net/sunrpc/cache.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * net/sunrpc/cache.c
   4  *
   5  * Generic code for various authentication-related caches
   6  * used by sunrpc clients and servers.
   7  *
   8  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   9  */
  10
  11 #include <linux/types.h>
  12 #include <linux/fs.h>
  13 #include <linux/file.h>
  14 #include <linux/slab.h>
  15 #include <linux/signal.h>
  16 #include <linux/sched.h>
  17 #include <linux/kmod.h>
  18 #include <linux/list.h>
  19 #include <linux/module.h>
  20 #include <linux/ctype.h>
  21 #include <linux/string_helpers.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/poll.h>
  24 #include <linux/seq_file.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/net.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/mutex.h>
  29 #include <linux/pagemap.h>
  30 #include <asm/ioctls.h>
  31 #include <linux/sunrpc/types.h>
  32 #include <linux/sunrpc/cache.h>
  33 #include <linux/sunrpc/stats.h>
  34 #include <linux/sunrpc/rpc_pipe_fs.h>
  35 #include "netns.h"
  36
  37 #define  RPCDBG_FACILITY RPCDBG_CACHE
  38
  39 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  40 static void cache_revisit_request(struct cache_head *item);
  41 static bool cache_listeners_exist(struct cache_detail *detail);
  42
  43 static void cache_init(struct cache_head *h, struct cache_detail *detail)
  44 {
  45         time64_t now = seconds_since_boot();
  46         INIT_HLIST_NODE(&h->cache_list);
  47         h->flags = 0;
  48         kref_init(&h->ref);
  49         h->expiry_time = now + CACHE_NEW_EXPIRY;
  50         if (now <= detail->flush_time)
  51                 /* ensure it isn't already expired */
  52                 now = detail->flush_time + 1;
  53         h->last_refresh = now;
  54 }
  55
  56 static void cache_fresh_unlocked(struct cache_head *head,
  57                                 struct cache_detail *detail);
  58
  59 static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
  60                                                 struct cache_head *key,
  61                                                 int hash)
  62 {
  63         struct hlist_head *head = &detail->hash_table[hash];
  64         struct cache_head *tmp;
  65
  66         rcu_read_lock();
  67         hlist_for_each_entry_rcu(tmp, head, cache_list) {
  68                 if (detail->match(tmp, key)) {
  69                         if (cache_is_expired(detail, tmp))
  70                                 continue;
  71                         tmp = cache_get_rcu(tmp);
  72                         rcu_read_unlock();
  73                         return tmp;
  74                 }
  75         }
  76         rcu_read_unlock();
  77         return NULL;
  78 }
  79
  80 static void sunrpc_begin_cache_remove_entry(struct cache_head *ch,
  81                                             struct cache_detail *cd)
  82 {
  83         /* Must be called under cd->hash_lock */
  84         hlist_del_init_rcu(&ch->cache_list);
  85         set_bit(CACHE_CLEANED, &ch->flags);
  86         cd->entries --;
  87 }
  88
  89 static void sunrpc_end_cache_remove_entry(struct cache_head *ch,
  90                                           struct cache_detail *cd)
  91 {
  92         cache_fresh_unlocked(ch, cd);
  93         cache_put(ch, cd);
  94 }
  95
  96 static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
  97                                                  struct cache_head *key,
  98                                                  int hash)
  99 {
 100         struct cache_head *new, *tmp, *freeme = NULL;
 101         struct hlist_head *head = &detail->hash_table[hash];
 102
 103         new = detail->alloc();
 104         if (!new)
 105                 return NULL;
 106         /* must fully initialise 'new', else
 107          * we might get lose if we need to
 108          * cache_put it soon.
 109          */
 110         cache_init(new, detail);
 111         detail->init(new, key);
 112
 113         spin_lock(&detail->hash_lock);
 114
 115         /* check if entry appeared while we slept */
 116         hlist_for_each_entry_rcu(tmp, head, cache_list) {
 117                 if (detail->match(tmp, key)) {
 118                         if (cache_is_expired(detail, tmp)) {
 119                                 sunrpc_begin_cache_remove_entry(tmp, detail);
 120                                 freeme = tmp;
 121                                 break;
 122                         }
 123                         cache_get(tmp);
 124                         spin_unlock(&detail->hash_lock);
 125                         cache_put(new, detail);
 126                         return tmp;
 127                 }
 128         }
 129
 130         hlist_add_head_rcu(&new->cache_list, head);
 131         detail->entries++;
 132         cache_get(new);
 133         spin_unlock(&detail->hash_lock);
 134
 135         if (freeme)
 136                 sunrpc_end_cache_remove_entry(freeme, detail);
 137         return new;
 138 }
 139
 140 struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
 141                                            struct cache_head *key, int hash)
 142 {
 143         struct cache_head *ret;
 144
 145         ret = sunrpc_cache_find_rcu(detail, key, hash);
 146         if (ret)
 147                 return ret;
 148         /* Didn't find anything, insert an empty entry */
 149         return sunrpc_cache_add_entry(detail, key, hash);
 150 }
 151 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
 152
 153 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 154
 155 static void cache_fresh_locked(struct cache_head *head, time64_t expiry,
 156                                struct cache_detail *detail)
 157 {
 158         time64_t now = seconds_since_boot();
 159         if (now <= detail->flush_time)
 160                 /* ensure it isn't immediately treated as expired */
 161                 now = detail->flush_time + 1;
 162         head->expiry_time = expiry;
 163         head->last_refresh = now;
 164         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 165         set_bit(CACHE_VALID, &head->flags);
 166 }
 167
 168 static void cache_fresh_unlocked(struct cache_head *head,
 169                                  struct cache_detail *detail)
 170 {
 171         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 172                 cache_revisit_request(head);
 173                 cache_dequeue(detail, head);
 174         }
 175 }
 176
 177 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 178                                        struct cache_head *new, struct cache_head *old, int hash)
 179 {
 180         /* The 'old' entry is to be replaced by 'new'.
 181          * If 'old' is not VALID, we update it directly,
 182          * otherwise we need to replace it
 183          */
 184         struct cache_head *tmp;
 185
 186         if (!test_bit(CACHE_VALID, &old->flags)) {
 187                 spin_lock(&detail->hash_lock);
 188                 if (!test_bit(CACHE_VALID, &old->flags)) {
 189                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 190                                 set_bit(CACHE_NEGATIVE, &old->flags);
 191                         else
 192                                 detail->update(old, new);
 193                         cache_fresh_locked(old, new->expiry_time, detail);
 194                         spin_unlock(&detail->hash_lock);
 195                         cache_fresh_unlocked(old, detail);
 196                         return old;
 197                 }
 198                 spin_unlock(&detail->hash_lock);
 199         }
 200         /* We need to insert a new entry */
 201         tmp = detail->alloc();
 202         if (!tmp) {
 203                 cache_put(old, detail);
 204                 return NULL;
 205         }
 206         cache_init(tmp, detail);
 207         detail->init(tmp, old);
 208
 209         spin_lock(&detail->hash_lock);
 210         if (test_bit(CACHE_NEGATIVE, &new->flags))
 211                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 212         else
 213                 detail->update(tmp, new);
 214         hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
 215         detail->entries++;
 216         cache_get(tmp);
 217         cache_fresh_locked(tmp, new->expiry_time, detail);
 218         cache_fresh_locked(old, 0, detail);
 219         spin_unlock(&detail->hash_lock);
 220         cache_fresh_unlocked(tmp, detail);
 221         cache_fresh_unlocked(old, detail);
 222         cache_put(old, detail);
 223         return tmp;
 224 }
 225 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 226
 227 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 228 {
 229         if (cd->cache_upcall)
 230                 return cd->cache_upcall(cd, h);
 231         return sunrpc_cache_pipe_upcall(cd, h);
 232 }
 233
 234 static inline int cache_is_valid(struct cache_head *h)
 235 {
 236         if (!test_bit(CACHE_VALID, &h->flags))
 237                 return -EAGAIN;
 238         else {
 239                 /* entry is valid */
 240                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 241                         return -ENOENT;
 242                 else {
 243                         /*
 244                          * In combination with write barrier in
 245                          * sunrpc_cache_update, ensures that anyone
 246                          * using the cache entry after this sees the
 247                          * updated contents:
 248                          */
 249                         smp_rmb();
 250                         return 0;
 251                 }
 252         }
 253 }
 254
 255 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 256 {
 257         int rv;
 258
 259         spin_lock(&detail->hash_lock);
 260         rv = cache_is_valid(h);
 261         if (rv == -EAGAIN) {
 262                 set_bit(CACHE_NEGATIVE, &h->flags);
 263                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
 264                                    detail);
 265                 rv = -ENOENT;
 266         }
 267         spin_unlock(&detail->hash_lock);
 268         cache_fresh_unlocked(h, detail);
 269         return rv;
 270 }
 271
 272 /*
 273  * This is the generic cache management routine for all
 274  * the authentication caches.
