net/sunrpc/cache.c

   1 /*
   2  * net/sunrpc/cache.c
   3  *
   4  * Generic code for various authentication-related caches
   5  * used by sunrpc clients and servers.
   6  *
   7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   8  *
   9  * Released under terms in GPL version 2.  See COPYING.
  10  *
  11  */
  12
  13 #include <linux/types.h>
  14 #include <linux/fs.h>
  15 #include <linux/file.h>
  16 #include <linux/slab.h>
  17 #include <linux/signal.h>
  18 #include <linux/sched.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/module.h>
  22 #include <linux/ctype.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/poll.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/net.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/mutex.h>
  30 #include <linux/pagemap.h>
  31 #include <asm/ioctls.h>
  32 #include <linux/sunrpc/types.h>
  33 #include <linux/sunrpc/cache.h>
  34 #include <linux/sunrpc/stats.h>
  35 #include <linux/sunrpc/rpc_pipe_fs.h>
  36 #include "netns.h"
  37
  38 #define  RPCDBG_FACILITY RPCDBG_CACHE
  39
  40 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  41 static void cache_revisit_request(struct cache_head *item);
  42
  43 static void cache_init(struct cache_head *h)
  44 {
  45         time_t now = seconds_since_boot();
  46         h->next = NULL;
  47         h->flags = 0;
  48         kref_init(&h->ref);
  49         h->expiry_time = now + CACHE_NEW_EXPIRY;
  50         h->last_refresh = now;
  51 }
  52
  53 static inline int cache_is_expired(struct cache_detail *detail, struct cache_head *h)
  54 {
  55         return  (h->expiry_time < seconds_since_boot()) ||
  56                 (detail->flush_time > h->last_refresh);
  57 }
  58
  59 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  60                                        struct cache_head *key, int hash)
  61 {
  62         struct cache_head **head,  **hp;
  63         struct cache_head *new = NULL, *freeme = NULL;
  64
  65         head = &detail->hash_table[hash];
  66
  67         read_lock(&detail->hash_lock);
  68
  69         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  70                 struct cache_head *tmp = *hp;
  71                 if (detail->match(tmp, key)) {
  72                         if (cache_is_expired(detail, tmp))
  73                                 /* This entry is expired, we will discard it. */
  74                                 break;
  75                         cache_get(tmp);
  76                         read_unlock(&detail->hash_lock);
  77                         return tmp;
  78                 }
  79         }
  80         read_unlock(&detail->hash_lock);
  81         /* Didn't find anything, insert an empty entry */
  82
  83         new = detail->alloc();
  84         if (!new)
  85                 return NULL;
  86         /* must fully initialise 'new', else
  87          * we might get lose if we need to
  88          * cache_put it soon.
  89          */
  90         cache_init(new);
  91         detail->init(new, key);
  92
  93         write_lock(&detail->hash_lock);
  94
  95         /* check if entry appeared while we slept */
  96         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  97                 struct cache_head *tmp = *hp;
  98                 if (detail->match(tmp, key)) {
  99                         if (cache_is_expired(detail, tmp)) {
 100                                 *hp = tmp->next;
 101                                 tmp->next = NULL;
 102                                 detail->entries --;
 103                                 freeme = tmp;
 104                                 break;
 105                         }
 106                         cache_get(tmp);
 107                         write_unlock(&detail->hash_lock);
 108                         cache_put(new, detail);
 109                         return tmp;
 110                 }
 111         }
 112         new->next = *head;
 113         *head = new;
 114         detail->entries++;
 115         cache_get(new);
 116         write_unlock(&detail->hash_lock);
 117
 118         if (freeme)
 119                 cache_put(freeme, detail);
 120         return new;
 121 }
 122 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
 123
 124
 125 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 126
 127 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
 128 {
 129         head->expiry_time = expiry;
 130         head->last_refresh = seconds_since_boot();
 131         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 132         set_bit(CACHE_VALID, &head->flags);
 133 }
 134
 135 static void cache_fresh_unlocked(struct cache_head *head,
 136                                  struct cache_detail *detail)
 137 {
 138         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 139                 cache_revisit_request(head);
 140                 cache_dequeue(detail, head);
 141         }
 142 }
 143
 144 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 145                                        struct cache_head *new, struct cache_head *old, int hash)
 146 {
 147         /* The 'old' entry is to be replaced by 'new'.
 148          * If 'old' is not VALID, we update it directly,
 149          * otherwise we need to replace it
 150          */
 151         struct cache_head **head;
 152         struct cache_head *tmp;
 153
 154         if (!test_bit(CACHE_VALID, &old->flags)) {
 155                 write_lock(&detail->hash_lock);
 156                 if (!test_bit(CACHE_VALID, &old->flags)) {
 157                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 158                                 set_bit(CACHE_NEGATIVE, &old->flags);
 159                         else
 160                                 detail->update(old, new);
 161                         cache_fresh_locked(old, new->expiry_time);
 162                         write_unlock(&detail->hash_lock);
 163                         cache_fresh_unlocked(old, detail);
 164                         return old;
 165                 }
 166                 write_unlock(&detail->hash_lock);
 167         }
 168         /* We need to insert a new entry */
 169         tmp = detail->alloc();
 170         if (!tmp) {
 171                 cache_put(old, detail);
 172                 return NULL;
 173         }
 174         cache_init(tmp);
 175         detail->init(tmp, old);
 176         head = &detail->hash_table[hash];
 177
 178         write_lock(&detail->hash_lock);
 179         if (test_bit(CACHE_NEGATIVE, &new->flags))
 180                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 181         else
 182                 detail->update(tmp, new);
 183         tmp->next = *head;
 184         *head = tmp;
 185         detail->entries++;
 186         cache_get(tmp);
 187         cache_fresh_locked(tmp, new->expiry_time);
 188         cache_fresh_locked(old, 0);
 189         write_unlock(&detail->hash_lock);
 190         cache_fresh_unlocked(tmp, detail);
 191         cache_fresh_unlocked(old, detail);
 192         cache_put(old, detail);
 193         return tmp;
 194 }
 195 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 196
 197 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 198 {
 199         if (!cd->cache_upcall)
 200                 return -EINVAL;
 201         return cd->cache_upcall(cd, h);
 202 }
 203
 204 static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
 205 {
 206         if (!test_bit(CACHE_VALID, &h->flags))
 207                 return -EAGAIN;
 208         else {
 209                 /* entry is valid */
 210                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 211                         return -ENOENT;
 212                 else {
 213                         /*
 214                          * In combination with write barrier in
 215                          * sunrpc_cache_update, ensures that anyone
 216                          * using the cache entry after this sees the
 217                          * updated contents:
 218                          */
 219                         smp_rmb();
 220                         return 0;
 221                 }
 222         }
 223 }
 224
 225 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 226 {
 227         int rv;
 228
 229         write_lock(&detail->hash_lock);
 230         rv = cache_is_valid(detail, h);
 231         if (rv != -EAGAIN) {
 232                 write_unlock(&detail->hash_lock);
 233                 return rv;
 234         }
 235         set_bit(CACHE_NEGATIVE, &h->flags);
 236         cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY);
 237         write_unlock(&detail->hash_lock);
 238         cache_fresh_unlocked(h, detail);
 239         return -ENOENT;
 240 }
 241
 242 /*
 243  * This is the generic cache management routine for all
 244  * the authentication caches.
