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         struct hlist_node *tmp = NULL;
 533         int i = 0;
 534
 535         spin_lock(&detail->hash_lock);
 536         if (!detail->entries) {
 537                 spin_unlock(&detail->hash_lock);
 538                 return;
 539         }
 540
 541         dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
 542         for (i = 0; i < detail->hash_size; i++) {
 543                 head = &detail->hash_table[i];
 544                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 545                         sunrpc_begin_cache_remove_entry(ch, detail);
 546                         spin_unlock(&detail->hash_lock);
 547                         sunrpc_end_cache_remove_entry(ch, detail);
 548                         spin_lock(&detail->hash_lock);
 549                 }
 550         }
 551         spin_unlock(&detail->hash_lock);
 552 }
 553 EXPORT_SYMBOL_GPL(cache_purge);
 554
 555
 556 /*
 557  * Deferral and Revisiting of Requests.
 558  *
 559  * If a cache lookup finds a pending entry, we
 560  * need to defer the request and revisit it later.
 561  * All deferred requests are stored in a hash table,
 562  * indexed by "struct cache_head *".
 563  * As it may be wasteful to store a whole request
 564  * structure, we allow the request to provide a
 565  * deferred form, which must contain a
 566  * 'struct cache_deferred_req'
 567  * This cache_deferred_req contains a method to allow
 568  * it to be revisited when cache info is available
 569  */
 570
 571 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 572 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 573
 574 #define DFR_MAX 300     /* ??? */
 575
 576 static DEFINE_SPINLOCK(cache_defer_lock);
 577 static LIST_HEAD(cache_defer_list);
 578 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 579 static int cache_defer_cnt;
 580
 581 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 582 {
 583         hlist_del_init(&dreq->hash);
 584         if (!list_empty(&dreq->recent)) {
 585                 list_del_init(&dreq->recent);
 586                 cache_defer_cnt--;
 587         }
 588 }
 589
 590 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 591 {
 592         int hash = DFR_HASH(item);
 593
 594         INIT_LIST_HEAD(&dreq->recent);
 595         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 596 }
 597
 598 static void setup_deferral(struct cache_deferred_req *dreq,
 599                            struct cache_head *item,
 600                            int count_me)
 601 {
 602
 603         dreq->item = item;
 604
 605         spin_lock(&cache_defer_lock);
 606
 607         __hash_deferred_req(dreq, item);
 608
 609         if (count_me) {
 610                 cache_defer_cnt++;
 611                 list_add(&dreq->recent, &cache_defer_list);
 612         }
 613
 614         spin_unlock(&cache_defer_lock);
 615
 616 }
 617
 618 struct thread_deferred_req {
 619         struct cache_deferred_req handle;
 620         struct completion completion;
 621 };
 622
 623 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 624 {
 625         struct thread_deferred_req *dr =
 626                 container_of(dreq, struct thread_deferred_req, handle);
 627         complete(&dr->completion);
 628 }
 629
 630 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 631 {
 632         struct thread_deferred_req sleeper;
 633         struct cache_deferred_req *dreq = &sleeper.handle;
 634
 635         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 636         dreq->revisit = cache_restart_thread;
 637
 638         setup_deferral(dreq, item, 0);
 639
 640         if (!test_bit(CACHE_PENDING, &item->flags) ||
 641             wait_for_completion_interruptible_timeout(
 642                     &sleeper.completion, req->thread_wait) <= 0) {
 643                 /* The completion wasn't completed, so we need
 644                  * to clean up
 645                  */
 646                 spin_lock(&cache_defer_lock);
 647                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 648                         __unhash_deferred_req(&sleeper.handle);
 649                         spin_unlock(&cache_defer_lock);
 650                 } else {
 651                         /* cache_revisit_request already removed
 652                          * this from the hash table, but hasn't
 653                          * called ->revisit yet.  It will very soon
 654                          * and we need to wait for it.
 655                          */
 656                         spin_unlock(&cache_defer_lock);
 657                         wait_for_completion(&sleeper.completion);
 658                 }
 659         }
 660 }
 661
 662 static void cache_limit_defers(void)
 663 {
 664         /* Make sure we haven't exceed the limit of allowed deferred
 665          * requests.
