linux/fs/nfsd/nfscache.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   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
  12#include <linux/sunrpc/svc_xprt.h>
  13#include <linux/slab.h>
  14#include <linux/vmalloc.h>
  15#include <linux/sunrpc/addr.h>
  16#include <linux/highmem.h>
  17#include <linux/log2.h>
  18#include <linux/hash.h>
  19#include <net/checksum.h>
  20
  21#include "nfsd.h"
  22#include "cache.h"
  23#include "trace.h"
  24
  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
  31
  32struct nfsd_drc_bucket {
  33        struct rb_root rb_head;
  34        struct list_head lru_head;
  35        spinlock_t cache_lock;
  36};
  37
  38static struct kmem_cache        *drc_slab;
  39
  40static int      nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *vec);
  41static unsigned long nfsd_reply_cache_count(struct shrinker *shrink,
  42                                            struct shrink_control *sc);
  43static unsigned long nfsd_reply_cache_scan(struct shrinker *shrink,
  44                                           struct shrink_control *sc);
  45
  46/*
  47 * Put a cap on the size of the DRC based on the amount of available
  48 * low memory in the machine.
  49 *
  50 *  64MB:    8192
  51 * 128MB:   11585
  52 * 256MB:   16384
  53 * 512MB:   23170
  54 *   1GB:   32768
  55 *   2GB:   46340
  56 *   4GB:   65536
  57 *   8GB:   92681
  58 *  16GB:  131072
  59 *
  60 * ...with a hard cap of 256k entries. In the worst case, each entry will be
  61 * ~1k, so the above numbers should give a rough max of the amount of memory
  62 * used in k.
  63 *
  64 * XXX: these limits are per-container, so memory used will increase
  65 * linearly with number of containers.  Maybe that's OK.
  66 */
  67static unsigned int
  68nfsd_cache_size_limit(void)
  69{
  70        unsigned int limit;
  71        unsigned long low_pages = totalram_pages() - totalhigh_pages();
  72
  73        limit = (16 * int_sqrt(low_pages)) << (PAGE_SHIFT-10);
  74        return min_t(unsigned int, limit, 256*1024);
  75}
  76
  77/*
  78 * Compute the number of hash buckets we need. Divide the max cachesize by
  79 * the "target" max bucket size, and round up to next power of two.
  80 */
  81static unsigned int
  82nfsd_hashsize(unsigned int limit)
  83{
  84        return roundup_pow_of_two(limit / TARGET_BUCKET_SIZE);
  85}
  86
  87static u32
  88nfsd_cache_hash(__be32 xid, struct nfsd_net *nn)
  89{
  90        return hash_32(be32_to_cpu(xid), nn->maskbits);
  91}
  92
  93static struct svc_cacherep *
  94nfsd_reply_cache_alloc(struct svc_rqst *rqstp, __wsum csum,
  95                        struct nfsd_net *nn)
  96{
  97        struct svc_cacherep     *rp;
  98
  99        rp = kmem_cache_alloc(drc_slab, GFP_KERNEL);
 100        if (rp) {
 101                rp->c_state = RC_UNUSED;
 102                rp->c_type = RC_NOCACHE;
 103                RB_CLEAR_NODE(&rp->c_node);
 104                INIT_LIST_HEAD(&rp->c_lru);
 105
 106                memset(&rp->c_key, 0, sizeof(rp->c_key));
 107                rp->c_key.k_xid = rqstp->rq_xid;
 108                rp->c_key.k_proc = rqstp->rq_proc;
 109                rpc_copy_addr((struct sockaddr *)&rp->c_key.k_addr, svc_addr(rqstp));
 110                rpc_set_port((struct sockaddr *)&rp->c_key.k_addr, rpc_get_port(svc_addr(rqstp)));
 111                rp->c_key.k_prot = rqstp->rq_prot;
 112                rp->c_key.k_vers = rqstp->rq_vers;
 113                rp->c_key.k_len = rqstp->rq_arg.len;
 114                rp->c_key.k_csum = csum;
 115        }
 116        return rp;
 117}
