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         INIT_HLIST_NODE(&h->cache_list);
  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 *new = NULL, *freeme = NULL, *tmp = NULL;
  58         struct hlist_head *head;
  59
  60         head = &detail->hash_table[hash];
  61
  62         read_lock(&detail->hash_lock);
  63
  64         hlist_for_each_entry(tmp, head, cache_list) {
  65                 if (detail->match(tmp, key)) {
  66                         if (cache_is_expired(detail, tmp))
  67                                 /* This entry is expired, we will discard it. */
  68                                 break;
  69                         cache_get(tmp);
  70                         read_unlock(&detail->hash_lock);
  71                         return tmp;
  72                 }
  73         }
  74         read_unlock(&detail->hash_lock);
  75         /* Didn't find anything, insert an empty entry */
  76
  77         new = detail->alloc();
  78         if (!new)
  79                 return NULL;
  80         /* must fully initialise 'new', else
  81          * we might get lose if we need to
  82          * cache_put it soon.
  83          */
  84         cache_init(new);
  85         detail->init(new, key);
  86
  87         write_lock(&detail->hash_lock);
  88
  89         /* check if entry appeared while we slept */
  90         hlist_for_each_entry(tmp, head, cache_list) {
  91                 if (detail->match(tmp, key)) {
  92                         if (cache_is_expired(detail, tmp)) {
  93                                 hlist_del_init(&tmp->cache_list);
  94                                 detail->entries --;
  95                                 freeme = tmp;
  96                                 break;
  97                         }
  98                         cache_get(tmp);
  99                         write_unlock(&detail->hash_lock);
 100                         cache_put(new, detail);
 101                         return tmp;
 102                 }
 103         }
 104
 105         hlist_add_head(&new->cache_list, head);
 106         detail->entries++;
 107         cache_get(new);
 108         write_unlock(&detail->hash_lock);
 109
 110         if (freeme)
 111                 cache_put(freeme, detail);
 112         return new;
 113 }
 114 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
 115
 116
 117 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 118
 119 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
 120 {
 121         head->expiry_time = expiry;
 122         head->last_refresh = seconds_since_boot();
 123         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 124         set_bit(CACHE_VALID, &head->flags);
 125 }
 126
 127 static void cache_fresh_unlocked(struct cache_head *head,
 128                                  struct cache_detail *detail)
 129 {
 130         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 131                 cache_revisit_request(head);
 132                 cache_dequeue(detail, head);
 133         }
 134 }
 135
 136 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 137                                        struct cache_head *new, struct cache_head *old, int hash)
 138 {
 139         /* The 'old' entry is to be replaced by 'new'.
 140          * If 'old' is not VALID, we update it directly,
 141          * otherwise we need to replace it
 142          */
 143         struct cache_head *tmp;
 144
 145         if (!test_bit(CACHE_VALID, &old->flags)) {
 146                 write_lock(&detail->hash_lock);
 147                 if (!test_bit(CACHE_VALID, &old->flags)) {
 148                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 149                                 set_bit(CACHE_NEGATIVE, &old->flags);
 150                         else
 151                                 detail->update(old, new);
 152                         cache_fresh_locked(old, new->expiry_time);
 153                         write_unlock(&detail->hash_lock);
 154                         cache_fresh_unlocked(old, detail);
 155                         return old;
 156                 }
 157                 write_unlock(&detail->hash_lock);
 158         }
 159         /* We need to insert a new entry */
 160         tmp = detail->alloc();
 161         if (!tmp) {
 162                 cache_put(old, detail);
 163                 return NULL;
 164         }
 165         cache_init(tmp);
 166         detail->init(tmp, old);
 167
 168         write_lock(&detail->hash_lock);
 169         if (test_bit(CACHE_NEGATIVE, &new->flags))
 170                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 171         else
 172                 detail->update(tmp, new);
 173         hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
 174         detail->entries++;
 175         cache_get(tmp);
 176         cache_fresh_locked(tmp, new->expiry_time);
 177         cache_fresh_locked(old, 0);
 178         write_unlock(&detail->hash_lock);
 179         cache_fresh_unlocked(tmp, detail);
 180         cache_fresh_unlocked(old, detail);
 181         cache_put(old, detail);
 182         return tmp;
 183 }
 184 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 185
 186 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 187 {
 188         if (cd->cache_upcall)
 189                 return cd->cache_upcall(cd, h);
 190         return sunrpc_cache_pipe_upcall(cd, h);
 191 }
 192
 193 static inline int cache_is_valid(struct cache_head *h)
 194 {
 195         if (!test_bit(CACHE_VALID, &h->flags))
 196                 return -EAGAIN;
 197         else {
 198                 /* entry is valid */
 199                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 200                         return -ENOENT;
 201                 else {
 202                         /*
 203                          * In combination with write barrier in
 204                          * sunrpc_cache_update, ensures that anyone
 205                          * using the cache entry after this sees the
 206                          * updated contents:
 207                          */
 208                         smp_rmb();
 209                         return 0;
 210                 }
 211         }
 212 }
 213
 214 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 215 {
 216         int rv;
 217
 218         write_lock(&detail->hash_lock);
 219         rv = cache_is_valid(h);
 220         if (rv == -EAGAIN) {
 221                 set_bit(CACHE_NEGATIVE, &h->flags);
 222                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
 223                 rv = -ENOENT;
 224         }
 225         write_unlock(&detail->hash_lock);
 226         cache_fresh_unlocked(h, detail);
 227         return rv;
 228 }
 229
 230 /*
 231  * This is the generic cache management routine for all
 232  * the authentication caches.
 233  * It checks the currency of a cache item and will (later)
 234  * initiate an upcall to fill it if needed.
 235  *
 236  *
 237  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 238  * -EAGAIN if upcall is pending and request has been queued
 239  * -ETIMEDOUT if upcall failed or request could not be queue or
 240  *           upcall completed but item is still invalid (implying that
 241  *           the cache item has been replaced with a newer one).
 242  * -ENOENT if cache entry was negative
 243  */
 244 int cache_check(struct cache_detail *detail,
 245                     struct cache_head *h, struct cache_req *rqstp)
 246 {
 247         int rv;
 248         long refresh_age, age;
 249
 250         /* First decide return status as best we can */
 251         rv = cache_is_valid(h);
 252
 253         /* now see if we want to start an upcall */
 254         refresh_age = (h->expiry_time - h->last_refresh);
 255         age = seconds_since_boot() - h->last_refresh;
 256
 257         if (rqstp == NULL) {
 258                 if (rv == -EAGAIN)
 259                         rv = -ENOENT;
 260         } else if (rv == -EAGAIN ||
 261                    (h->expiry_time != 0 && age > refresh_age/2)) {
 262                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 263                                 refresh_age, age);
 264                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 265                         switch (cache_make_upcall(detail, h)) {
 266                         case -EINVAL:
 267                                 rv = try_to_negate_entry(detail, h);
 268                                 break;
 269                         case -EAGAIN:
 270                                 cache_fresh_unlocked(h, detail);
 271                                 break;
 272                         }
 273                 }
 274         }
 275
 276         if (rv == -EAGAIN) {
 277                 if (!cache_defer_req(rqstp, h)) {
 278                         /*
 279                          * Request was not deferred; handle it as best
 280                          * we can ourselves:
 281                          */
 282                         rv = cache_is_valid(h);
 283                         if (rv == -EAGAIN)
 284                                 rv = -ETIMEDOUT;
 285                 }
 286         }
 287         if (rv)
 288                 cache_put(h, detail);
 289         return rv;
 290 }
 291 EXPORT_SYMBOL_GPL(cache_check);
 292
 293 /*
 294  * caches need to be periodically cleaned.
