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 <trace/events/sunrpc.h>
  36 #include "netns.h"
  37
  38 #define  RPCDBG_FACILITY RPCDBG_CACHE
  39
  40 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  41 static void cache_revisit_request(struct cache_head *item);
  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                         continue;
  70                 if (test_bit(CACHE_VALID, &tmp->flags) &&
  71                     cache_is_expired(detail, tmp))
  72                         continue;
  73                 tmp = cache_get_rcu(tmp);
  74                 rcu_read_unlock();
  75                 return tmp;
  76         }
  77         rcu_read_unlock();
  78         return NULL;
  79 }
  80
  81 static void sunrpc_begin_cache_remove_entry(struct cache_head *ch,
  82                                             struct cache_detail *cd)
  83 {
  84         /* Must be called under cd->hash_lock */
  85         hlist_del_init_rcu(&ch->cache_list);
  86         set_bit(CACHE_CLEANED, &ch->flags);
  87         cd->entries --;
  88 }
  89
  90 static void sunrpc_end_cache_remove_entry(struct cache_head *ch,
  91                                           struct cache_detail *cd)
  92 {
  93         cache_fresh_unlocked(ch, cd);
  94         cache_put(ch, cd);
  95 }
  96
  97 static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
  98                                                  struct cache_head *key,
  99                                                  int hash)
 100 {
 101         struct cache_head *new, *tmp, *freeme = NULL;
 102         struct hlist_head *head = &detail->hash_table[hash];
 103
 104         new = detail->alloc();
 105         if (!new)
 106                 return NULL;
 107         /* must fully initialise 'new', else
 108          * we might get lose if we need to
 109          * cache_put it soon.
 110          */
 111         cache_init(new, detail);
 112         detail->init(new, key);
 113
 114         spin_lock(&detail->hash_lock);
 115
 116         /* check if entry appeared while we slept */
 117         hlist_for_each_entry_rcu(tmp, head, cache_list,
 118                                  lockdep_is_held(&detail->hash_lock)) {
 119                 if (!detail->match(tmp, key))
 120                         continue;
 121                 if (test_bit(CACHE_VALID, &tmp->flags) &&
 122                     cache_is_expired(detail, tmp)) {
 123                         sunrpc_begin_cache_remove_entry(tmp, detail);
 124                         trace_cache_entry_expired(detail, tmp);
 125                         freeme = tmp;
 126                         break;
 127                 }
 128                 cache_get(tmp);
 129                 spin_unlock(&detail->hash_lock);
 130                 cache_put(new, detail);
 131                 return tmp;
 132         }
 133
 134         hlist_add_head_rcu(&new->cache_list, head);
 135         detail->entries++;
 136         cache_get(new);
 137         spin_unlock(&detail->hash_lock);
 138
 139         if (freeme)
 140                 sunrpc_end_cache_remove_entry(freeme, detail);
 141         return new;
 142 }
 143
 144 struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
 145                                            struct cache_head *key, int hash)
 146 {
 147         struct cache_head *ret;
 148
 149         ret = sunrpc_cache_find_rcu(detail, key, hash);
 150         if (ret)
 151                 return ret;
 152         /* Didn't find anything, insert an empty entry */
 153         return sunrpc_cache_add_entry(detail, key, hash);
 154 }
 155 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
 156
 157 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 158
 159 static void cache_fresh_locked(struct cache_head *head, time64_t expiry,
 160                                struct cache_detail *detail)
 161 {
 162         time64_t now = seconds_since_boot();
 163         if (now <= detail->flush_time)
 164                 /* ensure it isn't immediately treated as expired */
 165                 now = detail->flush_time + 1;
 166         head->expiry_time = expiry;
 167         head->last_refresh = now;
 168         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 169         set_bit(CACHE_VALID, &head->flags);
 170 }
 171
 172 static void cache_fresh_unlocked(struct cache_head *head,
 173                                  struct cache_detail *detail)
 174 {
 175         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 176                 cache_revisit_request(head);
 177                 cache_dequeue(detail, head);
 178         }
 179 }
 180
 181 static void cache_make_negative(struct cache_detail *detail,
 182                                 struct cache_head *h)
 183 {
 184         set_bit(CACHE_NEGATIVE, &h->flags);
 185         trace_cache_entry_make_negative(detail, h);
 186 }
 187
 188 static void cache_entry_update(struct cache_detail *detail,
 189                                struct cache_head *h,
 190                                struct cache_head *new)
 191 {
 192         if (!test_bit(CACHE_NEGATIVE, &new->flags)) {
 193                 detail->update(h, new);
 194                 trace_cache_entry_update(detail, h);
 195         } else {
 196                 cache_make_negative(detail, h);
 197         }
 198 }
 199
 200 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 201                                        struct cache_head *new, struct cache_head *old, int hash)
 202 {
 203         /* The 'old' entry is to be replaced by 'new'.
 204          * If 'old' is not VALID, we update it directly,
 205          * otherwise we need to replace it
 206          */
 207         struct cache_head *tmp;
 208
 209         if (!test_bit(CACHE_VALID, &old->flags)) {
 210                 spin_lock(&detail->hash_lock);
 211                 if (!test_bit(CACHE_VALID, &old->flags)) {
 212                         cache_entry_update(detail, old, new);
 213                         cache_fresh_locked(old, new->expiry_time, detail);
 214                         spin_unlock(&detail->hash_lock);
 215                         cache_fresh_unlocked(old, detail);
 216                         return old;
 217                 }
 218                 spin_unlock(&detail->hash_lock);
 219         }
 220         /* We need to insert a new entry */
 221         tmp = detail->alloc();
 222         if (!tmp) {
 223                 cache_put(old, detail);
 224                 return NULL;
 225         }
 226         cache_init(tmp, detail);
 227         detail->init(tmp, old);
 228
 229         spin_lock(&detail->hash_lock);
 230         cache_entry_update(detail, tmp, new);
 231         hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
 232         detail->entries++;
 233         cache_get(tmp);
 234         cache_fresh_locked(tmp, new->expiry_time, detail);
 235         cache_fresh_locked(old, 0, detail);
 236         spin_unlock(&detail->hash_lock);
 237         cache_fresh_unlocked(tmp, detail);
 238         cache_fresh_unlocked(old, detail);
 239         cache_put(old, detail);
 240         return tmp;
 241 }
 242 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 243
 244 static inline int cache_is_valid(struct cache_head *h)
 245 {
 246         if (!test_bit(CACHE_VALID, &h->flags))
 247                 return -EAGAIN;
 248         else {
 249                 /* entry is valid */
 250                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 251                         return -ENOENT;
 252                 else {
 253                         /*
 254                          * In combination with write barrier in
 255                          * sunrpc_cache_update, ensures that anyone
 256                          * using the cache entry after this sees the
 257                          * updated contents:
 258                          */
 259                         smp_rmb();
 260                         return 0;
 261                 }
 262         }
 263 }
 264
 265 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 266 {
 267         int rv;
 268
 269         spin_lock(&detail->hash_lock);
 270         rv = cache_is_valid(h);
 271         if (rv == -EAGAIN) {
 272                 cache_make_negative(detail, h);
 273                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
 274                                    detail);
 275                 rv = -ENOENT;
 276         }
 277         spin_unlock(&detail->hash_lock);
 278         cache_fresh_unlocked(h, detail);
 279         return rv;
 280 }
 281
 282 /*
 283  * This is the generic cache management routine for all
 284  * the authentication caches.