 118
 119static void
 120nfsd_reply_cache_free_locked(struct nfsd_drc_bucket *b, struct svc_cacherep *rp,
 121                                struct nfsd_net *nn)
 122{
 123        if (rp->c_type == RC_REPLBUFF && rp->c_replvec.iov_base) {
 124                nn->drc_mem_usage -= rp->c_replvec.iov_len;
 125                kfree(rp->c_replvec.iov_base);
 126        }
 127        if (rp->c_state != RC_UNUSED) {
 128                rb_erase(&rp->c_node, &b->rb_head);
 129                list_del(&rp->c_lru);
 130                atomic_dec(&nn->num_drc_entries);
 131                nn->drc_mem_usage -= sizeof(*rp);
 132        }
 133        kmem_cache_free(drc_slab, rp);
 134}
 135
 136static void
 137nfsd_reply_cache_free(struct nfsd_drc_bucket *b, struct svc_cacherep *rp,
 138                        struct nfsd_net *nn)
 139{
 140        spin_lock(&b->cache_lock);
 141        nfsd_reply_cache_free_locked(b, rp, nn);
 142        spin_unlock(&b->cache_lock);
 143}
 144
 145int nfsd_drc_slab_create(void)
 146{
 147        drc_slab = kmem_cache_create("nfsd_drc",
 148                                sizeof(struct svc_cacherep), 0, 0, NULL);
 149        return drc_slab ? 0: -ENOMEM;
 150}
 151
 152void nfsd_drc_slab_free(void)
 153{
 154        kmem_cache_destroy(drc_slab);
 155}
 156
 157int nfsd_reply_cache_init(struct nfsd_net *nn)
 158{
 159        unsigned int hashsize;
 160        unsigned int i;
 161        int status = 0;
 162
 163        nn->max_drc_entries = nfsd_cache_size_limit();
 164        atomic_set(&nn->num_drc_entries, 0);
 165        hashsize = nfsd_hashsize(nn->max_drc_entries);
 166        nn->maskbits = ilog2(hashsize);
 167
 168        nn->nfsd_reply_cache_shrinker.scan_objects = nfsd_reply_cache_scan;
 169        nn->nfsd_reply_cache_shrinker.count_objects = nfsd_reply_cache_count;
 170        nn->nfsd_reply_cache_shrinker.seeks = 1;
 171        status = register_shrinker(&nn->nfsd_reply_cache_shrinker);
 172        if (status)
 173                goto out_nomem;
 174
 175        nn->drc_hashtbl = kvzalloc(array_size(hashsize,
 176                                sizeof(*nn->drc_hashtbl)), GFP_KERNEL);
 177        if (!nn->drc_hashtbl)
 178                goto out_shrinker;
 179
 180        for (i = 0; i < hashsize; i++) {
 181                INIT_LIST_HEAD(&nn->drc_hashtbl[i].lru_head);
 182                spin_lock_init(&nn->drc_hashtbl[i].cache_lock);
 183        }
 184        nn->drc_hashsize = hashsize;
 185
 186        return 0;
 187out_shrinker:
 188        unregister_shrinker(&nn->nfsd_reply_cache_shrinker);
 189out_nomem:
 190        printk(KERN_ERR "nfsd: failed to allocate reply cache\n");
 191        return -ENOMEM;
 192}
 193
 194void nfsd_reply_cache_shutdown(struct nfsd_net *nn)
 195{
 196        struct svc_cacherep     *rp;
 197        unsigned int i;
 198
 199        unregister_shrinker(&nn->nfsd_reply_cache_shrinker);
 200
 201        for (i = 0; i < nn->drc_hashsize; i++) {
 202                struct list_head *head = &nn->drc_hashtbl[i].lru_head;
 203                while (!list_empty(head)) {
 204                        rp = list_first_entry(head, struct svc_cacherep, c_lru);
 205                        nfsd_reply_cache_free_locked(&nn->drc_hashtbl[i],
 206                                                                        rp, nn);
 207                }
 208        }
 209
 210        kvfree(nn->drc_hashtbl);
 211        nn->drc_hashtbl = NULL;
 212        nn->drc_hashsize = 0;
 213
 214}
 215
 216/*
 217 * Move cache entry to end of LRU list, and queue the cleaner to run if it's
 218 * not already scheduled.
