SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  38 #define  RPCDBG_FACILITY RPCDBG_CACHE
  39
  40 static int cache_defer_req(struct cache_req *req, struct cache_head *item);
  41 static void cache_revisit_request(struct cache_head *item);
  42
  43 static void cache_init(struct cache_head *h)
  44 {
  45         time_t now = get_seconds();
  46         h->next = NULL;
  47         h->flags = 0;
  48         kref_init(&h->ref);
  49         h->expiry_time = now + CACHE_NEW_EXPIRY;
  50         h->last_refresh = now;
  51 }
  52
  53 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  54                                        struct cache_head *key, int hash)
  55 {
  56         struct cache_head **head,  **hp;
  57         struct cache_head *new = NULL;
  58
  59         head = &detail->hash_table[hash];
  60
  61         read_lock(&detail->hash_lock);
  62
  63         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  64                 struct cache_head *tmp = *hp;
  65                 if (detail->match(tmp, key)) {
  66                         cache_get(tmp);
  67                         read_unlock(&detail->hash_lock);
  68                         return tmp;
  69                 }
  70         }
  71         read_unlock(&detail->hash_lock);
  72         /* Didn't find anything, insert an empty entry */
  73
  74         new = detail->alloc();
  75         if (!new)
  76                 return NULL;
  77         /* must fully initialise 'new', else
  78          * we might get lose if we need to
  79          * cache_put it soon.
  80          */
  81         cache_init(new);
  82         detail->init(new, key);
  83
  84         write_lock(&detail->hash_lock);
  85
  86         /* check if entry appeared while we slept */
  87         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  88                 struct cache_head *tmp = *hp;
  89                 if (detail->match(tmp, key)) {
  90                         cache_get(tmp);
  91                         write_unlock(&detail->hash_lock);
  92                         cache_put(new, detail);
  93                         return tmp;
  94                 }
  95         }
  96         new->next = *head;
  97         *head = new;
  98         detail->entries++;
  99         cache_get(new);
 100         write_unlock(&detail->hash_lock);
 101
 102         return new;
 103 }
 104 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
 105
 106
 107 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 108
 109 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
 110 {
 111         head->expiry_time = expiry;
 112         head->last_refresh = get_seconds();
 113         set_bit(CACHE_VALID, &head->flags);
 114 }
 115
 116 static void cache_fresh_unlocked(struct cache_head *head,
 117                                  struct cache_detail *detail)
 118 {
 119         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 120                 cache_revisit_request(head);
 121                 cache_dequeue(detail, head);
 122         }
 123 }
 124
 125 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 126                                        struct cache_head *new, struct cache_head *old, int hash)
 127 {
 128         /* The 'old' entry is to be replaced by 'new'.
 129          * If 'old' is not VALID, we update it directly,
 130          * otherwise we need to replace it
 131          */
 132         struct cache_head **head;
 133         struct cache_head *tmp;
 134
 135         if (!test_bit(CACHE_VALID, &old->flags)) {
 136                 write_lock(&detail->hash_lock);
 137                 if (!test_bit(CACHE_VALID, &old->flags)) {
 138                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 139                                 set_bit(CACHE_NEGATIVE, &old->flags);
 140                         else
 141                                 detail->update(old, new);
 142                         cache_fresh_locked(old, new->expiry_time);
 143                         write_unlock(&detail->hash_lock);
 144                         cache_fresh_unlocked(old, detail);
 145                         return old;
 146                 }
 147                 write_unlock(&detail->hash_lock);
 148         }
 149         /* We need to insert a new entry */
 150         tmp = detail->alloc();
 151         if (!tmp) {
 152                 cache_put(old, detail);
 153                 return NULL;
 154         }
 155         cache_init(tmp);
 156         detail->init(tmp, old);
 157         head = &detail->hash_table[hash];
 158
 159         write_lock(&detail->hash_lock);
 160         if (test_bit(CACHE_NEGATIVE, &new->flags))
 161                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 162         else
 163                 detail->update(tmp, new);
 164         tmp->next = *head;
 165         *head = tmp;
 166         detail->entries++;
 167         cache_get(tmp);
 168         cache_fresh_locked(tmp, new->expiry_time);
 169         cache_fresh_locked(old, 0);
 170         write_unlock(&detail->hash_lock);
 171         cache_fresh_unlocked(tmp, detail);
 172         cache_fresh_unlocked(old, detail);
 173         cache_put(old, detail);
 174         return tmp;
 175 }
 176 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 177
 178 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 179 {
 180         if (!cd->cache_upcall)
 181                 return -EINVAL;
 182         return cd->cache_upcall(cd, h);
 183 }
 184
 185 static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
 186 {
 187         if (!test_bit(CACHE_VALID, &h->flags) ||
 188             h->expiry_time < get_seconds())
 189                 return -EAGAIN;
 190         else if (detail->flush_time > h->last_refresh)
 191                 return -EAGAIN;
 192         else {
 193                 /* entry is valid */
 194                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 195                         return -ENOENT;
 196                 else
 197                         return 0;
 198         }
 199 }
 200
 201 /*
 202  * This is the generic cache management routine for all
 203  * the authentication caches.
 204  * It checks the currency of a cache item and will (later)
 205  * initiate an upcall to fill it if needed.
 206  *
 207  *
 208  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 209  * -EAGAIN if upcall is pending and request has been queued
 210  * -ETIMEDOUT if upcall failed or request could not be queue or
 211  *           upcall completed but item is still invalid (implying that
 212  *           the cache item has been replaced with a newer one).
