net/sunrpc/cache.c

   1 /*
   2  * net/sunrpc/cache.c
   3  *
   4  * Generic code for various authentication-related caches
   5  * used by sunrpc clients and servers.
   6  *
   7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   8  *
   9  * Released under terms in GPL version 2.  See COPYING.
  10  *
  11  */
  12
  13 #include <linux/types.h>
  14 #include <linux/fs.h>
  15 #include <linux/file.h>
  16 #include <linux/slab.h>
  17 #include <linux/signal.h>
  18 #include <linux/sched.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/module.h>
  22 #include <linux/ctype.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/poll.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/net.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/mutex.h>
  30 #include <linux/pagemap.h>
  31 #include <asm/ioctls.h>
  32 #include <linux/sunrpc/types.h>
  33 #include <linux/sunrpc/cache.h>
  34 #include <linux/sunrpc/stats.h>
  35 #include <linux/sunrpc/rpc_pipe_fs.h>
  36 #include "netns.h"
  37
  38 #define  RPCDBG_FACILITY RPCDBG_CACHE
  39
  40 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  41 static void cache_revisit_request(struct cache_head *item);
  42
  43 static void cache_init(struct cache_head *h)
  44 {
  45         time_t now = seconds_since_boot();
  46         h->next = NULL;
  47         h->flags = 0;
  48         kref_init(&h->ref);
  49         h->expiry_time = now + CACHE_NEW_EXPIRY;
  50         h->last_refresh = now;
  51 }
  52
  53 static inline int cache_is_expired(struct cache_detail *detail, struct cache_head *h)
  54 {
  55         return  (h->expiry_time < seconds_since_boot()) ||
  56                 (detail->flush_time > h->last_refresh);
  57 }
  58
  59 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  60                                        struct cache_head *key, int hash)
  61 {
  62         struct cache_head **head,  **hp;
  63         struct cache_head *new = NULL, *freeme = NULL;
  64
  65         head = &detail->hash_table[hash];
  66
  67         read_lock(&detail->hash_lock);
  68
  69         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  70                 struct cache_head *tmp = *hp;
  71                 if (detail->match(tmp, key)) {
  72                         if (cache_is_expired(detail, tmp))
  73                                 /* This entry is expired, we will discard it. */
  74                                 break;
  75                         cache_get(tmp);
  76                         read_unlock(&detail->hash_lock);
  77                         return tmp;
  78                 }
  79         }
  80         read_unlock(&detail->hash_lock);
  81         /* Didn't find anything, insert an empty entry */
  82
  83         new = detail->alloc();
  84         if (!new)
  85                 return NULL;
  86         /* must fully initialise 'new', else
  87          * we might get lose if we need to
  88          * cache_put it soon.
  89          */
  90         cache_init(new);
  91         detail->init(new, key);
  92
  93         write_lock(&detail->hash_lock);
  94
  95         /* check if entry appeared while we slept */
  96         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  97                 struct cache_head *tmp = *hp;
  98                 if (detail->match(tmp, key)) {
  99                         if (cache_is_expired(detail, tmp)) {
 100                                 *hp = tmp->next;
 101                                 tmp->next = NULL;
 102                                 detail->entries --;
 103                                 freeme = tmp;
 104                                 break;
 105                         }
 106                         cache_get(tmp);
 107                         write_unlock(&detail->hash_lock);
 108                         cache_put(new, detail);
 109                         return tmp;
 110                 }
 111         }
 112         new->next = *head;
 113         *head = new;
 114         detail->entries++;
 115         cache_get(new);
 116         write_unlock(&detail->hash_lock);
 117
 118         if (freeme)
 119                 cache_put(freeme, detail);
 120         return new;
 121 }
 122 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
 123
 124
 125 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 126
 127 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
 128 {
 129         head->expiry_time = expiry;
 130         head->last_refresh = seconds_since_boot();
 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 **head;
 152         struct cache_head *tmp;
 153
 154         if (!test_bit(CACHE_VALID, &old->flags)) {
 155                 write_lock(&detail->hash_lock);
 156                 if (!test_bit(CACHE_VALID, &old->flags)) {
 157                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 158                                 set_bit(CACHE_NEGATIVE, &old->flags);
 159                         else
 160                                 detail->update(old, new);
 161                         cache_fresh_locked(old, new->expiry_time);
 162                         write_unlock(&detail->hash_lock);
 163                         cache_fresh_unlocked(old, detail);
 164                         return old;
 165                 }
 166                 write_unlock(&detail->hash_lock);
 167         }
 168         /* We need to insert a new entry */
 169         tmp = detail->alloc();
 170         if (!tmp) {
 171                 cache_put(old, detail);
 172                 return NULL;
 173         }
 174         cache_init(tmp);
 175         detail->init(tmp, old);
 176         head = &detail->hash_table[hash];
 177
 178         write_lock(&detail->hash_lock);
 179         if (test_bit(CACHE_NEGATIVE, &new->flags))
 180                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 181         else
 182                 detail->update(tmp, new);
 183         tmp->next = *head;
 184         *head = tmp;
 185         detail->entries++;
 186         cache_get(tmp);
 187         cache_fresh_locked(tmp, new->expiry_time);
 188         cache_fresh_locked(old, 0);
 189         write_unlock(&detail->hash_lock);
 190         cache_fresh_unlocked(tmp, detail);
 191         cache_fresh_unlocked(old, detail);
 192         cache_put(old, detail);
 193         return tmp;
 194 }
 195 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 196
 197 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 198 {
 199         if (cd->cache_upcall)
 200                 return cd->cache_upcall(cd, h);
 201         return sunrpc_cache_pipe_upcall(cd, h);
 202 }
 203
 204 static inline int cache_is_valid(struct cache_head *h)
 205 {
 206         if (!test_bit(CACHE_VALID, &h->flags))
 207                 return -EAGAIN;
 208         else {
 209                 /* entry is valid */
 210                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 211                         return -ENOENT;
 212                 else {
 213                         /*
 214                          * In combination with write barrier in
 215                          * sunrpc_cache_update, ensures that anyone
 216                          * using the cache entry after this sees the
 217                          * updated contents:
 218                          */
 219                         smp_rmb();
 220                         return 0;
 221                 }
 222         }
 223 }
 224
 225 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 226 {
 227         int rv;
 228
 229         write_lock(&detail->hash_lock);
 230         rv = cache_is_valid(h);
 231         if (rv != -EAGAIN) {
 232                 write_unlock(&detail->hash_lock);
 233                 return rv;
 234         }
 235         set_bit(CACHE_NEGATIVE, &h->flags);
 236         cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
 237         write_unlock(&detail->hash_lock);
 238         cache_fresh_unlocked(h, detail);
 239         return -ENOENT;
 240 }
 241
 242 /*
 243  * This is the generic cache management routine for all
 244  * the authentication caches.
