blob: 122f69185ef511414f73e2bdcf2174ca8b2488c7 [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
Trond Myklebust7142b982014-08-06 13:44:20 -040030struct nfsd_drc_bucket {
Trond Myklebustbedd4b62014-08-06 13:44:21 -040031 struct list_head lru_head;
Trond Myklebust89a26b32014-08-06 13:44:24 -040032 spinlock_t cache_lock;
Trond Myklebust7142b982014-08-06 13:44:20 -040033};
34
35static struct nfsd_drc_bucket *drc_hashtbl;
Jeff Layton8a8bc402013-01-28 14:41:10 -050036static struct kmem_cache *drc_slab;
Jeff Layton9dc56142013-03-27 10:15:37 -040037
38/* max number of entries allowed in the cache */
Jeff Layton0338dd12013-02-04 08:18:02 -050039static unsigned int max_drc_entries;
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
Jeff Layton0733c7b2013-03-27 10:15:39 -040041/* number of significant bits in the hash value */
42static unsigned int maskbits;
Trond Myklebustbedd4b62014-08-06 13:44:21 -040043static unsigned int drc_hashsize;
Jeff Layton0733c7b2013-03-27 10:15:39 -040044
Greg Banksfca42172009-04-01 07:28:13 +110045/*
Jeff Layton9dc56142013-03-27 10:15:37 -040046 * Stats and other tracking of on the duplicate reply cache. All of these and
47 * the "rc" fields in nfsdstats are protected by the cache_lock
48 */
49
50/* total number of entries */
Trond Myklebust31e60f52014-08-06 13:44:23 -040051static atomic_t num_drc_entries;
Jeff Layton9dc56142013-03-27 10:15:37 -040052
53/* cache misses due only to checksum comparison failures */
54static unsigned int payload_misses;
55
Jeff Layton6c6910c2013-03-27 10:15:38 -040056/* amount of memory (in bytes) currently consumed by the DRC */
57static unsigned int drc_mem_usage;
58
Jeff Layton98d821b2013-03-27 10:15:39 -040059/* longest hash chain seen */
60static unsigned int longest_chain;
61
62/* size of cache when we saw the longest hash chain */
63static unsigned int longest_chain_cachesize;
64
Linus Torvalds1da177e2005-04-16 15:20:36 -070065static int nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec);
Jeff Laytonaca8a232013-02-04 08:18:05 -050066static void cache_cleaner_func(struct work_struct *unused);
Dave Chinner1ab6c492013-08-28 10:18:09 +100067static unsigned long nfsd_reply_cache_count(struct shrinker *shrink,
68 struct shrink_control *sc);
69static unsigned long nfsd_reply_cache_scan(struct shrinker *shrink,
70 struct shrink_control *sc);
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -050071
Wei Yongjunc8c797f2013-04-05 21:22:39 +080072static struct shrinker nfsd_reply_cache_shrinker = {
Dave Chinner1ab6c492013-08-28 10:18:09 +100073 .scan_objects = nfsd_reply_cache_scan,
74 .count_objects = nfsd_reply_cache_count,
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -050075 .seeks = 1,
76};
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Greg Banksfca42172009-04-01 07:28:13 +110078/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * locking for the reply cache:
80 * A cache entry is "single use" if c_state == RC_INPROG
81 * Otherwise, it when accessing _prev or _next, the lock must be held.
82 */
Jeff Laytonaca8a232013-02-04 08:18:05 -050083static DECLARE_DELAYED_WORK(cache_cleaner, cache_cleaner_func);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
Jeff Layton0338dd12013-02-04 08:18:02 -050085/*
86 * Put a cap on the size of the DRC based on the amount of available
87 * low memory in the machine.
88 *
89 * 64MB: 8192
90 * 128MB: 11585
91 * 256MB: 16384
92 * 512MB: 23170
93 * 1GB: 32768
94 * 2GB: 46340
95 * 4GB: 65536
96 * 8GB: 92681
97 * 16GB: 131072
98 *
99 * ...with a hard cap of 256k entries. In the worst case, each entry will be
100 * ~1k, so the above numbers should give a rough max of the amount of memory
101 * used in k.
