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 list_head cache_defer_hash[DFR_HASHSIZE];
 510 static int cache_defer_cnt;
 511
 512 struct thread_deferred_req {
 513         struct cache_deferred_req handle;
 514         struct completion completion;
 515 };
 516 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 517 {
 518         struct thread_deferred_req *dr =
 519                 container_of(dreq, struct thread_deferred_req, handle);
 520         complete(&dr->completion);
 521 }
 522
 523 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 524 {
 525         list_del_init(&dreq->recent);
 526         list_del_init(&dreq->hash);
 527         cache_defer_cnt--;
 528 }
 529
 530 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 531 {
 532         int hash = DFR_HASH(item);
 533
 534         list_add(&dreq->recent, &cache_defer_list);
 535         if (cache_defer_hash[hash].next == NULL)
 536                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 537         list_add(&dreq->hash, &cache_defer_hash[hash]);
 538 }
 539
 540 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 541 {
 542         struct cache_deferred_req *dreq, *discard;
 543         struct thread_deferred_req sleeper;
 544
 545         if (cache_defer_cnt >= DFR_MAX) {
 546                 /* too much in the cache, randomly drop this one,
 547                  * or continue and drop the oldest below
 548                  */
 549                 if (net_random()&1)
 550                         return -ENOMEM;
 551         }
 552         if (req->thread_wait) {
 553                 dreq = &sleeper.handle;
 554                 sleeper.completion =
 555                         COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 556                 dreq->revisit = cache_restart_thread;
 557         } else
 558                 dreq = req->defer(req);
 559
 560  retry:
 561         if (dreq == NULL)
 562                 return -ENOMEM;
 563
 564         dreq->item = item;
 565
 566         spin_lock(&cache_defer_lock);
 567
 568         __hash_deferred_req(dreq, item);
 569
 570         /* it is in, now maybe clean up */
 571         discard = NULL;
 572         if (++cache_defer_cnt > DFR_MAX) {
 573                 discard = list_entry(cache_defer_list.prev,
 574                                      struct cache_deferred_req, recent);
 575                 __unhash_deferred_req(discard);
 576         }
 577         spin_unlock(&cache_defer_lock);
 578
 579         if (discard)
 580                 /* there was one too many */
 581                 discard->revisit(discard, 1);
 582
 583         if (!test_bit(CACHE_PENDING, &item->flags)) {
 584                 /* must have just been validated... */
 585                 cache_revisit_request(item);
 586                 return -EAGAIN;
 587         }
 588
 589         if (dreq == &sleeper.handle) {
 590                 if (wait_for_completion_interruptible_timeout(
 591                             &sleeper.completion, req->thread_wait) <= 0) {
 592                         /* The completion wasn't completed, so we need
 593                          * to clean up
 594                          */
 595                         spin_lock(&cache_defer_lock);
 596                         if (!list_empty(&sleeper.handle.hash)) {
 597                                 __unhash_deferred_req(&sleeper.handle);
 598                                 spin_unlock(&cache_defer_lock);
 599                         } else {
 600                                 /* cache_revisit_request already removed
 601                                  * this from the hash table, but hasn't
 602                                  * called ->revisit yet.  It will very soon
 603                                  * and we need to wait for it.
 604                                  */
 605                                 spin_unlock(&cache_defer_lock);
 606                                 wait_for_completion(&sleeper.completion);
 607                         }
 608                 }
 609                 if (test_bit(CACHE_PENDING, &item->flags)) {
 610                         /* item is still pending, try request
 611                          * deferral
 612                          */
 613                         dreq = req->defer(req);
 614                         goto retry;
 615                 }
 616                 /* only return success if we actually deferred the
 617                  * request.  In this case we waited until it was
 618                  * answered so no deferral has happened - rather
 619                  * an answer already exists.
 620                  */
 621                 return -EEXIST;
 622         }
 623         return 0;
 624 }
 625
 626 static void cache_revisit_request(struct cache_head *item)
 627 {
 628         struct cache_deferred_req *dreq;
 629         struct list_head pending;
 630
 631         struct list_head *lp;
 632         int hash = DFR_HASH(item);
 633
 634         INIT_LIST_HEAD(&pending);
 635         spin_lock(&cache_defer_lock);
 636
 637         lp = cache_defer_hash[hash].next;
 638         if (lp) {
 639                 while (lp != &cache_defer_hash[hash]) {
 640                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 641                         lp = lp->next;
 642                         if (dreq->item == item) {
 643                                 __unhash_deferred_req(dreq);
 644                                 list_add(&dreq->recent, &pending);
 645                         }
 646                 }
 647         }
 648         spin_unlock(&cache_defer_lock);
 649
 650         while (!list_empty(&pending)) {
 651                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 652                 list_del_init(&dreq->recent);
 653                 dreq->revisit(dreq, 0);
 654         }
 655 }
 656
 657 void cache_clean_deferred(void *owner)
 658 {
 659         struct cache_deferred_req *dreq, *tmp;
 660         struct list_head pending;
 661
 662
 663         INIT_LIST_HEAD(&pending);
 664         spin_lock(&cache_defer_lock);
 665
 666         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 667                 if (dreq->owner == owner)
 668                         __unhash_deferred_req(dreq);
 669         }
 670         spin_unlock(&cache_defer_lock);
 671
 672         while (!list_empty(&pending)) {
 673                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 674                 list_del_init(&dreq->recent);
 675                 dreq->revisit(dreq, 1);
 676         }
 677 }
 678
 679 /*
 680  * communicate with user-space
 681  *
 682  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 683  * On read, you get a full request, or block.