 245  * It checks the currency of a cache item and will (later)
 246  * initiate an upcall to fill it if needed.
 247  *
 248  *
 249  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 250  * -EAGAIN if upcall is pending and request has been queued
 251  * -ETIMEDOUT if upcall failed or request could not be queue or
 252  *           upcall completed but item is still invalid (implying that
 253  *           the cache item has been replaced with a newer one).
 254  * -ENOENT if cache entry was negative
 255  */
 256 int cache_check(struct cache_detail *detail,
 257                     struct cache_head *h, struct cache_req *rqstp)
 258 {
 259         int rv;
 260         long refresh_age, age;
 261
 262         /* First decide return status as best we can */
 263         rv = cache_is_valid(detail, h);
 264
 265         /* now see if we want to start an upcall */
 266         refresh_age = (h->expiry_time - h->last_refresh);
 267         age = seconds_since_boot() - h->last_refresh;
 268
 269         if (rqstp == NULL) {
 270                 if (rv == -EAGAIN)
 271                         rv = -ENOENT;
 272         } else if (rv == -EAGAIN || age > refresh_age/2) {
 273                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 274                                 refresh_age, age);
 275                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 276                         switch (cache_make_upcall(detail, h)) {
 277                         case -EINVAL:
 278                                 clear_bit(CACHE_PENDING, &h->flags);
 279                                 cache_revisit_request(h);
 280                                 rv = try_to_negate_entry(detail, h);
 281                                 break;
 282                         case -EAGAIN:
 283                                 clear_bit(CACHE_PENDING, &h->flags);
 284                                 cache_revisit_request(h);
 285                                 break;
 286                         }
 287                 }
 288         }
 289
 290         if (rv == -EAGAIN) {
 291                 if (!cache_defer_req(rqstp, h)) {
 292                         /*
 293                          * Request was not deferred; handle it as best
 294                          * we can ourselves:
 295                          */
 296                         rv = cache_is_valid(detail, h);
 297                         if (rv == -EAGAIN)
 298                                 rv = -ETIMEDOUT;
 299                 }
 300         }
 301         if (rv)
 302                 cache_put(h, detail);
 303         return rv;
 304 }
 305 EXPORT_SYMBOL_GPL(cache_check);
 306
 307 /*
 308  * caches need to be periodically cleaned.
 309  * For this we maintain a list of cache_detail and
 310  * a current pointer into that list and into the table
 311  * for that entry.
 312  *
 313  * Each time clean_cache is called it finds the next non-empty entry
 314  * in the current table and walks the list in that entry
 315  * looking for entries that can be removed.
 316  *
 317  * An entry gets removed if:
 318  * - The expiry is before current time
 319  * - The last_refresh time is before the flush_time for that cache
 320  *
 321  * later we might drop old entries with non-NEVER expiry if that table
 322  * is getting 'full' for some definition of 'full'
 323  *
 324  * The question of "how often to scan a table" is an interesting one
 325  * and is answered in part by the use of the "nextcheck" field in the
 326  * cache_detail.
 327  * When a scan of a table begins, the nextcheck field is set to a time
 328  * that is well into the future.
 329  * While scanning, if an expiry time is found that is earlier than the
 330  * current nextcheck time, nextcheck is set to that expiry time.
 331  * If the flush_time is ever set to a time earlier than the nextcheck
 332  * time, the nextcheck time is then set to that flush_time.
 333  *
 334  * A table is then only scanned if the current time is at least
 335  * the nextcheck time.
 336  *
 337  */
 338
 339 static LIST_HEAD(cache_list);
 340 static DEFINE_SPINLOCK(cache_list_lock);
 341 static struct cache_detail *current_detail;
 342 static int current_index;
 343
 344 static void do_cache_clean(struct work_struct *work);
 345 static struct delayed_work cache_cleaner;
 346
 347 void sunrpc_init_cache_detail(struct cache_detail *cd)
 348 {
 349         rwlock_init(&cd->hash_lock);
 350         INIT_LIST_HEAD(&cd->queue);
 351         spin_lock(&cache_list_lock);
 352         cd->nextcheck = 0;
 353         cd->entries = 0;
 354         atomic_set(&cd->readers, 0);
 355         cd->last_close = 0;
 356         cd->last_warn = -1;
 357         list_add(&cd->others, &cache_list);
 358         spin_unlock(&cache_list_lock);
 359
 360         /* start the cleaning process */
 361         schedule_delayed_work(&cache_cleaner, 0);
 362 }
 363 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 364
 365 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 366 {
 367         cache_purge(cd);
 368         spin_lock(&cache_list_lock);
 369         write_lock(&cd->hash_lock);
 370         if (cd->entries || atomic_read(&cd->inuse)) {
 371                 write_unlock(&cd->hash_lock);
 372                 spin_unlock(&cache_list_lock);
 373                 goto out;
 374         }
 375         if (current_detail == cd)
 376                 current_detail = NULL;
 377         list_del_init(&cd->others);
 378         write_unlock(&cd->hash_lock);
 379         spin_unlock(&cache_list_lock);
 380         if (list_empty(&cache_list)) {
 381                 /* module must be being unloaded so its safe to kill the worker */
 382                 cancel_delayed_work_sync(&cache_cleaner);
 383         }
 384         return;
 385 out:
 386         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 387 }
 388 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 389
 390 /* clean cache tries to find something to clean
 391  * and cleans it.
 392  * It returns 1 if it cleaned something,
 393  *            0 if it didn't find anything this time
 394  *           -1 if it fell off the end of the list.
 395  */
 396 static int cache_clean(void)
 397 {
 398         int rv = 0;
 399         struct list_head *next;
 400
 401         spin_lock(&cache_list_lock);
 402
 403         /* find a suitable table if we don't already have one */
 404         while (current_detail == NULL ||
 405             current_index >= current_detail->hash_size) {
 406                 if (current_detail)
 407                         next = current_detail->others.next;
 408                 else
 409                         next = cache_list.next;
 410                 if (next == &cache_list) {
 411                         current_detail = NULL;
 412                         spin_unlock(&cache_list_lock);
 413                         return -1;
 414                 }
 415                 current_detail = list_entry(next, struct cache_detail, others);
 416                 if (current_detail->nextcheck > seconds_since_boot())
 417                         current_index = current_detail->hash_size;
 418                 else {
 419                         current_index = 0;
 420                         current_detail->nextcheck = seconds_since_boot()+30*60;
 421                 }
 422         }
 423
 424         /* find a non-empty bucket in the table */
 425         while (current_detail &&
 426                current_index < current_detail->hash_size &&
 427                current_detail->hash_table[current_index] == NULL)
 428                 current_index++;
 429
 430         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 431
 432         if (current_detail && current_index < current_detail->hash_size) {
 433                 struct cache_head *ch, **cp;
 434                 struct cache_detail *d;
 435
 436                 write_lock(&current_detail->hash_lock);
 437
 438                 /* Ok, now to clean this strand */
 439
 440                 cp = & current_detail->hash_table[current_index];
 441                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
 442                         if (current_detail->nextcheck > ch->expiry_time)
 443                                 current_detail->nextcheck = ch->expiry_time+1;
 444                         if (!cache_is_expired(current_detail, ch))
 445                                 continue;
 446
 447                         *cp = ch->next;
 448                         ch->next = NULL;
 449                         current_detail->entries--;
 450                         rv = 1;
 451                         break;
 452                 }
 453
 454                 write_unlock(&current_detail->hash_lock);
 455                 d = current_detail;
 456                 if (!ch)
 457                         current_index ++;
 458                 spin_unlock(&cache_list_lock);
 459                 if (ch) {
 460                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 461                                 cache_dequeue(current_detail, ch);
 462                         cache_revisit_request(ch);
 463                         cache_put(ch, d);
 464                 }
 465         } else
 466                 spin_unlock(&cache_list_lock);
 467
 468         return rv;
 469 }
 470
 471 /*
 472  * We want to regularly clean the cache, so we need to schedule some work ...
