blob: 9186c7ce0b141b187a8b127a6608005a05ad53d1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * 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
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090011#include <linux/slab.h>
Jeff Layton59766872013-02-04 12:50:00 -050012#include <linux/sunrpc/addr.h>
Jeff Layton0338dd12013-02-04 08:18:02 -050013#include <linux/highmem.h>
Jeff Layton0733c7b2013-03-27 10:15:39 -040014#include <linux/log2.h>
15#include <linux/hash.h>
Jeff Layton01a7dec2013-02-04 11:57:27 -050016#include <net/checksum.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090017
Boaz Harrosh9a74af22009-12-03 20:30:56 +020018#include "nfsd.h"
19#include "cache.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020
Jeff Layton0338dd12013-02-04 08:18:02 -050021#define NFSDDBG_FACILITY NFSDDBG_REPCACHE
22
Jeff Layton0733c7b2013-03-27 10:15:39 -040023/*
24 * We use this value to determine the number of hash buckets from the max
25 * cache size, the idea being that when the cache is at its maximum number
26 * of entries, then this should be the average number of entries per bucket.
27 */
28#define TARGET_BUCKET_SIZE 64
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
Greg Banksfca42172009-04-01 07:28:13 +110030static struct hlist_head * cache_hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -070031static struct list_head lru_head;
Jeff Layton8a8bc402013-01-28 14:41:10 -050032static struct kmem_cache *drc_slab;
Jeff Layton9dc56142013-03-27 10:15:37 -040033
34/* max number of entries allowed in the cache */
Jeff Layton0338dd12013-02-04 08:18:02 -050035static unsigned int max_drc_entries;
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Jeff Layton0733c7b2013-03-27 10:15:39 -040037/* number of significant bits in the hash value */
38static unsigned int maskbits;
39
Greg Banksfca42172009-04-01 07:28:13 +110040/*
Jeff Layton9dc56142013-03-27 10:15:37 -040041 * Stats and other tracking of on the duplicate reply cache. All of these and
42 * the "rc" fields in nfsdstats are protected by the cache_lock
43 */
44
45/* total number of entries */
46static unsigned int num_drc_entries;
47
48/* cache misses due only to checksum comparison failures */
49static unsigned int payload_misses;
50
Jeff Layton6c6910c2013-03-27 10:15:38 -040051/* amount of memory (in bytes) currently consumed by the DRC */
52static unsigned int drc_mem_usage;
53
Jeff Layton98d821b2013-03-27 10:15:39 -040054/* longest hash chain seen */
55static unsigned int longest_chain;
56
57/* size of cache when we saw the longest hash chain */
58static unsigned int longest_chain_cachesize;
59
Linus Torvalds1da177e2005-04-16 15:20:36 -070060static int nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec);
Jeff Laytonaca8a232013-02-04 08:18:05 -050061static void cache_cleaner_func(struct work_struct *unused);
Dave Chinner1ab6c492013-08-28 10:18:09 +100062static unsigned long nfsd_reply_cache_count(struct shrinker *shrink,
63 struct shrink_control *sc);
64static unsigned long nfsd_reply_cache_scan(struct shrinker *shrink,
65 struct shrink_control *sc);
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -050066
Wei Yongjunc8c797f2013-04-05 21:22:39 +080067static struct shrinker nfsd_reply_cache_shrinker = {
Dave Chinner1ab6c492013-08-28 10:18:09 +100068 .scan_objects = nfsd_reply_cache_scan,
69 .count_objects = nfsd_reply_cache_count,
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -050070 .seeks = 1,
71};
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Greg Banksfca42172009-04-01 07:28:13 +110073/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070074 * locking for the reply cache:
75 * A cache entry is "single use" if c_state == RC_INPROG
76 * Otherwise, it when accessing _prev or _next, the lock must be held.
77 */
78static DEFINE_SPINLOCK(cache_lock);
Jeff Laytonaca8a232013-02-04 08:18:05 -050079static DECLARE_DELAYED_WORK(cache_cleaner, cache_cleaner_func);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
Jeff Layton0338dd12013-02-04 08:18:02 -050081/*
82 * Put a cap on the size of the DRC based on the amount of available
83 * low memory in the machine.
