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 <linux/smp_lock.h>
  32 #include <asm/ioctls.h>
  33 #include <linux/sunrpc/types.h>
  34 #include <linux/sunrpc/cache.h>
  35 #include <linux/sunrpc/stats.h>
  36 #include <linux/sunrpc/rpc_pipe_fs.h>
  37
  38 #define  RPCDBG_FACILITY RPCDBG_CACHE
  39
  40 static int 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) < 0) {
 272                         /* Request is not deferred */
 273                         rv = cache_is_valid(detail, h);
 274                         if (rv == -EAGAIN)
 275                                 rv = -ETIMEDOUT;
 276                 }
 277         }
 278         if (rv)
 279                 cache_put(h, detail);
 280         return rv;
 281 }
 282 EXPORT_SYMBOL_GPL(cache_check);
 283
 284 /*
 285  * caches need to be periodically cleaned.
 286  * For this we maintain a list of cache_detail and
 287  * a current pointer into that list and into the table
 288  * for that entry.
 289  *
 290  * Each time clean_cache is called it finds the next non-empty entry
 291  * in the current table and walks the list in that entry
 292  * looking for entries that can be removed.
 293  *
 294  * An entry gets removed if:
 295  * - The expiry is before current time
 296  * - The last_refresh time is before the flush_time for that cache
 297  *
 298  * later we might drop old entries with non-NEVER expiry if that table
 299  * is getting 'full' for some definition of 'full'
 300  *
 301  * The question of "how often to scan a table" is an interesting one
 302  * and is answered in part by the use of the "nextcheck" field in the
 303  * cache_detail.
 304  * When a scan of a table begins, the nextcheck field is set to a time
 305  * that is well into the future.
 306  * While scanning, if an expiry time is found that is earlier than the
 307  * current nextcheck time, nextcheck is set to that expiry time.
 308  * If the flush_time is ever set to a time earlier than the nextcheck
 309  * time, the nextcheck time is then set to that flush_time.
 310  *
 311  * A table is then only scanned if the current time is at least
 312  * the nextcheck time.
 313  *
 314  */
 315
 316 static LIST_HEAD(cache_list);
 317 static DEFINE_SPINLOCK(cache_list_lock);
 318 static struct cache_detail *current_detail;
 319 static int current_index;
 320
 321 static void do_cache_clean(struct work_struct *work);
 322 static struct delayed_work cache_cleaner;
 323
 324 static void sunrpc_init_cache_detail(struct cache_detail *cd)
 325 {
 326         rwlock_init(&cd->hash_lock);
 327         INIT_LIST_HEAD(&cd->queue);
 328         spin_lock(&cache_list_lock);
 329         cd->nextcheck = 0;
 330         cd->entries = 0;
 331         atomic_set(&cd->readers, 0);
 332         cd->last_close = 0;
 333         cd->last_warn = -1;
 334         list_add(&cd->others, &cache_list);
 335         spin_unlock(&cache_list_lock);
 336
 337         /* start the cleaning process */
 338         schedule_delayed_work(&cache_cleaner, 0);
 339 }
 340
 341 static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 342 {
 343         cache_purge(cd);
 344         spin_lock(&cache_list_lock);
 345         write_lock(&cd->hash_lock);
 346         if (cd->entries || atomic_read(&cd->inuse)) {
 347                 write_unlock(&cd->hash_lock);
 348                 spin_unlock(&cache_list_lock);
 349                 goto out;
 350         }
 351         if (current_detail == cd)
 352                 current_detail = NULL;
 353         list_del_init(&cd->others);
 354         write_unlock(&cd->hash_lock);
 355         spin_unlock(&cache_list_lock);
 356         if (list_empty(&cache_list)) {
 357                 /* module must be being unloaded so its safe to kill the worker */
 358                 cancel_delayed_work_sync(&cache_cleaner);
 359         }
 360         return;
 361 out:
 362         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 363 }
 364
 365 /* clean cache tries to find something to clean
 366  * and cleans it.
 367  * It returns 1 if it cleaned something,
 368  *            0 if it didn't find anything this time
 369  *           -1 if it fell off the end of the list.
 370  */
 371 static int cache_clean(void)
 372 {
 373         int rv = 0;
 374         struct list_head *next;
 375
 376         spin_lock(&cache_list_lock);
 377
 378         /* find a suitable table if we don't already have one */
 379         while (current_detail == NULL ||
 380             current_index >= current_detail->hash_size) {
 381                 if (current_detail)
 382                         next = current_detail->others.next;
 383                 else
 384                         next = cache_list.next;
 385                 if (next == &cache_list) {
 386                         current_detail = NULL;
 387                         spin_unlock(&cache_list_lock);
 388                         return -1;
 389                 }
 390                 current_detail = list_entry(next, struct cache_detail, others);
 391                 if (current_detail->nextcheck > seconds_since_boot())
 392                         current_index = current_detail->hash_size;
 393                 else {
 394                         current_index = 0;
 395                         current_detail->nextcheck = seconds_since_boot()+30*60;
 396                 }
 397         }
 398
 399         /* find a non-empty bucket in the table */
 400         while (current_detail &&
 401                current_index < current_detail->hash_size &&
 402                current_detail->hash_table[current_index] == NULL)
 403                 current_index++;
 404
 405         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 406
 407         if (current_detail && current_index < current_detail->hash_size) {
 408                 struct cache_head *ch, **cp;
 409                 struct cache_detail *d;
 410
 411                 write_lock(&current_detail->hash_lock);
 412
 413                 /* Ok, now to clean this strand */
 414
 415                 cp = & current_detail->hash_table[current_index];
 416                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
 417                         if (current_detail->nextcheck > ch->expiry_time)
 418                                 current_detail->nextcheck = ch->expiry_time+1;
 419                         if (!cache_is_expired(current_detail, ch))
 420                                 continue;
 421
 422                         *cp = ch->next;
 423                         ch->next = NULL;
 424                         current_detail->entries--;
 425                         rv = 1;
 426                         break;
 427                 }
 428
 429                 write_unlock(&current_detail->hash_lock);
 430                 d = current_detail;
 431                 if (!ch)
 432                         current_index ++;
 433                 spin_unlock(&cache_list_lock);
 434                 if (ch) {
 435                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 436                                 cache_dequeue(current_detail, ch);
 437                         cache_revisit_request(ch);
 438                         cache_put(ch, d);
 439                 }
 440         } else
 441                 spin_unlock(&cache_list_lock);
 442
 443         return rv;
 444 }
 445
 446 /*
 447  * We want to regularly clean the cache, so we need to schedule some work ...
