]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * 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 | ||
5a0e3ad6 | 11 | #include <linux/slab.h> |
5976687a | 12 | #include <linux/sunrpc/addr.h> |
0338dd15 | 13 | #include <linux/highmem.h> |
0733c7ba JL |
14 | #include <linux/log2.h> |
15 | #include <linux/hash.h> | |
01a7decf | 16 | #include <net/checksum.h> |
5a0e3ad6 | 17 | |
9a74af21 BH |
18 | #include "nfsd.h" |
19 | #include "cache.h" | |
1da177e4 | 20 | |
0338dd15 JL |
21 | #define NFSDDBG_FACILITY NFSDDBG_REPCACHE |
22 | ||
0733c7ba JL |
23 | /* |
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 | |
1da177e4 | 29 | |
7142b98d | 30 | struct nfsd_drc_bucket { |
bedd4b61 | 31 | struct list_head lru_head; |
89a26b3d | 32 | spinlock_t cache_lock; |
7142b98d TM |
33 | }; |
34 | ||
35 | static struct nfsd_drc_bucket *drc_hashtbl; | |
8a8bc40d | 36 | static struct kmem_cache *drc_slab; |
9dc56143 JL |
37 | |
38 | /* max number of entries allowed in the cache */ | |
0338dd15 | 39 | static unsigned int max_drc_entries; |
1da177e4 | 40 | |
0733c7ba JL |
41 | /* number of significant bits in the hash value */ |
42 | static unsigned int maskbits; | |
bedd4b61 | 43 | static unsigned int drc_hashsize; |
0733c7ba | 44 | |
9dc56143 JL |
45 | /* |
46 | * 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 */ | |
31e60f52 | 51 | static atomic_t num_drc_entries; |
9dc56143 JL |
52 | |
53 | /* cache misses due only to checksum comparison failures */ | |
54 | static unsigned int payload_misses; | |
55 | ||
6c6910cd JL |
56 | /* amount of memory (in bytes) currently consumed by the DRC */ |
57 | static unsigned int drc_mem_usage; | |
58 | ||
98d821bd JL |
59 | /* longest hash chain seen */ |
60 | static unsigned int longest_chain; | |
61 | ||
62 | /* size of cache when we saw the longest hash chain */ | |
63 | static unsigned int longest_chain_cachesize; | |
64 | ||
1da177e4 | 65 | static int nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec); |
1ab6c499 DC |
66 | static unsigned long nfsd_reply_cache_count(struct shrinker *shrink, |
67 | struct shrink_control *sc); | |
68 | static unsigned long nfsd_reply_cache_scan(struct shrinker *shrink, | |
69 | struct shrink_control *sc); | |
b4e7f2c9 | 70 | |
c8c797f9 | 71 | static struct shrinker nfsd_reply_cache_shrinker = { |
1ab6c499 DC |
72 | .scan_objects = nfsd_reply_cache_scan, |
73 | .count_objects = nfsd_reply_cache_count, | |
b4e7f2c9 JL |
74 | .seeks = 1, |
75 | }; | |
1da177e4 | 76 | |
0338dd15 JL |
77 | /* |
78 | * Put a cap on the size of the DRC based on the amount of available | |
79 | * low memory in the machine. | |
80 | * | |
81 | * 64MB: 8192 | |
82 | * 128MB: 11585 | |
83 | * 256MB: 16384 | |
84 | * 512MB: 23170 | |
85 | * 1GB: 32768 | |
86 | * 2GB: 46340 | |
87 | * 4GB: 65536 | |
88 | * 8GB: 92681 | |
89 | * 16GB: 131072 | |
90 | * | |
91 | * ...with a hard cap of 256k entries. In the worst case, each entry will be | |
92 | * ~1k, so the above numbers should give a rough max of the amount of memory | |
93 | * used in k. | |
94 | */ | |
95 | static unsigned int | |
96 | nfsd_cache_size_limit(void) | |
97 | { | |
98 | unsigned int limit; | |
99 | unsigned long low_pages = totalram_pages - totalhigh_pages; | |
