[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/addr.h>
#include <linux/highmem.h>
#include <net/checksum.h>

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

#define NFSDDBG_FACILITY	NFSDDBG_REPCACHE

#define HASHSIZE		64

static struct hlist_head *	cache_hash;
static struct list_head 	lru_head;
static struct kmem_cache	*drc_slab;
static unsigned int		num_drc_entries;
static unsigned int		max_drc_entries;

/*
 * 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);
static void	cache_cleaner_func(struct work_struct *unused);
static int 	nfsd_reply_cache_shrink(struct shrinker *shrink,
					struct shrink_control *sc);

struct shrinker nfsd_reply_cache_shrinker = {
	.shrink	= nfsd_reply_cache_shrink,
	.seeks	= 1,
};

/*
 * 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);
static DECLARE_DELAYED_WORK(cache_cleaner, cache_cleaner_func);

/*
 * Put a cap on the size of the DRC based on the amount of available
 * low memory in the machine.
 *
 *  64MB:    8192
 * 128MB:   11585
 * 256MB:   16384
 * 512MB:   23170
 *   1GB:   32768
 *   2GB:   46340
 *   4GB:   65536
 *   8GB:   92681
 *  16GB:  131072
 *
 * ...with a hard cap of 256k entries. In the worst case, each entry will be
 * ~1k, so the above numbers should give a rough max of the amount of memory
 * used in k.
 */
static unsigned int
nfsd_cache_size_limit(void)
{
	unsigned int limit;
	unsigned long low_pages = totalram_pages - totalhigh_pages;

	limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10);
	return min_t(unsigned int, limit, 256*1024);
}

static struct svc_cacherep *
nfsd_reply_cache_alloc(void)
{
	struct svc_cacherep	*rp;

	rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
	if (rp) {
		rp->c_state = RC_UNUSED;
		rp->c_type = RC_NOCACHE;
		INIT_LIST_HEAD(&rp->c_lru);
		INIT_HLIST_NODE(&rp->c_hash);
	}
	return rp;
}

static void
nfsd_reply_cache_free_locked(struct svc_cacherep *rp)
{
	if (rp->c_type == RC_REPLBUFF)
		kfree(rp->c_replvec.iov_base);
	hlist_del(&rp->c_hash);
	list_del(&rp->c_lru);
	--num_drc_entries;
	kmem_cache_free(drc_slab, rp);
}

static void
nfsd_reply_cache_free(struct svc_cacherep *rp)
{
	spin_lock(&cache_lock);
	nfsd_reply_cache_free_locked(rp);
	spin_unlock(&cache_lock);
}

int nfsd_reply_cache_init(void)
{
	register_shrinker(&nfsd_reply_cache_shrinker);
	drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep),
					0, 0, NULL);
	if (!drc_slab)
		goto out_nomem;

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

	INIT_LIST_HEAD(&lru_head);
	max_drc_entries = nfsd_cache_size_limit();
	num_drc_entries = 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;

	unregister_shrinker(&nfsd_reply_cache_shrinker);
	cancel_delayed_work_sync(&cache_cleaner);

	while (!list_empty(&lru_head)) {
		rp = list_entry(lru_head.next, struct svc_cacherep, c_lru);
		nfsd_reply_cache_free_locked(rp);
	}

	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, and queue the cleaner to run if it's
 * not already scheduled.
 */
static void
lru_put_end(struct svc_cacherep *rp)
{
	rp->c_timestamp = jiffies;
	list_move_tail(&rp->c_lru, &lru_head);
	schedule_delayed_work(&cache_cleaner, RC_EXPIRE);
}

/*
 * 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));
}

static inline bool
nfsd_cache_entry_expired(struct svc_cacherep *rp)
{
	return rp->c_state != RC_INPROG &&
	       time_after(jiffies, rp->c_timestamp + RC_EXPIRE);
}