 219 */
 220static void
 221lru_put_end(struct nfsd_drc_bucket *b, struct svc_cacherep *rp)
 222{
 223        rp->c_timestamp = jiffies;
 224        list_move_tail(&rp->c_lru, &b->lru_head);
 225}
 226
 227static long
 228prune_bucket(struct nfsd_drc_bucket *b, struct nfsd_net *nn)
 229{
 230        struct svc_cacherep *rp, *tmp;
 231        long freed = 0;
 232
 233        list_for_each_entry_safe(rp, tmp, &b->lru_head, c_lru) {
 234                /*
 235                 * Don't free entries attached to calls that are still
 236                 * in-progress, but do keep scanning the list.
 237                 */
 238                if (rp->c_state == RC_INPROG)
 239                        continue;
 240                if (atomic_read(&nn->num_drc_entries) <= nn->max_drc_entries &&
 241                    time_before(jiffies, rp->c_timestamp + RC_EXPIRE))
 242                        break;
 243                nfsd_reply_cache_free_locked(b, rp, nn);
 244                freed++;
 245        }
 246        return freed;
 247}
 248
 249/*
 250 * Walk the LRU list and prune off entries that are older than RC_EXPIRE.
 251 * Also prune the oldest ones when the total exceeds the max number of entries.
 252 */
 253static long
 254prune_cache_entries(struct nfsd_net *nn)
 255{
 256        unsigned int i;
 257        long freed = 0;
 258
 259        for (i = 0; i < nn->drc_hashsize; i++) {
 260                struct nfsd_drc_bucket *b = &nn->drc_hashtbl[i];
 261
 262                if (list_empty(&b->lru_head))
 263                        continue;
 264                spin_lock(&b->cache_lock);
 265                freed += prune_bucket(b, nn);
 266                spin_unlock(&b->cache_lock);
 267        }
 268        return freed;
 269}
 270
 271static unsigned long
 272nfsd_reply_cache_count(struct shrinker *shrink, struct shrink_control *sc)
 273{
 274        struct nfsd_net *nn = container_of(shrink,
 275                                struct nfsd_net, nfsd_reply_cache_shrinker);
 276
 277        return atomic_read(&nn->num_drc_entries);
 278}
 279
 280static unsigned long
 281nfsd_reply_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
 282{
 283        struct nfsd_net *nn = container_of(shrink,
 284                                struct nfsd_net, nfsd_reply_cache_shrinker);
 285
 286        return prune_cache_entries(nn);
 287}
 288/*
 289 * Walk an xdr_buf and get a CRC for at most the first RC_CSUMLEN bytes
 290 */
 291static __wsum
 292nfsd_cache_csum(struct svc_rqst *rqstp)
 293{
 294        int idx;
 295        unsigned int base;
 296        __wsum csum;
 297        struct xdr_buf *buf = &rqstp->rq_arg;
 298        const unsigned char *p = buf->head[0].iov_base;
 299        size_t csum_len = min_t(size_t, buf->head[0].iov_len + buf->page_len,
 300                                RC_CSUMLEN);
 301        size_t len = min(buf->head[0].iov_len, csum_len);
 302
 303        /* rq_arg.head first */
 304        csum = csum_partial(p, len, 0);
 305        csum_len -= len;
 306
 307        /* Continue into page array */
 308        idx = buf->page_base / PAGE_SIZE;
 309        base = buf->page_base & ~PAGE_MASK;
 310        while (csum_len) {
 311                p = page_address(buf->pages[idx]) + base;
 312                len = min_t(size_t, PAGE_SIZE - base, csum_len);
 313                csum = csum_partial(p, len, csum);
 314                csum_len -= len;
 315                base = 0;
 316                ++idx;
 317        }
 318        return csum;
 319}
 320
 321static int
 322nfsd_cache_key_cmp(const struct svc_cacherep *key,
 323                        const struct svc_cacherep *rp, struct nfsd_net *nn)
 324{
 325        if (key->c_key.k_xid == rp->c_key.k_xid &&
 326            key->c_key.k_csum != rp->c_key.k_csum) {
 327                ++nn->payload_misses;
 328                trace_nfsd_drc_mismatch(nn, key, rp);
 329        }
 330
 331        return memcmp(&key->c_key, &rp->c_key, sizeof(key->c_key));
 332}
 333
 334/*
 335 * Search the request hash for an entry that matches the given rqstp.