 213  * -ENOENT if cache entry was negative
 214  */
 215 int cache_check(struct cache_detail *detail,
 216                     struct cache_head *h, struct cache_req *rqstp)
 217 {
 218         int rv;
 219         long refresh_age, age;
 220
 221         /* First decide return status as best we can */
 222         rv = cache_is_valid(detail, h);
 223
 224         /* now see if we want to start an upcall */
 225         refresh_age = (h->expiry_time - h->last_refresh);
 226         age = get_seconds() - h->last_refresh;
 227
 228         if (rqstp == NULL) {
 229                 if (rv == -EAGAIN)
 230                         rv = -ENOENT;
 231         } else if (rv == -EAGAIN || age > refresh_age/2) {
 232                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 233                                 refresh_age, age);
 234                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 235                         switch (cache_make_upcall(detail, h)) {
 236                         case -EINVAL:
 237                                 clear_bit(CACHE_PENDING, &h->flags);
 238                                 cache_revisit_request(h);
 239                                 if (rv == -EAGAIN) {
 240                                         set_bit(CACHE_NEGATIVE, &h->flags);
 241                                         cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY);
 242                                         cache_fresh_unlocked(h, detail);
 243                                         rv = -ENOENT;
 244                                 }
 245                                 break;
 246
 247                         case -EAGAIN:
 248                                 clear_bit(CACHE_PENDING, &h->flags);
 249                                 cache_revisit_request(h);
 250                                 break;
 251                         }
 252                 }
 253         }
 254
 255         if (rv == -EAGAIN) {
 256                 if (cache_defer_req(rqstp, h) < 0) {
 257                         /* Request is not deferred */
 258                         rv = cache_is_valid(detail, h);
 259                         if (rv == -EAGAIN)
 260                                 rv = -ETIMEDOUT;
 261                 }
 262         }
 263         if (rv)
 264                 cache_put(h, detail);
 265         return rv;
 266 }
 267 EXPORT_SYMBOL_GPL(cache_check);
 268
 269 /*
 270  * caches need to be periodically cleaned.
 271  * For this we maintain a list of cache_detail and
 272  * a current pointer into that list and into the table
 273  * for that entry.
 274  *
 275  * Each time clean_cache is called it finds the next non-empty entry
 276  * in the current table and walks the list in that entry
 277  * looking for entries that can be removed.
 278  *
 279  * An entry gets removed if:
 280  * - The expiry is before current time
 281  * - The last_refresh time is before the flush_time for that cache
 282  *
 283  * later we might drop old entries with non-NEVER expiry if that table
 284  * is getting 'full' for some definition of 'full'
 285  *
 286  * The question of "how often to scan a table" is an interesting one
 287  * and is answered in part by the use of the "nextcheck" field in the
 288  * cache_detail.
 289  * When a scan of a table begins, the nextcheck field is set to a time
 290  * that is well into the future.
 291  * While scanning, if an expiry time is found that is earlier than the
 292  * current nextcheck time, nextcheck is set to that expiry time.
 293  * If the flush_time is ever set to a time earlier than the nextcheck
 294  * time, the nextcheck time is then set to that flush_time.
 295  *
 296  * A table is then only scanned if the current time is at least
 297  * the nextcheck time.
 298  *
 299  */
 300
 301 static LIST_HEAD(cache_list);
 302 static DEFINE_SPINLOCK(cache_list_lock);
 303 static struct cache_detail *current_detail;
 304 static int current_index;
 305
 306 static void do_cache_clean(struct work_struct *work);
 307 static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
 308
 309 static void sunrpc_init_cache_detail(struct cache_detail *cd)
 310 {
 311         rwlock_init(&cd->hash_lock);
 312         INIT_LIST_HEAD(&cd->queue);
 313         spin_lock(&cache_list_lock);
 314         cd->nextcheck = 0;
 315         cd->entries = 0;
 316         atomic_set(&cd->readers, 0);
 317         cd->last_close = 0;
 318         cd->last_warn = -1;
 319         list_add(&cd->others, &cache_list);
 320         spin_unlock(&cache_list_lock);
 321
 322         /* start the cleaning process */
 323         schedule_delayed_work(&cache_cleaner, 0);
 324 }
 325
 326 static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 327 {
 328         cache_purge(cd);
 329         spin_lock(&cache_list_lock);
 330         write_lock(&cd->hash_lock);
 331         if (cd->entries || atomic_read(&cd->inuse)) {
 332                 write_unlock(&cd->hash_lock);
 333                 spin_unlock(&cache_list_lock);
 334                 goto out;
 335         }
 336         if (current_detail == cd)
 337                 current_detail = NULL;
 338         list_del_init(&cd->others);
 339         write_unlock(&cd->hash_lock);
 340         spin_unlock(&cache_list_lock);
 341         if (list_empty(&cache_list)) {
 342                 /* module must be being unloaded so its safe to kill the worker */
 343                 cancel_delayed_work_sync(&cache_cleaner);
 344         }
 345         return;
 346 out:
 347         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 348 }
 349
 350 /* clean cache tries to find something to clean
 351  * and cleans it.
 352  * It returns 1 if it cleaned something,
 353  *            0 if it didn't find anything this time
 354  *           -1 if it fell off the end of the list.
 355  */
 356 static int cache_clean(void)
 357 {
 358         int rv = 0;
 359         struct list_head *next;
 360
 361         spin_lock(&cache_list_lock);
 362
 363         /* find a suitable table if we don't already have one */
 364         while (current_detail == NULL ||
 365             current_index >= current_detail->hash_size) {
 366                 if (current_detail)
 367                         next = current_detail->others.next;
 368                 else
 369                         next = cache_list.next;
 370                 if (next == &cache_list) {
 371                         current_detail = NULL;
 372                         spin_unlock(&cache_list_lock);
 373                         return -1;
 374                 }
 375                 current_detail = list_entry(next, struct cache_detail, others);
 376                 if (current_detail->nextcheck > get_seconds())
 377                         current_index = current_detail->hash_size;
 378                 else {
 379                         current_index = 0;
 380                         current_detail->nextcheck = get_seconds()+30*60;
 381                 }
 382         }
 383
 384         /* find a non-empty bucket in the table */
 385         while (current_detail &&
 386                current_index < current_detail->hash_size &&
 387                current_detail->hash_table[current_index] == NULL)
 388                 current_index++;
 389
 390         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 391
 392         if (current_detail && current_index < current_detail->hash_size) {
 393                 struct cache_head *ch, **cp;
 394                 struct cache_detail *d;
 395
 396                 write_lock(&current_detail->hash_lock);
 397
 398                 /* Ok, now to clean this strand */
 399
 400                 cp = & current_detail->hash_table[current_index];
 401                 ch = *cp;
 402                 for (; ch; cp= & ch->next, ch= *cp) {
 403                         if (current_detail->nextcheck > ch->expiry_time)
 404                                 current_detail->nextcheck = ch->expiry_time+1;
 405                         if (ch->expiry_time >= get_seconds() &&
 406                             ch->last_refresh >= current_detail->flush_time)
 407                                 continue;
 408                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 409                                 cache_dequeue(current_detail, ch);
 410
 411                         if (atomic_read(&ch->ref.refcount) == 1)
 412                                 break;
 413                 }
 414                 if (ch) {
 415                         *cp = ch->next;
 416                         ch->next = NULL;
 417                         current_detail->entries--;
 418                         rv = 1;
 419                 }
 420                 write_unlock(&current_detail->hash_lock);
 421                 d = current_detail;
 422                 if (!ch)
 423                         current_index ++;
 424                 spin_unlock(&cache_list_lock);
 425                 if (ch) {
 426                         cache_revisit_request(ch);
 427                         cache_put(ch, d);
 428                 }
 429         } else
 430                 spin_unlock(&cache_list_lock);
 431
 432         return rv;
 433 }
 434
 435 /*
 436  * We want to regularly clean the cache, so we need to schedule some work ...