 473  */
 474 static void do_cache_clean(struct work_struct *work)
 475 {
 476         int delay = 5;
 477         if (cache_clean() == -1)
 478                 delay = round_jiffies_relative(30*HZ);
 479
 480         if (list_empty(&cache_list))
 481                 delay = 0;
 482
 483         if (delay)
 484                 schedule_delayed_work(&cache_cleaner, delay);
 485 }
 486
 487
 488 /*
 489  * Clean all caches promptly.  This just calls cache_clean
 490  * repeatedly until we are sure that every cache has had a chance to
 491  * be fully cleaned
 492  */
 493 void cache_flush(void)
 494 {
 495         while (cache_clean() != -1)
 496                 cond_resched();
 497         while (cache_clean() != -1)
 498                 cond_resched();
 499 }
 500 EXPORT_SYMBOL_GPL(cache_flush);
 501
 502 void cache_purge(struct cache_detail *detail)
 503 {
 504         detail->flush_time = LONG_MAX;
 505         detail->nextcheck = seconds_since_boot();
 506         cache_flush();
 507         detail->flush_time = 1;
 508 }
 509 EXPORT_SYMBOL_GPL(cache_purge);
 510
 511
 512 /*
 513  * Deferral and Revisiting of Requests.
 514  *
 515  * If a cache lookup finds a pending entry, we
 516  * need to defer the request and revisit it later.
 517  * All deferred requests are stored in a hash table,
 518  * indexed by "struct cache_head *".
 519  * As it may be wasteful to store a whole request
 520  * structure, we allow the request to provide a
 521  * deferred form, which must contain a
 522  * 'struct cache_deferred_req'
 523  * This cache_deferred_req contains a method to allow
 524  * it to be revisited when cache info is available
 525  */
 526
 527 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 528 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 529
 530 #define DFR_MAX 300     /* ??? */
 531
 532 static DEFINE_SPINLOCK(cache_defer_lock);
 533 static LIST_HEAD(cache_defer_list);
 534 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 535 static int cache_defer_cnt;
 536
 537 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 538 {
 539         hlist_del_init(&dreq->hash);
 540         if (!list_empty(&dreq->recent)) {
 541                 list_del_init(&dreq->recent);
 542                 cache_defer_cnt--;
 543         }
 544 }
 545
 546 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 547 {
 548         int hash = DFR_HASH(item);
 549
 550         INIT_LIST_HEAD(&dreq->recent);
 551         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 552 }
 553
 554 static void setup_deferral(struct cache_deferred_req *dreq,
 555                            struct cache_head *item,
 556                            int count_me)
 557 {
 558
 559         dreq->item = item;
 560
 561         spin_lock(&cache_defer_lock);
 562
 563         __hash_deferred_req(dreq, item);
 564
 565         if (count_me) {
 566                 cache_defer_cnt++;
 567                 list_add(&dreq->recent, &cache_defer_list);
 568         }
 569
 570         spin_unlock(&cache_defer_lock);
 571
 572 }
 573
 574 struct thread_deferred_req {
 575         struct cache_deferred_req handle;
 576         struct completion completion;
 577 };
 578
 579 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 580 {
 581         struct thread_deferred_req *dr =
 582                 container_of(dreq, struct thread_deferred_req, handle);
 583         complete(&dr->completion);
 584 }
 585
 586 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 587 {
 588         struct thread_deferred_req sleeper;
 589         struct cache_deferred_req *dreq = &sleeper.handle;
 590
 591         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 592         dreq->revisit = cache_restart_thread;
 593
 594         setup_deferral(dreq, item, 0);
 595
 596         if (!test_bit(CACHE_PENDING, &item->flags) ||
 597             wait_for_completion_interruptible_timeout(
 598                     &sleeper.completion, req->thread_wait) <= 0) {
 599                 /* The completion wasn't completed, so we need
 600                  * to clean up
 601                  */
 602                 spin_lock(&cache_defer_lock);
 603                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 604                         __unhash_deferred_req(&sleeper.handle);
 605                         spin_unlock(&cache_defer_lock);
 606                 } else {
 607                         /* cache_revisit_request already removed
 608                          * this from the hash table, but hasn't
 609                          * called ->revisit yet.  It will very soon
 610                          * and we need to wait for it.
 611                          */
 612                         spin_unlock(&cache_defer_lock);
 613                         wait_for_completion(&sleeper.completion);
 614                 }
 615         }
 616 }
 617
 618 static void cache_limit_defers(void)
 619 {
 620         /* Make sure we haven't exceed the limit of allowed deferred
 621          * requests.
