[net-next-2.6.git] / fs / nfsd / nfscache.c

/*
 * Request reply cache. This is currently a global cache, but this may
 * change in the future and be a per-client cache.
 *
 * This code is heavily inspired by the 44BSD implementation, although
 * it does things a bit differently.
 *
 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
 */

#include <linux/slab.h>

#include "nfsd.h"
#include "cache.h"

/* Size of reply cache. Common values are:
 * 4.3BSD:	128
 * 4.4BSD:	256
 * Solaris2:	1024
 * DEC Unix:	512-4096
 */
#define CACHESIZE		1024
#define HASHSIZE		64

static struct hlist_head *	cache_hash;
static struct list_head 	lru_head;
static int			cache_disabled = 1;

/*
 * Calculate the hash index from an XID.
 */
static inline u32 request_hash(u32 xid)
{
	u32 h = xid;
	h ^= (xid >> 24);
	return h & (HASHSIZE-1);
}

static int	nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec);

/*
 * locking for the reply cache:
 * A cache entry is "single use" if c_state == RC_INPROG
 * Otherwise, it when accessing _prev or _next, the lock must be held.
 */
static DEFINE_SPINLOCK(cache_lock);

int nfsd_reply_cache_init(void)
{
	struct svc_cacherep	*rp;
	int			i;

	INIT_LIST_HEAD(&lru_head);
	i = CACHESIZE;
	while (i) {
		rp = kmalloc(sizeof(*rp), GFP_KERNEL);
		if (!rp)
			goto out_nomem;
		list_add(&rp->c_lru, &lru_head);
		rp->c_state = RC_UNUSED;
		rp->c_type = RC_NOCACHE;
		INIT_HLIST_NODE(&rp->c_hash);
		i--;
	}

	cache_hash = kcalloc (HASHSIZE, sizeof(struct hlist_head), GFP_KERNEL);
	if (!cache_hash)
		goto out_nomem;

	cache_disabled = 0;
	return 0;
out_nomem:
	printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
	nfsd_reply_cache_shutdown();
	return -ENOMEM;
}

void nfsd_reply_cache_shutdown(void)
{
	struct svc_cacherep	*rp;

	while (!list_empty(&lru_head)) {
		rp = list_entry(lru_head.next, struct svc_cacherep, c_lru);
		if (rp->c_state == RC_DONE && rp->c_type == RC_REPLBUFF)
			kfree(rp->c_replvec.iov_base);
		list_del(&rp->c_lru);
		kfree(rp);
	}

	cache_disabled = 1;

	kfree (cache_hash);
	cache_hash = NULL;
}

/*
 * Move cache entry to end of LRU list
 */
static void
lru_put_end(struct svc_cacherep *rp)
{
	list_move_tail(&rp->c_lru, &lru_head);
}

/*
 * Move a cache entry from one hash list to another
 */
static void
hash_refile(struct svc_cacherep *rp)
{
	hlist_del_init(&rp->c_hash);
	hlist_add_head(&rp->c_hash, cache_hash + request_hash(rp->c_xid));
}

/*
 * Try to find an entry matching the current call in the cache. When none
 * is found, we grab the oldest unlocked entry off the LRU list.
 * Note that no operation within the loop may sleep.
 */
int
nfsd_cache_lookup(struct svc_rqst *rqstp, int type)
{
	struct hlist_node	*hn;
	struct hlist_head 	*rh;
	struct svc_cacherep	*rp;
	__be32			xid = rqstp->rq_xid;
	u32			proto =  rqstp->rq_prot,
				vers = rqstp->rq_vers,
				proc = rqstp->rq_proc;
	unsigned long		age;
	int rtn;

	rqstp->rq_cacherep = NULL;
	if (cache_disabled || type == RC_NOCACHE) {
		nfsdstats.rcnocache++;
		return RC_DOIT;
	}

	spin_lock(&cache_lock);
	rtn = RC_DOIT;

	rh = &cache_hash[request_hash(xid)];
	hlist_for_each_entry(rp, hn, rh, c_hash) {
		if (rp->c_state != RC_UNUSED &&
		    xid == rp->c_xid && proc == rp->c_proc &&
		    proto == rp->c_prot && vers == rp->c_vers &&
		    time_before(jiffies, rp->c_timestamp + 120*HZ) &&
		    memcmp((char*)&rqstp->rq_addr, (char*)&rp->c_addr, sizeof(rp->c_addr))==0) {
			nfsdstats.rchits++;
			goto found_entry;
		}
	}
	nfsdstats.rcmisses++;

