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         set_bit(CACHE_VALID, &head->flags);
 132 }
 133
 134 static void cache_fresh_unlocked(struct cache_head *head,
 135                                  struct cache_detail *detail)
 136 {
 137         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 138                 cache_revisit_request(head);
 139                 cache_dequeue(detail, head);
 140         }
 141 }
 142
 143 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 144                                        struct cache_head *new, struct cache_head *old, int hash)
 145 {
 146         /* The 'old' entry is to be replaced by 'new'.
 147          * If 'old' is not VALID, we update it directly,
 148          * otherwise we need to replace it
 149          */
 150         struct cache_head **head;
 151         struct cache_head *tmp;
 152
 153         if (!test_bit(CACHE_VALID, &old->flags)) {
 154                 write_lock(&detail->hash_lock);
 155                 if (!test_bit(CACHE_VALID, &old->flags)) {
 156                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 157                                 set_bit(CACHE_NEGATIVE, &old->flags);
 158                         else
 159                                 detail->update(old, new);
 160                         cache_fresh_locked(old, new->expiry_time);
 161                         write_unlock(&detail->hash_lock);
 162                         cache_fresh_unlocked(old, detail);
 163                         return old;
 164                 }
 165                 write_unlock(&detail->hash_lock);
 166         }
 167         /* We need to insert a new entry */
 168         tmp = detail->alloc();
 169         if (!tmp) {
 170                 cache_put(old, detail);
 171                 return NULL;
 172         }
 173         cache_init(tmp);
 174         detail->init(tmp, old);
 175         head = &detail->hash_table[hash];
 176
 177         write_lock(&detail->hash_lock);
 178         if (test_bit(CACHE_NEGATIVE, &new->flags))
 179                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 180         else
 181                 detail->update(tmp, new);
 182         tmp->next = *head;
 183         *head = tmp;
 184         detail->entries++;
 185         cache_get(tmp);
 186         cache_fresh_locked(tmp, new->expiry_time);
 187         cache_fresh_locked(old, 0);
 188         write_unlock(&detail->hash_lock);
 189         cache_fresh_unlocked(tmp, detail);
 190         cache_fresh_unlocked(old, detail);
 191         cache_put(old, detail);
 192         return tmp;
 193 }
 194 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 195
 196 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 197 {
 198         if (!cd->cache_upcall)
 199                 return -EINVAL;
 200         return cd->cache_upcall(cd, h);
 201 }
 202
 203 static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
 204 {
 205         if (!test_bit(CACHE_VALID, &h->flags))
 206                 return -EAGAIN;
 207         else {
 208                 /* entry is valid */
 209                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 210                         return -ENOENT;
 211                 else
 212                         return 0;
 213         }
 214 }
 215
 216 /*
 217  * This is the generic cache management routine for all
 218  * the authentication caches.
 219  * It checks the currency of a cache item and will (later)
 220  * initiate an upcall to fill it if needed.
 221  *
 222  *
 223  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 224  * -EAGAIN if upcall is pending and request has been queued
 225  * -ETIMEDOUT if upcall failed or request could not be queue or
 226  *           upcall completed but item is still invalid (implying that
 227  *           the cache item has been replaced with a newer one).
 228  * -ENOENT if cache entry was negative
 229  */
 230 int cache_check(struct cache_detail *detail,
 231                     struct cache_head *h, struct cache_req *rqstp)
 232 {
 233         int rv;
 234         long refresh_age, age;
 235
 236         /* First decide return status as best we can */
 237         rv = cache_is_valid(detail, h);
 238
 239         /* now see if we want to start an upcall */
 240         refresh_age = (h->expiry_time - h->last_refresh);
 241         age = seconds_since_boot() - h->last_refresh;
 242
 243         if (rqstp == NULL) {
 244                 if (rv == -EAGAIN)
 245                         rv = -ENOENT;
 246         } else if (rv == -EAGAIN || age > refresh_age/2) {
 247                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 248                                 refresh_age, age);
 249                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 250                         switch (cache_make_upcall(detail, h)) {
 251                         case -EINVAL:
 252                                 clear_bit(CACHE_PENDING, &h->flags);
 253                                 cache_revisit_request(h);
 254                                 if (rv == -EAGAIN) {
 255                                         set_bit(CACHE_NEGATIVE, &h->flags);
 256                                         cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
 257                                         cache_fresh_unlocked(h, detail);
 258                                         rv = -ENOENT;
 259                                 }
 260                                 break;
 261
 262                         case -EAGAIN:
 263                                 clear_bit(CACHE_PENDING, &h->flags);
 264                                 cache_revisit_request(h);
 265                                 break;
 266                         }
 267                 }
 268         }
 269
 270         if (rv == -EAGAIN) {
 271                 if (!cache_defer_req(rqstp, h)) {
 272                         /*
 273                          * Request was not deferred; handle it as best
 274                          * we can ourselves:
 275                          */
 276                         rv = cache_is_valid(detail, h);
 277                         if (rv == -EAGAIN)
 278                                 rv = -ETIMEDOUT;
 279                 }
 280         }
 281         if (rv)
 282                 cache_put(h, detail);
 283         return rv;
 284 }
 285 EXPORT_SYMBOL_GPL(cache_check);
 286
 287 /*
 288  * caches need to be periodically cleaned.
 289  * For this we maintain a list of cache_detail and
 290  * a current pointer into that list and into the table
 291  * for that entry.
 292  *
 293  * Each time clean_cache is called it finds the next non-empty entry
 294  * in the current table and walks the list in that entry
 295  * looking for entries that can be removed.
 296  *
 297  * An entry gets removed if:
 298  * - The expiry is before current time
 299  * - The last_refresh time is before the flush_time for that cache
 300  *
 301  * later we might drop old entries with non-NEVER expiry if that table
 302  * is getting 'full' for some definition of 'full'
 303  *
 304  * The question of "how often to scan a table" is an interesting one
 305  * and is answered in part by the use of the "nextcheck" field in the
 306  * cache_detail.
 307  * When a scan of a table begins, the nextcheck field is set to a time
 308  * that is well into the future.
 309  * While scanning, if an expiry time is found that is earlier than the
 310  * current nextcheck time, nextcheck is set to that expiry time.
 311  * If the flush_time is ever set to a time earlier than the nextcheck
 312  * time, the nextcheck time is then set to that flush_time.
 313  *
 314  * A table is then only scanned if the current time is at least
 315  * the nextcheck time.
