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 <linux/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 inline int cache_is_valid(struct cache_head *h);
  58 static void cache_fresh_locked(struct cache_head *head, time_t expiry,
  59                                 struct cache_detail *detail);
  60 static void cache_fresh_unlocked(struct cache_head *head,
  61                                 struct cache_detail *detail);
  62
  63 static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
  64                                                 struct cache_head *key,
  65                                                 int hash)
  66 {
  67         struct hlist_head *head = &detail->hash_table[hash];
  68         struct cache_head *tmp;
  69
  70         rcu_read_lock();
  71         hlist_for_each_entry_rcu(tmp, head, cache_list) {
  72                 if (detail->match(tmp, key)) {
  73                         if (cache_is_expired(detail, tmp))
  74                                 continue;
  75                         tmp = cache_get_rcu(tmp);
  76                         rcu_read_unlock();
  77                         return tmp;
  78                 }
  79         }
  80         rcu_read_unlock();
  81         return NULL;
  82 }
  83
  84 static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
  85                                                  struct cache_head *key,
  86                                                  int hash)
  87 {
  88         struct cache_head *new, *tmp, *freeme = NULL;
  89         struct hlist_head *head = &detail->hash_table[hash];
  90
  91         new = detail->alloc();
  92         if (!new)
  93                 return NULL;
  94         /* must fully initialise 'new', else
  95          * we might get lose if we need to
  96          * cache_put it soon.
  97          */
  98         cache_init(new, detail);
  99         detail->init(new, key);
 100
 101         spin_lock(&detail->hash_lock);
 102
 103         /* check if entry appeared while we slept */
 104         hlist_for_each_entry_rcu(tmp, head, cache_list) {
 105                 if (detail->match(tmp, key)) {
 106                         if (cache_is_expired(detail, tmp)) {
 107                                 hlist_del_init_rcu(&tmp->cache_list);
 108                                 detail->entries --;
 109                                 if (cache_is_valid(tmp) == -EAGAIN)
 110                                         set_bit(CACHE_NEGATIVE, &tmp->flags);
 111                                 cache_fresh_locked(tmp, 0, detail);
 112                                 freeme = tmp;
 113                                 break;
 114                         }
 115                         cache_get(tmp);
 116                         spin_unlock(&detail->hash_lock);
 117                         cache_put(new, detail);
 118                         return tmp;
 119                 }
 120         }
 121
 122         hlist_add_head_rcu(&new->cache_list, head);
 123         detail->entries++;
 124         cache_get(new);
 125         spin_unlock(&detail->hash_lock);
 126
 127         if (freeme) {
 128                 cache_fresh_unlocked(freeme, detail);
 129                 cache_put(freeme, detail);
 130         }
 131         return new;
 132 }
 133
 134 struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
 135                                            struct cache_head *key, int hash)
 136 {
 137         struct cache_head *ret;
 138
 139         ret = sunrpc_cache_find_rcu(detail, key, hash);
 140         if (ret)
 141                 return ret;
 142         /* Didn't find anything, insert an empty entry */
 143         return sunrpc_cache_add_entry(detail, key, hash);
 144 }
 145 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
 146
 147 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 148
 149 static void cache_fresh_locked(struct cache_head *head, time_t expiry,
 150                                struct cache_detail *detail)
 151 {
 152         time_t now = seconds_since_boot();
 153         if (now <= detail->flush_time)
 154                 /* ensure it isn't immediately treated as expired */
 155                 now = detail->flush_time + 1;
 156         head->expiry_time = expiry;
 157         head->last_refresh = now;
 158         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 159         set_bit(CACHE_VALID, &head->flags);
 160 }
 161
 162 static void cache_fresh_unlocked(struct cache_head *head,
 163                                  struct cache_detail *detail)
 164 {
 165         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 166                 cache_revisit_request(head);
 167                 cache_dequeue(detail, head);
 168         }
 169 }
 170
 171 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 172                                        struct cache_head *new, struct cache_head *old, int hash)
 173 {
 174         /* The 'old' entry is to be replaced by 'new'.
 175          * If 'old' is not VALID, we update it directly,
 176          * otherwise we need to replace it
 177          */
 178         struct cache_head *tmp;
 179
 180         if (!test_bit(CACHE_VALID, &old->flags)) {
 181                 spin_lock(&detail->hash_lock);
 182                 if (!test_bit(CACHE_VALID, &old->flags)) {
 183                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 184                                 set_bit(CACHE_NEGATIVE, &old->flags);
 185                         else
 186                                 detail->update(old, new);
 187                         cache_fresh_locked(old, new->expiry_time, detail);
 188                         spin_unlock(&detail->hash_lock);
 189                         cache_fresh_unlocked(old, detail);
 190                         return old;
 191                 }
 192                 spin_unlock(&detail->hash_lock);
 193         }
 194         /* We need to insert a new entry */
 195         tmp = detail->alloc();
 196         if (!tmp) {
 197                 cache_put(old, detail);
 198                 return NULL;
 199         }
 200         cache_init(tmp, detail);
 201         detail->init(tmp, old);
 202
 203         spin_lock(&detail->hash_lock);
 204         if (test_bit(CACHE_NEGATIVE, &new->flags))
 205                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 206         else
 207                 detail->update(tmp, new);
 208         hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
 209         detail->entries++;
 210         cache_get(tmp);
 211         cache_fresh_locked(tmp, new->expiry_time, detail);
 212         cache_fresh_locked(old, 0, detail);
 213         spin_unlock(&detail->hash_lock);
 214         cache_fresh_unlocked(tmp, detail);
 215         cache_fresh_unlocked(old, detail);
 216         cache_put(old, detail);
 217         return tmp;
 218 }
 219 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 220
 221 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 222 {
 223         if (cd->cache_upcall)
 224                 return cd->cache_upcall(cd, h);
 225         return sunrpc_cache_pipe_upcall(cd, h);
 226 }
 227
 228 static inline int cache_is_valid(struct cache_head *h)
 229 {
 230         if (!test_bit(CACHE_VALID, &h->flags))
 231                 return -EAGAIN;
 232         else {
 233                 /* entry is valid */
 234                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 235                         return -ENOENT;
 236                 else {
 237                         /*
 238                          * In combination with write barrier in
 239                          * sunrpc_cache_update, ensures that anyone
 240                          * using the cache entry after this sees the
 241                          * updated contents:
 242                          */
 243                         smp_rmb();
 244                         return 0;
 245                 }
 246         }
 247 }
 248
 249 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 250 {
 251         int rv;
 252
 253         spin_lock(&detail->hash_lock);
 254         rv = cache_is_valid(h);
 255         if (rv == -EAGAIN) {
 256                 set_bit(CACHE_NEGATIVE, &h->flags);
 257                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
 258                                    detail);
 259                 rv = -ENOENT;
 260         }
 261         spin_unlock(&detail->hash_lock);
 262         cache_fresh_unlocked(h, detail);
 263         return rv;
 264 }
 265
 266 /*
 267  * This is the generic cache management routine for all
 268  * the authentication caches.
 269  * It checks the currency of a cache item and will (later)
 270  * initiate an upcall to fill it if needed.