 245  * It checks the currency of a cache item and will (later)
 246  * initiate an upcall to fill it if needed.
 247  *
 248  *
 249  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 250  * -EAGAIN if upcall is pending and request has been queued
 251  * -ETIMEDOUT if upcall failed or request could not be queue or
 252  *           upcall completed but item is still invalid (implying that
 253  *           the cache item has been replaced with a newer one).
 254  * -ENOENT if cache entry was negative
 255  */
 256 int cache_check(struct cache_detail *detail,
 257                     struct cache_head *h, struct cache_req *rqstp)
 258 {
 259         int rv;
 260         long refresh_age, age;
 261
 262         /* First decide return status as best we can */
 263         rv = cache_is_valid(h);
 264
 265         /* now see if we want to start an upcall */
 266         refresh_age = (h->expiry_time - h->last_refresh);
 267         age = seconds_since_boot() - h->last_refresh;
 268
 269         if (rqstp == NULL) {
 270                 if (rv == -EAGAIN)
 271                         rv = -ENOENT;
 272         } else if (rv == -EAGAIN || age > refresh_age/2) {
 273                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 274                                 refresh_age, age);
 275                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 276                         switch (cache_make_upcall(detail, h)) {
 277                         case -EINVAL:
 278                                 clear_bit(CACHE_PENDING, &h->flags);
 279                                 cache_revisit_request(h);
 280                                 rv = try_to_negate_entry(detail, h);
 281                                 break;
 282                         case -EAGAIN:
 283                                 clear_bit(CACHE_PENDING, &h->flags);
 284                                 cache_revisit_request(h);
 285                                 break;
 286                         }
 287                 }
 288         }
 289
 290         if (rv == -EAGAIN) {
 291                 if (!cache_defer_req(rqstp, h)) {
 292                         /*
 293                          * Request was not deferred; handle it as best
 294                          * we can ourselves:
 295                          */
 296                         rv = cache_is_valid(h);
 297                         if (rv == -EAGAIN)
 298                                 rv = -ETIMEDOUT;
 299                 }
 300         }
 301         if (rv)
 302                 cache_put(h, detail);
 303         return rv;
 304 }
 305 EXPORT_SYMBOL_GPL(cache_check);
 306
 307 /*
 308  * caches need to be periodically cleaned.
 309  * For this we maintain a list of cache_detail and
 310  * a current pointer into that list and into the table
 311  * for that entry.
 312  *
 313  * Each time clean_cache is called it finds the next non-empty entry
 314  * in the current table and walks the list in that entry
 315  * looking for entries that can be removed.
 316  *
 317  * An entry gets removed if:
 318  * - The expiry is before current time
 319  * - The last_refresh time is before the flush_time for that cache
 320  *
 321  * later we might drop old entries with non-NEVER expiry if that table
 322  * is getting 'full' for some definition of 'full'
 323  *
 324  * The question of "how often to scan a table" is an interesting one
 325  * and is answered in part by the use of the "nextcheck" field in the
 326  * cache_detail.
 327  * When a scan of a table begins, the nextcheck field is set to a time
 328  * that is well into the future.
 329  * While scanning, if an expiry time is found that is earlier than the
 330  * current nextcheck time, nextcheck is set to that expiry time.
 331  * If the flush_time is ever set to a time earlier than the nextcheck
 332  * time, the nextcheck time is then set to that flush_time.
 333  *
 334  * A table is then only scanned if the current time is at least
 335  * the nextcheck time.
 336  *
 337  */
 338
 339 static LIST_HEAD(cache_list);
 340 static DEFINE_SPINLOCK(cache_list_lock);
 341 static struct cache_detail *current_detail;
 342 static int current_index;
 343
 344 static void do_cache_clean(struct work_struct *work);
 345 static struct delayed_work cache_cleaner;
 346
 347 void sunrpc_init_cache_detail(struct cache_detail *cd)
 348 {
 349         rwlock_init(&cd->hash_lock);
 350         INIT_LIST_HEAD(&cd->queue);
 351         spin_lock(&cache_list_lock);
 352         cd->nextcheck = 0;
 353         cd->entries = 0;
 354         atomic_set(&cd->readers, 0);
 355         cd->last_close = 0;
 356         cd->last_warn = -1;
 357         list_add(&cd->others, &cache_list);
 358         spin_unlock(&cache_list_lock);
 359
 360         /* start the cleaning process */
 361         schedule_delayed_work(&cache_cleaner, 0);
 362 }
 363 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 364
 365 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 366 {
 367         cache_purge(cd);
 368         spin_lock(&cache_list_lock);
 369         write_lock(&cd->hash_lock);
 370         if (cd->entries || atomic_read(&cd->inuse)) {
 371                 write_unlock(&cd->hash_lock);
 372                 spin_unlock(&cache_list_lock);
 373                 goto out;
 374         }
 375         if (current_detail == cd)
 376                 current_detail = NULL;
 377         list_del_init(&cd->others);
 378         write_unlock(&cd->hash_lock);
 379         spin_unlock(&cache_list_lock);
 380         if (list_empty(&cache_list)) {
 381                 /* module must be being unloaded so its safe to kill the worker */
 382                 cancel_delayed_work_sync(&cache_cleaner);
 383         }
 384         return;
 385 out:
 386         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 387 }
 388 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 389
 390 /* clean cache tries to find something to clean
 391  * and cleans it.
 392  * It returns 1 if it cleaned something,
 393  *            0 if it didn't find anything this time
 394  *           -1 if it fell off the end of the list.