 684  * On write, an update request is processed.
 685  * Poll works if anything to read, and always allows write.
 686  *
 687  * Implemented by linked list of requests.  Each open file has
 688  * a ->private that also exists in this list.  New requests are added
 689  * to the end and may wakeup and preceding readers.
 690  * New readers are added to the head.  If, on read, an item is found with
 691  * CACHE_UPCALLING clear, we free it from the list.
 692  *
 693  */
 694
 695 static DEFINE_SPINLOCK(queue_lock);
 696 static DEFINE_MUTEX(queue_io_mutex);
 697
 698 struct cache_queue {
 699         struct list_head        list;
 700         int                     reader; /* if 0, then request */
 701 };
 702 struct cache_request {
 703         struct cache_queue      q;
 704         struct cache_head       *item;
 705         char                    * buf;
 706         int                     len;
 707         int                     readers;
 708 };
 709 struct cache_reader {
 710         struct cache_queue      q;
 711         int                     offset; /* if non-0, we have a refcnt on next request */
 712 };
 713
 714 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 715                           loff_t *ppos, struct cache_detail *cd)
 716 {
 717         struct cache_reader *rp = filp->private_data;
 718         struct cache_request *rq;
 719         struct inode *inode = filp->f_path.dentry->d_inode;
 720         int err;
 721
 722         if (count == 0)
 723                 return 0;
 724
 725         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 726                               * readers on this file */
 727  again:
 728         spin_lock(&queue_lock);
 729         /* need to find next request */
 730         while (rp->q.list.next != &cd->queue &&
 731                list_entry(rp->q.list.next, struct cache_queue, list)
 732                ->reader) {
 733                 struct list_head *next = rp->q.list.next;
 734                 list_move(&rp->q.list, next);
 735         }
 736         if (rp->q.list.next == &cd->queue) {
 737                 spin_unlock(&queue_lock);
 738                 mutex_unlock(&inode->i_mutex);
 739                 BUG_ON(rp->offset);
 740                 return 0;
 741         }
 742         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 743         BUG_ON(rq->q.reader);
 744         if (rp->offset == 0)
 745                 rq->readers++;
 746         spin_unlock(&queue_lock);
 747
 748         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 749                 err = -EAGAIN;
 750                 spin_lock(&queue_lock);
 751                 list_move(&rp->q.list, &rq->q.list);
 752                 spin_unlock(&queue_lock);
 753         } else {
 754                 if (rp->offset + count > rq->len)
 755                         count = rq->len - rp->offset;
 756                 err = -EFAULT;
 757                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 758                         goto out;
 759                 rp->offset += count;
 760                 if (rp->offset >= rq->len) {
 761                         rp->offset = 0;
 762                         spin_lock(&queue_lock);
 763                         list_move(&rp->q.list, &rq->q.list);
 764                         spin_unlock(&queue_lock);
 765                 }
 766                 err = 0;
 767         }
 768  out:
 769         if (rp->offset == 0) {
 770                 /* need to release rq */
 771                 spin_lock(&queue_lock);
 772                 rq->readers--;
 773                 if (rq->readers == 0 &&
 774                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 775                         list_del(&rq->q.list);
 776                         spin_unlock(&queue_lock);
 777                         cache_put(rq->item, cd);
 778                         kfree(rq->buf);
 779                         kfree(rq);
 780                 } else
 781                         spin_unlock(&queue_lock);
 782         }
 783         if (err == -EAGAIN)
 784                 goto again;
 785         mutex_unlock(&inode->i_mutex);
 786         return err ? err :  count;
 787 }
 788
 789 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 790                                  size_t count, struct cache_detail *cd)
 791 {
 792         ssize_t ret;
 793
 794         if (copy_from_user(kaddr, buf, count))
 795                 return -EFAULT;
 796         kaddr[count] = '\0';
 797         ret = cd->cache_parse(cd, kaddr, count);
 798         if (!ret)
 799                 ret = count;
 800         return ret;
 801 }
 802
 803 static ssize_t cache_slow_downcall(const char __user *buf,
 804                                    size_t count, struct cache_detail *cd)
 805 {
 806         static char write_buf[8192]; /* protected by queue_io_mutex */
