xref: /openbmc/linux/fs/nfsd/nfscache.c (revision 2b076054)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Request reply cache. This is currently a global cache, but this may
4  * change in the future and be a per-client cache.
5  *
6  * This code is heavily inspired by the 44BSD implementation, although
7  * it does things a bit differently.
8  *
9  * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
10  */
11 
12 #include <linux/sunrpc/svc_xprt.h>
13 #include <linux/slab.h>
14 #include <linux/vmalloc.h>
15 #include <linux/sunrpc/addr.h>
16 #include <linux/highmem.h>
17 #include <linux/log2.h>
18 #include <linux/hash.h>
19 #include <net/checksum.h>
20 
21 #include "nfsd.h"
22 #include "cache.h"
23 #include "trace.h"
24 
25 /*
26  * We use this value to determine the number of hash buckets from the max
27  * cache size, the idea being that when the cache is at its maximum number
28  * of entries, then this should be the average number of entries per bucket.
29  */
30 #define TARGET_BUCKET_SIZE	64
31 
32 struct nfsd_drc_bucket {
33 	struct rb_root rb_head;
34 	struct list_head lru_head;
35 	spinlock_t cache_lock;
36 };
37 
38 static struct kmem_cache	*drc_slab;
39 
40 static int	nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec);
41 static unsigned long nfsd_reply_cache_count(struct shrinker *shrink,
42 					    struct shrink_control *sc);
43 static unsigned long nfsd_reply_cache_scan(struct shrinker *shrink,
44 					   struct shrink_control *sc);
45 
46 /*
47  * Put a cap on the size of the DRC based on the amount of available
48  * low memory in the machine.
49  *
50  *  64MB:    8192
51  * 128MB:   11585
52  * 256MB:   16384
53  * 512MB:   23170
54  *   1GB:   32768
55  *   2GB:   46340
56  *   4GB:   65536
57  *   8GB:   92681
58  *  16GB:  131072
59  *
60  * ...with a hard cap of 256k entries. In the worst case, each entry will be
61  * ~1k, so the above numbers should give a rough max of the amount of memory
62  * used in k.
63  *
64  * XXX: these limits are per-container, so memory used will increase
65  * linearly with number of containers.  Maybe that's OK.
66  */
67 static unsigned int
68 nfsd_cache_size_limit(void)
69 {
70 	unsigned int limit;
71 	unsigned long low_pages = totalram_pages() - totalhigh_pages();
72 
73 	limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10);
74 	return min_t(unsigned int, limit, 256*1024);
75 }
76 
77 /*
78  * Compute the number of hash buckets we need. Divide the max cachesize by
79  * the "target" max bucket size, and round up to next power of two.
80  */
81 static unsigned int
82 nfsd_hashsize(unsigned int limit)
83 {
84 	return roundup_pow_of_two(limit / TARGET_BUCKET_SIZE);
85 }
86 
87 static u32
88 nfsd_cache_hash(__be32 xid, struct nfsd_net *nn)
89 {
90 	return hash_32(be32_to_cpu(xid), nn->maskbits);
91 }
92 
93 static struct svc_cacherep *
94 nfsd_reply_cache_alloc(struct svc_rqst *rqstp, __wsum csum,
95 			struct nfsd_net *nn)
96 {
97 	struct svc_cacherep	*rp;
98 
99 	rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
100 	if (rp) {
101 		rp->c_state = RC_UNUSED;
102 		rp->c_type = RC_NOCACHE;
103 		RB_CLEAR_NODE(&rp->c_node);
104 		INIT_LIST_HEAD(&rp->c_lru);
105 
106 		memset(&rp->c_key, 0, sizeof(rp->c_key));
107 		rp->c_key.k_xid = rqstp->rq_xid;
108 		rp->c_key.k_proc = rqstp->rq_proc;
109 		rpc_copy_addr((struct sockaddr *)&rp->c_key.k_addr, svc_addr(rqstp));
110 		rpc_set_port((struct sockaddr *)&rp->c_key.k_addr, rpc_get_port(svc_addr(rqstp)));
111 		rp->c_key.k_prot = rqstp->rq_prot;
112 		rp->c_key.k_vers = rqstp->rq_vers;