100 | ||
101 | limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10); | |
102 | return min_t(unsigned int, limit, 256*1024); | |
103 | } | |
104 | ||
0733c7ba JL |
105 | /* |
106 | * Compute the number of hash buckets we need. Divide the max cachesize by | |
107 | * the "target" max bucket size, and round up to next power of two. | |
108 | */ | |
109 | static unsigned int | |
110 | nfsd_hashsize(unsigned int limit) | |
111 | { | |
112 | return roundup_pow_of_two(limit / TARGET_BUCKET_SIZE); | |
113 | } | |
114 | ||
7142b98d TM |
115 | static u32 |
116 | nfsd_cache_hash(__be32 xid) | |
117 | { | |
118 | return hash_32(be32_to_cpu(xid), maskbits); | |
119 | } | |
120 | ||
f09841fd JL |
121 | static struct svc_cacherep * |
122 | nfsd_reply_cache_alloc(void) | |
1da177e4 LT |
123 | { |
124 | struct svc_cacherep *rp; | |
1da177e4 | 125 | |
f09841fd JL |
126 | rp = kmem_cache_alloc(drc_slab, GFP_KERNEL); |
127 | if (rp) { | |
1da177e4 LT |
128 | rp->c_state = RC_UNUSED; |
129 | rp->c_type = RC_NOCACHE; | |
f09841fd | 130 | INIT_LIST_HEAD(&rp->c_lru); |
1da177e4 | 131 | } |
f09841fd JL |
132 | return rp; |
133 | } | |
1da177e4 | 134 | |
f09841fd JL |
135 | static void |
136 | nfsd_reply_cache_free_locked(struct svc_cacherep *rp) | |
137 | { | |
6c6910cd JL |
138 | if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) { |
139 | drc_mem_usage -= rp->c_replvec.iov_len; | |
f09841fd | 140 | kfree(rp->c_replvec.iov_base); |
6c6910cd | 141 | } |
f09841fd | 142 | list_del(&rp->c_lru); |
31e60f52 | 143 | atomic_dec(&num_drc_entries); |
6c6910cd | 144 | drc_mem_usage -= sizeof(*rp); |
f09841fd JL |
145 | kmem_cache_free(drc_slab, rp); |
146 | } | |
147 | ||
2c6b691c | 148 | static void |
89a26b3d | 149 | nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct svc_cacherep *rp) |
2c6b691c | 150 | { |
89a26b3d | 151 | spin_lock(&b->cache_lock); |
2c6b691c | 152 | nfsd_reply_cache_free_locked(rp); |
89a26b3d | 153 | spin_unlock(&b->cache_lock); |
2c6b691c JL |
154 | } |
155 | ||
f09841fd JL |
156 | int nfsd_reply_cache_init(void) |
157 | { | |
0733c7ba | 158 | unsigned int hashsize; |
bedd4b61 | 159 | unsigned int i; |
a68465c9 | 160 | int status = 0; |
0733c7ba | 161 | |
ac534ff2 | 162 | max_drc_entries = nfsd_cache_size_limit(); |
31e60f52 | 163 | atomic_set(&num_drc_entries, 0); |
0733c7ba JL |
164 | hashsize = nfsd_hashsize(max_drc_entries); |
165 | maskbits = ilog2(hashsize); | |
ac534ff2 | 166 | |
a68465c9 KM |
167 | status = register_shrinker(&nfsd_reply_cache_shrinker); |
168 | if (status) | |
169 | return status; | |
170 | ||
8a8bc40d JL |
171 | drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep), |
172 | 0, 0, NULL); | |
173 | if (!drc_slab) | |
174 | goto out_nomem; | |
175 | ||
7142b98d TM |
176 | drc_hashtbl = kcalloc(hashsize, sizeof(*drc_hashtbl), GFP_KERNEL); |
177 | if (!drc_hashtbl) | |
d5c3428b | 178 | goto out_nomem; |
89a26b3d | 179 | for (i = 0; i < hashsize; i++) { |
bedd4b61 | 180 | INIT_LIST_HEAD(&drc_hashtbl[i].lru_head); |
89a26b3d TM |
181 | spin_lock_init(&drc_hashtbl[i].cache_lock); |
182 | } | |
bedd4b61 | 183 | drc_hashsize = hashsize; |
1da177e4 | 184 | |
d5c3428b BF |
185 | return 0; |
186 | out_nomem: | |
187 | printk(KERN_ERR "nfsd: failed to allocate reply cache\n"); | |
