[linux-2.6-block.git] / fs / nfs / pagelist.c

/*
 * linux/fs/nfs/pagelist.c
 *
 * A set of helper functions for managing NFS read and write requests.
 * The main purpose of these routines is to provide support for the
 * coalescing of several requests into a single RPC call.
 *
 * Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
 *
 */

#include <linux/config.h>
#include <linux/slab.h>
#include <linux/file.h>
#include <linux/sunrpc/clnt.h>
#include <linux/nfs3.h>
#include <linux/nfs4.h>
#include <linux/nfs_page.h>
#include <linux/nfs_fs.h>
#include <linux/nfs_mount.h>

#define NFS_PARANOIA 1

static kmem_cache_t *nfs_page_cachep;

static inline struct nfs_page *
nfs_page_alloc(void)
{
	struct nfs_page	*p;
	p = kmem_cache_alloc(nfs_page_cachep, SLAB_KERNEL);
	if (p) {
		memset(p, 0, sizeof(*p));
		INIT_LIST_HEAD(&p->wb_list);
	}
	return p;
}

static inline void
nfs_page_free(struct nfs_page *p)
{
	kmem_cache_free(nfs_page_cachep, p);
}

/**
 * nfs_create_request - Create an NFS read/write request.
 * @file: file descriptor to use
 * @inode: inode to which the request is attached
 * @page: page to write
 * @offset: starting offset within the page for the write
 * @count: number of bytes to read/write
 *
 * The page must be locked by the caller. This makes sure we never
 * create two different requests for the same page, and avoids
 * a possible deadlock when we reach the hard limit on the number
 * of dirty pages.
 * User should ensure it is safe to sleep in this function.
 */
struct nfs_page *
nfs_create_request(struct nfs_open_context *ctx, struct inode *inode,
		   struct page *page,
		   unsigned int offset, unsigned int count)
{
	struct nfs_server *server = NFS_SERVER(inode);
	struct nfs_page		*req;

	/* Deal with hard limits.  */
	for (;;) {
		/* try to allocate the request struct */
		req = nfs_page_alloc();
		if (req != NULL)
			break;

		/* Try to free up at least one request in order to stay
		 * below the hard limit
		 */
		if (signalled() && (server->flags & NFS_MOUNT_INTR))
			return ERR_PTR(-ERESTARTSYS);
		yield();
	}

	/* Initialize the request struct. Initially, we assume a
	 * long write-back delay. This will be adjusted in
	 * update_nfs_request below if the region is not locked. */
	req->wb_page    = page;
	atomic_set(&req->wb_complete, 0);
	req->wb_index	= page->index;
	page_cache_get(page);
	req->wb_offset  = offset;
	req->wb_pgbase	= offset;
	req->wb_bytes   = count;
	atomic_set(&req->wb_count, 1);
	req->wb_context = get_nfs_open_context(ctx);

	return req;
}

/**
 * nfs_unlock_request - Unlock request and wake up sleepers.
 * @req:
 */
void nfs_unlock_request(struct nfs_page *req)
{
	if (!NFS_WBACK_BUSY(req)) {
		printk(KERN_ERR "NFS: Invalid unlock attempted\n");
		BUG();
	}
	smp_mb__before_clear_bit();
	clear_bit(PG_BUSY, &req->wb_flags);
	smp_mb__after_clear_bit();
	wake_up_bit(&req->wb_flags, PG_BUSY);
	nfs_release_request(req);
}

/**
 * nfs_set_page_writeback_locked - Lock a request for writeback
 * @req:
 */
int nfs_set_page_writeback_locked(struct nfs_page *req)
{
	struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);

	if (!nfs_lock_request(req))
		return 0;
	radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
	return 1;
}

/**
 * nfs_clear_page_writeback - Unlock request and wake up sleepers
 */
void nfs_clear_page_writeback(struct nfs_page *req)
{
	struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);

	spin_lock(&nfsi->req_lock);
	radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
	spin_unlock(&nfsi->req_lock);
	nfs_unlock_request(req);
}