 666          */
 667         struct cache_deferred_req *discard = NULL;
 668
 669         if (cache_defer_cnt <= DFR_MAX)
 670                 return;
 671
 672         spin_lock(&cache_defer_lock);
 673
 674         /* Consider removing either the first or the last */
 675         if (cache_defer_cnt > DFR_MAX) {
 676                 if (prandom_u32() & 1)
 677                         discard = list_entry(cache_defer_list.next,
 678                                              struct cache_deferred_req, recent);
 679                 else
 680                         discard = list_entry(cache_defer_list.prev,
 681                                              struct cache_deferred_req, recent);
 682                 __unhash_deferred_req(discard);
 683         }
 684         spin_unlock(&cache_defer_lock);
 685         if (discard)
 686                 discard->revisit(discard, 1);
 687 }
 688
 689 /* Return true if and only if a deferred request is queued. */
 690 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 691 {
 692         struct cache_deferred_req *dreq;
 693
 694         if (req->thread_wait) {
 695                 cache_wait_req(req, item);
 696                 if (!test_bit(CACHE_PENDING, &item->flags))
 697                         return false;
 698         }
 699         dreq = req->defer(req);
 700         if (dreq == NULL)
 701                 return false;
 702         setup_deferral(dreq, item, 1);
 703         if (!test_bit(CACHE_PENDING, &item->flags))
 704                 /* Bit could have been cleared before we managed to
 705                  * set up the deferral, so need to revisit just in case
 706                  */
 707                 cache_revisit_request(item);
 708
 709         cache_limit_defers();
 710         return true;
 711 }
 712
 713 static void cache_revisit_request(struct cache_head *item)
 714 {
 715         struct cache_deferred_req *dreq;
 716         struct list_head pending;
 717         struct hlist_node *tmp;
 718         int hash = DFR_HASH(item);
 719
 720         INIT_LIST_HEAD(&pending);
 721         spin_lock(&cache_defer_lock);
 722
 723         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 724                 if (dreq->item == item) {
 725                         __unhash_deferred_req(dreq);
 726                         list_add(&dreq->recent, &pending);
 727                 }
 728
 729         spin_unlock(&cache_defer_lock);
 730
 731         while (!list_empty(&pending)) {
 732                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 733                 list_del_init(&dreq->recent);
 734                 dreq->revisit(dreq, 0);
 735         }
 736 }
 737
 738 void cache_clean_deferred(void *owner)
 739 {
 740         struct cache_deferred_req *dreq, *tmp;
 741         struct list_head pending;
 742
 743
 744         INIT_LIST_HEAD(&pending);
 745         spin_lock(&cache_defer_lock);
 746
 747         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 748                 if (dreq->owner == owner) {
 749                         __unhash_deferred_req(dreq);
 750                         list_add(&dreq->recent, &pending);
 751                 }
 752         }
 753         spin_unlock(&cache_defer_lock);
 754
 755         while (!list_empty(&pending)) {
 756                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 757                 list_del_init(&dreq->recent);
 758                 dreq->revisit(dreq, 1);
 759         }
 760 }
 761
 762 /*
 763  * communicate with user-space
 764  *
 765  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
 766  * On read, you get a full request, or block.
 767  * On write, an update request is processed.
 768  * Poll works if anything to read, and always allows write.
 769  *
 770  * Implemented by linked list of requests.  Each open file has
 771  * a ->private that also exists in this list.  New requests are added
 772  * to the end and may wakeup and preceding readers.
 773  * New readers are added to the head.  If, on read, an item is found with
 774  * CACHE_UPCALLING clear, we free it from the list.
 775  *
 776  */
 777
 778 static DEFINE_SPINLOCK(queue_lock);
 779 static DEFINE_MUTEX(queue_io_mutex);
 780
 781 struct cache_queue {
 782         struct list_head        list;
 783         int                     reader; /* if 0, then request */
 784 };
 785 struct cache_request {
 786         struct cache_queue      q;
 787         struct cache_head       *item;
 788         char                    * buf;
 789         int                     len;
 790         int                     readers;
 791 };
 792 struct cache_reader {
 793         struct cache_queue      q;
 794         int                     offset; /* if non-0, we have a refcnt on next request */
 795 };
 796
 797 static int cache_request(struct cache_detail *detail,
 798                                struct cache_request *crq)
 799 {
 800         char *bp = crq->buf;
 801         int len = PAGE_SIZE;
 802
 803         detail->cache_request(detail, crq->item, &bp, &len);
 804         if (len < 0)
 805                 return -EAGAIN;
 806         return PAGE_SIZE - len;
 807 }
 808
 809 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 810                           loff_t *ppos, struct cache_detail *cd)
 811 {
 812         struct cache_reader *rp = filp->private_data;
 813         struct cache_request *rq;
 814         struct inode *inode = file_inode(filp);
 815         int err;
 816
 817         if (count == 0)
 818                 return 0;
 819
 820         inode_lock(inode); /* protect against multiple concurrent
 821                               * readers on this file */
 822  again:
 823         spin_lock(&queue_lock);
 824         /* need to find next request */
 825         while (rp->q.list.next != &cd->queue &&
 826                list_entry(rp->q.list.next, struct cache_queue, list)
 827                ->reader) {
 828                 struct list_head *next = rp->q.list.next;
 829                 list_move(&rp->q.list, next);
 830         }
 831         if (rp->q.list.next == &cd->queue) {
 832                 spin_unlock(&queue_lock);
 833                 inode_unlock(inode);
 834                 WARN_ON_ONCE(rp->offset);
 835                 return 0;
 836         }
 837         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 838         WARN_ON_ONCE(rq->q.reader);
