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); |
aca8a23d | 66 | static void cache_cleaner_func(struct work_struct *unused); |
1ab6c499 DC |
67 | static unsigned long nfsd_reply_cache_count(struct shrinker *shrink, |
68 | struct shrink_control *sc); | |
69 | static unsigned long nfsd_reply_cache_scan(struct shrinker *shrink, | |
70 | struct shrink_control *sc); | |
b4e7f2c9 | 71 | |
c8c797f9 | 72 | static struct shrinker nfsd_reply_cache_shrinker = { |
1ab6c499 DC |
73 | .scan_objects = nfsd_reply_cache_scan, |
74 | .count_objects = nfsd_reply_cache_count, | |
b4e7f2c9 JL |
75 | .seeks = 1, |
76 | }; | |
1da177e4 | 77 | |
fca4217c | 78 | /* |
1da177e4 LT |
79 | * 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 | */ | |
aca8a23d | 83 | static DECLARE_DELAYED_WORK(cache_cleaner, cache_cleaner_func); |
1da177e4 | 84 | |
0338dd15 JL |
85 | /* |
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 | */ | |
103 | static unsigned int | |
104 | nfsd_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 | ||
0733c7ba JL |
113 | /* |
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 | */ | |
117 | static unsigned int | |
118 | nfsd_hashsize(unsigned int limit) | |
119 | { | |
120 | return roundup_pow_of_two(limit / TARGET_BUCKET_SIZE); | |
121 | } | |
122 | ||
7142b98d TM |
123 | static u32 |
124 | nfsd_cache_hash(__be32 xid) | |
125 | { | |
126 | return hash_32(be32_to_cpu(xid), maskbits); | |
127 | } | |
128 | ||
f09841fd JL |
129 | static struct svc_cacherep * |
130 | nfsd_reply_cache_alloc(void) | |
1da177e4 LT |
131 | { |
132 | struct svc_cacherep *rp; | |
1da177e4 | 133 | |
f09841fd JL |
134 | rp = kmem_cache_alloc(drc_slab, GFP_KERNEL); |
135 | if (rp) { | |
1da177e4 LT |
136 | rp->c_state = RC_UNUSED; |
137 | rp->c_type = RC_NOCACHE; | |
f09841fd | 138 | INIT_LIST_HEAD(&rp->c_lru); |
1da177e4 | 139 | } |
f09841fd JL |
140 | return rp; |
141 | } | |
1da177e4 | 142 | |
f09841fd JL |
143 | static void |
144 | nfsd_reply_cache_free_locked(struct svc_cacherep *rp) | |
145 | { | |
6c6910cd JL |
146 | if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) { |
147 | drc_mem_usage -= rp->c_replvec.iov_len; | |
f09841fd | 148 | kfree(rp->c_replvec.iov_base); |
6c6910cd | 149 | } |
f09841fd | 150 | list_del(&rp->c_lru); |
31e60f52 | 151 | atomic_dec(&num_drc_entries); |
6c6910cd | 152 | drc_mem_usage -= sizeof(*rp); |
f09841fd JL |
153 | kmem_cache_free(drc_slab, rp); |
154 | } | |
155 | ||
2c6b691c | 156 | static void |
89a26b3d | 157 | nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct svc_cacherep *rp) |
2c6b691c | 158 | { |
89a26b3d | 159 | spin_lock(&b->cache_lock); |
2c6b691c | 160 | nfsd_reply_cache_free_locked(rp); |
89a26b3d | 161 | spin_unlock(&b->cache_lock); |
2c6b691c JL |
162 | } |
163 | ||
f09841fd JL |
164 | int nfsd_reply_cache_init(void) |
165 | { | |
0733c7ba | 166 | unsigned int hashsize; |
bedd4b61 | 167 | unsigned int i; |
a68465c9 | 168 | int status = 0; |
0733c7ba | 169 | |