84 *
85 * 64MB: 8192
86 * 128MB: 11585
87 * 256MB: 16384
88 * 512MB: 23170
89 * 1GB: 32768
90 * 2GB: 46340
91 * 4GB: 65536
92 * 8GB: 92681
93 * 16GB: 131072
94 *
95 * ...with a hard cap of 256k entries. In the worst case, each entry will be
96 * ~1k, so the above numbers should give a rough max of the amount of memory
97 * used in k.
98 */
99static unsigned int
100nfsd_cache_size_limit(void)
101{
102 unsigned int limit;
103 unsigned long low_pages = totalram_pages - totalhigh_pages;
104
105 limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10);
106 return min_t(unsigned int, limit, 256*1024);
107}
108
Jeff Layton0733c7b2013-03-27 10:15:39 -0400109/*
110 * Compute the number of hash buckets we need. Divide the max cachesize by
111 * the "target" max bucket size, and round up to next power of two.
112 */
113static unsigned int
114nfsd_hashsize(unsigned int limit)
115{
116 return roundup_pow_of_two(limit / TARGET_BUCKET_SIZE);
117}
118
Jeff Laytonf09841f2013-01-28 14:41:11 -0500119static struct svc_cacherep *
120nfsd_reply_cache_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
122 struct svc_cacherep *rp;
Jeff Laytonf09841f2013-01-28 14:41:11 -0500123
124 rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
125 if (rp) {
126 rp->c_state = RC_UNUSED;
127 rp->c_type = RC_NOCACHE;
128 INIT_LIST_HEAD(&rp->c_lru);
129 INIT_HLIST_NODE(&rp->c_hash);
130 }
131 return rp;
132}
133
134static void
135nfsd_reply_cache_free_locked(struct svc_cacherep *rp)
136{
Jeff Layton6c6910c2013-03-27 10:15:38 -0400137 if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) {
138 drc_mem_usage -= rp->c_replvec.iov_len;
Jeff Laytonf09841f2013-01-28 14:41:11 -0500139 kfree(rp->c_replvec.iov_base);
Jeff Layton6c6910c2013-03-27 10:15:38 -0400140 }
Jeff Laytona517b602013-03-18 10:49:07 -0400141 if (!hlist_unhashed(&rp->c_hash))
142 hlist_del(&rp->c_hash);
Jeff Laytonf09841f2013-01-28 14:41:11 -0500143 list_del(&rp->c_lru);
Jeff Layton0ee0bf72013-02-04 08:18:01 -0500144 --num_drc_entries;
Jeff Layton6c6910c2013-03-27 10:15:38 -0400145 drc_mem_usage -= sizeof(*rp);
Jeff Laytonf09841f2013-01-28 14:41:11 -0500146 kmem_cache_free(drc_slab, rp);
147}
148
Jeff Layton2c6b6912013-02-04 08:18:04 -0500149static void
150nfsd_reply_cache_free(struct svc_cacherep *rp)
151{
152 spin_lock(&cache_lock);
153 nfsd_reply_cache_free_locked(rp);
154 spin_unlock(&cache_lock);
155}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157int nfsd_reply_cache_init(void)
158{
Jeff Layton0733c7b2013-03-27 10:15:39 -0400159 unsigned int hashsize;
160
Jeff Laytonac534ff2013-03-15 09:16:29 -0400161 INIT_LIST_HEAD(&lru_head);
162 max_drc_entries = nfsd_cache_size_limit();
163 num_drc_entries = 0;
Jeff Layton0733c7b2013-03-27 10:15:39 -0400164 hashsize = nfsd_hashsize(max_drc_entries);
165 maskbits = ilog2(hashsize);
Jeff Laytonac534ff2013-03-15 09:16:29 -0400166
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500167 register_shrinker(&nfsd_reply_cache_shrinker);
Jeff Layton8a8bc402013-01-28 14:41:10 -0500168 drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep),
169 0, 0, NULL);
170 if (!drc_slab)
171 goto out_nomem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Jeff Layton0733c7b2013-03-27 10:15:39 -0400173 cache_hash = kcalloc(hashsize, sizeof(struct hlist_head), GFP_KERNEL);
Greg Banksfca42172009-04-01 07:28:13 +1100174 if (!cache_hash)
J. Bruce Fieldsd5c34282007-11-09 14:10:56 -0500175 goto out_nomem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