 807         ssize_t ret = -EINVAL;
 808
 809         if (count >= sizeof(write_buf))
 810                 goto out;
 811         mutex_lock(&queue_io_mutex);
 812         ret = cache_do_downcall(write_buf, buf, count, cd);
 813         mutex_unlock(&queue_io_mutex);
 814 out:
 815         return ret;
 816 }
 817
 818 static ssize_t cache_downcall(struct address_space *mapping,
 819                               const char __user *buf,
 820                               size_t count, struct cache_detail *cd)
 821 {
 822         struct page *page;
 823         char *kaddr;
 824         ssize_t ret = -ENOMEM;
 825
 826         if (count >= PAGE_CACHE_SIZE)
 827                 goto out_slow;
 828
 829         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 830         if (!page)
 831                 goto out_slow;
 832
 833         kaddr = kmap(page);
 834         ret = cache_do_downcall(kaddr, buf, count, cd);
 835         kunmap(page);
 836         unlock_page(page);
 837         page_cache_release(page);
 838         return ret;
 839 out_slow:
 840         return cache_slow_downcall(buf, count, cd);
 841 }
 842
 843 static ssize_t cache_write(struct file *filp, const char __user *buf,
 844                            size_t count, loff_t *ppos,
 845                            struct cache_detail *cd)
 846 {
 847         struct address_space *mapping = filp->f_mapping;
 848         struct inode *inode = filp->f_path.dentry->d_inode;
 849         ssize_t ret = -EINVAL;
 850
 851         if (!cd->cache_parse)
 852                 goto out;
 853
 854         mutex_lock(&inode->i_mutex);
 855         ret = cache_downcall(mapping, buf, count, cd);
 856         mutex_unlock(&inode->i_mutex);
 857 out:
 858         return ret;
 859 }
 860
 861 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 862
 863 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 864                                struct cache_detail *cd)
 865 {
 866         unsigned int mask;
 867         struct cache_reader *rp = filp->private_data;
 868         struct cache_queue *cq;
 869
 870         poll_wait(filp, &queue_wait, wait);
 871
 872         /* alway allow write */
 873         mask = POLL_OUT | POLLWRNORM;
 874
 875         if (!rp)
 876                 return mask;
 877
 878         spin_lock(&queue_lock);
 879
 880         for (cq= &rp->q; &cq->list != &cd->queue;
 881              cq = list_entry(cq->list.next, struct cache_queue, list))
 882                 if (!cq->reader) {
 883                         mask |= POLLIN | POLLRDNORM;
 884                         break;
 885                 }
 886         spin_unlock(&queue_lock);
 887         return mask;
 888 }
 889
 890 static int cache_ioctl(struct inode *ino, struct file *filp,
 891                        unsigned int cmd, unsigned long arg,
 892                        struct cache_detail *cd)
 893 {
 894         int len = 0;
 895         struct cache_reader *rp = filp->private_data;
 896         struct cache_queue *cq;
 897
 898         if (cmd != FIONREAD || !rp)
 899                 return -EINVAL;
 900
 901         spin_lock(&queue_lock);
 902
 903         /* only find the length remaining in current request,
 904          * or the length of the next request
 905          */
 906         for (cq= &rp->q; &cq->list != &cd->queue;
 907              cq = list_entry(cq->list.next, struct cache_queue, list))
 908                 if (!cq->reader) {
 909                         struct cache_request *cr =
 910                                 container_of(cq, struct cache_request, q);
 911                         len = cr->len - rp->offset;
 912                         break;
 913                 }
 914         spin_unlock(&queue_lock);
 915
 916         return put_user(len, (int __user *)arg);
 917 }
 918
 919 static int cache_open(struct inode *inode, struct file *filp,
 920                       struct cache_detail *cd)
 921 {
 922         struct cache_reader *rp = NULL;
 923
 924         if (!cd || !try_module_get(cd->owner))
 925                 return -EACCES;
 926         nonseekable_open(inode, filp);
 927         if (filp->f_mode & FMODE_READ) {
 928                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 929                 if (!rp)
 930                         return -ENOMEM;
 931                 rp->offset = 0;
 932                 rp->q.reader = 1;
 933                 atomic_inc(&cd->readers);
 934                 spin_lock(&queue_lock);
 935                 list_add(&rp->q.list, &cd->queue);
 936                 spin_unlock(&queue_lock);
 937         }
 938         filp->private_data = rp;
 939         return 0;
 940 }
 941
 942 static int cache_release(struct inode *inode, struct file *filp,
 943                          struct cache_detail *cd)
 944 {
