net/sunrpc/cache.c

   1 /*
   2  * net/sunrpc/cache.c
   3  *
   4  * Generic code for various authentication-related caches
   5  * used by sunrpc clients and servers.
   6  *
   7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   8  *
   9  * Released under terms in GPL version 2.  See COPYING.
  10  *
  11  */
  12
  13 #include <linux/types.h>
  14 #include <linux/fs.h>
  15 #include <linux/file.h>
  16 #include <linux/slab.h>
  17 #include <linux/signal.h>
  18 #include <linux/sched.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/module.h>
  22 #include <linux/ctype.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/poll.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/net.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/mutex.h>
  30 #include <linux/pagemap.h>
  31 #include <asm/ioctls.h>
  32 #include <linux/sunrpc/types.h>
  33 #include <linux/sunrpc/cache.h>
  34 #include <linux/sunrpc/stats.h>
  35 #include <linux/sunrpc/rpc_pipe_fs.h>
  36 #include "netns.h"
  37
  38 #define  RPCDBG_FACILITY RPCDBG_CACHE
  39
  40 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  41 static void cache_revisit_request(struct cache_head *item);
  42
  43 static void cache_init(struct cache_head *h)
  44 {
  45         time_t now = seconds_since_boot();
  46         h->next = NULL;
  47         h->flags = 0;
  48         kref_init(&h->ref);
  49         h->expiry_time = now + CACHE_NEW_EXPIRY;
  50         h->last_refresh = now;
  51 }
  52
  53 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  54                                        struct cache_head *key, int hash)
  55 {
  56         struct cache_head **head,  **hp;
  57         struct cache_head *new = NULL, *freeme = NULL;
  58
  59         head = &detail->hash_table[hash];
  60
  61         read_lock(&detail->hash_lock);
  62
  63         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  64                 struct cache_head *tmp = *hp;
  65                 if (detail->match(tmp, key)) {
  66                         if (cache_is_expired(detail, tmp))
  67                                 /* This entry is expired, we will discard it. */
  68                                 break;
  69                         cache_get(tmp);
  70                         read_unlock(&detail->hash_lock);
  71                         return tmp;
  72                 }
  73         }
  74         read_unlock(&detail->hash_lock);
  75         /* Didn't find anything, insert an empty entry */
  76
  77         new = detail->alloc();
  78         if (!new)
  79                 return NULL;
  80         /* must fully initialise 'new', else
  81          * we might get lose if we need to
  82          * cache_put it soon.
  83          */
  84         cache_init(new);
  85         detail->init(new, key);
  86
  87         write_lock(&detail->hash_lock);
  88
  89         /* check if entry appeared while we slept */
  90         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  91                 struct cache_head *tmp = *hp;
  92                 if (detail->match(tmp, key)) {
  93                         if (cache_is_expired(detail, tmp)) {
  94                                 *hp = tmp->next;
  95                                 tmp->next = NULL;
  96                                 detail->entries --;
  97                                 freeme = tmp;
  98                                 break;
  99                         }
 100                         cache_get(tmp);
 101                         write_unlock(&detail->hash_lock);
 102                         cache_put(new, detail);
 103                         return tmp;
 104                 }
 105         }
 106         new->next = *head;
 107         *head = new;
 108         detail->entries++;
 109         cache_get(new);
 110         write_unlock(&detail->hash_lock);
 111
 112         if (freeme)
 113                 cache_put(freeme, detail);
 114         return new;
 115 }
 116 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
 117
 118
 119 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 120
 121 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
 122 {
 123         head->expiry_time = expiry;
 124         head->last_refresh = seconds_since_boot();
 125         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 126         set_bit(CACHE_VALID, &head->flags);
 127 }
 128
 129 static void cache_fresh_unlocked(struct cache_head *head,
 130                                  struct cache_detail *detail)
 131 {
 132         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 133                 cache_revisit_request(head);
 134                 cache_dequeue(detail, head);
 135         }
 136 }
 137
 138 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 139                                        struct cache_head *new, struct cache_head *old, int hash)
 140 {
 141         /* The 'old' entry is to be replaced by 'new'.
 142          * If 'old' is not VALID, we update it directly,
 143          * otherwise we need to replace it
 144          */
 145         struct cache_head **head;
 146         struct cache_head *tmp;
 147
 148         if (!test_bit(CACHE_VALID, &old->flags)) {
 149                 write_lock(&detail->hash_lock);
 150                 if (!test_bit(CACHE_VALID, &old->flags)) {
 151                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 152                                 set_bit(CACHE_NEGATIVE, &old->flags);
 153                         else
 154                                 detail->update(old, new);
 155                         cache_fresh_locked(old, new->expiry_time);
 156                         write_unlock(&detail->hash_lock);
 157                         cache_fresh_unlocked(old, detail);
 158                         return old;
 159                 }
 160                 write_unlock(&detail->hash_lock);
 161         }
 162         /* We need to insert a new entry */
 163         tmp = detail->alloc();
 164         if (!tmp) {
 165                 cache_put(old, detail);
 166                 return NULL;
 167         }
 168         cache_init(tmp);
 169         detail->init(tmp, old);
 170         head = &detail->hash_table[hash];
 171
 172         write_lock(&detail->hash_lock);
 173         if (test_bit(CACHE_NEGATIVE, &new->flags))
 174                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 175         else
 176                 detail->update(tmp, new);
 177         tmp->next = *head;
 178         *head = tmp;
 179         detail->entries++;
 180         cache_get(tmp);
 181         cache_fresh_locked(tmp, new->expiry_time);
 182         cache_fresh_locked(old, 0);
 183         write_unlock(&detail->hash_lock);
 184         cache_fresh_unlocked(tmp, detail);
 185         cache_fresh_unlocked(old, detail);
 186         cache_put(old, detail);
 187         return tmp;
 188 }
 189 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 190
 191 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 192 {
 193         if (cd->cache_upcall)
 194                 return cd->cache_upcall(cd, h);
 195         return sunrpc_cache_pipe_upcall(cd, h);
 196 }
 197
 198 static inline int cache_is_valid(struct cache_head *h)
 199 {
 200         if (!test_bit(CACHE_VALID, &h->flags))
 201                 return -EAGAIN;
 202         else {
 203                 /* entry is valid */
 204                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 205                         return -ENOENT;
 206                 else {
 207                         /*
 208                          * In combination with write barrier in
 209                          * sunrpc_cache_update, ensures that anyone
 210                          * using the cache entry after this sees the
 211                          * updated contents:
 212                          */
 213                         smp_rmb();
 214                         return 0;
 215                 }
 216         }
 217 }
 218
 219 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 220 {
 221         int rv;
 222
 223         write_lock(&detail->hash_lock);
 224         rv = cache_is_valid(h);
 225         if (rv == -EAGAIN) {
 226                 set_bit(CACHE_NEGATIVE, &h->flags);
 227                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
 228                 rv = -ENOENT;
 229         }
 230         write_unlock(&detail->hash_lock);
 231         cache_fresh_unlocked(h, detail);
 232         return rv;
 233 }
 234
 235 /*
 236  * This is the generic cache management routine for all
 237  * the authentication caches.