 271  *
 272  *
 273  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 274  * -EAGAIN if upcall is pending and request has been queued
 275  * -ETIMEDOUT if upcall failed or request could not be queue or
 276  *           upcall completed but item is still invalid (implying that
 277  *           the cache item has been replaced with a newer one).
 278  * -ENOENT if cache entry was negative
 279  */
 280 int cache_check(struct cache_detail *detail,
 281                     struct cache_head *h, struct cache_req *rqstp)
 282 {
 283         int rv;
 284         long refresh_age, age;
 285
 286         /* First decide return status as best we can */
 287         rv = cache_is_valid(h);
 288
 289         /* now see if we want to start an upcall */
 290         refresh_age = (h->expiry_time - h->last_refresh);
 291         age = seconds_since_boot() - h->last_refresh;
 292
 293         if (rqstp == NULL) {
 294                 if (rv == -EAGAIN)
 295                         rv = -ENOENT;
 296         } else if (rv == -EAGAIN ||
 297                    (h->expiry_time != 0 && age > refresh_age/2)) {
 298                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 299                                 refresh_age, age);
 300                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 301                         switch (cache_make_upcall(detail, h)) {
 302                         case -EINVAL:
 303                                 rv = try_to_negate_entry(detail, h);
 304                                 break;
 305                         case -EAGAIN:
 306                                 cache_fresh_unlocked(h, detail);
 307                                 break;
 308                         }
 309                 }
 310         }
 311
 312         if (rv == -EAGAIN) {
 313                 if (!cache_defer_req(rqstp, h)) {
 314                         /*
 315                          * Request was not deferred; handle it as best
 316                          * we can ourselves:
 317                          */
 318                         rv = cache_is_valid(h);
 319                         if (rv == -EAGAIN)
 320                                 rv = -ETIMEDOUT;
 321                 }
 322         }
 323         if (rv)
 324                 cache_put(h, detail);
 325         return rv;
 326 }
 327 EXPORT_SYMBOL_GPL(cache_check);
 328
 329 /*
 330  * caches need to be periodically cleaned.
 331  * For this we maintain a list of cache_detail and
 332  * a current pointer into that list and into the table
 333  * for that entry.
 334  *
 335  * Each time cache_clean is called it finds the next non-empty entry
 336  * in the current table and walks the list in that entry
 337  * looking for entries that can be removed.
 338  *
 339  * An entry gets removed if:
 340  * - The expiry is before current time
 341  * - The last_refresh time is before the flush_time for that cache
 342  *
 343  * later we might drop old entries with non-NEVER expiry if that table
 344  * is getting 'full' for some definition of 'full'
 345  *
 346  * The question of "how often to scan a table" is an interesting one
 347  * and is answered in part by the use of the "nextcheck" field in the
 348  * cache_detail.
 349  * When a scan of a table begins, the nextcheck field is set to a time
 350  * that is well into the future.
 351  * While scanning, if an expiry time is found that is earlier than the
 352  * current nextcheck time, nextcheck is set to that expiry time.
 353  * If the flush_time is ever set to a time earlier than the nextcheck
 354  * time, the nextcheck time is then set to that flush_time.
 355  *
 356  * A table is then only scanned if the current time is at least
 357  * the nextcheck time.
 358  *
 359  */
 360
 361 static LIST_HEAD(cache_list);
 362 static DEFINE_SPINLOCK(cache_list_lock);
 363 static struct cache_detail *current_detail;
 364 static int current_index;
 365
 366 static void do_cache_clean(struct work_struct *work);
 367 static struct delayed_work cache_cleaner;
 368
 369 void sunrpc_init_cache_detail(struct cache_detail *cd)
 370 {
 371         spin_lock_init(&cd->hash_lock);
 372         INIT_LIST_HEAD(&cd->queue);
 373         spin_lock(&cache_list_lock);
 374         cd->nextcheck = 0;
 375         cd->entries = 0;
 376         atomic_set(&cd->readers, 0);
 377         cd->last_close = 0;
 378         cd->last_warn = -1;
 379         list_add(&cd->others, &cache_list);
 380         spin_unlock(&cache_list_lock);
 381
 382         /* start the cleaning process */
 383         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
 384 }
 385 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 386
 387 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 388 {
 389         cache_purge(cd);
 390         spin_lock(&cache_list_lock);
 391         spin_lock(&cd->hash_lock);
 392         if (current_detail == cd)
 393                 current_detail = NULL;
 394         list_del_init(&cd->others);
 395         spin_unlock(&cd->hash_lock);
 396         spin_unlock(&cache_list_lock);
 397         if (list_empty(&cache_list)) {
 398                 /* module must be being unloaded so its safe to kill the worker */
 399                 cancel_delayed_work_sync(&cache_cleaner);
 400         }
 401 }
 402 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 403
 404 /* clean cache tries to find something to clean
 405  * and cleans it.
 406  * It returns 1 if it cleaned something,
 407  *            0 if it didn't find anything this time
 408  *           -1 if it fell off the end of the list.
 409  */
 410 static int cache_clean(void)
 411 {
 412         int rv = 0;
 413         struct list_head *next;
 414
 415         spin_lock(&cache_list_lock);
 416
 417         /* find a suitable table if we don't already have one */
 418         while (current_detail == NULL ||
 419             current_index >= current_detail->hash_size) {
 420                 if (current_detail)
 421                         next = current_detail->others.next;
 422                 else
 423                         next = cache_list.next;
 424                 if (next == &cache_list) {
 425                         current_detail = NULL;
 426                         spin_unlock(&cache_list_lock);
 427                         return -1;
 428                 }
 429                 current_detail = list_entry(next, struct cache_detail, others);
 430                 if (current_detail->nextcheck > seconds_since_boot())
 431                         current_index = current_detail->hash_size;
 432                 else {
 433                         current_index = 0;
 434                         current_detail->nextcheck = seconds_since_boot()+30*60;
 435                 }
 436         }
 437
 438         /* find a non-empty bucket in the table */
 439         while (current_detail &&
 440                current_index < current_detail->hash_size &&
 441                hlist_empty(&current_detail->hash_table[current_index]))
 442                 current_index++;
 443
 444         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 445
 446         if (current_detail && current_index < current_detail->hash_size) {
 447                 struct cache_head *ch = NULL;
 448                 struct cache_detail *d;
 449                 struct hlist_head *head;
 450                 struct hlist_node *tmp;
 451
 452                 spin_lock(&current_detail->hash_lock);
 453
 454                 /* Ok, now to clean this strand */
 455
 456                 head = &current_detail->hash_table[current_index];
 457                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 458                         if (current_detail->nextcheck > ch->expiry_time)
 459                                 current_detail->nextcheck = ch->expiry_time+1;
 460                         if (!cache_is_expired(current_detail, ch))
 461                                 continue;
 462
 463                         hlist_del_init_rcu(&ch->cache_list);
 464                         current_detail->entries--;
 465                         rv = 1;
 466                         break;
 467                 }
 468
 469                 spin_unlock(&current_detail->hash_lock);
 470                 d = current_detail;
 471                 if (!ch)
 472                         current_index ++;
 473                 spin_unlock(&cache_list_lock);
 474                 if (ch) {
 475                         set_bit(CACHE_CLEANED, &ch->flags);
 476                         cache_fresh_unlocked(ch, d);
 477                         cache_put(ch, d);
 478                 }
 479         } else
 480                 spin_unlock(&cache_list_lock);
 481
 482         return rv;
 483 }
 484
 485 /*
 486  * We want to regularly clean the cache, so we need to schedule some work ...