J. Bruce Fieldsd5c34282007-11-09 14:10:56 -0500177 return 0;
178out_nomem:
179 printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
180 nfsd_reply_cache_shutdown();
181 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182}
183
J. Bruce Fieldsd5c34282007-11-09 14:10:56 -0500184void nfsd_reply_cache_shutdown(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185{
186 struct svc_cacherep *rp;
187
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500188 unregister_shrinker(&nfsd_reply_cache_shrinker);
Jeff Laytonaca8a232013-02-04 08:18:05 -0500189 cancel_delayed_work_sync(&cache_cleaner);
190
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 while (!list_empty(&lru_head)) {
192 rp = list_entry(lru_head.next, struct svc_cacherep, c_lru);
Jeff Laytonf09841f2013-01-28 14:41:11 -0500193 nfsd_reply_cache_free_locked(rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 }
195
Greg Banksfca42172009-04-01 07:28:13 +1100196 kfree (cache_hash);
197 cache_hash = NULL;
Jeff Layton8a8bc402013-01-28 14:41:10 -0500198
199 if (drc_slab) {
200 kmem_cache_destroy(drc_slab);
201 drc_slab = NULL;
202 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203}
204
205/*
Jeff Laytonaca8a232013-02-04 08:18:05 -0500206 * Move cache entry to end of LRU list, and queue the cleaner to run if it's
207 * not already scheduled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 */
209static void
210lru_put_end(struct svc_cacherep *rp)
211{
Jeff Layton56c25482013-02-04 08:18:00 -0500212 rp->c_timestamp = jiffies;
Akinobu Mitaf1166292006-06-26 00:24:46 -0700213 list_move_tail(&rp->c_lru, &lru_head);
Jeff Laytonaca8a232013-02-04 08:18:05 -0500214 schedule_delayed_work(&cache_cleaner, RC_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215}
216
217/*
218 * Move a cache entry from one hash list to another
219 */
220static void
221hash_refile(struct svc_cacherep *rp)
222{
223 hlist_del_init(&rp->c_hash);
Jeff Layton0733c7b2013-03-27 10:15:39 -0400224 hlist_add_head(&rp->c_hash, cache_hash + hash_32(rp->c_xid, maskbits));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
Jeff Laytond1a07742013-01-28 14:41:13 -0500227static inline bool
228nfsd_cache_entry_expired(struct svc_cacherep *rp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
Jeff Laytond1a07742013-01-28 14:41:13 -0500230 return rp->c_state != RC_INPROG &&
231 time_after(jiffies, rp->c_timestamp + RC_EXPIRE);
232}
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234/*
Jeff Laytonaca8a232013-02-04 08:18:05 -0500235 * Walk the LRU list and prune off entries that are older than RC_EXPIRE.
236 * Also prune the oldest ones when the total exceeds the max number of entries.
237 */
Dave Chinner1ab6c492013-08-28 10:18:09 +1000238static long
Jeff Laytonaca8a232013-02-04 08:18:05 -0500239prune_cache_entries(void)
240{
241 struct svc_cacherep *rp, *tmp;
Dave Chinner1ab6c492013-08-28 10:18:09 +1000242 long freed = 0;
Jeff Laytonaca8a232013-02-04 08:18:05 -0500243
244 list_for_each_entry_safe(rp, tmp, &lru_head, c_lru) {
245 if (!nfsd_cache_entry_expired(rp) &&
246 num_drc_entries <= max_drc_entries)
247 break;
248 nfsd_reply_cache_free_locked(rp);
Dave Chinner1ab6c492013-08-28 10:18:09 +1000249 freed++;
Jeff Laytonaca8a232013-02-04 08:18:05 -0500250 }
251
252 /*
253 * Conditionally rearm the job. If we cleaned out the list, then
254 * cancel any pending run (since there won't be any work to do).
255 * Otherwise, we rearm the job or modify the existing one to run in
256 * RC_EXPIRE since we just ran the pruner.
