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, struct cache_detail *detail)
  45 {
  46         time_t now = seconds_since_boot();
  47         INIT_HLIST_NODE(&h->cache_list);
  48         h->flags = 0;
  49         kref_init(&h->ref);
  50         h->expiry_time = now + CACHE_NEW_EXPIRY;
  51         if (now <= detail->flush_time)
  52                 /* ensure it isn't already expired */
  53                 now = detail->flush_time + 1;
  54         h->last_refresh = now;
  55 }
  56
  57 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  58                                        struct cache_head *key, int hash)
  59 {
  60         struct cache_head *new = NULL, *freeme = NULL, *tmp = NULL;
  61         struct hlist_head *head;
  62
  63         head = &detail->hash_table[hash];
  64
  65         read_lock(&detail->hash_lock);
  66
  67         hlist_for_each_entry(tmp, head, cache_list) {
  68                 if (detail->match(tmp, key)) {
  69                         if (cache_is_expired(detail, tmp))
  70                                 /* This entry is expired, we will discard it. */
  71                                 break;
  72                         cache_get(tmp);
  73                         read_unlock(&detail->hash_lock);
  74                         return tmp;
  75                 }
  76         }
  77         read_unlock(&detail->hash_lock);
  78         /* Didn't find anything, insert an empty entry */
  79
  80         new = detail->alloc();
  81         if (!new)
  82                 return NULL;
  83         /* must fully initialise 'new', else
  84          * we might get lose if we need to
  85          * cache_put it soon.
  86          */
  87         cache_init(new, detail);
  88         detail->init(new, key);
  89
  90         write_lock(&detail->hash_lock);
  91
  92         /* check if entry appeared while we slept */
  93         hlist_for_each_entry(tmp, head, cache_list) {
  94                 if (detail->match(tmp, key)) {
  95                         if (cache_is_expired(detail, tmp)) {
  96                                 hlist_del_init(&tmp->cache_list);
  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
 108         hlist_add_head(&new->cache_list, head);
 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                                struct cache_detail *detail)
 124 {
 125         time_t now = seconds_since_boot();
 126         if (now <= detail->flush_time)
 127                 /* ensure it isn't immediately treated as expired */
 128                 now = detail->flush_time + 1;
 129         head->expiry_time = expiry;
 130         head->last_refresh = now;
 131         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 132         set_bit(CACHE_VALID, &head->flags);
 133 }
 134
 135 static void cache_fresh_unlocked(struct cache_head *head,
 136                                  struct cache_detail *detail)
 137 {
 138         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 139                 cache_revisit_request(head);
 140                 cache_dequeue(detail, head);
 141         }
 142 }
 143
 144 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 145                                        struct cache_head *new, struct cache_head *old, int hash)
 146 {
 147         /* The 'old' entry is to be replaced by 'new'.
 148          * If 'old' is not VALID, we update it directly,
 149          * otherwise we need to replace it
 150          */
 151         struct cache_head *tmp;
 152
 153         if (!test_bit(CACHE_VALID, &old->flags)) {
 154                 write_lock(&detail->hash_lock);
 155                 if (!test_bit(CACHE_VALID, &old->flags)) {
 156                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 157                                 set_bit(CACHE_NEGATIVE, &old->flags);
 158                         else
 159                                 detail->update(old, new);
 160                         cache_fresh_locked(old, new->expiry_time, detail);
 161                         write_unlock(&detail->hash_lock);
 162                         cache_fresh_unlocked(old, detail);
 163                         return old;
 164                 }
 165                 write_unlock(&detail->hash_lock);
 166         }
 167         /* We need to insert a new entry */
 168         tmp = detail->alloc();
 169         if (!tmp) {
 170                 cache_put(old, detail);
 171                 return NULL;
 172         }
 173         cache_init(tmp, detail);
 174         detail->init(tmp, old);
 175
 176         write_lock(&detail->hash_lock);
 177         if (test_bit(CACHE_NEGATIVE, &new->flags))
 178                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 179         else
 180                 detail->update(tmp, new);
 181         hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
 182         detail->entries++;
 183         cache_get(tmp);
 184         cache_fresh_locked(tmp, new->expiry_time, detail);
 185         cache_fresh_locked(old, 0, detail);
 186         write_unlock(&detail->hash_lock);
 187         cache_fresh_unlocked(tmp, detail);
 188         cache_fresh_unlocked(old, detail);
 189         cache_put(old, detail);
 190         return tmp;
 191 }
 192 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 193
 194 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 195 {
 196         if (cd->cache_upcall)
 197                 return cd->cache_upcall(cd, h);
 198         return sunrpc_cache_pipe_upcall(cd, h);
 199 }
 200
 201 static inline int cache_is_valid(struct cache_head *h)
 202 {
 203         if (!test_bit(CACHE_VALID, &h->flags))
 204                 return -EAGAIN;
 205         else {
 206                 /* entry is valid */
 207                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 208                         return -ENOENT;
 209                 else {
 210                         /*
 211                          * In combination with write barrier in
 212                          * sunrpc_cache_update, ensures that anyone
 213                          * using the cache entry after this sees the
 214                          * updated contents:
 215                          */
 216                         smp_rmb();
 217                         return 0;
 218                 }
 219         }
 220 }
 221
 222 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 223 {
 224         int rv;
 225
 226         write_lock(&detail->hash_lock);
 227         rv = cache_is_valid(h);
 228         if (rv == -EAGAIN) {
 229                 set_bit(CACHE_NEGATIVE, &h->flags);
 230                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
 231                                    detail);
 232                 rv = -ENOENT;
 233         }
 234         write_unlock(&detail->hash_lock);
 235         cache_fresh_unlocked(h, detail);
 236         return rv;
 237 }
 238
 239 /*
 240  * This is the generic cache management routine for all
 241  * the authentication caches.
 242  * It checks the currency of a cache item and will (later)
 243  * initiate an upcall to fill it if needed.
 244  *
 245  *
 246  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 247  * -EAGAIN if upcall is pending and request has been queued
 248  * -ETIMEDOUT if upcall failed or request could not be queue or
 249  *           upcall completed but item is still invalid (implying that
 250  *           the cache item has been replaced with a newer one).
 251  * -ENOENT if cache entry was negative
 252  */
 253 int cache_check(struct cache_detail *detail,
 254                     struct cache_head *h, struct cache_req *rqstp)
 255 {
 256         int rv;
 257         long refresh_age, age;
 258
 259         /* First decide return status as best we can */
 260         rv = cache_is_valid(h);
 261
 262         /* now see if we want to start an upcall */
 263         refresh_age = (h->expiry_time - h->last_refresh);
 264         age = seconds_since_boot() - h->last_refresh;
 265
 266         if (rqstp == NULL) {
 267                 if (rv == -EAGAIN)
 268                         rv = -ENOENT;
 269         } else if (rv == -EAGAIN ||
 270                    (h->expiry_time != 0 && age > refresh_age/2)) {
 271                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 272                                 refresh_age, age);
 273                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 274                         switch (cache_make_upcall(detail, h)) {
 275                         case -EINVAL:
 276                                 rv = try_to_negate_entry(detail, h);
 277                                 break;
 278                         case -EAGAIN:
 279                                 cache_fresh_unlocked(h, detail);
 280                                 break;
 281                         }
 282                 }
 283         }
 284
 285         if (rv == -EAGAIN) {
 286                 if (!cache_defer_req(rqstp, h)) {
 287                         /*
 288                          * Request was not deferred; handle it as best
 289                          * we can ourselves:
 290                          */
 291                         rv = cache_is_valid(h);
 292                         if (rv == -EAGAIN)
 293                                 rv = -ETIMEDOUT;
 294                 }
 295         }
 296         if (rv)
 297                 cache_put(h, detail);
 298         return rv;
 299 }
 300 EXPORT_SYMBOL_GPL(cache_check);
 301
 302 /*
 303  * caches need to be periodically cleaned.