	/* This loop shouldn't take more than a few iterations normally */
	{
	int	safe = 0;
	list_for_each_entry(rp, &lru_head, c_lru) {
		if (rp->c_state != RC_INPROG)
			break;
		if (safe++ > CACHESIZE) {
			printk("nfsd: loop in repcache LRU list\n");
			cache_disabled = 1;
			goto out;
		}
	}
	}

	/* All entries on the LRU are in-progress. This should not happen */
	if (&rp->c_lru == &lru_head) {
		static int	complaints;

		printk(KERN_WARNING "nfsd: all repcache entries locked!\n");
		if (++complaints > 5) {
			printk(KERN_WARNING "nfsd: disabling repcache.\n");
			cache_disabled = 1;
		}
		goto out;
	}

	rqstp->rq_cacherep = rp;
	rp->c_state = RC_INPROG;
	rp->c_xid = xid;
	rp->c_proc = proc;
	memcpy(&rp->c_addr, svc_addr_in(rqstp), sizeof(rp->c_addr));
	rp->c_prot = proto;
	rp->c_vers = vers;
	rp->c_timestamp = jiffies;

	hash_refile(rp);

	/* release any buffer */
	if (rp->c_type == RC_REPLBUFF) {
		kfree(rp->c_replvec.iov_base);
		rp->c_replvec.iov_base = NULL;
	}
	rp->c_type = RC_NOCACHE;
 out:
	spin_unlock(&cache_lock);
	return rtn;

found_entry:
	/* We found a matching entry which is either in progress or done. */
	age = jiffies - rp->c_timestamp;
	rp->c_timestamp = jiffies;
	lru_put_end(rp);

	rtn = RC_DROPIT;
	/* Request being processed or excessive rexmits */
	if (rp->c_state == RC_INPROG || age < RC_DELAY)
		goto out;

	/* From the hall of fame of impractical attacks:
	 * Is this a user who tries to snoop on the cache? */
	rtn = RC_DOIT;
	if (!rqstp->rq_secure && rp->c_secure)
		goto out;

	/* Compose RPC reply header */
	switch (rp->c_type) {
	case RC_NOCACHE:
		break;
	case RC_REPLSTAT:
		svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
		rtn = RC_REPLY;
		break;
	case RC_REPLBUFF:
		if (!nfsd_cache_append(rqstp, &rp->c_replvec))
			goto out;	/* should not happen */
		rtn = RC_REPLY;
		break;
	default:
		printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
		rp->c_state = RC_UNUSED;
	}

	goto out;
}

/*
 * Update a cache entry. This is called from nfsd_dispatch when
 * the procedure has been executed and the complete reply is in
 * rqstp->rq_res.
 *
 * We're copying around data here rather than swapping buffers because
 * the toplevel loop requires max-sized buffers, which would be a waste
 * of memory for a cache with a max reply size of 100 bytes (diropokres).
 *
 * If we should start to use different types of cache entries tailored
 * specifically for attrstat and fh's, we may save even more space.
 *
 * Also note that a cachetype of RC_NOCACHE can legally be passed when
 * nfsd failed to encode a reply that otherwise would have been cached.
 * In this case, nfsd_cache_update is called with statp == NULL.
 */
void
nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
{
	struct svc_cacherep *rp;
	struct kvec	*resv = &rqstp->rq_res.head[0], *cachv;
	int		len;

	if (!(rp = rqstp->rq_cacherep) || cache_disabled)
		return;

	len = resv->iov_len - ((char*)statp - (char*)resv->iov_base);
	len >>= 2;

	/* Don't cache excessive amounts of data and XDR failures */
	if (!statp || len > (256 >> 2)) {
		rp->c_state = RC_UNUSED;
		return;
	}

	switch (cachetype) {
	case RC_REPLSTAT:
		if (len != 1)
			printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
		rp->c_replstat = *statp;
		break;
	case RC_REPLBUFF:
		cachv = &rp->c_replvec;
		cachv->iov_base = kmalloc(len << 2, GFP_KERNEL);
		if (!cachv->iov_base) {
			spin_lock(&cache_lock);
			rp->c_state = RC_UNUSED;
			spin_unlock(&cache_lock);
			return;
		}
		cachv->iov_len = len << 2;
		memcpy(cachv->iov_base, statp, len << 2);
		break;
	}
	spin_lock(&cache_lock);
	lru_put_end(rp);
	rp->c_secure = rqstp->rq_secure;
	rp->c_type = cachetype;
	rp->c_state = RC_DONE;
	rp->c_timestamp = jiffies;
	spin_unlock(&cache_lock);
	return;
}

