[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/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);

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

	/* 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),
};

/*
 * 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(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
1027abe8	9	#include <linux/module.h>
1da177e4 LT	10	#include <linux/fs.h>
	11	#include <linux/mm.h>
	12	#include <linux/time.h>
	13	#include <linux/slab.h>
	14	#include <linux/vmalloc.h>
	15	#include <linux/file.h>
9f3acc31	16	#include <linux/fdtable.h>
1da177e4	17	#include <linux/bitops.h>
ab2af1f5 DS	18	#include <linux/interrupt.h>
	19	#include <linux/spinlock.h>
	20	#include <linux/rcupdate.h>
	21	#include <linux/workqueue.h>
	22
	23	struct fdtable_defer {
	24	spinlock_t lock;
	25	struct work_struct wq;
ab2af1f5 DS	26	struct fdtable *next;
	27	};
	28
9cfe015a	29	int sysctl_nr_open __read_mostly = 1024*1024;
eceea0b3 AV	30	int sysctl_nr_open_min = BITS_PER_LONG;
eceea0b3 AV	31	int sysctl_nr_open_max = 1024 * 1024; /* raised later */
9cfe015a	32
ab2af1f5 DS	33	/*
	34	* We use this list to defer free fdtables that have vmalloced
	35	* sets/arrays. By keeping a per-cpu list, we avoid having to embed
	36	* the work_struct in fdtable itself which avoids a 64 byte (i386) increase in
	37	* this per-task structure.
	38	*/
	39	static DEFINE_PER_CPU(struct fdtable_defer, fdtable_defer_list);
1da177e4	40
5466b456	41	static inline void * alloc_fdmem(unsigned int size)
1da177e4	42	{
1da177e4	43	if (size <= PAGE_SIZE)
5466b456 VL	44	return kmalloc(size, GFP_KERNEL);
	45	else
	46	return vmalloc(size);
1da177e4 LT	47	}
1da177e4 LT	48
5466b456	49	static inline void free_fdarr(struct fdtable *fdt)
1da177e4	50	{
5466b456 VL	51	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *)))
5466b456 VL	52	kfree(fdt->fd);
1da177e4	53	else
5466b456	54	vfree(fdt->fd);
1da177e4 LT	55	}
1da177e4 LT	56
5466b456	57	static inline void free_fdset(struct fdtable *fdt)
1da177e4	58	{
5466b456 VL	59	if (fdt->max_fds <= (PAGE_SIZE * BITS_PER_BYTE / 2))
	60	kfree(fdt->open_fds);
	61	else
	62	vfree(fdt->open_fds);
ab2af1f5	63	}
1da177e4	64
65f27f38	65	static void free_fdtable_work(struct work_struct *work)
ab2af1f5	66	{
65f27f38 DH	67	struct fdtable_defer *f =
65f27f38 DH	68	container_of(work, struct fdtable_defer, wq);
ab2af1f5	69	struct fdtable *fdt;
1da177e4	70
ab2af1f5 DS	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;
5466b456 VL	77	vfree(fdt->fd);
	78	free_fdset(fdt);
	79	kfree(fdt);
ab2af1f5 DS	80	fdt = next;
	81	}
	82	}
1da177e4	83
4fd45812	84	void free_fdtable_rcu(struct rcu_head *rcu)
ab2af1f5 DS	85	{
ab2af1f5 DS	86	struct fdtable *fdt = container_of(rcu, struct fdtable, rcu);
ab2af1f5	87	struct fdtable_defer *fddef;
1da177e4	88
ab2af1f5	89	BUG_ON(!fdt);
ab2af1f5	90
4fd45812	91	if (fdt->max_fds <= NR_OPEN_DEFAULT) {
ab2af1f5	92	/*
4fd45812 VL	93	* This fdtable is embedded in the files structure and that
4fd45812 VL	94	* structure itself is getting destroyed.
