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 <linux/string_helpers.h>
  24 #include <asm/uaccess.h>
  25 #include <linux/poll.h>
  26 #include <linux/seq_file.h>
  27 #include <linux/proc_fs.h>
  28 #include <linux/net.h>
  29 #include <linux/workqueue.h>
  30 #include <linux/mutex.h>
  31 #include <linux/pagemap.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 #include "netns.h"
  38
  39 #define  RPCDBG_FACILITY RPCDBG_CACHE
  40
  41 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  42 static void cache_revisit_request(struct cache_head *item);
  43
  44 static void cache_init(struct cache_head *h, struct cache_detail *detail)
  45 {
  46         time_t now = seconds_since_boot();
  47         INIT_HLIST_NODE(&h->cache_list);
  48         h->flags = 0;
  49         kref_init(&h->ref);
  50         h->expiry_time = now + CACHE_NEW_EXPIRY;
  51         if (now <= detail->flush_time)
  52                 /* ensure it isn't already expired */
  53                 now = detail->flush_time + 1;
  54         h->last_refresh = now;
  55 }
  56
  57 static void cache_fresh_unlocked(struct cache_head *head,
  58                                 struct cache_detail *detail);
  59
  60 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  61                                        struct cache_head *key, int hash)
  62 {
  63         struct cache_head *new = NULL, *freeme = NULL, *tmp = NULL;
  64         struct hlist_head *head;
  65
  66         head = &detail->hash_table[hash];
  67
  68         read_lock(&detail->hash_lock);
  69
  70         hlist_for_each_entry(tmp, head, cache_list) {
  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, detail);
  91         detail->init(new, key);
  92
  93         write_lock(&detail->hash_lock);
  94
  95         /* check if entry appeared while we slept */
  96         hlist_for_each_entry(tmp, head, cache_list) {
  97                 if (detail->match(tmp, key)) {
  98                         if (cache_is_expired(detail, tmp)) {
  99                                 hlist_del_init(&tmp->cache_list);
 100                                 detail->entries --;
 101                                 freeme = tmp;
 102                                 break;
 103                         }
 104                         cache_get(tmp);
 105                         write_unlock(&detail->hash_lock);
 106                         cache_put(new, detail);
 107                         return tmp;
 108                 }
 109         }
 110
 111         hlist_add_head(&new->cache_list, head);
 112         detail->entries++;
 113         cache_get(new);
 114         write_unlock(&detail->hash_lock);
 115
 116         if (freeme) {
 117                 cache_fresh_unlocked(freeme, detail);
 118                 cache_put(freeme, detail);
 119         }
 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                                struct cache_detail *detail)
 129 {
 130         time_t now = seconds_since_boot();
 131         if (now <= detail->flush_time)
 132                 /* ensure it isn't immediately treated as expired */
 133                 now = detail->flush_time + 1;
 134         head->expiry_time = expiry;
 135         head->last_refresh = now;
 136         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 137         set_bit(CACHE_VALID, &head->flags);
 138 }
 139
 140 static void cache_fresh_unlocked(struct cache_head *head,
 141                                  struct cache_detail *detail)
 142 {
 143         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 144                 cache_revisit_request(head);
 145                 cache_dequeue(detail, head);
 146         }
 147 }
 148
 149 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 150                                        struct cache_head *new, struct cache_head *old, int hash)
 151 {
 152         /* The 'old' entry is to be replaced by 'new'.
 153          * If 'old' is not VALID, we update it directly,
 154          * otherwise we need to replace it
 155          */
 156         struct cache_head *tmp;
 157
 158         if (!test_bit(CACHE_VALID, &old->flags)) {
 159                 write_lock(&detail->hash_lock);
 160                 if (!test_bit(CACHE_VALID, &old->flags)) {
 161                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 162                                 set_bit(CACHE_NEGATIVE, &old->flags);
 163                         else
 164                                 detail->update(old, new);
 165                         cache_fresh_locked(old, new->expiry_time, detail);
 166                         write_unlock(&detail->hash_lock);
 167                         cache_fresh_unlocked(old, detail);
 168                         return old;
 169                 }
 170                 write_unlock(&detail->hash_lock);
 171         }
 172         /* We need to insert a new entry */
 173         tmp = detail->alloc();
 174         if (!tmp) {
 175                 cache_put(old, detail);
 176                 return NULL;
 177         }
 178         cache_init(tmp, detail);
 179         detail->init(tmp, old);
 180
 181         write_lock(&detail->hash_lock);
 182         if (test_bit(CACHE_NEGATIVE, &new->flags))
 183                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 184         else
 185                 detail->update(tmp, new);
 186         hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
 187         detail->entries++;
 188         cache_get(tmp);
 189         cache_fresh_locked(tmp, new->expiry_time, detail);
 190         cache_fresh_locked(old, 0, detail);
 191         write_unlock(&detail->hash_lock);
 192         cache_fresh_unlocked(tmp, detail);
 193         cache_fresh_unlocked(old, detail);
 194         cache_put(old, detail);
 195         return tmp;
 196 }
 197 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 198
 199 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 200 {
 201         if (cd->cache_upcall)
 202                 return cd->cache_upcall(cd, h);
 203         return sunrpc_cache_pipe_upcall(cd, h);
 204 }
 205
 206 static inline int cache_is_valid(struct cache_head *h)
 207 {
 208         if (!test_bit(CACHE_VALID, &h->flags))
 209                 return -EAGAIN;
 210         else {
 211                 /* entry is valid */
 212                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 213                         return -ENOENT;
 214                 else {
 215                         /*
 216                          * In combination with write barrier in
 217                          * sunrpc_cache_update, ensures that anyone
 218                          * using the cache entry after this sees the
 219                          * updated contents:
 220                          */
 221                         smp_rmb();
 222                         return 0;
 223                 }
 224         }
 225 }
 226
 227 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 228 {
 229         int rv;
 230
 231         write_lock(&detail->hash_lock);
 232         rv = cache_is_valid(h);
 233         if (rv == -EAGAIN) {
 234                 set_bit(CACHE_NEGATIVE, &h->flags);
 235                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
 236                                    detail);
 237                 rv = -ENOENT;
 238         }
 239         write_unlock(&detail->hash_lock);
 240         cache_fresh_unlocked(h, detail);
 241         return rv;
 242 }
 243
 244 /*
 245  * This is the generic cache management routine for all
 246  * the authentication caches.
 247  * It checks the currency of a cache item and will (later)
 248  * initiate an upcall to fill it if needed.
