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 <linux/smp_lock.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 = get_seconds();
  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 < get_seconds()) ||
  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 = get_seconds();
 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 = get_seconds() - 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, get_seconds()+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 DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
 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 > get_seconds())
 393                         current_index = current_detail->hash_size;
 394                 else {
 395                         current_index = 0;
 396                         current_detail->nextcheck = get_seconds()+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 = get_seconds();
 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 list_head cache_defer_hash[DFR_HASHSIZE];
 511 static int cache_defer_cnt;
 512
 513 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 514 {
 515         struct cache_deferred_req *dreq, *discard;
 516         int hash = DFR_HASH(item);
 517
 518         if (cache_defer_cnt >= DFR_MAX) {
 519                 /* too much in the cache, randomly drop this one,
 520                  * or continue and drop the oldest below
 521                  */
 522                 if (net_random()&1)
 523                         return -ENOMEM;
 524         }
 525         dreq = req->defer(req);
 526         if (dreq == NULL)
 527                 return -ENOMEM;
 528
 529         dreq->item = item;
 530
 531         spin_lock(&cache_defer_lock);
 532
 533         list_add(&dreq->recent, &cache_defer_list);
 534
 535         if (cache_defer_hash[hash].next == NULL)
 536                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 537         list_add(&dreq->hash, &cache_defer_hash[hash]);
 538
 539         /* it is in, now maybe clean up */
 540         discard = NULL;
 541         if (++cache_defer_cnt > DFR_MAX) {
 542                 discard = list_entry(cache_defer_list.prev,
 543                                      struct cache_deferred_req, recent);
 544                 list_del_init(&discard->recent);
 545                 list_del_init(&discard->hash);
 546                 cache_defer_cnt--;
 547         }
 548         spin_unlock(&cache_defer_lock);
 549
 550         if (discard)
 551                 /* there was one too many */
 552                 discard->revisit(discard, 1);
 553
 554         if (!test_bit(CACHE_PENDING, &item->flags)) {
 555                 /* must have just been validated... */
 556                 cache_revisit_request(item);
 557                 return -EAGAIN;
 558         }
 559         return 0;
 560 }
 561
 562 static void cache_revisit_request(struct cache_head *item)
 563 {
 564         struct cache_deferred_req *dreq;
 565         struct list_head pending;
 566
 567         struct list_head *lp;
 568         int hash = DFR_HASH(item);
 569
 570         INIT_LIST_HEAD(&pending);
 571         spin_lock(&cache_defer_lock);
 572
 573         lp = cache_defer_hash[hash].next;
 574         if (lp) {
 575                 while (lp != &cache_defer_hash[hash]) {
 576                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 577                         lp = lp->next;
 578                         if (dreq->item == item) {
 579                                 list_del_init(&dreq->hash);
 580                                 list_move(&dreq->recent, &pending);
 581                                 cache_defer_cnt--;
 582                         }
 583                 }
 584         }
 585         spin_unlock(&cache_defer_lock);
 586
 587         while (!list_empty(&pending)) {
 588                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 589                 list_del_init(&dreq->recent);
 590                 dreq->revisit(dreq, 0);
 591         }
 592 }
 593
 594 void cache_clean_deferred(void *owner)
 595 {
 596         struct cache_deferred_req *dreq, *tmp;
 597         struct list_head pending;
 598
 599
 600         INIT_LIST_HEAD(&pending);
 601         spin_lock(&cache_defer_lock);
 602
 603         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 604                 if (dreq->owner == owner) {
 605                         list_del_init(&dreq->hash);
 606                         list_move(&dreq->recent, &pending);
 607                         cache_defer_cnt--;
 608                 }
 609         }
 610         spin_unlock(&cache_defer_lock);
 611
 612         while (!list_empty(&pending)) {
 613                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 614                 list_del_init(&dreq->recent);
 615                 dreq->revisit(dreq, 1);
 616         }
 617 }
 618
 619 /*
 620  * communicate with user-space
 621  *
 622  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 623  * On read, you get a full request, or block.
 624  * On write, an update request is processed.
 625  * Poll works if anything to read, and always allows write.
 626  *
 627  * Implemented by linked list of requests.  Each open file has
 628  * a ->private that also exists in this list.  New requests are added
 629  * to the end and may wakeup and preceding readers.
 630  * New readers are added to the head.  If, on read, an item is found with
 631  * CACHE_UPCALLING clear, we free it from the list.