ab2af1f5	95	*/
4fd45812 VL	96	kmem_cache_free(files_cachep,
4fd45812 VL	97	container_of(fdt, struct files_struct, fdtab));
ab2af1f5 DS	98	return;
ab2af1f5 DS	99	}
5466b456	100	if (fdt->max_fds <= (PAGE_SIZE / sizeof(struct file *))) {
ab2af1f5	101	kfree(fdt->fd);
5466b456	102	kfree(fdt->open_fds);
ab2af1f5	103	kfree(fdt);
1da177e4	104	} else {
ab2af1f5 DS	105	fddef = &get_cpu_var(fdtable_defer_list);
	106	spin_lock(&fddef->lock);
	107	fdt->next = fddef->next;
	108	fddef->next = fdt;
593be07a TH	109	/* vmallocs are handled from the workqueue context */
593be07a TH	110	schedule_work(&fddef->wq);
ab2af1f5 DS	111	spin_unlock(&fddef->lock);
ab2af1f5 DS	112	put_cpu_var(fdtable_defer_list);
1da177e4	113	}
ab2af1f5 DS	114	}
ab2af1f5 DS	115
ab2af1f5 DS	116	/*
	117	* Expand the fdset in the files_struct. Called with the files spinlock
	118	* held for write.
	119	*/
5466b456	120	static void copy_fdtable(struct fdtable nfdt, struct fdtable ofdt)
ab2af1f5	121	{
5466b456	122	unsigned int cpy, set;
ab2af1f5	123
5466b456	124	BUG_ON(nfdt->max_fds < ofdt->max_fds);
5466b456 VL	125
	126	cpy = ofdt->max_fds * sizeof(struct file *);
	127	set = (nfdt->max_fds - ofdt->max_fds) * sizeof(struct file *);
	128	memcpy(nfdt->fd, ofdt->fd, cpy);
	129	memset((char *)(nfdt->fd) + cpy, 0, set);
	130
	131	cpy = ofdt->max_fds / BITS_PER_BYTE;
	132	set = (nfdt->max_fds - ofdt->max_fds) / BITS_PER_BYTE;
	133	memcpy(nfdt->open_fds, ofdt->open_fds, cpy);
	134	memset((char *)(nfdt->open_fds) + cpy, 0, set);
	135	memcpy(nfdt->close_on_exec, ofdt->close_on_exec, cpy);
	136	memset((char *)(nfdt->close_on_exec) + cpy, 0, set);
1da177e4 LT	137	}
1da177e4 LT	138
5466b456	139	static struct fdtable * alloc_fdtable(unsigned int nr)
1da177e4	140	{
5466b456 VL	141	struct fdtable *fdt;
5466b456 VL	142	char *data;
1da177e4	143
ab2af1f5	144	/*
5466b456 VL	145	* Figure out how many fds we actually want to support in this fdtable.
	146	* Allocation steps are keyed to the size of the fdarray, since it
	147	* grows far faster than any of the other dynamic data. We try to fit
	148	* the fdarray into comfortable page-tuned chunks: starting at 1024B
	149	* and growing in powers of two from there on.
ab2af1f5	150	*/
5466b456 VL	151	nr /= (1024 / sizeof(struct file *));
	152	nr = roundup_pow_of_two(nr + 1);
	153	nr = (1024 / sizeof(struct file ));
5c598b34 AV	154	/*
	155	* Note that this can drive nr below what we had passed if sysctl_nr_open
	156	* had been set lower between the check in expand_files() and here. Deal
	157	* with that in caller, it's cheaper that way.
	158	*
	159	* We make sure that nr remains a multiple of BITS_PER_LONG - otherwise
	160	* bitmaps handling below becomes unpleasant, to put it mildly...
	161	*/
	162	if (unlikely(nr > sysctl_nr_open))
	163	nr = ((sysctl_nr_open - 1) \| (BITS_PER_LONG - 1)) + 1;
bbea9f69	164
5466b456 VL	165	fdt = kmalloc(sizeof(struct fdtable), GFP_KERNEL);
5466b456 VL	166	if (!fdt)
bbea9f69	167	goto out;
5466b456 VL	168	fdt->max_fds = nr;
	169	data = alloc_fdmem(nr * sizeof(struct file *));
	170	if (!data)
	171	goto out_fdt;
	172	fdt->fd = (struct file **)data;
	173	data = alloc_fdmem(max_t(unsigned int,
	174	2 * nr / BITS_PER_BYTE, L1_CACHE_BYTES));
	175	if (!data)
	176	goto out_arr;
	177	fdt->open_fds = (fd_set *)data;
	178	data += nr / BITS_PER_BYTE;
	179	fdt->close_on_exec = (fd_set *)data;
	180	INIT_RCU_HEAD(&fdt->rcu);
	181	fdt->next = NULL;
	182
ab2af1f5	183	return fdt;
5466b456 VL	184
	185	out_arr:
	186	free_fdarr(fdt);
	187	out_fdt:
ab2af1f5	188	kfree(fdt);
5466b456	189	out:
ab2af1f5 DS	190	return NULL;
ab2af1f5 DS	191	}
1da177e4	192
ab2af1f5	193	/*
74d392aa VL	194	* Expand the file descriptor table.
