Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
1da177e4 LT |
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 | ||
3ba75830 | 12 | #include <linux/sunrpc/svc_xprt.h> |
5a0e3ad6 | 13 | #include <linux/slab.h> |
8f97514b | 14 | #include <linux/vmalloc.h> |
5976687a | 15 | #include <linux/sunrpc/addr.h> |
0338dd15 | 16 | #include <linux/highmem.h> |
0733c7ba JL |
17 | #include <linux/log2.h> |
18 | #include <linux/hash.h> | |
01a7decf | 19 | #include <net/checksum.h> |
5a0e3ad6 | 20 | |
9a74af21 BH |
21 | #include "nfsd.h" |
22 | #include "cache.h" | |
0b175b18 | 23 | #include "trace.h" |
0338dd15 | 24 | |
0733c7ba JL |
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 | |
1da177e4 | 31 | |
7142b98d | 32 | struct nfsd_drc_bucket { |
736c6625 | 33 | struct rb_root rb_head; |
bedd4b61 | 34 | struct list_head lru_head; |
89a26b3d | 35 | spinlock_t cache_lock; |
7142b98d TM |
36 | }; |
37 | ||
027690c7 BF |
38 | static struct kmem_cache *drc_slab; |
39 | ||
1da177e4 | 40 | static int nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec); |
1ab6c499 DC |
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); | |
b4e7f2c9 | 45 | |
0338dd15 JL |
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. | |
3ba75830 BF |
63 | * |
64 | * XXX: these limits are per-container, so memory used will increase | |
65 | * linearly with number of containers. Maybe that's OK. | |
0338dd15 JL |
66 | */ |
67 | static unsigned int | |
68 | nfsd_cache_size_limit(void) | |
69 | { | |
70 | unsigned int limit; | |
ca79b0c2 | 71 | unsigned long low_pages = totalram_pages() - totalhigh_pages(); |
0338dd15 JL |
72 | |
73 | limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10); | |
74 | return min_t(unsigned int, limit, 256*1024); | |
75 | } | |
76 | ||
0733c7ba JL |
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 | ||
f09841fd | 87 | static struct svc_cacherep * |
3ba75830 BF |
88 | nfsd_reply_cache_alloc(struct svc_rqst *rqstp, __wsum csum, |
89 | struct nfsd_net *nn) | |
1da177e4 LT |
90 | { |
91 | struct svc_cacherep *rp; | |
1da177e4 | 92 | |
027690c7 | 93 | rp = kmem_cache_alloc(drc_slab, GFP_KERNEL); |
f09841fd | 94 | if (rp) { |
1da177e4 LT |
95 | rp->c_state = RC_UNUSED; |
96 | rp->c_type = RC_NOCACHE; | |
736c6625 | 97 | RB_CLEAR_NODE(&rp->c_node); |
f09841fd | 98 | INIT_LIST_HEAD(&rp->c_lru); |
76ecec21 | 99 | |
ed00c2f6 TM |
100 | memset(&rp->c_key, 0, sizeof(rp->c_key)); |
101 | rp->c_key.k_xid = rqstp->rq_xid; | |
102 | rp->c_key.k_proc = rqstp->rq_proc; | |
103 | rpc_copy_addr((struct sockaddr *)&rp->c_key.k_addr, svc_addr(rqstp)); | |
104 | rpc_set_port((struct sockaddr *)&rp->c_key.k_addr, rpc_get_port(svc_addr(rqstp))); | |
105 | rp->c_key.k_prot = rqstp->rq_prot; | |
106 | rp->c_key.k_vers = rqstp->rq_vers; | |
107 | rp->c_key.k_len = rqstp->rq_arg.len; | |
108 | rp->c_key.k_csum = csum; | |
1da177e4 | 109 | } |
f09841fd JL |
110 | return rp; |
111 | } | |
1da177e4 | 112 | |
f09841fd | 113 | static void |
3ba75830 BF |