 336 * Must be called with cache_lock held. Returns the found entry or
 337 * inserts an empty key on failure.
 338 */
 339static struct svc_cacherep *
 340nfsd_cache_insert(struct nfsd_drc_bucket *b, struct svc_cacherep *key,
 341                        struct nfsd_net *nn)
 342{
 343        struct svc_cacherep     *rp, *ret = key;
 344        struct rb_node          **p = &b->rb_head.rb_node,
 345                                *parent = NULL;
 346        unsigned int            entries = 0;
 347        int cmp;
 348
 349        while (*p != NULL) {
 350                ++entries;
 351                parent = *p;
 352                rp = rb_entry(parent, struct svc_cacherep, c_node);
 353
 354                cmp = nfsd_cache_key_cmp(key, rp, nn);
 355                if (cmp < 0)
 356                        p = &parent->rb_left;
 357                else if (cmp > 0)
 358                        p = &parent->rb_right;
 359                else {
 360                        ret = rp;
 361                        goto out;
 362                }
 363        }
 364        rb_link_node(&key->c_node, parent, p);
 365        rb_insert_color(&key->c_node, &b->rb_head);
 366out:
 367        /* tally hash chain length stats */
 368        if (entries > nn->longest_chain) {
 369                nn->longest_chain = entries;
 370                nn->longest_chain_cachesize = atomic_read(&nn->num_drc_entries);
 371        } else if (entries == nn->longest_chain) {
 372                /* prefer to keep the smallest cachesize possible here */
 373                nn->longest_chain_cachesize = min_t(unsigned int,
 374                                nn->longest_chain_cachesize,
 375                                atomic_read(&nn->num_drc_entries));
 376        }
 377
 378        lru_put_end(b, ret);
 379        return ret;
 380}
 381
 382/**
 383 * nfsd_cache_lookup - Find an entry in the duplicate reply cache
 384 * @rqstp: Incoming Call to find
 385 *
 386 * Try to find an entry matching the current call in the cache. When none
 387 * is found, we try to grab the oldest expired entry off the LRU list. If
 388 * a suitable one isn't there, then drop the cache_lock and allocate a
 389 * new one, then search again in case one got inserted while this thread
 390 * didn't hold the lock.
 391 *
 392 * Return values:
 393 *   %RC_DOIT: Process the request normally
 394 *   %RC_REPLY: Reply from cache
 395 *   %RC_DROPIT: Do not process the request further
 396 */
 397int nfsd_cache_lookup(struct svc_rqst *rqstp)
 398{
 399        struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 400        struct svc_cacherep     *rp, *found;
 401        __be32                  xid = rqstp->rq_xid;
 402        __wsum                  csum;
 403        u32 hash = nfsd_cache_hash(xid, nn);
 404        struct nfsd_drc_bucket *b = &nn->drc_hashtbl[hash];
 405        int type = rqstp->rq_cachetype;
 406        int rtn = RC_DOIT;
 407
 408        rqstp->rq_cacherep = NULL;
 409        if (type == RC_NOCACHE) {
 410                nfsdstats.rcnocache++;
 411                goto out;
 412        }
 413
 414        csum = nfsd_cache_csum(rqstp);
 415
 416        /*
 417         * Since the common case is a cache miss followed by an insert,
 418         * preallocate an entry.
