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