/**
 * nfs_clear_request - Free up all resources allocated to the request
 * @req:
 *
 * Release page resources associated with a write request after it
 * has completed.
 */
void nfs_clear_request(struct nfs_page *req)
{
	if (req->wb_page) {
		page_cache_release(req->wb_page);
		req->wb_page = NULL;
	}
}


/**
 * nfs_release_request - Release the count on an NFS read/write request
 * @req: request to release
 *
 * Note: Should never be called with the spinlock held!
 */
void
nfs_release_request(struct nfs_page *req)
{
	if (!atomic_dec_and_test(&req->wb_count))
		return;

#ifdef NFS_PARANOIA
	BUG_ON (!list_empty(&req->wb_list));
	BUG_ON (NFS_WBACK_BUSY(req));
#endif

	/* Release struct file or cached credential */
	nfs_clear_request(req);
	put_nfs_open_context(req->wb_context);
	nfs_page_free(req);
}

static int nfs_wait_bit_interruptible(void *word)
{
	int ret = 0;

	if (signal_pending(current))
		ret = -ERESTARTSYS;
	else
		schedule();
	return ret;
}

/**
 * nfs_wait_on_request - Wait for a request to complete.
 * @req: request to wait upon.
 *
 * Interruptible by signals only if mounted with intr flag.
 * The user is responsible for holding a count on the request.
 */
int
nfs_wait_on_request(struct nfs_page *req)
{
        struct rpc_clnt	*clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
	sigset_t oldmask;
	int ret = 0;

	if (!test_bit(PG_BUSY, &req->wb_flags))
		goto out;
	/*
	 * Note: the call to rpc_clnt_sigmask() suffices to ensure that we
	 *	 are not interrupted if intr flag is not set
	 */
	rpc_clnt_sigmask(clnt, &oldmask);
	ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
			nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
	rpc_clnt_sigunmask(clnt, &oldmask);
out:
	return ret;
}

/**
 * nfs_coalesce_requests - Split coalesced requests out from a list.
 * @head: source list
 * @dst: destination list
 * @nmax: maximum number of requests to coalesce
 *
 * Moves a maximum of 'nmax' elements from one list to another.
 * The elements are checked to ensure that they form a contiguous set
 * of pages, and that the RPC credentials are the same.
 */
int
nfs_coalesce_requests(struct list_head *head, struct list_head *dst,
		      unsigned int nmax)
{
	struct nfs_page		*req = NULL;
	unsigned int		npages = 0;

	while (!list_empty(head)) {
		struct nfs_page	*prev = req;

		req = nfs_list_entry(head->next);
		if (prev) {
			if (req->wb_context->cred != prev->wb_context->cred)
				break;
			if (req->wb_context->lockowner != prev->wb_context->lockowner)
				break;
			if (req->wb_context->state != prev->wb_context->state)
				break;
			if (req->wb_index != (prev->wb_index + 1))
				break;

			if (req->wb_pgbase != 0)
				break;
		}
		nfs_list_remove_request(req);
		nfs_list_add_request(req, dst);
		npages++;
		if (req->wb_pgbase + req->wb_bytes != PAGE_CACHE_SIZE)
			break;
		if (npages >= nmax)
			break;
	}
	return npages;
}

#define NFS_SCAN_MAXENTRIES 16
/**
 * nfs_scan_lock_dirty - Scan the radix tree for dirty requests
 * @nfsi: NFS inode
 * @dst: Destination list
 * @idx_start: lower bound of page->index to scan
 * @npages: idx_start + npages sets the upper bound to scan.
 *
 * Moves elements from one of the inode request lists.
 * If the number of requests is set to 0, the entire address_space
 * starting at index idx_start, is scanned.
 * The requests are *not* checked to ensure that they form a contiguous set.
 * You must be holding the inode's req_lock when calling this function
 */
int
nfs_scan_lock_dirty(struct nfs_inode *nfsi, struct list_head *dst,
	      unsigned long idx_start, unsigned int npages)
{
	struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
	struct nfs_page *req;
	unsigned long idx_end;
	int found, i;
	int res;