 487  */
 488 static void do_cache_clean(struct work_struct *work)
 489 {
 490         int delay = 5;
 491         if (cache_clean() == -1)
 492                 delay = round_jiffies_relative(30*HZ);
 493
 494         if (list_empty(&cache_list))
 495                 delay = 0;
 496
 497         if (delay)
 498                 queue_delayed_work(system_power_efficient_wq,
 499                                    &cache_cleaner, delay);
 500 }
 501
 502
 503 /*
 504  * Clean all caches promptly.  This just calls cache_clean
 505  * repeatedly until we are sure that every cache has had a chance to
 506  * be fully cleaned
 507  */
 508 void cache_flush(void)
 509 {
 510         while (cache_clean() != -1)
 511                 cond_resched();
 512         while (cache_clean() != -1)
 513                 cond_resched();
 514 }
 515 EXPORT_SYMBOL_GPL(cache_flush);
 516
 517 void cache_purge(struct cache_detail *detail)
 518 {
 519         struct cache_head *ch = NULL;
 520         struct hlist_head *head = NULL;
 521         struct hlist_node *tmp = NULL;
 522         int i = 0;
 523
 524         spin_lock(&detail->hash_lock);
 525         if (!detail->entries) {
 526                 spin_unlock(&detail->hash_lock);
 527                 return;
 528         }
 529
 530         dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
 531         for (i = 0; i < detail->hash_size; i++) {
 532                 head = &detail->hash_table[i];
 533                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 534                         hlist_del_init_rcu(&ch->cache_list);
 535                         detail->entries--;
 536
 537                         set_bit(CACHE_CLEANED, &ch->flags);
 538                         spin_unlock(&detail->hash_lock);
 539                         cache_fresh_unlocked(ch, detail);
 540                         cache_put(ch, detail);
 541                         spin_lock(&detail->hash_lock);
 542                 }
 543         }
 544         spin_unlock(&detail->hash_lock);
 545 }
 546 EXPORT_SYMBOL_GPL(cache_purge);
 547
 548
 549 /*
 550  * Deferral and Revisiting of Requests.
 551  *
 552  * If a cache lookup finds a pending entry, we
 553  * need to defer the request and revisit it later.
 554  * All deferred requests are stored in a hash table,
 555  * indexed by "struct cache_head *".
 556  * As it may be wasteful to store a whole request
 557  * structure, we allow the request to provide a
 558  * deferred form, which must contain a
 559  * 'struct cache_deferred_req'
 560  * This cache_deferred_req contains a method to allow
 561  * it to be revisited when cache info is available
 562  */
 563
 564 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 565 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 566
 567 #define DFR_MAX 300     /* ??? */
 568
 569 static DEFINE_SPINLOCK(cache_defer_lock);
 570 static LIST_HEAD(cache_defer_list);
 571 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 572 static int cache_defer_cnt;
 573
 574 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 575 {
 576         hlist_del_init(&dreq->hash);
 577         if (!list_empty(&dreq->recent)) {
 578                 list_del_init(&dreq->recent);
 579                 cache_defer_cnt--;
 580         }
 581 }
 582
 583 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 584 {
 585         int hash = DFR_HASH(item);
 586
 587         INIT_LIST_HEAD(&dreq->recent);
 588         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 589 }
 590
 591 static void setup_deferral(struct cache_deferred_req *dreq,
 592                            struct cache_head *item,
 593                            int count_me)
 594 {
 595
 596         dreq->item = item;
 597
 598         spin_lock(&cache_defer_lock);
 599
 600         __hash_deferred_req(dreq, item);
 601
 602         if (count_me) {
 603                 cache_defer_cnt++;
 604                 list_add(&dreq->recent, &cache_defer_list);
 605         }
 606
 607         spin_unlock(&cache_defer_lock);
 608
 609 }
 610
 611 struct thread_deferred_req {
 612         struct cache_deferred_req handle;
 613         struct completion completion;
 614 };
 615
 616 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 617 {
 618         struct thread_deferred_req *dr =
 619                 container_of(dreq, struct thread_deferred_req, handle);
 620         complete(&dr->completion);
 621 }
 622
 623 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 624 {
 625         struct thread_deferred_req sleeper;
 626         struct cache_deferred_req *dreq = &sleeper.handle;
 627
 628         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 629         dreq->revisit = cache_restart_thread;
 630
 631         setup_deferral(dreq, item, 0);
 632
 633         if (!test_bit(CACHE_PENDING, &item->flags) ||
 634             wait_for_completion_interruptible_timeout(
 635                     &sleeper.completion, req->thread_wait) <= 0) {
 636                 /* The completion wasn't completed, so we need
 637                  * to clean up
 638                  */
 639                 spin_lock(&cache_defer_lock);
 640                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 641                         __unhash_deferred_req(&sleeper.handle);
 642                         spin_unlock(&cache_defer_lock);
 643                 } else {
 644                         /* cache_revisit_request already removed
 645                          * this from the hash table, but hasn't
 646                          * called ->revisit yet.  It will very soon
 647                          * and we need to wait for it.
 648                          */
 649                         spin_unlock(&cache_defer_lock);
 650                         wait_for_completion(&sleeper.completion);
 651                 }
 652         }
 653 }
 654
 655 static void cache_limit_defers(void)
 656 {
 657         /* Make sure we haven't exceed the limit of allowed deferred
 658          * requests.
 659          */
 660         struct cache_deferred_req *discard = NULL;
 661
 662         if (cache_defer_cnt <= DFR_MAX)
 663                 return;
 664
 665         spin_lock(&cache_defer_lock);
 666
 667         /* Consider removing either the first or the last */
 668         if (cache_defer_cnt > DFR_MAX) {
 669                 if (prandom_u32() & 1)
 670                         discard = list_entry(cache_defer_list.next,
 671                                              struct cache_deferred_req, recent);
 672                 else
 673                         discard = list_entry(cache_defer_list.prev,
 674                                              struct cache_deferred_req, recent);
 675                 __unhash_deferred_req(discard);
 676         }
 677         spin_unlock(&cache_defer_lock);
 678         if (discard)
 679                 discard->revisit(discard, 1);
 680 }
 681
 682 /* Return true if and only if a deferred request is queued. */
 683 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 684 {
 685         struct cache_deferred_req *dreq;
 686
 687         if (req->thread_wait) {
 688                 cache_wait_req(req, item);
 689                 if (!test_bit(CACHE_PENDING, &item->flags))
 690                         return false;
 691         }
 692         dreq = req->defer(req);
 693         if (dreq == NULL)
 694                 return false;
 695         setup_deferral(dreq, item, 1);
 696         if (!test_bit(CACHE_PENDING, &item->flags))
 697                 /* Bit could have been cleared before we managed to
 698                  * set up the deferral, so need to revisit just in case
 699                  */
 700                 cache_revisit_request(item);
 701
 702         cache_limit_defers();
 703         return true;
 704 }
 705
 706 static void cache_revisit_request(struct cache_head *item)
 707 {
 708         struct cache_deferred_req *dreq;
 709         struct list_head pending;
 710         struct hlist_node *tmp;
 711         int hash = DFR_HASH(item);
 712
 713         INIT_LIST_HEAD(&pending);
 714         spin_lock(&cache_defer_lock);
 715
 716         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 717                 if (dreq->item == item) {
 718                         __unhash_deferred_req(dreq);
 719                         list_add(&dreq->recent, &pending);
 720                 }
 721
 722         spin_unlock(&cache_defer_lock);
 723
 724         while (!list_empty(&pending)) {
 725                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 726                 list_del_init(&dreq->recent);
 727                 dreq->revisit(dreq, 0);
 728         }
 729 }
 730
 731 void cache_clean_deferred(void *owner)
 732 {
 733         struct cache_deferred_req *dreq, *tmp;
 734         struct list_head pending;
 735
 736
 737         INIT_LIST_HEAD(&pending);
 738         spin_lock(&cache_defer_lock);
 739
 740         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 741                 if (dreq->owner == owner) {
 742                         __unhash_deferred_req(dreq);
 743                         list_add(&dreq->recent, &pending);
 744                 }
 745         }
 746         spin_unlock(&cache_defer_lock);
 747
 748         while (!list_empty(&pending)) {
 749                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 750                 list_del_init(&dreq->recent);
 751                 dreq->revisit(dreq, 1);
 752         }
 753 }
 754
 755 /*
 756  * communicate with user-space
 757  *
 758  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
 759  * On read, you get a full request, or block.