 632  *
 633  */
 634
 635 static DEFINE_SPINLOCK(queue_lock);
 636 static DEFINE_MUTEX(queue_io_mutex);
 637
 638 struct cache_queue {
 639         struct list_head        list;
 640         int                     reader; /* if 0, then request */
 641 };
 642 struct cache_request {
 643         struct cache_queue      q;
 644         struct cache_head       *item;
 645         char                    * buf;
 646         int                     len;
 647         int                     readers;
 648 };
 649 struct cache_reader {
 650         struct cache_queue      q;
 651         int                     offset; /* if non-0, we have a refcnt on next request */
 652 };
 653
 654 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 655                           loff_t *ppos, struct cache_detail *cd)
 656 {
 657         struct cache_reader *rp = filp->private_data;
 658         struct cache_request *rq;
 659         struct inode *inode = filp->f_path.dentry->d_inode;
 660         int err;
 661
 662         if (count == 0)
 663                 return 0;
 664
 665         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 666                               * readers on this file */
 667  again:
 668         spin_lock(&queue_lock);
 669         /* need to find next request */
 670         while (rp->q.list.next != &cd->queue &&
 671                list_entry(rp->q.list.next, struct cache_queue, list)
 672                ->reader) {
 673                 struct list_head *next = rp->q.list.next;
 674                 list_move(&rp->q.list, next);
 675         }
 676         if (rp->q.list.next == &cd->queue) {
 677                 spin_unlock(&queue_lock);
 678                 mutex_unlock(&inode->i_mutex);
 679                 BUG_ON(rp->offset);
 680                 return 0;
 681         }
 682         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 683         BUG_ON(rq->q.reader);
 684         if (rp->offset == 0)
 685                 rq->readers++;
 686         spin_unlock(&queue_lock);
 687
 688         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 689                 err = -EAGAIN;
 690                 spin_lock(&queue_lock);
 691                 list_move(&rp->q.list, &rq->q.list);
 692                 spin_unlock(&queue_lock);
 693         } else {
 694                 if (rp->offset + count > rq->len)
 695                         count = rq->len - rp->offset;
 696                 err = -EFAULT;
 697                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 698                         goto out;
 699                 rp->offset += count;
 700                 if (rp->offset >= rq->len) {
 701                         rp->offset = 0;
 702                         spin_lock(&queue_lock);
 703                         list_move(&rp->q.list, &rq->q.list);
 704                         spin_unlock(&queue_lock);
 705                 }
 706                 err = 0;
 707         }
 708  out:
 709         if (rp->offset == 0) {
 710                 /* need to release rq */
 711                 spin_lock(&queue_lock);
 712                 rq->readers--;
 713                 if (rq->readers == 0 &&
 714                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 715                         list_del(&rq->q.list);
 716                         spin_unlock(&queue_lock);
 717                         cache_put(rq->item, cd);
 718                         kfree(rq->buf);
 719                         kfree(rq);
 720                 } else
 721                         spin_unlock(&queue_lock);
 722         }
 723         if (err == -EAGAIN)
 724                 goto again;
 725         mutex_unlock(&inode->i_mutex);
 726         return err ? err :  count;
 727 }
 728
 729 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 730                                  size_t count, struct cache_detail *cd)
 731 {
 732         ssize_t ret;
 733
 734         if (copy_from_user(kaddr, buf, count))
 735                 return -EFAULT;
 736         kaddr[count] = '\0';
 737         ret = cd->cache_parse(cd, kaddr, count);
 738         if (!ret)
 739                 ret = count;
 740         return ret;
 741 }
 742
 743 static ssize_t cache_slow_downcall(const char __user *buf,
 744                                    size_t count, struct cache_detail *cd)
 745 {
 746         static char write_buf[8192]; /* protected by queue_io_mutex */
 747         ssize_t ret = -EINVAL;
 748
 749         if (count >= sizeof(write_buf))
 750                 goto out;
 751         mutex_lock(&queue_io_mutex);
 752         ret = cache_do_downcall(write_buf, buf, count, cd);
 753         mutex_unlock(&queue_io_mutex);
 754 out:
 755         return ret;
 756 }
 757
 758 static ssize_t cache_downcall(struct address_space *mapping,
 759                               const char __user *buf,
 760                               size_t count, struct cache_detail *cd)
 761 {
 762         struct page *page;
 763         char *kaddr;
 764         ssize_t ret = -ENOMEM;
 765
 766         if (count >= PAGE_CACHE_SIZE)
 767                 goto out_slow;
 768
 769         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 770         if (!page)
 771                 goto out_slow;
 772
 773         kaddr = kmap(page);
 774         ret = cache_do_downcall(kaddr, buf, count, cd);
 775         kunmap(page);
 776         unlock_page(page);
 777         page_cache_release(page);
 778         return ret;
 779 out_slow:
 780         return cache_slow_downcall(buf, count, cd);
 781 }
 782
 783 static ssize_t cache_write(struct file *filp, const char __user *buf,