188 | nfsd_reply_cache_shutdown(); | |
189 | return -ENOMEM; | |
1da177e4 LT |
190 | } |
191 | ||
d5c3428b | 192 | void nfsd_reply_cache_shutdown(void) |
1da177e4 LT |
193 | { |
194 | struct svc_cacherep *rp; | |
bedd4b61 | 195 | unsigned int i; |
1da177e4 | 196 | |
b4e7f2c9 | 197 | unregister_shrinker(&nfsd_reply_cache_shrinker); |
aca8a23d | 198 | |
bedd4b61 TM |
199 | for (i = 0; i < drc_hashsize; i++) { |
200 | struct list_head *head = &drc_hashtbl[i].lru_head; | |
201 | while (!list_empty(head)) { | |
202 | rp = list_first_entry(head, struct svc_cacherep, c_lru); | |
203 | nfsd_reply_cache_free_locked(rp); | |
204 | } | |
1da177e4 LT |
205 | } |
206 | ||
7142b98d TM |
207 | kfree (drc_hashtbl); |
208 | drc_hashtbl = NULL; | |
bedd4b61 | 209 | drc_hashsize = 0; |
8a8bc40d | 210 | |
e79017dd JL |
211 | kmem_cache_destroy(drc_slab); |
212 | drc_slab = NULL; | |
1da177e4 LT |
213 | } |
214 | ||
215 | /* | |
aca8a23d JL |
216 | * Move cache entry to end of LRU list, and queue the cleaner to run if it's |
217 | * not already scheduled. | |
1da177e4 LT |
218 | */ |
219 | static void | |
bedd4b61 | 220 | lru_put_end(struct nfsd_drc_bucket *b, struct svc_cacherep *rp) |
1da177e4 | 221 | { |
56c2548b | 222 | rp->c_timestamp = jiffies; |
bedd4b61 | 223 | list_move_tail(&rp->c_lru, &b->lru_head); |
1da177e4 LT |
224 | } |
225 | ||
1ab6c499 | 226 | static long |
bedd4b61 | 227 | prune_bucket(struct nfsd_drc_bucket *b) |
aca8a23d JL |
228 | { |
229 | struct svc_cacherep *rp, *tmp; | |
1ab6c499 | 230 | long freed = 0; |
aca8a23d | 231 | |
bedd4b61 | 232 | list_for_each_entry_safe(rp, tmp, &b->lru_head, c_lru) { |
1b19453d JL |
233 | /* |
234 | * Don't free entries attached to calls that are still | |
235 | * in-progress, but do keep scanning the list. | |
236 | */ | |
237 | if (rp->c_state == RC_INPROG) | |
238 | continue; | |
31e60f52 | 239 | if (atomic_read(&num_drc_entries) <= max_drc_entries && |
1b19453d | 240 | time_before(jiffies, rp->c_timestamp + RC_EXPIRE)) |
aca8a23d JL |
241 | break; |
242 | nfsd_reply_cache_free_locked(rp); | |
1ab6c499 | 243 | freed++; |
aca8a23d | 244 | } |
bedd4b61 TM |
245 | return freed; |
246 | } | |
247 | ||
248 | /* | |
249 | * Walk the LRU list and prune off entries that are older than RC_EXPIRE. | |
250 | * Also prune the oldest ones when the total exceeds the max number of entries. | |
251 | */ | |
252 | static long | |
253 | prune_cache_entries(void) | |
254 | { | |
255 | unsigned int i; | |
256 | long freed = 0; | |
bedd4b61 TM |
257 | |
258 | for (i = 0; i < drc_hashsize; i++) { | |
259 | struct nfsd_drc_bucket *b = &drc_hashtbl[i]; | |
260 | ||
89a26b3d TM |
261 | if (list_empty(&b->lru_head)) |
262 | continue; | |
263 | spin_lock(&b->cache_lock); | |
bedd4b61 | 264 | freed += prune_bucket(b); |
89a26b3d | 265 | spin_unlock(&b->cache_lock); |
bedd4b61 | 266 | } |
1ab6c499 | 267 | return freed; |
aca8a23d JL |
268 | } |
269 | ||
1ab6c499 DC |
270 | static unsigned long |
271 | nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
b4e7f2c9 | 272 | { |
31e60f52 | 273 | return atomic_read(&num_drc_entries); |
b4e7f2c9 JL |
274 | } |
275 | ||
1ab6c499 DC |
276 | static unsigned long |