 295  * For this we maintain a list of cache_detail and
 296  * a current pointer into that list and into the table
 297  * for that entry.
 298  *
 299  * Each time cache_clean is called it finds the next non-empty entry
 300  * in the current table and walks the list in that entry
 301  * looking for entries that can be removed.
 302  *
 303  * An entry gets removed if:
 304  * - The expiry is before current time
 305  * - The last_refresh time is before the flush_time for that cache
 306  *
 307  * later we might drop old entries with non-NEVER expiry if that table
 308  * is getting 'full' for some definition of 'full'
 309  *
 310  * The question of "how often to scan a table" is an interesting one
 311  * and is answered in part by the use of the "nextcheck" field in the
 312  * cache_detail.
 313  * When a scan of a table begins, the nextcheck field is set to a time
 314  * that is well into the future.
 315  * While scanning, if an expiry time is found that is earlier than the
 316  * current nextcheck time, nextcheck is set to that expiry time.
 317  * If the flush_time is ever set to a time earlier than the nextcheck
 318  * time, the nextcheck time is then set to that flush_time.
 319  *
 320  * A table is then only scanned if the current time is at least
 321  * the nextcheck time.
 322  *
 323  */
 324
 325 static LIST_HEAD(cache_list);
 326 static DEFINE_SPINLOCK(cache_list_lock);
 327 static struct cache_detail *current_detail;
 328 static int current_index;
 329
 330 static void do_cache_clean(struct work_struct *work);
 331 static struct delayed_work cache_cleaner;
 332
 333 void sunrpc_init_cache_detail(struct cache_detail *cd)
 334 {
 335         rwlock_init(&cd->hash_lock);
 336         INIT_LIST_HEAD(&cd->queue);
 337         spin_lock(&cache_list_lock);
 338         cd->nextcheck = 0;
 339         cd->entries = 0;
 340         atomic_set(&cd->readers, 0);
 341         cd->last_close = 0;
 342         cd->last_warn = -1;
 343         list_add(&cd->others, &cache_list);
 344         spin_unlock(&cache_list_lock);
 345
 346         /* start the cleaning process */
 347         schedule_delayed_work(&cache_cleaner, 0);
 348 }
 349 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 350
 351 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 352 {
 353         cache_purge(cd);
 354         spin_lock(&cache_list_lock);
 355         write_lock(&cd->hash_lock);
 356         if (cd->entries || atomic_read(&cd->inuse)) {
 357                 write_unlock(&cd->hash_lock);
 358                 spin_unlock(&cache_list_lock);
 359                 goto out;
 360         }
 361         if (current_detail == cd)
 362                 current_detail = NULL;
 363         list_del_init(&cd->others);
 364         write_unlock(&cd->hash_lock);
 365         spin_unlock(&cache_list_lock);
 366         if (list_empty(&cache_list)) {
 367                 /* module must be being unloaded so its safe to kill the worker */
 368                 cancel_delayed_work_sync(&cache_cleaner);
 369         }
 370         return;
 371 out:
 372         printk(KERN_ERR "RPC: failed to unregister %s cache\n", cd->name);
 373 }
 374 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 375
 376 /* clean cache tries to find something to clean
 377  * and cleans it.
 378  * It returns 1 if it cleaned something,
 379  *            0 if it didn't find anything this time
 380  *           -1 if it fell off the end of the list.
 381  */
 382 static int cache_clean(void)
 383 {
 384         int rv = 0;
 385         struct list_head *next;
 386
 387         spin_lock(&cache_list_lock);
 388
 389         /* find a suitable table if we don't already have one */
 390         while (current_detail == NULL ||
 391             current_index >= current_detail->hash_size) {
 392                 if (current_detail)
 393                         next = current_detail->others.next;
 394                 else
 395                         next = cache_list.next;
 396                 if (next == &cache_list) {
 397                         current_detail = NULL;
 398                         spin_unlock(&cache_list_lock);
 399                         return -1;
 400                 }
 401                 current_detail = list_entry(next, struct cache_detail, others);
 402                 if (current_detail->nextcheck > seconds_since_boot())
 403                         current_index = current_detail->hash_size;
 404                 else {
 405                         current_index = 0;
 406                         current_detail->nextcheck = seconds_since_boot()+30*60;
 407                 }
 408         }
 409
 410         /* find a non-empty bucket in the table */
 411         while (current_detail &&
 412                current_index < current_detail->hash_size &&
 413                hlist_empty(&current_detail->hash_table[current_index]))
 414                 current_index++;
 415
 416         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 417
 418         if (current_detail && current_index < current_detail->hash_size) {
 419                 struct cache_head *ch = NULL;
 420                 struct cache_detail *d;
 421                 struct hlist_head *head;
 422                 struct hlist_node *tmp;
 423
 424                 write_lock(&current_detail->hash_lock);
 425
 426                 /* Ok, now to clean this strand */
 427
 428                 head = &current_detail->hash_table[current_index];
 429                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 430                         if (current_detail->nextcheck > ch->expiry_time)
 431                                 current_detail->nextcheck = ch->expiry_time+1;
 432                         if (!cache_is_expired(current_detail, ch))
 433                                 continue;
 434
 435                         hlist_del_init(&ch->cache_list);
 436                         current_detail->entries--;
 437                         rv = 1;
 438                         break;
 439                 }
 440
 441                 write_unlock(&current_detail->hash_lock);
 442                 d = current_detail;
 443                 if (!ch)
 444                         current_index ++;
 445                 spin_unlock(&cache_list_lock);
 446                 if (ch) {
 447                         set_bit(CACHE_CLEANED, &ch->flags);
 448                         cache_fresh_unlocked(ch, d);
 449                         cache_put(ch, d);
 450                 }
 451         } else
 452                 spin_unlock(&cache_list_lock);
 453
 454         return rv;
 455 }
 456
 457 /*
 458  * We want to regularly clean the cache, so we need to schedule some work ...