 784                            size_t count, loff_t *ppos,
 785                            struct cache_detail *cd)
 786 {
 787         struct address_space *mapping = filp->f_mapping;
 788         struct inode *inode = filp->f_path.dentry->d_inode;
 789         ssize_t ret = -EINVAL;
 790
 791         if (!cd->cache_parse)
 792                 goto out;
 793
 794         mutex_lock(&inode->i_mutex);
 795         ret = cache_downcall(mapping, buf, count, cd);
 796         mutex_unlock(&inode->i_mutex);
 797 out:
 798         return ret;
 799 }
 800
 801 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 802
 803 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 804                                struct cache_detail *cd)
 805 {
 806         unsigned int mask;
 807         struct cache_reader *rp = filp->private_data;
 808         struct cache_queue *cq;
 809
 810         poll_wait(filp, &queue_wait, wait);
 811
 812         /* alway allow write */
 813         mask = POLL_OUT | POLLWRNORM;
 814
 815         if (!rp)
 816                 return mask;
 817
 818         spin_lock(&queue_lock);
 819
 820         for (cq= &rp->q; &cq->list != &cd->queue;
 821              cq = list_entry(cq->list.next, struct cache_queue, list))
 822                 if (!cq->reader) {
 823                         mask |= POLLIN | POLLRDNORM;
 824                         break;
 825                 }
 826         spin_unlock(&queue_lock);
 827         return mask;
 828 }
 829
 830 static int cache_ioctl(struct inode *ino, struct file *filp,
 831                        unsigned int cmd, unsigned long arg,
 832                        struct cache_detail *cd)
 833 {
 834         int len = 0;
 835         struct cache_reader *rp = filp->private_data;
 836         struct cache_queue *cq;
 837
 838         if (cmd != FIONREAD || !rp)
 839                 return -EINVAL;
 840
 841         spin_lock(&queue_lock);
 842
 843         /* only find the length remaining in current request,
 844          * or the length of the next request
 845          */
 846         for (cq= &rp->q; &cq->list != &cd->queue;
 847              cq = list_entry(cq->list.next, struct cache_queue, list))
 848                 if (!cq->reader) {
 849                         struct cache_request *cr =
 850                                 container_of(cq, struct cache_request, q);
 851                         len = cr->len - rp->offset;
 852                         break;
 853                 }
 854         spin_unlock(&queue_lock);
 855
 856         return put_user(len, (int __user *)arg);
 857 }
 858
 859 static int cache_open(struct inode *inode, struct file *filp,
 860                       struct cache_detail *cd)
 861 {
 862         struct cache_reader *rp = NULL;
 863
 864         if (!cd || !try_module_get(cd->owner))
 865                 return -EACCES;
 866         nonseekable_open(inode, filp);
 867         if (filp->f_mode & FMODE_READ) {
 868                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 869                 if (!rp)
 870                         return -ENOMEM;
 871                 rp->offset = 0;
 872                 rp->q.reader = 1;
 873                 atomic_inc(&cd->readers);
 874                 spin_lock(&queue_lock);
 875                 list_add(&rp->q.list, &cd->queue);
 876                 spin_unlock(&queue_lock);
 877         }
 878         filp->private_data = rp;
 879         return 0;
 880 }
 881
 882 static int cache_release(struct inode *inode, struct file *filp,
 883                          struct cache_detail *cd)
 884 {
 885         struct cache_reader *rp = filp->private_data;
 886
 887         if (rp) {
 888                 spin_lock(&queue_lock);
 889                 if (rp->offset) {
 890                         struct cache_queue *cq;
 891                         for (cq= &rp->q; &cq->list != &cd->queue;
 892                              cq = list_entry(cq->list.next, struct cache_queue, list))
 893                                 if (!cq->reader) {
 894                                         container_of(cq, struct cache_request, q)
 895                                                 ->readers--;
 896                                         break;
 897                                 }
 898                         rp->offset = 0;
 899                 }
 900                 list_del(&rp->q.list);
 901                 spin_unlock(&queue_lock);
 902
 903                 filp->private_data = NULL;
 904                 kfree(rp);
 905
 906                 cd->last_close = get_seconds();
 907                 atomic_dec(&cd->readers);
 908         }
 909         module_put(cd->owner);
 910         return 0;
 911 }
 912
 913
 914
 915 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
 916 {
 917         struct cache_queue *cq;
 918         spin_lock(&queue_lock);
 919         list_for_each_entry(cq, &detail->queue, list)
 920                 if (!cq->reader) {
 921                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 922                         if (cr->item != ch)
 923                                 continue;
 924                         if (cr->readers != 0)
 925                                 continue;
 926                         list_del(&cr->q.list);
 927                         spin_unlock(&queue_lock);
 928                         cache_put(cr->item, detail);
 929                         kfree(cr->buf);
 930                         kfree(cr);
 931                         return;
 932                 }
 933         spin_unlock(&queue_lock);
 934 }
 935
 936 /*
 937  * Support routines for text-based upcalls.