 448  */
 449 static void do_cache_clean(struct work_struct *work)
 450 {
 451         int delay = 5;
 452         if (cache_clean() == -1)
 453                 delay = round_jiffies_relative(30*HZ);
 454
 455         if (list_empty(&cache_list))
 456                 delay = 0;
 457
 458         if (delay)
 459                 schedule_delayed_work(&cache_cleaner, delay);
 460 }
 461
 462
 463 /*
 464  * Clean all caches promptly.  This just calls cache_clean
 465  * repeatedly until we are sure that every cache has had a chance to
 466  * be fully cleaned
 467  */
 468 void cache_flush(void)
 469 {
 470         while (cache_clean() != -1)
 471                 cond_resched();
 472         while (cache_clean() != -1)
 473                 cond_resched();
 474 }
 475 EXPORT_SYMBOL_GPL(cache_flush);
 476
 477 void cache_purge(struct cache_detail *detail)
 478 {
 479         detail->flush_time = LONG_MAX;
 480         detail->nextcheck = seconds_since_boot();
 481         cache_flush();
 482         detail->flush_time = 1;
 483 }
 484 EXPORT_SYMBOL_GPL(cache_purge);
 485
 486
 487 /*
 488  * Deferral and Revisiting of Requests.
 489  *
 490  * If a cache lookup finds a pending entry, we
 491  * need to defer the request and revisit it later.
 492  * All deferred requests are stored in a hash table,
 493  * indexed by "struct cache_head *".
 494  * As it may be wasteful to store a whole request
 495  * structure, we allow the request to provide a
 496  * deferred form, which must contain a
 497  * 'struct cache_deferred_req'
 498  * This cache_deferred_req contains a method to allow
 499  * it to be revisited when cache info is available
 500  */
 501
 502 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 503 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 504
 505 #define DFR_MAX 300     /* ??? */
 506
 507 static DEFINE_SPINLOCK(cache_defer_lock);
 508 static LIST_HEAD(cache_defer_list);
 509 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 510 static int cache_defer_cnt;
 511
 512 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 513 {
 514         list_del_init(&dreq->recent);
 515         hlist_del_init(&dreq->hash);
 516         cache_defer_cnt--;
 517 }
 518
 519 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 520 {
 521         int hash = DFR_HASH(item);
 522
 523         list_add(&dreq->recent, &cache_defer_list);
 524         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 525 }
 526
 527 static int setup_deferral(struct cache_deferred_req *dreq, struct cache_head *item)
 528 {
 529         struct cache_deferred_req *discard;
 530
 531         dreq->item = item;
 532
 533         spin_lock(&cache_defer_lock);
 534
 535         __hash_deferred_req(dreq, item);
 536
 537         /* it is in, now maybe clean up */
 538         discard = NULL;
 539         if (++cache_defer_cnt > DFR_MAX) {
 540                 discard = list_entry(cache_defer_list.prev,
 541                                      struct cache_deferred_req, recent);
 542                 __unhash_deferred_req(discard);
 543         }
 544         spin_unlock(&cache_defer_lock);
 545
 546         if (discard)
 547                 /* there was one too many */
 548                 discard->revisit(discard, 1);
 549
 550         if (!test_bit(CACHE_PENDING, &item->flags)) {
 551                 /* must have just been validated... */
 552                 cache_revisit_request(item);
 553                 return -EAGAIN;
 554         }
 555         return 0;
 556 }
 557
 558 struct thread_deferred_req {
 559         struct cache_deferred_req handle;
 560         struct completion completion;
 561 };
 562
 563 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 564 {
 565         struct thread_deferred_req *dr =
 566                 container_of(dreq, struct thread_deferred_req, handle);
 567         complete(&dr->completion);
 568 }
 569
 570 static int cache_wait_req(struct cache_req *req, struct cache_head *item)
 571 {
 572         struct thread_deferred_req sleeper;
 573         struct cache_deferred_req *dreq = &sleeper.handle;
 574         int ret;
 575
 576         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 577         dreq->revisit = cache_restart_thread;
 578
 579         ret = setup_deferral(dreq, item);
 580         if (ret)
 581                 return ret;
 582
 583         if (wait_for_completion_interruptible_timeout(
 584                     &sleeper.completion, req->thread_wait) <= 0) {
 585                 /* The completion wasn't completed, so we need
 586                  * to clean up
 587                  */
 588                 spin_lock(&cache_defer_lock);
 589                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 590                         __unhash_deferred_req(&sleeper.handle);
 591                         spin_unlock(&cache_defer_lock);
 592                 } else {
 593                         /* cache_revisit_request already removed
 594                          * this from the hash table, but hasn't
 595                          * called ->revisit yet.  It will very soon
 596                          * and we need to wait for it.
 597                          */
 598                         spin_unlock(&cache_defer_lock);
 599                         wait_for_completion(&sleeper.completion);
 600                 }
 601         }
 602         if (test_bit(CACHE_PENDING, &item->flags)) {
 603                 /* item is still pending, try request
 604                  * deferral
 605                  */
 606                 return -ETIMEDOUT;
 607         }
 608         /* only return success if we actually deferred the
 609          * request.  In this case we waited until it was
 610          * answered so no deferral has happened - rather
 611          * an answer already exists.