 316  *
 317  */
 318
 319 static LIST_HEAD(cache_list);
 320 static DEFINE_SPINLOCK(cache_list_lock);
 321 static struct cache_detail *current_detail;
 322 static int current_index;
 323
 324 static void do_cache_clean(struct work_struct *work);
 325 static struct delayed_work cache_cleaner;
 326
 327 static void sunrpc_init_cache_detail(struct cache_detail *cd)
 328 {
 329         rwlock_init(&cd->hash_lock);
 330         INIT_LIST_HEAD(&cd->queue);
 331         spin_lock(&cache_list_lock);
 332         cd->nextcheck = 0;
 333         cd->entries = 0;
 334         atomic_set(&cd->readers, 0);
 335         cd->last_close = 0;
 336         cd->last_warn = -1;
 337         list_add(&cd->others, &cache_list);
 338         spin_unlock(&cache_list_lock);
 339
 340         /* start the cleaning process */
 341         schedule_delayed_work(&cache_cleaner, 0);
 342 }
 343
 344 static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 345 {
 346         cache_purge(cd);
 347         spin_lock(&cache_list_lock);
 348         write_lock(&cd->hash_lock);
 349         if (cd->entries || atomic_read(&cd->inuse)) {
 350                 write_unlock(&cd->hash_lock);
 351                 spin_unlock(&cache_list_lock);
 352                 goto out;
 353         }
 354         if (current_detail == cd)
 355                 current_detail = NULL;
 356         list_del_init(&cd->others);
 357         write_unlock(&cd->hash_lock);
 358         spin_unlock(&cache_list_lock);
 359         if (list_empty(&cache_list)) {
 360                 /* module must be being unloaded so its safe to kill the worker */
 361                 cancel_delayed_work_sync(&cache_cleaner);
 362         }
 363         return;
 364 out:
 365         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 366 }
 367
 368 /* clean cache tries to find something to clean
 369  * and cleans it.
 370  * It returns 1 if it cleaned something,
 371  *            0 if it didn't find anything this time
 372  *           -1 if it fell off the end of the list.
 373  */
 374 static int cache_clean(void)
 375 {
 376         int rv = 0;
 377         struct list_head *next;
 378
 379         spin_lock(&cache_list_lock);
 380
 381         /* find a suitable table if we don't already have one */
 382         while (current_detail == NULL ||
 383             current_index >= current_detail->hash_size) {
 384                 if (current_detail)
 385                         next = current_detail->others.next;
 386                 else
 387                         next = cache_list.next;
 388                 if (next == &cache_list) {
 389                         current_detail = NULL;
 390                         spin_unlock(&cache_list_lock);
 391                         return -1;
 392                 }
 393                 current_detail = list_entry(next, struct cache_detail, others);
 394                 if (current_detail->nextcheck > seconds_since_boot())
 395                         current_index = current_detail->hash_size;
 396                 else {
 397                         current_index = 0;
 398                         current_detail->nextcheck = seconds_since_boot()+30*60;
 399                 }
 400         }
 401
 402         /* find a non-empty bucket in the table */
 403         while (current_detail &&
 404                current_index < current_detail->hash_size &&
 405                current_detail->hash_table[current_index] == NULL)
 406                 current_index++;
 407
 408         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 409
 410         if (current_detail && current_index < current_detail->hash_size) {
 411                 struct cache_head *ch, **cp;
 412                 struct cache_detail *d;
 413
 414                 write_lock(&current_detail->hash_lock);
 415
 416                 /* Ok, now to clean this strand */
 417
 418                 cp = & current_detail->hash_table[current_index];
 419                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
 420                         if (current_detail->nextcheck > ch->expiry_time)
 421                                 current_detail->nextcheck = ch->expiry_time+1;
 422                         if (!cache_is_expired(current_detail, ch))
 423                                 continue;
 424
 425                         *cp = ch->next;
 426                         ch->next = NULL;
 427                         current_detail->entries--;
 428                         rv = 1;
 429                         break;
 430                 }
 431
 432                 write_unlock(&current_detail->hash_lock);
 433                 d = current_detail;
 434                 if (!ch)
 435                         current_index ++;
 436                 spin_unlock(&cache_list_lock);
 437                 if (ch) {
 438                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 439                                 cache_dequeue(current_detail, ch);
 440                         cache_revisit_request(ch);
 441                         cache_put(ch, d);
 442                 }
 443         } else
 444                 spin_unlock(&cache_list_lock);
 445
 446         return rv;
 447 }
 448
 449 /*
 450  * We want to regularly clean the cache, so we need to schedule some work ...
 451  */
 452 static void do_cache_clean(struct work_struct *work)
 453 {
 454         int delay = 5;
 455         if (cache_clean() == -1)
 456                 delay = round_jiffies_relative(30*HZ);
 457
 458         if (list_empty(&cache_list))
 459                 delay = 0;
 460
 461         if (delay)
 462                 schedule_delayed_work(&cache_cleaner, delay);
 463 }
 464
 465
 466 /*
 467  * Clean all caches promptly.  This just calls cache_clean
 468  * repeatedly until we are sure that every cache has had a chance to
 469  * be fully cleaned
 470  */
 471 void cache_flush(void)
 472 {
 473         while (cache_clean() != -1)
 474                 cond_resched();
 475         while (cache_clean() != -1)
 476                 cond_resched();
 477 }
 478 EXPORT_SYMBOL_GPL(cache_flush);
 479
 480 void cache_purge(struct cache_detail *detail)
 481 {
 482         detail->flush_time = LONG_MAX;
 483         detail->nextcheck = seconds_since_boot();
 484         cache_flush();
 485         detail->flush_time = 1;
 486 }
 487 EXPORT_SYMBOL_GPL(cache_purge);
 488
 489
 490 /*
 491  * Deferral and Revisiting of Requests.
 492  *
 493  * If a cache lookup finds a pending entry, we
 494  * need to defer the request and revisit it later.
 495  * All deferred requests are stored in a hash table,
 496  * indexed by "struct cache_head *".