 760  * On write, an update request is processed.
 761  * Poll works if anything to read, and always allows write.
 762  *
 763  * Implemented by linked list of requests.  Each open file has
 764  * a ->private that also exists in this list.  New requests are added
 765  * to the end and may wakeup and preceding readers.
 766  * New readers are added to the head.  If, on read, an item is found with
 767  * CACHE_UPCALLING clear, we free it from the list.
 768  *
 769  */
 770
 771 static DEFINE_SPINLOCK(queue_lock);
 772 static DEFINE_MUTEX(queue_io_mutex);
 773
 774 struct cache_queue {
 775         struct list_head        list;
 776         int                     reader; /* if 0, then request */
 777 };
 778 struct cache_request {
 779         struct cache_queue      q;
 780         struct cache_head       *item;
 781         char                    * buf;
 782         int                     len;
 783         int                     readers;
 784 };
 785 struct cache_reader {
 786         struct cache_queue      q;
 787         int                     offset; /* if non-0, we have a refcnt on next request */
 788 };
 789
 790 static int cache_request(struct cache_detail *detail,
 791                                struct cache_request *crq)
 792 {
 793         char *bp = crq->buf;
 794         int len = PAGE_SIZE;
 795
 796         detail->cache_request(detail, crq->item, &bp, &len);
 797         if (len < 0)
 798                 return -EAGAIN;
 799         return PAGE_SIZE - len;
 800 }
 801
 802 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 803                           loff_t *ppos, struct cache_detail *cd)
 804 {
 805         struct cache_reader *rp = filp->private_data;
 806         struct cache_request *rq;
 807         struct inode *inode = file_inode(filp);
 808         int err;
 809
 810         if (count == 0)
 811                 return 0;
 812
 813         inode_lock(inode); /* protect against multiple concurrent
 814                               * readers on this file */
 815  again:
 816         spin_lock(&queue_lock);
 817         /* need to find next request */
 818         while (rp->q.list.next != &cd->queue &&
 819                list_entry(rp->q.list.next, struct cache_queue, list)
 820                ->reader) {
 821                 struct list_head *next = rp->q.list.next;
 822                 list_move(&rp->q.list, next);
 823         }
 824         if (rp->q.list.next == &cd->queue) {
 825                 spin_unlock(&queue_lock);
 826                 inode_unlock(inode);
 827                 WARN_ON_ONCE(rp->offset);
 828                 return 0;
 829         }
 830         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 831         WARN_ON_ONCE(rq->q.reader);
 832         if (rp->offset == 0)
 833                 rq->readers++;
 834         spin_unlock(&queue_lock);
 835
 836         if (rq->len == 0) {
 837                 err = cache_request(cd, rq);
 838                 if (err < 0)
 839                         goto out;
 840                 rq->len = err;
 841         }
 842
 843         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 844                 err = -EAGAIN;
 845                 spin_lock(&queue_lock);
 846                 list_move(&rp->q.list, &rq->q.list);
 847                 spin_unlock(&queue_lock);
 848         } else {
 849                 if (rp->offset + count > rq->len)
 850                         count = rq->len - rp->offset;
 851                 err = -EFAULT;
 852                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 853                         goto out;
 854                 rp->offset += count;
 855                 if (rp->offset >= rq->len) {
 856                         rp->offset = 0;
 857                         spin_lock(&queue_lock);
 858                         list_move(&rp->q.list, &rq->q.list);
 859                         spin_unlock(&queue_lock);
 860                 }
 861                 err = 0;
 862         }
 863  out:
 864         if (rp->offset == 0) {
 865                 /* need to release rq */
 866                 spin_lock(&queue_lock);
 867                 rq->readers--;
 868                 if (rq->readers == 0 &&
 869                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 870                         list_del(&rq->q.list);
 871                         spin_unlock(&queue_lock);
 872                         cache_put(rq->item, cd);
 873                         kfree(rq->buf);
 874                         kfree(rq);
 875                 } else
 876                         spin_unlock(&queue_lock);
 877         }
 878         if (err == -EAGAIN)
 879                 goto again;
 880         inode_unlock(inode);
 881         return err ? err :  count;
 882 }
 883
 884 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 885                                  size_t count, struct cache_detail *cd)
 886 {
 887         ssize_t ret;
 888
 889         if (count == 0)
 890                 return -EINVAL;
 891         if (copy_from_user(kaddr, buf, count))
 892                 return -EFAULT;
 893         kaddr[count] = '\0';
 894         ret = cd->cache_parse(cd, kaddr, count);
 895         if (!ret)
 896                 ret = count;
 897         return ret;
 898 }
 899
 900 static ssize_t cache_slow_downcall(const char __user *buf,
 901                                    size_t count, struct cache_detail *cd)
 902 {
 903         static char write_buf[8192]; /* protected by queue_io_mutex */
 904         ssize_t ret = -EINVAL;
 905
 906         if (count >= sizeof(write_buf))
 907                 goto out;
 908         mutex_lock(&queue_io_mutex);
 909         ret = cache_do_downcall(write_buf, buf, count, cd);
 910         mutex_unlock(&queue_io_mutex);
 911 out:
 912         return ret;
 913 }
 914
 915 static ssize_t cache_downcall(struct address_space *mapping,
 916                               const char __user *buf,
 917                               size_t count, struct cache_detail *cd)
 918 {
 919         struct page *page;
 920         char *kaddr;
 921         ssize_t ret = -ENOMEM;
 922
 923         if (count >= PAGE_SIZE)
 924                 goto out_slow;
 925
 926         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 927         if (!page)
 928                 goto out_slow;
 929
 930         kaddr = kmap(page);
 931         ret = cache_do_downcall(kaddr, buf, count, cd);
 932         kunmap(page);
 933         unlock_page(page);
 934         put_page(page);
 935         return ret;
 936 out_slow:
 937         return cache_slow_downcall(buf, count, cd);
 938 }
 939
 940 static ssize_t cache_write(struct file *filp, const char __user *buf,
 941                            size_t count, loff_t *ppos,
 942                            struct cache_detail *cd)
 943 {
 944         struct address_space *mapping = filp->f_mapping;
 945         struct inode *inode = file_inode(filp);
 946         ssize_t ret = -EINVAL;
 947
 948         if (!cd->cache_parse)
 949                 goto out;
 950
 951         inode_lock(inode);
 952         ret = cache_downcall(mapping, buf, count, cd);
 953         inode_unlock(inode);
 954 out:
 955         return ret;
 956 }
 957
 958 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 959
 960 static __poll_t cache_poll(struct file *filp, poll_table *wait,
 961                                struct cache_detail *cd)
 962 {
 963         __poll_t mask;
 964         struct cache_reader *rp = filp->private_data;
 965         struct cache_queue *cq;
 966
 967         poll_wait(filp, &queue_wait, wait);
 968
 969         /* alway allow write */