 945         struct cache_reader *rp = filp->private_data;
 946
 947         if (rp) {
 948                 spin_lock(&queue_lock);
 949                 if (rp->offset) {
 950                         struct cache_queue *cq;
 951                         for (cq= &rp->q; &cq->list != &cd->queue;
 952                              cq = list_entry(cq->list.next, struct cache_queue, list))
 953                                 if (!cq->reader) {
 954                                         container_of(cq, struct cache_request, q)
 955                                                 ->readers--;
 956                                         break;
 957                                 }
 958                         rp->offset = 0;
 959                 }
 960                 list_del(&rp->q.list);
 961                 spin_unlock(&queue_lock);
 962
 963                 filp->private_data = NULL;
 964                 kfree(rp);
 965
 966                 cd->last_close = seconds_since_boot();
 967                 atomic_dec(&cd->readers);
 968         }
 969         module_put(cd->owner);
 970         return 0;
 971 }
 972
 973
 974
 975 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
 976 {
 977         struct cache_queue *cq;
 978         spin_lock(&queue_lock);
 979         list_for_each_entry(cq, &detail->queue, list)
 980                 if (!cq->reader) {
 981                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 982                         if (cr->item != ch)
 983                                 continue;
 984                         if (cr->readers != 0)
 985                                 continue;
 986                         list_del(&cr->q.list);
 987                         spin_unlock(&queue_lock);
 988                         cache_put(cr->item, detail);
 989                         kfree(cr->buf);
 990                         kfree(cr);
 991                         return;
 992                 }
 993         spin_unlock(&queue_lock);
 994 }
 995
 996 /*
 997  * Support routines for text-based upcalls.
 998  * Fields are separated by spaces.
 999  * Fields are either mangled to quote space tab newline slosh with slosh
1000  * or a hexified with a leading \x
1001  * Record is terminated with newline.
1002  *
1003  */
1004
1005 void qword_add(char **bpp, int *lp, char *str)
1006 {
1007         char *bp = *bpp;
1008         int len = *lp;
1009         char c;
1010
1011         if (len < 0) return;
1012
1013         while ((c=*str++) && len)
1014                 switch(c) {
1015                 case ' ':
1016                 case '\t':
1017                 case '\n':
1018                 case '\\':
1019                         if (len >= 4) {
1020                                 *bp++ = '\\';
1021                                 *bp++ = '0' + ((c & 0300)>>6);
1022                                 *bp++ = '0' + ((c & 0070)>>3);
1023                                 *bp++ = '0' + ((c & 0007)>>0);
1024                         }
1025                         len -= 4;
1026                         break;
1027                 default:
1028                         *bp++ = c;
1029                         len--;
1030                 }
1031         if (c || len <1) len = -1;
1032         else {
1033                 *bp++ = ' ';
1034                 len--;
1035         }
1036         *bpp = bp;
1037         *lp = len;
1038 }
1039 EXPORT_SYMBOL_GPL(qword_add);
1040
1041 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1042 {
1043         char *bp = *bpp;
1044         int len = *lp;
1045
1046         if (len < 0) return;
1047
1048         if (len > 2) {
1049                 *bp++ = '\\';
1050                 *bp++ = 'x';
1051                 len -= 2;
1052                 while (blen && len >= 2) {
1053                         unsigned char c = *buf++;
1054                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1055                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1056                         len -= 2;
1057                         blen--;
1058                 }
1059         }
1060         if (blen || len<1) len = -1;
1061         else {
1062                 *bp++ = ' ';
1063                 len--;
1064         }
1065         *bpp = bp;
1066         *lp = len;
1067 }
1068 EXPORT_SYMBOL_GPL(qword_addhex);
1069
1070 static void warn_no_listener(struct cache_detail *detail)
1071 {
1072         if (detail->last_warn != detail->last_close) {
1073                 detail->last_warn = detail->last_close;
1074                 if (detail->warn_no_listener)
1075                         detail->warn_no_listener(detail, detail->last_close != 0);
1076         }
1077 }
1078
1079 /*
1080  * register an upcall request to user-space and queue it up for read() by the
1081  * upcall daemon.
1082  *
1083  * Each request is at most one page long.