 249  *
 250  *
 251  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 252  * -EAGAIN if upcall is pending and request has been queued
 253  * -ETIMEDOUT if upcall failed or request could not be queue or
 254  *           upcall completed but item is still invalid (implying that
 255  *           the cache item has been replaced with a newer one).
 256  * -ENOENT if cache entry was negative
 257  */
 258 int cache_check(struct cache_detail *detail,
 259                     struct cache_head *h, struct cache_req *rqstp)
 260 {
 261         int rv;
 262         long refresh_age, age;
 263
 264         /* First decide return status as best we can */
 265         rv = cache_is_valid(h);
 266
 267         /* now see if we want to start an upcall */
 268         refresh_age = (h->expiry_time - h->last_refresh);
 269         age = seconds_since_boot() - h->last_refresh;
 270
 271         if (rqstp == NULL) {
 272                 if (rv == -EAGAIN)
 273                         rv = -ENOENT;
 274         } else if (rv == -EAGAIN ||
 275                    (h->expiry_time != 0 && age > refresh_age/2)) {
 276                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 277                                 refresh_age, age);
 278                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 279                         switch (cache_make_upcall(detail, h)) {
 280                         case -EINVAL:
 281                                 rv = try_to_negate_entry(detail, h);
 282                                 break;
 283                         case -EAGAIN:
 284                                 cache_fresh_unlocked(h, detail);
 285                                 break;
 286                         }
 287                 }
 288         }
 289
 290         if (rv == -EAGAIN) {
 291                 if (!cache_defer_req(rqstp, h)) {
 292                         /*
 293                          * Request was not deferred; handle it as best
 294                          * we can ourselves:
 295                          */
 296                         rv = cache_is_valid(h);
 297                         if (rv == -EAGAIN)
 298                                 rv = -ETIMEDOUT;
 299                 }
 300         }
 301         if (rv)
 302                 cache_put(h, detail);
 303         return rv;
 304 }
 305 EXPORT_SYMBOL_GPL(cache_check);
 306
 307 /*
 308  * caches need to be periodically cleaned.
 309  * For this we maintain a list of cache_detail and
 310  * a current pointer into that list and into the table
 311  * for that entry.
 312  *
 313  * Each time cache_clean 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 "RPC: 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                hlist_empty(&current_detail->hash_table[current_index]))
 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 = NULL;
 434                 struct cache_detail *d;
 435                 struct hlist_head *head;
 436                 struct hlist_node *tmp;
 437
 438                 write_lock(&current_detail->hash_lock);
 439
 440                 /* Ok, now to clean this strand */
 441
 442                 head = &current_detail->hash_table[current_index];
 443                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 444                         if (current_detail->nextcheck > ch->expiry_time)
 445                                 current_detail->nextcheck = ch->expiry_time+1;
 446                         if (!cache_is_expired(current_detail, ch))
 447                                 continue;
 448
 449                         hlist_del_init(&ch->cache_list);
 450                         current_detail->entries--;
 451                         rv = 1;
 452                         break;
 453                 }
 454
 455                 write_unlock(&current_detail->hash_lock);
 456                 d = current_detail;
 457                 if (!ch)
 458                         current_index ++;
 459                 spin_unlock(&cache_list_lock);
 460                 if (ch) {
 461                         set_bit(CACHE_CLEANED, &ch->flags);
 462                         cache_fresh_unlocked(ch, d);
 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         time_t now = seconds_since_boot();
 505         if (detail->flush_time >= now)
 506                 now = detail->flush_time + 1;
 507         /* 'now' is the maximum value any 'last_refresh' can have */
 508         detail->flush_time = now;
 509         detail->nextcheck = seconds_since_boot();
 510         cache_flush();
 511 }
 512 EXPORT_SYMBOL_GPL(cache_purge);
 513
 514
 515 /*
 516  * Deferral and Revisiting of Requests.
 517  *
 518  * If a cache lookup finds a pending entry, we
 519  * need to defer the request and revisit it later.
 520  * All deferred requests are stored in a hash table,
 521  * indexed by "struct cache_head *".
 522  * As it may be wasteful to store a whole request
 523  * structure, we allow the request to provide a
 524  * deferred form, which must contain a
 525  * 'struct cache_deferred_req'
 526  * This cache_deferred_req contains a method to allow
 527  * it to be revisited when cache info is available
 528  */
 529
 530 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 531 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 532
 533 #define DFR_MAX 300     /* ??? */
 534
 535 static DEFINE_SPINLOCK(cache_defer_lock);
 536 static LIST_HEAD(cache_defer_list);
 537 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 538 static int cache_defer_cnt;
 539
 540 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 541 {
 542         hlist_del_init(&dreq->hash);
 543         if (!list_empty(&dreq->recent)) {
 544                 list_del_init(&dreq->recent);
 545                 cache_defer_cnt--;
 546         }
 547 }
 548
 549 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 550 {
 551         int hash = DFR_HASH(item);
 552
 553         INIT_LIST_HEAD(&dreq->recent);
 554         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 555 }
 556
 557 static void setup_deferral(struct cache_deferred_req *dreq,
 558                            struct cache_head *item,
 559                            int count_me)
 560 {
 561
 562         dreq->item = item;
 563
 564         spin_lock(&cache_defer_lock);
 565
 566         __hash_deferred_req(dreq, item);
 567
 568         if (count_me) {
 569                 cache_defer_cnt++;
 570                 list_add(&dreq->recent, &cache_defer_list);
 571         }
 572
 573         spin_unlock(&cache_defer_lock);
 574
 575 }
 576
 577 struct thread_deferred_req {
 578         struct cache_deferred_req handle;
 579         struct completion completion;
 580 };
 581
 582 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 583 {
 584         struct thread_deferred_req *dr =
 585                 container_of(dreq, struct thread_deferred_req, handle);
 586         complete(&dr->completion);
 587 }
 588
 589 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 590 {
 591         struct thread_deferred_req sleeper;
 592         struct cache_deferred_req *dreq = &sleeper.handle;
 593
 594         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 595         dreq->revisit = cache_restart_thread;
 596
 597         setup_deferral(dreq, item, 0);
 598
 599         if (!test_bit(CACHE_PENDING, &item->flags) ||
 600             wait_for_completion_interruptible_timeout(
 601                     &sleeper.completion, req->thread_wait) <= 0) {
 602                 /* The completion wasn't completed, so we need
 603                  * to clean up
 604                  */
 605                 spin_lock(&cache_defer_lock);
 606                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 607                         __unhash_deferred_req(&sleeper.handle);
 608                         spin_unlock(&cache_defer_lock);
 609                 } else {
 610                         /* cache_revisit_request already removed
 611                          * this from the hash table, but hasn't
 612                          * called ->revisit yet.  It will very soon
 613                          * and we need to wait for it.
