kernel/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 <linux/string_helpers.h>
  24 #include <asm/uaccess.h>
  25 #include <linux/poll.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/net.h>
  29 #include <linux/workqueue.h>
  30 #include <linux/mutex.h>
  31 #include <linux/pagemap.h>
  32 #include <asm/ioctls.h>
  33 #include <linux/sunrpc/types.h>
  34 #include <linux/sunrpc/cache.h>
  35 #include <linux/sunrpc/stats.h>
  36 #include <linux/sunrpc/rpc_pipe_fs.h>
  37 #include "netns.h"
  38
  39 #define  RPCDBG_FACILITY RPCDBG_CACHE
  40
  41 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  42 static void cache_revisit_request(struct cache_head *item);
  43
  44 static void cache_init(struct cache_head *h)
  45 {
  46         time_t now = seconds_since_boot();
  47         h->next = NULL;
  48         h->flags = 0;
  49         kref_init(&h->ref);
  50         h->expiry_time = now + CACHE_NEW_EXPIRY;
  51         h->last_refresh = now;
  52 }
  53
  54 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  55                                        struct cache_head *key, int hash)
  56 {
  57         struct cache_head **head,  **hp;
  58         struct cache_head *new = NULL, *freeme = NULL;
  59
  60         head = &detail->hash_table[hash];
  61
  62         read_lock(&detail->hash_lock);
  63
  64         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  65                 struct cache_head *tmp = *hp;
  66                 if (detail->match(tmp, key)) {
  67                         if (cache_is_expired(detail, tmp))
  68                                 /* This entry is expired, we will discard it. */
  69                                 break;
  70                         cache_get(tmp);
  71                         read_unlock(&detail->hash_lock);
  72                         return tmp;
  73                 }
  74         }
  75         read_unlock(&detail->hash_lock);
  76         /* Didn't find anything, insert an empty entry */
  77
  78         new = detail->alloc();
  79         if (!new)
  80                 return NULL;
  81         /* must fully initialise 'new', else
  82          * we might get lose if we need to
  83          * cache_put it soon.
  84          */
  85         cache_init(new);
  86         detail->init(new, key);
  87
  88         write_lock(&detail->hash_lock);
  89
  90         /* check if entry appeared while we slept */
  91         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  92                 struct cache_head *tmp = *hp;
  93                 if (detail->match(tmp, key)) {
  94                         if (cache_is_expired(detail, tmp)) {
  95                                 *hp = tmp->next;
  96                                 tmp->next = NULL;
  97                                 detail->entries --;
  98                                 freeme = tmp;
  99                                 break;
 100                         }
 101                         cache_get(tmp);
 102                         write_unlock(&detail->hash_lock);
 103                         cache_put(new, detail);
 104                         return tmp;
 105                 }
 106         }
 107         new->next = *head;
 108         *head = new;
 109         detail->entries++;
 110         cache_get(new);
 111         write_unlock(&detail->hash_lock);
 112
 113         if (freeme)
 114                 cache_put(freeme, detail);
 115         return new;
 116 }
 117 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
 118
 119
 120 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 121
 122 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
 123 {
 124         head->expiry_time = expiry;
 125         head->last_refresh = seconds_since_boot();
 126         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 127         set_bit(CACHE_VALID, &head->flags);
 128 }
 129
 130 static void cache_fresh_unlocked(struct cache_head *head,
 131                                  struct cache_detail *detail)
 132 {
 133         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 134                 cache_revisit_request(head);
 135                 cache_dequeue(detail, head);
 136         }
 137 }
 138
 139 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 140                                        struct cache_head *new, struct cache_head *old, int hash)
 141 {
 142         /* The 'old' entry is to be replaced by 'new'.
 143          * If 'old' is not VALID, we update it directly,
 144          * otherwise we need to replace it
 145          */
 146         struct cache_head **head;
 147         struct cache_head *tmp;
 148
 149         if (!test_bit(CACHE_VALID, &old->flags)) {
 150                 write_lock(&detail->hash_lock);
 151                 if (!test_bit(CACHE_VALID, &old->flags)) {
 152                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 153                                 set_bit(CACHE_NEGATIVE, &old->flags);
 154                         else
 155                                 detail->update(old, new);
 156                         cache_fresh_locked(old, new->expiry_time);
 157                         write_unlock(&detail->hash_lock);
 158                         cache_fresh_unlocked(old, detail);
 159                         return old;
 160                 }
 161                 write_unlock(&detail->hash_lock);
 162         }
 163         /* We need to insert a new entry */
 164         tmp = detail->alloc();
 165         if (!tmp) {
 166                 cache_put(old, detail);
 167                 return NULL;
 168         }
 169         cache_init(tmp);
 170         detail->init(tmp, old);
 171         head = &detail->hash_table[hash];
 172
 173         write_lock(&detail->hash_lock);
 174         if (test_bit(CACHE_NEGATIVE, &new->flags))
 175                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 176         else
 177                 detail->update(tmp, new);
 178         tmp->next = *head;
 179         *head = tmp;
 180         detail->entries++;
 181         cache_get(tmp);
 182         cache_fresh_locked(tmp, new->expiry_time);
 183         cache_fresh_locked(old, 0);
 184         write_unlock(&detail->hash_lock);
 185         cache_fresh_unlocked(tmp, detail);
 186         cache_fresh_unlocked(old, detail);
 187         cache_put(old, detail);
 188         return tmp;
 189 }
 190 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 191
 192 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 193 {
 194         if (cd->cache_upcall)
 195                 return cd->cache_upcall(cd, h);
 196         return sunrpc_cache_pipe_upcall(cd, h);
 197 }
 198
 199 static inline int cache_is_valid(struct cache_head *h)
 200 {
 201         if (!test_bit(CACHE_VALID, &h->flags))
 202                 return -EAGAIN;
 203         else {
 204                 /* entry is valid */
 205                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 206                         return -ENOENT;
 207                 else {
 208                         /*
 209                          * In combination with write barrier in
 210                          * sunrpc_cache_update, ensures that anyone
 211                          * using the cache entry after this sees the
 212                          * updated contents:
 213                          */
 214                         smp_rmb();
 215                         return 0;
 216                 }
 217         }
 218 }
 219
 220 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 221 {
 222         int rv;
 223
 224         write_lock(&detail->hash_lock);
 225         rv = cache_is_valid(h);
 226         if (rv == -EAGAIN) {
 227                 set_bit(CACHE_NEGATIVE, &h->flags);
 228                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
 229                 rv = -ENOENT;
 230         }
 231         write_unlock(&detail->hash_lock);
 232         cache_fresh_unlocked(h, detail);
 233         return rv;
 234 }
 235
 236 /*
 237  * This is the generic cache management routine for all
 238  * the authentication caches.
 239  * It checks the currency of a cache item and will (later)
 240  * initiate an upcall to fill it if needed.
 241  *
 242  *
 243  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 244  * -EAGAIN if upcall is pending and request has been queued
 245  * -ETIMEDOUT if upcall failed or request could not be queue or
 246  *           upcall completed but item is still invalid (implying that
 247  *           the cache item has been replaced with a newer one).