113 		rp->c_key.k_len = rqstp->rq_arg.len;
114 		rp->c_key.k_csum = csum;
115 	}
116 	return rp;
117 }
118 
119 static void
120 nfsd_reply_cache_free_locked(struct nfsd_drc_bucket *b, struct svc_cacherep *rp,
121 				struct nfsd_net *nn)
122 {
123 	if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) {
124 		nn->drc_mem_usage -= rp->c_replvec.iov_len;
125 		kfree(rp->c_replvec.iov_base);
126 	}
127 	if (rp->c_state != RC_UNUSED) {
128 		rb_erase(&rp->c_node, &b->rb_head);
129 		list_del(&rp->c_lru);
130 		atomic_dec(&nn->num_drc_entries);
131 		nn->drc_mem_usage -= sizeof(*rp);
132 	}
133 	kmem_cache_free(drc_slab, rp);
134 }
135 
136 static void
137 nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct svc_cacherep *rp,
138 			struct nfsd_net *nn)
139 {
140 	spin_lock(&b->cache_lock);
141 	nfsd_reply_cache_free_locked(b, rp, nn);
142 	spin_unlock(&b->cache_lock);
143 }
144 
145 int nfsd_drc_slab_create(void)
146 {
147 	drc_slab = kmem_cache_create("nfsd_drc",
148 				sizeof(struct svc_cacherep), 0, 0, NULL);
149 	return drc_slab ? 0: -ENOMEM;
150 }
151 
152 void nfsd_drc_slab_free(void)
153 {
154 	kmem_cache_destroy(drc_slab);
155 }
156 
157 int nfsd_reply_cache_init(struct nfsd_net *nn)
158 {
159 	unsigned int hashsize;
160 	unsigned int i;
161 	int status = 0;
162 
163 	nn->max_drc_entries = nfsd_cache_size_limit();
164 	atomic_set(&nn->num_drc_entries, 0);
165 	hashsize = nfsd_hashsize(nn->max_drc_entries);
166 	nn->maskbits = ilog2(hashsize);
167 
168 	nn->nfsd_reply_cache_shrinker.scan_objects = nfsd_reply_cache_scan;
169 	nn->nfsd_reply_cache_shrinker.count_objects = nfsd_reply_cache_count;
170 	nn->nfsd_reply_cache_shrinker.seeks = 1;
171 	status = register_shrinker(&nn->nfsd_reply_cache_shrinker);
172 	if (status)
173 		goto out_nomem;
174 
175 	nn->drc_hashtbl = kvzalloc(array_size(hashsize,
176 				sizeof(*nn->drc_hashtbl)), GFP_KERNEL);
177 	if (!nn->drc_hashtbl)
178 		goto out_shrinker;
179 
180 	for (i = 0; i < hashsize; i++) {
181 		INIT_LIST_HEAD(&nn->drc_hashtbl[i].lru_head);
182 		spin_lock_init(&nn->drc_hashtbl[i].cache_lock);
183 	}
184 	nn->drc_hashsize = hashsize;
185 
186 	return 0;
187 out_shrinker:
188 	unregister_shrinker(&nn->nfsd_reply_cache_shrinker);
189 out_nomem:
190 	printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
191 	return -ENOMEM;
192 }
193 
194 void nfsd_reply_cache_shutdown(struct nfsd_net *nn)
195 {
196 	struct svc_cacherep	*rp;
197 	unsigned int i;
198 
199 	unregister_shrinker(&nn->nfsd_reply_cache_shrinker);
200 
201 	for (i = 0; i < nn->drc_hashsize; i++) {
202 		struct list_head *head = &nn->drc_hashtbl[i].lru_head;
203 		while (!list_empty(head)) {
204 			rp = list_first_entry(head, struct svc_cacherep, c_lru);
205 			nfsd_reply_cache_free_locked(&nn->drc_hashtbl[i],
206 									rp, nn);
207 		}
208 	}
209 
210 	kvfree(nn->drc_hashtbl);
211 	nn->drc_hashtbl = NULL;
212 	nn->drc_hashsize = 0;
213 
214 }
215 
216 /*
217  * Move cache entry to end of LRU list, and queue the cleaner to run if it's
218  * not already scheduled.
219  */
220 static void
221 lru_put_end(struct nfsd_drc_bucket *b, struct svc_cacherep *rp)
222 {
223 	rp->c_timestamp = jiffies;
224 	list_move_tail(&rp->c_lru, &b->lru_head);
225 }
226 
227 static long
228 prune_bucket(struct nfsd_drc_bucket *b, struct nfsd_net *nn)
229 {
230 	struct svc_cacherep *rp, *tmp;
231 	long freed = 0;
232 
233 	list_for_each_entry_safe(rp, tmp, &b->lru_head, c_lru) {
234 		/*
235 		 * Don't free entries attached to calls that are still
236 		 * in-progress, but do keep scanning the list.