 839         if (rp->offset == 0)
 840                 rq->readers++;
 841         spin_unlock(&queue_lock);
 842
 843         if (rq->len == 0) {
 844                 err = cache_request(cd, rq);
 845                 if (err < 0)
 846                         goto out;
 847                 rq->len = err;
 848         }
 849
 850         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 851                 err = -EAGAIN;
 852                 spin_lock(&queue_lock);
 853                 list_move(&rp->q.list, &rq->q.list);
 854                 spin_unlock(&queue_lock);
 855         } else {
 856                 if (rp->offset + count > rq->len)
 857                         count = rq->len - rp->offset;
 858                 err = -EFAULT;
 859                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 860                         goto out;
 861                 rp->offset += count;
 862                 if (rp->offset >= rq->len) {
 863                         rp->offset = 0;
 864                         spin_lock(&queue_lock);
 865                         list_move(&rp->q.list, &rq->q.list);
 866                         spin_unlock(&queue_lock);
 867                 }
 868                 err = 0;
 869         }
 870  out:
 871         if (rp->offset == 0) {
 872                 /* need to release rq */
 873                 spin_lock(&queue_lock);
 874                 rq->readers--;
 875                 if (rq->readers == 0 &&
 876                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 877                         list_del(&rq->q.list);
 878                         spin_unlock(&queue_lock);
 879                         cache_put(rq->item, cd);
 880                         kfree(rq->buf);
 881                         kfree(rq);
 882                 } else
 883                         spin_unlock(&queue_lock);
 884         }
 885         if (err == -EAGAIN)
 886                 goto again;
 887         inode_unlock(inode);
 888         return err ? err :  count;
 889 }
 890
 891 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 892                                  size_t count, struct cache_detail *cd)
 893 {
 894         ssize_t ret;
 895
 896         if (count == 0)
 897                 return -EINVAL;
 898         if (copy_from_user(kaddr, buf, count))
 899                 return -EFAULT;
 900         kaddr[count] = '\0';
 901         ret = cd->cache_parse(cd, kaddr, count);
 902         if (!ret)
 903                 ret = count;
 904         return ret;
 905 }
 906
 907 static ssize_t cache_slow_downcall(const char __user *buf,
 908                                    size_t count, struct cache_detail *cd)
 909 {
 910         static char write_buf[8192]; /* protected by queue_io_mutex */
 911         ssize_t ret = -EINVAL;
 912
 913         if (count >= sizeof(write_buf))
 914                 goto out;
 915         mutex_lock(&queue_io_mutex);
 916         ret = cache_do_downcall(write_buf, buf, count, cd);
 917         mutex_unlock(&queue_io_mutex);
 918 out:
 919         return ret;
 920 }
 921
 922 static ssize_t cache_downcall(struct address_space *mapping,
 923                               const char __user *buf,
 924                               size_t count, struct cache_detail *cd)
 925 {
 926         struct page *page;
 927         char *kaddr;
 928         ssize_t ret = -ENOMEM;
 929
 930         if (count >= PAGE_SIZE)
 931                 goto out_slow;
 932
 933         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 934         if (!page)
 935                 goto out_slow;
 936
 937         kaddr = kmap(page);
 938         ret = cache_do_downcall(kaddr, buf, count, cd);
 939         kunmap(page);
 940         unlock_page(page);
 941         put_page(page);
 942         return ret;
 943 out_slow:
 944         return cache_slow_downcall(buf, count, cd);
 945 }
 946
 947 static ssize_t cache_write(struct file *filp, const char __user *buf,
 948                            size_t count, loff_t *ppos,
 949                            struct cache_detail *cd)
 950 {
 951         struct address_space *mapping = filp->f_mapping;
 952         struct inode *inode = file_inode(filp);
 953         ssize_t ret = -EINVAL;
 954
 955         if (!cd->cache_parse)
 956                 goto out;
 957
 958         inode_lock(inode);
 959         ret = cache_downcall(mapping, buf, count, cd);
 960         inode_unlock(inode);
 961 out:
 962         return ret;
 963 }
 964
 965 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 966
 967 static __poll_t cache_poll(struct file *filp, poll_table *wait,
 968                                struct cache_detail *cd)
 969 {
 970         __poll_t mask;
 971         struct cache_reader *rp = filp->private_data;
 972         struct cache_queue *cq;
 973
 974         poll_wait(filp, &queue_wait, wait);
 975
 976         /* alway allow write */
 977         mask = EPOLLOUT | EPOLLWRNORM;
 978
 979         if (!rp)
 980                 return mask;
 981
 982         spin_lock(&queue_lock);
 983
 984         for (cq= &rp->q; &cq->list != &cd->queue;
 985              cq = list_entry(cq->list.next, struct cache_queue, list))
 986                 if (!cq->reader) {
 987                         mask |= EPOLLIN | EPOLLRDNORM;
 988                         break;
 989                 }
 990         spin_unlock(&queue_lock);
 991         return mask;
 992 }
 993
 994 static int cache_ioctl(struct inode *ino, struct file *filp,
 995                        unsigned int cmd, unsigned long arg,
 996                        struct cache_detail *cd)
 997 {
 998         int len = 0;
 999         struct cache_reader *rp = filp->private_data;
1000         struct cache_queue *cq;
1001
1002         if (cmd != FIONREAD || !rp)
1003                 return -EINVAL;
1004
1005         spin_lock(&queue_lock);
1006
1007         /* only find the length remaining in current request,
1008          * or the length of the next request
1009          */
1010         for (cq= &rp->q; &cq->list != &cd->queue;
1011              cq = list_entry(cq->list.next, struct cache_queue, list))
1012                 if (!cq->reader) {
1013                         struct cache_request *cr =
1014                                 container_of(cq, struct cache_request, q);
1015                         len = cr->len - rp->offset;