 395  */
 396 static int cache_clean(void)
 397 {
 398         int rv = 0;
 399         struct list_head *next;
 400
 401         spin_lock(&cache_list_lock);
 402
 403         /* find a suitable table if we don't already have one */
 404         while (current_detail == NULL ||
 405             current_index >= current_detail->hash_size) {
 406                 if (current_detail)
 407                         next = current_detail->others.next;
 408                 else
 409                         next = cache_list.next;
 410                 if (next == &cache_list) {
 411                         current_detail = NULL;
 412                         spin_unlock(&cache_list_lock);
 413                         return -1;
 414                 }
 415                 current_detail = list_entry(next, struct cache_detail, others);
 416                 if (current_detail->nextcheck > seconds_since_boot())
 417                         current_index = current_detail->hash_size;
 418                 else {
 419                         current_index = 0;
 420                         current_detail->nextcheck = seconds_since_boot()+30*60;
 421                 }
 422         }
 423
 424         /* find a non-empty bucket in the table */
 425         while (current_detail &&
 426                current_index < current_detail->hash_size &&
 427                current_detail->hash_table[current_index] == NULL)
 428                 current_index++;
 429
 430         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 431
 432         if (current_detail && current_index < current_detail->hash_size) {
 433                 struct cache_head *ch, **cp;
 434                 struct cache_detail *d;
 435
 436                 write_lock(&current_detail->hash_lock);
 437
 438                 /* Ok, now to clean this strand */
 439
 440                 cp = & current_detail->hash_table[current_index];
 441                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
 442                         if (current_detail->nextcheck > ch->expiry_time)
 443                                 current_detail->nextcheck = ch->expiry_time+1;
 444                         if (!cache_is_expired(current_detail, ch))
 445                                 continue;
 446
 447                         *cp = ch->next;
 448                         ch->next = NULL;
 449                         current_detail->entries--;
 450                         rv = 1;
 451                         break;
 452                 }
 453
 454                 write_unlock(&current_detail->hash_lock);
 455                 d = current_detail;
 456                 if (!ch)
 457                         current_index ++;
 458                 spin_unlock(&cache_list_lock);
 459                 if (ch) {
 460                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 461                                 cache_dequeue(current_detail, ch);
 462                         cache_revisit_request(ch);
 463                         cache_put(ch, d);
 464                 }
 465         } else
 466                 spin_unlock(&cache_list_lock);
 467
 468         return rv;
 469 }
 470
 471 /*
 472  * We want to regularly clean the cache, so we need to schedule some work ...
 473  */
 474 static void do_cache_clean(struct work_struct *work)
 475 {
 476         int delay = 5;
 477         if (cache_clean() == -1)
 478                 delay = round_jiffies_relative(30*HZ);
 479
 480         if (list_empty(&cache_list))
 481                 delay = 0;
 482
 483         if (delay)
 484                 schedule_delayed_work(&cache_cleaner, delay);
 485 }
 486
 487
 488 /*
 489  * Clean all caches promptly.  This just calls cache_clean
 490  * repeatedly until we are sure that every cache has had a chance to
 491  * be fully cleaned
 492  */
 493 void cache_flush(void)
 494 {
 495         while (cache_clean() != -1)
 496                 cond_resched();
 497         while (cache_clean() != -1)
 498                 cond_resched();
 499 }
 500 EXPORT_SYMBOL_GPL(cache_flush);
 501
 502 void cache_purge(struct cache_detail *detail)
 503 {
 504         detail->flush_time = LONG_MAX;
 505         detail->nextcheck = seconds_since_boot();
 506         cache_flush();
 507         detail->flush_time = 1;
 508 }
 509 EXPORT_SYMBOL_GPL(cache_purge);
 510
 511
 512 /*
 513  * Deferral and Revisiting of Requests.
 514  *
 515  * If a cache lookup finds a pending entry, we
 516  * need to defer the request and revisit it later.
 517  * All deferred requests are stored in a hash table,
 518  * indexed by "struct cache_head *".
 519  * As it may be wasteful to store a whole request
 520  * structure, we allow the request to provide a
 521  * deferred form, which must contain a
 522  * 'struct cache_deferred_req'
 523  * This cache_deferred_req contains a method to allow
 524  * it to be revisited when cache info is available
 525  */
 526
 527 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 528 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 529
 530 #define DFR_MAX 300     /* ??? */
 531
 532 static DEFINE_SPINLOCK(cache_defer_lock);
 533 static LIST_HEAD(cache_defer_list);
 534 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 535 static int cache_defer_cnt;
 536
 537 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 538 {
 539         hlist_del_init(&dreq->hash);
 540         if (!list_empty(&dreq->recent)) {
 541                 list_del_init(&dreq->recent);
 542                 cache_defer_cnt--;
 543         }
 544 }
 545
 546 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 547 {
 548         int hash = DFR_HASH(item);
 549
 550         INIT_LIST_HEAD(&dreq->recent);
 551         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 552 }
 553
 554 static void setup_deferral(struct cache_deferred_req *dreq,
 555                            struct cache_head *item,
 556                            int count_me)
 557 {
 558
 559         dreq->item = item;
 560
 561         spin_lock(&cache_defer_lock);
 562
 563         __hash_deferred_req(dreq, item);
 564
 565         if (count_me) {
 566                 cache_defer_cnt++;
 567                 list_add(&dreq->recent, &cache_defer_list);
 568         }
 569
 570         spin_unlock(&cache_defer_lock);
 571
 572 }
 573
 574 struct thread_deferred_req {
 575         struct cache_deferred_req handle;
 576         struct completion completion;
 577 };
 578
 579 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 580 {
 581         struct thread_deferred_req *dr =
 582                 container_of(dreq, struct thread_deferred_req, handle);
 583         complete(&dr->completion);
 584 }
 585
 586 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 587 {
 588         struct thread_deferred_req sleeper;
 589         struct cache_deferred_req *dreq = &sleeper.handle;
 590
 591         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 592         dreq->revisit = cache_restart_thread;
 593
 594         setup_deferral(dreq, item, 0);
 595
 596         if (!test_bit(CACHE_PENDING, &item->flags) ||
 597             wait_for_completion_interruptible_timeout(
 598                     &sleeper.completion, req->thread_wait) <= 0) {
 599                 /* The completion wasn't completed, so we need
 600                  * to clean up
 601                  */
 602                 spin_lock(&cache_defer_lock);
 603                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 604                         __unhash_deferred_req(&sleeper.handle);
 605                         spin_unlock(&cache_defer_lock);
 606                 } else {
 607                         /* cache_revisit_request already removed
 608                          * this from the hash table, but hasn't
 609                          * called ->revisit yet.  It will very soon
 610                          * and we need to wait for it.
 611                          */
 612                         spin_unlock(&cache_defer_lock);
 613                         wait_for_completion(&sleeper.completion);
 614                 }
 615         }
 616 }
 617
 618 static void cache_limit_defers(void)
 619 {
 620         /* Make sure we haven't exceed the limit of allowed deferred
 621          * requests.