 938  * Fields are separated by spaces.
 939  * Fields are either mangled to quote space tab newline slosh with slosh
 940  * or a hexified with a leading \x
 941  * Record is terminated with newline.
 942  *
 943  */
 944
 945 void qword_add(char **bpp, int *lp, char *str)
 946 {
 947         char *bp = *bpp;
 948         int len = *lp;
 949         char c;
 950
 951         if (len < 0) return;
 952
 953         while ((c=*str++) && len)
 954                 switch(c) {
 955                 case ' ':
 956                 case '\t':
 957                 case '\n':
 958                 case '\\':
 959                         if (len >= 4) {
 960                                 *bp++ = '\\';
 961                                 *bp++ = '0' + ((c & 0300)>>6);
 962                                 *bp++ = '0' + ((c & 0070)>>3);
 963                                 *bp++ = '0' + ((c & 0007)>>0);
 964                         }
 965                         len -= 4;
 966                         break;
 967                 default:
 968                         *bp++ = c;
 969                         len--;
 970                 }
 971         if (c || len <1) len = -1;
 972         else {
 973                 *bp++ = ' ';
 974                 len--;
 975         }
 976         *bpp = bp;
 977         *lp = len;
 978 }
 979 EXPORT_SYMBOL_GPL(qword_add);
 980
 981 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
 982 {
 983         char *bp = *bpp;
 984         int len = *lp;
 985
 986         if (len < 0) return;
 987
 988         if (len > 2) {
 989                 *bp++ = '\\';
 990                 *bp++ = 'x';
 991                 len -= 2;
 992                 while (blen && len >= 2) {
 993                         unsigned char c = *buf++;
 994                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
 995                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
 996                         len -= 2;
 997                         blen--;
 998                 }
 999         }
1000         if (blen || len<1) len = -1;
1001         else {
1002                 *bp++ = ' ';
1003                 len--;
1004         }
1005         *bpp = bp;
1006         *lp = len;
1007 }
1008 EXPORT_SYMBOL_GPL(qword_addhex);
1009
1010 static void warn_no_listener(struct cache_detail *detail)
1011 {
1012         if (detail->last_warn != detail->last_close) {
1013                 detail->last_warn = detail->last_close;
1014                 if (detail->warn_no_listener)
1015                         detail->warn_no_listener(detail, detail->last_close != 0);
1016         }
1017 }
1018
1019 /*
1020  * register an upcall request to user-space and queue it up for read() by the
1021  * upcall daemon.
1022  *
1023  * Each request is at most one page long.
1024  */
1025 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1026                 void (*cache_request)(struct cache_detail *,
1027                                       struct cache_head *,
1028                                       char **,
1029                                       int *))
1030 {
1031
1032         char *buf;
1033         struct cache_request *crq;
1034         char *bp;
1035         int len;
1036
1037         if (atomic_read(&detail->readers) == 0 &&
1038             detail->last_close < get_seconds() - 30) {
1039                         warn_no_listener(detail);
1040                         return -EINVAL;
1041         }
1042
1043         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1044         if (!buf)
1045                 return -EAGAIN;
1046
1047         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1048         if (!crq) {
1049                 kfree(buf);
1050                 return -EAGAIN;
1051         }
1052
1053         bp = buf; len = PAGE_SIZE;
1054
1055         cache_request(detail, h, &bp, &len);
1056
1057         if (len < 0) {
1058                 kfree(buf);
1059                 kfree(crq);
1060                 return -EAGAIN;
1061         }
1062         crq->q.reader = 0;
1063         crq->item = cache_get(h);
1064         crq->buf = buf;
1065         crq->len = PAGE_SIZE - len;
1066         crq->readers = 0;
1067         spin_lock(&queue_lock);
1068         list_add_tail(&crq->q.list, &detail->queue);
1069         spin_unlock(&queue_lock);
1070         wake_up(&queue_wait);
1071         return 0;
1072 }
1073 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1074
1075 /*
1076  * parse a message from user-space and pass it
1077  * to an appropriate cache
1078  * Messages are, like requests, separated into fields by
1079  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1080  *
1081  * Message is
1082  *   reply cachename expiry key ... content....