 622          */
 623         struct cache_deferred_req *discard = NULL;
 624
 625         if (cache_defer_cnt <= DFR_MAX)
 626                 return;
 627
 628         spin_lock(&cache_defer_lock);
 629
 630         /* Consider removing either the first or the last */
 631         if (cache_defer_cnt > DFR_MAX) {
 632                 if (net_random() & 1)
 633                         discard = list_entry(cache_defer_list.next,
 634                                              struct cache_deferred_req, recent);
 635                 else
 636                         discard = list_entry(cache_defer_list.prev,
 637                                              struct cache_deferred_req, recent);
 638                 __unhash_deferred_req(discard);
 639         }
 640         spin_unlock(&cache_defer_lock);
 641         if (discard)
 642                 discard->revisit(discard, 1);
 643 }
 644
 645 /* Return true if and only if a deferred request is queued. */
 646 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 647 {
 648         struct cache_deferred_req *dreq;
 649
 650         if (req->thread_wait) {
 651                 cache_wait_req(req, item);
 652                 if (!test_bit(CACHE_PENDING, &item->flags))
 653                         return false;
 654         }
 655         dreq = req->defer(req);
 656         if (dreq == NULL)
 657                 return false;
 658         setup_deferral(dreq, item, 1);
 659         if (!test_bit(CACHE_PENDING, &item->flags))
 660                 /* Bit could have been cleared before we managed to
 661                  * set up the deferral, so need to revisit just in case
 662                  */
 663                 cache_revisit_request(item);
 664
 665         cache_limit_defers();
 666         return true;
 667 }
 668
 669 static void cache_revisit_request(struct cache_head *item)
 670 {
 671         struct cache_deferred_req *dreq;
 672         struct list_head pending;
 673         struct hlist_node *lp, *tmp;
 674         int hash = DFR_HASH(item);
 675
 676         INIT_LIST_HEAD(&pending);
 677         spin_lock(&cache_defer_lock);
 678
 679         hlist_for_each_entry_safe(dreq, lp, tmp, &cache_defer_hash[hash], hash)
 680                 if (dreq->item == item) {
 681                         __unhash_deferred_req(dreq);
 682                         list_add(&dreq->recent, &pending);
 683                 }
 684
 685         spin_unlock(&cache_defer_lock);
 686
 687         while (!list_empty(&pending)) {
 688                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 689                 list_del_init(&dreq->recent);
 690                 dreq->revisit(dreq, 0);
 691         }
 692 }
 693
 694 void cache_clean_deferred(void *owner)
 695 {
 696         struct cache_deferred_req *dreq, *tmp;
 697         struct list_head pending;
 698
 699
 700         INIT_LIST_HEAD(&pending);
 701         spin_lock(&cache_defer_lock);
 702
 703         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 704                 if (dreq->owner == owner) {
 705                         __unhash_deferred_req(dreq);
 706                         list_add(&dreq->recent, &pending);
 707                 }
 708         }
 709         spin_unlock(&cache_defer_lock);
 710
 711         while (!list_empty(&pending)) {
 712                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 713                 list_del_init(&dreq->recent);
 714                 dreq->revisit(dreq, 1);
 715         }
 716 }
 717
 718 /*
 719  * communicate with user-space
 720  *
 721  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 722  * On read, you get a full request, or block.
 723  * On write, an update request is processed.
 724  * Poll works if anything to read, and always allows write.
 725  *
 726  * Implemented by linked list of requests.  Each open file has
 727  * a ->private that also exists in this list.  New requests are added
 728  * to the end and may wakeup and preceding readers.
 729  * New readers are added to the head.  If, on read, an item is found with
 730  * CACHE_UPCALLING clear, we free it from the list.
 731  *
 732  */
 733
 734 static DEFINE_SPINLOCK(queue_lock);
 735 static DEFINE_MUTEX(queue_io_mutex);
 736
 737 struct cache_queue {
 738         struct list_head        list;
 739         int                     reader; /* if 0, then request */
 740 };
 741 struct cache_request {
 742         struct cache_queue      q;
 743         struct cache_head       *item;
 744         char                    * buf;
 745         int                     len;
 746         int                     readers;
 747 };
 748 struct cache_reader {
 749         struct cache_queue      q;
 750         int                     offset; /* if non-0, we have a refcnt on next request */
 751 };
 752
 753 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 754                           loff_t *ppos, struct cache_detail *cd)
 755 {
 756         struct cache_reader *rp = filp->private_data;
 757         struct cache_request *rq;
 758         struct inode *inode = filp->f_path.dentry->d_inode;
 759         int err;
 760
 761         if (count == 0)
 762                 return 0;
 763
 764         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 765                               * readers on this file */
 766  again:
 767         spin_lock(&queue_lock);
 768         /* need to find next request */
 769         while (rp->q.list.next != &cd->queue &&
 770                list_entry(rp->q.list.next, struct cache_queue, list)
 771                ->reader) {
 772                 struct list_head *next = rp->q.list.next;
 773                 list_move(&rp->q.list, next);
 774         }
 775         if (rp->q.list.next == &cd->queue) {
 776                 spin_unlock(&queue_lock);
 777                 mutex_unlock(&inode->i_mutex);
 778                 WARN_ON_ONCE(rp->offset);
 779                 return 0;
 780         }
 781         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 782         WARN_ON_ONCE(rq->q.reader);
 783         if (rp->offset == 0)
 784                 rq->readers++;
 785         spin_unlock(&queue_lock);
 786
 787         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 788                 err = -EAGAIN;
 789                 spin_lock(&queue_lock);
 790                 list_move(&rp->q.list, &rq->q.list);
 791                 spin_unlock(&queue_lock);
 792         } else {
 793                 if (rp->offset + count > rq->len)
 794                         count = rq->len - rp->offset;
 795                 err = -EFAULT;
 796                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 797                         goto out;
 798                 rp->offset += count;
 799                 if (rp->offset >= rq->len) {
 800                         rp->offset = 0;
 801                         spin_lock(&queue_lock);
 802                         list_move(&rp->q.list, &rq->q.list);
 803                         spin_unlock(&queue_lock);
 804                 }
 805                 err = 0;
 806         }
 807  out:
 808         if (rp->offset == 0) {
 809                 /* need to release rq */
 810                 spin_lock(&queue_lock);
 811                 rq->readers--;
 812                 if (rq->readers == 0 &&
 813                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 814                         list_del(&rq->q.list);
 815                         spin_unlock(&queue_lock);
 816                         cache_put(rq->item, cd);
 817                         kfree(rq->buf);
 818                         kfree(rq);
 819                 } else
 820                         spin_unlock(&queue_lock);
 821         }
 822         if (err == -EAGAIN)
 823                 goto again;
 824         mutex_unlock(&inode->i_mutex);
 825         return err ? err :  count;
 826 }
 827
 828 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 829                                  size_t count, struct cache_detail *cd)
 830 {
 831         ssize_t ret;
 832
 833         if (count == 0)
 834                 return -EINVAL;
 835         if (copy_from_user(kaddr, buf, count))
 836                 return -EFAULT;
 837         kaddr[count] = '\0';
 838         ret = cd->cache_parse(cd, kaddr, count);
 839         if (!ret)
 840                 ret = count;
 841         return ret;
 842 }
 843
 844 static ssize_t cache_slow_downcall(const char __user *buf,
 845                                    size_t count, struct cache_detail *cd)
 846 {
 847         static char write_buf[8192]; /* protected by queue_io_mutex */
 848         ssize_t ret = -EINVAL;
 849
 850         if (count >= sizeof(write_buf))
 851                 goto out;
 852         mutex_lock(&queue_io_mutex);
 853         ret = cache_do_downcall(write_buf, buf, count, cd);
 854         mutex_unlock(&queue_io_mutex);
 855 out:
 856         return ret;
 857 }
 858
 859 static ssize_t cache_downcall(struct address_space *mapping,
 860                               const char __user *buf,
 861                               size_t count, struct cache_detail *cd)
 862 {
 863         struct page *page;
 864         char *kaddr;
 865         ssize_t ret = -ENOMEM;
 866
 867         if (count >= PAGE_CACHE_SIZE)
 868                 goto out_slow;
 869