 285  * It checks the currency of a cache item and will (later)
 286  * initiate an upcall to fill it if needed.
 287  *
 288  *
 289  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 290  * -EAGAIN if upcall is pending and request has been queued
 291  * -ETIMEDOUT if upcall failed or request could not be queue or
 292  *           upcall completed but item is still invalid (implying that
 293  *           the cache item has been replaced with a newer one).
 294  * -ENOENT if cache entry was negative
 295  */
 296 int cache_check(struct cache_detail *detail,
 297                     struct cache_head *h, struct cache_req *rqstp)
 298 {
 299         int rv;
 300         time64_t refresh_age, age;
 301
 302         /* First decide return status as best we can */
 303         rv = cache_is_valid(h);
 304
 305         /* now see if we want to start an upcall */
 306         refresh_age = (h->expiry_time - h->last_refresh);
 307         age = seconds_since_boot() - h->last_refresh;
 308
 309         if (rqstp == NULL) {
 310                 if (rv == -EAGAIN)
 311                         rv = -ENOENT;
 312         } else if (rv == -EAGAIN ||
 313                    (h->expiry_time != 0 && age > refresh_age/2)) {
 314                 dprintk("RPC:       Want update, refage=%lld, age=%lld\n",
 315                                 refresh_age, age);
 316                 switch (detail->cache_upcall(detail, h)) {
 317                 case -EINVAL:
 318                         rv = try_to_negate_entry(detail, h);
 319                         break;
 320                 case -EAGAIN:
 321                         cache_fresh_unlocked(h, detail);
 322                         break;
 323                 }
 324         }
 325
 326         if (rv == -EAGAIN) {
 327                 if (!cache_defer_req(rqstp, h)) {
 328                         /*
 329                          * Request was not deferred; handle it as best
 330                          * we can ourselves:
 331                          */
 332                         rv = cache_is_valid(h);
 333                         if (rv == -EAGAIN)
 334                                 rv = -ETIMEDOUT;
 335                 }
 336         }
 337         if (rv)
 338                 cache_put(h, detail);
 339         return rv;
 340 }
 341 EXPORT_SYMBOL_GPL(cache_check);
 342
 343 /*
 344  * caches need to be periodically cleaned.
 345  * For this we maintain a list of cache_detail and
 346  * a current pointer into that list and into the table
 347  * for that entry.
 348  *
 349  * Each time cache_clean is called it finds the next non-empty entry
 350  * in the current table and walks the list in that entry
 351  * looking for entries that can be removed.
 352  *
 353  * An entry gets removed if:
 354  * - The expiry is before current time
 355  * - The last_refresh time is before the flush_time for that cache
 356  *
 357  * later we might drop old entries with non-NEVER expiry if that table
 358  * is getting 'full' for some definition of 'full'
 359  *
 360  * The question of "how often to scan a table" is an interesting one
 361  * and is answered in part by the use of the "nextcheck" field in the
 362  * cache_detail.
 363  * When a scan of a table begins, the nextcheck field is set to a time
 364  * that is well into the future.
 365  * While scanning, if an expiry time is found that is earlier than the
 366  * current nextcheck time, nextcheck is set to that expiry time.
 367  * If the flush_time is ever set to a time earlier than the nextcheck
 368  * time, the nextcheck time is then set to that flush_time.
 369  *
 370  * A table is then only scanned if the current time is at least
 371  * the nextcheck time.
 372  *
 373  */
 374
 375 static LIST_HEAD(cache_list);
 376 static DEFINE_SPINLOCK(cache_list_lock);
 377 static struct cache_detail *current_detail;
 378 static int current_index;
 379
 380 static void do_cache_clean(struct work_struct *work);
 381 static struct delayed_work cache_cleaner;
 382
 383 void sunrpc_init_cache_detail(struct cache_detail *cd)
 384 {
 385         spin_lock_init(&cd->hash_lock);
 386         INIT_LIST_HEAD(&cd->queue);
 387         spin_lock(&cache_list_lock);
 388         cd->nextcheck = 0;
 389         cd->entries = 0;
 390         atomic_set(&cd->writers, 0);
 391         cd->last_close = 0;
 392         cd->last_warn = -1;
 393         list_add(&cd->others, &cache_list);
 394         spin_unlock(&cache_list_lock);
 395
 396         /* start the cleaning process */
 397         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
 398 }
 399 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 400
 401 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 402 {
 403         cache_purge(cd);
 404         spin_lock(&cache_list_lock);
 405         spin_lock(&cd->hash_lock);
 406         if (current_detail == cd)
 407                 current_detail = NULL;
 408         list_del_init(&cd->others);
 409         spin_unlock(&cd->hash_lock);
 410         spin_unlock(&cache_list_lock);
 411         if (list_empty(&cache_list)) {
 412                 /* module must be being unloaded so its safe to kill the worker */
 413                 cancel_delayed_work_sync(&cache_cleaner);
 414         }
 415 }
 416 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 417
 418 /* clean cache tries to find something to clean
 419  * and cleans it.
 420  * It returns 1 if it cleaned something,
 421  *            0 if it didn't find anything this time
 422  *           -1 if it fell off the end of the list.
 423  */
 424 static int cache_clean(void)
 425 {
 426         int rv = 0;
 427         struct list_head *next;
 428
 429         spin_lock(&cache_list_lock);
 430
 431         /* find a suitable table if we don't already have one */
 432         while (current_detail == NULL ||
 433             current_index >= current_detail->hash_size) {
 434                 if (current_detail)
 435                         next = current_detail->others.next;
 436                 else
 437                         next = cache_list.next;
 438                 if (next == &cache_list) {
 439                         current_detail = NULL;
 440                         spin_unlock(&cache_list_lock);
 441                         return -1;
 442                 }
 443                 current_detail = list_entry(next, struct cache_detail, others);
 444                 if (current_detail->nextcheck > seconds_since_boot())
 445                         current_index = current_detail->hash_size;
 446                 else {
 447                         current_index = 0;
 448                         current_detail->nextcheck = seconds_since_boot()+30*60;
 449                 }
 450         }
 451
 452         /* find a non-empty bucket in the table */
 453         while (current_detail &&
 454                current_index < current_detail->hash_size &&
 455                hlist_empty(&current_detail->hash_table[current_index]))
 456                 current_index++;
 457
 458         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 459
 460         if (current_detail && current_index < current_detail->hash_size) {
 461                 struct cache_head *ch = NULL;
 462                 struct cache_detail *d;
 463                 struct hlist_head *head;
 464                 struct hlist_node *tmp;
 465
 466                 spin_lock(&current_detail->hash_lock);
 467
 468                 /* Ok, now to clean this strand */
 469
 470                 head = &current_detail->hash_table[current_index];
 471                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 472                         if (current_detail->nextcheck > ch->expiry_time)
 473                                 current_detail->nextcheck = ch->expiry_time+1;
 474                         if (!cache_is_expired(current_detail, ch))
 475                                 continue;
 476
 477                         sunrpc_begin_cache_remove_entry(ch, current_detail);
 478                         trace_cache_entry_expired(current_detail, ch);
 479                         rv = 1;
 480                         break;
 481                 }
 482
 483                 spin_unlock(&current_detail->hash_lock);
 484                 d = current_detail;
 485                 if (!ch)
 486                         current_index ++;
 487                 spin_unlock(&cache_list_lock);
 488                 if (ch)
 489                         sunrpc_end_cache_remove_entry(ch, d);
 490         } else
 491                 spin_unlock(&cache_list_lock);
 492
 493         return rv;
 494 }
 495
 496 /*
 497  * We want to regularly clean the cache, so we need to schedule some work ...