 497  * As it may be wasteful to store a whole request
 498  * structure, we allow the request to provide a
 499  * deferred form, which must contain a
 500  * 'struct cache_deferred_req'
 501  * This cache_deferred_req contains a method to allow
 502  * it to be revisited when cache info is available
 503  */
 504
 505 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 506 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 507
 508 #define DFR_MAX 300     /* ??? */
 509
 510 static DEFINE_SPINLOCK(cache_defer_lock);
 511 static LIST_HEAD(cache_defer_list);
 512 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 513 static int cache_defer_cnt;
 514
 515 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 516 {
 517         hlist_del_init(&dreq->hash);
 518         if (!list_empty(&dreq->recent)) {
 519                 list_del_init(&dreq->recent);
 520                 cache_defer_cnt--;
 521         }
 522 }
 523
 524 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 525 {
 526         int hash = DFR_HASH(item);
 527
 528         INIT_LIST_HEAD(&dreq->recent);
 529         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 530 }
 531
 532 static void setup_deferral(struct cache_deferred_req *dreq,
 533                            struct cache_head *item,
 534                            int count_me)
 535 {
 536
 537         dreq->item = item;
 538
 539         spin_lock(&cache_defer_lock);
 540
 541         __hash_deferred_req(dreq, item);
 542
 543         if (count_me) {
 544                 cache_defer_cnt++;
 545                 list_add(&dreq->recent, &cache_defer_list);
 546         }
 547
 548         spin_unlock(&cache_defer_lock);
 549
 550 }
 551
 552 struct thread_deferred_req {
 553         struct cache_deferred_req handle;
 554         struct completion completion;
 555 };
 556
 557 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 558 {
 559         struct thread_deferred_req *dr =
 560                 container_of(dreq, struct thread_deferred_req, handle);
 561         complete(&dr->completion);
 562 }
 563
 564 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 565 {
 566         struct thread_deferred_req sleeper;
 567         struct cache_deferred_req *dreq = &sleeper.handle;
 568
 569         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 570         dreq->revisit = cache_restart_thread;
 571
 572         setup_deferral(dreq, item, 0);
 573
 574         if (!test_bit(CACHE_PENDING, &item->flags) ||
 575             wait_for_completion_interruptible_timeout(
 576                     &sleeper.completion, req->thread_wait) <= 0) {
 577                 /* The completion wasn't completed, so we need
 578                  * to clean up
 579                  */
 580                 spin_lock(&cache_defer_lock);
 581                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 582                         __unhash_deferred_req(&sleeper.handle);
 583                         spin_unlock(&cache_defer_lock);
 584                 } else {
 585                         /* cache_revisit_request already removed
 586                          * this from the hash table, but hasn't
 587                          * called ->revisit yet.  It will very soon
 588                          * and we need to wait for it.
 589                          */
 590                         spin_unlock(&cache_defer_lock);
 591                         wait_for_completion(&sleeper.completion);
 592                 }
 593         }
 594 }
 595
 596 static void cache_limit_defers(void)
 597 {
 598         /* Make sure we haven't exceed the limit of allowed deferred
 599          * requests.
 600          */
 601         struct cache_deferred_req *discard = NULL;
 602
 603         if (cache_defer_cnt <= DFR_MAX)
 604                 return;
 605
 606         spin_lock(&cache_defer_lock);
 607
 608         /* Consider removing either the first or the last */
 609         if (cache_defer_cnt > DFR_MAX) {
 610                 if (net_random() & 1)
 611                         discard = list_entry(cache_defer_list.next,
 612                                              struct cache_deferred_req, recent);
 613                 else
 614                         discard = list_entry(cache_defer_list.prev,
 615                                              struct cache_deferred_req, recent);
 616                 __unhash_deferred_req(discard);
 617         }
 618         spin_unlock(&cache_defer_lock);
 619         if (discard)
 620                 discard->revisit(discard, 1);
 621 }
 622
 623 /* Return true if and only if a deferred request is queued. */
 624 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 625 {
 626         struct cache_deferred_req *dreq;
 627
 628         if (req->thread_wait) {
 629                 cache_wait_req(req, item);
 630                 if (!test_bit(CACHE_PENDING, &item->flags))
 631                         return false;
 632         }
 633         dreq = req->defer(req);
 634         if (dreq == NULL)
 635                 return false;
 636         setup_deferral(dreq, item, 1);
 637         if (!test_bit(CACHE_PENDING, &item->flags))
 638                 /* Bit could have been cleared before we managed to
 639                  * set up the deferral, so need to revisit just in case
 640                  */
 641                 cache_revisit_request(item);
 642
 643         cache_limit_defers();
 644         return true;
 645 }
 646
 647 static void cache_revisit_request(struct cache_head *item)
 648 {
 649         struct cache_deferred_req *dreq;
 650         struct list_head pending;
 651         struct hlist_node *lp, *tmp;
 652         int hash = DFR_HASH(item);
 653
 654         INIT_LIST_HEAD(&pending);
 655         spin_lock(&cache_defer_lock);
 656
 657         hlist_for_each_entry_safe(dreq, lp, tmp, &cache_defer_hash[hash], hash)
 658                 if (dreq->item == item) {
 659                         __unhash_deferred_req(dreq);
 660                         list_add(&dreq->recent, &pending);
 661                 }
 662
 663         spin_unlock(&cache_defer_lock);
 664
 665         while (!list_empty(&pending)) {
 666                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 667                 list_del_init(&dreq->recent);
 668                 dreq->revisit(dreq, 0);
 669         }
 670 }
 671
 672 void cache_clean_deferred(void *owner)
 673 {
 674         struct cache_deferred_req *dreq, *tmp;
 675         struct list_head pending;
 676
 677
 678         INIT_LIST_HEAD(&pending);
 679         spin_lock(&cache_defer_lock);
 680
 681         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 682                 if (dreq->owner == owner) {
 683                         __unhash_deferred_req(dreq);
 684                         list_add(&dreq->recent, &pending);
 685                 }
 686         }
 687         spin_unlock(&cache_defer_lock);
 688
 689         while (!list_empty(&pending)) {
 690                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 691                 list_del_init(&dreq->recent);
 692                 dreq->revisit(dreq, 1);
 693         }
 694 }
 695
 696 /*
 697  * communicate with user-space
 698  *
 699  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 700  * On read, you get a full request, or block.
 701  * On write, an update request is processed.
 702  * Poll works if anything to read, and always allows write.
 703  *
 704  * Implemented by linked list of requests.  Each open file has
 705  * a ->private that also exists in this list.  New requests are added
 706  * to the end and may wakeup and preceding readers.
 707  * New readers are added to the head.  If, on read, an item is found with
 708  * CACHE_UPCALLING clear, we free it from the list.