 304  * For this we maintain a list of cache_detail and
 305  * a current pointer into that list and into the table
 306  * for that entry.
 307  *
 308  * Each time cache_clean is called it finds the next non-empty entry
 309  * in the current table and walks the list in that entry
 310  * looking for entries that can be removed.
 311  *
 312  * An entry gets removed if:
 313  * - The expiry is before current time
 314  * - The last_refresh time is before the flush_time for that cache
 315  *
 316  * later we might drop old entries with non-NEVER expiry if that table
 317  * is getting 'full' for some definition of 'full'
 318  *
 319  * The question of "how often to scan a table" is an interesting one
 320  * and is answered in part by the use of the "nextcheck" field in the
 321  * cache_detail.
 322  * When a scan of a table begins, the nextcheck field is set to a time
 323  * that is well into the future.
 324  * While scanning, if an expiry time is found that is earlier than the
 325  * current nextcheck time, nextcheck is set to that expiry time.
 326  * If the flush_time is ever set to a time earlier than the nextcheck
 327  * time, the nextcheck time is then set to that flush_time.
 328  *
 329  * A table is then only scanned if the current time is at least
 330  * the nextcheck time.
 331  *
 332  */
 333
 334 static LIST_HEAD(cache_list);
 335 static DEFINE_SPINLOCK(cache_list_lock);
 336 static struct cache_detail *current_detail;
 337 static int current_index;
 338
 339 static void do_cache_clean(struct work_struct *work);
 340 static struct delayed_work cache_cleaner;
 341
 342 void sunrpc_init_cache_detail(struct cache_detail *cd)
 343 {
 344         rwlock_init(&cd->hash_lock);
 345         INIT_LIST_HEAD(&cd->queue);
 346         spin_lock(&cache_list_lock);
 347         cd->nextcheck = 0;
 348         cd->entries = 0;
 349         atomic_set(&cd->readers, 0);
 350         cd->last_close = 0;
 351         cd->last_warn = -1;
 352         list_add(&cd->others, &cache_list);
 353         spin_unlock(&cache_list_lock);
 354
 355         /* start the cleaning process */
 356         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
 357 }
 358 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 359
 360 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 361 {
 362         cache_purge(cd);
 363         spin_lock(&cache_list_lock);
 364         write_lock(&cd->hash_lock);
 365         if (cd->entries) {
 366                 write_unlock(&cd->hash_lock);
 367                 spin_unlock(&cache_list_lock);
 368                 goto out;
 369         }
 370         if (current_detail == cd)
 371                 current_detail = NULL;
 372         list_del_init(&cd->others);
 373         write_unlock(&cd->hash_lock);
 374         spin_unlock(&cache_list_lock);
 375         if (list_empty(&cache_list)) {
 376                 /* module must be being unloaded so its safe to kill the worker */
 377                 cancel_delayed_work_sync(&cache_cleaner);
 378         }
 379         return;
 380 out:
 381         printk(KERN_ERR "RPC: failed to unregister %s cache\n", cd->name);
 382 }
 383 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 384
 385 /* clean cache tries to find something to clean
 386  * and cleans it.
 387  * It returns 1 if it cleaned something,
 388  *            0 if it didn't find anything this time
 389  *           -1 if it fell off the end of the list.
 390  */
 391 static int cache_clean(void)
 392 {
 393         int rv = 0;
 394         struct list_head *next;
 395
 396         spin_lock(&cache_list_lock);
 397
 398         /* find a suitable table if we don't already have one */
 399         while (current_detail == NULL ||
 400             current_index >= current_detail->hash_size) {
 401                 if (current_detail)
 402                         next = current_detail->others.next;
 403                 else
 404                         next = cache_list.next;
 405                 if (next == &cache_list) {
 406                         current_detail = NULL;
 407                         spin_unlock(&cache_list_lock);
 408                         return -1;
 409                 }
 410                 current_detail = list_entry(next, struct cache_detail, others);
 411                 if (current_detail->nextcheck > seconds_since_boot())
 412                         current_index = current_detail->hash_size;
 413                 else {
 414                         current_index = 0;
 415                         current_detail->nextcheck = seconds_since_boot()+30*60;
 416                 }
 417         }
 418
 419         /* find a non-empty bucket in the table */
 420         while (current_detail &&
 421                current_index < current_detail->hash_size &&
 422                hlist_empty(&current_detail->hash_table[current_index]))
 423                 current_index++;
 424
 425         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 426
 427         if (current_detail && current_index < current_detail->hash_size) {
 428                 struct cache_head *ch = NULL;
 429                 struct cache_detail *d;
 430                 struct hlist_head *head;
 431                 struct hlist_node *tmp;
 432
 433                 write_lock(&current_detail->hash_lock);
 434
 435                 /* Ok, now to clean this strand */
 436
 437                 head = &current_detail->hash_table[current_index];
 438                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 439                         if (current_detail->nextcheck > ch->expiry_time)
 440                                 current_detail->nextcheck = ch->expiry_time+1;
 441                         if (!cache_is_expired(current_detail, ch))
 442                                 continue;
 443
 444                         hlist_del_init(&ch->cache_list);
 445                         current_detail->entries--;
 446                         rv = 1;
 447                         break;
 448                 }
 449
 450                 write_unlock(&current_detail->hash_lock);
 451                 d = current_detail;
 452                 if (!ch)
 453                         current_index ++;
 454                 spin_unlock(&cache_list_lock);
 455                 if (ch) {
 456                         set_bit(CACHE_CLEANED, &ch->flags);
 457                         cache_fresh_unlocked(ch, d);
 458                         cache_put(ch, d);
 459                 }
 460         } else
 461                 spin_unlock(&cache_list_lock);
 462
 463         return rv;
 464 }
 465
 466 /*
 467  * We want to regularly clean the cache, so we need to schedule some work ...