 622          */
 623         struct cache_deferred_req *discard = NULL;
 624
 625         if (cache_defer_cnt <= DFR_MAX)
 626                 return;
 627
 628         spin_lock(&cache_defer_lock);
 629
 630         /* Consider removing either the first or the last */
 631         if (cache_defer_cnt > DFR_MAX) {
 632                 if (net_random() & 1)
 633                         discard = list_entry(cache_defer_list.next,
 634                                              struct cache_deferred_req, recent);
 635                 else
 636                         discard = list_entry(cache_defer_list.prev,
 637                                              struct cache_deferred_req, recent);
 638                 __unhash_deferred_req(discard);
 639         }
 640         spin_unlock(&cache_defer_lock);
 641         if (discard)
 642                 discard->revisit(discard, 1);
 643 }
 644
 645 /* Return true if and only if a deferred request is queued. */
 646 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 647 {
 648         struct cache_deferred_req *dreq;
 649
 650         if (req->thread_wait) {
 651                 cache_wait_req(req, item);
 652                 if (!test_bit(CACHE_PENDING, &item->flags))
 653                         return false;
 654         }
 655         dreq = req->defer(req);
 656         if (dreq == NULL)
 657                 return false;
 658         setup_deferral(dreq, item, 1);
 659         if (!test_bit(CACHE_PENDING, &item->flags))
 660                 /* Bit could have been cleared before we managed to
 661                  * set up the deferral, so need to revisit just in case
 662                  */
 663                 cache_revisit_request(item);
 664
 665         cache_limit_defers();
 666         return true;
 667 }
 668
 669 static void cache_revisit_request(struct cache_head *item)
 670 {
 671         struct cache_deferred_req *dreq;
 672         struct list_head pending;
 673         struct hlist_node *tmp;
 674         int hash = DFR_HASH(item);
 675
 676         INIT_LIST_HEAD(&pending);
 677         spin_lock(&cache_defer_lock);
 678
 679         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 680                 if (dreq->item == item) {
 681                         __unhash_deferred_req(dreq);
 682                         list_add(&dreq->recent, &pending);
 683                 }
 684
 685         spin_unlock(&cache_defer_lock);
 686
 687         while (!list_empty(&pending)) {
 688                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 689                 list_del_init(&dreq->recent);
 690                 dreq->revisit(dreq, 0);
 691         }
 692 }
 693
 694 void cache_clean_deferred(void *owner)
 695 {
 696         struct cache_deferred_req *dreq, *tmp;
 697         struct list_head pending;
 698
 699
 700         INIT_LIST_HEAD(&pending);
 701         spin_lock(&cache_defer_lock);
 702
 703         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 704                 if (dreq->owner == owner) {
 705                         __unhash_deferred_req(dreq);
 706                         list_add(&dreq->recent, &pending);
 707                 }
 708         }
 709         spin_unlock(&cache_defer_lock);
 710
 711         while (!list_empty(&pending)) {
 712                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 713                 list_del_init(&dreq->recent);
 714                 dreq->revisit(dreq, 1);
 715         }
 716 }
 717
 718 /*
 719  * communicate with user-space
 720  *
 721  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 722  * On read, you get a full request, or block.
 723  * On write, an update request is processed.
 724  * Poll works if anything to read, and always allows write.
 725  *
 726  * Implemented by linked list of requests.  Each open file has
 727  * a ->private that also exists in this list.  New requests are added
 728  * to the end and may wakeup and preceding readers.
 729  * New readers are added to the head.  If, on read, an item is found with
 730  * CACHE_UPCALLING clear, we free it from the list.
 731  *
 732  */
 733
 734 static DEFINE_SPINLOCK(queue_lock);
 735 static DEFINE_MUTEX(queue_io_mutex);
 736
 737 struct cache_queue {
 738         struct list_head        list;
 739         int                     reader; /* if 0, then request */
 740 };
 741 struct cache_request {
 742         struct cache_queue      q;
 743         struct cache_head       *item;
 744         char                    * buf;
 745         int                     len;
 746         int                     readers;
 747 };
 748 struct cache_reader {
 749         struct cache_queue      q;
 750         int                     offset; /* if non-0, we have a refcnt on next request */
 751 };
 752
 753 static int cache_request(struct cache_detail *detail,
 754                                struct cache_request *crq)
 755 {
 756         char *bp = crq->buf;
 757         int len = PAGE_SIZE;
 758
 759         detail->cache_request(detail, crq->item, &bp, &len);
 760         if (len < 0)
 761                 return -EAGAIN;
 762         return PAGE_SIZE - len;
 763 }
 764
 765 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 766                           loff_t *ppos, struct cache_detail *cd)
 767 {
 768         struct cache_reader *rp = filp->private_data;
 769         struct cache_request *rq;
 770         struct inode *inode = file_inode(filp);
 771         int err;
 772
 773         if (count == 0)
 774                 return 0;
 775
 776         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 777                               * readers on this file */
 778  again:
 779         spin_lock(&queue_lock);
 780         /* need to find next request */
 781         while (rp->q.list.next != &cd->queue &&
 782                list_entry(rp->q.list.next, struct cache_queue, list)
 783                ->reader) {
 784                 struct list_head *next = rp->q.list.next;
 785                 list_move(&rp->q.list, next);
 786         }
 787         if (rp->q.list.next == &cd->queue) {
 788                 spin_unlock(&queue_lock);
 789                 mutex_unlock(&inode->i_mutex);
 790                 WARN_ON_ONCE(rp->offset);
 791                 return 0;
 792         }
 793         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 794         WARN_ON_ONCE(rq->q.reader);
 795         if (rp->offset == 0)
 796                 rq->readers++;
 797         spin_unlock(&queue_lock);
 798
 799         if (rq->len == 0) {
 800                 err = cache_request(cd, rq);
 801                 if (err < 0)
 802                         goto out;
 803                 rq->len = err;
 804         }
 805
 806         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 807                 err = -EAGAIN;
 808                 spin_lock(&queue_lock);
 809                 list_move(&rp->q.list, &rq->q.list);
 810                 spin_unlock(&queue_lock);
 811         } else {
 812                 if (rp->offset + count > rq->len)
 813                         count = rq->len - rp->offset;
 814                 err = -EFAULT;
 815                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 816                         goto out;
 817                 rp->offset += count;
 818                 if (rp->offset >= rq->len) {
 819                         rp->offset = 0;
 820                         spin_lock(&queue_lock);
 821                         list_move(&rp->q.list, &rq->q.list);
 822                         spin_unlock(&queue_lock);
 823                 }
 824                 err = 0;
 825         }
 826  out:
 827         if (rp->offset == 0) {
 828                 /* need to release rq */
 829                 spin_lock(&queue_lock);
 830                 rq->readers--;
 831                 if (rq->readers == 0 &&
 832                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 833                         list_del(&rq->q.list);
 834                         spin_unlock(&queue_lock);