	195	* This function will allocate a new fdtable and both fd array and fdset, of
	196	* the given size.
	197	* Return <0 error code on error; 1 on successful completion.
	198	* The files->file_lock should be held on entry, and will be held on exit.
ab2af1f5 DS	199	*/
	200	static int expand_fdtable(struct files_struct *files, int nr)
	201	__releases(files->file_lock)
	202	__acquires(files->file_lock)
	203	{
74d392aa	204	struct fdtable new_fdt, cur_fdt;
ab2af1f5 DS	205
ab2af1f5 DS	206	spin_unlock(&files->file_lock);
74d392aa	207	new_fdt = alloc_fdtable(nr);
ab2af1f5	208	spin_lock(&files->file_lock);
74d392aa VL	209	if (!new_fdt)
74d392aa VL	210	return -ENOMEM;
5c598b34 AV	211	/*
	212	* extremely unlikely race - sysctl_nr_open decreased between the check in
	213	* caller and alloc_fdtable(). Cheaper to catch it here...
	214	*/
	215	if (unlikely(new_fdt->max_fds <= nr)) {
	216	free_fdarr(new_fdt);
	217	free_fdset(new_fdt);
	218	kfree(new_fdt);
	219	return -EMFILE;
	220	}
ab2af1f5	221	/*
74d392aa VL	222	* Check again since another task may have expanded the fd table while
74d392aa VL	223	* we dropped the lock
ab2af1f5	224	*/
74d392aa	225	cur_fdt = files_fdtable(files);
bbea9f69	226	if (nr >= cur_fdt->max_fds) {
74d392aa VL	227	/* Continue as planned */
	228	copy_fdtable(new_fdt, cur_fdt);
	229	rcu_assign_pointer(files->fdt, new_fdt);
4fd45812	230	if (cur_fdt->max_fds > NR_OPEN_DEFAULT)
01b2d93c	231	free_fdtable(cur_fdt);
ab2af1f5	232	} else {
74d392aa	233	/* Somebody else expanded, so undo our attempt */
5466b456 VL	234	free_fdarr(new_fdt);
	235	free_fdset(new_fdt);
	236	kfree(new_fdt);
ab2af1f5	237	}
74d392aa	238	return 1;
1da177e4 LT	239	}
	240
	241	/*
	242	* Expand files.
74d392aa VL	243	* This function will expand the file structures, if the requested size exceeds
	244	* the current capacity and there is room for expansion.
	245	* Return <0 error code on error; 0 when nothing done; 1 when files were
	246	* expanded and execution may have blocked.
	247	* The files->file_lock should be held on entry, and will be held on exit.
1da177e4 LT	248	*/
	249	int expand_files(struct files_struct *files, int nr)
	250	{
badf1662	251	struct fdtable *fdt;
1da177e4	252
badf1662	253	fdt = files_fdtable(files);
4e1e018e AV	254
	255	/*
	256	* N.B. For clone tasks sharing a files structure, this test
	257	* will limit the total number of files that can be opened.
	258	*/
	259	if (nr >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
	260	return -EMFILE;
	261
74d392aa	262	/* Do we need to expand? */
bbea9f69	263	if (nr < fdt->max_fds)
74d392aa	264	return 0;
4e1e018e	265
74d392aa	266	/* Can we expand? */
9cfe015a	267	if (nr >= sysctl_nr_open)
74d392aa VL	268	return -EMFILE;
	269
	270	/* All good, so we try */
	271	return expand_fdtable(files, nr);
1da177e4	272	}
ab2af1f5	273
02afc626 AV	274	static int count_open_files(struct fdtable *fdt)
	275	{
	276	int size = fdt->max_fds;
	277	int i;
	278
	279	/* Find the last open fd */
	280	for (i = size/(8*sizeof(long)); i > 0; ) {
	281	if (fdt->open_fds->fds_bits[--i])
	282	break;
	283	}
	284	i = (i+1) * 8 * sizeof(long);
	285	return i;
	286	}
	287
02afc626 AV	288	/*
	289	* Allocate a new files structure and copy contents from the
	290	* passed in files structure.
