net/sunrpc/cache.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * net/sunrpc/cache.c
   4  *
   5  * Generic code for various authentication-related caches
   6  * used by sunrpc clients and servers.
   7  *
   8  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   9  */
  10
  11 #include <linux/types.h>
  12 #include <linux/fs.h>
  13 #include <linux/file.h>
  14 #include <linux/slab.h>
  15 #include <linux/signal.h>
  16 #include <linux/sched.h>
  17 #include <linux/kmod.h>
  18 #include <linux/list.h>
  19 #include <linux/module.h>
  20 #include <linux/ctype.h>
  21 #include <linux/string_helpers.h>
  22 #include <linux/uaccess.h>
  23 #include <linux/poll.h>
  24 #include <linux/seq_file.h>
  25 #include <linux/proc_fs.h>
  26 #include <linux/net.h>
  27 #include <linux/workqueue.h>
  28 #include <linux/mutex.h>
  29 #include <linux/pagemap.h>
  30 #include <asm/ioctls.h>
  31 #include <linux/sunrpc/types.h>
  32 #include <linux/sunrpc/cache.h>
  33 #include <linux/sunrpc/stats.h>
  34 #include <linux/sunrpc/rpc_pipe_fs.h>
  35 #include <trace/events/sunrpc.h>
  36 #include "netns.h"
  37
  38 #define  RPCDBG_FACILITY RPCDBG_CACHE
  39
  40 static bool cache_defer_req(struct cache_req *req, struct cache_head *item);
  41 static void cache_revisit_request(struct cache_head *item);
  42
  43 static void cache_init(struct cache_head *h, struct cache_detail *detail)
  44 {
  45         time64_t now = seconds_since_boot();
  46         INIT_HLIST_NODE(&h->cache_list);
  47         h->flags = 0;
  48         kref_init(&h->ref);
  49         h->expiry_time = now + CACHE_NEW_EXPIRY;
  50         if (now <= detail->flush_time)
  51                 /* ensure it isn't already expired */
  52                 now = detail->flush_time + 1;
  53         h->last_refresh = now;
  54 }
  55
  56 static void cache_fresh_unlocked(struct cache_head *head,
  57                                 struct cache_detail *detail);
  58
  59 static struct cache_head *sunrpc_cache_find_rcu(struct cache_detail *detail,
  60                                                 struct cache_head *key,
  61                                                 int hash)
  62 {
  63         struct hlist_head *head = &detail->hash_table[hash];
  64         struct cache_head *tmp;
  65
  66         rcu_read_lock();
  67         hlist_for_each_entry_rcu(tmp, head, cache_list) {
  68                 if (!detail->match(tmp, key))
  69                         continue;
  70                 if (test_bit(CACHE_VALID, &tmp->flags) &&
  71                     cache_is_expired(detail, tmp))
  72                         continue;
  73                 tmp = cache_get_rcu(tmp);
  74                 rcu_read_unlock();
  75                 return tmp;
  76         }
  77         rcu_read_unlock();
  78         return NULL;
  79 }
  80
  81 static void sunrpc_begin_cache_remove_entry(struct cache_head *ch,
  82                                             struct cache_detail *cd)
  83 {
  84         /* Must be called under cd->hash_lock */
  85         hlist_del_init_rcu(&ch->cache_list);
  86         set_bit(CACHE_CLEANED, &ch->flags);
  87         cd->entries --;
  88 }
  89
  90 static void sunrpc_end_cache_remove_entry(struct cache_head *ch,
  91                                           struct cache_detail *cd)
  92 {
  93         cache_fresh_unlocked(ch, cd);
  94         cache_put(ch, cd);
  95 }
  96
  97 static struct cache_head *sunrpc_cache_add_entry(struct cache_detail *detail,
  98                                                  struct cache_head *key,
  99                                                  int hash)
 100 {
 101         struct cache_head *new, *tmp, *freeme = NULL;
 102         struct hlist_head *head = &detail->hash_table[hash];
 103
 104         new = detail->alloc();
 105         if (!new)
 106                 return NULL;
 107         /* must fully initialise 'new', else
 108          * we might get lose if we need to
 109          * cache_put it soon.