 614                          */
 615                         spin_unlock(&cache_defer_lock);
 616                         wait_for_completion(&sleeper.completion);
 617                 }
 618         }
 619 }
 620
 621 static void cache_limit_defers(void)
 622 {
 623         /* Make sure we haven't exceed the limit of allowed deferred
 624          * requests.
 625          */
 626         struct cache_deferred_req *discard = NULL;
 627
 628         if (cache_defer_cnt <= DFR_MAX)
 629                 return;
 630
 631         spin_lock(&cache_defer_lock);
 632
 633         /* Consider removing either the first or the last */
 634         if (cache_defer_cnt > DFR_MAX) {
 635                 if (prandom_u32() & 1)
 636                         discard = list_entry(cache_defer_list.next,
 637                                              struct cache_deferred_req, recent);
 638                 else
 639                         discard = list_entry(cache_defer_list.prev,
 640                                              struct cache_deferred_req, recent);
 641                 __unhash_deferred_req(discard);
 642         }
 643         spin_unlock(&cache_defer_lock);
 644         if (discard)
 645                 discard->revisit(discard, 1);
 646 }
 647
 648 /* Return true if and only if a deferred request is queued. */
 649 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 650 {
 651         struct cache_deferred_req *dreq;
 652
 653         if (req->thread_wait) {
 654                 cache_wait_req(req, item);
 655                 if (!test_bit(CACHE_PENDING, &item->flags))
 656                         return false;
 657         }
 658         dreq = req->defer(req);
 659         if (dreq == NULL)
 660                 return false;
 661         setup_deferral(dreq, item, 1);
 662         if (!test_bit(CACHE_PENDING, &item->flags))
 663                 /* Bit could have been cleared before we managed to
 664                  * set up the deferral, so need to revisit just in case
 665                  */
 666                 cache_revisit_request(item);
 667
 668         cache_limit_defers();
 669         return true;
 670 }
 671
 672 static void cache_revisit_request(struct cache_head *item)
 673 {
 674         struct cache_deferred_req *dreq;
 675         struct list_head pending;
 676         struct hlist_node *tmp;
 677         int hash = DFR_HASH(item);
 678
 679         INIT_LIST_HEAD(&pending);
 680         spin_lock(&cache_defer_lock);
 681
 682         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 683                 if (dreq->item == item) {
 684                         __unhash_deferred_req(dreq);
 685                         list_add(&dreq->recent, &pending);
 686                 }
 687
 688         spin_unlock(&cache_defer_lock);
 689
 690         while (!list_empty(&pending)) {
 691                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 692                 list_del_init(&dreq->recent);
 693                 dreq->revisit(dreq, 0);
 694         }
 695 }
 696
 697 void cache_clean_deferred(void *owner)
 698 {
 699         struct cache_deferred_req *dreq, *tmp;
 700         struct list_head pending;
 701
 702
 703         INIT_LIST_HEAD(&pending);
 704         spin_lock(&cache_defer_lock);
 705
 706         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 707                 if (dreq->owner == owner) {
 708                         __unhash_deferred_req(dreq);
 709                         list_add(&dreq->recent, &pending);
 710                 }
 711         }
 712         spin_unlock(&cache_defer_lock);
 713
 714         while (!list_empty(&pending)) {
 715                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 716                 list_del_init(&dreq->recent);
 717                 dreq->revisit(dreq, 1);
 718         }
 719 }
 720
 721 /*
 722  * communicate with user-space
 723  *
 724  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 725  * On read, you get a full request, or block.
 726  * On write, an update request is processed.
 727  * Poll works if anything to read, and always allows write.
 728  *
 729  * Implemented by linked list of requests.  Each open file has
 730  * a ->private that also exists in this list.  New requests are added
 731  * to the end and may wakeup and preceding readers.
 732  * New readers are added to the head.  If, on read, an item is found with
 733  * CACHE_UPCALLING clear, we free it from the list.
 734  *
 735  */
 736
 737 static DEFINE_SPINLOCK(queue_lock);
 738 static DEFINE_MUTEX(queue_io_mutex);
 739
 740 struct cache_queue {
 741         struct list_head        list;
 742         int                     reader; /* if 0, then request */
 743 };
 744 struct cache_request {
 745         struct cache_queue      q;
 746         struct cache_head       *item;
 747         char                    * buf;
 748         int                     len;
 749         int                     readers;
 750 };
 751 struct cache_reader {
 752         struct cache_queue      q;
 753         int                     offset; /* if non-0, we have a refcnt on next request */
 754 };
 755
 756 static int cache_request(struct cache_detail *detail,
 757                                struct cache_request *crq)
 758 {
 759         char *bp = crq->buf;
 760         int len = PAGE_SIZE;
 761
 762         detail->cache_request(detail, crq->item, &bp, &len);
 763         if (len < 0)
 764                 return -EAGAIN;
 765         return PAGE_SIZE - len;
 766 }
 767
 768 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 769                           loff_t *ppos, struct cache_detail *cd)
 770 {
 771         struct cache_reader *rp = filp->private_data;
 772         struct cache_request *rq;
 773         struct inode *inode = file_inode(filp);
 774         int err;
 775
 776         if (count == 0)
 777                 return 0;
 778
 779         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 780                               * readers on this file */
 781  again:
 782         spin_lock(&queue_lock);
 783         /* need to find next request */
 784         while (rp->q.list.next != &cd->queue &&
 785                list_entry(rp->q.list.next, struct cache_queue, list)
 786                ->reader) {
 787                 struct list_head *next = rp->q.list.next;
 788                 list_move(&rp->q.list, next);
 789         }
 790         if (rp->q.list.next == &cd->queue) {
 791                 spin_unlock(&queue_lock);
 792                 mutex_unlock(&inode->i_mutex);
 793                 WARN_ON_ONCE(rp->offset);
 794                 return 0;
 795         }
 796         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 797         WARN_ON_ONCE(rq->q.reader);
 798         if (rp->offset == 0)
 799                 rq->readers++;
 800         spin_unlock(&queue_lock);
 801
 802         if (rq->len == 0) {
 803                 err = cache_request(cd, rq);
 804                 if (err < 0)
 805                         goto out;
 806                 rq->len = err;
 807         }
 808
 809         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 810                 err = -EAGAIN;
 811                 spin_lock(&queue_lock);
 812                 list_move(&rp->q.list, &rq->q.list);
 813                 spin_unlock(&queue_lock);
 814         } else {
 815                 if (rp->offset + count > rq->len)
 816                         count = rq->len - rp->offset;
 817                 err = -EFAULT;
 818                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 819                         goto out;
 820                 rp->offset += count;
 821                 if (rp->offset >= rq->len) {
 822                         rp->offset = 0;
 823                         spin_lock(&queue_lock);
 824                         list_move(&rp->q.list, &rq->q.list);
 825                         spin_unlock(&queue_lock);
 826                 }
 827                 err = 0;
 828         }
 829  out:
 830         if (rp->offset == 0) {
 831                 /* need to release rq */
 832                 spin_lock(&queue_lock);
 833                 rq->readers--;
 834                 if (rq->readers == 0 &&
 835                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 836                         list_del(&rq->q.list);
 837                         spin_unlock(&queue_lock);