 459  */
 460 static void do_cache_clean(struct work_struct *work)
 461 {
 462         int delay = 5;
 463         if (cache_clean() == -1)
 464                 delay = round_jiffies_relative(30*HZ);
 465
 466         if (list_empty(&cache_list))
 467                 delay = 0;
 468
 469         if (delay)
 470                 schedule_delayed_work(&cache_cleaner, delay);
 471 }
 472
 473
 474 /*
 475  * Clean all caches promptly.  This just calls cache_clean
 476  * repeatedly until we are sure that every cache has had a chance to
 477  * be fully cleaned
 478  */
 479 void cache_flush(void)
 480 {
 481         while (cache_clean() != -1)
 482                 cond_resched();
 483         while (cache_clean() != -1)
 484                 cond_resched();
 485 }
 486 EXPORT_SYMBOL_GPL(cache_flush);
 487
 488 void cache_purge(struct cache_detail *detail)
 489 {
 490         detail->flush_time = LONG_MAX;
 491         detail->nextcheck = seconds_since_boot();
 492         cache_flush();
 493         detail->flush_time = 1;
 494 }
 495 EXPORT_SYMBOL_GPL(cache_purge);
 496
 497
 498 /*
 499  * Deferral and Revisiting of Requests.
 500  *
 501  * If a cache lookup finds a pending entry, we
 502  * need to defer the request and revisit it later.
 503  * All deferred requests are stored in a hash table,
 504  * indexed by "struct cache_head *".
 505  * As it may be wasteful to store a whole request
 506  * structure, we allow the request to provide a
 507  * deferred form, which must contain a
 508  * 'struct cache_deferred_req'
 509  * This cache_deferred_req contains a method to allow
 510  * it to be revisited when cache info is available
 511  */
 512
 513 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 514 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 515
 516 #define DFR_MAX 300     /* ??? */
 517
 518 static DEFINE_SPINLOCK(cache_defer_lock);
 519 static LIST_HEAD(cache_defer_list);
 520 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 521 static int cache_defer_cnt;
 522
 523 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 524 {
 525         hlist_del_init(&dreq->hash);
 526         if (!list_empty(&dreq->recent)) {
 527                 list_del_init(&dreq->recent);
 528                 cache_defer_cnt--;
 529         }
 530 }
 531
 532 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 533 {
 534         int hash = DFR_HASH(item);
 535
 536         INIT_LIST_HEAD(&dreq->recent);
 537         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 538 }
 539
 540 static void setup_deferral(struct cache_deferred_req *dreq,
 541                            struct cache_head *item,
 542                            int count_me)
 543 {
 544
 545         dreq->item = item;
 546
 547         spin_lock(&cache_defer_lock);
 548
 549         __hash_deferred_req(dreq, item);
 550
 551         if (count_me) {
 552                 cache_defer_cnt++;
 553                 list_add(&dreq->recent, &cache_defer_list);
 554         }
 555
 556         spin_unlock(&cache_defer_lock);
 557
 558 }
 559
 560 struct thread_deferred_req {
 561         struct cache_deferred_req handle;
 562         struct completion completion;
 563 };
 564
 565 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 566 {
 567         struct thread_deferred_req *dr =
 568                 container_of(dreq, struct thread_deferred_req, handle);
 569         complete(&dr->completion);
 570 }
 571
 572 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 573 {
 574         struct thread_deferred_req sleeper;
 575         struct cache_deferred_req *dreq = &sleeper.handle;
 576
 577         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 578         dreq->revisit = cache_restart_thread;
 579
 580         setup_deferral(dreq, item, 0);
 581
 582         if (!test_bit(CACHE_PENDING, &item->flags) ||
 583             wait_for_completion_interruptible_timeout(
 584                     &sleeper.completion, req->thread_wait) <= 0) {
 585                 /* The completion wasn't completed, so we need
 586                  * to clean up
 587                  */
 588                 spin_lock(&cache_defer_lock);
 589                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 590                         __unhash_deferred_req(&sleeper.handle);
 591                         spin_unlock(&cache_defer_lock);
 592                 } else {
 593                         /* cache_revisit_request already removed
 594                          * this from the hash table, but hasn't
 595                          * called ->revisit yet.  It will very soon
 596                          * and we need to wait for it.
 597                          */
 598                         spin_unlock(&cache_defer_lock);
 599                         wait_for_completion(&sleeper.completion);
 600                 }
 601         }
 602 }
 603
 604 static void cache_limit_defers(void)
 605 {
 606         /* Make sure we haven't exceed the limit of allowed deferred
 607          * requests.
 608          */
 609         struct cache_deferred_req *discard = NULL;
 610
 611         if (cache_defer_cnt <= DFR_MAX)
 612                 return;
 613
 614         spin_lock(&cache_defer_lock);
 615
 616         /* Consider removing either the first or the last */
 617         if (cache_defer_cnt > DFR_MAX) {
 618                 if (prandom_u32() & 1)
 619                         discard = list_entry(cache_defer_list.next,
 620                                              struct cache_deferred_req, recent);
 621                 else
 622                         discard = list_entry(cache_defer_list.prev,
 623                                              struct cache_deferred_req, recent);
 624                 __unhash_deferred_req(discard);
 625         }
 626         spin_unlock(&cache_defer_lock);
 627         if (discard)
 628                 discard->revisit(discard, 1);
 629 }
 630
 631 /* Return true if and only if a deferred request is queued. */
 632 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 633 {
 634         struct cache_deferred_req *dreq;
 635
 636         if (req->thread_wait) {
 637                 cache_wait_req(req, item);
 638                 if (!test_bit(CACHE_PENDING, &item->flags))
 639                         return false;
 640         }
 641         dreq = req->defer(req);
 642         if (dreq == NULL)
 643                 return false;
 644         setup_deferral(dreq, item, 1);
 645         if (!test_bit(CACHE_PENDING, &item->flags))
 646                 /* Bit could have been cleared before we managed to
 647                  * set up the deferral, so need to revisit just in case
 648                  */
 649                 cache_revisit_request(item);
 650
 651         cache_limit_defers();
 652         return true;
 653 }
 654
 655 static void cache_revisit_request(struct cache_head *item)
 656 {
 657         struct cache_deferred_req *dreq;
 658         struct list_head pending;
 659         struct hlist_node *tmp;
 660         int hash = DFR_HASH(item);
 661
 662         INIT_LIST_HEAD(&pending);
 663         spin_lock(&cache_defer_lock);
 664
 665         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 666                 if (dreq->item == item) {
 667                         __unhash_deferred_req(dreq);
 668                         list_add(&dreq->recent, &pending);
 669                 }
 670
 671         spin_unlock(&cache_defer_lock);
 672
 673         while (!list_empty(&pending)) {
 674                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 675                 list_del_init(&dreq->recent);
 676                 dreq->revisit(dreq, 0);
 677         }
 678 }
 679
 680 void cache_clean_deferred(void *owner)
 681 {
 682         struct cache_deferred_req *dreq, *tmp;
 683         struct list_head pending;
 684
 685
 686         INIT_LIST_HEAD(&pending);
 687         spin_lock(&cache_defer_lock);
 688
 689         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 690                 if (dreq->owner == owner) {
 691                         __unhash_deferred_req(dreq);
 692                         list_add(&dreq->recent, &pending);
 693                 }
 694         }
 695         spin_unlock(&cache_defer_lock);
 696
 697         while (!list_empty(&pending)) {
 698                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 699                 list_del_init(&dreq->recent);
 700                 dreq->revisit(dreq, 1);
 701         }
 702 }
 703
 704 /*
 705  * communicate with user-space
 706  *
 707  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 708  * On read, you get a full request, or block.