 498  */
 499 static void do_cache_clean(struct work_struct *work)
 500 {
 501         int delay = 5;
 502         if (cache_clean() == -1)
 503                 delay = round_jiffies_relative(30*HZ);
 504
 505         if (list_empty(&cache_list))
 506                 delay = 0;
 507
 508         if (delay)
 509                 queue_delayed_work(system_power_efficient_wq,
 510                                    &cache_cleaner, delay);
 511 }
 512
 513
 514 /*
 515  * Clean all caches promptly.  This just calls cache_clean
 516  * repeatedly until we are sure that every cache has had a chance to
 517  * be fully cleaned
 518  */
 519 void cache_flush(void)
 520 {
 521         while (cache_clean() != -1)
 522                 cond_resched();
 523         while (cache_clean() != -1)
 524                 cond_resched();
 525 }
 526 EXPORT_SYMBOL_GPL(cache_flush);
 527
 528 void cache_purge(struct cache_detail *detail)
 529 {
 530         struct cache_head *ch = NULL;
 531         struct hlist_head *head = NULL;
 532         int i = 0;
 533
 534         spin_lock(&detail->hash_lock);
 535         if (!detail->entries) {
 536                 spin_unlock(&detail->hash_lock);
 537                 return;
 538         }
 539
 540         dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
 541         for (i = 0; i < detail->hash_size; i++) {
 542                 head = &detail->hash_table[i];
 543                 while (!hlist_empty(head)) {
 544                         ch = hlist_entry(head->first, struct cache_head,
 545                                          cache_list);
 546                         sunrpc_begin_cache_remove_entry(ch, detail);
 547                         spin_unlock(&detail->hash_lock);
 548                         sunrpc_end_cache_remove_entry(ch, detail);
 549                         spin_lock(&detail->hash_lock);
 550                 }
 551         }
 552         spin_unlock(&detail->hash_lock);
 553 }
 554 EXPORT_SYMBOL_GPL(cache_purge);
 555
 556
 557 /*
 558  * Deferral and Revisiting of Requests.
 559  *
 560  * If a cache lookup finds a pending entry, we
 561  * need to defer the request and revisit it later.
 562  * All deferred requests are stored in a hash table,
 563  * indexed by "struct cache_head *".
 564  * As it may be wasteful to store a whole request
 565  * structure, we allow the request to provide a
 566  * deferred form, which must contain a
 567  * 'struct cache_deferred_req'
 568  * This cache_deferred_req contains a method to allow
 569  * it to be revisited when cache info is available
 570  */
 571
 572 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 573 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 574
 575 #define DFR_MAX 300     /* ??? */
 576
 577 static DEFINE_SPINLOCK(cache_defer_lock);
 578 static LIST_HEAD(cache_defer_list);
 579 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 580 static int cache_defer_cnt;
 581
 582 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 583 {
 584         hlist_del_init(&dreq->hash);
 585         if (!list_empty(&dreq->recent)) {
 586                 list_del_init(&dreq->recent);
 587                 cache_defer_cnt--;
 588         }
 589 }
 590
 591 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 592 {
 593         int hash = DFR_HASH(item);
 594
 595         INIT_LIST_HEAD(&dreq->recent);
 596         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 597 }
 598
 599 static void setup_deferral(struct cache_deferred_req *dreq,
 600                            struct cache_head *item,
 601                            int count_me)
 602 {
 603
 604         dreq->item = item;
 605
 606         spin_lock(&cache_defer_lock);
 607
 608         __hash_deferred_req(dreq, item);
 609
 610         if (count_me) {
 611                 cache_defer_cnt++;
 612                 list_add(&dreq->recent, &cache_defer_list);
 613         }
 614
 615         spin_unlock(&cache_defer_lock);
 616
 617 }
 618
 619 struct thread_deferred_req {
 620         struct cache_deferred_req handle;
 621         struct completion completion;
 622 };
 623
 624 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 625 {
 626         struct thread_deferred_req *dr =
 627                 container_of(dreq, struct thread_deferred_req, handle);
 628         complete(&dr->completion);
 629 }
 630
 631 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 632 {
 633         struct thread_deferred_req sleeper;
 634         struct cache_deferred_req *dreq = &sleeper.handle;
 635
 636         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 637         dreq->revisit = cache_restart_thread;
 638
 639         setup_deferral(dreq, item, 0);
 640
 641         if (!test_bit(CACHE_PENDING, &item->flags) ||
 642             wait_for_completion_interruptible_timeout(
 643                     &sleeper.completion, req->thread_wait) <= 0) {
 644                 /* The completion wasn't completed, so we need
 645                  * to clean up
 646                  */
 647                 spin_lock(&cache_defer_lock);
 648                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 649                         __unhash_deferred_req(&sleeper.handle);
 650                         spin_unlock(&cache_defer_lock);
 651                 } else {
 652                         /* cache_revisit_request already removed
 653                          * this from the hash table, but hasn't
 654                          * called ->revisit yet.  It will very soon
 655                          * and we need to wait for it.
 656                          */
 657                         spin_unlock(&cache_defer_lock);
 658                         wait_for_completion(&sleeper.completion);
 659                 }
 660         }
 661 }
 662
 663 static void cache_limit_defers(void)
 664 {
 665         /* Make sure we haven't exceed the limit of allowed deferred
 666          * requests.
 667          */
 668         struct cache_deferred_req *discard = NULL;
 669
 670         if (cache_defer_cnt <= DFR_MAX)
 671                 return;
 672
 673         spin_lock(&cache_defer_lock);
 674
 675         /* Consider removing either the first or the last */
 676         if (cache_defer_cnt > DFR_MAX) {
 677                 if (prandom_u32() & 1)
 678                         discard = list_entry(cache_defer_list.next,
 679                                              struct cache_deferred_req, recent);
 680                 else
 681                         discard = list_entry(cache_defer_list.prev,
 682                                              struct cache_deferred_req, recent);
 683                 __unhash_deferred_req(discard);
 684         }
 685         spin_unlock(&cache_defer_lock);
 686         if (discard)
 687                 discard->revisit(discard, 1);
 688 }
 689
 690 /* Return true if and only if a deferred request is queued. */
 691 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 692 {
 693         struct cache_deferred_req *dreq;
 694
 695         if (req->thread_wait) {
 696                 cache_wait_req(req, item);
 697                 if (!test_bit(CACHE_PENDING, &item->flags))
 698                         return false;
 699         }
 700         dreq = req->defer(req);
 701         if (dreq == NULL)
 702                 return false;
 703         setup_deferral(dreq, item, 1);
 704         if (!test_bit(CACHE_PENDING, &item->flags))
 705                 /* Bit could have been cleared before we managed to
 706                  * set up the deferral, so need to revisit just in case
 707                  */
 708                 cache_revisit_request(item);
 709
 710         cache_limit_defers();
 711         return true;
 712 }
 713
 714 static void cache_revisit_request(struct cache_head *item)
 715 {
 716         struct cache_deferred_req *dreq;
 717         struct list_head pending;
 718         struct hlist_node *tmp;
 719         int hash = DFR_HASH(item);
 720
 721         INIT_LIST_HEAD(&pending);
 722         spin_lock(&cache_defer_lock);
 723
 724         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 725                 if (dreq->item == item) {
 726                         __unhash_deferred_req(dreq);
 727                         list_add(&dreq->recent, &pending);
 728                 }
 729
 730         spin_unlock(&cache_defer_lock);
 731
 732         while (!list_empty(&pending)) {
 733                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 734                 list_del_init(&dreq->recent);
 735                 dreq->revisit(dreq, 0);
 736         }
 737 }
 738
 739 void cache_clean_deferred(void *owner)
 740 {
 741         struct cache_deferred_req *dreq, *tmp;
 742         struct list_head pending;
 743
 744
 745         INIT_LIST_HEAD(&pending);
 746         spin_lock(&cache_defer_lock);
 747
 748         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 749                 if (dreq->owner == owner) {
 750                         __unhash_deferred_req(dreq);
 751                         list_add(&dreq->recent, &pending);
 752                 }
 753         }
 754         spin_unlock(&cache_defer_lock);
 755
 756         while (!list_empty(&pending)) {
 757                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 758                 list_del_init(&dreq->recent);
 759                 dreq->revisit(dreq, 1);
 760         }
 761 }
 762
 763 /*
 764  * communicate with user-space
 765  *
 766  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
 767  * On read, you get a full request, or block.
