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