ac534ff2 | 170 | max_drc_entries = nfsd_cache_size_limit(); |
31e60f52 | 171 | atomic_set(&num_drc_entries, 0); |
0733c7ba JL |
172 | hashsize = nfsd_hashsize(max_drc_entries); |
173 | maskbits = ilog2(hashsize); | |
ac534ff2 | 174 | |
a68465c9 KM |
175 | status = register_shrinker(&nfsd_reply_cache_shrinker); |
176 | if (status) | |
177 | return status; | |
178 | ||
8a8bc40d JL |
179 | drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep), |
180 | 0, 0, NULL); | |
181 | if (!drc_slab) | |
182 | goto out_nomem; | |
183 | ||
7142b98d TM |
184 | drc_hashtbl = kcalloc(hashsize, sizeof(*drc_hashtbl), GFP_KERNEL); |
185 | if (!drc_hashtbl) | |
d5c3428b | 186 | goto out_nomem; |
89a26b3d | 187 | for (i = 0; i < hashsize; i++) { |
bedd4b61 | 188 | INIT_LIST_HEAD(&drc_hashtbl[i].lru_head); |
89a26b3d TM |
189 | spin_lock_init(&drc_hashtbl[i].cache_lock); |
190 | } | |
bedd4b61 | 191 | drc_hashsize = hashsize; |
1da177e4 | 192 | |
d5c3428b BF |
193 | return 0; |
194 | out_nomem: | |
195 | printk(KERN_ERR "nfsd: failed to allocate reply cache\n"); | |
196 | nfsd_reply_cache_shutdown(); | |
197 | return -ENOMEM; | |
1da177e4 LT |
198 | } |
199 | ||
d5c3428b | 200 | void nfsd_reply_cache_shutdown(void) |
1da177e4 LT |
201 | { |
202 | struct svc_cacherep *rp; | |
bedd4b61 | 203 | unsigned int i; |
1da177e4 | 204 | |
b4e7f2c9 | 205 | unregister_shrinker(&nfsd_reply_cache_shrinker); |
aca8a23d JL |
206 | cancel_delayed_work_sync(&cache_cleaner); |
207 | ||
bedd4b61 TM |
208 | for (i = 0; i < drc_hashsize; i++) { |
209 | struct list_head *head = &drc_hashtbl[i].lru_head; | |
210 | while (!list_empty(head)) { | |
211 | rp = list_first_entry(head, struct svc_cacherep, c_lru); | |
212 | nfsd_reply_cache_free_locked(rp); | |
213 | } | |
1da177e4 LT |
214 | } |
215 | ||
7142b98d TM |
216 | kfree (drc_hashtbl); |
217 | drc_hashtbl = NULL; | |
bedd4b61 | 218 | drc_hashsize = 0; |
8a8bc40d JL |
219 | |
220 | if (drc_slab) { | |
221 | kmem_cache_destroy(drc_slab); | |
222 | drc_slab = NULL; | |
223 | } | |
1da177e4 LT |
224 | } |
225 | ||
226 | /* | |
aca8a23d JL |
227 | * Move cache entry to end of LRU list, and queue the cleaner to run if it's |
228 | * not already scheduled. | |
1da177e4 LT |
229 | */ |
230 | static void | |
bedd4b61 | 231 | lru_put_end(struct nfsd_drc_bucket *b, struct svc_cacherep *rp) |
1da177e4 | 232 | { |
56c2548b | 233 | rp->c_timestamp = jiffies; |
bedd4b61 | 234 | list_move_tail(&rp->c_lru, &b->lru_head); |
aca8a23d | 235 | schedule_delayed_work(&cache_cleaner, RC_EXPIRE); |
1da177e4 LT |
236 | } |
237 | ||
1ab6c499 | 238 | static long |
bedd4b61 | 239 | prune_bucket(struct nfsd_drc_bucket *b) |
aca8a23d JL |
240 | { |
241 | struct svc_cacherep *rp, *tmp; | |
1ab6c499 | 242 | long freed = 0; |
aca8a23d | 243 | |
bedd4b61 | 244 | list_for_each_entry_safe(rp, tmp, &b->lru_head, c_lru) { |
1b19453d JL |
245 | /* |
246 | * Don't free entries attached to calls that are still | |
247 | * in-progress, but do keep scanning the list. | |
