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;
 502
 503         if (list_empty(&cache_list))
 504                 return;
 505
 506         if (cache_clean() == -1)
 507                 delay = round_jiffies_relative(30*HZ);
 508         else
 509                 delay = 5;
 510
 511         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, delay);
 512 }
 513
 514
 515 /*
 516  * Clean all caches promptly.  This just calls cache_clean
 517  * repeatedly until we are sure that every cache has had a chance to
 518  * be fully cleaned
 519  */
 520 void cache_flush(void)
 521 {
 522         while (cache_clean() != -1)
 523                 cond_resched();
 524         while (cache_clean() != -1)
 525                 cond_resched();
 526 }
 527 EXPORT_SYMBOL_GPL(cache_flush);
 528
 529 void cache_purge(struct cache_detail *detail)
 530 {
 531         struct cache_head *ch = NULL;
 532         struct hlist_head *head = 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                 while (!hlist_empty(head)) {
 545                         ch = hlist_entry(head->first, struct cache_head,
 546                                          cache_list);
 547                         sunrpc_begin_cache_remove_entry(ch, detail);
 548                         spin_unlock(&detail->hash_lock);
 549                         sunrpc_end_cache_remove_entry(ch, detail);
 550                         spin_lock(&detail->hash_lock);
 551                 }
 552         }
 553         spin_unlock(&detail->hash_lock);
 554 }
 555 EXPORT_SYMBOL_GPL(cache_purge);
 556
 557
 558 /*
 559  * Deferral and Revisiting of Requests.
 560  *
 561  * If a cache lookup finds a pending entry, we
 562  * need to defer the request and revisit it later.
 563  * All deferred requests are stored in a hash table,
 564  * indexed by "struct cache_head *".
 565  * As it may be wasteful to store a whole request
 566  * structure, we allow the request to provide a
 567  * deferred form, which must contain a
 568  * 'struct cache_deferred_req'
 569  * This cache_deferred_req contains a method to allow
 570  * it to be revisited when cache info is available
 571  */
 572
 573 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 574 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 575
 576 #define DFR_MAX 300     /* ??? */
 577
 578 static DEFINE_SPINLOCK(cache_defer_lock);
 579 static LIST_HEAD(cache_defer_list);
 580 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 581 static int cache_defer_cnt;
 582
 583 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 584 {
 585         hlist_del_init(&dreq->hash);
 586         if (!list_empty(&dreq->recent)) {
 587                 list_del_init(&dreq->recent);
 588                 cache_defer_cnt--;
 589         }
 590 }
 591
 592 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 593 {
 594         int hash = DFR_HASH(item);
 595
 596         INIT_LIST_HEAD(&dreq->recent);
 597         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 598 }
 599
 600 static void setup_deferral(struct cache_deferred_req *dreq,
 601                            struct cache_head *item,
 602                            int count_me)
 603 {
 604
 605         dreq->item = item;
 606
 607         spin_lock(&cache_defer_lock);
 608
 609         __hash_deferred_req(dreq, item);
 610
 611         if (count_me) {
 612                 cache_defer_cnt++;
 613                 list_add(&dreq->recent, &cache_defer_list);
 614         }
 615
 616         spin_unlock(&cache_defer_lock);
 617
 618 }
 619
 620 struct thread_deferred_req {
 621         struct cache_deferred_req handle;
 622         struct completion completion;
 623 };
 624
 625 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 626 {
 627         struct thread_deferred_req *dr =
 628                 container_of(dreq, struct thread_deferred_req, handle);
 629         complete(&dr->completion);
 630 }
 631
 632 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 633 {
 634         struct thread_deferred_req sleeper;
 635         struct cache_deferred_req *dreq = &sleeper.handle;
 636
 637         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 638         dreq->revisit = cache_restart_thread;
 639
 640         setup_deferral(dreq, item, 0);
 641
 642         if (!test_bit(CACHE_PENDING, &item->flags) ||
 643             wait_for_completion_interruptible_timeout(
 644                     &sleeper.completion, req->thread_wait) <= 0) {
 645                 /* The completion wasn't completed, so we need
 646                  * to clean up
 647                  */
 648                 spin_lock(&cache_defer_lock);
 649                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 650                         __unhash_deferred_req(&sleeper.handle);
 651                         spin_unlock(&cache_defer_lock);
 652                 } else {
 653                         /* cache_revisit_request already removed
 654                          * this from the hash table, but hasn't
 655                          * called ->revisit yet.  It will very soon
 656                          * and we need to wait for it.
 657                          */
 658                         spin_unlock(&cache_defer_lock);
 659                         wait_for_completion(&sleeper.completion);
 660                 }
 661         }
 662 }
 663
 664 static void cache_limit_defers(void)
 665 {
 666         /* Make sure we haven't exceed the limit of allowed deferred
 667          * requests.
