[linux-2.6-block.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;
int sysctl_nr_open_min = BITS_PER_LONG;
int sysctl_nr_open_max = 1024 * 1024; /* raised later */

/*
 * 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);

	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;
}

/*
 * 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 = 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;
	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;

	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.
	 */
	while (unlikely(open_files > new_fdt->max_fds)) {
		spin_unlock(&oldf->file_lock);

		if (new_fdt != &newf->fdtab) {
			free_fdarr(new_fdt);
			free_fdset(new_fdt);
			kfree(new_fdt);
		}

		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;
			goto out_release;
		}

		/*
		 * 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);
		open_files = count_open_files(old_fdt);
	}

	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);
	}

	rcu_assign_pointer(newf->fdt, new_fdt);

	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);
	sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
			     -BITS_PER_LONG;
}

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	28	int sysctl_nr_open __read_mostly = 1024*1024;
eceea0b3 AV	29	int sysctl_nr_open_min = BITS_PER_LONG;
eceea0b3 AV	30	int sysctl_nr_open_max = 1024 * 1024; /* raised later */
9cfe015a	31
ab2af1f5 DS	32	/*
	33	* We use this list to defer free fdtables that have vmalloced
	34	* sets/arrays. By keeping a per-cpu list, we avoid having to embed
	35	* the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
	36	* this per-task structure.
	37	*/
	38	static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
1da177e4	39
5466b456	40	static inline void * alloc_fdmem(unsigned int size)
1da177e4	41	{
1da177e4	42	if (size <= PAGE_SIZE)
5466b456 VL	43	return kmalloc(size, GFP_KERNEL);
	44	else
	45	return vmalloc(size);
1da177e4 LT	46	}
1da177e4 LT	47
5466b456	48	static inline void free_fdarr(struct fdtable *fdt)
1da177e4	49	{
5466b456 VL	50	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
5466b456 VL	51	kfree(fdt->fd);
1da177e4	52	else
5466b456	53	vfree(fdt->fd);
1da177e4 LT	54	}
1da177e4 LT	55
5466b456	56	static inline void free_fdset(struct fdtable *fdt)
1da177e4	57	{
5466b456 VL	58	if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
	59	kfree(fdt->open_fds);
	60	else
	61	vfree(fdt->open_fds);
ab2af1f5	62	}
1da177e4	63
65f27f38	64	static void free_fdtable_work(struct work_struct *work)
ab2af1f5	65	{
65f27f38 DH	66	struct fdtable_defer *f =
65f27f38 DH	67	container_of(work, struct fdtable_defer, wq);
ab2af1f5	68	struct fdtable *fdt;
1da177e4	69
ab2af1f5 DS	70	spin_lock_bh(&f->lock);
	71	fdt = f->next;
	72	f->next = NULL;
	73	spin_unlock_bh(&f->lock);
	74	while(fdt) {
	75	struct fdtable *next = fdt->next;
5466b456 VL	76	vfree(fdt->fd);
	77	free_fdset(fdt);
	78	kfree(fdt);
ab2af1f5 DS	79	fdt = next;
	80	}
	81	}
1da177e4	82
4fd45812	83	void free_fdtable_rcu(struct rcu_head *rcu)
ab2af1f5 DS	84	{
ab2af1f5 DS	85	struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
ab2af1f5	86	struct fdtable_defer *fddef;
1da177e4	87
ab2af1f5	88	BUG_ON(!fdt);
ab2af1f5	89
4fd45812	90	if (fdt->max_fds <= NR_OPEN_DEFAULT) {
ab2af1f5	91	/*
4fd45812 VL	92	* This fdtable is embedded in the files structure and that
4fd45812 VL	93	* structure itself is getting destroyed.
ab2af1f5	94	*/
4fd45812 VL	95	kmem_cache_free(files_cachep,
4fd45812 VL	96	container_of(fdt, struct files_struct, fdtab));
ab2af1f5 DS	97	return;
ab2af1f5 DS	98	}
5466b456	99	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
ab2af1f5	100	kfree(fdt->fd);
5466b456	101	kfree(fdt->open_fds);
ab2af1f5	102	kfree(fdt);
1da177e4	103	} else {
ab2af1f5 DS	104	fddef = &get_cpu_var(fdtable_defer_list);
	105	spin_lock(&fddef->lock);
	106	fdt->next = fddef->next;
	107	fddef->next = fdt;
593be07a TH	108	/* vmallocs are handled from the workqueue context */
593be07a TH	109	schedule_work(&fddef->wq);
ab2af1f5 DS	110	spin_unlock(&fddef->lock);
ab2af1f5 DS	111	put_cpu_var(fdtable_defer_list);
1da177e4	112	}
ab2af1f5 DS	113	}
ab2af1f5 DS	114
ab2af1f5 DS	115	/*
	116	* Expand the fdset in the files_struct. Called with the files spinlock
	117	* held for write.
	118	*/
5466b456	119	static void copy_fdtable(struct fdtable nfdt, struct fdtable ofdt)
ab2af1f5	120	{
5466b456	121	unsigned int cpy, set;
ab2af1f5	122
5466b456	123	BUG_ON(nfdt->max_fds < ofdt->max_fds);