 835                         cache_put(rq->item, cd);
 836                         kfree(rq->buf);
 837                         kfree(rq);
 838                 } else
 839                         spin_unlock(&queue_lock);
 840         }
 841         if (err == -EAGAIN)
 842                 goto again;
 843         mutex_unlock(&inode->i_mutex);
 844         return err ? err :  count;
 845 }
 846
 847 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 848                                  size_t count, struct cache_detail *cd)
 849 {
 850         ssize_t ret;
 851
 852         if (count == 0)
 853                 return -EINVAL;
 854         if (copy_from_user(kaddr, buf, count))
 855                 return -EFAULT;
 856         kaddr[count] = '\0';
 857         ret = cd->cache_parse(cd, kaddr, count);
 858         if (!ret)
 859                 ret = count;
 860         return ret;
 861 }
 862
 863 static ssize_t cache_slow_downcall(const char __user *buf,
 864                                    size_t count, struct cache_detail *cd)
 865 {
 866         static char write_buf[8192]; /* protected by queue_io_mutex */
 867         ssize_t ret = -EINVAL;
 868
 869         if (count >= sizeof(write_buf))
 870                 goto out;
 871         mutex_lock(&queue_io_mutex);
 872         ret = cache_do_downcall(write_buf, buf, count, cd);
 873         mutex_unlock(&queue_io_mutex);
 874 out:
 875         return ret;
 876 }
 877
 878 static ssize_t cache_downcall(struct address_space *mapping,
 879                               const char __user *buf,
 880                               size_t count, struct cache_detail *cd)
 881 {
 882         struct page *page;
 883         char *kaddr;
 884         ssize_t ret = -ENOMEM;
 885
 886         if (count >= PAGE_CACHE_SIZE)
 887                 goto out_slow;
 888
 889         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 890         if (!page)
 891                 goto out_slow;
 892
 893         kaddr = kmap(page);
 894         ret = cache_do_downcall(kaddr, buf, count, cd);
 895         kunmap(page);
 896         unlock_page(page);
 897         page_cache_release(page);
 898         return ret;
 899 out_slow:
 900         return cache_slow_downcall(buf, count, cd);
 901 }
 902
 903 static ssize_t cache_write(struct file *filp, const char __user *buf,
 904                            size_t count, loff_t *ppos,
 905                            struct cache_detail *cd)
 906 {
 907         struct address_space *mapping = filp->f_mapping;
 908         struct inode *inode = file_inode(filp);
 909         ssize_t ret = -EINVAL;
 910
 911         if (!cd->cache_parse)
 912                 goto out;
 913
 914         mutex_lock(&inode->i_mutex);
 915         ret = cache_downcall(mapping, buf, count, cd);
 916         mutex_unlock(&inode->i_mutex);
 917 out:
 918         return ret;
 919 }
 920
 921 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 922
 923 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 924                                struct cache_detail *cd)
 925 {
 926         unsigned int mask;
 927         struct cache_reader *rp = filp->private_data;
 928         struct cache_queue *cq;
 929
 930         poll_wait(filp, &queue_wait, wait);
 931
 932         /* alway allow write */
 933         mask = POLL_OUT | POLLWRNORM;
 934
 935         if (!rp)
 936                 return mask;
 937
 938         spin_lock(&queue_lock);
 939
 940         for (cq= &rp->q; &cq->list != &cd->queue;
 941              cq = list_entry(cq->list.next, struct cache_queue, list))
 942                 if (!cq->reader) {
 943                         mask |= POLLIN | POLLRDNORM;
 944                         break;
 945                 }
 946         spin_unlock(&queue_lock);
 947         return mask;
 948 }
 949
 950 static int cache_ioctl(struct inode *ino, struct file *filp,
 951                        unsigned int cmd, unsigned long arg,
 952                        struct cache_detail *cd)
 953 {
 954         int len = 0;
 955         struct cache_reader *rp = filp->private_data;
 956         struct cache_queue *cq;
 957
 958         if (cmd != FIONREAD || !rp)
 959                 return -EINVAL;
 960
 961         spin_lock(&queue_lock);
 962
 963         /* only find the length remaining in current request,
 964          * or the length of the next request
 965          */
 966         for (cq= &rp->q; &cq->list != &cd->queue;
 967              cq = list_entry(cq->list.next, struct cache_queue, list))
 968                 if (!cq->reader) {
 969                         struct cache_request *cr =
 970                                 container_of(cq, struct cache_request, q);
 971                         len = cr->len - rp->offset;
 972                         break;
 973                 }
 974         spin_unlock(&queue_lock);
 975
 976         return put_user(len, (int __user *)arg);
 977 }
 978
 979 static int cache_open(struct inode *inode, struct file *filp,
 980                       struct cache_detail *cd)
 981 {
 982         struct cache_reader *rp = NULL;
 983
 984         if (!cd || !try_module_get(cd->owner))
 985                 return -EACCES;
 986         nonseekable_open(inode, filp);
 987         if (filp->f_mode & FMODE_READ) {
 988                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 989                 if (!rp) {
 990                         module_put(cd->owner);
 991                         return -ENOMEM;
 992                 }
 993                 rp->offset = 0;
 994                 rp->q.reader = 1;
 995                 atomic_inc(&cd->readers);
 996                 spin_lock(&queue_lock);
 997                 list_add(&rp->q.list, &cd->queue);
 998                 spin_unlock(&queue_lock);
 999         }
1000         filp->private_data = rp;
1001         return 0;
1002 }
1003
1004 static int cache_release(struct inode *inode, struct file *filp,
1005                          struct cache_detail *cd)
1006 {
1007         struct cache_reader *rp = filp->private_data;
1008
1009         if (rp) {
1010                 spin_lock(&queue_lock);
1011                 if (rp->offset) {
1012                         struct cache_queue *cq;
1013                         for (cq= &rp->q; &cq->list != &cd->queue;
1014                              cq = list_entry(cq->list.next, struct cache_queue, list))
1015                                 if (!cq->reader) {
1016                                         container_of(cq, struct cache_request, q)
1017                                                 ->readers--;
1018                                         break;
1019                                 }
1020                         rp->offset = 0;
1021                 }
1022                 list_del(&rp->q.list);
1023                 spin_unlock(&queue_lock);
1024
1025                 filp->private_data = NULL;
1026                 kfree(rp);
1027
1028                 cd->last_close = seconds_since_boot();
1029                 atomic_dec(&cd->readers);
1030         }
1031         module_put(cd->owner);
1032         return 0;
1033 }
1034
1035
1036
1037 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1038 {
1039         struct cache_queue *cq;
1040         spin_lock(&queue_lock);
1041         list_for_each_entry(cq, &detail->queue, list)
1042                 if (!cq->reader) {
1043                         struct cache_request *cr = container_of(cq, struct cache_request, q);
1044                         if (cr->item != ch)
1045                                 continue;
1046                         if (cr->readers != 0)
1047                                 continue;
1048                         list_del(&cr->q.list);
1049                         spin_unlock(&queue_lock);
1050                         cache_put(cr->item, detail);
1051                         kfree(cr->buf);
1052                         kfree(cr);
1053                         return;
1054                 }
1055         spin_unlock(&queue_lock);
1056 }
1057
1058 /*
1059  * Support routines for text-based upcalls.