 110          */
 111         cache_init(new, detail);
 112         detail->init(new, key);
 113
 114         spin_lock(&detail->hash_lock);
 115
 116         /* check if entry appeared while we slept */
 117         hlist_for_each_entry_rcu(tmp, head, cache_list,
 118                                  lockdep_is_held(&detail->hash_lock)) {
 119                 if (!detail->match(tmp, key))
 120                         continue;
 121                 if (test_bit(CACHE_VALID, &tmp->flags) &&
 122                     cache_is_expired(detail, tmp)) {
 123                         sunrpc_begin_cache_remove_entry(tmp, detail);
 124                         trace_cache_entry_expired(detail, tmp);
 125                         freeme = tmp;
 126                         break;
 127                 }
 128                 cache_get(tmp);
 129                 spin_unlock(&detail->hash_lock);
 130                 cache_put(new, detail);
 131                 return tmp;
 132         }
 133
 134         hlist_add_head_rcu(&new->cache_list, head);
 135         detail->entries++;
 136         cache_get(new);
 137         spin_unlock(&detail->hash_lock);
 138
 139         if (freeme)
 140                 sunrpc_end_cache_remove_entry(freeme, detail);
 141         return new;
 142 }
 143
 144 struct cache_head *sunrpc_cache_lookup_rcu(struct cache_detail *detail,
 145                                            struct cache_head *key, int hash)
 146 {
 147         struct cache_head *ret;
 148
 149         ret = sunrpc_cache_find_rcu(detail, key, hash);
 150         if (ret)
 151                 return ret;
 152         /* Didn't find anything, insert an empty entry */
 153         return sunrpc_cache_add_entry(detail, key, hash);
 154 }
 155 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup_rcu);
 156
 157 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 158
 159 static void cache_fresh_locked(struct cache_head *head, time64_t expiry,
 160                                struct cache_detail *detail)
 161 {
 162         time64_t now = seconds_since_boot();
 163         if (now <= detail->flush_time)
 164                 /* ensure it isn't immediately treated as expired */
 165                 now = detail->flush_time + 1;
 166         head->expiry_time = expiry;
 167         head->last_refresh = now;
 168         smp_wmb(); /* paired with smp_rmb() in cache_is_valid() */
 169         set_bit(CACHE_VALID, &head->flags);
 170 }
 171
 172 static void cache_fresh_unlocked(struct cache_head *head,
 173                                  struct cache_detail *detail)
 174 {
 175         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 176                 cache_revisit_request(head);
 177                 cache_dequeue(detail, head);
 178         }
 179 }
 180
 181 static void cache_make_negative(struct cache_detail *detail,
 182                                 struct cache_head *h)
 183 {
 184         set_bit(CACHE_NEGATIVE, &h->flags);
 185         trace_cache_entry_make_negative(detail, h);
 186 }
 187
 188 static void cache_entry_update(struct cache_detail *detail,
 189                                struct cache_head *h,
 190                                struct cache_head *new)
 191 {
 192         if (!test_bit(CACHE_NEGATIVE, &new->flags)) {
 193                 detail->update(h, new);
 194                 trace_cache_entry_update(detail, h);
 195         } else {
 196                 cache_make_negative(detail, h);
 197         }
 198 }
 199
 200 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 201                                        struct cache_head *new, struct cache_head *old, int hash)
 202 {
 203         /* The 'old' entry is to be replaced by 'new'.