	res = 0;
	if (npages == 0)
		idx_end = ~0;
	else
		idx_end = idx_start + npages - 1;

	for (;;) {
		found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
				(void **)&pgvec[0], idx_start, NFS_SCAN_MAXENTRIES,
				NFS_PAGE_TAG_DIRTY);
		if (found <= 0)
			break;
		for (i = 0; i < found; i++) {
			req = pgvec[i];
			if (req->wb_index > idx_end)
				goto out;

			idx_start = req->wb_index + 1;

			if (nfs_set_page_writeback_locked(req)) {
				radix_tree_tag_clear(&nfsi->nfs_page_tree,
						req->wb_index, NFS_PAGE_TAG_DIRTY);
				nfs_list_remove_request(req);
				nfs_list_add_request(req, dst);
				res++;
			}
		}
	}
out:
	return res;
}

/**
 * nfs_scan_list - Scan a list for matching requests
 * @head: One of the NFS inode request lists
 * @dst: Destination list
 * @idx_start: lower bound of page->index to scan
 * @npages: idx_start + npages sets the upper bound to scan.
 *
 * Moves elements from one of the inode request lists.
 * If the number of requests is set to 0, the entire address_space
 * starting at index idx_start, is scanned.
 * The requests are *not* checked to ensure that they form a contiguous set.
 * You must be holding the inode's req_lock when calling this function
 */
int
nfs_scan_list(struct list_head *head, struct list_head *dst,
	      unsigned long idx_start, unsigned int npages)
{
	struct list_head	*pos, *tmp;
	struct nfs_page		*req;
	unsigned long		idx_end;
	int			res;

	res = 0;
	if (npages == 0)
		idx_end = ~0;
	else
		idx_end = idx_start + npages - 1;

	list_for_each_safe(pos, tmp, head) {

		req = nfs_list_entry(pos);

		if (req->wb_index < idx_start)
			continue;
		if (req->wb_index > idx_end)
			break;

		if (!nfs_set_page_writeback_locked(req))
			continue;
		nfs_list_remove_request(req);
		nfs_list_add_request(req, dst);
		res++;
	}
	return res;
}

int nfs_init_nfspagecache(void)
{
	nfs_page_cachep = kmem_cache_create("nfs_page",
					    sizeof(struct nfs_page),
					    0, SLAB_HWCACHE_ALIGN,
					    NULL, NULL);
	if (nfs_page_cachep == NULL)
		return -ENOMEM;

