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 <[email protected]>
   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 (get_random_u32_below(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 hlist_node *tmp;
 735         int hash = DFR_HASH(item);
 736         LIST_HEAD(pending);
 737
 738         spin_lock(&cache_defer_lock);
 739
 740         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 741                 if (dreq->item == item) {
 742                         __unhash_deferred_req(dreq);
 743                         list_add(&dreq->recent, &pending);
 744                 }
 745
 746         spin_unlock(&cache_defer_lock);
 747
 748         while (!list_empty(&pending)) {
 749                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 750                 list_del_init(&dreq->recent);
 751                 dreq->revisit(dreq, 0);
 752         }
 753 }
 754
 755 void cache_clean_deferred(void *owner)
 756 {
 757         struct cache_deferred_req *dreq, *tmp;
 758         LIST_HEAD(pending);
 759
 760         spin_lock(&cache_defer_lock);
 761
 762         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 763                 if (dreq->owner == owner) {
 764                         __unhash_deferred_req(dreq);
 765                         list_add(&dreq->recent, &pending);
 766                 }
 767         }
 768         spin_unlock(&cache_defer_lock);
 769
 770         while (!list_empty(&pending)) {
 771                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 772                 list_del_init(&dreq->recent);
 773                 dreq->revisit(dreq, 1);
 774         }
 775 }
 776
 777 /*
 778  * communicate with user-space
 779  *
 780  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
 781  * On read, you get a full request, or block.
 782  * On write, an update request is processed.
 783  * Poll works if anything to read, and always allows write.
 784  *
 785  * Implemented by linked list of requests.  Each open file has
 786  * a ->private that also exists in this list.  New requests are added
 787  * to the end and may wakeup and preceding readers.
 788  * New readers are added to the head.  If, on read, an item is found with
 789  * CACHE_UPCALLING clear, we free it from the list.
 790  *
 791  */
 792
 793 static DEFINE_SPINLOCK(queue_lock);
 794
 795 struct cache_queue {
 796         struct list_head        list;
 797         int                     reader; /* if 0, then request */
 798 };
 799 struct cache_request {
 800         struct cache_queue      q;
 801         struct cache_head       *item;
 802         char                    * buf;
 803         int                     len;
 804         int                     readers;
 805 };
 806 struct cache_reader {
 807         struct cache_queue      q;
 808         int                     offset; /* if non-0, we have a refcnt on next request */
 809 };
 810
 811 static int cache_request(struct cache_detail *detail,
 812                                struct cache_request *crq)
 813 {
 814         char *bp = crq->buf;
 815         int len = PAGE_SIZE;
 816
 817         detail->cache_request(detail, crq->item, &bp, &len);
 818         if (len < 0)
 819                 return -E2BIG;
 820         return PAGE_SIZE - len;
 821 }
 822
 823 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 824                           loff_t *ppos, struct cache_detail *cd)
 825 {
 826         struct cache_reader *rp = filp->private_data;
 827         struct cache_request *rq;
 828         struct inode *inode = file_inode(filp);
 829         int err;
 830
 831         if (count == 0)
 832                 return 0;
 833
 834         inode_lock(inode); /* protect against multiple concurrent
 835                               * readers on this file */
 836  again:
 837         spin_lock(&queue_lock);
 838         /* need to find next request */
 839         while (rp->q.list.next != &cd->queue &&
 840                list_entry(rp->q.list.next, struct cache_queue, list)
 841                ->reader) {
 842                 struct list_head *next = rp->q.list.next;
 843                 list_move(&rp->q.list, next);
 844         }
 845         if (rp->q.list.next == &cd->queue) {
 846                 spin_unlock(&queue_lock);
 847                 inode_unlock(inode);
 848                 WARN_ON_ONCE(rp->offset);
 849                 return 0;
 850         }
 851         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 852         WARN_ON_ONCE(rq->q.reader);
 853         if (rp->offset == 0)
 854                 rq->readers++;
 855         spin_unlock(&queue_lock);
 856
 857         if (rq->len == 0) {
 858                 err = cache_request(cd, rq);
 859                 if (err < 0)
 860                         goto out;
 861                 rq->len = err;
 862         }
 863
 864         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 865                 err = -EAGAIN;
 866                 spin_lock(&queue_lock);
 867                 list_move(&rp->q.list, &rq->q.list);
 868                 spin_unlock(&queue_lock);
 869         } else {
 870                 if (rp->offset + count > rq->len)
 871                         count = rq->len - rp->offset;
 872                 err = -EFAULT;
 873                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 874                         goto out;
 875                 rp->offset += count;
 876                 if (rp->offset >= rq->len) {
 877                         rp->offset = 0;