1060  * Fields are separated by spaces.
1061  * Fields are either mangled to quote space tab newline slosh with slosh
1062  * or a hexified with a leading \x
1063  * Record is terminated with newline.
1064  *
1065  */
1066
1067 void qword_add(char **bpp, int *lp, char *str)
1068 {
1069         char *bp = *bpp;
1070         int len = *lp;
1071         char c;
1072
1073         if (len < 0) return;
1074
1075         while ((c=*str++) && len)
1076                 switch(c) {
1077                 case ' ':
1078                 case '\t':
1079                 case '\n':
1080                 case '\\':
1081                         if (len >= 4) {
1082                                 *bp++ = '\\';
1083                                 *bp++ = '0' + ((c & 0300)>>6);
1084                                 *bp++ = '0' + ((c & 0070)>>3);
1085                                 *bp++ = '0' + ((c & 0007)>>0);
1086                         }
1087                         len -= 4;
1088                         break;
1089                 default:
1090                         *bp++ = c;
1091                         len--;
1092                 }
1093         if (c || len <1) len = -1;
1094         else {
1095                 *bp++ = ' ';
1096                 len--;
1097         }
1098         *bpp = bp;
1099         *lp = len;
1100 }
1101 EXPORT_SYMBOL_GPL(qword_add);
1102
1103 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1104 {
1105         char *bp = *bpp;
1106         int len = *lp;
1107
1108         if (len < 0) return;
1109
1110         if (len > 2) {
1111                 *bp++ = '\\';
1112                 *bp++ = 'x';
1113                 len -= 2;
1114                 while (blen && len >= 2) {
1115                         unsigned char c = *buf++;
1116                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1117                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1118                         len -= 2;
1119                         blen--;
1120                 }
1121         }
1122         if (blen || len<1) len = -1;
1123         else {
1124                 *bp++ = ' ';
1125                 len--;
1126         }
1127         *bpp = bp;
1128         *lp = len;
1129 }
1130 EXPORT_SYMBOL_GPL(qword_addhex);
1131
1132 static void warn_no_listener(struct cache_detail *detail)
1133 {
1134         if (detail->last_warn != detail->last_close) {
1135                 detail->last_warn = detail->last_close;
1136                 if (detail->warn_no_listener)
1137                         detail->warn_no_listener(detail, detail->last_close != 0);
1138         }
1139 }
1140
1141 static bool cache_listeners_exist(struct cache_detail *detail)
1142 {
1143         if (atomic_read(&detail->readers))
1144                 return true;
1145         if (detail->last_close == 0)
1146                 /* This cache was never opened */
1147                 return false;
1148         if (detail->last_close < seconds_since_boot() - 30)
1149                 /*
1150                  * We allow for the possibility that someone might
1151                  * restart a userspace daemon without restarting the
1152                  * server; but after 30 seconds, we give up.
1153                  */
1154                  return false;
1155         return true;
1156 }
1157
1158 /*
1159  * register an upcall request to user-space and queue it up for read() by the
1160  * upcall daemon.
1161  *
1162  * Each request is at most one page long.
1163  */
1164 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1165 {
1166
1167         char *buf;
1168         struct cache_request *crq;
1169
1170         if (!detail->cache_request)
1171                 return -EINVAL;
1172
1173         if (!cache_listeners_exist(detail)) {
1174                 warn_no_listener(detail);
1175                 return -EINVAL;
1176         }
1177
1178         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1179         if (!buf)
1180                 return -EAGAIN;
1181
1182         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1183         if (!crq) {
1184                 kfree(buf);
1185                 return -EAGAIN;
1186         }
1187
1188         crq->q.reader = 0;
1189         crq->item = cache_get(h);
1190         crq->buf = buf;
1191         crq->len = 0;
1192         crq->readers = 0;
1193         spin_lock(&queue_lock);
1194         list_add_tail(&crq->q.list, &detail->queue);
1195         spin_unlock(&queue_lock);
1196         wake_up(&queue_wait);
1197         return 0;
1198 }
1199 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1200
1201 /*
1202  * parse a message from user-space and pass it
1203  * to an appropriate cache
1204  * Messages are, like requests, separated into fields by
1205  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1206  *
1207  * Message is
1208  *   reply cachename expiry key ... content....
1209  *
1210  * key and content are both parsed by cache
1211  */
1212
1213 int qword_get(char **bpp, char *dest, int bufsize)
1214 {
1215         /* return bytes copied, or -1 on error */
1216         char *bp = *bpp;
1217         int len = 0;
1218
1219         while (*bp == ' ') bp++;
1220
1221         if (bp[0] == '\\' && bp[1] == 'x') {
1222                 /* HEX STRING */
1223                 bp += 2;
1224                 while (len < bufsize) {
1225                         int h, l;
1226
1227                         h = hex_to_bin(bp[0]);
1228                         if (h < 0)
1229                                 break;
1230
1231                         l = hex_to_bin(bp[1]);
1232                         if (l < 0)
1233                                 break;
1234
1235                         *dest++ = (h << 4) | l;
1236                         bp += 2;
1237                         len++;
1238                 }
1239         } else {
1240                 /* text with \nnn octal quoting */
1241                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1242                         if (*bp == '\\' &&
1243                             isodigit(bp[1]) && (bp[1] <= '3') &&
1244                             isodigit(bp[2]) &&
1245                             isodigit(bp[3])) {
1246                                 int byte = (*++bp -'0');
1247                                 bp++;
1248                                 byte = (byte << 3) | (*bp++ - '0');
1249                                 byte = (byte << 3) | (*bp++ - '0');
1250                                 *dest++ = byte;
1251                                 len++;
1252                         } else {
1253                                 *dest++ = *bp++;
1254                                 len++;
1255                         }
1256                 }
1257         }
1258
1259         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1260                 return -1;
1261         while (*bp == ' ') bp++;
1262         *bpp = bp;
1263         *dest = '\0';
1264         return len;
1265 }
1266 EXPORT_SYMBOL_GPL(qword_get);
1267
1268
1269 /*
1270  * support /proc/sunrpc/cache/$CACHENAME/content
1271  * as a seqfile.