 238  * It checks the currency of a cache item and will (later)
 239  * initiate an upcall to fill it if needed.
 240  *
 241  *
 242  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 243  * -EAGAIN if upcall is pending and request has been queued
 244  * -ETIMEDOUT if upcall failed or request could not be queue or
 245  *           upcall completed but item is still invalid (implying that
 246  *           the cache item has been replaced with a newer one).
 247  * -ENOENT if cache entry was negative
 248  */
 249 int cache_check(struct cache_detail *detail,
 250                     struct cache_head *h, struct cache_req *rqstp)
 251 {
 252         int rv;
 253         long refresh_age, age;
 254
 255         /* First decide return status as best we can */
 256         rv = cache_is_valid(h);
 257
 258         /* now see if we want to start an upcall */
 259         refresh_age = (h->expiry_time - h->last_refresh);
 260         age = seconds_since_boot() - h->last_refresh;
 261
 262         if (rqstp == NULL) {
 263                 if (rv == -EAGAIN)
 264                         rv = -ENOENT;
 265         } else if (rv == -EAGAIN ||
 266                    (h->expiry_time != 0 && age > refresh_age/2)) {
 267                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 268                                 refresh_age, age);
 269                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 270                         switch (cache_make_upcall(detail, h)) {
 271                         case -EINVAL:
 272                                 rv = try_to_negate_entry(detail, h);
 273                                 break;
 274                         case -EAGAIN:
 275                                 cache_fresh_unlocked(h, detail);
 276                                 break;
 277                         }
 278                 }
 279         }
 280
 281         if (rv == -EAGAIN) {
 282                 if (!cache_defer_req(rqstp, h)) {
 283                         /*
 284                          * Request was not deferred; handle it as best
 285                          * we can ourselves:
 286                          */
 287                         rv = cache_is_valid(h);
 288                         if (rv == -EAGAIN)
 289                                 rv = -ETIMEDOUT;
 290                 }
 291         }
 292         if (rv)
 293                 cache_put(h, detail);
 294         return rv;
 295 }
 296 EXPORT_SYMBOL_GPL(cache_check);
 297
 298 /*
 299  * caches need to be periodically cleaned.
 300  * For this we maintain a list of cache_detail and
 301  * a current pointer into that list and into the table
 302  * for that entry.
 303  *
 304  * Each time cache_clean is called it finds the next non-empty entry
 305  * in the current table and walks the list in that entry
 306  * looking for entries that can be removed.
 307  *
 308  * An entry gets removed if:
 309  * - The expiry is before current time
 310  * - The last_refresh time is before the flush_time for that cache
 311  *
 312  * later we might drop old entries with non-NEVER expiry if that table
 313  * is getting 'full' for some definition of 'full'
 314  *
 315  * The question of "how often to scan a table" is an interesting one
 316  * and is answered in part by the use of the "nextcheck" field in the
 317  * cache_detail.
 318  * When a scan of a table begins, the nextcheck field is set to a time
 319  * that is well into the future.
 320  * While scanning, if an expiry time is found that is earlier than the
 321  * current nextcheck time, nextcheck is set to that expiry time.
 322  * If the flush_time is ever set to a time earlier than the nextcheck
 323  * time, the nextcheck time is then set to that flush_time.
 324  *
 325  * A table is then only scanned if the current time is at least
 326  * the nextcheck time.
 327  *
 328  */
 329
 330 static LIST_HEAD(cache_list);
 331 static DEFINE_SPINLOCK(cache_list_lock);
 332 static struct cache_detail *current_detail;
 333 static int current_index;
 334
 335 static void do_cache_clean(struct work_struct *work);
 336 static struct delayed_work cache_cleaner;
 337
 338 void sunrpc_init_cache_detail(struct cache_detail *cd)
 339 {
 340         rwlock_init(&cd->hash_lock);
 341         INIT_LIST_HEAD(&cd->queue);
 342         spin_lock(&cache_list_lock);
 343         cd->nextcheck = 0;
 344         cd->entries = 0;
 345         atomic_set(&cd->readers, 0);
 346         cd->last_close = 0;
 347         cd->last_warn = -1;
 348         list_add(&cd->others, &cache_list);
 349         spin_unlock(&cache_list_lock);
 350
 351         /* start the cleaning process */
 352         schedule_delayed_work(&cache_cleaner, 0);
 353 }
 354 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 355
 356 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 357 {
 358         cache_purge(cd);
 359         spin_lock(&cache_list_lock);
 360         write_lock(&cd->hash_lock);
 361         if (cd->entries || atomic_read(&cd->inuse)) {
 362                 write_unlock(&cd->hash_lock);
 363                 spin_unlock(&cache_list_lock);
 364                 goto out;
 365         }
 366         if (current_detail == cd)
 367                 current_detail = NULL;
 368         list_del_init(&cd->others);
 369         write_unlock(&cd->hash_lock);
 370         spin_unlock(&cache_list_lock);
 371         if (list_empty(&cache_list)) {
 372                 /* module must be being unloaded so its safe to kill the worker */
 373                 cancel_delayed_work_sync(&cache_cleaner);
 374         }
 375         return;
 376 out:
 377         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 378 }
 379 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 380
 381 /* clean cache tries to find something to clean
 382  * and cleans it.
 383  * It returns 1 if it cleaned something,
 384  *            0 if it didn't find anything this time
 385  *           -1 if it fell off the end of the list.
 386  */
 387 static int cache_clean(void)
 388 {
 389         int rv = 0;
 390         struct list_head *next;
 391
 392         spin_lock(&cache_list_lock);
 393
 394         /* find a suitable table if we don't already have one */
 395         while (current_detail == NULL ||
 396             current_index >= current_detail->hash_size) {
 397                 if (current_detail)
 398                         next = current_detail->others.next;
 399                 else
 400                         next = cache_list.next;
 401                 if (next == &cache_list) {
 402                         current_detail = NULL;
 403                         spin_unlock(&cache_list_lock);
 404                         return -1;
 405                 }
 406                 current_detail = list_entry(next, struct cache_detail, others);
 407                 if (current_detail->nextcheck > seconds_since_boot())
 408                         current_index = current_detail->hash_size;
 409                 else {
 410                         current_index = 0;
 411                         current_detail->nextcheck = seconds_since_boot()+30*60;
 412                 }
 413         }
 414
 415         /* find a non-empty bucket in the table */
 416         while (current_detail &&
 417                current_index < current_detail->hash_size &&
 418                current_detail->hash_table[current_index] == NULL)
 419                 current_index++;
 420
 421         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 422
 423         if (current_detail && current_index < current_detail->hash_size) {
 424                 struct cache_head *ch, **cp;
 425                 struct cache_detail *d;
 426
 427                 write_lock(&current_detail->hash_lock);
 428
 429                 /* Ok, now to clean this strand */
 430
 431                 cp = & current_detail->hash_table[current_index];
 432                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
 433                         if (current_detail->nextcheck > ch->expiry_time)
 434                                 current_detail->nextcheck = ch->expiry_time+1;
 435                         if (!cache_is_expired(current_detail, ch))
 436                                 continue;
 437
 438                         *cp = ch->next;
 439                         ch->next = NULL;
 440                         current_detail->entries--;
 441                         rv = 1;
 442                         break;
 443                 }
 444
 445                 write_unlock(&current_detail->hash_lock);
 446                 d = current_detail;
 447                 if (!ch)
 448                         current_index ++;
 449                 spin_unlock(&cache_list_lock);
 450                 if (ch) {
 451                         set_bit(CACHE_CLEANED, &ch->flags);
 452                         cache_fresh_unlocked(ch, d);
 453                         cache_put(ch, d);
 454                 }
 455         } else
 456                 spin_unlock(&cache_list_lock);
 457
 458         return rv;
 459 }
 460
 461 /*
 462  * We want to regularly clean the cache, so we need to schedule some work ...
