[linux.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/export.h>
#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/mmzone.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 void *alloc_fdmem(size_t size)
{
	/*
	 * Very large allocations can stress page reclaim, so fall back to
	 * vmalloc() if the allocation size will be considered "large" by the VM.
	 */
	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
		void *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;
	void *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 = data;

	data = alloc_fdmem(max_t(size_t,
				 2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
	if (!data)
		goto out_arr;
	fdt->open_fds = data;
	data += nr / BITS_PER_BYTE;
	fdt->close_on_exec = 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 / BITS_PER_LONG; i > 0; ) {
		if (fdt->open_fds[--i])
			break;
	}
	i = (i + 1) * BITS_PER_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 = newf->close_on_exec_init;
	new_fdt->open_fds = 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, old_fdt->open_fds, open_files / 8);
	memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, 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.
			 */
			__clear_open_fd(open_files - i, new_fdt);
		}
		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 / BITS_PER_LONG;

		memset(&new_fdt->open_fds[start], 0, left);
		memset(&new_fdt->close_on_exec[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	= init_files.close_on_exec_init,
		.open_fds	= 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, 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;

	__set_open_fd(fd, fdt);
	if (flags & O_CLOEXEC)
		__set_close_on_exec(fd, fdt);
	else
		__clear_close_on_exec(fd, fdt);
	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
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
630d9c47	9	#include <linux/export.h>
1da177e4 LT	10	#include <linux/fs.h>
1da177e4 LT	11	#include <linux/mm.h>
6d4831c2	12	#include <linux/mmzone.h>
1da177e4	13	#include <linux/time.h>
d43c36dc	14	#include <linux/sched.h>
1da177e4 LT	15	#include <linux/slab.h>
	16	#include <linux/vmalloc.h>
	17	#include <linux/file.h>
9f3acc31	18	#include <linux/fdtable.h>
1da177e4	19	#include <linux/bitops.h>
ab2af1f5 DS	20	#include <linux/interrupt.h>
	21	#include <linux/spinlock.h>
	22	#include <linux/rcupdate.h>
	23	#include <linux/workqueue.h>
	24
	25	struct fdtable_defer {
	26	spinlock_t lock;
	27	struct work_struct wq;
ab2af1f5 DS	28	struct fdtable *next;
	29	};
	30
9cfe015a	31	int sysctl_nr_open __read_mostly = 1024*1024;
eceea0b3 AV	32	int sysctl_nr_open_min = BITS_PER_LONG;
eceea0b3 AV	33	int sysctl_nr_open_max = 1024 * 1024; /* raised later */
9cfe015a	34
ab2af1f5 DS	35	/*
	36	* We use this list to defer free fdtables that have vmalloced
	37	* sets/arrays. By keeping a per-cpu list, we avoid having to embed
	38	* the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
	39	* this per-task structure.
	40	*/
	41	static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
1da177e4	42
1fd36adc	43	static void *alloc_fdmem(size_t size)
1da177e4	44	{
6d4831c2 AM	45	/*
	46	* Very large allocations can stress page reclaim, so fall back to
	47	* vmalloc() if the allocation size will be considered "large" by the VM.
	48	*/
	49	if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) {
	50	void *data = kmalloc(size, GFP_KERNEL\|__GFP_NOWARN);
	51	if (data != NULL)
	52	return data;
	53	}
a892e2d7	54	return vmalloc(size);
1da177e4 LT	55	}
1da177e4 LT	56
a892e2d7	57	static void free_fdmem(void *ptr)
1da177e4	58	{
a892e2d7	59	is_vmalloc_addr(ptr) ? vfree(ptr) : kfree(ptr);
1da177e4 LT	60	}
1da177e4 LT	61
a892e2d7	62	static void __free_fdtable(struct fdtable *fdt)
1da177e4	63	{
a892e2d7 CG	64	free_fdmem(fdt->fd);
	65	free_fdmem(fdt->open_fds);
	66	kfree(fdt);
ab2af1f5	67	}
1da177e4	68
65f27f38	69	static void free_fdtable_work(struct work_struct *work)
ab2af1f5	70	{
65f27f38 DH	71	struct fdtable_defer *f =
65f27f38 DH	72	container_of(work, struct fdtable_defer, wq);
ab2af1f5	73	struct fdtable *fdt;
1da177e4	74
ab2af1f5 DS	75	spin_lock_bh(&f->lock);
	76	fdt = f->next;
	77	f->next = NULL;
	78	spin_unlock_bh(&f->lock);
	79	while(fdt) {
	80	struct fdtable *next = fdt->next;
a892e2d7 CG	81
a892e2d7 CG	82	__free_fdtable(fdt);
ab2af1f5 DS	83	fdt = next;
	84	}
	85	}
1da177e4	86
4fd45812	87	void free_fdtable_rcu(struct rcu_head *rcu)
ab2af1f5 DS	88	{
ab2af1f5 DS	89	struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
ab2af1f5	90	struct fdtable_defer *fddef;
1da177e4	91
ab2af1f5	92	BUG_ON(!fdt);
ab2af1f5	93
4fd45812	94	if (fdt->max_fds <= NR_OPEN_DEFAULT) {
ab2af1f5	95	/*
4fd45812 VL	96	* This fdtable is embedded in the files structure and that
4fd45812 VL	97	* structure itself is getting destroyed.