 275  * It checks the currency of a cache item and will (later)
 276  * initiate an upcall to fill it if needed.
 277  *
 278  *
 279  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 280  * -EAGAIN if upcall is pending and request has been queued
 281  * -ETIMEDOUT if upcall failed or request could not be queue or
 282  *           upcall completed but item is still invalid (implying that
 283  *           the cache item has been replaced with a newer one).
 284  * -ENOENT if cache entry was negative
 285  */
 286 int cache_check(struct cache_detail *detail,
 287                     struct cache_head *h, struct cache_req *rqstp)
 288 {
 289         int rv;
 290         time64_t refresh_age, age;
 291
 292         /* First decide return status as best we can */
 293         rv = cache_is_valid(h);
 294
 295         /* now see if we want to start an upcall */
 296         refresh_age = (h->expiry_time - h->last_refresh);
 297         age = seconds_since_boot() - h->last_refresh;
 298
 299         if (rqstp == NULL) {
 300                 if (rv == -EAGAIN)
 301                         rv = -ENOENT;
 302         } else if (rv == -EAGAIN ||
 303                    (h->expiry_time != 0 && age > refresh_age/2)) {
 304                 dprintk("RPC:       Want update, refage=%lld, age=%lld\n",
 305                                 refresh_age, age);
 306                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 307                         switch (cache_make_upcall(detail, h)) {
 308                         case -EINVAL:
 309                                 rv = try_to_negate_entry(detail, h);
 310                                 break;
 311                         case -EAGAIN:
 312                                 cache_fresh_unlocked(h, detail);
 313                                 break;
 314                         }
 315                 } else if (!cache_listeners_exist(detail))
 316                         rv = try_to_negate_entry(detail, h);
 317         }
 318
 319         if (rv == -EAGAIN) {
 320                 if (!cache_defer_req(rqstp, h)) {
 321                         /*
 322                          * Request was not deferred; handle it as best
 323                          * we can ourselves:
 324                          */
 325                         rv = cache_is_valid(h);
 326                         if (rv == -EAGAIN)
 327                                 rv = -ETIMEDOUT;
 328                 }
 329         }
 330         if (rv)
 331                 cache_put(h, detail);
 332         return rv;
 333 }
 334 EXPORT_SYMBOL_GPL(cache_check);
 335
 336 /*
 337  * caches need to be periodically cleaned.
 338  * For this we maintain a list of cache_detail and
 339  * a current pointer into that list and into the table
 340  * for that entry.
 341  *
 342  * Each time cache_clean is called it finds the next non-empty entry
 343  * in the current table and walks the list in that entry
 344  * looking for entries that can be removed.
 345  *
 346  * An entry gets removed if:
 347  * - The expiry is before current time
 348  * - The last_refresh time is before the flush_time for that cache
 349  *
 350  * later we might drop old entries with non-NEVER expiry if that table
 351  * is getting 'full' for some definition of 'full'
 352  *
 353  * The question of "how often to scan a table" is an interesting one
 354  * and is answered in part by the use of the "nextcheck" field in the
 355  * cache_detail.
 356  * When a scan of a table begins, the nextcheck field is set to a time
 357  * that is well into the future.
 358  * While scanning, if an expiry time is found that is earlier than the
 359  * current nextcheck time, nextcheck is set to that expiry time.
 360  * If the flush_time is ever set to a time earlier than the nextcheck
 361  * time, the nextcheck time is then set to that flush_time.
 362  *
 363  * A table is then only scanned if the current time is at least
 364  * the nextcheck time.
 365  *
 366  */
 367
 368 static LIST_HEAD(cache_list);
 369 static DEFINE_SPINLOCK(cache_list_lock);
 370 static struct cache_detail *current_detail;
 371 static int current_index;
 372
 373 static void do_cache_clean(struct work_struct *work);
 374 static struct delayed_work cache_cleaner;
 375
 376 void sunrpc_init_cache_detail(struct cache_detail *cd)
 377 {
 378         spin_lock_init(&cd->hash_lock);
 379         INIT_LIST_HEAD(&cd->queue);
 380         spin_lock(&cache_list_lock);
 381         cd->nextcheck = 0;
 382         cd->entries = 0;
 383         atomic_set(&cd->writers, 0);
 384         cd->last_close = 0;
 385         cd->last_warn = -1;
 386         list_add(&cd->others, &cache_list);
 387         spin_unlock(&cache_list_lock);
 388
 389         /* start the cleaning process */
 390         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
 391 }
 392 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 393
 394 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 395 {
 396         cache_purge(cd);
 397         spin_lock(&cache_list_lock);
 398         spin_lock(&cd->hash_lock);
 399         if (current_detail == cd)
 400                 current_detail = NULL;
 401         list_del_init(&cd->others);
 402         spin_unlock(&cd->hash_lock);
 403         spin_unlock(&cache_list_lock);
 404         if (list_empty(&cache_list)) {
 405                 /* module must be being unloaded so its safe to kill the worker */
 406                 cancel_delayed_work_sync(&cache_cleaner);
 407         }
 408 }
 409 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 410
 411 /* clean cache tries to find something to clean
 412  * and cleans it.
 413  * It returns 1 if it cleaned something,
 414  *            0 if it didn't find anything this time
 415  *           -1 if it fell off the end of the list.