1272  * We call ->cache_show passing NULL for the item to
1273  * get a header, then pass each real item in the cache
1274  */
1275
1276 struct handle {
1277         struct cache_detail *cd;
1278 };
1279
1280 static void *c_start(struct seq_file *m, loff_t *pos)
1281         __acquires(cd->hash_lock)
1282 {
1283         loff_t n = *pos;
1284         unsigned int hash, entry;
1285         struct cache_head *ch;
1286         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1287
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         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
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                 cd->hash_table[hash]==NULL);
1304         if (hash >= cd->hash_size)
1305                 return NULL;
1306         *pos = n+1;
1307         return cd->hash_table[hash];
1308 }
1309
1310 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1311 {
1312         struct cache_head *ch = p;
1313         int hash = (*pos >> 32);
1314         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1315
1316         if (p == SEQ_START_TOKEN)
1317                 hash = 0;
1318         else if (ch->next == NULL) {
1319                 hash++;
1320                 *pos += 1LL<<32;
1321         } else {
1322                 ++*pos;
1323                 return ch->next;
1324         }
1325         *pos &= ~((1LL<<32) - 1);
1326         while (hash < cd->hash_size &&
1327                cd->hash_table[hash] == NULL) {
1328                 hash++;
1329                 *pos += 1LL<<32;
1330         }
1331         if (hash >= cd->hash_size)
1332                 return NULL;
1333         ++*pos;
1334         return cd->hash_table[hash];
1335 }
1336
1337 static void c_stop(struct seq_file *m, void *p)
1338         __releases(cd->hash_lock)
1339 {
1340         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1341         read_unlock(&cd->hash_lock);
1342 }
1343
1344 static int c_show(struct seq_file *m, void *p)
1345 {
1346         struct cache_head *cp = p;
1347         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1348
1349         if (p == SEQ_START_TOKEN)
1350                 return cd->cache_show(m, cd, NULL);
1351
1352         ifdebug(CACHE)
1353                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1354                            convert_to_wallclock(cp->expiry_time),
1355                            atomic_read(&cp->ref.refcount), cp->flags);
1356         cache_get(cp);
1357         if (cache_check(cd, cp, NULL))
1358                 /* cache_check does a cache_put on failure */
1359                 seq_printf(m, "# ");
1360         else {
1361                 if (cache_is_expired(cd, cp))
1362                         seq_printf(m, "# ");
1363                 cache_put(cp, cd);
1364         }
1365
1366         return cd->cache_show(m, cd, cp);
1367 }
1368
1369 static const struct seq_operations cache_content_op = {
1370         .start  = c_start,
1371         .next   = c_next,
1372         .stop   = c_stop,
1373         .show   = c_show,
1374 };
1375
1376 static int content_open(struct inode *inode, struct file *file,
1377                         struct cache_detail *cd)
1378 {
1379         struct handle *han;
1380
1381         if (!cd || !try_module_get(cd->owner))
1382                 return -EACCES;
1383         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1384         if (han == NULL) {
1385                 module_put(cd->owner);
1386                 return -ENOMEM;
1387         }
1388
1389         han->cd = cd;
1390         return 0;
1391 }
1392
1393 static int content_release(struct inode *inode, struct file *file,
1394                 struct cache_detail *cd)
1395 {
1396         int ret = seq_release_private(inode, file);
1397         module_put(cd->owner);
1398         return ret;
1399 }
1400
1401 static int open_flush(struct inode *inode, struct file *file,
1402                         struct cache_detail *cd)
1403 {
1404         if (!cd || !try_module_get(cd->owner))
1405                 return -EACCES;
1406         return nonseekable_open(inode, file);
1407 }
1408
1409 static int release_flush(struct inode *inode, struct file *file,
1410                         struct cache_detail *cd)
1411 {
1412         module_put(cd->owner);
1413         return 0;
1414 }
1415
1416 static ssize_t read_flush(struct file *file, char __user *buf,
1417                           size_t count, loff_t *ppos,
1418                           struct cache_detail *cd)
1419 {
1420         char tbuf[22];
1421         unsigned long p = *ppos;
1422         size_t len;
1423
1424         snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1425         len = strlen(tbuf);
1426         if (p >= len)
1427                 return 0;
1428         len -= p;
1429         if (len > count)
1430                 len = count;
1431         if (copy_to_user(buf, (void*)(tbuf+p), len))
1432                 return -EFAULT;
1433         *ppos += len;
1434         return len;
1435 }
1436
1437 static ssize_t write_flush(struct file *file, const char __user *buf,
1438                            size_t count, loff_t *ppos,
1439                            struct cache_detail *cd)
1440 {
1441         char tbuf[20];
1442         char *bp, *ep;
1443
1444         if (*ppos || count > sizeof(tbuf)-1)
1445                 return -EINVAL;
1446         if (copy_from_user(tbuf, buf, count))
1447                 return -EFAULT;
1448         tbuf[count] = 0;
1449         simple_strtoul(tbuf, &ep, 0);
1450         if (*ep && *ep != '\n')
1451                 return -EINVAL;
1452
1453         bp = tbuf;
1454         cd->flush_time = get_expiry(&bp);
1455         cd->nextcheck = seconds_since_boot();
1456         cache_flush();
1457
1458         *ppos += count;
1459         return count;
1460 }
1461
1462 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1463                                  size_t count, loff_t *ppos)
1464 {
1465         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1466
1467         return cache_read(filp, buf, count, ppos, cd);
1468 }
1469
1470 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1471                                   size_t count, loff_t *ppos)
1472 {
1473         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1474
1475         return cache_write(filp, buf, count, ppos, cd);
1476 }
1477
1478 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1479 {
1480         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1481
1482         return cache_poll(filp, wait, cd);
1483 }
1484
1485 static long cache_ioctl_procfs(struct file *filp,
1486                                unsigned int cmd, unsigned long arg)
1487 {
1488         struct inode *inode = file_inode(filp);
1489         struct cache_detail *cd = PDE_DATA(inode);
1490
1491         return cache_ioctl(inode, filp, cmd, arg, cd);
1492 }
1493
1494 static int cache_open_procfs(struct inode *inode, struct file *filp)
1495 {
1496         struct cache_detail *cd = PDE_DATA(inode);
1497
1498         return cache_open(inode, filp, cd);
1499 }
1500
1501 static int cache_release_procfs(struct inode *inode, struct file *filp)
1502 {
