fs/nfsd/nfscache.c

   1 /*
   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
  11 #include <linux/slab.h>
  12 #include <linux/sunrpc/addr.h>
  13 #include <linux/highmem.h>
  14 #include <linux/log2.h>
  15 #include <linux/hash.h>
  16 #include <net/checksum.h>
  17
  18 #include "nfsd.h"
  19 #include "cache.h"
  20
  21 #define NFSDDBG_FACILITY        NFSDDBG_REPCACHE
  22
  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
  29
  30 struct nfsd_drc_bucket {
  31         struct list_head lru_head;
  32         spinlock_t cache_lock;
  33 };
  34
  35 static struct nfsd_drc_bucket   *drc_hashtbl;
  36 static struct kmem_cache        *drc_slab;
  37
  38 /* max number of entries allowed in the cache */
  39 static unsigned int             max_drc_entries;
  40
  41 /* number of significant bits in the hash value */
  42 static unsigned int             maskbits;
  43 static unsigned int             drc_hashsize;
  44
  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 */
  51 static atomic_t                 num_drc_entries;
  52
  53 /* cache misses due only to checksum comparison failures */
  54 static unsigned int             payload_misses;
  55
  56 /* amount of memory (in bytes) currently consumed by the DRC */
  57 static unsigned int             drc_mem_usage;
  58
  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
  65 static int      nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec);
  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);
  70
  71 static struct shrinker nfsd_reply_cache_shrinker = {
  72         .scan_objects = nfsd_reply_cache_scan,
  73         .count_objects = nfsd_reply_cache_count,
  74         .seeks  = 1,
  75 };
  76
  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
 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
 115 static u32
 116 nfsd_cache_hash(__be32 xid)
 117 {
 118         return hash_32(be32_to_cpu(xid), maskbits);
 119 }
 120
 121 static struct svc_cacherep *
 122 nfsd_reply_cache_alloc(void)
 123 {
 124         struct svc_cacherep     *rp;
 125
 126         rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
 127         if (rp) {
 128                 rp->c_state = RC_UNUSED;
 129                 rp->c_type = RC_NOCACHE;
 130                 INIT_LIST_HEAD(&rp->c_lru);
 131         }
 132         return rp;
 133 }
 134
 135 static void
 136 nfsd_reply_cache_free_locked(struct svc_cacherep *rp)
 137 {
 138         if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) {
 139                 drc_mem_usage -= rp->c_replvec.iov_len;
 140                 kfree(rp->c_replvec.iov_base);
 141         }
 142         list_del(&rp->c_lru);
 143         atomic_dec(&num_drc_entries);
 144         drc_mem_usage -= sizeof(*rp);
 145         kmem_cache_free(drc_slab, rp);
 146 }
 147
 148 static void
 149 nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct svc_cacherep *rp)
 150 {
 151         spin_lock(&b->cache_lock);
 152         nfsd_reply_cache_free_locked(rp);
 153         spin_unlock(&b->cache_lock);
 154 }
 155
 156 int nfsd_reply_cache_init(void)
 157 {
 158         unsigned int hashsize;
 159         unsigned int i;
 160         int status = 0;
 161
 162         max_drc_entries = nfsd_cache_size_limit();
 163         atomic_set(&num_drc_entries, 0);
 164         hashsize = nfsd_hashsize(max_drc_entries);
 165         maskbits = ilog2(hashsize);
 166
 167         status = register_shrinker(&nfsd_reply_cache_shrinker);
 168         if (status)
 169                 return status;
 170
 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
 176         drc_hashtbl = kcalloc(hashsize, sizeof(*drc_hashtbl), GFP_KERNEL);
 177         if (!drc_hashtbl)
 178                 goto out_nomem;
 179         for (i = 0; i < hashsize; i++) {
 180                 INIT_LIST_HEAD(&drc_hashtbl[i].lru_head);
 181                 spin_lock_init(&drc_hashtbl[i].cache_lock);
 182         }
 183         drc_hashsize = hashsize;
 184
 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;
 190 }
 191
 192 void nfsd_reply_cache_shutdown(void)
 193 {
 194         struct svc_cacherep     *rp;
 195         unsigned int i;
 196
 197         unregister_shrinker(&nfsd_reply_cache_shrinker);
 198
 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                 }
 205         }
 206
 207         kfree (drc_hashtbl);
 208         drc_hashtbl = NULL;
 209         drc_hashsize = 0;
 210
 211         kmem_cache_destroy(drc_slab);
 212         drc_slab = NULL;
 213 }
 214
 215 /*
 216  * Move cache entry to end of LRU list, and queue the cleaner to run if it's
 217  * not already scheduled.
