af1f5 DS	82	{
ab2af1f5 DS	83	struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
ab2af1f5	84	struct fdtable_defer *fddef;
1da177e4	85
ab2af1f5	86	BUG_ON(!fdt);
ab2af1f5	87
4fd45812	88	if (fdt->max_fds <= NR_OPEN_DEFAULT) {
ab2af1f5	89	/*
4fd45812 VL	90	* This fdtable is embedded in the files structure and that
4fd45812 VL	91	* structure itself is getting destroyed.
ab2af1f5	92	*/
4fd45812 VL	93	kmem_cache_free(files_cachep,
4fd45812 VL	94	container_of(fdt, struct files_struct, fdtab));
ab2af1f5 DS	95	return;
ab2af1f5 DS	96	}
5466b456	97	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
ab2af1f5	98	kfree(fdt->fd);
5466b456	99	kfree(fdt->open_fds);
ab2af1f5	100	kfree(fdt);
1da177e4	101	} else {
ab2af1f5 DS	102	fddef = &get_cpu_var(fdtable_defer_list);
	103	spin_lock(&fddef->lock);
	104	fdt->next = fddef->next;
	105	fddef->next = fdt;
593be07a TH	106	/* vmallocs are handled from the workqueue context */
593be07a TH	107	schedule_work(&fddef->wq);
ab2af1f5 DS	108	spin_unlock(&fddef->lock);
ab2af1f5 DS	109	put_cpu_var(fdtable_defer_list);
1da177e4	110	}
ab2af1f5 DS	111	}
ab2af1f5 DS	112
ab2af1f5 DS	113	/*
	114	* Expand the fdset in the files_struct. Called with the files spinlock
	115	* held for write.
	116	*/
5466b456	117	static void copy_fdtable(struct fdtable nfdt, struct fdtable ofdt)
ab2af1f5	118	{
5466b456	119	unsigned int cpy, set;
ab2af1f5	120
5466b456 VL	121	BUG_ON(nfdt->max_fds < ofdt->max_fds);
5466b456 VL	122	if (ofdt->max_fds == 0)
1da177e4	123	return;
5466b456 VL	124
	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);
1da177e4 LT	136	}
1da177e4 LT	137
5466b456	138	static struct fdtable * alloc_fdtable(unsigned int nr)
1da177e4	139	{
5466b456 VL	140	struct fdtable *fdt;
5466b456 VL	141	char *data;
1da177e4	142
ab2af1f5	143	/*
5466b456 VL	144	* 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.
ab2af1f5	149	*/
5466b456 VL	150	nr /= (1024 / sizeof(struct file *));
	151	nr = roundup_pow_of_two(nr + 1);
	152	nr = (1024 / sizeof(struct file ));
5c598b34 AV	153	/*
	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;
bbea9f69	163
5466b456 VL	164	fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
5466b456 VL	165	if (!fdt)
bbea9f69	166	goto out;
5466b456 VL	167	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
ab2af1f5	182	return fdt;
5466b456 VL	183
	184	out_arr:
	185	free_fdarr(fdt);
	186	out_fdt:
ab2af1f5	187	kfree(fdt);
5466b456	188	out:
ab2af1f5 DS	189	return NULL;
ab2af1f5 DS	190	}
1da177e4	191
ab2af1f5	192	/*
74d392aa VL	193	* 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.
ab2af1f5 DS	198	*/
	199	static int expand_fdtable(struct files_struct *files, int nr)
	200	__releases(files->file_lock)
	201	__acquires(files->file_lock)
	202	{
74d392aa	203	struct fdtable new_fdt, cur_fdt;
ab2af1f5 DS	204
ab2af1f5 DS	205	spin_unlock(&files->file_lock);
74d392aa	206	new_fdt = alloc_fdtable(nr);
ab2af1f5	207	spin_lock(&files->file_lock);
74d392aa VL	208	if (!new_fdt)
74d392aa VL	209	return -ENOMEM;
5c598b34 AV	210	/*
	211	* 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	}
ab2af1f5	220	/*
74d392aa VL	221	* Check again since another task may have expanded the fd table while
74d392aa VL	222	* we dropped the lock
ab2af1f5	223	*/
74d392aa	224	cur_fdt = files_fdtable(files);
bbea9f69	225	if (nr >= cur_fdt->max_fds) {
74d392aa VL	226	/* Continue as planned */
	227	copy_fdtable(new_fdt, cur_fdt);
	228	rcu_assign_pointer(files->fdt, new_fdt);
4fd45812	229	if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
01b2d93c	230	free_fdtable(cur_fdt);
ab2af1f5	231	} else {
74d392aa	232	/* Somebody else expanded, so undo our attempt */
5466b456 VL	233	free_fdarr(new_fdt);
	234	free_fdset(new_fdt);
	235	kfree(new_fdt);
ab2af1f5	236	}
74d392aa	237	return 1;
1da177e4 LT	238	}
	239
	240	/*
	241	* Expand files.
74d392aa VL	242	* 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.
