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 | ||
5a0e3ad6 | 12 | #include <linux/slab.h> |
8f97514b | 13 | #include <linux/vmalloc.h> |
5976687a | 14 | #include <linux/sunrpc/addr.h> |
0338dd15 | 15 | #include <linux/highmem.h> |
0733c7ba JL |
16 | #include <linux/log2.h> |
17 | #include <linux/hash.h> | |
01a7decf | 18 | #include <net/checksum.h> |
5a0e3ad6 | 19 | |
9a74af21 BH |
20 | #include "nfsd.h" |
21 | #include "cache.h" | |
1da177e4 | 22 | |
0338dd15 JL |
23 | #define NFSDDBG_FACILITY NFSDDBG_REPCACHE |
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 { |
bedd4b61 | 33 | struct list_head lru_head; |
89a26b3d | 34 | spinlock_t cache_lock; |
7142b98d TM |
35 | }; |
36 | ||
37 | static struct nfsd_drc_bucket *drc_hashtbl; | |
8a8bc40d | 38 | static struct kmem_cache *drc_slab; |
9dc56143 JL |
39 | |
40 | /* max number of entries allowed in the cache */ | |
0338dd15 | 41 | static unsigned int max_drc_entries; |
1da177e4 | 42 | |
0733c7ba JL |
43 | /* number of significant bits in the hash value */ |
44 | static unsigned int maskbits; | |
bedd4b61 | 45 | static unsigned int drc_hashsize; |
0733c7ba | 46 | |
9dc56143 JL |
47 | /* |
48 | * Stats and other tracking of on the duplicate reply cache. All of these and | |
49 | * the "rc" fields in nfsdstats are protected by the cache_lock | |
50 | */ | |
51 | ||
52 | /* total number of entries */ | |
31e60f52 | 53 | static atomic_t num_drc_entries; |
9dc56143 JL |
54 | |
55 | /* cache misses due only to checksum comparison failures */ | |
56 | static unsigned int payload_misses; | |
57 | ||
6c6910cd JL |
58 | /* amount of memory (in bytes) currently consumed by the DRC */ |
59 | static unsigned int drc_mem_usage; | |
60 | ||
98d821bd JL |
61 | /* longest hash chain seen */ |
62 | static unsigned int longest_chain; | |
63 | ||
64 | /* size of cache when we saw the longest hash chain */ | |
65 | static unsigned int longest_chain_cachesize; | |
66 | ||
1da177e4 | 67 | static int nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec); |
1ab6c499 DC |
68 | static unsigned long nfsd_reply_cache_count(struct shrinker *shrink, |
69 | struct shrink_control *sc); | |
70 | static unsigned long nfsd_reply_cache_scan(struct shrinker *shrink, | |
71 | struct shrink_control *sc); | |
b4e7f2c9 | 72 | |
c8c797f9 | 73 | static struct shrinker nfsd_reply_cache_shrinker = { |
1ab6c499 DC |
74 | .scan_objects = nfsd_reply_cache_scan, |
75 | .count_objects = nfsd_reply_cache_count, | |
b4e7f2c9 JL |
76 | .seeks = 1, |
77 | }; | |
1da177e4 | 78 | |
0338dd15 JL |
79 | /* |
80 | * Put a cap on the size of the DRC based on the amount of available | |
81 | * low memory in the machine. | |
82 | * | |
83 | * 64MB: 8192 | |
84 | * 128MB: 11585 | |
85 | * 256MB: 16384 | |
86 | * 512MB: 23170 | |
87 | * 1GB: 32768 | |
88 | * 2GB: 46340 | |
89 | * 4GB: 65536 | |
90 | * 8GB: 92681 | |
91 | * 16GB: 131072 | |
92 | * | |
93 | * ...with a hard cap of 256k entries. In the worst case, each entry will be | |
94 | * ~1k, so the above numbers should give a rough max of the amount of memory | |
95 | * used in k. | |
96 | */ | |
97 | static unsigned int | |
98 | nfsd_cache_size_limit(void) | |
99 | { | |
100 | unsigned int limit; | |
101 | unsigned long low_pages = totalram_pages - totalhigh_pages; | |