 768  * On write, an update request is processed.
 769  * Poll works if anything to read, and always allows write.
 770  *
 771  * Implemented by linked list of requests.  Each open file has
 772  * a ->private that also exists in this list.  New requests are added
 773  * to the end and may wakeup and preceding readers.
 774  * New readers are added to the head.  If, on read, an item is found with
 775  * CACHE_UPCALLING clear, we free it from the list.
 776  *
 777  */
 778
 779 static DEFINE_SPINLOCK(queue_lock);
 780 static DEFINE_MUTEX(queue_io_mutex);
 781
 782 struct cache_queue {
 783         struct list_head        list;
 784         int                     reader; /* if 0, then request */
 785 };
 786 struct cache_request {
 787         struct cache_queue      q;
 788         struct cache_head       *item;
 789         char                    * buf;
 790         int                     len;
 791         int                     readers;
 792 };
 793 struct cache_reader {
 794         struct cache_queue      q;
 795         int                     offset; /* if non-0, we have a refcnt on next request */
 796 };
 797
 798 static int cache_request(struct cache_detail *detail,
 799                                struct cache_request *crq)
 800 {
 801         char *bp = crq->buf;
 802         int len = PAGE_SIZE;
 803
 804         detail->cache_request(detail, crq->item, &bp, &len);
 805         if (len < 0)
 806                 return -EAGAIN;
 807         return PAGE_SIZE - len;
 808 }
 809
 810 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 811                           loff_t *ppos, struct cache_detail *cd)
 812 {
 813         struct cache_reader *rp = filp->private_data;
 814         struct cache_request *rq;
 815         struct inode *inode = file_inode(filp);
 816         int err;
 817
 818         if (count == 0)
 819                 return 0;
 820
 821         inode_lock(inode); /* protect against multiple concurrent
 822                               * readers on this file */
 823  again:
 824         spin_lock(&queue_lock);
 825         /* need to find next request */
 826         while (rp->q.list.next != &cd->queue &&
 827                list_entry(rp->q.list.next, struct cache_queue, list)
 828                ->reader) {
 829                 struct list_head *next = rp->q.list.next;
 830                 list_move(&rp->q.list, next);
 831         }
 832         if (rp->q.list.next == &cd->queue) {
 833                 spin_unlock(&queue_lock);
 834                 inode_unlock(inode);
 835                 WARN_ON_ONCE(rp->offset);
 836                 return 0;
 837         }
 838         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 839         WARN_ON_ONCE(rq->q.reader);
 840         if (rp->offset == 0)
 841                 rq->readers++;
 842         spin_unlock(&queue_lock);
 843
 844         if (rq->len == 0) {
 845                 err = cache_request(cd, rq);
 846                 if (err < 0)
 847                         goto out;
 848                 rq->len = err;
 849         }
 850
 851         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 852                 err = -EAGAIN;
 853                 spin_lock(&queue_lock);
 854                 list_move(&rp->q.list, &rq->q.list);
 855                 spin_unlock(&queue_lock);
 856         } else {
 857                 if (rp->offset + count > rq->len)
 858                         count = rq->len - rp->offset;
 859                 err = -EFAULT;
 860                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 861                         goto out;
 862                 rp->offset += count;
 863                 if (rp->offset >= rq->len) {
 864                         rp->offset = 0;
 865                         spin_lock(&queue_lock);
 866                         list_move(&rp->q.list, &rq->q.list);
 867                         spin_unlock(&queue_lock);
 868                 }
 869                 err = 0;
 870         }
 871  out:
 872         if (rp->offset == 0) {
 873                 /* need to release rq */
 874                 spin_lock(&queue_lock);
 875                 rq->readers--;
 876                 if (rq->readers == 0 &&
 877                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 878                         list_del(&rq->q.list);
 879                         spin_unlock(&queue_lock);
 880                         cache_put(rq->item, cd);
 881                         kfree(rq->buf);
 882                         kfree(rq);
 883                 } else
 884                         spin_unlock(&queue_lock);
 885         }
 886         if (err == -EAGAIN)
 887                 goto again;
 888         inode_unlock(inode);
 889         return err ? err :  count;
 890 }
 891
 892 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 893                                  size_t count, struct cache_detail *cd)
 894 {
 895         ssize_t ret;
 896
 897         if (count == 0)
 898                 return -EINVAL;
 899         if (copy_from_user(kaddr, buf, count))
 900                 return -EFAULT;
 901         kaddr[count] = '\0';
 902         ret = cd->cache_parse(cd, kaddr, count);
 903         if (!ret)
 904                 ret = count;
 905         return ret;
 906 }
 907
 908 static ssize_t cache_slow_downcall(const char __user *buf,
 909                                    size_t count, struct cache_detail *cd)
 910 {
 911         static char write_buf[8192]; /* protected by queue_io_mutex */
 912         ssize_t ret = -EINVAL;
 913
 914         if (count >= sizeof(write_buf))
 915                 goto out;
 916         mutex_lock(&queue_io_mutex);
 917         ret = cache_do_downcall(write_buf, buf, count, cd);
 918         mutex_unlock(&queue_io_mutex);
 919 out:
 920         return ret;
 921 }
 922
 923 static ssize_t cache_downcall(struct address_space *mapping,
 924                               const char __user *buf,
 925                               size_t count, struct cache_detail *cd)
 926 {
 927         struct page *page;
 928         char *kaddr;
 929         ssize_t ret = -ENOMEM;
 930
 931         if (count >= PAGE_SIZE)
 932                 goto out_slow;
 933
 934         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 935         if (!page)
 936                 goto out_slow;
 937
 938         kaddr = kmap(page);
 939         ret = cache_do_downcall(kaddr, buf, count, cd);
 940         kunmap(page);
 941         unlock_page(page);
 942         put_page(page);
 943         return ret;
 944 out_slow:
 945         return cache_slow_downcall(buf, count, cd);
 946 }
 947
 948 static ssize_t cache_write(struct file *filp, const char __user *buf,
 949                            size_t count, loff_t *ppos,
 950                            struct cache_detail *cd)
 951 {
 952         struct address_space *mapping = filp->f_mapping;
 953         struct inode *inode = file_inode(filp);
 954         ssize_t ret = -EINVAL;
 955
 956         if (!cd->cache_parse)
 957                 goto out;
 958
 959         inode_lock(inode);
 960         ret = cache_downcall(mapping, buf, count, cd);
 961         inode_unlock(inode);
 962 out:
 963         return ret;
 964 }
 965
 966 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 967
 968 static __poll_t cache_poll(struct file *filp, poll_table *wait,
 969                                struct cache_detail *cd)
 970 {
 971         __poll_t mask;
 972         struct cache_reader *rp = filp->private_data;
 973         struct cache_queue *cq;
 974
 975         poll_wait(filp, &queue_wait, wait);
 976
 977         /* alway allow write */
 978         mask = EPOLLOUT | EPOLLWRNORM;
 979
 980         if (!rp)
