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
  37 #include "netns.h"
  38 #include "fail.h"
  39
  40 #define  RPCDBG_FACILITY RPCDBG_CACHE
  41
  42 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  43 static void cache_revisit_request(struct cache_head *item);
  44
  45 static void cache_init(struct cache_head *h, struct cache_detail *detail)
  46 {
  47         time64_t now = seconds_since_boot();
  48         INIT_HLIST_NODE(&h->cache_list);
  49         h->flags = 0;
  50         kref_init(&h->ref);
  51         h->expiry_time = now + CACHE_NEW_EXPIRY;
  52         if (now <= detail->flush_time)
  53                 /* ensure it isn't already expired */
  54                 now = detail->flush_time + 1;
  55         h->last_refresh = now;
  56 }
  57
  58 static void cache_fresh_unlocked(struct cache_head *head,
  59                                 struct cache_detail *detail);
  60
  61 static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
  62                                                 struct cache_head *key,
  63                                                 int hash)
  64 {
  65         struct hlist_head *head = &detail->hash_table[hash];
  66         struct cache_head *tmp;
  67
  68         rcu_read_lock();
  69         hlist_for_each_entry_rcu(tmp, head, cache_list) {
  70                 if (!detail->match(tmp, key))
  71                         continue;
  72                 if (test_bit(CACHE_VALID, &tmp->flags) &&
  73                     cache_is_expired(detail, tmp))
  74                         continue;
  75                 tmp = cache_get_rcu(tmp);
  76                 rcu_read_unlock();
  77                 return tmp;
  78         }
  79         rcu_read_unlock();
  80         return NULL;
  81 }
  82
  83 static void sunrpc_begin_cache_remove_entry(struct cache_head *ch,
  84                                             struct cache_detail *cd)
  85 {
  86         /* Must be called under cd->hash_lock */
  87         hlist_del_init_rcu(&ch->cache_list);
  88         set_bit(CACHE_CLEANED, &ch->flags);
  89         cd->entries --;
  90 }
  91
  92 static void sunrpc_end_cache_remove_entry(struct cache_head *ch,
  93                                           struct cache_detail *cd)
  94 {
  95         cache_fresh_unlocked(ch, cd);
  96         cache_put(ch, cd);
  97 }
  98
  99 static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
 100                                                  struct cache_head *key,
 101                                                  int hash)
 102 {
 103         struct cache_head *new, *tmp, *freeme = NULL;
 104         struct hlist_head *head = &detail->hash_table[hash];
 105
 106         new = detail->alloc();
 107         if (!new)
 108                 return NULL;
 109         /* must fully initialise 'new', else
 110          * we might get lose if we need to
 111          * cache_put it soon.
 112          */
 113         cache_init(new, detail);
 114         detail->init(new, key);
 115
 116         spin_lock(&detail->hash_lock);
 117
 118         /* check if entry appeared while we slept */
 119         hlist_for_each_entry_rcu(tmp, head, cache_list,
 120                                  lockdep_is_held(&detail->hash_lock)) {
 121                 if (!detail->match(tmp, key))
 122                         continue;
 123                 if (test_bit(CACHE_VALID, &tmp->flags) &&
 124                     cache_is_expired(detail, tmp)) {
 125                         sunrpc_begin_cache_remove_entry(tmp, detail);
 126                         trace_cache_entry_expired(detail, tmp);
 127                         freeme = tmp;
 128                         break;
 129                 }
 130                 cache_get(tmp);
 131                 spin_unlock(&detail->hash_lock);
 132                 cache_put(new, detail);
 133                 return tmp;
 134         }
 135
 136         hlist_add_head_rcu(&new->cache_list, head);
 137         detail->entries++;
 138         cache_get(new);
 139         spin_unlock(&detail->hash_lock);
 140
 141         if (freeme)
 142                 sunrpc_end_cache_remove_entry(freeme, detail);
 143         return new;
 144 }
 145
 146 struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
 147                                            struct cache_head *key, int hash)
 148 {
 149         struct cache_head *ret;
 150
 151         ret = sunrpc_cache_find_rcu(detail, key, hash);
 152         if (ret)
 153                 return ret;
 154         /* Didn't find anything, insert an empty entry */
 155         return sunrpc_cache_add_entry(detail, key, hash);
 156 }
 157 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
 158
 159 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 160
 161 static void cache_fresh_locked(struct cache_head *head, time64_t expiry,
 162                                struct cache_detail *detail)
 163 {
 164         time64_t now = seconds_since_boot();
 165         if (now <= detail->flush_time)
 166                 /* ensure it isn't immediately treated as expired */
 167                 now = detail->flush_time + 1;
 168         head->expiry_time = expiry;
 169         head->last_refresh = now;
 170         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 171         set_bit(CACHE_VALID, &head->flags);
 172 }
 173
 174 static void cache_fresh_unlocked(struct cache_head *head,
 175                                  struct cache_detail *detail)
 176 {
 177         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 178                 cache_revisit_request(head);
 179                 cache_dequeue(detail, head);
 180         }
 181 }
 182
 183 static void cache_make_negative(struct cache_detail *detail,
 184                                 struct cache_head *h)
 185 {
 186         set_bit(CACHE_NEGATIVE, &h->flags);
 187         trace_cache_entry_make_negative(detail, h);
 188 }
 189
 190 static void cache_entry_update(struct cache_detail *detail,
 191                                struct cache_head *h,
 192                                struct cache_head *new)
 193 {
 194         if (!test_bit(CACHE_NEGATIVE, &new->flags)) {
 195                 detail->update(h, new);
 196                 trace_cache_entry_update(detail, h);
 197         } else {
 198                 cache_make_negative(detail, h);
 199         }
 200 }
 201
 202 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 203                                        struct cache_head *new, struct cache_head *old, int hash)
 204 {
 205         /* The 'old' entry is to be replaced by 'new'.
 206          * If 'old' is not VALID, we update it directly,
 207          * otherwise we need to replace it
 208          */
 209         struct cache_head *tmp;
 210
 211         if (!test_bit(CACHE_VALID, &old->flags)) {
 212                 spin_lock(&detail->hash_lock);
 213                 if (!test_bit(CACHE_VALID, &old->flags)) {
 214                         cache_entry_update(detail, old, new);
 215                         cache_fresh_locked(old, new->expiry_time, detail);
 216                         spin_unlock(&detail->hash_lock);
 217                         cache_fresh_unlocked(old, detail);
 218                         return old;
 219                 }
 220                 spin_unlock(&detail->hash_lock);
 221         }
 222         /* We need to insert a new entry */
 223         tmp = detail->alloc();
 224         if (!tmp) {
 225                 cache_put(old, detail);
 226                 return NULL;
 227         }
 228         cache_init(tmp, detail);
 229         detail->init(tmp, old);
 230
 231         spin_lock(&detail->hash_lock);
 232         cache_entry_update(detail, tmp, new);
 233         hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
 234         detail->entries++;
 235         cache_get(tmp);
 236         cache_fresh_locked(tmp, new->expiry_time, detail);
 237         cache_fresh_locked(old, 0, detail);
 238         spin_unlock(&detail->hash_lock);
 239         cache_fresh_unlocked(tmp, detail);
 240         cache_fresh_unlocked(old, detail);
 241         cache_put(old, detail);
 242         return tmp;
 243 }
 244 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 245
 246 static inline int cache_is_valid(struct cache_head *h)
 247 {
 248         if (!test_bit(CACHE_VALID, &h->flags))
 249                 return -EAGAIN;
 250         else {
 251                 /* entry is valid */
 252                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 253                         return -ENOENT;
 254                 else {
 255                         /*
 256                          * In combination with write barrier in
 257                          * sunrpc_cache_update, ensures that anyone
 258                          * using the cache entry after this sees the
 259                          * updated contents:
 260                          */
 261                         smp_rmb();
 262                         return 0;
 263                 }
 264         }
 265 }
 266
 267 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 268 {
 269         int rv;
 270
 271         spin_lock(&detail->hash_lock);
 272         rv = cache_is_valid(h);
 273         if (rv == -EAGAIN) {
 274                 cache_make_negative(detail, h);
 275                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
 276                                    detail);
 277                 rv = -ENOENT;
 278         }
 279         spin_unlock(&detail->hash_lock);
 280         cache_fresh_unlocked(h, detail);
 281         return rv;
 282 }
 283
 284 /*
 285  * This is the generic cache management routine for all
 286  * the authentication caches.