 612          */
 613         return -EEXIST;
 614 }
 615
 616 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 617 {
 618         struct cache_deferred_req *dreq;
 619         int ret;
 620
 621         if (cache_defer_cnt >= DFR_MAX) {
 622                 /* too much in the cache, randomly drop this one,
 623                  * or continue and drop the oldest
 624                  */
 625                 if (net_random()&1)
 626                         return -ENOMEM;
 627         }
 628         if (req->thread_wait) {
 629                 ret = cache_wait_req(req, item);
 630                 if (ret != -ETIMEDOUT)
 631                         return ret;
 632         }
 633         dreq = req->defer(req);
 634         if (dreq == NULL)
 635                 return -ENOMEM;
 636         return setup_deferral(dreq, item);
 637 }
 638
 639 static void cache_revisit_request(struct cache_head *item)
 640 {
 641         struct cache_deferred_req *dreq;
 642         struct list_head pending;
 643         struct hlist_node *lp, *tmp;
 644         int hash = DFR_HASH(item);
 645
 646         INIT_LIST_HEAD(&pending);
 647         spin_lock(&cache_defer_lock);
 648
 649         hlist_for_each_entry_safe(dreq, lp, tmp, &cache_defer_hash[hash], hash)
 650                 if (dreq->item == item) {
 651                         __unhash_deferred_req(dreq);
 652                         list_add(&dreq->recent, &pending);
 653                 }
 654
 655         spin_unlock(&cache_defer_lock);
 656
 657         while (!list_empty(&pending)) {
 658                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 659                 list_del_init(&dreq->recent);
 660                 dreq->revisit(dreq, 0);
 661         }
 662 }
 663
 664 void cache_clean_deferred(void *owner)
 665 {
 666         struct cache_deferred_req *dreq, *tmp;
 667         struct list_head pending;
 668
 669
 670         INIT_LIST_HEAD(&pending);
 671         spin_lock(&cache_defer_lock);
 672
 673         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 674                 if (dreq->owner == owner) {
 675                         __unhash_deferred_req(dreq);
 676                         list_add(&dreq->recent, &pending);
 677                 }
 678         }
 679         spin_unlock(&cache_defer_lock);
 680
 681         while (!list_empty(&pending)) {
 682                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 683                 list_del_init(&dreq->recent);
 684                 dreq->revisit(dreq, 1);
 685         }
 686 }
 687
 688 /*
 689  * communicate with user-space
 690  *
 691  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 692  * On read, you get a full request, or block.
 693  * On write, an update request is processed.
 694  * Poll works if anything to read, and always allows write.
 695  *
 696  * Implemented by linked list of requests.  Each open file has
 697  * a ->private that also exists in this list.  New requests are added
 698  * to the end and may wakeup and preceding readers.
 699  * New readers are added to the head.  If, on read, an item is found with
 700  * CACHE_UPCALLING clear, we free it from the list.
 701  *
 702  */
 703
 704 static DEFINE_SPINLOCK(queue_lock);
 705 static DEFINE_MUTEX(queue_io_mutex);
 706
 707 struct cache_queue {
 708         struct list_head        list;
 709         int                     reader; /* if 0, then request */
 710 };
 711 struct cache_request {
 712         struct cache_queue      q;
 713         struct cache_head       *item;
 714         char                    * buf;
 715         int                     len;
 716         int                     readers;
 717 };
 718 struct cache_reader {
 719         struct cache_queue      q;
 720         int                     offset; /* if non-0, we have a refcnt on next request */
 721 };
 722
 723 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 724                           loff_t *ppos, struct cache_detail *cd)
 725 {
 726         struct cache_reader *rp = filp->private_data;
 727         struct cache_request *rq;
 728         struct inode *inode = filp->f_path.dentry->d_inode;
 729         int err;
 730
 731         if (count == 0)
 732                 return 0;
 733
 734         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 735                               * readers on this file */
 736  again:
 737         spin_lock(&queue_lock);
 738         /* need to find next request */
 739         while (rp->q.list.next != &cd->queue &&
 740                list_entry(rp->q.list.next, struct cache_queue, list)
 741                ->reader) {
 742                 struct list_head *next = rp->q.list.next;
 743                 list_move(&rp->q.list, next);
 744         }
 745         if (rp->q.list.next == &cd->queue) {
 746                 spin_unlock(&queue_lock);
 747                 mutex_unlock(&inode->i_mutex);
 748                 BUG_ON(rp->offset);
 749                 return 0;
 750         }
 751         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 752         BUG_ON(rq->q.reader);
 753         if (rp->offset == 0)
 754                 rq->readers++;
 755         spin_unlock(&queue_lock);
 756
 757         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 758                 err = -EAGAIN;
 759                 spin_lock(&queue_lock);
 760                 list_move(&rp->q.list, &rq->q.list);
 761                 spin_unlock(&queue_lock);
 762         } else {
 763                 if (rp->offset + count > rq->len)
 764                         count = rq->len - rp->offset;
 765                 err = -EFAULT;
 766                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 767                         goto out;
 768                 rp->offset += count;
 769                 if (rp->offset >= rq->len) {
 770                         rp->offset = 0;
 771                         spin_lock(&queue_lock);
 772                         list_move(&rp->q.list, &rq->q.list);
 773                         spin_unlock(&queue_lock);
 774                 }
 775                 err = 0;
 776         }
 777  out:
 778         if (rp->offset == 0) {
 779                 /* need to release rq */
 780                 spin_lock(&queue_lock);
 781                 rq->readers--;
 782                 if (rq->readers == 0 &&
 783                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 784                         list_del(&rq->q.list);
 785                         spin_unlock(&queue_lock);
 786                         cache_put(rq->item, cd);
 787                         kfree(rq->buf);
 788                         kfree(rq);
 789                 } else
 790                         spin_unlock(&queue_lock);
 791         }
 792         if (err == -EAGAIN)
 793                 goto again;
 794         mutex_unlock(&inode->i_mutex);
 795         return err ? err :  count;
 796 }
 797
 798 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 799                                  size_t count, struct cache_detail *cd)
 800 {
 801         ssize_t ret;
 802
 803         if (copy_from_user(kaddr, buf, count))
 804                 return -EFAULT;
 805         kaddr[count] = '\0';