 248  * -ENOENT if cache entry was negative
 249  */
 250 int cache_check(struct cache_detail *detail,
 251                     struct cache_head *h, struct cache_req *rqstp)
 252 {
 253         int rv;
 254         long refresh_age, age;
 255
 256         /* First decide return status as best we can */
 257         rv = cache_is_valid(h);
 258
 259         /* now see if we want to start an upcall */
 260         refresh_age = (h->expiry_time - h->last_refresh);
 261         age = seconds_since_boot() - h->last_refresh;
 262
 263         if (rqstp == NULL) {
 264                 if (rv == -EAGAIN)
 265                         rv = -ENOENT;
 266         } else if (rv == -EAGAIN ||
 267                    (h->expiry_time != 0 && age > refresh_age/2)) {
 268                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 269                                 refresh_age, age);
 270                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 271                         switch (cache_make_upcall(detail, h)) {
 272                         case -EINVAL:
 273                                 rv = try_to_negate_entry(detail, h);
 274                                 break;
 275                         case -EAGAIN:
 276                                 cache_fresh_unlocked(h, detail);
 277                                 break;
 278                         }
 279                 }
 280         }
 281
 282         if (rv == -EAGAIN) {
 283                 if (!cache_defer_req(rqstp, h)) {
 284                         /*
 285                          * Request was not deferred; handle it as best
 286                          * we can ourselves:
 287                          */
 288                         rv = cache_is_valid(h);
 289                         if (rv == -EAGAIN)
 290                                 rv = -ETIMEDOUT;
 291                 }
 292         }
 293         if (rv)
 294                 cache_put(h, detail);
 295         return rv;
 296 }
 297 EXPORT_SYMBOL_GPL(cache_check);
 298
 299 /*
 300  * caches need to be periodically cleaned.
 301  * For this we maintain a list of cache_detail and
 302  * a current pointer into that list and into the table
 303  * for that entry.
 304  *
 305  * Each time cache_clean is called it finds the next non-empty entry
 306  * in the current table and walks the list in that entry
 307  * looking for entries that can be removed.
 308  *
 309  * An entry gets removed if:
 310  * - The expiry is before current time
 311  * - The last_refresh time is before the flush_time for that cache
 312  *
 313  * later we might drop old entries with non-NEVER expiry if that table
 314  * is getting 'full' for some definition of 'full'
 315  *
 316  * The question of "how often to scan a table" is an interesting one
 317  * and is answered in part by the use of the "nextcheck" field in the
 318  * cache_detail.
 319  * When a scan of a table begins, the nextcheck field is set to a time
 320  * that is well into the future.
 321  * While scanning, if an expiry time is found that is earlier than the
 322  * current nextcheck time, nextcheck is set to that expiry time.
 323  * If the flush_time is ever set to a time earlier than the nextcheck
 324  * time, the nextcheck time is then set to that flush_time.
 325  *
 326  * A table is then only scanned if the current time is at least
 327  * the nextcheck time.
 328  *
 329  */
 330
 331 static LIST_HEAD(cache_list);
 332 static DEFINE_SPINLOCK(cache_list_lock);
 333 static struct cache_detail *current_detail;
 334 static int current_index;
 335
 336 static void do_cache_clean(struct work_struct *work);
 337 static struct delayed_work cache_cleaner;
 338
 339 void sunrpc_init_cache_detail(struct cache_detail *cd)
 340 {
 341         rwlock_init(&cd->hash_lock);
 342         INIT_LIST_HEAD(&cd->queue);
 343         spin_lock(&cache_list_lock);
 344         cd->nextcheck = 0;
 345         cd->entries = 0;
 346         atomic_set(&cd->readers, 0);
 347         cd->last_close = 0;
 348         cd->last_warn = -1;
 349         list_add(&cd->others, &cache_list);
 350         spin_unlock(&cache_list_lock);
 351
 352         /* start the cleaning process */
 353         schedule_delayed_work(&cache_cleaner, 0);
 354 }
 355 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 356
 357 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 358 {
 359         cache_purge(cd);
 360         spin_lock(&cache_list_lock);
 361         write_lock(&cd->hash_lock);
 362         if (cd->entries || atomic_read(&cd->inuse)) {
 363                 write_unlock(&cd->hash_lock);
 364                 spin_unlock(&cache_list_lock);
 365                 goto out;
 366         }
 367         if (current_detail == cd)
 368                 current_detail = NULL;
 369         list_del_init(&cd->others);
 370         write_unlock(&cd->hash_lock);
 371         spin_unlock(&cache_list_lock);
 372         if (list_empty(&cache_list)) {
 373                 /* module must be being unloaded so its safe to kill the worker */
 374                 cancel_delayed_work_sync(&cache_cleaner);
 375         }
 376         return;
 377 out:
 378         printk(KERN_ERR "RPC: failed to unregister %s cache\n", cd->name);
 379 }
 380 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 381
 382 /* clean cache tries to find something to clean
 383  * and cleans it.
 384  * It returns 1 if it cleaned something,
 385  *            0 if it didn't find anything this time
 386  *           -1 if it fell off the end of the list.
 387  */
 388 static int cache_clean(void)
 389 {
 390         int rv = 0;
 391         struct list_head *next;
 392
 393         spin_lock(&cache_list_lock);
 394
 395         /* find a suitable table if we don't already have one */
 396         while (current_detail == NULL ||
 397             current_index >= current_detail->hash_size) {
 398                 if (current_detail)
 399                         next = current_detail->others.next;
 400                 else
 401                         next = cache_list.next;
 402                 if (next == &cache_list) {
 403                         current_detail = NULL;
 404                         spin_unlock(&cache_list_lock);
 405                         return -1;
 406                 }
 407                 current_detail = list_entry(next, struct cache_detail, others);
 408                 if (current_detail->nextcheck > seconds_since_boot())
 409                         current_index = current_detail->hash_size;
 410                 else {
 411                         current_index = 0;
 412                         current_detail->nextcheck = seconds_since_boot()+30*60;
 413                 }
 414         }
 415
 416         /* find a non-empty bucket in the table */
 417         while (current_detail &&
 418                current_index < current_detail->hash_size &&
 419                current_detail->hash_table[current_index] == NULL)
 420                 current_index++;
 421
 422         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 423
 424         if (current_detail && current_index < current_detail->hash_size) {
 425                 struct cache_head *ch, **cp;
 426                 struct cache_detail *d;
 427
 428                 write_lock(&current_detail->hash_lock);
 429
 430                 /* Ok, now to clean this strand */
 431
 432                 cp = & current_detail->hash_table[current_index];
 433                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
 434                         if (current_detail->nextcheck > ch->expiry_time)
 435                                 current_detail->nextcheck = ch->expiry_time+1;
 436                         if (!cache_is_expired(current_detail, ch))
 437                                 continue;
 438
 439                         *cp = ch->next;
 440                         ch->next = NULL;
 441                         current_detail->entries--;
 442                         rv = 1;
 443                         break;
 444                 }
 445
 446                 write_unlock(&current_detail->hash_lock);
 447                 d = current_detail;
 448                 if (!ch)
 449                         current_index ++;
 450                 spin_unlock(&cache_list_lock);
 451                 if (ch) {
 452                         set_bit(CACHE_CLEANED, &ch->flags);
 453                         cache_fresh_unlocked(ch, d);
 454                         cache_put(ch, d);
 455                 }
 456         } else
 457                 spin_unlock(&cache_list_lock);
 458
 459         return rv;
 460 }
 461
 462 /*
 463  * We want to regularly clean the cache, so we need to schedule some work ...