 709  * On write, an update request is processed.
 710  * Poll works if anything to read, and always allows write.
 711  *
 712  * Implemented by linked list of requests.  Each open file has
 713  * a ->private that also exists in this list.  New requests are added
 714  * to the end and may wakeup and preceding readers.
 715  * New readers are added to the head.  If, on read, an item is found with
 716  * CACHE_UPCALLING clear, we free it from the list.
 717  *
 718  */
 719
 720 static DEFINE_SPINLOCK(queue_lock);
 721 static DEFINE_MUTEX(queue_io_mutex);
 722
 723 struct cache_queue {
 724         struct list_head        list;
 725         int                     reader; /* if 0, then request */
 726 };
 727 struct cache_request {
 728         struct cache_queue      q;
 729         struct cache_head       *item;
 730         char                    * buf;
 731         int                     len;
 732         int                     readers;
 733 };
 734 struct cache_reader {
 735         struct cache_queue      q;
 736         int                     offset; /* if non-0, we have a refcnt on next request */
 737 };
 738
 739 static int cache_request(struct cache_detail *detail,
 740                                struct cache_request *crq)
 741 {
 742         char *bp = crq->buf;
 743         int len = PAGE_SIZE;
 744
 745         detail->cache_request(detail, crq->item, &bp, &len);
 746         if (len < 0)
 747                 return -EAGAIN;
 748         return PAGE_SIZE - len;
 749 }
 750
 751 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 752                           loff_t *ppos, struct cache_detail *cd)
 753 {
 754         struct cache_reader *rp = filp->private_data;
 755         struct cache_request *rq;
 756         struct inode *inode = file_inode(filp);
 757         int err;
 758
 759         if (count == 0)
 760                 return 0;
 761
 762         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 763                               * readers on this file */
 764  again:
 765         spin_lock(&queue_lock);
 766         /* need to find next request */
 767         while (rp->q.list.next != &cd->queue &&
 768                list_entry(rp->q.list.next, struct cache_queue, list)
 769                ->reader) {
 770                 struct list_head *next = rp->q.list.next;
 771                 list_move(&rp->q.list, next);
 772         }
 773         if (rp->q.list.next == &cd->queue) {
 774                 spin_unlock(&queue_lock);
 775                 mutex_unlock(&inode->i_mutex);
 776                 WARN_ON_ONCE(rp->offset);
 777                 return 0;
 778         }
 779         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 780         WARN_ON_ONCE(rq->q.reader);
 781         if (rp->offset == 0)
 782                 rq->readers++;
 783         spin_unlock(&queue_lock);
 784
 785         if (rq->len == 0) {
 786                 err = cache_request(cd, rq);
 787                 if (err < 0)
 788                         goto out;
 789                 rq->len = err;
 790         }
 791
 792         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 793                 err = -EAGAIN;
 794                 spin_lock(&queue_lock);
 795                 list_move(&rp->q.list, &rq->q.list);
 796                 spin_unlock(&queue_lock);
 797         } else {
 798                 if (rp->offset + count > rq->len)
 799                         count = rq->len - rp->offset;
 800                 err = -EFAULT;
 801                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 802                         goto out;
 803                 rp->offset += count;
 804                 if (rp->offset >= rq->len) {
 805                         rp->offset = 0;
 806                         spin_lock(&queue_lock);
 807                         list_move(&rp->q.list, &rq->q.list);
 808                         spin_unlock(&queue_lock);
 809                 }
 810                 err = 0;
 811         }
 812  out:
 813         if (rp->offset == 0) {
 814                 /* need to release rq */
 815                 spin_lock(&queue_lock);
 816                 rq->readers--;
 817                 if (rq->readers == 0 &&
 818                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 819                         list_del(&rq->q.list);
 820                         spin_unlock(&queue_lock);
 821                         cache_put(rq->item, cd);
 822                         kfree(rq->buf);
 823                         kfree(rq);
 824                 } else
 825                         spin_unlock(&queue_lock);
 826         }
 827         if (err == -EAGAIN)
 828                 goto again;
 829         mutex_unlock(&inode->i_mutex);
 830         return err ? err :  count;
 831 }
 832
 833 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 834                                  size_t count, struct cache_detail *cd)
 835 {
 836         ssize_t ret;
 837
 838         if (count == 0)
 839                 return -EINVAL;
 840         if (copy_from_user(kaddr, buf, count))
 841                 return -EFAULT;
 842         kaddr[count] = '\0';
 843         ret = cd->cache_parse(cd, kaddr, count);
 844         if (!ret)
 845                 ret = count;
 846         return ret;
 847 }
 848
 849 static ssize_t cache_slow_downcall(const char __user *buf,
 850                                    size_t count, struct cache_detail *cd)
 851 {
 852         static char write_buf[8192]; /* protected by queue_io_mutex */
 853         ssize_t ret = -EINVAL;
 854
 855         if (count >= sizeof(write_buf))
 856                 goto out;
 857         mutex_lock(&queue_io_mutex);
 858         ret = cache_do_downcall(write_buf, buf, count, cd);
 859         mutex_unlock(&queue_io_mutex);
 860 out:
 861         return ret;
 862 }
 863
 864 static ssize_t cache_downcall(struct address_space *mapping,
 865                               const char __user *buf,
 866                               size_t count, struct cache_detail *cd)
 867 {
 868         struct page *page;
 869         char *kaddr;
 870         ssize_t ret = -ENOMEM;
 871
 872         if (count >= PAGE_CACHE_SIZE)
 873                 goto out_slow;
 874
 875         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 876         if (!page)
 877                 goto out_slow;
 878
 879         kaddr = kmap(page);
 880         ret = cache_do_downcall(kaddr, buf, count, cd);
 881         kunmap(page);
 882         unlock_page(page);
 883         page_cache_release(page);
 884         return ret;
 885 out_slow:
 886         return cache_slow_downcall(buf, count, cd);
 887 }
 888
 889 static ssize_t cache_write(struct file *filp, const char __user *buf,
 890                            size_t count, loff_t *ppos,
 891                            struct cache_detail *cd)
 892 {
 893         struct address_space *mapping = filp->f_mapping;
 894         struct inode *inode = file_inode(filp);
 895         ssize_t ret = -EINVAL;
 896
 897         if (!cd->cache_parse)
 898                 goto out;
 899
 900         mutex_lock(&inode->i_mutex);
 901         ret = cache_downcall(mapping, buf, count, cd);
 902         mutex_unlock(&inode->i_mutex);
 903 out:
 904         return ret;
 905 }
 906
 907 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 908
 909 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 910                                struct cache_detail *cd)
 911 {
 912         unsigned int mask;
 913         struct cache_reader *rp = filp->private_data;
 914         struct cache_queue *cq;
 915
 916         poll_wait(filp, &queue_wait, wait);
 917
 918         /* alway allow write */
 919         mask = POLLOUT | POLLWRNORM;
 920
 921         if (!rp)
 922                 return mask;