237 		 */
238 		if (rp->c_state == RC_INPROG)
239 			continue;
240 		if (atomic_read(&nn->num_drc_entries) <= nn->max_drc_entries &&
241 		    time_before(jiffies, rp->c_timestamp + RC_EXPIRE))
242 			break;
243 		nfsd_reply_cache_free_locked(b, rp, nn);
244 		freed++;
245 	}
246 	return freed;
247 }
248 
249 /*
250  * Walk the LRU list and prune off entries that are older than RC_EXPIRE.
251  * Also prune the oldest ones when the total exceeds the max number of entries.
252  */
253 static long
254 prune_cache_entries(struct nfsd_net *nn)
255 {
256 	unsigned int i;
257 	long freed = 0;
258 
259 	for (i = 0; i < nn->drc_hashsize; i++) {
260 		struct nfsd_drc_bucket *b = &nn->drc_hashtbl[i];
261 
262 		if (list_empty(&b->lru_head))
263 			continue;
264 		spin_lock(&b->cache_lock);
265 		freed += prune_bucket(b, nn);
266 		spin_unlock(&b->cache_lock);
267 	}
268 	return freed;
269 }
270 
271 static unsigned long
272 nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc)
273 {
274 	struct nfsd_net *nn = container_of(shrink,
275 				struct nfsd_net, nfsd_reply_cache_shrinker);
276 
277 	return atomic_read(&nn->num_drc_entries);
278 }
279 
280 static unsigned long
281 nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
282 {
283 	struct nfsd_net *nn = container_of(shrink,
284 				struct nfsd_net, nfsd_reply_cache_shrinker);
285 
286 	return prune_cache_entries(nn);
287 }
288 /*
289  * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes
290  */
291 static __wsum
292 nfsd_cache_csum(struct svc_rqst *rqstp)
293 {
294 	int idx;
295 	unsigned int base;
296 	__wsum csum;
297 	struct xdr_buf *buf = &rqstp->rq_arg;
298 	const unsigned char *p = buf->head[0].iov_base;
299 	size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len,
300 				RC_CSUMLEN);
301 	size_t len = min(buf->head[0].iov_len, csum_len);
302 
303 	/* rq_arg.head first */
304 	csum = csum_partial(p, len, 0);
305 	csum_len -= len;
306 
307 	/* Continue into page array */
308 	idx = buf->page_base / PAGE_SIZE;
309 	base = buf->page_base & ~PAGE_MASK;
310 	while (csum_len) {
311 		p = page_address(buf->pages[idx]) + base;
312 		len = min_t(size_t, PAGE_SIZE - base, csum_len);
313 		csum = csum_partial(p, len, csum);
314 		csum_len -= len;
315 		base = 0;
316 		++idx;
317 	}
318 	return csum;
319 }
320 
321 static int
322 nfsd_cache_key_cmp(const struct svc_cacherep *key,
323 			const struct svc_cacherep *rp, struct nfsd_net *nn)
324 {
325 	if (key->c_key.k_xid == rp->c_key.k_xid &&
326 	    key->c_key.k_csum != rp->c_key.k_csum) {
327 		++nn->payload_misses;
328 		trace_nfsd_drc_mismatch(nn, key, rp);
329 	}
330 
331 	return memcmp(&key->c_key, &rp->c_key, sizeof(key->c_key));
332 }
333 
334 /*
335  * Search the request hash for an entry that matches the given rqstp.
336  * Must be called with cache_lock held. Returns the found entry or
337  * inserts an empty key on failure.
338  */
339 static struct svc_cacherep *
340 nfsd_cache_insert(struct nfsd_drc_bucket *b, struct svc_cacherep *key,
341 			struct nfsd_net *nn)
342 {
343 	struct svc_cacherep	*rp, *ret = key;
344 	struct rb_node		**p = &b->rb_head.rb_node,
345 				*parent = NULL;
346 	unsigned int		entries = 0;
347 	int cmp;
348 
349 	while (*p != NULL) {
350 		++entries;
351 		parent = *p;
352 		rp = rb_entry(parent, struct svc_cacherep, c_node);
353 
354 		cmp = nfsd_cache_key_cmp(key, rp, nn);
355 		if (cmp < 0)
356 			p = &parent->rb_left;
357 		else if (cmp > 0)
358 			p = &parent->rb_right;
359 		else {
360 			ret = rp;
361 			goto out;
362 		}
363 	}
364 	rb_link_node(&key->c_node, parent, p);
365 	rb_insert_color(&key->c_node, &b->rb_head);
366 out:
367 	/* tally hash chain length stats */
368 	if (entries > nn->longest_chain) {
369 		nn->longest_chain = entries;
370 		nn->longest_chain_cachesize = atomic_read(&nn->num_drc_entries);
371 	} else if (entries == nn->longest_chain) {
372 		/* prefer to keep the smallest cachesize possible here */
373 		nn->longest_chain_cachesize = min_t(unsigned int,
374 				nn->longest_chain_cachesize,
375 				atomic_read(&nn->num_drc_entries));
376 	}
377 
378 	lru_put_end(b, ret);
379 	return ret;
380 }
381 
382 /**
383  * nfsd_cache_lookup - Find an entry in the duplicate reply cache
384  * @rqstp: Incoming Call to find
385  *
386  * Try to find an entry matching the current call in the cache. When none
387  * is found, we try to grab the oldest expired entry off the LRU list. If
388  * a suitable one isn't there, then drop the cache_lock and allocate a
389  * new one, then search again in case one got inserted while this thread
390  * didn't hold the lock.