277 | nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
278 | { | |
89a26b3d | 279 | return prune_cache_entries(); |
1ab6c499 | 280 | } |
01a7decf JL |
281 | /* |
282 | * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes | |
283 | */ | |
284 | static __wsum | |
285 | nfsd_cache_csum(struct svc_rqst *rqstp) | |
286 | { | |
287 | int idx; | |
288 | unsigned int base; | |
289 | __wsum csum; | |
290 | struct xdr_buf *buf = &rqstp->rq_arg; | |
291 | const unsigned char *p = buf->head[0].iov_base; | |
292 | size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len, | |
293 | RC_CSUMLEN); | |
294 | size_t len = min(buf->head[0].iov_len, csum_len); | |
295 | ||
296 | /* rq_arg.head first */ | |
297 | csum = csum_partial(p, len, 0); | |
298 | csum_len -= len; | |
299 | ||
300 | /* Continue into page array */ | |
301 | idx = buf->page_base / PAGE_SIZE; | |
302 | base = buf->page_base & ~PAGE_MASK; | |
303 | while (csum_len) { | |
304 | p = page_address(buf->pages[idx]) + base; | |
56edc86b | 305 | len = min_t(size_t, PAGE_SIZE - base, csum_len); |
01a7decf JL |
306 | csum = csum_partial(p, len, csum); |
307 | csum_len -= len; | |
308 | base = 0; | |
309 | ++idx; | |
310 | } | |
311 | return csum; | |
312 | } | |
313 | ||
9dc56143 JL |
314 | static bool |
315 | nfsd_cache_match(struct svc_rqst *rqstp, __wsum csum, struct svc_cacherep *rp) | |
316 | { | |
ef9b16dc TM |
317 | /* Check RPC XID first */ |
318 | if (rqstp->rq_xid != rp->c_xid) | |
9dc56143 | 319 | return false; |
9dc56143 JL |
320 | /* compare checksum of NFS data */ |
321 | if (csum != rp->c_csum) { | |
322 | ++payload_misses; | |
323 | return false; | |
324 | } | |
325 | ||
ef9b16dc TM |
326 | /* Other discriminators */ |
327 | if (rqstp->rq_proc != rp->c_proc || | |
328 | rqstp->rq_prot != rp->c_prot || | |
329 | rqstp->rq_vers != rp->c_vers || | |
330 | rqstp->rq_arg.len != rp->c_len || | |
331 | !rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) || | |
332 | rpc_get_port(svc_addr(rqstp)) != rpc_get_port((struct sockaddr *)&rp->c_addr)) | |
333 | return false; | |
334 | ||
9dc56143 JL |
335 | return true; |
336 | } | |
337 | ||
a4a3ec32 JL |
338 | /* |
339 | * Search the request hash for an entry that matches the given rqstp. | |
340 | * Must be called with cache_lock held. Returns the found entry or | |
341 | * NULL on failure. | |
342 | */ | |
343 | static struct svc_cacherep * | |
7142b98d TM |
344 | nfsd_cache_search(struct nfsd_drc_bucket *b, struct svc_rqst *rqstp, |
345 | __wsum csum) | |
a4a3ec32 | 346 | { |
98d821bd | 347 | struct svc_cacherep *rp, *ret = NULL; |
11acf6ef | 348 | struct list_head *rh = &b->lru_head; |
98d821bd | 349 | unsigned int entries = 0; |
a4a3ec32 | 350 | |
11acf6ef | 351 | list_for_each_entry(rp, rh, c_lru) { |
98d821bd JL |
352 | ++entries; |
353 | if (nfsd_cache_match(rqstp, csum, rp)) { | |
354 | ret = rp; | |
355 | break; | |
356 | } | |
357 | } | |
358 | ||
359 | /* tally hash chain length stats */ | |
360 | if (entries > longest_chain) { | |
361 | longest_chain = entries; | |
31e60f52 | 362 | longest_chain_cachesize = atomic_read(&num_drc_entries); |
98d821bd JL |
363 | } else if (entries == longest_chain) { |
364 | /* prefer to keep the smallest cachesize possible here */ | |
31e60f52 TM |