/*
 * Walk the LRU list and prune off entries that are older than RC_EXPIRE.
 * Also prune the oldest ones when the total exceeds the max number of entries.
 */
static void
prune_cache_entries(void)
{
	struct svc_cacherep *rp, *tmp;

	list_for_each_entry_safe(rp, tmp, &lru_head, c_lru) {
		if (!nfsd_cache_entry_expired(rp) &&
		    num_drc_entries <= max_drc_entries)
			break;
		nfsd_reply_cache_free_locked(rp);
	}

	/*
	 * Conditionally rearm the job. If we cleaned out the list, then
	 * cancel any pending run (since there won't be any work to do).
	 * Otherwise, we rearm the job or modify the existing one to run in
	 * RC_EXPIRE since we just ran the pruner.
	 */
	if (list_empty(&lru_head))
		cancel_delayed_work(&cache_cleaner);
	else
		mod_delayed_work(system_wq, &cache_cleaner, RC_EXPIRE);
}

static void
cache_cleaner_func(struct work_struct *unused)
{
	spin_lock(&cache_lock);
	prune_cache_entries();
	spin_unlock(&cache_lock);
}

static int
nfsd_reply_cache_shrink(struct shrinker *shrink, struct shrink_control *sc)
{
	unsigned int num;

	spin_lock(&cache_lock);
	if (sc->nr_to_scan)
		prune_cache_entries();
	num = num_drc_entries;
	spin_unlock(&cache_lock);

	return num;
}

/*
 * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes
 */
static __wsum
nfsd_cache_csum(struct svc_rqst *rqstp)
{
	int idx;
	unsigned int base;
	__wsum csum;
	struct xdr_buf *buf = &rqstp->rq_arg;
	const unsigned char *p = buf->head[0].iov_base;
	size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len,
				RC_CSUMLEN);
	size_t len = min(buf->head[0].iov_len, csum_len);

	/* rq_arg.head first */
	csum = csum_partial(p, len, 0);
	csum_len -= len;

	/* Continue into page array */
	idx = buf->page_base / PAGE_SIZE;
	base = buf->page_base & ~PAGE_MASK;
	while (csum_len) {
		p = page_address(buf->pages[idx]) + base;
		len = min_t(size_t, PAGE_SIZE - base, csum_len);
		csum = csum_partial(p, len, csum);
		csum_len -= len;
		base = 0;
		++idx;
	}
	return csum;
}

/*
 * Search the request hash for an entry that matches the given rqstp.
 * Must be called with cache_lock held. Returns the found entry or
 * NULL on failure.
 */
static struct svc_cacherep *
nfsd_cache_search(struct svc_rqst *rqstp, __wsum csum)
{
	struct svc_cacherep	*rp;
	struct hlist_head 	*rh;
	__be32			xid = rqstp->rq_xid;
	u32			proto =  rqstp->rq_prot,
				vers = rqstp->rq_vers,
				proc = rqstp->rq_proc;

	rh = &cache_hash[request_hash(xid)];
	hlist_for_each_entry(rp, rh, c_hash) {
		if (xid == rp->c_xid && proc == rp->c_proc &&
		    proto == rp->c_prot && vers == rp->c_vers &&
		    rqstp->rq_arg.len == rp->c_len && csum == rp->c_csum &&
		    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))
			return rp;
	}
	return NULL;
}

/*
 * Try to find an entry matching the current call in the cache. When none
 * is found, we try to grab the oldest expired entry off the LRU list. If
 * a suitable one isn't there, then drop the cache_lock and allocate a
 * new one, then search again in case one got inserted while this thread
 * didn't hold the lock.
 */
int
nfsd_cache_lookup(struct svc_rqst *rqstp)
{
	struct svc_cacherep	*rp, *found;
	__be32			xid = rqstp->rq_xid;
	u32			proto =  rqstp->rq_prot,
				vers = rqstp->rq_vers,
				proc = rqstp->rq_proc;
	__wsum			csum;
	unsigned long		age;
	int type = rqstp->rq_cachetype;
	int rtn;

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

	csum = nfsd_cache_csum(rqstp);

	spin_lock(&cache_lock);
	rtn = RC_DOIT;

	rp = nfsd_cache_search(rqstp, csum);
	if (rp)
		goto found_entry;