ab2af1f5	98	*/
4fd45812 VL	99	kmem_cache_free(files_cachep,
4fd45812 VL	100	container_of(fdt, struct files_struct, fdtab));
ab2af1f5 DS	101	return;
ab2af1f5 DS	102	}
a892e2d7	103	if (!is_vmalloc_addr(fdt->fd) && !is_vmalloc_addr(fdt->open_fds)) {
ab2af1f5	104	kfree(fdt->fd);
5466b456	105	kfree(fdt->open_fds);
ab2af1f5	106	kfree(fdt);
1da177e4	107	} else {
ab2af1f5 DS	108	fddef = &get_cpu_var(fdtable_defer_list);
	109	spin_lock(&fddef->lock);
	110	fdt->next = fddef->next;
	111	fddef->next = fdt;
593be07a TH	112	/* vmallocs are handled from the workqueue context */
593be07a TH	113	schedule_work(&fddef->wq);
ab2af1f5 DS	114	spin_unlock(&fddef->lock);
ab2af1f5 DS	115	put_cpu_var(fdtable_defer_list);
1da177e4	116	}
ab2af1f5 DS	117	}
ab2af1f5 DS	118
ab2af1f5 DS	119	/*
	120	* Expand the fdset in the files_struct. Called with the files spinlock
	121	* held for write.
	122	*/
5466b456	123	static void copy_fdtable(struct fdtable nfdt, struct fdtable ofdt)
ab2af1f5	124	{
5466b456	125	unsigned int cpy, set;
ab2af1f5	126
5466b456	127	BUG_ON(nfdt->max_fds < ofdt->max_fds);
5466b456 VL	128
	129	cpy = ofdt->max_fds * sizeof(struct file *);
	130	set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
	131	memcpy(nfdt->fd, ofdt->fd, cpy);
	132	memset((char *)(nfdt->fd) + cpy, 0, set);
	133
	134	cpy = ofdt->max_fds / BITS_PER_BYTE;
	135	set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
	136	memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
	137	memset((char *)(nfdt->open_fds) + cpy, 0, set);
	138	memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
	139	memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
1da177e4 LT	140	}
1da177e4 LT	141
5466b456	142	static struct fdtable * alloc_fdtable(unsigned int nr)
1da177e4	143	{
5466b456	144	struct fdtable *fdt;
1fd36adc	145	void *data;
1da177e4	146
ab2af1f5	147	/*
5466b456 VL	148	* Figure out how many fds we actually want to support in this fdtable.
	149	* Allocation steps are keyed to the size of the fdarray, since it
	150	* grows far faster than any of the other dynamic data. We try to fit
	151	* the fdarray into comfortable page-tuned chunks: starting at 1024B
	152	* and growing in powers of two from there on.
ab2af1f5	153	*/
5466b456 VL	154	nr /= (1024 / sizeof(struct file *));
	155	nr = roundup_pow_of_two(nr + 1);
	156	nr = (1024 / sizeof(struct file ));
5c598b34 AV	157	/*
	158	* Note that this can drive nr below what we had passed if sysctl_nr_open
	159	* had been set lower between the check in expand_files() and here. Deal
	160	* with that in caller, it's cheaper that way.
