[linux.git] / fs / nfsd / nfscache.c

/*
 * linux/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 <[email protected]>
 */

#include <linux/kernel.h>
#include <linux/time.h>
#include <linux/slab.h>
#include <linux/string.h>
#include <linux/spinlock.h>
#include <linux/list.h>

#include <linux/sunrpc/svc.h>
#include <linux/nfsd/nfsd.h>
#include <linux/nfsd/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
#define REQHASH(xid)		((((xid) >> 24) ^ (xid)) & (HASHSIZE-1))

static struct hlist_head *	hash_list;
static struct list_head 	lru_head;
static int			cache_disabled = 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);

void
nfsd_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) break;
		list_add(&rp->c_lru, &lru_head);
		rp->c_state = RC_UNUSED;
		rp->c_type = RC_NOCACHE;
		INIT_HLIST_NODE(&rp->c_hash);
		i--;
	}

	if (i)
		printk (KERN_ERR "nfsd: cannot allocate all %d cache entries, only got %d\n",
			CACHESIZE, CACHESIZE-i);

	hash_list = kmalloc (HASHSIZE * sizeof(struct hlist_head), GFP_KERNEL);
	if (!hash_list) {
		nfsd_cache_shutdown();
		printk (KERN_ERR "nfsd: cannot allocate %Zd bytes for hash list\n",
			HASHSIZE * sizeof(struct hlist_head));
		return;
	}
	memset(hash_list, 0, HASHSIZE * sizeof(struct hlist_head));

	cache_disabled = 0;
}

void
nfsd_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 (hash_list);
	hash_list = NULL;
}

/*
 * Move cache entry to end of LRU list
 */
static void
lru_put_end(struct svc_cacherep *rp)
{
	list_del(&rp->c_lru);
	list_add_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, hash_list + REQHASH(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;
	u32			xid = rqstp->rq_xid,
				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 = &hash_list[REQHASH(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;
		}
	}
	}