102 | ||
103 | limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10); | |
104 | return min_t(unsigned int, limit, 256*1024); | |
105 | } | |
106 | ||
0733c7ba JL |
107 | /* |
108 | * Compute the number of hash buckets we need. Divide the max cachesize by | |
109 | * the "target" max bucket size, and round up to next power of two. | |
110 | */ | |
111 | static unsigned int | |
112 | nfsd_hashsize(unsigned int limit) | |
113 | { | |
114 | return roundup_pow_of_two(limit / TARGET_BUCKET_SIZE); | |
115 | } | |
116 | ||
7142b98d TM |
117 | static u32 |
118 | nfsd_cache_hash(__be32 xid) | |
119 | { | |
120 | return hash_32(be32_to_cpu(xid), maskbits); | |
121 | } | |
122 | ||
f09841fd JL |
123 | static struct svc_cacherep * |
124 | nfsd_reply_cache_alloc(void) | |
1da177e4 LT |
125 | { |
126 | struct svc_cacherep *rp; | |
1da177e4 | 127 | |
f09841fd JL |
128 | rp = kmem_cache_alloc(drc_slab, GFP_KERNEL); |
129 | if (rp) { | |
1da177e4 LT |
130 | rp->c_state = RC_UNUSED; |
131 | rp->c_type = RC_NOCACHE; | |
f09841fd | 132 | INIT_LIST_HEAD(&rp->c_lru); |
1da177e4 | 133 | } |
f09841fd JL |
134 | return rp; |
135 | } | |
1da177e4 | 136 | |
f09841fd JL |
137 | static void |
138 | nfsd_reply_cache_free_locked(struct svc_cacherep *rp) | |
139 | { | |
6c6910cd JL |
140 | if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) { |
141 | drc_mem_usage -= rp->c_replvec.iov_len; | |
f09841fd | 142 | kfree(rp->c_replvec.iov_base); |
6c6910cd | 143 | } |
f09841fd | 144 | list_del(&rp->c_lru); |
31e60f52 | 145 | atomic_dec(&num_drc_entries); |
6c6910cd | 146 | drc_mem_usage -= sizeof(*rp); |
f09841fd JL |
147 | kmem_cache_free(drc_slab, rp); |
148 | } | |
149 | ||
2c6b691c | 150 | static void |
89a26b3d | 151 | nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct svc_cacherep *rp) |
2c6b691c | 152 | { |
89a26b3d | 153 | spin_lock(&b->cache_lock); |
2c6b691c | 154 | nfsd_reply_cache_free_locked(rp); |
89a26b3d | 155 | spin_unlock(&b->cache_lock); |
2c6b691c JL |
156 | } |
157 | ||
f09841fd JL |
158 | int nfsd_reply_cache_init(void) |
159 | { | |
0733c7ba | 160 | unsigned int hashsize; |
bedd4b61 | 161 | unsigned int i; |
a68465c9 | 162 | int status = 0; |
0733c7ba | 163 | |
ac534ff2 | 164 | max_drc_entries = nfsd_cache_size_limit(); |
31e60f52 | 165 | atomic_set(&num_drc_entries, 0); |
0733c7ba JL |
166 | hashsize = nfsd_hashsize(max_drc_entries); |
167 | maskbits = ilog2(hashsize); | |
ac534ff2 | 168 | |
a68465c9 KM |
169 | status = register_shrinker(&nfsd_reply_cache_shrinker); |
170 | if (status) | |
171 | return status; | |
172 | ||
8a8bc40d JL |
173 | drc_slab = kmem_cache_create("nfsd_drc", sizeof(struct svc_cacherep), |
174 | 0, 0, NULL); | |
175 | if (!drc_slab) | |
176 | goto out_nomem; | |
177 | ||
7142b98d | 178 | drc_hashtbl = kcalloc(hashsize, sizeof(*drc_hashtbl), GFP_KERNEL); |
8f97514b JL |
179 | if (!drc_hashtbl) { |
180 | drc_hashtbl = vzalloc(hashsize * sizeof(*drc_hashtbl)); | |
181 | if (!drc_hashtbl) | |
182 | goto out_nomem; | |
183 | } | |
184 | ||
89a26b3d | 185 | for (i = 0; i < hashsize; i++) { |
bedd4b61 | 186 | INIT_LIST_HEAD(&drc_hashtbl[i].lru_head); |
89a26b3d TM |
187 | spin_lock_init(&drc_hashtbl[i].cache_lock); |
188 | } | |
bedd4b61 | 189 | drc_hashsize = hashsize; |