 870         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 871         if (!page)
 872                 goto out_slow;
 873
 874         kaddr = kmap(page);
 875         ret = cache_do_downcall(kaddr, buf, count, cd);
 876         kunmap(page);
 877         unlock_page(page);
 878         page_cache_release(page);
 879         return ret;
 880 out_slow:
 881         return cache_slow_downcall(buf, count, cd);
 882 }
 883
 884 static ssize_t cache_write(struct file *filp, const char __user *buf,
 885                            size_t count, loff_t *ppos,
 886                            struct cache_detail *cd)
 887 {
 888         struct address_space *mapping = filp->f_mapping;
 889         struct inode *inode = filp->f_path.dentry->d_inode;
 890         ssize_t ret = -EINVAL;
 891
 892         if (!cd->cache_parse)
 893                 goto out;
 894
 895         mutex_lock(&inode->i_mutex);
 896         ret = cache_downcall(mapping, buf, count, cd);
 897         mutex_unlock(&inode->i_mutex);
 898 out:
 899         return ret;
 900 }
 901
 902 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 903
 904 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 905                                struct cache_detail *cd)
 906 {
 907         unsigned int mask;
 908         struct cache_reader *rp = filp->private_data;
 909         struct cache_queue *cq;
 910
 911         poll_wait(filp, &queue_wait, wait);
 912
 913         /* alway allow write */
 914         mask = POLL_OUT | POLLWRNORM;
 915
 916         if (!rp)
 917                 return mask;
 918
 919         spin_lock(&queue_lock);
 920
 921         for (cq= &rp->q; &cq->list != &cd->queue;
 922              cq = list_entry(cq->list.next, struct cache_queue, list))
 923                 if (!cq->reader) {
 924                         mask |= POLLIN | POLLRDNORM;
 925                         break;
 926                 }
 927         spin_unlock(&queue_lock);
 928         return mask;
 929 }
 930
 931 static int cache_ioctl(struct inode *ino, struct file *filp,
 932                        unsigned int cmd, unsigned long arg,
 933                        struct cache_detail *cd)
 934 {
 935         int len = 0;
 936         struct cache_reader *rp = filp->private_data;
 937         struct cache_queue *cq;
 938
 939         if (cmd != FIONREAD || !rp)
 940                 return -EINVAL;
 941
 942         spin_lock(&queue_lock);
 943
 944         /* only find the length remaining in current request,
 945          * or the length of the next request
 946          */
 947         for (cq= &rp->q; &cq->list != &cd->queue;
 948              cq = list_entry(cq->list.next, struct cache_queue, list))
 949                 if (!cq->reader) {
 950                         struct cache_request *cr =
 951                                 container_of(cq, struct cache_request, q);
 952                         len = cr->len - rp->offset;
 953                         break;
 954                 }
 955         spin_unlock(&queue_lock);
 956
 957         return put_user(len, (int __user *)arg);
 958 }
 959
 960 static int cache_open(struct inode *inode, struct file *filp,
 961                       struct cache_detail *cd)
 962 {
 963         struct cache_reader *rp = NULL;
 964
 965         if (!cd || !try_module_get(cd->owner))
 966                 return -EACCES;
 967         nonseekable_open(inode, filp);
 968         if (filp->f_mode & FMODE_READ) {
 969                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 970                 if (!rp)
 971                         return -ENOMEM;
 972                 rp->offset = 0;
 973                 rp->q.reader = 1;
 974                 atomic_inc(&cd->readers);
 975                 spin_lock(&queue_lock);
 976                 list_add(&rp->q.list, &cd->queue);
 977                 spin_unlock(&queue_lock);
 978         }
 979         filp->private_data = rp;
 980         return 0;
 981 }
 982
 983 static int cache_release(struct inode *inode, struct file *filp,
 984                          struct cache_detail *cd)
 985 {
 986         struct cache_reader *rp = filp->private_data;
 987
 988         if (rp) {
 989                 spin_lock(&queue_lock);
 990                 if (rp->offset) {
 991                         struct cache_queue *cq;
 992                         for (cq= &rp->q; &cq->list != &cd->queue;
 993                              cq = list_entry(cq->list.next, struct cache_queue, list))
 994                                 if (!cq->reader) {
 995                                         container_of(cq, struct cache_request, q)
 996                                                 ->readers--;
 997                                         break;
 998                                 }
 999                         rp->offset = 0;
1000                 }
1001                 list_del(&rp->q.list);
1002                 spin_unlock(&queue_lock);
1003
1004                 filp->private_data = NULL;
1005                 kfree(rp);
1006
1007                 cd->last_close = seconds_since_boot();
1008                 atomic_dec(&cd->readers);
1009         }
1010         module_put(cd->owner);
1011         return 0;
1012 }
1013
1014
1015
1016 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1017 {
1018         struct cache_queue *cq;
1019         spin_lock(&queue_lock);
1020         list_for_each_entry(cq, &detail->queue, list)
1021                 if (!cq->reader) {
1022                         struct cache_request *cr = container_of(cq, struct cache_request, q);
1023                         if (cr->item != ch)
1024                                 continue;
1025                         if (cr->readers != 0)
1026                                 continue;
1027                         list_del(&cr->q.list);
1028                         spin_unlock(&queue_lock);
1029                         cache_put(cr->item, detail);
1030                         kfree(cr->buf);
1031                         kfree(cr);
1032                         return;
1033                 }
1034         spin_unlock(&queue_lock);
1035 }
1036
1037 /*
1038  * Support routines for text-based upcalls.
1039  * Fields are separated by spaces.
1040  * Fields are either mangled to quote space tab newline slosh with slosh
1041  * or a hexified with a leading \x
1042  * Record is terminated with newline.
1043  *
1044  */
1045
1046 void qword_add(char **bpp, int *lp, char *str)
1047 {
1048         char *bp = *bpp;
1049         int len = *lp;
1050         char c;
1051
1052         if (len < 0) return;
1053
1054         while ((c=*str++) && len)
1055                 switch(c) {
1056                 case ' ':
1057                 case '\t':
1058                 case '\n':
1059                 case '\\':
1060                         if (len >= 4) {
1061                                 *bp++ = '\\';
1062                                 *bp++ = '0' + ((c & 0300)>>6);
1063                                 *bp++ = '0' + ((c & 0070)>>3);
1064                                 *bp++ = '0' + ((c & 0007)>>0);
1065                         }
1066                         len -= 4;
1067                         break;
1068                 default:
1069                         *bp++ = c;
1070                         len--;
1071                 }
1072         if (c || len <1) len = -1;
1073         else {
1074                 *bp++ = ' ';
1075                 len--;
1076         }
1077         *bpp = bp;
1078         *lp = len;
1079 }
1080 EXPORT_SYMBOL_GPL(qword_add);
1081
1082 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1083 {
1084         char *bp = *bpp;
1085         int len = *lp;
1086
1087         if (len < 0) return;
1088
1089         if (len > 2) {
1090                 *bp++ = '\\';
1091                 *bp++ = 'x';
1092                 len -= 2;
1093                 while (blen && len >= 2) {
1094                         unsigned char c = *buf++;
1095                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1096                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1097                         len -= 2;
1098                         blen--;
1099                 }
1100         }
1101         if (blen || len<1) len = -1;
1102         else {
1103                 *bp++ = ' ';
1104                 len--;
1105         }
1106         *bpp = bp;
1107         *lp = len;
1108 }
1109 EXPORT_SYMBOL_GPL(qword_addhex);
1110
1111 static void warn_no_listener(struct cache_detail *detail)
1112 {
1113         if (detail->last_warn != detail->last_close) {
1114                 detail->last_warn = detail->last_close;
1115                 if (detail->warn_no_listener)
1116                         detail->warn_no_listener(detail, detail->last_close != 0);
1117         }
1118 }
1119
1120 static bool cache_listeners_exist(struct cache_detail *detail)
1121 {
1122         if (atomic_read(&detail->readers))
1123                 return true;
1124         if (detail->last_close == 0)
1125                 /* This cache was never opened */
1126                 return false;
1127         if (detail->last_close < seconds_since_boot() - 30)
1128                 /*
1129                  * We allow for the possibility that someone might
1130                  * restart a userspace daemon without restarting the
1131                  * server; but after 30 seconds, we give up.