114 | nfsd_reply_cache_free_locked(struct nfsd_drc_bucket *b, struct svc_cacherep *rp, |
115 | struct nfsd_net *nn) | |
f09841fd | 116 | { |
6c6910cd | 117 | if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) { |
e567b98c | 118 | nfsd_stats_drc_mem_usage_sub(nn, rp->c_replvec.iov_len); |
f09841fd | 119 | kfree(rp->c_replvec.iov_base); |
6c6910cd | 120 | } |
76ecec21 | 121 | if (rp->c_state != RC_UNUSED) { |
736c6625 | 122 | rb_erase(&rp->c_node, &b->rb_head); |
76ecec21 | 123 | list_del(&rp->c_lru); |
3ba75830 | 124 | atomic_dec(&nn->num_drc_entries); |
e567b98c | 125 | nfsd_stats_drc_mem_usage_sub(nn, sizeof(*rp)); |
76ecec21 | 126 | } |
027690c7 | 127 | kmem_cache_free(drc_slab, rp); |
f09841fd JL |
128 | } |
129 | ||
2c6b691c | 130 | static void |
3ba75830 BF |
131 | nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct svc_cacherep *rp, |
132 | struct nfsd_net *nn) | |
2c6b691c | 133 | { |
89a26b3d | 134 | spin_lock(&b->cache_lock); |
3ba75830 | 135 | nfsd_reply_cache_free_locked(b, rp, nn); |
89a26b3d | 136 | spin_unlock(&b->cache_lock); |
2c6b691c JL |
137 | } |
138 | ||
027690c7 BF |
139 | int nfsd_drc_slab_create(void) |
140 | { | |
141 | drc_slab = kmem_cache_create("nfsd_drc", | |
142 | sizeof(struct svc_cacherep), 0, 0, NULL); | |
143 | return drc_slab ? 0: -ENOMEM; | |
144 | } | |
145 | ||
146 | void nfsd_drc_slab_free(void) | |
147 | { | |
148 | kmem_cache_destroy(drc_slab); | |
149 | } | |
150 | ||
e567b98c AG |
151 | static int nfsd_reply_cache_stats_init(struct nfsd_net *nn) |
152 | { | |
153 | return nfsd_percpu_counters_init(nn->counter, NFSD_NET_COUNTERS_NUM); | |
154 | } | |
155 | ||
156 | static void nfsd_reply_cache_stats_destroy(struct nfsd_net *nn) | |
157 | { | |
158 | nfsd_percpu_counters_destroy(nn->counter, NFSD_NET_COUNTERS_NUM); | |
159 | } | |
160 | ||
3ba75830 | 161 | int nfsd_reply_cache_init(struct nfsd_net *nn) |
f09841fd | 162 | { |
0733c7ba | 163 | unsigned int hashsize; |
bedd4b61 | 164 | unsigned int i; |
a68465c9 | 165 | int status = 0; |
0733c7ba | 166 | |
3ba75830 BF |
167 | nn->max_drc_entries = nfsd_cache_size_limit(); |
168 | atomic_set(&nn->num_drc_entries, 0); | |
169 | hashsize = nfsd_hashsize(nn->max_drc_entries); | |
170 | nn->maskbits = ilog2(hashsize); | |
ac534ff2 | 171 | |
e567b98c AG |
172 | status = nfsd_reply_cache_stats_init(nn); |
173 | if (status) | |
174 | goto out_nomem; | |
175 | ||
3ba75830 BF |
176 | nn->nfsd_reply_cache_shrinker.scan_objects = nfsd_reply_cache_scan; |
177 | nn->nfsd_reply_cache_shrinker.count_objects = nfsd_reply_cache_count; | |
178 | nn->nfsd_reply_cache_shrinker.seeks = 1; | |
e33c267a RG |
179 | status = register_shrinker(&nn->nfsd_reply_cache_shrinker, |
180 | "nfsd-reply:%s", nn->nfsd_name); | |
a68465c9 | 181 | if (status) |
e567b98c | 182 | goto out_stats_destroy; |
a68465c9 | 183 | |
8c38b705 RR |
184 | nn->drc_hashtbl = kvzalloc(array_size(hashsize, |
185 | sizeof(*nn->drc_hashtbl)), GFP_KERNEL); | |
186 | if (!nn->drc_hashtbl) | |
187 | goto out_shrinker; | |