 437  */
 438 static void do_cache_clean(struct work_struct *work)
 439 {
 440         int delay = 5;
 441         if (cache_clean() == -1)
 442                 delay = round_jiffies_relative(30*HZ);
 443
 444         if (list_empty(&cache_list))
 445                 delay = 0;
 446
 447         if (delay)
 448                 schedule_delayed_work(&cache_cleaner, delay);
 449 }
 450
 451
 452 /*
 453  * Clean all caches promptly.  This just calls cache_clean
 454  * repeatedly until we are sure that every cache has had a chance to
 455  * be fully cleaned
 456  */
 457 void cache_flush(void)
 458 {
 459         while (cache_clean() != -1)
 460                 cond_resched();
 461         while (cache_clean() != -1)
 462                 cond_resched();
 463 }
 464 EXPORT_SYMBOL_GPL(cache_flush);
 465
 466 void cache_purge(struct cache_detail *detail)
 467 {
 468         detail->flush_time = LONG_MAX;
 469         detail->nextcheck = get_seconds();
 470         cache_flush();
 471         detail->flush_time = 1;
 472 }
 473 EXPORT_SYMBOL_GPL(cache_purge);
 474
 475
 476 /*
 477  * Deferral and Revisiting of Requests.
 478  *
 479  * If a cache lookup finds a pending entry, we
 480  * need to defer the request and revisit it later.
 481  * All deferred requests are stored in a hash table,
 482  * indexed by "struct cache_head *".
 483  * As it may be wasteful to store a whole request
 484  * structure, we allow the request to provide a
 485  * deferred form, which must contain a
 486  * 'struct cache_deferred_req'
 487  * This cache_deferred_req contains a method to allow
 488  * it to be revisited when cache info is available
 489  */
 490
 491 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 492 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 493
 494 #define DFR_MAX 300     /* ??? */
 495
 496 static DEFINE_SPINLOCK(cache_defer_lock);
 497 static LIST_HEAD(cache_defer_list);
 498 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 499 static int cache_defer_cnt;
 500
 501 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 502 {
 503         struct cache_deferred_req *dreq, *discard;
 504         int hash = DFR_HASH(item);
 505
 506         if (cache_defer_cnt >= DFR_MAX) {
 507                 /* too much in the cache, randomly drop this one,
 508                  * or continue and drop the oldest below
 509                  */
 510                 if (net_random()&1)
 511                         return -ENOMEM;
 512         }
 513         dreq = req->defer(req);
 514         if (dreq == NULL)
 515                 return -ENOMEM;
 516
 517         dreq->item = item;
 518
 519         spin_lock(&cache_defer_lock);
 520
 521         list_add(&dreq->recent, &cache_defer_list);
 522
 523         if (cache_defer_hash[hash].next == NULL)
 524                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 525         list_add(&dreq->hash, &cache_defer_hash[hash]);
 526
 527         /* it is in, now maybe clean up */
 528         discard = NULL;
 529         if (++cache_defer_cnt > DFR_MAX) {
 530                 discard = list_entry(cache_defer_list.prev,
 531                                      struct cache_deferred_req, recent);
 532                 list_del_init(&discard->recent);
 533                 list_del_init(&discard->hash);
 534                 cache_defer_cnt--;
 535         }
 536         spin_unlock(&cache_defer_lock);
 537
 538         if (discard)
 539                 /* there was one too many */
 540                 discard->revisit(discard, 1);
 541
 542         if (!test_bit(CACHE_PENDING, &item->flags)) {
 543                 /* must have just been validated... */
 544                 cache_revisit_request(item);
 545                 return -EAGAIN;
 546         }
 547         return 0;
 548 }
 549
 550 static void cache_revisit_request(struct cache_head *item)
 551 {
 552         struct cache_deferred_req *dreq;
 553         struct list_head pending;
 554
 555         struct list_head *lp;
 556         int hash = DFR_HASH(item);
 557
 558         INIT_LIST_HEAD(&pending);
 559         spin_lock(&cache_defer_lock);
 560
 561         lp = cache_defer_hash[hash].next;
 562         if (lp) {
 563                 while (lp != &cache_defer_hash[hash]) {
 564                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 565                         lp = lp->next;
 566                         if (dreq->item == item) {
 567                                 list_del_init(&dreq->hash);
 568                                 list_move(&dreq->recent, &pending);
 569                                 cache_defer_cnt--;
 570                         }
 571                 }
 572         }
 573         spin_unlock(&cache_defer_lock);
 574
 575         while (!list_empty(&pending)) {
 576                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 577                 list_del_init(&dreq->recent);
 578                 dreq->revisit(dreq, 0);
 579         }
 580 }
 581
 582 void cache_clean_deferred(void *owner)
 583 {
 584         struct cache_deferred_req *dreq, *tmp;
 585         struct list_head pending;
 586
 587
 588         INIT_LIST_HEAD(&pending);
 589         spin_lock(&cache_defer_lock);
 590
 591         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 592                 if (dreq->owner == owner) {
 593                         list_del_init(&dreq->hash);
 594                         list_move(&dreq->recent, &pending);
 595                         cache_defer_cnt--;
 596                 }
 597         }
 598         spin_unlock(&cache_defer_lock);
 599
 600         while (!list_empty(&pending)) {
 601                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 602                 list_del_init(&dreq->recent);
 603                 dreq->revisit(dreq, 1);
 604         }
 605 }
 606
 607 /*
 608  * communicate with user-space
 609  *
 610  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 611  * On read, you get a full request, or block.
 612  * On write, an update request is processed.
 613  * Poll works if anything to read, and always allows write.
 614  *
 615  * Implemented by linked list of requests.  Each open file has
 616  * a ->private that also exists in this list.  New requests are added
 617  * to the end and may wakeup and preceding readers.