	/* Try to use the first entry on the LRU */
	if (!list_empty(&lru_head)) {
		rp = list_first_entry(&lru_head, struct svc_cacherep, c_lru);
		if (nfsd_cache_entry_expired(rp) ||
		    num_drc_entries >= max_drc_entries) {
			lru_put_end(rp);
			prune_cache_entries();
			goto setup_entry;
		}
	}

	/* Drop the lock and allocate a new entry */
	spin_unlock(&cache_lock);
	rp = nfsd_reply_cache_alloc();
	if (!rp) {
		dprintk("nfsd: unable to allocate DRC entry!\n");
		return RC_DOIT;
	}
	spin_lock(&cache_lock);
	++num_drc_entries;

	/*
	 * Must search again just in case someone inserted one
	 * after we dropped the lock above.
	 */
	found = nfsd_cache_search(rqstp, csum);
	if (found) {
		nfsd_reply_cache_free_locked(rp);
		rp = found;
		goto found_entry;
	}

	/*
	 * We're keeping the one we just allocated. Are we now over the
	 * limit? Prune one off the tip of the LRU in trade for the one we
	 * just allocated if so.
	 */
	if (num_drc_entries >= max_drc_entries)
		nfsd_reply_cache_free_locked(list_first_entry(&lru_head,
						struct svc_cacherep, c_lru));

setup_entry:
	nfsdstats.rcmisses++;
	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_len = rqstp->rq_arg.len;
	rp->c_csum = csum;

	hash_refile(rp);
	lru_put_end(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:
	nfsdstats.rchits++;
	/* We found a matching entry which is either in progress or done. */
	age = jiffies - rp->c_timestamp;
	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);
		nfsd_reply_cache_free_locked(rp);
	}

	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 = rqstp->rq_cacherep;
	struct kvec	*resv = &rqstp->rq_res.head[0], *cachv;
	int		len;