	161	*
	162	* We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
	163	* bitmaps handling below becomes unpleasant, to put it mildly...
	164	*/
	165	if (unlikely(nr > sysctl_nr_open))
	166	nr = ((sysctl_nr_open - 1) \| (BITS_PER_LONG - 1)) + 1;
bbea9f69	167
5466b456 VL	168	fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
5466b456 VL	169	if (!fdt)
bbea9f69	170	goto out;
5466b456 VL	171	fdt->max_fds = nr;
	172	data = alloc_fdmem(nr * sizeof(struct file *));
	173	if (!data)
	174	goto out_fdt;
1fd36adc DH	175	fdt->fd = data;
	176
	177	data = alloc_fdmem(max_t(size_t,
5466b456 VL	178	2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
	179	if (!data)
	180	goto out_arr;
1fd36adc	181	fdt->open_fds = data;
5466b456	182	data += nr / BITS_PER_BYTE;
1fd36adc	183	fdt->close_on_exec = data;
5466b456 VL	184	fdt->next = NULL;
5466b456 VL	185
ab2af1f5	186	return fdt;
5466b456 VL	187
5466b456 VL	188	out_arr:
a892e2d7	189	free_fdmem(fdt->fd);
5466b456	190	out_fdt:
ab2af1f5	191	kfree(fdt);
5466b456	192	out:
ab2af1f5 DS	193	return NULL;
ab2af1f5 DS	194	}
1da177e4	195
ab2af1f5	196	/*
74d392aa VL	197	* Expand the file descriptor table.
	198	* This function will allocate a new fdtable and both fd array and fdset, of
	199	* the given size.
	200	* Return <0 error code on error; 1 on successful completion.
	201	* The files->file_lock should be held on entry, and will be held on exit.
ab2af1f5 DS	202	*/
	203	static int expand_fdtable(struct files_struct *files, int nr)
	204	__releases(files->file_lock)
	205	__acquires(files->file_lock)
	206	{
74d392aa	207	struct fdtable new_fdt, cur_fdt;
ab2af1f5 DS	208
ab2af1f5 DS	209	spin_unlock(&files->file_lock);
74d392aa	210	new_fdt = alloc_fdtable(nr);
ab2af1f5	211	spin_lock(&files->file_lock);
74d392aa VL	212	if (!new_fdt)
74d392aa VL	213	return -ENOMEM;
5c598b34 AV	214	/*
	215	* extremely unlikely race - sysctl_nr_open decreased between the check in
	216	* caller and alloc_fdtable(). Cheaper to catch it here...
	217	*/
	218	if (unlikely(new_fdt->max_fds <= nr)) {
a892e2d7	219	__free_fdtable(new_fdt);
5c598b34 AV	220	return -EMFILE;
5c598b34 AV	221	}
ab2af1f5	222	/*
74d392aa VL	223	* Check again since another task may have expanded the fd table while
74d392aa VL	224	* we dropped the lock
ab2af1f5	225	*/
74d392aa	226	cur_fdt = files_fdtable(files);
bbea9f69	227	if (nr >= cur_fdt->max_fds) {
74d392aa VL	228	/* Continue as planned */
	229	copy_fdtable(new_fdt, cur_fdt);
	230	rcu_assign_pointer(files->fdt, new_fdt);
4fd45812	231	if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
01b2d93c	232	free_fdtable(cur_fdt);
ab2af1f5	233	} else {
74d392aa	234	/* Somebody else expanded, so undo our attempt */
a892e2d7	235	__free_fdtable(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);
4e1e018e AV	253
	254	/*
	255	* N.B. For clone tasks sharing a files structure, this test
	256	* will limit the total number of files that can be opened.