 838                         cache_put(rq->item, cd);
 839                         kfree(rq->buf);
 840                         kfree(rq);
 841                 } else
 842                         spin_unlock(&queue_lock);
 843         }
 844         if (err == -EAGAIN)
 845                 goto again;
 846         mutex_unlock(&inode->i_mutex);
 847         return err ? err :  count;
 848 }
 849
 850 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 851                                  size_t count, struct cache_detail *cd)
 852 {
 853         ssize_t ret;
 854
 855         if (count == 0)
 856                 return -EINVAL;
 857         if (copy_from_user(kaddr, buf, count))
 858                 return -EFAULT;
 859         kaddr[count] = '\0';
 860         ret = cd->cache_parse(cd, kaddr, count);
 861         if (!ret)
 862                 ret = count;
 863         return ret;
 864 }
 865
 866 static ssize_t cache_slow_downcall(const char __user *buf,
 867                                    size_t count, struct cache_detail *cd)
 868 {
 869         static char write_buf[8192]; /* protected by queue_io_mutex */
 870         ssize_t ret = -EINVAL;
 871
 872         if (count >= sizeof(write_buf))
 873                 goto out;
 874         mutex_lock(&queue_io_mutex);
 875         ret = cache_do_downcall(write_buf, buf, count, cd);
 876         mutex_unlock(&queue_io_mutex);
 877 out:
 878         return ret;
 879 }
 880
 881 static ssize_t cache_downcall(struct address_space *mapping,
 882                               const char __user *buf,
 883                               size_t count, struct cache_detail *cd)
 884 {
 885         struct page *page;
 886         char *kaddr;
 887         ssize_t ret = -ENOMEM;
 888
 889         if (count >= PAGE_CACHE_SIZE)
 890                 goto out_slow;
 891
 892         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 893         if (!page)
 894                 goto out_slow;
 895
 896         kaddr = kmap(page);
 897         ret = cache_do_downcall(kaddr, buf, count, cd);
 898         kunmap(page);
 899         unlock_page(page);
 900         page_cache_release(page);
 901         return ret;
 902 out_slow:
 903         return cache_slow_downcall(buf, count, cd);
 904 }
 905
 906 static ssize_t cache_write(struct file *filp, const char __user *buf,
 907                            size_t count, loff_t *ppos,
 908                            struct cache_detail *cd)
 909 {
 910         struct address_space *mapping = filp->f_mapping;
 911         struct inode *inode = file_inode(filp);
 912         ssize_t ret = -EINVAL;
 913
 914         if (!cd->cache_parse)
 915                 goto out;
 916
 917         mutex_lock(&inode->i_mutex);
 918         ret = cache_downcall(mapping, buf, count, cd);
 919         mutex_unlock(&inode->i_mutex);
 920 out:
 921         return ret;
 922 }
 923
 924 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 925
 926 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 927                                struct cache_detail *cd)
 928 {
 929         unsigned int mask;
 930         struct cache_reader *rp = filp->private_data;
 931         struct cache_queue *cq;
 932
 933         poll_wait(filp, &queue_wait, wait);
 934
 935         /* alway allow write */
 936         mask = POLLOUT | POLLWRNORM;
 937
 938         if (!rp)
 939                 return mask;
 940
 941         spin_lock(&queue_lock);
 942
 943         for (cq= &rp->q; &cq->list != &cd->queue;
 944              cq = list_entry(cq->list.next, struct cache_queue, list))
 945                 if (!cq->reader) {
 946                         mask |= POLLIN | POLLRDNORM;
 947                         break;
 948                 }
 949         spin_unlock(&queue_lock);
 950         return mask;
 951 }
 952
 953 static int cache_ioctl(struct inode *ino, struct file *filp,
 954                        unsigned int cmd, unsigned long arg,
 955                        struct cache_detail *cd)
 956 {
 957         int len = 0;
 958         struct cache_reader *rp = filp->private_data;
 959         struct cache_queue *cq;
 960
 961         if (cmd != FIONREAD || !rp)
 962                 return -EINVAL;
 963
 964         spin_lock(&queue_lock);
 965
 966         /* only find the length remaining in current request,
 967          * or the length of the next request
 968          */
 969         for (cq= &rp->q; &cq->list != &cd->queue;
 970              cq = list_entry(cq->list.next, struct cache_queue, list))
 971                 if (!cq->reader) {
 972                         struct cache_request *cr =
 973                                 container_of(cq, struct cache_request, q);
 974                         len = cr->len - rp->offset;
 975                         break;
 976                 }
 977         spin_unlock(&queue_lock);
 978
 979         return put_user(len, (int __user *)arg);
 980 }
 981
 982 static int cache_open(struct inode *inode, struct file *filp,
 983                       struct cache_detail *cd)
 984 {
 985         struct cache_reader *rp = NULL;
 986
 987         if (!cd || !try_module_get(cd->owner))
 988                 return -EACCES;
 989         nonseekable_open(inode, filp);
 990         if (filp->f_mode & FMODE_READ) {
 991                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 992                 if (!rp) {
 993                         module_put(cd->owner);
 994                         return -ENOMEM;
 995                 }
 996                 rp->offset = 0;
 997                 rp->q.reader = 1;
 998                 atomic_inc(&cd->readers);
 999                 spin_lock(&queue_lock);
1000                 list_add(&rp->q.list, &cd->queue);
1001                 spin_unlock(&queue_lock);
1002         }
1003         filp->private_data = rp;
1004         return 0;
1005 }
1006
1007 static int cache_release(struct inode *inode, struct file *filp,
1008                          struct cache_detail *cd)
1009 {
1010         struct cache_reader *rp = filp->private_data;
1011
1012         if (rp) {
1013                 spin_lock(&queue_lock);
1014                 if (rp->offset) {
1015                         struct cache_queue *cq;
1016                         for (cq= &rp->q; &cq->list != &cd->queue;
1017                              cq = list_entry(cq->list.next, struct cache_queue, list))
1018                                 if (!cq->reader) {
1019                                         container_of(cq, struct cache_request, q)
1020                                                 ->readers--;
1021                                         break;
1022                                 }
1023                         rp->offset = 0;
1024                 }
1025                 list_del(&rp->q.list);
1026                 spin_unlock(&queue_lock);
1027
1028                 filp->private_data = NULL;
1029                 kfree(rp);
1030
1031                 cd->last_close = seconds_since_boot();
1032                 atomic_dec(&cd->readers);
1033         }
1034         module_put(cd->owner);
1035         return 0;
1036 }
1037
1038
1039
1040 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1041 {
1042         struct cache_queue *cq, *tmp;
1043         struct cache_request *cr;
1044         struct list_head dequeued;
1045
1046         INIT_LIST_HEAD(&dequeued);
1047         spin_lock(&queue_lock);
1048         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1049                 if (!cq->reader) {
1050                         cr = container_of(cq, struct cache_request, q);
1051                         if (cr->item != ch)
1052                                 continue;
1053                         if (test_bit(CACHE_PENDING, &ch->flags))
1054                                 /* Lost a race and it is pending again */
1055                                 break;
1056                         if (cr->readers != 0)
1057                                 continue;
1058                         list_move(&cr->q.list, &dequeued);
1059                 }
1060         spin_unlock(&queue_lock);
1061         while (!list_empty(&dequeued)) {
1062                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1063                 list_del(&cr->q.list);
1064                 cache_put(cr->item, detail);
1065                 kfree(cr->buf);
1066                 kfree(cr);
1067         }
1068 }
1069
1070 /*
1071  * Support routines for text-based upcalls.