 709  *
 710  */
 711
 712 static DEFINE_SPINLOCK(queue_lock);
 713 static DEFINE_MUTEX(queue_io_mutex);
 714
 715 struct cache_queue {
 716         struct list_head        list;
 717         int                     reader; /* if 0, then request */
 718 };
 719 struct cache_request {
 720         struct cache_queue      q;
 721         struct cache_head       *item;
 722         char                    * buf;
 723         int                     len;
 724         int                     readers;
 725 };
 726 struct cache_reader {
 727         struct cache_queue      q;
 728         int                     offset; /* if non-0, we have a refcnt on next request */
 729 };
 730
 731 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 732                           loff_t *ppos, struct cache_detail *cd)
 733 {
 734         struct cache_reader *rp = filp->private_data;
 735         struct cache_request *rq;
 736         struct inode *inode = filp->f_path.dentry->d_inode;
 737         int err;
 738
 739         if (count == 0)
 740                 return 0;
 741
 742         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 743                               * readers on this file */
 744  again:
 745         spin_lock(&queue_lock);
 746         /* need to find next request */
 747         while (rp->q.list.next != &cd->queue &&
 748                list_entry(rp->q.list.next, struct cache_queue, list)
 749                ->reader) {
 750                 struct list_head *next = rp->q.list.next;
 751                 list_move(&rp->q.list, next);
 752         }
 753         if (rp->q.list.next == &cd->queue) {
 754                 spin_unlock(&queue_lock);
 755                 mutex_unlock(&inode->i_mutex);
 756                 BUG_ON(rp->offset);
 757                 return 0;
 758         }
 759         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 760         BUG_ON(rq->q.reader);
 761         if (rp->offset == 0)
 762                 rq->readers++;
 763         spin_unlock(&queue_lock);
 764
 765         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 766                 err = -EAGAIN;
 767                 spin_lock(&queue_lock);
 768                 list_move(&rp->q.list, &rq->q.list);
 769                 spin_unlock(&queue_lock);
 770         } else {
 771                 if (rp->offset + count > rq->len)
 772                         count = rq->len - rp->offset;
 773                 err = -EFAULT;
 774                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 775                         goto out;
 776                 rp->offset += count;
 777                 if (rp->offset >= rq->len) {
 778                         rp->offset = 0;
 779                         spin_lock(&queue_lock);
 780                         list_move(&rp->q.list, &rq->q.list);
 781                         spin_unlock(&queue_lock);
 782                 }
 783                 err = 0;
 784         }
 785  out:
 786         if (rp->offset == 0) {
 787                 /* need to release rq */
 788                 spin_lock(&queue_lock);
 789                 rq->readers--;
 790                 if (rq->readers == 0 &&
 791                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 792                         list_del(&rq->q.list);
 793                         spin_unlock(&queue_lock);
 794                         cache_put(rq->item, cd);
 795                         kfree(rq->buf);
 796                         kfree(rq);
 797                 } else
 798                         spin_unlock(&queue_lock);
 799         }
 800         if (err == -EAGAIN)
 801                 goto again;
 802         mutex_unlock(&inode->i_mutex);
 803         return err ? err :  count;
 804 }
 805
 806 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 807                                  size_t count, struct cache_detail *cd)
 808 {
 809         ssize_t ret;
 810
 811         if (copy_from_user(kaddr, buf, count))
 812                 return -EFAULT;
 813         kaddr[count] = '\0';
 814         ret = cd->cache_parse(cd, kaddr, count);
 815         if (!ret)
 816                 ret = count;
 817         return ret;
 818 }
 819
 820 static ssize_t cache_slow_downcall(const char __user *buf,
 821                                    size_t count, struct cache_detail *cd)
 822 {
 823         static char write_buf[8192]; /* protected by queue_io_mutex */
 824         ssize_t ret = -EINVAL;
 825
 826         if (count >= sizeof(write_buf))
 827                 goto out;
 828         mutex_lock(&queue_io_mutex);
 829         ret = cache_do_downcall(write_buf, buf, count, cd);
 830         mutex_unlock(&queue_io_mutex);
 831 out:
 832         return ret;
 833 }
 834
 835 static ssize_t cache_downcall(struct address_space *mapping,
 836                               const char __user *buf,
 837                               size_t count, struct cache_detail *cd)
 838 {
 839         struct page *page;
 840         char *kaddr;
 841         ssize_t ret = -ENOMEM;
 842
 843         if (count >= PAGE_CACHE_SIZE)
 844                 goto out_slow;
 845
 846         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 847         if (!page)
 848                 goto out_slow;
 849
 850         kaddr = kmap(page);
 851         ret = cache_do_downcall(kaddr, buf, count, cd);
 852         kunmap(page);
 853         unlock_page(page);
 854         page_cache_release(page);
 855         return ret;
 856 out_slow:
 857         return cache_slow_downcall(buf, count, cd);
 858 }
 859
 860 static ssize_t cache_write(struct file *filp, const char __user *buf,
 861                            size_t count, loff_t *ppos,
 862                            struct cache_detail *cd)
 863 {
 864         struct address_space *mapping = filp->f_mapping;
 865         struct inode *inode = filp->f_path.dentry->d_inode;
 866         ssize_t ret = -EINVAL;
 867
 868         if (!cd->cache_parse)
 869                 goto out;
 870
 871         mutex_lock(&inode->i_mutex);
 872         ret = cache_downcall(mapping, buf, count, cd);
 873         mutex_unlock(&inode->i_mutex);
 874 out:
 875         return ret;
 876 }
 877
 878 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 879
 880 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 881                                struct cache_detail *cd)
 882 {
 883         unsigned int mask;
 884         struct cache_reader *rp = filp->private_data;
 885         struct cache_queue *cq;
 886
 887         poll_wait(filp, &queue_wait, wait);
 888
 889         /* alway allow write */
 890         mask = POLL_OUT | POLLWRNORM;
 891
 892         if (!rp)
 893                 return mask;
 894
 895         spin_lock(&queue_lock);
 896
 897         for (cq= &rp->q; &cq->list != &cd->queue;
 898              cq = list_entry(cq->list.next, struct cache_queue, list))
 899                 if (!cq->reader) {
 900                         mask |= POLLIN | POLLRDNORM;
 901                         break;
 902                 }
 903         spin_unlock(&queue_lock);
 904         return mask;
 905 }
 906
 907 static int cache_ioctl(struct inode *ino, struct file *filp,
 908                        unsigned int cmd, unsigned long arg,
 909                        struct cache_detail *cd)
 910 {
 911         int len = 0;
 912         struct cache_reader *rp = filp->private_data;
