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