 668          */
 669         struct cache_deferred_req *discard = NULL;
 670
 671         if (cache_defer_cnt <= DFR_MAX)
 672                 return;
 673
 674         spin_lock(&cache_defer_lock);
 675
 676         /* Consider removing either the first or the last */
 677         if (cache_defer_cnt > DFR_MAX) {
 678                 if (prandom_u32() & 1)
 679                         discard = list_entry(cache_defer_list.next,
 680                                              struct cache_deferred_req, recent);
 681                 else
 682                         discard = list_entry(cache_defer_list.prev,
 683                                              struct cache_deferred_req, recent);
 684                 __unhash_deferred_req(discard);
 685         }
 686         spin_unlock(&cache_defer_lock);
 687         if (discard)
 688                 discard->revisit(discard, 1);
 689 }
 690
 691 /* Return true if and only if a deferred request is queued. */
 692 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 693 {
 694         struct cache_deferred_req *dreq;
 695
 696         if (req->thread_wait) {
 697                 cache_wait_req(req, item);
 698                 if (!test_bit(CACHE_PENDING, &item->flags))
 699                         return false;
 700         }
 701         dreq = req->defer(req);
 702         if (dreq == NULL)
 703                 return false;
 704         setup_deferral(dreq, item, 1);
 705         if (!test_bit(CACHE_PENDING, &item->flags))
 706                 /* Bit could have been cleared before we managed to
 707                  * set up the deferral, so need to revisit just in case
 708                  */
 709                 cache_revisit_request(item);
 710
 711         cache_limit_defers();
 712         return true;
 713 }
 714
 715 static void cache_revisit_request(struct cache_head *item)
 716 {
 717         struct cache_deferred_req *dreq;
 718         struct list_head pending;
 719         struct hlist_node *tmp;
 720         int hash = DFR_HASH(item);
 721
 722         INIT_LIST_HEAD(&pending);
 723         spin_lock(&cache_defer_lock);
 724
 725         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 726                 if (dreq->item == item) {
 727                         __unhash_deferred_req(dreq);
 728                         list_add(&dreq->recent, &pending);
 729                 }
 730
 731         spin_unlock(&cache_defer_lock);
 732
 733         while (!list_empty(&pending)) {
 734                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 735                 list_del_init(&dreq->recent);
 736                 dreq->revisit(dreq, 0);
 737         }
 738 }
 739
 740 void cache_clean_deferred(void *owner)
 741 {
 742         struct cache_deferred_req *dreq, *tmp;
 743         struct list_head pending;
 744
 745
 746         INIT_LIST_HEAD(&pending);
 747         spin_lock(&cache_defer_lock);
 748
 749         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 750                 if (dreq->owner == owner) {
 751                         __unhash_deferred_req(dreq);
 752                         list_add(&dreq->recent, &pending);
 753                 }
 754         }
 755         spin_unlock(&cache_defer_lock);
 756
 757         while (!list_empty(&pending)) {
 758                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 759                 list_del_init(&dreq->recent);
 760                 dreq->revisit(dreq, 1);
 761         }
 762 }
 763
 764 /*
 765  * communicate with user-space
 766  *
 767  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
 768  * On read, you get a full request, or block.
 769  * On write, an update request is processed.
 770  * Poll works if anything to read, and always allows write.
 771  *
 772  * Implemented by linked list of requests.  Each open file has
 773  * a ->private that also exists in this list.  New requests are added
 774  * to the end and may wakeup and preceding readers.
 775  * New readers are added to the head.  If, on read, an item is found with
 776  * CACHE_UPCALLING clear, we free it from the list.
 777  *
 778  */
 779
 780 static DEFINE_SPINLOCK(queue_lock);
 781 static DEFINE_MUTEX(queue_io_mutex);
 782
 783 struct cache_queue {
 784         struct list_head        list;
 785         int                     reader; /* if 0, then request */
 786 };
 787 struct cache_request {
 788         struct cache_queue      q;
 789         struct cache_head       *item;
 790         char                    * buf;
 791         int                     len;
 792         int                     readers;
 793 };
 794 struct cache_reader {
 795         struct cache_queue      q;
 796         int                     offset; /* if non-0, we have a refcnt on next request */
 797 };
 798
 799 static int cache_request(struct cache_detail *detail,
 800                                struct cache_request *crq)
 801 {
 802         char *bp = crq->buf;
 803         int len = PAGE_SIZE;
 804
 805         detail->cache_request(detail, crq->item, &bp, &len);
 806         if (len < 0)
 807                 return -EAGAIN;
 808         return PAGE_SIZE - len;
 809 }
 810
 811 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 812                           loff_t *ppos, struct cache_detail *cd)
 813 {
 814         struct cache_reader *rp = filp->private_data;
 815         struct cache_request *rq;
 816         struct inode *inode = file_inode(filp);
 817         int err;
 818
 819         if (count == 0)
 820                 return 0;
 821
 822         inode_lock(inode); /* protect against multiple concurrent
 823                               * readers on this file */
 824  again:
 825         spin_lock(&queue_lock);
 826         /* need to find next request */
 827         while (rp->q.list.next != &cd->queue &&
 828                list_entry(rp->q.list.next, struct cache_queue, list)
 829                ->reader) {
 830                 struct list_head *next = rp->q.list.next;
 831                 list_move(&rp->q.list, next);
 832         }
 833         if (rp->q.list.next == &cd->queue) {
 834                 spin_unlock(&queue_lock);
 835                 inode_unlock(inode);
 836                 WARN_ON_ONCE(rp->offset);
 837                 return 0;
 838         }
 839         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 840         WARN_ON_ONCE(rq->q.reader);