1084  */
1085 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1086                 void (*cache_request)(struct cache_detail *,
1087                                       struct cache_head *,
1088                                       char **,
1089                                       int *))
1090 {
1091
1092         char *buf;
1093         struct cache_request *crq;
1094         char *bp;
1095         int len;
1096
1097         if (atomic_read(&detail->readers) == 0 &&
1098             detail->last_close < seconds_since_boot() - 30) {
1099                         warn_no_listener(detail);
1100                         return -EINVAL;
1101         }
1102
1103         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1104         if (!buf)
1105                 return -EAGAIN;
1106
1107         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1108         if (!crq) {
1109                 kfree(buf);
1110                 return -EAGAIN;
1111         }
1112
1113         bp = buf; len = PAGE_SIZE;
1114
1115         cache_request(detail, h, &bp, &len);
1116
1117         if (len < 0) {
1118                 kfree(buf);
1119                 kfree(crq);
1120                 return -EAGAIN;
1121         }
1122         crq->q.reader = 0;
1123         crq->item = cache_get(h);
1124         crq->buf = buf;
1125         crq->len = PAGE_SIZE - len;
1126         crq->readers = 0;
1127         spin_lock(&queue_lock);
1128         list_add_tail(&crq->q.list, &detail->queue);
1129         spin_unlock(&queue_lock);
1130         wake_up(&queue_wait);
1131         return 0;
1132 }
1133 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1134
1135 /*
1136  * parse a message from user-space and pass it
1137  * to an appropriate cache
1138  * Messages are, like requests, separated into fields by
1139  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1140  *
1141  * Message is
1142  *   reply cachename expiry key ... content....
1143  *
1144  * key and content are both parsed by cache
1145  */
1146
1147 #define isodigit(c) (isdigit(c) && c <= '7')
1148 int qword_get(char **bpp, char *dest, int bufsize)
1149 {
1150         /* return bytes copied, or -1 on error */
1151         char *bp = *bpp;
1152         int len = 0;
1153
1154         while (*bp == ' ') bp++;
1155
1156         if (bp[0] == '\\' && bp[1] == 'x') {
1157                 /* HEX STRING */
1158                 bp += 2;
1159                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1160                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1161                         bp++;
1162                         byte <<= 4;
1163                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1164                         *dest++ = byte;
1165                         bp++;
1166                         len++;
1167                 }
1168         } else {
1169                 /* text with \nnn octal quoting */
1170                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1171                         if (*bp == '\\' &&
1172                             isodigit(bp[1]) && (bp[1] <= '3') &&
1173                             isodigit(bp[2]) &&
1174                             isodigit(bp[3])) {
1175                                 int byte = (*++bp -'0');
1176                                 bp++;
1177                                 byte = (byte << 3) | (*bp++ - '0');
1178                                 byte = (byte << 3) | (*bp++ - '0');
1179                                 *dest++ = byte;
1180                                 len++;
1181                         } else {
1182                                 *dest++ = *bp++;
1183                                 len++;
1184                         }
1185                 }
1186         }
1187
1188         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1189                 return -1;
1190         while (*bp == ' ') bp++;
1191         *bpp = bp;
1192         *dest = '\0';
1193         return len;
1194 }
1195 EXPORT_SYMBOL_GPL(qword_get);
1196
1197
1198 /*
1199  * support /proc/sunrpc/cache/$CACHENAME/content
1200  * as a seqfile.
1201  * We call ->cache_show passing NULL for the item to
1202  * get a header, then pass each real item in the cache
1203  */
1204
1205 struct handle {
1206         struct cache_detail *cd;
1207 };
1208
1209 static void *c_start(struct seq_file *m, loff_t *pos)
1210         __acquires(cd->hash_lock)
1211 {
1212         loff_t n = *pos;
1213         unsigned hash, entry;
1214         struct cache_head *ch;
1215         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1216
1217
1218         read_lock(&cd->hash_lock);
1219         if (!n--)
1220                 return SEQ_START_TOKEN;
1221         hash = n >> 32;
1222         entry = n & ((1LL<<32) - 1);
1223
1224         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1225                 if (!entry--)
1226                         return ch;
1227         n &= ~((1LL<<32) - 1);
1228         do {
1229                 hash++;
1230                 n += 1LL<<32;
1231         } while(hash < cd->hash_size &&
1232                 cd->hash_table[hash]==NULL);
1233         if (hash >= cd->hash_size)
1234                 return NULL;
1235         *pos = n+1;
1236         return cd->hash_table[hash];
1237 }
1238
1239 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1240 {
1241         struct cache_head *ch = p;
1242         int hash = (*pos >> 32);
1243         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1244
1245         if (p == SEQ_START_TOKEN)
1246                 hash = 0;
1247         else if (ch->next == NULL) {