 204          * If 'old' is not VALID, we update it directly,
 205          * otherwise we need to replace it
 206          */
 207         struct cache_head *tmp;
 208
 209         if (!test_bit(CACHE_VALID, &old->flags)) {
 210                 spin_lock(&detail->hash_lock);
 211                 if (!test_bit(CACHE_VALID, &old->flags)) {
 212                         cache_entry_update(detail, old, new);
 213                         cache_fresh_locked(old, new->expiry_time, detail);
 214                         spin_unlock(&detail->hash_lock);
 215                         cache_fresh_unlocked(old, detail);
 216                         return old;
 217                 }
 218                 spin_unlock(&detail->hash_lock);
 219         }
 220         /* We need to insert a new entry */
 221         tmp = detail->alloc();
 222         if (!tmp) {
 223                 cache_put(old, detail);
 224                 return NULL;
 225         }
 226         cache_init(tmp, detail);
 227         detail->init(tmp, old);
 228
 229         spin_lock(&detail->hash_lock);
 230         cache_entry_update(detail, tmp, new);
 231         hlist_add_head(&tmp->cache_list, &detail->hash_table[hash]);
 232         detail->entries++;
 233         cache_get(tmp);
 234         cache_fresh_locked(tmp, new->expiry_time, detail);
 235         cache_fresh_locked(old, 0, detail);
 236         spin_unlock(&detail->hash_lock);
 237         cache_fresh_unlocked(tmp, detail);
 238         cache_fresh_unlocked(old, detail);
 239         cache_put(old, detail);
 240         return tmp;
 241 }
 242 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 243
 244 static inline int cache_is_valid(struct cache_head *h)
 245 {
 246         if (!test_bit(CACHE_VALID, &h->flags))
 247                 return -EAGAIN;
 248         else {
 249                 /* entry is valid */
 250                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 251                         return -ENOENT;
 252                 else {
 253                         /*
 254                          * In combination with write barrier in
 255                          * sunrpc_cache_update, ensures that anyone
 256                          * using the cache entry after this sees the
 257                          * updated contents:
 258                          */
 259                         smp_rmb();
 260                         return 0;
 261                 }
 262         }
 263 }
 264
 265 static int try_to_negate_entry(struct cache_detail *detail, struct cache_head *h)
 266 {
 267         int rv;
 268
 269         spin_lock(&detail->hash_lock);
 270         rv = cache_is_valid(h);
 271         if (rv == -EAGAIN) {
 272                 cache_make_negative(detail, h);
 273                 cache_fresh_locked(h, seconds_since_boot()+CACHE_NEW_EXPIRY,
 274                                    detail);
 275                 rv = -ENOENT;
 276         }
 277         spin_unlock(&detail->hash_lock);
 278         cache_fresh_unlocked(h, detail);
 279         return rv;
 280 }
 281
 282 /*
 283  * This is the generic cache management routine for all
 284  * the authentication caches.
 285  * It checks the currency of a cache item and will (later)
 286  * initiate an upcall to fill it if needed.
 287  *
 288  *
 289  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 290  * -EAGAIN if upcall is pending and request has been queued
 291  * -ETIMEDOUT if upcall failed or request could not be queue or
 292  *           upcall completed but item is still invalid (implying that
 293  *           the cache item has been replaced with a newer one).