 878                         spin_lock(&queue_lock);
 879                         list_move(&rp->q.list, &rq->q.list);
 880                         spin_unlock(&queue_lock);
 881                 }
 882                 err = 0;
 883         }
 884  out:
 885         if (rp->offset == 0) {
 886                 /* need to release rq */
 887                 spin_lock(&queue_lock);
 888                 rq->readers--;
 889                 if (rq->readers == 0 &&
 890                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 891                         list_del(&rq->q.list);
 892                         spin_unlock(&queue_lock);
 893                         cache_put(rq->item, cd);
 894                         kfree(rq->buf);
 895                         kfree(rq);
 896                 } else
 897                         spin_unlock(&queue_lock);
 898         }
 899         if (err == -EAGAIN)
 900                 goto again;
 901         inode_unlock(inode);
 902         return err ? err :  count;
 903 }
 904
 905 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 906                                  size_t count, struct cache_detail *cd)
 907 {
 908         ssize_t ret;
 909
 910         if (count == 0)
 911                 return -EINVAL;
 912         if (copy_from_user(kaddr, buf, count))
 913                 return -EFAULT;
 914         kaddr[count] = '\0';
 915         ret = cd->cache_parse(cd, kaddr, count);
 916         if (!ret)
 917                 ret = count;
 918         return ret;
 919 }
 920
 921 static ssize_t cache_downcall(struct address_space *mapping,
 922                               const char __user *buf,
 923                               size_t count, struct cache_detail *cd)
 924 {
 925         char *write_buf;
 926         ssize_t ret = -ENOMEM;
 927
 928         if (count >= 32768) { /* 32k is max userland buffer, lets check anyway */
 929                 ret = -EINVAL;
 930                 goto out;
 931         }
 932
 933         write_buf = kvmalloc(count + 1, GFP_KERNEL);
 934         if (!write_buf)
 935                 goto out;
 936
 937         ret = cache_do_downcall(write_buf, buf, count, cd);
 938         kvfree(write_buf);
 939 out:
 940         return ret;
 941 }
 942
 943 static ssize_t cache_write(struct file *filp, const char __user *buf,
 944                            size_t count, loff_t *ppos,
 945                            struct cache_detail *cd)
 946 {
 947         struct address_space *mapping = filp->f_mapping;
 948         struct inode *inode = file_inode(filp);
 949         ssize_t ret = -EINVAL;
 950
 951         if (!cd->cache_parse)
 952                 goto out;
 953
 954         inode_lock(inode);
 955         ret = cache_downcall(mapping, buf, count, cd);
 956         inode_unlock(inode);
 957 out:
 958         return ret;
 959 }
 960
 961 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 962
 963 static __poll_t cache_poll(struct file *filp, poll_table *wait,
 964                                struct cache_detail *cd)
 965 {
 966         __poll_t mask;
 967         struct cache_reader *rp = filp->private_data;
 968         struct cache_queue *cq;
 969
 970         poll_wait(filp, &queue_wait, wait);
 971
 972         /* alway allow write */
 973         mask = EPOLLOUT | EPOLLWRNORM;
 974
 975         if (!rp)
 976                 return mask;
 977
 978         spin_lock(&queue_lock);
 979
 980         for (cq= &rp->q; &cq->list != &cd->queue;
 981              cq = list_entry(cq->list.next, struct cache_queue, list))
 982                 if (!cq->reader) {
 983                         mask |= EPOLLIN | EPOLLRDNORM;
 984                         break;
 985                 }
 986         spin_unlock(&queue_lock);
 987         return mask;
 988 }
 989
 990 static int cache_ioctl(struct inode *ino, struct file *filp,
 991                        unsigned int cmd, unsigned long arg,
 992                        struct cache_detail *cd)
 993 {
 994         int len = 0;
 995         struct cache_reader *rp = filp->private_data;
 996         struct cache_queue *cq;
 997
 998         if (cmd != FIONREAD || !rp)
 999                 return -EINVAL;
1000
1001         spin_lock(&queue_lock);
1002
1003         /* only find the length remaining in current request,
1004          * or the length of the next request
1005          */
1006         for (cq= &rp->q; &cq->list != &cd->queue;
1007              cq = list_entry(cq->list.next, struct cache_queue, list))
1008                 if (!cq->reader) {
1009                         struct cache_request *cr =
1010                                 container_of(cq, struct cache_request, q);
1011                         len = cr->len - rp->offset;
1012                         break;
1013                 }
1014         spin_unlock(&queue_lock);
1015
1016         return put_user(len, (int __user *)arg);
1017 }
1018
1019 static int cache_open(struct inode *inode, struct file *filp,
1020                       struct cache_detail *cd)
1021 {
1022         struct cache_reader *rp = NULL;
1023
1024         if (!cd || !try_module_get(cd->owner))
1025                 return -EACCES;
1026         nonseekable_open(inode, filp);
1027         if (filp->f_mode & FMODE_READ) {
1028                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1029                 if (!rp) {
1030                         module_put(cd->owner);
1031                         return -ENOMEM;
1032                 }
1033                 rp->offset = 0;