	/* This should not happen */
	if (rp == NULL) {
		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;
	rp->c_addr = rqstp->rq_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, u32 *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 LT	1	/*
	2	* linux/fs/nfsd/nfscache.c
	3	*
	4	* Request reply cache. This is currently a global cache, but this may
	5	* change in the future and be a per-client cache.
	6	*
	7	* This code is heavily inspired by the 44BSD implementation, although
	8	* it does things a bit differently.
	9	*
	10	* Copyright (C) 1995, 1996 Olaf Kirch <[email protected]>
	11	*/
	12
	13	#include <linux/kernel.h>
	14	#include <linux/time.h>
	15	#include <linux/slab.h>
	16	#include <linux/string.h>
	17	#include <linux/spinlock.h>
	18	#include <linux/list.h>
	19
	20	#include <linux/sunrpc/svc.h>
	21	#include <linux/nfsd/nfsd.h>
	22	#include <linux/nfsd/cache.h>
	23
	24	/* Size of reply cache. Common values are:
	25	* 4.3BSD: 128
	26	* 4.4BSD: 256
	27	* Solaris2: 1024
	28	* DEC Unix: 512-4096
	29	*/
	30	#define CACHESIZE 1024
	31	#define HASHSIZE 64
	32	#define REQHASH(xid) ((((xid) >> 24) ^ (xid)) & (HASHSIZE-1))
	33
	34	static struct hlist_head * hash_list;
	35	static struct list_head lru_head;
	36	static int cache_disabled = 1;
	37
	38	static int nfsd_cache_append(struct svc_rqst rqstp, struct kvec vec);
	39
	40	/*
	41	* locking for the reply cache:
	42	* A cache entry is "single use" if c_state == RC_INPROG
	43	* Otherwise, it when accessing _prev or _next, the lock must be held.
	44	*/
	45	static DEFINE_SPINLOCK(cache_lock);
	46
	47	void
	48	nfsd_cache_init(void)
	49	{
	50	struct svc_cacherep *rp;
	51	int i;
	52
	53	INIT_LIST_HEAD(&lru_head);
	54	i = CACHESIZE;
	55	while(i) {
	56	rp = kmalloc(sizeof(*rp), GFP_KERNEL);
	57	if (!rp) break;
	58	list_add(&rp->c_lru, &lru_head);
	59	rp->c_state = RC_UNUSED;
	60	rp->c_type = RC_NOCACHE;
	61	INIT_HLIST_NODE(&rp->c_hash);
	62	i--;
	63	}
	64
65	if (i)
66	printk (KERN_ERR "nfsd: cannot allocate all %d cache entries, only got %d\n",
67	CACHESIZE, CACHESIZE-i);
68
69	hash_list = kmalloc (HASHSIZE * sizeof(struct hlist_head), GFP_KERNEL);
70	if (!hash_list) {
71	nfsd_cache_shutdown();
72	printk (KERN_ERR "nfsd: cannot allocate %Zd bytes for hash list\n",
73	HASHSIZE * sizeof(struct hlist_head));
74	return;
75	}
76	memset(hash_list, 0, HASHSIZE * sizeof(struct hlist_head));
77
78	cache_disabled = 0;
79	}
80
81	void
82	nfsd_cache_shutdown(void)
83	{
84	struct svc_cacherep *rp;
85
86	while (!list_empty(&lru_head)) {
87	rp = list_entry(lru_head.next, struct svc_cacherep, c_lru);
88	if (rp->c_state == RC_DONE && rp->c_type == RC_REPLBUFF)
89	kfree(rp->c_replvec.iov_base);
90	list_del(&rp->c_lru);
91	kfree(rp);
92	}
93
94	cache_disabled = 1;
95
f99d49ad	96	kfree (hash_list);
1da177e4 LT	97	hash_list = NULL;
	98	}
	99
	100	/*
	101	* Move cache entry to end of LRU list
	102	*/
	103	static void
	104	lru_put_end(struct svc_cacherep *rp)
	105	{
	106	list_del(&rp->c_lru);
	107	list_add_tail(&rp->c_lru, &lru_head);
	108	}
	109
	110	/*
	111	* Move a cache entry from one hash list to another
	112	*/
	113	static void
	114	hash_refile(struct svc_cacherep *rp)
	115	{
	116	hlist_del_init(&rp->c_hash);
	117	hlist_add_head(&rp->c_hash, hash_list + REQHASH(rp->c_xid));
	118	}
	119
	120	/*
	121	* Try to find an entry matching the current call in the cache. When none
	122	* is found, we grab the oldest unlocked entry off the LRU list.
	123	* Note that no operation within the loop may sleep.
	124	*/
	125	int
	126	nfsd_cache_lookup(struct svc_rqst *rqstp, int type)
	127	{
	128	struct hlist_node *hn;
	129	struct hlist_head *rh;
	130	struct svc_cacherep *rp;
	131	u32 xid = rqstp->rq_xid,
	132	proto = rqstp->rq_prot,
	133	vers = rqstp->rq_vers,
	134	proc = rqstp->rq_proc;
	135	unsigned long age;
	136	int rtn;
	137
	138	rqstp->rq_cacherep = NULL;
	139	if (cache_disabled \|\| type == RC_NOCACHE) {
	140	nfsdstats.rcnocache++;
	141	return RC_DOIT;
	142	}
	143
	144	spin_lock(&cache_lock);
	145	rtn = RC_DOIT;
	146
	147	rh = &hash_list[REQHASH(xid)];
	148	hlist_for_each_entry(rp, hn, rh, c_hash) {
	149	if (rp->c_state != RC_UNUSED &&
	150	xid == rp->c_xid && proc == rp->c_proc &&
	151	proto == rp->c_prot && vers == rp->c_vers &&
	152	time_before(jiffies, rp->c_timestamp + 120*HZ) &&
	153	memcmp((char)&rqstp->rq_addr, (char)&rp->c_addr, sizeof(rp->c_addr))==0) {
	154	nfsdstats.rchits++;
	155	goto found_entry;
	156	}
	157	}
	158	nfsdstats.rcmisses++;
	159
	160	/* This loop shouldn't take more than a few iterations normally */
161	{
162	int safe = 0;