 218  */
 219 static void
 220 lru_put_end(struct nfsd_drc_bucket *b, struct svc_cacherep *rp)
 221 {
 222         rp->c_timestamp = jiffies;
 223         list_move_tail(&rp->c_lru, &b->lru_head);
 224 }
 225
 226 static long
 227 prune_bucket(struct nfsd_drc_bucket *b)
 228 {
 229         struct svc_cacherep *rp, *tmp;
 230         long freed = 0;
 231
 232         list_for_each_entry_safe(rp, tmp, &b->lru_head, c_lru) {
 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;
 239                 if (atomic_read(&num_drc_entries) <= max_drc_entries &&
 240                     time_before(jiffies, rp->c_timestamp + RC_EXPIRE))
 241                         break;
 242                 nfsd_reply_cache_free_locked(rp);
 243                 freed++;
 244         }
 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;
 257
 258         for (i = 0; i < drc_hashsize; i++) {
 259                 struct nfsd_drc_bucket *b = &drc_hashtbl[i];
 260
 261                 if (list_empty(&b->lru_head))
 262                         continue;
 263                 spin_lock(&b->cache_lock);
 264                 freed += prune_bucket(b);
 265                 spin_unlock(&b->cache_lock);
 266         }
 267         return freed;
 268 }
 269
 270 static unsigned long
 271 nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc)
 272 {
 273         return atomic_read(&num_drc_entries);
 274 }
 275
 276 static unsigned long
 277 nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
 278 {
 279         return prune_cache_entries();
 280 }
 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;
 305                 len = min_t(size_t, PAGE_SIZE - base, csum_len);
 306                 csum = csum_partial(p, len, csum);
 307                 csum_len -= len;
 308                 base = 0;
 309                 ++idx;
 310         }
 311         return csum;
 312 }
 313
 314 static bool
 315 nfsd_cache_match(struct svc_rqst *rqstp, __wsum csum, struct svc_cacherep *rp)
 316 {
 317         /* Check RPC XID first */
 318         if (rqstp->rq_xid != rp->c_xid)
 319                 return false;
 320         /* compare checksum of NFS data */
 321         if (csum != rp->c_csum) {
 322                 ++payload_misses;
 323                 return false;
 324         }
 325
 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
 335         return true;
 336 }
 337
 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 *
 344 nfsd_cache_search(struct nfsd_drc_bucket *b, struct svc_rqst *rqstp,
 345                 __wsum csum)
 346 {
 347         struct svc_cacherep     *rp, *ret = NULL;
 348         struct list_head        *rh = &b->lru_head;
 349         unsigned int            entries = 0;
 350
 351         list_for_each_entry(rp, rh, c_lru) {
 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;
 362                 longest_chain_cachesize = atomic_read(&num_drc_entries);
 363         } else if (entries == longest_chain) {
 364                 /* prefer to keep the smallest cachesize possible here */
 365                 longest_chain_cachesize = min_t(unsigned int,
 366                                 longest_chain_cachesize,
 367                                 atomic_read(&num_drc_entries));
 368         }
 369
 370         return ret;
 371 }
 372
 373 /*
 374  * Try to find an entry matching the current call in the cache. When none
 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.
 379  */
 380 int
 381 nfsd_cache_lookup(struct svc_rqst *rqstp)
 382 {
 383         struct svc_cacherep     *rp, *found;
 384         __be32                  xid = rqstp->rq_xid;
 385         u32                     proto =  rqstp->rq_prot,
 386                                 vers = rqstp->rq_vers,
 387                                 proc = rqstp->rq_proc;
 388         __wsum                  csum;
 389         u32 hash = nfsd_cache_hash(xid);
 390         struct nfsd_drc_bucket *b = &drc_hashtbl[hash];
 391         unsigned long           age;
 392         int type = rqstp->rq_cachetype;
 393         int rtn = RC_DOIT;
 394
 395         rqstp->rq_cacherep = NULL;
 396         if (type == RC_NOCACHE) {
 397                 nfsdstats.rcnocache++;
 398                 return rtn;
 399         }
 400
 401         csum = nfsd_cache_csum(rqstp);
 402
 403         /*
 404          * Since the common case is a cache miss followed by an insert,
 405          * preallocate an entry.