365 | longest_chain_cachesize = min_t(unsigned int, |
366 | longest_chain_cachesize, | |
367 | atomic_read(&num_drc_entries)); | |
a4a3ec32 | 368 | } |
98d821bd JL |
369 | |
370 | return ret; | |
a4a3ec32 JL |
371 | } |
372 | ||
1da177e4 LT |
373 | /* |
374 | * Try to find an entry matching the current call in the cache. When none | |
1ac83629 JL |
375 | * is found, we try to grab the oldest expired entry off the LRU list. If |
376 | * a suitable one isn't there, then drop the cache_lock and allocate a | |
377 | * new one, then search again in case one got inserted while this thread | |
378 | * didn't hold the lock. | |
1da177e4 LT |
379 | */ |
380 | int | |
1091006c | 381 | nfsd_cache_lookup(struct svc_rqst *rqstp) |
1da177e4 | 382 | { |
0338dd15 | 383 | struct svc_cacherep *rp, *found; |
c7afef1f AV |
384 | __be32 xid = rqstp->rq_xid; |
385 | u32 proto = rqstp->rq_prot, | |
1da177e4 LT |
386 | vers = rqstp->rq_vers, |
387 | proc = rqstp->rq_proc; | |
01a7decf | 388 | __wsum csum; |
7142b98d TM |
389 | u32 hash = nfsd_cache_hash(xid); |
390 | struct nfsd_drc_bucket *b = &drc_hashtbl[hash]; | |
1da177e4 | 391 | unsigned long age; |
1091006c | 392 | int type = rqstp->rq_cachetype; |
0b9ea37f | 393 | int rtn = RC_DOIT; |
1da177e4 LT |
394 | |
395 | rqstp->rq_cacherep = NULL; | |
13cc8a78 | 396 | if (type == RC_NOCACHE) { |
1da177e4 | 397 | nfsdstats.rcnocache++; |
0b9ea37f | 398 | return rtn; |
1da177e4 LT |
399 | } |
400 | ||
01a7decf JL |
401 | csum = nfsd_cache_csum(rqstp); |
402 | ||
0b9ea37f JL |
403 | /* |
404 | * Since the common case is a cache miss followed by an insert, | |
a0ef5e19 | 405 | * preallocate an entry. |
0b9ea37f | 406 | */ |
0338dd15 | 407 | rp = nfsd_reply_cache_alloc(); |
89a26b3d | 408 | spin_lock(&b->cache_lock); |
6c6910cd | 409 | if (likely(rp)) { |
31e60f52 | 410 | atomic_inc(&num_drc_entries); |
6c6910cd JL |
411 | drc_mem_usage += sizeof(*rp); |
412 | } | |
0338dd15 | 413 | |
a0ef5e19 | 414 | /* go ahead and prune the cache */ |
89a26b3d | 415 | prune_bucket(b); |
a0ef5e19 | 416 | |
7142b98d | 417 | found = nfsd_cache_search(b, rqstp, csum); |
0338dd15 | 418 | if (found) { |
0b9ea37f JL |
419 | if (likely(rp)) |
420 | nfsd_reply_cache_free_locked(rp); | |
0338dd15 JL |
421 | rp = found; |
422 | goto found_entry; | |
1da177e4 LT |
423 | } |
424 | ||
0b9ea37f JL |
425 | if (!rp) { |
426 | dprintk("nfsd: unable to allocate DRC entry!\n"); | |
427 | goto out; | |
428 | } | |
429 | ||
0338dd15 | 430 | nfsdstats.rcmisses++; |
1da177e4 LT |
431 | rqstp->rq_cacherep = rp; |
432 | rp->c_state = RC_INPROG; | |
433 | rp->c_xid = xid; | |
434 | rp->c_proc = proc; | |
7b9e8522 JL |
435 | rpc_copy_addr((struct sockaddr *)&rp->c_addr, svc_addr(rqstp)); |
436 | rpc_set_port((struct sockaddr *)&rp->c_addr, rpc_get_port(svc_addr(rqstp))); | |
1da177e4 LT |
437 | rp->c_prot = proto; |
438 | rp->c_vers = vers; | |
01a7decf JL |
439 | rp->c_len = rqstp->rq_arg.len; |
440 | rp->c_csum = csum; | |
1da177e4 | 441 | |
bedd4b61 | 442 | lru_put_end(b, rp); |
1da177e4 LT |
443 | |
444 | /* release any buffer */ | |
445 | if (rp->c_type == RC_REPLBUFF) { | |
6c6910cd | 446 | drc_mem_usage -= rp->c_replvec.iov_len; |
1da177e4 LT |
447 | kfree(rp->c_replvec.iov_base); |