1034                 rp->q.reader = 1;
1035
1036                 spin_lock(&queue_lock);
1037                 list_add(&rp->q.list, &cd->queue);
1038                 spin_unlock(&queue_lock);
1039         }
1040         if (filp->f_mode & FMODE_WRITE)
1041                 atomic_inc(&cd->writers);
1042         filp->private_data = rp;
1043         return 0;
1044 }
1045
1046 static int cache_release(struct inode *inode, struct file *filp,
1047                          struct cache_detail *cd)
1048 {
1049         struct cache_reader *rp = filp->private_data;
1050
1051         if (rp) {
1052                 spin_lock(&queue_lock);
1053                 if (rp->offset) {
1054                         struct cache_queue *cq;
1055                         for (cq= &rp->q; &cq->list != &cd->queue;
1056                              cq = list_entry(cq->list.next, struct cache_queue, list))
1057                                 if (!cq->reader) {
1058                                         container_of(cq, struct cache_request, q)
1059                                                 ->readers--;
1060                                         break;
1061                                 }
1062                         rp->offset = 0;
1063                 }
1064                 list_del(&rp->q.list);
1065                 spin_unlock(&queue_lock);
1066
1067                 filp->private_data = NULL;
1068                 kfree(rp);
1069
1070         }
1071         if (filp->f_mode & FMODE_WRITE) {
1072                 atomic_dec(&cd->writers);
1073                 cd->last_close = seconds_since_boot();
1074         }
1075         module_put(cd->owner);
1076         return 0;
1077 }
1078
1079
1080
1081 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1082 {
1083         struct cache_queue *cq, *tmp;
1084         struct cache_request *cr;
1085         LIST_HEAD(dequeued);
1086
1087         spin_lock(&queue_lock);
1088         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1089                 if (!cq->reader) {
1090                         cr = container_of(cq, struct cache_request, q);
1091                         if (cr->item != ch)
1092                                 continue;
1093                         if (test_bit(CACHE_PENDING, &ch->flags))
1094                                 /* Lost a race and it is pending again */
1095                                 break;
1096                         if (cr->readers != 0)
1097                                 continue;
1098                         list_move(&cr->q.list, &dequeued);
1099                 }
1100         spin_unlock(&queue_lock);
1101         while (!list_empty(&dequeued)) {
1102                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1103                 list_del(&cr->q.list);
1104                 cache_put(cr->item, detail);
1105                 kfree(cr->buf);
1106                 kfree(cr);
1107         }
1108 }
1109
1110 /*
1111  * Support routines for text-based upcalls.
1112  * Fields are separated by spaces.
1113  * Fields are either mangled to quote space tab newline slosh with slosh
1114  * or a hexified with a leading \x
1115  * Record is terminated with newline.
1116  *
1117  */
1118
1119 void qword_add(char **bpp, int *lp, char *str)
1120 {
1121         char *bp = *bpp;
1122         int len = *lp;
1123         int ret;
1124
1125         if (len < 0) return;
1126
1127         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1128         if (ret >= len) {
1129                 bp += len;
1130                 len = -1;
1131         } else {
1132                 bp += ret;
1133                 len -= ret;
1134                 *bp++ = ' ';
1135                 len--;
1136         }
1137         *bpp = bp;
1138         *lp = len;
1139 }
1140 EXPORT_SYMBOL_GPL(qword_add);
1141
1142 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1143 {
1144         char *bp = *bpp;
1145         int len = *lp;
1146
1147         if (len < 0) return;
1148
1149         if (len > 2) {
1150                 *bp++ = '\\';
1151                 *bp++ = 'x';
1152                 len -= 2;
1153                 while (blen && len >= 2) {
1154                         bp = hex_byte_pack(bp, *buf++);
1155                         len -= 2;
1156                         blen--;
1157                 }
1158         }
1159         if (blen || len<1) len = -1;
1160         else {
1161                 *bp++ = ' ';
1162                 len--;
1163         }
1164         *bpp = bp;
1165         *lp = len;
1166 }
1167 EXPORT_SYMBOL_GPL(qword_addhex);
1168
1169 static void warn_no_listener(struct cache_detail *detail)
1170 {
1171         if (detail->last_warn != detail->last_close) {
1172                 detail->last_warn = detail->last_close;
1173                 if (detail->warn_no_listener)
1174                         detail->warn_no_listener(detail, detail->last_close != 0);
1175         }
1176 }
1177
1178 static bool cache_listeners_exist(struct cache_detail *detail)
1179 {
1180         if (atomic_read(&detail->writers))
1181                 return true;
1182         if (detail->last_close == 0)
1183                 /* This cache was never opened */
1184                 return false;
1185         if (detail->last_close < seconds_since_boot() - 30)
1186                 /*
1187                  * We allow for the possibility that someone might
1188                  * restart a userspace daemon without restarting the
1189                  * server; but after 30 seconds, we give up.