 913         struct cache_queue *cq;
 914
 915         if (cmd != FIONREAD || !rp)
 916                 return -EINVAL;
 917
 918         spin_lock(&queue_lock);
 919
 920         /* only find the length remaining in current request,
 921          * or the length of the next request
 922          */
 923         for (cq= &rp->q; &cq->list != &cd->queue;
 924              cq = list_entry(cq->list.next, struct cache_queue, list))
 925                 if (!cq->reader) {
 926                         struct cache_request *cr =
 927                                 container_of(cq, struct cache_request, q);
 928                         len = cr->len - rp->offset;
 929                         break;
 930                 }
 931         spin_unlock(&queue_lock);
 932
 933         return put_user(len, (int __user *)arg);
 934 }
 935
 936 static int cache_open(struct inode *inode, struct file *filp,
 937                       struct cache_detail *cd)
 938 {
 939         struct cache_reader *rp = NULL;
 940
 941         if (!cd || !try_module_get(cd->owner))
 942                 return -EACCES;
 943         nonseekable_open(inode, filp);
 944         if (filp->f_mode & FMODE_READ) {
 945                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 946                 if (!rp)
 947                         return -ENOMEM;
 948                 rp->offset = 0;
 949                 rp->q.reader = 1;
 950                 atomic_inc(&cd->readers);
 951                 spin_lock(&queue_lock);
 952                 list_add(&rp->q.list, &cd->queue);
 953                 spin_unlock(&queue_lock);
 954         }
 955         filp->private_data = rp;
 956         return 0;
 957 }
 958
 959 static int cache_release(struct inode *inode, struct file *filp,
 960                          struct cache_detail *cd)
 961 {
 962         struct cache_reader *rp = filp->private_data;
 963
 964         if (rp) {
 965                 spin_lock(&queue_lock);
 966                 if (rp->offset) {
 967                         struct cache_queue *cq;
 968                         for (cq= &rp->q; &cq->list != &cd->queue;
 969                              cq = list_entry(cq->list.next, struct cache_queue, list))
 970                                 if (!cq->reader) {
 971                                         container_of(cq, struct cache_request, q)
 972                                                 ->readers--;
 973                                         break;
 974                                 }
 975                         rp->offset = 0;
 976                 }
 977                 list_del(&rp->q.list);
 978                 spin_unlock(&queue_lock);
 979
 980                 filp->private_data = NULL;
 981                 kfree(rp);
 982
 983                 cd->last_close = seconds_since_boot();
 984                 atomic_dec(&cd->readers);
 985         }
 986         module_put(cd->owner);
 987         return 0;
 988 }
 989
 990
 991
 992 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
 993 {
 994         struct cache_queue *cq;
 995         spin_lock(&queue_lock);
 996         list_for_each_entry(cq, &detail->queue, list)
 997                 if (!cq->reader) {
 998                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 999                         if (cr->item != ch)
1000                                 continue;
1001                         if (cr->readers != 0)
1002                                 continue;
1003                         list_del(&cr->q.list);
1004                         spin_unlock(&queue_lock);
1005                         cache_put(cr->item, detail);
1006                         kfree(cr->buf);
1007                         kfree(cr);
1008                         return;
1009                 }
1010         spin_unlock(&queue_lock);
1011 }
1012
1013 /*
1014  * Support routines for text-based upcalls.
1015  * Fields are separated by spaces.
1016  * Fields are either mangled to quote space tab newline slosh with slosh
1017  * or a hexified with a leading \x
1018  * Record is terminated with newline.
1019  *
1020  */
1021
1022 void qword_add(char **bpp, int *lp, char *str)
1023 {
1024         char *bp = *bpp;
1025         int len = *lp;
1026         char c;
1027
1028         if (len < 0) return;
1029
1030         while ((c=*str++) && len)
1031                 switch(c) {
1032                 case ' ':
1033                 case '\t':
1034                 case '\n':
1035                 case '\\':
1036                         if (len >= 4) {
1037                                 *bp++ = '\\';
1038                                 *bp++ = '0' + ((c & 0300)>>6);
1039                                 *bp++ = '0' + ((c & 0070)>>3);
1040                                 *bp++ = '0' + ((c & 0007)>>0);
1041                         }
1042                         len -= 4;
1043                         break;
1044                 default:
1045                         *bp++ = c;
1046                         len--;
1047                 }
1048         if (c || len <1) len = -1;
1049         else {
1050                 *bp++ = ' ';
1051                 len--;
1052         }
1053         *bpp = bp;
1054         *lp = len;
1055 }
1056 EXPORT_SYMBOL_GPL(qword_add);
1057
1058 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1059 {
1060         char *bp = *bpp;
1061         int len = *lp;
1062
1063         if (len < 0) return;
1064
1065         if (len > 2) {
1066                 *bp++ = '\\';
1067                 *bp++ = 'x';
1068                 len -= 2;
1069                 while (blen && len >= 2) {
1070                         unsigned char c = *buf++;
1071                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1072                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1073                         len -= 2;
1074                         blen--;
1075                 }
1076         }
1077         if (blen || len<1) len = -1;
1078         else {
1079                 *bp++ = ' ';
1080                 len--;
1081         }
1082         *bpp = bp;
1083         *lp = len;
1084 }
1085 EXPORT_SYMBOL_GPL(qword_addhex);
1086
1087 static void warn_no_listener(struct cache_detail *detail)
1088 {
1089         if (detail->last_warn != detail->last_close) {
1090                 detail->last_warn = detail->last_close;
1091                 if (detail->warn_no_listener)
1092                         detail->warn_no_listener(detail, detail->last_close != 0);
1093         }
1094 }
1095
1096 static bool cache_listeners_exist(struct cache_detail *detail)
1097 {
1098         if (atomic_read(&detail->readers))
1099                 return true;
1100         if (detail->last_close == 0)
1101                 /* This cache was never opened */
1102                 return false;
1103         if (detail->last_close < seconds_since_boot() - 30)
1104                 /*
1105                  * We allow for the possibility that someone might
1106                  * restart a userspace daemon without restarting the
1107                  * server; but after 30 seconds, we give up.
1108                  */
1109                  return false;
1110         return true;
1111 }
1112
1113 /*
1114  * register an upcall request to user-space and queue it up for read() by the
1115  * upcall daemon.
1116  *
1117  * Each request is at most one page long.