1083  *
1084  * key and content are both parsed by cache
1085  */
1086
1087 #define isodigit(c) (isdigit(c) && c <= '7')
1088 int qword_get(char **bpp, char *dest, int bufsize)
1089 {
1090         /* return bytes copied, or -1 on error */
1091         char *bp = *bpp;
1092         int len = 0;
1093
1094         while (*bp == ' ') bp++;
1095
1096         if (bp[0] == '\\' && bp[1] == 'x') {
1097                 /* HEX STRING */
1098                 bp += 2;
1099                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1100                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1101                         bp++;
1102                         byte <<= 4;
1103                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1104                         *dest++ = byte;
1105                         bp++;
1106                         len++;
1107                 }
1108         } else {
1109                 /* text with \nnn octal quoting */
1110                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1111                         if (*bp == '\\' &&
1112                             isodigit(bp[1]) && (bp[1] <= '3') &&
1113                             isodigit(bp[2]) &&
1114                             isodigit(bp[3])) {
1115                                 int byte = (*++bp -'0');
1116                                 bp++;
1117                                 byte = (byte << 3) | (*bp++ - '0');
1118                                 byte = (byte << 3) | (*bp++ - '0');
1119                                 *dest++ = byte;
1120                                 len++;
1121                         } else {
1122                                 *dest++ = *bp++;
1123                                 len++;
1124                         }
1125                 }
1126         }
1127
1128         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1129                 return -1;
1130         while (*bp == ' ') bp++;
1131         *bpp = bp;
1132         *dest = '\0';
1133         return len;
1134 }
1135 EXPORT_SYMBOL_GPL(qword_get);
1136
1137
1138 /*
1139  * support /proc/sunrpc/cache/$CACHENAME/content
1140  * as a seqfile.
1141  * We call ->cache_show passing NULL for the item to
1142  * get a header, then pass each real item in the cache
1143  */
1144
1145 struct handle {
1146         struct cache_detail *cd;
1147 };
1148
1149 static void *c_start(struct seq_file *m, loff_t *pos)
1150         __acquires(cd->hash_lock)
1151 {
1152         loff_t n = *pos;
1153         unsigned hash, entry;
1154         struct cache_head *ch;
1155         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1156
1157
1158         read_lock(&cd->hash_lock);
1159         if (!n--)
1160                 return SEQ_START_TOKEN;
1161         hash = n >> 32;
1162         entry = n & ((1LL<<32) - 1);
1163
1164         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1165                 if (!entry--)
1166                         return ch;
1167         n &= ~((1LL<<32) - 1);
1168         do {
1169                 hash++;
1170                 n += 1LL<<32;
1171         } while(hash < cd->hash_size &&
1172                 cd->hash_table[hash]==NULL);
1173         if (hash >= cd->hash_size)
1174                 return NULL;
1175         *pos = n+1;
1176         return cd->hash_table[hash];
1177 }
1178
1179 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1180 {
1181         struct cache_head *ch = p;
1182         int hash = (*pos >> 32);
1183         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1184
1185         if (p == SEQ_START_TOKEN)
1186                 hash = 0;
1187         else if (ch->next == NULL) {
1188                 hash++;
1189                 *pos += 1LL<<32;
1190         } else {
1191                 ++*pos;
1192                 return ch->next;
1193         }
1194         *pos &= ~((1LL<<32) - 1);
1195         while (hash < cd->hash_size &&
1196                cd->hash_table[hash] == NULL) {
1197                 hash++;
1198                 *pos += 1LL<<32;
1199         }
1200         if (hash >= cd->hash_size)
1201                 return NULL;
1202         ++*pos;
1203         return cd->hash_table[hash];
1204 }
1205
1206 static void c_stop(struct seq_file *m, void *p)
1207         __releases(cd->hash_lock)
1208 {
1209         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1210         read_unlock(&cd->hash_lock);
1211 }
1212
1213 static int c_show(struct seq_file *m, void *p)
1214 {
1215         struct cache_head *cp = p;
1216         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1217
1218         if (p == SEQ_START_TOKEN)
1219                 return cd->cache_show(m, cd, NULL);
1220
1221         ifdebug(CACHE)
1222                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1223                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1224         cache_get(cp);
1225         if (cache_check(cd, cp, NULL))
1226                 /* cache_check does a cache_put on failure */
1227                 seq_printf(m, "# ");