 419         */
 420        rp = nfsd_reply_cache_alloc(rqstp, csum, nn);
 421        if (!rp)
 422                goto out;
 423
 424        spin_lock(&b->cache_lock);
 425        found = nfsd_cache_insert(b, rp, nn);
 426        if (found != rp) {
 427                nfsd_reply_cache_free_locked(NULL, rp, nn);
 428                rp = found;
 429                goto found_entry;
 430        }
 431
 432        nfsdstats.rcmisses++;
 433        rqstp->rq_cacherep = rp;
 434        rp->c_state = RC_INPROG;
 435
 436        atomic_inc(&nn->num_drc_entries);
 437        nn->drc_mem_usage += sizeof(*rp);
 438
 439        /* go ahead and prune the cache */
 440        prune_bucket(b, nn);
 441
 442out_unlock:
 443        spin_unlock(&b->cache_lock);
 444out:
 445        return rtn;
 446
 447found_entry:
 448        /* We found a matching entry which is either in progress or done. */
 449        nfsdstats.rchits++;
 450        rtn = RC_DROPIT;
 451
 452        /* Request being processed */
 453        if (rp->c_state == RC_INPROG)
 454                goto out_trace;
 455
 456        /* From the hall of fame of impractical attacks:
 457         * Is this a user who tries to snoop on the cache? */
 458        rtn = RC_DOIT;
 459        if (!test_bit(RQ_SECURE, &rqstp->rq_flags) && rp->c_secure)
 460                goto out_trace;
 461
 462        /* Compose RPC reply header */
 463        switch (rp->c_type) {
 464        case RC_NOCACHE:
 465                break;
 466        case RC_REPLSTAT:
 467                svc_putu32(&rqstp->rq_res.head[0], rp->c_replstat);
 468                rtn = RC_REPLY;
 469                break;
 470        case RC_REPLBUFF:
 471                if (!nfsd_cache_append(rqstp, &rp->c_replvec))
 472                        goto out_unlock; /* should not happen */
 473                rtn = RC_REPLY;
 474                break;
 475        default:
 476                WARN_ONCE(1, "nfsd: bad repcache type %d\n", rp->c_type);
 477        }
 478
 479out_trace:
 480        trace_nfsd_drc_found(nn, rqstp, rtn);
 481        goto out_unlock;
 482}
 483
 484/**
 485 * nfsd_cache_update - Update an entry in the duplicate reply cache.
 486 * @rqstp: svc_rqst with a finished Reply
 487 * @cachetype: which cache to update
 488 * @statp: Reply's status code
 489 *
 490 * This is called from nfsd_dispatch when the procedure has been
 491 * executed and the complete reply is in rqstp->rq_res.
 492 *
 493 * We're copying around data here rather than swapping buffers because
 494 * the toplevel loop requires max-sized buffers, which would be a waste
 495 * of memory for a cache with a max reply size of 100 bytes (diropokres).
 496 *
 497 * If we should start to use different types of cache entries tailored
 498 * specifically for attrstat and fh's, we may save even more space.
 499 *
 500 * Also note that a cachetype of RC_NOCACHE can legally be passed when
 501 * nfsd failed to encode a reply that otherwise would have been cached.
 502 * In this case, nfsd_cache_update is called with statp == NULL.