 618  * New readers are added to the head.  If, on read, an item is found with
 619  * CACHE_UPCALLING clear, we free it from the list.
 620  *
 621  */
 622
 623 static DEFINE_SPINLOCK(queue_lock);
 624 static DEFINE_MUTEX(queue_io_mutex);
 625
 626 struct cache_queue {
 627         struct list_head        list;
 628         int                     reader; /* if 0, then request */
 629 };
 630 struct cache_request {
 631         struct cache_queue      q;
 632         struct cache_head       *item;
 633         char                    * buf;
 634         int                     len;
 635         int                     readers;
 636 };
 637 struct cache_reader {
 638         struct cache_queue      q;
 639         int                     offset; /* if non-0, we have a refcnt on next request */
 640 };
 641
 642 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 643                           loff_t *ppos, struct cache_detail *cd)
 644 {
 645         struct cache_reader *rp = filp->private_data;
 646         struct cache_request *rq;
 647         struct inode *inode = filp->f_path.dentry->d_inode;
 648         int err;
 649
 650         if (count == 0)
 651                 return 0;
 652
 653         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 654                               * readers on this file */
 655  again:
 656         spin_lock(&queue_lock);
 657         /* need to find next request */
 658         while (rp->q.list.next != &cd->queue &&
 659                list_entry(rp->q.list.next, struct cache_queue, list)
 660                ->reader) {
 661                 struct list_head *next = rp->q.list.next;
 662                 list_move(&rp->q.list, next);
 663         }
 664         if (rp->q.list.next == &cd->queue) {
 665                 spin_unlock(&queue_lock);
 666                 mutex_unlock(&inode->i_mutex);
 667                 BUG_ON(rp->offset);
 668                 return 0;
 669         }
 670         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 671         BUG_ON(rq->q.reader);
 672         if (rp->offset == 0)
 673                 rq->readers++;
 674         spin_unlock(&queue_lock);
 675
 676         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 677                 err = -EAGAIN;
 678                 spin_lock(&queue_lock);
 679                 list_move(&rp->q.list, &rq->q.list);
 680                 spin_unlock(&queue_lock);
 681         } else {
 682                 if (rp->offset + count > rq->len)
 683                         count = rq->len - rp->offset;
 684                 err = -EFAULT;
 685                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 686                         goto out;
 687                 rp->offset += count;
 688                 if (rp->offset >= rq->len) {
 689                         rp->offset = 0;
 690                         spin_lock(&queue_lock);
 691                         list_move(&rp->q.list, &rq->q.list);
 692                         spin_unlock(&queue_lock);
 693                 }
 694                 err = 0;
 695         }
 696  out:
 697         if (rp->offset == 0) {
 698                 /* need to release rq */
 699                 spin_lock(&queue_lock);
 700                 rq->readers--;
 701                 if (rq->readers == 0 &&
 702                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 703                         list_del(&rq->q.list);
 704                         spin_unlock(&queue_lock);
 705                         cache_put(rq->item, cd);
 706                         kfree(rq->buf);
 707                         kfree(rq);
 708                 } else
 709                         spin_unlock(&queue_lock);
 710         }
 711         if (err == -EAGAIN)
 712                 goto again;
 713         mutex_unlock(&inode->i_mutex);
 714         return err ? err :  count;
 715 }
 716
 717 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 718                                  size_t count, struct cache_detail *cd)
 719 {
 720         ssize_t ret;
 721
 722         if (copy_from_user(kaddr, buf, count))
 723                 return -EFAULT;
 724         kaddr[count] = '\0';
 725         ret = cd->cache_parse(cd, kaddr, count);
 726         if (!ret)
 727                 ret = count;
 728         return ret;
 729 }
 730
 731 static ssize_t cache_slow_downcall(const char __user *buf,
 732                                    size_t count, struct cache_detail *cd)
 733 {
 734         static char write_buf[8192]; /* protected by queue_io_mutex */
 735         ssize_t ret = -EINVAL;
 736
 737         if (count >= sizeof(write_buf))
 738                 goto out;
 739         mutex_lock(&queue_io_mutex);
 740         ret = cache_do_downcall(write_buf, buf, count, cd);
 741         mutex_unlock(&queue_io_mutex);
 742 out:
 743         return ret;
 744 }
 745
 746 static ssize_t cache_downcall(struct address_space *mapping,
 747                               const char __user *buf,
 748                               size_t count, struct cache_detail *cd)
 749 {
 750         struct page *page;
 751         char *kaddr;
 752         ssize_t ret = -ENOMEM;
 753
 754         if (count >= PAGE_CACHE_SIZE)
 755                 goto out_slow;
 756
 757         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 758         if (!page)
 759                 goto out_slow;
 760
 761         kaddr = kmap(page);
 762         ret = cache_do_downcall(kaddr, buf, count, cd);
 763         kunmap(page);
 764         unlock_page(page);
 765         page_cache_release(page);
 766         return ret;
 767 out_slow:
 768         return cache_slow_downcall(buf, count, cd);
 769 }
 770
 771 static ssize_t cache_write(struct file *filp, const char __user *buf,
 772                            size_t count, loff_t *ppos,
 773                            struct cache_detail *cd)
 774 {
 775         struct address_space *mapping = filp->f_mapping;
 776         struct inode *inode = filp->f_path.dentry->d_inode;
 777         ssize_t ret = -EINVAL;
 778
 779         if (!cd->cache_parse)
 780                 goto out;
 781
 782         mutex_lock(&inode->i_mutex);
 783         ret = cache_downcall(mapping, buf, count, cd);
 784         mutex_unlock(&inode->i_mutex);
 785 out:
 786         return ret;
 787 }
 788
 789 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 790
 791 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 792                                struct cache_detail *cd)
 793 {
 794         unsigned int mask;
 795         struct cache_reader *rp = filp->private_data;
 796         struct cache_queue *cq;
 797
 798         poll_wait(filp, &queue_wait, wait);
 799
 800         /* alway allow write */
 801         mask = POLL_OUT | POLLWRNORM;
 802
 803         if (!rp)
 804                 return mask;
 805
 806         spin_lock(&queue_lock);
 807
 808         for (cq= &rp->q; &cq->list != &cd->queue;
 809              cq = list_entry(cq->list.next, struct cache_queue, list))
 810                 if (!cq->reader) {
 811                         mask |= POLLIN | POLLRDNORM;