1118  */
1119 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1120                 void (*cache_request)(struct cache_detail *,
1121                                       struct cache_head *,
1122                                       char **,
1123                                       int *))
1124 {
1125
1126         char *buf;
1127         struct cache_request *crq;
1128         char *bp;
1129         int len;
1130
1131         if (!cache_listeners_exist(detail)) {
1132                 warn_no_listener(detail);
1133                 return -EINVAL;
1134         }
1135
1136         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1137         if (!buf)
1138                 return -EAGAIN;
1139
1140         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1141         if (!crq) {
1142                 kfree(buf);
1143                 return -EAGAIN;
1144         }
1145
1146         bp = buf; len = PAGE_SIZE;
1147
1148         cache_request(detail, h, &bp, &len);
1149
1150         if (len < 0) {
1151                 kfree(buf);
1152                 kfree(crq);
1153                 return -EAGAIN;
1154         }
1155         crq->q.reader = 0;
1156         crq->item = cache_get(h);
1157         crq->buf = buf;
1158         crq->len = PAGE_SIZE - len;
1159         crq->readers = 0;
1160         spin_lock(&queue_lock);
1161         list_add_tail(&crq->q.list, &detail->queue);
1162         spin_unlock(&queue_lock);
1163         wake_up(&queue_wait);
1164         return 0;
1165 }
1166 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1167
1168 /*
1169  * parse a message from user-space and pass it
1170  * to an appropriate cache
1171  * Messages are, like requests, separated into fields by
1172  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1173  *
1174  * Message is
1175  *   reply cachename expiry key ... content....
1176  *
1177  * key and content are both parsed by cache
1178  */
1179
1180 #define isodigit(c) (isdigit(c) && c <= '7')
1181 int qword_get(char **bpp, char *dest, int bufsize)
1182 {
1183         /* return bytes copied, or -1 on error */
1184         char *bp = *bpp;
1185         int len = 0;
1186
1187         while (*bp == ' ') bp++;
1188
1189         if (bp[0] == '\\' && bp[1] == 'x') {
1190                 /* HEX STRING */
1191                 bp += 2;
1192                 while (len < bufsize) {
1193                         int h, l;
1194
1195                         h = hex_to_bin(bp[0]);
1196                         if (h < 0)
1197                                 break;
1198
1199                         l = hex_to_bin(bp[1]);
1200                         if (l < 0)
1201                                 break;
1202
1203                         *dest++ = (h << 4) | l;
1204                         bp += 2;
1205                         len++;
1206                 }
1207         } else {
1208                 /* text with \nnn octal quoting */
1209                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1210                         if (*bp == '\\' &&
1211                             isodigit(bp[1]) && (bp[1] <= '3') &&
1212                             isodigit(bp[2]) &&
1213                             isodigit(bp[3])) {
1214                                 int byte = (*++bp -'0');
1215                                 bp++;
1216                                 byte = (byte << 3) | (*bp++ - '0');
1217                                 byte = (byte << 3) | (*bp++ - '0');
1218                                 *dest++ = byte;
1219                                 len++;
1220                         } else {
1221                                 *dest++ = *bp++;
1222                                 len++;
1223                         }
1224                 }
1225         }
1226
1227         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1228                 return -1;
1229         while (*bp == ' ') bp++;
1230         *bpp = bp;
1231         *dest = '\0';
1232         return len;
1233 }
1234 EXPORT_SYMBOL_GPL(qword_get);
1235
1236
1237 /*
1238  * support /proc/sunrpc/cache/$CACHENAME/content
1239  * as a seqfile.
1240  * We call ->cache_show passing NULL for the item to
1241  * get a header, then pass each real item in the cache
1242  */
1243
1244 struct handle {
1245         struct cache_detail *cd;
1246 };
1247
1248 static void *c_start(struct seq_file *m, loff_t *pos)
1249         __acquires(cd->hash_lock)
1250 {
1251         loff_t n = *pos;
1252         unsigned hash, entry;
1253         struct cache_head *ch;
1254         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1255
1256
1257         read_lock(&cd->hash_lock);
1258         if (!n--)
1259                 return SEQ_START_TOKEN;
1260         hash = n >> 32;
1261         entry = n & ((1LL<<32) - 1);
1262
1263         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1264                 if (!entry--)
1265                         return ch;
1266         n &= ~((1LL<<32) - 1);
1267         do {
1268                 hash++;
1269                 n += 1LL<<32;
1270         } while(hash < cd->hash_size &&
1271                 cd->hash_table[hash]==NULL);
1272         if (hash >= cd->hash_size)
1273                 return NULL;
1274         *pos = n+1;
1275         return cd->hash_table[hash];
1276 }
1277
1278 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1279 {
1280         struct cache_head *ch = p;
1281         int hash = (*pos >> 32);
1282         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1283
1284         if (p == SEQ_START_TOKEN)
1285                 hash = 0;
1286         else if (ch->next == NULL) {
1287                 hash++;
1288                 *pos += 1LL<<32;
1289         } else {
1290                 ++*pos;
1291                 return ch->next;
1292         }
1293         *pos &= ~((1LL<<32) - 1);
1294         while (hash < cd->hash_size &&
1295                cd->hash_table[hash] == NULL) {
1296                 hash++;
1297                 *pos += 1LL<<32;
1298         }
1299         if (hash >= cd->hash_size)
1300                 return NULL;
1301         ++*pos;
1302         return cd->hash_table[hash];
1303 }
1304
1305 static void c_stop(struct seq_file *m, void *p)
1306         __releases(cd->hash_lock)
1307 {
1308         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1309         read_unlock(&cd->hash_lock);
1310 }
1311
1312 static int c_show(struct seq_file *m, void *p)
1313 {
1314         struct cache_head *cp = p;
1315         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1316
1317         if (p == SEQ_START_TOKEN)
1318                 return cd->cache_show(m, cd, NULL);
1319
1320         ifdebug(CACHE)
1321                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1322                            convert_to_wallclock(cp->expiry_time),
1323                            atomic_read(&cp->ref.refcount), cp->flags);
1324         cache_get(cp);
1325         if (cache_check(cd, cp, NULL))
1326                 /* cache_check does a cache_put on failure */
1327                 seq_printf(m, "# ");
1328         else
1329                 cache_put(cp, cd);
1330
1331         return cd->cache_show(m, cd, cp);
1332 }
1333
1334 static const struct seq_operations cache_content_op = {
1335         .start  = c_start,
1336         .next   = c_next,
1337         .stop   = c_stop,
1338         .show   = c_show,
1339 };
1340
1341 static int content_open(struct inode *inode, struct file *file,
1342                         struct cache_detail *cd)
1343 {
1344         struct handle *han;
1345