 287  * It checks the currency of a cache item and will (later)
 288  * initiate an upcall to fill it if needed.
 289  *
 290  *
 291  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 292  * -EAGAIN if upcall is pending and request has been queued
 293  * -ETIMEDOUT if upcall failed or request could not be queue or
 294  *           upcall completed but item is still invalid (implying that
 295  *           the cache item has been replaced with a newer one).
 296  * -ENOENT if cache entry was negative
 297  */
 298 int cache_check(struct cache_detail *detail,
 299                     struct cache_head *h, struct cache_req *rqstp)
 300 {
 301         int rv;
 302         time64_t refresh_age, age;
 303
 304         /* First decide return status as best we can */
 305         rv = cache_is_valid(h);
 306
 307         /* now see if we want to start an upcall */
 308         refresh_age = (h->expiry_time - h->last_refresh);
 309         age = seconds_since_boot() - h->last_refresh;
 310
 311         if (rqstp == NULL) {
 312                 if (rv == -EAGAIN)
 313                         rv = -ENOENT;
 314         } else if (rv == -EAGAIN ||
 315                    (h->expiry_time != 0 && age > refresh_age/2)) {
 316                 dprintk("RPC:       Want update, refage=%lld, age=%lld\n",
 317                                 refresh_age, age);
 318                 switch (detail->cache_upcall(detail, h)) {
 319                 case -EINVAL:
 320                         rv = try_to_negate_entry(detail, h);
 321                         break;
 322                 case -EAGAIN:
 323                         cache_fresh_unlocked(h, detail);
 324                         break;
 325                 }
 326         }
 327
 328         if (rv == -EAGAIN) {
 329                 if (!cache_defer_req(rqstp, h)) {
 330                         /*
 331                          * Request was not deferred; handle it as best
 332                          * we can ourselves:
 333                          */
 334                         rv = cache_is_valid(h);
 335                         if (rv == -EAGAIN)
 336                                 rv = -ETIMEDOUT;
 337                 }
 338         }
 339         if (rv)
 340                 cache_put(h, detail);
 341         return rv;
 342 }
 343 EXPORT_SYMBOL_GPL(cache_check);
 344
 345 /*
 346  * caches need to be periodically cleaned.
 347  * For this we maintain a list of cache_detail and
 348  * a current pointer into that list and into the table
 349  * for that entry.
 350  *
 351  * Each time cache_clean is called it finds the next non-empty entry
 352  * in the current table and walks the list in that entry
 353  * looking for entries that can be removed.
 354  *
 355  * An entry gets removed if:
 356  * - The expiry is before current time
 357  * - The last_refresh time is before the flush_time for that cache
 358  *
 359  * later we might drop old entries with non-NEVER expiry if that table
 360  * is getting 'full' for some definition of 'full'
 361  *
 362  * The question of "how often to scan a table" is an interesting one
 363  * and is answered in part by the use of the "nextcheck" field in the
 364  * cache_detail.
 365  * When a scan of a table begins, the nextcheck field is set to a time
 366  * that is well into the future.
 367  * While scanning, if an expiry time is found that is earlier than the
 368  * current nextcheck time, nextcheck is set to that expiry time.
 369  * If the flush_time is ever set to a time earlier than the nextcheck
 370  * time, the nextcheck time is then set to that flush_time.
 371  *
 372  * A table is then only scanned if the current time is at least
 373  * the nextcheck time.
 374  *
 375  */
 376
 377 static LIST_HEAD(cache_list);
 378 static DEFINE_SPINLOCK(cache_list_lock);
 379 static struct cache_detail *current_detail;
 380 static int current_index;
 381
 382 static void do_cache_clean(struct work_struct *work);
 383 static struct delayed_work cache_cleaner;
 384
 385 void sunrpc_init_cache_detail(struct cache_detail *cd)
 386 {
 387         spin_lock_init(&cd->hash_lock);
 388         INIT_LIST_HEAD(&cd->queue);
 389         spin_lock(&cache_list_lock);
 390         cd->nextcheck = 0;
 391         cd->entries = 0;
 392         atomic_set(&cd->writers, 0);
 393         cd->last_close = 0;
 394         cd->last_warn = -1;
 395         list_add(&cd->others, &cache_list);
 396         spin_unlock(&cache_list_lock);
 397
 398         /* start the cleaning process */
 399         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
 400 }
 401 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 402
 403 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 404 {
 405         cache_purge(cd);
 406         spin_lock(&cache_list_lock);
 407         spin_lock(&cd->hash_lock);
 408         if (current_detail == cd)
 409                 current_detail = NULL;
 410         list_del_init(&cd->others);
 411         spin_unlock(&cd->hash_lock);
 412         spin_unlock(&cache_list_lock);
 413         if (list_empty(&cache_list)) {
 414                 /* module must be being unloaded so its safe to kill the worker */
 415                 cancel_delayed_work_sync(&cache_cleaner);
 416         }
 417 }
 418 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 419
 420 /* clean cache tries to find something to clean
 421  * and cleans it.
 422  * It returns 1 if it cleaned something,
 423  *            0 if it didn't find anything this time
 424  *           -1 if it fell off the end of the list.
 425  */
 426 static int cache_clean(void)
 427 {
 428         int rv = 0;
 429         struct list_head *next;
 430
 431         spin_lock(&cache_list_lock);
 432
 433         /* find a suitable table if we don't already have one */
 434         while (current_detail == NULL ||
 435             current_index >= current_detail->hash_size) {
 436                 if (current_detail)
 437                         next = current_detail->others.next;
 438                 else
 439                         next = cache_list.next;
 440                 if (next == &cache_list) {
 441                         current_detail = NULL;
 442                         spin_unlock(&cache_list_lock);
 443                         return -1;
 444                 }
 445                 current_detail = list_entry(next, struct cache_detail, others);
 446                 if (current_detail->nextcheck > seconds_since_boot())
 447                         current_index = current_detail->hash_size;
 448                 else {
 449                         current_index = 0;
 450                         current_detail->nextcheck = seconds_since_boot()+30*60;
 451                 }
 452         }
 453
 454         /* find a non-empty bucket in the table */
 455         while (current_detail &&
 456                current_index < current_detail->hash_size &&
 457                hlist_empty(&current_detail->hash_table[current_index]))
 458                 current_index++;
 459
 460         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 461
 462         if (current_detail && current_index < current_detail->hash_size) {
 463                 struct cache_head *ch = NULL;
 464                 struct cache_detail *d;
 465                 struct hlist_head *head;
 466                 struct hlist_node *tmp;
 467
 468                 spin_lock(&current_detail->hash_lock);
 469
 470                 /* Ok, now to clean this strand */
 471
 472                 head = &current_detail->hash_table[current_index];
 473                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 474                         if (current_detail->nextcheck > ch->expiry_time)
 475                                 current_detail->nextcheck = ch->expiry_time+1;
 476                         if (!cache_is_expired(current_detail, ch))
 477                                 continue;
 478
 479                         sunrpc_begin_cache_remove_entry(ch, current_detail);
 480                         trace_cache_entry_expired(current_detail, ch);
 481                         rv = 1;
 482                         break;
 483                 }
 484
 485                 spin_unlock(&current_detail->hash_lock);
 486                 d = current_detail;
 487                 if (!ch)
 488                         current_index ++;
 489                 spin_unlock(&cache_list_lock);
 490                 if (ch)
 491                         sunrpc_end_cache_remove_entry(ch, d);
 492         } else
 493                 spin_unlock(&cache_list_lock);
 494
 495         return rv;
 496 }
 497
 498 /*
 499  * We want to regularly clean the cache, so we need to schedule some work ...
