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