1da177e4 | 190 | |
d5c3428b BF |
191 | return 0; |
192 | out_nomem: | |
193 | printk(KERN_ERR "nfsd: failed to allocate reply cache\n"); | |
194 | nfsd_reply_cache_shutdown(); | |
195 | return -ENOMEM; | |
1da177e4 LT |
196 | } |
197 | ||
d5c3428b | 198 | void nfsd_reply_cache_shutdown(void) |
1da177e4 LT |
199 | { |
200 | struct svc_cacherep *rp; | |
bedd4b61 | 201 | unsigned int i; |
1da177e4 | 202 | |
b4e7f2c9 | 203 | unregister_shrinker(&nfsd_reply_cache_shrinker); |
aca8a23d | 204 | |
bedd4b61 TM |
205 | for (i = 0; i < drc_hashsize; i++) { |
206 | struct list_head *head = &drc_hashtbl[i].lru_head; | |
207 | while (!list_empty(head)) { | |
208 | rp = list_first_entry(head, struct svc_cacherep, c_lru); | |
209 | nfsd_reply_cache_free_locked(rp); | |
210 | } | |
1da177e4 LT |
211 | } |
212 | ||
8f97514b | 213 | kvfree(drc_hashtbl); |
7142b98d | 214 | drc_hashtbl = NULL; |
bedd4b61 | 215 | drc_hashsize = 0; |
8a8bc40d | 216 | |
e79017dd JL |
217 | kmem_cache_destroy(drc_slab); |
218 | drc_slab = NULL; | |
1da177e4 LT |
219 | } |
220 | ||
221 | /* | |
aca8a23d JL |
222 | * Move cache entry to end of LRU list, and queue the cleaner to run if it's |
223 | * not already scheduled. | |
1da177e4 LT |
224 | */ |
225 | static void | |
bedd4b61 | 226 | lru_put_end(struct nfsd_drc_bucket *b, struct svc_cacherep *rp) |
1da177e4 | 227 | { |
56c2548b | 228 | rp->c_timestamp = jiffies; |
bedd4b61 | 229 | list_move_tail(&rp->c_lru, &b->lru_head); |
1da177e4 LT |
230 | } |
231 | ||
1ab6c499 | 232 | static long |
bedd4b61 | 233 | prune_bucket(struct nfsd_drc_bucket *b) |
aca8a23d JL |
234 | { |
235 | struct svc_cacherep *rp, *tmp; | |
1ab6c499 | 236 | long freed = 0; |
aca8a23d | 237 | |
bedd4b61 | 238 | list_for_each_entry_safe(rp, tmp, &b->lru_head, c_lru) { |
1b19453d JL |
239 | /* |
240 | * Don't free entries attached to calls that are still | |
241 | * in-progress, but do keep scanning the list. | |
242 | */ | |
243 | if (rp->c_state == RC_INPROG) | |
244 | continue; | |
31e60f52 | 245 | if (atomic_read(&num_drc_entries) <= max_drc_entries && |
1b19453d | 246 | time_before(jiffies, rp->c_timestamp + RC_EXPIRE)) |
aca8a23d JL |
247 | break; |
248 | nfsd_reply_cache_free_locked(rp); | |
1ab6c499 | 249 | freed++; |
aca8a23d | 250 | } |
bedd4b61 TM |
251 | return freed; |
252 | } | |
253 | ||
254 | /* | |
255 | * Walk the LRU list and prune off entries that are older than RC_EXPIRE. | |
256 | * Also prune the oldest ones when the total exceeds the max number of entries. | |
257 | */ | |
258 | static long | |
259 | prune_cache_entries(void) | |
260 | { | |
261 | unsigned int i; | |
262 | long freed = 0; | |
bedd4b61 TM |
263 | |
264 | for (i = 0; i < drc_hashsize; i++) { | |
265 | struct nfsd_drc_bucket *b = &drc_hashtbl[i]; | |
266 | ||
89a26b3d TM |
267 | if (list_empty(&b->lru_head)) |
268 | continue; | |
269 | spin_lock(&b->cache_lock); | |
bedd4b61 | 270 | freed += prune_bucket(b); |
89a26b3d | 271 | spin_unlock(&b->cache_lock); |
bedd4b61 | 272 | } |
1ab6c499 | 273 | return freed; |
aca8a23d JL |
274 | } |
275 | ||
1ab6c499 DC |
276 | static unsigned long |
277 | nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc) | |
b4e7f2c9 | 278 | { |
31e60f52 | 279 | return atomic_read(&num_drc_entries); |
b4e7f2c9 JL |