/*
 * Copy cached reply to current reply buffer. Should always fit.
 * FIXME as reply is in a page, we should just attach the page, and
 * keep a refcount....
 */
static int
nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data)
{
	struct kvec	*vec = &rqstp->rq_res.head[0];

	if (vec->iov_len + data->iov_len > PAGE_SIZE) {
		printk(KERN_WARNING "nfsd: cached reply too large (%Zd).\n",
				data->iov_len);
		return 0;
	}
	memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len);
	vec->iov_len += data->iov_len;
	return 1;
}
Commit	Line	Data
1da177e4	1	/*
1da177e4 LT	2	* Request reply cache. This is currently a global cache, but this may
	3	* change in the future and be a per-client cache.
	4	*
	5	* This code is heavily inspired by the 44BSD implementation, although
	6	* it does things a bit differently.
	7	*
	8	* Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
	9	*/
	10
5a0e3ad6 TH	11	#include <linux/slab.h>
5a0e3ad6 TH	12
9a74af21 BH	13	#include "nfsd.h"
9a74af21 BH	14	#include "cache.h"
1da177e4 LT	15
	16	/* Size of reply cache. Common values are:
	17	* 4.3BSD: 128
	18	* 4.4BSD: 256
	19	* Solaris2: 1024
	20	* DEC Unix: 512-4096
	21	*/
	22	#define CACHESIZE 1024
	23	#define HASHSIZE 64
1da177e4	24
fca4217c	25	static struct hlist_head * cache_hash;
1da177e4 LT	26	static struct list_head lru_head;
	27	static int cache_disabled = 1;
	28
fca4217c GB	29	/*
	30	* Calculate the hash index from an XID.
	31	*/
	32	static inline u32 request_hash(u32 xid)
	33	{
	34	u32 h = xid;
	35	h ^= (xid >> 24);
	36	return h & (HASHSIZE-1);
	37	}
	38
1da177e4 LT	39	static int nfsd_cache_append(struct svc_rqst rqstp, struct kvec vec);
1da177e4 LT	40
fca4217c	41	/*
1da177e4 LT	42	* locking for the reply cache:
	43	* A cache entry is "single use" if c_state == RC_INPROG
	44	* Otherwise, it when accessing _prev or _next, the lock must be held.
	45	*/
	46	static DEFINE_SPINLOCK(cache_lock);
	47
d5c3428b	48	int nfsd_reply_cache_init(void)
1da177e4 LT	49	{
	50	struct svc_cacherep *rp;
	51	int i;
	52
	53	INIT_LIST_HEAD(&lru_head);
	54	i = CACHESIZE;
d5c3428b	55	while (i) {
1da177e4	56	rp = kmalloc(sizeof(*rp), GFP_KERNEL);
d5c3428b BF	57	if (!rp)
d5c3428b BF	58	goto out_nomem;
1da177e4 LT	59	list_add(&rp->c_lru, &lru_head);
	60	rp->c_state = RC_UNUSED;
	61	rp->c_type = RC_NOCACHE;
	62	INIT_HLIST_NODE(&rp->c_hash);
	63	i--;
	64	}
	65
fca4217c GB	66	cache_hash = kcalloc (HASHSIZE, sizeof(struct hlist_head), GFP_KERNEL);
fca4217c GB	67	if (!cache_hash)
d5c3428b	68	goto out_nomem;
1da177e4 LT	69
1da177e4 LT	70	cache_disabled = 0;
d5c3428b BF	71	return 0;
	72	out_nomem:
	73	printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
	74	nfsd_reply_cache_shutdown();
	75	return -ENOMEM;
1da177e4 LT	76	}
1da177e4 LT	77
d5c3428b	78	void nfsd_reply_cache_shutdown(void)
1da177e4 LT	79	{
	80	struct svc_cacherep *rp;
	81
	82	while (!list_empty(&lru_head)) {
	83	rp = list_entry(lru_head.next, struct svc_cacherep, c_lru);
	84	if (rp->c_state == RC_DONE && rp->c_type == RC_REPLBUFF)
	85	kfree(rp->c_replvec.iov_base);
	86	list_del(&rp->c_lru);
	87	kfree(rp);
	88	}
	89
	90	cache_disabled = 1;
	91
fca4217c GB	92	kfree (cache_hash);
fca4217c GB	93	cache_hash = NULL;
1da177e4 LT	94	}
	95
	96	/*
	97	* Move cache entry to end of LRU list
	98	*/
	99	static void
	100	lru_put_end(struct svc_cacherep *rp)
