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