 463  */
 464 static void do_cache_clean(struct work_struct *work)
 465 {
 466         int delay = 5;
 467         if (cache_clean() == -1)
 468                 delay = round_jiffies_relative(30*HZ);
 469
 470         if (list_empty(&cache_list))
 471                 delay = 0;
 472
 473         if (delay)
 474                 schedule_delayed_work(&cache_cleaner, delay);
 475 }
 476
 477
 478 /*
 479  * Clean all caches promptly.  This just calls cache_clean
 480  * repeatedly until we are sure that every cache has had a chance to
 481  * be fully cleaned
 482  */
 483 void cache_flush(void)
 484 {
 485         while (cache_clean() != -1)
 486                 cond_resched();
 487         while (cache_clean() != -1)
 488                 cond_resched();
 489 }
 490 EXPORT_SYMBOL_GPL(cache_flush);
 491
 492 void cache_purge(struct cache_detail *detail)
 493 {
 494         detail->flush_time = LONG_MAX;
 495         detail->nextcheck = seconds_since_boot();
 496         cache_flush();
 497         detail->flush_time = 1;
 498 }
 499 EXPORT_SYMBOL_GPL(cache_purge);
 500
 501
 502 /*
 503  * Deferral and Revisiting of Requests.
 504  *
 505  * If a cache lookup finds a pending entry, we
 506  * need to defer the request and revisit it later.
 507  * All deferred requests are stored in a hash table,
 508  * indexed by "struct cache_head *".
 509  * As it may be wasteful to store a whole request
 510  * structure, we allow the request to provide a
 511  * deferred form, which must contain a
 512  * 'struct cache_deferred_req'
 513  * This cache_deferred_req contains a method to allow
 514  * it to be revisited when cache info is available
 515  */
 516
 517 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 518 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 519
 520 #define DFR_MAX 300     /* ??? */
 521
 522 static DEFINE_SPINLOCK(cache_defer_lock);
 523 static LIST_HEAD(cache_defer_list);
 524 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 525 static int cache_defer_cnt;
 526
 527 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 528 {
 529         hlist_del_init(&dreq->hash);
 530         if (!list_empty(&dreq->recent)) {
 531                 list_del_init(&dreq->recent);
 532                 cache_defer_cnt--;
 533         }
 534 }
 535
 536 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 537 {
 538         int hash = DFR_HASH(item);
 539
 540         INIT_LIST_HEAD(&dreq->recent);
 541         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 542 }
 543
 544 static void setup_deferral(struct cache_deferred_req *dreq,
 545                            struct cache_head *item,
 546                            int count_me)
 547 {
 548
 549         dreq->item = item;
 550
 551         spin_lock(&cache_defer_lock);
 552
 553         __hash_deferred_req(dreq, item);
 554
 555         if (count_me) {
 556                 cache_defer_cnt++;
 557                 list_add(&dreq->recent, &cache_defer_list);
 558         }
 559
 560         spin_unlock(&cache_defer_lock);
 561
 562 }
 563
 564 struct thread_deferred_req {
 565         struct cache_deferred_req handle;
 566         struct completion completion;
 567 };
 568
 569 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 570 {
 571         struct thread_deferred_req *dr =
 572                 container_of(dreq, struct thread_deferred_req, handle);
 573         complete(&dr->completion);
 574 }
 575
 576 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 577 {
 578         struct thread_deferred_req sleeper;
 579         struct cache_deferred_req *dreq = &sleeper.handle;
 580
 581         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 582         dreq->revisit = cache_restart_thread;
 583
 584         setup_deferral(dreq, item, 0);
 585
 586         if (!test_bit(CACHE_PENDING, &item->flags) ||
 587             wait_for_completion_interruptible_timeout(
 588                     &sleeper.completion, req->thread_wait) <= 0) {
 589                 /* The completion wasn't completed, so we need
 590                  * to clean up
 591                  */
 592                 spin_lock(&cache_defer_lock);
 593                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 594                         __unhash_deferred_req(&sleeper.handle);
 595                         spin_unlock(&cache_defer_lock);
 596                 } else {
 597                         /* cache_revisit_request already removed
 598                          * this from the hash table, but hasn't
 599                          * called ->revisit yet.  It will very soon
 600                          * and we need to wait for it.
 601                          */
 602                         spin_unlock(&cache_defer_lock);
 603                         wait_for_completion(&sleeper.completion);
 604                 }
 605         }
 606 }
 607
 608 static void cache_limit_defers(void)
 609 {
 610         /* Make sure we haven't exceed the limit of allowed deferred
 611          * requests.
 612          */
 613         struct cache_deferred_req *discard = NULL;
 614
 615         if (cache_defer_cnt <= DFR_MAX)
 616                 return;
 617
 618         spin_lock(&cache_defer_lock);
 619
 620         /* Consider removing either the first or the last */
 621         if (cache_defer_cnt > DFR_MAX) {
 622                 if (net_random() & 1)
 623                         discard = list_entry(cache_defer_list.next,
 624                                              struct cache_deferred_req, recent);
 625                 else
 626                         discard = list_entry(cache_defer_list.prev,
 627                                              struct cache_deferred_req, recent);
 628                 __unhash_deferred_req(discard);
 629         }
 630         spin_unlock(&cache_defer_lock);
 631         if (discard)
 632                 discard->revisit(discard, 1);
 633 }
 634
 635 /* Return true if and only if a deferred request is queued. */
 636 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 637 {
 638         struct cache_deferred_req *dreq;
 639
 640         if (req->thread_wait) {
 641                 cache_wait_req(req, item);
 642                 if (!test_bit(CACHE_PENDING, &item->flags))
 643                         return false;
 644         }
 645         dreq = req->defer(req);
 646         if (dreq == NULL)
 647                 return false;
 648         setup_deferral(dreq, item, 1);
 649         if (!test_bit(CACHE_PENDING, &item->flags))
 650                 /* Bit could have been cleared before we managed to
 651                  * set up the deferral, so need to revisit just in case
 652                  */
 653                 cache_revisit_request(item);
 654
 655         cache_limit_defers();
 656         return true;
 657 }
 658
 659 static void cache_revisit_request(struct cache_head *item)
 660 {
 661         struct cache_deferred_req *dreq;
 662         struct list_head pending;
 663         struct hlist_node *tmp;
 664         int hash = DFR_HASH(item);
 665
 666         INIT_LIST_HEAD(&pending);
 667         spin_lock(&cache_defer_lock);
 668
 669         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 670                 if (dreq->item == item) {
 671                         __unhash_deferred_req(dreq);
 672                         list_add(&dreq->recent, &pending);
 673                 }
 674
 675         spin_unlock(&cache_defer_lock);
 676
 677         while (!list_empty(&pending)) {
 678                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 679                 list_del_init(&dreq->recent);
 680                 dreq->revisit(dreq, 0);
 681         }
 682 }
 683
 684 void cache_clean_deferred(void *owner)
 685 {
 686         struct cache_deferred_req *dreq, *tmp;
 687         struct list_head pending;
 688
 689
 690         INIT_LIST_HEAD(&pending);
 691         spin_lock(&cache_defer_lock);
 692
 693         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 694                 if (dreq->owner == owner) {
 695                         __unhash_deferred_req(dreq);
 696                         list_add(&dreq->recent, &pending);
 697                 }
 698         }
 699         spin_unlock(&cache_defer_lock);
 700
 701         while (!list_empty(&pending)) {
 702                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 703                 list_del_init(&dreq->recent);
 704                 dreq->revisit(dreq, 1);
 705         }
 706 }
 707
 708 /*
 709  * communicate with user-space
 710  *
 711  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 712  * On read, you get a full request, or block.