 923
 924         spin_lock(&queue_lock);
 925
 926         for (cq= &rp->q; &cq->list != &cd->queue;
 927              cq = list_entry(cq->list.next, struct cache_queue, list))
 928                 if (!cq->reader) {
 929                         mask |= POLLIN | POLLRDNORM;
 930                         break;
 931                 }
 932         spin_unlock(&queue_lock);
 933         return mask;
 934 }
 935
 936 static int cache_ioctl(struct inode *ino, struct file *filp,
 937                        unsigned int cmd, unsigned long arg,
 938                        struct cache_detail *cd)
 939 {
 940         int len = 0;
 941         struct cache_reader *rp = filp->private_data;
 942         struct cache_queue *cq;
 943
 944         if (cmd != FIONREAD || !rp)
 945                 return -EINVAL;
 946
 947         spin_lock(&queue_lock);
 948
 949         /* only find the length remaining in current request,
 950          * or the length of the next request
 951          */
 952         for (cq= &rp->q; &cq->list != &cd->queue;
 953              cq = list_entry(cq->list.next, struct cache_queue, list))
 954                 if (!cq->reader) {
 955                         struct cache_request *cr =
 956                                 container_of(cq, struct cache_request, q);
 957                         len = cr->len - rp->offset;
 958                         break;
 959                 }
 960         spin_unlock(&queue_lock);
 961
 962         return put_user(len, (int __user *)arg);
 963 }
 964
 965 static int cache_open(struct inode *inode, struct file *filp,
 966                       struct cache_detail *cd)
 967 {
 968         struct cache_reader *rp = NULL;
 969
 970         if (!cd || !try_module_get(cd->owner))
 971                 return -EACCES;
 972         nonseekable_open(inode, filp);
 973         if (filp->f_mode & FMODE_READ) {
 974                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 975                 if (!rp) {
 976                         module_put(cd->owner);
 977                         return -ENOMEM;
 978                 }
 979                 rp->offset = 0;
 980                 rp->q.reader = 1;
 981                 atomic_inc(&cd->readers);
 982                 spin_lock(&queue_lock);
 983                 list_add(&rp->q.list, &cd->queue);
 984                 spin_unlock(&queue_lock);
 985         }
 986         filp->private_data = rp;
 987         return 0;
 988 }
 989
 990 static int cache_release(struct inode *inode, struct file *filp,
 991                          struct cache_detail *cd)
 992 {
 993         struct cache_reader *rp = filp->private_data;
 994
 995         if (rp) {
 996                 spin_lock(&queue_lock);
 997                 if (rp->offset) {
 998                         struct cache_queue *cq;
 999                         for (cq= &rp->q; &cq->list != &cd->queue;
1000                              cq = list_entry(cq->list.next, struct cache_queue, list))
1001                                 if (!cq->reader) {
1002                                         container_of(cq, struct cache_request, q)
1003                                                 ->readers--;
1004                                         break;
1005                                 }
1006                         rp->offset = 0;
1007                 }
1008                 list_del(&rp->q.list);
1009                 spin_unlock(&queue_lock);
1010
1011                 filp->private_data = NULL;
1012                 kfree(rp);
1013
1014                 cd->last_close = seconds_since_boot();
1015                 atomic_dec(&cd->readers);
1016         }
1017         module_put(cd->owner);
1018         return 0;
1019 }
1020
1021
1022
1023 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1024 {
1025         struct cache_queue *cq, *tmp;
1026         struct cache_request *cr;
1027         struct list_head dequeued;
1028
1029         INIT_LIST_HEAD(&dequeued);
1030         spin_lock(&queue_lock);
1031         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1032                 if (!cq->reader) {
1033                         cr = container_of(cq, struct cache_request, q);
1034                         if (cr->item != ch)
1035                                 continue;
1036                         if (test_bit(CACHE_PENDING, &ch->flags))
1037                                 /* Lost a race and it is pending again */
1038                                 break;
1039                         if (cr->readers != 0)
1040                                 continue;
1041                         list_move(&cr->q.list, &dequeued);
1042                 }
1043         spin_unlock(&queue_lock);
1044         while (!list_empty(&dequeued)) {
1045                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1046                 list_del(&cr->q.list);
1047                 cache_put(cr->item, detail);
1048                 kfree(cr->buf);
1049                 kfree(cr);
1050         }
1051 }
1052
1053 /*
1054  * Support routines for text-based upcalls.
1055  * Fields are separated by spaces.
1056  * Fields are either mangled to quote space tab newline slosh with slosh
1057  * or a hexified with a leading \x
1058  * Record is terminated with newline.
1059  *
1060  */
1061
1062 void qword_add(char **bpp, int *lp, char *str)
1063 {
1064         char *bp = *bpp;
1065         int len = *lp;
1066         int ret;
1067
1068         if (len < 0) return;
1069
1070         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1071         if (ret >= len) {
1072                 bp += len;
1073                 len = -1;
1074         } else {
1075                 bp += ret;
1076                 len -= ret;
1077                 *bp++ = ' ';
1078                 len--;
1079         }
1080         *bpp = bp;
1081         *lp = len;
1082 }
1083 EXPORT_SYMBOL_GPL(qword_add);
1084
1085 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1086 {
1087         char *bp = *bpp;
1088         int len = *lp;
1089
1090         if (len < 0) return;
1091
1092         if (len > 2) {
1093                 *bp++ = '\\';
1094                 *bp++ = 'x';
1095                 len -= 2;
1096                 while (blen && len >= 2) {
1097                         bp = hex_byte_pack(bp, *buf++);
1098                         len -= 2;
1099                         blen--;
1100                 }
1101         }
1102         if (blen || len<1) len = -1;
1103         else {
1104                 *bp++ = ' ';
1105                 len--;
1106         }
1107         *bpp = bp;
1108         *lp = len;
1109 }
1110 EXPORT_SYMBOL_GPL(qword_addhex);
1111
1112 static void warn_no_listener(struct cache_detail *detail)
1113 {
1114         if (detail->last_warn != detail->last_close) {
1115                 detail->last_warn = detail->last_close;
1116                 if (detail->warn_no_listener)
1117                         detail->warn_no_listener(detail, detail->last_close != 0);
1118         }
1119 }
1120
1121 static bool cache_listeners_exist(struct cache_detail *detail)
1122 {
1123         if (atomic_read(&detail->readers))
1124                 return true;
1125         if (detail->last_close == 0)
1126                 /* This cache was never opened */
1127                 return false;
1128         if (detail->last_close < seconds_since_boot() - 30)
1129                 /*
1130                  * We allow for the possibility that someone might
1131                  * restart a userspace daemon without restarting the
1132                  * server; but after 30 seconds, we give up.
1133                  */
1134                  return false;
1135         return true;
1136 }
1137
1138 /*
1139  * register an upcall request to user-space and queue it up for read() by the
1140  * upcall daemon.