 970         mask = EPOLLOUT | EPOLLWRNORM;
 971
 972         if (!rp)
 973                 return mask;
 974
 975         spin_lock(&queue_lock);
 976
 977         for (cq= &rp->q; &cq->list != &cd->queue;
 978              cq = list_entry(cq->list.next, struct cache_queue, list))
 979                 if (!cq->reader) {
 980                         mask |= EPOLLIN | EPOLLRDNORM;
 981                         break;
 982                 }
 983         spin_unlock(&queue_lock);
 984         return mask;
 985 }
 986
 987 static int cache_ioctl(struct inode *ino, struct file *filp,
 988                        unsigned int cmd, unsigned long arg,
 989                        struct cache_detail *cd)
 990 {
 991         int len = 0;
 992         struct cache_reader *rp = filp->private_data;
 993         struct cache_queue *cq;
 994
 995         if (cmd != FIONREAD || !rp)
 996                 return -EINVAL;
 997
 998         spin_lock(&queue_lock);
 999
1000         /* only find the length remaining in current request,
1001          * or the length of the next request
1002          */
1003         for (cq= &rp->q; &cq->list != &cd->queue;
1004              cq = list_entry(cq->list.next, struct cache_queue, list))
1005                 if (!cq->reader) {
1006                         struct cache_request *cr =
1007                                 container_of(cq, struct cache_request, q);
1008                         len = cr->len - rp->offset;
1009                         break;
1010                 }
1011         spin_unlock(&queue_lock);
1012
1013         return put_user(len, (int __user *)arg);
1014 }
1015
1016 static int cache_open(struct inode *inode, struct file *filp,
1017                       struct cache_detail *cd)
1018 {
1019         struct cache_reader *rp = NULL;
1020
1021         if (!cd || !try_module_get(cd->owner))
1022                 return -EACCES;
1023         nonseekable_open(inode, filp);
1024         if (filp->f_mode & FMODE_READ) {
1025                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1026                 if (!rp) {
1027                         module_put(cd->owner);
1028                         return -ENOMEM;
1029                 }
1030                 rp->offset = 0;
1031                 rp->q.reader = 1;
1032                 atomic_inc(&cd->readers);
1033                 spin_lock(&queue_lock);
1034                 list_add(&rp->q.list, &cd->queue);
1035                 spin_unlock(&queue_lock);
1036         }
1037         filp->private_data = rp;
1038         return 0;
1039 }
1040
1041 static int cache_release(struct inode *inode, struct file *filp,
1042                          struct cache_detail *cd)
1043 {
1044         struct cache_reader *rp = filp->private_data;
1045
1046         if (rp) {
1047                 spin_lock(&queue_lock);
1048                 if (rp->offset) {
1049                         struct cache_queue *cq;
1050                         for (cq= &rp->q; &cq->list != &cd->queue;
1051                              cq = list_entry(cq->list.next, struct cache_queue, list))
1052                                 if (!cq->reader) {
1053                                         container_of(cq, struct cache_request, q)
1054                                                 ->readers--;
1055                                         break;
1056                                 }
1057                         rp->offset = 0;
1058                 }
1059                 list_del(&rp->q.list);
1060                 spin_unlock(&queue_lock);
1061
1062                 filp->private_data = NULL;
1063                 kfree(rp);
1064
1065                 cd->last_close = seconds_since_boot();
1066                 atomic_dec(&cd->readers);
1067         }
1068         module_put(cd->owner);
1069         return 0;
1070 }
1071
1072
1073
1074 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1075 {
1076         struct cache_queue *cq, *tmp;
1077         struct cache_request *cr;
1078         struct list_head dequeued;
1079
1080         INIT_LIST_HEAD(&dequeued);
1081         spin_lock(&queue_lock);
1082         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1083                 if (!cq->reader) {
1084                         cr = container_of(cq, struct cache_request, q);
1085                         if (cr->item != ch)
1086                                 continue;
1087                         if (test_bit(CACHE_PENDING, &ch->flags))
1088                                 /* Lost a race and it is pending again */
1089                                 break;
1090                         if (cr->readers != 0)
1091                                 continue;
1092                         list_move(&cr->q.list, &dequeued);
1093                 }
1094         spin_unlock(&queue_lock);
1095         while (!list_empty(&dequeued)) {
1096                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1097                 list_del(&cr->q.list);
1098                 cache_put(cr->item, detail);
1099                 kfree(cr->buf);
1100                 kfree(cr);
1101         }
1102 }
1103
1104 /*
1105  * Support routines for text-based upcalls.
1106  * Fields are separated by spaces.
1107  * Fields are either mangled to quote space tab newline slosh with slosh
1108  * or a hexified with a leading \x
1109  * Record is terminated with newline.
1110  *
1111  */
1112
1113 void qword_add(char **bpp, int *lp, char *str)
1114 {
1115         char *bp = *bpp;
1116         int len = *lp;
1117         int ret;
1118
1119         if (len < 0) return;
1120
1121         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1122         if (ret >= len) {
1123                 bp += len;
1124                 len = -1;
1125         } else {
1126                 bp += ret;
1127                 len -= ret;
1128                 *bp++ = ' ';
1129                 len--;
1130         }
1131         *bpp = bp;
1132         *lp = len;
1133 }
1134 EXPORT_SYMBOL_GPL(qword_add);
1135
1136 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1137 {
1138         char *bp = *bpp;
1139         int len = *lp;
1140
1141         if (len < 0) return;
1142
1143         if (len > 2) {
1144                 *bp++ = '\\';
1145                 *bp++ = 'x';
1146                 len -= 2;
1147                 while (blen && len >= 2) {
1148                         bp = hex_byte_pack(bp, *buf++);
1149                         len -= 2;
1150                         blen--;
1151                 }
1152         }
1153         if (blen || len<1) len = -1;
1154         else {
1155                 *bp++ = ' ';
1156                 len--;
1157         }
1158         *bpp = bp;
1159         *lp = len;
1160 }
1161 EXPORT_SYMBOL_GPL(qword_addhex);
1162
1163 static void warn_no_listener(struct cache_detail *detail)
1164 {
1165         if (detail->last_warn != detail->last_close) {
1166                 detail->last_warn = detail->last_close;
1167                 if (detail->warn_no_listener)
1168                         detail->warn_no_listener(detail, detail->last_close != 0);
1169         }
1170 }
1171
1172 static bool cache_listeners_exist(struct cache_detail *detail)
1173 {
1174         if (atomic_read(&detail->readers))
1175                 return true;
1176         if (detail->last_close == 0)
1177                 /* This cache was never opened */
1178                 return false;
1179         if (detail->last_close < seconds_since_boot() - 30)
1180                 /*
1181                  * We allow for the possibility that someone might
1182                  * restart a userspace daemon without restarting the
1183                  * server; but after 30 seconds, we give up.