 464  */
 465 static void do_cache_clean(struct work_struct *work)
 466 {
 467         int delay = 5;
 468         if (cache_clean() == -1)
 469                 delay = round_jiffies_relative(30*HZ);
 470
 471         if (list_empty(&cache_list))
 472                 delay = 0;
 473
 474         if (delay)
 475                 schedule_delayed_work(&cache_cleaner, delay);
 476 }
 477
 478
 479 /*
 480  * Clean all caches promptly.  This just calls cache_clean
 481  * repeatedly until we are sure that every cache has had a chance to
 482  * be fully cleaned
 483  */
 484 void cache_flush(void)
 485 {
 486         while (cache_clean() != -1)
 487                 cond_resched();
 488         while (cache_clean() != -1)
 489                 cond_resched();
 490 }
 491 EXPORT_SYMBOL_GPL(cache_flush);
 492
 493 void cache_purge(struct cache_detail *detail)
 494 {
 495         detail->flush_time = LONG_MAX;
 496         detail->nextcheck = seconds_since_boot();
 497         cache_flush();
 498         detail->flush_time = 1;
 499 }
 500 EXPORT_SYMBOL_GPL(cache_purge);
 501
 502
 503 /*
 504  * Deferral and Revisiting of Requests.
 505  *
 506  * If a cache lookup finds a pending entry, we
 507  * need to defer the request and revisit it later.
 508  * All deferred requests are stored in a hash table,
 509  * indexed by "struct cache_head *".
 510  * As it may be wasteful to store a whole request
 511  * structure, we allow the request to provide a
 512  * deferred form, which must contain a
 513  * 'struct cache_deferred_req'
 514  * This cache_deferred_req contains a method to allow
 515  * it to be revisited when cache info is available
 516  */
 517
 518 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 519 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 520
 521 #define DFR_MAX 300     /* ??? */
 522
 523 static DEFINE_SPINLOCK(cache_defer_lock);
 524 static LIST_HEAD(cache_defer_list);
 525 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 526 static int cache_defer_cnt;
 527
 528 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 529 {
 530         hlist_del_init(&dreq->hash);
 531         if (!list_empty(&dreq->recent)) {
 532                 list_del_init(&dreq->recent);
 533                 cache_defer_cnt--;
 534         }
 535 }
 536
 537 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 538 {
 539         int hash = DFR_HASH(item);
 540
 541         INIT_LIST_HEAD(&dreq->recent);
 542         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 543 }
 544
 545 static void setup_deferral(struct cache_deferred_req *dreq,
 546                            struct cache_head *item,
 547                            int count_me)
 548 {
 549
 550         dreq->item = item;
 551
 552         spin_lock(&cache_defer_lock);
 553
 554         __hash_deferred_req(dreq, item);
 555
 556         if (count_me) {
 557                 cache_defer_cnt++;
 558                 list_add(&dreq->recent, &cache_defer_list);
 559         }
 560
 561         spin_unlock(&cache_defer_lock);
 562
 563 }
 564
 565 struct thread_deferred_req {
 566         struct cache_deferred_req handle;
 567         struct completion completion;
 568 };
 569
 570 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 571 {
 572         struct thread_deferred_req *dr =
 573                 container_of(dreq, struct thread_deferred_req, handle);
 574         complete(&dr->completion);
 575 }
 576
 577 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 578 {
 579         struct thread_deferred_req sleeper;
 580         struct cache_deferred_req *dreq = &sleeper.handle;
 581
 582         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 583         dreq->revisit = cache_restart_thread;
 584
 585         setup_deferral(dreq, item, 0);
 586
 587         if (!test_bit(CACHE_PENDING, &item->flags) ||
 588             wait_for_completion_interruptible_timeout(
 589                     &sleeper.completion, req->thread_wait) <= 0) {
 590                 /* The completion wasn't completed, so we need
 591                  * to clean up
 592                  */
 593                 spin_lock(&cache_defer_lock);
 594                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 595                         __unhash_deferred_req(&sleeper.handle);
 596                         spin_unlock(&cache_defer_lock);
 597                 } else {
 598                         /* cache_revisit_request already removed
 599                          * this from the hash table, but hasn't
 600                          * called ->revisit yet.  It will very soon
 601                          * and we need to wait for it.
 602                          */
 603                         spin_unlock(&cache_defer_lock);
 604                         wait_for_completion(&sleeper.completion);
 605                 }
 606         }
 607 }
 608
 609 static void cache_limit_defers(void)
 610 {
 611         /* Make sure we haven't exceed the limit of allowed deferred
 612          * requests.
 613          */
 614         struct cache_deferred_req *discard = NULL;
 615
 616         if (cache_defer_cnt <= DFR_MAX)
 617                 return;
 618
 619         spin_lock(&cache_defer_lock);
 620
 621         /* Consider removing either the first or the last */
 622         if (cache_defer_cnt > DFR_MAX) {
 623                 if (prandom_u32() & 1)
 624                         discard = list_entry(cache_defer_list.next,
 625                                              struct cache_deferred_req, recent);
 626                 else
 627                         discard = list_entry(cache_defer_list.prev,
 628                                              struct cache_deferred_req, recent);
 629                 __unhash_deferred_req(discard);
 630         }
 631         spin_unlock(&cache_defer_lock);
 632         if (discard)
 633                 discard->revisit(discard, 1);
 634 }
 635
 636 /* Return true if and only if a deferred request is queued. */
 637 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 638 {
 639         struct cache_deferred_req *dreq;
 640
 641         if (req->thread_wait) {
 642                 cache_wait_req(req, item);
 643                 if (!test_bit(CACHE_PENDING, &item->flags))
 644                         return false;
 645         }
 646         dreq = req->defer(req);
 647         if (dreq == NULL)
 648                 return false;
 649         setup_deferral(dreq, item, 1);
 650         if (!test_bit(CACHE_PENDING, &item->flags))
 651                 /* Bit could have been cleared before we managed to
 652                  * set up the deferral, so need to revisit just in case
 653                  */
 654                 cache_revisit_request(item);
 655
 656         cache_limit_defers();
 657         return true;
 658 }
 659
 660 static void cache_revisit_request(struct cache_head *item)
 661 {
 662         struct cache_deferred_req *dreq;
 663         struct list_head pending;
 664         struct hlist_node *tmp;
 665         int hash = DFR_HASH(item);
 666
 667         INIT_LIST_HEAD(&pending);
 668         spin_lock(&cache_defer_lock);
 669
 670         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 671                 if (dreq->item == item) {
 672                         __unhash_deferred_req(dreq);
 673                         list_add(&dreq->recent, &pending);
 674                 }
 675
 676         spin_unlock(&cache_defer_lock);
 677
 678         while (!list_empty(&pending)) {
 679                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 680                 list_del_init(&dreq->recent);
 681                 dreq->revisit(dreq, 0);
 682         }
 683 }
 684
 685 void cache_clean_deferred(void *owner)
 686 {
 687         struct cache_deferred_req *dreq, *tmp;
 688         struct list_head pending;
 689
 690
 691         INIT_LIST_HEAD(&pending);
 692         spin_lock(&cache_defer_lock);
 693
 694         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 695                 if (dreq->owner == owner) {
 696                         __unhash_deferred_req(dreq);
 697                         list_add(&dreq->recent, &pending);
 698                 }
 699         }
 700         spin_unlock(&cache_defer_lock);
 701
 702         while (!list_empty(&pending)) {
 703                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 704                 list_del_init(&dreq->recent);
 705                 dreq->revisit(dreq, 1);
 706         }
 707 }
 708
 709 /*
 710  * communicate with user-space
 711  *
 712  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 713  * On read, you get a full request, or block.