1346         if (!cd || !try_module_get(cd->owner))
1347                 return -EACCES;
1348         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1349         if (han == NULL) {
1350                 module_put(cd->owner);
1351                 return -ENOMEM;
1352         }
1353
1354         han->cd = cd;
1355         return 0;
1356 }
1357
1358 static int content_release(struct inode *inode, struct file *file,
1359                 struct cache_detail *cd)
1360 {
1361         int ret = seq_release_private(inode, file);
1362         module_put(cd->owner);
1363         return ret;
1364 }
1365
1366 static int open_flush(struct inode *inode, struct file *file,
1367                         struct cache_detail *cd)
1368 {
1369         if (!cd || !try_module_get(cd->owner))
1370                 return -EACCES;
1371         return nonseekable_open(inode, file);
1372 }
1373
1374 static int release_flush(struct inode *inode, struct file *file,
1375                         struct cache_detail *cd)
1376 {
1377         module_put(cd->owner);
1378         return 0;
1379 }
1380
1381 static ssize_t read_flush(struct file *file, char __user *buf,
1382                           size_t count, loff_t *ppos,
1383                           struct cache_detail *cd)
1384 {
1385         char tbuf[20];
1386         unsigned long p = *ppos;
1387         size_t len;
1388
1389         sprintf(tbuf, "%lu\n", convert_to_wallclock(cd->flush_time));
1390         len = strlen(tbuf);
1391         if (p >= len)
1392                 return 0;
1393         len -= p;
1394         if (len > count)
1395                 len = count;
1396         if (copy_to_user(buf, (void*)(tbuf+p), len))
1397                 return -EFAULT;
1398         *ppos += len;
1399         return len;
1400 }
1401
1402 static ssize_t write_flush(struct file *file, const char __user *buf,
1403                            size_t count, loff_t *ppos,
1404                            struct cache_detail *cd)
1405 {
1406         char tbuf[20];
1407         char *bp, *ep;
1408
1409         if (*ppos || count > sizeof(tbuf)-1)
1410                 return -EINVAL;
1411         if (copy_from_user(tbuf, buf, count))
1412                 return -EFAULT;
1413         tbuf[count] = 0;
1414         simple_strtoul(tbuf, &ep, 0);
1415         if (*ep && *ep != '\n')
1416                 return -EINVAL;
1417
1418         bp = tbuf;
1419         cd->flush_time = get_expiry(&bp);
1420         cd->nextcheck = seconds_since_boot();
1421         cache_flush();
1422
1423         *ppos += count;
1424         return count;
1425 }
1426
1427 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1428                                  size_t count, loff_t *ppos)
1429 {
1430         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1431
1432         return cache_read(filp, buf, count, ppos, cd);
1433 }
1434
1435 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1436                                   size_t count, loff_t *ppos)
1437 {
1438         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1439
1440         return cache_write(filp, buf, count, ppos, cd);
1441 }
1442
1443 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1444 {
1445         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1446
1447         return cache_poll(filp, wait, cd);
1448 }
1449
1450 static long cache_ioctl_procfs(struct file *filp,
1451                                unsigned int cmd, unsigned long arg)
1452 {
1453         struct inode *inode = filp->f_path.dentry->d_inode;
1454         struct cache_detail *cd = PDE(inode)->data;
1455
1456         return cache_ioctl(inode, filp, cmd, arg, cd);
1457 }
1458
1459 static int cache_open_procfs(struct inode *inode, struct file *filp)
1460 {
1461         struct cache_detail *cd = PDE(inode)->data;
1462
1463         return cache_open(inode, filp, cd);
1464 }
1465
1466 static int cache_release_procfs(struct inode *inode, struct file *filp)
1467 {
1468         struct cache_detail *cd = PDE(inode)->data;
1469
1470         return cache_release(inode, filp, cd);
1471 }
1472
1473 static const struct file_operations cache_file_operations_procfs = {
1474         .owner          = THIS_MODULE,
1475         .llseek         = no_llseek,
1476         .read           = cache_read_procfs,
1477         .write          = cache_write_procfs,
1478         .poll           = cache_poll_procfs,
1479         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1480         .open           = cache_open_procfs,
1481         .release        = cache_release_procfs,
1482 };
1483
1484 static int content_open_procfs(struct inode *inode, struct file *filp)
1485 {
1486         struct cache_detail *cd = PDE(inode)->data;
1487
1488         return content_open(inode, filp, cd);
1489 }
1490
1491 static int content_release_procfs(struct inode *inode, struct file *filp)
1492 {
1493         struct cache_detail *cd = PDE(inode)->data;
1494
1495         return content_release(inode, filp, cd);
1496 }
1497
1498 static const struct file_operations content_file_operations_procfs = {
1499         .open           = content_open_procfs,
1500         .read           = seq_read,
1501         .llseek         = seq_lseek,
1502         .release        = content_release_procfs,
1503 };
1504
1505 static int open_flush_procfs(struct inode *inode, struct file *filp)
1506 {
1507         struct cache_detail *cd = PDE(inode)->data;
1508
1509         return open_flush(inode, filp, cd);
1510 }
1511
1512 static int release_flush_procfs(struct inode *inode, struct file *filp)
1513 {
1514         struct cache_detail *cd = PDE(inode)->data;
1515
1516         return release_flush(inode, filp, cd);
1517 }
1518
1519 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1520                             size_t count, loff_t *ppos)
1521 {
1522         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1523
1524         return read_flush(filp, buf, count, ppos, cd);
1525 }
1526
1527 static ssize_t write_flush_procfs(struct file *filp,
1528                                   const char __user *buf,
1529                                   size_t count, loff_t *ppos)
1530 {
1531         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1532
1533         return write_flush(filp, buf, count, ppos, cd);
1534 }
1535
1536 static const struct file_operations cache_flush_operations_procfs = {
1537         .open           = open_flush_procfs,
1538         .read           = read_flush_procfs,
1539         .write          = write_flush_procfs,
1540         .release        = release_flush_procfs,
1541         .llseek         = no_llseek,
1542 };
1543
1544 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1545 {
1546         struct sunrpc_net *sn;
1547
1548         if (cd->u.procfs.proc_ent == NULL)
1549                 return;
1550         if (cd->u.procfs.flush_ent)
1551                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1552         if (cd->u.procfs.channel_ent)
1553                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1554         if (cd->u.procfs.content_ent)
1555                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1556         cd->u.procfs.proc_ent = NULL;
1557         sn = net_generic(net, sunrpc_net_id);
1558         remove_proc_entry(cd->name, sn->proc_net_rpc);
1559 }
1560
1561 #ifdef CONFIG_PROC_FS
1562 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1563 {
1564         struct proc_dir_entry *p;
1565         struct sunrpc_net *sn;
1566
1567         sn = net_generic(net, sunrpc_net_id);
