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 <okir@monad.swb.de>
11  */
12 
13 #include <linux/kernel.h>
14 #include <linux/sched.h>
15 #include <linux/slab.h>
16 #include <linux/string.h>
17 
18 #include <linux/sunrpc/svc.h>
19 #include <linux/nfsd/nfsd.h>
20 #include <linux/nfsd/cache.h>
21 
22 /* Size of reply cache. Common values are:
23  * 4.3BSD:	128
24  * 4.4BSD:	256
25  * Solaris2:	1024
26  * DEC Unix:	512-4096
27  */
28 #define CACHESIZE		1024
29 #define HASHSIZE		64
30 #define REQHASH(xid)		((((xid) >> 24) ^ (xid)) & (HASHSIZE-1))
31 
32 struct nfscache_head {
33 	struct svc_cacherep *	next;
34 	struct svc_cacherep *	prev;
35 };
36 
37 static struct nfscache_head *	hash_list;
38 static struct svc_cacherep *	lru_head;
39 static struct svc_cacherep *	lru_tail;
40 static struct svc_cacherep *	nfscache;
41 static int			cache_disabled = 1;
42 
43 static int	nfsd_cache_append(struct svc_rqst *rqstp, struct svc_buf *data);
44 
45 void
nfsd_cache_init(void)46 nfsd_cache_init(void)
47 {
48 	struct svc_cacherep	*rp;
49 	struct nfscache_head	*rh;
50 	size_t			i;
51 	unsigned long		order;
52 
53 
54 	i = CACHESIZE * sizeof (struct svc_cacherep);
55 	for (order = 0; (PAGE_SIZE << order) < i; order++)
56 		;
57 	nfscache = (struct svc_cacherep *)
58 		__get_free_pages(GFP_KERNEL, order);
59 	if (!nfscache) {
60 		printk (KERN_ERR "nfsd: cannot allocate %Zd bytes for reply cache\n", i);
61 		return;
62 	}
63 	memset(nfscache, 0, i);
64 
65 	i = HASHSIZE * sizeof (struct nfscache_head);
66 	hash_list = kmalloc (i, GFP_KERNEL);
67 	if (!hash_list) {
68 		free_pages ((unsigned long)nfscache, order);
69 		nfscache = NULL;
70 		printk (KERN_ERR "nfsd: cannot allocate %Zd bytes for hash list\n", i);
71 		return;
72 	}
73 
74 	for (i = 0, rh = hash_list; i < HASHSIZE; i++, rh++)
75 		rh->next = rh->prev = (struct svc_cacherep *) rh;
76 
77 	for (i = 0, rp = nfscache; i < CACHESIZE; i++, rp++) {
78 		rp->c_state = RC_UNUSED;
79 		rp->c_type = RC_NOCACHE;
80 		rp->c_hash_next =
81 		rp->c_hash_prev = rp;
82 		rp->c_lru_next = rp + 1;
83 		rp->c_lru_prev = rp - 1;
84 	}
85 	lru_head = nfscache;
86 	lru_tail = nfscache + CACHESIZE - 1;
87 	lru_head->c_lru_prev = NULL;
88 	lru_tail->c_lru_next = NULL;
89 
90 	cache_disabled = 0;
91 }
92 
93 void
nfsd_cache_shutdown(void)94 nfsd_cache_shutdown(void)
95 {
96 	struct svc_cacherep	*rp;
97 	size_t			i;
98 	unsigned long		order;
99 
100 	for (rp = lru_head; rp; rp = rp->c_lru_next) {
101 		if (rp->c_state == RC_DONE && rp->c_type == RC_REPLBUFF)
102 			kfree(rp->c_replbuf.buf);
103 	}
104 
105 	cache_disabled = 1;
106 
107 	i = CACHESIZE * sizeof (struct svc_cacherep);
108 	for (order = 0; (PAGE_SIZE << order) < i; order++)
109 		;
110 	free_pages ((unsigned long)nfscache, order);
111 	nfscache = NULL;
112 	kfree (hash_list);
113 	hash_list = NULL;
114 }
115 
116 /*
117  * Move cache entry to front of LRU list
118  */
119 static void
lru_put_front(struct svc_cacherep * rp)120 lru_put_front(struct svc_cacherep *rp)
121 {
122 	struct svc_cacherep	*prev = rp->c_lru_prev,
123 				*next = rp->c_lru_next;
124 
125 	if (prev)
126 		prev->c_lru_next = next;
127 	else
128 		lru_head = next;
129 	if (next)
130 		next->c_lru_prev = prev;
131 	else
132 		lru_tail = prev;
133 
134 	rp->c_lru_next = lru_head;