8f97514b | 188 | |
89a26b3d | 189 | for (i = 0; i < hashsize; i++) { |
3ba75830 BF |
190 | INIT_LIST_HEAD(&nn->drc_hashtbl[i].lru_head); |
191 | spin_lock_init(&nn->drc_hashtbl[i].cache_lock); | |
89a26b3d | 192 | } |
3ba75830 | 193 | nn->drc_hashsize = hashsize; |
1da177e4 | 194 | |
d5c3428b | 195 | return 0; |
689d7ba4 BF |
196 | out_shrinker: |
197 | unregister_shrinker(&nn->nfsd_reply_cache_shrinker); | |
e567b98c AG |
198 | out_stats_destroy: |
199 | nfsd_reply_cache_stats_destroy(nn); | |
d5c3428b BF |
200 | out_nomem: |
201 | printk(KERN_ERR "nfsd: failed to allocate reply cache\n"); | |
d5c3428b | 202 | return -ENOMEM; |
1da177e4 LT |
203 | } |
204 | ||
3ba75830 | 205 | void nfsd_reply_cache_shutdown(struct nfsd_net *nn) |
1da177e4 LT |
206 | { |
207 | struct svc_cacherep *rp; | |
bedd4b61 | 208 | unsigned int i; |
1da177e4 | 209 | |
3ba75830 | 210 | unregister_shrinker(&nn->nfsd_reply_cache_shrinker); |
aca8a23d | 211 | |
3ba75830 BF |
212 | for (i = 0; i < nn->drc_hashsize; i++) { |
213 | struct list_head *head = &nn->drc_hashtbl[i].lru_head; | |
bedd4b61 TM |
214 | while (!list_empty(head)) { |
215 | rp = list_first_entry(head, struct svc_cacherep, c_lru); | |
3ba75830 BF |
216 | nfsd_reply_cache_free_locked(&nn->drc_hashtbl[i], |
217 | rp, nn); | |
bedd4b61 | 218 | } |
1da177e4 | 219 | } |
fd5e363e | 220 | nfsd_reply_cache_stats_destroy(nn); |
1da177e4 | 221 | |
3ba75830 BF |
222 | kvfree(nn->drc_hashtbl); |
223 | nn->drc_hashtbl = NULL; | |
224 | nn->drc_hashsize = 0; | |
8a8bc40d | 225 | |
1da177e4 LT |
226 | } |
227 | ||
228 | /* | |
aca8a23d JL |
229 | * Move cache entry to end of LRU list, and queue the cleaner to run if it's |
230 | * not already scheduled. | |
1da177e4 LT |
231 | */ |
232 | static void | |
bedd4b61 | 233 | lru_put_end(struct nfsd_drc_bucket *b, struct svc_cacherep *rp) |
1da177e4 | 234 | { |
56c2548b | 235 | rp->c_timestamp = jiffies; |
bedd4b61 | 236 | list_move_tail(&rp->c_lru, &b->lru_head); |
1da177e4 LT |
237 | } |
238 | ||
378a6109 CL |
239 | static noinline struct nfsd_drc_bucket * |
240 | nfsd_cache_bucket_find(__be32 xid, struct nfsd_net *nn) | |
241 | { | |
242 | unsigned int hash = hash_32((__force u32)xid, nn->maskbits); | |
243 | ||
244 | return &nn->drc_hashtbl[hash]; | |
245 | } | |
246 | ||
8847ecc9 CL |
247 | static long prune_bucket(struct nfsd_drc_bucket *b, struct nfsd_net *nn, |
248 | unsigned int max) | |
aca8a23d JL |
249 | { |
250 | struct svc_cacherep *rp, *tmp; | |
1ab6c499 | 251 | long freed = 0; |
aca8a23d | 252 | |
bedd4b61 | 253 | list_for_each_entry_safe(rp, tmp, &b->lru_head, c_lru) { |
1b19453d JL |
254 | /* |
255 | * Don't free entries attached to calls that are still | |
256 | * in-progress, but do keep scanning the list. | |
257 | */ | |
258 | if (rp->c_state == RC_INPROG) | |
259 | continue; | |
3ba75830 | 260 | if (atomic_read(&nn->num_drc_entries) <= nn->max_drc_entries && |
1b19453d | 261 | time_before(jiffies, rp->c_timestamp + RC_EXPIRE)) |
aca8a23d | 262 | break; |