 468  */
 469 static void do_cache_clean(struct work_struct *work)
 470 {
 471         int delay = 5;
 472         if (cache_clean() == -1)
 473                 delay = round_jiffies_relative(30*HZ);
 474
 475         if (list_empty(&cache_list))
 476                 delay = 0;
 477
 478         if (delay)
 479                 queue_delayed_work(system_power_efficient_wq,
 480                                    &cache_cleaner, delay);
 481 }
 482
 483
 484 /*
 485  * Clean all caches promptly.  This just calls cache_clean
 486  * repeatedly until we are sure that every cache has had a chance to
 487  * be fully cleaned
 488  */
 489 void cache_flush(void)
 490 {
 491         while (cache_clean() != -1)
 492                 cond_resched();
 493         while (cache_clean() != -1)
 494                 cond_resched();
 495 }
 496 EXPORT_SYMBOL_GPL(cache_flush);
 497
 498 void cache_purge(struct cache_detail *detail)
 499 {
 500         time_t now = seconds_since_boot();
 501         if (detail->flush_time >= now)
 502                 now = detail->flush_time + 1;
 503         /* 'now' is the maximum value any 'last_refresh' can have */
 504         detail->flush_time = now;
 505         detail->nextcheck = seconds_since_boot();
 506         cache_flush();
 507 }
 508 EXPORT_SYMBOL_GPL(cache_purge);
 509
 510
 511 /*
 512  * Deferral and Revisiting of Requests.
 513  *
 514  * If a cache lookup finds a pending entry, we
 515  * need to defer the request and revisit it later.
 516  * All deferred requests are stored in a hash table,
 517  * indexed by "struct cache_head *".
 518  * As it may be wasteful to store a whole request
 519  * structure, we allow the request to provide a
 520  * deferred form, which must contain a
 521  * 'struct cache_deferred_req'
 522  * This cache_deferred_req contains a method to allow
 523  * it to be revisited when cache info is available
 524  */
 525
 526 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 527 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 528
 529 #define DFR_MAX 300     /* ??? */
 530
 531 static DEFINE_SPINLOCK(cache_defer_lock);
 532 static LIST_HEAD(cache_defer_list);
 533 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 534 static int cache_defer_cnt;
 535
 536 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 537 {
 538         hlist_del_init(&dreq->hash);
 539         if (!list_empty(&dreq->recent)) {
 540                 list_del_init(&dreq->recent);
 541                 cache_defer_cnt--;
 542         }
 543 }
 544
 545 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 546 {
 547         int hash = DFR_HASH(item);
 548
 549         INIT_LIST_HEAD(&dreq->recent);
 550         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 551 }
 552
 553 static void setup_deferral(struct cache_deferred_req *dreq,
 554                            struct cache_head *item,
 555                            int count_me)
 556 {
 557
 558         dreq->item = item;
 559
 560         spin_lock(&cache_defer_lock);
 561
 562         __hash_deferred_req(dreq, item);
 563
 564         if (count_me) {
 565                 cache_defer_cnt++;
 566                 list_add(&dreq->recent, &cache_defer_list);
 567         }
 568
 569         spin_unlock(&cache_defer_lock);
 570
 571 }
 572
 573 struct thread_deferred_req {
 574         struct cache_deferred_req handle;
 575         struct completion completion;
 576 };
 577
 578 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 579 {
 580         struct thread_deferred_req *dr =
 581                 container_of(dreq, struct thread_deferred_req, handle);
 582         complete(&dr->completion);
 583 }
 584
 585 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 586 {
 587         struct thread_deferred_req sleeper;
 588         struct cache_deferred_req *dreq = &sleeper.handle;
 589
 590         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 591         dreq->revisit = cache_restart_thread;
 592
 593         setup_deferral(dreq, item, 0);
 594
 595         if (!test_bit(CACHE_PENDING, &item->flags) ||
 596             wait_for_completion_interruptible_timeout(
 597                     &sleeper.completion, req->thread_wait) <= 0) {
 598                 /* The completion wasn't completed, so we need
 599                  * to clean up
 600                  */
 601                 spin_lock(&cache_defer_lock);
 602                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 603                         __unhash_deferred_req(&sleeper.handle);
 604                         spin_unlock(&cache_defer_lock);
 605                 } else {
 606                         /* cache_revisit_request already removed
 607                          * this from the hash table, but hasn't
 608                          * called ->revisit yet.  It will very soon
 609                          * and we need to wait for it.
 610                          */
 611                         spin_unlock(&cache_defer_lock);
 612                         wait_for_completion(&sleeper.completion);
 613                 }
 614         }
 615 }
 616
 617 static void cache_limit_defers(void)
 618 {
 619         /* Make sure we haven't exceed the limit of allowed deferred
 620          * requests.
 621          */
 622         struct cache_deferred_req *discard = NULL;
 623
 624         if (cache_defer_cnt <= DFR_MAX)
 625                 return;
 626
 627         spin_lock(&cache_defer_lock);
 628
 629         /* Consider removing either the first or the last */
 630         if (cache_defer_cnt > DFR_MAX) {
 631                 if (prandom_u32() & 1)
 632                         discard = list_entry(cache_defer_list.next,
 633                                              struct cache_deferred_req, recent);
 634                 else
 635                         discard = list_entry(cache_defer_list.prev,
 636                                              struct cache_deferred_req, recent);
 637                 __unhash_deferred_req(discard);
 638         }
 639         spin_unlock(&cache_defer_lock);
 640         if (discard)
 641                 discard->revisit(discard, 1);
 642 }
 643
 644 /* Return true if and only if a deferred request is queued. */
 645 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 646 {
 647         struct cache_deferred_req *dreq;
 648
 649         if (req->thread_wait) {
 650                 cache_wait_req(req, item);
 651                 if (!test_bit(CACHE_PENDING, &item->flags))
 652                         return false;
 653         }
 654         dreq = req->defer(req);
 655         if (dreq == NULL)
 656                 return false;
 657         setup_deferral(dreq, item, 1);
 658         if (!test_bit(CACHE_PENDING, &item->flags))
 659                 /* Bit could have been cleared before we managed to
 660                  * set up the deferral, so need to revisit just in case
 661                  */
 662                 cache_revisit_request(item);
 663
 664         cache_limit_defers();
 665         return true;
 666 }
 667
 668 static void cache_revisit_request(struct cache_head *item)
 669 {
 670         struct cache_deferred_req *dreq;
 671         struct list_head pending;
 672         struct hlist_node *tmp;
 673         int hash = DFR_HASH(item);
 674
 675         INIT_LIST_HEAD(&pending);
 676         spin_lock(&cache_defer_lock);
 677
 678         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 679                 if (dreq->item == item) {
 680                         __unhash_deferred_req(dreq);
 681                         list_add(&dreq->recent, &pending);
 682                 }
 683
 684         spin_unlock(&cache_defer_lock);
 685
 686         while (!list_empty(&pending)) {
 687                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 688                 list_del_init(&dreq->recent);
 689                 dreq->revisit(dreq, 0);
 690         }
 691 }
 692
 693 void cache_clean_deferred(void *owner)
 694 {
 695         struct cache_deferred_req *dreq, *tmp;
 696         struct list_head pending;
 697
 698
 699         INIT_LIST_HEAD(&pending);
 700         spin_lock(&cache_defer_lock);
 701
 702         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 703                 if (dreq->owner == owner) {
 704                         __unhash_deferred_req(dreq);
 705                         list_add(&dreq->recent, &pending);
 706                 }
 707         }
 708         spin_unlock(&cache_defer_lock);
 709
 710         while (!list_empty(&pending)) {
 711                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 712                 list_del_init(&dreq->recent);
 713                 dreq->revisit(dreq, 1);
 714         }
 715 }
 716
 717 /*
 718  * communicate with user-space
 719  *
 720  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 721  * On read, you get a full request, or block.