 806         ret = cd->cache_parse(cd, kaddr, count);
 807         if (!ret)
 808                 ret = count;
 809         return ret;
 810 }
 811
 812 static ssize_t cache_slow_downcall(const char __user *buf,
 813                                    size_t count, struct cache_detail *cd)
 814 {
 815         static char write_buf[8192]; /* protected by queue_io_mutex */
 816         ssize_t ret = -EINVAL;
 817
 818         if (count >= sizeof(write_buf))
 819                 goto out;
 820         mutex_lock(&queue_io_mutex);
 821         ret = cache_do_downcall(write_buf, buf, count, cd);
 822         mutex_unlock(&queue_io_mutex);
 823 out:
 824         return ret;
 825 }
 826
 827 static ssize_t cache_downcall(struct address_space *mapping,
 828                               const char __user *buf,
 829                               size_t count, struct cache_detail *cd)
 830 {
 831         struct page *page;
 832         char *kaddr;
 833         ssize_t ret = -ENOMEM;
 834
 835         if (count >= PAGE_CACHE_SIZE)
 836                 goto out_slow;
 837
 838         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 839         if (!page)
 840                 goto out_slow;
 841
 842         kaddr = kmap(page);
 843         ret = cache_do_downcall(kaddr, buf, count, cd);
 844         kunmap(page);
 845         unlock_page(page);
 846         page_cache_release(page);
 847         return ret;
 848 out_slow:
 849         return cache_slow_downcall(buf, count, cd);
 850 }
 851
 852 static ssize_t cache_write(struct file *filp, const char __user *buf,
 853                            size_t count, loff_t *ppos,
 854                            struct cache_detail *cd)
 855 {
 856         struct address_space *mapping = filp->f_mapping;
 857         struct inode *inode = filp->f_path.dentry->d_inode;
 858         ssize_t ret = -EINVAL;
 859
 860         if (!cd->cache_parse)
 861                 goto out;
 862
 863         mutex_lock(&inode->i_mutex);
 864         ret = cache_downcall(mapping, buf, count, cd);
 865         mutex_unlock(&inode->i_mutex);
 866 out:
 867         return ret;
 868 }
 869
 870 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 871
 872 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 873                                struct cache_detail *cd)
 874 {
 875         unsigned int mask;
 876         struct cache_reader *rp = filp->private_data;
 877         struct cache_queue *cq;
 878
 879         poll_wait(filp, &queue_wait, wait);
 880
 881         /* alway allow write */
 882         mask = POLL_OUT | POLLWRNORM;
 883
 884         if (!rp)
 885                 return mask;
 886
 887         spin_lock(&queue_lock);
 888
 889         for (cq= &rp->q; &cq->list != &cd->queue;
 890              cq = list_entry(cq->list.next, struct cache_queue, list))
 891                 if (!cq->reader) {
 892                         mask |= POLLIN | POLLRDNORM;
 893                         break;
 894                 }
 895         spin_unlock(&queue_lock);
 896         return mask;
 897 }
 898
 899 static int cache_ioctl(struct inode *ino, struct file *filp,
 900                        unsigned int cmd, unsigned long arg,
 901                        struct cache_detail *cd)
 902 {
 903         int len = 0;
 904         struct cache_reader *rp = filp->private_data;
 905         struct cache_queue *cq;
 906
 907         if (cmd != FIONREAD || !rp)
 908                 return -EINVAL;
 909
 910         spin_lock(&queue_lock);
 911
 912         /* only find the length remaining in current request,
 913          * or the length of the next request
 914          */
 915         for (cq= &rp->q; &cq->list != &cd->queue;
 916              cq = list_entry(cq->list.next, struct cache_queue, list))
 917                 if (!cq->reader) {
 918                         struct cache_request *cr =
 919                                 container_of(cq, struct cache_request, q);
 920                         len = cr->len - rp->offset;
 921                         break;
 922                 }
 923         spin_unlock(&queue_lock);
 924
 925         return put_user(len, (int __user *)arg);
 926 }
 927
 928 static int cache_open(struct inode *inode, struct file *filp,
 929                       struct cache_detail *cd)
 930 {
 931         struct cache_reader *rp = NULL;
 932
 933         if (!cd || !try_module_get(cd->owner))
 934                 return -EACCES;
 935         nonseekable_open(inode, filp);
 936         if (filp->f_mode & FMODE_READ) {
 937                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 938                 if (!rp)
 939                         return -ENOMEM;
 940                 rp->offset = 0;
 941                 rp->q.reader = 1;
 942                 atomic_inc(&cd->readers);
 943                 spin_lock(&queue_lock);
 944                 list_add(&rp->q.list, &cd->queue);
 945                 spin_unlock(&queue_lock);
 946         }
 947         filp->private_data = rp;
 948         return 0;
 949 }
 950
 951 static int cache_release(struct inode *inode, struct file *filp,
 952                          struct cache_detail *cd)
 953 {
 954         struct cache_reader *rp = filp->private_data;
 955
 956         if (rp) {
 957                 spin_lock(&queue_lock);
 958                 if (rp->offset) {
 959                         struct cache_queue *cq;
 960                         for (cq= &rp->q; &cq->list != &cd->queue;
 961                              cq = list_entry(cq->list.next, struct cache_queue, list))
 962                                 if (!cq->reader) {
 963                                         container_of(cq, struct cache_request, q)
 964                                                 ->readers--;
 965                                         break;
 966                                 }
 967                         rp->offset = 0;
 968                 }
 969                 list_del(&rp->q.list);
 970                 spin_unlock(&queue_lock);
 971
 972                 filp->private_data = NULL;
 973                 kfree(rp);
 974
 975                 cd->last_close = seconds_since_boot();
 976                 atomic_dec(&cd->readers);
 977         }
 978         module_put(cd->owner);
 979         return 0;
 980 }
 981
 982
 983
 984 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
 985 {
 986         struct cache_queue *cq;
 987         spin_lock(&queue_lock);
 988         list_for_each_entry(cq, &detail->queue, list)
 989                 if (!cq->reader) {
 990                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 991                         if (cr->item != ch)
 992                                 continue;
 993                         if (cr->readers != 0)
 994                                 continue;
 995                         list_del(&cr->q.list);
 996                         spin_unlock(&queue_lock);
 997                         cache_put(cr->item, detail);
 998                         kfree(cr->buf);
 999                         kfree(cr);
1000                         return;
1001                 }
1002         spin_unlock(&queue_lock);
1003 }
1004
1005 /*
1006  * Support routines for text-based upcalls.