1503         struct cache_detail *cd = PDE_DATA(inode);
1504
1505         return cache_release(inode, filp, cd);
1506 }
1507
1508 static const struct file_operations cache_file_operations_procfs = {
1509         .owner          = THIS_MODULE,
1510         .llseek         = no_llseek,
1511         .read           = cache_read_procfs,
1512         .write          = cache_write_procfs,
1513         .poll           = cache_poll_procfs,
1514         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1515         .open           = cache_open_procfs,
1516         .release        = cache_release_procfs,
1517 };
1518
1519 static int content_open_procfs(struct inode *inode, struct file *filp)
1520 {
1521         struct cache_detail *cd = PDE_DATA(inode);
1522
1523         return content_open(inode, filp, cd);
1524 }
1525
1526 static int content_release_procfs(struct inode *inode, struct file *filp)
1527 {
1528         struct cache_detail *cd = PDE_DATA(inode);
1529
1530         return content_release(inode, filp, cd);
1531 }
1532
1533 static const struct file_operations content_file_operations_procfs = {
1534         .open           = content_open_procfs,
1535         .read           = seq_read,
1536         .llseek         = seq_lseek,
1537         .release        = content_release_procfs,
1538 };
1539
1540 static int open_flush_procfs(struct inode *inode, struct file *filp)
1541 {
1542         struct cache_detail *cd = PDE_DATA(inode);
1543
1544         return open_flush(inode, filp, cd);
1545 }
1546
1547 static int release_flush_procfs(struct inode *inode, struct file *filp)
1548 {
1549         struct cache_detail *cd = PDE_DATA(inode);
1550
1551         return release_flush(inode, filp, cd);
1552 }
1553
1554 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1555                             size_t count, loff_t *ppos)
1556 {
1557         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1558
1559         return read_flush(filp, buf, count, ppos, cd);
1560 }
1561
1562 static ssize_t write_flush_procfs(struct file *filp,
1563                                   const char __user *buf,
1564                                   size_t count, loff_t *ppos)
1565 {
1566         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1567
1568         return write_flush(filp, buf, count, ppos, cd);
1569 }
1570
1571 static const struct file_operations cache_flush_operations_procfs = {
1572         .open           = open_flush_procfs,
1573         .read           = read_flush_procfs,
1574         .write          = write_flush_procfs,
1575         .release        = release_flush_procfs,
1576         .llseek         = no_llseek,
1577 };
1578
1579 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1580 {
1581         struct sunrpc_net *sn;
1582
1583         if (cd->u.procfs.proc_ent == NULL)
1584                 return;
1585         if (cd->u.procfs.flush_ent)
1586                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1587         if (cd->u.procfs.channel_ent)
1588                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1589         if (cd->u.procfs.content_ent)
1590                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1591         cd->u.procfs.proc_ent = NULL;
1592         sn = net_generic(net, sunrpc_net_id);
1593         remove_proc_entry(cd->name, sn->proc_net_rpc);
1594 }
1595
1596 #ifdef CONFIG_PROC_FS
1597 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1598 {
1599         struct proc_dir_entry *p;
1600         struct sunrpc_net *sn;
1601
1602         sn = net_generic(net, sunrpc_net_id);
1603         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1604         if (cd->u.procfs.proc_ent == NULL)
1605                 goto out_nomem;
1606         cd->u.procfs.channel_ent = NULL;
1607         cd->u.procfs.content_ent = NULL;
1608
1609         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1610                              cd->u.procfs.proc_ent,
1611                              &cache_flush_operations_procfs, cd);
1612         cd->u.procfs.flush_ent = p;
1613         if (p == NULL)
1614                 goto out_nomem;
1615
1616         if (cd->cache_request || cd->cache_parse) {
1617                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1618                                      cd->u.procfs.proc_ent,
1619                                      &cache_file_operations_procfs, cd);
1620                 cd->u.procfs.channel_ent = p;
1621                 if (p == NULL)
1622                         goto out_nomem;
1623         }
1624         if (cd->cache_show) {
1625                 p = proc_create_data("content", S_IFREG|S_IRUSR,
1626                                 cd->u.procfs.proc_ent,
1627                                 &content_file_operations_procfs, cd);
1628                 cd->u.procfs.content_ent = p;
1629                 if (p == NULL)
1630                         goto out_nomem;
1631         }
1632         return 0;
1633 out_nomem:
1634         remove_cache_proc_entries(cd, net);
1635         return -ENOMEM;
1636 }
1637 #else /* CONFIG_PROC_FS */
1638 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1639 {
1640         return 0;
1641 }
1642 #endif
1643
1644 void __init cache_initialize(void)
1645 {
1646         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1647 }
1648
1649 int cache_register_net(struct cache_detail *cd, struct net *net)
1650 {
1651         int ret;
1652
1653         sunrpc_init_cache_detail(cd);
1654         ret = create_cache_proc_entries(cd, net);
1655         if (ret)
1656                 sunrpc_destroy_cache_detail(cd);
1657         return ret;
1658 }
1659 EXPORT_SYMBOL_GPL(cache_register_net);
1660
1661 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1662 {
1663         remove_cache_proc_entries(cd, net);
1664         sunrpc_destroy_cache_detail(cd);
1665 }
1666 EXPORT_SYMBOL_GPL(cache_unregister_net);
1667
1668 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1669 {
1670         struct cache_detail *cd;
1671
1672         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1673         if (cd == NULL)
1674                 return ERR_PTR(-ENOMEM);
1675
1676         cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1677                                  GFP_KERNEL);
1678         if (cd->hash_table == NULL) {
1679                 kfree(cd);
1680                 return ERR_PTR(-ENOMEM);
1681         }
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 qstr q;
1816         struct dentry *dir;
1817         int ret = 0;
1818
1819         q.name = name;
1820         q.len = strlen(name);
1821         q.hash = full_name_hash(q.name, q.len);
1822         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1823         if (!IS_ERR(dir))
1824                 cd->u.pipefs.dir = dir;
1825         else
1826                 ret = PTR_ERR(dir);
1827         return ret;
1828 }
1829 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1830
1831 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1832 {
1833         rpc_remove_cache_dir(cd->u.pipefs.dir);
1834         cd->u.pipefs.dir = NULL;
1835 }
1836 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1837