280 | } |
281 | ||
1ab6c499 DC |
282 | static unsigned long |
283 | nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc) | |
284 | { | |
89a26b3d | 285 | return prune_cache_entries(); |
1ab6c499 | 286 | } |
01a7decf JL |
287 | /* |
288 | * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes | |
289 | */ | |
290 | static __wsum | |
291 | nfsd_cache_csum(struct svc_rqst *rqstp) | |
292 | { | |
293 | int idx; | |
294 | unsigned int base; | |
295 | __wsum csum; | |
296 | struct xdr_buf *buf = &rqstp->rq_arg; | |
297 | const unsigned char *p = buf->head[0].iov_base; | |
298 | size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len, | |
299 | RC_CSUMLEN); | |
300 | size_t len = min(buf->head[0].iov_len, csum_len); | |
301 | ||
302 | /* rq_arg.head first */ | |
303 | csum = csum_partial(p, len, 0); | |
304 | csum_len -= len; | |
305 | ||
306 | /* Continue into page array */ | |
307 | idx = buf->page_base / PAGE_SIZE; | |
308 | base = buf->page_base & ~PAGE_MASK; | |
309 | while (csum_len) { | |
310 | p = page_address(buf->pages[idx]) + base; | |
56edc86b | 311 | len = min_t(size_t, PAGE_SIZE - base, csum_len); |
01a7decf JL |
312 | csum = csum_partial(p, len, csum); |
313 | csum_len -= len; | |
314 | base = 0; | |
315 | ++idx; | |
316 | } | |
317 | return csum; | |
318 | } | |
319 | ||
9dc56143 JL |
320 | static bool |
321 | nfsd_cache_match(struct svc_rqst *rqstp, __wsum csum, struct svc_cacherep *rp) | |
322 | { | |
ef9b16dc TM |
323 | /* Check RPC XID first */ |
324 | if (rqstp->rq_xid != rp->c_xid) | |
9dc56143 | 325 | return false; |
9dc56143 JL |
326 | /* compare checksum of NFS data */ |
327 | if (csum != rp->c_csum) { | |
328 | ++payload_misses; | |
329 | return false; | |
330 | } | |
331 | ||
ef9b16dc TM |
332 | /* Other discriminators */ |
333 | if (rqstp->rq_proc != rp->c_proc || | |
334 | rqstp->rq_prot != rp->c_prot || | |
335 | rqstp->rq_vers != rp->c_vers || | |
336 | rqstp->rq_arg.len != rp->c_len || | |
337 | !rpc_cmp_addr(svc_addr(rqstp), (struct sockaddr *)&rp->c_addr) || | |
338 | rpc_get_port(svc_addr(rqstp)) != rpc_get_port((struct sockaddr *)&rp->c_addr)) | |
339 | return false; | |
340 | ||
9dc56143 JL |
341 | return true; |
342 | } | |
343 | ||
a4a3ec32 JL |
344 | /* |
345 | * Search the request hash for an entry that matches the given rqstp. | |
346 | * Must be called with cache_lock held. Returns the found entry or | |
347 | * NULL on failure. | |
348 | */ | |
349 | static struct svc_cacherep * | |
7142b98d TM |
350 | nfsd_cache_search(struct nfsd_drc_bucket *b, struct svc_rqst *rqstp, |
351 | __wsum csum) | |
a4a3ec32 | 352 | { |
98d821bd | 353 | struct svc_cacherep *rp, *ret = NULL; |
11acf6ef | 354 | struct list_head *rh = &b->lru_head; |
98d821bd | 355 | unsigned int entries = 0; |
a4a3ec32 | 356 | |
11acf6ef | 357 | list_for_each_entry(rp, rh, c_lru) { |
98d821bd JL |
358 | ++entries; |
359 | if (nfsd_cache_match(rqstp, csum, rp)) { | |
360 | ret = rp; | |
361 | break; | |
362 | } | |
363 | } | |
364 | ||
365 | /* tally hash chain length stats */ | |
366 | if (entries > longest_chain) { | |
367 | longest_chain = entries; | |
31e60f52 | 368 | longest_chain_cachesize = atomic_read(&num_drc_entries); |
98d821bd JL |
369 | } else if (entries == longest_chain) { |