 416  */
 417 static int cache_clean(void)
 418 {
 419         int rv = 0;
 420         struct list_head *next;
 421
 422         spin_lock(&cache_list_lock);
 423
 424         /* find a suitable table if we don't already have one */
 425         while (current_detail == NULL ||
 426             current_index >= current_detail->hash_size) {
 427                 if (current_detail)
 428                         next = current_detail->others.next;
 429                 else
 430                         next = cache_list.next;
 431                 if (next == &cache_list) {
 432                         current_detail = NULL;
 433                         spin_unlock(&cache_list_lock);
 434                         return -1;
 435                 }
 436                 current_detail = list_entry(next, struct cache_detail, others);
 437                 if (current_detail->nextcheck > seconds_since_boot())
 438                         current_index = current_detail->hash_size;
 439                 else {
 440                         current_index = 0;
 441                         current_detail->nextcheck = seconds_since_boot()+30*60;
 442                 }
 443         }
 444
 445         /* find a non-empty bucket in the table */
 446         while (current_detail &&
 447                current_index < current_detail->hash_size &&
 448                hlist_empty(&current_detail->hash_table[current_index]))
 449                 current_index++;
 450
 451         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 452
 453         if (current_detail && current_index < current_detail->hash_size) {
 454                 struct cache_head *ch = NULL;
 455                 struct cache_detail *d;
 456                 struct hlist_head *head;
 457                 struct hlist_node *tmp;
 458
 459                 spin_lock(&current_detail->hash_lock);
 460
 461                 /* Ok, now to clean this strand */
 462
 463                 head = &current_detail->hash_table[current_index];
 464                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 465                         if (current_detail->nextcheck > ch->expiry_time)
 466                                 current_detail->nextcheck = ch->expiry_time+1;
 467                         if (!cache_is_expired(current_detail, ch))
 468                                 continue;
 469
 470                         sunrpc_begin_cache_remove_entry(ch, current_detail);
 471                         rv = 1;
 472                         break;
 473                 }
 474
 475                 spin_unlock(&current_detail->hash_lock);
 476                 d = current_detail;
 477                 if (!ch)
 478                         current_index ++;
 479                 spin_unlock(&cache_list_lock);
 480                 if (ch)
 481                         sunrpc_end_cache_remove_entry(ch, d);
 482         } else
 483                 spin_unlock(&cache_list_lock);
 484
 485         return rv;
 486 }
 487
 488 /*
 489  * We want to regularly clean the cache, so we need to schedule some work ...
 490  */
 491 static void do_cache_clean(struct work_struct *work)
 492 {
 493         int delay = 5;
 494         if (cache_clean() == -1)
 495                 delay = round_jiffies_relative(30*HZ);
 496
 497         if (list_empty(&cache_list))
 498                 delay = 0;
 499
 500         if (delay)
 501                 queue_delayed_work(system_power_efficient_wq,
 502                                    &cache_cleaner, delay);
 503 }
 504
 505
 506 /*
 507  * Clean all caches promptly.  This just calls cache_clean
 508  * repeatedly until we are sure that every cache has had a chance to
 509  * be fully cleaned
 510  */
 511 void cache_flush(void)
 512 {
 513         while (cache_clean() != -1)
 514                 cond_resched();
 515         while (cache_clean() != -1)
 516                 cond_resched();
 517 }
 518 EXPORT_SYMBOL_GPL(cache_flush);
 519
 520 void cache_purge(struct cache_detail *detail)
 521 {
 522         struct cache_head *ch = NULL;
 523         struct hlist_head *head = NULL;
 524         struct hlist_node *tmp = NULL;
 525         int i = 0;
 526
 527         spin_lock(&detail->hash_lock);
 528         if (!detail->entries) {
 529                 spin_unlock(&detail->hash_lock);
 530                 return;
 531         }
 532
 533         dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
 534         for (i = 0; i < detail->hash_size; i++) {
 535                 head = &detail->hash_table[i];
 536                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 537                         sunrpc_begin_cache_remove_entry(ch, detail);
 538                         spin_unlock(&detail->hash_lock);
 539                         sunrpc_end_cache_remove_entry(ch, detail);
 540                         spin_lock(&detail->hash_lock);
 541                 }
 542         }
 543         spin_unlock(&detail->hash_lock);
 544 }
 545 EXPORT_SYMBOL_GPL(cache_purge);
 546
 547
 548 /*
 549  * Deferral and Revisiting of Requests.
 550  *
 551  * If a cache lookup finds a pending entry, we
 552  * need to defer the request and revisit it later.
 553  * All deferred requests are stored in a hash table,
 554  * indexed by "struct cache_head *".
 555  * As it may be wasteful to store a whole request
 556  * structure, we allow the request to provide a
 557  * deferred form, which must contain a
 558  * 'struct cache_deferred_req'
 559  * This cache_deferred_req contains a method to allow
 560  * it to be revisited when cache info is available
 561  */
 562
 563 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 564 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 565
 566 #define DFR_MAX 300     /* ??? */
 567
 568 static DEFINE_SPINLOCK(cache_defer_lock);
 569 static LIST_HEAD(cache_defer_list);
 570 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 571 static int cache_defer_cnt;
 572
 573 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 574 {
 575         hlist_del_init(&dreq->hash);
 576         if (!list_empty(&dreq->recent)) {
 577                 list_del_init(&dreq->recent);
 578                 cache_defer_cnt--;
 579         }
 580 }
 581
 582 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 583 {
 584         int hash = DFR_HASH(item);
 585
 586         INIT_LIST_HEAD(&dreq->recent);
 587         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 588 }
 589
 590 static void setup_deferral(struct cache_deferred_req *dreq,
 591                            struct cache_head *item,
 592                            int count_me)
 593 {
 594
 595         dreq->item = item;
 596
 597         spin_lock(&cache_defer_lock);
 598
 599         __hash_deferred_req(dreq, item);
 600
 601         if (count_me) {
 602                 cache_defer_cnt++;
 603                 list_add(&dreq->recent, &cache_defer_list);
 604         }
 605
 606         spin_unlock(&cache_defer_lock);
 607
 608 }
 609
 610 struct thread_deferred_req {
 611         struct cache_deferred_req handle;
 612         struct completion completion;
 613 };
 614
 615 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 616 {
 617         struct thread_deferred_req *dr =
 618                 container_of(dreq, struct thread_deferred_req, handle);
 619         complete(&dr->completion);
 620 }
 621
 622 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 623 {
 624         struct thread_deferred_req sleeper;
 625         struct cache_deferred_req *dreq = &sleeper.handle;
 626
 627         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 628         dreq->revisit = cache_restart_thread;
 629
 630         setup_deferral(dreq, item, 0);
 631
 632         if (!test_bit(CACHE_PENDING, &item->flags) ||
 633             wait_for_completion_interruptible_timeout(
 634                     &sleeper.completion, req->thread_wait) <= 0) {
 635                 /* The completion wasn't completed, so we need
 636                  * to clean up
 637                  */
 638                 spin_lock(&cache_defer_lock);
 639                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 640                         __unhash_deferred_req(&sleeper.handle);
 641                         spin_unlock(&cache_defer_lock);
 642                 } else {
 643                         /* cache_revisit_request already removed
 644                          * this from the hash table, but hasn't
 645                          * called ->revisit yet.  It will very soon
 646                          * and we need to wait for it.
 647                          */
 648                         spin_unlock(&cache_defer_lock);
 649                         wait_for_completion(&sleeper.completion);
 650                 }
 651         }
 652 }
 653
 654 static void cache_limit_defers(void)
 655 {
 656         /* Make sure we haven't exceed the limit of allowed deferred
 657          * requests.