1190                  */
1191                  return false;
1192         return true;
1193 }
1194
1195 /*
1196  * register an upcall request to user-space and queue it up for read() by the
1197  * upcall daemon.
1198  *
1199  * Each request is at most one page long.
1200  */
1201 static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1202 {
1203         char *buf;
1204         struct cache_request *crq;
1205         int ret = 0;
1206
1207         if (test_bit(CACHE_CLEANED, &h->flags))
1208                 /* Too late to make an upcall */
1209                 return -EAGAIN;
1210
1211         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1212         if (!buf)
1213                 return -EAGAIN;
1214
1215         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1216         if (!crq) {
1217                 kfree(buf);
1218                 return -EAGAIN;
1219         }
1220
1221         crq->q.reader = 0;
1222         crq->buf = buf;
1223         crq->len = 0;
1224         crq->readers = 0;
1225         spin_lock(&queue_lock);
1226         if (test_bit(CACHE_PENDING, &h->flags)) {
1227                 crq->item = cache_get(h);
1228                 list_add_tail(&crq->q.list, &detail->queue);
1229                 trace_cache_entry_upcall(detail, h);
1230         } else
1231                 /* Lost a race, no longer PENDING, so don't enqueue */
1232                 ret = -EAGAIN;
1233         spin_unlock(&queue_lock);
1234         wake_up(&queue_wait);
1235         if (ret == -EAGAIN) {
1236                 kfree(buf);
1237                 kfree(crq);
1238         }
1239         return ret;
1240 }
1241
1242 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1243 {
1244         if (test_and_set_bit(CACHE_PENDING, &h->flags))
1245                 return 0;
1246         return cache_pipe_upcall(detail, h);
1247 }
1248 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1249
1250 int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1251                                      struct cache_head *h)
1252 {
1253         if (!cache_listeners_exist(detail)) {
1254                 warn_no_listener(detail);
1255                 trace_cache_entry_no_listener(detail, h);
1256                 return -EINVAL;
1257         }
1258         return sunrpc_cache_pipe_upcall(detail, h);
1259 }
1260 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1261
1262 /*
1263  * parse a message from user-space and pass it
1264  * to an appropriate cache
1265  * Messages are, like requests, separated into fields by
1266  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1267  *
1268  * Message is
1269  *   reply cachename expiry key ... content....
1270  *
1271  * key and content are both parsed by cache
1272  */
1273
1274 int qword_get(char **bpp, char *dest, int bufsize)
1275 {
1276         /* return bytes copied, or -1 on error */
1277         char *bp = *bpp;
1278         int len = 0;
1279
1280         while (*bp == ' ') bp++;
1281
1282         if (bp[0] == '\\' && bp[1] == 'x') {
1283                 /* HEX STRING */
1284                 bp += 2;
1285                 while (len < bufsize - 1) {
1286                         int h, l;
1287
1288                         h = hex_to_bin(bp[0]);
1289                         if (h < 0)
1290                                 break;
1291
1292                         l = hex_to_bin(bp[1]);
1293                         if (l < 0)
1294                                 break;
1295
1296                         *dest++ = (h << 4) | l;
1297                         bp += 2;
1298                         len++;
1299                 }
1300         } else {
1301                 /* text with \nnn octal quoting */
1302                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1303                         if (*bp == '\\' &&
1304                             isodigit(bp[1]) && (bp[1] <= '3') &&
1305                             isodigit(bp[2]) &&
1306                             isodigit(bp[3])) {
1307                                 int byte = (*++bp -'0');
1308                                 bp++;
1309                                 byte = (byte << 3) | (*bp++ - '0');
1310                                 byte = (byte << 3) | (*bp++ - '0');
1311                                 *dest++ = byte;
1312                                 len++;
1313                         } else {
1314                                 *dest++ = *bp++;
1315                                 len++;
1316                         }
1317                 }
1318         }
1319
1320         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1321                 return -1;
1322         while (*bp == ' ') bp++;
1323         *bpp = bp;
1324         *dest = '\0';
1325         return len;
1326 }
1327 EXPORT_SYMBOL_GPL(qword_get);
1328
1329
1330 /*
1331  * support /proc/net/rpc/$CACHENAME/content
1332  * as a seqfile.