102 */
103static unsigned int
104nfsd_cache_size_limit(void)
105{
106 unsigned int limit;
107 unsigned long low_pages = totalram_pages - totalhigh_pages;
108
109 limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10);
110 return min_t(unsigned int, limit, 256*1024);
111}
112
Jeff Layton0733c7b2013-03-27 10:15:39 -0400113/*
114 * Compute the number of hash buckets we need. Divide the max cachesize by
115 * the "target" max bucket size, and round up to next power of two.
116 */
117static unsigned int
118nfsd_hashsize(unsigned int limit)
119{
120 return roundup_pow_of_two(limit / TARGET_BUCKET_SIZE);
121}
122
Trond Myklebust7142b982014-08-06 13:44:20 -0400123static u32
124nfsd_cache_hash(__be32 xid)
125{
126 return hash_32(be32_to_cpu(xid), maskbits);
127}
128
Jeff Laytonf09841f2013-01-28 14:41:11 -0500129static struct svc_cacherep *
130nfsd_reply_cache_alloc(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131{
132 struct svc_cacherep *rp;
Jeff Laytonf09841f2013-01-28 14:41:11 -0500133
134 rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
135 if (rp) {
136 rp->c_state = RC_UNUSED;
137 rp->c_type = RC_NOCACHE;
138 INIT_LIST_HEAD(&rp->c_lru);
Jeff Laytonf09841f2013-01-28 14:41:11 -0500139 }
140 return rp;
141}
142
143static void
144nfsd_reply_cache_free_locked(struct svc_cacherep *rp)
145{
Jeff Layton6c6910c2013-03-27 10:15:38 -0400146 if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) {
147 drc_mem_usage -= rp->c_replvec.iov_len;
Jeff Laytonf09841f2013-01-28 14:41:11 -0500148 kfree(rp->c_replvec.iov_base);
Jeff Layton6c6910c2013-03-27 10:15:38 -0400149 }
Jeff Laytonf09841f2013-01-28 14:41:11 -0500150 list_del(&rp->c_lru);
Trond Myklebust31e60f52014-08-06 13:44:23 -0400151 atomic_dec(&num_drc_entries);
Jeff Layton6c6910c2013-03-27 10:15:38 -0400152 drc_mem_usage -= sizeof(*rp);
Jeff Laytonf09841f2013-01-28 14:41:11 -0500153 kmem_cache_free(drc_slab, rp);
154}
155
Jeff Layton2c6b6912013-02-04 08:18:04 -0500156static void
Trond Myklebust89a26b32014-08-06 13:44:24 -0400157nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct svc_cacherep *rp)
Jeff Layton2c6b6912013-02-04 08:18:04 -0500158{
Trond Myklebust89a26b32014-08-06 13:44:24 -0400159 spin_lock(&b->cache_lock);
Jeff Layton2c6b6912013-02-04 08:18:04 -0500160 nfsd_reply_cache_free_locked(rp);
Trond Myklebust89a26b32014-08-06 13:44:24 -0400161 spin_unlock(&b->cache_lock);
Jeff Layton2c6b6912013-02-04 08:18:04 -0500162}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164int nfsd_reply_cache_init(void)
165{
Jeff Layton0733c7b2013-03-27 10:15:39 -0400166 unsigned int hashsize;
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400167 unsigned int i;
Jeff Layton0733c7b2013-03-27 10:15:39 -0400168
Jeff Laytonac534ff2013-03-15 09:16:29 -0400169 max_drc_entries = nfsd_cache_size_limit();
Trond Myklebust31e60f52014-08-06 13:44:23 -0400170 atomic_set(&num_drc_entries, 0);
Jeff Layton0733c7b2013-03-27 10:15:39 -0400171 hashsize = nfsd_hashsize(max_drc_entries);
172 maskbits = ilog2(hashsize);
Jeff Laytonac534ff2013-03-15 09:16:29 -0400173
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500174 register_shrinker(&nfsd_reply_cache_shrinker);
Jeff Layton8a8bc402013-01-28 14:41:10 -0500175 drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep),
176 0, 0, NULL);
177 if (!drc_slab)
178 goto out_nomem;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
Trond Myklebust7142b982014-08-06 13:44:20 -0400180 drc_hashtbl = kcalloc(hashsize, sizeof(*drc_hashtbl), GFP_KERNEL);
181 if (!drc_hashtbl)
J. Bruce Fieldsd5c34282007-11-09 14:10:56 -0500182 goto out_nomem;
Trond Myklebust89a26b32014-08-06 13:44:24 -0400183 for (i = 0; i < hashsize; i++) {