248 | */ | |
249 | if (rp->c_state == RC_INPROG) | |
250 | continue; | |
31e60f52 | 251 | if (atomic_read(&num_drc_entries) <= max_drc_entries && |
1b19453d | 252 | time_before(jiffies, rp->c_timestamp + RC_EXPIRE)) |
aca8a23d JL |
253 | break; |
254 | nfsd_reply_cache_free_locked(rp); | |
1ab6c499 | 255 | freed++; |
aca8a23d | 256 | } |
bedd4b61 TM |
257 | return freed; |
258 | } | |
259 | ||
260 | /* | |
261 | * Walk the LRU list and prune off entries that are older than RC_EXPIRE. | |
262 | * Also prune the oldest ones when the total exceeds the max number of entries. | |
263 | */ | |
264 | static long | |
265 | prune_cache_entries(void) | |
266 | { | |
267 | unsigned int i; | |
268 | long freed = 0; | |
269 | bool cancel = true; | |
270 | ||
271 | for (i = 0; i < drc_hashsize; i++) { | |
272 | struct nfsd_drc_bucket *b = &drc_hashtbl[i]; | |
273 | ||
89a26b3d TM |
274 | if (list_empty(&b->lru_head)) |
275 | continue; | |
276 | spin_lock(&b->cache_lock); | |
bedd4b61 TM |
277 | freed += prune_bucket(b); |
278 | if (!list_empty(&b->lru_head)) | |
279 | cancel = false; | |
89a26b3d | 280 | spin_unlock(&b->cache_lock); |
bedd4b61 | 281 | } |
aca8a23d JL |
282 | |
283 | /* | |
bedd4b61 TM |
284 | * Conditionally rearm the job to run in RC_EXPIRE since we just |
285 | * ran the pruner. | |
aca8a23d | 286 | */ |
bedd4b61 | 287 | if (!cancel) |
aca8a23d | 288 | mod_delayed_work(system_wq, &cache_cleaner, RC_EXPIRE); |
1ab6c499 | 289 | return freed; |
aca8a23d JL |
290 | } |
291 | ||
292 | static void | |
293 | cache_cleaner_func(struct work_struct *unused) | |
294 | { | |
aca8a23d | 295 | prune_cache_entries(); |
aca8a23d JL |
296 | } |
297 | ||
1ab6c499 DC |
298 | static unsigned long |
299 | nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
b4e7f2c9 | 300 | { |
31e60f52 | 301 | return atomic_read(&num_drc_entries); |
b4e7f2c9 JL |
302 | } |
303 | ||
1ab6c499 DC |
304 | static unsigned long |
305 | nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
306 | { | |
89a26b3d | 307 | return prune_cache_entries(); |
1ab6c499 | 308 | } |
01a7decf JL |
309 | /* |
310 | * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes | |
311 | */ | |
312 | static __wsum | |
313 | nfsd_cache_csum(struct svc_rqst *rqstp) | |
314 | { | |
315 | int idx; | |
316 | unsigned int base; | |
317 | __wsum csum; | |
318 | struct xdr_buf *buf = &rqstp->rq_arg; | |
319 | const unsigned char *p = buf->head[0].iov_base; | |
320 | size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len, | |
321 | RC_CSUMLEN); | |
322 | size_t len = min(buf->head[0].iov_len, csum_len); | |
323 | ||
324 | /* rq_arg.head first */ | |
325 | csum = csum_partial(p, len, 0); | |
326 | csum_len -= len; | |
327 | ||
328 | /* Continue into page array */ | |
329 | idx = buf->page_base / PAGE_SIZE; | |
330 | base = buf->page_base & ~PAGE_MASK; | |
331 | while (csum_len) { | |
332 | p = page_address(buf->pages[idx]) + base; | |