1228         else
1229                 cache_put(cp, cd);
1230
1231         return cd->cache_show(m, cd, cp);
1232 }
1233
1234 static const struct seq_operations cache_content_op = {
1235         .start  = c_start,
1236         .next   = c_next,
1237         .stop   = c_stop,
1238         .show   = c_show,
1239 };
1240
1241 static int content_open(struct inode *inode, struct file *file,
1242                         struct cache_detail *cd)
1243 {
1244         struct handle *han;
1245
1246         if (!cd || !try_module_get(cd->owner))
1247                 return -EACCES;
1248         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1249         if (han == NULL) {
1250                 module_put(cd->owner);
1251                 return -ENOMEM;
1252         }
1253
1254         han->cd = cd;
1255         return 0;
1256 }
1257
1258 static int content_release(struct inode *inode, struct file *file,
1259                 struct cache_detail *cd)
1260 {
1261         int ret = seq_release_private(inode, file);
1262         module_put(cd->owner);
1263         return ret;
1264 }
1265
1266 static int open_flush(struct inode *inode, struct file *file,
1267                         struct cache_detail *cd)
1268 {
1269         if (!cd || !try_module_get(cd->owner))
1270                 return -EACCES;
1271         return nonseekable_open(inode, file);
1272 }
1273
1274 static int release_flush(struct inode *inode, struct file *file,
1275                         struct cache_detail *cd)
1276 {
1277         module_put(cd->owner);
1278         return 0;
1279 }
1280
1281 static ssize_t read_flush(struct file *file, char __user *buf,
1282                           size_t count, loff_t *ppos,
1283                           struct cache_detail *cd)
1284 {
1285         char tbuf[20];
1286         unsigned long p = *ppos;
1287         size_t len;
1288
1289         sprintf(tbuf, "%lu\n", cd->flush_time);
1290         len = strlen(tbuf);
1291         if (p >= len)
1292                 return 0;
1293         len -= p;
1294         if (len > count)
1295                 len = count;
1296         if (copy_to_user(buf, (void*)(tbuf+p), len))
1297                 return -EFAULT;
1298         *ppos += len;
1299         return len;
1300 }
1301
1302 static ssize_t write_flush(struct file *file, const char __user *buf,
1303                            size_t count, loff_t *ppos,
1304                            struct cache_detail *cd)
1305 {
1306         char tbuf[20];
1307         char *ep;
1308         long flushtime;
1309         if (*ppos || count > sizeof(tbuf)-1)
1310                 return -EINVAL;
1311         if (copy_from_user(tbuf, buf, count))
1312                 return -EFAULT;
1313         tbuf[count] = 0;
1314         flushtime = simple_strtoul(tbuf, &ep, 0);
1315         if (*ep && *ep != '\n')
1316                 return -EINVAL;
1317
1318         cd->flush_time = flushtime;
1319         cd->nextcheck = get_seconds();
1320         cache_flush();
1321
1322         *ppos += count;
1323         return count;
1324 }
1325
1326 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1327                                  size_t count, loff_t *ppos)
1328 {
1329         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1330
1331         return cache_read(filp, buf, count, ppos, cd);
1332 }
1333
1334 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1335                                   size_t count, loff_t *ppos)
1336 {
1337         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1338
1339         return cache_write(filp, buf, count, ppos, cd);
1340 }
1341
1342 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1343 {
1344         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1345
1346         return cache_poll(filp, wait, cd);
1347 }
1348
1349 static long cache_ioctl_procfs(struct file *filp,
1350                                unsigned int cmd, unsigned long arg)
1351 {
1352         long ret;
1353         struct inode *inode = filp->f_path.dentry->d_inode;
1354         struct cache_detail *cd = PDE(inode)->data;
1355
1356         lock_kernel();
1357         ret = cache_ioctl(inode, filp, cmd, arg, cd);
1358         unlock_kernel();
1359
1360         return ret;
1361 }
1362
1363 static int cache_open_procfs(struct inode *inode, struct file *filp)
1364 {
1365         struct cache_detail *cd = PDE(inode)->data;
1366
1367         return cache_open(inode, filp, cd);
1368 }
1369
1370 static int cache_release_procfs(struct inode *inode, struct file *filp)
1371 {
1372         struct cache_detail *cd = PDE(inode)->data;
1373
1374         return cache_release(inode, filp, cd);
1375 }
1376
1377 static const struct file_operations cache_file_operations_procfs = {
1378         .owner          = THIS_MODULE,
1379         .llseek         = no_llseek,
1380         .read           = cache_read_procfs,
1381         .write          = cache_write_procfs,
1382         .poll           = cache_poll_procfs,