370 | /* prefer to keep the smallest cachesize possible here */ | |
31e60f52 TM |
371 | longest_chain_cachesize = min_t(unsigned int, |
372 | longest_chain_cachesize, | |
373 | atomic_read(&num_drc_entries)); | |
a4a3ec32 | 374 | } |
98d821bd JL |
375 | |
376 | return ret; | |
a4a3ec32 JL |
377 | } |
378 | ||
1da177e4 LT |
379 | /* |
380 | * Try to find an entry matching the current call in the cache. When none | |
1ac83629 JL |
381 | * is found, we try to grab the oldest expired entry off the LRU list. If |
382 | * a suitable one isn't there, then drop the cache_lock and allocate a | |
383 | * new one, then search again in case one got inserted while this thread | |
384 | * didn't hold the lock. | |
1da177e4 LT |
385 | */ |
386 | int | |
1091006c | 387 | nfsd_cache_lookup(struct svc_rqst *rqstp) |
1da177e4 | 388 | { |
0338dd15 | 389 | struct svc_cacherep *rp, *found; |
c7afef1f AV |
390 | __be32 xid = rqstp->rq_xid; |
391 | u32 proto = rqstp->rq_prot, | |
1da177e4 LT |
392 | vers = rqstp->rq_vers, |
393 | proc = rqstp->rq_proc; | |
01a7decf | 394 | __wsum csum; |
7142b98d TM |
395 | u32 hash = nfsd_cache_hash(xid); |
396 | struct nfsd_drc_bucket *b = &drc_hashtbl[hash]; | |
1da177e4 | 397 | unsigned long age; |
1091006c | 398 | int type = rqstp->rq_cachetype; |
0b9ea37f | 399 | int rtn = RC_DOIT; |
1da177e4 LT |
400 | |
401 | rqstp->rq_cacherep = NULL; | |
13cc8a78 | 402 | if (type == RC_NOCACHE) { |
1da177e4 | 403 | nfsdstats.rcnocache++; |
0b9ea37f | 404 | return rtn; |
1da177e4 LT |
405 | } |
406 | ||
01a7decf JL |
407 | csum = nfsd_cache_csum(rqstp); |
408 | ||
0b9ea37f JL |
409 | /* |
410 | * Since the common case is a cache miss followed by an insert, | |
a0ef5e19 | 411 | * preallocate an entry. |
0b9ea37f | 412 | */ |
0338dd15 | 413 | rp = nfsd_reply_cache_alloc(); |
89a26b3d | 414 | spin_lock(&b->cache_lock); |
6c6910cd | 415 | if (likely(rp)) { |
31e60f52 | 416 | atomic_inc(&num_drc_entries); |
6c6910cd JL |
417 | drc_mem_usage += sizeof(*rp); |
418 | } | |
0338dd15 | 419 | |
a0ef5e19 | 420 | /* go ahead and prune the cache */ |
89a26b3d | 421 | prune_bucket(b); |
a0ef5e19 | 422 | |
7142b98d | 423 | found = nfsd_cache_search(b, rqstp, csum); |
0338dd15 | 424 | if (found) { |
0b9ea37f JL |
425 | if (likely(rp)) |
426 | nfsd_reply_cache_free_locked(rp); | |
0338dd15 JL |
427 | rp = found; |
428 | goto found_entry; | |
1da177e4 LT |
429 | } |
430 | ||
0b9ea37f JL |
431 | if (!rp) { |
432 | dprintk("nfsd: unable to allocate DRC entry!\n"); | |
433 | goto out; | |
434 | } | |
435 | ||
0338dd15 | 436 | nfsdstats.rcmisses++; |
1da177e4 LT |
437 | rqstp->rq_cacherep = rp; |
438 | rp->c_state = RC_INPROG; | |
439 | rp->c_xid = xid; | |
440 | rp->c_proc = proc; | |
7b9e8522 JL |
441 | rpc_copy_addr((struct sockaddr *)&rp->c_addr, svc_addr(rqstp)); |
442 | rpc_set_port((struct sockaddr *)&rp->c_addr, rpc_get_port(svc_addr(rqstp))); | |
1da177e4 LT |
443 | rp->c_prot = proto; |
444 | rp->c_vers = vers; | |
01a7decf JL |
445 | rp->c_len = rqstp->rq_arg.len; |
446 | rp->c_csum = csum; | |
1da177e4 | 447 | |
bedd4b61 | 448 | lru_put_end(b, rp); |
1da177e4 LT |
449 | |
450 | /* release any buffer */ | |
451 | if (rp->c_type == RC_REPLBUFF) { | |
6c6910cd | 452 | drc_mem_usage -= rp->c_replvec.iov_len; |
1da177e4 LT |