 500  */
 501 static void do_cache_clean(struct work_struct *work)
 502 {
 503         int delay;
 504
 505         if (list_empty(&cache_list))
 506                 return;
 507
 508         if (cache_clean() == -1)
 509                 delay = round_jiffies_relative(30*HZ);
 510         else
 511                 delay = 5;
 512
 513         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, delay);
 514 }
 515
 516
 517 /*
 518  * Clean all caches promptly.  This just calls cache_clean
 519  * repeatedly until we are sure that every cache has had a chance to
 520  * be fully cleaned
 521  */
 522 void cache_flush(void)
 523 {
 524         while (cache_clean() != -1)
 525                 cond_resched();
 526         while (cache_clean() != -1)
 527                 cond_resched();
 528 }
 529 EXPORT_SYMBOL_GPL(cache_flush);
 530
 531 void cache_purge(struct cache_detail *detail)
 532 {
 533         struct cache_head *ch = NULL;
 534         struct hlist_head *head = NULL;
 535         int i = 0;
 536
 537         spin_lock(&detail->hash_lock);
 538         if (!detail->entries) {
 539                 spin_unlock(&detail->hash_lock);
 540                 return;
 541         }
 542
 543         dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
 544         for (i = 0; i < detail->hash_size; i++) {
 545                 head = &detail->hash_table[i];
 546                 while (!hlist_empty(head)) {
 547                         ch = hlist_entry(head->first, struct cache_head,
 548                                          cache_list);
 549                         sunrpc_begin_cache_remove_entry(ch, detail);
 550                         spin_unlock(&detail->hash_lock);
 551                         sunrpc_end_cache_remove_entry(ch, detail);
 552                         spin_lock(&detail->hash_lock);
 553                 }
 554         }
 555         spin_unlock(&detail->hash_lock);
 556 }
 557 EXPORT_SYMBOL_GPL(cache_purge);
 558
 559
 560 /*
 561  * Deferral and Revisiting of Requests.
 562  *
 563  * If a cache lookup finds a pending entry, we
 564  * need to defer the request and revisit it later.
 565  * All deferred requests are stored in a hash table,
 566  * indexed by "struct cache_head *".
 567  * As it may be wasteful to store a whole request
 568  * structure, we allow the request to provide a
 569  * deferred form, which must contain a
 570  * 'struct cache_deferred_req'
 571  * This cache_deferred_req contains a method to allow
 572  * it to be revisited when cache info is available
 573  */
 574
 575 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 576 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 577
 578 #define DFR_MAX 300     /* ??? */
 579
 580 static DEFINE_SPINLOCK(cache_defer_lock);
 581 static LIST_HEAD(cache_defer_list);
 582 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 583 static int cache_defer_cnt;
 584
 585 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 586 {
 587         hlist_del_init(&dreq->hash);
 588         if (!list_empty(&dreq->recent)) {
 589                 list_del_init(&dreq->recent);
 590                 cache_defer_cnt--;
 591         }
 592 }
 593
 594 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 595 {
 596         int hash = DFR_HASH(item);
 597
 598         INIT_LIST_HEAD(&dreq->recent);
 599         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 600 }
 601
 602 static void setup_deferral(struct cache_deferred_req *dreq,
 603                            struct cache_head *item,
 604                            int count_me)
 605 {
 606
 607         dreq->item = item;
 608
 609         spin_lock(&cache_defer_lock);
 610
 611         __hash_deferred_req(dreq, item);
 612
 613         if (count_me) {
 614                 cache_defer_cnt++;
 615                 list_add(&dreq->recent, &cache_defer_list);
 616         }
 617
 618         spin_unlock(&cache_defer_lock);
 619
 620 }
 621
 622 struct thread_deferred_req {
 623         struct cache_deferred_req handle;
 624         struct completion completion;
 625 };
 626
 627 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 628 {
 629         struct thread_deferred_req *dr =
 630                 container_of(dreq, struct thread_deferred_req, handle);
 631         complete(&dr->completion);
 632 }
 633
 634 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 635 {
 636         struct thread_deferred_req sleeper;
 637         struct cache_deferred_req *dreq = &sleeper.handle;
 638
 639         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 640         dreq->revisit = cache_restart_thread;
 641
 642         setup_deferral(dreq, item, 0);
 643
 644         if (!test_bit(CACHE_PENDING, &item->flags) ||
 645             wait_for_completion_interruptible_timeout(
 646                     &sleeper.completion, req->thread_wait) <= 0) {
 647                 /* The completion wasn't completed, so we need
 648                  * to clean up
 649                  */
 650                 spin_lock(&cache_defer_lock);
 651                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 652                         __unhash_deferred_req(&sleeper.handle);
 653                         spin_unlock(&cache_defer_lock);
 654                 } else {
 655                         /* cache_revisit_request already removed
 656                          * this from the hash table, but hasn't
 657                          * called ->revisit yet.  It will very soon
 658                          * and we need to wait for it.
 659                          */
 660                         spin_unlock(&cache_defer_lock);
 661                         wait_for_completion(&sleeper.completion);
 662                 }
 663         }
 664 }
 665
 666 static void cache_limit_defers(void)
 667 {
 668         /* Make sure we haven't exceed the limit of allowed deferred
 669          * requests.