448 | rp->c_replvec.iov_base = NULL; | |
449 | } | |
450 | rp->c_type = RC_NOCACHE; | |
451 | out: | |
89a26b3d | 452 | spin_unlock(&b->cache_lock); |
1da177e4 LT |
453 | return rtn; |
454 | ||
455 | found_entry: | |
0338dd15 | 456 | nfsdstats.rchits++; |
1da177e4 LT |
457 | /* We found a matching entry which is either in progress or done. */ |
458 | age = jiffies - rp->c_timestamp; | |
bedd4b61 | 459 | lru_put_end(b, rp); |
1da177e4 LT |
460 | |
461 | rtn = RC_DROPIT; | |
462 | /* Request being processed or excessive rexmits */ | |
463 | if (rp->c_state == RC_INPROG || age < RC_DELAY) | |
464 | goto out; | |
465 | ||
466 | /* From the hall of fame of impractical attacks: | |
467 | * Is this a user who tries to snoop on the cache? */ | |
468 | rtn = RC_DOIT; | |
4d152e2c | 469 | if (!test_bit(RQ_SECURE, &rqstp->rq_flags) && rp->c_secure) |
1da177e4 LT |
470 | goto out; |
471 | ||
472 | /* Compose RPC reply header */ | |
473 | switch (rp->c_type) { | |
474 | case RC_NOCACHE: | |
475 | break; | |
476 | case RC_REPLSTAT: | |
477 | svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat); | |
478 | rtn = RC_REPLY; | |
479 | break; | |
480 | case RC_REPLBUFF: | |
481 | if (!nfsd_cache_append(rqstp, &rp->c_replvec)) | |
482 | goto out; /* should not happen */ | |
483 | rtn = RC_REPLY; | |
484 | break; | |
485 | default: | |
486 | printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type); | |
0338dd15 | 487 | nfsd_reply_cache_free_locked(rp); |
1da177e4 LT |
488 | } |
489 | ||
490 | goto out; | |
491 | } | |
492 | ||
493 | /* | |
494 | * Update a cache entry. This is called from nfsd_dispatch when | |
495 | * the procedure has been executed and the complete reply is in | |
496 | * rqstp->rq_res. | |
497 | * | |
498 | * We're copying around data here rather than swapping buffers because | |
499 | * the toplevel loop requires max-sized buffers, which would be a waste | |
500 | * of memory for a cache with a max reply size of 100 bytes (diropokres). | |
501 | * | |
502 | * If we should start to use different types of cache entries tailored | |
503 | * specifically for attrstat and fh's, we may save even more space. | |
504 | * | |
505 | * Also note that a cachetype of RC_NOCACHE can legally be passed when | |
506 | * nfsd failed to encode a reply that otherwise would have been cached. | |
507 | * In this case, nfsd_cache_update is called with statp == NULL. | |
508 | */ | |
509 | void | |
c7afef1f | 510 | nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp) |
1da177e4 | 511 | { |
13cc8a78 | 512 | struct svc_cacherep *rp = rqstp->rq_cacherep; |
1da177e4 | 513 | struct kvec *resv = &rqstp->rq_res.head[0], *cachv; |
bedd4b61 TM |
514 | u32 hash; |
515 | struct nfsd_drc_bucket *b; | |
1da177e4 | 516 | int len; |
6c6910cd | 517 | size_t bufsize = 0; |
1da177e4 | 518 | |
13cc8a78 | 519 | if (!rp) |
1da177e4 LT |
520 | return; |
521 | ||
bedd4b61 TM |
522 | hash = nfsd_cache_hash(rp->c_xid); |
523 | b = &drc_hashtbl[hash]; | |
524 | ||
1da177e4 LT |
525 | len = resv->iov_len - ((char*)statp - (char*)resv->iov_base); |
526 | len >>= 2; | |
fca4217c | 527 | |
1da177e4 LT |
528 | /* Don't cache excessive amounts of data and XDR failures */ |
529 | if (!statp || len > (256 >> 2)) { | |