1016                         break;
1017                 }
1018         spin_unlock(&queue_lock);
1019
1020         return put_user(len, (int __user *)arg);
1021 }
1022
1023 static int cache_open(struct inode *inode, struct file *filp,
1024                       struct cache_detail *cd)
1025 {
1026         struct cache_reader *rp = NULL;
1027
1028         if (!cd || !try_module_get(cd->owner))
1029                 return -EACCES;
1030         nonseekable_open(inode, filp);
1031         if (filp->f_mode & FMODE_READ) {
1032                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1033                 if (!rp) {
1034                         module_put(cd->owner);
1035                         return -ENOMEM;
1036                 }
1037                 rp->offset = 0;
1038                 rp->q.reader = 1;
1039
1040                 spin_lock(&queue_lock);
1041                 list_add(&rp->q.list, &cd->queue);
1042                 spin_unlock(&queue_lock);
1043         }
1044         if (filp->f_mode & FMODE_WRITE)
1045                 atomic_inc(&cd->writers);
1046         filp->private_data = rp;
1047         return 0;
1048 }
1049
1050 static int cache_release(struct inode *inode, struct file *filp,
1051                          struct cache_detail *cd)
1052 {
1053         struct cache_reader *rp = filp->private_data;
1054
1055         if (rp) {
1056                 spin_lock(&queue_lock);
1057                 if (rp->offset) {
1058                         struct cache_queue *cq;
1059                         for (cq= &rp->q; &cq->list != &cd->queue;
1060                              cq = list_entry(cq->list.next, struct cache_queue, list))
1061                                 if (!cq->reader) {
1062                                         container_of(cq, struct cache_request, q)
1063                                                 ->readers--;
1064                                         break;
1065                                 }
1066                         rp->offset = 0;
1067                 }
1068                 list_del(&rp->q.list);
1069                 spin_unlock(&queue_lock);
1070
1071                 filp->private_data = NULL;
1072                 kfree(rp);
1073
1074         }
1075         if (filp->f_mode & FMODE_WRITE) {
1076                 atomic_dec(&cd->writers);
1077                 cd->last_close = seconds_since_boot();
1078         }
1079         module_put(cd->owner);
1080         return 0;
1081 }
1082
1083
1084
1085 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1086 {
1087         struct cache_queue *cq, *tmp;
1088         struct cache_request *cr;
1089         struct list_head dequeued;
1090
1091         INIT_LIST_HEAD(&dequeued);
1092         spin_lock(&queue_lock);
1093         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1094                 if (!cq->reader) {
1095                         cr = container_of(cq, struct cache_request, q);
1096                         if (cr->item != ch)
1097                                 continue;
1098                         if (test_bit(CACHE_PENDING, &ch->flags))
1099                                 /* Lost a race and it is pending again */
1100                                 break;
1101                         if (cr->readers != 0)
1102                                 continue;
1103                         list_move(&cr->q.list, &dequeued);
1104                 }
1105         spin_unlock(&queue_lock);
1106         while (!list_empty(&dequeued)) {
1107                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1108                 list_del(&cr->q.list);
1109                 cache_put(cr->item, detail);
1110                 kfree(cr->buf);
1111                 kfree(cr);
1112         }
1113 }
1114
1115 /*
1116  * Support routines for text-based upcalls.
1117  * Fields are separated by spaces.
1118  * Fields are either mangled to quote space tab newline slosh with slosh
1119  * or a hexified with a leading \x
1120  * Record is terminated with newline.
1121  *
1122  */
1123
1124 void qword_add(char **bpp, int *lp, char *str)
1125 {
1126         char *bp = *bpp;
1127         int len = *lp;
1128         int ret;
1129
1130         if (len < 0) return;
1131
1132         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1133         if (ret >= len) {
1134                 bp += len;
1135                 len = -1;
1136         } else {
1137                 bp += ret;
1138                 len -= ret;
1139                 *bp++ = ' ';
1140                 len--;
1141         }
1142         *bpp = bp;
1143         *lp = len;
1144 }
1145 EXPORT_SYMBOL_GPL(qword_add);
1146
1147 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1148 {
1149         char *bp = *bpp;
1150         int len = *lp;
1151
1152         if (len < 0) return;
1153
1154         if (len > 2) {
1155                 *bp++ = '\\';
1156                 *bp++ = 'x';
1157                 len -= 2;
1158                 while (blen && len >= 2) {
1159                         bp = hex_byte_pack(bp, *buf++);
1160                         len -= 2;
1161                         blen--;
1162                 }
1163         }
1164         if (blen || len<1) len = -1;
1165         else {
1166                 *bp++ = ' ';
1167                 len--;
1168         }
1169         *bpp = bp;
1170         *lp = len;
1171 }
1172 EXPORT_SYMBOL_GPL(qword_addhex);
1173
1174 static void warn_no_listener(struct cache_detail *detail)
1175 {
1176         if (detail->last_warn != detail->last_close) {
1177                 detail->last_warn = detail->last_close;
1178                 if (detail->warn_no_listener)
1179                         detail->warn_no_listener(detail, detail->last_close != 0);
1180         }
1181 }
1182
1183 static bool cache_listeners_exist(struct cache_detail *detail)
1184 {
1185         if (atomic_read(&detail->writers))
1186                 return true;
1187         if (detail->last_close == 0)
1188                 /* This cache was never opened */
1189                 return false;
1190         if (detail->last_close < seconds_since_boot() - 30)
1191                 /*
1192                  * We allow for the possibility that someone might
1193                  * restart a userspace daemon without restarting the
1194                  * server; but after 30 seconds, we give up.