 294  * -ENOENT if cache entry was negative
 295  */
 296 int cache_check(struct cache_detail *detail,
 297                     struct cache_head *h, struct cache_req *rqstp)
 298 {
 299         int rv;
 300         time64_t refresh_age, age;
 301
 302         /* First decide return status as best we can */
 303         rv = cache_is_valid(h);
 304
 305         /* now see if we want to start an upcall */
 306         refresh_age = (h->expiry_time - h->last_refresh);
 307         age = seconds_since_boot() - h->last_refresh;
 308
 309         if (rqstp == NULL) {
 310                 if (rv == -EAGAIN)
 311                         rv = -ENOENT;
 312         } else if (rv == -EAGAIN ||
 313                    (h->expiry_time != 0 && age > refresh_age/2)) {
 314                 dprintk("RPC:       Want update, refage=%lld, age=%lld\n",
 315                                 refresh_age, age);
 316                 switch (detail->cache_upcall(detail, h)) {
 317                 case -EINVAL:
 318                         rv = try_to_negate_entry(detail, h);
 319                         break;
 320                 case -EAGAIN:
 321                         cache_fresh_unlocked(h, detail);
 322                         break;
 323                 }
 324         }
 325
 326         if (rv == -EAGAIN) {
 327                 if (!cache_defer_req(rqstp, h)) {
 328                         /*
 329                          * Request was not deferred; handle it as best
 330                          * we can ourselves:
 331                          */
 332                         rv = cache_is_valid(h);
 333                         if (rv == -EAGAIN)
 334                                 rv = -ETIMEDOUT;
 335                 }
 336         }
 337         if (rv)
 338                 cache_put(h, detail);
 339         return rv;
 340 }
 341 EXPORT_SYMBOL_GPL(cache_check);
 342
 343 /*
 344  * caches need to be periodically cleaned.
 345  * For this we maintain a list of cache_detail and
 346  * a current pointer into that list and into the table
 347  * for that entry.
 348  *
 349  * Each time cache_clean is called it finds the next non-empty entry
 350  * in the current table and walks the list in that entry
 351  * looking for entries that can be removed.
 352  *
 353  * An entry gets removed if:
 354  * - The expiry is before current time
 355  * - The last_refresh time is before the flush_time for that cache
 356  *
 357  * later we might drop old entries with non-NEVER expiry if that table
 358  * is getting 'full' for some definition of 'full'
 359  *
 360  * The question of "how often to scan a table" is an interesting one
 361  * and is answered in part by the use of the "nextcheck" field in the
 362  * cache_detail.
 363  * When a scan of a table begins, the nextcheck field is set to a time
 364  * that is well into the future.
 365  * While scanning, if an expiry time is found that is earlier than the
 366  * current nextcheck time, nextcheck is set to that expiry time.
 367  * If the flush_time is ever set to a time earlier than the nextcheck
 368  * time, the nextcheck time is then set to that flush_time.
 369  *
 370  * A table is then only scanned if the current time is at least
 371  * the nextcheck time.
 372  *
 373  */
 374
 375 static LIST_HEAD(cache_list);
 376 static DEFINE_SPINLOCK(cache_list_lock);
 377 static struct cache_detail *current_detail;
 378 static int current_index;
 379
 380 static void do_cache_clean(struct work_struct *work);
 381 static struct delayed_work cache_cleaner;
 382
 383 void sunrpc_init_cache_detail(struct cache_detail *cd)
 384 {
 385         spin_lock_init(&cd->hash_lock);
 386         INIT_LIST_HEAD(&cd->queue);
 387         spin_lock(&cache_list_lock);
 388         cd->nextcheck = 0;
 389         cd->entries = 0;
 390         atomic_set(&cd->writers, 0);
 391         cd->last_close = 0;
 392         cd->last_warn = -1;
 393         list_add(&cd->others, &cache_list);
 394         spin_unlock(&cache_list_lock);
 395
 396         /* start the cleaning process */
 397         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, 0);
 398 }
 399 EXPORT_SYMBOL_GPL(sunrpc_init_cache_detail);
 400
 401 void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 402 {
 403         cache_purge(cd);
 404         spin_lock(&cache_list_lock);
 405         spin_lock(&cd->hash_lock);
 406         if (current_detail == cd)
 407                 current_detail = NULL;
 408         list_del_init(&cd->others);
 409         spin_unlock(&cd->hash_lock);
 410         spin_unlock(&cache_list_lock);
 411         if (list_empty(&cache_list)) {
 412                 /* module must be being unloaded so its safe to kill the worker */
 413                 cancel_delayed_work_sync(&cache_cleaner);
 414         }
 415 }
 416 EXPORT_SYMBOL_GPL(sunrpc_destroy_cache_detail);
 417
 418 /* clean cache tries to find something to clean
 419  * and cleans it.