391  *
392  * Return values:
393  *   %RC_DOIT: Process the request normally
394  *   %RC_REPLY: Reply from cache
395  *   %RC_DROPIT: Do not process the request further
396  */
397 int nfsd_cache_lookup(struct svc_rqst *rqstp)
398 {
399 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
400 	struct svc_cacherep	*rp, *found;
401 	__be32			xid = rqstp->rq_xid;
402 	__wsum			csum;
403 	u32 hash = nfsd_cache_hash(xid, nn);
404 	struct nfsd_drc_bucket *b = &nn->drc_hashtbl[hash];
405 	int type = rqstp->rq_cachetype;
406 	int rtn = RC_DOIT;
407 
408 	rqstp->rq_cacherep = NULL;
409 	if (type == RC_NOCACHE) {
410 		nfsdstats.rcnocache++;
411 		goto out;
412 	}
413 
414 	csum = nfsd_cache_csum(rqstp);
415 
416 	/*
417 	 * Since the common case is a cache miss followed by an insert,
418 	 * preallocate an entry.
419 	 */
420 	rp = nfsd_reply_cache_alloc(rqstp, csum, nn);
421 	if (!rp)
422 		goto out;
423 
424 	spin_lock(&b->cache_lock);
425 	found = nfsd_cache_insert(b, rp, nn);
426 	if (found != rp) {
427 		nfsd_reply_cache_free_locked(NULL, rp, nn);
428 		rp = found;
429 		goto found_entry;
430 	}
431 
432 	nfsdstats.rcmisses++;
433 	rqstp->rq_cacherep = rp;
434 	rp->c_state = RC_INPROG;
435 
436 	atomic_inc(&nn->num_drc_entries);
437 	nn->drc_mem_usage += sizeof(*rp);
438 
439 	/* go ahead and prune the cache */
440 	prune_bucket(b, nn);
441 
442 out_unlock:
443 	spin_unlock(&b->cache_lock);
444 out:
445 	return rtn;
446 
447 found_entry:
448 	/* We found a matching entry which is either in progress or done. */
449 	nfsdstats.rchits++;
450 	rtn = RC_DROPIT;
451 
452 	/* Request being processed */
453 	if (rp->c_state == RC_INPROG)
454 		goto out_trace;
455 
456 	/* From the hall of fame of impractical attacks:
457 	 * Is this a user who tries to snoop on the cache? */
458 	rtn = RC_DOIT;
459 	if (!test_bit(RQ_SECURE, &rqstp->rq_flags) && rp->c_secure)
460 		goto out_trace;
461 
462 	/* Compose RPC reply header */
463 	switch (rp->c_type) {
464 	case RC_NOCACHE:
465 		break;
466 	case RC_REPLSTAT:
467 		svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
468 		rtn = RC_REPLY;
469 		break;
470 	case RC_REPLBUFF:
471 		if (!nfsd_cache_append(rqstp, &rp->c_replvec))
472 			goto out_unlock; /* should not happen */
473 		rtn = RC_REPLY;
474 		break;
475 	default:
476 		WARN_ONCE(1, "nfsd: bad repcache type %d\n", rp->c_type);
477 	}
478 
479 out_trace:
480 	trace_nfsd_drc_found(nn, rqstp, rtn);
481 	goto out_unlock;
482 }
483 
484 /**
485  * nfsd_cache_update - Update an entry in the duplicate reply cache.
486  * @rqstp: svc_rqst with a finished Reply
487  * @cachetype: which cache to update
488  * @statp: Reply's status code
489  *
490  * This is called from nfsd_dispatch when the procedure has been
491  * executed and the complete reply is in rqstp->rq_res.
492  *
493  * We're copying around data here rather than swapping buffers because
494  * the toplevel loop requires max-sized buffers, which would be a waste
495  * of memory for a cache with a max reply size of 100 bytes (diropokres).
