[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/module.h>
#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/time.h>
#include <linux/sched.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)
{
	void *data;

	data = kmalloc(size, GFP_KERNEL|__GFP_NOWARN);
	if (data != NULL)
		return data;

	return vmalloc(size);
}

static void free_fdmem(void *ptr)
{
	is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr);
}

static void __free_fdtable(struct fdtable *fdt)
{
	free_fdmem(fdt->fd);
	free_fdmem(fdt->open_fds);
	kfree(fdt);
}

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;

		__free_fdtable(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 (!is_vmalloc_addr(fdt->fd) && !is_vmalloc_addr(fdt->open_fds)) {
		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;
	fdt->next = NULL;

	return fdt;

out_arr:
	free_fdmem(fdt->fd);
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_fdtable(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_fdtable(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);

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

	/* 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];
	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_fdtable(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_fdtable(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,
	},
	.file_lock	= __SPIN_LOCK_UNLOCKED(init_task.file_lock),
};

/*
 * allocate a file descriptor, mark it busy.
 */
int alloc_fd(unsigned start, unsigned flags)
{
	struct files_struct *files = current->files;
	unsigned int fd;
	int error;
	struct fdtable *fdt;

	spin_lock(&files->file_lock);
repeat:
	fdt = files_fdtable(files);
	fd = start;
	if (fd < files->next_fd)
		fd = files->next_fd;

	if (fd < fdt->max_fds)
		fd = find_next_zero_bit(fdt->open_fds->fds_bits,
					   fdt->max_fds, fd);

	error = expand_files(files, fd);
	if (error < 0)
		goto out;

	/*
	 * If we needed to expand the fs array we
	 * might have blocked - try again.
	 */
	if (error)
		goto repeat;

	if (start <= files->next_fd)
		files->next_fd = fd + 1;

	FD_SET(fd, fdt->open_fds);
	if (flags & O_CLOEXEC)
		FD_SET(fd, fdt->close_on_exec);
	else
		FD_CLR(fd, fdt->close_on_exec);
	error = fd;
#if 1
	/* Sanity check */
	if (rcu_dereference_raw(fdt->fd[fd]) != NULL) {
		printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
		rcu_assign_pointer(fdt->fd[fd], NULL);
	}
#endif