56edc86b | 333 | len = min_t(size_t, PAGE_SIZE - base, csum_len); |
01a7decf JL |
334 | csum = csum_partial(p, len, csum); |
335 | csum_len -= len; | |
336 | base = 0; | |
337 | ++idx; | |
338 | } | |
339 | return csum; | |
340 | } | |
341 | ||
9dc56143 JL |
342 | static bool |
343 | nfsd_cache_match(struct svc_rqst *rqstp, __wsum csum, struct svc_cacherep *rp) | |
344 | { | |
ef9b16dc TM |
345 | /* Check RPC XID first */ |
346 | if (rqstp->rq_xid != rp->c_xid) | |
9dc56143 | 347 | return false; |
9dc56143 JL |
348 | /* compare checksum of NFS data */ |
349 | if (csum != rp->c_csum) { | |
350 | ++payload_misses; | |
351 | return false; | |
352 | } | |
353 | ||
ef9b16dc TM |
354 | /* Other discriminators */ |
355 | if (rqstp->rq_proc != rp->c_proc || | |
356 | rqstp->rq_prot != rp->c_prot || | |
357 | rqstp->rq_vers != rp->c_vers || | |
358 | rqstp->rq_arg.len != rp->c_len || | |
359 | !rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) || | |
360 | rpc_get_port(svc_addr(rqstp)) != rpc_get_port((struct sockaddr *)&rp->c_addr)) | |
361 | return false; | |
362 | ||
9dc56143 JL |
363 | return true; |
364 | } | |
365 | ||
a4a3ec32 JL |
366 | /* |
367 | * Search the request hash for an entry that matches the given rqstp. | |
368 | * Must be called with cache_lock held. Returns the found entry or | |
369 | * NULL on failure. | |
370 | */ | |
371 | static struct svc_cacherep * | |
7142b98d TM |
372 | nfsd_cache_search(struct nfsd_drc_bucket *b, struct svc_rqst *rqstp, |
373 | __wsum csum) | |
a4a3ec32 | 374 | { |
98d821bd | 375 | struct svc_cacherep *rp, *ret = NULL; |
11acf6ef | 376 | struct list_head *rh = &b->lru_head; |
98d821bd | 377 | unsigned int entries = 0; |
a4a3ec32 | 378 | |
11acf6ef | 379 | list_for_each_entry(rp, rh, c_lru) { |
98d821bd JL |
380 | ++entries; |
381 | if (nfsd_cache_match(rqstp, csum, rp)) { | |
382 | ret = rp; | |
383 | break; | |
384 | } | |
385 | } | |
386 | ||
387 | /* tally hash chain length stats */ | |
388 | if (entries > longest_chain) { | |
389 | longest_chain = entries; | |
31e60f52 | 390 | longest_chain_cachesize = atomic_read(&num_drc_entries); |
98d821bd JL |
391 | } else if (entries == longest_chain) { |
392 | /* prefer to keep the smallest cachesize possible here */ | |
31e60f52 TM |
393 | longest_chain_cachesize = min_t(unsigned int, |
394 | longest_chain_cachesize, | |
395 | atomic_read(&num_drc_entries)); | |
a4a3ec32 | 396 | } |
98d821bd JL |
397 | |
398 | return ret; | |
a4a3ec32 JL |
399 | } |
400 | ||
1da177e4 LT |
401 | /* |
402 | * Try to find an entry matching the current call in the cache. When none | |
1ac83629 JL |
403 | * is found, we try to grab the oldest expired entry off the LRU list. If |
404 | * a suitable one isn't there, then drop the cache_lock and allocate a | |
405 | * new one, then search again in case one got inserted while this thread | |
406 | * didn't hold the lock. | |
1da177e4 LT |
407 | */ |
408 | int | |
1091006c | 409 | nfsd_cache_lookup(struct svc_rqst *rqstp) |