163	list_for_each_entry(rp, &lru_head, c_lru) {
164	if (rp->c_state != RC_INPROG)
165	break;
166	if (safe++ > CACHESIZE) {
167	printk("nfsd: loop in repcache LRU list\n");
168	cache_disabled = 1;
169	goto out;
170	}
171	}
172	}
173
174	/* This should not happen */
175	if (rp == NULL) {
176	static int complaints;
177
178	printk(KERN_WARNING "nfsd: all repcache entries locked!\n");
179	if (++complaints > 5) {
180	printk(KERN_WARNING "nfsd: disabling repcache.\n");
181	cache_disabled = 1;
182	}
183	goto out;
184	}
185
186	rqstp->rq_cacherep = rp;
187	rp->c_state = RC_INPROG;
188	rp->c_xid = xid;
189	rp->c_proc = proc;
190	rp->c_addr = rqstp->rq_addr;
191	rp->c_prot = proto;
192	rp->c_vers = vers;
193	rp->c_timestamp = jiffies;
194
195	hash_refile(rp);
196
197	/* release any buffer */
198	if (rp->c_type == RC_REPLBUFF) {
199	kfree(rp->c_replvec.iov_base);
200	rp->c_replvec.iov_base = NULL;
201	}
202	rp->c_type = RC_NOCACHE;
203	out:
204	spin_unlock(&cache_lock);
205	return rtn;
206
207	found_entry:
208	/* We found a matching entry which is either in progress or done. */
209	age = jiffies - rp->c_timestamp;
210	rp->c_timestamp = jiffies;
211	lru_put_end(rp);
212
213	rtn = RC_DROPIT;
214	/* Request being processed or excessive rexmits */
215	if (rp->c_state == RC_INPROG \|\| age < RC_DELAY)
216	goto out;
217
218	/* From the hall of fame of impractical attacks:
219	* Is this a user who tries to snoop on the cache? */
220	rtn = RC_DOIT;
221	if (!rqstp->rq_secure && rp->c_secure)
222	goto out;
223
224	/* Compose RPC reply header */
225	switch (rp->c_type) {
226	case RC_NOCACHE:
227	break;
228	case RC_REPLSTAT:
229	svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
230	rtn = RC_REPLY;
231	break;
232	case RC_REPLBUFF:
233	if (!nfsd_cache_append(rqstp, &rp->c_replvec))
234	goto out; /* should not happen */
235	rtn = RC_REPLY;
236	break;
237	default:
238	printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
239	rp->c_state = RC_UNUSED;
240	}
241
242	goto out;
243	}
244
245	/*
246	* Update a cache entry. This is called from nfsd_dispatch when
247	* the procedure has been executed and the complete reply is in
248	* rqstp->rq_res.
249	*
250	* We're copying around data here rather than swapping buffers because
251	* the toplevel loop requires max-sized buffers, which would be a waste
252	* of memory for a cache with a max reply size of 100 bytes (diropokres).
253	*
254	* If we should start to use different types of cache entries tailored
255	* specifically for attrstat and fh's, we may save even more space.
256	*
257	* Also note that a cachetype of RC_NOCACHE can legally be passed when
258	* nfsd failed to encode a reply that otherwise would have been cached.
259	* In this case, nfsd_cache_update is called with statp == NULL.
260	*/
261	void
262	nfsd_cache_update(struct svc_rqst rqstp, int cachetype, u32 statp)
263	{
264	struct svc_cacherep *rp;
265	struct kvec resv = &rqstp->rq_res.head[0], cachv;
266	int len;
267
268	if (!(rp = rqstp->rq_cacherep) \|\| cache_disabled)
269	return;
270
271	len = resv->iov_len - ((char)statp - (char)resv->iov_base);
272	len >>= 2;
273
274	/* Don't cache excessive amounts of data and XDR failures */
275	if (!statp \|\| len > (256 >> 2)) {
276	rp->c_state = RC_UNUSED;
277	return;
278	}
279
280	switch (cachetype) {
281	case RC_REPLSTAT:
282	if (len != 1)
283	printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
284	rp->c_replstat = *statp;
285	break;
286	case RC_REPLBUFF:
287	cachv = &rp->c_replvec;
288	cachv->iov_base = kmalloc(len << 2, GFP_KERNEL);
289	if (!cachv->iov_base) {
290	spin_lock(&cache_lock);
291	rp->c_state = RC_UNUSED;
292	spin_unlock(&cache_lock);
293	return;
294	}
295	cachv->iov_len = len << 2;
296	memcpy(cachv->iov_base, statp, len << 2);
297	break;
298	}
299	spin_lock(&cache_lock);
300	lru_put_end(rp);
301	rp->c_secure = rqstp->rq_secure;
302	rp->c_type = cachetype;
303	rp->c_state = RC_DONE;
304	rp->c_timestamp = jiffies;
305	spin_unlock(&cache_lock);
306	return;
307	}
308
309	/*
310	* Copy cached reply to current reply buffer. Should always fit.
311	* FIXME as reply is in a page, we should just attach the page, and
312	* keep a refcount....
313	*/
314	static int
315	nfsd_cache_append(struct svc_rqst rqstp, struct kvec data)
316	{
317	struct kvec *vec = &rqstp->rq_res.head[0];
318
319	if (vec->iov_len + data->iov_len > PAGE_SIZE) {
320	printk(KERN_WARNING "nfsd: cached reply too large (%Zd).\n",
321	data->iov_len);
322	return 0;
323	}
324	memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len);
325	vec->iov_len += data->iov_len;
326	return 1;
327	}