1007  * Fields are separated by spaces.
1008  * Fields are either mangled to quote space tab newline slosh with slosh
1009  * or a hexified with a leading \x
1010  * Record is terminated with newline.
1011  *
1012  */
1013
1014 void qword_add(char **bpp, int *lp, char *str)
1015 {
1016         char *bp = *bpp;
1017         int len = *lp;
1018         char c;
1019
1020         if (len < 0) return;
1021
1022         while ((c=*str++) && len)
1023                 switch(c) {
1024                 case ' ':
1025                 case '\t':
1026                 case '\n':
1027                 case '\\':
1028                         if (len >= 4) {
1029                                 *bp++ = '\\';
1030                                 *bp++ = '0' + ((c & 0300)>>6);
1031                                 *bp++ = '0' + ((c & 0070)>>3);
1032                                 *bp++ = '0' + ((c & 0007)>>0);
1033                         }
1034                         len -= 4;
1035                         break;
1036                 default:
1037                         *bp++ = c;
1038                         len--;
1039                 }
1040         if (c || len <1) len = -1;
1041         else {
1042                 *bp++ = ' ';
1043                 len--;
1044         }
1045         *bpp = bp;
1046         *lp = len;
1047 }
1048 EXPORT_SYMBOL_GPL(qword_add);
1049
1050 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1051 {
1052         char *bp = *bpp;
1053         int len = *lp;
1054
1055         if (len < 0) return;
1056
1057         if (len > 2) {
1058                 *bp++ = '\\';
1059                 *bp++ = 'x';
1060                 len -= 2;
1061                 while (blen && len >= 2) {
1062                         unsigned char c = *buf++;
1063                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1064                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1065                         len -= 2;
1066                         blen--;
1067                 }
1068         }
1069         if (blen || len<1) len = -1;
1070         else {
1071                 *bp++ = ' ';
1072                 len--;
1073         }
1074         *bpp = bp;
1075         *lp = len;
1076 }
1077 EXPORT_SYMBOL_GPL(qword_addhex);
1078
1079 static void warn_no_listener(struct cache_detail *detail)
1080 {
1081         if (detail->last_warn != detail->last_close) {
1082                 detail->last_warn = detail->last_close;
1083                 if (detail->warn_no_listener)
1084                         detail->warn_no_listener(detail, detail->last_close != 0);
1085         }
1086 }
1087
1088 static bool cache_listeners_exist(struct cache_detail *detail)
1089 {
1090         if (atomic_read(&detail->readers))
1091                 return true;
1092         if (detail->last_close == 0)
1093                 /* This cache was never opened */
1094                 return false;
1095         if (detail->last_close < seconds_since_boot() - 30)
1096                 /*
1097                  * We allow for the possibility that someone might
1098                  * restart a userspace daemon without restarting the
1099                  * server; but after 30 seconds, we give up.
1100                  */
1101                  return false;
1102         return true;
1103 }
1104
1105 /*
1106  * register an upcall request to user-space and queue it up for read() by the
1107  * upcall daemon.
1108  *
1109  * Each request is at most one page long.
1110  */
1111 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1112                 void (*cache_request)(struct cache_detail *,
1113                                       struct cache_head *,
1114                                       char **,
1115                                       int *))
1116 {
1117
1118         char *buf;
1119         struct cache_request *crq;
1120         char *bp;
1121         int len;
1122
1123         if (!cache_listeners_exist(detail)) {
1124                 warn_no_listener(detail);
1125                 return -EINVAL;
1126         }
1127
1128         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1129         if (!buf)
1130                 return -EAGAIN;
1131
1132         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1133         if (!crq) {
1134                 kfree(buf);
1135                 return -EAGAIN;
1136         }
1137
1138         bp = buf; len = PAGE_SIZE;
1139
1140         cache_request(detail, h, &bp, &len);
1141
1142         if (len < 0) {
1143                 kfree(buf);
1144                 kfree(crq);
1145                 return -EAGAIN;
1146         }
1147         crq->q.reader = 0;
1148         crq->item = cache_get(h);
1149         crq->buf = buf;
1150         crq->len = PAGE_SIZE - len;
1151         crq->readers = 0;
1152         spin_lock(&queue_lock);
1153         list_add_tail(&crq->q.list, &detail->queue);
1154         spin_unlock(&queue_lock);
1155         wake_up(&queue_wait);
1156         return 0;
1157 }
1158 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1159
1160 /*
1161  * parse a message from user-space and pass it
1162  * to an appropriate cache
1163  * Messages are, like requests, separated into fields by
1164  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1165  *
1166  * Message is
1167  *   reply cachename expiry key ... content....