 722  * On write, an update request is processed.
 723  * Poll works if anything to read, and always allows write.
 724  *
 725  * Implemented by linked list of requests.  Each open file has
 726  * a ->private that also exists in this list.  New requests are added
 727  * to the end and may wakeup and preceding readers.
 728  * New readers are added to the head.  If, on read, an item is found with
 729  * CACHE_UPCALLING clear, we free it from the list.
 730  *
 731  */
 732
 733 static DEFINE_SPINLOCK(queue_lock);
 734 static DEFINE_MUTEX(queue_io_mutex);
 735
 736 struct cache_queue {
 737         struct list_head        list;
 738         int                     reader; /* if 0, then request */
 739 };
 740 struct cache_request {
 741         struct cache_queue      q;
 742         struct cache_head       *item;
 743         char                    * buf;
 744         int                     len;
 745         int                     readers;
 746 };
 747 struct cache_reader {
 748         struct cache_queue      q;
 749         int                     offset; /* if non-0, we have a refcnt on next request */
 750 };
 751
 752 static int cache_request(struct cache_detail *detail,
 753                                struct cache_request *crq)
 754 {
 755         char *bp = crq->buf;
 756         int len = PAGE_SIZE;
 757
 758         detail->cache_request(detail, crq->item, &bp, &len);
 759         if (len < 0)
 760                 return -EAGAIN;
 761         return PAGE_SIZE - len;
 762 }
 763
 764 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 765                           loff_t *ppos, struct cache_detail *cd)
 766 {
 767         struct cache_reader *rp = filp->private_data;
 768         struct cache_request *rq;
 769         struct inode *inode = file_inode(filp);
 770         int err;
 771
 772         if (count == 0)
 773                 return 0;
 774
 775         inode_lock(inode); /* protect against multiple concurrent
 776                               * readers on this file */
 777  again:
 778         spin_lock(&queue_lock);
 779         /* need to find next request */
 780         while (rp->q.list.next != &cd->queue &&
 781                list_entry(rp->q.list.next, struct cache_queue, list)
 782                ->reader) {
 783                 struct list_head *next = rp->q.list.next;
 784                 list_move(&rp->q.list, next);
 785         }
 786         if (rp->q.list.next == &cd->queue) {
 787                 spin_unlock(&queue_lock);
 788                 inode_unlock(inode);
 789                 WARN_ON_ONCE(rp->offset);
 790                 return 0;
 791         }
 792         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 793         WARN_ON_ONCE(rq->q.reader);
 794         if (rp->offset == 0)
 795                 rq->readers++;
 796         spin_unlock(&queue_lock);
 797
 798         if (rq->len == 0) {
 799                 err = cache_request(cd, rq);
 800                 if (err < 0)
 801                         goto out;
 802                 rq->len = err;
 803         }
 804
 805         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 806                 err = -EAGAIN;
 807                 spin_lock(&queue_lock);
 808                 list_move(&rp->q.list, &rq->q.list);
 809                 spin_unlock(&queue_lock);
 810         } else {
 811                 if (rp->offset + count > rq->len)
 812                         count = rq->len - rp->offset;
 813                 err = -EFAULT;
 814                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 815                         goto out;
 816                 rp->offset += count;
 817                 if (rp->offset >= rq->len) {
 818                         rp->offset = 0;
 819                         spin_lock(&queue_lock);
 820                         list_move(&rp->q.list, &rq->q.list);
 821                         spin_unlock(&queue_lock);
 822                 }
 823                 err = 0;
 824         }
 825  out:
 826         if (rp->offset == 0) {
 827                 /* need to release rq */
 828                 spin_lock(&queue_lock);
 829                 rq->readers--;
 830                 if (rq->readers == 0 &&
 831                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 832                         list_del(&rq->q.list);
 833                         spin_unlock(&queue_lock);
 834                         cache_put(rq->item, cd);
 835                         kfree(rq->buf);
 836                         kfree(rq);
 837                 } else
 838                         spin_unlock(&queue_lock);
 839         }
 840         if (err == -EAGAIN)
 841                 goto again;
 842         inode_unlock(inode);
 843         return err ? err :  count;
 844 }
 845
 846 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 847                                  size_t count, struct cache_detail *cd)
 848 {
 849         ssize_t ret;
 850
 851         if (count == 0)
 852                 return -EINVAL;
 853         if (copy_from_user(kaddr, buf, count))
 854                 return -EFAULT;
 855         kaddr[count] = '\0';
 856         ret = cd->cache_parse(cd, kaddr, count);
 857         if (!ret)
 858                 ret = count;
 859         return ret;
 860 }
 861
 862 static ssize_t cache_slow_downcall(const char __user *buf,
 863                                    size_t count, struct cache_detail *cd)
 864 {
 865         static char write_buf[8192]; /* protected by queue_io_mutex */
 866         ssize_t ret = -EINVAL;
 867
 868         if (count >= sizeof(write_buf))
 869                 goto out;
 870         mutex_lock(&queue_io_mutex);
 871         ret = cache_do_downcall(write_buf, buf, count, cd);
 872         mutex_unlock(&queue_io_mutex);
 873 out:
 874         return ret;
 875 }
 876
 877 static ssize_t cache_downcall(struct address_space *mapping,
 878                               const char __user *buf,
 879                               size_t count, struct cache_detail *cd)
 880 {
 881         struct page *page;
 882         char *kaddr;
 883         ssize_t ret = -ENOMEM;
 884
 885         if (count >= PAGE_SIZE)
 886                 goto out_slow;
 887
 888         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 889         if (!page)
 890                 goto out_slow;
 891
 892         kaddr = kmap(page);
 893         ret = cache_do_downcall(kaddr, buf, count, cd);
 894         kunmap(page);
 895         unlock_page(page);
 896         put_page(page);
 897         return ret;
 898 out_slow:
 899         return cache_slow_downcall(buf, count, cd);
 900 }
 901
 902 static ssize_t cache_write(struct file *filp, const char __user *buf,
 903                            size_t count, loff_t *ppos,
 904                            struct cache_detail *cd)
 905 {
 906         struct address_space *mapping = filp->f_mapping;
 907         struct inode *inode = file_inode(filp);
 908         ssize_t ret = -EINVAL;
 909
 910         if (!cd->cache_parse)
 911                 goto out;
 912
 913         inode_lock(inode);
 914         ret = cache_downcall(mapping, buf, count, cd);
 915         inode_unlock(inode);
 916 out:
 917         return ret;
 918 }
 919
 920 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 921
 922 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 923                                struct cache_detail *cd)
 924 {
 925         unsigned int mask;
 926         struct cache_reader *rp = filp->private_data;
 927         struct cache_queue *cq;
 928
 929         poll_wait(filp, &queue_wait, wait);
 930
 931         /* alway allow write */
 932         mask = POLLOUT | POLLWRNORM;
 933
 934         if (!rp)
 935                 return mask;
 936
 937         spin_lock(&queue_lock);
 938