 670          */
 671         struct cache_deferred_req *discard = NULL;
 672
 673         if (cache_defer_cnt <= DFR_MAX)
 674                 return;
 675
 676         spin_lock(&cache_defer_lock);
 677
 678         /* Consider removing either the first or the last */
 679         if (cache_defer_cnt > DFR_MAX) {
 680                 if (prandom_u32_max(2))
 681                         discard = list_entry(cache_defer_list.next,
 682                                              struct cache_deferred_req, recent);
 683                 else
 684                         discard = list_entry(cache_defer_list.prev,
 685                                              struct cache_deferred_req, recent);
 686                 __unhash_deferred_req(discard);
 687         }
 688         spin_unlock(&cache_defer_lock);
 689         if (discard)
 690                 discard->revisit(discard, 1);
 691 }
 692
 693 #if IS_ENABLED(CONFIG_FAIL_SUNRPC)
 694 static inline bool cache_defer_immediately(void)
 695 {
 696         return !fail_sunrpc.ignore_cache_wait &&
 697                 should_fail(&fail_sunrpc.attr, 1);
 698 }
 699 #else
 700 static inline bool cache_defer_immediately(void)
 701 {
 702         return false;
 703 }
 704 #endif
 705
 706 /* Return true if and only if a deferred request is queued. */
 707 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 708 {
 709         struct cache_deferred_req *dreq;
 710
 711         if (!cache_defer_immediately()) {
 712                 cache_wait_req(req, item);
 713                 if (!test_bit(CACHE_PENDING, &item->flags))
 714                         return false;
 715         }
 716
 717         dreq = req->defer(req);
 718         if (dreq == NULL)
 719                 return false;
 720         setup_deferral(dreq, item, 1);
 721         if (!test_bit(CACHE_PENDING, &item->flags))
 722                 /* Bit could have been cleared before we managed to
 723                  * set up the deferral, so need to revisit just in case
 724                  */
 725                 cache_revisit_request(item);
 726
 727         cache_limit_defers();
 728         return true;
 729 }
 730
 731 static void cache_revisit_request(struct cache_head *item)
 732 {
 733         struct cache_deferred_req *dreq;
 734         struct list_head pending;
 735         struct hlist_node *tmp;
 736         int hash = DFR_HASH(item);
 737
 738         INIT_LIST_HEAD(&pending);
 739         spin_lock(&cache_defer_lock);
 740
 741         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 742                 if (dreq->item == item) {
 743                         __unhash_deferred_req(dreq);
 744                         list_add(&dreq->recent, &pending);
 745                 }
 746
 747         spin_unlock(&cache_defer_lock);
 748
 749         while (!list_empty(&pending)) {
 750                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 751                 list_del_init(&dreq->recent);
 752                 dreq->revisit(dreq, 0);
 753         }
 754 }
 755
 756 void cache_clean_deferred(void *owner)
 757 {
 758         struct cache_deferred_req *dreq, *tmp;
 759         struct list_head pending;
 760
 761
 762         INIT_LIST_HEAD(&pending);
 763         spin_lock(&cache_defer_lock);
 764
 765         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 766                 if (dreq->owner == owner) {
 767                         __unhash_deferred_req(dreq);
 768                         list_add(&dreq->recent, &pending);
 769                 }
 770         }
 771         spin_unlock(&cache_defer_lock);
 772
 773         while (!list_empty(&pending)) {
 774                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 775                 list_del_init(&dreq->recent);
 776                 dreq->revisit(dreq, 1);
 777         }
 778 }
 779
 780 /*
 781  * communicate with user-space
 782  *
 783  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
 784  * On read, you get a full request, or block.
 785  * On write, an update request is processed.
 786  * Poll works if anything to read, and always allows write.
 787  *
 788  * Implemented by linked list of requests.  Each open file has
 789  * a ->private that also exists in this list.  New requests are added
 790  * to the end and may wakeup and preceding readers.
 791  * New readers are added to the head.  If, on read, an item is found with
 792  * CACHE_UPCALLING clear, we free it from the list.
 793  *
 794  */
 795
 796 static DEFINE_SPINLOCK(queue_lock);
 797
 798 struct cache_queue {
 799         struct list_head        list;
 800         int                     reader; /* if 0, then request */
 801 };
 802 struct cache_request {
 803         struct cache_queue      q;
 804         struct cache_head       *item;
 805         char                    * buf;
 806         int                     len;
 807         int                     readers;
 808 };
 809 struct cache_reader {
 810         struct cache_queue      q;
 811         int                     offset; /* if non-0, we have a refcnt on next request */
 812 };
 813
 814 static int cache_request(struct cache_detail *detail,
 815                                struct cache_request *crq)
 816 {
 817         char *bp = crq->buf;
 818         int len = PAGE_SIZE;
 819
 820         detail->cache_request(detail, crq->item, &bp, &len);
 821         if (len < 0)
 822                 return -E2BIG;
 823         return PAGE_SIZE - len;
 824 }
 825
 826 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 827                           loff_t *ppos, struct cache_detail *cd)
 828 {
 829         struct cache_reader *rp = filp->private_data;
 830         struct cache_request *rq;
 831         struct inode *inode = file_inode(filp);
 832         int err;
 833
 834         if (count == 0)
 835                 return 0;
 836
 837         inode_lock(inode); /* protect against multiple concurrent
 838                               * readers on this file */
 839  again:
 840         spin_lock(&queue_lock);
 841         /* need to find next request */
 842         while (rp->q.list.next != &cd->queue &&
 843                list_entry(rp->q.list.next, struct cache_queue, list)
 844                ->reader) {
 845                 struct list_head *next = rp->q.list.next;
 846                 list_move(&rp->q.list, next);
 847         }
 848         if (rp->q.list.next == &cd->queue) {
 849                 spin_unlock(&queue_lock);
 850                 inode_unlock(inode);
 851                 WARN_ON_ONCE(rp->offset);
 852                 return 0;
 853         }
 854         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 855         WARN_ON_ONCE(rq->q.reader);
 856         if (rp->offset == 0)
 857                 rq->readers++;
 858         spin_unlock(&queue_lock);
 859
 860         if (rq->len == 0) {
 861                 err = cache_request(cd, rq);
 862                 if (err < 0)
 863                         goto out;
 864                 rq->len = err;
 865         }
 866
 867         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 868                 err = -EAGAIN;
 869                 spin_lock(&queue_lock);
 870                 list_move(&rp->q.list, &rq->q.list);
 871                 spin_unlock(&queue_lock);
 872         } else {
 873                 if (rp->offset + count > rq->len)
 874                         count = rq->len - rp->offset;
 875                 err = -EFAULT;
 876                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 877                         goto out;
 878                 rp->offset += count;
 879                 if (rp->offset >= rq->len) {
 880                         rp->offset = 0;
 881                         spin_lock(&queue_lock);
 882                         list_move(&rp->q.list, &rq->q.list);
 883                         spin_unlock(&queue_lock);
 884                 }
 885                 err = 0;
 886         }
 887  out:
 888         if (rp->offset == 0) {
 889                 /* need to release rq */
 890                 spin_lock(&queue_lock);
 891                 rq->readers--;
 892                 if (rq->readers == 0 &&
 893                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 894                         list_del(&rq->q.list);
 895                         spin_unlock(&queue_lock);
 896                         cache_put(rq->item, cd);
 897                         kfree(rq->buf);
 898                         kfree(rq);
 899                 } else
 900                         spin_unlock(&queue_lock);
 901         }
 902         if (err == -EAGAIN)
 903                 goto again;
 904         inode_unlock(inode);
 905         return err ? err :  count;
 906 }
 907
 908 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 909                                  size_t count, struct cache_detail *cd)
 910 {
 911         ssize_t ret;
 912
 913         if (count == 0)
 914                 return -EINVAL;
 915         if (copy_from_user(kaddr, buf, count))
 916                 return -EFAULT;
 917         kaddr[count] = '\0';
 918         ret = cd->cache_parse(cd, kaddr, count);
 919         if (!ret)
 920                 ret = count;
 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         char *write_buf;