1184                  */
1185                  return false;
1186         return true;
1187 }
1188
1189 /*
1190  * register an upcall request to user-space and queue it up for read() by the
1191  * upcall daemon.
1192  *
1193  * Each request is at most one page long.
1194  */
1195 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1196 {
1197
1198         char *buf;
1199         struct cache_request *crq;
1200         int ret = 0;
1201
1202         if (!detail->cache_request)
1203                 return -EINVAL;
1204
1205         if (!cache_listeners_exist(detail)) {
1206                 warn_no_listener(detail);
1207                 return -EINVAL;
1208         }
1209         if (test_bit(CACHE_CLEANED, &h->flags))
1210                 /* Too late to make an upcall */
1211                 return -EAGAIN;
1212
1213         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1214         if (!buf)
1215                 return -EAGAIN;
1216
1217         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1218         if (!crq) {
1219                 kfree(buf);
1220                 return -EAGAIN;
1221         }
1222
1223         crq->q.reader = 0;
1224         crq->buf = buf;
1225         crq->len = 0;
1226         crq->readers = 0;
1227         spin_lock(&queue_lock);
1228         if (test_bit(CACHE_PENDING, &h->flags)) {
1229                 crq->item = cache_get(h);
1230                 list_add_tail(&crq->q.list, &detail->queue);
1231         } else
1232                 /* Lost a race, no longer PENDING, so don't enqueue */
1233                 ret = -EAGAIN;
1234         spin_unlock(&queue_lock);
1235         wake_up(&queue_wait);
1236         if (ret == -EAGAIN) {
1237                 kfree(buf);
1238                 kfree(crq);
1239         }
1240         return ret;
1241 }
1242 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1243
1244 /*
1245  * parse a message from user-space and pass it
1246  * to an appropriate cache
1247  * Messages are, like requests, separated into fields by
1248  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1249  *
1250  * Message is
1251  *   reply cachename expiry key ... content....
1252  *
1253  * key and content are both parsed by cache
1254  */
1255
1256 int qword_get(char **bpp, char *dest, int bufsize)
1257 {
1258         /* return bytes copied, or -1 on error */
1259         char *bp = *bpp;
1260         int len = 0;
1261
1262         while (*bp == ' ') bp++;
1263
1264         if (bp[0] == '\\' && bp[1] == 'x') {
1265                 /* HEX STRING */
1266                 bp += 2;
1267                 while (len < bufsize - 1) {
1268                         int h, l;
1269
1270                         h = hex_to_bin(bp[0]);
1271                         if (h < 0)
1272                                 break;
1273
1274                         l = hex_to_bin(bp[1]);
1275                         if (l < 0)
1276                                 break;
1277
1278                         *dest++ = (h << 4) | l;
1279                         bp += 2;
1280                         len++;
1281                 }
1282         } else {
1283                 /* text with \nnn octal quoting */
1284                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1285                         if (*bp == '\\' &&
1286                             isodigit(bp[1]) && (bp[1] <= '3') &&
1287                             isodigit(bp[2]) &&
1288                             isodigit(bp[3])) {
1289                                 int byte = (*++bp -'0');
1290                                 bp++;
1291                                 byte = (byte << 3) | (*bp++ - '0');
1292                                 byte = (byte << 3) | (*bp++ - '0');
1293                                 *dest++ = byte;
1294                                 len++;
1295                         } else {
1296                                 *dest++ = *bp++;
1297                                 len++;
1298                         }
1299                 }
1300         }
1301
1302         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1303                 return -1;
1304         while (*bp == ' ') bp++;
1305         *bpp = bp;
1306         *dest = '\0';
1307         return len;
1308 }
1309 EXPORT_SYMBOL_GPL(qword_get);
1310
1311
1312 /*
1313  * support /proc/net/rpc/$CACHENAME/content
1314  * as a seqfile.
1315  * We call ->cache_show passing NULL for the item to
1316  * get a header, then pass each real item in the cache
1317  */
1318
1319 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1320 {
1321         loff_t n = *pos;
1322         unsigned int hash, entry;
1323         struct cache_head *ch;
1324         struct cache_detail *cd = m->private;
1325
1326         if (!n--)
1327                 return SEQ_START_TOKEN;
1328         hash = n >> 32;
1329         entry = n & ((1LL<<32) - 1);
1330
1331         hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1332                 if (!entry--)
1333                         return ch;
1334         n &= ~((1LL<<32) - 1);
1335         do {
1336                 hash++;
1337                 n += 1LL<<32;
1338         } while(hash < cd->hash_size &&
1339                 hlist_empty(&cd->hash_table[hash]));
1340         if (hash >= cd->hash_size)
1341                 return NULL;
1342         *pos = n+1;
1343         return hlist_entry_safe(rcu_dereference_raw(
1344                                 hlist_first_rcu(&cd->hash_table[hash])),
1345                                 struct cache_head, cache_list);
1346 }
1347
1348 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1349 {
1350         struct cache_head *ch = p;
1351         int hash = (*pos >> 32);
1352         struct cache_detail *cd = m->private;
1353
1354         if (p == SEQ_START_TOKEN)
1355                 hash = 0;
1356         else if (ch->cache_list.next == NULL) {
1357                 hash++;
1358                 *pos += 1LL<<32;
1359         } else {
1360                 ++*pos;
1361                 return hlist_entry_safe(rcu_dereference_raw(
1362                                         hlist_next_rcu(&ch->cache_list)),
1363                                         struct cache_head, cache_list);
1364         }
1365         *pos &= ~((1LL<<32) - 1);
1366         while (hash < cd->hash_size &&
1367                hlist_empty(&cd->hash_table[hash])) {
1368                 hash++;
1369                 *pos += 1LL<<32;
1370         }
1371         if (hash >= cd->hash_size)
1372                 return NULL;
1373         ++*pos;
1374         return hlist_entry_safe(rcu_dereference_raw(
1375                                 hlist_first_rcu(&cd->hash_table[hash])),
1376                                 struct cache_head, cache_list);
1377 }
1378 EXPORT_SYMBOL_GPL(cache_seq_next);
1379
1380 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1381         __acquires(RCU)
1382 {
1383         rcu_read_lock();
1384         return __cache_seq_start(m, pos);
1385 }
1386 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1387
1388 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1389 {
1390         return cache_seq_next(file, p, pos);
1391 }
1392 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1393
1394 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1395         __releases(RCU)
1396 {
1397         rcu_read_unlock();
1398 }
1399 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1400
1401 static int c_show(struct seq_file *m, void *p)
1402 {
1403         struct cache_head *cp = p;
1404         struct cache_detail *cd = m->private;
1405
1406         if (p == SEQ_START_TOKEN)
1407                 return cd->cache_show(m, cd, NULL);
1408
1409         ifdebug(CACHE)
1410                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1411                            convert_to_wallclock(cp->expiry_time),
1412                            kref_read(&cp->ref), cp->flags);
1413         cache_get(cp);
1414         if (cache_check(cd, cp, NULL))
1415                 /* cache_check does a cache_put on failure */
1416                 seq_printf(m, "# ");
1417         else {