1da177e4 | 410 | { |
0338dd15 | 411 | struct svc_cacherep *rp, *found; |
c7afef1f AV |
412 | __be32 xid = rqstp->rq_xid; |
413 | u32 proto = rqstp->rq_prot, | |
1da177e4 LT |
414 | vers = rqstp->rq_vers, |
415 | proc = rqstp->rq_proc; | |
01a7decf | 416 | __wsum csum; |
7142b98d TM |
417 | u32 hash = nfsd_cache_hash(xid); |
418 | struct nfsd_drc_bucket *b = &drc_hashtbl[hash]; | |
1da177e4 | 419 | unsigned long age; |
1091006c | 420 | int type = rqstp->rq_cachetype; |
0b9ea37f | 421 | int rtn = RC_DOIT; |
1da177e4 LT |
422 | |
423 | rqstp->rq_cacherep = NULL; | |
13cc8a78 | 424 | if (type == RC_NOCACHE) { |
1da177e4 | 425 | nfsdstats.rcnocache++; |
0b9ea37f | 426 | return rtn; |
1da177e4 LT |
427 | } |
428 | ||
01a7decf JL |
429 | csum = nfsd_cache_csum(rqstp); |
430 | ||
0b9ea37f JL |
431 | /* |
432 | * Since the common case is a cache miss followed by an insert, | |
a0ef5e19 | 433 | * preallocate an entry. |
0b9ea37f | 434 | */ |
0338dd15 | 435 | rp = nfsd_reply_cache_alloc(); |
89a26b3d | 436 | spin_lock(&b->cache_lock); |
6c6910cd | 437 | if (likely(rp)) { |
31e60f52 | 438 | atomic_inc(&num_drc_entries); |
6c6910cd JL |
439 | drc_mem_usage += sizeof(*rp); |
440 | } | |
0338dd15 | 441 | |
a0ef5e19 | 442 | /* go ahead and prune the cache */ |
89a26b3d | 443 | prune_bucket(b); |
a0ef5e19 | 444 | |
7142b98d | 445 | found = nfsd_cache_search(b, rqstp, csum); |
0338dd15 | 446 | if (found) { |
0b9ea37f JL |
447 | if (likely(rp)) |
448 | nfsd_reply_cache_free_locked(rp); | |
0338dd15 JL |
449 | rp = found; |
450 | goto found_entry; | |
1da177e4 LT |
451 | } |
452 | ||
0b9ea37f JL |
453 | if (!rp) { |
454 | dprintk("nfsd: unable to allocate DRC entry!\n"); | |
455 | goto out; | |
456 | } | |
457 | ||
0338dd15 | 458 | nfsdstats.rcmisses++; |
1da177e4 LT |
459 | rqstp->rq_cacherep = rp; |
460 | rp->c_state = RC_INPROG; | |
461 | rp->c_xid = xid; | |
462 | rp->c_proc = proc; | |
7b9e8522 JL |
463 | 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))); | |
1da177e4 LT |
465 | rp->c_prot = proto; |
466 | rp->c_vers = vers; | |
01a7decf JL |
467 | rp->c_len = rqstp->rq_arg.len; |
468 | rp->c_csum = csum; | |
1da177e4 | 469 | |
bedd4b61 | 470 | lru_put_end(b, rp); |
1da177e4 LT |
471 | |
472 | /* release any buffer */ | |
473 | if (rp->c_type == RC_REPLBUFF) { | |
6c6910cd | 474 | drc_mem_usage -= rp->c_replvec.iov_len; |
1da177e4 LT |
475 | kfree(rp->c_replvec.iov_base); |
476 | rp->c_replvec.iov_base = NULL; | |
477 | } | |
478 | rp->c_type = RC_NOCACHE; | |
479 | out: | |
89a26b3d | 480 | spin_unlock(&b->cache_lock); |
1da177e4 LT |
481 | return rtn; |
482 | ||
483 | found_entry: | |
0338dd15 | 484 | nfsdstats.rchits++; |
1da177e4 LT |
485 | /* We found a matching entry which is either in progress or done. */ |
486 | age = jiffies - rp->c_timestamp; | |
bedd4b61 | 487 | lru_put_end(b, rp); |
1da177e4 LT |
488 | |
489 | rtn = RC_DROPIT; | |
490 | /* Request being processed or excessive rexmits */ | |