 713  * On write, an update request is processed.
 714  * Poll works if anything to read, and always allows write.
 715  *
 716  * Implemented by linked list of requests.  Each open file has
 717  * a ->private that also exists in this list.  New requests are added
 718  * to the end and may wakeup and preceding readers.
 719  * New readers are added to the head.  If, on read, an item is found with
 720  * CACHE_UPCALLING clear, we free it from the list.
 721  *
 722  */
 723
 724 static DEFINE_SPINLOCK(queue_lock);
 725 static DEFINE_MUTEX(queue_io_mutex);
 726
 727 struct cache_queue {
 728         struct list_head        list;
 729         int                     reader; /* if 0, then request */
 730 };
 731 struct cache_request {
 732         struct cache_queue      q;
 733         struct cache_head       *item;
 734         char                    * buf;
 735         int                     len;
 736         int                     readers;
 737 };
 738 struct cache_reader {
 739         struct cache_queue      q;
 740         int                     offset; /* if non-0, we have a refcnt on next request */
 741 };
 742
 743 static int cache_request(struct cache_detail *detail,
 744                                struct cache_request *crq)
 745 {
 746         char *bp = crq->buf;
 747         int len = PAGE_SIZE;
 748
 749         detail->cache_request(detail, crq->item, &bp, &len);
 750         if (len < 0)
 751                 return -EAGAIN;
 752         return PAGE_SIZE - len;
 753 }
 754
 755 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 756                           loff_t *ppos, struct cache_detail *cd)
 757 {
 758         struct cache_reader *rp = filp->private_data;
 759         struct cache_request *rq;
 760         struct inode *inode = file_inode(filp);
 761         int err;
 762
 763         if (count == 0)
 764                 return 0;
 765
 766         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 767                               * readers on this file */
 768  again:
 769         spin_lock(&queue_lock);
 770         /* need to find next request */
 771         while (rp->q.list.next != &cd->queue &&
 772                list_entry(rp->q.list.next, struct cache_queue, list)
 773                ->reader) {
 774                 struct list_head *next = rp->q.list.next;
 775                 list_move(&rp->q.list, next);
 776         }
 777         if (rp->q.list.next == &cd->queue) {
 778                 spin_unlock(&queue_lock);
 779                 mutex_unlock(&inode->i_mutex);
 780                 WARN_ON_ONCE(rp->offset);
 781                 return 0;
 782         }
 783         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 784         WARN_ON_ONCE(rq->q.reader);
 785         if (rp->offset == 0)
 786                 rq->readers++;
 787         spin_unlock(&queue_lock);
 788
 789         if (rq->len == 0) {
 790                 err = cache_request(cd, rq);
 791                 if (err < 0)
 792                         goto out;
 793                 rq->len = err;
 794         }
 795
 796         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 797                 err = -EAGAIN;
 798                 spin_lock(&queue_lock);
 799                 list_move(&rp->q.list, &rq->q.list);
 800                 spin_unlock(&queue_lock);
 801         } else {
 802                 if (rp->offset + count > rq->len)
 803                         count = rq->len - rp->offset;
 804                 err = -EFAULT;
 805                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 806                         goto out;
 807                 rp->offset += count;
 808                 if (rp->offset >= rq->len) {
 809                         rp->offset = 0;
 810                         spin_lock(&queue_lock);
 811                         list_move(&rp->q.list, &rq->q.list);
 812                         spin_unlock(&queue_lock);
 813                 }
 814                 err = 0;
 815         }
 816  out:
 817         if (rp->offset == 0) {
 818                 /* need to release rq */
 819                 spin_lock(&queue_lock);
 820                 rq->readers--;
 821                 if (rq->readers == 0 &&
 822                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 823                         list_del(&rq->q.list);
 824                         spin_unlock(&queue_lock);
 825                         cache_put(rq->item, cd);
 826                         kfree(rq->buf);
 827                         kfree(rq);
 828                 } else
 829                         spin_unlock(&queue_lock);
 830         }
 831         if (err == -EAGAIN)
 832                 goto again;
 833         mutex_unlock(&inode->i_mutex);
 834         return err ? err :  count;
 835 }
 836
 837 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 838                                  size_t count, struct cache_detail *cd)
 839 {
 840         ssize_t ret;
 841
 842         if (count == 0)
 843                 return -EINVAL;
 844         if (copy_from_user(kaddr, buf, count))
 845                 return -EFAULT;
 846         kaddr[count] = '\0';
 847         ret = cd->cache_parse(cd, kaddr, count);
 848         if (!ret)
 849                 ret = count;
 850         return ret;
 851 }
 852
 853 static ssize_t cache_slow_downcall(const char __user *buf,
 854                                    size_t count, struct cache_detail *cd)
 855 {
 856         static char write_buf[8192]; /* protected by queue_io_mutex */
 857         ssize_t ret = -EINVAL;
 858
 859         if (count >= sizeof(write_buf))
 860                 goto out;
 861         mutex_lock(&queue_io_mutex);
 862         ret = cache_do_downcall(write_buf, buf, count, cd);
 863         mutex_unlock(&queue_io_mutex);
 864 out:
 865         return ret;
 866 }
 867
 868 static ssize_t cache_downcall(struct address_space *mapping,
 869                               const char __user *buf,
 870                               size_t count, struct cache_detail *cd)
 871 {
 872         struct page *page;
 873         char *kaddr;
 874         ssize_t ret = -ENOMEM;
 875
 876         if (count >= PAGE_CACHE_SIZE)
 877                 goto out_slow;
 878
 879         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 880         if (!page)
 881                 goto out_slow;
 882
 883         kaddr = kmap(page);
 884         ret = cache_do_downcall(kaddr, buf, count, cd);
 885         kunmap(page);
 886         unlock_page(page);
 887         page_cache_release(page);
 888         return ret;
 889 out_slow:
 890         return cache_slow_downcall(buf, count, cd);
 891 }
 892
 893 static ssize_t cache_write(struct file *filp, const char __user *buf,
 894                            size_t count, loff_t *ppos,
 895                            struct cache_detail *cd)
 896 {
 897         struct address_space *mapping = filp->f_mapping;
 898         struct inode *inode = file_inode(filp);
 899         ssize_t ret = -EINVAL;
 900
 901         if (!cd->cache_parse)
 902                 goto out;
 903
 904         mutex_lock(&inode->i_mutex);
 905         ret = cache_downcall(mapping, buf, count, cd);
 906         mutex_unlock(&inode->i_mutex);
 907 out:
 908         return ret;
 909 }
 910
 911 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 912
 913 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 914                                struct cache_detail *cd)
 915 {
 916         unsigned int mask;
 917         struct cache_reader *rp = filp->private_data;
 918         struct cache_queue *cq;
 919
 920         poll_wait(filp, &queue_wait, wait);
 921
 922         /* alway allow write */
 923         mask = POLL_OUT | POLLWRNORM;
 924
 925         if (!rp)
 926                 return mask;
 927
 928         spin_lock(&queue_lock);
 929
 930         for (cq= &rp->q; &cq->list != &cd->queue;
 931              cq = list_entry(cq->list.next, struct cache_queue, list))
 932                 if (!cq->reader) {