 939         for (cq= &rp->q; &cq->list != &cd->queue;
 940              cq = list_entry(cq->list.next, struct cache_queue, list))
 941                 if (!cq->reader) {
 942                         mask |= POLLIN | POLLRDNORM;
 943                         break;
 944                 }
 945         spin_unlock(&queue_lock);
 946         return mask;
 947 }
 948
 949 static int cache_ioctl(struct inode *ino, struct file *filp,
 950                        unsigned int cmd, unsigned long arg,
 951                        struct cache_detail *cd)
 952 {
 953         int len = 0;
 954         struct cache_reader *rp = filp->private_data;
 955         struct cache_queue *cq;
 956
 957         if (cmd != FIONREAD || !rp)
 958                 return -EINVAL;
 959
 960         spin_lock(&queue_lock);
 961
 962         /* only find the length remaining in current request,
 963          * or the length of the next request
 964          */
 965         for (cq= &rp->q; &cq->list != &cd->queue;
 966              cq = list_entry(cq->list.next, struct cache_queue, list))
 967                 if (!cq->reader) {
 968                         struct cache_request *cr =
 969                                 container_of(cq, struct cache_request, q);
 970                         len = cr->len - rp->offset;
 971                         break;
 972                 }
 973         spin_unlock(&queue_lock);
 974
 975         return put_user(len, (int __user *)arg);
 976 }
 977
 978 static int cache_open(struct inode *inode, struct file *filp,
 979                       struct cache_detail *cd)
 980 {
 981         struct cache_reader *rp = NULL;
 982
 983         if (!cd || !try_module_get(cd->owner))
 984                 return -EACCES;
 985         nonseekable_open(inode, filp);
 986         if (filp->f_mode & FMODE_READ) {
 987                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 988                 if (!rp) {
 989                         module_put(cd->owner);
 990                         return -ENOMEM;
 991                 }
 992                 rp->offset = 0;
 993                 rp->q.reader = 1;
 994                 atomic_inc(&cd->readers);
 995                 spin_lock(&queue_lock);
 996                 list_add(&rp->q.list, &cd->queue);
 997                 spin_unlock(&queue_lock);
 998         }
 999         filp->private_data = rp;
1000         return 0;
1001 }
1002
1003 static int cache_release(struct inode *inode, struct file *filp,
1004                          struct cache_detail *cd)
1005 {
1006         struct cache_reader *rp = filp->private_data;
1007
1008         if (rp) {
1009                 spin_lock(&queue_lock);
1010                 if (rp->offset) {
1011                         struct cache_queue *cq;
1012                         for (cq= &rp->q; &cq->list != &cd->queue;
1013                              cq = list_entry(cq->list.next, struct cache_queue, list))
1014                                 if (!cq->reader) {
1015                                         container_of(cq, struct cache_request, q)
1016                                                 ->readers--;
1017                                         break;
1018                                 }
1019                         rp->offset = 0;
1020                 }
1021                 list_del(&rp->q.list);
1022                 spin_unlock(&queue_lock);
1023
1024                 filp->private_data = NULL;
1025                 kfree(rp);
1026
1027                 cd->last_close = seconds_since_boot();
1028                 atomic_dec(&cd->readers);
1029         }
1030         module_put(cd->owner);
1031         return 0;
1032 }
1033
1034
1035
1036 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1037 {
1038         struct cache_queue *cq, *tmp;
1039         struct cache_request *cr;
1040         struct list_head dequeued;
1041
1042         INIT_LIST_HEAD(&dequeued);
1043         spin_lock(&queue_lock);
1044         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1045                 if (!cq->reader) {
1046                         cr = container_of(cq, struct cache_request, q);
1047                         if (cr->item != ch)
1048                                 continue;
1049                         if (test_bit(CACHE_PENDING, &ch->flags))
1050                                 /* Lost a race and it is pending again */
1051                                 break;
1052                         if (cr->readers != 0)
1053                                 continue;
1054                         list_move(&cr->q.list, &dequeued);
1055                 }
1056         spin_unlock(&queue_lock);
1057         while (!list_empty(&dequeued)) {
1058                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1059                 list_del(&cr->q.list);
1060                 cache_put(cr->item, detail);
1061                 kfree(cr->buf);
1062                 kfree(cr);
1063         }
1064 }
1065
1066 /*
1067  * Support routines for text-based upcalls.
1068  * Fields are separated by spaces.
1069  * Fields are either mangled to quote space tab newline slosh with slosh
1070  * or a hexified with a leading \x
1071  * Record is terminated with newline.
1072  *
1073  */
1074
1075 void qword_add(char **bpp, int *lp, char *str)
1076 {
1077         char *bp = *bpp;
1078         int len = *lp;
1079         int ret;
1080
1081         if (len < 0) return;
1082
1083         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1084         if (ret >= len) {
1085                 bp += len;
1086                 len = -1;
1087         } else {
1088                 bp += ret;
1089                 len -= ret;
1090                 *bp++ = ' ';
1091                 len--;
1092         }
1093         *bpp = bp;
1094         *lp = len;
1095 }
1096 EXPORT_SYMBOL_GPL(qword_add);
1097
1098 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1099 {
1100         char *bp = *bpp;
1101         int len = *lp;
1102
1103         if (len < 0) return;
1104
1105         if (len > 2) {
1106                 *bp++ = '\\';
1107                 *bp++ = 'x';
1108                 len -= 2;
1109                 while (blen && len >= 2) {
1110                         bp = hex_byte_pack(bp, *buf++);
1111                         len -= 2;
1112                         blen--;
1113                 }
1114         }
1115         if (blen || len<1) len = -1;
1116         else {
1117                 *bp++ = ' ';
1118                 len--;
1119         }
1120         *bpp = bp;
1121         *lp = len;
1122 }
1123 EXPORT_SYMBOL_GPL(qword_addhex);
1124
1125 static void warn_no_listener(struct cache_detail *detail)
1126 {
1127         if (detail->last_warn != detail->last_close) {
1128                 detail->last_warn = detail->last_close;
1129                 if (detail->warn_no_listener)
1130                         detail->warn_no_listener(detail, detail->last_close != 0);
1131         }
1132 }
1133
1134 static bool cache_listeners_exist(struct cache_detail *detail)
1135 {
1136         if (atomic_read(&detail->readers))
1137                 return true;
1138         if (detail->last_close == 0)
1139                 /* This cache was never opened */
1140                 return false;
1141         if (detail->last_close < seconds_since_boot() - 30)
1142                 /*
1143                  * We allow for the possibility that someone might
1144                  * restart a userspace daemon without restarting the
1145                  * server; but after 30 seconds, we give up.
1146                  */
1147                  return false;
1148         return true;
1149 }
1150
1151 /*
1152  * register an upcall request to user-space and queue it up for read() by the
1153  * upcall daemon.
1154  *
1155  * Each request is at most one page long.