 981                 return mask;
 982
 983         spin_lock(&queue_lock);
 984
 985         for (cq= &rp->q; &cq->list != &cd->queue;
 986              cq = list_entry(cq->list.next, struct cache_queue, list))
 987                 if (!cq->reader) {
 988                         mask |= EPOLLIN | EPOLLRDNORM;
 989                         break;
 990                 }
 991         spin_unlock(&queue_lock);
 992         return mask;
 993 }
 994
 995 static int cache_ioctl(struct inode *ino, struct file *filp,
 996                        unsigned int cmd, unsigned long arg,
 997                        struct cache_detail *cd)
 998 {
 999         int len = 0;
1000         struct cache_reader *rp = filp->private_data;
1001         struct cache_queue *cq;
1002
1003         if (cmd != FIONREAD || !rp)
1004                 return -EINVAL;
1005
1006         spin_lock(&queue_lock);
1007
1008         /* only find the length remaining in current request,
1009          * or the length of the next request
1010          */
1011         for (cq= &rp->q; &cq->list != &cd->queue;
1012              cq = list_entry(cq->list.next, struct cache_queue, list))
1013                 if (!cq->reader) {
1014                         struct cache_request *cr =
1015                                 container_of(cq, struct cache_request, q);
1016                         len = cr->len - rp->offset;
1017                         break;
1018                 }
1019         spin_unlock(&queue_lock);
1020
1021         return put_user(len, (int __user *)arg);
1022 }
1023
1024 static int cache_open(struct inode *inode, struct file *filp,
1025                       struct cache_detail *cd)
1026 {
1027         struct cache_reader *rp = NULL;
1028
1029         if (!cd || !try_module_get(cd->owner))
1030                 return -EACCES;
1031         nonseekable_open(inode, filp);
1032         if (filp->f_mode & FMODE_READ) {
1033                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1034                 if (!rp) {
1035                         module_put(cd->owner);
1036                         return -ENOMEM;
1037                 }
1038                 rp->offset = 0;
1039                 rp->q.reader = 1;
1040
1041                 spin_lock(&queue_lock);
1042                 list_add(&rp->q.list, &cd->queue);
1043                 spin_unlock(&queue_lock);
1044         }
1045         if (filp->f_mode & FMODE_WRITE)
1046                 atomic_inc(&cd->writers);
1047         filp->private_data = rp;
1048         return 0;
1049 }
1050
1051 static int cache_release(struct inode *inode, struct file *filp,
1052                          struct cache_detail *cd)
1053 {
1054         struct cache_reader *rp = filp->private_data;
1055
1056         if (rp) {
1057                 spin_lock(&queue_lock);
1058                 if (rp->offset) {
1059                         struct cache_queue *cq;
1060                         for (cq= &rp->q; &cq->list != &cd->queue;
1061                              cq = list_entry(cq->list.next, struct cache_queue, list))
1062                                 if (!cq->reader) {
1063                                         container_of(cq, struct cache_request, q)
1064                                                 ->readers--;
1065                                         break;
1066                                 }
1067                         rp->offset = 0;
1068                 }
1069                 list_del(&rp->q.list);
1070                 spin_unlock(&queue_lock);
1071
1072                 filp->private_data = NULL;
1073                 kfree(rp);
1074
1075         }
1076         if (filp->f_mode & FMODE_WRITE) {
1077                 atomic_dec(&cd->writers);
1078                 cd->last_close = seconds_since_boot();
1079         }
1080         module_put(cd->owner);
1081         return 0;
1082 }
1083
1084
1085
1086 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1087 {
1088         struct cache_queue *cq, *tmp;
1089         struct cache_request *cr;
1090         struct list_head dequeued;
1091
1092         INIT_LIST_HEAD(&dequeued);
1093         spin_lock(&queue_lock);
1094         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1095                 if (!cq->reader) {
1096                         cr = container_of(cq, struct cache_request, q);
1097                         if (cr->item != ch)
1098                                 continue;
1099                         if (test_bit(CACHE_PENDING, &ch->flags))
1100                                 /* Lost a race and it is pending again */
1101                                 break;
1102                         if (cr->readers != 0)
1103                                 continue;
1104                         list_move(&cr->q.list, &dequeued);
1105                 }
1106         spin_unlock(&queue_lock);
1107         while (!list_empty(&dequeued)) {
1108                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1109                 list_del(&cr->q.list);
1110                 cache_put(cr->item, detail);
1111                 kfree(cr->buf);
1112                 kfree(cr);
1113         }
1114 }
1115
1116 /*
1117  * Support routines for text-based upcalls.
1118  * Fields are separated by spaces.
1119  * Fields are either mangled to quote space tab newline slosh with slosh
1120  * or a hexified with a leading \x
1121  * Record is terminated with newline.
1122  *
1123  */
1124
1125 void qword_add(char **bpp, int *lp, char *str)
1126 {
1127         char *bp = *bpp;
1128         int len = *lp;
1129         int ret;
1130
1131         if (len < 0) return;
1132
1133         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1134         if (ret >= len) {
1135                 bp += len;
1136                 len = -1;
1137         } else {
1138                 bp += ret;
1139                 len -= ret;
1140                 *bp++ = ' ';
1141                 len--;
1142         }
1143         *bpp = bp;
1144         *lp = len;
1145 }
1146 EXPORT_SYMBOL_GPL(qword_add);
1147
1148 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1149 {
1150         char *bp = *bpp;
1151         int len = *lp;
1152
1153         if (len < 0) return;
1154
1155         if (len > 2) {
1156                 *bp++ = '\\';
1157                 *bp++ = 'x';
1158                 len -= 2;
1159                 while (blen && len >= 2) {
1160                         bp = hex_byte_pack(bp, *buf++);
1161                         len -= 2;
1162                         blen--;
1163                 }
1164         }
1165         if (blen || len<1) len = -1;
1166         else {
1167                 *bp++ = ' ';
1168                 len--;
1169         }
1170         *bpp = bp;
1171         *lp = len;
1172 }
1173 EXPORT_SYMBOL_GPL(qword_addhex);
1174
1175 static void warn_no_listener(struct cache_detail *detail)
1176 {
1177         if (detail->last_warn != detail->last_close) {
1178                 detail->last_warn = detail->last_close;
1179                 if (detail->warn_no_listener)
1180                         detail->warn_no_listener(detail, detail->last_close != 0);
1181         }
1182 }
1183
1184 static bool cache_listeners_exist(struct cache_detail *detail)
1185 {
1186         if (atomic_read(&detail->writers))
1187                 return true;
1188         if (detail->last_close == 0)
1189                 /* This cache was never opened */
1190                 return false;
1191         if (detail->last_close < seconds_since_boot() - 30)
1192                 /*
1193                  * We allow for the possibility that someone might
1194                  * restart a userspace daemon without restarting the
1195                  * server; but after 30 seconds, we give up.