1072  * Fields are separated by spaces.
1073  * Fields are either mangled to quote space tab newline slosh with slosh
1074  * or a hexified with a leading \x
1075  * Record is terminated with newline.
1076  *
1077  */
1078
1079 void qword_add(char **bpp, int *lp, char *str)
1080 {
1081         char *bp = *bpp;
1082         int len = *lp;
1083         int ret;
1084
1085         if (len < 0) return;
1086
1087         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1088         if (ret >= len) {
1089                 bp += len;
1090                 len = -1;
1091         } else {
1092                 bp += ret;
1093                 len -= ret;
1094                 *bp++ = ' ';
1095                 len--;
1096         }
1097         *bpp = bp;
1098         *lp = len;
1099 }
1100 EXPORT_SYMBOL_GPL(qword_add);
1101
1102 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1103 {
1104         char *bp = *bpp;
1105         int len = *lp;
1106
1107         if (len < 0) return;
1108
1109         if (len > 2) {
1110                 *bp++ = '\\';
1111                 *bp++ = 'x';
1112                 len -= 2;
1113                 while (blen && len >= 2) {
1114                         bp = hex_byte_pack(bp, *buf++);
1115                         len -= 2;
1116                         blen--;
1117                 }
1118         }
1119         if (blen || len<1) len = -1;
1120         else {
1121                 *bp++ = ' ';
1122                 len--;
1123         }
1124         *bpp = bp;
1125         *lp = len;
1126 }
1127 EXPORT_SYMBOL_GPL(qword_addhex);
1128
1129 static void warn_no_listener(struct cache_detail *detail)
1130 {
1131         if (detail->last_warn != detail->last_close) {
1132                 detail->last_warn = detail->last_close;
1133                 if (detail->warn_no_listener)
1134                         detail->warn_no_listener(detail, detail->last_close != 0);
1135         }
1136 }
1137
1138 static bool cache_listeners_exist(struct cache_detail *detail)
1139 {
1140         if (atomic_read(&detail->readers))
1141                 return true;
1142         if (detail->last_close == 0)
1143                 /* This cache was never opened */
1144                 return false;
1145         if (detail->last_close < seconds_since_boot() - 30)
1146                 /*
1147                  * We allow for the possibility that someone might
1148                  * restart a userspace daemon without restarting the
1149                  * server; but after 30 seconds, we give up.
1150                  */
1151                  return false;
1152         return true;
1153 }
1154
1155 /*
1156  * register an upcall request to user-space and queue it up for read() by the
1157  * upcall daemon.
1158  *
1159  * Each request is at most one page long.
1160  */
1161 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1162 {
1163
1164         char *buf;
1165         struct cache_request *crq;
1166         int ret = 0;
1167
1168         if (!detail->cache_request)
1169                 return -EINVAL;
1170
1171         if (!cache_listeners_exist(detail)) {
1172                 warn_no_listener(detail);
1173                 return -EINVAL;
1174         }
1175         if (test_bit(CACHE_CLEANED, &h->flags))
1176                 /* Too late to make an upcall */
1177                 return -EAGAIN;
1178
1179         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1180         if (!buf)
1181                 return -EAGAIN;
1182
1183         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1184         if (!crq) {
1185                 kfree(buf);
1186                 return -EAGAIN;
1187         }
1188
1189         crq->q.reader = 0;
1190         crq->buf = buf;
1191         crq->len = 0;
1192         crq->readers = 0;
1193         spin_lock(&queue_lock);
1194         if (test_bit(CACHE_PENDING, &h->flags)) {
1195                 crq->item = cache_get(h);
1196                 list_add_tail(&crq->q.list, &detail->queue);
1197         } else
1198                 /* Lost a race, no longer PENDING, so don't enqueue */
1199                 ret = -EAGAIN;
1200         spin_unlock(&queue_lock);
1201         wake_up(&queue_wait);
1202         if (ret == -EAGAIN) {
1203                 kfree(buf);
1204                 kfree(crq);
1205         }
1206         return ret;
1207 }
1208 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1209
1210 /*
1211  * parse a message from user-space and pass it
1212  * to an appropriate cache
1213  * Messages are, like requests, separated into fields by
1214  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1215  *
1216  * Message is
1217  *   reply cachename expiry key ... content....