 812                         break;
 813                 }
 814         spin_unlock(&queue_lock);
 815         return mask;
 816 }
 817
 818 static int cache_ioctl(struct inode *ino, struct file *filp,
 819                        unsigned int cmd, unsigned long arg,
 820                        struct cache_detail *cd)
 821 {
 822         int len = 0;
 823         struct cache_reader *rp = filp->private_data;
 824         struct cache_queue *cq;
 825
 826         if (cmd != FIONREAD || !rp)
 827                 return -EINVAL;
 828
 829         spin_lock(&queue_lock);
 830
 831         /* only find the length remaining in current request,
 832          * or the length of the next request
 833          */
 834         for (cq= &rp->q; &cq->list != &cd->queue;
 835              cq = list_entry(cq->list.next, struct cache_queue, list))
 836                 if (!cq->reader) {
 837                         struct cache_request *cr =
 838                                 container_of(cq, struct cache_request, q);
 839                         len = cr->len - rp->offset;
 840                         break;
 841                 }
 842         spin_unlock(&queue_lock);
 843
 844         return put_user(len, (int __user *)arg);
 845 }
 846
 847 static int cache_open(struct inode *inode, struct file *filp,
 848                       struct cache_detail *cd)
 849 {
 850         struct cache_reader *rp = NULL;
 851
 852         if (!cd || !try_module_get(cd->owner))
 853                 return -EACCES;
 854         nonseekable_open(inode, filp);
 855         if (filp->f_mode & FMODE_READ) {
 856                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 857                 if (!rp)
 858                         return -ENOMEM;
 859                 rp->offset = 0;
 860                 rp->q.reader = 1;
 861                 atomic_inc(&cd->readers);
 862                 spin_lock(&queue_lock);
 863                 list_add(&rp->q.list, &cd->queue);
 864                 spin_unlock(&queue_lock);
 865         }
 866         filp->private_data = rp;
 867         return 0;
 868 }
 869
 870 static int cache_release(struct inode *inode, struct file *filp,
 871                          struct cache_detail *cd)
 872 {
 873         struct cache_reader *rp = filp->private_data;
 874
 875         if (rp) {
 876                 spin_lock(&queue_lock);
 877                 if (rp->offset) {
 878                         struct cache_queue *cq;
 879                         for (cq= &rp->q; &cq->list != &cd->queue;
 880                              cq = list_entry(cq->list.next, struct cache_queue, list))
 881                                 if (!cq->reader) {
 882                                         container_of(cq, struct cache_request, q)
 883                                                 ->readers--;
 884                                         break;
 885                                 }
 886                         rp->offset = 0;
 887                 }
 888                 list_del(&rp->q.list);
 889                 spin_unlock(&queue_lock);
 890
 891                 filp->private_data = NULL;
 892                 kfree(rp);
 893
 894                 cd->last_close = get_seconds();
 895                 atomic_dec(&cd->readers);
 896         }
 897         module_put(cd->owner);
 898         return 0;
 899 }
 900
 901
 902
 903 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
 904 {
 905         struct cache_queue *cq;
 906         spin_lock(&queue_lock);
 907         list_for_each_entry(cq, &detail->queue, list)
 908                 if (!cq->reader) {
 909                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 910                         if (cr->item != ch)
 911                                 continue;
 912                         if (cr->readers != 0)
 913                                 continue;
 914                         list_del(&cr->q.list);
 915                         spin_unlock(&queue_lock);
 916                         cache_put(cr->item, detail);
 917                         kfree(cr->buf);
 918                         kfree(cr);
 919                         return;
 920                 }
 921         spin_unlock(&queue_lock);
 922 }
 923
 924 /*
 925  * Support routines for text-based upcalls.
 926  * Fields are separated by spaces.
 927  * Fields are either mangled to quote space tab newline slosh with slosh
 928  * or a hexified with a leading \x
 929  * Record is terminated with newline.
 930  *
 931  */
 932
 933 void qword_add(char **bpp, int *lp, char *str)
 934 {
 935         char *bp = *bpp;
 936         int len = *lp;
 937         char c;
 938
 939         if (len < 0) return;
 940
 941         while ((c=*str++) && len)
 942                 switch(c) {
 943                 case ' ':
 944                 case '\t':
 945                 case '\n':
 946                 case '\\':
 947                         if (len >= 4) {
 948                                 *bp++ = '\\';
 949                                 *bp++ = '0' + ((c & 0300)>>6);
 950                                 *bp++ = '0' + ((c & 0070)>>3);
 951                                 *bp++ = '0' + ((c & 0007)>>0);
 952                         }
 953                         len -= 4;
 954                         break;
 955                 default:
 956                         *bp++ = c;
 957                         len--;
 958                 }
 959         if (c || len <1) len = -1;
 960         else {
 961                 *bp++ = ' ';
 962                 len--;
 963         }
 964         *bpp = bp;
 965         *lp = len;
 966 }
 967 EXPORT_SYMBOL_GPL(qword_add);
 968
 969 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
 970 {
 971         char *bp = *bpp;
 972         int len = *lp;
 973
 974         if (len < 0) return;
 975
 976         if (len > 2) {
 977                 *bp++ = '\\';
 978                 *bp++ = 'x';
 979                 len -= 2;
 980                 while (blen && len >= 2) {
 981                         unsigned char c = *buf++;
 982                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
 983                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
 984                         len -= 2;
 985                         blen--;
 986                 }
 987         }
 988         if (blen || len<1) len = -1;
 989         else {
 990                 *bp++ = ' ';
 991                 len--;
 992         }
 993         *bpp = bp;
 994         *lp = len;
 995 }
 996 EXPORT_SYMBOL_GPL(qword_addhex);
 997
 998 static void warn_no_listener(struct cache_detail *detail)
 999 {
1000         if (detail->last_warn != detail->last_close) {
1001                 detail->last_warn = detail->last_close;
1002                 if (detail->warn_no_listener)
1003                         detail->warn_no_listener(detail, detail->last_close != 0);
1004         }
1005 }
1006
1007 /*
1008  * register an upcall request to user-space and queue it up for read() by the
1009  * upcall daemon.
1010  *
1011  * Each request is at most one page long.