1196                  */
1197                  return false;
1198         return true;
1199 }
1200
1201 /*
1202  * register an upcall request to user-space and queue it up for read() by the
1203  * upcall daemon.
1204  *
1205  * Each request is at most one page long.
1206  */
1207 static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1208 {
1209         char *buf;
1210         struct cache_request *crq;
1211         int ret = 0;
1212
1213         if (test_bit(CACHE_CLEANED, &h->flags))
1214                 /* Too late to make an upcall */
1215                 return -EAGAIN;
1216
1217         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1218         if (!buf)
1219                 return -EAGAIN;
1220
1221         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1222         if (!crq) {
1223                 kfree(buf);
1224                 return -EAGAIN;
1225         }
1226
1227         crq->q.reader = 0;
1228         crq->buf = buf;
1229         crq->len = 0;
1230         crq->readers = 0;
1231         spin_lock(&queue_lock);
1232         if (test_bit(CACHE_PENDING, &h->flags)) {
1233                 crq->item = cache_get(h);
1234                 list_add_tail(&crq->q.list, &detail->queue);
1235                 trace_cache_entry_upcall(detail, h);
1236         } else
1237                 /* Lost a race, no longer PENDING, so don't enqueue */
1238                 ret = -EAGAIN;
1239         spin_unlock(&queue_lock);
1240         wake_up(&queue_wait);
1241         if (ret == -EAGAIN) {
1242                 kfree(buf);
1243                 kfree(crq);
1244         }
1245         return ret;
1246 }
1247
1248 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1249 {
1250         if (test_and_set_bit(CACHE_PENDING, &h->flags))
1251                 return 0;
1252         return cache_pipe_upcall(detail, h);
1253 }
1254 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1255
1256 int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1257                                      struct cache_head *h)
1258 {
1259         if (!cache_listeners_exist(detail)) {
1260                 warn_no_listener(detail);
1261                 trace_cache_entry_no_listener(detail, h);
1262                 return -EINVAL;
1263         }
1264         return sunrpc_cache_pipe_upcall(detail, h);
1265 }
1266 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1267
1268 /*
1269  * parse a message from user-space and pass it
1270  * to an appropriate cache
1271  * Messages are, like requests, separated into fields by
1272  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1273  *
1274  * Message is
1275  *   reply cachename expiry key ... content....
1276  *
1277  * key and content are both parsed by cache
1278  */
1279
1280 int qword_get(char **bpp, char *dest, int bufsize)
1281 {
1282         /* return bytes copied, or -1 on error */
1283         char *bp = *bpp;
1284         int len = 0;
1285
1286         while (*bp == ' ') bp++;
1287
1288         if (bp[0] == '\\' && bp[1] == 'x') {
1289                 /* HEX STRING */
1290                 bp += 2;
1291                 while (len < bufsize - 1) {
1292                         int h, l;
1293
1294                         h = hex_to_bin(bp[0]);
1295                         if (h < 0)
1296                                 break;
1297
1298                         l = hex_to_bin(bp[1]);
1299                         if (l < 0)
1300                                 break;
1301
1302                         *dest++ = (h << 4) | l;
1303                         bp += 2;
1304                         len++;
1305                 }
1306         } else {
1307                 /* text with \nnn octal quoting */
1308                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1309                         if (*bp == '\\' &&
1310                             isodigit(bp[1]) && (bp[1] <= '3') &&
1311                             isodigit(bp[2]) &&
1312                             isodigit(bp[3])) {
1313                                 int byte = (*++bp -'0');
1314                                 bp++;
1315                                 byte = (byte << 3) | (*bp++ - '0');
1316                                 byte = (byte << 3) | (*bp++ - '0');
1317                                 *dest++ = byte;
1318                                 len++;
1319                         } else {
1320                                 *dest++ = *bp++;
1321                                 len++;
1322                         }
1323                 }
1324         }
1325
1326         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1327                 return -1;
1328         while (*bp == ' ') bp++;
1329         *bpp = bp;
1330         *dest = '\0';
1331         return len;
1332 }
1333 EXPORT_SYMBOL_GPL(qword_get);
1334
1335
1336 /*
1337  * support /proc/net/rpc/$CACHENAME/content
1338  * as a seqfile.
1339  * We call ->cache_show passing NULL for the item to
1340  * get a header, then pass each real item in the cache
1341  */
1342
1343 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1344 {
1345         loff_t n = *pos;
1346         unsigned int hash, entry;
1347         struct cache_head *ch;
1348         struct cache_detail *cd = m->private;
1349
1350         if (!n--)
1351                 return SEQ_START_TOKEN;
1352         hash = n >> 32;
1353         entry = n & ((1LL<<32) - 1);
1354
1355         hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1356                 if (!entry--)
1357                         return ch;
1358         n &= ~((1LL<<32) - 1);
1359         do {
1360                 hash++;
1361                 n += 1LL<<32;
1362         } while(hash < cd->hash_size &&
1363                 hlist_empty(&cd->hash_table[hash]));
1364         if (hash >= cd->hash_size)
1365                 return NULL;
1366         *pos = n+1;
1367         return hlist_entry_safe(rcu_dereference_raw(
1368                                 hlist_first_rcu(&cd->hash_table[hash])),
1369                                 struct cache_head, cache_list);
1370 }
1371
1372 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1373 {
1374         struct cache_head *ch = p;
1375         int hash = (*pos >> 32);
1376         struct cache_detail *cd = m->private;
1377
1378         if (p == SEQ_START_TOKEN)
1379                 hash = 0;
1380         else if (ch->cache_list.next == NULL) {
1381                 hash++;
1382                 *pos += 1LL<<32;
1383         } else {
1384                 ++*pos;
1385                 return hlist_entry_safe(rcu_dereference_raw(
1386                                         hlist_next_rcu(&ch->cache_list)),
1387                                         struct cache_head, cache_list);
1388         }
1389         *pos &= ~((1LL<<32) - 1);
1390         while (hash < cd->hash_size &&
1391                hlist_empty(&cd->hash_table[hash])) {
1392                 hash++;
1393                 *pos += 1LL<<32;
1394         }
1395         if (hash >= cd->hash_size)
1396                 return NULL;
1397         ++*pos;
1398         return hlist_entry_safe(rcu_dereference_raw(
1399                                 hlist_first_rcu(&cd->hash_table[hash])),
1400                                 struct cache_head, cache_list);
1401 }
1402
1403 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1404         __acquires(RCU)
1405 {
1406         rcu_read_lock();
1407         return __cache_seq_start(m, pos);
1408 }
1409 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1410
1411 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1412 {
1413         return cache_seq_next(file, p, pos);
1414 }
1415 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1416
1417 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1418         __releases(RCU)
1419 {
1420         rcu_read_unlock();
1421 }
1422 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1423
1424 static int c_show(struct seq_file *m, void *p)
1425 {
1426         struct cache_head *cp = p;
1427         struct cache_detail *cd = m->private;
1428
1429         if (p == SEQ_START_TOKEN)
1430                 return cd->cache_show(m, cd, NULL);
1431
1432         ifdebug(CACHE)