	291	* errorp will be valid only when the returned files_struct is NULL.
	292	*/
	293	struct files_struct dup_fd(struct files_struct oldf, int *errorp)
	294	{
	295	struct files_struct *newf;
	296	struct file old_fds, new_fds;
	297	int open_files, size, i;
	298	struct fdtable old_fdt, new_fdt;
	299
	300	*errorp = -ENOMEM;
afbec7ff	301	newf = kmem_cache_alloc(files_cachep, GFP_KERNEL);
02afc626 AV	302	if (!newf)
	303	goto out;
	304
afbec7ff AV	305	atomic_set(&newf->count, 1);
	306
	307	spin_lock_init(&newf->file_lock);
	308	newf->next_fd = 0;
	309	new_fdt = &newf->fdtab;
	310	new_fdt->max_fds = NR_OPEN_DEFAULT;
	311	new_fdt->close_on_exec = (fd_set *)&newf->close_on_exec_init;
	312	new_fdt->open_fds = (fd_set *)&newf->open_fds_init;
	313	new_fdt->fd = &newf->fd_array[0];
	314	INIT_RCU_HEAD(&new_fdt->rcu);
	315	new_fdt->next = NULL;
	316
02afc626 AV	317	spin_lock(&oldf->file_lock);
02afc626 AV	318	old_fdt = files_fdtable(oldf);
02afc626 AV	319	open_files = count_open_files(old_fdt);
	320
	321	/*
	322	* Check whether we need to allocate a larger fd array and fd set.
02afc626	323	*/
adbecb12	324	while (unlikely(open_files > new_fdt->max_fds)) {
02afc626	325	spin_unlock(&oldf->file_lock);
9dec3c4d	326
adbecb12 AV	327	if (new_fdt != &newf->fdtab) {
	328	free_fdarr(new_fdt);
	329	free_fdset(new_fdt);
	330	kfree(new_fdt);
	331	}
	332
9dec3c4d AV	333	new_fdt = alloc_fdtable(open_files - 1);
	334	if (!new_fdt) {
	335	*errorp = -ENOMEM;
	336	goto out_release;
	337	}
	338
	339	/* beyond sysctl_nr_open; nothing to do */
	340	if (unlikely(new_fdt->max_fds < open_files)) {
	341	free_fdarr(new_fdt);
	342	free_fdset(new_fdt);
	343	kfree(new_fdt);
	344	*errorp = -EMFILE;
02afc626	345	goto out_release;
9dec3c4d	346	}
9dec3c4d	347
02afc626 AV	348	/*
	349	* Reacquire the oldf lock and a pointer to its fd table
	350	* who knows it may have a new bigger fd table. We need
	351	* the latest pointer.
	352	*/
	353	spin_lock(&oldf->file_lock);
	354	old_fdt = files_fdtable(oldf);
adbecb12	355	open_files = count_open_files(old_fdt);
02afc626 AV	356	}
	357
	358	old_fds = old_fdt->fd;
	359	new_fds = new_fdt->fd;
	360
	361	memcpy(new_fdt->open_fds->fds_bits,
	362	old_fdt->open_fds->fds_bits, open_files/8);
	363	memcpy(new_fdt->close_on_exec->fds_bits,
	364	old_fdt->close_on_exec->fds_bits, open_files/8);
	365
	366	for (i = open_files; i != 0; i--) {
	367	struct file f = old_fds++;
	368	if (f) {
	369	get_file(f);
	370	} else {
	371	/*
	372	* The fd may be claimed in the fd bitmap but not yet
	373	* instantiated in the files array if a sibling thread
	374	* is partway through open(). So make sure that this
	375	* fd is available to the new process.