	101	{
f116629d	102	list_move_tail(&rp->c_lru, &lru_head);
1da177e4 LT	103	}
	104
	105	/*
	106	* Move a cache entry from one hash list to another
	107	*/
	108	static void
	109	hash_refile(struct svc_cacherep *rp)
	110	{
	111	hlist_del_init(&rp->c_hash);
fca4217c	112	hlist_add_head(&rp->c_hash, cache_hash + request_hash(rp->c_xid));
1da177e4 LT	113	}
	114
	115	/*
	116	* Try to find an entry matching the current call in the cache. When none
	117	* is found, we grab the oldest unlocked entry off the LRU list.
	118	* Note that no operation within the loop may sleep.
	119	*/
	120	int
	121	nfsd_cache_lookup(struct svc_rqst *rqstp, int type)
	122	{
	123	struct hlist_node *hn;
	124	struct hlist_head *rh;
	125	struct svc_cacherep *rp;
c7afef1f AV	126	__be32 xid = rqstp->rq_xid;
c7afef1f AV	127	u32 proto = rqstp->rq_prot,
1da177e4 LT	128	vers = rqstp->rq_vers,
	129	proc = rqstp->rq_proc;
	130	unsigned long age;
	131	int rtn;
	132
	133	rqstp->rq_cacherep = NULL;
	134	if (cache_disabled \|\| type == RC_NOCACHE) {
	135	nfsdstats.rcnocache++;
	136	return RC_DOIT;
	137	}
	138
	139	spin_lock(&cache_lock);
	140	rtn = RC_DOIT;
	141
fca4217c	142	rh = &cache_hash[request_hash(xid)];
1da177e4 LT	143	hlist_for_each_entry(rp, hn, rh, c_hash) {
	144	if (rp->c_state != RC_UNUSED &&
	145	xid == rp->c_xid && proc == rp->c_proc &&
	146	proto == rp->c_prot && vers == rp->c_vers &&
	147	time_before(jiffies, rp->c_timestamp + 120*HZ) &&
	148	memcmp((char)&rqstp->rq_addr, (char)&rp->c_addr, sizeof(rp->c_addr))==0) {
	149	nfsdstats.rchits++;
	150	goto found_entry;
	151	}
	152	}
	153	nfsdstats.rcmisses++;
	154
	155	/* This loop shouldn't take more than a few iterations normally */
	156	{
	157	int safe = 0;
	158	list_for_each_entry(rp, &lru_head, c_lru) {
	159	if (rp->c_state != RC_INPROG)
	160	break;
	161	if (safe++ > CACHESIZE) {
	162	printk("nfsd: loop in repcache LRU list\n");
	163	cache_disabled = 1;
	164	goto out;
	165	}
	166	}
	167	}
	168
cf0a586c GB	169	/* All entries on the LRU are in-progress. This should not happen */
cf0a586c GB	170	if (&rp->c_lru == &lru_head) {
1da177e4 LT	171	static int complaints;
	172
	173	printk(KERN_WARNING "nfsd: all repcache entries locked!\n");
	174	if (++complaints > 5) {
	175	printk(KERN_WARNING "nfsd: disabling repcache.\n");
	176	cache_disabled = 1;
	177	}
	178	goto out;
	179	}
	180
	181	rqstp->rq_cacherep = rp;
	182	rp->c_state = RC_INPROG;
	183	rp->c_xid = xid;
	184	rp->c_proc = proc;
27459f09	185	memcpy(&rp->c_addr, svc_addr_in(rqstp), sizeof(rp->c_addr));
1da177e4 LT	186	rp->c_prot = proto;
	187	rp->c_vers = vers;
	188	rp->c_timestamp = jiffies;
	189
	190	hash_refile(rp);
	191
	192	/* release any buffer */
	193	if (rp->c_type == RC_REPLBUFF) {
	194	kfree(rp->c_replvec.iov_base);
	195	rp->c_replvec.iov_base = NULL;
	196	}
	197	rp->c_type = RC_NOCACHE;
	198	out:
	199	spin_unlock(&cache_lock);
	200	return rtn;
	201
	202	found_entry:
	203	/* We found a matching entry which is either in progress or done. */
	204	age = jiffies - rp->c_timestamp;
	205	rp->c_timestamp = jiffies;
	206	lru_put_end(rp);
	207
	208	rtn = RC_DROPIT;
	209	/* Request being processed or excessive rexmits */
	210	if (rp->c_state == RC_INPROG \|\| age < RC_DELAY)
	211	goto out;
	212
	213	/* From the hall of fame of impractical attacks:
	214	* Is this a user who tries to snoop on the cache? */