1433                 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1434                            convert_to_wallclock(cp->expiry_time),
1435                            kref_read(&cp->ref), cp->flags);
1436         cache_get(cp);
1437         if (cache_check(cd, cp, NULL))
1438                 /* cache_check does a cache_put on failure */
1439                 seq_printf(m, "# ");
1440         else {
1441                 if (cache_is_expired(cd, cp))
1442                         seq_printf(m, "# ");
1443                 cache_put(cp, cd);
1444         }
1445
1446         return cd->cache_show(m, cd, cp);
1447 }
1448
1449 static const struct seq_operations cache_content_op = {
1450         .start  = cache_seq_start_rcu,
1451         .next   = cache_seq_next_rcu,
1452         .stop   = cache_seq_stop_rcu,
1453         .show   = c_show,
1454 };
1455
1456 static int content_open(struct inode *inode, struct file *file,
1457                         struct cache_detail *cd)
1458 {
1459         struct seq_file *seq;
1460         int err;
1461
1462         if (!cd || !try_module_get(cd->owner))
1463                 return -EACCES;
1464
1465         err = seq_open(file, &cache_content_op);
1466         if (err) {
1467                 module_put(cd->owner);
1468                 return err;
1469         }
1470
1471         seq = file->private_data;
1472         seq->private = cd;
1473         return 0;
1474 }
1475
1476 static int content_release(struct inode *inode, struct file *file,
1477                 struct cache_detail *cd)
1478 {
1479         int ret = seq_release(inode, file);
1480         module_put(cd->owner);
1481         return ret;
1482 }
1483
1484 static int open_flush(struct inode *inode, struct file *file,
1485                         struct cache_detail *cd)
1486 {
1487         if (!cd || !try_module_get(cd->owner))
1488                 return -EACCES;
1489         return nonseekable_open(inode, file);
1490 }
1491
1492 static int release_flush(struct inode *inode, struct file *file,
1493                         struct cache_detail *cd)
1494 {
1495         module_put(cd->owner);
1496         return 0;
1497 }
1498
1499 static ssize_t read_flush(struct file *file, char __user *buf,
1500                           size_t count, loff_t *ppos,
1501                           struct cache_detail *cd)
1502 {
1503         char tbuf[22];
1504         size_t len;
1505
1506         len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1507                         convert_to_wallclock(cd->flush_time));
1508         return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1509 }
1510
1511 static ssize_t write_flush(struct file *file, const char __user *buf,
1512                            size_t count, loff_t *ppos,
1513                            struct cache_detail *cd)
1514 {
1515         char tbuf[20];
1516         char *ep;
1517         time64_t now;
1518
1519         if (*ppos || count > sizeof(tbuf)-1)
1520                 return -EINVAL;
1521         if (copy_from_user(tbuf, buf, count))
1522                 return -EFAULT;
1523         tbuf[count] = 0;
1524         simple_strtoul(tbuf, &ep, 0);
1525         if (*ep && *ep != '\n')
1526                 return -EINVAL;
1527         /* Note that while we check that 'buf' holds a valid number,
1528          * we always ignore the value and just flush everything.
1529          * Making use of the number leads to races.
1530          */
1531
1532         now = seconds_since_boot();
1533         /* Always flush everything, so behave like cache_purge()
1534          * Do this by advancing flush_time to the current time,
1535          * or by one second if it has already reached the current time.
1536          * Newly added cache entries will always have ->last_refresh greater
1537          * that ->flush_time, so they don't get flushed prematurely.
1538          */
1539
1540         if (cd->flush_time >= now)
1541                 now = cd->flush_time + 1;
1542
1543         cd->flush_time = now;
1544         cd->nextcheck = now;
1545         cache_flush();
1546
1547         if (cd->flush)
1548                 cd->flush();
1549
1550         *ppos += count;
1551         return count;
1552 }
1553
1554 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1555                                  size_t count, loff_t *ppos)
1556 {
1557         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1558
1559         return cache_read(filp, buf, count, ppos, cd);
1560 }
1561
1562 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1563                                   size_t count, loff_t *ppos)
1564 {
1565         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1566
1567         return cache_write(filp, buf, count, ppos, cd);
1568 }
1569
1570 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1571 {
1572         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1573
1574         return cache_poll(filp, wait, cd);
1575 }
1576
1577 static long cache_ioctl_procfs(struct file *filp,
1578                                unsigned int cmd, unsigned long arg)
1579 {
1580         struct inode *inode = file_inode(filp);
1581         struct cache_detail *cd = PDE_DATA(inode);
1582
1583         return cache_ioctl(inode, filp, cmd, arg, cd);
1584 }
1585
1586 static int cache_open_procfs(struct inode *inode, struct file *filp)
1587 {
1588         struct cache_detail *cd = PDE_DATA(inode);
1589
1590         return cache_open(inode, filp, cd);
1591 }
1592
1593 static int cache_release_procfs(struct inode *inode, struct file *filp)
1594 {
1595         struct cache_detail *cd = PDE_DATA(inode);
1596
1597         return cache_release(inode, filp, cd);
1598 }
1599
1600 static const struct proc_ops cache_channel_proc_ops = {
1601         .proc_lseek     = no_llseek,
1602         .proc_read      = cache_read_procfs,
1603         .proc_write     = cache_write_procfs,
1604         .proc_poll      = cache_poll_procfs,
1605         .proc_ioctl     = cache_ioctl_procfs, /* for FIONREAD */
1606         .proc_open      = cache_open_procfs,
1607         .proc_release   = cache_release_procfs,
1608 };
1609
1610 static int content_open_procfs(struct inode *inode, struct file *filp)
1611 {
1612         struct cache_detail *cd = PDE_DATA(inode);
1613
1614         return content_open(inode, filp, cd);
1615 }
1616
1617 static int content_release_procfs(struct inode *inode, struct file *filp)
1618 {
1619         struct cache_detail *cd = PDE_DATA(inode);
1620
1621         return content_release(inode, filp, cd);
1622 }
1623
1624 static const struct proc_ops content_proc_ops = {
1625         .proc_open      = content_open_procfs,
1626         .proc_read      = seq_read,
1627         .proc_lseek     = seq_lseek,
1628         .proc_release   = content_release_procfs,
1629 };
1630
1631 static int open_flush_procfs(struct inode *inode, struct file *filp)
1632 {
1633         struct cache_detail *cd = PDE_DATA(inode);
1634
1635         return open_flush(inode, filp, cd);
1636 }
1637
1638 static int release_flush_procfs(struct inode *inode, struct file *filp)
1639 {
1640         struct cache_detail *cd = PDE_DATA(inode);
1641
1642         return release_flush(inode, filp, cd);
1643 }
1644
1645 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1646                             size_t count, loff_t *ppos)
1647 {
1648         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1649
1650         return read_flush(filp, buf, count, ppos, cd);
1651 }
1652
1653 static ssize_t write_flush_procfs(struct file *filp,
1654                                   const char __user *buf,
1655                                   size_t count, loff_t *ppos)
1656 {
1657         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1658
1659         return write_flush(filp, buf, count, ppos, cd);
1660 }
1661
1662 static const struct proc_ops cache_flush_proc_ops = {
1663         .proc_open      = open_flush_procfs,
1664         .proc_read      = read_flush_procfs,
1665         .proc_write     = write_flush_procfs,
1666         .proc_release   = release_flush_procfs,
1667         .proc_lseek     = no_llseek,
1668 };
1669
1670 static void remove_cache_proc_entries(struct cache_detail *cd)
1671 {
1672         if (cd->procfs) {
1673                 proc_remove(cd->procfs);
1674                 cd->procfs = NULL;