3ba75830 | 263 | nfsd_reply_cache_free_locked(b, rp, nn); |
8847ecc9 CL |
264 | if (max && freed++ > max) |
265 | break; | |
aca8a23d | 266 | } |
bedd4b61 TM |
267 | return freed; |
268 | } | |
269 | ||
8847ecc9 CL |
270 | static long nfsd_prune_bucket(struct nfsd_drc_bucket *b, struct nfsd_net *nn) |
271 | { | |
272 | return prune_bucket(b, nn, 3); | |
273 | } | |
274 | ||
bedd4b61 TM |
275 | /* |
276 | * Walk the LRU list and prune off entries that are older than RC_EXPIRE. | |
277 | * Also prune the oldest ones when the total exceeds the max number of entries. | |
278 | */ | |
279 | static long | |
3ba75830 | 280 | prune_cache_entries(struct nfsd_net *nn) |
bedd4b61 TM |
281 | { |
282 | unsigned int i; | |
283 | long freed = 0; | |
bedd4b61 | 284 | |
3ba75830 BF |
285 | for (i = 0; i < nn->drc_hashsize; i++) { |
286 | struct nfsd_drc_bucket *b = &nn->drc_hashtbl[i]; | |
bedd4b61 | 287 | |
89a26b3d TM |
288 | if (list_empty(&b->lru_head)) |
289 | continue; | |
290 | spin_lock(&b->cache_lock); | |
8847ecc9 | 291 | freed += prune_bucket(b, nn, 0); |
89a26b3d | 292 | spin_unlock(&b->cache_lock); |
bedd4b61 | 293 | } |
1ab6c499 | 294 | return freed; |
aca8a23d JL |
295 | } |
296 | ||
1ab6c499 DC |
297 | static unsigned long |
298 | nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
b4e7f2c9 | 299 | { |
3ba75830 BF |
300 | struct nfsd_net *nn = container_of(shrink, |
301 | struct nfsd_net, nfsd_reply_cache_shrinker); | |
302 | ||
303 | return atomic_read(&nn->num_drc_entries); | |
b4e7f2c9 JL |
304 | } |
305 | ||
1ab6c499 DC |
306 | static unsigned long |
307 | nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
308 | { | |
3ba75830 BF |
309 | struct nfsd_net *nn = container_of(shrink, |
310 | struct nfsd_net, nfsd_reply_cache_shrinker); | |
311 | ||
312 | return prune_cache_entries(nn); | |
1ab6c499 | 313 | } |
01a7decf JL |
314 | /* |
315 | * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes | |
316 | */ | |
317 | static __wsum | |
318 | nfsd_cache_csum(struct svc_rqst *rqstp) | |
319 | { | |
320 | int idx; | |
321 | unsigned int base; | |
322 | __wsum csum; | |
323 | struct xdr_buf *buf = &rqstp->rq_arg; | |
324 | const unsigned char *p = buf->head[0].iov_base; | |
325 | size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len, | |
326 | RC_CSUMLEN); | |
327 | size_t len = min(buf->head[0].iov_len, csum_len); | |
328 | ||
329 | /* rq_arg.head first */ | |
330 | csum = csum_partial(p, len, 0); | |
331 | csum_len -= len; | |
332 | ||
333 | /* Continue into page array */ | |
334 | idx = buf->page_base / PAGE_SIZE; | |
335 | base = buf->page_base & ~PAGE_MASK; | |
336 | while (csum_len) { | |
337 | p = page_address(buf->pages[idx]) + base; | |
56edc86b | 338 | len = min_t(size_t, PAGE_SIZE - base, csum_len); |
01a7decf JL |
339 | csum = csum_partial(p, len, csum); |
340 | csum_len -= len; | |
341 | base = 0; | |
342 | ++idx; | |
343 | } | |
344 | return csum; | |
345 | } | |
346 | ||
ed00c2f6 | 347 | static int |
3ba75830 BF |