1333  * We call ->cache_show passing NULL for the item to
1334  * get a header, then pass each real item in the cache
1335  */
1336
1337 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1338 {
1339         loff_t n = *pos;
1340         unsigned int hash, entry;
1341         struct cache_head *ch;
1342         struct cache_detail *cd = m->private;
1343
1344         if (!n--)
1345                 return SEQ_START_TOKEN;
1346         hash = n >> 32;
1347         entry = n & ((1LL<<32) - 1);
1348
1349         hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1350                 if (!entry--)
1351                         return ch;
1352         n &= ~((1LL<<32) - 1);
1353         do {
1354                 hash++;
1355                 n += 1LL<<32;
1356         } while(hash < cd->hash_size &&
1357                 hlist_empty(&cd->hash_table[hash]));
1358         if (hash >= cd->hash_size)
1359                 return NULL;
1360         *pos = n+1;
1361         return hlist_entry_safe(rcu_dereference_raw(
1362                                 hlist_first_rcu(&cd->hash_table[hash])),
1363                                 struct cache_head, cache_list);
1364 }
1365
1366 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1367 {
1368         struct cache_head *ch = p;
1369         int hash = (*pos >> 32);
1370         struct cache_detail *cd = m->private;
1371
1372         if (p == SEQ_START_TOKEN)
1373                 hash = 0;
1374         else if (ch->cache_list.next == NULL) {
1375                 hash++;
1376                 *pos += 1LL<<32;
1377         } else {
1378                 ++*pos;
1379                 return hlist_entry_safe(rcu_dereference_raw(
1380                                         hlist_next_rcu(&ch->cache_list)),
1381                                         struct cache_head, cache_list);
1382         }
1383         *pos &= ~((1LL<<32) - 1);
1384         while (hash < cd->hash_size &&
1385                hlist_empty(&cd->hash_table[hash])) {
1386                 hash++;
1387                 *pos += 1LL<<32;
1388         }
1389         if (hash >= cd->hash_size)
1390                 return NULL;
1391         ++*pos;
1392         return hlist_entry_safe(rcu_dereference_raw(
1393                                 hlist_first_rcu(&cd->hash_table[hash])),
1394                                 struct cache_head, cache_list);
1395 }
1396
1397 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1398         __acquires(RCU)
1399 {
1400         rcu_read_lock();
1401         return __cache_seq_start(m, pos);
1402 }
1403 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1404
1405 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1406 {
1407         return cache_seq_next(file, p, pos);
1408 }
1409 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1410
1411 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1412         __releases(RCU)
1413 {
1414         rcu_read_unlock();
1415 }
1416 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1417
1418 static int c_show(struct seq_file *m, void *p)
1419 {
1420         struct cache_head *cp = p;
1421         struct cache_detail *cd = m->private;
1422
1423         if (p == SEQ_START_TOKEN)
1424                 return cd->cache_show(m, cd, NULL);
1425
1426         ifdebug(CACHE)
1427                 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1428                            convert_to_wallclock(cp->expiry_time),
1429                            kref_read(&cp->ref), cp->flags);
1430         if (!cache_get_rcu(cp))
1431                 return 0;
1432
1433         if (cache_check(cd, cp, NULL))
1434                 /* cache_check does a cache_put on failure */
1435                 seq_puts(m, "# ");
1436         else {
1437                 if (cache_is_expired(cd, cp))
1438                         seq_puts(m, "# ");
1439                 cache_put(cp, cd);
1440         }
1441
1442         return cd->cache_show(m, cd, cp);
1443 }
1444
1445 static const struct seq_operations cache_content_op = {
1446         .start  = cache_seq_start_rcu,
1447         .next   = cache_seq_next_rcu,
1448         .stop   = cache_seq_stop_rcu,
1449         .show   = c_show,
1450 };
1451
1452 static int content_open(struct inode *inode, struct file *file,
1453                         struct cache_detail *cd)
1454 {
1455         struct seq_file *seq;
1456         int err;
1457
1458         if (!cd || !try_module_get(cd->owner))
1459                 return -EACCES;
1460
1461         err = seq_open(file, &cache_content_op);
1462         if (err) {
1463                 module_put(cd->owner);
1464                 return err;
1465         }
1466
1467         seq = file->private_data;
1468         seq->private = cd;
1469         return 0;
1470 }
1471
1472 static int content_release(struct inode *inode, struct file *file,
1473                 struct cache_detail *cd)
1474 {
1475         int ret = seq_release(inode, file);