 841         if (rp->offset == 0)
 842                 rq->readers++;
 843         spin_unlock(&queue_lock);
 844
 845         if (rq->len == 0) {
 846                 err = cache_request(cd, rq);
 847                 if (err < 0)
 848                         goto out;
 849                 rq->len = err;
 850         }
 851
 852         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 853                 err = -EAGAIN;
 854                 spin_lock(&queue_lock);
 855                 list_move(&rp->q.list, &rq->q.list);
 856                 spin_unlock(&queue_lock);
 857         } else {
 858                 if (rp->offset + count > rq->len)
 859                         count = rq->len - rp->offset;
 860                 err = -EFAULT;
 861                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 862                         goto out;
 863                 rp->offset += count;
 864                 if (rp->offset >= rq->len) {
 865                         rp->offset = 0;
 866                         spin_lock(&queue_lock);
 867                         list_move(&rp->q.list, &rq->q.list);
 868                         spin_unlock(&queue_lock);
 869                 }
 870                 err = 0;
 871         }
 872  out:
 873         if (rp->offset == 0) {
 874                 /* need to release rq */
 875                 spin_lock(&queue_lock);
 876                 rq->readers--;
 877                 if (rq->readers == 0 &&
 878                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 879                         list_del(&rq->q.list);
 880                         spin_unlock(&queue_lock);
 881                         cache_put(rq->item, cd);
 882                         kfree(rq->buf);
 883                         kfree(rq);
 884                 } else
 885                         spin_unlock(&queue_lock);
 886         }
 887         if (err == -EAGAIN)
 888                 goto again;
 889         inode_unlock(inode);
 890         return err ? err :  count;
 891 }
 892
 893 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 894                                  size_t count, struct cache_detail *cd)
 895 {
 896         ssize_t ret;
 897
 898         if (count == 0)
 899                 return -EINVAL;
 900         if (copy_from_user(kaddr, buf, count))
 901                 return -EFAULT;
 902         kaddr[count] = '\0';
 903         ret = cd->cache_parse(cd, kaddr, count);
 904         if (!ret)
 905                 ret = count;
 906         return ret;
 907 }
 908
 909 static ssize_t cache_slow_downcall(const char __user *buf,
 910                                    size_t count, struct cache_detail *cd)
 911 {
 912         static char write_buf[32768]; /* protected by queue_io_mutex */
 913         ssize_t ret = -EINVAL;
 914
 915         if (count >= sizeof(write_buf))
 916                 goto out;
 917         mutex_lock(&queue_io_mutex);
 918         ret = cache_do_downcall(write_buf, buf, count, cd);
 919         mutex_unlock(&queue_io_mutex);
 920 out:
 921         return ret;
 922 }
 923
 924 static ssize_t cache_downcall(struct address_space *mapping,
 925                               const char __user *buf,
 926                               size_t count, struct cache_detail *cd)
 927 {
 928         struct page *page;
 929         char *kaddr;
 930         ssize_t ret = -ENOMEM;
 931
 932         if (count >= PAGE_SIZE)
 933                 goto out_slow;
 934
 935         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 936         if (!page)
 937                 goto out_slow;
 938
 939         kaddr = kmap(page);
 940         ret = cache_do_downcall(kaddr, buf, count, cd);
 941         kunmap(page);
 942         unlock_page(page);
 943         put_page(page);
 944         return ret;
 945 out_slow:
 946         return cache_slow_downcall(buf, count, cd);
 947 }
 948
 949 static ssize_t cache_write(struct file *filp, const char __user *buf,
 950                            size_t count, loff_t *ppos,
 951                            struct cache_detail *cd)
 952 {
 953         struct address_space *mapping = filp->f_mapping;
 954         struct inode *inode = file_inode(filp);
 955         ssize_t ret = -EINVAL;
 956
 957         if (!cd->cache_parse)
 958                 goto out;
 959
 960         inode_lock(inode);
 961         ret = cache_downcall(mapping, buf, count, cd);
 962         inode_unlock(inode);
 963 out:
 964         return ret;
 965 }
 966
 967 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 968
 969 static __poll_t cache_poll(struct file *filp, poll_table *wait,
 970                                struct cache_detail *cd)
 971 {
 972         __poll_t mask;
 973         struct cache_reader *rp = filp->private_data;
 974         struct cache_queue *cq;
 975
 976         poll_wait(filp, &queue_wait, wait);
 977
 978         /* alway allow write */
 979         mask = EPOLLOUT | EPOLLWRNORM;
 980
 981         if (!rp)
 982                 return mask;
 983
 984         spin_lock(&queue_lock);
 985
 986         for (cq= &rp->q; &cq->list != &cd->queue;
 987              cq = list_entry(cq->list.next, struct cache_queue, list))
 988                 if (!cq->reader) {
 989                         mask |= EPOLLIN | EPOLLRDNORM;
 990                         break;
 991                 }
 992         spin_unlock(&queue_lock);
 993         return mask;
 994 }
 995
 996 static int cache_ioctl(struct inode *ino, struct file *filp,
 997                        unsigned int cmd, unsigned long arg,
 998                        struct cache_detail *cd)
 999 {
1000         int len = 0;
1001         struct cache_reader *rp = filp->private_data;
1002         struct cache_queue *cq;
1003
1004         if (cmd != FIONREAD || !rp)
1005                 return -EINVAL;
1006
1007         spin_lock(&queue_lock);
1008
1009         /* only find the length remaining in current request,
1010          * or the length of the next request
1011          */
1012         for (cq= &rp->q; &cq->list != &cd->queue;
1013              cq = list_entry(cq->list.next, struct cache_queue, list))
1014                 if (!cq->reader) {
1015                         struct cache_request *cr =
1016                                 container_of(cq, struct cache_request, q);
1017                         len = cr->len - rp->offset;