1da177e4 LT	247	*/
	248	int expand_files(struct files_struct *files, int nr)
	249	{
badf1662	250	struct fdtable *fdt;
1da177e4	251
badf1662	252	fdt = files_fdtable(files);
74d392aa	253	/* Do we need to expand? */
bbea9f69	254	if (nr < fdt->max_fds)
74d392aa VL	255	return 0;
74d392aa VL	256	/* Can we expand? */
9cfe015a	257	if (nr >= sysctl_nr_open)
74d392aa VL	258	return -EMFILE;
	259
	260	/* All good, so we try */
	261	return expand_fdtable(files, nr);
1da177e4	262	}
ab2af1f5	263
02afc626 AV	264	static int count_open_files(struct fdtable *fdt)
	265	{
	266	int size = fdt->max_fds;
	267	int i;
	268
	269	/* Find the last open fd */
	270	for (i = size/(8*sizeof(long)); i > 0; ) {
	271	if (fdt->open_fds->fds_bits[--i])
	272	break;
	273	}
	274	i = (i+1) * 8 * sizeof(long);
	275	return i;
	276	}
	277
	278	static struct files_struct *alloc_files(void)
	279	{
	280	struct files_struct *newf;
	281	struct fdtable *fdt;
	282
	283	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
	284	if (!newf)
	285	goto out;
	286
	287	atomic_set(&newf->count, 1);
	288
	289	spin_lock_init(&newf->file_lock);
	290	newf->next_fd = 0;
	291	fdt = &newf->fdtab;
	292	fdt->max_fds = NR_OPEN_DEFAULT;
	293	fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
	294	fdt->open_fds = (fd_set *)&newf->open_fds_init;
	295	fdt->fd = &newf->fd_array[0];
	296	INIT_RCU_HEAD(&fdt->rcu);
	297	fdt->next = NULL;
	298	rcu_assign_pointer(newf->fdt, fdt);
	299	out:
	300	return newf;
	301	}
	302
	303	/*
	304	* Allocate a new files structure and copy contents from the
	305	* passed in files structure.
	306	* errorp will be valid only when the returned files_struct is NULL.
	307	*/
	308	struct files_struct dup_fd(struct files_struct oldf, int *errorp)
	309	{
	310	struct files_struct *newf;
	311	struct file old_fds, new_fds;
	312	int open_files, size, i;
	313	struct fdtable old_fdt, new_fdt;
	314
	315	*errorp = -ENOMEM;
	316	newf = alloc_files();
	317	if (!newf)
	318	goto out;
	319
	320	spin_lock(&oldf->file_lock);
	321	old_fdt = files_fdtable(oldf);
	322	new_fdt = files_fdtable(newf);
	323	open_files = count_open_files(old_fdt);
	324
	325	/*
	326	* Check whether we need to allocate a larger fd array and fd set.
	327	* Note: we're not a clone task, so the open count won't change.
328	*/
329	if (open_files > new_fdt->max_fds) {
330	new_fdt->max_fds = 0;
331	spin_unlock(&oldf->file_lock);
332	spin_lock(&newf->file_lock);
333	*errorp = expand_files(newf, open_files-1);
334	spin_unlock(&newf->file_lock);
335	if (*errorp < 0)
336	goto out_release;
337	new_fdt = files_fdtable(newf);
338	/*
339	* Reacquire the oldf lock and a pointer to its fd table
340	* who knows it may have a new bigger fd table. We need
341	* the latest pointer.
342	*/
343	spin_lock(&oldf->file_lock);
344	old_fdt = files_fdtable(oldf);
345	}
346
347	old_fds = old_fdt->fd;
348	new_fds = new_fdt->fd;
349
350	memcpy(new_fdt->open_fds->fds_bits,
351	old_fdt->open_fds->fds_bits, open_files/8);
352	memcpy(new_fdt->close_on_exec->fds_bits,
353	old_fdt->close_on_exec->fds_bits, open_files/8);
354
355	for (i = open_files; i != 0; i--) {
356	struct file f = old_fds++;
357	if (f) {
358	get_file(f);
359	} else {
360	/*
361	* The fd may be claimed in the fd bitmap but not yet
362	* instantiated in the files array if a sibling thread
363	* is partway through open(). So make sure that this
364	* fd is available to the new process.
365	*/
366	FD_CLR(open_files - i, new_fdt->open_fds);
367	}
368	rcu_assign_pointer(*new_fds++, f);
369	}
370	spin_unlock(&oldf->file_lock);
371
372	/* compute the remainder to be cleared */
373	size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
374
375	/* This is long word aligned thus could use a optimized version */
376	memset(new_fds, 0, size);
377
378	if (new_fdt->max_fds > open_files) {
379	int left = (new_fdt->max_fds-open_files)/8;
380	int start = open_files / (8 * sizeof(unsigned long));
381
382	memset(&new_fdt->open_fds->fds_bits[start], 0, left);
383	memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
384	}
385
386	return newf;
387
388	out_release:
389	kmem_cache_free(files_cachep, newf);
390	out:
391	return NULL;
392	}
393
ab2af1f5 DS	394	static void __devinit fdtable_defer_list_init(int cpu)
	395	{
	396	struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
	397	spin_lock_init(&fddef->lock);
65f27f38	398	INIT_WORK(&fddef->wq, free_fdtable_work);
ab2af1f5 DS	399	fddef->next = NULL;
	400	}
	401
	402	void __init files_defer_init(void)
	403	{
	404	int i;
0a945022	405	for_each_possible_cpu(i)
ab2af1f5 DS	406	fdtable_defer_list_init(i);
ab2af1f5 DS	407	}
f52111b1 AV	408
	409	struct files_struct init_files = {
	410	.count = ATOMIC_INIT(1),
	411	.fdt = &init_files.fdtab,
	412	.fdtab = {
	413	.max_fds = NR_OPEN_DEFAULT,
	414	.fd = &init_files.fd_array[0],
	415	.close_on_exec = (fd_set *)&init_files.close_on_exec_init,
	416	.open_fds = (fd_set *)&init_files.open_fds_init,
	417	.rcu = RCU_HEAD_INIT,
	418	},
	419	.file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock),
	420	};