1218  *
1219  * key and content are both parsed by cache
1220  */
1221
1222 int qword_get(char **bpp, char *dest, int bufsize)
1223 {
1224         /* return bytes copied, or -1 on error */
1225         char *bp = *bpp;
1226         int len = 0;
1227
1228         while (*bp == ' ') bp++;
1229
1230         if (bp[0] == '\\' && bp[1] == 'x') {
1231                 /* HEX STRING */
1232                 bp += 2;
1233                 while (len < bufsize - 1) {
1234                         int h, l;
1235
1236                         h = hex_to_bin(bp[0]);
1237                         if (h < 0)
1238                                 break;
1239
1240                         l = hex_to_bin(bp[1]);
1241                         if (l < 0)
1242                                 break;
1243
1244                         *dest++ = (h << 4) | l;
1245                         bp += 2;
1246                         len++;
1247                 }
1248         } else {
1249                 /* text with \nnn octal quoting */
1250                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1251                         if (*bp == '\\' &&
1252                             isodigit(bp[1]) && (bp[1] <= '3') &&
1253                             isodigit(bp[2]) &&
1254                             isodigit(bp[3])) {
1255                                 int byte = (*++bp -'0');
1256                                 bp++;
1257                                 byte = (byte << 3) | (*bp++ - '0');
1258                                 byte = (byte << 3) | (*bp++ - '0');
1259                                 *dest++ = byte;
1260                                 len++;
1261                         } else {
1262                                 *dest++ = *bp++;
1263                                 len++;
1264                         }
1265                 }
1266         }
1267
1268         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1269                 return -1;
1270         while (*bp == ' ') bp++;
1271         *bpp = bp;
1272         *dest = '\0';
1273         return len;
1274 }
1275 EXPORT_SYMBOL_GPL(qword_get);
1276
1277
1278 /*
1279  * support /proc/sunrpc/cache/$CACHENAME/content
1280  * as a seqfile.
1281  * We call ->cache_show passing NULL for the item to
1282  * get a header, then pass each real item in the cache
1283  */
1284
1285 void *cache_seq_start(struct seq_file *m, loff_t *pos)
1286         __acquires(cd->hash_lock)
1287 {
1288         loff_t n = *pos;
1289         unsigned int hash, entry;
1290         struct cache_head *ch;
1291         struct cache_detail *cd = m->private;
1292
1293         read_lock(&cd->hash_lock);
1294         if (!n--)
1295                 return SEQ_START_TOKEN;
1296         hash = n >> 32;
1297         entry = n & ((1LL<<32) - 1);
1298
1299         hlist_for_each_entry(ch, &cd->hash_table[hash], cache_list)
1300                 if (!entry--)
1301                         return ch;
1302         n &= ~((1LL<<32) - 1);
1303         do {
1304                 hash++;
1305                 n += 1LL<<32;
1306         } while(hash < cd->hash_size &&
1307                 hlist_empty(&cd->hash_table[hash]));
1308         if (hash >= cd->hash_size)
1309                 return NULL;
1310         *pos = n+1;
1311         return hlist_entry_safe(cd->hash_table[hash].first,
1312                                 struct cache_head, cache_list);
1313 }
1314 EXPORT_SYMBOL_GPL(cache_seq_start);
1315
1316 void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1317 {
1318         struct cache_head *ch = p;
1319         int hash = (*pos >> 32);
1320         struct cache_detail *cd = m->private;
1321
1322         if (p == SEQ_START_TOKEN)
1323                 hash = 0;
1324         else if (ch->cache_list.next == NULL) {
1325                 hash++;
1326                 *pos += 1LL<<32;
1327         } else {
1328                 ++*pos;
1329                 return hlist_entry_safe(ch->cache_list.next,
1330                                         struct cache_head, cache_list);
1331         }
1332         *pos &= ~((1LL<<32) - 1);
1333         while (hash < cd->hash_size &&
1334                hlist_empty(&cd->hash_table[hash])) {
1335                 hash++;
1336                 *pos += 1LL<<32;
1337         }
1338         if (hash >= cd->hash_size)
1339                 return NULL;
1340         ++*pos;
1341         return hlist_entry_safe(cd->hash_table[hash].first,
1342                                 struct cache_head, cache_list);
1343 }
1344 EXPORT_SYMBOL_GPL(cache_seq_next);
1345
1346 void cache_seq_stop(struct seq_file *m, void *p)
1347         __releases(cd->hash_lock)
1348 {
1349         struct cache_detail *cd = m->private;
1350         read_unlock(&cd->hash_lock);
1351 }
1352 EXPORT_SYMBOL_GPL(cache_seq_stop);
1353
1354 static int c_show(struct seq_file *m, void *p)
1355 {
1356         struct cache_head *cp = p;
1357         struct cache_detail *cd = m->private;
1358
1359         if (p == SEQ_START_TOKEN)
1360                 return cd->cache_show(m, cd, NULL);
1361
1362         ifdebug(CACHE)
1363                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1364                            convert_to_wallclock(cp->expiry_time),
1365                            atomic_read(&cp->ref.refcount), cp->flags);
1366         cache_get(cp);
1367         if (cache_check(cd, cp, NULL))
1368                 /* cache_check does a cache_put on failure */
1369                 seq_printf(m, "# ");
1370         else {
1371                 if (cache_is_expired(cd, cp))
1372                         seq_printf(m, "# ");
1373                 cache_put(cp, cd);
1374         }
1375
1376         return cd->cache_show(m, cd, cp);
1377 }
1378
1379 static const struct seq_operations cache_content_op = {
1380         .start  = cache_seq_start,
1381         .next   = cache_seq_next,
1382         .stop   = cache_seq_stop,
1383         .show   = c_show,
1384 };
1385
1386 static int content_open(struct inode *inode, struct file *file,
1387                         struct cache_detail *cd)
1388 {
1389         struct seq_file *seq;
1390         int err;
1391
1392         if (!cd || !try_module_get(cd->owner))
1393                 return -EACCES;
1394
1395         err = seq_open(file, &cache_content_op);
1396         if (err) {
1397                 module_put(cd->owner);
1398                 return err;
1399         }
1400
1401         seq = file->private_data;
1402         seq->private = cd;
1403         return 0;
1404 }
1405
1406 static int content_release(struct inode *inode, struct file *file,
1407                 struct cache_detail *cd)
1408 {
1409         int ret = seq_release(inode, file);
1410         module_put(cd->owner);
1411         return ret;
1412 }
1413
1414 static int open_flush(struct inode *inode, struct file *file,
1415                         struct cache_detail *cd)
1416 {
1417         if (!cd || !try_module_get(cd->owner))
1418                 return -EACCES;
1419         return nonseekable_open(inode, file);
1420 }
1421
1422 static int release_flush(struct inode *inode, struct file *file,
1423                         struct cache_detail *cd)
1424 {
1425         module_put(cd->owner);
1426         return 0;
1427 }
1428
1429 static ssize_t read_flush(struct file *file, char __user *buf,
1430                           size_t count, loff_t *ppos,
1431                           struct cache_detail *cd)
1432 {
1433         char tbuf[22];
1434         unsigned long p = *ppos;
1435         size_t len;
1436
1437         snprintf(tbuf, sizeof(tbuf), "%lu\n", convert_to_wallclock(cd->flush_time));
1438         len = strlen(tbuf);
1439         if (p >= len)
1440                 return 0;
1441         len -= p;
1442         if (len > count)
1443                 len = count;
1444         if (copy_to_user(buf, (void*)(tbuf+p), len))
1445                 return -EFAULT;
1446         *ppos += len;
1447         return len;
1448 }
1449
1450 static ssize_t write_flush(struct file *file, const char __user *buf,
1451                            size_t count, loff_t *ppos,
1452                            struct cache_detail *cd)
1453 {
1454         char tbuf[20];
1455         char *bp, *ep;
1456         time_t then, now;
1457
1458         if (*ppos || count > sizeof(tbuf)-1)
1459                 return -EINVAL;
1460         if (copy_from_user(tbuf, buf, count))
1461                 return -EFAULT;
1462         tbuf[count] = 0;
1463         simple_strtoul(tbuf, &ep, 0);
1464         if (*ep && *ep != '\n')
1465                 return -EINVAL;
1466
1467         bp = tbuf;
1468         then = get_expiry(&bp);
1469         now = seconds_since_boot();
1470         cd->nextcheck = now;
1471         /* Can only set flush_time to 1 second beyond "now", or
1472          * possibly 1 second beyond flushtime.  This is because
1473          * flush_time never goes backwards so it mustn't get too far
1474          * ahead of time.