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400184 INIT_LIST_HEAD(&drc_hashtbl[i].lru_head);
Trond Myklebust89a26b32014-08-06 13:44:24 -0400185 spin_lock_init(&drc_hashtbl[i].cache_lock);
186 }
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400187 drc_hashsize = hashsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
J. Bruce Fieldsd5c34282007-11-09 14:10:56 -0500189 return 0;
190out_nomem:
191 printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
192 nfsd_reply_cache_shutdown();
193 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194}
195
J. Bruce Fieldsd5c34282007-11-09 14:10:56 -0500196void nfsd_reply_cache_shutdown(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
198 struct svc_cacherep *rp;
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400199 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500201 unregister_shrinker(&nfsd_reply_cache_shrinker);
Jeff Laytonaca8a232013-02-04 08:18:05 -0500202 cancel_delayed_work_sync(&cache_cleaner);
203
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400204 for (i = 0; i < drc_hashsize; i++) {
205 struct list_head *head = &drc_hashtbl[i].lru_head;
206 while (!list_empty(head)) {
207 rp = list_first_entry(head, struct svc_cacherep, c_lru);
208 nfsd_reply_cache_free_locked(rp);
209 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 }
211
Trond Myklebust7142b982014-08-06 13:44:20 -0400212 kfree (drc_hashtbl);
213 drc_hashtbl = NULL;
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400214 drc_hashsize = 0;
Jeff Layton8a8bc402013-01-28 14:41:10 -0500215
216 if (drc_slab) {
217 kmem_cache_destroy(drc_slab);
218 drc_slab = NULL;
219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220}
221
222/*
Jeff Laytonaca8a232013-02-04 08:18:05 -0500223 * Move cache entry to end of LRU list, and queue the cleaner to run if it's
224 * not already scheduled.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 */
226static void
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400227lru_put_end(struct nfsd_drc_bucket *b, struct svc_cacherep *rp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228{
Jeff Layton56c25482013-02-04 08:18:00 -0500229 rp->c_timestamp = jiffies;
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400230 list_move_tail(&rp->c_lru, &b->lru_head);
Jeff Laytonaca8a232013-02-04 08:18:05 -0500231 schedule_delayed_work(&cache_cleaner, RC_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
Dave Chinner1ab6c492013-08-28 10:18:09 +1000234static long
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400235prune_bucket(struct nfsd_drc_bucket *b)
Jeff Laytonaca8a232013-02-04 08:18:05 -0500236{
237 struct svc_cacherep *rp, *tmp;
Dave Chinner1ab6c492013-08-28 10:18:09 +1000238 long freed = 0;
Jeff Laytonaca8a232013-02-04 08:18:05 -0500239
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400240 list_for_each_entry_safe(rp, tmp, &b->lru_head, c_lru) {
Jeff Layton1b194532014-06-05 09:45:00 -0400241 /*
242 * Don't free entries attached to calls that are still
243 * in-progress, but do keep scanning the list.
244 */
245 if (rp->c_state == RC_INPROG)
246 continue;
Trond Myklebust31e60f52014-08-06 13:44:23 -0400247 if (atomic_read(&num_drc_entries) <= max_drc_entries &&
Jeff Layton1b194532014-06-05 09:45:00 -0400248 time_before(jiffies, rp->c_timestamp + RC_EXPIRE))
Jeff Laytonaca8a232013-02-04 08:18:05 -0500249 break;
250 nfsd_reply_cache_free_locked(rp);
Dave Chinner1ab6c492013-08-28 10:18:09 +1000251 freed++;
Jeff Laytonaca8a232013-02-04 08:18:05 -0500252 }
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400253 return freed;
254}
255
256/*
257 * Walk the LRU list and prune off entries that are older than RC_EXPIRE.