491 | if (rp->c_state == RC_INPROG || age < RC_DELAY) | |
492 | goto out; | |
493 | ||
494 | /* From the hall of fame of impractical attacks: | |
495 | * Is this a user who tries to snoop on the cache? */ | |
496 | rtn = RC_DOIT; | |
4d152e2c | 497 | if (!test_bit(RQ_SECURE, &rqstp->rq_flags) && rp->c_secure) |
1da177e4 LT |
498 | goto out; |
499 | ||
500 | /* Compose RPC reply header */ | |
501 | switch (rp->c_type) { | |
502 | case RC_NOCACHE: | |
503 | break; | |
504 | case RC_REPLSTAT: | |
505 | svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat); | |
506 | rtn = RC_REPLY; | |
507 | break; | |
508 | case RC_REPLBUFF: | |
509 | if (!nfsd_cache_append(rqstp, &rp->c_replvec)) | |
510 | goto out; /* should not happen */ | |
511 | rtn = RC_REPLY; | |
512 | break; | |
513 | default: | |
514 | printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type); | |
0338dd15 | 515 | nfsd_reply_cache_free_locked(rp); |
1da177e4 LT |
516 | } |
517 | ||
518 | goto out; | |
519 | } | |
520 | ||
521 | /* | |
522 | * Update a cache entry. This is called from nfsd_dispatch when | |
523 | * the procedure has been executed and the complete reply is in | |
524 | * rqstp->rq_res. | |
525 | * | |
526 | * We're copying around data here rather than swapping buffers because | |
527 | * the toplevel loop requires max-sized buffers, which would be a waste | |
528 | * of memory for a cache with a max reply size of 100 bytes (diropokres). | |
529 | * | |
530 | * If we should start to use different types of cache entries tailored | |
531 | * specifically for attrstat and fh's, we may save even more space. | |
532 | * | |
533 | * Also note that a cachetype of RC_NOCACHE can legally be passed when | |
534 | * nfsd failed to encode a reply that otherwise would have been cached. | |
535 | * In this case, nfsd_cache_update is called with statp == NULL. | |
536 | */ | |
537 | void | |
c7afef1f | 538 | nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp) |
1da177e4 | 539 | { |
13cc8a78 | 540 | struct svc_cacherep *rp = rqstp->rq_cacherep; |
1da177e4 | 541 | struct kvec *resv = &rqstp->rq_res.head[0], *cachv; |
bedd4b61 TM |
542 | u32 hash; |
543 | struct nfsd_drc_bucket *b; | |
1da177e4 | 544 | int len; |
6c6910cd | 545 | size_t bufsize = 0; |
1da177e4 | 546 | |
13cc8a78 | 547 | if (!rp) |
1da177e4 LT |
548 | return; |
549 | ||
bedd4b61 TM |
550 | hash = nfsd_cache_hash(rp->c_xid); |
551 | b = &drc_hashtbl[hash]; | |
552 | ||
1da177e4 LT |
553 | len = resv->iov_len - ((char*)statp - (char*)resv->iov_base); |
554 | len >>= 2; | |
fca4217c | 555 | |
1da177e4 LT |
556 | /* Don't cache excessive amounts of data and XDR failures */ |
557 | if (!statp || len > (256 >> 2)) { | |
89a26b3d | 558 | nfsd_reply_cache_free(b, rp); |
1da177e4 LT |
559 | return; |
560 | } | |
561 | ||
562 | switch (cachetype) { | |
563 | case RC_REPLSTAT: | |
564 | if (len != 1) | |
565 | printk("nfsd: RC_REPLSTAT/reply len %d!\n",len); | |
566 | rp->c_replstat = *statp; | |