	if (!rp)
		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)) {
		nfsd_reply_cache_free(rp);
		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) {
			nfsd_reply_cache_free(rp);
			return;
		}
		cachv->iov_len = len << 2;
		memcpy(cachv->iov_base, statp, len << 2);
		break;
	case RC_NOCACHE:
		nfsd_reply_cache_free(rp);
		return;
	}
	spin_lock(&cache_lock);
	lru_put_end(rp);
	rp->c_secure = rqstp->rq_secure;
	rp->c_type = cachetype;
	rp->c_state = RC_DONE;
	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>
5976687a	12	#include <linux/sunrpc/addr.h>
0338dd15	13	#include <linux/highmem.h>
01a7decf	14	#include <net/checksum.h>
5a0e3ad6	15
9a74af21 BH	16	#include "nfsd.h"
9a74af21 BH	17	#include "cache.h"
1da177e4	18
0338dd15 JL	19	#define NFSDDBG_FACILITY NFSDDBG_REPCACHE
0338dd15 JL	20
1da177e4	21	#define HASHSIZE 64
1da177e4	22
fca4217c	23	static struct hlist_head * cache_hash;
1da177e4	24	static struct list_head lru_head;
8a8bc40d	25	static struct kmem_cache *drc_slab;
0ee0bf7e	26	static unsigned int num_drc_entries;
0338dd15	27	static unsigned int max_drc_entries;
1da177e4	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	39	static int nfsd_cache_append(struct svc_rqst rqstp, struct kvec vec);
aca8a23d	40	static void cache_cleaner_func(struct work_struct *unused);
b4e7f2c9 JL	41	static int nfsd_reply_cache_shrink(struct shrinker *shrink,
	42	struct shrink_control *sc);
	43
	44	struct shrinker nfsd_reply_cache_shrinker = {
	45	.shrink = nfsd_reply_cache_shrink,
	46	.seeks = 1,
	47	};
1da177e4	48
fca4217c	49	/*
1da177e4 LT	50	* locking for the reply cache:
	51	* A cache entry is "single use" if c_state == RC_INPROG
	52	* Otherwise, it when accessing _prev or _next, the lock must be held.
	53	*/
	54	static DEFINE_SPINLOCK(cache_lock);
aca8a23d	55	static DECLARE_DELAYED_WORK(cache_cleaner, cache_cleaner_func);
1da177e4	56
0338dd15 JL	57	/*
	58	* Put a cap on the size of the DRC based on the amount of available
	59	* low memory in the machine.
	60	*
	61	* 64MB: 8192
	62	* 128MB: 11585
	63	* 256MB: 16384
	64	* 512MB: 23170
	65	* 1GB: 32768
	66	* 2GB: 46340
	67	* 4GB: 65536
	68	* 8GB: 92681
	69	* 16GB: 131072
	70	*
	71	* ...with a hard cap of 256k entries. In the worst case, each entry will be
	72	* ~1k, so the above numbers should give a rough max of the amount of memory
	73	* used in k.
	74	*/
	75	static unsigned int
	76	nfsd_cache_size_limit(void)
	77	{
	78	unsigned int limit;
	79	unsigned long low_pages = totalram_pages - totalhigh_pages;
	80
	81	limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10);
	82	return min_t(unsigned int, limit, 256*1024);
	83	}
	84
f09841fd JL	85	static struct svc_cacherep *
f09841fd JL	86	nfsd_reply_cache_alloc(void)
1da177e4 LT	87	{
1da177e4 LT	88	struct svc_cacherep *rp;
1da177e4	89
f09841fd JL	90	rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
f09841fd JL	91	if (rp) {
1da177e4 LT	92	rp->c_state = RC_UNUSED;
1da177e4 LT	93	rp->c_type = RC_NOCACHE;
f09841fd	94	INIT_LIST_HEAD(&rp->c_lru);
1da177e4	95	INIT_HLIST_NODE(&rp->c_hash);
1da177e4	96	}
f09841fd JL	97	return rp;
f09841fd JL	98	}
1da177e4	99
f09841fd JL	100	static void
	101	nfsd_reply_cache_free_locked(struct svc_cacherep *rp)
	102	{
25e6b8b0	103	if (rp->c_type == RC_REPLBUFF)
f09841fd	104	kfree(rp->c_replvec.iov_base);
0338dd15	105	hlist_del(&rp->c_hash);