1475          */
1476         if (then >= now) {
1477                 /* Want to flush everything, so behave like cache_purge() */
1478                 if (cd->flush_time >= now)
1479                         now = cd->flush_time + 1;
1480                 then = now;
1481         }
1482
1483         cd->flush_time = then;
1484         cache_flush();
1485
1486         *ppos += count;
1487         return count;
1488 }
1489
1490 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1491                                  size_t count, loff_t *ppos)
1492 {
1493         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1494
1495         return cache_read(filp, buf, count, ppos, cd);
1496 }
1497
1498 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1499                                   size_t count, loff_t *ppos)
1500 {
1501         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1502
1503         return cache_write(filp, buf, count, ppos, cd);
1504 }
1505
1506 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1507 {
1508         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1509
1510         return cache_poll(filp, wait, cd);
1511 }
1512
1513 static long cache_ioctl_procfs(struct file *filp,
1514                                unsigned int cmd, unsigned long arg)
1515 {
1516         struct inode *inode = file_inode(filp);
1517         struct cache_detail *cd = PDE_DATA(inode);
1518
1519         return cache_ioctl(inode, filp, cmd, arg, cd);
1520 }
1521
1522 static int cache_open_procfs(struct inode *inode, struct file *filp)
1523 {
1524         struct cache_detail *cd = PDE_DATA(inode);
1525
1526         return cache_open(inode, filp, cd);
1527 }
1528
1529 static int cache_release_procfs(struct inode *inode, struct file *filp)
1530 {
1531         struct cache_detail *cd = PDE_DATA(inode);
1532
1533         return cache_release(inode, filp, cd);
1534 }
1535
1536 static const struct file_operations cache_file_operations_procfs = {
1537         .owner          = THIS_MODULE,
1538         .llseek         = no_llseek,
1539         .read           = cache_read_procfs,
1540         .write          = cache_write_procfs,
1541         .poll           = cache_poll_procfs,
1542         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1543         .open           = cache_open_procfs,
1544         .release        = cache_release_procfs,
1545 };
1546
1547 static int content_open_procfs(struct inode *inode, struct file *filp)
1548 {
1549         struct cache_detail *cd = PDE_DATA(inode);
1550
1551         return content_open(inode, filp, cd);
1552 }
1553
1554 static int content_release_procfs(struct inode *inode, struct file *filp)
1555 {
1556         struct cache_detail *cd = PDE_DATA(inode);
1557
1558         return content_release(inode, filp, cd);
1559 }
1560
1561 static const struct file_operations content_file_operations_procfs = {
1562         .open           = content_open_procfs,
1563         .read           = seq_read,
1564         .llseek         = seq_lseek,
1565         .release        = content_release_procfs,
1566 };
1567
1568 static int open_flush_procfs(struct inode *inode, struct file *filp)
1569 {
1570         struct cache_detail *cd = PDE_DATA(inode);
1571
1572         return open_flush(inode, filp, cd);
1573 }
1574
1575 static int release_flush_procfs(struct inode *inode, struct file *filp)
1576 {
1577         struct cache_detail *cd = PDE_DATA(inode);
1578
1579         return release_flush(inode, filp, cd);
1580 }
1581
1582 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1583                             size_t count, loff_t *ppos)
1584 {
1585         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1586
1587         return read_flush(filp, buf, count, ppos, cd);
1588 }
1589
1590 static ssize_t write_flush_procfs(struct file *filp,
1591                                   const char __user *buf,
1592                                   size_t count, loff_t *ppos)
1593 {
1594         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1595
1596         return write_flush(filp, buf, count, ppos, cd);
1597 }
1598
1599 static const struct file_operations cache_flush_operations_procfs = {
1600         .open           = open_flush_procfs,
1601         .read           = read_flush_procfs,
1602         .write          = write_flush_procfs,
1603         .release        = release_flush_procfs,
1604         .llseek         = no_llseek,
1605 };
1606
1607 static void remove_cache_proc_entries(struct cache_detail *cd, struct net *net)
1608 {
1609         struct sunrpc_net *sn;
1610
1611         if (cd->u.procfs.proc_ent == NULL)
1612                 return;
1613         if (cd->u.procfs.flush_ent)
1614                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1615         if (cd->u.procfs.channel_ent)
1616                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1617         if (cd->u.procfs.content_ent)
1618                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1619         cd->u.procfs.proc_ent = NULL;
1620         sn = net_generic(net, sunrpc_net_id);
1621         remove_proc_entry(cd->name, sn->proc_net_rpc);
1622 }
1623
1624 #ifdef CONFIG_PROC_FS
1625 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1626 {
1627         struct proc_dir_entry *p;
1628         struct sunrpc_net *sn;
1629
1630         sn = net_generic(net, sunrpc_net_id);
1631         cd->u.procfs.proc_ent = proc_mkdir(cd->name, sn->proc_net_rpc);
1632         if (cd->u.procfs.proc_ent == NULL)
1633                 goto out_nomem;
1634         cd->u.procfs.channel_ent = NULL;
1635         cd->u.procfs.content_ent = NULL;
1636
1637         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1638                              cd->u.procfs.proc_ent,
1639                              &cache_flush_operations_procfs, cd);
1640         cd->u.procfs.flush_ent = p;
1641         if (p == NULL)
1642                 goto out_nomem;
1643
1644         if (cd->cache_request || cd->cache_parse) {
1645                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1646                                      cd->u.procfs.proc_ent,
1647                                      &cache_file_operations_procfs, cd);
1648                 cd->u.procfs.channel_ent = p;
1649                 if (p == NULL)
1650                         goto out_nomem;
1651         }
1652         if (cd->cache_show) {
1653                 p = proc_create_data("content", S_IFREG|S_IRUSR,
1654                                 cd->u.procfs.proc_ent,
1655                                 &content_file_operations_procfs, cd);
1656                 cd->u.procfs.content_ent = p;
1657                 if (p == NULL)