out:
	spin_unlock(&files->file_lock);
	return error;
}

int get_unused_fd(void)
{
	return alloc_fd(0, 0);
}
EXPORT_SYMBOL(get_unused_fd);
Commit	Line	Data
	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/module.h>
	10	#include <linux/fs.h>
	11	#include <linux/mm.h>
	12	#include <linux/time.h>
	13	#include <linux/sched.h>
	14	#include <linux/slab.h>
	15	#include <linux/vmalloc.h>
	16	#include <linux/file.h>
	17	#include <linux/fdtable.h>
	18	#include <linux/bitops.h>
	19	#include <linux/interrupt.h>
	20	#include <linux/spinlock.h>
	21	#include <linux/rcupdate.h>
	22	#include <linux/workqueue.h>
	23
	24	struct fdtable_defer {
	25	spinlock_t lock;
	26	struct work_struct wq;
	27	struct fdtable *next;
	28	};
	29
	30	int sysctl_nr_open __read_mostly = 1024*1024;
	31	int sysctl_nr_open_min = BITS_PER_LONG;
	32	int sysctl_nr_open_max = 1024 * 1024; /* raised later */
	33
	34	/*
	35	* We use this list to defer free fdtables that have vmalloced
	36	* sets/arrays. By keeping a per-cpu list, we avoid having to embed
	37	* the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
	38	* this per-task structure.
	39	*/
	40	static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
	41
	42	static inline void *alloc_fdmem(unsigned int size)
	43	{
	44	void *data;
	45
	46	data = kmalloc(size, GFP_KERNEL\|__GFP_NOWARN);
	47	if (data != NULL)
	48	return data;
	49
	50	return vmalloc(size);
	51	}
	52
	53	static void free_fdmem(void *ptr)
	54	{
	55	is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr);
	56	}
	57
	58	static void __free_fdtable(struct fdtable *fdt)
	59	{
	60	free_fdmem(fdt->fd);
	61	free_fdmem(fdt->open_fds);
	62	kfree(fdt);
	63	}
	64
	65	static void free_fdtable_work(struct work_struct *work)
	66	{
	67	struct fdtable_defer *f =
	68	container_of(work, struct fdtable_defer, wq);
	69	struct fdtable *fdt;
	70
	71	spin_lock_bh(&f->lock);
	72	fdt = f->next;
	73	f->next = NULL;
	74	spin_unlock_bh(&f->lock);
	75	while(fdt) {
	76	struct fdtable *next = fdt->next;
	77
	78	__free_fdtable(fdt);
	79	fdt = next;
	80	}
	81	}
	82
	83	void free_fdtable_rcu(struct rcu_head *rcu)
	84	{
	85	struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
	86	struct fdtable_defer *fddef;
	87
	88	BUG_ON(!fdt);
	89
	90	if (fdt->max_fds <= NR_OPEN_DEFAULT) {
	91	/*
	92	* This fdtable is embedded in the files structure and that
	93	* structure itself is getting destroyed.
	94	*/
	95	kmem_cache_free(files_cachep,
	96	container_of(fdt, struct files_struct, fdtab));
	97	return;
	98	}
	99	if (!is_vmalloc_addr(fdt->fd) && !is_vmalloc_addr(fdt->open_fds)) {
	100	kfree(fdt->fd);
	101	kfree(fdt->open_fds);
	102	kfree(fdt);
	103	} else {
	104	fddef = &get_cpu_var(fdtable_defer_list);
	105	spin_lock(&fddef->lock);
	106	fdt->next = fddef->next;
	107	fddef->next = fdt;
	108	/* vmallocs are handled from the workqueue context */
	109	schedule_work(&fddef->wq);
	110	spin_unlock(&fddef->lock);
	111	put_cpu_var(fdtable_defer_list);
	112	}
	113	}
	114
	115	/*
	116	* Expand the fdset in the files_struct. Called with the files spinlock
	117	* held for write.
	118	*/
	119	static void copy_fdtable(struct fdtable nfdt, struct fdtable ofdt)
	120	{
	121	unsigned int cpy, set;
	122
	123	BUG_ON(nfdt->max_fds < ofdt->max_fds);
	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);
	136	}
	137
	138	static struct fdtable * alloc_fdtable(unsigned int nr)
	139	{
	140	struct fdtable *fdt;
	141	char *data;
	142
	143	/*
	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.
	149	*/
	150	nr /= (1024 / sizeof(struct file *));
	151	nr = roundup_pow_of_two(nr + 1);
	152	nr = (1024 / sizeof(struct file ));
	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;
	163
	164	fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
	165	if (!fdt)
	166	goto out;
	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	fdt->next = NULL;
	180
	181	return fdt;
	182
	183	out_arr:
	184	free_fdmem(fdt->fd);
	185	out_fdt:
	186	kfree(fdt);
	187	out:
	188	return NULL;
	189	}
	190
	191	/*
	192	* Expand the file descriptor table.