f09841fd	106	list_del(&rp->c_lru);
0ee0bf7e	107	--num_drc_entries;
f09841fd JL	108	kmem_cache_free(drc_slab, rp);
	109	}
	110
2c6b691c JL	111	static void
	112	nfsd_reply_cache_free(struct svc_cacherep *rp)
	113	{
	114	spin_lock(&cache_lock);
	115	nfsd_reply_cache_free_locked(rp);
	116	spin_unlock(&cache_lock);
	117	}
	118
f09841fd JL	119	int nfsd_reply_cache_init(void)
f09841fd JL	120	{
b4e7f2c9	121	register_shrinker(&nfsd_reply_cache_shrinker);
8a8bc40d JL	122	drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep),
	123	0, 0, NULL);
	124	if (!drc_slab)
	125	goto out_nomem;
	126
0338dd15	127	cache_hash = kcalloc(HASHSIZE, sizeof(struct hlist_head), GFP_KERNEL);
fca4217c	128	if (!cache_hash)
d5c3428b	129	goto out_nomem;
1da177e4	130
0338dd15 JL	131	INIT_LIST_HEAD(&lru_head);
	132	max_drc_entries = nfsd_cache_size_limit();
	133	num_drc_entries = 0;
01a7decf	134
d5c3428b BF	135	return 0;
	136	out_nomem:
	137	printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
	138	nfsd_reply_cache_shutdown();
	139	return -ENOMEM;
1da177e4 LT	140	}
1da177e4 LT	141
d5c3428b	142	void nfsd_reply_cache_shutdown(void)
1da177e4 LT	143	{
	144	struct svc_cacherep *rp;
	145
b4e7f2c9	146	unregister_shrinker(&nfsd_reply_cache_shrinker);
aca8a23d JL	147	cancel_delayed_work_sync(&cache_cleaner);
aca8a23d JL	148
1da177e4 LT	149	while (!list_empty(&lru_head)) {
1da177e4 LT	150	rp = list_entry(lru_head.next, struct svc_cacherep, c_lru);
f09841fd	151	nfsd_reply_cache_free_locked(rp);
1da177e4 LT	152	}
1da177e4 LT	153
fca4217c GB	154	kfree (cache_hash);
fca4217c GB	155	cache_hash = NULL;
8a8bc40d JL	156
	157	if (drc_slab) {
	158	kmem_cache_destroy(drc_slab);
	159	drc_slab = NULL;
	160	}
1da177e4 LT	161	}
	162
	163	/*
aca8a23d JL	164	* Move cache entry to end of LRU list, and queue the cleaner to run if it's
aca8a23d JL	165	* not already scheduled.
1da177e4 LT	166	*/
	167	static void
	168	lru_put_end(struct svc_cacherep *rp)
	169	{
56c2548b	170	rp->c_timestamp = jiffies;
f116629d	171	list_move_tail(&rp->c_lru, &lru_head);
aca8a23d	172	schedule_delayed_work(&cache_cleaner, RC_EXPIRE);
1da177e4 LT	173	}
	174
	175	/*
	176	* Move a cache entry from one hash list to another
	177	*/
	178	static void
	179	hash_refile(struct svc_cacherep *rp)
	180	{
	181	hlist_del_init(&rp->c_hash);
fca4217c	182	hlist_add_head(&rp->c_hash, cache_hash + request_hash(rp->c_xid));
1da177e4 LT	183	}
1da177e4 LT	184
d1a0774d JL	185	static inline bool
	186	nfsd_cache_entry_expired(struct svc_cacherep *rp)
	187	{
	188	return rp->c_state != RC_INPROG &&
	189	time_after(jiffies, rp->c_timestamp + RC_EXPIRE);
	190	}
	191
aca8a23d JL	192	/*
	193	* Walk the LRU list and prune off entries that are older than RC_EXPIRE.
	194	* Also prune the oldest ones when the total exceeds the max number of entries.
	195	*/
	196	static void
	197	prune_cache_entries(void)
	198	{
	199	struct svc_cacherep rp, tmp;
	200
	201	list_for_each_entry_safe(rp, tmp, &lru_head, c_lru) {
	202	if (!nfsd_cache_entry_expired(rp) &&
	203	num_drc_entries <= max_drc_entries)
	204	break;
	205	nfsd_reply_cache_free_locked(rp);
	206	}
	207
	208	/*
	209	* Conditionally rearm the job. If we cleaned out the list, then
	210	* cancel any pending run (since there won't be any work to do).
	211	* Otherwise, we rearm the job or modify the existing one to run in