1195                  */
1196                  return false;
1197         return true;
1198 }
1199
1200 /*
1201  * register an upcall request to user-space and queue it up for read() by the
1202  * upcall daemon.
1203  *
1204  * Each request is at most one page long.
1205  */
1206 static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1207 {
1208         char *buf;
1209         struct cache_request *crq;
1210         int ret = 0;
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                 trace_cache_entry_upcall(detail, h);
1235         } else
1236                 /* Lost a race, no longer PENDING, so don't enqueue */
1237                 ret = -EAGAIN;
1238         spin_unlock(&queue_lock);
1239         wake_up(&queue_wait);
1240         if (ret == -EAGAIN) {
1241                 kfree(buf);
1242                 kfree(crq);
1243         }
1244         return ret;
1245 }
1246
1247 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1248 {
1249         if (test_and_set_bit(CACHE_PENDING, &h->flags))
1250                 return 0;
1251         return cache_pipe_upcall(detail, h);
1252 }
1253 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1254
1255 int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1256                                      struct cache_head *h)
1257 {
1258         if (!cache_listeners_exist(detail)) {
1259                 warn_no_listener(detail);
1260                 trace_cache_entry_no_listener(detail, h);
1261                 return -EINVAL;
1262         }
1263         return sunrpc_cache_pipe_upcall(detail, h);
1264 }
1265 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1266
1267 /*
1268  * parse a message from user-space and pass it
1269  * to an appropriate cache
1270  * Messages are, like requests, separated into fields by
1271  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1272  *
1273  * Message is
1274  *   reply cachename expiry key ... content....
1275  *
1276  * key and content are both parsed by cache
1277  */
1278
1279 int qword_get(char **bpp, char *dest, int bufsize)
1280 {
1281         /* return bytes copied, or -1 on error */
1282         char *bp = *bpp;
1283         int len = 0;
1284
1285         while (*bp == ' ') bp++;
1286
1287         if (bp[0] == '\\' && bp[1] == 'x') {
1288                 /* HEX STRING */
1289                 bp += 2;
1290                 while (len < bufsize - 1) {
1291                         int h, l;
1292
1293                         h = hex_to_bin(bp[0]);
1294                         if (h < 0)
1295                                 break;
1296
1297                         l = hex_to_bin(bp[1]);
1298                         if (l < 0)
1299                                 break;
1300
1301                         *dest++ = (h << 4) | l;
1302                         bp += 2;
1303                         len++;
1304                 }
1305         } else {
1306                 /* text with \nnn octal quoting */
1307                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1308                         if (*bp == '\\' &&
1309                             isodigit(bp[1]) && (bp[1] <= '3') &&
1310                             isodigit(bp[2]) &&
1311                             isodigit(bp[3])) {
1312                                 int byte = (*++bp -'0');
1313                                 bp++;
1314                                 byte = (byte << 3) | (*bp++ - '0');
1315                                 byte = (byte << 3) | (*bp++ - '0');
1316                                 *dest++ = byte;
1317                                 len++;
1318                         } else {
1319                                 *dest++ = *bp++;
1320                                 len++;
1321                         }
1322                 }
1323         }
1324
1325         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1326                 return -1;
1327         while (*bp == ' ') bp++;
1328         *bpp = bp;
1329         *dest = '\0';
1330         return len;
1331 }
1332 EXPORT_SYMBOL_GPL(qword_get);
1333
1334
1335 /*
1336  * support /proc/net/rpc/$CACHENAME/content
1337  * as a seqfile.
1338  * We call ->cache_show passing NULL for the item to
1339  * get a header, then pass each real item in the cache
1340  */
1341
1342 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1343 {
1344         loff_t n = *pos;
1345         unsigned int hash, entry;
1346         struct cache_head *ch;
1347         struct cache_detail *cd = m->private;
1348
1349         if (!n--)
1350                 return SEQ_START_TOKEN;
1351         hash = n >> 32;
1352         entry = n & ((1LL<<32) - 1);
1353
1354         hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1355                 if (!entry--)
1356                         return ch;
1357         n &= ~((1LL<<32) - 1);
1358         do {
1359                 hash++;
1360                 n += 1LL<<32;
1361         } while(hash < cd->hash_size &&
1362                 hlist_empty(&cd->hash_table[hash]));
1363         if (hash >= cd->hash_size)
1364                 return NULL;
1365         *pos = n+1;
1366         return hlist_entry_safe(rcu_dereference_raw(
1367                                 hlist_first_rcu(&cd->hash_table[hash])),
1368                                 struct cache_head, cache_list);
1369 }
1370
1371 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1372 {
1373         struct cache_head *ch = p;
1374         int hash = (*pos >> 32);
1375         struct cache_detail *cd = m->private;
1376
1377         if (p == SEQ_START_TOKEN)
1378                 hash = 0;
1379         else if (ch->cache_list.next == NULL) {