1418                 if (cache_is_expired(cd, cp))
1419                         seq_printf(m, "# ");
1420                 cache_put(cp, cd);
1421         }
1422
1423         return cd->cache_show(m, cd, cp);
1424 }
1425
1426 static const struct seq_operations cache_content_op = {
1427         .start  = cache_seq_start_rcu,
1428         .next   = cache_seq_next_rcu,
1429         .stop   = cache_seq_stop_rcu,
1430         .show   = c_show,
1431 };
1432
1433 static int content_open(struct inode *inode, struct file *file,
1434                         struct cache_detail *cd)
1435 {
1436         struct seq_file *seq;
1437         int err;
1438
1439         if (!cd || !try_module_get(cd->owner))
1440                 return -EACCES;
1441
1442         err = seq_open(file, &cache_content_op);
1443         if (err) {
1444                 module_put(cd->owner);
1445                 return err;
1446         }
1447
1448         seq = file->private_data;
1449         seq->private = cd;
1450         return 0;
1451 }
1452
1453 static int content_release(struct inode *inode, struct file *file,
1454                 struct cache_detail *cd)
1455 {
1456         int ret = seq_release(inode, file);
1457         module_put(cd->owner);
1458         return ret;
1459 }
1460
1461 static int open_flush(struct inode *inode, struct file *file,
1462                         struct cache_detail *cd)
1463 {
1464         if (!cd || !try_module_get(cd->owner))
1465                 return -EACCES;
1466         return nonseekable_open(inode, file);
1467 }
1468
1469 static int release_flush(struct inode *inode, struct file *file,
1470                         struct cache_detail *cd)
1471 {
1472         module_put(cd->owner);
1473         return 0;
1474 }
1475
1476 static ssize_t read_flush(struct file *file, char __user *buf,
1477                           size_t count, loff_t *ppos,
1478                           struct cache_detail *cd)
1479 {
1480         char tbuf[22];
1481         size_t len;
1482
1483         len = snprintf(tbuf, sizeof(tbuf), "%lu\n",
1484                         convert_to_wallclock(cd->flush_time));
1485         return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1486 }
1487
1488 static ssize_t write_flush(struct file *file, const char __user *buf,
1489                            size_t count, loff_t *ppos,
1490                            struct cache_detail *cd)
1491 {
1492         char tbuf[20];
1493         char *ep;
1494         time_t now;
1495
1496         if (*ppos || count > sizeof(tbuf)-1)
1497                 return -EINVAL;
1498         if (copy_from_user(tbuf, buf, count))
1499                 return -EFAULT;
1500         tbuf[count] = 0;
1501         simple_strtoul(tbuf, &ep, 0);
1502         if (*ep && *ep != '\n')
1503                 return -EINVAL;
1504         /* Note that while we check that 'buf' holds a valid number,
1505          * we always ignore the value and just flush everything.
1506          * Making use of the number leads to races.
1507          */
1508
1509         now = seconds_since_boot();
1510         /* Always flush everything, so behave like cache_purge()
1511          * Do this by advancing flush_time to the current time,
1512          * or by one second if it has already reached the current time.
1513          * Newly added cache entries will always have ->last_refresh greater
1514          * that ->flush_time, so they don't get flushed prematurely.
1515          */
1516
1517         if (cd->flush_time >= now)
1518                 now = cd->flush_time + 1;
1519
1520         cd->flush_time = now;
1521         cd->nextcheck = now;
1522         cache_flush();
1523
1524         *ppos += count;
1525         return count;
1526 }
1527
1528 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1529                                  size_t count, loff_t *ppos)
1530 {
1531         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1532
1533         return cache_read(filp, buf, count, ppos, cd);
1534 }
1535
1536 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1537                                   size_t count, loff_t *ppos)
1538 {
1539         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1540
1541         return cache_write(filp, buf, count, ppos, cd);
1542 }
1543
1544 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1545 {
1546         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1547
1548         return cache_poll(filp, wait, cd);
1549 }
1550
1551 static long cache_ioctl_procfs(struct file *filp,
1552                                unsigned int cmd, unsigned long arg)
1553 {
1554         struct inode *inode = file_inode(filp);
1555         struct cache_detail *cd = PDE_DATA(inode);
1556
1557         return cache_ioctl(inode, filp, cmd, arg, cd);
1558 }
1559
1560 static int cache_open_procfs(struct inode *inode, struct file *filp)
1561 {
1562         struct cache_detail *cd = PDE_DATA(inode);
1563
1564         return cache_open(inode, filp, cd);
1565 }
1566
1567 static int cache_release_procfs(struct inode *inode, struct file *filp)
1568 {
1569         struct cache_detail *cd = PDE_DATA(inode);
1570
1571         return cache_release(inode, filp, cd);
1572 }
1573
1574 static const struct file_operations cache_file_operations_procfs = {
1575         .owner          = THIS_MODULE,
1576         .llseek         = no_llseek,
1577         .read           = cache_read_procfs,
1578         .write          = cache_write_procfs,
1579         .poll           = cache_poll_procfs,
1580         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1581         .open           = cache_open_procfs,
1582         .release        = cache_release_procfs,
1583 };
1584
1585 static int content_open_procfs(struct inode *inode, struct file *filp)
1586 {
1587         struct cache_detail *cd = PDE_DATA(inode);
1588
1589         return content_open(inode, filp, cd);
1590 }
1591
1592 static int content_release_procfs(struct inode *inode, struct file *filp)
1593 {
1594         struct cache_detail *cd = PDE_DATA(inode);
1595
1596         return content_release(inode, filp, cd);
1597 }
1598
1599 static const struct file_operations content_file_operations_procfs = {
1600         .open           = content_open_procfs,
1601         .read           = seq_read,
1602         .llseek         = seq_lseek,
1603         .release        = content_release_procfs,
1604 };
1605
1606 static int open_flush_procfs(struct inode *inode, struct file *filp)
1607 {
1608         struct cache_detail *cd = PDE_DATA(inode);
1609
1610         return open_flush(inode, filp, cd);
1611 }
1612
1613 static int release_flush_procfs(struct inode *inode, struct file *filp)
1614 {
1615         struct cache_detail *cd = PDE_DATA(inode);
1616
1617         return release_flush(inode, filp, cd);
1618 }
1619
1620 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1621                             size_t count, loff_t *ppos)
1622 {
1623         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1624
1625         return read_flush(filp, buf, count, ppos, cd);
1626 }
1627
1628 static ssize_t write_flush_procfs(struct file *filp,
1629                                   const char __user *buf,
1630                                   size_t count, loff_t *ppos)
1631 {
1632         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1633
1634         return write_flush(filp, buf, count, ppos, cd);
1635 }
1636
1637 static const struct file_operations cache_flush_operations_procfs = {
1638         .open           = open_flush_procfs,
1639         .read           = read_flush_procfs,
1640         .write          = write_flush_procfs,
1641         .release        = release_flush_procfs,
1642         .llseek         = no_llseek,
1643 };
1644
1645 static void remove_cache_proc_entries(struct cache_detail *cd)
1646 {
1647         if (cd->procfs) {
1648                 proc_remove(cd->procfs);
1649                 cd->procfs = NULL;
1650         }
1651 }
1652
1653 #ifdef CONFIG_PROC_FS
1654 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1655 {
1656         struct proc_dir_entry *p;