 658          */
 659         struct cache_deferred_req *discard = NULL;
 660
 661         if (cache_defer_cnt <= DFR_MAX)
 662                 return;
 663
 664         spin_lock(&cache_defer_lock);
 665
 666         /* Consider removing either the first or the last */
 667         if (cache_defer_cnt > DFR_MAX) {
 668                 if (prandom_u32() & 1)
 669                         discard = list_entry(cache_defer_list.next,
 670                                              struct cache_deferred_req, recent);
 671                 else
 672                         discard = list_entry(cache_defer_list.prev,
 673                                              struct cache_deferred_req, recent);
 674                 __unhash_deferred_req(discard);
 675         }
 676         spin_unlock(&cache_defer_lock);
 677         if (discard)
 678                 discard->revisit(discard, 1);
 679 }
 680
 681 /* Return true if and only if a deferred request is queued. */
 682 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 683 {
 684         struct cache_deferred_req *dreq;
 685
 686         if (req->thread_wait) {
 687                 cache_wait_req(req, item);
 688                 if (!test_bit(CACHE_PENDING, &item->flags))
 689                         return false;
 690         }
 691         dreq = req->defer(req);
 692         if (dreq == NULL)
 693                 return false;
 694         setup_deferral(dreq, item, 1);
 695         if (!test_bit(CACHE_PENDING, &item->flags))
 696                 /* Bit could have been cleared before we managed to
 697                  * set up the deferral, so need to revisit just in case
 698                  */
 699                 cache_revisit_request(item);
 700
 701         cache_limit_defers();
 702         return true;
 703 }
 704
 705 static void cache_revisit_request(struct cache_head *item)
 706 {
 707         struct cache_deferred_req *dreq;
 708         struct list_head pending;
 709         struct hlist_node *tmp;
 710         int hash = DFR_HASH(item);
 711
 712         INIT_LIST_HEAD(&pending);
 713         spin_lock(&cache_defer_lock);
 714
 715         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 716                 if (dreq->item == item) {
 717                         __unhash_deferred_req(dreq);
 718                         list_add(&dreq->recent, &pending);
 719                 }
 720
 721         spin_unlock(&cache_defer_lock);
 722
 723         while (!list_empty(&pending)) {
 724                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 725                 list_del_init(&dreq->recent);
 726                 dreq->revisit(dreq, 0);
 727         }
 728 }
 729
 730 void cache_clean_deferred(void *owner)
 731 {
 732         struct cache_deferred_req *dreq, *tmp;
 733         struct list_head pending;
 734
 735
 736         INIT_LIST_HEAD(&pending);
 737         spin_lock(&cache_defer_lock);
 738
 739         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 740                 if (dreq->owner == owner) {
 741                         __unhash_deferred_req(dreq);
 742                         list_add(&dreq->recent, &pending);
 743                 }
 744         }
 745         spin_unlock(&cache_defer_lock);
 746
 747         while (!list_empty(&pending)) {
 748                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 749                 list_del_init(&dreq->recent);
 750                 dreq->revisit(dreq, 1);
 751         }
 752 }
 753
 754 /*
 755  * communicate with user-space
 756  *
 757  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
 758  * On read, you get a full request, or block.
 759  * On write, an update request is processed.
 760  * Poll works if anything to read, and always allows write.
 761  *
 762  * Implemented by linked list of requests.  Each open file has
 763  * a ->private that also exists in this list.  New requests are added
 764  * to the end and may wakeup and preceding readers.
 765  * New readers are added to the head.  If, on read, an item is found with
 766  * CACHE_UPCALLING clear, we free it from the list.
 767  *
 768  */
 769
 770 static DEFINE_SPINLOCK(queue_lock);
 771 static DEFINE_MUTEX(queue_io_mutex);
 772
 773 struct cache_queue {
 774         struct list_head        list;
 775         int                     reader; /* if 0, then request */
 776 };
 777 struct cache_request {
 778         struct cache_queue      q;
 779         struct cache_head       *item;
 780         char                    * buf;
 781         int                     len;
 782         int                     readers;
 783 };
 784 struct cache_reader {
 785         struct cache_queue      q;
 786         int                     offset; /* if non-0, we have a refcnt on next request */
 787 };
 788
 789 static int cache_request(struct cache_detail *detail,
 790                                struct cache_request *crq)
 791 {
 792         char *bp = crq->buf;
 793         int len = PAGE_SIZE;
 794
 795         detail->cache_request(detail, crq->item, &bp, &len);
 796         if (len < 0)
 797                 return -EAGAIN;
 798         return PAGE_SIZE - len;
 799 }
 800
 801 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 802                           loff_t *ppos, struct cache_detail *cd)
 803 {
 804         struct cache_reader *rp = filp->private_data;
 805         struct cache_request *rq;
 806         struct inode *inode = file_inode(filp);
 807         int err;
 808
 809         if (count == 0)
 810                 return 0;
 811
 812         inode_lock(inode); /* protect against multiple concurrent
 813                               * readers on this file */
 814  again:
 815         spin_lock(&queue_lock);
 816         /* need to find next request */
 817         while (rp->q.list.next != &cd->queue &&
 818                list_entry(rp->q.list.next, struct cache_queue, list)
 819                ->reader) {
 820                 struct list_head *next = rp->q.list.next;
 821                 list_move(&rp->q.list, next);
 822         }
 823         if (rp->q.list.next == &cd->queue) {
 824                 spin_unlock(&queue_lock);
 825                 inode_unlock(inode);
 826                 WARN_ON_ONCE(rp->offset);
 827                 return 0;
 828         }
 829         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 830         WARN_ON_ONCE(rq->q.reader);
 831         if (rp->offset == 0)
 832                 rq->readers++;
 833         spin_unlock(&queue_lock);
 834
 835         if (rq->len == 0) {
 836                 err = cache_request(cd, rq);
 837                 if (err < 0)
 838                         goto out;
 839                 rq->len = err;
 840         }
 841
 842         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 843                 err = -EAGAIN;
 844                 spin_lock(&queue_lock);
 845                 list_move(&rp->q.list, &rq->q.list);
 846                 spin_unlock(&queue_lock);
 847         } else {
 848                 if (rp->offset + count > rq->len)
 849                         count = rq->len - rp->offset;
 850                 err = -EFAULT;
 851                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 852                         goto out;
 853                 rp->offset += count;
 854                 if (rp->offset >= rq->len) {
 855                         rp->offset = 0;
 856                         spin_lock(&queue_lock);
 857                         list_move(&rp->q.list, &rq->q.list);
 858                         spin_unlock(&queue_lock);
 859                 }
 860                 err = 0;
 861         }
 862  out:
 863         if (rp->offset == 0) {
 864                 /* need to release rq */
 865                 spin_lock(&queue_lock);
 866                 rq->readers--;
 867                 if (rq->readers == 0 &&
 868                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 869                         list_del(&rq->q.list);
 870                         spin_unlock(&queue_lock);
 871                         cache_put(rq->item, cd);
 872                         kfree(rq->buf);
 873                         kfree(rq);
 874                 } else
 875                         spin_unlock(&queue_lock);
 876         }
 877         if (err == -EAGAIN)
 878                 goto again;