258 * Also prune the oldest ones when the total exceeds the max number of entries.
259 */
260static long
261prune_cache_entries(void)
262{
263 unsigned int i;
264 long freed = 0;
265 bool cancel = true;
266
267 for (i = 0; i < drc_hashsize; i++) {
268 struct nfsd_drc_bucket *b = &drc_hashtbl[i];
269
Trond Myklebust89a26b32014-08-06 13:44:24 -0400270 if (list_empty(&b->lru_head))
271 continue;
272 spin_lock(&b->cache_lock);
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400273 freed += prune_bucket(b);
274 if (!list_empty(&b->lru_head))
275 cancel = false;
Trond Myklebust89a26b32014-08-06 13:44:24 -0400276 spin_unlock(&b->cache_lock);
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400277 }
Jeff Laytonaca8a232013-02-04 08:18:05 -0500278
279 /*
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400280 * Conditionally rearm the job to run in RC_EXPIRE since we just
281 * ran the pruner.
Jeff Laytonaca8a232013-02-04 08:18:05 -0500282 */
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400283 if (!cancel)
Jeff Laytonaca8a232013-02-04 08:18:05 -0500284 mod_delayed_work(system_wq, &cache_cleaner, RC_EXPIRE);
Dave Chinner1ab6c492013-08-28 10:18:09 +1000285 return freed;
Jeff Laytonaca8a232013-02-04 08:18:05 -0500286}
287
288static void
289cache_cleaner_func(struct work_struct *unused)
290{
Jeff Laytonaca8a232013-02-04 08:18:05 -0500291 prune_cache_entries();
Jeff Laytonaca8a232013-02-04 08:18:05 -0500292}
293
Dave Chinner1ab6c492013-08-28 10:18:09 +1000294static unsigned long
295nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc)
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500296{
Trond Myklebust31e60f52014-08-06 13:44:23 -0400297 return atomic_read(&num_drc_entries);
Jeff Laytonb4e7f2c2013-02-04 08:18:06 -0500298}
299
Dave Chinner1ab6c492013-08-28 10:18:09 +1000300static unsigned long
301nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
302{
Trond Myklebust89a26b32014-08-06 13:44:24 -0400303 return prune_cache_entries();
Dave Chinner1ab6c492013-08-28 10:18:09 +1000304}
Jeff Laytonaca8a232013-02-04 08:18:05 -0500305/*
Jeff Layton01a7dec2013-02-04 11:57:27 -0500306 * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes
307 */
308static __wsum
309nfsd_cache_csum(struct svc_rqst *rqstp)
310{
311 int idx;
312 unsigned int base;
313 __wsum csum;
314 struct xdr_buf *buf = &rqstp->rq_arg;
315 const unsigned char *p = buf->head[0].iov_base;
316 size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len,
317 RC_CSUMLEN);
318 size_t len = min(buf->head[0].iov_len, csum_len);
319
320 /* rq_arg.head first */
321 csum = csum_partial(p, len, 0);
322 csum_len -= len;
323
324 /* Continue into page array */
325 idx = buf->page_base / PAGE_SIZE;
326 base = buf->page_base & ~PAGE_MASK;
327 while (csum_len) {
328 p = page_address(buf->pages[idx]) + base;
Jeff Layton56edc862013-02-15 13:36:34 -0500329 len = min_t(size_t, PAGE_SIZE - base, csum_len);
Jeff Layton01a7dec2013-02-04 11:57:27 -0500330 csum = csum_partial(p, len, csum);
331 csum_len -= len;
332 base = 0;
333 ++idx;
334 }
335 return csum;
336}
337
Jeff Layton9dc56142013-03-27 10:15:37 -0400338static bool
339nfsd_cache_match(struct svc_rqst *rqstp, __wsum csum, struct svc_cacherep *rp)
340{
Trond Myklebustef9b16d2014-08-06 13:44:25 -0400341 /* Check RPC XID first */
342 if (rqstp->rq_xid != rp->c_xid)
Jeff Layton9dc56142013-03-27 10:15:37 -0400343 return false;
Jeff Layton9dc56142013-03-27 10:15:37 -0400344 /* compare checksum of NFS data */
345 if (csum != rp->c_csum) {
346 ++payload_misses;
347 return false;
348 }
349
Trond Myklebustef9b16d2014-08-06 13:44:25 -0400350 /* Other discriminators */
351 if (rqstp->rq_proc != rp->c_proc ||
352 rqstp->rq_prot != rp->c_prot ||
353 rqstp->rq_vers != rp->c_vers ||
354 rqstp->rq_arg.len != rp->c_len ||
355 !rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) ||
356 rpc_get_port(svc_addr(rqstp)) != rpc_get_port((struct sockaddr *)&rp->c_addr))
357 return false;
358
Jeff Layton9dc56142013-03-27 10:15:37 -0400359 return true;
360}
361
Jeff Layton01a7dec2013-02-04 11:57:27 -0500362/*
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500363 * Search the request hash for an entry that matches the given rqstp.