 714  * On write, an update request is processed.
 715  * Poll works if anything to read, and always allows write.
 716  *
 717  * Implemented by linked list of requests.  Each open file has
 718  * a ->private that also exists in this list.  New requests are added
 719  * to the end and may wakeup and preceding readers.
 720  * New readers are added to the head.  If, on read, an item is found with
 721  * CACHE_UPCALLING clear, we free it from the list.
 722  *
 723  */
 724
 725 static DEFINE_SPINLOCK(queue_lock);
 726 static DEFINE_MUTEX(queue_io_mutex);
 727
 728 struct cache_queue {
 729         struct list_head        list;
 730         int                     reader; /* if 0, then request */
 731 };
 732 struct cache_request {
 733         struct cache_queue      q;
 734         struct cache_head       *item;
 735         char                    * buf;
 736         int                     len;
 737         int                     readers;
 738 };
 739 struct cache_reader {
 740         struct cache_queue      q;
 741         int                     offset; /* if non-0, we have a refcnt on next request */
 742 };
 743
 744 static int cache_request(struct cache_detail *detail,
 745                                struct cache_request *crq)
 746 {
 747         char *bp = crq->buf;
 748         int len = PAGE_SIZE;
 749
 750         detail->cache_request(detail, crq->item, &bp, &len);
 751         if (len < 0)
 752                 return -EAGAIN;
 753         return PAGE_SIZE - len;
 754 }
 755
 756 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 757                           loff_t *ppos, struct cache_detail *cd)
 758 {
 759         struct cache_reader *rp = filp->private_data;
 760         struct cache_request *rq;
 761         struct inode *inode = file_inode(filp);
 762         int err;
 763
 764         if (count == 0)
 765                 return 0;
 766
 767         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 768                               * readers on this file */
 769  again:
 770         spin_lock(&queue_lock);
 771         /* need to find next request */
 772         while (rp->q.list.next != &cd->queue &&
 773                list_entry(rp->q.list.next, struct cache_queue, list)
 774                ->reader) {
 775                 struct list_head *next = rp->q.list.next;
 776                 list_move(&rp->q.list, next);
 777         }
 778         if (rp->q.list.next == &cd->queue) {
 779                 spin_unlock(&queue_lock);
 780                 mutex_unlock(&inode->i_mutex);
 781                 WARN_ON_ONCE(rp->offset);
 782                 return 0;
 783         }
 784         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 785         WARN_ON_ONCE(rq->q.reader);
 786         if (rp->offset == 0)
 787                 rq->readers++;
 788         spin_unlock(&queue_lock);
 789
 790         if (rq->len == 0) {
 791                 err = cache_request(cd, rq);
 792                 if (err < 0)
 793                         goto out;
 794                 rq->len = err;
 795         }
 796
 797         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 798                 err = -EAGAIN;
 799                 spin_lock(&queue_lock);
 800                 list_move(&rp->q.list, &rq->q.list);
 801                 spin_unlock(&queue_lock);
 802         } else {
 803                 if (rp->offset + count > rq->len)
 804                         count = rq->len - rp->offset;
 805                 err = -EFAULT;
 806                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 807                         goto out;
 808                 rp->offset += count;
 809                 if (rp->offset >= rq->len) {
 810                         rp->offset = 0;
 811                         spin_lock(&queue_lock);
 812                         list_move(&rp->q.list, &rq->q.list);
 813                         spin_unlock(&queue_lock);
 814                 }
 815                 err = 0;
 816         }
 817  out:
 818         if (rp->offset == 0) {
 819                 /* need to release rq */
 820                 spin_lock(&queue_lock);
 821                 rq->readers--;
 822                 if (rq->readers == 0 &&
 823                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 824                         list_del(&rq->q.list);
 825                         spin_unlock(&queue_lock);
 826                         cache_put(rq->item, cd);
 827                         kfree(rq->buf);
 828                         kfree(rq);
 829                 } else
 830                         spin_unlock(&queue_lock);
 831         }
 832         if (err == -EAGAIN)
 833                 goto again;
 834         mutex_unlock(&inode->i_mutex);
 835         return err ? err :  count;
 836 }
 837
 838 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 839                                  size_t count, struct cache_detail *cd)
 840 {
 841         ssize_t ret;
 842
 843         if (count == 0)
 844                 return -EINVAL;
 845         if (copy_from_user(kaddr, buf, count))
 846                 return -EFAULT;
 847         kaddr[count] = '\0';
 848         ret = cd->cache_parse(cd, kaddr, count);
 849         if (!ret)
 850                 ret = count;
 851         return ret;
 852 }
 853
 854 static ssize_t cache_slow_downcall(const char __user *buf,
 855                                    size_t count, struct cache_detail *cd)
 856 {
 857         static char write_buf[8192]; /* protected by queue_io_mutex */
 858         ssize_t ret = -EINVAL;
 859
 860         if (count >= sizeof(write_buf))
 861                 goto out;
 862         mutex_lock(&queue_io_mutex);
 863         ret = cache_do_downcall(write_buf, buf, count, cd);
 864         mutex_unlock(&queue_io_mutex);
 865 out:
 866         return ret;
 867 }
 868
 869 static ssize_t cache_downcall(struct address_space *mapping,
 870                               const char __user *buf,
 871                               size_t count, struct cache_detail *cd)
 872 {
 873         struct page *page;
 874         char *kaddr;
 875         ssize_t ret = -ENOMEM;
 876
 877         if (count >= PAGE_CACHE_SIZE)
 878                 goto out_slow;
 879
 880         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 881         if (!page)
 882                 goto out_slow;
 883
 884         kaddr = kmap(page);
 885         ret = cache_do_downcall(kaddr, buf, count, cd);
 886         kunmap(page);
 887         unlock_page(page);
 888         page_cache_release(page);
 889         return ret;
 890 out_slow:
 891         return cache_slow_downcall(buf, count, cd);
 892 }
 893
 894 static ssize_t cache_write(struct file *filp, const char __user *buf,
 895                            size_t count, loff_t *ppos,
 896                            struct cache_detail *cd)
 897 {
 898         struct address_space *mapping = filp->f_mapping;
 899         struct inode *inode = file_inode(filp);
 900         ssize_t ret = -EINVAL;
 901
 902         if (!cd->cache_parse)
 903                 goto out;
 904
 905         mutex_lock(&inode->i_mutex);
 906         ret = cache_downcall(mapping, buf, count, cd);
 907         mutex_unlock(&inode->i_mutex);
 908 out:
 909         return ret;
 910 }
 911
 912 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 913
 914 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 915                                struct cache_detail *cd)
 916 {
 917         unsigned int mask;
 918         struct cache_reader *rp = filp->private_data;
 919         struct cache_queue *cq;
 920
 921         poll_wait(filp, &queue_wait, wait);
 922
 923         /* alway allow write */