1018                         break;
1019                 }
1020         spin_unlock(&queue_lock);
1021
1022         return put_user(len, (int __user *)arg);
1023 }
1024
1025 static int cache_open(struct inode *inode, struct file *filp,
1026                       struct cache_detail *cd)
1027 {
1028         struct cache_reader *rp = NULL;
1029
1030         if (!cd || !try_module_get(cd->owner))
1031                 return -EACCES;
1032         nonseekable_open(inode, filp);
1033         if (filp->f_mode & FMODE_READ) {
1034                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1035                 if (!rp) {
1036                         module_put(cd->owner);
1037                         return -ENOMEM;
1038                 }
1039                 rp->offset = 0;
1040                 rp->q.reader = 1;
1041
1042                 spin_lock(&queue_lock);
1043                 list_add(&rp->q.list, &cd->queue);
1044                 spin_unlock(&queue_lock);
1045         }
1046         if (filp->f_mode & FMODE_WRITE)
1047                 atomic_inc(&cd->writers);
1048         filp->private_data = rp;
1049         return 0;
1050 }
1051
1052 static int cache_release(struct inode *inode, struct file *filp,
1053                          struct cache_detail *cd)
1054 {
1055         struct cache_reader *rp = filp->private_data;
1056
1057         if (rp) {
1058                 spin_lock(&queue_lock);
1059                 if (rp->offset) {
1060                         struct cache_queue *cq;
1061                         for (cq= &rp->q; &cq->list != &cd->queue;
1062                              cq = list_entry(cq->list.next, struct cache_queue, list))
1063                                 if (!cq->reader) {
1064                                         container_of(cq, struct cache_request, q)
1065                                                 ->readers--;
1066                                         break;
1067                                 }
1068                         rp->offset = 0;
1069                 }
1070                 list_del(&rp->q.list);
1071                 spin_unlock(&queue_lock);
1072
1073                 filp->private_data = NULL;
1074                 kfree(rp);
1075
1076         }
1077         if (filp->f_mode & FMODE_WRITE) {
1078                 atomic_dec(&cd->writers);
1079                 cd->last_close = seconds_since_boot();
1080         }
1081         module_put(cd->owner);
1082         return 0;
1083 }
1084
1085
1086
1087 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1088 {
1089         struct cache_queue *cq, *tmp;
1090         struct cache_request *cr;
1091         struct list_head dequeued;
1092
1093         INIT_LIST_HEAD(&dequeued);
1094         spin_lock(&queue_lock);
1095         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1096                 if (!cq->reader) {
1097                         cr = container_of(cq, struct cache_request, q);
1098                         if (cr->item != ch)
1099                                 continue;
1100                         if (test_bit(CACHE_PENDING, &ch->flags))
1101                                 /* Lost a race and it is pending again */
1102                                 break;
1103                         if (cr->readers != 0)
1104                                 continue;
1105                         list_move(&cr->q.list, &dequeued);
1106                 }
1107         spin_unlock(&queue_lock);
1108         while (!list_empty(&dequeued)) {
1109                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1110                 list_del(&cr->q.list);
1111                 cache_put(cr->item, detail);
1112                 kfree(cr->buf);
1113                 kfree(cr);
1114         }
1115 }
1116
1117 /*
1118  * Support routines for text-based upcalls.
1119  * Fields are separated by spaces.
1120  * Fields are either mangled to quote space tab newline slosh with slosh
1121  * or a hexified with a leading \x
1122  * Record is terminated with newline.
1123  *
1124  */
1125
1126 void qword_add(char **bpp, int *lp, char *str)
1127 {
1128         char *bp = *bpp;
1129         int len = *lp;
1130         int ret;
1131
1132         if (len < 0) return;
1133
1134         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1135         if (ret >= len) {
1136                 bp += len;
1137                 len = -1;
1138         } else {
1139                 bp += ret;
1140                 len -= ret;
1141                 *bp++ = ' ';
1142                 len--;
1143         }
1144         *bpp = bp;
1145         *lp = len;
1146 }
1147 EXPORT_SYMBOL_GPL(qword_add);
1148
1149 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1150 {
1151         char *bp = *bpp;
1152         int len = *lp;
1153
1154         if (len < 0) return;
1155
1156         if (len > 2) {
1157                 *bp++ = '\\';
1158                 *bp++ = 'x';
1159                 len -= 2;
1160                 while (blen && len >= 2) {
1161                         bp = hex_byte_pack(bp, *buf++);
1162                         len -= 2;
1163                         blen--;
1164                 }
1165         }
1166         if (blen || len<1) len = -1;
1167         else {
1168                 *bp++ = ' ';
1169                 len--;
1170         }
1171         *bpp = bp;
1172         *lp = len;
1173 }
1174 EXPORT_SYMBOL_GPL(qword_addhex);
1175
1176 static void warn_no_listener(struct cache_detail *detail)
1177 {
1178         if (detail->last_warn != detail->last_close) {
1179                 detail->last_warn = detail->last_close;
1180                 if (detail->warn_no_listener)
1181                         detail->warn_no_listener(detail, detail->last_close != 0);
1182         }
1183 }
1184
1185 static bool cache_listeners_exist(struct cache_detail *detail)
1186 {
1187         if (atomic_read(&detail->writers))
1188                 return true;
1189         if (detail->last_close == 0)
1190                 /* This cache was never opened */
1191                 return false;
1192         if (detail->last_close < seconds_since_boot() - 30)
1193                 /*
1194                  * We allow for the possibility that someone might
1195                  * restart a userspace daemon without restarting the
1196                  * server; but after 30 seconds, we give up.