1380                 hash++;
1381                 *pos += 1LL<<32;
1382         } else {
1383                 ++*pos;
1384                 return hlist_entry_safe(rcu_dereference_raw(
1385                                         hlist_next_rcu(&ch->cache_list)),
1386                                         struct cache_head, cache_list);
1387         }
1388         *pos &= ~((1LL<<32) - 1);
1389         while (hash < cd->hash_size &&
1390                hlist_empty(&cd->hash_table[hash])) {
1391                 hash++;
1392                 *pos += 1LL<<32;
1393         }
1394         if (hash >= cd->hash_size)
1395                 return NULL;
1396         ++*pos;
1397         return hlist_entry_safe(rcu_dereference_raw(
1398                                 hlist_first_rcu(&cd->hash_table[hash])),
1399                                 struct cache_head, cache_list);
1400 }
1401
1402 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1403         __acquires(RCU)
1404 {
1405         rcu_read_lock();
1406         return __cache_seq_start(m, pos);
1407 }
1408 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1409
1410 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1411 {
1412         return cache_seq_next(file, p, pos);
1413 }
1414 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1415
1416 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1417         __releases(RCU)
1418 {
1419         rcu_read_unlock();
1420 }
1421 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1422
1423 static int c_show(struct seq_file *m, void *p)
1424 {
1425         struct cache_head *cp = p;
1426         struct cache_detail *cd = m->private;
1427
1428         if (p == SEQ_START_TOKEN)
1429                 return cd->cache_show(m, cd, NULL);
1430
1431         ifdebug(CACHE)
1432                 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1433                            convert_to_wallclock(cp->expiry_time),
1434                            kref_read(&cp->ref), cp->flags);
1435         cache_get(cp);
1436         if (cache_check(cd, cp, NULL))
1437                 /* cache_check does a cache_put on failure */
1438                 seq_printf(m, "# ");
1439         else {
1440                 if (cache_is_expired(cd, cp))
1441                         seq_printf(m, "# ");
1442                 cache_put(cp, cd);
1443         }
1444
1445         return cd->cache_show(m, cd, cp);
1446 }
1447
1448 static const struct seq_operations cache_content_op = {
1449         .start  = cache_seq_start_rcu,
1450         .next   = cache_seq_next_rcu,
1451         .stop   = cache_seq_stop_rcu,
1452         .show   = c_show,
1453 };
1454
1455 static int content_open(struct inode *inode, struct file *file,
1456                         struct cache_detail *cd)
1457 {
1458         struct seq_file *seq;
1459         int err;
1460
1461         if (!cd || !try_module_get(cd->owner))
1462                 return -EACCES;
1463
1464         err = seq_open(file, &cache_content_op);
1465         if (err) {
1466                 module_put(cd->owner);
1467                 return err;
1468         }
1469
1470         seq = file->private_data;
1471         seq->private = cd;
1472         return 0;
1473 }
1474
1475 static int content_release(struct inode *inode, struct file *file,
1476                 struct cache_detail *cd)
1477 {
1478         int ret = seq_release(inode, file);
1479         module_put(cd->owner);
1480         return ret;
1481 }
1482
1483 static int open_flush(struct inode *inode, struct file *file,
1484                         struct cache_detail *cd)
1485 {
1486         if (!cd || !try_module_get(cd->owner))
1487                 return -EACCES;
1488         return nonseekable_open(inode, file);
1489 }
1490
1491 static int release_flush(struct inode *inode, struct file *file,
1492                         struct cache_detail *cd)
1493 {
1494         module_put(cd->owner);
1495         return 0;
1496 }
1497
1498 static ssize_t read_flush(struct file *file, char __user *buf,
1499                           size_t count, loff_t *ppos,
1500                           struct cache_detail *cd)
1501 {
1502         char tbuf[22];
1503         size_t len;
1504
1505         len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1506                         convert_to_wallclock(cd->flush_time));
1507         return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1508 }
1509
1510 static ssize_t write_flush(struct file *file, const char __user *buf,
1511                            size_t count, loff_t *ppos,
1512                            struct cache_detail *cd)
1513 {
1514         char tbuf[20];
1515         char *ep;
1516         time64_t now;
1517
1518         if (*ppos || count > sizeof(tbuf)-1)
1519                 return -EINVAL;
1520         if (copy_from_user(tbuf, buf, count))
1521                 return -EFAULT;
1522         tbuf[count] = 0;
1523         simple_strtoul(tbuf, &ep, 0);
1524         if (*ep && *ep != '\n')
1525                 return -EINVAL;
1526         /* Note that while we check that 'buf' holds a valid number,
1527          * we always ignore the value and just flush everything.
1528          * Making use of the number leads to races.
1529          */
1530
1531         now = seconds_since_boot();
1532         /* Always flush everything, so behave like cache_purge()
1533          * Do this by advancing flush_time to the current time,
1534          * or by one second if it has already reached the current time.
1535          * Newly added cache entries will always have ->last_refresh greater
1536          * that ->flush_time, so they don't get flushed prematurely.