 420  * It returns 1 if it cleaned something,
 421  *            0 if it didn't find anything this time
 422  *           -1 if it fell off the end of the list.
 423  */
 424 static int cache_clean(void)
 425 {
 426         int rv = 0;
 427         struct list_head *next;
 428
 429         spin_lock(&cache_list_lock);
 430
 431         /* find a suitable table if we don't already have one */
 432         while (current_detail == NULL ||
 433             current_index >= current_detail->hash_size) {
 434                 if (current_detail)
 435                         next = current_detail->others.next;
 436                 else
 437                         next = cache_list.next;
 438                 if (next == &cache_list) {
 439                         current_detail = NULL;
 440                         spin_unlock(&cache_list_lock);
 441                         return -1;
 442                 }
 443                 current_detail = list_entry(next, struct cache_detail, others);
 444                 if (current_detail->nextcheck > seconds_since_boot())
 445                         current_index = current_detail->hash_size;
 446                 else {
 447                         current_index = 0;
 448                         current_detail->nextcheck = seconds_since_boot()+30*60;
 449                 }
 450         }
 451
 452         /* find a non-empty bucket in the table */
 453         while (current_detail &&
 454                current_index < current_detail->hash_size &&
 455                hlist_empty(&current_detail->hash_table[current_index]))
 456                 current_index++;
 457
 458         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 459
 460         if (current_detail && current_index < current_detail->hash_size) {
 461                 struct cache_head *ch = NULL;
 462                 struct cache_detail *d;
 463                 struct hlist_head *head;
 464                 struct hlist_node *tmp;
 465
 466                 spin_lock(&current_detail->hash_lock);
 467
 468                 /* Ok, now to clean this strand */
 469
 470                 head = &current_detail->hash_table[current_index];
 471                 hlist_for_each_entry_safe(ch, tmp, head, cache_list) {
 472                         if (current_detail->nextcheck > ch->expiry_time)
 473                                 current_detail->nextcheck = ch->expiry_time+1;
 474                         if (!cache_is_expired(current_detail, ch))
 475                                 continue;
 476
 477                         sunrpc_begin_cache_remove_entry(ch, current_detail);
 478                         trace_cache_entry_expired(current_detail, ch);
 479                         rv = 1;
 480                         break;
 481                 }
 482
 483                 spin_unlock(&current_detail->hash_lock);
 484                 d = current_detail;
 485                 if (!ch)
 486                         current_index ++;
 487                 spin_unlock(&cache_list_lock);
 488                 if (ch)
 489                         sunrpc_end_cache_remove_entry(ch, d);
 490         } else
 491                 spin_unlock(&cache_list_lock);
 492
 493         return rv;
 494 }
 495
 496 /*
 497  * We want to regularly clean the cache, so we need to schedule some work ...