1132                  */
1133                  return false;
1134         return true;
1135 }
1136
1137 /*
1138  * register an upcall request to user-space and queue it up for read() by the
1139  * upcall daemon.
1140  *
1141  * Each request is at most one page long.
1142  */
1143 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1144                 void (*cache_request)(struct cache_detail *,
1145                                       struct cache_head *,
1146                                       char **,
1147                                       int *))
1148 {
1149
1150         char *buf;
1151         struct cache_request *crq;
1152         char *bp;
1153         int len;
1154
1155         if (!cache_listeners_exist(detail)) {
1156                 warn_no_listener(detail);
1157                 return -EINVAL;
1158         }
1159
1160         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1161         if (!buf)
1162                 return -EAGAIN;
1163
1164         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1165         if (!crq) {
1166                 kfree(buf);
1167                 return -EAGAIN;
1168         }
1169
1170         bp = buf; len = PAGE_SIZE;
1171
1172         cache_request(detail, h, &bp, &len);
1173
1174         if (len < 0) {
1175                 kfree(buf);
1176                 kfree(crq);
1177                 return -EAGAIN;
1178         }
1179         crq->q.reader = 0;
1180         crq->item = cache_get(h);
1181         crq->buf = buf;
1182         crq->len = PAGE_SIZE - len;
1183         crq->readers = 0;
1184         spin_lock(&queue_lock);
1185         list_add_tail(&crq->q.list, &detail->queue);
1186         spin_unlock(&queue_lock);
1187         wake_up(&queue_wait);
1188         return 0;
1189 }
1190 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1191
1192 /*
1193  * parse a message from user-space and pass it
1194  * to an appropriate cache
1195  * Messages are, like requests, separated into fields by
1196  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1197  *
1198  * Message is
1199  *   reply cachename expiry key ... content....
1200  *
1201  * key and content are both parsed by cache
1202  */
1203
1204 #define isodigit(c) (isdigit(c) && c <= '7')
1205 int qword_get(char **bpp, char *dest, int bufsize)
1206 {
1207         /* return bytes copied, or -1 on error */
1208         char *bp = *bpp;
1209         int len = 0;
1210
1211         while (*bp == ' ') bp++;
1212
1213         if (bp[0] == '\\' && bp[1] == 'x') {
1214                 /* HEX STRING */
1215                 bp += 2;
1216                 while (len < bufsize) {
1217                         int h, l;
1218
1219                         h = hex_to_bin(bp[0]);
1220                         if (h < 0)
1221                                 break;
1222
1223                         l = hex_to_bin(bp[1]);
1224                         if (l < 0)
1225                                 break;
1226
1227                         *dest++ = (h << 4) | l;
1228                         bp += 2;
1229                         len++;
1230                 }
1231         } else {
1232                 /* text with \nnn octal quoting */
1233                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1234                         if (*bp == '\\' &&
1235                             isodigit(bp[1]) && (bp[1] <= '3') &&
1236                             isodigit(bp[2]) &&
1237                             isodigit(bp[3])) {
1238                                 int byte = (*++bp -'0');
1239                                 bp++;
1240                                 byte = (byte << 3) | (*bp++ - '0');
1241                                 byte = (byte << 3) | (*bp++ - '0');
1242                                 *dest++ = byte;
1243                                 len++;
1244                         } else {
1245                                 *dest++ = *bp++;
1246                                 len++;
1247                         }
1248                 }
1249         }
1250
1251         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1252                 return -1;
1253         while (*bp == ' ') bp++;
1254         *bpp = bp;
1255         *dest = '\0';
1256         return len;
1257 }
1258 EXPORT_SYMBOL_GPL(qword_get);
1259
1260
1261 /*
1262  * support /proc/sunrpc/cache/$CACHENAME/content
1263  * as a seqfile.
1264  * We call ->cache_show passing NULL for the item to
1265  * get a header, then pass each real item in the cache
1266  */
1267
1268 struct handle {
1269         struct cache_detail *cd;
1270 };
1271
1272 static void *c_start(struct seq_file *m, loff_t *pos)
1273         __acquires(cd->hash_lock)
1274 {
1275         loff_t n = *pos;
1276         unsigned int hash, entry;
1277         struct cache_head *ch;
1278         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1279
1280
1281         read_lock(&cd->hash_lock);
1282         if (!n--)
1283                 return SEQ_START_TOKEN;
1284         hash = n >> 32;
1285         entry = n & ((1LL<<32) - 1);
1286
1287         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1288                 if (!entry--)
1289                         return ch;
1290         n &= ~((1LL<<32) - 1);
1291         do {
1292                 hash++;
1293                 n += 1LL<<32;
1294         } while(hash < cd->hash_size &&
1295                 cd->hash_table[hash]==NULL);
1296         if (hash >= cd->hash_size)
1297                 return NULL;
1298         *pos = n+1;
1299         return cd->hash_table[hash];
1300 }
1301
1302 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1303 {
1304         struct cache_head *ch = p;
1305         int hash = (*pos >> 32);
1306         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1307
1308         if (p == SEQ_START_TOKEN)
1309                 hash = 0;
1310         else if (ch->next == NULL) {
1311                 hash++;
1312                 *pos += 1LL<<32;
1313         } else {
1314                 ++*pos;
1315                 return ch->next;
1316         }
1317         *pos &= ~((1LL<<32) - 1);
1318         while (hash < cd->hash_size &&
1319                cd->hash_table[hash] == NULL) {
1320                 hash++;
1321                 *pos += 1LL<<32;
1322         }
1323         if (hash >= cd->hash_size)
1324                 return NULL;
1325         ++*pos;
1326         return cd->hash_table[hash];
1327 }
1328
1329 static void c_stop(struct seq_file *m, void *p)
1330         __releases(cd->hash_lock)
1331 {
1332         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1333         read_unlock(&cd->hash_lock);
1334 }
1335
1336 static int c_show(struct seq_file *m, void *p)
1337 {
1338         struct cache_head *cp = p;
1339         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1340
1341         if (p == SEQ_START_TOKEN)
1342                 return cd->cache_show(m, cd, NULL);
1343
1344         ifdebug(CACHE)
1345                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1346                            convert_to_wallclock(cp->expiry_time),
1347                            atomic_read(&cp->ref.refcount), cp->flags);
1348         cache_get(cp);
1349         if (cache_check(cd, cp, NULL))
1350                 /* cache_check does a cache_put on failure */
1351                 seq_printf(m, "# ");
1352         else {
1353                 if (cache_is_expired(cd, cp))
1354                         seq_printf(m, "# ");
1355                 cache_put(cp, cd);
1356         }
1357
1358         return cd->cache_show(m, cd, cp);
1359 }
1360
1361 static const struct seq_operations cache_content_op = {
1362         .start  = c_start,
1363         .next   = c_next,
1364         .stop   = c_stop,
1365         .show   = c_show,
1366 };
1367
1368 static int content_open(struct inode *inode, struct file *file,