348 | nfsd_cache_key_cmp(const struct svc_cacherep *key, |
349 | const struct svc_cacherep *rp, struct nfsd_net *nn) | |
9dc56143 | 350 | { |
ed00c2f6 | 351 | if (key->c_key.k_xid == rp->c_key.k_xid && |
0b175b18 | 352 | key->c_key.k_csum != rp->c_key.k_csum) { |
e567b98c | 353 | nfsd_stats_payload_misses_inc(nn); |
0b175b18 CL |
354 | trace_nfsd_drc_mismatch(nn, key, rp); |
355 | } | |
ef9b16dc | 356 | |
ed00c2f6 | 357 | return memcmp(&key->c_key, &rp->c_key, sizeof(key->c_key)); |
9dc56143 JL |
358 | } |
359 | ||
a4a3ec32 JL |
360 | /* |
361 | * Search the request hash for an entry that matches the given rqstp. | |
362 | * Must be called with cache_lock held. Returns the found entry or | |
76ecec21 | 363 | * inserts an empty key on failure. |
a4a3ec32 JL |
364 | */ |
365 | static struct svc_cacherep * | |
3ba75830 BF |
366 | nfsd_cache_insert(struct nfsd_drc_bucket *b, struct svc_cacherep *key, |
367 | struct nfsd_net *nn) | |
a4a3ec32 | 368 | { |
76ecec21 | 369 | struct svc_cacherep *rp, *ret = key; |
736c6625 TM |
370 | struct rb_node **p = &b->rb_head.rb_node, |
371 | *parent = NULL; | |
98d821bd | 372 | unsigned int entries = 0; |
736c6625 | 373 | int cmp; |
a4a3ec32 | 374 | |
736c6625 | 375 | while (*p != NULL) { |
98d821bd | 376 | ++entries; |
736c6625 TM |
377 | parent = *p; |
378 | rp = rb_entry(parent, struct svc_cacherep, c_node); | |
379 | ||
3ba75830 | 380 | cmp = nfsd_cache_key_cmp(key, rp, nn); |
736c6625 TM |
381 | if (cmp < 0) |
382 | p = &parent->rb_left; | |
383 | else if (cmp > 0) | |
384 | p = &parent->rb_right; | |
385 | else { | |
98d821bd | 386 | ret = rp; |
736c6625 | 387 | goto out; |
98d821bd JL |
388 | } |
389 | } | |
736c6625 TM |
390 | rb_link_node(&key->c_node, parent, p); |
391 | rb_insert_color(&key->c_node, &b->rb_head); | |
392 | out: | |
98d821bd | 393 | /* tally hash chain length stats */ |
3ba75830 BF |
394 | if (entries > nn->longest_chain) { |
395 | nn->longest_chain = entries; | |
396 | nn->longest_chain_cachesize = atomic_read(&nn->num_drc_entries); | |
397 | } else if (entries == nn->longest_chain) { | |
98d821bd | 398 | /* prefer to keep the smallest cachesize possible here */ |
3ba75830 BF |
399 | nn->longest_chain_cachesize = min_t(unsigned int, |
400 | nn->longest_chain_cachesize, | |
401 | atomic_read(&nn->num_drc_entries)); | |
a4a3ec32 | 402 | } |
98d821bd | 403 | |
76ecec21 | 404 | lru_put_end(b, ret); |
98d821bd | 405 | return ret; |
a4a3ec32 JL |
406 | } |
407 | ||
0b175b18 CL |
408 | /** |
409 | * nfsd_cache_lookup - Find an entry in the duplicate reply cache | |
410 | * @rqstp: Incoming Call to find | |
411 | * | |
1da177e4 | 412 | * Try to find an entry matching the current call in the cache. When none |
1ac83629 JL |
413 | * is found, we try to grab the oldest expired entry off the LRU list. If |
414 | * a suitable one isn't there, then drop the cache_lock and allocate a | |
415 | * new one, then search again in case one got inserted while this thread | |
416 | * didn't hold the lock. | |
0b175b18 CL |