	212	* RC_EXPIRE since we just ran the pruner.
	213	*/
	214	if (list_empty(&lru_head))
	215	cancel_delayed_work(&cache_cleaner);
	216	else
	217	mod_delayed_work(system_wq, &cache_cleaner, RC_EXPIRE);
	218	}
	219
	220	static void
	221	cache_cleaner_func(struct work_struct *unused)
	222	{
	223	spin_lock(&cache_lock);
	224	prune_cache_entries();
	225	spin_unlock(&cache_lock);
	226	}
	227
b4e7f2c9 JL	228	static int
	229	nfsd_reply_cache_shrink(struct shrinker shrink, struct shrink_control sc)
	230	{
	231	unsigned int num;
	232
	233	spin_lock(&cache_lock);
	234	if (sc->nr_to_scan)
	235	prune_cache_entries();
	236	num = num_drc_entries;
	237	spin_unlock(&cache_lock);
	238
	239	return num;
	240	}
	241
01a7decf JL	242	/*
	243	* Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes
	244	*/
	245	static __wsum
	246	nfsd_cache_csum(struct svc_rqst *rqstp)
	247	{
	248	int idx;
	249	unsigned int base;
	250	__wsum csum;
	251	struct xdr_buf *buf = &rqstp->rq_arg;
	252	const unsigned char *p = buf->head[0].iov_base;
	253	size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len,
	254	RC_CSUMLEN);
	255	size_t len = min(buf->head[0].iov_len, csum_len);
	256
	257	/* rq_arg.head first */
	258	csum = csum_partial(p, len, 0);
	259	csum_len -= len;
	260
	261	/* Continue into page array */
	262	idx = buf->page_base / PAGE_SIZE;
	263	base = buf->page_base & ~PAGE_MASK;
	264	while (csum_len) {
	265	p = page_address(buf->pages[idx]) + base;
56edc86b	266	len = min_t(size_t, PAGE_SIZE - base, csum_len);
01a7decf JL	267	csum = csum_partial(p, len, csum);
	268	csum_len -= len;
	269	base = 0;
	270	++idx;
	271	}
	272	return csum;
	273	}
	274
a4a3ec32 JL	275	/*
	276	* Search the request hash for an entry that matches the given rqstp.
	277	* Must be called with cache_lock held. Returns the found entry or
	278	* NULL on failure.
	279	*/
	280	static struct svc_cacherep *
01a7decf	281	nfsd_cache_search(struct svc_rqst *rqstp, __wsum csum)
a4a3ec32 JL	282	{
a4a3ec32 JL	283	struct svc_cacherep *rp;
a4a3ec32 JL	284	struct hlist_head *rh;
	285	__be32 xid = rqstp->rq_xid;
	286	u32 proto = rqstp->rq_prot,
	287	vers = rqstp->rq_vers,
	288	proc = rqstp->rq_proc;
	289
	290	rh = &cache_hash[request_hash(xid)];
b6669737	291	hlist_for_each_entry(rp, rh, c_hash) {
2c6b691c	292	if (xid == rp->c_xid && proc == rp->c_proc &&
a4a3ec32	293	proto == rp->c_prot && vers == rp->c_vers &&
01a7decf	294	rqstp->rq_arg.len == rp->c_len && csum == rp->c_csum &&
a4a3ec32 JL	295	rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) &&
	296	rpc_get_port(svc_addr(rqstp)) == rpc_get_port((struct sockaddr *)&rp->c_addr))
	297	return rp;
	298	}
	299	return NULL;
	300	}
	301
1da177e4 LT	302	/*
1da177e4 LT	303	* Try to find an entry matching the current call in the cache. When none
1ac83629 JL	304	* is found, we try to grab the oldest expired entry off the LRU list. If
	305	* a suitable one isn't there, then drop the cache_lock and allocate a
	306	* new one, then search again in case one got inserted while this thread
	307	* didn't hold the lock.
1da177e4 LT	308	*/
1da177e4 LT	309	int
1091006c	310	nfsd_cache_lookup(struct svc_rqst *rqstp)
1da177e4	311	{
0338dd15	312	struct svc_cacherep rp, found;
c7afef1f AV	313	__be32 xid = rqstp->rq_xid;
c7afef1f AV	314	u32 proto = rqstp->rq_prot,
1da177e4 LT	315	vers = rqstp->rq_vers,