 933                         mask |= POLLIN | POLLRDNORM;
 934                         break;
 935                 }
 936         spin_unlock(&queue_lock);
 937         return mask;
 938 }
 939
 940 static int cache_ioctl(struct inode *ino, struct file *filp,
 941                        unsigned int cmd, unsigned long arg,
 942                        struct cache_detail *cd)
 943 {
 944         int len = 0;
 945         struct cache_reader *rp = filp->private_data;
 946         struct cache_queue *cq;
 947
 948         if (cmd != FIONREAD || !rp)
 949                 return -EINVAL;
 950
 951         spin_lock(&queue_lock);
 952
 953         /* only find the length remaining in current request,
 954          * or the length of the next request
 955          */
 956         for (cq= &rp->q; &cq->list != &cd->queue;
 957              cq = list_entry(cq->list.next, struct cache_queue, list))
 958                 if (!cq->reader) {
 959                         struct cache_request *cr =
 960                                 container_of(cq, struct cache_request, q);
 961                         len = cr->len - rp->offset;
 962                         break;
 963                 }
 964         spin_unlock(&queue_lock);
 965
 966         return put_user(len, (int __user *)arg);
 967 }
 968
 969 static int cache_open(struct inode *inode, struct file *filp,
 970                       struct cache_detail *cd)
 971 {
 972         struct cache_reader *rp = NULL;
 973
 974         if (!cd || !try_module_get(cd->owner))
 975                 return -EACCES;
 976         nonseekable_open(inode, filp);
 977         if (filp->f_mode & FMODE_READ) {
 978                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 979                 if (!rp) {
 980                         module_put(cd->owner);
 981                         return -ENOMEM;
 982                 }
 983                 rp->offset = 0;
 984                 rp->q.reader = 1;
 985                 atomic_inc(&cd->readers);
 986                 spin_lock(&queue_lock);
 987                 list_add(&rp->q.list, &cd->queue);
 988                 spin_unlock(&queue_lock);
 989         }
 990         filp->private_data = rp;
 991         return 0;
 992 }
 993
 994 static int cache_release(struct inode *inode, struct file *filp,
 995                          struct cache_detail *cd)
 996 {
 997         struct cache_reader *rp = filp->private_data;
 998
 999         if (rp) {
1000                 spin_lock(&queue_lock);
1001                 if (rp->offset) {
1002                         struct cache_queue *cq;
1003                         for (cq= &rp->q; &cq->list != &cd->queue;
1004                              cq = list_entry(cq->list.next, struct cache_queue, list))
1005                                 if (!cq->reader) {
1006                                         container_of(cq, struct cache_request, q)
1007                                                 ->readers--;
1008                                         break;
1009                                 }
1010                         rp->offset = 0;
1011                 }
1012                 list_del(&rp->q.list);
1013                 spin_unlock(&queue_lock);
1014
1015                 filp->private_data = NULL;
1016                 kfree(rp);
1017
1018                 cd->last_close = seconds_since_boot();
1019                 atomic_dec(&cd->readers);
1020         }
1021         module_put(cd->owner);
1022         return 0;
1023 }
1024
1025
1026
1027 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1028 {
1029         struct cache_queue *cq, *tmp;
1030         struct cache_request *cr;
1031         struct list_head dequeued;
1032
1033         INIT_LIST_HEAD(&dequeued);
1034         spin_lock(&queue_lock);
1035         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1036                 if (!cq->reader) {
1037                         cr = container_of(cq, struct cache_request, q);
1038                         if (cr->item != ch)
1039                                 continue;
1040                         if (test_bit(CACHE_PENDING, &ch->flags))
1041                                 /* Lost a race and it is pending again */
1042                                 break;
1043                         if (cr->readers != 0)
1044                                 continue;
1045                         list_move(&cr->q.list, &dequeued);
1046                 }
1047         spin_unlock(&queue_lock);
1048         while (!list_empty(&dequeued)) {
1049                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1050                 list_del(&cr->q.list);
1051                 cache_put(cr->item, detail);
1052                 kfree(cr->buf);
1053                 kfree(cr);
1054         }
1055 }
1056
1057 /*
1058  * Support routines for text-based upcalls.
1059  * Fields are separated by spaces.
1060  * Fields are either mangled to quote space tab newline slosh with slosh
1061  * or a hexified with a leading \x
1062  * Record is terminated with newline.
1063  *
1064  */
1065
1066 void qword_add(char **bpp, int *lp, char *str)
1067 {
1068         char *bp = *bpp;
1069         int len = *lp;
1070         char c;
1071
1072         if (len < 0) return;
1073
1074         while ((c=*str++) && len)
1075                 switch(c) {
1076                 case ' ':
1077                 case '\t':
1078                 case '\n':
1079                 case '\\':
1080                         if (len >= 4) {
1081                                 *bp++ = '\\';
1082                                 *bp++ = '0' + ((c & 0300)>>6);
1083                                 *bp++ = '0' + ((c & 0070)>>3);
1084                                 *bp++ = '0' + ((c & 0007)>>0);
1085                         }
1086                         len -= 4;
1087                         break;
1088                 default:
1089                         *bp++ = c;
1090                         len--;
1091                 }
1092         if (c || len <1) len = -1;
1093         else {
1094                 *bp++ = ' ';
1095                 len--;
1096         }
1097         *bpp = bp;
1098         *lp = len;
1099 }
1100 EXPORT_SYMBOL_GPL(qword_add);
1101
1102 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1103 {
1104         char *bp = *bpp;
1105         int len = *lp;
1106
1107         if (len < 0) return;
1108
1109         if (len > 2) {
1110                 *bp++ = '\\';
1111                 *bp++ = 'x';
1112                 len -= 2;
1113                 while (blen && len >= 2) {
1114                         unsigned char c = *buf++;
1115                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1116                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1117                         len -= 2;
1118                         blen--;
1119                 }
1120         }
1121         if (blen || len<1) len = -1;
1122         else {
1123                 *bp++ = ' ';
1124                 len--;
1125         }
1126         *bpp = bp;
1127         *lp = len;
1128 }
1129 EXPORT_SYMBOL_GPL(qword_addhex);
1130
1131 static void warn_no_listener(struct cache_detail *detail)
1132 {
1133         if (detail->last_warn != detail->last_close) {
1134                 detail->last_warn = detail->last_close;
1135                 if (detail->warn_no_listener)
1136                         detail->warn_no_listener(detail, detail->last_close != 0);
1137         }
1138 }
1139
1140 static bool cache_listeners_exist(struct cache_detail *detail)
1141 {
1142         if (atomic_read(&detail->readers))
1143                 return true;
1144         if (detail->last_close == 0)
1145                 /* This cache was never opened */
1146                 return false;
1147         if (detail->last_close < seconds_since_boot() - 30)
1148                 /*
1149                  * We allow for the possibility that someone might
1150                  * restart a userspace daemon without restarting the
1151                  * server; but after 30 seconds, we give up.