1675         }
1676 }
1677
1678 #ifdef CONFIG_PROC_FS
1679 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1680 {
1681         struct proc_dir_entry *p;
1682         struct sunrpc_net *sn;
1683
1684         sn = net_generic(net, sunrpc_net_id);
1685         cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1686         if (cd->procfs == NULL)
1687                 goto out_nomem;
1688
1689         p = proc_create_data("flush", S_IFREG | 0600,
1690                              cd->procfs, &cache_flush_proc_ops, cd);
1691         if (p == NULL)
1692                 goto out_nomem;
1693
1694         if (cd->cache_request || cd->cache_parse) {
1695                 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1696                                      &cache_channel_proc_ops, cd);
1697                 if (p == NULL)
1698                         goto out_nomem;
1699         }
1700         if (cd->cache_show) {
1701                 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1702                                      &content_proc_ops, cd);
1703                 if (p == NULL)
1704                         goto out_nomem;
1705         }
1706         return 0;
1707 out_nomem:
1708         remove_cache_proc_entries(cd);
1709         return -ENOMEM;
1710 }
1711 #else /* CONFIG_PROC_FS */
1712 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1713 {
1714         return 0;
1715 }
1716 #endif
1717
1718 void __init cache_initialize(void)
1719 {
1720         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1721 }
1722
1723 int cache_register_net(struct cache_detail *cd, struct net *net)
1724 {
1725         int ret;
1726
1727         sunrpc_init_cache_detail(cd);
1728         ret = create_cache_proc_entries(cd, net);
1729         if (ret)
1730                 sunrpc_destroy_cache_detail(cd);
1731         return ret;
1732 }
1733 EXPORT_SYMBOL_GPL(cache_register_net);
1734
1735 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1736 {
1737         remove_cache_proc_entries(cd);
1738         sunrpc_destroy_cache_detail(cd);
1739 }
1740 EXPORT_SYMBOL_GPL(cache_unregister_net);
1741
1742 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1743 {
1744         struct cache_detail *cd;
1745         int i;
1746
1747         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1748         if (cd == NULL)
1749                 return ERR_PTR(-ENOMEM);
1750
1751         cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1752                                  GFP_KERNEL);
1753         if (cd->hash_table == NULL) {
1754                 kfree(cd);
1755                 return ERR_PTR(-ENOMEM);
1756         }
1757
1758         for (i = 0; i < cd->hash_size; i++)
1759                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1760         cd->net = net;
1761         return cd;
1762 }
1763 EXPORT_SYMBOL_GPL(cache_create_net);
1764
1765 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1766 {
1767         kfree(cd->hash_table);
1768         kfree(cd);
1769 }
1770 EXPORT_SYMBOL_GPL(cache_destroy_net);
1771
1772 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1773                                  size_t count, loff_t *ppos)
1774 {
1775         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1776
1777         return cache_read(filp, buf, count, ppos, cd);
1778 }
1779
1780 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1781                                   size_t count, loff_t *ppos)
1782 {
1783         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1784
1785         return cache_write(filp, buf, count, ppos, cd);
1786 }
1787
1788 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1789 {
1790         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1791
1792         return cache_poll(filp, wait, cd);
1793 }
1794
1795 static long cache_ioctl_pipefs(struct file *filp,
1796                               unsigned int cmd, unsigned long arg)
1797 {
1798         struct inode *inode = file_inode(filp);
1799         struct cache_detail *cd = RPC_I(inode)->private;
1800
1801         return cache_ioctl(inode, filp, cmd, arg, cd);
1802 }
1803
1804 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1805 {
1806         struct cache_detail *cd = RPC_I(inode)->private;
1807
1808         return cache_open(inode, filp, cd);
1809 }
1810
1811 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1812 {
1813         struct cache_detail *cd = RPC_I(inode)->private;
1814
1815         return cache_release(inode, filp, cd);
1816 }
1817
1818 const struct file_operations cache_file_operations_pipefs = {
1819         .owner          = THIS_MODULE,
1820         .llseek         = no_llseek,
1821         .read           = cache_read_pipefs,
1822         .write          = cache_write_pipefs,
1823         .poll           = cache_poll_pipefs,
1824         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1825         .open           = cache_open_pipefs,
1826         .release        = cache_release_pipefs,
1827 };
1828
1829 static int content_open_pipefs(struct inode *inode, struct file *filp)
1830 {
1831         struct cache_detail *cd = RPC_I(inode)->private;
1832
1833         return content_open(inode, filp, cd);
1834 }
1835
1836 static int content_release_pipefs(struct inode *inode, struct file *filp)
1837 {
1838         struct cache_detail *cd = RPC_I(inode)->private;
1839
1840         return content_release(inode, filp, cd);
1841 }
1842
1843 const struct file_operations content_file_operations_pipefs = {
1844         .open           = content_open_pipefs,
1845         .read           = seq_read,
1846         .llseek         = seq_lseek,
1847         .release        = content_release_pipefs,
1848 };
1849
1850 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1851 {
1852         struct cache_detail *cd = RPC_I(inode)->private;
1853
1854         return open_flush(inode, filp, cd);
1855 }
1856
1857 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1858 {
1859         struct cache_detail *cd = RPC_I(inode)->private;
1860
1861         return release_flush(inode, filp, cd);
1862 }
1863
1864 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1865                             size_t count, loff_t *ppos)
1866 {
1867         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1868
1869         return read_flush(filp, buf, count, ppos, cd);
1870 }
1871
1872 static ssize_t write_flush_pipefs(struct file *filp,
1873                                   const char __user *buf,
1874                                   size_t count, loff_t *ppos)
1875 {
1876         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1877
1878         return write_flush(filp, buf, count, ppos, cd);
1879 }
1880
1881 const struct file_operations cache_flush_operations_pipefs = {
1882         .open           = open_flush_pipefs,
1883         .read           = read_flush_pipefs,
1884         .write          = write_flush_pipefs,
1885         .release        = release_flush_pipefs,
1886         .llseek         = no_llseek,
1887 };
1888
1889 int sunrpc_cache_register_pipefs(struct dentry *parent,
1890                                  const char *name, umode_t umode,
1891                                  struct cache_detail *cd)
1892 {
1893         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1894         if (IS_ERR(dir))
1895                 return PTR_ERR(dir);
1896         cd->pipefs = dir;
1897         return 0;
1898 }
1899 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1900
1901 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1902 {
1903         if (cd->pipefs) {
1904                 rpc_remove_cache_dir(cd->pipefs);
1905                 cd->pipefs = NULL;
1906         }
1907 }
1908 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1909
1910 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1911 {
1912         spin_lock(&cd->hash_lock);
1913         if (!hlist_unhashed(&h->cache_list)){
1914                 sunrpc_begin_cache_remove_entry(h, cd);
1915                 spin_unlock(&cd->hash_lock);
1916                 sunrpc_end_cache_remove_entry(h, cd);
1917         } else
1918                 spin_unlock(&cd->hash_lock);
1919 }
1920 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);