1369                         struct cache_detail *cd)
1370 {
1371         struct handle *han;
1372
1373         if (!cd || !try_module_get(cd->owner))
1374                 return -EACCES;
1375         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1376         if (han == NULL) {
1377                 module_put(cd->owner);
1378                 return -ENOMEM;
1379         }
1380
1381         han->cd = cd;
1382         return 0;
1383 }
1384
1385 static int content_release(struct inode *inode, struct file *file,
1386                 struct cache_detail *cd)
1387 {
1388         int ret = seq_release_private(inode, file);
1389         module_put(cd->owner);
1390         return ret;
1391 }
1392
1393 static int open_flush(struct inode *inode, struct file *file,
1394                         struct cache_detail *cd)
1395 {
1396         if (!cd || !try_module_get(cd->owner))
1397                 return -EACCES;
1398         return nonseekable_open(inode, file);
1399 }
1400
1401 static int release_flush(struct inode *inode, struct file *file,
1402                         struct cache_detail *cd)
1403 {
1404         module_put(cd->owner);
1405         return 0;
1406 }
1407
1408 static ssize_t read_flush(struct file *file, char __user *buf,
1409                           size_t count, loff_t *ppos,
1410                           struct cache_detail *cd)
1411 {
1412         char tbuf[22];
1413         unsigned long p = *ppos;
1414         size_t len;
1415
1416         snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1417         len = strlen(tbuf);
1418         if (p >= len)
1419                 return 0;
1420         len -= p;
1421         if (len > count)
1422                 len = count;
1423         if (copy_to_user(buf, (void*)(tbuf+p), len))
1424                 return -EFAULT;
1425         *ppos += len;
1426         return len;
1427 }
1428
1429 static ssize_t write_flush(struct file *file, const char __user *buf,
1430                            size_t count, loff_t *ppos,
1431                            struct cache_detail *cd)
1432 {
1433         char tbuf[20];
1434         char *bp, *ep;
1435
1436         if (*ppos || count > sizeof(tbuf)-1)
1437                 return -EINVAL;
1438         if (copy_from_user(tbuf, buf, count))
1439                 return -EFAULT;
1440         tbuf[count] = 0;
1441         simple_strtoul(tbuf, &ep, 0);
1442         if (*ep && *ep != '\n')
1443                 return -EINVAL;
1444
1445         bp = tbuf;
1446         cd->flush_time = get_expiry(&bp);
1447         cd->nextcheck = seconds_since_boot();
1448         cache_flush();
1449
1450         *ppos += count;
1451         return count;
1452 }
1453
1454 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1455                                  size_t count, loff_t *ppos)
1456 {
1457         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1458
1459         return cache_read(filp, buf, count, ppos, cd);
1460 }
1461
1462 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1463                                   size_t count, loff_t *ppos)
1464 {
1465         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1466
1467         return cache_write(filp, buf, count, ppos, cd);
1468 }
1469
1470 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1471 {
1472         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1473
1474         return cache_poll(filp, wait, cd);
1475 }
1476
1477 static long cache_ioctl_procfs(struct file *filp,
1478                                unsigned int cmd, unsigned long arg)
1479 {
1480         struct inode *inode = filp->f_path.dentry->d_inode;
1481         struct cache_detail *cd = PDE(inode)->data;
1482
1483         return cache_ioctl(inode, filp, cmd, arg, cd);
1484 }
1485
1486 static int cache_open_procfs(struct inode *inode, struct file *filp)
1487 {
1488         struct cache_detail *cd = PDE(inode)->data;
1489
1490         return cache_open(inode, filp, cd);
1491 }
1492
1493 static int cache_release_procfs(struct inode *inode, struct file *filp)
1494 {
1495         struct cache_detail *cd = PDE(inode)->data;
1496
1497         return cache_release(inode, filp, cd);
1498 }
1499
1500 static const struct file_operations cache_file_operations_procfs = {
1501         .owner          = THIS_MODULE,
1502         .llseek         = no_llseek,
1503         .read           = cache_read_procfs,
1504         .write          = cache_write_procfs,
1505         .poll           = cache_poll_procfs,
1506         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1507         .open           = cache_open_procfs,
1508         .release        = cache_release_procfs,
1509 };
1510
1511 static int content_open_procfs(struct inode *inode, struct file *filp)
1512 {
1513         struct cache_detail *cd = PDE(inode)->data;
1514
1515         return content_open(inode, filp, cd);
1516 }
1517
1518 static int content_release_procfs(struct inode *inode, struct file *filp)
1519 {
1520         struct cache_detail *cd = PDE(inode)->data;
1521
1522         return content_release(inode, filp, cd);
1523 }
1524
1525 static const struct file_operations content_file_operations_procfs = {
1526         .open           = content_open_procfs,
1527         .read           = seq_read,
1528         .llseek         = seq_lseek,
1529         .release        = content_release_procfs,
1530 };
1531
1532 static int open_flush_procfs(struct inode *inode, struct file *filp)
1533 {
1534         struct cache_detail *cd = PDE(inode)->data;
1535
1536         return open_flush(inode, filp, cd);
1537 }
1538
1539 static int release_flush_procfs(struct inode *inode, struct file *filp)
1540 {
1541         struct cache_detail *cd = PDE(inode)->data;
1542
1543         return release_flush(inode, filp, cd);
1544 }
1545
1546 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1547                             size_t count, loff_t *ppos)
1548 {
1549         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1550
1551         return read_flush(filp, buf, count, ppos, cd);
1552 }
1553
1554 static ssize_t write_flush_procfs(struct file *filp,
1555                                   const char __user *buf,
1556                                   size_t count, loff_t *ppos)
1557 {
1558         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1559
1560         return write_flush(filp, buf, count, ppos, cd);
1561 }
1562
1563 static const struct file_operations cache_flush_operations_procfs = {
1564         .open           = open_flush_procfs,
1565         .read           = read_flush_procfs,
1566         .write          = write_flush_procfs,
1567         .release        = release_flush_procfs,
1568         .llseek         = no_llseek,
1569 };
1570
1571 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1572 {
1573         struct sunrpc_net *sn;
1574
1575         if (cd->u.procfs.proc_ent == NULL)
1576                 return;
1577         if (cd->u.procfs.flush_ent)
1578                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1579         if (cd->u.procfs.channel_ent)
1580                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1581         if (cd->u.procfs.content_ent)
1582                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1583         cd->u.procfs.proc_ent = NULL;
1584         sn = net_generic(net, sunrpc_net_id);
1585         remove_proc_entry(cd->name, sn->proc_net_rpc);
1586 }
1587
1588 #ifdef CONFIG_PROC_FS
1589 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1590 {
1591         struct proc_dir_entry *p;
1592         struct sunrpc_net *sn;
1593
1594         sn = net_generic(net, sunrpc_net_id);
1595         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1596         if (cd->u.procfs.proc_ent == NULL)
1597                 goto out_nomem;
1598         cd->u.procfs.channel_ent = NULL;
