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