257 */
258 if (list_empty(&lru_head))
259 cancel_delayed_work(&cache_cleaner);
260 else
261 mod_delayed_work(system_wq, &cache_cleaner, RC_EXPIRE);
Dave Chinner1ab6c492013-08-28 10:18:09 +1000262 return freed;
Jeff Laytonaca8a232013-02-04 08:18:05 -0500263}
264
265static void
266cache_cleaner_func(struct work_struct *unused)
267{
268 spin_lock(&cache_lock);
269 prune_cache_entries();
270 spin_unlock(&cache_lock);
271}
272
Dave Chinner1ab6c492013-08-28 10:18:09 +1000273static unsigned long
274nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc)
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500275{
Dave Chinner1ab6c492013-08-28 10:18:09 +1000276 unsigned long num;
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500277
278 spin_lock(&cache_lock);
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500279 num = num_drc_entries;
280 spin_unlock(&cache_lock);
281
282 return num;
283}
284
Dave Chinner1ab6c492013-08-28 10:18:09 +1000285static unsigned long
286nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
287{
288 unsigned long freed;
289
290 spin_lock(&cache_lock);
291 freed = prune_cache_entries();
292 spin_unlock(&cache_lock);
293 return freed;
294}
Jeff Laytonaca8a232013-02-04 08:18:05 -0500295/*
Jeff Layton01a7dec2013-02-04 11:57:27 -0500296 * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes
297 */
298static __wsum
299nfsd_cache_csum(struct svc_rqst *rqstp)
300{
301 int idx;
302 unsigned int base;
303 __wsum csum;
304 struct xdr_buf *buf = &rqstp->rq_arg;
305 const unsigned char *p = buf->head[0].iov_base;
306 size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len,
307 RC_CSUMLEN);
308 size_t len = min(buf->head[0].iov_len, csum_len);
309
310 /* rq_arg.head first */
311 csum = csum_partial(p, len, 0);
312 csum_len -= len;
313
314 /* Continue into page array */
315 idx = buf->page_base / PAGE_SIZE;
316 base = buf->page_base & ~PAGE_MASK;
317 while (csum_len) {
318 p = page_address(buf->pages[idx]) + base;
Jeff Layton56edc862013-02-15 13:36:34 -0500319 len = min_t(size_t, PAGE_SIZE - base, csum_len);
Jeff Layton01a7dec2013-02-04 11:57:27 -0500320 csum = csum_partial(p, len, csum);
321 csum_len -= len;
322 base = 0;
323 ++idx;
324 }
325 return csum;
326}
327
Jeff Layton9dc56142013-03-27 10:15:37 -0400328static bool
329nfsd_cache_match(struct svc_rqst *rqstp, __wsum csum, struct svc_cacherep *rp)
330{
331 /* Check RPC header info first */
332 if (rqstp->rq_xid != rp->c_xid || rqstp->rq_proc != rp->c_proc ||
333 rqstp->rq_prot != rp->c_prot || rqstp->rq_vers != rp->c_vers ||
334 rqstp->rq_arg.len != rp->c_len ||
335 !rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) ||
336 rpc_get_port(svc_addr(rqstp)) != rpc_get_port((struct sockaddr *)&rp->c_addr))
337 return false;
338
339 /* compare checksum of NFS data */
340 if (csum != rp->c_csum) {
341 ++payload_misses;
342 return false;
343 }
344
345 return true;
346}
347
Jeff Layton01a7dec2013-02-04 11:57:27 -0500348/*
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500349 * Search the request hash for an entry that matches the given rqstp.
350 * Must be called with cache_lock held. Returns the found entry or
351 * NULL on failure.