1168  *
1169  * key and content are both parsed by cache
1170  */
1171
1172 #define isodigit(c) (isdigit(c) && c <= '7')
1173 int qword_get(char **bpp, char *dest, int bufsize)
1174 {
1175         /* return bytes copied, or -1 on error */
1176         char *bp = *bpp;
1177         int len = 0;
1178
1179         while (*bp == ' ') bp++;
1180
1181         if (bp[0] == '\\' && bp[1] == 'x') {
1182                 /* HEX STRING */
1183                 bp += 2;
1184                 while (len < bufsize) {
1185                         int h, l;
1186
1187                         h = hex_to_bin(bp[0]);
1188                         if (h < 0)
1189                                 break;
1190
1191                         l = hex_to_bin(bp[1]);
1192                         if (l < 0)
1193                                 break;
1194
1195                         *dest++ = (h << 4) | l;
1196                         bp += 2;
1197                         len++;
1198                 }
1199         } else {
1200                 /* text with \nnn octal quoting */
1201                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1202                         if (*bp == '\\' &&
1203                             isodigit(bp[1]) && (bp[1] <= '3') &&
1204                             isodigit(bp[2]) &&
1205                             isodigit(bp[3])) {
1206                                 int byte = (*++bp -'0');
1207                                 bp++;
1208                                 byte = (byte << 3) | (*bp++ - '0');
1209                                 byte = (byte << 3) | (*bp++ - '0');
1210                                 *dest++ = byte;
1211                                 len++;
1212                         } else {
1213                                 *dest++ = *bp++;
1214                                 len++;
1215                         }
1216                 }
1217         }
1218
1219         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1220                 return -1;
1221         while (*bp == ' ') bp++;
1222         *bpp = bp;
1223         *dest = '\0';
1224         return len;
1225 }
1226 EXPORT_SYMBOL_GPL(qword_get);
1227
1228
1229 /*
1230  * support /proc/sunrpc/cache/$CACHENAME/content
1231  * as a seqfile.
1232  * We call ->cache_show passing NULL for the item to
1233  * get a header, then pass each real item in the cache
1234  */
1235
1236 struct handle {
1237         struct cache_detail *cd;
1238 };
1239
1240 static void *c_start(struct seq_file *m, loff_t *pos)
1241         __acquires(cd->hash_lock)
1242 {
1243         loff_t n = *pos;
1244         unsigned hash, entry;
1245         struct cache_head *ch;
1246         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1247
1248
1249         read_lock(&cd->hash_lock);
1250         if (!n--)
1251                 return SEQ_START_TOKEN;
1252         hash = n >> 32;
1253         entry = n & ((1LL<<32) - 1);
1254
1255         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1256                 if (!entry--)
1257                         return ch;
1258         n &= ~((1LL<<32) - 1);
1259         do {
1260                 hash++;
1261                 n += 1LL<<32;
1262         } while(hash < cd->hash_size &&
1263                 cd->hash_table[hash]==NULL);
1264         if (hash >= cd->hash_size)
1265                 return NULL;
1266         *pos = n+1;
1267         return cd->hash_table[hash];
1268 }
1269
1270 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1271 {
1272         struct cache_head *ch = p;
1273         int hash = (*pos >> 32);
1274         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1275
1276         if (p == SEQ_START_TOKEN)
1277                 hash = 0;
1278         else if (ch->next == NULL) {
1279                 hash++;
1280                 *pos += 1LL<<32;
1281         } else {
1282                 ++*pos;
1283                 return ch->next;
1284         }
1285         *pos &= ~((1LL<<32) - 1);
1286         while (hash < cd->hash_size &&
1287                cd->hash_table[hash] == NULL) {
1288                 hash++;
1289                 *pos += 1LL<<32;
1290         }
1291         if (hash >= cd->hash_size)
1292                 return NULL;
1293         ++*pos;
1294         return cd->hash_table[hash];
1295 }
1296
1297 static void c_stop(struct seq_file *m, void *p)
1298         __releases(cd->hash_lock)
1299 {
1300         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1301         read_unlock(&cd->hash_lock);
1302 }
1303
1304 static int c_show(struct seq_file *m, void *p)
1305 {
1306         struct cache_head *cp = p;
1307         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1308
1309         if (p == SEQ_START_TOKEN)
1310                 return cd->cache_show(m, cd, NULL);
1311
1312         ifdebug(CACHE)
1313                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1314                            convert_to_wallclock(cp->expiry_time),
1315                            atomic_read(&cp->ref.refcount), cp->flags);
1316         cache_get(cp);
1317         if (cache_check(cd, cp, NULL))
1318                 /* cache_check does a cache_put on failure */
1319                 seq_printf(m, "# ");
1320         else
1321                 cache_put(cp, cd);
1322
1323         return cd->cache_show(m, cd, cp);
1324 }
1325
1326 static const struct seq_operations cache_content_op = {
1327         .start  = c_start,
1328         .next   = c_next,
1329         .stop   = c_stop,
1330         .show   = c_show,
1331 };
1332
1333 static int content_open(struct inode *inode, struct file *file,
1334                         struct cache_detail *cd)
1335 {
1336         struct handle *han;
1337
1338         if (!cd || !try_module_get(cd->owner))
1339                 return -EACCES;
1340         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1341         if (han == NULL) {
1342                 module_put(cd->owner);
1343                 return -ENOMEM;
1344         }
1345
1346         han->cd = cd;
1347         return 0;
1348 }
1349
1350 static int content_release(struct inode *inode, struct file *file,
1351                 struct cache_detail *cd)
1352 {
1353         int ret = seq_release_private(inode, file);
1354         module_put(cd->owner);
1355         return ret;
1356 }
1357
1358 static int open_flush(struct inode *inode, struct file *file,
1359                         struct cache_detail *cd)
1360 {
1361         if (!cd || !try_module_get(cd->owner))
1362                 return -EACCES;
1363         return nonseekable_open(inode, file);
1364 }
1365
1366 static int release_flush(struct inode *inode, struct file *file,
1367                         struct cache_detail *cd)
1368 {
1369         module_put(cd->owner);
1370         return 0;
1371 }
1372
1373 static ssize_t read_flush(struct file *file, char __user *buf,