364 * Must be called with cache_lock held. Returns the found entry or
365 * NULL on failure.
366 */
367static struct svc_cacherep *
Trond Myklebust7142b982014-08-06 13:44:20 -0400368nfsd_cache_search(struct nfsd_drc_bucket *b, struct svc_rqst *rqstp,
369 __wsum csum)
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500370{
Jeff Layton98d821b2013-03-27 10:15:39 -0400371 struct svc_cacherep *rp, *ret = NULL;
Trond Myklebust11acf6e2014-08-06 13:44:22 -0400372 struct list_head *rh = &b->lru_head;
Jeff Layton98d821b2013-03-27 10:15:39 -0400373 unsigned int entries = 0;
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500374
Trond Myklebust11acf6e2014-08-06 13:44:22 -0400375 list_for_each_entry(rp, rh, c_lru) {
Jeff Layton98d821b2013-03-27 10:15:39 -0400376 ++entries;
377 if (nfsd_cache_match(rqstp, csum, rp)) {
378 ret = rp;
379 break;
380 }
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500381 }
Jeff Layton98d821b2013-03-27 10:15:39 -0400382
383 /* tally hash chain length stats */
384 if (entries > longest_chain) {
385 longest_chain = entries;
Trond Myklebust31e60f52014-08-06 13:44:23 -0400386 longest_chain_cachesize = atomic_read(&num_drc_entries);
Jeff Layton98d821b2013-03-27 10:15:39 -0400387 } else if (entries == longest_chain) {
388 /* prefer to keep the smallest cachesize possible here */
Trond Myklebust31e60f52014-08-06 13:44:23 -0400389 longest_chain_cachesize = min_t(unsigned int,
390 longest_chain_cachesize,
391 atomic_read(&num_drc_entries));
Jeff Layton98d821b2013-03-27 10:15:39 -0400392 }
393
394 return ret;
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500395}
396
397/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 * Try to find an entry matching the current call in the cache. When none
Jeff Layton1ac83622013-02-14 16:45:13 -0500399 * is found, we try to grab the oldest expired entry off the LRU list. If
400 * a suitable one isn't there, then drop the cache_lock and allocate a
401 * new one, then search again in case one got inserted while this thread
402 * didn't hold the lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 */
404int
J. Bruce Fields10910062011-01-24 12:11:02 -0500405nfsd_cache_lookup(struct svc_rqst *rqstp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
Jeff Layton0338dd12013-02-04 08:18:02 -0500407 struct svc_cacherep *rp, *found;
Al Viroc7afef12006-10-19 23:29:02 -0700408 __be32 xid = rqstp->rq_xid;
409 u32 proto = rqstp->rq_prot,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 vers = rqstp->rq_vers,
411 proc = rqstp->rq_proc;
Jeff Layton01a7dec2013-02-04 11:57:27 -0500412 __wsum csum;
Trond Myklebust7142b982014-08-06 13:44:20 -0400413 u32 hash = nfsd_cache_hash(xid);
414 struct nfsd_drc_bucket *b = &drc_hashtbl[hash];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 unsigned long age;
J. Bruce Fields10910062011-01-24 12:11:02 -0500416 int type = rqstp->rq_cachetype;
Jeff Layton0b9ea372013-03-27 10:15:37 -0400417 int rtn = RC_DOIT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
419 rqstp->rq_cacherep = NULL;
Jeff Layton13cc8a72013-02-04 08:18:03 -0500420 if (type == RC_NOCACHE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 nfsdstats.rcnocache++;
Jeff Layton0b9ea372013-03-27 10:15:37 -0400422 return rtn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 }
424
Jeff Layton01a7dec2013-02-04 11:57:27 -0500425 csum = nfsd_cache_csum(rqstp);
426
Jeff Layton0b9ea372013-03-27 10:15:37 -0400427 /*
428 * Since the common case is a cache miss followed by an insert,
Jeff Laytona0ef5e192013-12-05 06:00:51 -0500429 * preallocate an entry.