1599         cd->u.procfs.content_ent = NULL;
1600
1601         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1602                              cd->u.procfs.proc_ent,
1603                              &cache_flush_operations_procfs, cd);
1604         cd->u.procfs.flush_ent = p;
1605         if (p == NULL)
1606                 goto out_nomem;
1607
1608         if (cd->cache_upcall || cd->cache_parse) {
1609                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1610                                      cd->u.procfs.proc_ent,
1611                                      &cache_file_operations_procfs, cd);
1612                 cd->u.procfs.channel_ent = p;
1613                 if (p == NULL)
1614                         goto out_nomem;
1615         }
1616         if (cd->cache_show) {
1617                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1618                                 cd->u.procfs.proc_ent,
1619                                 &content_file_operations_procfs, cd);
1620                 cd->u.procfs.content_ent = p;
1621                 if (p == NULL)
1622                         goto out_nomem;
1623         }
1624         return 0;
1625 out_nomem:
1626         remove_cache_proc_entries(cd, net);
1627         return -ENOMEM;
1628 }
1629 #else /* CONFIG_PROC_FS */
1630 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1631 {
1632         return 0;
1633 }
1634 #endif
1635
1636 void __init cache_initialize(void)
1637 {
1638         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1639 }
1640
1641 int cache_register_net(struct cache_detail *cd, struct net *net)
1642 {
1643         int ret;
1644
1645         sunrpc_init_cache_detail(cd);
1646         ret = create_cache_proc_entries(cd, net);
1647         if (ret)
1648                 sunrpc_destroy_cache_detail(cd);
1649         return ret;
1650 }
1651 EXPORT_SYMBOL_GPL(cache_register_net);
1652
1653 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1654 {
1655         remove_cache_proc_entries(cd, net);
1656         sunrpc_destroy_cache_detail(cd);
1657 }
1658 EXPORT_SYMBOL_GPL(cache_unregister_net);
1659
1660 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1661 {
1662         struct cache_detail *cd;
1663
1664         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1665         if (cd == NULL)
1666                 return ERR_PTR(-ENOMEM);
1667
1668         cd->hash_table = kzalloc(cd->hash_size * sizeof(struct cache_head *),
1669                                  GFP_KERNEL);
1670         if (cd->hash_table == NULL) {
1671                 kfree(cd);
1672                 return ERR_PTR(-ENOMEM);
1673         }
1674         cd->net = net;
1675         return cd;
1676 }
1677 EXPORT_SYMBOL_GPL(cache_create_net);
1678
1679 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1680 {
1681         kfree(cd->hash_table);
1682         kfree(cd);
1683 }
1684 EXPORT_SYMBOL_GPL(cache_destroy_net);
1685
1686 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1687                                  size_t count, loff_t *ppos)
1688 {
1689         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1690
1691         return cache_read(filp, buf, count, ppos, cd);
1692 }
1693
1694 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1695                                   size_t count, loff_t *ppos)
1696 {
1697         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1698
1699         return cache_write(filp, buf, count, ppos, cd);
1700 }
1701
1702 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1703 {
1704         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1705
1706         return cache_poll(filp, wait, cd);
1707 }
1708
1709 static long cache_ioctl_pipefs(struct file *filp,
1710                               unsigned int cmd, unsigned long arg)
1711 {
1712         struct inode *inode = filp->f_dentry->d_inode;
1713         struct cache_detail *cd = RPC_I(inode)->private;
1714
1715         return cache_ioctl(inode, filp, cmd, arg, cd);
1716 }
1717
1718 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1719 {
1720         struct cache_detail *cd = RPC_I(inode)->private;
1721
1722         return cache_open(inode, filp, cd);
1723 }
1724
1725 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1726 {
1727         struct cache_detail *cd = RPC_I(inode)->private;
1728
1729         return cache_release(inode, filp, cd);
1730 }
1731
1732 const struct file_operations cache_file_operations_pipefs = {
1733         .owner          = THIS_MODULE,
1734         .llseek         = no_llseek,
1735         .read           = cache_read_pipefs,
1736         .write          = cache_write_pipefs,
1737         .poll           = cache_poll_pipefs,
1738         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1739         .open           = cache_open_pipefs,
1740         .release        = cache_release_pipefs,
1741 };
1742
1743 static int content_open_pipefs(struct inode *inode, struct file *filp)
1744 {
1745         struct cache_detail *cd = RPC_I(inode)->private;
1746
1747         return content_open(inode, filp, cd);
1748 }
1749
1750 static int content_release_pipefs(struct inode *inode, struct file *filp)
1751 {
1752         struct cache_detail *cd = RPC_I(inode)->private;
1753
1754         return content_release(inode, filp, cd);
1755 }
1756
1757 const struct file_operations content_file_operations_pipefs = {
1758         .open           = content_open_pipefs,
1759         .read           = seq_read,
1760         .llseek         = seq_lseek,
1761         .release        = content_release_pipefs,
1762 };
1763
1764 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1765 {
1766         struct cache_detail *cd = RPC_I(inode)->private;
1767
1768         return open_flush(inode, filp, cd);
1769 }
1770
1771 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1772 {
1773         struct cache_detail *cd = RPC_I(inode)->private;
1774
1775         return release_flush(inode, filp, cd);
1776 }
1777
1778 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1779                             size_t count, loff_t *ppos)
1780 {
1781         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1782
1783         return read_flush(filp, buf, count, ppos, cd);
1784 }
1785
1786 static ssize_t write_flush_pipefs(struct file *filp,
1787                                   const char __user *buf,
1788                                   size_t count, loff_t *ppos)
1789 {
1790         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1791
1792         return write_flush(filp, buf, count, ppos, cd);
1793 }
1794
1795 const struct file_operations cache_flush_operations_pipefs = {
1796         .open           = open_flush_pipefs,
1797         .read           = read_flush_pipefs,
1798         .write          = write_flush_pipefs,
1799         .release        = release_flush_pipefs,
1800         .llseek         = no_llseek,
1801 };
1802
1803 int sunrpc_cache_register_pipefs(struct dentry *parent,
1804                                  const char *name, umode_t umode,
1805                                  struct cache_detail *cd)
1806 {
1807         struct qstr q;
1808         struct dentry *dir;
1809         int ret = 0;
1810
1811         q.name = name;
1812         q.len = strlen(name);
1813         q.hash = full_name_hash(q.name, q.len);
1814         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1815         if (!IS_ERR(dir))
1816                 cd->u.pipefs.dir = dir;
1817         else
1818                 ret = PTR_ERR(dir);
1819         return ret;
1820 }
1821 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1822
1823 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1824 {
1825         rpc_remove_cache_dir(cd->u.pipefs.dir);
1826         cd->u.pipefs.dir = NULL;
1827 }
1828 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1829