417 | * |
418 | * Return values: | |
419 | * %RC_DOIT: Process the request normally | |
420 | * %RC_REPLY: Reply from cache | |
421 | * %RC_DROPIT: Do not process the request further | |
1da177e4 | 422 | */ |
0b175b18 | 423 | int nfsd_cache_lookup(struct svc_rqst *rqstp) |
1da177e4 | 424 | { |
0f29ce32 | 425 | struct nfsd_net *nn; |
0338dd15 | 426 | struct svc_cacherep *rp, *found; |
01a7decf | 427 | __wsum csum; |
0f29ce32 | 428 | struct nfsd_drc_bucket *b; |
1091006c | 429 | int type = rqstp->rq_cachetype; |
0b9ea37f | 430 | int rtn = RC_DOIT; |
1da177e4 LT |
431 | |
432 | rqstp->rq_cacherep = NULL; | |
13cc8a78 | 433 | if (type == RC_NOCACHE) { |
e567b98c | 434 | nfsd_stats_rc_nocache_inc(); |
0b175b18 | 435 | goto out; |
1da177e4 LT |
436 | } |
437 | ||
01a7decf JL |
438 | csum = nfsd_cache_csum(rqstp); |
439 | ||
0b9ea37f JL |
440 | /* |
441 | * Since the common case is a cache miss followed by an insert, | |
a0ef5e19 | 442 | * preallocate an entry. |
0b9ea37f | 443 | */ |
0f29ce32 | 444 | nn = net_generic(SVC_NET(rqstp), nfsd_net_id); |
3ba75830 | 445 | rp = nfsd_reply_cache_alloc(rqstp, csum, nn); |
0b175b18 CL |
446 | if (!rp) |
447 | goto out; | |
0338dd15 | 448 | |
0f29ce32 | 449 | b = nfsd_cache_bucket_find(rqstp->rq_xid, nn); |
76ecec21 | 450 | spin_lock(&b->cache_lock); |
3ba75830 | 451 | found = nfsd_cache_insert(b, rp, nn); |
add1511c | 452 | if (found != rp) |
0338dd15 | 453 | goto found_entry; |
1da177e4 | 454 | |
e567b98c | 455 | nfsd_stats_rc_misses_inc(); |
1da177e4 LT |
456 | rqstp->rq_cacherep = rp; |
457 | rp->c_state = RC_INPROG; | |
76ecec21 | 458 | |
3ba75830 | 459 | atomic_inc(&nn->num_drc_entries); |
e567b98c | 460 | nfsd_stats_drc_mem_usage_add(nn, sizeof(*rp)); |
76ecec21 | 461 | |
8847ecc9 | 462 | nfsd_prune_bucket(b, nn); |
0b175b18 CL |
463 | |
464 | out_unlock: | |
89a26b3d | 465 | spin_unlock(&b->cache_lock); |
0b175b18 | 466 | out: |
1da177e4 LT |
467 | return rtn; |
468 | ||
469 | found_entry: | |
470 | /* We found a matching entry which is either in progress or done. */ | |
add1511c | 471 | nfsd_reply_cache_free_locked(NULL, rp, nn); |
e567b98c | 472 | nfsd_stats_rc_hits_inc(); |
1da177e4 | 473 | rtn = RC_DROPIT; |
add1511c | 474 | rp = found; |
76ecec21 | 475 | |
7e5d0e0d TM |
476 | /* Request being processed */ |
477 | if (rp->c_state == RC_INPROG) | |
0b175b18 | 478 | goto out_trace; |
1da177e4 LT |
479 | |
480 | /* From the hall of fame of impractical attacks: | |
481 | * Is this a user who tries to snoop on the cache? */ | |
482 | rtn = RC_DOIT; | |
4d152e2c | 483 | if (!test_bit(RQ_SECURE, &rqstp->rq_flags) && rp->c_secure) |
0b175b18 | 484 | goto out_trace; |
1da177e4 LT |
485 | |
486 | /* Compose RPC reply header */ | |
487 | switch (rp->c_type) { | |
488 | case RC_NOCACHE: | |
489 | break; | |
490 | case RC_REPLSTAT: | |
8dd41d70 | 491 | xdr_stream_encode_be32(&rqstp->rq_res_stream, rp->c_replstat); |
1da177e4 LT |
492 | rtn = RC_REPLY; |
493 | break; | |
494 | case RC_REPLBUFF: | |