 929         ssize_t ret = -ENOMEM;
 930
 931         if (count >= 32768) { /* 32k is max userland buffer, lets check anyway */
 932                 ret = -EINVAL;
 933                 goto out;
 934         }
 935
 936         write_buf = kvmalloc(count + 1, GFP_KERNEL);
 937         if (!write_buf)
 938                 goto out;
 939
 940         ret = cache_do_downcall(write_buf, buf, count, cd);
 941         kvfree(write_buf);
 942 out:
 943         return ret;
 944 }
 945
 946 static ssize_t cache_write(struct file *filp, const char __user *buf,
 947                            size_t count, loff_t *ppos,
 948                            struct cache_detail *cd)
 949 {
 950         struct address_space *mapping = filp->f_mapping;
 951         struct inode *inode = file_inode(filp);
 952         ssize_t ret = -EINVAL;
 953
 954         if (!cd->cache_parse)
 955                 goto out;
 956
 957         inode_lock(inode);
 958         ret = cache_downcall(mapping, buf, count, cd);
 959         inode_unlock(inode);
 960 out:
 961         return ret;
 962 }
 963
 964 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 965
 966 static __poll_t cache_poll(struct file *filp, poll_table *wait,
 967                                struct cache_detail *cd)
 968 {
 969         __poll_t mask;
 970         struct cache_reader *rp = filp->private_data;
 971         struct cache_queue *cq;
 972
 973         poll_wait(filp, &queue_wait, wait);
 974
 975         /* alway allow write */
 976         mask = EPOLLOUT | EPOLLWRNORM;
 977
 978         if (!rp)
 979                 return mask;
 980
 981         spin_lock(&queue_lock);
 982
 983         for (cq= &rp->q; &cq->list != &cd->queue;
 984              cq = list_entry(cq->list.next, struct cache_queue, list))
 985                 if (!cq->reader) {
 986                         mask |= EPOLLIN | EPOLLRDNORM;
 987                         break;
 988                 }
 989         spin_unlock(&queue_lock);
 990         return mask;
 991 }
 992
 993 static int cache_ioctl(struct inode *ino, struct file *filp,
 994                        unsigned int cmd, unsigned long arg,
 995                        struct cache_detail *cd)
 996 {
 997         int len = 0;
 998         struct cache_reader *rp = filp->private_data;
 999         struct cache_queue *cq;
1000
1001         if (cmd != FIONREAD || !rp)
1002                 return -EINVAL;
1003
1004         spin_lock(&queue_lock);
1005
1006         /* only find the length remaining in current request,
1007          * or the length of the next request
1008          */
1009         for (cq= &rp->q; &cq->list != &cd->queue;
1010              cq = list_entry(cq->list.next, struct cache_queue, list))
1011                 if (!cq->reader) {
1012                         struct cache_request *cr =
1013                                 container_of(cq, struct cache_request, q);
1014                         len = cr->len - rp->offset;
1015                         break;
1016                 }
1017         spin_unlock(&queue_lock);
1018
1019         return put_user(len, (int __user *)arg);
1020 }
1021
1022 static int cache_open(struct inode *inode, struct file *filp,
1023                       struct cache_detail *cd)
1024 {
1025         struct cache_reader *rp = NULL;
1026
1027         if (!cd || !try_module_get(cd->owner))
1028                 return -EACCES;
1029         nonseekable_open(inode, filp);
1030         if (filp->f_mode & FMODE_READ) {
1031                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1032                 if (!rp) {
1033                         module_put(cd->owner);
1034                         return -ENOMEM;
1035                 }
1036                 rp->offset = 0;
1037                 rp->q.reader = 1;
1038
1039                 spin_lock(&queue_lock);
1040                 list_add(&rp->q.list, &cd->queue);
1041                 spin_unlock(&queue_lock);
1042         }
1043         if (filp->f_mode & FMODE_WRITE)
1044                 atomic_inc(&cd->writers);
1045         filp->private_data = rp;
1046         return 0;
1047 }
1048
1049 static int cache_release(struct inode *inode, struct file *filp,
1050                          struct cache_detail *cd)
1051 {
1052         struct cache_reader *rp = filp->private_data;
1053
1054         if (rp) {
1055                 spin_lock(&queue_lock);
1056                 if (rp->offset) {
1057                         struct cache_queue *cq;
1058                         for (cq= &rp->q; &cq->list != &cd->queue;
1059                              cq = list_entry(cq->list.next, struct cache_queue, list))
1060                                 if (!cq->reader) {
1061                                         container_of(cq, struct cache_request, q)
1062                                                 ->readers--;
1063                                         break;
1064                                 }
1065                         rp->offset = 0;
1066                 }
1067                 list_del(&rp->q.list);
1068                 spin_unlock(&queue_lock);
1069
1070                 filp->private_data = NULL;
1071                 kfree(rp);
1072
1073         }
1074         if (filp->f_mode & FMODE_WRITE) {
1075                 atomic_dec(&cd->writers);
1076                 cd->last_close = seconds_since_boot();
1077         }
1078         module_put(cd->owner);
1079         return 0;
1080 }
1081
1082
1083
1084 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1085 {
1086         struct cache_queue *cq, *tmp;
1087         struct cache_request *cr;
1088         struct list_head dequeued;
1089
1090         INIT_LIST_HEAD(&dequeued);
1091         spin_lock(&queue_lock);
1092         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1093                 if (!cq->reader) {
1094                         cr = container_of(cq, struct cache_request, q);
1095                         if (cr->item != ch)
1096                                 continue;
1097                         if (test_bit(CACHE_PENDING, &ch->flags))
1098                                 /* Lost a race and it is pending again */
1099                                 break;
1100                         if (cr->readers != 0)
1101                                 continue;
1102                         list_move(&cr->q.list, &dequeued);
1103                 }
1104         spin_unlock(&queue_lock);
1105         while (!list_empty(&dequeued)) {
1106                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1107                 list_del(&cr->q.list);
1108                 cache_put(cr->item, detail);
1109                 kfree(cr->buf);
1110                 kfree(cr);
1111         }
1112 }
1113
1114 /*
1115  * Support routines for text-based upcalls.
1116  * Fields are separated by spaces.
1117  * Fields are either mangled to quote space tab newline slosh with slosh
1118  * or a hexified with a leading \x
1119  * Record is terminated with newline.
1120  *
1121  */
1122
1123 void qword_add(char **bpp, int *lp, char *str)
1124 {
1125         char *bp = *bpp;
1126         int len = *lp;
1127         int ret;
1128
1129         if (len < 0) return;
1130
1131         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1132         if (ret >= len) {
1133                 bp += len;
1134                 len = -1;
1135         } else {
1136                 bp += ret;
1137                 len -= ret;
1138                 *bp++ = ' ';
1139                 len--;
1140         }
1141         *bpp = bp;
1142         *lp = len;
1143 }
1144 EXPORT_SYMBOL_GPL(qword_add);
1145
1146 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1147 {
1148         char *bp = *bpp;
1149         int len = *lp;
1150
1151         if (len < 0) return;
1152
1153         if (len > 2) {
1154                 *bp++ = '\\';
1155                 *bp++ = 'x';
1156                 len -= 2;
1157                 while (blen && len >= 2) {
1158                         bp = hex_byte_pack(bp, *buf++);
1159                         len -= 2;
1160                         blen--;
1161                 }
1162         }
1163         if (blen || len<1) len = -1;
1164         else {
1165                 *bp++ = ' ';
1166                 len--;
1167         }
1168         *bpp = bp;
1169         *lp = len;
1170 }
1171 EXPORT_SYMBOL_GPL(qword_addhex);
1172
1173 static void warn_no_listener(struct cache_detail *detail)
1174 {
1175         if (detail->last_warn != detail->last_close) {
1176                 detail->last_warn = detail->last_close;
1177                 if (detail->warn_no_listener)
1178                         detail->warn_no_listener(detail, detail->last_close != 0);
1179         }
1180 }
1181
1182 static bool cache_listeners_exist(struct cache_detail *detail)
1183 {
1184         if (atomic_read(&detail->writers))
1185                 return true;
1186         if (detail->last_close == 0)
1187                 /* This cache was never opened */
1188                 return false;
1189         if (detail->last_close < seconds_since_boot() - 30)
1190                 /*
1191                  * We allow for the possibility that someone might
1192                  * restart a userspace daemon without restarting the
1193                  * server; but after 30 seconds, we give up.