1476         module_put(cd->owner);
1477         return ret;
1478 }
1479
1480 static int open_flush(struct inode *inode, struct file *file,
1481                         struct cache_detail *cd)
1482 {
1483         if (!cd || !try_module_get(cd->owner))
1484                 return -EACCES;
1485         return nonseekable_open(inode, file);
1486 }
1487
1488 static int release_flush(struct inode *inode, struct file *file,
1489                         struct cache_detail *cd)
1490 {
1491         module_put(cd->owner);
1492         return 0;
1493 }
1494
1495 static ssize_t read_flush(struct file *file, char __user *buf,
1496                           size_t count, loff_t *ppos,
1497                           struct cache_detail *cd)
1498 {
1499         char tbuf[22];
1500         size_t len;
1501
1502         len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1503                         convert_to_wallclock(cd->flush_time));
1504         return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1505 }
1506
1507 static ssize_t write_flush(struct file *file, const char __user *buf,
1508                            size_t count, loff_t *ppos,
1509                            struct cache_detail *cd)
1510 {
1511         char tbuf[20];
1512         char *ep;
1513         time64_t now;
1514
1515         if (*ppos || count > sizeof(tbuf)-1)
1516                 return -EINVAL;
1517         if (copy_from_user(tbuf, buf, count))
1518                 return -EFAULT;
1519         tbuf[count] = 0;
1520         simple_strtoul(tbuf, &ep, 0);
1521         if (*ep && *ep != '\n')
1522                 return -EINVAL;
1523         /* Note that while we check that 'buf' holds a valid number,
1524          * we always ignore the value and just flush everything.
1525          * Making use of the number leads to races.
1526          */
1527
1528         now = seconds_since_boot();
1529         /* Always flush everything, so behave like cache_purge()
1530          * Do this by advancing flush_time to the current time,
1531          * or by one second if it has already reached the current time.
1532          * Newly added cache entries will always have ->last_refresh greater
1533          * that ->flush_time, so they don't get flushed prematurely.
1534          */
1535
1536         if (cd->flush_time >= now)
1537                 now = cd->flush_time + 1;
1538
1539         cd->flush_time = now;
1540         cd->nextcheck = now;
1541         cache_flush();
1542
1543         if (cd->flush)
1544                 cd->flush();
1545
1546         *ppos += count;
1547         return count;
1548 }
1549
1550 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1551                                  size_t count, loff_t *ppos)
1552 {
1553         struct cache_detail *cd = pde_data(file_inode(filp));
1554
1555         return cache_read(filp, buf, count, ppos, cd);
1556 }
1557
1558 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1559                                   size_t count, loff_t *ppos)
1560 {
1561         struct cache_detail *cd = pde_data(file_inode(filp));
1562
1563         return cache_write(filp, buf, count, ppos, cd);
1564 }
1565
1566 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1567 {
1568         struct cache_detail *cd = pde_data(file_inode(filp));
1569
1570         return cache_poll(filp, wait, cd);
1571 }
1572
1573 static long cache_ioctl_procfs(struct file *filp,
1574                                unsigned int cmd, unsigned long arg)
1575 {
1576         struct inode *inode = file_inode(filp);
1577         struct cache_detail *cd = pde_data(inode);
1578
1579         return cache_ioctl(inode, filp, cmd, arg, cd);
1580 }
1581
1582 static int cache_open_procfs(struct inode *inode, struct file *filp)
1583 {
1584         struct cache_detail *cd = pde_data(inode);
1585
1586         return cache_open(inode, filp, cd);
1587 }
1588
1589 static int cache_release_procfs(struct inode *inode, struct file *filp)
1590 {
1591         struct cache_detail *cd = pde_data(inode);
1592
1593         return cache_release(inode, filp, cd);
1594 }
1595
1596 static const struct proc_ops cache_channel_proc_ops = {
1597         .proc_read      = cache_read_procfs,
1598         .proc_write     = cache_write_procfs,
1599         .proc_poll      = cache_poll_procfs,
1600         .proc_ioctl     = cache_ioctl_procfs, /* for FIONREAD */
1601         .proc_open      = cache_open_procfs,
1602         .proc_release   = cache_release_procfs,
1603 };
1604
1605 static int content_open_procfs(struct inode *inode, struct file *filp)
1606 {
1607         struct cache_detail *cd = pde_data(inode);
1608
1609         return content_open(inode, filp, cd);
1610 }
1611
1612 static int content_release_procfs(struct inode *inode, struct file *filp)
1613 {
1614         struct cache_detail *cd = pde_data(inode);
1615
1616         return content_release(inode, filp, cd);
1617 }
1618