	215	rtn = RC_DOIT;
	216	if (!rqstp->rq_secure && rp->c_secure)
	217	goto out;
	218
	219	/* Compose RPC reply header */
	220	switch (rp->c_type) {
	221	case RC_NOCACHE:
	222	break;
	223	case RC_REPLSTAT:
	224	svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
	225	rtn = RC_REPLY;
	226	break;
	227	case RC_REPLBUFF:
	228	if (!nfsd_cache_append(rqstp, &rp->c_replvec))
	229	goto out; /* should not happen */
	230	rtn = RC_REPLY;
	231	break;
	232	default:
	233	printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
	234	rp->c_state = RC_UNUSED;
	235	}
	236
	237	goto out;
	238	}
	239
	240	/*
	241	* Update a cache entry. This is called from nfsd_dispatch when
	242	* the procedure has been executed and the complete reply is in
	243	* rqstp->rq_res.
	244	*
	245	* We're copying around data here rather than swapping buffers because
	246	* the toplevel loop requires max-sized buffers, which would be a waste
	247	* of memory for a cache with a max reply size of 100 bytes (diropokres).
	248	*
	249	* If we should start to use different types of cache entries tailored
250	* specifically for attrstat and fh's, we may save even more space.
251	*
252	* Also note that a cachetype of RC_NOCACHE can legally be passed when
253	* nfsd failed to encode a reply that otherwise would have been cached.
254	* In this case, nfsd_cache_update is called with statp == NULL.
255	*/
256	void
c7afef1f	257	nfsd_cache_update(struct svc_rqst rqstp, int cachetype, __be32 statp)
1da177e4 LT	258	{
	259	struct svc_cacherep *rp;
	260	struct kvec resv = &rqstp->rq_res.head[0], cachv;
	261	int len;
	262
	263	if (!(rp = rqstp->rq_cacherep) \|\| cache_disabled)
	264	return;
	265
	266	len = resv->iov_len - ((char)statp - (char)resv->iov_base);
	267	len >>= 2;
fca4217c	268
1da177e4 LT	269	/* Don't cache excessive amounts of data and XDR failures */
	270	if (!statp \|\| len > (256 >> 2)) {
	271	rp->c_state = RC_UNUSED;
	272	return;
	273	}
	274
	275	switch (cachetype) {
	276	case RC_REPLSTAT:
	277	if (len != 1)
	278	printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
	279	rp->c_replstat = *statp;
	280	break;
	281	case RC_REPLBUFF:
	282	cachv = &rp->c_replvec;
	283	cachv->iov_base = kmalloc(len << 2, GFP_KERNEL);
	284	if (!cachv->iov_base) {
	285	spin_lock(&cache_lock);
	286	rp->c_state = RC_UNUSED;
	287	spin_unlock(&cache_lock);
	288	return;
	289	}
	290	cachv->iov_len = len << 2;
	291	memcpy(cachv->iov_base, statp, len << 2);
	292	break;
	293	}
	294	spin_lock(&cache_lock);
	295	lru_put_end(rp);
	296	rp->c_secure = rqstp->rq_secure;
	297	rp->c_type = cachetype;
	298	rp->c_state = RC_DONE;
	299	rp->c_timestamp = jiffies;
	300	spin_unlock(&cache_lock);
	301	return;
	302	}
	303
	304	/*
	305	* Copy cached reply to current reply buffer. Should always fit.
	306	* FIXME as reply is in a page, we should just attach the page, and
	307	* keep a refcount....
	308	*/
	309	static int
	310	nfsd_cache_append(struct svc_rqst rqstp, struct kvec data)
	311	{
	312	struct kvec *vec = &rqstp->rq_res.head[0];
	313
	314	if (vec->iov_len + data->iov_len > PAGE_SIZE) {
	315	printk(KERN_WARNING "nfsd: cached reply too large (%Zd).\n",
	316	data->iov_len);
	317	return 0;
	318	}
	319	memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len);
	320	vec->iov_len += data->iov_len;
	321	return 1;
	322	}