1658                         goto out_nomem;
1659         }
1660         return 0;
1661 out_nomem:
1662         remove_cache_proc_entries(cd, net);
1663         return -ENOMEM;
1664 }
1665 #else /* CONFIG_PROC_FS */
1666 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1667 {
1668         return 0;
1669 }
1670 #endif
1671
1672 void __init cache_initialize(void)
1673 {
1674         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1675 }
1676
1677 int cache_register_net(struct cache_detail *cd, struct net *net)
1678 {
1679         int ret;
1680
1681         sunrpc_init_cache_detail(cd);
1682         ret = create_cache_proc_entries(cd, net);
1683         if (ret)
1684                 sunrpc_destroy_cache_detail(cd);
1685         return ret;
1686 }
1687 EXPORT_SYMBOL_GPL(cache_register_net);
1688
1689 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1690 {
1691         remove_cache_proc_entries(cd, net);
1692         sunrpc_destroy_cache_detail(cd);
1693 }
1694 EXPORT_SYMBOL_GPL(cache_unregister_net);
1695
1696 struct cache_detail *cache_create_net(struct cache_detail *tmpl, struct net *net)
1697 {
1698         struct cache_detail *cd;
1699         int i;
1700
1701         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1702         if (cd == NULL)
1703                 return ERR_PTR(-ENOMEM);
1704
1705         cd->hash_table = kzalloc(cd->hash_size * sizeof(struct hlist_head),
1706                                  GFP_KERNEL);
1707         if (cd->hash_table == NULL) {
1708                 kfree(cd);
1709                 return ERR_PTR(-ENOMEM);
1710         }
1711
1712         for (i = 0; i < cd->hash_size; i++)
1713                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1714         cd->net = net;
1715         return cd;
1716 }
1717 EXPORT_SYMBOL_GPL(cache_create_net);
1718
1719 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1720 {
1721         kfree(cd->hash_table);
1722         kfree(cd);
1723 }
1724 EXPORT_SYMBOL_GPL(cache_destroy_net);
1725
1726 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1727                                  size_t count, loff_t *ppos)
1728 {
1729         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1730
1731         return cache_read(filp, buf, count, ppos, cd);
1732 }
1733
1734 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1735                                   size_t count, loff_t *ppos)
1736 {
1737         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1738
1739         return cache_write(filp, buf, count, ppos, cd);
1740 }
1741
1742 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1743 {
1744         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1745
1746         return cache_poll(filp, wait, cd);
1747 }
1748
1749 static long cache_ioctl_pipefs(struct file *filp,
1750                               unsigned int cmd, unsigned long arg)
1751 {
1752         struct inode *inode = file_inode(filp);
1753         struct cache_detail *cd = RPC_I(inode)->private;
1754
1755         return cache_ioctl(inode, filp, cmd, arg, cd);
1756 }
1757
1758 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1759 {
1760         struct cache_detail *cd = RPC_I(inode)->private;
1761
1762         return cache_open(inode, filp, cd);
1763 }
1764
1765 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1766 {
1767         struct cache_detail *cd = RPC_I(inode)->private;
1768
1769         return cache_release(inode, filp, cd);
1770 }
1771
1772 const struct file_operations cache_file_operations_pipefs = {
1773         .owner          = THIS_MODULE,
1774         .llseek         = no_llseek,
1775         .read           = cache_read_pipefs,
1776         .write          = cache_write_pipefs,
1777         .poll           = cache_poll_pipefs,
1778         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1779         .open           = cache_open_pipefs,
1780         .release        = cache_release_pipefs,
1781 };
1782
1783 static int content_open_pipefs(struct inode *inode, struct file *filp)
1784 {
1785         struct cache_detail *cd = RPC_I(inode)->private;
1786
1787         return content_open(inode, filp, cd);
1788 }
1789
1790 static int content_release_pipefs(struct inode *inode, struct file *filp)
1791 {
1792         struct cache_detail *cd = RPC_I(inode)->private;
1793
1794         return content_release(inode, filp, cd);
1795 }
1796
1797 const struct file_operations content_file_operations_pipefs = {
1798         .open           = content_open_pipefs,
1799         .read           = seq_read,
1800         .llseek         = seq_lseek,
1801         .release        = content_release_pipefs,
1802 };
1803
1804 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1805 {
1806         struct cache_detail *cd = RPC_I(inode)->private;
1807
1808         return open_flush(inode, filp, cd);
1809 }
1810
1811 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1812 {
1813         struct cache_detail *cd = RPC_I(inode)->private;
1814
1815         return release_flush(inode, filp, cd);
1816 }
1817
1818 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1819                             size_t count, loff_t *ppos)
1820 {
1821         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1822
1823         return read_flush(filp, buf, count, ppos, cd);
1824 }
1825
1826 static ssize_t write_flush_pipefs(struct file *filp,
1827                                   const char __user *buf,
1828                                   size_t count, loff_t *ppos)
1829 {
1830         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1831
1832         return write_flush(filp, buf, count, ppos, cd);
1833 }
1834
1835 const struct file_operations cache_flush_operations_pipefs = {
1836         .open           = open_flush_pipefs,
1837         .read           = read_flush_pipefs,
1838         .write          = write_flush_pipefs,
1839         .release        = release_flush_pipefs,
1840         .llseek         = no_llseek,
1841 };
1842
1843 int sunrpc_cache_register_pipefs(struct dentry *parent,
1844                                  const char *name, umode_t umode,
1845                                  struct cache_detail *cd)
1846 {
1847         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1848         if (IS_ERR(dir))
1849                 return PTR_ERR(dir);
1850         cd->u.pipefs.dir = dir;
1851         return 0;
1852 }
1853 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1854
1855 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1856 {
1857         rpc_remove_cache_dir(cd->u.pipefs.dir);
1858         cd->u.pipefs.dir = NULL;
1859 }
1860 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1861