352 */
353static struct svc_cacherep *
Jeff Layton01a7dec2013-02-04 11:57:27 -0500354nfsd_cache_search(struct svc_rqst *rqstp, __wsum csum)
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500355{
Jeff Layton98d821b2013-03-27 10:15:39 -0400356 struct svc_cacherep *rp, *ret = NULL;
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500357 struct hlist_head *rh;
Jeff Layton98d821b2013-03-27 10:15:39 -0400358 unsigned int entries = 0;
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500359
Jeff Layton0733c7b2013-03-27 10:15:39 -0400360 rh = &cache_hash[hash_32(rqstp->rq_xid, maskbits)];
Linus Torvaldsb6669732013-02-28 18:02:55 -0800361 hlist_for_each_entry(rp, rh, c_hash) {
Jeff Layton98d821b2013-03-27 10:15:39 -0400362 ++entries;
363 if (nfsd_cache_match(rqstp, csum, rp)) {
364 ret = rp;
365 break;
366 }
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500367 }
Jeff Layton98d821b2013-03-27 10:15:39 -0400368
369 /* tally hash chain length stats */
370 if (entries > longest_chain) {
371 longest_chain = entries;
372 longest_chain_cachesize = num_drc_entries;
373 } else if (entries == longest_chain) {
374 /* prefer to keep the smallest cachesize possible here */
375 longest_chain_cachesize = min(longest_chain_cachesize,
376 num_drc_entries);
377 }
378
379 return ret;
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500380}
381
382/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 * Try to find an entry matching the current call in the cache. When none
Jeff Layton1ac83622013-02-14 16:45:13 -0500384 * is found, we try to grab the oldest expired entry off the LRU list. If
385 * a suitable one isn't there, then drop the cache_lock and allocate a
386 * new one, then search again in case one got inserted while this thread
387 * didn't hold the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 */
389int
390nfsd_cache_lookup(struct svc_rqst *rqstp)
391{
Jeff Layton0338dd12013-02-04 08:18:02 -0500392 struct svc_cacherep *rp, *found;
Al Viroc7afef12006-10-19 23:29:02 -0700393 __be32 xid = rqstp->rq_xid;
394 u32 proto = rqstp->rq_prot,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 vers = rqstp->rq_vers,
396 proc = rqstp->rq_proc;
Jeff Layton01a7dec2013-02-04 11:57:27 -0500397 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 unsigned long age;
J. Bruce Fields10910062011-01-24 12:11:02 -0500399 int type = rqstp->rq_cachetype;
Jeff Layton0b9ea372013-03-27 10:15:37 -0400400 int rtn = RC_DOIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
402 rqstp->rq_cacherep = NULL;
Jeff Layton13cc8a72013-02-04 08:18:03 -0500403 if (type == RC_NOCACHE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 nfsdstats.rcnocache++;
Jeff Layton0b9ea372013-03-27 10:15:37 -0400405 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 }
407
Jeff Layton01a7dec2013-02-04 11:57:27 -0500408 csum = nfsd_cache_csum(rqstp);
409
Jeff Layton0b9ea372013-03-27 10:15:37 -0400410 /*
411 * Since the common case is a cache miss followed by an insert,
412 * preallocate an entry. First, try to reuse the first entry on the LRU
413 * if it works, then go ahead and prune the LRU list.
414 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 spin_lock(&cache_lock);
Jeff Layton0338dd12013-02-04 08:18:02 -0500416 if (!list_empty(&lru_head)) {
417 rp = list_first_entry(&lru_head, struct svc_cacherep, c_lru);
418 if (nfsd_cache_entry_expired(rp) ||
Jeff Laytonaca8a232013-02-04 08:18:05 -0500419 num_drc_entries >= max_drc_entries) {
420 lru_put_end(rp);
421 prune_cache_entries();
Jeff Layton0b9ea372013-03-27 10:15:37 -0400422 goto search_cache;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
424 }
Jeff Layton0338dd12013-02-04 08:18:02 -0500425
Jeff Layton0b9ea372013-03-27 10:15:37 -0400426 /* No expired ones available, allocate a new one. */
Jeff Layton0338dd12013-02-04 08:18:02 -0500427 spin_unlock(&cache_lock);
428 rp = nfsd_reply_cache_alloc();
Jeff Layton0338dd12013-02-04 08:18:02 -0500429 spin_lock(&cache_lock);
Jeff Layton6c6910c2013-03-27 10:15:38 -0400430 if (likely(rp)) {
Jeff Layton0b9ea372013-03-27 10:15:37 -0400431 ++num_drc_entries;
Jeff Layton6c6910c2013-03-27 10:15:38 -0400432 drc_mem_usage += sizeof(*rp);
433 }
Jeff Layton0338dd12013-02-04 08:18:02 -0500434
Jeff Layton0b9ea372013-03-27 10:15:37 -0400435search_cache:
Jeff Layton01a7dec2013-02-04 11:57:27 -0500436 found = nfsd_cache_search(rqstp, csum);
Jeff Layton0338dd12013-02-04 08:18:02 -0500437 if (found) {
Jeff Layton0b9ea372013-03-27 10:15:37 -0400438 if (likely(rp))
439 nfsd_reply_cache_free_locked(rp);
Jeff Layton0338dd12013-02-04 08:18:02 -0500440 rp = found;
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500441 goto found_entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 }
Jeff Layton0338dd12013-02-04 08:18:02 -0500443
Jeff Layton0b9ea372013-03-27 10:15:37 -0400444 if (!rp) {
445 dprintk("nfsd: unable to allocate DRC entry!\n");
446 goto out;
447 }
448
Jeff Layton0338dd12013-02-04 08:18:02 -0500449 /*
450 * We're keeping the one we just allocated. Are we now over the
451 * limit? Prune one off the tip of the LRU in trade for the one we
452 * just allocated if so.