495 | if (!nfsd_cache_append(rqstp, &rp->c_replvec)) | |
0b175b18 | 496 | goto out_unlock; /* should not happen */ |
1da177e4 LT |
497 | rtn = RC_REPLY; |
498 | break; | |
499 | default: | |
c25bf185 | 500 | WARN_ONCE(1, "nfsd: bad repcache type %d\n", rp->c_type); |
1da177e4 LT |
501 | } |
502 | ||
0b175b18 CL |
503 | out_trace: |
504 | trace_nfsd_drc_found(nn, rqstp, rtn); | |
505 | goto out_unlock; | |
1da177e4 LT |
506 | } |
507 | ||
0b175b18 CL |
508 | /** |
509 | * nfsd_cache_update - Update an entry in the duplicate reply cache. | |
510 | * @rqstp: svc_rqst with a finished Reply | |
511 | * @cachetype: which cache to update | |
cee4db19 | 512 | * @statp: pointer to Reply's NFS status code, or NULL |
0b175b18 CL |
513 | * |
514 | * This is called from nfsd_dispatch when the procedure has been | |
515 | * executed and the complete reply is in rqstp->rq_res. | |
1da177e4 LT |
516 | * |
517 | * We're copying around data here rather than swapping buffers because | |
518 | * the toplevel loop requires max-sized buffers, which would be a waste | |
519 | * of memory for a cache with a max reply size of 100 bytes (diropokres). | |
520 | * | |
521 | * If we should start to use different types of cache entries tailored | |
522 | * specifically for attrstat and fh's, we may save even more space. | |
523 | * | |
524 | * Also note that a cachetype of RC_NOCACHE can legally be passed when | |
525 | * nfsd failed to encode a reply that otherwise would have been cached. | |
526 | * In this case, nfsd_cache_update is called with statp == NULL. | |
527 | */ | |
0b175b18 | 528 | void nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp) |
1da177e4 | 529 | { |
3ba75830 | 530 | struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id); |
13cc8a78 | 531 | struct svc_cacherep *rp = rqstp->rq_cacherep; |
1da177e4 | 532 | struct kvec *resv = &rqstp->rq_res.head[0], *cachv; |
bedd4b61 | 533 | struct nfsd_drc_bucket *b; |
1da177e4 | 534 | int len; |
6c6910cd | 535 | size_t bufsize = 0; |
1da177e4 | 536 | |
13cc8a78 | 537 | if (!rp) |
1da177e4 LT |
538 | return; |
539 | ||
378a6109 | 540 | b = nfsd_cache_bucket_find(rp->c_key.k_xid, nn); |
bedd4b61 | 541 | |
1da177e4 LT |
542 | len = resv->iov_len - ((char*)statp - (char*)resv->iov_base); |
543 | len >>= 2; | |
fca4217c | 544 | |
1da177e4 LT |
545 | /* Don't cache excessive amounts of data and XDR failures */ |
546 | if (!statp || len > (256 >> 2)) { | |
3ba75830 | 547 | nfsd_reply_cache_free(b, rp, nn); |
1da177e4 LT |
548 | return; |
549 | } | |
550 | ||
551 | switch (cachetype) { | |
552 | case RC_REPLSTAT: | |
553 | if (len != 1) | |
554 | printk("nfsd: RC_REPLSTAT/reply len %d!\n",len); | |
555 | rp->c_replstat = *statp; | |
556 | break; | |
557 | case RC_REPLBUFF: | |
558 | cachv = &rp->c_replvec; | |
6c6910cd JL |
559 | bufsize = len << 2; |
560 | cachv->iov_base = kmalloc(bufsize, GFP_KERNEL); | |
1da177e4 | 561 | if (!cachv->iov_base) { |
3ba75830 | 562 | nfsd_reply_cache_free(b, rp, nn); |
1da177e4 LT |
563 | return; |