1619 static const struct proc_ops content_proc_ops = {
1620         .proc_open      = content_open_procfs,
1621         .proc_read      = seq_read,
1622         .proc_lseek     = seq_lseek,
1623         .proc_release   = content_release_procfs,
1624 };
1625
1626 static int open_flush_procfs(struct inode *inode, struct file *filp)
1627 {
1628         struct cache_detail *cd = pde_data(inode);
1629
1630         return open_flush(inode, filp, cd);
1631 }
1632
1633 static int release_flush_procfs(struct inode *inode, struct file *filp)
1634 {
1635         struct cache_detail *cd = pde_data(inode);
1636
1637         return release_flush(inode, filp, cd);
1638 }
1639
1640 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1641                             size_t count, loff_t *ppos)
1642 {
1643         struct cache_detail *cd = pde_data(file_inode(filp));
1644
1645         return read_flush(filp, buf, count, ppos, cd);
1646 }
1647
1648 static ssize_t write_flush_procfs(struct file *filp,
1649                                   const char __user *buf,
1650                                   size_t count, loff_t *ppos)
1651 {
1652         struct cache_detail *cd = pde_data(file_inode(filp));
1653
1654         return write_flush(filp, buf, count, ppos, cd);
1655 }
1656
1657 static const struct proc_ops cache_flush_proc_ops = {
1658         .proc_open      = open_flush_procfs,
1659         .proc_read      = read_flush_procfs,
1660         .proc_write     = write_flush_procfs,
1661         .proc_release   = release_flush_procfs,
1662 };
1663
1664 static void remove_cache_proc_entries(struct cache_detail *cd)
1665 {
1666         if (cd->procfs) {
1667                 proc_remove(cd->procfs);
1668                 cd->procfs = NULL;
1669         }
1670 }
1671
1672 #ifdef CONFIG_PROC_FS
1673 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1674 {
1675         struct proc_dir_entry *p;
1676         struct sunrpc_net *sn;
1677
1678         sn = net_generic(net, sunrpc_net_id);
1679         cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1680         if (cd->procfs == NULL)
1681                 goto out_nomem;
1682
1683         p = proc_create_data("flush", S_IFREG | 0600,
1684                              cd->procfs, &cache_flush_proc_ops, cd);
1685         if (p == NULL)
1686                 goto out_nomem;
1687
1688         if (cd->cache_request || cd->cache_parse) {
1689                 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1690                                      &cache_channel_proc_ops, cd);
1691                 if (p == NULL)
1692                         goto out_nomem;
1693         }
1694         if (cd->cache_show) {
1695                 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1696                                      &content_proc_ops, cd);
1697                 if (p == NULL)
1698                         goto out_nomem;
1699         }
1700         return 0;
1701 out_nomem:
1702         remove_cache_proc_entries(cd);
1703         return -ENOMEM;
1704 }
1705 #else /* CONFIG_PROC_FS */
1706 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1707 {
1708         return 0;
1709 }
1710 #endif
1711
1712 void __init cache_initialize(void)
1713 {
1714         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1715 }
1716
1717 int cache_register_net(struct cache_detail *cd, struct net *net)
1718 {
1719         int ret;
1720
1721         sunrpc_init_cache_detail(cd);
1722         ret = create_cache_proc_entries(cd, net);
1723         if (ret)
1724                 sunrpc_destroy_cache_detail(cd);
1725         return ret;
1726 }
1727 EXPORT_SYMBOL_GPL(cache_register_net);
1728
1729 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1730 {
1731         remove_cache_proc_entries(cd);
1732         sunrpc_destroy_cache_detail(cd);
1733 }
1734 EXPORT_SYMBOL_GPL(cache_unregister_net);
1735
1736 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1737 {
1738         struct cache_detail *cd;
1739         int i;
1740
1741         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1742         if (cd == NULL)
1743                 return ERR_PTR(-ENOMEM);
1744
1745         cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1746                                  GFP_KERNEL);
1747         if (cd->hash_table == NULL) {
1748                 kfree(cd);
1749                 return ERR_PTR(-ENOMEM);
1750         }
1751
1752         for (i = 0; i < cd->hash_size; i++)
1753                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1754         cd->net = net;
1755         return cd;
1756 }
1757 EXPORT_SYMBOL_GPL(cache_create_net);
1758
1759 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1760 {
1761         kfree(cd->hash_table);
1762         kfree(cd);
1763 }
1764 EXPORT_SYMBOL_GPL(cache_destroy_net);
1765
1766 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1767                                  size_t count, loff_t *ppos)
1768 {