	257	*/
d554ed89	258	if (nr >= rlimit(RLIMIT_NOFILE))
4e1e018e AV	259	return -EMFILE;
4e1e018e AV	260
74d392aa	261	/* Do we need to expand? */
bbea9f69	262	if (nr < fdt->max_fds)
74d392aa	263	return 0;
4e1e018e	264
74d392aa	265	/* Can we expand? */
9cfe015a	266	if (nr >= sysctl_nr_open)
74d392aa VL	267	return -EMFILE;
	268
	269	/* All good, so we try */
	270	return expand_fdtable(files, nr);
1da177e4	271	}
ab2af1f5	272
02afc626 AV	273	static int count_open_files(struct fdtable *fdt)
	274	{
	275	int size = fdt->max_fds;
	276	int i;
	277
	278	/* Find the last open fd */
1fd36adc DH	279	for (i = size / BITS_PER_LONG; i > 0; ) {
1fd36adc DH	280	if (fdt->open_fds[--i])
02afc626 AV	281	break;
02afc626 AV	282	}
1fd36adc	283	i = (i + 1) * BITS_PER_LONG;
02afc626 AV	284	return i;
	285	}
	286
02afc626 AV	287	/*
	288	* Allocate a new files structure and copy contents from the
	289	* passed in files structure.
	290	* errorp will be valid only when the returned files_struct is NULL.
	291	*/
	292	struct files_struct dup_fd(struct files_struct oldf, int *errorp)
	293	{
	294	struct files_struct *newf;
	295	struct file old_fds, new_fds;
	296	int open_files, size, i;
	297	struct fdtable old_fdt, new_fdt;
	298
	299	*errorp = -ENOMEM;
afbec7ff	300	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
02afc626 AV	301	if (!newf)
	302	goto out;
	303
afbec7ff AV	304	atomic_set(&newf->count, 1);
	305
	306	spin_lock_init(&newf->file_lock);
	307	newf->next_fd = 0;
	308	new_fdt = &newf->fdtab;
	309	new_fdt->max_fds = NR_OPEN_DEFAULT;
1fd36adc DH	310	new_fdt->close_on_exec = newf->close_on_exec_init;
1fd36adc DH	311	new_fdt->open_fds = newf->open_fds_init;
afbec7ff	312	new_fdt->fd = &newf->fd_array[0];
afbec7ff AV	313	new_fdt->next = NULL;
afbec7ff AV	314
02afc626 AV	315	spin_lock(&oldf->file_lock);
02afc626 AV	316	old_fdt = files_fdtable(oldf);
02afc626 AV	317	open_files = count_open_files(old_fdt);
	318
	319	/*
	320	* Check whether we need to allocate a larger fd array and fd set.
02afc626	321	*/
adbecb12	322	while (unlikely(open_files > new_fdt->max_fds)) {
02afc626	323	spin_unlock(&oldf->file_lock);
9dec3c4d	324
a892e2d7 CG	325	if (new_fdt != &newf->fdtab)
a892e2d7 CG	326	__free_fdtable(new_fdt);
adbecb12	327
9dec3c4d AV	328	new_fdt = alloc_fdtable(open_files - 1);
	329	if (!new_fdt) {
	330	*errorp = -ENOMEM;
	331	goto out_release;
	332	}
	333
	334	/* beyond sysctl_nr_open; nothing to do */
	335	if (unlikely(new_fdt->max_fds < open_files)) {
a892e2d7	336	__free_fdtable(new_fdt);
9dec3c4d	337	*errorp = -EMFILE;
02afc626	338	goto out_release;
9dec3c4d	339	}
9dec3c4d	340
02afc626 AV	341	/*
	342	* Reacquire the oldf lock and a pointer to its fd table
	343	* who knows it may have a new bigger fd table. We need
	344	* the latest pointer.
	345	*/
	346	spin_lock(&oldf->file_lock);
	347	old_fdt = files_fdtable(oldf);
adbecb12	348	open_files = count_open_files(old_fdt);
02afc626 AV	349	}
	350
	351	old_fds = old_fdt->fd;
	352	new_fds = new_fdt->fd;
	353
1fd36adc DH	354	memcpy(new_fdt->open_fds, old_fdt->open_fds, open_files / 8);
1fd36adc DH	355	memcpy(new_fdt->close_on_exec, old_fdt->close_on_exec, open_files / 8);
02afc626 AV	356
	357	for (i = open_files; i != 0; i--) {
	358	struct file f = old_fds++;
	359	if (f) {
	360	get_file(f);
	361	} else {
	362	/*
	363	* The fd may be claimed in the fd bitmap but not yet
	364	* instantiated in the files array if a sibling thread
	365	* is partway through open(). So make sure that this
	366	* fd is available to the new process.