	return 0;
}

void nfs_destroy_nfspagecache(void)
{
	if (kmem_cache_destroy(nfs_page_cachep))
		printk(KERN_INFO "nfs_page: not all structures were freed\n");
}
Commit	Line	Data
1da177e4 LT	1	/*
	2	* linux/fs/nfs/pagelist.c
	3	*
	4	* A set of helper functions for managing NFS read and write requests.
	5	* The main purpose of these routines is to provide support for the
	6	* coalescing of several requests into a single RPC call.
	7	*
	8	* Copyright 2000, 2001 (c) Trond Myklebust <trond.myklebust@fys.uio.no>
	9	*
	10	*/
	11
	12	#include <linux/config.h>
	13	#include <linux/slab.h>
	14	#include <linux/file.h>
	15	#include <linux/sunrpc/clnt.h>
	16	#include <linux/nfs3.h>
	17	#include <linux/nfs4.h>
	18	#include <linux/nfs_page.h>
	19	#include <linux/nfs_fs.h>
	20	#include <linux/nfs_mount.h>
	21
	22	#define NFS_PARANOIA 1
	23
	24	static kmem_cache_t *nfs_page_cachep;
	25
	26	static inline struct nfs_page *
	27	nfs_page_alloc(void)
	28	{
	29	struct nfs_page *p;
	30	p = kmem_cache_alloc(nfs_page_cachep, SLAB_KERNEL);
	31	if (p) {
	32	memset(p, 0, sizeof(*p));
	33	INIT_LIST_HEAD(&p->wb_list);
	34	}
	35	return p;
	36	}
	37
	38	static inline void
	39	nfs_page_free(struct nfs_page *p)
	40	{
	41	kmem_cache_free(nfs_page_cachep, p);
	42	}
	43
	44	/**
	45	* nfs_create_request - Create an NFS read/write request.
	46	* @file: file descriptor to use
	47	* @inode: inode to which the request is attached
	48	* @page: page to write
	49	* @offset: starting offset within the page for the write
	50	* @count: number of bytes to read/write
	51	*
	52	* The page must be locked by the caller. This makes sure we never
	53	* create two different requests for the same page, and avoids
	54	* a possible deadlock when we reach the hard limit on the number
	55	* of dirty pages.
	56	* User should ensure it is safe to sleep in this function.
	57	*/
	58	struct nfs_page *
	59	nfs_create_request(struct nfs_open_context ctx, struct inode inode,
	60	struct page *page,
	61	unsigned int offset, unsigned int count)
	62	{
	63	struct nfs_server *server = NFS_SERVER(inode);
	64	struct nfs_page *req;
65
66	/* Deal with hard limits. */
67	for (;;) {
68	/* try to allocate the request struct */
69	req = nfs_page_alloc();
70	if (req != NULL)
71	break;
72
73	/* Try to free up at least one request in order to stay
74	* below the hard limit
75	*/
76	if (signalled() && (server->flags & NFS_MOUNT_INTR))
77	return ERR_PTR(-ERESTARTSYS);
78	yield();
79	}
80
81	/* Initialize the request struct. Initially, we assume a
82	* long write-back delay. This will be adjusted in
83	* update_nfs_request below if the region is not locked. */
84	req->wb_page = page;
85	atomic_set(&req->wb_complete, 0);
86	req->wb_index = page->index;
87	page_cache_get(page);
88	req->wb_offset = offset;
89	req->wb_pgbase = offset;
90	req->wb_bytes = count;
91	atomic_set(&req->wb_count, 1);
92	req->wb_context = get_nfs_open_context(ctx);
93
94	return req;
95	}
96
97	/**
98	* nfs_unlock_request - Unlock request and wake up sleepers.
99	* @req:
100	*/
101	void nfs_unlock_request(struct nfs_page *req)
102	{
103	if (!NFS_WBACK_BUSY(req)) {
104	printk(KERN_ERR "NFS: Invalid unlock attempted\n");
105	BUG();
106	}
107	smp_mb__before_clear_bit();
108	clear_bit(PG_BUSY, &req->wb_flags);
109	smp_mb__after_clear_bit();
464a98bd	110	wake_up_bit(&req->wb_flags, PG_BUSY);
1da177e4 LT	111	nfs_release_request(req);
	112	}
	113
c6a556b8 TM	114	/**
	115	* nfs_set_page_writeback_locked - Lock a request for writeback
	116	* @req:
	117	*/
	118	int nfs_set_page_writeback_locked(struct nfs_page *req)
	119	{
	120	struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
	121
	122	if (!nfs_lock_request(req))
	123	return 0;