1383         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1384         .open           = cache_open_procfs,
1385         .release        = cache_release_procfs,
1386 };
1387
1388 static int content_open_procfs(struct inode *inode, struct file *filp)
1389 {
1390         struct cache_detail *cd = PDE(inode)->data;
1391
1392         return content_open(inode, filp, cd);
1393 }
1394
1395 static int content_release_procfs(struct inode *inode, struct file *filp)
1396 {
1397         struct cache_detail *cd = PDE(inode)->data;
1398
1399         return content_release(inode, filp, cd);
1400 }
1401
1402 static const struct file_operations content_file_operations_procfs = {
1403         .open           = content_open_procfs,
1404         .read           = seq_read,
1405         .llseek         = seq_lseek,
1406         .release        = content_release_procfs,
1407 };
1408
1409 static int open_flush_procfs(struct inode *inode, struct file *filp)
1410 {
1411         struct cache_detail *cd = PDE(inode)->data;
1412
1413         return open_flush(inode, filp, cd);
1414 }
1415
1416 static int release_flush_procfs(struct inode *inode, struct file *filp)
1417 {
1418         struct cache_detail *cd = PDE(inode)->data;
1419
1420         return release_flush(inode, filp, cd);
1421 }
1422
1423 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1424                             size_t count, loff_t *ppos)
1425 {
1426         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1427
1428         return read_flush(filp, buf, count, ppos, cd);
1429 }
1430
1431 static ssize_t write_flush_procfs(struct file *filp,
1432                                   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 write_flush(filp, buf, count, ppos, cd);
1438 }
1439
1440 static const struct file_operations cache_flush_operations_procfs = {
1441         .open           = open_flush_procfs,
1442         .read           = read_flush_procfs,
1443         .write          = write_flush_procfs,
1444         .release        = release_flush_procfs,
1445 };
1446
1447 static void remove_cache_proc_entries(struct cache_detail *cd)
1448 {
1449         if (cd->u.procfs.proc_ent == NULL)
1450                 return;
1451         if (cd->u.procfs.flush_ent)
1452                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1453         if (cd->u.procfs.channel_ent)
1454                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1455         if (cd->u.procfs.content_ent)
1456                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1457         cd->u.procfs.proc_ent = NULL;
1458         remove_proc_entry(cd->name, proc_net_rpc);
1459 }
1460
1461 #ifdef CONFIG_PROC_FS
1462 static int create_cache_proc_entries(struct cache_detail *cd)
1463 {
1464         struct proc_dir_entry *p;
1465
1466         cd->u.procfs.proc_ent = proc_mkdir(cd->name, proc_net_rpc);
1467         if (cd->u.procfs.proc_ent == NULL)
1468                 goto out_nomem;
1469         cd->u.procfs.channel_ent = NULL;
1470         cd->u.procfs.content_ent = NULL;
1471
1472         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1473                              cd->u.procfs.proc_ent,
1474                              &cache_flush_operations_procfs, cd);
1475         cd->u.procfs.flush_ent = p;
1476         if (p == NULL)
1477                 goto out_nomem;
1478
1479         if (cd->cache_upcall || cd->cache_parse) {
1480                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1481                                      cd->u.procfs.proc_ent,
1482                                      &cache_file_operations_procfs, cd);
1483                 cd->u.procfs.channel_ent = p;
1484                 if (p == NULL)
1485                         goto out_nomem;
1486         }
1487         if (cd->cache_show) {
1488                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1489                                 cd->u.procfs.proc_ent,
1490                                 &content_file_operations_procfs, cd);
1491                 cd->u.procfs.content_ent = p;
1492                 if (p == NULL)
1493                         goto out_nomem;
1494         }
1495         return 0;
1496 out_nomem:
1497         remove_cache_proc_entries(cd);
1498         return -ENOMEM;
1499 }
1500 #else /* CONFIG_PROC_FS */
1501 static int create_cache_proc_entries(struct cache_detail *cd)
1502 {
1503         return 0;
1504 }
1505 #endif
1506
1507 int cache_register(struct cache_detail *cd)
1508 {
1509         int ret;
1510
1511         sunrpc_init_cache_detail(cd);
1512         ret = create_cache_proc_entries(cd);
1513         if (ret)
1514                 sunrpc_destroy_cache_detail(cd);
1515         return ret;
1516 }
1517 EXPORT_SYMBOL_GPL(cache_register);
1518
1519 void cache_unregister(struct cache_detail *cd)
1520 {
1521         remove_cache_proc_entries(cd);
1522         sunrpc_destroy_cache_detail(cd);
1523 }
1524 EXPORT_SYMBOL_GPL(cache_unregister);
1525
1526 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1527                                  size_t count, loff_t *ppos)