1197                  */
1198                  return false;
1199         return true;
1200 }
1201
1202 /*
1203  * register an upcall request to user-space and queue it up for read() by the
1204  * upcall daemon.
1205  *
1206  * Each request is at most one page long.
1207  */
1208 static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1209 {
1210         char *buf;
1211         struct cache_request *crq;
1212         int ret = 0;
1213
1214         if (test_bit(CACHE_CLEANED, &h->flags))
1215                 /* Too late to make an upcall */
1216                 return -EAGAIN;
1217
1218         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1219         if (!buf)
1220                 return -EAGAIN;
1221
1222         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1223         if (!crq) {
1224                 kfree(buf);
1225                 return -EAGAIN;
1226         }
1227
1228         crq->q.reader = 0;
1229         crq->buf = buf;
1230         crq->len = 0;
1231         crq->readers = 0;
1232         spin_lock(&queue_lock);
1233         if (test_bit(CACHE_PENDING, &h->flags)) {
1234                 crq->item = cache_get(h);
1235                 list_add_tail(&crq->q.list, &detail->queue);
1236                 trace_cache_entry_upcall(detail, h);
1237         } else
1238                 /* Lost a race, no longer PENDING, so don't enqueue */
1239                 ret = -EAGAIN;
1240         spin_unlock(&queue_lock);
1241         wake_up(&queue_wait);
1242         if (ret == -EAGAIN) {
1243                 kfree(buf);
1244                 kfree(crq);
1245         }
1246         return ret;
1247 }
1248
1249 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1250 {
1251         if (test_and_set_bit(CACHE_PENDING, &h->flags))
1252                 return 0;
1253         return cache_pipe_upcall(detail, h);
1254 }
1255 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1256
1257 int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1258                                      struct cache_head *h)
1259 {
1260         if (!cache_listeners_exist(detail)) {
1261                 warn_no_listener(detail);
1262                 trace_cache_entry_no_listener(detail, h);
1263                 return -EINVAL;
1264         }
1265         return sunrpc_cache_pipe_upcall(detail, h);
1266 }
1267 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1268
1269 /*
1270  * parse a message from user-space and pass it
1271  * to an appropriate cache
1272  * Messages are, like requests, separated into fields by
1273  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1274  *
1275  * Message is
1276  *   reply cachename expiry key ... content....
1277  *
1278  * key and content are both parsed by cache
1279  */
1280
1281 int qword_get(char **bpp, char *dest, int bufsize)
1282 {
1283         /* return bytes copied, or -1 on error */
1284         char *bp = *bpp;
1285         int len = 0;
1286
1287         while (*bp == ' ') bp++;
1288
1289         if (bp[0] == '\\' && bp[1] == 'x') {
1290                 /* HEX STRING */
1291                 bp += 2;
1292                 while (len < bufsize - 1) {
1293                         int h, l;
1294
1295                         h = hex_to_bin(bp[0]);
1296                         if (h < 0)
1297                                 break;
1298
1299                         l = hex_to_bin(bp[1]);
1300                         if (l < 0)
1301                                 break;
1302
1303                         *dest++ = (h << 4) | l;
1304                         bp += 2;
1305                         len++;
1306                 }
1307         } else {
1308                 /* text with \nnn octal quoting */
1309                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1310                         if (*bp == '\\' &&
1311                             isodigit(bp[1]) && (bp[1] <= '3') &&
1312                             isodigit(bp[2]) &&
1313                             isodigit(bp[3])) {
1314                                 int byte = (*++bp -'0');
1315                                 bp++;
1316                                 byte = (byte << 3) | (*bp++ - '0');
1317                                 byte = (byte << 3) | (*bp++ - '0');
1318                                 *dest++ = byte;
1319                                 len++;
1320                         } else {
1321                                 *dest++ = *bp++;
1322                                 len++;
1323                         }
1324                 }
1325         }
1326
1327         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1328                 return -1;
1329         while (*bp == ' ') bp++;
1330         *bpp = bp;
1331         *dest = '\0';
1332         return len;
1333 }
1334 EXPORT_SYMBOL_GPL(qword_get);
1335
1336
1337 /*
1338  * support /proc/net/rpc/$CACHENAME/content
1339  * as a seqfile.