 406          */
 407         rp = nfsd_reply_cache_alloc();
 408         spin_lock(&b->cache_lock);
 409         if (likely(rp)) {
 410                 atomic_inc(&num_drc_entries);
 411                 drc_mem_usage += sizeof(*rp);
 412         }
 413
 414         /* go ahead and prune the cache */
 415         prune_bucket(b);
 416
 417         found = nfsd_cache_search(b, rqstp, csum);
 418         if (found) {
 419                 if (likely(rp))
 420                         nfsd_reply_cache_free_locked(rp);
 421                 rp = found;
 422                 goto found_entry;
 423         }
 424
 425         if (!rp) {
 426                 dprintk("nfsd: unable to allocate DRC entry!\n");
 427                 goto out;
 428         }
 429
 430         nfsdstats.rcmisses++;
 431         rqstp->rq_cacherep = rp;
 432         rp->c_state = RC_INPROG;
 433         rp->c_xid = xid;
 434         rp->c_proc = proc;
 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)));
 437         rp->c_prot = proto;
 438         rp->c_vers = vers;
 439         rp->c_len = rqstp->rq_arg.len;
 440         rp->c_csum = csum;
 441
 442         lru_put_end(b, rp);
 443
 444         /* release any buffer */
 445         if (rp->c_type == RC_REPLBUFF) {
 446                 drc_mem_usage -= rp->c_replvec.iov_len;
 447                 kfree(rp->c_replvec.iov_base);
 448                 rp->c_replvec.iov_base = NULL;
 449         }
 450         rp->c_type = RC_NOCACHE;
 451  out:
 452         spin_unlock(&b->cache_lock);
 453         return rtn;
 454
 455 found_entry:
 456         nfsdstats.rchits++;
 457         /* We found a matching entry which is either in progress or done. */
 458         age = jiffies - rp->c_timestamp;
 459         lru_put_end(b, rp);
 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;
 469         if (!test_bit(RQ_SECURE, &rqstp->rq_flags) && rp->c_secure)
 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);
 487                 nfsd_reply_cache_free_locked(rp);
 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
 510 nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
 511 {
 512         struct svc_cacherep *rp = rqstp->rq_cacherep;
 513         struct kvec     *resv = &rqstp->rq_res.head[0], *cachv;
 514         u32             hash;
 515         struct nfsd_drc_bucket *b;
 516         int             len;
 517         size_t          bufsize = 0;
 518
 519         if (!rp)
 520                 return;
 521
 522         hash = nfsd_cache_hash(rp->c_xid);
 523         b = &drc_hashtbl[hash];
 524
 525         len = resv->iov_len - ((char*)statp - (char*)resv->iov_base);
 526         len >>= 2;
 527
 528         /* Don't cache excessive amounts of data and XDR failures */
 529         if (!statp || len > (256 >> 2)) {
 530                 nfsd_reply_cache_free(b, rp);
 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;
 542                 bufsize = len << 2;
 543                 cachv->iov_base = kmalloc(bufsize, GFP_KERNEL);
 544                 if (!cachv->iov_base) {
 545                         nfsd_reply_cache_free(b, rp);
 546                         return;
 547                 }
 548                 cachv->iov_len = bufsize;
 549                 memcpy(cachv->iov_base, statp, bufsize);
 550                 break;
 551         case RC_NOCACHE:
 552                 nfsd_reply_cache_free(b, rp);
 553                 return;
 554         }
 555         spin_lock(&b->cache_lock);
 556         drc_mem_usage += bufsize;
 557         lru_put_end(b, rp);
 558         rp->c_secure = test_bit(RQ_SECURE, &rqstp->rq_flags);
 559         rp->c_type = cachetype;
 560         rp->c_state = RC_DONE;
 561         spin_unlock(&b->cache_lock);
 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 }
 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 {
 592         seq_printf(m, "max entries:           %u\n", max_drc_entries);
 593         seq_printf(m, "num entries:           %u\n",
 594                         atomic_read(&num_drc_entries));
 595         seq_printf(m, "hash buckets:          %u\n", 1 << maskbits);
 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);
 601         seq_printf(m, "longest chain len:     %u\n", longest_chain);
 602         seq_printf(m, "cachesize at longest:  %u\n", longest_chain_cachesize);
 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 }