 498  */
 499 static void do_cache_clean(struct work_struct *work)
 500 {
 501         int delay;
 502
 503         if (list_empty(&cache_list))
 504                 return;
 505
 506         if (cache_clean() == -1)
 507                 delay = round_jiffies_relative(30*HZ);
 508         else
 509                 delay = 5;
 510
 511         queue_delayed_work(system_power_efficient_wq, &cache_cleaner, delay);
 512 }
 513
 514
 515 /*
 516  * Clean all caches promptly.  This just calls cache_clean
 517  * repeatedly until we are sure that every cache has had a chance to
 518  * be fully cleaned
 519  */
 520 void cache_flush(void)
 521 {
 522         while (cache_clean() != -1)
 523                 cond_resched();
 524         while (cache_clean() != -1)
 525                 cond_resched();
 526 }
 527 EXPORT_SYMBOL_GPL(cache_flush);
 528
 529 void cache_purge(struct cache_detail *detail)
 530 {
 531         struct cache_head *ch = NULL;
 532         struct hlist_head *head = NULL;
 533         int i = 0;
 534
 535         spin_lock(&detail->hash_lock);
 536         if (!detail->entries) {
 537                 spin_unlock(&detail->hash_lock);
 538                 return;
 539         }
 540
 541         dprintk("RPC: %d entries in %s cache\n", detail->entries, detail->name);
 542         for (i = 0; i < detail->hash_size; i++) {
 543                 head = &detail->hash_table[i];
 544                 while (!hlist_empty(head)) {
 545                         ch = hlist_entry(head->first, struct cache_head,
 546                                          cache_list);
 547                         sunrpc_begin_cache_remove_entry(ch, detail);
 548                         spin_unlock(&detail->hash_lock);
 549                         sunrpc_end_cache_remove_entry(ch, detail);
 550                         spin_lock(&detail->hash_lock);
 551                 }
 552         }
 553         spin_unlock(&detail->hash_lock);
 554 }
 555 EXPORT_SYMBOL_GPL(cache_purge);
 556
 557
 558 /*
 559  * Deferral and Revisiting of Requests.
 560  *
 561  * If a cache lookup finds a pending entry, we
 562  * need to defer the request and revisit it later.
 563  * All deferred requests are stored in a hash table,
 564  * indexed by "struct cache_head *".
 565  * As it may be wasteful to store a whole request
 566  * structure, we allow the request to provide a
 567  * deferred form, which must contain a
 568  * 'struct cache_deferred_req'
 569  * This cache_deferred_req contains a method to allow
 570  * it to be revisited when cache info is available
 571  */
 572
 573 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 574 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 575
 576 #define DFR_MAX 300     /* ??? */
 577
 578 static DEFINE_SPINLOCK(cache_defer_lock);
 579 static LIST_HEAD(cache_defer_list);
 580 static struct hlist_head cache_defer_hash[DFR_HASHSIZE];
 581 static int cache_defer_cnt;
 582
 583 static void __unhash_deferred_req(struct cache_deferred_req *dreq)
 584 {
 585         hlist_del_init(&dreq->hash);
 586         if (!list_empty(&dreq->recent)) {
 587                 list_del_init(&dreq->recent);
 588                 cache_defer_cnt--;
 589         }
 590 }
 591
 592 static void __hash_deferred_req(struct cache_deferred_req *dreq, struct cache_head *item)
 593 {
 594         int hash = DFR_HASH(item);
 595
 596         INIT_LIST_HEAD(&dreq->recent);
 597         hlist_add_head(&dreq->hash, &cache_defer_hash[hash]);
 598 }
 599
 600 static void setup_deferral(struct cache_deferred_req *dreq,
 601                            struct cache_head *item,
 602                            int count_me)
 603 {
 604
 605         dreq->item = item;
 606
 607         spin_lock(&cache_defer_lock);
 608
 609         __hash_deferred_req(dreq, item);
 610
 611         if (count_me) {
 612                 cache_defer_cnt++;
 613                 list_add(&dreq->recent, &cache_defer_list);
 614         }
 615
 616         spin_unlock(&cache_defer_lock);
 617
 618 }
 619
 620 struct thread_deferred_req {
 621         struct cache_deferred_req handle;
 622         struct completion completion;
 623 };
 624
 625 static void cache_restart_thread(struct cache_deferred_req *dreq, int too_many)
 626 {
 627         struct thread_deferred_req *dr =
 628                 container_of(dreq, struct thread_deferred_req, handle);
 629         complete(&dr->completion);
 630 }
 631
 632 static void cache_wait_req(struct cache_req *req, struct cache_head *item)
 633 {
 634         struct thread_deferred_req sleeper;
 635         struct cache_deferred_req *dreq = &sleeper.handle;
 636
 637         sleeper.completion = COMPLETION_INITIALIZER_ONSTACK(sleeper.completion);
 638         dreq->revisit = cache_restart_thread;
 639
 640         setup_deferral(dreq, item, 0);
 641
 642         if (!test_bit(CACHE_PENDING, &item->flags) ||
 643             wait_for_completion_interruptible_timeout(
 644                     &sleeper.completion, req->thread_wait) <= 0) {
 645                 /* The completion wasn't completed, so we need
 646                  * to clean up
 647                  */
 648                 spin_lock(&cache_defer_lock);
 649                 if (!hlist_unhashed(&sleeper.handle.hash)) {
 650                         __unhash_deferred_req(&sleeper.handle);
 651                         spin_unlock(&cache_defer_lock);
 652                 } else {
 653                         /* cache_revisit_request already removed
 654                          * this from the hash table, but hasn't
 655                          * called ->revisit yet.  It will very soon
 656                          * and we need to wait for it.