453 */
454 if (num_drc_entries >= max_drc_entries)
455 nfsd_reply_cache_free_locked(list_first_entry(&lru_head,
456 struct svc_cacherep, c_lru));
457
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 nfsdstats.rcmisses++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 rqstp->rq_cacherep = rp;
460 rp->c_state = RC_INPROG;
461 rp->c_xid = xid;
462 rp->c_proc = proc;
Jeff Layton7b9e8522013-01-28 14:41:07 -0500463 rpc_copy_addr((struct sockaddr *)&rp->c_addr, svc_addr(rqstp));
464 rpc_set_port((struct sockaddr *)&rp->c_addr, rpc_get_port(svc_addr(rqstp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 rp->c_prot = proto;
466 rp->c_vers = vers;
Jeff Layton01a7dec2013-02-04 11:57:27 -0500467 rp->c_len = rqstp->rq_arg.len;
468 rp->c_csum = csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 hash_refile(rp);
Jeff Layton56c25482013-02-04 08:18:00 -0500471 lru_put_end(rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 /* release any buffer */
474 if (rp->c_type == RC_REPLBUFF) {
Jeff Layton6c6910c2013-03-27 10:15:38 -0400475 drc_mem_usage -= rp->c_replvec.iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 kfree(rp->c_replvec.iov_base);
477 rp->c_replvec.iov_base = NULL;
478 }
479 rp->c_type = RC_NOCACHE;
480 out:
481 spin_unlock(&cache_lock);
482 return rtn;
483
484found_entry:
Jeff Layton0338dd12013-02-04 08:18:02 -0500485 nfsdstats.rchits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 /* We found a matching entry which is either in progress or done. */
487 age = jiffies - rp->c_timestamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 lru_put_end(rp);
489
490 rtn = RC_DROPIT;
491 /* Request being processed or excessive rexmits */
492 if (rp->c_state == RC_INPROG || age < RC_DELAY)
493 goto out;
494
495 /* From the hall of fame of impractical attacks:
496 * Is this a user who tries to snoop on the cache? */
497 rtn = RC_DOIT;
498 if (!rqstp->rq_secure && rp->c_secure)
499 goto out;
500
501 /* Compose RPC reply header */
502 switch (rp->c_type) {
503 case RC_NOCACHE:
504 break;
505 case RC_REPLSTAT:
506 svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
507 rtn = RC_REPLY;
508 break;
509 case RC_REPLBUFF:
510 if (!nfsd_cache_append(rqstp, &rp->c_replvec))
511 goto out; /* should not happen */
512 rtn = RC_REPLY;
513 break;
514 default:
515 printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
Jeff Layton0338dd12013-02-04 08:18:02 -0500516 nfsd_reply_cache_free_locked(rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 }
518
519 goto out;
520}
521
522/*
523 * Update a cache entry. This is called from nfsd_dispatch when
524 * the procedure has been executed and the complete reply is in
525 * rqstp->rq_res.
526 *
527 * We're copying around data here rather than swapping buffers because
528 * the toplevel loop requires max-sized buffers, which would be a waste
529 * of memory for a cache with a max reply size of 100 bytes (diropokres).
530 *
531 * If we should start to use different types of cache entries tailored
532 * specifically for attrstat and fh's, we may save even more space.
533 *
534 * Also note that a cachetype of RC_NOCACHE can legally be passed when
535 * nfsd failed to encode a reply that otherwise would have been cached.
536 * In this case, nfsd_cache_update is called with statp == NULL.