1152                  */
1153                  return false;
1154         return true;
1155 }
1156
1157 /*
1158  * register an upcall request to user-space and queue it up for read() by the
1159  * upcall daemon.
1160  *
1161  * Each request is at most one page long.
1162  */
1163 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1164 {
1165
1166         char *buf;
1167         struct cache_request *crq;
1168         int ret = 0;
1169
1170         if (!detail->cache_request)
1171                 return -EINVAL;
1172
1173         if (!cache_listeners_exist(detail)) {
1174                 warn_no_listener(detail);
1175                 return -EINVAL;
1176         }
1177         if (test_bit(CACHE_CLEANED, &h->flags))
1178                 /* Too late to make an upcall */
1179                 return -EAGAIN;
1180
1181         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1182         if (!buf)
1183                 return -EAGAIN;
1184
1185         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1186         if (!crq) {
1187                 kfree(buf);
1188                 return -EAGAIN;
1189         }
1190
1191         crq->q.reader = 0;
1192         crq->item = cache_get(h);
1193         crq->buf = buf;
1194         crq->len = 0;
1195         crq->readers = 0;
1196         spin_lock(&queue_lock);
1197         if (test_bit(CACHE_PENDING, &h->flags))
1198                 list_add_tail(&crq->q.list, &detail->queue);
1199         else
1200                 /* Lost a race, no longer PENDING, so don't enqueue */
1201                 ret = -EAGAIN;
1202         spin_unlock(&queue_lock);
1203         wake_up(&queue_wait);
1204         if (ret == -EAGAIN) {
1205                 kfree(buf);
1206                 kfree(crq);
1207         }
1208         return ret;
1209 }
1210 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1211
1212 /*
1213  * parse a message from user-space and pass it
1214  * to an appropriate cache
1215  * Messages are, like requests, separated into fields by
1216  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1217  *
1218  * Message is
1219  *   reply cachename expiry key ... content....
1220  *
1221  * key and content are both parsed by cache
1222  */
1223
1224 int qword_get(char **bpp, char *dest, int bufsize)
1225 {
1226         /* return bytes copied, or -1 on error */
1227         char *bp = *bpp;
1228         int len = 0;
1229
1230         while (*bp == ' ') bp++;
1231
1232         if (bp[0] == '\\' && bp[1] == 'x') {
1233                 /* HEX STRING */
1234                 bp += 2;
1235                 while (len < bufsize) {
1236                         int h, l;
1237
1238                         h = hex_to_bin(bp[0]);
1239                         if (h < 0)
1240                                 break;
1241
1242                         l = hex_to_bin(bp[1]);
1243                         if (l < 0)
1244                                 break;
1245
1246                         *dest++ = (h << 4) | l;
1247                         bp += 2;
1248                         len++;
1249                 }
1250         } else {
1251                 /* text with \nnn octal quoting */
1252                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1253                         if (*bp == '\\' &&
1254                             isodigit(bp[1]) && (bp[1] <= '3') &&
1255                             isodigit(bp[2]) &&
1256                             isodigit(bp[3])) {
1257                                 int byte = (*++bp -'0');
1258                                 bp++;
1259                                 byte = (byte << 3) | (*bp++ - '0');
1260                                 byte = (byte << 3) | (*bp++ - '0');
1261                                 *dest++ = byte;
1262                                 len++;
1263                         } else {
1264                                 *dest++ = *bp++;
1265                                 len++;
1266                         }
1267                 }
1268         }
1269
1270         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1271                 return -1;
1272         while (*bp == ' ') bp++;
1273         *bpp = bp;
1274         *dest = '\0';
1275         return len;
1276 }
1277 EXPORT_SYMBOL_GPL(qword_get);
1278
1279
1280 /*
1281  * support /proc/sunrpc/cache/$CACHENAME/content
1282  * as a seqfile.
1283  * We call ->cache_show passing NULL for the item to
1284  * get a header, then pass each real item in the cache
1285  */
1286
1287 struct handle {
1288         struct cache_detail *cd;
1289 };
1290
1291 static void *c_start(struct seq_file *m, loff_t *pos)
1292         __acquires(cd->hash_lock)
1293 {
1294         loff_t n = *pos;
1295         unsigned int hash, entry;
1296         struct cache_head *ch;
1297         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1298
1299
1300         read_lock(&cd->hash_lock);
1301         if (!n--)
1302                 return SEQ_START_TOKEN;
1303         hash = n >> 32;
1304         entry = n & ((1LL<<32) - 1);
1305
1306         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1307                 if (!entry--)
1308                         return ch;
1309         n &= ~((1LL<<32) - 1);
1310         do {
1311                 hash++;
1312                 n += 1LL<<32;
1313         } while(hash < cd->hash_size &&
1314                 cd->hash_table[hash]==NULL);
1315         if (hash >= cd->hash_size)
1316                 return NULL;
1317         *pos = n+1;
1318         return cd->hash_table[hash];
1319 }
1320
1321 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1322 {
1323         struct cache_head *ch = p;
1324         int hash = (*pos >> 32);
1325         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1326
1327         if (p == SEQ_START_TOKEN)
1328                 hash = 0;
1329         else if (ch->next == NULL) {
1330                 hash++;
1331                 *pos += 1LL<<32;
1332         } else {
1333                 ++*pos;
1334                 return ch->next;
1335         }
1336         *pos &= ~((1LL<<32) - 1);
1337         while (hash < cd->hash_size &&
1338                cd->hash_table[hash] == NULL) {
1339                 hash++;
1340                 *pos += 1LL<<32;
1341         }
1342         if (hash >= cd->hash_size)
1343                 return NULL;
1344         ++*pos;
1345         return cd->hash_table[hash];
1346 }
1347
1348 static void c_stop(struct seq_file *m, void *p)
1349         __releases(cd->hash_lock)
1350 {
1351         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1352         read_unlock(&cd->hash_lock);
1353 }
1354
1355 static int c_show(struct seq_file *m, void *p)
1356 {
1357         struct cache_head *cp = p;
1358         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1359
1360         if (p == SEQ_START_TOKEN)
1361                 return cd->cache_show(m, cd, NULL);
1362
1363         ifdebug(CACHE)
1364                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1365                            convert_to_wallclock(cp->expiry_time),
1366                            atomic_read(&cp->ref.refcount), cp->flags);
1367         cache_get(cp);
1368         if (cache_check(cd, cp, NULL))
1369                 /* cache_check does a cache_put on failure */
1370                 seq_printf(m, "# ");
1371         else {
1372                 if (cache_is_expired(cd, cp))
1373                         seq_printf(m, "# ");
1374                 cache_put(cp, cd);
1375         }
1376
1377         return cd->cache_show(m, cd, cp);
1378 }
1379
1380 static const struct seq_operations cache_content_op = {
1381         .start  = c_start,
1382         .next   = c_next,
1383         .stop   = c_stop,
1384         .show   = c_show,