496  *
497  * If we should start to use different types of cache entries tailored
498  * specifically for attrstat and fh's, we may save even more space.
499  *
500  * Also note that a cachetype of RC_NOCACHE can legally be passed when
501  * nfsd failed to encode a reply that otherwise would have been cached.
502  * In this case, nfsd_cache_update is called with statp == NULL.
503  */
504 void nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
505 {
506 	struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
507 	struct svc_cacherep *rp = rqstp->rq_cacherep;
508 	struct kvec	*resv = &rqstp->rq_res.head[0], *cachv;
509 	u32		hash;
510 	struct nfsd_drc_bucket *b;
511 	int		len;
512 	size_t		bufsize = 0;
513 
514 	if (!rp)
515 		return;
516 
517 	hash = nfsd_cache_hash(rp->c_key.k_xid, nn);
518 	b = &nn->drc_hashtbl[hash];
519 
520 	len = resv->iov_len - ((char*)statp - (char*)resv->iov_base);
521 	len >>= 2;
522 
523 	/* Don't cache excessive amounts of data and XDR failures */
524 	if (!statp || len > (256 >> 2)) {
525 		nfsd_reply_cache_free(b, rp, nn);
526 		return;
527 	}
528 
529 	switch (cachetype) {
530 	case RC_REPLSTAT:
531 		if (len != 1)
532 			printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
533 		rp->c_replstat = *statp;
534 		break;
535 	case RC_REPLBUFF:
536 		cachv = &rp->c_replvec;
537 		bufsize = len << 2;
538 		cachv->iov_base = kmalloc(bufsize, GFP_KERNEL);
539 		if (!cachv->iov_base) {
540 			nfsd_reply_cache_free(b, rp, nn);
541 			return;
542 		}
543 		cachv->iov_len = bufsize;
544 		memcpy(cachv->iov_base, statp, bufsize);
545 		break;
546 	case RC_NOCACHE:
547 		nfsd_reply_cache_free(b, rp, nn);
548 		return;
549 	}
550 	spin_lock(&b->cache_lock);
551 	nn->drc_mem_usage += bufsize;
552 	lru_put_end(b, rp);
553 	rp->c_secure = test_bit(RQ_SECURE, &rqstp->rq_flags);
554 	rp->c_type = cachetype;
555 	rp->c_state = RC_DONE;
556 	spin_unlock(&b->cache_lock);
557 	return;
558 }
559 
560 /*
561  * Copy cached reply to current reply buffer. Should always fit.
562  * FIXME as reply is in a page, we should just attach the page, and
563  * keep a refcount....
564  */
565 static int
566 nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data)
567 {
568 	struct kvec	*vec = &rqstp->rq_res.head[0];
569 
570 	if (vec->iov_len + data->iov_len > PAGE_SIZE) {
571 		printk(KERN_WARNING "nfsd: cached reply too large (%zd).\n",
572 				data->iov_len);
573 		return 0;
574 	}
575 	memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len);
576 	vec->iov_len += data->iov_len;
577 	return 1;
578 }
579 
580 /*
581  * Note that fields may be added, removed or reordered in the future. Programs
582  * scraping this file for info should test the labels to ensure they're
583  * getting the correct field.
584  */
585 static int nfsd_reply_cache_stats_show(struct seq_file *m, void *v)
586 {
587 	struct nfsd_net *nn = m->private;
588 
589 	seq_printf(m, "max entries:           %u\n", nn->max_drc_entries);
590 	seq_printf(m, "num entries:           %u\n",
591 			atomic_read(&nn->num_drc_entries));
592 	seq_printf(m, "hash buckets:          %u\n", 1 << nn->maskbits);
593 	seq_printf(m, "mem usage:             %u\n", nn->drc_mem_usage);
594 	seq_printf(m, "cache hits:            %u\n", nfsdstats.rchits);
595 	seq_printf(m, "cache misses:          %u\n", nfsdstats.rcmisses);
596 	seq_printf(m, "not cached:            %u\n", nfsdstats.rcnocache);
597 	seq_printf(m, "payload misses:        %u\n", nn->payload_misses);
598 	seq_printf(m, "longest chain len:     %u\n", nn->longest_chain);
599 	seq_printf(m, "cachesize at longest:  %u\n", nn->longest_chain_cachesize);
600 	return 0;
601 }
602 
603 int nfsd_reply_cache_stats_open(struct inode *inode, struct file *file)
604 {
605 	struct nfsd_net *nn = net_generic(file_inode(file)->i_sb->s_fs_info,
606 								nfsd_net_id);
607 
608 	return single_open(file, nfsd_reply_cache_stats_show, nn);
609 }
610