567 | break; | |
568 | case RC_REPLBUFF: | |
569 | cachv = &rp->c_replvec; | |
6c6910cd JL |
570 | bufsize = len << 2; |
571 | cachv->iov_base = kmalloc(bufsize, GFP_KERNEL); | |
1da177e4 | 572 | if (!cachv->iov_base) { |
89a26b3d | 573 | nfsd_reply_cache_free(b, rp); |
1da177e4 LT |
574 | return; |
575 | } | |
6c6910cd JL |
576 | cachv->iov_len = bufsize; |
577 | memcpy(cachv->iov_base, statp, bufsize); | |
1da177e4 | 578 | break; |
2c6b691c | 579 | case RC_NOCACHE: |
89a26b3d | 580 | nfsd_reply_cache_free(b, rp); |
2c6b691c | 581 | return; |
1da177e4 | 582 | } |
89a26b3d | 583 | spin_lock(&b->cache_lock); |
6c6910cd | 584 | drc_mem_usage += bufsize; |
bedd4b61 | 585 | lru_put_end(b, rp); |
4d152e2c | 586 | rp->c_secure = test_bit(RQ_SECURE, &rqstp->rq_flags); |
1da177e4 LT |
587 | rp->c_type = cachetype; |
588 | rp->c_state = RC_DONE; | |
89a26b3d | 589 | spin_unlock(&b->cache_lock); |
1da177e4 LT |
590 | return; |
591 | } | |
592 | ||
593 | /* | |
594 | * Copy cached reply to current reply buffer. Should always fit. | |
595 | * FIXME as reply is in a page, we should just attach the page, and | |
596 | * keep a refcount.... | |
597 | */ | |
598 | static int | |
599 | nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data) | |
600 | { | |
601 | struct kvec *vec = &rqstp->rq_res.head[0]; | |
602 | ||
603 | if (vec->iov_len + data->iov_len > PAGE_SIZE) { | |
604 | printk(KERN_WARNING "nfsd: cached reply too large (%Zd).\n", | |
605 | data->iov_len); | |
606 | return 0; | |
607 | } | |
608 | memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len); | |
609 | vec->iov_len += data->iov_len; | |
610 | return 1; | |
611 | } | |
a2f999a3 JL |
612 | |
613 | /* | |
614 | * Note that fields may be added, removed or reordered in the future. Programs | |
615 | * scraping this file for info should test the labels to ensure they're | |
616 | * getting the correct field. | |
617 | */ | |
618 | static int nfsd_reply_cache_stats_show(struct seq_file *m, void *v) | |
619 | { | |
a2f999a3 | 620 | seq_printf(m, "max entries: %u\n", max_drc_entries); |
31e60f52 TM |
621 | seq_printf(m, "num entries: %u\n", |
622 | atomic_read(&num_drc_entries)); | |
0733c7ba | 623 | seq_printf(m, "hash buckets: %u\n", 1 << maskbits); |
a2f999a3 JL |
624 | seq_printf(m, "mem usage: %u\n", drc_mem_usage); |
625 | seq_printf(m, "cache hits: %u\n", nfsdstats.rchits); | |
626 | seq_printf(m, "cache misses: %u\n", nfsdstats.rcmisses); | |
627 | seq_printf(m, "not cached: %u\n", nfsdstats.rcnocache); | |
628 | seq_printf(m, "payload misses: %u\n", payload_misses); | |
98d821bd JL |
629 | seq_printf(m, "longest chain len: %u\n", longest_chain); |
630 | seq_printf(m, "cachesize at longest: %u\n", longest_chain_cachesize); | |
a2f999a3 JL |
631 | return 0; |
632 | } | |
633 | ||
634 | int nfsd_reply_cache_stats_open(struct inode *inode, struct file *file) | |
635 | { | |
636 | return single_open(file, nfsd_reply_cache_stats_show, NULL); | |
637 | } |