1385 };
1386
1387 static int content_open(struct inode *inode, struct file *file,
1388                         struct cache_detail *cd)
1389 {
1390         struct handle *han;
1391
1392         if (!cd || !try_module_get(cd->owner))
1393                 return -EACCES;
1394         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1395         if (han == NULL) {
1396                 module_put(cd->owner);
1397                 return -ENOMEM;
1398         }
1399
1400         han->cd = cd;
1401         return 0;
1402 }
1403
1404 static int content_release(struct inode *inode, struct file *file,
1405                 struct cache_detail *cd)
1406 {
1407         int ret = seq_release_private(inode, file);
1408         module_put(cd->owner);
1409         return ret;
1410 }
1411
1412 static int open_flush(struct inode *inode, struct file *file,
1413                         struct cache_detail *cd)
1414 {
1415         if (!cd || !try_module_get(cd->owner))
1416                 return -EACCES;
1417         return nonseekable_open(inode, file);
1418 }
1419
1420 static int release_flush(struct inode *inode, struct file *file,
1421                         struct cache_detail *cd)
1422 {
1423         module_put(cd->owner);
1424         return 0;
1425 }
1426
1427 static ssize_t read_flush(struct file *file, char __user *buf,
1428                           size_t count, loff_t *ppos,
1429                           struct cache_detail *cd)
1430 {
1431         char tbuf[22];
1432         unsigned long p = *ppos;
1433         size_t len;
1434
1435         snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1436         len = strlen(tbuf);
1437         if (p >= len)
1438                 return 0;
1439         len -= p;
1440         if (len > count)
1441                 len = count;
1442         if (copy_to_user(buf, (void*)(tbuf+p), len))
1443                 return -EFAULT;
1444         *ppos += len;
1445         return len;
1446 }
1447
1448 static ssize_t write_flush(struct file *file, const char __user *buf,
1449                            size_t count, loff_t *ppos,
1450                            struct cache_detail *cd)
1451 {
1452         char tbuf[20];
1453         char *bp, *ep;
1454
1455         if (*ppos || count > sizeof(tbuf)-1)
1456                 return -EINVAL;
1457         if (copy_from_user(tbuf, buf, count))
1458                 return -EFAULT;
1459         tbuf[count] = 0;
1460         simple_strtoul(tbuf, &ep, 0);
1461         if (*ep && *ep != '\n')
1462                 return -EINVAL;
1463
1464         bp = tbuf;
1465         cd->flush_time = get_expiry(&bp);
1466         cd->nextcheck = seconds_since_boot();
1467         cache_flush();
1468
1469         *ppos += count;
1470         return count;
1471 }
1472
1473 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1474                                  size_t count, loff_t *ppos)
1475 {
1476         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1477
1478         return cache_read(filp, buf, count, ppos, cd);
1479 }
1480
1481 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1482                                   size_t count, loff_t *ppos)
1483 {
1484         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1485
1486         return cache_write(filp, buf, count, ppos, cd);
1487 }
1488
1489 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1490 {
1491         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1492
1493         return cache_poll(filp, wait, cd);
1494 }
1495
1496 static long cache_ioctl_procfs(struct file *filp,
1497                                unsigned int cmd, unsigned long arg)
1498 {
1499         struct inode *inode = file_inode(filp);
1500         struct cache_detail *cd = PDE_DATA(inode);
1501
1502         return cache_ioctl(inode, filp, cmd, arg, cd);
1503 }
1504
1505 static int cache_open_procfs(struct inode *inode, struct file *filp)
1506 {
1507         struct cache_detail *cd = PDE_DATA(inode);
1508
1509         return cache_open(inode, filp, cd);
1510 }
1511
1512 static int cache_release_procfs(struct inode *inode, struct file *filp)
1513 {
1514         struct cache_detail *cd = PDE_DATA(inode);
1515
1516         return cache_release(inode, filp, cd);
1517 }
1518
1519 static const struct file_operations cache_file_operations_procfs = {
1520         .owner          = THIS_MODULE,
1521         .llseek         = no_llseek,
1522         .read           = cache_read_procfs,
1523         .write          = cache_write_procfs,
1524         .poll           = cache_poll_procfs,
1525         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1526         .open           = cache_open_procfs,
1527         .release        = cache_release_procfs,
1528 };
1529
1530 static int content_open_procfs(struct inode *inode, struct file *filp)
1531 {
1532         struct cache_detail *cd = PDE_DATA(inode);
1533
1534         return content_open(inode, filp, cd);
1535 }
1536
1537 static int content_release_procfs(struct inode *inode, struct file *filp)
1538 {
1539         struct cache_detail *cd = PDE_DATA(inode);
1540
1541         return content_release(inode, filp, cd);
1542 }
1543
1544 static const struct file_operations content_file_operations_procfs = {
1545         .open           = content_open_procfs,
1546         .read           = seq_read,
1547         .llseek         = seq_lseek,
1548         .release        = content_release_procfs,
1549 };
1550
1551 static int open_flush_procfs(struct inode *inode, struct file *filp)
1552 {
1553         struct cache_detail *cd = PDE_DATA(inode);
1554
1555         return open_flush(inode, filp, cd);
1556 }
1557
1558 static int release_flush_procfs(struct inode *inode, struct file *filp)
1559 {
1560         struct cache_detail *cd = PDE_DATA(inode);
1561
1562         return release_flush(inode, filp, cd);
1563 }
1564
1565 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1566                             size_t count, loff_t *ppos)
1567 {
1568         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1569
1570         return read_flush(filp, buf, count, ppos, cd);
1571 }
1572
1573 static ssize_t write_flush_procfs(struct file *filp,
1574                                   const char __user *buf,
1575                                   size_t count, loff_t *ppos)
1576 {
1577         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1578
1579         return write_flush(filp, buf, count, ppos, cd);
1580 }
1581
1582 static const struct file_operations cache_flush_operations_procfs = {
1583         .open           = open_flush_procfs,
1584         .read           = read_flush_procfs,
1585         .write          = write_flush_procfs,
1586         .release        = release_flush_procfs,
1587         .llseek         = no_llseek,
1588 };
1589
1590 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1591 {
1592         struct sunrpc_net *sn;
1593
1594         if (cd->u.procfs.proc_ent == NULL)
1595                 return;
1596         if (cd->u.procfs.flush_ent)
1597                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1598         if (cd->u.procfs.channel_ent)
1599                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1600         if (cd->u.procfs.content_ent)
1601                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1602         cd->u.procfs.proc_ent = NULL;
1603         sn = net_generic(net, sunrpc_net_id);
1604         remove_proc_entry(cd->name, sn->proc_net_rpc);
1605 }
1606
1607 #ifdef CONFIG_PROC_FS
1608 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1609 {
1610         struct proc_dir_entry *p;
1611         struct sunrpc_net *sn;
1612
1613         sn = net_generic(net, sunrpc_net_id);