 879         inode_unlock(inode);
 880         return err ? err :  count;
 881 }
 882
 883 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 884                                  size_t count, struct cache_detail *cd)
 885 {
 886         ssize_t ret;
 887
 888         if (count == 0)
 889                 return -EINVAL;
 890         if (copy_from_user(kaddr, buf, count))
 891                 return -EFAULT;
 892         kaddr[count] = '\0';
 893         ret = cd->cache_parse(cd, kaddr, count);
 894         if (!ret)
 895                 ret = count;
 896         return ret;
 897 }
 898
 899 static ssize_t cache_slow_downcall(const char __user *buf,
 900                                    size_t count, struct cache_detail *cd)
 901 {
 902         static char write_buf[8192]; /* protected by queue_io_mutex */
 903         ssize_t ret = -EINVAL;
 904
 905         if (count >= sizeof(write_buf))
 906                 goto out;
 907         mutex_lock(&queue_io_mutex);
 908         ret = cache_do_downcall(write_buf, buf, count, cd);
 909         mutex_unlock(&queue_io_mutex);
 910 out:
 911         return ret;
 912 }
 913
 914 static ssize_t cache_downcall(struct address_space *mapping,
 915                               const char __user *buf,
 916                               size_t count, struct cache_detail *cd)
 917 {
 918         struct page *page;
 919         char *kaddr;
 920         ssize_t ret = -ENOMEM;
 921
 922         if (count >= PAGE_SIZE)
 923                 goto out_slow;
 924
 925         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 926         if (!page)
 927                 goto out_slow;
 928
 929         kaddr = kmap(page);
 930         ret = cache_do_downcall(kaddr, buf, count, cd);
 931         kunmap(page);
 932         unlock_page(page);
 933         put_page(page);
 934         return ret;
 935 out_slow:
 936         return cache_slow_downcall(buf, count, cd);
 937 }
 938
 939 static ssize_t cache_write(struct file *filp, const char __user *buf,
 940                            size_t count, loff_t *ppos,
 941                            struct cache_detail *cd)
 942 {
 943         struct address_space *mapping = filp->f_mapping;
 944         struct inode *inode = file_inode(filp);
 945         ssize_t ret = -EINVAL;
 946
 947         if (!cd->cache_parse)
 948                 goto out;
 949
 950         inode_lock(inode);
 951         ret = cache_downcall(mapping, buf, count, cd);
 952         inode_unlock(inode);
 953 out:
 954         return ret;
 955 }
 956
 957 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 958
 959 static __poll_t cache_poll(struct file *filp, poll_table *wait,
 960                                struct cache_detail *cd)
 961 {
 962         __poll_t mask;
 963         struct cache_reader *rp = filp->private_data;
 964         struct cache_queue *cq;
 965
 966         poll_wait(filp, &queue_wait, wait);
 967
 968         /* alway allow write */
 969         mask = EPOLLOUT | EPOLLWRNORM;
 970
 971         if (!rp)
 972                 return mask;
 973
 974         spin_lock(&queue_lock);
 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                         mask |= EPOLLIN | EPOLLRDNORM;
 980                         break;
 981                 }
 982         spin_unlock(&queue_lock);
 983         return mask;
 984 }
 985
 986 static int cache_ioctl(struct inode *ino, struct file *filp,
 987                        unsigned int cmd, unsigned long arg,
 988                        struct cache_detail *cd)
 989 {
 990         int len = 0;
 991         struct cache_reader *rp = filp->private_data;
 992         struct cache_queue *cq;
 993
 994         if (cmd != FIONREAD || !rp)
 995                 return -EINVAL;
 996
 997         spin_lock(&queue_lock);
 998
 999         /* only find the length remaining in current request,
1000          * or the length of the next request
1001          */
1002         for (cq= &rp->q; &cq->list != &cd->queue;
1003              cq = list_entry(cq->list.next, struct cache_queue, list))
1004                 if (!cq->reader) {
1005                         struct cache_request *cr =
1006                                 container_of(cq, struct cache_request, q);
1007                         len = cr->len - rp->offset;
1008                         break;
1009                 }
1010         spin_unlock(&queue_lock);
1011
1012         return put_user(len, (int __user *)arg);
1013 }
1014
1015 static int cache_open(struct inode *inode, struct file *filp,
1016                       struct cache_detail *cd)
1017 {
1018         struct cache_reader *rp = NULL;
1019
1020         if (!cd || !try_module_get(cd->owner))
1021                 return -EACCES;
1022         nonseekable_open(inode, filp);
1023         if (filp->f_mode & FMODE_READ) {
1024                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1025                 if (!rp) {
1026                         module_put(cd->owner);
1027                         return -ENOMEM;
1028                 }
1029                 rp->offset = 0;
1030                 rp->q.reader = 1;
1031
1032                 spin_lock(&queue_lock);
1033                 list_add(&rp->q.list, &cd->queue);
1034                 spin_unlock(&queue_lock);
1035         }
1036         if (filp->f_mode & FMODE_WRITE)
1037                 atomic_inc(&cd->writers);
1038         filp->private_data = rp;
1039         return 0;
1040 }
1041
1042 static int cache_release(struct inode *inode, struct file *filp,
1043                          struct cache_detail *cd)
1044 {
1045         struct cache_reader *rp = filp->private_data;
1046
1047         if (rp) {
1048                 spin_lock(&queue_lock);
1049                 if (rp->offset) {
1050                         struct cache_queue *cq;
1051                         for (cq= &rp->q; &cq->list != &cd->queue;
1052                              cq = list_entry(cq->list.next, struct cache_queue, list))
1053                                 if (!cq->reader) {
1054                                         container_of(cq, struct cache_request, q)
1055                                                 ->readers--;
1056                                         break;
1057                                 }
1058                         rp->offset = 0;
1059                 }
1060                 list_del(&rp->q.list);
1061                 spin_unlock(&queue_lock);
1062
1063                 filp->private_data = NULL;
1064                 kfree(rp);
1065
1066         }
1067         if (filp->f_mode & FMODE_WRITE) {
1068                 atomic_dec(&cd->writers);
1069                 cd->last_close = seconds_since_boot();
1070         }
1071         module_put(cd->owner);
1072         return 0;
1073 }
1074
1075
1076
1077 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1078 {
1079         struct cache_queue *cq, *tmp;
1080         struct cache_request *cr;
1081         struct list_head dequeued;
1082
1083         INIT_LIST_HEAD(&dequeued);
1084         spin_lock(&queue_lock);
1085         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1086                 if (!cq->reader) {
1087                         cr = container_of(cq, struct cache_request, q);
1088                         if (cr->item != ch)
1089                                 continue;
1090                         if (test_bit(CACHE_PENDING, &ch->flags))
1091                                 /* Lost a race and it is pending again */
1092                                 break;
1093                         if (cr->readers != 0)
1094                                 continue;
1095                         list_move(&cr->q.list, &dequeued);
1096                 }
1097         spin_unlock(&queue_lock);
1098         while (!list_empty(&dequeued)) {
1099                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1100                 list_del(&cr->q.list);
1101                 cache_put(cr->item, detail);
1102                 kfree(cr->buf);
1103                 kfree(cr);
1104         }
1105 }
1106
1107 /*
1108  * Support routines for text-based upcalls.
1109  * Fields are separated by spaces.
1110  * Fields are either mangled to quote space tab newline slosh with slosh
1111  * or a hexified with a leading \x
1112  * Record is terminated with newline.