1248                 hash++;
1249                 *pos += 1LL<<32;
1250         } else {
1251                 ++*pos;
1252                 return ch->next;
1253         }
1254         *pos &= ~((1LL<<32) - 1);
1255         while (hash < cd->hash_size &&
1256                cd->hash_table[hash] == NULL) {
1257                 hash++;
1258                 *pos += 1LL<<32;
1259         }
1260         if (hash >= cd->hash_size)
1261                 return NULL;
1262         ++*pos;
1263         return cd->hash_table[hash];
1264 }
1265
1266 static void c_stop(struct seq_file *m, void *p)
1267         __releases(cd->hash_lock)
1268 {
1269         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1270         read_unlock(&cd->hash_lock);
1271 }
1272
1273 static int c_show(struct seq_file *m, void *p)
1274 {
1275         struct cache_head *cp = p;
1276         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1277
1278         if (p == SEQ_START_TOKEN)
1279                 return cd->cache_show(m, cd, NULL);
1280
1281         ifdebug(CACHE)
1282                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1283                            convert_to_wallclock(cp->expiry_time),
1284                            atomic_read(&cp->ref.refcount), cp->flags);
1285         cache_get(cp);
1286         if (cache_check(cd, cp, NULL))
1287                 /* cache_check does a cache_put on failure */
1288                 seq_printf(m, "# ");
1289         else
1290                 cache_put(cp, cd);
1291
1292         return cd->cache_show(m, cd, cp);
1293 }
1294
1295 static const struct seq_operations cache_content_op = {
1296         .start  = c_start,
1297         .next   = c_next,
1298         .stop   = c_stop,
1299         .show   = c_show,
1300 };
1301
1302 static int content_open(struct inode *inode, struct file *file,
1303                         struct cache_detail *cd)
1304 {
1305         struct handle *han;
1306
1307         if (!cd || !try_module_get(cd->owner))
1308                 return -EACCES;
1309         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1310         if (han == NULL) {
1311                 module_put(cd->owner);
1312                 return -ENOMEM;
1313         }
1314
1315         han->cd = cd;
1316         return 0;
1317 }
1318
1319 static int content_release(struct inode *inode, struct file *file,
1320                 struct cache_detail *cd)
1321 {
1322         int ret = seq_release_private(inode, file);
1323         module_put(cd->owner);
1324         return ret;
1325 }
1326
1327 static int open_flush(struct inode *inode, struct file *file,
1328                         struct cache_detail *cd)
1329 {
1330         if (!cd || !try_module_get(cd->owner))
1331                 return -EACCES;
1332         return nonseekable_open(inode, file);
1333 }
1334
1335 static int release_flush(struct inode *inode, struct file *file,
1336                         struct cache_detail *cd)
1337 {
1338         module_put(cd->owner);
1339         return 0;
1340 }
1341
1342 static ssize_t read_flush(struct file *file, char __user *buf,
1343                           size_t count, loff_t *ppos,
1344                           struct cache_detail *cd)
1345 {
1346         char tbuf[20];
1347         unsigned long p = *ppos;
1348         size_t len;
1349
1350         sprintf(tbuf, "%lu\n", convert_to_wallclock(cd->flush_time));
1351         len = strlen(tbuf);
1352         if (p >= len)
1353                 return 0;
1354         len -= p;
1355         if (len > count)
1356                 len = count;
1357         if (copy_to_user(buf, (void*)(tbuf+p), len))
1358                 return -EFAULT;
1359         *ppos += len;
1360         return len;
1361 }
1362
1363 static ssize_t write_flush(struct file *file, const char __user *buf,
1364                            size_t count, loff_t *ppos,
1365                            struct cache_detail *cd)
1366 {
1367         char tbuf[20];
1368         char *bp, *ep;
1369
1370         if (*ppos || count > sizeof(tbuf)-1)
1371                 return -EINVAL;
1372         if (copy_from_user(tbuf, buf, count))
1373                 return -EFAULT;
1374         tbuf[count] = 0;
1375         simple_strtoul(tbuf, &ep, 0);
1376         if (*ep && *ep != '\n')
1377                 return -EINVAL;
1378
1379         bp = tbuf;
1380         cd->flush_time = get_expiry(&bp);
1381         cd->nextcheck = seconds_since_boot();
1382         cache_flush();
1383
1384         *ppos += count;
1385         return count;
1386 }
1387
1388 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1389                                  size_t count, loff_t *ppos)
1390 {
1391         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1392
1393         return cache_read(filp, buf, count, ppos, cd);
1394 }
1395
1396 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1397                                   size_t count, loff_t *ppos)
1398 {
1399         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1400
1401         return cache_write(filp, buf, count, ppos, cd);
1402 }
1403
1404 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1405 {
1406         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1407
1408         return cache_poll(filp, wait, cd);
1409 }
1410
1411 static long cache_ioctl_procfs(struct file *filp,
1412                                unsigned int cmd, unsigned long arg)
1413 {
1414         long ret;
1415         struct inode *inode = filp->f_path.dentry->d_inode;
1416         struct cache_detail *cd = PDE(inode)->data;
1417
1418         lock_kernel();