1141  *
1142  * Each request is at most one page long.
1143  */
1144 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1145 {
1146
1147         char *buf;
1148         struct cache_request *crq;
1149         int ret = 0;
1150
1151         if (!detail->cache_request)
1152                 return -EINVAL;
1153
1154         if (!cache_listeners_exist(detail)) {
1155                 warn_no_listener(detail);
1156                 return -EINVAL;
1157         }
1158         if (test_bit(CACHE_CLEANED, &h->flags))
1159                 /* Too late to make an upcall */
1160                 return -EAGAIN;
1161
1162         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1163         if (!buf)
1164                 return -EAGAIN;
1165
1166         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1167         if (!crq) {
1168                 kfree(buf);
1169                 return -EAGAIN;
1170         }
1171
1172         crq->q.reader = 0;
1173         crq->item = cache_get(h);
1174         crq->buf = buf;
1175         crq->len = 0;
1176         crq->readers = 0;
1177         spin_lock(&queue_lock);
1178         if (test_bit(CACHE_PENDING, &h->flags))
1179                 list_add_tail(&crq->q.list, &detail->queue);
1180         else
1181                 /* Lost a race, no longer PENDING, so don't enqueue */
1182                 ret = -EAGAIN;
1183         spin_unlock(&queue_lock);
1184         wake_up(&queue_wait);
1185         if (ret == -EAGAIN) {
1186                 kfree(buf);
1187                 kfree(crq);
1188         }
1189         return ret;
1190 }
1191 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1192
1193 /*
1194  * parse a message from user-space and pass it
1195  * to an appropriate cache
1196  * Messages are, like requests, separated into fields by
1197  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1198  *
1199  * Message is
1200  *   reply cachename expiry key ... content....
1201  *
1202  * key and content are both parsed by cache
1203  */
1204
1205 int qword_get(char **bpp, char *dest, int bufsize)
1206 {
1207         /* return bytes copied, or -1 on error */
1208         char *bp = *bpp;
1209         int len = 0;
1210
1211         while (*bp == ' ') bp++;
1212
1213         if (bp[0] == '\\' && bp[1] == 'x') {
1214                 /* HEX STRING */
1215                 bp += 2;
1216                 while (len < bufsize) {
1217                         int h, l;
1218
1219                         h = hex_to_bin(bp[0]);
1220                         if (h < 0)
1221                                 break;
1222
1223                         l = hex_to_bin(bp[1]);
1224                         if (l < 0)
1225                                 break;
1226
1227                         *dest++ = (h << 4) | l;
1228                         bp += 2;
1229                         len++;
1230                 }
1231         } else {
1232                 /* text with \nnn octal quoting */
1233                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1234                         if (*bp == '\\' &&
1235                             isodigit(bp[1]) && (bp[1] <= '3') &&
1236                             isodigit(bp[2]) &&
1237                             isodigit(bp[3])) {
1238                                 int byte = (*++bp -'0');
1239                                 bp++;
1240                                 byte = (byte << 3) | (*bp++ - '0');
1241                                 byte = (byte << 3) | (*bp++ - '0');
1242                                 *dest++ = byte;
1243                                 len++;
1244                         } else {
1245                                 *dest++ = *bp++;
1246                                 len++;
1247                         }
1248                 }
1249         }
1250
1251         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1252                 return -1;
1253         while (*bp == ' ') bp++;
1254         *bpp = bp;
1255         *dest = '\0';
1256         return len;
1257 }
1258 EXPORT_SYMBOL_GPL(qword_get);
1259
1260
1261 /*
1262  * support /proc/sunrpc/cache/$CACHENAME/content
1263  * as a seqfile.
1264  * We call ->cache_show passing NULL for the item to
1265  * get a header, then pass each real item in the cache
1266  */
1267
1268 void *cache_seq_start(struct seq_file *m, loff_t *pos)
1269         __acquires(cd->hash_lock)
1270 {
1271         loff_t n = *pos;
1272         unsigned int hash, entry;
1273         struct cache_head *ch;
1274         struct cache_detail *cd = m->private;
1275
1276         read_lock(&cd->hash_lock);
1277         if (!n--)
1278                 return SEQ_START_TOKEN;
1279         hash = n >> 32;
1280         entry = n & ((1LL<<32) - 1);
1281
1282         hlist_for_each_entry(ch, &cd->hash_table[hash], cache_list)
1283                 if (!entry--)
1284                         return ch;
1285         n &= ~((1LL<<32) - 1);
1286         do {
1287                 hash++;
1288                 n += 1LL<<32;
1289         } while(hash < cd->hash_size &&
1290                 hlist_empty(&cd->hash_table[hash]));
1291         if (hash >= cd->hash_size)
1292                 return NULL;
1293         *pos = n+1;
1294         return hlist_entry_safe(cd->hash_table[hash].first,
1295                                 struct cache_head, cache_list);
1296 }
1297 EXPORT_SYMBOL_GPL(cache_seq_start);
1298
1299 void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1300 {
1301         struct cache_head *ch = p;
1302         int hash = (*pos >> 32);
1303         struct cache_detail *cd = m->private;
1304
1305         if (p == SEQ_START_TOKEN)
1306                 hash = 0;
1307         else if (ch->cache_list.next == NULL) {
1308                 hash++;
1309                 *pos += 1LL<<32;
1310         } else {
1311                 ++*pos;
1312                 return hlist_entry_safe(ch->cache_list.next,
1313                                         struct cache_head, cache_list);
1314         }
1315         *pos &= ~((1LL<<32) - 1);
1316         while (hash < cd->hash_size &&
1317                hlist_empty(&cd->hash_table[hash])) {
1318                 hash++;
1319                 *pos += 1LL<<32;
1320         }
1321         if (hash >= cd->hash_size)
1322                 return NULL;
1323         ++*pos;
1324         return hlist_entry_safe(cd->hash_table[hash].first,
1325                                 struct cache_head, cache_list);
1326 }
1327 EXPORT_SYMBOL_GPL(cache_seq_next);
1328
1329 void cache_seq_stop(struct seq_file *m, void *p)
1330         __releases(cd->hash_lock)
1331 {
1332         struct cache_detail *cd = m->private;
1333         read_unlock(&cd->hash_lock);
1334 }
1335 EXPORT_SYMBOL_GPL(cache_seq_stop);
1336
1337 static int c_show(struct seq_file *m, void *p)
1338 {
1339         struct cache_head *cp = p;
1340         struct cache_detail *cd = m->private;
1341
1342         if (p == SEQ_START_TOKEN)
1343                 return cd->cache_show(m, cd, NULL);
1344
1345         ifdebug(CACHE)
1346                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1347                            convert_to_wallclock(cp->expiry_time),
1348                            atomic_read(&cp->ref.refcount), cp->flags);
1349         cache_get(cp);
1350         if (cache_check(cd, cp, NULL))
1351                 /* cache_check does a cache_put on failure */
1352                 seq_printf(m, "# ");
1353         else {
1354                 if (cache_is_expired(cd, cp))
1355                         seq_printf(m, "# ");
1356                 cache_put(cp, cd);
1357         }
1358
1359         return cd->cache_show(m, cd, cp);
1360 }
1361
1362 static const struct seq_operations cache_content_op = {
1363         .start  = cache_seq_start,
1364         .next   = cache_seq_next,