453 | kfree(rp->c_replvec.iov_base); |
454 | rp->c_replvec.iov_base = NULL; | |
455 | } | |
456 | rp->c_type = RC_NOCACHE; | |
457 | out: | |
89a26b3d | 458 | spin_unlock(&b->cache_lock); |
1da177e4 LT |
459 | return rtn; |
460 | ||
461 | found_entry: | |
0338dd15 | 462 | nfsdstats.rchits++; |
1da177e4 LT |
463 | /* We found a matching entry which is either in progress or done. */ |
464 | age = jiffies - rp->c_timestamp; | |
bedd4b61 | 465 | lru_put_end(b, rp); |
1da177e4 LT |
466 | |
467 | rtn = RC_DROPIT; | |
468 | /* Request being processed or excessive rexmits */ | |
469 | if (rp->c_state == RC_INPROG || age < RC_DELAY) | |
470 | goto out; | |
471 | ||
472 | /* From the hall of fame of impractical attacks: | |
473 | * Is this a user who tries to snoop on the cache? */ | |
474 | rtn = RC_DOIT; | |
4d152e2c | 475 | if (!test_bit(RQ_SECURE, &rqstp->rq_flags) && rp->c_secure) |
1da177e4 LT |
476 | goto out; |
477 | ||
478 | /* Compose RPC reply header */ | |
479 | switch (rp->c_type) { | |
480 | case RC_NOCACHE: | |
481 | break; | |
482 | case RC_REPLSTAT: | |
483 | svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat); | |
484 | rtn = RC_REPLY; | |
485 | break; | |
486 | case RC_REPLBUFF: | |
487 | if (!nfsd_cache_append(rqstp, &rp->c_replvec)) | |
488 | goto out; /* should not happen */ | |
489 | rtn = RC_REPLY; | |
490 | break; | |
491 | default: | |
492 | printk(KERN_WARNING "nfsd: bad repcache type %d\n", rp->c_type); | |
0338dd15 | 493 | nfsd_reply_cache_free_locked(rp); |
1da177e4 LT |
494 | } |
495 | ||
496 | goto out; | |
497 | } | |
498 | ||
499 | /* | |
500 | * Update a cache entry. This is called from nfsd_dispatch when | |
501 | * the procedure has been executed and the complete reply is in | |
502 | * rqstp->rq_res. | |
503 | * | |
504 | * We're copying around data here rather than swapping buffers because | |
505 | * the toplevel loop requires max-sized buffers, which would be a waste | |
506 | * of memory for a cache with a max reply size of 100 bytes (diropokres). | |
507 | * | |
508 | * If we should start to use different types of cache entries tailored | |
509 | * specifically for attrstat and fh's, we may save even more space. | |
510 | * | |
511 | * Also note that a cachetype of RC_NOCACHE can legally be passed when | |
512 | * nfsd failed to encode a reply that otherwise would have been cached. | |
513 | * In this case, nfsd_cache_update is called with statp == NULL. | |
514 | */ | |
515 | void | |
c7afef1f | 516 | nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp) |
1da177e4 | 517 | { |
13cc8a78 | 518 | struct svc_cacherep *rp = rqstp->rq_cacherep; |
1da177e4 | 519 | struct kvec *resv = &rqstp->rq_res.head[0], *cachv; |
bedd4b61 TM |
520 | u32 hash; |
521 | struct nfsd_drc_bucket *b; | |
1da177e4 | 522 | int len; |
6c6910cd | 523 | size_t bufsize = 0; |
1da177e4 | 524 | |
13cc8a78 | 525 | if (!rp) |
1da177e4 LT |
526 | return; |
527 | ||
bedd4b61 TM |
528 | hash = nfsd_cache_hash(rp->c_xid); |
529 | b = &drc_hashtbl[hash]; | |
530 | ||
1da177e4 LT |
531 | len = resv->iov_len - ((char*)statp - (char*)resv->iov_base); |
532 | len >>= 2; | |
fca4217c | 533 | |
1da177e4 LT |
534 | /* Don't cache excessive amounts of data and XDR failures */ |
535 | if (!statp || len > (256 >> 2)) { | |