	376	*/
	377	FD_CLR(open_files - i, new_fdt->open_fds);
	378	}
	379	rcu_assign_pointer(*new_fds++, f);
	380	}
	381	spin_unlock(&oldf->file_lock);
	382
	383	/* compute the remainder to be cleared */
	384	size = (new_fdt->max_fds - open_files) * sizeof(struct file *);
	385
	386	/* This is long word aligned thus could use a optimized version */
	387	memset(new_fds, 0, size);
	388
	389	if (new_fdt->max_fds > open_files) {
	390	int left = (new_fdt->max_fds-open_files)/8;
	391	int start = open_files / (8 * sizeof(unsigned long));
	392
	393	memset(&new_fdt->open_fds->fds_bits[start], 0, left);
	394	memset(&new_fdt->close_on_exec->fds_bits[start], 0, left);
	395	}
	396
afbec7ff AV	397	rcu_assign_pointer(newf->fdt, new_fdt);
afbec7ff AV	398
02afc626 AV	399	return newf;
	400
	401	out_release:
	402	kmem_cache_free(files_cachep, newf);
	403	out:
	404	return NULL;
	405	}
	406
ab2af1f5 DS	407	static void __devinit fdtable_defer_list_init(int cpu)
	408	{
	409	struct fdtable_defer *fddef = &per_cpu(fdtable_defer_list, cpu);
	410	spin_lock_init(&fddef->lock);
65f27f38	411	INIT_WORK(&fddef->wq, free_fdtable_work);
ab2af1f5 DS	412	fddef->next = NULL;
	413	}
	414
	415	void __init files_defer_init(void)
	416	{
	417	int i;
0a945022	418	for_each_possible_cpu(i)
ab2af1f5	419	fdtable_defer_list_init(i);
eceea0b3 AV	420	sysctl_nr_open_max = min((size_t)INT_MAX, ~(size_t)0/sizeof(void *)) &
eceea0b3 AV	421	-BITS_PER_LONG;
ab2af1f5	422	}
f52111b1 AV	423
	424	struct files_struct init_files = {
	425	.count = ATOMIC_INIT(1),
	426	.fdt = &init_files.fdtab,
	427	.fdtab = {
	428	.max_fds = NR_OPEN_DEFAULT,
	429	.fd = &init_files.fd_array[0],
	430	.close_on_exec = (fd_set *)&init_files.close_on_exec_init,
	431	.open_fds = (fd_set *)&init_files.open_fds_init,
	432	.rcu = RCU_HEAD_INIT,
	433	},
	434	.file_lock = __SPIN_LOCK_UNLOCKED(init_task.file_lock),
	435	};
1027abe8 AV	436
	437	/*
	438	* allocate a file descriptor, mark it busy.
	439	*/
	440	int alloc_fd(unsigned start, unsigned flags)
	441	{
	442	struct files_struct *files = current->files;
	443	unsigned int fd;
	444	int error;
	445	struct fdtable *fdt;
	446
	447	spin_lock(&files->file_lock);
	448	repeat:
	449	fdt = files_fdtable(files);
	450	fd = start;
	451	if (fd < files->next_fd)
	452	fd = files->next_fd;
	453
	454	if (fd < fdt->max_fds)
	455	fd = find_next_zero_bit(fdt->open_fds->fds_bits,
	456	fdt->max_fds, fd);
	457
	458	error = expand_files(files, fd);
	459	if (error < 0)
	460	goto out;
	461
	462	/*
	463	* If we needed to expand the fs array we
	464	* might have blocked - try again.
	465	*/
	466	if (error)
	467	goto repeat;
	468
	469	if (start <= files->next_fd)
	470	files->next_fd = fd + 1;
	471
	472	FD_SET(fd, fdt->open_fds);
	473	if (flags & O_CLOEXEC)
	474	FD_SET(fd, fdt->close_on_exec);
	475	else
	476	FD_CLR(fd, fdt->close_on_exec);
	477	error = fd;
	478	#if 1
	479	/* Sanity check */
	480	if (rcu_dereference(fdt->fd[fd]) != NULL) {
	481	printk(KERN_WARNING "alloc_fd: slot %d not NULL!\n", fd);
	482	rcu_assign_pointer(fdt->fd[fd], NULL);
	483	}
	484	#endif
	485
	486	out:
	487	spin_unlock(&files->file_lock);
	488	return error;
	489	}
	490
	491	int get_unused_fd(void)
	492	{
	493	return alloc_fd(0, 0);
	494	}
	495	EXPORT_SYMBOL(get_unused_fd);