1113  *
1114  */
1115
1116 void qword_add(char **bpp, int *lp, char *str)
1117 {
1118         char *bp = *bpp;
1119         int len = *lp;
1120         int ret;
1121
1122         if (len < 0) return;
1123
1124         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1125         if (ret >= len) {
1126                 bp += len;
1127                 len = -1;
1128         } else {
1129                 bp += ret;
1130                 len -= ret;
1131                 *bp++ = ' ';
1132                 len--;
1133         }
1134         *bpp = bp;
1135         *lp = len;
1136 }
1137 EXPORT_SYMBOL_GPL(qword_add);
1138
1139 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1140 {
1141         char *bp = *bpp;
1142         int len = *lp;
1143
1144         if (len < 0) return;
1145
1146         if (len > 2) {
1147                 *bp++ = '\\';
1148                 *bp++ = 'x';
1149                 len -= 2;
1150                 while (blen && len >= 2) {
1151                         bp = hex_byte_pack(bp, *buf++);
1152                         len -= 2;
1153                         blen--;
1154                 }
1155         }
1156         if (blen || len<1) len = -1;
1157         else {
1158                 *bp++ = ' ';
1159                 len--;
1160         }
1161         *bpp = bp;
1162         *lp = len;
1163 }
1164 EXPORT_SYMBOL_GPL(qword_addhex);
1165
1166 static void warn_no_listener(struct cache_detail *detail)
1167 {
1168         if (detail->last_warn != detail->last_close) {
1169                 detail->last_warn = detail->last_close;
1170                 if (detail->warn_no_listener)
1171                         detail->warn_no_listener(detail, detail->last_close != 0);
1172         }
1173 }
1174
1175 static bool cache_listeners_exist(struct cache_detail *detail)
1176 {
1177         if (atomic_read(&detail->writers))
1178                 return true;
1179         if (detail->last_close == 0)
1180                 /* This cache was never opened */
1181                 return false;
1182         if (detail->last_close < seconds_since_boot() - 30)
1183                 /*
1184                  * We allow for the possibility that someone might
1185                  * restart a userspace daemon without restarting the
1186                  * server; but after 30 seconds, we give up.
1187                  */
1188                  return false;
1189         return true;
1190 }
1191
1192 /*
1193  * register an upcall request to user-space and queue it up for read() by the
1194  * upcall daemon.
1195  *
1196  * Each request is at most one page long.
1197  */
1198 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1199 {
1200
1201         char *buf;
1202         struct cache_request *crq;
1203         int ret = 0;
1204
1205         if (!detail->cache_request)
1206                 return -EINVAL;
1207
1208         if (!cache_listeners_exist(detail)) {
1209                 warn_no_listener(detail);
1210                 return -EINVAL;
1211         }
1212         if (test_bit(CACHE_CLEANED, &h->flags))
1213                 /* Too late to make an upcall */
1214                 return -EAGAIN;
1215
1216         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1217         if (!buf)
1218                 return -EAGAIN;
1219
1220         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1221         if (!crq) {
1222                 kfree(buf);
1223                 return -EAGAIN;
1224         }
1225
1226         crq->q.reader = 0;
1227         crq->buf = buf;
1228         crq->len = 0;
1229         crq->readers = 0;
1230         spin_lock(&queue_lock);
1231         if (test_bit(CACHE_PENDING, &h->flags)) {
1232                 crq->item = cache_get(h);
1233                 list_add_tail(&crq->q.list, &detail->queue);
1234         } else
1235                 /* Lost a race, no longer PENDING, so don't enqueue */
1236                 ret = -EAGAIN;
1237         spin_unlock(&queue_lock);
1238         wake_up(&queue_wait);
1239         if (ret == -EAGAIN) {
1240                 kfree(buf);
1241                 kfree(crq);
1242         }
1243         return ret;
1244 }
1245 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1246
1247 /*
1248  * parse a message from user-space and pass it
1249  * to an appropriate cache
1250  * Messages are, like requests, separated into fields by
1251  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1252  *
1253  * Message is
1254  *   reply cachename expiry key ... content....
1255  *
1256  * key and content are both parsed by cache
1257  */
1258
1259 int qword_get(char **bpp, char *dest, int bufsize)
1260 {
1261         /* return bytes copied, or -1 on error */
1262         char *bp = *bpp;
1263         int len = 0;
1264
1265         while (*bp == ' ') bp++;
1266
1267         if (bp[0] == '\\' && bp[1] == 'x') {
1268                 /* HEX STRING */
1269                 bp += 2;
1270                 while (len < bufsize - 1) {
1271                         int h, l;
1272
1273                         h = hex_to_bin(bp[0]);
1274                         if (h < 0)
1275                                 break;
1276
1277                         l = hex_to_bin(bp[1]);
1278                         if (l < 0)
1279                                 break;
1280
1281                         *dest++ = (h << 4) | l;
1282                         bp += 2;
1283                         len++;
1284                 }
1285         } else {
1286                 /* text with \nnn octal quoting */
1287                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1288                         if (*bp == '\\' &&
1289                             isodigit(bp[1]) && (bp[1] <= '3') &&
1290                             isodigit(bp[2]) &&
1291                             isodigit(bp[3])) {
1292                                 int byte = (*++bp -'0');
1293                                 bp++;
1294                                 byte = (byte << 3) | (*bp++ - '0');
1295                                 byte = (byte << 3) | (*bp++ - '0');
1296                                 *dest++ = byte;
1297                                 len++;
1298                         } else {
1299                                 *dest++ = *bp++;
1300                                 len++;
1301                         }
1302                 }
1303         }
1304
1305         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1306                 return -1;
1307         while (*bp == ' ') bp++;
1308         *bpp = bp;
1309         *dest = '\0';
1310         return len;
1311 }
1312 EXPORT_SYMBOL_GPL(qword_get);
1313
1314
1315 /*
1316  * support /proc/net/rpc/$CACHENAME/content
1317  * as a seqfile.
1318  * We call ->cache_show passing NULL for the item to
1319  * get a header, then pass each real item in the cache
1320  */
1321
1322 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1323 {
1324         loff_t n = *pos;
1325         unsigned int hash, entry;
1326         struct cache_head *ch;
1327         struct cache_detail *cd = m->private;
1328
1329         if (!n--)
1330                 return SEQ_START_TOKEN;
1331         hash = n >> 32;
1332         entry = n & ((1LL<<32) - 1);
1333
1334         hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1335                 if (!entry--)
1336                         return ch;
1337         n &= ~((1LL<<32) - 1);
1338         do {
1339                 hash++;
1340                 n += 1LL<<32;
1341         } while(hash < cd->hash_size &&
1342                 hlist_empty(&cd->hash_table[hash]));
1343         if (hash >= cd->hash_size)
1344                 return NULL;
1345         *pos = n+1;
1346         return hlist_entry_safe(rcu_dereference_raw(
1347                                 hlist_first_rcu(&cd->hash_table[hash])),
1348                                 struct cache_head, cache_list);
1349 }
1350
1351 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1352 {
1353         struct cache_head *ch = p;
1354         int hash = (*pos >> 32);
1355         struct cache_detail *cd = m->private;
1356
1357         if (p == SEQ_START_TOKEN)
1358                 hash = 0;
1359         else if (ch->cache_list.next == NULL) {
1360                 hash++;
1361                 *pos += 1LL<<32;
1362         } else {
1363                 ++*pos;
1364                 return hlist_entry_safe(rcu_dereference_raw(
1365                                         hlist_next_rcu(&ch->cache_list)),
1366                                         struct cache_head, cache_list);
1367         }
1368         *pos &= ~((1LL<<32) - 1);
1369         while (hash < cd->hash_size &&
1370                hlist_empty(&cd->hash_table[hash])) {
1371                 hash++;
1372                 *pos += 1LL<<32;
1373         }
1374         if (hash >= cd->hash_size)