1194                  */
1195                  return false;
1196         return true;
1197 }
1198
1199 /*
1200  * register an upcall request to user-space and queue it up for read() by the
1201  * upcall daemon.
1202  *
1203  * Each request is at most one page long.
1204  */
1205 static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1206 {
1207         char *buf;
1208         struct cache_request *crq;
1209         int ret = 0;
1210
1211         if (test_bit(CACHE_CLEANED, &h->flags))
1212                 /* Too late to make an upcall */
1213                 return -EAGAIN;
1214
1215         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1216         if (!buf)
1217                 return -EAGAIN;
1218
1219         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1220         if (!crq) {
1221                 kfree(buf);
1222                 return -EAGAIN;
1223         }
1224
1225         crq->q.reader = 0;
1226         crq->buf = buf;
1227         crq->len = 0;
1228         crq->readers = 0;
1229         spin_lock(&queue_lock);
1230         if (test_bit(CACHE_PENDING, &h->flags)) {
1231                 crq->item = cache_get(h);
1232                 list_add_tail(&crq->q.list, &detail->queue);
1233                 trace_cache_entry_upcall(detail, h);
1234         } else
1235                 /* Lost a race, no longer PENDING, so don't enqueue */
1236                 ret = -EAGAIN;
1237         spin_unlock(&queue_lock);
1238         wake_up(&queue_wait);
1239         if (ret == -EAGAIN) {
1240                 kfree(buf);
1241                 kfree(crq);
1242         }
1243         return ret;
1244 }
1245
1246 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1247 {
1248         if (test_and_set_bit(CACHE_PENDING, &h->flags))
1249                 return 0;
1250         return cache_pipe_upcall(detail, h);
1251 }
1252 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1253
1254 int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1255                                      struct cache_head *h)
1256 {
1257         if (!cache_listeners_exist(detail)) {
1258                 warn_no_listener(detail);
1259                 trace_cache_entry_no_listener(detail, h);
1260                 return -EINVAL;
1261         }
1262         return sunrpc_cache_pipe_upcall(detail, h);
1263 }
1264 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1265
1266 /*
1267  * parse a message from user-space and pass it
1268  * to an appropriate cache
1269  * Messages are, like requests, separated into fields by
1270  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1271  *
1272  * Message is
1273  *   reply cachename expiry key ... content....
1274  *
1275  * key and content are both parsed by cache
1276  */
1277
1278 int qword_get(char **bpp, char *dest, int bufsize)
1279 {
1280         /* return bytes copied, or -1 on error */
1281         char *bp = *bpp;
1282         int len = 0;
1283
1284         while (*bp == ' ') bp++;
1285
1286         if (bp[0] == '\\' && bp[1] == 'x') {
1287                 /* HEX STRING */
1288                 bp += 2;
1289                 while (len < bufsize - 1) {
1290                         int h, l;
1291
1292                         h = hex_to_bin(bp[0]);
1293                         if (h < 0)
1294                                 break;
1295
1296                         l = hex_to_bin(bp[1]);
1297                         if (l < 0)
1298                                 break;
1299
1300                         *dest++ = (h << 4) | l;
1301                         bp += 2;
1302                         len++;
1303                 }
1304         } else {
1305                 /* text with \nnn octal quoting */
1306                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1307                         if (*bp == '\\' &&
1308                             isodigit(bp[1]) && (bp[1] <= '3') &&
1309                             isodigit(bp[2]) &&
1310                             isodigit(bp[3])) {
1311                                 int byte = (*++bp -'0');
1312                                 bp++;
1313                                 byte = (byte << 3) | (*bp++ - '0');
1314                                 byte = (byte << 3) | (*bp++ - '0');
1315                                 *dest++ = byte;
1316                                 len++;
1317                         } else {
1318                                 *dest++ = *bp++;
1319                                 len++;
1320                         }
1321                 }
1322         }
1323
1324         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1325                 return -1;
1326         while (*bp == ' ') bp++;
1327         *bpp = bp;
1328         *dest = '\0';
1329         return len;
1330 }
1331 EXPORT_SYMBOL_GPL(qword_get);
1332
1333
1334 /*
1335  * support /proc/net/rpc/$CACHENAME/content
1336  * as a seqfile.
1337  * We call ->cache_show passing NULL for the item to
1338  * get a header, then pass each real item in the cache
1339  */
1340
1341 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1342 {
1343         loff_t n = *pos;
1344         unsigned int hash, entry;
1345         struct cache_head *ch;
1346         struct cache_detail *cd = m->private;
1347
1348         if (!n--)
1349                 return SEQ_START_TOKEN;
1350         hash = n >> 32;
1351         entry = n & ((1LL<<32) - 1);
1352
1353         hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1354                 if (!entry--)
1355                         return ch;
1356         n &= ~((1LL<<32) - 1);
1357         do {
1358                 hash++;
1359                 n += 1LL<<32;
1360         } while(hash < cd->hash_size &&
1361                 hlist_empty(&cd->hash_table[hash]));
1362         if (hash >= cd->hash_size)
1363                 return NULL;
1364         *pos = n+1;
1365         return hlist_entry_safe(rcu_dereference_raw(
1366                                 hlist_first_rcu(&cd->hash_table[hash])),
1367                                 struct cache_head, cache_list);
1368 }
1369
1370 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1371 {
1372         struct cache_head *ch = p;
1373         int hash = (*pos >> 32);
1374         struct cache_detail *cd = m->private;
1375
1376         if (p == SEQ_START_TOKEN)
1377                 hash = 0;
1378         else if (ch->cache_list.next == NULL) {
1379                 hash++;
1380                 *pos += 1LL<<32;
1381         } else {
1382                 ++*pos;
1383                 return hlist_entry_safe(rcu_dereference_raw(
1384                                         hlist_next_rcu(&ch->cache_list)),
1385                                         struct cache_head, cache_list);
1386         }
1387         *pos &= ~((1LL<<32) - 1);
1388         while (hash < cd->hash_size &&
1389                hlist_empty(&cd->hash_table[hash])) {
1390                 hash++;
1391                 *pos += 1LL<<32;
1392         }
1393         if (hash >= cd->hash_size)
1394                 return NULL;
1395         ++*pos;
1396         return hlist_entry_safe(rcu_dereference_raw(
1397                                 hlist_first_rcu(&cd->hash_table[hash])),
1398                                 struct cache_head, cache_list);
1399 }
1400
1401 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1402         __acquires(RCU)
1403 {
1404         rcu_read_lock();
1405         return __cache_seq_start(m, pos);
1406 }
1407 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1408
1409 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1410 {
1411         return cache_seq_next(file, p, pos);
1412 }
1413 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1414
1415 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1416         __releases(RCU)
1417 {
1418         rcu_read_unlock();
1419 }
1420 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1421
1422 static int c_show(struct seq_file *m, void *p)
1423 {
1424         struct cache_head *cp = p;
1425         struct cache_detail *cd = m->private;
1426
1427         if (p == SEQ_START_TOKEN)
1428                 return cd->cache_show(m, cd, NULL);
1429
1430         ifdebug(CACHE)