1012  */
1013 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1014                 void (*cache_request)(struct cache_detail *,
1015                                       struct cache_head *,
1016                                       char **,
1017                                       int *))
1018 {
1019
1020         char *buf;
1021         struct cache_request *crq;
1022         char *bp;
1023         int len;
1024
1025         if (atomic_read(&detail->readers) == 0 &&
1026             detail->last_close < get_seconds() - 30) {
1027                         warn_no_listener(detail);
1028                         return -EINVAL;
1029         }
1030
1031         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1032         if (!buf)
1033                 return -EAGAIN;
1034
1035         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1036         if (!crq) {
1037                 kfree(buf);
1038                 return -EAGAIN;
1039         }
1040
1041         bp = buf; len = PAGE_SIZE;
1042
1043         cache_request(detail, h, &bp, &len);
1044
1045         if (len < 0) {
1046                 kfree(buf);
1047                 kfree(crq);
1048                 return -EAGAIN;
1049         }
1050         crq->q.reader = 0;
1051         crq->item = cache_get(h);
1052         crq->buf = buf;
1053         crq->len = PAGE_SIZE - len;
1054         crq->readers = 0;
1055         spin_lock(&queue_lock);
1056         list_add_tail(&crq->q.list, &detail->queue);
1057         spin_unlock(&queue_lock);
1058         wake_up(&queue_wait);
1059         return 0;
1060 }
1061 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1062
1063 /*
1064  * parse a message from user-space and pass it
1065  * to an appropriate cache
1066  * Messages are, like requests, separated into fields by
1067  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1068  *
1069  * Message is
1070  *   reply cachename expiry key ... content....
1071  *
1072  * key and content are both parsed by cache
1073  */
1074
1075 #define isodigit(c) (isdigit(c) && c <= '7')
1076 int qword_get(char **bpp, char *dest, int bufsize)
1077 {
1078         /* return bytes copied, or -1 on error */
1079         char *bp = *bpp;
1080         int len = 0;
1081
1082         while (*bp == ' ') bp++;
1083
1084         if (bp[0] == '\\' && bp[1] == 'x') {
1085                 /* HEX STRING */
1086                 bp += 2;
1087                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1088                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1089                         bp++;
1090                         byte <<= 4;
1091                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1092                         *dest++ = byte;
1093                         bp++;
1094                         len++;
1095                 }
1096         } else {
1097                 /* text with \nnn octal quoting */
1098                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1099                         if (*bp == '\\' &&
1100                             isodigit(bp[1]) && (bp[1] <= '3') &&
1101                             isodigit(bp[2]) &&
1102                             isodigit(bp[3])) {
1103                                 int byte = (*++bp -'0');
1104                                 bp++;
1105                                 byte = (byte << 3) | (*bp++ - '0');
1106                                 byte = (byte << 3) | (*bp++ - '0');
1107                                 *dest++ = byte;
1108                                 len++;
1109                         } else {
1110                                 *dest++ = *bp++;
1111                                 len++;
1112                         }
1113                 }
1114         }
1115
1116         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1117                 return -1;
1118         while (*bp == ' ') bp++;
1119         *bpp = bp;
1120         *dest = '\0';
1121         return len;
1122 }
1123 EXPORT_SYMBOL_GPL(qword_get);
1124
1125
1126 /*
1127  * support /proc/sunrpc/cache/$CACHENAME/content
1128  * as a seqfile.
1129  * We call ->cache_show passing NULL for the item to
1130  * get a header, then pass each real item in the cache
1131  */
1132
1133 struct handle {
1134         struct cache_detail *cd;
1135 };
1136
1137 static void *c_start(struct seq_file *m, loff_t *pos)
1138         __acquires(cd->hash_lock)
1139 {
1140         loff_t n = *pos;
1141         unsigned hash, entry;
1142         struct cache_head *ch;
1143         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1144
1145
1146         read_lock(&cd->hash_lock);
1147         if (!n--)
1148                 return SEQ_START_TOKEN;
1149         hash = n >> 32;
1150         entry = n & ((1LL<<32) - 1);
1151
1152         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1153                 if (!entry--)
1154                         return ch;
1155         n &= ~((1LL<<32) - 1);
1156         do {
1157                 hash++;
1158                 n += 1LL<<32;
1159         } while(hash < cd->hash_size &&
1160                 cd->hash_table[hash]==NULL);
1161         if (hash >= cd->hash_size)
1162                 return NULL;
1163         *pos = n+1;
1164         return cd->hash_table[hash];
1165 }
1166
1167 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1168 {
1169         struct cache_head *ch = p;
1170         int hash = (*pos >> 32);
1171         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1172
1173         if (p == SEQ_START_TOKEN)
1174                 hash = 0;
1175         else if (ch->next == NULL) {
1176                 hash++;
1177                 *pos += 1LL<<32;
1178         } else {
1179                 ++*pos;
1180                 return ch->next;
1181         }
1182         *pos &= ~((1LL<<32) - 1);
1183         while (hash < cd->hash_size &&
1184                cd->hash_table[hash] == NULL) {
1185                 hash++;
1186                 *pos += 1LL<<32;
1187         }
1188         if (hash >= cd->hash_size)
1189                 return NULL;
1190         ++*pos;
1191         return cd->hash_table[hash];
1192 }
1193
1194 static void c_stop(struct seq_file *m, void *p)
1195         __releases(cd->hash_lock)
1196 {
1197         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1198         read_unlock(&cd->hash_lock);
1199 }
1200
1201 static int c_show(struct seq_file *m, void *p)
1202 {
1203         struct cache_head *cp = p;
1204         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1205
1206         if (p == SEQ_START_TOKEN)
1207                 return cd->cache_show(m, cd, NULL);
1208
1209         ifdebug(CACHE)
1210                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1211                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1212         cache_get(cp);
1213         if (cache_check(cd, cp, NULL))
1214                 /* cache_check does a cache_put on failure */
1215                 seq_printf(m, "# ");
1216         else
1217                 cache_put(cp, cd);
1218
1219         return cd->cache_show(m, cd, cp);
1220 }
1221
1222 static const struct seq_operations cache_content_op = {
1223         .start  = c_start,
1224         .next   = c_next,
1225         .stop   = c_stop,
1226         .show   = c_show,
1227 };
1228