 503 */
 504void nfsd_cache_update(struct svc_rqst *rqstp, int cachetype, __be32 *statp)
 505{
 506        struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
 507        struct svc_cacherep *rp = rqstp->rq_cacherep;
 508        struct kvec     *resv = &rqstp->rq_res.head[0], *cachv;
 509        u32             hash;
 510        struct nfsd_drc_bucket *b;
 511        int             len;
 512        size_t          bufsize = 0;
 513
 514        if (!rp)
 515                return;
 516
 517        hash = nfsd_cache_hash(rp->c_key.k_xid, nn);
 518        b = &nn->drc_hashtbl[hash];
 519
 520        len = resv->iov_len - ((char*)statp - (char*)resv->iov_base);
 521        len >>= 2;
 522
 523        /* Don't cache excessive amounts of data and XDR failures */
 524        if (!statp || len > (256 >> 2)) {
 525                nfsd_reply_cache_free(b, rp, nn);
 526                return;
 527        }
 528
 529        switch (cachetype) {
 530        case RC_REPLSTAT:
 531                if (len != 1)
 532                        printk("nfsd: RC_REPLSTAT/reply len %d!\n",len);
 533                rp->c_replstat = *statp;
 534                break;
 535        case RC_REPLBUFF:
 536                cachv = &rp->c_replvec;
 537                bufsize = len << 2;
 538                cachv->iov_base = kmalloc(bufsize, GFP_KERNEL);
 539                if (!cachv->iov_base) {
 540                        nfsd_reply_cache_free(b, rp, nn);
 541                        return;
 542                }
 543                cachv->iov_len = bufsize;
 544                memcpy(cachv->iov_base, statp, bufsize);
 545                break;
 546        case RC_NOCACHE:
 547                nfsd_reply_cache_free(b, rp, nn);
 548                return;
 549        }
 550        spin_lock(&b->cache_lock);
 551        nn->drc_mem_usage += bufsize;
 552        lru_put_end(b, rp);
 553        rp->c_secure = test_bit(RQ_SECURE, &rqstp->rq_flags);
 554        rp->c_type = cachetype;
 555        rp->c_state = RC_DONE;
 556        spin_unlock(&b->cache_lock);
 557        return;
 558}
 559
 560/*
 561 * Copy cached reply to current reply buffer. Should always fit.
 562 * FIXME as reply is in a page, we should just attach the page, and
 563 * keep a refcount....
 564 */
 565static int
 566nfsd_cache_append(struct svc_rqst *rqstp, struct kvec *data)
 567{
 568        struct kvec     *vec = &rqstp->rq_res.head[0];
 569
 570        if (vec->iov_len + data->iov_len > PAGE_SIZE) {
 571                printk(KERN_WARNING "nfsd: cached reply too large (%zd).\n",
 572                                data->iov_len);
 573                return 0;
 574        }
 575        memcpy((char*)vec->iov_base + vec->iov_len, data->iov_base, data->iov_len);
 576        vec->iov_len += data->iov_len;
 577        return 1;
 578}
 579
 580/*
 581 * Note that fields may be added, removed or reordered in the future. Programs
 582 * scraping this file for info should test the labels to ensure they're
 583 * getting the correct field.
 584 */
 585static int nfsd_reply_cache_stats_show(struct seq_file *m, void *v)
 586{
 587        struct nfsd_net *nn = m->private;
 588
 589        seq_printf(m, "max entries:           %u\n", nn->max_drc_entries);
 590        seq_printf(m, "num entries:           %u\n",
 591                        atomic_read(&nn->num_drc_entries));
 592        seq_printf(m, "hash buckets:          %u\n", 1 << nn->maskbits);
 593        seq_printf(m, "mem usage:             %u\n", nn->drc_mem_usage);
 594        seq_printf(m, "cache hits:            %u\n", nfsdstats.rchits);
 595        seq_printf(m, "cache misses:          %u\n", nfsdstats.rcmisses);
 596        seq_printf(m, "not cached:            %u\n", nfsdstats.rcnocache);
 597        seq_printf(m, "payload misses:        %u\n", nn->payload_misses);
 598        seq_printf(m, "longest chain len:     %u\n", nn->longest_chain);
 599        seq_printf(m, "cachesize at longest:  %u\n", nn->longest_chain_cachesize);
 600        return 0;
 601}
 602
 603int nfsd_reply_cache_stats_open(struct inode *inode, struct file *file)
 604{
 605        struct nfsd_net *nn = net_generic(file_inode(file)->i_sb->s_fs_info,
 606                                                                nfsd_net_id);
 607
 608        return single_open(file, nfsd_reply_cache_stats_show, nn);
 609}
 610