1340  * We call ->cache_show passing NULL for the item to
1341  * get a header, then pass each real item in the cache
1342  */
1343
1344 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1345 {
1346         loff_t n = *pos;
1347         unsigned int hash, entry;
1348         struct cache_head *ch;
1349         struct cache_detail *cd = m->private;
1350
1351         if (!n--)
1352                 return SEQ_START_TOKEN;
1353         hash = n >> 32;
1354         entry = n & ((1LL<<32) - 1);
1355
1356         hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1357                 if (!entry--)
1358                         return ch;
1359         n &= ~((1LL<<32) - 1);
1360         do {
1361                 hash++;
1362                 n += 1LL<<32;
1363         } while(hash < cd->hash_size &&
1364                 hlist_empty(&cd->hash_table[hash]));
1365         if (hash >= cd->hash_size)
1366                 return NULL;
1367         *pos = n+1;
1368         return hlist_entry_safe(rcu_dereference_raw(
1369                                 hlist_first_rcu(&cd->hash_table[hash])),
1370                                 struct cache_head, cache_list);
1371 }
1372
1373 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1374 {
1375         struct cache_head *ch = p;
1376         int hash = (*pos >> 32);
1377         struct cache_detail *cd = m->private;
1378
1379         if (p == SEQ_START_TOKEN)
1380                 hash = 0;
1381         else if (ch->cache_list.next == NULL) {
1382                 hash++;
1383                 *pos += 1LL<<32;
1384         } else {
1385                 ++*pos;
1386                 return hlist_entry_safe(rcu_dereference_raw(
1387                                         hlist_next_rcu(&ch->cache_list)),
1388                                         struct cache_head, cache_list);
1389         }
1390         *pos &= ~((1LL<<32) - 1);
1391         while (hash < cd->hash_size &&
1392                hlist_empty(&cd->hash_table[hash])) {
1393                 hash++;
1394                 *pos += 1LL<<32;
1395         }
1396         if (hash >= cd->hash_size)
1397                 return NULL;
1398         ++*pos;
1399         return hlist_entry_safe(rcu_dereference_raw(
1400                                 hlist_first_rcu(&cd->hash_table[hash])),
1401                                 struct cache_head, cache_list);
1402 }
1403
1404 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1405         __acquires(RCU)
1406 {
1407         rcu_read_lock();
1408         return __cache_seq_start(m, pos);
1409 }
1410 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1411
1412 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1413 {
1414         return cache_seq_next(file, p, pos);
1415 }
1416 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1417
1418 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1419         __releases(RCU)
1420 {
1421         rcu_read_unlock();
1422 }
1423 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1424
1425 static int c_show(struct seq_file *m, void *p)
1426 {
1427         struct cache_head *cp = p;
1428         struct cache_detail *cd = m->private;
1429
1430         if (p == SEQ_START_TOKEN)
1431                 return cd->cache_show(m, cd, NULL);
1432
1433         ifdebug(CACHE)
1434                 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1435                            convert_to_wallclock(cp->expiry_time),
1436                            kref_read(&cp->ref), cp->flags);
1437         cache_get(cp);
1438         if (cache_check(cd, cp, NULL))
1439                 /* cache_check does a cache_put on failure */
1440                 seq_puts(m, "# ");
1441         else {
1442                 if (cache_is_expired(cd, cp))
1443                         seq_puts(m, "# ");
1444                 cache_put(cp, cd);
1445         }
1446
1447         return cd->cache_show(m, cd, cp);
1448 }
1449
1450 static const struct seq_operations cache_content_op = {
1451         .start  = cache_seq_start_rcu,
1452         .next   = cache_seq_next_rcu,
1453         .stop   = cache_seq_stop_rcu,
1454         .show   = c_show,
1455 };
1456
1457 static int content_open(struct inode *inode, struct file *file,
1458                         struct cache_detail *cd)
1459 {
1460         struct seq_file *seq;
1461         int err;
1462
1463         if (!cd || !try_module_get(cd->owner))
1464                 return -EACCES;
1465
1466         err = seq_open(file, &cache_content_op);
1467         if (err) {
1468                 module_put(cd->owner);
1469                 return err;
1470         }
1471
1472         seq = file->private_data;
1473         seq->private = cd;
1474         return 0;
1475 }
1476
1477 static int content_release(struct inode *inode, struct file *file,
1478                 struct cache_detail *cd)
1479 {
1480         int ret = seq_release(inode, file);
1481         module_put(cd->owner);
1482         return ret;
1483 }
1484
1485 static int open_flush(struct inode *inode, struct file *file,
1486                         struct cache_detail *cd)
1487 {
1488         if (!cd || !try_module_get(cd->owner))
1489                 return -EACCES;
1490         return nonseekable_open(inode, file);
1491 }
1492
1493 static int release_flush(struct inode *inode, struct file *file,
1494                         struct cache_detail *cd)
1495 {
1496         module_put(cd->owner);
1497         return 0;
1498 }
1499
1500 static ssize_t read_flush(struct file *file, char __user *buf,
1501                           size_t count, loff_t *ppos,
1502                           struct cache_detail *cd)
1503 {
1504         char tbuf[22];
1505         size_t len;
1506
1507         len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1508                         convert_to_wallclock(cd->flush_time));
1509         return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1510 }
1511
1512 static ssize_t write_flush(struct file *file, const char __user *buf,
1513                            size_t count, loff_t *ppos,
1514                            struct cache_detail *cd)
1515 {
1516         char tbuf[20];
1517         char *ep;
1518         time64_t now;
1519
1520         if (*ppos || count > sizeof(tbuf)-1)
1521                 return -EINVAL;
1522         if (copy_from_user(tbuf, buf, count))
1523                 return -EFAULT;
1524         tbuf[count] = 0;
1525         simple_strtoul(tbuf, &ep, 0);
1526         if (*ep && *ep != '\n')
1527                 return -EINVAL;
1528         /* Note that while we check that 'buf' holds a valid number,
1529          * we always ignore the value and just flush everything.
1530          * Making use of the number leads to races.
1531          */
1532
1533         now = seconds_since_boot();
1534         /* Always flush everything, so behave like cache_purge()
1535          * Do this by advancing flush_time to the current time,
1536          * or by one second if it has already reached the current time.
1537          * Newly added cache entries will always have ->last_refresh greater
1538          * that ->flush_time, so they don't get flushed prematurely.