	124	radix_tree_tag_set(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
	125	return 1;
	126	}
	127
	128	/**
	129	* nfs_clear_page_writeback - Unlock request and wake up sleepers
	130	*/
	131	void nfs_clear_page_writeback(struct nfs_page *req)
	132	{
	133	struct nfs_inode *nfsi = NFS_I(req->wb_context->dentry->d_inode);
	134
	135	spin_lock(&nfsi->req_lock);
	136	radix_tree_tag_clear(&nfsi->nfs_page_tree, req->wb_index, NFS_PAGE_TAG_WRITEBACK);
	137	spin_unlock(&nfsi->req_lock);
	138	nfs_unlock_request(req);
	139	}
	140
1da177e4 LT	141	/**
	142	* nfs_clear_request - Free up all resources allocated to the request
	143	* @req:
	144	*
	145	* Release page resources associated with a write request after it
	146	* has completed.
	147	*/
	148	void nfs_clear_request(struct nfs_page *req)
	149	{
	150	if (req->wb_page) {
	151	page_cache_release(req->wb_page);
	152	req->wb_page = NULL;
	153	}
	154	}
	155
	156
	157	/**
	158	* nfs_release_request - Release the count on an NFS read/write request
	159	* @req: request to release
	160	*
	161	* Note: Should never be called with the spinlock held!
	162	*/
	163	void
	164	nfs_release_request(struct nfs_page *req)
	165	{
	166	if (!atomic_dec_and_test(&req->wb_count))
	167	return;
	168
	169	#ifdef NFS_PARANOIA
	170	BUG_ON (!list_empty(&req->wb_list));
	171	BUG_ON (NFS_WBACK_BUSY(req));
	172	#endif
	173
	174	/* Release struct file or cached credential */
	175	nfs_clear_request(req);
	176	put_nfs_open_context(req->wb_context);
	177	nfs_page_free(req);
	178	}
	179
464a98bd TM	180	static int nfs_wait_bit_interruptible(void *word)
	181	{
	182	int ret = 0;
	183
	184	if (signal_pending(current))
	185	ret = -ERESTARTSYS;
	186	else
	187	schedule();
	188	return ret;
	189	}
	190
1da177e4 LT	191	/**
	192	* nfs_wait_on_request - Wait for a request to complete.
	193	* @req: request to wait upon.
	194	*
	195	* Interruptible by signals only if mounted with intr flag.
	196	* The user is responsible for holding a count on the request.
	197	*/
	198	int
	199	nfs_wait_on_request(struct nfs_page *req)
	200	{
464a98bd TM	201	struct rpc_clnt *clnt = NFS_CLIENT(req->wb_context->dentry->d_inode);
	202	sigset_t oldmask;
	203	int ret = 0;
	204
	205	if (!test_bit(PG_BUSY, &req->wb_flags))
	206	goto out;
	207	/*
	208	* Note: the call to rpc_clnt_sigmask() suffices to ensure that we
	209	* are not interrupted if intr flag is not set
	210	*/
	211	rpc_clnt_sigmask(clnt, &oldmask);
	212	ret = out_of_line_wait_on_bit(&req->wb_flags, PG_BUSY,
	213	nfs_wait_bit_interruptible, TASK_INTERRUPTIBLE);
	214	rpc_clnt_sigunmask(clnt, &oldmask);
	215	out:
	216	return ret;
1da177e4 LT	217	}
	218
	219	/**
	220	* nfs_coalesce_requests - Split coalesced requests out from a list.
	221	* @head: source list
	222	* @dst: destination list
	223	* @nmax: maximum number of requests to coalesce
	224	*
	225	* Moves a maximum of 'nmax' elements from one list to another.
	226	* The elements are checked to ensure that they form a contiguous set
	227	* of pages, and that the RPC credentials are the same.
	228	*/
	229	int
	230	nfs_coalesce_requests(struct list_head head, struct list_head dst,
	231	unsigned int nmax)
	232	{
	233	struct nfs_page *req = NULL;
	234	unsigned int npages = 0;
	235
	236	while (!list_empty(head)) {
	237	struct nfs_page *prev = req;
	238
	239	req = nfs_list_entry(head->next);
	240	if (prev) {
	241	if (req->wb_context->cred != prev->wb_context->cred)
	242	break;
	243	if (req->wb_context->lockowner != prev->wb_context->lockowner)
	244	break;
	245	if (req->wb_context->state != prev->wb_context->state)
	246	break;
	247	if (req->wb_index != (prev->wb_index + 1))
	248	break;
	249
	250	if (req->wb_pgbase != 0)
	251	break;
	252	}