	367	*/
1dce27c5	368	__clear_open_fd(open_files - i, new_fdt);
02afc626 AV	369	}
	370	rcu_assign_pointer(*new_fds++, f);
	371	}
	372	spin_unlock(&oldf->file_lock);
	373
	374	/* compute the remainder to be cleared */
	375	size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
	376
	377	/* This is long word aligned thus could use a optimized version */
	378	memset(new_fds, 0, size);
	379
	380	if (new_fdt->max_fds > open_files) {
1fd36adc DH	381	int left = (new_fdt->max_fds - open_files) / 8;
1fd36adc DH	382	int start = open_files / BITS_PER_LONG;
02afc626	383
1fd36adc DH	384	memset(&new_fdt->open_fds[start], 0, left);
1fd36adc DH	385	memset(&new_fdt->close_on_exec[start], 0, left);
02afc626 AV	386	}
02afc626 AV	387
afbec7ff AV	388	rcu_assign_pointer(newf->fdt, new_fdt);
afbec7ff AV	389
02afc626 AV	390	return newf;
	391
	392	out_release:
	393	kmem_cache_free(files_cachep, newf);
	394	out:
	395	return NULL;
	396	}
	397
ab2af1f5 DS	398	static void __devinit fdtable_defer_list_init(int cpu)
	399	{
	400	struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
	401	spin_lock_init(&fddef->lock);
65f27f38	402	INIT_WORK(&fddef->wq, free_fdtable_work);
ab2af1f5 DS	403	fddef->next = NULL;
	404	}
	405
	406	void __init files_defer_init(void)
	407	{
	408	int i;
0a945022	409	for_each_possible_cpu(i)
ab2af1f5	410	fdtable_defer_list_init(i);
eceea0b3 AV	411	sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
eceea0b3 AV	412	-BITS_PER_LONG;
ab2af1f5	413	}
f52111b1 AV	414
	415	struct files_struct init_files = {
	416	.count = ATOMIC_INIT(1),
	417	.fdt = &init_files.fdtab,
	418	.fdtab = {
	419	.max_fds = NR_OPEN_DEFAULT,
	420	.fd = &init_files.fd_array[0],
1fd36adc DH	421	.close_on_exec = init_files.close_on_exec_init,
1fd36adc DH	422	.open_fds = init_files.open_fds_init,
f52111b1 AV	423	},
	424	.file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock),
	425	};
1027abe8 AV	426
	427	/*
	428	* allocate a file descriptor, mark it busy.
	429	*/
	430	int alloc_fd(unsigned start, unsigned flags)
	431	{
	432	struct files_struct *files = current->files;
	433	unsigned int fd;
	434	int error;
	435	struct fdtable *fdt;
	436
	437	spin_lock(&files->file_lock);
	438	repeat:
	439	fdt = files_fdtable(files);
	440	fd = start;
	441	if (fd < files->next_fd)
	442	fd = files->next_fd;
	443
	444	if (fd < fdt->max_fds)
1fd36adc	445	fd = find_next_zero_bit(fdt->open_fds, fdt->max_fds, fd);
1027abe8 AV	446
	447	error = expand_files(files, fd);
	448	if (error < 0)
	449	goto out;
	450
	451	/*
	452	* If we needed to expand the fs array we
	453	* might have blocked - try again.
	454	*/
	455	if (error)
	456	goto repeat;
	457
	458	if (start <= files->next_fd)
	459	files->next_fd = fd + 1;
	460
1dce27c5	461	__set_open_fd(fd, fdt);
1027abe8	462	if (flags & O_CLOEXEC)
1dce27c5	463	__set_close_on_exec(fd, fdt);
1027abe8	464	else
1dce27c5	465	__clear_close_on_exec(fd, fdt);
1027abe8 AV	466	error = fd;
	467	#if 1
	468	/* Sanity check */
7dc52157	469	if (rcu_dereference_raw(fdt->fd[fd]) != NULL) {
1027abe8 AV	470	printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
	471	rcu_assign_pointer(fdt->fd[fd], NULL);
	472	}
	473	#endif
	474
	475	out:
	476	spin_unlock(&files->file_lock);
	477	return error;
	478	}
	479
	480	int get_unused_fd(void)
	481	{
	482	return alloc_fd(0, 0);
	483	}
	484	EXPORT_SYMBOL(get_unused_fd);