89a26b3d | 536 | nfsd_reply_cache_free(b, rp); |
1da177e4 LT |
537 | return; |
538 | } | |
539 | ||
540 | switch (cachetype) { | |
541 | case RC_REPLSTAT: | |
542 | if (len != 1) | |
543 | printk("nfsd: RC_REPLSTAT/reply len %d!\n",len); | |
544 | rp->c_replstat = *statp; | |
545 | break; | |
546 | case RC_REPLBUFF: | |
547 | cachv = &rp->c_replvec; | |
6c6910cd JL |
548 | bufsize = len << 2; |
549 | cachv->iov_base = kmalloc(bufsize, GFP_KERNEL); | |
1da177e4 | 550 | if (!cachv->iov_base) { |
89a26b3d | 551 | nfsd_reply_cache_free(b, rp); |
1da177e4 LT |
552 | return; |
553 | } | |
6c6910cd JL |
554 | cachv->iov_len = bufsize; |
555 | memcpy(cachv->iov_base, statp, bufsize); | |
1da177e4 | 556 | break; |
2c6b691c | 557 | case RC_NOCACHE: |
89a26b3d | 558 | nfsd_reply_cache_free(b, rp); |
2c6b691c | 559 | return; |
1da177e4 | 560 | } |
89a26b3d | 561 | spin_lock(&b->cache_lock); |
6c6910cd | 562 | drc_mem_usage += bufsize; |
bedd4b61 | 563 | lru_put_end(b, rp); |
4d152e2c | 564 | rp->c_secure = test_bit(RQ_SECURE, &rqstp->rq_flags); |
1da177e4 LT |
565 | rp->c_type = cachetype; |
566 | rp->c_state = RC_DONE; | |
89a26b3d | 567 | spin_unlock(&b->cache_lock); |
1da177e4 LT |
568 | return; |
569 | } | |
570 | ||
571 | /* | |
572 | * Copy cached reply to current reply buffer. Should always fit. | |
573 | * FIXME as reply is in a page, we should just attach the page, and | |
574 | * keep a refcount.... | |
575 | */ | |
576 | static int | |
577 | nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data) | |
578 | { | |
579 | struct kvec *vec = &rqstp->rq_res.head[0]; | |
580 | ||
581 | if (vec->iov_len + data->iov_len > PAGE_SIZE) { | |
5b5e0928 | 582 | printk(KERN_WARNING "nfsd: cached reply too large (%zd).\n", |
1da177e4 LT |
583 | data->iov_len); |
584 | return 0; | |
585 | } | |
586 | memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len); | |
587 | vec->iov_len += data->iov_len; | |
588 | return 1; | |
589 | } | |
a2f999a3 JL |
590 | |
591 | /* | |
592 | * Note that fields may be added, removed or reordered in the future. Programs | |
593 | * scraping this file for info should test the labels to ensure they're | |
594 | * getting the correct field. | |
595 | */ | |
596 | static int nfsd_reply_cache_stats_show(struct seq_file *m, void *v) | |
597 | { | |
a2f999a3 | 598 | seq_printf(m, "max entries: %u\n", max_drc_entries); |
31e60f52 TM |
599 | seq_printf(m, "num entries: %u\n", |
600 | atomic_read(&num_drc_entries)); | |
0733c7ba | 601 | seq_printf(m, "hash buckets: %u\n", 1 << maskbits); |
a2f999a3 JL |
602 | seq_printf(m, "mem usage: %u\n", drc_mem_usage); |
603 | seq_printf(m, "cache hits: %u\n", nfsdstats.rchits); | |
604 | seq_printf(m, "cache misses: %u\n", nfsdstats.rcmisses); | |
605 | seq_printf(m, "not cached: %u\n", nfsdstats.rcnocache); | |
606 | seq_printf(m, "payload misses: %u\n", payload_misses); | |
98d821bd JL |
607 | seq_printf(m, "longest chain len: %u\n", longest_chain); |
608 | seq_printf(m, "cachesize at longest: %u\n", longest_chain_cachesize); | |
a2f999a3 JL |
609 | return 0; |
610 | } | |
611 | ||
612 | int nfsd_reply_cache_stats_open(struct inode *inode, struct file *file) | |
613 | { | |
614 | return single_open(file, nfsd_reply_cache_stats_show, NULL); | |
615 | } |