1156  */
1157 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1158 {
1159
1160         char *buf;
1161         struct cache_request *crq;
1162         int ret = 0;
1163
1164         if (!detail->cache_request)
1165                 return -EINVAL;
1166
1167         if (!cache_listeners_exist(detail)) {
1168                 warn_no_listener(detail);
1169                 return -EINVAL;
1170         }
1171         if (test_bit(CACHE_CLEANED, &h->flags))
1172                 /* Too late to make an upcall */
1173                 return -EAGAIN;
1174
1175         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1176         if (!buf)
1177                 return -EAGAIN;
1178
1179         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1180         if (!crq) {
1181                 kfree(buf);
1182                 return -EAGAIN;
1183         }
1184
1185         crq->q.reader = 0;
1186         crq->buf = buf;
1187         crq->len = 0;
1188         crq->readers = 0;
1189         spin_lock(&queue_lock);
1190         if (test_bit(CACHE_PENDING, &h->flags)) {
1191                 crq->item = cache_get(h);
1192                 list_add_tail(&crq->q.list, &detail->queue);
1193         } else
1194                 /* Lost a race, no longer PENDING, so don't enqueue */
1195                 ret = -EAGAIN;
1196         spin_unlock(&queue_lock);
1197         wake_up(&queue_wait);
1198         if (ret == -EAGAIN) {
1199                 kfree(buf);
1200                 kfree(crq);
1201         }
1202         return ret;
1203 }
1204 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1205
1206 /*
1207  * parse a message from user-space and pass it
1208  * to an appropriate cache
1209  * Messages are, like requests, separated into fields by
1210  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1211  *
1212  * Message is
1213  *   reply cachename expiry key ... content....
1214  *
1215  * key and content are both parsed by cache
1216  */
1217
1218 int qword_get(char **bpp, char *dest, int bufsize)
1219 {
1220         /* return bytes copied, or -1 on error */
1221         char *bp = *bpp;
1222         int len = 0;
1223
1224         while (*bp == ' ') bp++;
1225
1226         if (bp[0] == '\\' && bp[1] == 'x') {
1227                 /* HEX STRING */
1228                 bp += 2;
1229                 while (len < bufsize - 1) {
1230                         int h, l;
1231
1232                         h = hex_to_bin(bp[0]);
1233                         if (h < 0)
1234                                 break;
1235
1236                         l = hex_to_bin(bp[1]);
1237                         if (l < 0)
1238                                 break;
1239
1240                         *dest++ = (h << 4) | l;
1241                         bp += 2;
1242                         len++;
1243                 }
1244         } else {
1245                 /* text with \nnn octal quoting */
1246                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1247                         if (*bp == '\\' &&
1248                             isodigit(bp[1]) && (bp[1] <= '3') &&
1249                             isodigit(bp[2]) &&
1250                             isodigit(bp[3])) {
1251                                 int byte = (*++bp -'0');
1252                                 bp++;
1253                                 byte = (byte << 3) | (*bp++ - '0');
1254                                 byte = (byte << 3) | (*bp++ - '0');
1255                                 *dest++ = byte;
1256                                 len++;
1257                         } else {
1258                                 *dest++ = *bp++;
1259                                 len++;
1260                         }
1261                 }
1262         }
1263
1264         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1265                 return -1;
1266         while (*bp == ' ') bp++;
1267         *bpp = bp;
1268         *dest = '\0';
1269         return len;
1270 }
1271 EXPORT_SYMBOL_GPL(qword_get);
1272
1273
1274 /*
1275  * support /proc/sunrpc/cache/$CACHENAME/content
1276  * as a seqfile.
1277  * We call ->cache_show passing NULL for the item to
1278  * get a header, then pass each real item in the cache
1279  */
1280
1281 void *cache_seq_start(struct seq_file *m, loff_t *pos)
1282         __acquires(cd->hash_lock)
1283 {
1284         loff_t n = *pos;
1285         unsigned int hash, entry;
1286         struct cache_head *ch;
1287         struct cache_detail *cd = m->private;
1288
1289         read_lock(&cd->hash_lock);
1290         if (!n--)
1291                 return SEQ_START_TOKEN;
1292         hash = n >> 32;
1293         entry = n & ((1LL<<32) - 1);
1294
1295         hlist_for_each_entry(ch, &cd->hash_table[hash], cache_list)
1296                 if (!entry--)
1297                         return ch;
1298         n &= ~((1LL<<32) - 1);
1299         do {
1300                 hash++;
1301                 n += 1LL<<32;
1302         } while(hash < cd->hash_size &&
1303                 hlist_empty(&cd->hash_table[hash]));
1304         if (hash >= cd->hash_size)
1305                 return NULL;
1306         *pos = n+1;
1307         return hlist_entry_safe(cd->hash_table[hash].first,
1308                                 struct cache_head, cache_list);
1309 }
1310 EXPORT_SYMBOL_GPL(cache_seq_start);
1311
1312 void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1313 {
1314         struct cache_head *ch = p;
1315         int hash = (*pos >> 32);
1316         struct cache_detail *cd = m->private;
1317
1318         if (p == SEQ_START_TOKEN)
1319                 hash = 0;
1320         else if (ch->cache_list.next == NULL) {
1321                 hash++;
1322                 *pos += 1LL<<32;
1323         } else {
1324                 ++*pos;
1325                 return hlist_entry_safe(ch->cache_list.next,
1326                                         struct cache_head, cache_list);
1327         }
1328         *pos &= ~((1LL<<32) - 1);
1329         while (hash < cd->hash_size &&
1330                hlist_empty(&cd->hash_table[hash])) {
1331                 hash++;
1332                 *pos += 1LL<<32;
1333         }
1334         if (hash >= cd->hash_size)
1335                 return NULL;
1336         ++*pos;
1337         return hlist_entry_safe(cd->hash_table[hash].first,
1338                                 struct cache_head, cache_list);
1339 }
1340 EXPORT_SYMBOL_GPL(cache_seq_next);
1341
1342 void cache_seq_stop(struct seq_file *m, void *p)
1343         __releases(cd->hash_lock)
1344 {
1345         struct cache_detail *cd = m->private;
1346         read_unlock(&cd->hash_lock);
1347 }
1348 EXPORT_SYMBOL_GPL(cache_seq_stop);
1349
1350 static int c_show(struct seq_file *m, void *p)
1351 {
1352         struct cache_head *cp = p;
1353         struct cache_detail *cd = m->private;
1354
1355         if (p == SEQ_START_TOKEN)
1356                 return cd->cache_show(m, cd, NULL);
1357
1358         ifdebug(CACHE)
1359                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1360                            convert_to_wallclock(cp->expiry_time),
1361                            atomic_read(&cp->ref.refcount), cp->flags);
1362         cache_get(cp);
1363         if (cache_check(cd, cp, NULL))
1364                 /* cache_check does a cache_put on failure */
1365                 seq_printf(m, "# ");
1366         else {
1367                 if (cache_is_expired(cd, cp))
1368                         seq_printf(m, "# ");
1369                 cache_put(cp, cd);
1370         }
1371
1372         return cd->cache_show(m, cd, cp);
1373 }
1374
1375 static const struct seq_operations cache_content_op = {
1376         .start  = cache_seq_start,
1377         .next   = cache_seq_next,
1378         .stop   = cache_seq_stop,
1379         .show   = c_show,
1380 };
1381
1382 static int content_open(struct inode *inode, struct file *file,
1383                         struct cache_detail *cd)
1384 {
1385         struct seq_file *seq;
1386         int err;
1387