1537          */
1538
1539         if (cd->flush_time >= now)
1540                 now = cd->flush_time + 1;
1541
1542         cd->flush_time = now;
1543         cd->nextcheck = now;
1544         cache_flush();
1545
1546         if (cd->flush)
1547                 cd->flush();
1548
1549         *ppos += count;
1550         return count;
1551 }
1552
1553 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1554                                  size_t count, loff_t *ppos)
1555 {
1556         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1557
1558         return cache_read(filp, buf, count, ppos, cd);
1559 }
1560
1561 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1562                                   size_t count, loff_t *ppos)
1563 {
1564         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1565
1566         return cache_write(filp, buf, count, ppos, cd);
1567 }
1568
1569 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1570 {
1571         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1572
1573         return cache_poll(filp, wait, cd);
1574 }
1575
1576 static long cache_ioctl_procfs(struct file *filp,
1577                                unsigned int cmd, unsigned long arg)
1578 {
1579         struct inode *inode = file_inode(filp);
1580         struct cache_detail *cd = PDE_DATA(inode);
1581
1582         return cache_ioctl(inode, filp, cmd, arg, cd);
1583 }
1584
1585 static int cache_open_procfs(struct inode *inode, struct file *filp)
1586 {
1587         struct cache_detail *cd = PDE_DATA(inode);
1588
1589         return cache_open(inode, filp, cd);
1590 }
1591
1592 static int cache_release_procfs(struct inode *inode, struct file *filp)
1593 {
1594         struct cache_detail *cd = PDE_DATA(inode);
1595
1596         return cache_release(inode, filp, cd);
1597 }
1598
1599 static const struct proc_ops cache_channel_proc_ops = {
1600         .proc_lseek     = no_llseek,
1601         .proc_read      = cache_read_procfs,
1602         .proc_write     = cache_write_procfs,
1603         .proc_poll      = cache_poll_procfs,
1604         .proc_ioctl     = cache_ioctl_procfs, /* for FIONREAD */
1605         .proc_open      = cache_open_procfs,
1606         .proc_release   = cache_release_procfs,
1607 };
1608
1609 static int content_open_procfs(struct inode *inode, struct file *filp)
1610 {
1611         struct cache_detail *cd = PDE_DATA(inode);
1612
1613         return content_open(inode, filp, cd);
1614 }
1615
1616 static int content_release_procfs(struct inode *inode, struct file *filp)
1617 {
1618         struct cache_detail *cd = PDE_DATA(inode);
1619
1620         return content_release(inode, filp, cd);
1621 }
1622
1623 static const struct proc_ops content_proc_ops = {
1624         .proc_open      = content_open_procfs,
1625         .proc_read      = seq_read,
1626         .proc_lseek     = seq_lseek,
1627         .proc_release   = content_release_procfs,
1628 };
1629
1630 static int open_flush_procfs(struct inode *inode, struct file *filp)
1631 {
1632         struct cache_detail *cd = PDE_DATA(inode);
1633
1634         return open_flush(inode, filp, cd);
1635 }
1636
1637 static int release_flush_procfs(struct inode *inode, struct file *filp)
1638 {
1639         struct cache_detail *cd = PDE_DATA(inode);
1640
1641         return release_flush(inode, filp, cd);
1642 }
1643
1644 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1645                             size_t count, loff_t *ppos)
1646 {
1647         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1648
1649         return read_flush(filp, buf, count, ppos, cd);
1650 }
1651
1652 static ssize_t write_flush_procfs(struct file *filp,
1653                                   const char __user *buf,
1654                                   size_t count, loff_t *ppos)
1655 {
1656         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1657
1658         return write_flush(filp, buf, count, ppos, cd);
1659 }
1660
1661 static const struct proc_ops cache_flush_proc_ops = {
1662         .proc_open      = open_flush_procfs,
1663         .proc_read      = read_flush_procfs,
1664         .proc_write     = write_flush_procfs,
1665         .proc_release   = release_flush_procfs,
1666         .proc_lseek     = no_llseek,
1667 };
1668
1669 static void remove_cache_proc_entries(struct cache_detail *cd)
1670 {
1671         if (cd->procfs) {
1672                 proc_remove(cd->procfs);
1673                 cd->procfs = NULL;
1674         }
1675 }
1676
1677 #ifdef CONFIG_PROC_FS
1678 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1679 {
1680         struct proc_dir_entry *p;
1681         struct sunrpc_net *sn;
1682
1683         sn = net_generic(net, sunrpc_net_id);
1684         cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1685         if (cd->procfs == NULL)
1686                 goto out_nomem;
1687
1688         p = proc_create_data("flush", S_IFREG | 0600,
1689                              cd->procfs, &cache_flush_proc_ops, cd);
1690         if (p == NULL)
1691                 goto out_nomem;
1692
1693         if (cd->cache_request || cd->cache_parse) {
1694                 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1695                                      &cache_channel_proc_ops, cd);
1696                 if (p == NULL)
1697                         goto out_nomem;
1698         }
1699         if (cd->cache_show) {
1700                 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1701                                      &content_proc_ops, cd);
1702                 if (p == NULL)
1703                         goto out_nomem;
1704         }
1705         return 0;
1706 out_nomem:
1707         remove_cache_proc_entries(cd);
1708         return -ENOMEM;
1709 }
1710 #else /* CONFIG_PROC_FS */
1711 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1712 {
1713         return 0;
1714 }
1715 #endif
1716
1717 void __init cache_initialize(void)
1718 {
1719         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1720 }
1721
1722 int cache_register_net(struct cache_detail *cd, struct net *net)
1723 {
1724         int ret;
1725
1726         sunrpc_init_cache_detail(cd);
1727         ret = create_cache_proc_entries(cd, net);
1728         if (ret)
1729                 sunrpc_destroy_cache_detail(cd);
1730         return ret;
1731 }