 924         mask = POLLOUT | POLLWRNORM;
 925
 926         if (!rp)
 927                 return mask;
 928
 929         spin_lock(&queue_lock);
 930
 931         for (cq= &rp->q; &cq->list != &cd->queue;
 932              cq = list_entry(cq->list.next, struct cache_queue, list))
 933                 if (!cq->reader) {
 934                         mask |= POLLIN | POLLRDNORM;
 935                         break;
 936                 }
 937         spin_unlock(&queue_lock);
 938         return mask;
 939 }
 940
 941 static int cache_ioctl(struct inode *ino, struct file *filp,
 942                        unsigned int cmd, unsigned long arg,
 943                        struct cache_detail *cd)
 944 {
 945         int len = 0;
 946         struct cache_reader *rp = filp->private_data;
 947         struct cache_queue *cq;
 948
 949         if (cmd != FIONREAD || !rp)
 950                 return -EINVAL;
 951
 952         spin_lock(&queue_lock);
 953
 954         /* only find the length remaining in current request,
 955          * or the length of the next request
 956          */
 957         for (cq= &rp->q; &cq->list != &cd->queue;
 958              cq = list_entry(cq->list.next, struct cache_queue, list))
 959                 if (!cq->reader) {
 960                         struct cache_request *cr =
 961                                 container_of(cq, struct cache_request, q);
 962                         len = cr->len - rp->offset;
 963                         break;
 964                 }
 965         spin_unlock(&queue_lock);
 966
 967         return put_user(len, (int __user *)arg);
 968 }
 969
 970 static int cache_open(struct inode *inode, struct file *filp,
 971                       struct cache_detail *cd)
 972 {
 973         struct cache_reader *rp = NULL;
 974
 975         if (!cd || !try_module_get(cd->owner))
 976                 return -EACCES;
 977         nonseekable_open(inode, filp);
 978         if (filp->f_mode & FMODE_READ) {
 979                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 980                 if (!rp) {
 981                         module_put(cd->owner);
 982                         return -ENOMEM;
 983                 }
 984                 rp->offset = 0;
 985                 rp->q.reader = 1;
 986                 atomic_inc(&cd->readers);
 987                 spin_lock(&queue_lock);
 988                 list_add(&rp->q.list, &cd->queue);
 989                 spin_unlock(&queue_lock);
 990         }
 991         filp->private_data = rp;
 992         return 0;
 993 }
 994
 995 static int cache_release(struct inode *inode, struct file *filp,
 996                          struct cache_detail *cd)
 997 {
 998         struct cache_reader *rp = filp->private_data;
 999
1000         if (rp) {
1001                 spin_lock(&queue_lock);
1002                 if (rp->offset) {
1003                         struct cache_queue *cq;
1004                         for (cq= &rp->q; &cq->list != &cd->queue;
1005                              cq = list_entry(cq->list.next, struct cache_queue, list))
1006                                 if (!cq->reader) {
1007                                         container_of(cq, struct cache_request, q)
1008                                                 ->readers--;
1009                                         break;
1010                                 }
1011                         rp->offset = 0;
1012                 }
1013                 list_del(&rp->q.list);
1014                 spin_unlock(&queue_lock);
1015
1016                 filp->private_data = NULL;
1017                 kfree(rp);
1018
1019                 cd->last_close = seconds_since_boot();
1020                 atomic_dec(&cd->readers);
1021         }
1022         module_put(cd->owner);
1023         return 0;
1024 }
1025
1026
1027
1028 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1029 {
1030         struct cache_queue *cq, *tmp;
1031         struct cache_request *cr;
1032         struct list_head dequeued;
1033
1034         INIT_LIST_HEAD(&dequeued);
1035         spin_lock(&queue_lock);
1036         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1037                 if (!cq->reader) {
1038                         cr = container_of(cq, struct cache_request, q);
1039                         if (cr->item != ch)
1040                                 continue;
1041                         if (test_bit(CACHE_PENDING, &ch->flags))
1042                                 /* Lost a race and it is pending again */
1043                                 break;
1044                         if (cr->readers != 0)
1045                                 continue;
1046                         list_move(&cr->q.list, &dequeued);
1047                 }
1048         spin_unlock(&queue_lock);
1049         while (!list_empty(&dequeued)) {
1050                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1051                 list_del(&cr->q.list);
1052                 cache_put(cr->item, detail);
1053                 kfree(cr->buf);
1054                 kfree(cr);
1055         }
1056 }
1057
1058 /*
1059  * Support routines for text-based upcalls.
1060  * Fields are separated by spaces.
1061  * Fields are either mangled to quote space tab newline slosh with slosh
1062  * or a hexified with a leading \x
1063  * Record is terminated with newline.
1064  *
1065  */
1066
1067 void qword_add(char **bpp, int *lp, char *str)
1068 {
1069         char *bp = *bpp;
1070         int len = *lp;
1071         int ret;
1072
1073         if (len < 0) return;
1074
1075         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1076         if (ret >= len) {
1077                 bp += len;
1078                 len = -1;
1079         } else {
1080                 bp += ret;
1081                 len -= ret;
1082                 *bp++ = ' ';
1083                 len--;
1084         }
1085         *bpp = bp;
1086         *lp = len;
1087 }
1088 EXPORT_SYMBOL_GPL(qword_add);
1089
1090 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1091 {
1092         char *bp = *bpp;
1093         int len = *lp;
1094
1095         if (len < 0) return;
1096
1097         if (len > 2) {
1098                 *bp++ = '\\';
1099                 *bp++ = 'x';
1100                 len -= 2;
1101                 while (blen && len >= 2) {
1102                         bp = hex_byte_pack(bp, *buf++);
1103                         len -= 2;
1104                         blen--;
1105                 }
1106         }
1107         if (blen || len<1) len = -1;
1108         else {
1109                 *bp++ = ' ';
1110                 len--;
1111         }
1112         *bpp = bp;
1113         *lp = len;
1114 }
1115 EXPORT_SYMBOL_GPL(qword_addhex);
1116
1117 static void warn_no_listener(struct cache_detail *detail)
1118 {
1119         if (detail->last_warn != detail->last_close) {
1120                 detail->last_warn = detail->last_close;
1121                 if (detail->warn_no_listener)
1122                         detail->warn_no_listener(detail, detail->last_close != 0);
1123         }
1124 }
1125
1126 static bool cache_listeners_exist(struct cache_detail *detail)
1127 {
1128         if (atomic_read(&detail->readers))
1129                 return true;
1130         if (detail->last_close == 0)
1131                 /* This cache was never opened */
1132                 return false;
1133         if (detail->last_close < seconds_since_boot() - 30)
1134                 /*
1135                  * We allow for the possibility that someone might
1136                  * restart a userspace daemon without restarting the
1137                  * server; but after 30 seconds, we give up.