1375                 return NULL;
1376         ++*pos;
1377         return hlist_entry_safe(rcu_dereference_raw(
1378                                 hlist_first_rcu(&cd->hash_table[hash])),
1379                                 struct cache_head, cache_list);
1380 }
1381
1382 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1383         __acquires(RCU)
1384 {
1385         rcu_read_lock();
1386         return __cache_seq_start(m, pos);
1387 }
1388 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1389
1390 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1391 {
1392         return cache_seq_next(file, p, pos);
1393 }
1394 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1395
1396 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1397         __releases(RCU)
1398 {
1399         rcu_read_unlock();
1400 }
1401 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1402
1403 static int c_show(struct seq_file *m, void *p)
1404 {
1405         struct cache_head *cp = p;
1406         struct cache_detail *cd = m->private;
1407
1408         if (p == SEQ_START_TOKEN)
1409                 return cd->cache_show(m, cd, NULL);
1410
1411         ifdebug(CACHE)
1412                 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1413                            convert_to_wallclock(cp->expiry_time),
1414                            kref_read(&cp->ref), cp->flags);
1415         cache_get(cp);
1416         if (cache_check(cd, cp, NULL))
1417                 /* cache_check does a cache_put on failure */
1418                 seq_printf(m, "# ");
1419         else {
1420                 if (cache_is_expired(cd, cp))
1421                         seq_printf(m, "# ");
1422                 cache_put(cp, cd);
1423         }
1424
1425         return cd->cache_show(m, cd, cp);
1426 }
1427
1428 static const struct seq_operations cache_content_op = {
1429         .start  = cache_seq_start_rcu,
1430         .next   = cache_seq_next_rcu,
1431         .stop   = cache_seq_stop_rcu,
1432         .show   = c_show,
1433 };
1434
1435 static int content_open(struct inode *inode, struct file *file,
1436                         struct cache_detail *cd)
1437 {
1438         struct seq_file *seq;
1439         int err;
1440
1441         if (!cd || !try_module_get(cd->owner))
1442                 return -EACCES;
1443
1444         err = seq_open(file, &cache_content_op);
1445         if (err) {
1446                 module_put(cd->owner);
1447                 return err;
1448         }
1449
1450         seq = file->private_data;
1451         seq->private = cd;
1452         return 0;
1453 }
1454
1455 static int content_release(struct inode *inode, struct file *file,
1456                 struct cache_detail *cd)
1457 {
1458         int ret = seq_release(inode, file);
1459         module_put(cd->owner);
1460         return ret;
1461 }
1462
1463 static int open_flush(struct inode *inode, struct file *file,
1464                         struct cache_detail *cd)
1465 {
1466         if (!cd || !try_module_get(cd->owner))
1467                 return -EACCES;
1468         return nonseekable_open(inode, file);
1469 }
1470
1471 static int release_flush(struct inode *inode, struct file *file,
1472                         struct cache_detail *cd)
1473 {
1474         module_put(cd->owner);
1475         return 0;
1476 }
1477
1478 static ssize_t read_flush(struct file *file, char __user *buf,
1479                           size_t count, loff_t *ppos,
1480                           struct cache_detail *cd)
1481 {
1482         char tbuf[22];
1483         size_t len;
1484
1485         len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1486                         convert_to_wallclock(cd->flush_time));
1487         return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1488 }
1489
1490 static ssize_t write_flush(struct file *file, const char __user *buf,
1491                            size_t count, loff_t *ppos,
1492                            struct cache_detail *cd)
1493 {
1494         char tbuf[20];
1495         char *ep;
1496         time64_t now;
1497
1498         if (*ppos || count > sizeof(tbuf)-1)
1499                 return -EINVAL;
1500         if (copy_from_user(tbuf, buf, count))
1501                 return -EFAULT;
1502         tbuf[count] = 0;
1503         simple_strtoul(tbuf, &ep, 0);
1504         if (*ep && *ep != '\n')
1505                 return -EINVAL;
1506         /* Note that while we check that 'buf' holds a valid number,
1507          * we always ignore the value and just flush everything.
1508          * Making use of the number leads to races.
1509          */
1510
1511         now = seconds_since_boot();
1512         /* Always flush everything, so behave like cache_purge()
1513          * Do this by advancing flush_time to the current time,
1514          * or by one second if it has already reached the current time.
1515          * Newly added cache entries will always have ->last_refresh greater
1516          * that ->flush_time, so they don't get flushed prematurely.
1517          */
1518
1519         if (cd->flush_time >= now)
1520                 now = cd->flush_time + 1;
1521
1522         cd->flush_time = now;
1523         cd->nextcheck = now;
1524         cache_flush();
1525
1526         if (cd->flush)
1527                 cd->flush();
1528
1529         *ppos += count;
1530         return count;
1531 }
1532
1533 static ssize_t cache_read_procfs(struct file *filp, 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_read(filp, buf, count, ppos, cd);
1539 }
1540
1541 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1542                                   size_t count, loff_t *ppos)
1543 {
1544         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1545
1546         return cache_write(filp, buf, count, ppos, cd);
1547 }
1548
1549 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1550 {
1551         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1552
1553         return cache_poll(filp, wait, cd);
1554 }
1555
1556 static long cache_ioctl_procfs(struct file *filp,
1557                                unsigned int cmd, unsigned long arg)
1558 {
1559         struct inode *inode = file_inode(filp);
1560         struct cache_detail *cd = PDE_DATA(inode);
1561
1562         return cache_ioctl(inode, filp, cmd, arg, cd);
1563 }
1564
1565 static int cache_open_procfs(struct inode *inode, struct file *filp)
1566 {
1567         struct cache_detail *cd = PDE_DATA(inode);
1568
1569         return cache_open(inode, filp, cd);
1570 }
1571
1572 static int cache_release_procfs(struct inode *inode, struct file *filp)
1573 {
1574         struct cache_detail *cd = PDE_DATA(inode);
1575
1576         return cache_release(inode, filp, cd);
1577 }
1578
1579 static const struct proc_ops cache_channel_proc_ops = {
1580         .proc_lseek     = no_llseek,
1581         .proc_read      = cache_read_procfs,
1582         .proc_write     = cache_write_procfs,
1583         .proc_poll      = cache_poll_procfs,
1584         .proc_ioctl     = cache_ioctl_procfs, /* for FIONREAD */
1585         .proc_open      = cache_open_procfs,
1586         .proc_release   = cache_release_procfs,
1587 };
1588
1589 static int content_open_procfs(struct inode *inode, struct file *filp)
1590 {
1591         struct cache_detail *cd = PDE_DATA(inode);
1592
1593         return content_open(inode, filp, cd);
1594 }
1595
1596 static int content_release_procfs(struct inode *inode, struct file *filp)
1597 {
1598         struct cache_detail *cd = PDE_DATA(inode);
1599
1600         return content_release(inode, filp, cd);
1601 }
1602
1603 static const struct proc_ops content_proc_ops = {
1604         .proc_open      = content_open_procfs,
1605         .proc_read      = seq_read,
1606         .proc_lseek     = seq_lseek,
1607         .proc_release   = content_release_procfs,
1608 };
1609
1610 static int open_flush_procfs(struct inode *inode, struct file *filp)
1611 {
1612         struct cache_detail *cd = PDE_DATA(inode);
1613
1614         return open_flush(inode, filp, cd);
1615 }
1616
1617 static int release_flush_procfs(struct inode *inode, struct file *filp)
1618 {
1619         struct cache_detail *cd = PDE_DATA(inode);
1620
1621         return release_flush(inode, filp, cd);
1622 }
1623
1624 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1625                             size_t count, loff_t *ppos)
1626 {
1627         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1628
1629         return read_flush(filp, buf, count, ppos, cd);
1630 }
1631
1632 static ssize_t write_flush_procfs(struct file *filp,
1633                                   const char __user *buf,
1634                                   size_t count, loff_t *ppos)
1635 {
1636         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1637