1da177e4 LT	316	proc = rqstp->rq_proc;
01a7decf	317	__wsum csum;
1da177e4	318	unsigned long age;
1091006c	319	int type = rqstp->rq_cachetype;
1da177e4 LT	320	int rtn;
	321
	322	rqstp->rq_cacherep = NULL;
13cc8a78	323	if (type == RC_NOCACHE) {
1da177e4 LT	324	nfsdstats.rcnocache++;
	325	return RC_DOIT;
	326	}
	327
01a7decf JL	328	csum = nfsd_cache_csum(rqstp);
01a7decf JL	329
1da177e4 LT	330	spin_lock(&cache_lock);
	331	rtn = RC_DOIT;
	332
01a7decf	333	rp = nfsd_cache_search(rqstp, csum);
0338dd15	334	if (rp)
a4a3ec32	335	goto found_entry;
0338dd15 JL	336
	337	/* Try to use the first entry on the LRU */
	338	if (!list_empty(&lru_head)) {
	339	rp = list_first_entry(&lru_head, struct svc_cacherep, c_lru);
	340	if (nfsd_cache_entry_expired(rp) \|\|
aca8a23d JL	341	num_drc_entries >= max_drc_entries) {
	342	lru_put_end(rp);
	343	prune_cache_entries();
0338dd15	344	goto setup_entry;
1da177e4 LT	345	}
1da177e4 LT	346	}
1da177e4	347
1ac83629	348	/* Drop the lock and allocate a new entry */
0338dd15 JL	349	spin_unlock(&cache_lock);
	350	rp = nfsd_reply_cache_alloc();
	351	if (!rp) {
	352	dprintk("nfsd: unable to allocate DRC entry!\n");
	353	return RC_DOIT;
1da177e4	354	}
0338dd15 JL	355	spin_lock(&cache_lock);
	356	++num_drc_entries;
	357
	358	/*
	359	* Must search again just in case someone inserted one
	360	* after we dropped the lock above.
	361	*/
01a7decf	362	found = nfsd_cache_search(rqstp, csum);
0338dd15 JL	363	if (found) {
	364	nfsd_reply_cache_free_locked(rp);
	365	rp = found;
	366	goto found_entry;
1da177e4 LT	367	}
1da177e4 LT	368
0338dd15 JL	369	/*
	370	* We're keeping the one we just allocated. Are we now over the
	371	* limit? Prune one off the tip of the LRU in trade for the one we
	372	* just allocated if so.
	373	*/
	374	if (num_drc_entries >= max_drc_entries)
	375	nfsd_reply_cache_free_locked(list_first_entry(&lru_head,
	376	struct svc_cacherep, c_lru));
1da177e4	377
0338dd15 JL	378	setup_entry:
0338dd15 JL	379	nfsdstats.rcmisses++;
1da177e4 LT	380	rqstp->rq_cacherep = rp;
	381	rp->c_state = RC_INPROG;
	382	rp->c_xid = xid;
	383	rp->c_proc = proc;
7b9e8522 JL	384	rpc_copy_addr((struct sockaddr *)&rp->c_addr, svc_addr(rqstp));
7b9e8522 JL	385	rpc_set_port((struct sockaddr *)&rp->c_addr, rpc_get_port(svc_addr(rqstp)));
1da177e4 LT	386	rp->c_prot = proto;
1da177e4 LT	387	rp->c_vers = vers;
01a7decf JL	388	rp->c_len = rqstp->rq_arg.len;
01a7decf JL	389	rp->c_csum = csum;
1da177e4 LT	390
1da177e4 LT	391	hash_refile(rp);
56c2548b	392	lru_put_end(rp);
1da177e4 LT	393
	394	/* release any buffer */
	395	if (rp->c_type == RC_REPLBUFF) {
	396	kfree(rp->c_replvec.iov_base);
	397	rp->c_replvec.iov_base = NULL;
	398	}
	399	rp->c_type = RC_NOCACHE;
	400	out:
	401	spin_unlock(&cache_lock);
	402	return rtn;
	403
	404	found_entry:
0338dd15	405	nfsdstats.rchits++;
1da177e4 LT	406	/* We found a matching entry which is either in progress or done. */
1da177e4 LT	407	age = jiffies - rp->c_timestamp;
1da177e4 LT	408	lru_put_end(rp);
	409
	410	rtn = RC_DROPIT;
	411	/* Request being processed or excessive rexmits */
	412	if (rp->c_state == RC_INPROG \|\| age < RC_DELAY)
	413	goto out;
	414
	415	/* From the hall of fame of impractical attacks:
	416	* Is this a user who tries to snoop on the cache? */
	417	rtn = RC_DOIT;
	418	if (!rqstp->rq_secure && rp->c_secure)
	419	goto out;
	420