135 	rp->c_lru_prev = NULL;
136 	if (lru_head)
137 		lru_head->c_lru_prev = rp;
138 	lru_head = rp;
139 }
140 
141 /*
142  * Move a cache entry from one hash list to another
143  */
144 static void
hash_refile(struct svc_cacherep * rp)145 hash_refile(struct svc_cacherep *rp)
146 {
147 	struct svc_cacherep	*prev = rp->c_hash_prev,
148 				*next = rp->c_hash_next;
149 	struct nfscache_head	*head = hash_list + REQHASH(rp->c_xid);
150 
151 	prev->c_hash_next = next;
152 	next->c_hash_prev = prev;
153 
154 	rp->c_hash_next = head->next;
155 	rp->c_hash_prev = (struct svc_cacherep *) head;
156 	head->next->c_hash_prev = rp;
157 	head->next = rp;
158 }
159 
160 /*
161  * Try to find an entry matching the current call in the cache. When none
162  * is found, we grab the oldest unlocked entry off the LRU list.
163  * Note that no operation within the loop may sleep.
164  */
165 int
nfsd_cache_lookup(struct svc_rqst * rqstp,int type)166 nfsd_cache_lookup(struct svc_rqst *rqstp, int type)
167 {
168 	struct svc_cacherep	*rh, *rp;
169 	u32			xid = rqstp->rq_xid,
170 				proto =  rqstp->rq_prot,
171 				vers = rqstp->rq_vers,
172 				proc = rqstp->rq_proc;
173 	unsigned long		age;
174 
175 	rqstp->rq_cacherep = NULL;
176 	if (cache_disabled || type == RC_NOCACHE) {
177 		nfsdstats.rcnocache++;
178 		return RC_DOIT;
179 	}
180 
181 	rp = rh = (struct svc_cacherep *) &hash_list[REQHASH(xid)];
182 	while ((rp = rp->c_hash_next) != rh) {
183 		if (rp->c_state != RC_UNUSED &&
184 		    xid == rp->c_xid && proc == rp->c_proc &&
185 		    proto == rp->c_prot && vers == rp->c_vers &&
186 		    time_before(jiffies, rp->c_timestamp + 120*HZ) &&
187 		    memcmp((char*)&rqstp->rq_addr, (char*)&rp->c_addr, sizeof(rp->c_addr))==0) {
188 			nfsdstats.rchits++;
189 			goto found_entry;
190 		}
191 	}
192 	nfsdstats.rcmisses++;
193 
194 	/* This loop shouldn't take more than a few iterations normally */
195 	{
196 	int	safe = 0;
197 	for (rp = lru_tail; rp; rp = rp->c_lru_prev) {
198 		if (rp->c_state != RC_INPROG)
199 			break;
200 		if (safe++ > CACHESIZE) {
201 			printk("nfsd: loop in repcache LRU list\n");
202 			cache_disabled = 1;
203 			return RC_DOIT;
204 		}
205 	}
206 	}
207 
208 	/* This should not happen */
209 	if (rp == NULL) {
210 		static int	complaints;
211 
212 		printk(KERN_WARNING "nfsd: all repcache entries locked!\n");
213 		if (++complaints > 5) {
214 			printk(KERN_WARNING "nfsd: disabling repcache.\n");
215 			cache_disabled = 1;
216 		}
217 		return RC_DOIT;
218 	}
219 
220 	rqstp->rq_cacherep = rp;
221 	rp->c_state = RC_INPROG;
222 	rp->c_xid = xid;
223 	rp->c_proc = proc;
224 	rp->c_addr = rqstp->rq_addr;
225 	rp->c_prot = proto;
226 	rp->c_vers = vers;
227 	rp->c_timestamp = jiffies;
228 
229 	hash_refile(rp);
230 
231 	/* release any buffer */
232 	if (rp->c_type == RC_REPLBUFF) {
233 		kfree(rp->c_replbuf.buf);
234 		rp->c_replbuf.buf = NULL;
235 	}
236 	rp->c_type = RC_NOCACHE;
237 
238 	return RC_DOIT;
239 
240 found_entry:
241 	/* We found a matching entry which is either in progress or done. */
242 	age = jiffies - rp->c_timestamp;
243 	rp->c_timestamp = jiffies;
244 	lru_put_front(rp);
245 
246 	/* Request being processed or excessive rexmits */
247 	if (rp->c_state == RC_INPROG || age < RC_DELAY)