1138                  */
1139                  return false;
1140         return true;
1141 }
1142
1143 /*
1144  * register an upcall request to user-space and queue it up for read() by the
1145  * upcall daemon.
1146  *
1147  * Each request is at most one page long.
1148  */
1149 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1150 {
1151
1152         char *buf;
1153         struct cache_request *crq;
1154         int ret = 0;
1155
1156         if (!detail->cache_request)
1157                 return -EINVAL;
1158
1159         if (!cache_listeners_exist(detail)) {
1160                 warn_no_listener(detail);
1161                 return -EINVAL;
1162         }
1163         if (test_bit(CACHE_CLEANED, &h->flags))
1164                 /* Too late to make an upcall */
1165                 return -EAGAIN;
1166
1167         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1168         if (!buf)
1169                 return -EAGAIN;
1170
1171         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1172         if (!crq) {
1173                 kfree(buf);
1174                 return -EAGAIN;
1175         }
1176
1177         crq->q.reader = 0;
1178         crq->item = cache_get(h);
1179         crq->buf = buf;
1180         crq->len = 0;
1181         crq->readers = 0;
1182         spin_lock(&queue_lock);
1183         if (test_bit(CACHE_PENDING, &h->flags))
1184                 list_add_tail(&crq->q.list, &detail->queue);
1185         else
1186                 /* Lost a race, no longer PENDING, so don't enqueue */
1187                 ret = -EAGAIN;
1188         spin_unlock(&queue_lock);
1189         wake_up(&queue_wait);
1190         if (ret == -EAGAIN) {
1191                 kfree(buf);
1192                 kfree(crq);
1193         }
1194         return ret;
1195 }
1196 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1197
1198 /*
1199  * parse a message from user-space and pass it
1200  * to an appropriate cache
1201  * Messages are, like requests, separated into fields by
1202  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1203  *
1204  * Message is
1205  *   reply cachename expiry key ... content....
1206  *
1207  * key and content are both parsed by cache
1208  */
1209
1210 int qword_get(char **bpp, char *dest, int bufsize)
1211 {
1212         /* return bytes copied, or -1 on error */
1213         char *bp = *bpp;
1214         int len = 0;
1215
1216         while (*bp == ' ') bp++;
1217
1218         if (bp[0] == '\\' && bp[1] == 'x') {
1219                 /* HEX STRING */
1220                 bp += 2;
1221                 while (len < bufsize) {
1222                         int h, l;
1223
1224                         h = hex_to_bin(bp[0]);
1225                         if (h < 0)
1226                                 break;
1227
1228                         l = hex_to_bin(bp[1]);
1229                         if (l < 0)
1230                                 break;
1231
1232                         *dest++ = (h << 4) | l;
1233                         bp += 2;
1234                         len++;
1235                 }
1236         } else {
1237                 /* text with \nnn octal quoting */
1238                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1239                         if (*bp == '\\' &&
1240                             isodigit(bp[1]) && (bp[1] <= '3') &&
1241                             isodigit(bp[2]) &&
1242                             isodigit(bp[3])) {
1243                                 int byte = (*++bp -'0');
1244                                 bp++;
1245                                 byte = (byte << 3) | (*bp++ - '0');
1246                                 byte = (byte << 3) | (*bp++ - '0');
1247                                 *dest++ = byte;
1248                                 len++;
1249                         } else {
1250                                 *dest++ = *bp++;
1251                                 len++;
1252                         }
1253                 }
1254         }
1255
1256         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1257                 return -1;
1258         while (*bp == ' ') bp++;
1259         *bpp = bp;
1260         *dest = '\0';
1261         return len;
1262 }
1263 EXPORT_SYMBOL_GPL(qword_get);
1264
1265
1266 /*
1267  * support /proc/sunrpc/cache/$CACHENAME/content
1268  * as a seqfile.
1269  * We call ->cache_show passing NULL for the item to
1270  * get a header, then pass each real item in the cache
1271  */
1272
1273 struct handle {
1274         struct cache_detail *cd;
1275 };
1276
1277 static void *c_start(struct seq_file *m, loff_t *pos)
1278         __acquires(cd->hash_lock)
1279 {
1280         loff_t n = *pos;
1281         unsigned int hash, entry;
1282         struct cache_head *ch;
1283         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1284
1285
1286         read_lock(&cd->hash_lock);
1287         if (!n--)
1288                 return SEQ_START_TOKEN;
1289         hash = n >> 32;
1290         entry = n & ((1LL<<32) - 1);
1291
1292         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1293                 if (!entry--)
1294                         return ch;
1295         n &= ~((1LL<<32) - 1);
1296         do {
1297                 hash++;
1298                 n += 1LL<<32;
1299         } while(hash < cd->hash_size &&
1300                 cd->hash_table[hash]==NULL);
1301         if (hash >= cd->hash_size)
1302                 return NULL;
1303         *pos = n+1;
1304         return cd->hash_table[hash];
1305 }
1306
1307 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1308 {
1309         struct cache_head *ch = p;
1310         int hash = (*pos >> 32);
1311         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1312
1313         if (p == SEQ_START_TOKEN)
1314                 hash = 0;
1315         else if (ch->next == NULL) {
1316                 hash++;
1317                 *pos += 1LL<<32;
1318         } else {
1319                 ++*pos;
1320                 return ch->next;
1321         }
1322         *pos &= ~((1LL<<32) - 1);
1323         while (hash < cd->hash_size &&
1324                cd->hash_table[hash] == NULL) {
1325                 hash++;
1326                 *pos += 1LL<<32;
1327         }
1328         if (hash >= cd->hash_size)
1329                 return NULL;
1330         ++*pos;
1331         return cd->hash_table[hash];
1332 }
1333
1334 static void c_stop(struct seq_file *m, void *p)
1335         __releases(cd->hash_lock)
1336 {
1337         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1338         read_unlock(&cd->hash_lock);
1339 }
1340
1341 static int c_show(struct seq_file *m, void *p)
1342 {
1343         struct cache_head *cp = p;
1344         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1345
1346         if (p == SEQ_START_TOKEN)
1347                 return cd->cache_show(m, cd, NULL);
1348
1349         ifdebug(CACHE)
1350                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1351                            convert_to_wallclock(cp->expiry_time),
1352                            atomic_read(&cp->ref.refcount), cp->flags);
1353         cache_get(cp);
1354         if (cache_check(cd, cp, NULL))
1355                 /* cache_check does a cache_put on failure */
1356                 seq_printf(m, "# ");
1357         else {
1358                 if (cache_is_expired(cd, cp))
1359                         seq_printf(m, "# ");
1360                 cache_put(cp, cd);
1361         }
1362
1363         return cd->cache_show(m, cd, cp);
1364 }
1365
1366 static const struct seq_operations cache_content_op = {
1367         .start  = c_start,
1368         .next   = c_next,
1369         .stop   = c_stop,
1370         .show   = c_show,