	421	/* Compose RPC reply header */
	422	switch (rp->c_type) {
	423	case RC_NOCACHE:
	424	break;
	425	case RC_REPLSTAT:
	426	svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
	427	rtn = RC_REPLY;
	428	break;
	429	case RC_REPLBUFF:
	430	if (!nfsd_cache_append(rqstp, &rp->c_replvec))
	431	goto out; /* should not happen */
	432	rtn = RC_REPLY;
	433	break;
	434	default:
	435	printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
0338dd15	436	nfsd_reply_cache_free_locked(rp);
1da177e4 LT	437	}
	438
	439	goto out;
	440	}
	441
	442	/*
	443	* Update a cache entry. This is called from nfsd_dispatch when
	444	* the procedure has been executed and the complete reply is in
	445	* rqstp->rq_res.
	446	*
	447	* We're copying around data here rather than swapping buffers because
	448	* the toplevel loop requires max-sized buffers, which would be a waste
	449	* of memory for a cache with a max reply size of 100 bytes (diropokres).
	450	*
	451	* If we should start to use different types of cache entries tailored
	452	* specifically for attrstat and fh's, we may save even more space.
	453	*
	454	* Also note that a cachetype of RC_NOCACHE can legally be passed when
	455	* nfsd failed to encode a reply that otherwise would have been cached.
	456	* In this case, nfsd_cache_update is called with statp == NULL.
	457	*/
	458	void
c7afef1f	459	nfsd_cache_update(struct svc_rqst rqstp, int cachetype, __be32 statp)
1da177e4	460	{
13cc8a78	461	struct svc_cacherep *rp = rqstp->rq_cacherep;
1da177e4 LT	462	struct kvec resv = &rqstp->rq_res.head[0], cachv;
	463	int len;
	464
13cc8a78	465	if (!rp)
1da177e4 LT	466	return;
	467
	468	len = resv->iov_len - ((char)statp - (char)resv->iov_base);
	469	len >>= 2;
fca4217c	470
1da177e4 LT	471	/* Don't cache excessive amounts of data and XDR failures */
1da177e4 LT	472	if (!statp \|\| len > (256 >> 2)) {
2c6b691c	473	nfsd_reply_cache_free(rp);
1da177e4 LT	474	return;
	475	}
	476
	477	switch (cachetype) {
	478	case RC_REPLSTAT:
	479	if (len != 1)
	480	printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
	481	rp->c_replstat = *statp;
	482	break;
	483	case RC_REPLBUFF:
	484	cachv = &rp->c_replvec;
	485	cachv->iov_base = kmalloc(len << 2, GFP_KERNEL);
	486	if (!cachv->iov_base) {
2c6b691c	487	nfsd_reply_cache_free(rp);
1da177e4 LT	488	return;
	489	}
	490	cachv->iov_len = len << 2;
	491	memcpy(cachv->iov_base, statp, len << 2);
	492	break;
2c6b691c JL	493	case RC_NOCACHE:
	494	nfsd_reply_cache_free(rp);
	495	return;
1da177e4 LT	496	}
	497	spin_lock(&cache_lock);
	498	lru_put_end(rp);
	499	rp->c_secure = rqstp->rq_secure;
	500	rp->c_type = cachetype;
	501	rp->c_state = RC_DONE;
1da177e4 LT	502	spin_unlock(&cache_lock);
	503	return;
	504	}
	505
	506	/*
	507	* Copy cached reply to current reply buffer. Should always fit.
	508	* FIXME as reply is in a page, we should just attach the page, and
	509	* keep a refcount....
	510	*/
	511	static int
	512	nfsd_cache_append(struct svc_rqst rqstp, struct kvec data)
	513	{
	514	struct kvec *vec = &rqstp->rq_res.head[0];
	515
	516	if (vec->iov_len + data->iov_len > PAGE_SIZE) {
	517	printk(KERN_WARNING "nfsd: cached reply too large (%Zd).\n",
	518	data->iov_len);
	519	return 0;
	520	}
	521	memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len);
	522	vec->iov_len += data->iov_len;
	523	return 1;
	524	}