	193	* This function will allocate a new fdtable and both fd array and fdset, of
	194	* the given size.
	195	* Return <0 error code on error; 1 on successful completion.
	196	* The files->file_lock should be held on entry, and will be held on exit.
	197	*/
	198	static int expand_fdtable(struct files_struct *files, int nr)
	199	__releases(files->file_lock)
	200	__acquires(files->file_lock)
	201	{
	202	struct fdtable new_fdt, cur_fdt;
	203
	204	spin_unlock(&files->file_lock);
	205	new_fdt = alloc_fdtable(nr);
	206	spin_lock(&files->file_lock);
	207	if (!new_fdt)
	208	return -ENOMEM;
	209	/*
	210	* extremely unlikely race - sysctl_nr_open decreased between the check in
	211	* caller and alloc_fdtable(). Cheaper to catch it here...
	212	*/
	213	if (unlikely(new_fdt->max_fds <= nr)) {
	214	__free_fdtable(new_fdt);
	215	return -EMFILE;
	216	}
	217	/*
	218	* Check again since another task may have expanded the fd table while
	219	* we dropped the lock
	220	*/
	221	cur_fdt = files_fdtable(files);
	222	if (nr >= cur_fdt->max_fds) {
	223	/* Continue as planned */
	224	copy_fdtable(new_fdt, cur_fdt);
	225	rcu_assign_pointer(files->fdt, new_fdt);
	226	if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
	227	free_fdtable(cur_fdt);
	228	} else {
	229	/* Somebody else expanded, so undo our attempt */
	230	__free_fdtable(new_fdt);
	231	}
	232	return 1;
	233	}
	234
	235	/*
	236	* Expand files.
	237	* This function will expand the file structures, if the requested size exceeds
	238	* the current capacity and there is room for expansion.
	239	* Return <0 error code on error; 0 when nothing done; 1 when files were
	240	* expanded and execution may have blocked.
	241	* The files->file_lock should be held on entry, and will be held on exit.
	242	*/
	243	int expand_files(struct files_struct *files, int nr)
	244	{
	245	struct fdtable *fdt;
	246
	247	fdt = files_fdtable(files);
	248
	249	/*
	250	* N.B. For clone tasks sharing a files structure, this test
	251	* will limit the total number of files that can be opened.
	252	*/
	253	if (nr >= rlimit(RLIMIT_NOFILE))
	254	return -EMFILE;
	255
	256	/* Do we need to expand? */
	257	if (nr < fdt->max_fds)
	258	return 0;
	259
	260	/* Can we expand? */
	261	if (nr >= sysctl_nr_open)
	262	return -EMFILE;
	263
	264	/* All good, so we try */
	265	return expand_fdtable(files, nr);
	266	}
	267
	268	static int count_open_files(struct fdtable *fdt)
	269	{
	270	int size = fdt->max_fds;
	271	int i;
	272
	273	/* Find the last open fd */
	274	for (i = size/(8*sizeof(long)); i > 0; ) {
	275	if (fdt->open_fds->fds_bits[--i])
	276	break;
	277	}
	278	i = (i+1) * 8 * sizeof(long);
	279	return i;
	280	}
	281
	282	/*
	283	* Allocate a new files structure and copy contents from the
	284	* passed in files structure.
	285	* errorp will be valid only when the returned files_struct is NULL.
	286	*/
	287	struct files_struct dup_fd(struct files_struct oldf, int *errorp)
	288	{
	289	struct files_struct *newf;
	290	struct file old_fds, new_fds;
	291	int open_files, size, i;
	292	struct fdtable old_fdt, new_fdt;
	293
	294	*errorp = -ENOMEM;
	295	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
	296	if (!newf)
	297	goto out;
	298
	299	atomic_set(&newf->count, 1);
	300
	301	spin_lock_init(&newf->file_lock);
	302	newf->next_fd = 0;
	303	new_fdt = &newf->fdtab;
	304	new_fdt->max_fds = NR_OPEN_DEFAULT;
	305	new_fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
	306	new_fdt->open_fds = (fd_set *)&newf->open_fds_init;
	307	new_fdt->fd = &newf->fd_array[0];
	308	new_fdt->next = NULL;
	309
	310	spin_lock(&oldf->file_lock);
	311	old_fdt = files_fdtable(oldf);
	312	open_files = count_open_files(old_fdt);
	313
	314	/*
	315	* Check whether we need to allocate a larger fd array and fd set.
	316	*/
	317	while (unlikely(open_files > new_fdt->max_fds)) {
	318	spin_unlock(&oldf->file_lock);
	319
	320	if (new_fdt != &newf->fdtab)
	321	__free_fdtable(new_fdt);
	322
	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_fdtable(new_fdt);
