[deliverable/linux.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 <linux/sunrpc/clnt.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;
static struct kmem_cache	*drc_slab;

/*
 * 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;

	drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep),
					0, 0, NULL);
	if (!drc_slab)
		goto out_nomem;

	INIT_LIST_HEAD(&lru_head);
	i = CACHESIZE;
	while (i) {
		rp = kmem_cache_alloc(drc_slab, 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);
		kmem_cache_free(drc_slab, rp);
	}

	cache_disabled = 1;

	kfree (cache_hash);
	cache_hash = NULL;

	if (drc_slab) {
		kmem_cache_destroy(drc_slab);
		drc_slab = 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)
{
	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 type = rqstp->rq_cachetype;
	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) &&
		    rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) &&
		    rpc_get_port(svc_addr(rqstp)) == rpc_get_port((struct sockaddr *)&rp->c_addr)) {
			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;
	rpc_copy_addr((struct sockaddr *)&rp->c_addr, svc_addr(rqstp));
	rpc_set_port((struct sockaddr *)&rp->c_addr, rpc_get_port(svc_addr(rqstp)));
	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) {
			rp->c_state = RC_UNUSED;
			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	11	#include <linux/slab.h>
7b9e8522	12	#include <linux/sunrpc/clnt.h>
5a0e3ad6	13
9a74af21 BH	14	#include "nfsd.h"
9a74af21 BH	15	#include "cache.h"
1da177e4 LT	16
	17	/* Size of reply cache. Common values are:
	18	* 4.3BSD: 128
	19	* 4.4BSD: 256
	20	* Solaris2: 1024
	21	* DEC Unix: 512-4096
	22	*/
	23	#define CACHESIZE 1024
	24	#define HASHSIZE 64
1da177e4	25
fca4217c	26	static struct hlist_head * cache_hash;
1da177e4 LT	27	static struct list_head lru_head;
1da177e4 LT	28	static int cache_disabled = 1;
8a8bc40d	29	static struct kmem_cache *drc_slab;
1da177e4	30
fca4217c GB	31	/*
	32	* Calculate the hash index from an XID.
	33	*/
	34	static inline u32 request_hash(u32 xid)
	35	{
	36	u32 h = xid;
	37	h ^= (xid >> 24);
	38	return h & (HASHSIZE-1);
	39	}
	40
1da177e4 LT	41	static int nfsd_cache_append(struct svc_rqst rqstp, struct kvec vec);
1da177e4 LT	42
fca4217c	43	/*
1da177e4 LT	44	* locking for the reply cache:
	45	* A cache entry is "single use" if c_state == RC_INPROG
	46	* Otherwise, it when accessing _prev or _next, the lock must be held.
	47	*/
	48	static DEFINE_SPINLOCK(cache_lock);
	49
d5c3428b	50	int nfsd_reply_cache_init(void)
1da177e4 LT	51	{
	52	struct svc_cacherep *rp;
	53	int i;
	54
8a8bc40d JL	55	drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep),
	56	0, 0, NULL);
	57	if (!drc_slab)
	58	goto out_nomem;
	59
1da177e4 LT	60	INIT_LIST_HEAD(&lru_head);
1da177e4 LT	61	i = CACHESIZE;
d5c3428b	62	while (i) {
8a8bc40d	63	rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
d5c3428b BF	64	if (!rp)
d5c3428b BF	65	goto out_nomem;
1da177e4 LT	66	list_add(&rp->c_lru, &lru_head);
	67	rp->c_state = RC_UNUSED;
	68	rp->c_type = RC_NOCACHE;
	69	INIT_HLIST_NODE(&rp->c_hash);
	70	i--;
	71	}
	72
fca4217c GB	73	cache_hash = kcalloc (HASHSIZE, sizeof(struct hlist_head), GFP_KERNEL);
fca4217c GB	74	if (!cache_hash)
d5c3428b	75	goto out_nomem;
1da177e4 LT	76
1da177e4 LT	77	cache_disabled = 0;
d5c3428b BF	78	return 0;
	79	out_nomem:
	80	printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
	81	nfsd_reply_cache_shutdown();
	82	return -ENOMEM;
1da177e4 LT	83	}
1da177e4 LT	84
d5c3428b	85	void nfsd_reply_cache_shutdown(void)
1da177e4 LT	86	{
	87	struct svc_cacherep *rp;
	88
	89	while (!list_empty(&lru_head)) {
	90	rp = list_entry(lru_head.next, struct svc_cacherep, c_lru);