1614         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1615         if (cd->u.procfs.proc_ent == NULL)
1616                 goto out_nomem;
1617         cd->u.procfs.channel_ent = NULL;
1618         cd->u.procfs.content_ent = NULL;
1619
1620         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1621                              cd->u.procfs.proc_ent,
1622                              &cache_flush_operations_procfs, cd);
1623         cd->u.procfs.flush_ent = p;
1624         if (p == NULL)
1625                 goto out_nomem;
1626
1627         if (cd->cache_request || cd->cache_parse) {
1628                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1629                                      cd->u.procfs.proc_ent,
1630                                      &cache_file_operations_procfs, cd);
1631                 cd->u.procfs.channel_ent = p;
1632                 if (p == NULL)
1633                         goto out_nomem;
1634         }
1635         if (cd->cache_show) {
1636                 p = proc_create_data("content", S_IFREG|S_IRUSR,
1637                                 cd->u.procfs.proc_ent,
1638                                 &content_file_operations_procfs, cd);
1639                 cd->u.procfs.content_ent = p;
1640                 if (p == NULL)
1641                         goto out_nomem;
1642         }
1643         return 0;
1644 out_nomem:
1645         remove_cache_proc_entries(cd, net);
1646         return -ENOMEM;
1647 }
1648 #else /* CONFIG_PROC_FS */
1649 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1650 {
1651         return 0;
1652 }
1653 #endif
1654
1655 void __init cache_initialize(void)
1656 {
1657         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1658 }
1659
1660 int cache_register_net(struct cache_detail *cd, struct net *net)
1661 {
1662         int ret;
1663
1664         sunrpc_init_cache_detail(cd);
1665         ret = create_cache_proc_entries(cd, net);
1666         if (ret)
1667                 sunrpc_destroy_cache_detail(cd);
1668         return ret;
1669 }
1670 EXPORT_SYMBOL_GPL(cache_register_net);
1671
1672 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1673 {
1674         remove_cache_proc_entries(cd, net);
1675         sunrpc_destroy_cache_detail(cd);
1676 }
1677 EXPORT_SYMBOL_GPL(cache_unregister_net);
1678
1679 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1680 {
1681         struct cache_detail *cd;
1682
1683         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1684         if (cd == NULL)
1685                 return ERR_PTR(-ENOMEM);
1686
1687         cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1688                                  GFP_KERNEL);
1689         if (cd->hash_table == NULL) {
1690                 kfree(cd);
1691                 return ERR_PTR(-ENOMEM);
1692         }
1693         cd->net = net;
1694         return cd;
1695 }
1696 EXPORT_SYMBOL_GPL(cache_create_net);
1697
1698 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1699 {
1700         kfree(cd->hash_table);
1701         kfree(cd);
1702 }
1703 EXPORT_SYMBOL_GPL(cache_destroy_net);
1704
1705 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1706                                  size_t count, loff_t *ppos)
1707 {
1708         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1709
1710         return cache_read(filp, buf, count, ppos, cd);
1711 }
1712
1713 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1714                                   size_t count, loff_t *ppos)
1715 {
1716         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1717
1718         return cache_write(filp, buf, count, ppos, cd);
1719 }
1720
1721 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1722 {
1723         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1724
1725         return cache_poll(filp, wait, cd);
1726 }
1727
1728 static long cache_ioctl_pipefs(struct file *filp,
1729                               unsigned int cmd, unsigned long arg)
1730 {
1731         struct inode *inode = file_inode(filp);
1732         struct cache_detail *cd = RPC_I(inode)->private;
1733
1734         return cache_ioctl(inode, filp, cmd, arg, cd);
1735 }
1736
1737 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1738 {
1739         struct cache_detail *cd = RPC_I(inode)->private;
1740
1741         return cache_open(inode, filp, cd);
1742 }
1743
1744 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1745 {
1746         struct cache_detail *cd = RPC_I(inode)->private;
1747
1748         return cache_release(inode, filp, cd);
1749 }
1750
1751 const struct file_operations cache_file_operations_pipefs = {
1752         .owner          = THIS_MODULE,
1753         .llseek         = no_llseek,
1754         .read           = cache_read_pipefs,
1755         .write          = cache_write_pipefs,
1756         .poll           = cache_poll_pipefs,
1757         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1758         .open           = cache_open_pipefs,
1759         .release        = cache_release_pipefs,
1760 };
1761
1762 static int content_open_pipefs(struct inode *inode, struct file *filp)
1763 {
1764         struct cache_detail *cd = RPC_I(inode)->private;
1765
1766         return content_open(inode, filp, cd);
1767 }
1768
1769 static int content_release_pipefs(struct inode *inode, struct file *filp)
1770 {
1771         struct cache_detail *cd = RPC_I(inode)->private;
1772
1773         return content_release(inode, filp, cd);
1774 }
1775
1776 const struct file_operations content_file_operations_pipefs = {
1777         .open           = content_open_pipefs,
1778         .read           = seq_read,
1779         .llseek         = seq_lseek,
1780         .release        = content_release_pipefs,
1781 };
1782
1783 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1784 {
1785         struct cache_detail *cd = RPC_I(inode)->private;
1786
1787         return open_flush(inode, filp, cd);
1788 }
1789
1790 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1791 {
1792         struct cache_detail *cd = RPC_I(inode)->private;
1793
1794         return release_flush(inode, filp, cd);
1795 }
1796
1797 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1798                             size_t count, loff_t *ppos)
1799 {
1800         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1801
1802         return read_flush(filp, buf, count, ppos, cd);
1803 }
1804
1805 static ssize_t write_flush_pipefs(struct file *filp,
1806                                   const char __user *buf,
1807                                   size_t count, loff_t *ppos)
1808 {
1809         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1810
1811         return write_flush(filp, buf, count, ppos, cd);
1812 }
1813
1814 const struct file_operations cache_flush_operations_pipefs = {
1815         .open           = open_flush_pipefs,
1816         .read           = read_flush_pipefs,
1817         .write          = write_flush_pipefs,
1818         .release        = release_flush_pipefs,
1819         .llseek         = no_llseek,
1820 };
1821
1822 int sunrpc_cache_register_pipefs(struct dentry *parent,
1823                                  const char *name, umode_t umode,
1824                                  struct cache_detail *cd)
1825 {
1826         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1827         if (IS_ERR(dir))
1828                 return PTR_ERR(dir);
1829         cd->u.pipefs.dir = dir;
1830         return 0;
1831 }
1832 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1833
1834 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1835 {
1836         rpc_remove_cache_dir(cd->u.pipefs.dir);
1837         cd->u.pipefs.dir = NULL;
1838 }
1839 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1840