Jeff Layton0b9ea372013-03-27 10:15:37 -0400430 */
Jeff Layton0338dd12013-02-04 08:18:02 -0500431 rp = nfsd_reply_cache_alloc();
Trond Myklebust89a26b32014-08-06 13:44:24 -0400432 spin_lock(&b->cache_lock);
Jeff Layton6c6910c2013-03-27 10:15:38 -0400433 if (likely(rp)) {
Trond Myklebust31e60f52014-08-06 13:44:23 -0400434 atomic_inc(&num_drc_entries);
Jeff Layton6c6910c2013-03-27 10:15:38 -0400435 drc_mem_usage += sizeof(*rp);
436 }
Jeff Layton0338dd12013-02-04 08:18:02 -0500437
Jeff Laytona0ef5e192013-12-05 06:00:51 -0500438 /* go ahead and prune the cache */
Trond Myklebust89a26b32014-08-06 13:44:24 -0400439 prune_bucket(b);
Jeff Laytona0ef5e192013-12-05 06:00:51 -0500440
Trond Myklebust7142b982014-08-06 13:44:20 -0400441 found = nfsd_cache_search(b, rqstp, csum);
Jeff Layton0338dd12013-02-04 08:18:02 -0500442 if (found) {
Jeff Layton0b9ea372013-03-27 10:15:37 -0400443 if (likely(rp))
444 nfsd_reply_cache_free_locked(rp);
Jeff Layton0338dd12013-02-04 08:18:02 -0500445 rp = found;
Jeff Laytona4a3ec32013-01-28 14:41:14 -0500446 goto found_entry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 }
Jeff Layton0338dd12013-02-04 08:18:02 -0500448
Jeff Layton0b9ea372013-03-27 10:15:37 -0400449 if (!rp) {
450 dprintk("nfsd: unable to allocate DRC entry!\n");
451 goto out;
452 }
453
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 nfsdstats.rcmisses++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 rqstp->rq_cacherep = rp;
456 rp->c_state = RC_INPROG;
457 rp->c_xid = xid;
458 rp->c_proc = proc;
Jeff Layton7b9e8522013-01-28 14:41:07 -0500459 rpc_copy_addr((struct sockaddr *)&rp->c_addr, svc_addr(rqstp));
460 rpc_set_port((struct sockaddr *)&rp->c_addr, rpc_get_port(svc_addr(rqstp)));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 rp->c_prot = proto;
462 rp->c_vers = vers;
Jeff Layton01a7dec2013-02-04 11:57:27 -0500463 rp->c_len = rqstp->rq_arg.len;
464 rp->c_csum = csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400466 lru_put_end(b, rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
468 /* release any buffer */
469 if (rp->c_type == RC_REPLBUFF) {
Jeff Layton6c6910c2013-03-27 10:15:38 -0400470 drc_mem_usage -= rp->c_replvec.iov_len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 kfree(rp->c_replvec.iov_base);
472 rp->c_replvec.iov_base = NULL;
473 }
474 rp->c_type = RC_NOCACHE;
475 out:
Trond Myklebust89a26b32014-08-06 13:44:24 -0400476 spin_unlock(&b->cache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 return rtn;
478
479found_entry:
Jeff Layton0338dd12013-02-04 08:18:02 -0500480 nfsdstats.rchits++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 /* We found a matching entry which is either in progress or done. */
482 age = jiffies - rp->c_timestamp;
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400483 lru_put_end(b, rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
485 rtn = RC_DROPIT;
486 /* Request being processed or excessive rexmits */
487 if (rp->c_state == RC_INPROG || age < RC_DELAY)
488 goto out;
489
490 /* From the hall of fame of impractical attacks:
491 * Is this a user who tries to snoop on the cache? */
492 rtn = RC_DOIT;
493 if (!rqstp->rq_secure && rp->c_secure)
494 goto out;
495
496 /* Compose RPC reply header */
497 switch (rp->c_type) {
498 case RC_NOCACHE:
499 break;
500 case RC_REPLSTAT:
501 svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
502 rtn = RC_REPLY;
503 break;
504 case RC_REPLBUFF:
505 if (!nfsd_cache_append(rqstp, &rp->c_replvec))