1371 };
1372
1373 static int content_open(struct inode *inode, struct file *file,
1374                         struct cache_detail *cd)
1375 {
1376         struct handle *han;
1377
1378         if (!cd || !try_module_get(cd->owner))
1379                 return -EACCES;
1380         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1381         if (han == NULL) {
1382                 module_put(cd->owner);
1383                 return -ENOMEM;
1384         }
1385
1386         han->cd = cd;
1387         return 0;
1388 }
1389
1390 static int content_release(struct inode *inode, struct file *file,
1391                 struct cache_detail *cd)
1392 {
1393         int ret = seq_release_private(inode, file);
1394         module_put(cd->owner);
1395         return ret;
1396 }
1397
1398 static int open_flush(struct inode *inode, struct file *file,
1399                         struct cache_detail *cd)
1400 {
1401         if (!cd || !try_module_get(cd->owner))
1402                 return -EACCES;
1403         return nonseekable_open(inode, file);
1404 }
1405
1406 static int release_flush(struct inode *inode, struct file *file,
1407                         struct cache_detail *cd)
1408 {
1409         module_put(cd->owner);
1410         return 0;
1411 }
1412
1413 static ssize_t read_flush(struct file *file, char __user *buf,
1414                           size_t count, loff_t *ppos,
1415                           struct cache_detail *cd)
1416 {
1417         char tbuf[22];
1418         unsigned long p = *ppos;
1419         size_t len;
1420
1421         snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1422         len = strlen(tbuf);
1423         if (p >= len)
1424                 return 0;
1425         len -= p;
1426         if (len > count)
1427                 len = count;
1428         if (copy_to_user(buf, (void*)(tbuf+p), len))
1429                 return -EFAULT;
1430         *ppos += len;
1431         return len;
1432 }
1433
1434 static ssize_t write_flush(struct file *file, const char __user *buf,
1435                            size_t count, loff_t *ppos,
1436                            struct cache_detail *cd)
1437 {
1438         char tbuf[20];
1439         char *bp, *ep;
1440
1441         if (*ppos || count > sizeof(tbuf)-1)
1442                 return -EINVAL;
1443         if (copy_from_user(tbuf, buf, count))
1444                 return -EFAULT;
1445         tbuf[count] = 0;
1446         simple_strtoul(tbuf, &ep, 0);
1447         if (*ep && *ep != '\n')
1448                 return -EINVAL;
1449
1450         bp = tbuf;
1451         cd->flush_time = get_expiry(&bp);
1452         cd->nextcheck = seconds_since_boot();
1453         cache_flush();
1454
1455         *ppos += count;
1456         return count;
1457 }
1458
1459 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1460                                  size_t count, loff_t *ppos)
1461 {
1462         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1463
1464         return cache_read(filp, buf, count, ppos, cd);
1465 }
1466
1467 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1468                                   size_t count, loff_t *ppos)
1469 {
1470         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1471
1472         return cache_write(filp, buf, count, ppos, cd);
1473 }
1474
1475 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1476 {
1477         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1478
1479         return cache_poll(filp, wait, cd);
1480 }
1481
1482 static long cache_ioctl_procfs(struct file *filp,
1483                                unsigned int cmd, unsigned long arg)
1484 {
1485         struct inode *inode = file_inode(filp);
1486         struct cache_detail *cd = PDE_DATA(inode);
1487
1488         return cache_ioctl(inode, filp, cmd, arg, cd);
1489 }
1490
1491 static int cache_open_procfs(struct inode *inode, struct file *filp)
1492 {
1493         struct cache_detail *cd = PDE_DATA(inode);
1494
1495         return cache_open(inode, filp, cd);
1496 }
1497
1498 static int cache_release_procfs(struct inode *inode, struct file *filp)
1499 {
1500         struct cache_detail *cd = PDE_DATA(inode);
1501
1502         return cache_release(inode, filp, cd);
1503 }
1504
1505 static const struct file_operations cache_file_operations_procfs = {
1506         .owner          = THIS_MODULE,
1507         .llseek         = no_llseek,
1508         .read           = cache_read_procfs,
1509         .write          = cache_write_procfs,
1510         .poll           = cache_poll_procfs,
1511         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1512         .open           = cache_open_procfs,
1513         .release        = cache_release_procfs,
1514 };
1515
1516 static int content_open_procfs(struct inode *inode, struct file *filp)
1517 {
1518         struct cache_detail *cd = PDE_DATA(inode);
1519
1520         return content_open(inode, filp, cd);
1521 }
1522
1523 static int content_release_procfs(struct inode *inode, struct file *filp)
1524 {
1525         struct cache_detail *cd = PDE_DATA(inode);
1526
1527         return content_release(inode, filp, cd);
1528 }
1529
1530 static const struct file_operations content_file_operations_procfs = {
1531         .open           = content_open_procfs,
1532         .read           = seq_read,
1533         .llseek         = seq_lseek,
1534         .release        = content_release_procfs,
1535 };
1536
1537 static int open_flush_procfs(struct inode *inode, struct file *filp)
1538 {
1539         struct cache_detail *cd = PDE_DATA(inode);
1540
1541         return open_flush(inode, filp, cd);
1542 }
1543
1544 static int release_flush_procfs(struct inode *inode, struct file *filp)
1545 {
1546         struct cache_detail *cd = PDE_DATA(inode);
1547
1548         return release_flush(inode, filp, cd);
1549 }
1550
1551 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1552                             size_t count, loff_t *ppos)
1553 {
1554         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1555
1556         return read_flush(filp, buf, count, ppos, cd);
1557 }
1558
1559 static ssize_t write_flush_procfs(struct file *filp,
1560                                   const char __user *buf,
1561                                   size_t count, loff_t *ppos)
1562 {
1563         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1564
1565         return write_flush(filp, buf, count, ppos, cd);
1566 }
1567
1568 static const struct file_operations cache_flush_operations_procfs = {
1569         .open           = open_flush_procfs,
1570         .read           = read_flush_procfs,
1571         .write          = write_flush_procfs,
1572         .release        = release_flush_procfs,
1573         .llseek         = no_llseek,
1574 };
1575
1576 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1577 {
1578         struct sunrpc_net *sn;
1579
1580         if (cd->u.procfs.proc_ent == NULL)
1581                 return;
1582         if (cd->u.procfs.flush_ent)
1583                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1584         if (cd->u.procfs.channel_ent)
1585                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1586         if (cd->u.procfs.content_ent)
1587                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1588         cd->u.procfs.proc_ent = NULL;
1589         sn = net_generic(net, sunrpc_net_id);
1590         remove_proc_entry(cd->name, sn->proc_net_rpc);
1591 }
1592
1593 #ifdef CONFIG_PROC_FS
1594 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1595 {
1596         struct proc_dir_entry *p;
1597         struct sunrpc_net *sn;
1598
1599         sn = net_generic(net, sunrpc_net_id);