1365         .stop   = cache_seq_stop,
1366         .show   = c_show,
1367 };
1368
1369 static int content_open(struct inode *inode, struct file *file,
1370                         struct cache_detail *cd)
1371 {
1372         struct seq_file *seq;
1373         int err;
1374
1375         if (!cd || !try_module_get(cd->owner))
1376                 return -EACCES;
1377
1378         err = seq_open(file, &cache_content_op);
1379         if (err) {
1380                 module_put(cd->owner);
1381                 return err;
1382         }
1383
1384         seq = file->private_data;
1385         seq->private = cd;
1386         return 0;
1387 }
1388
1389 static int content_release(struct inode *inode, struct file *file,
1390                 struct cache_detail *cd)
1391 {
1392         int ret = seq_release(inode, file);
1393         module_put(cd->owner);
1394         return ret;
1395 }
1396
1397 static int open_flush(struct inode *inode, struct file *file,
1398                         struct cache_detail *cd)
1399 {
1400         if (!cd || !try_module_get(cd->owner))
1401                 return -EACCES;
1402         return nonseekable_open(inode, file);
1403 }
1404
1405 static int release_flush(struct inode *inode, struct file *file,
1406                         struct cache_detail *cd)
1407 {
1408         module_put(cd->owner);
1409         return 0;
1410 }
1411
1412 static ssize_t read_flush(struct file *file, char __user *buf,
1413                           size_t count, loff_t *ppos,
1414                           struct cache_detail *cd)
1415 {
1416         char tbuf[22];
1417         unsigned long p = *ppos;
1418         size_t len;
1419
1420         snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1421         len = strlen(tbuf);
1422         if (p >= len)
1423                 return 0;
1424         len -= p;
1425         if (len > count)
1426                 len = count;
1427         if (copy_to_user(buf, (void*)(tbuf+p), len))
1428                 return -EFAULT;
1429         *ppos += len;
1430         return len;
1431 }
1432
1433 static ssize_t write_flush(struct file *file, const char __user *buf,
1434                            size_t count, loff_t *ppos,
1435                            struct cache_detail *cd)
1436 {
1437         char tbuf[20];
1438         char *bp, *ep;
1439
1440         if (*ppos || count > sizeof(tbuf)-1)
1441                 return -EINVAL;
1442         if (copy_from_user(tbuf, buf, count))
1443                 return -EFAULT;
1444         tbuf[count] = 0;
1445         simple_strtoul(tbuf, &ep, 0);
1446         if (*ep && *ep != '\n')
1447                 return -EINVAL;
1448
1449         bp = tbuf;
1450         cd->flush_time = get_expiry(&bp);
1451         cd->nextcheck = seconds_since_boot();
1452         cache_flush();
1453
1454         *ppos += count;
1455         return count;
1456 }
1457
1458 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1459                                  size_t count, loff_t *ppos)
1460 {
1461         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1462
1463         return cache_read(filp, buf, count, ppos, cd);
1464 }
1465
1466 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1467                                   size_t count, loff_t *ppos)
1468 {
1469         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1470
1471         return cache_write(filp, buf, count, ppos, cd);
1472 }
1473
1474 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1475 {
1476         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1477
1478         return cache_poll(filp, wait, cd);
1479 }
1480
1481 static long cache_ioctl_procfs(struct file *filp,
1482                                unsigned int cmd, unsigned long arg)
1483 {
1484         struct inode *inode = file_inode(filp);
1485         struct cache_detail *cd = PDE_DATA(inode);
1486
1487         return cache_ioctl(inode, filp, cmd, arg, cd);
1488 }
1489
1490 static int cache_open_procfs(struct inode *inode, struct file *filp)
1491 {
1492         struct cache_detail *cd = PDE_DATA(inode);
1493
1494         return cache_open(inode, filp, cd);
1495 }
1496
1497 static int cache_release_procfs(struct inode *inode, struct file *filp)
1498 {
1499         struct cache_detail *cd = PDE_DATA(inode);
1500
1501         return cache_release(inode, filp, cd);
1502 }
1503
1504 static const struct file_operations cache_file_operations_procfs = {
1505         .owner          = THIS_MODULE,
1506         .llseek         = no_llseek,
1507         .read           = cache_read_procfs,
1508         .write          = cache_write_procfs,
1509         .poll           = cache_poll_procfs,
1510         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1511         .open           = cache_open_procfs,
1512         .release        = cache_release_procfs,
1513 };
1514
1515 static int content_open_procfs(struct inode *inode, struct file *filp)
1516 {
1517         struct cache_detail *cd = PDE_DATA(inode);
1518
1519         return content_open(inode, filp, cd);
1520 }
1521
1522 static int content_release_procfs(struct inode *inode, struct file *filp)
1523 {
1524         struct cache_detail *cd = PDE_DATA(inode);
1525
1526         return content_release(inode, filp, cd);
1527 }
1528
1529 static const struct file_operations content_file_operations_procfs = {
1530         .open           = content_open_procfs,
1531         .read           = seq_read,
1532         .llseek         = seq_lseek,
1533         .release        = content_release_procfs,
1534 };
1535
1536 static int open_flush_procfs(struct inode *inode, struct file *filp)
1537 {
1538         struct cache_detail *cd = PDE_DATA(inode);
1539
1540         return open_flush(inode, filp, cd);
1541 }
1542
1543 static int release_flush_procfs(struct inode *inode, struct file *filp)
1544 {
1545         struct cache_detail *cd = PDE_DATA(inode);
1546
1547         return release_flush(inode, filp, cd);
1548 }
1549
1550 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1551                             size_t count, loff_t *ppos)
1552 {
1553         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1554
1555         return read_flush(filp, buf, count, ppos, cd);
1556 }
1557
1558 static ssize_t write_flush_procfs(struct file *filp,
1559                                   const char __user *buf,
1560                                   size_t count, loff_t *ppos)
1561 {
1562         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1563
1564         return write_flush(filp, buf, count, ppos, cd);
1565 }
1566
1567 static const struct file_operations cache_flush_operations_procfs = {
1568         .open           = open_flush_procfs,
1569         .read           = read_flush_procfs,
1570         .write          = write_flush_procfs,
1571         .release        = release_flush_procfs,
1572         .llseek         = no_llseek,
1573 };
1574
1575 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1576 {
1577         struct sunrpc_net *sn;
1578
1579         if (cd->u.procfs.proc_ent == NULL)
1580                 return;
1581         if (cd->u.procfs.flush_ent)
1582                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1583         if (cd->u.procfs.channel_ent)
1584                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1585         if (cd->u.procfs.content_ent)
1586                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1587         cd->u.procfs.proc_ent = NULL;
1588         sn = net_generic(net, sunrpc_net_id);
1589         remove_proc_entry(cd->name, sn->proc_net_rpc);
1590 }
1591
1592 #ifdef CONFIG_PROC_FS
1593 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1594 {
1595         struct proc_dir_entry *p;
1596         struct sunrpc_net *sn;
1597
1598         sn = net_generic(net, sunrpc_net_id);