	253	nfs_list_remove_request(req);
	254	nfs_list_add_request(req, dst);
	255	npages++;
	256	if (req->wb_pgbase + req->wb_bytes != PAGE_CACHE_SIZE)
	257	break;
	258	if (npages >= nmax)
	259	break;
	260	}
	261	return npages;
	262	}
	263
3da28eb1 TM	264	#define NFS_SCAN_MAXENTRIES 16
	265	/**
	266	* nfs_scan_lock_dirty - Scan the radix tree for dirty requests
	267	* @nfsi: NFS inode
	268	* @dst: Destination list
	269	* @idx_start: lower bound of page->index to scan
	270	* @npages: idx_start + npages sets the upper bound to scan.
	271	*
	272	* Moves elements from one of the inode request lists.
	273	* If the number of requests is set to 0, the entire address_space
	274	* starting at index idx_start, is scanned.
	275	* The requests are not checked to ensure that they form a contiguous set.
	276	* You must be holding the inode's req_lock when calling this function
	277	*/
	278	int
	279	nfs_scan_lock_dirty(struct nfs_inode nfsi, struct list_head dst,
	280	unsigned long idx_start, unsigned int npages)
	281	{
	282	struct nfs_page *pgvec[NFS_SCAN_MAXENTRIES];
	283	struct nfs_page *req;
	284	unsigned long idx_end;
	285	int found, i;
	286	int res;
	287
	288	res = 0;
	289	if (npages == 0)
	290	idx_end = ~0;
	291	else
	292	idx_end = idx_start + npages - 1;
	293
	294	for (;;) {
	295	found = radix_tree_gang_lookup_tag(&nfsi->nfs_page_tree,
	296	(void **)&pgvec[0], idx_start, NFS_SCAN_MAXENTRIES,
	297	NFS_PAGE_TAG_DIRTY);
	298	if (found <= 0)
	299	break;
	300	for (i = 0; i < found; i++) {
	301	req = pgvec[i];
	302	if (req->wb_index > idx_end)
	303	goto out;
	304
	305	idx_start = req->wb_index + 1;
	306
	307	if (nfs_set_page_writeback_locked(req)) {
	308	radix_tree_tag_clear(&nfsi->nfs_page_tree,
	309	req->wb_index, NFS_PAGE_TAG_DIRTY);
	310	nfs_list_remove_request(req);
	311	nfs_list_add_request(req, dst);
	312	res++;
	313	}
	314	}
	315	}
	316	out:
	317	return res;
	318	}
	319
1da177e4 LT	320	/**
	321	* nfs_scan_list - Scan a list for matching requests
	322	* @head: One of the NFS inode request lists
	323	* @dst: Destination list
	324	* @idx_start: lower bound of page->index to scan
	325	* @npages: idx_start + npages sets the upper bound to scan.
	326	*
	327	* Moves elements from one of the inode request lists.
	328	* If the number of requests is set to 0, the entire address_space
	329	* starting at index idx_start, is scanned.
	330	* The requests are not checked to ensure that they form a contiguous set.
	331	* You must be holding the inode's req_lock when calling this function
	332	*/
	333	int
	334	nfs_scan_list(struct list_head head, struct list_head dst,
	335	unsigned long idx_start, unsigned int npages)
	336	{
	337	struct list_head pos, tmp;
	338	struct nfs_page *req;
	339	unsigned long idx_end;
	340	int res;
	341
	342	res = 0;
	343	if (npages == 0)
	344	idx_end = ~0;
	345	else
	346	idx_end = idx_start + npages - 1;
	347
	348	list_for_each_safe(pos, tmp, head) {
	349
	350	req = nfs_list_entry(pos);
	351
	352	if (req->wb_index < idx_start)
	353	continue;
	354	if (req->wb_index > idx_end)
	355	break;
	356
c6a556b8	357	if (!nfs_set_page_writeback_locked(req))
1da177e4 LT	358	continue;
	359	nfs_list_remove_request(req);
	360	nfs_list_add_request(req, dst);
	361	res++;
	362	}
	363	return res;
	364	}
	365
	366	int nfs_init_nfspagecache(void)
	367	{
	368	nfs_page_cachep = kmem_cache_create("nfs_page",
	369	sizeof(struct nfs_page),
	370	0, SLAB_HWCACHE_ALIGN,
	371	NULL, NULL);
	372	if (nfs_page_cachep == NULL)
	373	return -ENOMEM;
	374
	375	return 0;
	376	}
	377
	378	void nfs_destroy_nfspagecache(void)
	379	{
	380	if (kmem_cache_destroy(nfs_page_cachep))
	381	printk(KERN_INFO "nfs_page: not all structures were freed\n");
	382	}
	383