1388         if (!cd || !try_module_get(cd->owner))
1389                 return -EACCES;
1390
1391         err = seq_open(file, &cache_content_op);
1392         if (err) {
1393                 module_put(cd->owner);
1394                 return err;
1395         }
1396
1397         seq = file->private_data;
1398         seq->private = cd;
1399         return 0;
1400 }
1401
1402 static int content_release(struct inode *inode, struct file *file,
1403                 struct cache_detail *cd)
1404 {
1405         int ret = seq_release(inode, file);
1406         module_put(cd->owner);
1407         return ret;
1408 }
1409
1410 static int open_flush(struct inode *inode, struct file *file,
1411                         struct cache_detail *cd)
1412 {
1413         if (!cd || !try_module_get(cd->owner))
1414                 return -EACCES;
1415         return nonseekable_open(inode, file);
1416 }
1417
1418 static int release_flush(struct inode *inode, struct file *file,
1419                         struct cache_detail *cd)
1420 {
1421         module_put(cd->owner);
1422         return 0;
1423 }
1424
1425 static ssize_t read_flush(struct file *file, char __user *buf,
1426                           size_t count, loff_t *ppos,
1427                           struct cache_detail *cd)
1428 {
1429         char tbuf[22];
1430         unsigned long p = *ppos;
1431         size_t len;
1432
1433         snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1434         len = strlen(tbuf);
1435         if (p >= len)
1436                 return 0;
1437         len -= p;
1438         if (len > count)
1439                 len = count;
1440         if (copy_to_user(buf, (void*)(tbuf+p), len))
1441                 return -EFAULT;
1442         *ppos += len;
1443         return len;
1444 }
1445
1446 static ssize_t write_flush(struct file *file, const char __user *buf,
1447                            size_t count, loff_t *ppos,
1448                            struct cache_detail *cd)
1449 {
1450         char tbuf[20];
1451         char *bp, *ep;
1452         time_t then, now;
1453
1454         if (*ppos || count > sizeof(tbuf)-1)
1455                 return -EINVAL;
1456         if (copy_from_user(tbuf, buf, count))
1457                 return -EFAULT;
1458         tbuf[count] = 0;
1459         simple_strtoul(tbuf, &ep, 0);
1460         if (*ep && *ep != '\n')
1461                 return -EINVAL;
1462
1463         bp = tbuf;
1464         then = get_expiry(&bp);
1465         now = seconds_since_boot();
1466         cd->nextcheck = now;
1467         /* Can only set flush_time to 1 second beyond "now", or
1468          * possibly 1 second beyond flushtime.  This is because
1469          * flush_time never goes backwards so it mustn't get too far
1470          * ahead of time.
1471          */
1472         if (then >= now) {
1473                 /* Want to flush everything, so behave like cache_purge() */
1474                 if (cd->flush_time >= now)
1475                         now = cd->flush_time + 1;
1476                 then = now;
1477         }
1478
1479         cd->flush_time = then;
1480         cache_flush();
1481
1482         *ppos += count;
1483         return count;
1484 }
1485
1486 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1487                                  size_t count, loff_t *ppos)
1488 {
1489         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1490
1491         return cache_read(filp, buf, count, ppos, cd);
1492 }
1493
1494 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1495                                   size_t count, loff_t *ppos)
1496 {
1497         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1498
1499         return cache_write(filp, buf, count, ppos, cd);
1500 }
1501
1502 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1503 {
1504         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1505
1506         return cache_poll(filp, wait, cd);
1507 }
1508
1509 static long cache_ioctl_procfs(struct file *filp,
1510                                unsigned int cmd, unsigned long arg)
1511 {
1512         struct inode *inode = file_inode(filp);
1513         struct cache_detail *cd = PDE_DATA(inode);
1514
1515         return cache_ioctl(inode, filp, cmd, arg, cd);
1516 }
1517
1518 static int cache_open_procfs(struct inode *inode, struct file *filp)
1519 {
1520         struct cache_detail *cd = PDE_DATA(inode);
1521
1522         return cache_open(inode, filp, cd);
1523 }
1524
1525 static int cache_release_procfs(struct inode *inode, struct file *filp)
1526 {
1527         struct cache_detail *cd = PDE_DATA(inode);
1528
1529         return cache_release(inode, filp, cd);
1530 }
1531
1532 static const struct file_operations cache_file_operations_procfs = {
1533         .owner          = THIS_MODULE,
1534         .llseek         = no_llseek,
1535         .read           = cache_read_procfs,
1536         .write          = cache_write_procfs,
1537         .poll           = cache_poll_procfs,
1538         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1539         .open           = cache_open_procfs,
1540         .release        = cache_release_procfs,
1541 };
1542
1543 static int content_open_procfs(struct inode *inode, struct file *filp)
1544 {
1545         struct cache_detail *cd = PDE_DATA(inode);
1546
1547         return content_open(inode, filp, cd);
1548 }
1549
1550 static int content_release_procfs(struct inode *inode, struct file *filp)
1551 {
1552         struct cache_detail *cd = PDE_DATA(inode);
1553
1554         return content_release(inode, filp, cd);
1555 }
1556
1557 static const struct file_operations content_file_operations_procfs = {
1558         .open           = content_open_procfs,
1559         .read           = seq_read,
1560         .llseek         = seq_lseek,
1561         .release        = content_release_procfs,
1562 };
1563
1564 static int open_flush_procfs(struct inode *inode, struct file *filp)
1565 {
1566         struct cache_detail *cd = PDE_DATA(inode);
1567
1568         return open_flush(inode, filp, cd);
1569 }
1570
1571 static int release_flush_procfs(struct inode *inode, struct file *filp)
1572 {
1573         struct cache_detail *cd = PDE_DATA(inode);
1574
1575         return release_flush(inode, filp, cd);
1576 }
1577
1578 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1579                             size_t count, loff_t *ppos)
1580 {
1581         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1582
1583         return read_flush(filp, buf, count, ppos, cd);
1584 }
1585
1586 static ssize_t write_flush_procfs(struct file *filp,
1587                                   const char __user *buf,
1588                                   size_t count, loff_t *ppos)
1589 {
1590         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1591
1592         return write_flush(filp, buf, count, ppos, cd);
1593 }
1594
1595 static const struct file_operations cache_flush_operations_procfs = {
1596         .open           = open_flush_procfs,
1597         .read           = read_flush_procfs,
1598         .write          = write_flush_procfs,
1599         .release        = release_flush_procfs,
1600         .llseek         = no_llseek,
1601 };
1602
1603 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1604 {
1605         struct sunrpc_net *sn;
1606
1607         if (cd->u.procfs.proc_ent == NULL)
1608                 return;
1609         if (cd->u.procfs.flush_ent)
1610                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1611         if (cd->u.procfs.channel_ent)
1612                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1613         if (cd->u.procfs.content_ent)
1614                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1615         cd->u.procfs.proc_ent = NULL;
1616         sn = net_generic(net, sunrpc_net_id);
1617         remove_proc_entry(cd->name, sn->proc_net_rpc);
1618 }
1619
1620 #ifdef CONFIG_PROC_FS
1621 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1622 {
1623         struct proc_dir_entry *p;