1419         ret = cache_ioctl(inode, filp, cmd, arg, cd);
1420         unlock_kernel();
1421
1422         return ret;
1423 }
1424
1425 static int cache_open_procfs(struct inode *inode, struct file *filp)
1426 {
1427         struct cache_detail *cd = PDE(inode)->data;
1428
1429         return cache_open(inode, filp, cd);
1430 }
1431
1432 static int cache_release_procfs(struct inode *inode, struct file *filp)
1433 {
1434         struct cache_detail *cd = PDE(inode)->data;
1435
1436         return cache_release(inode, filp, cd);
1437 }
1438
1439 static const struct file_operations cache_file_operations_procfs = {
1440         .owner          = THIS_MODULE,
1441         .llseek         = no_llseek,
1442         .read           = cache_read_procfs,
1443         .write          = cache_write_procfs,
1444         .poll           = cache_poll_procfs,
1445         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1446         .open           = cache_open_procfs,
1447         .release        = cache_release_procfs,
1448 };
1449
1450 static int content_open_procfs(struct inode *inode, struct file *filp)
1451 {
1452         struct cache_detail *cd = PDE(inode)->data;
1453
1454         return content_open(inode, filp, cd);
1455 }
1456
1457 static int content_release_procfs(struct inode *inode, struct file *filp)
1458 {
1459         struct cache_detail *cd = PDE(inode)->data;
1460
1461         return content_release(inode, filp, cd);
1462 }
1463
1464 static const struct file_operations content_file_operations_procfs = {
1465         .open           = content_open_procfs,
1466         .read           = seq_read,
1467         .llseek         = seq_lseek,
1468         .release        = content_release_procfs,
1469 };
1470
1471 static int open_flush_procfs(struct inode *inode, struct file *filp)
1472 {
1473         struct cache_detail *cd = PDE(inode)->data;
1474
1475         return open_flush(inode, filp, cd);
1476 }
1477
1478 static int release_flush_procfs(struct inode *inode, struct file *filp)
1479 {
1480         struct cache_detail *cd = PDE(inode)->data;
1481
1482         return release_flush(inode, filp, cd);
1483 }
1484
1485 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1486                             size_t count, loff_t *ppos)
1487 {
1488         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1489
1490         return read_flush(filp, buf, count, ppos, cd);
1491 }
1492
1493 static ssize_t write_flush_procfs(struct file *filp,
1494                                   const char __user *buf,
1495                                   size_t count, loff_t *ppos)
1496 {
1497         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1498
1499         return write_flush(filp, buf, count, ppos, cd);
1500 }
1501
1502 static const struct file_operations cache_flush_operations_procfs = {
1503         .open           = open_flush_procfs,
1504         .read           = read_flush_procfs,
1505         .write          = write_flush_procfs,
1506         .release        = release_flush_procfs,
1507 };
1508
1509 static void remove_cache_proc_entries(struct cache_detail *cd)
1510 {
1511         if (cd->u.procfs.proc_ent == NULL)
1512                 return;
1513         if (cd->u.procfs.flush_ent)
1514                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1515         if (cd->u.procfs.channel_ent)
1516                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1517         if (cd->u.procfs.content_ent)
1518                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1519         cd->u.procfs.proc_ent = NULL;
1520         remove_proc_entry(cd->name, proc_net_rpc);
1521 }
1522
1523 #ifdef CONFIG_PROC_FS
1524 static int create_cache_proc_entries(struct cache_detail *cd)
1525 {
1526         struct proc_dir_entry *p;
1527
1528         cd->u.procfs.proc_ent = proc_mkdir(cd->name, proc_net_rpc);
1529         if (cd->u.procfs.proc_ent == NULL)
1530                 goto out_nomem;
1531         cd->u.procfs.channel_ent = NULL;
1532         cd->u.procfs.content_ent = NULL;
1533
1534         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1535                              cd->u.procfs.proc_ent,
1536                              &cache_flush_operations_procfs, cd);
1537         cd->u.procfs.flush_ent = p;
1538         if (p == NULL)
1539                 goto out_nomem;
1540
1541         if (cd->cache_upcall || cd->cache_parse) {
1542                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1543                                      cd->u.procfs.proc_ent,
1544                                      &cache_file_operations_procfs, cd);
1545                 cd->u.procfs.channel_ent = p;
1546                 if (p == NULL)
1547                         goto out_nomem;
1548         }
1549         if (cd->cache_show) {
1550                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1551                                 cd->u.procfs.proc_ent,
1552                                 &content_file_operations_procfs, cd);
1553                 cd->u.procfs.content_ent = p;
1554                 if (p == NULL)
1555                         goto out_nomem;
1556         }
1557         return 0;
1558 out_nomem:
1559         remove_cache_proc_entries(cd);
1560         return -ENOMEM;
1561 }
1562 #else /* CONFIG_PROC_FS */
1563 static int create_cache_proc_entries(struct cache_detail *cd)
1564 {
1565         return 0;
1566 }
1567 #endif
1568
1569 void __init cache_initialize(void)
1570 {
1571         INIT_DELAYED_WORK_DEFERRABLE(&cache_cleaner, do_cache_clean);
1572 }
1573
1574 int cache_register(struct cache_detail *cd)