1431                 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1432                            convert_to_wallclock(cp->expiry_time),
1433                            kref_read(&cp->ref), cp->flags);
1434         cache_get(cp);
1435         if (cache_check(cd, cp, NULL))
1436                 /* cache_check does a cache_put on failure */
1437                 seq_puts(m, "# ");
1438         else {
1439                 if (cache_is_expired(cd, cp))
1440                         seq_puts(m, "# ");
1441                 cache_put(cp, cd);
1442         }
1443
1444         return cd->cache_show(m, cd, cp);
1445 }
1446
1447 static const struct seq_operations cache_content_op = {
1448         .start  = cache_seq_start_rcu,
1449         .next   = cache_seq_next_rcu,
1450         .stop   = cache_seq_stop_rcu,
1451         .show   = c_show,
1452 };
1453
1454 static int content_open(struct inode *inode, struct file *file,
1455                         struct cache_detail *cd)
1456 {
1457         struct seq_file *seq;
1458         int err;
1459
1460         if (!cd || !try_module_get(cd->owner))
1461                 return -EACCES;
1462
1463         err = seq_open(file, &cache_content_op);
1464         if (err) {
1465                 module_put(cd->owner);
1466                 return err;
1467         }
1468
1469         seq = file->private_data;
1470         seq->private = cd;
1471         return 0;
1472 }
1473
1474 static int content_release(struct inode *inode, struct file *file,
1475                 struct cache_detail *cd)
1476 {
1477         int ret = seq_release(inode, file);
1478         module_put(cd->owner);
1479         return ret;
1480 }
1481
1482 static int open_flush(struct inode *inode, struct file *file,
1483                         struct cache_detail *cd)
1484 {
1485         if (!cd || !try_module_get(cd->owner))
1486                 return -EACCES;
1487         return nonseekable_open(inode, file);
1488 }
1489
1490 static int release_flush(struct inode *inode, struct file *file,
1491                         struct cache_detail *cd)
1492 {
1493         module_put(cd->owner);
1494         return 0;
1495 }
1496
1497 static ssize_t read_flush(struct file *file, char __user *buf,
1498                           size_t count, loff_t *ppos,
1499                           struct cache_detail *cd)
1500 {
1501         char tbuf[22];
1502         size_t len;
1503
1504         len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1505                         convert_to_wallclock(cd->flush_time));
1506         return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1507 }
1508
1509 static ssize_t write_flush(struct file *file, const char __user *buf,
1510                            size_t count, loff_t *ppos,
1511                            struct cache_detail *cd)
1512 {
1513         char tbuf[20];
1514         char *ep;
1515         time64_t now;
1516
1517         if (*ppos || count > sizeof(tbuf)-1)
1518                 return -EINVAL;
1519         if (copy_from_user(tbuf, buf, count))
1520                 return -EFAULT;
1521         tbuf[count] = 0;
1522         simple_strtoul(tbuf, &ep, 0);
1523         if (*ep && *ep != '\n')
1524                 return -EINVAL;
1525         /* Note that while we check that 'buf' holds a valid number,
1526          * we always ignore the value and just flush everything.
1527          * Making use of the number leads to races.
1528          */
1529
1530         now = seconds_since_boot();
1531         /* Always flush everything, so behave like cache_purge()
1532          * Do this by advancing flush_time to the current time,
1533          * or by one second if it has already reached the current time.
1534          * Newly added cache entries will always have ->last_refresh greater
1535          * that ->flush_time, so they don't get flushed prematurely.
1536          */
1537
1538         if (cd->flush_time >= now)
1539                 now = cd->flush_time + 1;
1540
1541         cd->flush_time = now;
1542         cd->nextcheck = now;
1543         cache_flush();
1544
1545         if (cd->flush)
1546                 cd->flush();
1547
1548         *ppos += count;
1549         return count;
1550 }
1551
1552 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1553                                  size_t count, loff_t *ppos)
1554 {
1555         struct cache_detail *cd = pde_data(file_inode(filp));
1556
1557         return cache_read(filp, buf, count, ppos, cd);
1558 }
1559
1560 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1561                                   size_t count, loff_t *ppos)
1562 {
1563         struct cache_detail *cd = pde_data(file_inode(filp));
1564
1565         return cache_write(filp, buf, count, ppos, cd);
1566 }
1567
1568 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1569 {
1570         struct cache_detail *cd = pde_data(file_inode(filp));
1571
1572         return cache_poll(filp, wait, cd);
1573 }
1574
1575 static long cache_ioctl_procfs(struct file *filp,
1576                                unsigned int cmd, unsigned long arg)
1577 {
1578         struct inode *inode = file_inode(filp);
1579         struct cache_detail *cd = pde_data(inode);
1580
1581         return cache_ioctl(inode, filp, cmd, arg, cd);
1582 }
1583
1584 static int cache_open_procfs(struct inode *inode, struct file *filp)
1585 {
1586         struct cache_detail *cd = pde_data(inode);
1587
1588         return cache_open(inode, filp, cd);
1589 }
1590
1591 static int cache_release_procfs(struct inode *inode, struct file *filp)
1592 {
1593         struct cache_detail *cd = pde_data(inode);
1594
1595         return cache_release(inode, filp, cd);
1596 }
1597
1598 static const struct proc_ops cache_channel_proc_ops = {
1599         .proc_lseek     = no_llseek,
1600         .proc_read      = cache_read_procfs,
1601         .proc_write     = cache_write_procfs,
1602         .proc_poll      = cache_poll_procfs,
1603         .proc_ioctl     = cache_ioctl_procfs, /* for FIONREAD */
1604         .proc_open      = cache_open_procfs,
1605         .proc_release   = cache_release_procfs,
1606 };
1607
1608 static int content_open_procfs(struct inode *inode, struct file *filp)
1609 {
1610         struct cache_detail *cd = pde_data(inode);
1611
1612         return content_open(inode, filp, cd);
1613 }
1614
1615 static int content_release_procfs(struct inode *inode, struct file *filp)
1616 {
1617         struct cache_detail *cd = pde_data(inode);
1618
1619         return content_release(inode, filp, cd);
1620 }
1621
1622 static const struct proc_ops content_proc_ops = {
1623         .proc_open      = content_open_procfs,
1624         .proc_read      = seq_read,
1625         .proc_lseek     = seq_lseek,
1626         .proc_release   = content_release_procfs,
1627 };
1628
1629 static int open_flush_procfs(struct inode *inode, struct file *filp)
1630 {
1631         struct cache_detail *cd = pde_data(inode);
1632
1633         return open_flush(inode, filp, cd);
1634 }
1635
1636 static int release_flush_procfs(struct inode *inode, struct file *filp)
1637 {
1638         struct cache_detail *cd = pde_data(inode);
1639
1640         return release_flush(inode, filp, cd);
1641 }
1642
1643 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1644                             size_t count, loff_t *ppos)
1645 {
1646         struct cache_detail *cd = pde_data(file_inode(filp));
1647
1648         return read_flush(filp, buf, count, ppos, cd);
1649 }
1650
1651 static ssize_t write_flush_procfs(struct file *filp,
1652                                   const char __user *buf,
1653                                   size_t count, loff_t *ppos)
1654 {
1655         struct cache_detail *cd = pde_data(file_inode(filp));
1656
1657         return write_flush(filp, buf, count, ppos, cd);
1658 }
1659
1660 static const struct proc_ops cache_flush_proc_ops = {
1661         .proc_open      = open_flush_procfs,
1662         .proc_read      = read_flush_procfs,
1663         .proc_write     = write_flush_procfs,
1664         .proc_release   = release_flush_procfs,
1665         .proc_lseek     = no_llseek,
1666 };
1667
1668 static void remove_cache_proc_entries(struct cache_detail *cd)
1669 {
1670         if (cd->procfs) {
1671                 proc_remove(cd->procfs);
1672                 cd->procfs = NULL;