1624         struct sunrpc_net *sn;
1625
1626         sn = net_generic(net, sunrpc_net_id);
1627         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1628         if (cd->u.procfs.proc_ent == NULL)
1629                 goto out_nomem;
1630         cd->u.procfs.channel_ent = NULL;
1631         cd->u.procfs.content_ent = NULL;
1632
1633         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1634                              cd->u.procfs.proc_ent,
1635                              &cache_flush_operations_procfs, cd);
1636         cd->u.procfs.flush_ent = p;
1637         if (p == NULL)
1638                 goto out_nomem;
1639
1640         if (cd->cache_request || cd->cache_parse) {
1641                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1642                                      cd->u.procfs.proc_ent,
1643                                      &cache_file_operations_procfs, cd);
1644                 cd->u.procfs.channel_ent = p;
1645                 if (p == NULL)
1646                         goto out_nomem;
1647         }
1648         if (cd->cache_show) {
1649                 p = proc_create_data("content", S_IFREG|S_IRUSR,
1650                                 cd->u.procfs.proc_ent,
1651                                 &content_file_operations_procfs, cd);
1652                 cd->u.procfs.content_ent = p;
1653                 if (p == NULL)
1654                         goto out_nomem;
1655         }
1656         return 0;
1657 out_nomem:
1658         remove_cache_proc_entries(cd, net);
1659         return -ENOMEM;
1660 }
1661 #else /* CONFIG_PROC_FS */
1662 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1663 {
1664         return 0;
1665 }
1666 #endif
1667
1668 void __init cache_initialize(void)
1669 {
1670         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1671 }
1672
1673 int cache_register_net(struct cache_detail *cd, struct net *net)
1674 {
1675         int ret;
1676
1677         sunrpc_init_cache_detail(cd);
1678         ret = create_cache_proc_entries(cd, net);
1679         if (ret)
1680                 sunrpc_destroy_cache_detail(cd);
1681         return ret;
1682 }
1683 EXPORT_SYMBOL_GPL(cache_register_net);
1684
1685 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1686 {
1687         remove_cache_proc_entries(cd, net);
1688         sunrpc_destroy_cache_detail(cd);
1689 }
1690 EXPORT_SYMBOL_GPL(cache_unregister_net);
1691
1692 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1693 {
1694         struct cache_detail *cd;
1695         int i;
1696
1697         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1698         if (cd == NULL)
1699                 return ERR_PTR(-ENOMEM);
1700
1701         cd->hash_table = kzalloc(cd->hash_size * sizeof(struct hlist_head),
1702                                  GFP_KERNEL);
1703         if (cd->hash_table == NULL) {
1704                 kfree(cd);
1705                 return ERR_PTR(-ENOMEM);
1706         }
1707
1708         for (i = 0; i < cd->hash_size; i++)
1709                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1710         cd->net = net;
1711         return cd;
1712 }
1713 EXPORT_SYMBOL_GPL(cache_create_net);
1714
1715 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1716 {
1717         kfree(cd->hash_table);
1718         kfree(cd);
1719 }
1720 EXPORT_SYMBOL_GPL(cache_destroy_net);
1721
1722 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1723                                  size_t count, loff_t *ppos)
1724 {
1725         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1726
1727         return cache_read(filp, buf, count, ppos, cd);
1728 }
1729
1730 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1731                                   size_t count, loff_t *ppos)
1732 {
1733         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1734
1735         return cache_write(filp, buf, count, ppos, cd);
1736 }
1737
1738 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1739 {
1740         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1741
1742         return cache_poll(filp, wait, cd);
1743 }
1744
1745 static long cache_ioctl_pipefs(struct file *filp,
1746                               unsigned int cmd, unsigned long arg)
1747 {
1748         struct inode *inode = file_inode(filp);
1749         struct cache_detail *cd = RPC_I(inode)->private;
1750
1751         return cache_ioctl(inode, filp, cmd, arg, cd);
1752 }
1753
1754 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1755 {
1756         struct cache_detail *cd = RPC_I(inode)->private;
1757
1758         return cache_open(inode, filp, cd);
1759 }
1760
1761 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1762 {
1763         struct cache_detail *cd = RPC_I(inode)->private;
1764
1765         return cache_release(inode, filp, cd);
1766 }
1767
1768 const struct file_operations cache_file_operations_pipefs = {
1769         .owner          = THIS_MODULE,
1770         .llseek         = no_llseek,
1771         .read           = cache_read_pipefs,
1772         .write          = cache_write_pipefs,
1773         .poll           = cache_poll_pipefs,
1774         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1775         .open           = cache_open_pipefs,
1776         .release        = cache_release_pipefs,
1777 };
1778
1779 static int content_open_pipefs(struct inode *inode, struct file *filp)
1780 {
1781         struct cache_detail *cd = RPC_I(inode)->private;
1782
1783         return content_open(inode, filp, cd);
1784 }
1785
1786 static int content_release_pipefs(struct inode *inode, struct file *filp)
1787 {
1788         struct cache_detail *cd = RPC_I(inode)->private;
1789
1790         return content_release(inode, filp, cd);
1791 }
1792
1793 const struct file_operations content_file_operations_pipefs = {
1794         .open           = content_open_pipefs,
1795         .read           = seq_read,
1796         .llseek         = seq_lseek,
1797         .release        = content_release_pipefs,
1798 };
1799
1800 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1801 {
1802         struct cache_detail *cd = RPC_I(inode)->private;
1803
1804         return open_flush(inode, filp, cd);
1805 }
1806
1807 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1808 {
1809         struct cache_detail *cd = RPC_I(inode)->private;
1810
1811         return release_flush(inode, filp, cd);
1812 }
1813
1814 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1815                             size_t count, loff_t *ppos)
1816 {
1817         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1818
1819         return read_flush(filp, buf, count, ppos, cd);
1820 }
1821
1822 static ssize_t write_flush_pipefs(struct file *filp,
1823                                   const char __user *buf,
1824                                   size_t count, loff_t *ppos)
1825 {
1826         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1827
1828         return write_flush(filp, buf, count, ppos, cd);
1829 }
1830
1831 const struct file_operations cache_flush_operations_pipefs = {
1832         .open           = open_flush_pipefs,
1833         .read           = read_flush_pipefs,
1834         .write          = write_flush_pipefs,
1835         .release        = release_flush_pipefs,
1836         .llseek         = no_llseek,
1837 };
1838
1839 int sunrpc_cache_register_pipefs(struct dentry *parent,
1840                                  const char *name, umode_t umode,
1841                                  struct cache_detail *cd)
1842 {
1843         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1844         if (IS_ERR(dir))
1845                 return PTR_ERR(dir);
1846         cd->u.pipefs.dir = dir;
1847         return 0;
1848 }
1849 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1850
1851 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1852 {
1853         rpc_remove_cache_dir(cd->u.pipefs.dir);
1854         cd->u.pipefs.dir = NULL;
1855 }
1856 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1857