506 goto out; /* should not happen */
507 rtn = RC_REPLY;
508 break;
509 default:
510 printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type);
Jeff Layton0338dd12013-02-04 08:18:02 -0500511 nfsd_reply_cache_free_locked(rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 }
513
514 goto out;
515}
516
517/*
518 * Update a cache entry. This is called from nfsd_dispatch when
519 * the procedure has been executed and the complete reply is in
520 * rqstp->rq_res.
521 *
522 * We're copying around data here rather than swapping buffers because
523 * the toplevel loop requires max-sized buffers, which would be a waste
524 * of memory for a cache with a max reply size of 100 bytes (diropokres).
525 *
526 * If we should start to use different types of cache entries tailored
527 * specifically for attrstat and fh's, we may save even more space.
528 *
529 * Also note that a cachetype of RC_NOCACHE can legally be passed when
530 * nfsd failed to encode a reply that otherwise would have been cached.
531 * In this case, nfsd_cache_update is called with statp == NULL.
532 */
533void
Al Viroc7afef12006-10-19 23:29:02 -0700534nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535{
Jeff Layton13cc8a72013-02-04 08:18:03 -0500536 struct svc_cacherep *rp = rqstp->rq_cacherep;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 struct kvec *resv = &rqstp->rq_res.head[0], *cachv;
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400538 u32 hash;
539 struct nfsd_drc_bucket *b;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 int len;
Jeff Layton6c6910c2013-03-27 10:15:38 -0400541 size_t bufsize = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
Jeff Layton13cc8a72013-02-04 08:18:03 -0500543 if (!rp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 return;
545
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400546 hash = nfsd_cache_hash(rp->c_xid);
547 b = &drc_hashtbl[hash];
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 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)) {
Trond Myklebust89a26b32014-08-06 13:44:24 -0400554 nfsd_reply_cache_free(b, 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) {
Trond Myklebust89a26b32014-08-06 13:44:24 -0400569 nfsd_reply_cache_free(b, 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:
Trond Myklebust89a26b32014-08-06 13:44:24 -0400576 nfsd_reply_cache_free(b, rp);
Jeff Layton2c6b6912013-02-04 08:18:04 -0500577 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 }
Trond Myklebust89a26b32014-08-06 13:44:24 -0400579 spin_lock(&b->cache_lock);
Jeff Layton6c6910c2013-03-27 10:15:38 -0400580 drc_mem_usage += bufsize;
Trond Myklebustbedd4b62014-08-06 13:44:21 -0400581 lru_put_end(b, rp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 rp->c_secure = rqstp->rq_secure;
583 rp->c_type = cachetype;
584 rp->c_state = RC_DONE;
Trond Myklebust89a26b32014-08-06 13:44:24 -0400585 spin_unlock(&b->cache_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 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{
Jeff Laytona2f999a2013-03-27 10:15:38 -0400616 seq_printf(m, "max entries: %u\n", max_drc_entries);
Trond Myklebust31e60f52014-08-06 13:44:23 -0400617 seq_printf(m, "num entries: %u\n",
618 atomic_read(&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 return 0;
628}
629
630int nfsd_reply_cache_stats_open(struct inode *inode, struct file *file)
631{
632 return single_open(file, nfsd_reply_cache_stats_show, NULL);
633}