	91	if (rp->c_state == RC_DONE && rp->c_type == RC_REPLBUFF)
	92	kfree(rp->c_replvec.iov_base);
	93	list_del(&rp->c_lru);
8a8bc40d	94	kmem_cache_free(drc_slab, rp);
1da177e4 LT	95	}
	96
	97	cache_disabled = 1;
	98
fca4217c GB	99	kfree (cache_hash);
fca4217c GB	100	cache_hash = NULL;
8a8bc40d JL	101
	102	if (drc_slab) {
	103	kmem_cache_destroy(drc_slab);
	104	drc_slab = NULL;
	105	}
1da177e4 LT	106	}
	107
	108	/*
	109	* Move cache entry to end of LRU list
	110	*/
	111	static void
	112	lru_put_end(struct svc_cacherep *rp)
	113	{
f116629d	114	list_move_tail(&rp->c_lru, &lru_head);
1da177e4 LT	115	}
	116
	117	/*
	118	* Move a cache entry from one hash list to another
	119	*/
	120	static void
	121	hash_refile(struct svc_cacherep *rp)
	122	{
	123	hlist_del_init(&rp->c_hash);
fca4217c	124	hlist_add_head(&rp->c_hash, cache_hash + request_hash(rp->c_xid));
1da177e4 LT	125	}
	126
	127	/*
	128	* Try to find an entry matching the current call in the cache. When none
	129	* is found, we grab the oldest unlocked entry off the LRU list.
	130	* Note that no operation within the loop may sleep.
	131	*/
	132	int
1091006c	133	nfsd_cache_lookup(struct svc_rqst *rqstp)
1da177e4 LT	134	{
	135	struct hlist_node *hn;
	136	struct hlist_head *rh;
	137	struct svc_cacherep *rp;
c7afef1f AV	138	__be32 xid = rqstp->rq_xid;
c7afef1f AV	139	u32 proto = rqstp->rq_prot,
1da177e4 LT	140	vers = rqstp->rq_vers,
	141	proc = rqstp->rq_proc;
	142	unsigned long age;
1091006c	143	int type = rqstp->rq_cachetype;
1da177e4 LT	144	int rtn;
	145
	146	rqstp->rq_cacherep = NULL;
	147	if (cache_disabled \|\| type == RC_NOCACHE) {
	148	nfsdstats.rcnocache++;
	149	return RC_DOIT;
	150	}
	151
	152	spin_lock(&cache_lock);
	153	rtn = RC_DOIT;
	154
fca4217c	155	rh = &cache_hash[request_hash(xid)];
1da177e4 LT	156	hlist_for_each_entry(rp, hn, rh, c_hash) {
	157	if (rp->c_state != RC_UNUSED &&
	158	xid == rp->c_xid && proc == rp->c_proc &&
	159	proto == rp->c_prot && vers == rp->c_vers &&
	160	time_before(jiffies, rp->c_timestamp + 120*HZ) &&
7b9e8522 JL	161	rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) &&
7b9e8522 JL	162	rpc_get_port(svc_addr(rqstp)) == rpc_get_port((struct sockaddr *)&rp->c_addr)) {
1da177e4 LT	163	nfsdstats.rchits++;
	164	goto found_entry;
	165	}
	166	}
	167	nfsdstats.rcmisses++;
	168
	169	/* This loop shouldn't take more than a few iterations normally */
	170	{
	171	int safe = 0;
	172	list_for_each_entry(rp, &lru_head, c_lru) {
	173	if (rp->c_state != RC_INPROG)
	174	break;
	175	if (safe++ > CACHESIZE) {
	176	printk("nfsd: loop in repcache LRU list\n");
	177	cache_disabled = 1;
	178	goto out;
	179	}
	180	}
	181	}
	182
cf0a586c GB	183	/* All entries on the LRU are in-progress. This should not happen */
cf0a586c GB	184	if (&rp->c_lru == &lru_head) {
1da177e4 LT	185	static int complaints;
	186
	187	printk(KERN_WARNING "nfsd: all repcache entries locked!\n");
	188	if (++complaints > 5) {
	189	printk(KERN_WARNING "nfsd: disabling repcache.\n");
	190	cache_disabled = 1;
	191	}
	192	goto out;
	193	}
	194
	195	rqstp->rq_cacherep = rp;
	196	rp->c_state = RC_INPROG;
	197	rp->c_xid = xid;
	198	rp->c_proc = proc;
7b9e8522 JL	199	rpc_copy_addr((struct sockaddr *)&rp->c_addr, svc_addr(rqstp));
7b9e8522 JL	200	rpc_set_port((struct sockaddr *)&rp->c_addr, rpc_get_port(svc_addr(rqstp)));
1da177e4 LT	201	rp->c_prot = proto;
	202	rp->c_vers = vers;
	203	rp->c_timestamp = jiffies;
	204
	205	hash_refile(rp);
	206
	207	/* release any buffer */
	208	if (rp->c_type == RC_REPLBUFF) {
	209	kfree(rp->c_replvec.iov_base);
	210	rp->c_replvec.iov_base = NULL;
	211	}
	212	rp->c_type = RC_NOCACHE;
	213	out:
	214	spin_unlock(&cache_lock);
	215	return rtn;
	216
	217	found_entry:
	218	/* We found a matching entry which is either in progress or done. */
	219	age = jiffies - rp->c_timestamp;
	220	rp->c_timestamp = jiffies;
	221	lru_put_end(rp);
	222
	223	rtn = RC_DROPIT;
	224	/* Request being processed or excessive rexmits */
	225	if (rp->c_state == RC_INPROG \|\| age < RC_DELAY)
	226	goto out;
	227
	228	/* From the hall of fame of impractical attacks:
	229	* Is this a user who tries to snoop on the cache? */
	230	rtn = RC_DOIT;
	231	if (!rqstp->rq_secure && rp->c_secure)
	232	goto out;
	233
	234	/* Compose RPC reply header */
	235	switch (rp->c_type) {
	236	case RC_NOCACHE:
	237	break;
	238	case RC_REPLSTAT:
	239	svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
	240	rtn = RC_REPLY;
	241	break;
	242	case RC_REPLBUFF:
	243	if (!nfsd_cache_append(rqstp, &rp->c_replvec))
	244	goto out; /* should not happen */
	245	rtn = RC_REPLY;
	246	break;
	247	default:
	248	printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
	249	rp->c_state = RC_UNUSED;
	250	}
	251
	252	goto out;
	253	}
	254
	255	/*
	256	* Update a cache entry. This is called from nfsd_dispatch when
	257	* the procedure has been executed and the complete reply is in
	258	* rqstp->rq_res.
	259	*
	260	* We're copying around data here rather than swapping buffers because
	261	* the toplevel loop requires max-sized buffers, which would be a waste
	262	* of memory for a cache with a max reply size of 100 bytes (diropokres).
	263	*
	264	* If we should start to use different types of cache entries tailored
265	* specifically for attrstat and fh's, we may save even more space.
266	*
267	* Also note that a cachetype of RC_NOCACHE can legally be passed when
268	* nfsd failed to encode a reply that otherwise would have been cached.
269	* In this case, nfsd_cache_update is called with statp == NULL.
270	*/
271	void
c7afef1f	272	nfsd_cache_update(struct svc_rqst rqstp, int cachetype, __be32 statp)
1da177e4 LT	273	{
	274	struct svc_cacherep *rp;
	275	struct kvec resv = &rqstp->rq_res.head[0], cachv;
	276	int len;
	277
	278	if (!(rp = rqstp->rq_cacherep) \|\| cache_disabled)
	279	return;
	280
	281	len = resv->iov_len - ((char)statp - (char)resv->iov_base);
	282	len >>= 2;
fca4217c	283
1da177e4 LT	284	/* Don't cache excessive amounts of data and XDR failures */
	285	if (!statp \|\| len > (256 >> 2)) {
	286	rp->c_state = RC_UNUSED;
	287	return;
	288	}
	289
	290	switch (cachetype) {
	291	case RC_REPLSTAT:
	292	if (len != 1)
	293	printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
	294	rp->c_replstat = *statp;
	295	break;
	296	case RC_REPLBUFF:
	297	cachv = &rp->c_replvec;
	298	cachv->iov_base = kmalloc(len << 2, GFP_KERNEL);
	299	if (!cachv->iov_base) {
1da177e4	300	rp->c_state = RC_UNUSED;
1da177e4 LT	301	return;
	302	}
	303	cachv->iov_len = len << 2;
	304	memcpy(cachv->iov_base, statp, len << 2);
	305	break;
	306	}
	307	spin_lock(&cache_lock);
	308	lru_put_end(rp);
	309	rp->c_secure = rqstp->rq_secure;
	310	rp->c_type = cachetype;
	311	rp->c_state = RC_DONE;
	312	rp->c_timestamp = jiffies;
	313	spin_unlock(&cache_lock);
	314	return;
	315	}
	316
	317	/*
	318	* Copy cached reply to current reply buffer. Should always fit.
	319	* FIXME as reply is in a page, we should just attach the page, and
	320	* keep a refcount....
	321	*/
	322	static int
	323	nfsd_cache_append(struct svc_rqst rqstp, struct kvec data)
	324	{
	325	struct kvec *vec = &rqstp->rq_res.head[0];
	326
	327	if (vec->iov_len + data->iov_len > PAGE_SIZE) {
	328	printk(KERN_WARNING "nfsd: cached reply too large (%Zd).\n",
	329	data->iov_len);
	330	return 0;
	331	}
	332	memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len);
	333	vec->iov_len += data->iov_len;
	334	return 1;
	335	}