89a26b3d | 530 | nfsd_reply_cache_free(b, rp); |
1da177e4 LT |
531 | return; |
532 | } | |
533 | ||
534 | switch (cachetype) { | |
535 | case RC_REPLSTAT: | |
536 | if (len != 1) | |
537 | printk("nfsd: RC_REPLSTAT/reply len %d!\n",len); | |
538 | rp->c_replstat = *statp; | |
539 | break; | |
540 | case RC_REPLBUFF: | |
541 | cachv = &rp->c_replvec; | |
6c6910cd JL |
542 | bufsize = len << 2; |
543 | cachv->iov_base = kmalloc(bufsize, GFP_KERNEL); | |
1da177e4 | 544 | if (!cachv->iov_base) { |
89a26b3d | 545 | nfsd_reply_cache_free(b, rp); |
1da177e4 LT |
546 | return; |
547 | } | |
6c6910cd JL |
548 | cachv->iov_len = bufsize; |
549 | memcpy(cachv->iov_base, statp, bufsize); | |
1da177e4 | 550 | break; |
2c6b691c | 551 | case RC_NOCACHE: |
89a26b3d | 552 | nfsd_reply_cache_free(b, rp); |
2c6b691c | 553 | return; |
1da177e4 | 554 | } |
89a26b3d | 555 | spin_lock(&b->cache_lock); |
6c6910cd | 556 | drc_mem_usage += bufsize; |
bedd4b61 | 557 | lru_put_end(b, rp); |
4d152e2c | 558 | rp->c_secure = test_bit(RQ_SECURE, &rqstp->rq_flags); |
1da177e4 LT |
559 | rp->c_type = cachetype; |
560 | rp->c_state = RC_DONE; | |
89a26b3d | 561 | spin_unlock(&b->cache_lock); |
1da177e4 LT |
562 | return; |
563 | } | |
564 | ||
565 | /* | |
566 | * Copy cached reply to current reply buffer. Should always fit. | |
567 | * FIXME as reply is in a page, we should just attach the page, and | |
568 | * keep a refcount.... | |
569 | */ | |
570 | static int | |
571 | nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data) | |
572 | { | |
573 | struct kvec *vec = &rqstp->rq_res.head[0]; | |
574 | ||
575 | if (vec->iov_len + data->iov_len > PAGE_SIZE) { | |
576 | printk(KERN_WARNING "nfsd: cached reply too large (%Zd).\n", | |
577 | data->iov_len); | |
578 | return 0; | |
579 | } | |
580 | memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len); | |
581 | vec->iov_len += data->iov_len; | |
582 | return 1; | |
583 | } | |
a2f999a3 JL |
584 | |
585 | /* | |
586 | * Note that fields may be added, removed or reordered in the future. Programs | |
587 | * scraping this file for info should test the labels to ensure they're | |
588 | * getting the correct field. | |
589 | */ | |
590 | static int nfsd_reply_cache_stats_show(struct seq_file *m, void *v) | |
591 | { | |
a2f999a3 | 592 | seq_printf(m, "max entries: %u\n", max_drc_entries); |
31e60f52 TM |
593 | seq_printf(m, "num entries: %u\n", |
594 | atomic_read(&num_drc_entries)); | |
0733c7ba | 595 | seq_printf(m, "hash buckets: %u\n", 1 << maskbits); |
a2f999a3 JL |
596 | seq_printf(m, "mem usage: %u\n", drc_mem_usage); |
597 | seq_printf(m, "cache hits: %u\n", nfsdstats.rchits); | |
598 | seq_printf(m, "cache misses: %u\n", nfsdstats.rcmisses); | |
599 | seq_printf(m, "not cached: %u\n", nfsdstats.rcnocache); | |
600 | seq_printf(m, "payload misses: %u\n", payload_misses); | |
98d821bd JL |
601 | seq_printf(m, "longest chain len: %u\n", longest_chain); |
602 | seq_printf(m, "cachesize at longest: %u\n", longest_chain_cachesize); | |
a2f999a3 JL |
603 | return 0; |
604 | } | |
605 | ||
606 | int nfsd_reply_cache_stats_open(struct inode *inode, struct file *file) | |
607 | { | |
608 | return single_open(file, nfsd_reply_cache_stats_show, NULL); | |
609 | } |