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
02afc626 AV	278	/*
	279	* Allocate a new files structure and copy contents from the
	280	* passed in files structure.
	281	* errorp will be valid only when the returned files_struct is NULL.
	282	*/
	283	struct files_struct dup_fd(struct files_struct oldf, int *errorp)
	284	{
	285	struct files_struct *newf;
	286	struct file old_fds, new_fds;
	287	int open_files, size, i;
	288	struct fdtable old_fdt, new_fdt;
	289
	290	*errorp = -ENOMEM;
afbec7ff	291	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
02afc626 AV	292	if (!newf)
	293	goto out;
	294
afbec7ff AV	295	atomic_set(&newf->count, 1);
	296
	297	spin_lock_init(&newf->file_lock);
	298	newf->next_fd = 0;
	299	new_fdt = &newf->fdtab;
	300	new_fdt->max_fds = NR_OPEN_DEFAULT;
	301	new_fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
	302	new_fdt->open_fds = (fd_set *)&newf->open_fds_init;
	303	new_fdt->fd = &newf->fd_array[0];
	304	INIT_RCU_HEAD(&new_fdt->rcu);
	305	new_fdt->next = NULL;
	306
02afc626 AV	307	spin_lock(&oldf->file_lock);
02afc626 AV	308	old_fdt = files_fdtable(oldf);
02afc626 AV	309	open_files = count_open_files(old_fdt);
	310
	311	/*
	312	* Check whether we need to allocate a larger fd array and fd set.
02afc626	313	*/
adbecb12	314	while (unlikely(open_files > new_fdt->max_fds)) {
02afc626	315	spin_unlock(&oldf->file_lock);
9dec3c4d	316
adbecb12 AV	317	if (new_fdt != &newf->fdtab) {
	318	free_fdarr(new_fdt);
	319	free_fdset(new_fdt);
	320	kfree(new_fdt);
	321	}
	322
9dec3c4d AV	323	new_fdt = alloc_fdtable(open_files - 1);
	324	if (!new_fdt) {
	325	*errorp = -ENOMEM;
	326	goto out_release;
	327	}
	328
	329	/* beyond sysctl_nr_open; nothing to do */
	330	if (unlikely(new_fdt->max_fds < open_files)) {
	331	free_fdarr(new_fdt);
	332	free_fdset(new_fdt);
	333	kfree(new_fdt);
	334	*errorp = -EMFILE;
02afc626	335	goto out_release;
9dec3c4d	336	}
9dec3c4d	337
02afc626 AV	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);
adbecb12	345	open_files = count_open_files(old_fdt);
02afc626 AV	346	}
	347
	348	old_fds = old_fdt->fd;
	349	new_fds = new_fdt->fd;
	350
	351	memcpy(new_fdt->open_fds->fds_bits,
	352	old_fdt->open_fds->fds_bits, open_files/8);
	353	memcpy(new_fdt->close_on_exec->fds_bits,
	354	old_fdt->close_on_exec->fds_bits, open_files/8);
	355
	356	for (i = open_files; i != 0; i--) {
	357	struct file f = old_fds++;
	358	if (f) {
	359	get_file(f);
	360	} else {
	361	/*
	362	* The fd may be claimed in the fd bitmap but not yet
	363	* instantiated in the files array if a sibling thread
	364	* is partway through open(). So make sure that this
	365	* fd is available to the new process.
	366	*/
	367	FD_CLR(open_files - i, new_fdt->open_fds);
	368	}
	369	rcu_assign_pointer(*new_fds++, f);
	370	}
	371	spin_unlock(&oldf->file_lock);
	372
	373	/* compute the remainder to be cleared */
	374	size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
	375
	376	/* This is long word aligned thus could use a optimized version */
	377	memset(new_fds, 0, size);
	378
	379	if (new_fdt->max_fds > open_files) {
	380	int left = (new_fdt->max_fds-open_files)/8;
	381	int start = open_files / (8 * sizeof(unsigned long));
	382
	383	memset(&new_fdt->open_fds->fds_bits[start], 0, left);
	384	memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
	385	}
	386
afbec7ff AV	387	rcu_assign_pointer(newf->fdt, new_fdt);
afbec7ff AV	388
02afc626 AV	389	return newf;
	390
	391	out_release:
	392	kmem_cache_free(files_cachep, newf);
	393	out:
	394	return NULL;
	395	}
	396
ab2af1f5 DS	397	static void __devinit fdtable_defer_list_init(int cpu)
	398	{
	399	struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
	400	spin_lock_init(&fddef->lock);
65f27f38	401	INIT_WORK(&fddef->wq, free_fdtable_work);
ab2af1f5 DS	402	fddef->next = NULL;
	403	}
	404
	405	void __init files_defer_init(void)
	406	{
	407	int i;
0a945022	408	for_each_possible_cpu(i)
ab2af1f5	409	fdtable_defer_list_init(i);
eceea0b3 AV	410	sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
eceea0b3 AV	411	-BITS_PER_LONG;
ab2af1f5	412	}
f52111b1 AV	413
	414	struct files_struct init_files = {
	415	.count = ATOMIC_INIT(1),
	416	.fdt = &init_files.fdtab,
	417	.fdtab = {
	418	.max_fds = NR_OPEN_DEFAULT,
	419	.fd = &init_files.fd_array[0],
	420	.close_on_exec = (fd_set *)&init_files.close_on_exec_init,
	421	.open_fds = (fd_set *)&init_files.open_fds_init,
	422	.rcu = RCU_HEAD_INIT,
	423	},
	424	.file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock),
	425	};