1575 {
1576         int ret;
1577
1578         sunrpc_init_cache_detail(cd);
1579         ret = create_cache_proc_entries(cd);
1580         if (ret)
1581                 sunrpc_destroy_cache_detail(cd);
1582         return ret;
1583 }
1584 EXPORT_SYMBOL_GPL(cache_register);
1585
1586 void cache_unregister(struct cache_detail *cd)
1587 {
1588         remove_cache_proc_entries(cd);
1589         sunrpc_destroy_cache_detail(cd);
1590 }
1591 EXPORT_SYMBOL_GPL(cache_unregister);
1592
1593 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1594                                  size_t count, loff_t *ppos)
1595 {
1596         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1597
1598         return cache_read(filp, buf, count, ppos, cd);
1599 }
1600
1601 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1602                                   size_t count, loff_t *ppos)
1603 {
1604         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1605
1606         return cache_write(filp, buf, count, ppos, cd);
1607 }
1608
1609 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1610 {
1611         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1612
1613         return cache_poll(filp, wait, cd);
1614 }
1615
1616 static long cache_ioctl_pipefs(struct file *filp,
1617                               unsigned int cmd, unsigned long arg)
1618 {
1619         struct inode *inode = filp->f_dentry->d_inode;
1620         struct cache_detail *cd = RPC_I(inode)->private;
1621         long ret;
1622
1623         lock_kernel();
1624         ret = cache_ioctl(inode, filp, cmd, arg, cd);
1625         unlock_kernel();
1626
1627         return ret;
1628 }
1629
1630 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1631 {
1632         struct cache_detail *cd = RPC_I(inode)->private;
1633
1634         return cache_open(inode, filp, cd);
1635 }
1636
1637 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1638 {
1639         struct cache_detail *cd = RPC_I(inode)->private;
1640
1641         return cache_release(inode, filp, cd);
1642 }
1643
1644 const struct file_operations cache_file_operations_pipefs = {
1645         .owner          = THIS_MODULE,
1646         .llseek         = no_llseek,
1647         .read           = cache_read_pipefs,
1648         .write          = cache_write_pipefs,
1649         .poll           = cache_poll_pipefs,
1650         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1651         .open           = cache_open_pipefs,
1652         .release        = cache_release_pipefs,
1653 };
1654
1655 static int content_open_pipefs(struct inode *inode, struct file *filp)
1656 {
1657         struct cache_detail *cd = RPC_I(inode)->private;
1658
1659         return content_open(inode, filp, cd);
1660 }
1661
1662 static int content_release_pipefs(struct inode *inode, struct file *filp)
1663 {
1664         struct cache_detail *cd = RPC_I(inode)->private;
1665
1666         return content_release(inode, filp, cd);
1667 }
1668
1669 const struct file_operations content_file_operations_pipefs = {
1670         .open           = content_open_pipefs,
1671         .read           = seq_read,
1672         .llseek         = seq_lseek,
1673         .release        = content_release_pipefs,
1674 };
1675
1676 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1677 {
1678         struct cache_detail *cd = RPC_I(inode)->private;
1679
1680         return open_flush(inode, filp, cd);
1681 }
1682
1683 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1684 {
1685         struct cache_detail *cd = RPC_I(inode)->private;
1686
1687         return release_flush(inode, filp, cd);
1688 }
1689
1690 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1691                             size_t count, loff_t *ppos)
1692 {
1693         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1694
1695         return read_flush(filp, buf, count, ppos, cd);
1696 }
1697
1698 static ssize_t write_flush_pipefs(struct file *filp,
1699                                   const char __user *buf,
1700                                   size_t count, loff_t *ppos)
1701 {
1702         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1703
1704         return write_flush(filp, buf, count, ppos, cd);
1705 }
1706
1707 const struct file_operations cache_flush_operations_pipefs = {
1708         .open           = open_flush_pipefs,
1709         .read           = read_flush_pipefs,
1710         .write          = write_flush_pipefs,
1711         .release        = release_flush_pipefs,
1712 };
1713
1714 int sunrpc_cache_register_pipefs(struct dentry *parent,
1715                                  const char *name, mode_t umode,
1716                                  struct cache_detail *cd)
1717 {
1718         struct qstr q;
1719         struct dentry *dir;
1720         int ret = 0;
1721
1722         sunrpc_init_cache_detail(cd);
1723         q.name = name;
1724         q.len = strlen(name);
1725         q.hash = full_name_hash(q.name, q.len);
1726         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1727         if (!IS_ERR(dir))
1728                 cd->u.pipefs.dir = dir;
1729         else {
1730                 sunrpc_destroy_cache_detail(cd);
1731                 ret = PTR_ERR(dir);
1732         }
1733         return ret;
1734 }
1735 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1736
1737 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1738 {
1739         rpc_remove_cache_dir(cd->u.pipefs.dir);
1740         cd->u.pipefs.dir = NULL;
1741         sunrpc_destroy_cache_detail(cd);
1742 }
1743 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1744