1539          */
1540
1541         if (cd->flush_time >= now)
1542                 now = cd->flush_time + 1;
1543
1544         cd->flush_time = now;
1545         cd->nextcheck = now;
1546         cache_flush();
1547
1548         if (cd->flush)
1549                 cd->flush();
1550
1551         *ppos += count;
1552         return count;
1553 }
1554
1555 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1556                                  size_t count, loff_t *ppos)
1557 {
1558         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1559
1560         return cache_read(filp, buf, count, ppos, cd);
1561 }
1562
1563 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1564                                   size_t count, loff_t *ppos)
1565 {
1566         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1567
1568         return cache_write(filp, buf, count, ppos, cd);
1569 }
1570
1571 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1572 {
1573         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1574
1575         return cache_poll(filp, wait, cd);
1576 }
1577
1578 static long cache_ioctl_procfs(struct file *filp,
1579                                unsigned int cmd, unsigned long arg)
1580 {
1581         struct inode *inode = file_inode(filp);
1582         struct cache_detail *cd = PDE_DATA(inode);
1583
1584         return cache_ioctl(inode, filp, cmd, arg, cd);
1585 }
1586
1587 static int cache_open_procfs(struct inode *inode, struct file *filp)
1588 {
1589         struct cache_detail *cd = PDE_DATA(inode);
1590
1591         return cache_open(inode, filp, cd);
1592 }
1593
1594 static int cache_release_procfs(struct inode *inode, struct file *filp)
1595 {
1596         struct cache_detail *cd = PDE_DATA(inode);
1597
1598         return cache_release(inode, filp, cd);
1599 }
1600
1601 static const struct proc_ops cache_channel_proc_ops = {
1602         .proc_lseek     = no_llseek,
1603         .proc_read      = cache_read_procfs,
1604         .proc_write     = cache_write_procfs,
1605         .proc_poll      = cache_poll_procfs,
1606         .proc_ioctl     = cache_ioctl_procfs, /* for FIONREAD */
1607         .proc_open      = cache_open_procfs,
1608         .proc_release   = cache_release_procfs,
1609 };
1610
1611 static int content_open_procfs(struct inode *inode, struct file *filp)
1612 {
1613         struct cache_detail *cd = PDE_DATA(inode);
1614
1615         return content_open(inode, filp, cd);
1616 }
1617
1618 static int content_release_procfs(struct inode *inode, struct file *filp)
1619 {
1620         struct cache_detail *cd = PDE_DATA(inode);
1621
1622         return content_release(inode, filp, cd);
1623 }
1624
1625 static const struct proc_ops content_proc_ops = {
1626         .proc_open      = content_open_procfs,
1627         .proc_read      = seq_read,
1628         .proc_lseek     = seq_lseek,
1629         .proc_release   = content_release_procfs,
1630 };
1631
1632 static int open_flush_procfs(struct inode *inode, struct file *filp)
1633 {
1634         struct cache_detail *cd = PDE_DATA(inode);
1635
1636         return open_flush(inode, filp, cd);
1637 }
1638
1639 static int release_flush_procfs(struct inode *inode, struct file *filp)
1640 {
1641         struct cache_detail *cd = PDE_DATA(inode);
1642
1643         return release_flush(inode, filp, cd);
1644 }
1645
1646 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1647                             size_t count, loff_t *ppos)
1648 {
1649         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1650
1651         return read_flush(filp, buf, count, ppos, cd);
1652 }
1653
1654 static ssize_t write_flush_procfs(struct file *filp,
1655                                   const char __user *buf,
1656                                   size_t count, loff_t *ppos)
1657 {
1658         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1659
1660         return write_flush(filp, buf, count, ppos, cd);
1661 }
1662
1663 static const struct proc_ops cache_flush_proc_ops = {
1664         .proc_open      = open_flush_procfs,
1665         .proc_read      = read_flush_procfs,
1666         .proc_write     = write_flush_procfs,
1667         .proc_release   = release_flush_procfs,
1668         .proc_lseek     = no_llseek,
1669 };
1670
1671 static void remove_cache_proc_entries(struct cache_detail *cd)
1672 {
1673         if (cd->procfs) {
1674                 proc_remove(cd->procfs);
1675                 cd->procfs = NULL;
1676         }
1677 }
1678
1679 #ifdef CONFIG_PROC_FS
1680 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1681 {
1682         struct proc_dir_entry *p;
1683         struct sunrpc_net *sn;
1684
1685         sn = net_generic(net, sunrpc_net_id);
1686         cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1687         if (cd->procfs == NULL)
1688                 goto out_nomem;
1689
1690         p = proc_create_data("flush", S_IFREG | 0600,
1691                              cd->procfs, &cache_flush_proc_ops, cd);
1692         if (p == NULL)
1693                 goto out_nomem;
1694
1695         if (cd->cache_request || cd->cache_parse) {
1696                 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1697                                      &cache_channel_proc_ops, cd);
1698                 if (p == NULL)
1699                         goto out_nomem;
1700         }
1701         if (cd->cache_show) {
1702                 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1703                                      &content_proc_ops, cd);
1704                 if (p == NULL)
1705                         goto out_nomem;
1706         }
1707         return 0;
1708 out_nomem:
1709         remove_cache_proc_entries(cd);
1710         return -ENOMEM;
1711 }
1712 #else /* CONFIG_PROC_FS */
1713 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1714 {
1715         return 0;
1716 }
1717 #endif
1718
1719 void __init cache_initialize(void)
1720 {
1721         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1722 }
1723
1724 int cache_register_net(struct cache_detail *cd, struct net *net)
1725 {
1726         int ret;
1727
1728         sunrpc_init_cache_detail(cd);
1729         ret = create_cache_proc_entries(cd, net);
1730         if (ret)
1731                 sunrpc_destroy_cache_detail(cd);
1732         return ret;
1733 }