1769         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1770
1771         return cache_read(filp, buf, count, ppos, cd);
1772 }
1773
1774 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1775                                   size_t count, loff_t *ppos)
1776 {
1777         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1778
1779         return cache_write(filp, buf, count, ppos, cd);
1780 }
1781
1782 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1783 {
1784         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1785
1786         return cache_poll(filp, wait, cd);
1787 }
1788
1789 static long cache_ioctl_pipefs(struct file *filp,
1790                               unsigned int cmd, unsigned long arg)
1791 {
1792         struct inode *inode = file_inode(filp);
1793         struct cache_detail *cd = RPC_I(inode)->private;
1794
1795         return cache_ioctl(inode, filp, cmd, arg, cd);
1796 }
1797
1798 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1799 {
1800         struct cache_detail *cd = RPC_I(inode)->private;
1801
1802         return cache_open(inode, filp, cd);
1803 }
1804
1805 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1806 {
1807         struct cache_detail *cd = RPC_I(inode)->private;
1808
1809         return cache_release(inode, filp, cd);
1810 }
1811
1812 const struct file_operations cache_file_operations_pipefs = {
1813         .owner          = THIS_MODULE,
1814         .read           = cache_read_pipefs,
1815         .write          = cache_write_pipefs,
1816         .poll           = cache_poll_pipefs,
1817         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1818         .open           = cache_open_pipefs,
1819         .release        = cache_release_pipefs,
1820 };
1821
1822 static int content_open_pipefs(struct inode *inode, struct file *filp)
1823 {
1824         struct cache_detail *cd = RPC_I(inode)->private;
1825
1826         return content_open(inode, filp, cd);
1827 }
1828
1829 static int content_release_pipefs(struct inode *inode, struct file *filp)
1830 {
1831         struct cache_detail *cd = RPC_I(inode)->private;
1832
1833         return content_release(inode, filp, cd);
1834 }
1835
1836 const struct file_operations content_file_operations_pipefs = {
1837         .open           = content_open_pipefs,
1838         .read           = seq_read,
1839         .llseek         = seq_lseek,
1840         .release        = content_release_pipefs,
1841 };
1842
1843 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1844 {
1845         struct cache_detail *cd = RPC_I(inode)->private;
1846
1847         return open_flush(inode, filp, cd);
1848 }
1849
1850 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1851 {
1852         struct cache_detail *cd = RPC_I(inode)->private;
1853
1854         return release_flush(inode, filp, cd);
1855 }
1856
1857 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1858                             size_t count, loff_t *ppos)
1859 {
1860         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1861
1862         return read_flush(filp, buf, count, ppos, cd);
1863 }
1864
1865 static ssize_t write_flush_pipefs(struct file *filp,
1866                                   const char __user *buf,
1867                                   size_t count, loff_t *ppos)
1868 {
1869         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1870
1871         return write_flush(filp, buf, count, ppos, cd);
1872 }
1873
1874 const struct file_operations cache_flush_operations_pipefs = {
1875         .open           = open_flush_pipefs,
1876         .read           = read_flush_pipefs,
1877         .write          = write_flush_pipefs,
1878         .release        = release_flush_pipefs,
1879 };
1880
1881 int sunrpc_cache_register_pipefs(struct dentry *parent,
1882                                  const char *name, umode_t umode,
1883                                  struct cache_detail *cd)
1884 {
1885         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1886         if (IS_ERR(dir))
1887                 return PTR_ERR(dir);
1888         cd->pipefs = dir;
1889         return 0;
1890 }
1891 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1892
1893 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1894 {
1895         if (cd->pipefs) {
1896                 rpc_remove_cache_dir(cd->pipefs);
1897                 cd->pipefs = NULL;
1898         }
1899 }
1900 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1901
1902 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1903 {
1904         spin_lock(&cd->hash_lock);
1905         if (!hlist_unhashed(&h->cache_list)){
1906                 sunrpc_begin_cache_remove_entry(h, cd);
1907                 spin_unlock(&cd->hash_lock);
1908                 sunrpc_end_cache_remove_entry(h, cd);
1909         } else
1910                 spin_unlock(&cd->hash_lock);
1911 }
1912 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);