1568         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1569         if (cd->u.procfs.proc_ent == NULL)
1570                 goto out_nomem;
1571         cd->u.procfs.channel_ent = NULL;
1572         cd->u.procfs.content_ent = NULL;
1573
1574         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1575                              cd->u.procfs.proc_ent,
1576                              &cache_flush_operations_procfs, cd);
1577         cd->u.procfs.flush_ent = p;
1578         if (p == NULL)
1579                 goto out_nomem;
1580
1581         if (cd->cache_upcall || cd->cache_parse) {
1582                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1583                                      cd->u.procfs.proc_ent,
1584                                      &cache_file_operations_procfs, cd);
1585                 cd->u.procfs.channel_ent = p;
1586                 if (p == NULL)
1587                         goto out_nomem;
1588         }
1589         if (cd->cache_show) {
1590                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1591                                 cd->u.procfs.proc_ent,
1592                                 &content_file_operations_procfs, cd);
1593                 cd->u.procfs.content_ent = p;
1594                 if (p == NULL)
1595                         goto out_nomem;
1596         }
1597         return 0;
1598 out_nomem:
1599         remove_cache_proc_entries(cd, net);
1600         return -ENOMEM;
1601 }
1602 #else /* CONFIG_PROC_FS */
1603 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1604 {
1605         return 0;
1606 }
1607 #endif
1608
1609 void __init cache_initialize(void)
1610 {
1611         INIT_DELAYED_WORK_DEFERRABLE(&cache_cleaner, do_cache_clean);
1612 }
1613
1614 int cache_register_net(struct cache_detail *cd, struct net *net)
1615 {
1616         int ret;
1617
1618         sunrpc_init_cache_detail(cd);
1619         ret = create_cache_proc_entries(cd, net);
1620         if (ret)
1621                 sunrpc_destroy_cache_detail(cd);
1622         return ret;
1623 }
1624
1625 int cache_register(struct cache_detail *cd)
1626 {
1627         return cache_register_net(cd, &init_net);
1628 }
1629 EXPORT_SYMBOL_GPL(cache_register);
1630
1631 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1632 {
1633         remove_cache_proc_entries(cd, net);
1634         sunrpc_destroy_cache_detail(cd);
1635 }
1636
1637 void cache_unregister(struct cache_detail *cd)
1638 {
1639         cache_unregister_net(cd, &init_net);
1640 }
1641 EXPORT_SYMBOL_GPL(cache_unregister);
1642
1643 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1644                                  size_t count, loff_t *ppos)
1645 {
1646         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1647
1648         return cache_read(filp, buf, count, ppos, cd);
1649 }
1650
1651 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1652                                   size_t count, loff_t *ppos)
1653 {
1654         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1655
1656         return cache_write(filp, buf, count, ppos, cd);
1657 }
1658
1659 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1660 {
1661         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1662
1663         return cache_poll(filp, wait, cd);
1664 }
1665
1666 static long cache_ioctl_pipefs(struct file *filp,
1667                               unsigned int cmd, unsigned long arg)
1668 {
1669         struct inode *inode = filp->f_dentry->d_inode;
1670         struct cache_detail *cd = RPC_I(inode)->private;
1671
1672         return cache_ioctl(inode, filp, cmd, arg, cd);
1673 }
1674
1675 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1676 {
1677         struct cache_detail *cd = RPC_I(inode)->private;
1678
1679         return cache_open(inode, filp, cd);
1680 }
1681
1682 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1683 {
1684         struct cache_detail *cd = RPC_I(inode)->private;
1685
1686         return cache_release(inode, filp, cd);
1687 }
1688
1689 const struct file_operations cache_file_operations_pipefs = {
1690         .owner          = THIS_MODULE,
1691         .llseek         = no_llseek,
1692         .read           = cache_read_pipefs,
1693         .write          = cache_write_pipefs,
1694         .poll           = cache_poll_pipefs,
1695         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1696         .open           = cache_open_pipefs,
1697         .release        = cache_release_pipefs,
1698 };
1699
1700 static int content_open_pipefs(struct inode *inode, struct file *filp)
1701 {
1702         struct cache_detail *cd = RPC_I(inode)->private;
1703
1704         return content_open(inode, filp, cd);
1705 }
1706
1707 static int content_release_pipefs(struct inode *inode, struct file *filp)
1708 {
1709         struct cache_detail *cd = RPC_I(inode)->private;
1710
1711         return content_release(inode, filp, cd);
1712 }
1713
1714 const struct file_operations content_file_operations_pipefs = {
1715         .open           = content_open_pipefs,
1716         .read           = seq_read,
1717         .llseek         = seq_lseek,
1718         .release        = content_release_pipefs,
1719 };
1720
1721 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1722 {
1723         struct cache_detail *cd = RPC_I(inode)->private;
1724
1725         return open_flush(inode, filp, cd);
1726 }
1727
1728 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1729 {
1730         struct cache_detail *cd = RPC_I(inode)->private;
1731
1732         return release_flush(inode, filp, cd);
1733 }
1734
1735 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1736                             size_t count, loff_t *ppos)
1737 {
1738         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1739
1740         return read_flush(filp, buf, count, ppos, cd);
1741 }
1742
1743 static ssize_t write_flush_pipefs(struct file *filp,
1744                                   const char __user *buf,
1745                                   size_t count, loff_t *ppos)
1746 {
1747         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1748
1749         return write_flush(filp, buf, count, ppos, cd);
1750 }
1751
1752 const struct file_operations cache_flush_operations_pipefs = {
1753         .open           = open_flush_pipefs,
1754         .read           = read_flush_pipefs,
1755         .write          = write_flush_pipefs,
1756         .release        = release_flush_pipefs,
1757         .llseek         = no_llseek,
1758 };
1759
1760 int sunrpc_cache_register_pipefs(struct dentry *parent,
1761                                  const char *name, mode_t umode,
1762                                  struct cache_detail *cd)
1763 {
1764         struct qstr q;
1765         struct dentry *dir;
1766         int ret = 0;
1767
1768         sunrpc_init_cache_detail(cd);
1769         q.name = name;
1770         q.len = strlen(name);
1771         q.hash = full_name_hash(q.name, q.len);
1772         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1773         if (!IS_ERR(dir))
1774                 cd->u.pipefs.dir = dir;
1775         else {
1776                 sunrpc_destroy_cache_detail(cd);
1777                 ret = PTR_ERR(dir);
1778         }
1779         return ret;
1780 }
1781 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1782
1783 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1784 {
1785         rpc_remove_cache_dir(cd->u.pipefs.dir);
1786         cd->u.pipefs.dir = NULL;
1787         sunrpc_destroy_cache_detail(cd);
1788 }
1789 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1790