1599         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1600         if (cd->u.procfs.proc_ent == NULL)
1601                 goto out_nomem;
1602         cd->u.procfs.channel_ent = NULL;
1603         cd->u.procfs.content_ent = NULL;
1604
1605         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1606                              cd->u.procfs.proc_ent,
1607                              &cache_flush_operations_procfs, cd);
1608         cd->u.procfs.flush_ent = p;
1609         if (p == NULL)
1610                 goto out_nomem;
1611
1612         if (cd->cache_request || cd->cache_parse) {
1613                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1614                                      cd->u.procfs.proc_ent,
1615                                      &cache_file_operations_procfs, cd);
1616                 cd->u.procfs.channel_ent = p;
1617                 if (p == NULL)
1618                         goto out_nomem;
1619         }
1620         if (cd->cache_show) {
1621                 p = proc_create_data("content", S_IFREG|S_IRUSR,
1622                                 cd->u.procfs.proc_ent,
1623                                 &content_file_operations_procfs, cd);
1624                 cd->u.procfs.content_ent = p;
1625                 if (p == NULL)
1626                         goto out_nomem;
1627         }
1628         return 0;
1629 out_nomem:
1630         remove_cache_proc_entries(cd, net);
1631         return -ENOMEM;
1632 }
1633 #else /* CONFIG_PROC_FS */
1634 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1635 {
1636         return 0;
1637 }
1638 #endif
1639
1640 void __init cache_initialize(void)
1641 {
1642         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1643 }
1644
1645 int cache_register_net(struct cache_detail *cd, struct net *net)
1646 {
1647         int ret;
1648
1649         sunrpc_init_cache_detail(cd);
1650         ret = create_cache_proc_entries(cd, net);
1651         if (ret)
1652                 sunrpc_destroy_cache_detail(cd);
1653         return ret;
1654 }
1655 EXPORT_SYMBOL_GPL(cache_register_net);
1656
1657 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1658 {
1659         remove_cache_proc_entries(cd, net);
1660         sunrpc_destroy_cache_detail(cd);
1661 }
1662 EXPORT_SYMBOL_GPL(cache_unregister_net);
1663
1664 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1665 {
1666         struct cache_detail *cd;
1667         int i;
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 hlist_head),
1674                                  GFP_KERNEL);
1675         if (cd->hash_table == NULL) {
1676                 kfree(cd);
1677                 return ERR_PTR(-ENOMEM);
1678         }
1679
1680         for (i = 0; i < cd->hash_size; i++)
1681                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1682         cd->net = net;
1683         return cd;
1684 }
1685 EXPORT_SYMBOL_GPL(cache_create_net);
1686
1687 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1688 {
1689         kfree(cd->hash_table);
1690         kfree(cd);
1691 }
1692 EXPORT_SYMBOL_GPL(cache_destroy_net);
1693
1694 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1695                                  size_t count, loff_t *ppos)
1696 {
1697         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1698
1699         return cache_read(filp, buf, count, ppos, cd);
1700 }
1701
1702 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1703                                   size_t count, loff_t *ppos)
1704 {
1705         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1706
1707         return cache_write(filp, buf, count, ppos, cd);
1708 }
1709
1710 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1711 {
1712         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1713
1714         return cache_poll(filp, wait, cd);
1715 }
1716
1717 static long cache_ioctl_pipefs(struct file *filp,
1718                               unsigned int cmd, unsigned long arg)
1719 {
1720         struct inode *inode = file_inode(filp);
1721         struct cache_detail *cd = RPC_I(inode)->private;
1722
1723         return cache_ioctl(inode, filp, cmd, arg, cd);
1724 }
1725
1726 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1727 {
1728         struct cache_detail *cd = RPC_I(inode)->private;
1729
1730         return cache_open(inode, filp, cd);
1731 }
1732
1733 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1734 {
1735         struct cache_detail *cd = RPC_I(inode)->private;
1736
1737         return cache_release(inode, filp, cd);
1738 }
1739
1740 const struct file_operations cache_file_operations_pipefs = {
1741         .owner          = THIS_MODULE,
1742         .llseek         = no_llseek,
1743         .read           = cache_read_pipefs,
1744         .write          = cache_write_pipefs,
1745         .poll           = cache_poll_pipefs,
1746         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1747         .open           = cache_open_pipefs,
1748         .release        = cache_release_pipefs,
1749 };
1750
1751 static int content_open_pipefs(struct inode *inode, struct file *filp)
1752 {
1753         struct cache_detail *cd = RPC_I(inode)->private;
1754
1755         return content_open(inode, filp, cd);
1756 }
1757
1758 static int content_release_pipefs(struct inode *inode, struct file *filp)
1759 {
1760         struct cache_detail *cd = RPC_I(inode)->private;
1761
1762         return content_release(inode, filp, cd);
1763 }
1764
1765 const struct file_operations content_file_operations_pipefs = {
1766         .open           = content_open_pipefs,
1767         .read           = seq_read,
1768         .llseek         = seq_lseek,
1769         .release        = content_release_pipefs,
1770 };
1771
1772 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1773 {
1774         struct cache_detail *cd = RPC_I(inode)->private;
1775
1776         return open_flush(inode, filp, cd);
1777 }
1778
1779 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1780 {
1781         struct cache_detail *cd = RPC_I(inode)->private;
1782
1783         return release_flush(inode, filp, cd);
1784 }
1785
1786 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1787                             size_t count, loff_t *ppos)
1788 {
1789         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1790
1791         return read_flush(filp, buf, count, ppos, cd);
1792 }
1793
1794 static ssize_t write_flush_pipefs(struct file *filp,
1795                                   const char __user *buf,
1796                                   size_t count, loff_t *ppos)
1797 {
1798         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1799
1800         return write_flush(filp, buf, count, ppos, cd);
1801 }
1802
1803 const struct file_operations cache_flush_operations_pipefs = {
1804         .open           = open_flush_pipefs,
1805         .read           = read_flush_pipefs,
1806         .write          = write_flush_pipefs,
1807         .release        = release_flush_pipefs,
1808         .llseek         = no_llseek,
1809 };
1810
1811 int sunrpc_cache_register_pipefs(struct dentry *parent,
1812                                  const char *name, umode_t umode,
1813                                  struct cache_detail *cd)
1814 {
1815         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1816         if (IS_ERR(dir))
1817                 return PTR_ERR(dir);
1818         cd->u.pipefs.dir = dir;
1819         return 0;
1820 }
1821 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1822
1823 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1824 {
1825         rpc_remove_cache_dir(cd->u.pipefs.dir);
1826         cd->u.pipefs.dir = NULL;
1827 }
1828 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1829