1374                           size_t count, loff_t *ppos,
1375                           struct cache_detail *cd)
1376 {
1377         char tbuf[20];
1378         unsigned long p = *ppos;
1379         size_t len;
1380
1381         sprintf(tbuf, "%lu\n", convert_to_wallclock(cd->flush_time));
1382         len = strlen(tbuf);
1383         if (p >= len)
1384                 return 0;
1385         len -= p;
1386         if (len > count)
1387                 len = count;
1388         if (copy_to_user(buf, (void*)(tbuf+p), len))
1389                 return -EFAULT;
1390         *ppos += len;
1391         return len;
1392 }
1393
1394 static ssize_t write_flush(struct file *file, const char __user *buf,
1395                            size_t count, loff_t *ppos,
1396                            struct cache_detail *cd)
1397 {
1398         char tbuf[20];
1399         char *bp, *ep;
1400
1401         if (*ppos || count > sizeof(tbuf)-1)
1402                 return -EINVAL;
1403         if (copy_from_user(tbuf, buf, count))
1404                 return -EFAULT;
1405         tbuf[count] = 0;
1406         simple_strtoul(tbuf, &ep, 0);
1407         if (*ep && *ep != '\n')
1408                 return -EINVAL;
1409
1410         bp = tbuf;
1411         cd->flush_time = get_expiry(&bp);
1412         cd->nextcheck = seconds_since_boot();
1413         cache_flush();
1414
1415         *ppos += count;
1416         return count;
1417 }
1418
1419 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1420                                  size_t count, loff_t *ppos)
1421 {
1422         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1423
1424         return cache_read(filp, buf, count, ppos, cd);
1425 }
1426
1427 static ssize_t cache_write_procfs(struct file *filp, const 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_write(filp, buf, count, ppos, cd);
1433 }
1434
1435 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1436 {
1437         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1438
1439         return cache_poll(filp, wait, cd);
1440 }
1441
1442 static long cache_ioctl_procfs(struct file *filp,
1443                                unsigned int cmd, unsigned long arg)
1444 {
1445         long ret;
1446         struct inode *inode = filp->f_path.dentry->d_inode;
1447         struct cache_detail *cd = PDE(inode)->data;
1448
1449         lock_kernel();
1450         ret = cache_ioctl(inode, filp, cmd, arg, cd);
1451         unlock_kernel();
1452
1453         return ret;
1454 }
1455
1456 static int cache_open_procfs(struct inode *inode, struct file *filp)
1457 {
1458         struct cache_detail *cd = PDE(inode)->data;
1459
1460         return cache_open(inode, filp, cd);
1461 }
1462
1463 static int cache_release_procfs(struct inode *inode, struct file *filp)
1464 {
1465         struct cache_detail *cd = PDE(inode)->data;
1466
1467         return cache_release(inode, filp, cd);
1468 }
1469
1470 static const struct file_operations cache_file_operations_procfs = {
1471         .owner          = THIS_MODULE,
1472         .llseek         = no_llseek,
1473         .read           = cache_read_procfs,
1474         .write          = cache_write_procfs,
1475         .poll           = cache_poll_procfs,
1476         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1477         .open           = cache_open_procfs,
1478         .release        = cache_release_procfs,
1479 };
1480
1481 static int content_open_procfs(struct inode *inode, struct file *filp)
1482 {
1483         struct cache_detail *cd = PDE(inode)->data;
1484
1485         return content_open(inode, filp, cd);
1486 }
1487
1488 static int content_release_procfs(struct inode *inode, struct file *filp)
1489 {
1490         struct cache_detail *cd = PDE(inode)->data;
1491
1492         return content_release(inode, filp, cd);
1493 }
1494
1495 static const struct file_operations content_file_operations_procfs = {
1496         .open           = content_open_procfs,
1497         .read           = seq_read,
1498         .llseek         = seq_lseek,
1499         .release        = content_release_procfs,
1500 };
1501
1502 static int open_flush_procfs(struct inode *inode, struct file *filp)
1503 {
1504         struct cache_detail *cd = PDE(inode)->data;
1505
1506         return open_flush(inode, filp, cd);
1507 }
1508
1509 static int release_flush_procfs(struct inode *inode, struct file *filp)
1510 {
1511         struct cache_detail *cd = PDE(inode)->data;
1512
1513         return release_flush(inode, filp, cd);
1514 }
1515
1516 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1517                             size_t count, loff_t *ppos)
1518 {
1519         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1520
1521         return read_flush(filp, buf, count, ppos, cd);
1522 }
1523
1524 static ssize_t write_flush_procfs(struct file *filp,
1525                                   const char __user *buf,
1526                                   size_t count, loff_t *ppos)
1527 {
1528         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1529
1530         return write_flush(filp, buf, count, ppos, cd);
1531 }
1532
1533 static const struct file_operations cache_flush_operations_procfs = {
1534         .open           = open_flush_procfs,
1535         .read           = read_flush_procfs,
1536         .write          = write_flush_procfs,
1537         .release        = release_flush_procfs,
1538 };
1539
1540 static void remove_cache_proc_entries(struct cache_detail *cd)
1541 {
1542         if (cd->u.procfs.proc_ent == NULL)
1543                 return;
1544         if (cd->u.procfs.flush_ent)
1545                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1546         if (cd->u.procfs.channel_ent)
1547                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1548         if (cd->u.procfs.content_ent)
1549                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1550         cd->u.procfs.proc_ent = NULL;
1551         remove_proc_entry(cd->name, proc_net_rpc);
1552 }
1553
1554 #ifdef CONFIG_PROC_FS
1555 static int create_cache_proc_entries(struct cache_detail *cd)
1556 {
1557         struct proc_dir_entry *p;
1558
1559         cd->u.procfs.proc_ent = proc_mkdir(cd->name, proc_net_rpc);
1560         if (cd->u.procfs.proc_ent == NULL)
1561                 goto out_nomem;
1562         cd->u.procfs.channel_ent = NULL;
1563         cd->u.procfs.content_ent = NULL;
1564
1565         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1566                              cd->u.procfs.proc_ent,
1567                              &cache_flush_operations_procfs, cd);
1568         cd->u.procfs.flush_ent = p;
1569         if (p == NULL)
1570                 goto out_nomem;
1571
1572         if (cd->cache_upcall || cd->cache_parse) {