1229 static int content_open(struct inode *inode, struct file *file,
1230                         struct cache_detail *cd)
1231 {
1232         struct handle *han;
1233
1234         if (!cd || !try_module_get(cd->owner))
1235                 return -EACCES;
1236         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1237         if (han == NULL)
1238                 return -ENOMEM;
1239
1240         han->cd = cd;
1241         return 0;
1242 }
1243
1244 static int content_release(struct inode *inode, struct file *file,
1245                 struct cache_detail *cd)
1246 {
1247         int ret = seq_release_private(inode, file);
1248         module_put(cd->owner);
1249         return ret;
1250 }
1251
1252 static int open_flush(struct inode *inode, struct file *file,
1253                         struct cache_detail *cd)
1254 {
1255         if (!cd || !try_module_get(cd->owner))
1256                 return -EACCES;
1257         return nonseekable_open(inode, file);
1258 }
1259
1260 static int release_flush(struct inode *inode, struct file *file,
1261                         struct cache_detail *cd)
1262 {
1263         module_put(cd->owner);
1264         return 0;
1265 }
1266
1267 static ssize_t read_flush(struct file *file, char __user *buf,
1268                           size_t count, loff_t *ppos,
1269                           struct cache_detail *cd)
1270 {
1271         char tbuf[20];
1272         unsigned long p = *ppos;
1273         size_t len;
1274
1275         sprintf(tbuf, "%lu\n", cd->flush_time);
1276         len = strlen(tbuf);
1277         if (p >= len)
1278                 return 0;
1279         len -= p;
1280         if (len > count)
1281                 len = count;
1282         if (copy_to_user(buf, (void*)(tbuf+p), len))
1283                 return -EFAULT;
1284         *ppos += len;
1285         return len;
1286 }
1287
1288 static ssize_t write_flush(struct file *file, const char __user *buf,
1289                            size_t count, loff_t *ppos,
1290                            struct cache_detail *cd)
1291 {
1292         char tbuf[20];
1293         char *ep;
1294         long flushtime;
1295         if (*ppos || count > sizeof(tbuf)-1)
1296                 return -EINVAL;
1297         if (copy_from_user(tbuf, buf, count))
1298                 return -EFAULT;
1299         tbuf[count] = 0;
1300         flushtime = simple_strtoul(tbuf, &ep, 0);
1301         if (*ep && *ep != '\n')
1302                 return -EINVAL;
1303
1304         cd->flush_time = flushtime;
1305         cd->nextcheck = get_seconds();
1306         cache_flush();
1307
1308         *ppos += count;
1309         return count;
1310 }
1311
1312 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1313                                  size_t count, loff_t *ppos)
1314 {
1315         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1316
1317         return cache_read(filp, buf, count, ppos, cd);
1318 }
1319
1320 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1321                                   size_t count, loff_t *ppos)
1322 {
1323         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1324
1325         return cache_write(filp, buf, count, ppos, cd);
1326 }
1327
1328 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1329 {
1330         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1331
1332         return cache_poll(filp, wait, cd);
1333 }
1334
1335 static long cache_ioctl_procfs(struct file *filp,
1336                                unsigned int cmd, unsigned long arg)
1337 {
1338         long ret;
1339         struct inode *inode = filp->f_path.dentry->d_inode;
1340         struct cache_detail *cd = PDE(inode)->data;
1341
1342         lock_kernel();
1343         ret = cache_ioctl(inode, filp, cmd, arg, cd);
1344         unlock_kernel();
1345
1346         return ret;
1347 }
1348
1349 static int cache_open_procfs(struct inode *inode, struct file *filp)
1350 {
1351         struct cache_detail *cd = PDE(inode)->data;
1352
1353         return cache_open(inode, filp, cd);
1354 }
1355
1356 static int cache_release_procfs(struct inode *inode, struct file *filp)
1357 {
1358         struct cache_detail *cd = PDE(inode)->data;
1359
1360         return cache_release(inode, filp, cd);
1361 }
1362
1363 static const struct file_operations cache_file_operations_procfs = {
1364         .owner          = THIS_MODULE,
1365         .llseek         = no_llseek,
1366         .read           = cache_read_procfs,
1367         .write          = cache_write_procfs,
1368         .poll           = cache_poll_procfs,
1369         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1370         .open           = cache_open_procfs,
1371         .release        = cache_release_procfs,
1372 };
1373
1374 static int content_open_procfs(struct inode *inode, struct file *filp)
1375 {
1376         struct cache_detail *cd = PDE(inode)->data;
1377
1378         return content_open(inode, filp, cd);
1379 }
1380
1381 static int content_release_procfs(struct inode *inode, struct file *filp)
1382 {
1383         struct cache_detail *cd = PDE(inode)->data;
1384
1385         return content_release(inode, filp, cd);
1386 }
1387
1388 static const struct file_operations content_file_operations_procfs = {
1389         .open           = content_open_procfs,
1390         .read           = seq_read,
1391         .llseek         = seq_lseek,
1392         .release        = content_release_procfs,
1393 };
1394
1395 static int open_flush_procfs(struct inode *inode, struct file *filp)
1396 {
1397         struct cache_detail *cd = PDE(inode)->data;
1398
1399         return open_flush(inode, filp, cd);
1400 }
1401
1402 static int release_flush_procfs(struct inode *inode, struct file *filp)
1403 {
1404         struct cache_detail *cd = PDE(inode)->data;
1405
1406         return release_flush(inode, filp, cd);
1407 }
1408
1409 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1410                             size_t count, loff_t *ppos)
1411 {
1412         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1413
1414         return read_flush(filp, buf, count, ppos, cd);
1415 }
1416
1417 static ssize_t write_flush_procfs(struct file *filp,
1418                                   const char __user *buf,
1419                                   size_t count, loff_t *ppos)
1420 {
1421         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1422
1423         return write_flush(filp, buf, count, ppos, cd);
1424 }
1425
1426 static const struct file_operations cache_flush_operations_procfs = {
1427         .open           = open_flush_procfs,
1428         .read           = read_flush_procfs,
1429         .write          = write_flush_procfs,
1430         .release        = release_flush_procfs,
1431 };
1432
1433 static void remove_cache_proc_entries(struct cache_detail *cd)
1434 {
1435         if (cd->u.procfs.proc_ent == NULL)
1436                 return;
1437         if (cd->u.procfs.flush_ent)
1438                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1439         if (cd->u.procfs.channel_ent)
1440                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1441         if (cd->u.procfs.content_ent)
1442                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1443         cd->u.procfs.proc_ent = NULL;
1444         remove_proc_entry(cd->name, proc_net_rpc);