1600         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1601         if (cd->u.procfs.proc_ent == NULL)
1602                 goto out_nomem;
1603         cd->u.procfs.channel_ent = NULL;
1604         cd->u.procfs.content_ent = NULL;
1605
1606         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1607                              cd->u.procfs.proc_ent,
1608                              &cache_flush_operations_procfs, cd);
1609         cd->u.procfs.flush_ent = p;
1610         if (p == NULL)
1611                 goto out_nomem;
1612
1613         if (cd->cache_request || cd->cache_parse) {
1614                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1615                                      cd->u.procfs.proc_ent,
1616                                      &cache_file_operations_procfs, cd);
1617                 cd->u.procfs.channel_ent = p;
1618                 if (p == NULL)
1619                         goto out_nomem;
1620         }
1621         if (cd->cache_show) {
1622                 p = proc_create_data("content", S_IFREG|S_IRUSR,
1623                                 cd->u.procfs.proc_ent,
1624                                 &content_file_operations_procfs, cd);
1625                 cd->u.procfs.content_ent = p;
1626                 if (p == NULL)
1627                         goto out_nomem;
1628         }
1629         return 0;
1630 out_nomem:
1631         remove_cache_proc_entries(cd, net);
1632         return -ENOMEM;
1633 }
1634 #else /* CONFIG_PROC_FS */
1635 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1636 {
1637         return 0;
1638 }
1639 #endif
1640
1641 void __init cache_initialize(void)
1642 {
1643         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1644 }
1645
1646 int cache_register_net(struct cache_detail *cd, struct net *net)
1647 {
1648         int ret;
1649
1650         sunrpc_init_cache_detail(cd);
1651         ret = create_cache_proc_entries(cd, net);
1652         if (ret)
1653                 sunrpc_destroy_cache_detail(cd);
1654         return ret;
1655 }
1656 EXPORT_SYMBOL_GPL(cache_register_net);
1657
1658 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1659 {
1660         remove_cache_proc_entries(cd, net);
1661         sunrpc_destroy_cache_detail(cd);
1662 }
1663 EXPORT_SYMBOL_GPL(cache_unregister_net);
1664
1665 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1666 {
1667         struct cache_detail *cd;
1668
1669         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1670         if (cd == NULL)
1671                 return ERR_PTR(-ENOMEM);
1672
1673         cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1674                                  GFP_KERNEL);
1675         if (cd->hash_table == NULL) {
1676                 kfree(cd);
1677                 return ERR_PTR(-ENOMEM);
1678         }
1679         cd->net = net;
1680         return cd;
1681 }
1682 EXPORT_SYMBOL_GPL(cache_create_net);
1683
1684 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1685 {
1686         kfree(cd->hash_table);
1687         kfree(cd);
1688 }
1689 EXPORT_SYMBOL_GPL(cache_destroy_net);
1690
1691 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1692                                  size_t count, loff_t *ppos)
1693 {
1694         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1695
1696         return cache_read(filp, buf, count, ppos, cd);
1697 }
1698
1699 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1700                                   size_t count, loff_t *ppos)
1701 {
1702         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1703
1704         return cache_write(filp, buf, count, ppos, cd);
1705 }
1706
1707 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1708 {
1709         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1710
1711         return cache_poll(filp, wait, cd);
1712 }
1713
1714 static long cache_ioctl_pipefs(struct file *filp,
1715                               unsigned int cmd, unsigned long arg)
1716 {
1717         struct inode *inode = file_inode(filp);
1718         struct cache_detail *cd = RPC_I(inode)->private;
1719
1720         return cache_ioctl(inode, filp, cmd, arg, cd);
1721 }
1722
1723 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1724 {
1725         struct cache_detail *cd = RPC_I(inode)->private;
1726
1727         return cache_open(inode, filp, cd);
1728 }
1729
1730 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1731 {
1732         struct cache_detail *cd = RPC_I(inode)->private;
1733
1734         return cache_release(inode, filp, cd);
1735 }
1736
1737 const struct file_operations cache_file_operations_pipefs = {
1738         .owner          = THIS_MODULE,
1739         .llseek         = no_llseek,
1740         .read           = cache_read_pipefs,
1741         .write          = cache_write_pipefs,
1742         .poll           = cache_poll_pipefs,
1743         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1744         .open           = cache_open_pipefs,
1745         .release        = cache_release_pipefs,
1746 };
1747
1748 static int content_open_pipefs(struct inode *inode, struct file *filp)
1749 {
1750         struct cache_detail *cd = RPC_I(inode)->private;
1751
1752         return content_open(inode, filp, cd);
1753 }
1754
1755 static int content_release_pipefs(struct inode *inode, struct file *filp)
1756 {
1757         struct cache_detail *cd = RPC_I(inode)->private;
1758
1759         return content_release(inode, filp, cd);
1760 }
1761
1762 const struct file_operations content_file_operations_pipefs = {
1763         .open           = content_open_pipefs,
1764         .read           = seq_read,
1765         .llseek         = seq_lseek,
1766         .release        = content_release_pipefs,
1767 };
1768
1769 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1770 {
1771         struct cache_detail *cd = RPC_I(inode)->private;
1772
1773         return open_flush(inode, filp, cd);
1774 }
1775
1776 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1777 {
1778         struct cache_detail *cd = RPC_I(inode)->private;
1779
1780         return release_flush(inode, filp, cd);
1781 }
1782
1783 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1784                             size_t count, loff_t *ppos)
1785 {
1786         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1787
1788         return read_flush(filp, buf, count, ppos, cd);
1789 }
1790
1791 static ssize_t write_flush_pipefs(struct file *filp,
1792                                   const char __user *buf,
1793                                   size_t count, loff_t *ppos)
1794 {
1795         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1796
1797         return write_flush(filp, buf, count, ppos, cd);
1798 }
1799
1800 const struct file_operations cache_flush_operations_pipefs = {
1801         .open           = open_flush_pipefs,
1802         .read           = read_flush_pipefs,
1803         .write          = write_flush_pipefs,
1804         .release        = release_flush_pipefs,
1805         .llseek         = no_llseek,
1806 };
1807
1808 int sunrpc_cache_register_pipefs(struct dentry *parent,
1809                                  const char *name, umode_t umode,
1810                                  struct cache_detail *cd)
1811 {
1812         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1813         if (IS_ERR(dir))
1814                 return PTR_ERR(dir);
1815         cd->u.pipefs.dir = dir;
1816         return 0;
1817 }
1818 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1819
1820 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1821 {
1822         rpc_remove_cache_dir(cd->u.pipefs.dir);
1823         cd->u.pipefs.dir = NULL;
1824 }
1825 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1826