1657         struct sunrpc_net *sn;
1658
1659         sn = net_generic(net, sunrpc_net_id);
1660         cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1661         if (cd->procfs == NULL)
1662                 goto out_nomem;
1663
1664         p = proc_create_data("flush", S_IFREG | 0600,
1665                              cd->procfs, &cache_flush_operations_procfs, cd);
1666         if (p == NULL)
1667                 goto out_nomem;
1668
1669         if (cd->cache_request || cd->cache_parse) {
1670                 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1671                                      &cache_file_operations_procfs, cd);
1672                 if (p == NULL)
1673                         goto out_nomem;
1674         }
1675         if (cd->cache_show) {
1676                 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1677                                      &content_file_operations_procfs, cd);
1678                 if (p == NULL)
1679                         goto out_nomem;
1680         }
1681         return 0;
1682 out_nomem:
1683         remove_cache_proc_entries(cd);
1684         return -ENOMEM;
1685 }
1686 #else /* CONFIG_PROC_FS */
1687 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1688 {
1689         return 0;
1690 }
1691 #endif
1692
1693 void __init cache_initialize(void)
1694 {
1695         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1696 }
1697
1698 int cache_register_net(struct cache_detail *cd, struct net *net)
1699 {
1700         int ret;
1701
1702         sunrpc_init_cache_detail(cd);
1703         ret = create_cache_proc_entries(cd, net);
1704         if (ret)
1705                 sunrpc_destroy_cache_detail(cd);
1706         return ret;
1707 }
1708 EXPORT_SYMBOL_GPL(cache_register_net);
1709
1710 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1711 {
1712         remove_cache_proc_entries(cd);
1713         sunrpc_destroy_cache_detail(cd);
1714 }
1715 EXPORT_SYMBOL_GPL(cache_unregister_net);
1716
1717 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1718 {
1719         struct cache_detail *cd;
1720         int i;
1721
1722         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1723         if (cd == NULL)
1724                 return ERR_PTR(-ENOMEM);
1725
1726         cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1727                                  GFP_KERNEL);
1728         if (cd->hash_table == NULL) {
1729                 kfree(cd);
1730                 return ERR_PTR(-ENOMEM);
1731         }
1732
1733         for (i = 0; i < cd->hash_size; i++)
1734                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1735         cd->net = net;
1736         return cd;
1737 }
1738 EXPORT_SYMBOL_GPL(cache_create_net);
1739
1740 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1741 {
1742         kfree(cd->hash_table);
1743         kfree(cd);
1744 }
1745 EXPORT_SYMBOL_GPL(cache_destroy_net);
1746
1747 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1748                                  size_t count, loff_t *ppos)
1749 {
1750         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1751
1752         return cache_read(filp, buf, count, ppos, cd);
1753 }
1754
1755 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1756                                   size_t count, loff_t *ppos)
1757 {
1758         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1759
1760         return cache_write(filp, buf, count, ppos, cd);
1761 }
1762
1763 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1764 {
1765         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1766
1767         return cache_poll(filp, wait, cd);
1768 }
1769
1770 static long cache_ioctl_pipefs(struct file *filp,
1771                               unsigned int cmd, unsigned long arg)
1772 {
1773         struct inode *inode = file_inode(filp);
1774         struct cache_detail *cd = RPC_I(inode)->private;
1775
1776         return cache_ioctl(inode, filp, cmd, arg, cd);
1777 }
1778
1779 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1780 {
1781         struct cache_detail *cd = RPC_I(inode)->private;
1782
1783         return cache_open(inode, filp, cd);
1784 }
1785
1786 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1787 {
1788         struct cache_detail *cd = RPC_I(inode)->private;
1789
1790         return cache_release(inode, filp, cd);
1791 }
1792
1793 const struct file_operations cache_file_operations_pipefs = {
1794         .owner          = THIS_MODULE,
1795         .llseek         = no_llseek,
1796         .read           = cache_read_pipefs,
1797         .write          = cache_write_pipefs,
1798         .poll           = cache_poll_pipefs,
1799         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1800         .open           = cache_open_pipefs,
1801         .release        = cache_release_pipefs,
1802 };
1803
1804 static int content_open_pipefs(struct inode *inode, struct file *filp)
1805 {
1806         struct cache_detail *cd = RPC_I(inode)->private;
1807
1808         return content_open(inode, filp, cd);
1809 }
1810
1811 static int content_release_pipefs(struct inode *inode, struct file *filp)
1812 {
1813         struct cache_detail *cd = RPC_I(inode)->private;
1814
1815         return content_release(inode, filp, cd);
1816 }
1817
1818 const struct file_operations content_file_operations_pipefs = {
1819         .open           = content_open_pipefs,
1820         .read           = seq_read,
1821         .llseek         = seq_lseek,
1822         .release        = content_release_pipefs,
1823 };
1824
1825 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1826 {
1827         struct cache_detail *cd = RPC_I(inode)->private;
1828
1829         return open_flush(inode, filp, cd);
1830 }
1831
1832 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1833 {
1834         struct cache_detail *cd = RPC_I(inode)->private;
1835
1836         return release_flush(inode, filp, cd);
1837 }
1838
1839 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1840                             size_t count, loff_t *ppos)
1841 {
1842         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1843
1844         return read_flush(filp, buf, count, ppos, cd);
1845 }
1846
1847 static ssize_t write_flush_pipefs(struct file *filp,
1848                                   const char __user *buf,
1849                                   size_t count, loff_t *ppos)
1850 {
1851         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1852
1853         return write_flush(filp, buf, count, ppos, cd);
1854 }
1855
1856 const struct file_operations cache_flush_operations_pipefs = {
1857         .open           = open_flush_pipefs,
1858         .read           = read_flush_pipefs,
1859         .write          = write_flush_pipefs,
1860         .release        = release_flush_pipefs,
1861         .llseek         = no_llseek,
1862 };
1863
1864 int sunrpc_cache_register_pipefs(struct dentry *parent,
1865                                  const char *name, umode_t umode,
1866                                  struct cache_detail *cd)
1867 {
1868         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1869         if (IS_ERR(dir))
1870                 return PTR_ERR(dir);
1871         cd->pipefs = dir;
1872         return 0;
1873 }
1874 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1875
1876 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1877 {
1878         if (cd->pipefs) {
1879                 rpc_remove_cache_dir(cd->pipefs);
1880                 cd->pipefs = NULL;
1881         }
1882 }
1883 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1884
1885 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1886 {
1887         spin_lock(&cd->hash_lock);
1888         if (!hlist_unhashed(&h->cache_list)){
1889                 hlist_del_init_rcu(&h->cache_list);
1890                 cd->entries--;
1891                 spin_unlock(&cd->hash_lock);
1892                 cache_put(h, cd);
1893         } else
1894                 spin_unlock(&cd->hash_lock);
1895 }
1896 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);