1734 EXPORT_SYMBOL_GPL(cache_register_net);
1735
1736 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1737 {
1738         remove_cache_proc_entries(cd);
1739         sunrpc_destroy_cache_detail(cd);
1740 }
1741 EXPORT_SYMBOL_GPL(cache_unregister_net);
1742
1743 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1744 {
1745         struct cache_detail *cd;
1746         int i;
1747
1748         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1749         if (cd == NULL)
1750                 return ERR_PTR(-ENOMEM);
1751
1752         cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1753                                  GFP_KERNEL);
1754         if (cd->hash_table == NULL) {
1755                 kfree(cd);
1756                 return ERR_PTR(-ENOMEM);
1757         }
1758
1759         for (i = 0; i < cd->hash_size; i++)
1760                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1761         cd->net = net;
1762         return cd;
1763 }
1764 EXPORT_SYMBOL_GPL(cache_create_net);
1765
1766 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1767 {
1768         kfree(cd->hash_table);
1769         kfree(cd);
1770 }
1771 EXPORT_SYMBOL_GPL(cache_destroy_net);
1772
1773 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1774                                  size_t count, loff_t *ppos)
1775 {
1776         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1777
1778         return cache_read(filp, buf, count, ppos, cd);
1779 }
1780
1781 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1782                                   size_t count, loff_t *ppos)
1783 {
1784         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1785
1786         return cache_write(filp, buf, count, ppos, cd);
1787 }
1788
1789 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1790 {
1791         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1792
1793         return cache_poll(filp, wait, cd);
1794 }
1795
1796 static long cache_ioctl_pipefs(struct file *filp,
1797                               unsigned int cmd, unsigned long arg)
1798 {
1799         struct inode *inode = file_inode(filp);
1800         struct cache_detail *cd = RPC_I(inode)->private;
1801
1802         return cache_ioctl(inode, filp, cmd, arg, cd);
1803 }
1804
1805 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1806 {
1807         struct cache_detail *cd = RPC_I(inode)->private;
1808
1809         return cache_open(inode, filp, cd);
1810 }
1811
1812 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1813 {
1814         struct cache_detail *cd = RPC_I(inode)->private;
1815
1816         return cache_release(inode, filp, cd);
1817 }
1818
1819 const struct file_operations cache_file_operations_pipefs = {
1820         .owner          = THIS_MODULE,
1821         .llseek         = no_llseek,
1822         .read           = cache_read_pipefs,
1823         .write          = cache_write_pipefs,
1824         .poll           = cache_poll_pipefs,
1825         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1826         .open           = cache_open_pipefs,
1827         .release        = cache_release_pipefs,
1828 };
1829
1830 static int content_open_pipefs(struct inode *inode, struct file *filp)
1831 {
1832         struct cache_detail *cd = RPC_I(inode)->private;
1833
1834         return content_open(inode, filp, cd);
1835 }
1836
1837 static int content_release_pipefs(struct inode *inode, struct file *filp)
1838 {
1839         struct cache_detail *cd = RPC_I(inode)->private;
1840
1841         return content_release(inode, filp, cd);
1842 }
1843
1844 const struct file_operations content_file_operations_pipefs = {
1845         .open           = content_open_pipefs,
1846         .read           = seq_read,
1847         .llseek         = seq_lseek,
1848         .release        = content_release_pipefs,
1849 };
1850
1851 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1852 {
1853         struct cache_detail *cd = RPC_I(inode)->private;
1854
1855         return open_flush(inode, filp, cd);
1856 }
1857
1858 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1859 {
1860         struct cache_detail *cd = RPC_I(inode)->private;
1861
1862         return release_flush(inode, filp, cd);
1863 }
1864
1865 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1866                             size_t count, loff_t *ppos)
1867 {
1868         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1869
1870         return read_flush(filp, buf, count, ppos, cd);
1871 }
1872
1873 static ssize_t write_flush_pipefs(struct file *filp,
1874                                   const char __user *buf,
1875                                   size_t count, loff_t *ppos)
1876 {
1877         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1878
1879         return write_flush(filp, buf, count, ppos, cd);
1880 }
1881
1882 const struct file_operations cache_flush_operations_pipefs = {
1883         .open           = open_flush_pipefs,
1884         .read           = read_flush_pipefs,
1885         .write          = write_flush_pipefs,
1886         .release        = release_flush_pipefs,
1887         .llseek         = no_llseek,
1888 };
1889
1890 int sunrpc_cache_register_pipefs(struct dentry *parent,
1891                                  const char *name, umode_t umode,
1892                                  struct cache_detail *cd)
1893 {
1894         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1895         if (IS_ERR(dir))
1896                 return PTR_ERR(dir);
1897         cd->pipefs = dir;
1898         return 0;
1899 }
1900 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1901
1902 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1903 {
1904         if (cd->pipefs) {
1905                 rpc_remove_cache_dir(cd->pipefs);
1906                 cd->pipefs = NULL;
1907         }
1908 }
1909 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1910
1911 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1912 {
1913         spin_lock(&cd->hash_lock);
1914         if (!hlist_unhashed(&h->cache_list)){
1915                 sunrpc_begin_cache_remove_entry(h, cd);
1916                 spin_unlock(&cd->hash_lock);
1917                 sunrpc_end_cache_remove_entry(h, cd);
1918         } else
1919                 spin_unlock(&cd->hash_lock);
1920 }
1921 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);