	332	*errorp = -EMFILE;
	333	goto out_release;
	334	}
	335
	336	/*
	337	* Reacquire the oldf lock and a pointer to its fd table
	338	* who knows it may have a new bigger fd table. We need
	339	* the latest pointer.
	340	*/
	341	spin_lock(&oldf->file_lock);
	342	old_fdt = files_fdtable(oldf);
	343	open_files = count_open_files(old_fdt);
	344	}
	345
	346	old_fds = old_fdt->fd;
	347	new_fds = new_fdt->fd;
	348
	349	memcpy(new_fdt->open_fds->fds_bits,
	350	old_fdt->open_fds->fds_bits, open_files/8);
	351	memcpy(new_fdt->close_on_exec->fds_bits,
	352	old_fdt->close_on_exec->fds_bits, open_files/8);
	353
	354	for (i = open_files; i != 0; i--) {
	355	struct file f = old_fds++;
	356	if (f) {
	357	get_file(f);
	358	} else {
	359	/*
	360	* The fd may be claimed in the fd bitmap but not yet
	361	* instantiated in the files array if a sibling thread
	362	* is partway through open(). So make sure that this
	363	* fd is available to the new process.
	364	*/
	365	FD_CLR(open_files - i, new_fdt->open_fds);
	366	}
	367	rcu_assign_pointer(*new_fds++, f);
	368	}
	369	spin_unlock(&oldf->file_lock);
	370
	371	/* compute the remainder to be cleared */
	372	size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
	373
	374	/* This is long word aligned thus could use a optimized version */
	375	memset(new_fds, 0, size);
	376
	377	if (new_fdt->max_fds > open_files) {
	378	int left = (new_fdt->max_fds-open_files)/8;
	379	int start = open_files / (8 * sizeof(unsigned long));
	380
	381	memset(&new_fdt->open_fds->fds_bits[start], 0, left);
	382	memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
	383	}
	384
	385	rcu_assign_pointer(newf->fdt, new_fdt);
	386
	387	return newf;
	388
	389	out_release:
	390	kmem_cache_free(files_cachep, newf);
	391	out:
	392	return NULL;
	393	}
	394
	395	static void __devinit fdtable_defer_list_init(int cpu)
	396	{
	397	struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
	398	spin_lock_init(&fddef->lock);
	399	INIT_WORK(&fddef->wq, free_fdtable_work);
	400	fddef->next = NULL;
	401	}
	402
	403	void __init files_defer_init(void)
	404	{
	405	int i;
	406	for_each_possible_cpu(i)
	407	fdtable_defer_list_init(i);
	408	sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
	409	-BITS_PER_LONG;
	410	}
	411
	412	struct files_struct init_files = {
	413	.count = ATOMIC_INIT(1),
	414	.fdt = &init_files.fdtab,
	415	.fdtab = {
	416	.max_fds = NR_OPEN_DEFAULT,
	417	.fd = &init_files.fd_array[0],
	418	.close_on_exec = (fd_set *)&init_files.close_on_exec_init,
	419	.open_fds = (fd_set *)&init_files.open_fds_init,
	420	},
	421	.file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock),
	422	};
	423
	424	/*
	425	* allocate a file descriptor, mark it busy.
	426	*/
	427	int alloc_fd(unsigned start, unsigned flags)
	428	{
	429	struct files_struct *files = current->files;
	430	unsigned int fd;
	431	int error;
	432	struct fdtable *fdt;
	433
	434	spin_lock(&files->file_lock);
	435	repeat:
	436	fdt = files_fdtable(files);
	437	fd = start;
	438	if (fd < files->next_fd)
	439	fd = files->next_fd;
	440
	441	if (fd < fdt->max_fds)
	442	fd = find_next_zero_bit(fdt->open_fds->fds_bits,
	443	fdt->max_fds, fd);
	444
	445	error = expand_files(files, fd);
	446	if (error < 0)
	447	goto out;
	448
	449	/*
	450	* If we needed to expand the fs array we
	451	* might have blocked - try again.
	452	*/
	453	if (error)
	454	goto repeat;
	455
	456	if (start <= files->next_fd)
	457	files->next_fd = fd + 1;
	458
	459	FD_SET(fd, fdt->open_fds);
	460	if (flags & O_CLOEXEC)
	461	FD_SET(fd, fdt->close_on_exec);
	462	else
	463	FD_CLR(fd, fdt->close_on_exec);
	464	error = fd;
	465	#if 1
	466	/* Sanity check */
	467	if (rcu_dereference_raw(fdt->fd[fd]) != NULL) {
	468	printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
	469	rcu_assign_pointer(fdt->fd[fd], NULL);
	470	}
	471	#endif
	472
	473	out:
	474	spin_unlock(&files->file_lock);
	475	return error;
	476	}
	477
	478	int get_unused_fd(void)
	479	{
	480	return alloc_fd(0, 0);
	481	}
	482	EXPORT_SYMBOL(get_unused_fd);