1638         return write_flush(filp, buf, count, ppos, cd);
1639 }
1640
1641 static const struct proc_ops cache_flush_proc_ops = {
1642         .proc_open      = open_flush_procfs,
1643         .proc_read      = read_flush_procfs,
1644         .proc_write     = write_flush_procfs,
1645         .proc_release   = release_flush_procfs,
1646         .proc_lseek     = no_llseek,
1647 };
1648
1649 static void remove_cache_proc_entries(struct cache_detail *cd)
1650 {
1651         if (cd->procfs) {
1652                 proc_remove(cd->procfs);
1653                 cd->procfs = NULL;
1654         }
1655 }
1656
1657 #ifdef CONFIG_PROC_FS
1658 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1659 {
1660         struct proc_dir_entry *p;
1661         struct sunrpc_net *sn;
1662
1663         sn = net_generic(net, sunrpc_net_id);
1664         cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1665         if (cd->procfs == NULL)
1666                 goto out_nomem;
1667
1668         p = proc_create_data("flush", S_IFREG | 0600,
1669                              cd->procfs, &cache_flush_proc_ops, cd);
1670         if (p == NULL)
1671                 goto out_nomem;
1672
1673         if (cd->cache_request || cd->cache_parse) {
1674                 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1675                                      &cache_channel_proc_ops, cd);
1676                 if (p == NULL)
1677                         goto out_nomem;
1678         }
1679         if (cd->cache_show) {
1680                 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1681                                      &content_proc_ops, cd);
1682                 if (p == NULL)
1683                         goto out_nomem;
1684         }
1685         return 0;
1686 out_nomem:
1687         remove_cache_proc_entries(cd);
1688         return -ENOMEM;
1689 }
1690 #else /* CONFIG_PROC_FS */
1691 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1692 {
1693         return 0;
1694 }
1695 #endif
1696
1697 void __init cache_initialize(void)
1698 {
1699         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1700 }
1701
1702 int cache_register_net(struct cache_detail *cd, struct net *net)
1703 {
1704         int ret;
1705
1706         sunrpc_init_cache_detail(cd);
1707         ret = create_cache_proc_entries(cd, net);
1708         if (ret)
1709                 sunrpc_destroy_cache_detail(cd);
1710         return ret;
1711 }
1712 EXPORT_SYMBOL_GPL(cache_register_net);
1713
1714 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1715 {
1716         remove_cache_proc_entries(cd);
1717         sunrpc_destroy_cache_detail(cd);
1718 }
1719 EXPORT_SYMBOL_GPL(cache_unregister_net);
1720
1721 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1722 {
1723         struct cache_detail *cd;
1724         int i;
1725
1726         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1727         if (cd == NULL)
1728                 return ERR_PTR(-ENOMEM);
1729
1730         cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1731                                  GFP_KERNEL);
1732         if (cd->hash_table == NULL) {
1733                 kfree(cd);
1734                 return ERR_PTR(-ENOMEM);
1735         }
1736
1737         for (i = 0; i < cd->hash_size; i++)
1738                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1739         cd->net = net;
1740         return cd;
1741 }
1742 EXPORT_SYMBOL_GPL(cache_create_net);
1743
1744 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1745 {
1746         kfree(cd->hash_table);
1747         kfree(cd);
1748 }
1749 EXPORT_SYMBOL_GPL(cache_destroy_net);
1750
1751 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1752                                  size_t count, loff_t *ppos)
1753 {
1754         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1755
1756         return cache_read(filp, buf, count, ppos, cd);
1757 }
1758
1759 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1760                                   size_t count, loff_t *ppos)
1761 {
1762         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1763
1764         return cache_write(filp, buf, count, ppos, cd);
1765 }
1766
1767 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1768 {
1769         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1770
1771         return cache_poll(filp, wait, cd);
1772 }
1773
1774 static long cache_ioctl_pipefs(struct file *filp,
1775                               unsigned int cmd, unsigned long arg)
1776 {
1777         struct inode *inode = file_inode(filp);
1778         struct cache_detail *cd = RPC_I(inode)->private;
1779
1780         return cache_ioctl(inode, filp, cmd, arg, cd);
1781 }
1782
1783 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1784 {
1785         struct cache_detail *cd = RPC_I(inode)->private;
1786
1787         return cache_open(inode, filp, cd);
1788 }
1789
1790 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1791 {
1792         struct cache_detail *cd = RPC_I(inode)->private;
1793
1794         return cache_release(inode, filp, cd);
1795 }
1796
1797 const struct file_operations cache_file_operations_pipefs = {
1798         .owner          = THIS_MODULE,
1799         .llseek         = no_llseek,
1800         .read           = cache_read_pipefs,
1801         .write          = cache_write_pipefs,
1802         .poll           = cache_poll_pipefs,
1803         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1804         .open           = cache_open_pipefs,
1805         .release        = cache_release_pipefs,
1806 };
1807
1808 static int content_open_pipefs(struct inode *inode, struct file *filp)
1809 {
1810         struct cache_detail *cd = RPC_I(inode)->private;
1811
1812         return content_open(inode, filp, cd);
1813 }
1814
1815 static int content_release_pipefs(struct inode *inode, struct file *filp)
1816 {
1817         struct cache_detail *cd = RPC_I(inode)->private;
1818
1819         return content_release(inode, filp, cd);
1820 }
1821
1822 const struct file_operations content_file_operations_pipefs = {
1823         .open           = content_open_pipefs,
1824         .read           = seq_read,
1825         .llseek         = seq_lseek,
1826         .release        = content_release_pipefs,
1827 };
1828
1829 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1830 {
1831         struct cache_detail *cd = RPC_I(inode)->private;
1832
1833         return open_flush(inode, filp, cd);
1834 }
1835
1836 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1837 {
1838         struct cache_detail *cd = RPC_I(inode)->private;
1839
1840         return release_flush(inode, filp, cd);
1841 }
1842
1843 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1844                             size_t count, loff_t *ppos)
1845 {
1846         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1847
1848         return read_flush(filp, buf, count, ppos, cd);
1849 }
1850
1851 static ssize_t write_flush_pipefs(struct file *filp,
1852                                   const char __user *buf,
1853                                   size_t count, loff_t *ppos)
1854 {
1855         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1856
1857         return write_flush(filp, buf, count, ppos, cd);
1858 }
1859
1860 const struct file_operations cache_flush_operations_pipefs = {
1861         .open           = open_flush_pipefs,
1862         .read           = read_flush_pipefs,
1863         .write          = write_flush_pipefs,
1864         .release        = release_flush_pipefs,
1865         .llseek         = no_llseek,
1866 };
1867
1868 int sunrpc_cache_register_pipefs(struct dentry *parent,
1869                                  const char *name, umode_t umode,
1870                                  struct cache_detail *cd)
1871 {
1872         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1873         if (IS_ERR(dir))
1874                 return PTR_ERR(dir);
1875         cd->pipefs = dir;
1876         return 0;
1877 }
1878 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1879
1880 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1881 {
1882         if (cd->pipefs) {
1883                 rpc_remove_cache_dir(cd->pipefs);
1884                 cd->pipefs = NULL;
1885         }
1886 }
1887 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1888
1889 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1890 {
1891         spin_lock(&cd->hash_lock);
1892         if (!hlist_unhashed(&h->cache_list)){
1893                 sunrpc_begin_cache_remove_entry(h, cd);
1894                 spin_unlock(&cd->hash_lock);
1895                 sunrpc_end_cache_remove_entry(h, cd);
1896         } else
1897                 spin_unlock(&cd->hash_lock);
1898 }
1899 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);