1573                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1574                                      cd->u.procfs.proc_ent,
1575                                      &cache_file_operations_procfs, cd);
1576                 cd->u.procfs.channel_ent = p;
1577                 if (p == NULL)
1578                         goto out_nomem;
1579         }
1580         if (cd->cache_show) {
1581                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1582                                 cd->u.procfs.proc_ent,
1583                                 &content_file_operations_procfs, cd);
1584                 cd->u.procfs.content_ent = p;
1585                 if (p == NULL)
1586                         goto out_nomem;
1587         }
1588         return 0;
1589 out_nomem:
1590         remove_cache_proc_entries(cd);
1591         return -ENOMEM;
1592 }
1593 #else /* CONFIG_PROC_FS */
1594 static int create_cache_proc_entries(struct cache_detail *cd)
1595 {
1596         return 0;
1597 }
1598 #endif
1599
1600 void __init cache_initialize(void)
1601 {
1602         INIT_DELAYED_WORK_DEFERRABLE(&cache_cleaner, do_cache_clean);
1603 }
1604
1605 int cache_register(struct cache_detail *cd)
1606 {
1607         int ret;
1608
1609         sunrpc_init_cache_detail(cd);
1610         ret = create_cache_proc_entries(cd);
1611         if (ret)
1612                 sunrpc_destroy_cache_detail(cd);
1613         return ret;
1614 }
1615 EXPORT_SYMBOL_GPL(cache_register);
1616
1617 void cache_unregister(struct cache_detail *cd)
1618 {
1619         remove_cache_proc_entries(cd);
1620         sunrpc_destroy_cache_detail(cd);
1621 }
1622 EXPORT_SYMBOL_GPL(cache_unregister);
1623
1624 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1625                                  size_t count, loff_t *ppos)
1626 {
1627         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1628
1629         return cache_read(filp, buf, count, ppos, cd);
1630 }
1631
1632 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1633                                   size_t count, loff_t *ppos)
1634 {
1635         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1636
1637         return cache_write(filp, buf, count, ppos, cd);
1638 }
1639
1640 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1641 {
1642         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1643
1644         return cache_poll(filp, wait, cd);
1645 }
1646
1647 static long cache_ioctl_pipefs(struct file *filp,
1648                               unsigned int cmd, unsigned long arg)
1649 {
1650         struct inode *inode = filp->f_dentry->d_inode;
1651         struct cache_detail *cd = RPC_I(inode)->private;
1652         long ret;
1653
1654         lock_kernel();
1655         ret = cache_ioctl(inode, filp, cmd, arg, cd);
1656         unlock_kernel();
1657
1658         return ret;
1659 }
1660
1661 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1662 {
1663         struct cache_detail *cd = RPC_I(inode)->private;
1664
1665         return cache_open(inode, filp, cd);
1666 }
1667
1668 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1669 {
1670         struct cache_detail *cd = RPC_I(inode)->private;
1671
1672         return cache_release(inode, filp, cd);
1673 }
1674
1675 const struct file_operations cache_file_operations_pipefs = {
1676         .owner          = THIS_MODULE,
1677         .llseek         = no_llseek,
1678         .read           = cache_read_pipefs,
1679         .write          = cache_write_pipefs,
1680         .poll           = cache_poll_pipefs,
1681         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1682         .open           = cache_open_pipefs,
1683         .release        = cache_release_pipefs,
1684 };
1685
1686 static int content_open_pipefs(struct inode *inode, struct file *filp)
1687 {
1688         struct cache_detail *cd = RPC_I(inode)->private;
1689
1690         return content_open(inode, filp, cd);
1691 }
1692
1693 static int content_release_pipefs(struct inode *inode, struct file *filp)
1694 {
1695         struct cache_detail *cd = RPC_I(inode)->private;
1696
1697         return content_release(inode, filp, cd);
1698 }
1699
1700 const struct file_operations content_file_operations_pipefs = {
1701         .open           = content_open_pipefs,
1702         .read           = seq_read,
1703         .llseek         = seq_lseek,
1704         .release        = content_release_pipefs,
1705 };
1706
1707 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1708 {
1709         struct cache_detail *cd = RPC_I(inode)->private;
1710
1711         return open_flush(inode, filp, cd);
1712 }
1713
1714 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1715 {
1716         struct cache_detail *cd = RPC_I(inode)->private;
1717
1718         return release_flush(inode, filp, cd);
1719 }
1720
1721 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1722                             size_t count, loff_t *ppos)
1723 {
1724         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1725
1726         return read_flush(filp, buf, count, ppos, cd);
1727 }
1728
1729 static ssize_t write_flush_pipefs(struct file *filp,
1730                                   const char __user *buf,
1731                                   size_t count, loff_t *ppos)
1732 {
1733         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1734
1735         return write_flush(filp, buf, count, ppos, cd);
1736 }
1737
1738 const struct file_operations cache_flush_operations_pipefs = {
1739         .open           = open_flush_pipefs,
1740         .read           = read_flush_pipefs,
1741         .write          = write_flush_pipefs,
1742         .release        = release_flush_pipefs,
1743 };
1744
1745 int sunrpc_cache_register_pipefs(struct dentry *parent,
1746                                  const char *name, mode_t umode,
1747                                  struct cache_detail *cd)
1748 {
1749         struct qstr q;
1750         struct dentry *dir;
1751         int ret = 0;
1752
1753         sunrpc_init_cache_detail(cd);
1754         q.name = name;
1755         q.len = strlen(name);
1756         q.hash = full_name_hash(q.name, q.len);
1757         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1758         if (!IS_ERR(dir))
1759                 cd->u.pipefs.dir = dir;
1760         else {
1761                 sunrpc_destroy_cache_detail(cd);
1762                 ret = PTR_ERR(dir);
1763         }
1764         return ret;
1765 }
1766 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1767
1768 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1769 {
1770         rpc_remove_cache_dir(cd->u.pipefs.dir);
1771         cd->u.pipefs.dir = NULL;
1772         sunrpc_destroy_cache_detail(cd);
1773 }
1774 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1775