1445 }
1446
1447 #ifdef CONFIG_PROC_FS
1448 static int create_cache_proc_entries(struct cache_detail *cd)
1449 {
1450         struct proc_dir_entry *p;
1451
1452         cd->u.procfs.proc_ent = proc_mkdir(cd->name, proc_net_rpc);
1453         if (cd->u.procfs.proc_ent == NULL)
1454                 goto out_nomem;
1455         cd->u.procfs.channel_ent = NULL;
1456         cd->u.procfs.content_ent = NULL;
1457
1458         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1459                              cd->u.procfs.proc_ent,
1460                              &cache_flush_operations_procfs, cd);
1461         cd->u.procfs.flush_ent = p;
1462         if (p == NULL)
1463                 goto out_nomem;
1464
1465         if (cd->cache_upcall || cd->cache_parse) {
1466                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1467                                      cd->u.procfs.proc_ent,
1468                                      &cache_file_operations_procfs, cd);
1469                 cd->u.procfs.channel_ent = p;
1470                 if (p == NULL)
1471                         goto out_nomem;
1472         }
1473         if (cd->cache_show) {
1474                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1475                                 cd->u.procfs.proc_ent,
1476                                 &content_file_operations_procfs, cd);
1477                 cd->u.procfs.content_ent = p;
1478                 if (p == NULL)
1479                         goto out_nomem;
1480         }
1481         return 0;
1482 out_nomem:
1483         remove_cache_proc_entries(cd);
1484         return -ENOMEM;
1485 }
1486 #else /* CONFIG_PROC_FS */
1487 static int create_cache_proc_entries(struct cache_detail *cd)
1488 {
1489         return 0;
1490 }
1491 #endif
1492
1493 int cache_register(struct cache_detail *cd)
1494 {
1495         int ret;
1496
1497         sunrpc_init_cache_detail(cd);
1498         ret = create_cache_proc_entries(cd);
1499         if (ret)
1500                 sunrpc_destroy_cache_detail(cd);
1501         return ret;
1502 }
1503 EXPORT_SYMBOL_GPL(cache_register);
1504
1505 void cache_unregister(struct cache_detail *cd)
1506 {
1507         remove_cache_proc_entries(cd);
1508         sunrpc_destroy_cache_detail(cd);
1509 }
1510 EXPORT_SYMBOL_GPL(cache_unregister);
1511
1512 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1513                                  size_t count, loff_t *ppos)
1514 {
1515         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1516
1517         return cache_read(filp, buf, count, ppos, cd);
1518 }
1519
1520 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1521                                   size_t count, loff_t *ppos)
1522 {
1523         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1524
1525         return cache_write(filp, buf, count, ppos, cd);
1526 }
1527
1528 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1529 {
1530         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1531
1532         return cache_poll(filp, wait, cd);
1533 }
1534
1535 static int cache_ioctl_pipefs(struct inode *inode, struct file *filp,
1536                               unsigned int cmd, unsigned long arg)
1537 {
1538         struct cache_detail *cd = RPC_I(inode)->private;
1539
1540         return cache_ioctl(inode, filp, cmd, arg, cd);
1541 }
1542
1543 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1544 {
1545         struct cache_detail *cd = RPC_I(inode)->private;
1546
1547         return cache_open(inode, filp, cd);
1548 }
1549
1550 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1551 {
1552         struct cache_detail *cd = RPC_I(inode)->private;
1553
1554         return cache_release(inode, filp, cd);
1555 }
1556
1557 const struct file_operations cache_file_operations_pipefs = {
1558         .owner          = THIS_MODULE,
1559         .llseek         = no_llseek,
1560         .read           = cache_read_pipefs,
1561         .write          = cache_write_pipefs,
1562         .poll           = cache_poll_pipefs,
1563         .ioctl          = cache_ioctl_pipefs, /* for FIONREAD */
1564         .open           = cache_open_pipefs,
1565         .release        = cache_release_pipefs,
1566 };
1567
1568 static int content_open_pipefs(struct inode *inode, struct file *filp)
1569 {
1570         struct cache_detail *cd = RPC_I(inode)->private;
1571
1572         return content_open(inode, filp, cd);
1573 }
1574
1575 static int content_release_pipefs(struct inode *inode, struct file *filp)
1576 {
1577         struct cache_detail *cd = RPC_I(inode)->private;
1578
1579         return content_release(inode, filp, cd);
1580 }
1581
1582 const struct file_operations content_file_operations_pipefs = {
1583         .open           = content_open_pipefs,
1584         .read           = seq_read,
1585         .llseek         = seq_lseek,
1586         .release        = content_release_pipefs,
1587 };
1588
1589 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1590 {
1591         struct cache_detail *cd = RPC_I(inode)->private;
1592
1593         return open_flush(inode, filp, cd);
1594 }
1595
1596 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1597 {
1598         struct cache_detail *cd = RPC_I(inode)->private;
1599
1600         return release_flush(inode, filp, cd);
1601 }
1602
1603 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1604                             size_t count, loff_t *ppos)
1605 {
1606         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1607
1608         return read_flush(filp, buf, count, ppos, cd);
1609 }
1610
1611 static ssize_t write_flush_pipefs(struct file *filp,
1612                                   const char __user *buf,
1613                                   size_t count, loff_t *ppos)
1614 {
1615         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1616
1617         return write_flush(filp, buf, count, ppos, cd);
1618 }
1619
1620 const struct file_operations cache_flush_operations_pipefs = {
1621         .open           = open_flush_pipefs,
1622         .read           = read_flush_pipefs,
1623         .write          = write_flush_pipefs,
1624         .release        = release_flush_pipefs,
1625 };
1626
1627 int sunrpc_cache_register_pipefs(struct dentry *parent,
1628                                  const char *name, mode_t umode,
1629                                  struct cache_detail *cd)
1630 {
1631         struct qstr q;
1632         struct dentry *dir;
1633         int ret = 0;
1634
1635         sunrpc_init_cache_detail(cd);
1636         q.name = name;
1637         q.len = strlen(name);
1638         q.hash = full_name_hash(q.name, q.len);
1639         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1640         if (!IS_ERR(dir))
1641                 cd->u.pipefs.dir = dir;
1642         else {
1643                 sunrpc_destroy_cache_detail(cd);
1644                 ret = PTR_ERR(dir);
1645         }
1646         return ret;
1647 }
1648 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1649
1650 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1651 {
1652         rpc_remove_cache_dir(cd->u.pipefs.dir);
1653         cd->u.pipefs.dir = NULL;
1654         sunrpc_destroy_cache_detail(cd);
1655 }
1656 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1657