1528 {
1529         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1530
1531         return cache_read(filp, buf, count, ppos, cd);
1532 }
1533
1534 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1535                                   size_t count, loff_t *ppos)
1536 {
1537         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1538
1539         return cache_write(filp, buf, count, ppos, cd);
1540 }
1541
1542 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1543 {
1544         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1545
1546         return cache_poll(filp, wait, cd);
1547 }
1548
1549 static long cache_ioctl_pipefs(struct file *filp,
1550                               unsigned int cmd, unsigned long arg)
1551 {
1552         struct inode *inode = filp->f_dentry->d_inode;
1553         struct cache_detail *cd = RPC_I(inode)->private;
1554         long ret;
1555
1556         lock_kernel();
1557         ret = cache_ioctl(inode, filp, cmd, arg, cd);
1558         unlock_kernel();
1559
1560         return ret;
1561 }
1562
1563 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1564 {
1565         struct cache_detail *cd = RPC_I(inode)->private;
1566
1567         return cache_open(inode, filp, cd);
1568 }
1569
1570 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1571 {
1572         struct cache_detail *cd = RPC_I(inode)->private;
1573
1574         return cache_release(inode, filp, cd);
1575 }
1576
1577 const struct file_operations cache_file_operations_pipefs = {
1578         .owner          = THIS_MODULE,
1579         .llseek         = no_llseek,
1580         .read           = cache_read_pipefs,
1581         .write          = cache_write_pipefs,
1582         .poll           = cache_poll_pipefs,
1583         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1584         .open           = cache_open_pipefs,
1585         .release        = cache_release_pipefs,
1586 };
1587
1588 static int content_open_pipefs(struct inode *inode, struct file *filp)
1589 {
1590         struct cache_detail *cd = RPC_I(inode)->private;
1591
1592         return content_open(inode, filp, cd);
1593 }
1594
1595 static int content_release_pipefs(struct inode *inode, struct file *filp)
1596 {
1597         struct cache_detail *cd = RPC_I(inode)->private;
1598
1599         return content_release(inode, filp, cd);
1600 }
1601
1602 const struct file_operations content_file_operations_pipefs = {
1603         .open           = content_open_pipefs,
1604         .read           = seq_read,
1605         .llseek         = seq_lseek,
1606         .release        = content_release_pipefs,
1607 };
1608
1609 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1610 {
1611         struct cache_detail *cd = RPC_I(inode)->private;
1612
1613         return open_flush(inode, filp, cd);
1614 }
1615
1616 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1617 {
1618         struct cache_detail *cd = RPC_I(inode)->private;
1619
1620         return release_flush(inode, filp, cd);
1621 }
1622
1623 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1624                             size_t count, loff_t *ppos)
1625 {
1626         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1627
1628         return read_flush(filp, buf, count, ppos, cd);
1629 }
1630
1631 static ssize_t write_flush_pipefs(struct file *filp,
1632                                   const char __user *buf,
1633                                   size_t count, loff_t *ppos)
1634 {
1635         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1636
1637         return write_flush(filp, buf, count, ppos, cd);
1638 }
1639
1640 const struct file_operations cache_flush_operations_pipefs = {
1641         .open           = open_flush_pipefs,
1642         .read           = read_flush_pipefs,
1643         .write          = write_flush_pipefs,
1644         .release        = release_flush_pipefs,
1645 };
1646
1647 int sunrpc_cache_register_pipefs(struct dentry *parent,
1648                                  const char *name, mode_t umode,
1649                                  struct cache_detail *cd)
1650 {
1651         struct qstr q;
1652         struct dentry *dir;
1653         int ret = 0;
1654
1655         sunrpc_init_cache_detail(cd);
1656         q.name = name;
1657         q.len = strlen(name);
1658         q.hash = full_name_hash(q.name, q.len);
1659         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1660         if (!IS_ERR(dir))
1661                 cd->u.pipefs.dir = dir;
1662         else {
1663                 sunrpc_destroy_cache_detail(cd);
1664                 ret = PTR_ERR(dir);
1665         }
1666         return ret;
1667 }
1668 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1669
1670 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1671 {
1672         rpc_remove_cache_dir(cd->u.pipefs.dir);
1673         cd->u.pipefs.dir = NULL;
1674         sunrpc_destroy_cache_detail(cd);
1675 }
1676 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1677