1673         }
1674 }
1675
1676 #ifdef CONFIG_PROC_FS
1677 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1678 {
1679         struct proc_dir_entry *p;
1680         struct sunrpc_net *sn;
1681
1682         sn = net_generic(net, sunrpc_net_id);
1683         cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1684         if (cd->procfs == NULL)
1685                 goto out_nomem;
1686
1687         p = proc_create_data("flush", S_IFREG | 0600,
1688                              cd->procfs, &cache_flush_proc_ops, cd);
1689         if (p == NULL)
1690                 goto out_nomem;
1691
1692         if (cd->cache_request || cd->cache_parse) {
1693                 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1694                                      &cache_channel_proc_ops, cd);
1695                 if (p == NULL)
1696                         goto out_nomem;
1697         }
1698         if (cd->cache_show) {
1699                 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1700                                      &content_proc_ops, cd);
1701                 if (p == NULL)
1702                         goto out_nomem;
1703         }
1704         return 0;
1705 out_nomem:
1706         remove_cache_proc_entries(cd);
1707         return -ENOMEM;
1708 }
1709 #else /* CONFIG_PROC_FS */
1710 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1711 {
1712         return 0;
1713 }
1714 #endif
1715
1716 void __init cache_initialize(void)
1717 {
1718         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1719 }
1720
1721 int cache_register_net(struct cache_detail *cd, struct net *net)
1722 {
1723         int ret;
1724
1725         sunrpc_init_cache_detail(cd);
1726         ret = create_cache_proc_entries(cd, net);
1727         if (ret)
1728                 sunrpc_destroy_cache_detail(cd);
1729         return ret;
1730 }
1731 EXPORT_SYMBOL_GPL(cache_register_net);
1732
1733 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1734 {
1735         remove_cache_proc_entries(cd);
1736         sunrpc_destroy_cache_detail(cd);
1737 }
1738 EXPORT_SYMBOL_GPL(cache_unregister_net);
1739
1740 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1741 {
1742         struct cache_detail *cd;
1743         int i;
1744
1745         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1746         if (cd == NULL)
1747                 return ERR_PTR(-ENOMEM);
1748
1749         cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1750                                  GFP_KERNEL);
1751         if (cd->hash_table == NULL) {
1752                 kfree(cd);
1753                 return ERR_PTR(-ENOMEM);
1754         }
1755
1756         for (i = 0; i < cd->hash_size; i++)
1757                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1758         cd->net = net;
1759         return cd;
1760 }
1761 EXPORT_SYMBOL_GPL(cache_create_net);
1762
1763 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1764 {
1765         kfree(cd->hash_table);
1766         kfree(cd);
1767 }
1768 EXPORT_SYMBOL_GPL(cache_destroy_net);
1769
1770 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1771                                  size_t count, loff_t *ppos)
1772 {
1773         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1774
1775         return cache_read(filp, buf, count, ppos, cd);
1776 }
1777
1778 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1779                                   size_t count, loff_t *ppos)
1780 {
1781         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1782
1783         return cache_write(filp, buf, count, ppos, cd);
1784 }
1785
1786 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1787 {
1788         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1789
1790         return cache_poll(filp, wait, cd);
1791 }
1792
1793 static long cache_ioctl_pipefs(struct file *filp,
1794                               unsigned int cmd, unsigned long arg)
1795 {
1796         struct inode *inode = file_inode(filp);
1797         struct cache_detail *cd = RPC_I(inode)->private;
1798
1799         return cache_ioctl(inode, filp, cmd, arg, cd);
1800 }
1801
1802 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1803 {
1804         struct cache_detail *cd = RPC_I(inode)->private;
1805
1806         return cache_open(inode, filp, cd);
1807 }
1808
1809 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1810 {
1811         struct cache_detail *cd = RPC_I(inode)->private;
1812
1813         return cache_release(inode, filp, cd);
1814 }
1815
1816 const struct file_operations cache_file_operations_pipefs = {
1817         .owner          = THIS_MODULE,
1818         .llseek         = no_llseek,
1819         .read           = cache_read_pipefs,
1820         .write          = cache_write_pipefs,
1821         .poll           = cache_poll_pipefs,
1822         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1823         .open           = cache_open_pipefs,
1824         .release        = cache_release_pipefs,
1825 };
1826
1827 static int content_open_pipefs(struct inode *inode, struct file *filp)
1828 {
1829         struct cache_detail *cd = RPC_I(inode)->private;
1830
1831         return content_open(inode, filp, cd);
1832 }
1833
1834 static int content_release_pipefs(struct inode *inode, struct file *filp)
1835 {
1836         struct cache_detail *cd = RPC_I(inode)->private;
1837
1838         return content_release(inode, filp, cd);
1839 }
1840
1841 const struct file_operations content_file_operations_pipefs = {
1842         .open           = content_open_pipefs,
1843         .read           = seq_read,
1844         .llseek         = seq_lseek,
1845         .release        = content_release_pipefs,
1846 };
1847
1848 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1849 {
1850         struct cache_detail *cd = RPC_I(inode)->private;
1851
1852         return open_flush(inode, filp, cd);
1853 }
1854
1855 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1856 {
1857         struct cache_detail *cd = RPC_I(inode)->private;
1858
1859         return release_flush(inode, filp, cd);
1860 }
1861
1862 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1863                             size_t count, loff_t *ppos)
1864 {
1865         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1866
1867         return read_flush(filp, buf, count, ppos, cd);
1868 }
1869
1870 static ssize_t write_flush_pipefs(struct file *filp,
1871                                   const char __user *buf,
1872                                   size_t count, loff_t *ppos)
1873 {
1874         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1875
1876         return write_flush(filp, buf, count, ppos, cd);
1877 }
1878
1879 const struct file_operations cache_flush_operations_pipefs = {
1880         .open           = open_flush_pipefs,
1881         .read           = read_flush_pipefs,
1882         .write          = write_flush_pipefs,
1883         .release        = release_flush_pipefs,
1884         .llseek         = no_llseek,
1885 };
1886
1887 int sunrpc_cache_register_pipefs(struct dentry *parent,
1888                                  const char *name, umode_t umode,
1889                                  struct cache_detail *cd)
1890 {
1891         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1892         if (IS_ERR(dir))
1893                 return PTR_ERR(dir);
1894         cd->pipefs = dir;
1895         return 0;
1896 }
1897 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1898
1899 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1900 {
1901         if (cd->pipefs) {
1902                 rpc_remove_cache_dir(cd->pipefs);
1903                 cd->pipefs = NULL;
1904         }
1905 }
1906 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1907
1908 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1909 {
1910         spin_lock(&cd->hash_lock);
1911         if (!hlist_unhashed(&h->cache_list)){
1912                 sunrpc_begin_cache_remove_entry(h, cd);
1913                 spin_unlock(&cd->hash_lock);
1914                 sunrpc_end_cache_remove_entry(h, cd);
1915         } else
1916                 spin_unlock(&cd->hash_lock);
1917 }
1918 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);