1732 EXPORT_SYMBOL_GPL(cache_register_net);
1733
1734 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1735 {
1736         remove_cache_proc_entries(cd);
1737         sunrpc_destroy_cache_detail(cd);
1738 }
1739 EXPORT_SYMBOL_GPL(cache_unregister_net);
1740
1741 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1742 {
1743         struct cache_detail *cd;
1744         int i;
1745
1746         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1747         if (cd == NULL)
1748                 return ERR_PTR(-ENOMEM);
1749
1750         cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1751                                  GFP_KERNEL);
1752         if (cd->hash_table == NULL) {
1753                 kfree(cd);
1754                 return ERR_PTR(-ENOMEM);
1755         }
1756
1757         for (i = 0; i < cd->hash_size; i++)
1758                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1759         cd->net = net;
1760         return cd;
1761 }
1762 EXPORT_SYMBOL_GPL(cache_create_net);
1763
1764 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1765 {
1766         kfree(cd->hash_table);
1767         kfree(cd);
1768 }
1769 EXPORT_SYMBOL_GPL(cache_destroy_net);
1770
1771 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1772                                  size_t count, loff_t *ppos)
1773 {
1774         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1775
1776         return cache_read(filp, buf, count, ppos, cd);
1777 }
1778
1779 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1780                                   size_t count, loff_t *ppos)
1781 {
1782         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1783
1784         return cache_write(filp, buf, count, ppos, cd);
1785 }
1786
1787 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1788 {
1789         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1790
1791         return cache_poll(filp, wait, cd);
1792 }
1793
1794 static long cache_ioctl_pipefs(struct file *filp,
1795                               unsigned int cmd, unsigned long arg)
1796 {
1797         struct inode *inode = file_inode(filp);
1798         struct cache_detail *cd = RPC_I(inode)->private;
1799
1800         return cache_ioctl(inode, filp, cmd, arg, cd);
1801 }
1802
1803 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1804 {
1805         struct cache_detail *cd = RPC_I(inode)->private;
1806
1807         return cache_open(inode, filp, cd);
1808 }
1809
1810 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1811 {
1812         struct cache_detail *cd = RPC_I(inode)->private;
1813
1814         return cache_release(inode, filp, cd);
1815 }
1816
1817 const struct file_operations cache_file_operations_pipefs = {
1818         .owner          = THIS_MODULE,
1819         .llseek         = no_llseek,
1820         .read           = cache_read_pipefs,
1821         .write          = cache_write_pipefs,
1822         .poll           = cache_poll_pipefs,
1823         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1824         .open           = cache_open_pipefs,
1825         .release        = cache_release_pipefs,
1826 };
1827
1828 static int content_open_pipefs(struct inode *inode, struct file *filp)
1829 {
1830         struct cache_detail *cd = RPC_I(inode)->private;
1831
1832         return content_open(inode, filp, cd);
1833 }
1834
1835 static int content_release_pipefs(struct inode *inode, struct file *filp)
1836 {
1837         struct cache_detail *cd = RPC_I(inode)->private;
1838
1839         return content_release(inode, filp, cd);
1840 }
1841
1842 const struct file_operations content_file_operations_pipefs = {
1843         .open           = content_open_pipefs,
1844         .read           = seq_read,
1845         .llseek         = seq_lseek,
1846         .release        = content_release_pipefs,
1847 };
1848
1849 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1850 {
1851         struct cache_detail *cd = RPC_I(inode)->private;
1852
1853         return open_flush(inode, filp, cd);
1854 }
1855
1856 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1857 {
1858         struct cache_detail *cd = RPC_I(inode)->private;
1859
1860         return release_flush(inode, filp, cd);
1861 }
1862
1863 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1864                             size_t count, loff_t *ppos)
1865 {
1866         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1867
1868         return read_flush(filp, buf, count, ppos, cd);
1869 }
1870
1871 static ssize_t write_flush_pipefs(struct file *filp,
1872                                   const char __user *buf,
1873                                   size_t count, loff_t *ppos)
1874 {
1875         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1876
1877         return write_flush(filp, buf, count, ppos, cd);
1878 }
1879
1880 const struct file_operations cache_flush_operations_pipefs = {
1881         .open           = open_flush_pipefs,
1882         .read           = read_flush_pipefs,
1883         .write          = write_flush_pipefs,
1884         .release        = release_flush_pipefs,
1885         .llseek         = no_llseek,
1886 };
1887
1888 int sunrpc_cache_register_pipefs(struct dentry *parent,
1889                                  const char *name, umode_t umode,
1890                                  struct cache_detail *cd)
1891 {
1892         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1893         if (IS_ERR(dir))
1894                 return PTR_ERR(dir);
1895         cd->pipefs = dir;
1896         return 0;
1897 }
1898 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1899
1900 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1901 {
1902         if (cd->pipefs) {
1903                 rpc_remove_cache_dir(cd->pipefs);
1904                 cd->pipefs = NULL;
1905         }
1906 }
1907 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1908
1909 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1910 {
1911         spin_lock(&cd->hash_lock);
1912         if (!hlist_unhashed(&h->cache_list)){
1913                 sunrpc_begin_cache_remove_entry(h, cd);
1914                 spin_unlock(&cd->hash_lock);
1915                 sunrpc_end_cache_remove_entry(h, cd);
1916         } else
1917                 spin_unlock(&cd->hash_lock);
1918 }
1919 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);