537 */
538void
Al Viroc7afef12006-10-19 23:29:02 -0700539nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540{
Jeff Layton13cc8a72013-02-04 08:18:03 -0500541 struct svc_cacherep *rp = rqstp->rq_cacherep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 struct kvec *resv = &rqstp->rq_res.head[0], *cachv;
543 int len;
Jeff Layton6c6910c2013-03-27 10:15:38 -0400544 size_t bufsize = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Jeff Layton13cc8a72013-02-04 08:18:03 -0500546 if (!rp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 return;
548
549 len = resv->iov_len - ((char*)statp - (char*)resv->iov_base);
550 len >>= 2;
Greg Banksfca42172009-04-01 07:28:13 +1100551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 /* Don't cache excessive amounts of data and XDR failures */
553 if (!statp || len > (256 >> 2)) {
Jeff Layton2c6b6912013-02-04 08:18:04 -0500554 nfsd_reply_cache_free(rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 return;
556 }
557
558 switch (cachetype) {
559 case RC_REPLSTAT:
560 if (len != 1)
561 printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
562 rp->c_replstat = *statp;
563 break;
564 case RC_REPLBUFF:
565 cachv = &rp->c_replvec;
Jeff Layton6c6910c2013-03-27 10:15:38 -0400566 bufsize = len << 2;
567 cachv->iov_base = kmalloc(bufsize, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 if (!cachv->iov_base) {
Jeff Layton2c6b6912013-02-04 08:18:04 -0500569 nfsd_reply_cache_free(rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 return;
571 }
Jeff Layton6c6910c2013-03-27 10:15:38 -0400572 cachv->iov_len = bufsize;
573 memcpy(cachv->iov_base, statp, bufsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 break;
Jeff Layton2c6b6912013-02-04 08:18:04 -0500575 case RC_NOCACHE:
576 nfsd_reply_cache_free(rp);
577 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
579 spin_lock(&cache_lock);
Jeff Layton6c6910c2013-03-27 10:15:38 -0400580 drc_mem_usage += bufsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 lru_put_end(rp);
582 rp->c_secure = rqstp->rq_secure;
583 rp->c_type = cachetype;
584 rp->c_state = RC_DONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 spin_unlock(&cache_lock);
586 return;
587}
588
589/*
590 * Copy cached reply to current reply buffer. Should always fit.
591 * FIXME as reply is in a page, we should just attach the page, and
592 * keep a refcount....
593 */
594static int
595nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data)
596{
597 struct kvec *vec = &rqstp->rq_res.head[0];
598
599 if (vec->iov_len + data->iov_len > PAGE_SIZE) {
600 printk(KERN_WARNING "nfsd: cached reply too large (%Zd).\n",
601 data->iov_len);
602 return 0;
603 }
604 memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len);
605 vec->iov_len += data->iov_len;
606 return 1;
607}
Jeff Laytona2f999a2013-03-27 10:15:38 -0400608
609/*
610 * Note that fields may be added, removed or reordered in the future. Programs
611 * scraping this file for info should test the labels to ensure they're
612 * getting the correct field.
613 */
614static int nfsd_reply_cache_stats_show(struct seq_file *m, void *v)
615{
616 spin_lock(&cache_lock);
617 seq_printf(m, "max entries: %u\n", max_drc_entries);
618 seq_printf(m, "num entries: %u\n", num_drc_entries);
Jeff Layton0733c7b2013-03-27 10:15:39 -0400619 seq_printf(m, "hash buckets: %u\n", 1 << maskbits);
Jeff Laytona2f999a2013-03-27 10:15:38 -0400620 seq_printf(m, "mem usage: %u\n", drc_mem_usage);
621 seq_printf(m, "cache hits: %u\n", nfsdstats.rchits);
622 seq_printf(m, "cache misses: %u\n", nfsdstats.rcmisses);
623 seq_printf(m, "not cached: %u\n", nfsdstats.rcnocache);
624 seq_printf(m, "payload misses: %u\n", payload_misses);
Jeff Layton98d821b2013-03-27 10:15:39 -0400625 seq_printf(m, "longest chain len: %u\n", longest_chain);
626 seq_printf(m, "cachesize at longest: %u\n", longest_chain_cachesize);
Jeff Laytona2f999a2013-03-27 10:15:38 -0400627 spin_unlock(&cache_lock);
628 return 0;
629}
630
631int nfsd_reply_cache_stats_open(struct inode *inode, struct file *file)
632{
633 return single_open(file, nfsd_reply_cache_stats_show, NULL);
634}