[mirror_ubuntu-hirsute-kernel.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;

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