564 | } | |
6c6910cd JL |
565 | cachv->iov_len = bufsize; |
566 | memcpy(cachv->iov_base, statp, bufsize); | |
1da177e4 | 567 | break; |
2c6b691c | 568 | case RC_NOCACHE: |
3ba75830 | 569 | nfsd_reply_cache_free(b, rp, nn); |
2c6b691c | 570 | return; |
1da177e4 | 571 | } |
89a26b3d | 572 | spin_lock(&b->cache_lock); |
e567b98c | 573 | nfsd_stats_drc_mem_usage_add(nn, bufsize); |
bedd4b61 | 574 | lru_put_end(b, rp); |
4d152e2c | 575 | rp->c_secure = test_bit(RQ_SECURE, &rqstp->rq_flags); |
1da177e4 LT |
576 | rp->c_type = cachetype; |
577 | rp->c_state = RC_DONE; | |
89a26b3d | 578 | spin_unlock(&b->cache_lock); |
1da177e4 LT |
579 | return; |
580 | } | |
581 | ||
582 | /* | |
583 | * Copy cached reply to current reply buffer. Should always fit. | |
584 | * FIXME as reply is in a page, we should just attach the page, and | |
585 | * keep a refcount.... | |
586 | */ | |
587 | static int | |
588 | nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data) | |
589 | { | |
590 | struct kvec *vec = &rqstp->rq_res.head[0]; | |
591 | ||
592 | if (vec->iov_len + data->iov_len > PAGE_SIZE) { | |
5b5e0928 | 593 | printk(KERN_WARNING "nfsd: cached reply too large (%zd).\n", |
1da177e4 LT |
594 | data->iov_len); |
595 | return 0; | |
596 | } | |
597 | memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len); | |
598 | vec->iov_len += data->iov_len; | |
599 | return 1; | |
600 | } | |
a2f999a3 JL |
601 | |
602 | /* | |
603 | * Note that fields may be added, removed or reordered in the future. Programs | |
604 | * scraping this file for info should test the labels to ensure they're | |
605 | * getting the correct field. | |
606 | */ | |
64776611 | 607 | int nfsd_reply_cache_stats_show(struct seq_file *m, void *v) |
a2f999a3 | 608 | { |
64776611 C |
609 | struct nfsd_net *nn = net_generic(file_inode(m->file)->i_sb->s_fs_info, |
610 | nfsd_net_id); | |
3ba75830 BF |
611 | |
612 | seq_printf(m, "max entries: %u\n", nn->max_drc_entries); | |
31e60f52 | 613 | seq_printf(m, "num entries: %u\n", |
e567b98c | 614 | atomic_read(&nn->num_drc_entries)); |
3ba75830 | 615 | seq_printf(m, "hash buckets: %u\n", 1 << nn->maskbits); |
e567b98c AG |
616 | seq_printf(m, "mem usage: %lld\n", |
617 | percpu_counter_sum_positive(&nn->counter[NFSD_NET_DRC_MEM_USAGE])); | |
618 | seq_printf(m, "cache hits: %lld\n", | |
619 | percpu_counter_sum_positive(&nfsdstats.counter[NFSD_STATS_RC_HITS])); | |
620 | seq_printf(m, "cache misses: %lld\n", | |
621 | percpu_counter_sum_positive(&nfsdstats.counter[NFSD_STATS_RC_MISSES])); | |
622 | seq_printf(m, "not cached: %lld\n", | |
623 | percpu_counter_sum_positive(&nfsdstats.counter[NFSD_STATS_RC_NOCACHE])); | |
624 | seq_printf(m, "payload misses: %lld\n", | |
625 | percpu_counter_sum_positive(&nn->counter[NFSD_NET_PAYLOAD_MISSES])); | |
3ba75830 BF |
626 | seq_printf(m, "longest chain len: %u\n", nn->longest_chain); |
627 | seq_printf(m, "cachesize at longest: %u\n", nn->longest_chain_cachesize); | |
a2f999a3 JL |
628 | return 0; |
629 | } |