248 		return RC_DROPIT;
249 
250 	/* From the hall of fame of impractical attacks:
251 	 * Is this a user who tries to snoop on the cache? */
252 	if (!rqstp->rq_secure && rp->c_secure)
253 		return RC_DOIT;
254 
255 	/* Compose RPC reply header */
256 	switch (rp->c_type) {
257 	case RC_NOCACHE:
258 		return RC_DOIT;
259 	case RC_REPLSTAT:
260 		svc_putlong(&rqstp->rq_resbuf, rp->c_replstat);
261 		break;
262 	case RC_REPLBUFF:
263 		if (!nfsd_cache_append(rqstp, &rp->c_replbuf))
264 			return RC_DOIT;	/* should not happen */
265 		break;
266 	default:
267 		printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
268 		rp->c_state = RC_UNUSED;
269 		return RC_DOIT;
270 	}
271 
272 	return RC_REPLY;
273 }
274 
275 /*
276  * Update a cache entry. This is called from nfsd_dispatch when
277  * the procedure has been executed and the complete reply is in
278  * rqstp->rq_res.
279  *
280  * We're copying around data here rather than swapping buffers because
281  * the toplevel loop requires max-sized buffers, which would be a waste
282  * of memory for a cache with a max reply size of 100 bytes (diropokres).
283  *
284  * If we should start to use different types of cache entries tailored
285  * specifically for attrstat and fh's, we may save even more space.
286  *
287  * Also note that a cachetype of RC_NOCACHE can legally be passed when
288  * nfsd failed to encode a reply that otherwise would have been cached.
289  * In this case, nfsd_cache_update is called with statp == NULL.
290  */
291 void
nfsd_cache_update(struct svc_rqst * rqstp,int cachetype,u32 * statp)292 nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, u32 *statp)
293 {
294 	struct svc_cacherep *rp;
295 	struct svc_buf	*resp = &rqstp->rq_resbuf, *cachp;
296 	int		len;
297 
298 	if (!(rp = rqstp->rq_cacherep) || cache_disabled)
299 		return;
300 
301 	len = resp->len - (statp - resp->base);
302 
303 	/* Don't cache excessive amounts of data and XDR failures */
304 	if (!statp || len > (256 >> 2)) {
305 		rp->c_state = RC_UNUSED;
306 		return;
307 	}
308 
309 	switch (cachetype) {
310 	case RC_REPLSTAT:
311 		if (len != 1)
312 			printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
313 		rp->c_replstat = *statp;
314 		break;
315 	case RC_REPLBUFF:
316 		cachp = &rp->c_replbuf;
317 		cachp->buf = (u32 *) kmalloc(len << 2, GFP_KERNEL);
318 		if (!cachp->buf) {
319 			rp->c_state = RC_UNUSED;
320 			return;
321 		}
322 		cachp->len = len;
323 		memcpy(cachp->buf, statp, len << 2);
324 		break;
325 	}
326 
327 	lru_put_front(rp);
328 	rp->c_secure = rqstp->rq_secure;
329 	rp->c_type = cachetype;
330 	rp->c_state = RC_DONE;
331 	rp->c_timestamp = jiffies;
332 
333 	return;
334 }
335 
336 /*
337  * Copy cached reply to current reply buffer. Should always fit.
338  */
339 static int
nfsd_cache_append(struct svc_rqst * rqstp,struct svc_buf * data)340 nfsd_cache_append(struct svc_rqst *rqstp, struct svc_buf *data)
341 {
342 	struct svc_buf	*resp = &rqstp->rq_resbuf;
343 
344 	if (resp->len + data->len > resp->buflen) {
345 		printk(KERN_WARNING "nfsd: cached reply too large (%d).\n",
346 				data->len);
347 		return 0;
348 	}
349 	memcpy(resp->buf, data->buf, data->len << 2);
350 	resp->buf += data->len;
351 	resp->len += data->len;
352 	return 1;
353 }
354