 657                          */
 658                         spin_unlock(&cache_defer_lock);
 659                         wait_for_completion(&sleeper.completion);
 660                 }
 661         }
 662 }
 663
 664 static void cache_limit_defers(void)
 665 {
 666         /* Make sure we haven't exceed the limit of allowed deferred
 667          * requests.
 668          */
 669         struct cache_deferred_req *discard = NULL;
 670
 671         if (cache_defer_cnt <= DFR_MAX)
 672                 return;
 673
 674         spin_lock(&cache_defer_lock);
 675
 676         /* Consider removing either the first or the last */
 677         if (cache_defer_cnt > DFR_MAX) {
 678                 if (prandom_u32() & 1)
 679                         discard = list_entry(cache_defer_list.next,
 680                                              struct cache_deferred_req, recent);
 681                 else
 682                         discard = list_entry(cache_defer_list.prev,
 683                                              struct cache_deferred_req, recent);
 684                 __unhash_deferred_req(discard);
 685         }
 686         spin_unlock(&cache_defer_lock);
 687         if (discard)
 688                 discard->revisit(discard, 1);
 689 }
 690
 691 /* Return true if and only if a deferred request is queued. */
 692 static bool cache_defer_req(struct cache_req *req, struct cache_head *item)
 693 {
 694         struct cache_deferred_req *dreq;
 695
 696         if (req->thread_wait) {
 697                 cache_wait_req(req, item);
 698                 if (!test_bit(CACHE_PENDING, &item->flags))
 699                         return false;
 700         }
 701         dreq = req->defer(req);
 702         if (dreq == NULL)
 703                 return false;
 704         setup_deferral(dreq, item, 1);
 705         if (!test_bit(CACHE_PENDING, &item->flags))
 706                 /* Bit could have been cleared before we managed to
 707                  * set up the deferral, so need to revisit just in case
 708                  */
 709                 cache_revisit_request(item);
 710
 711         cache_limit_defers();
 712         return true;
 713 }
 714
 715 static void cache_revisit_request(struct cache_head *item)
 716 {
 717         struct cache_deferred_req *dreq;
 718         struct list_head pending;
 719         struct hlist_node *tmp;
 720         int hash = DFR_HASH(item);
 721
 722         INIT_LIST_HEAD(&pending);
 723         spin_lock(&cache_defer_lock);
 724
 725         hlist_for_each_entry_safe(dreq, tmp, &cache_defer_hash[hash], hash)
 726                 if (dreq->item == item) {
 727                         __unhash_deferred_req(dreq);
 728                         list_add(&dreq->recent, &pending);
 729                 }
 730
 731         spin_unlock(&cache_defer_lock);
 732
 733         while (!list_empty(&pending)) {
 734                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 735                 list_del_init(&dreq->recent);
 736                 dreq->revisit(dreq, 0);
 737         }
 738 }
 739
 740 void cache_clean_deferred(void *owner)
 741 {
 742         struct cache_deferred_req *dreq, *tmp;
 743         struct list_head pending;
 744
 745
 746         INIT_LIST_HEAD(&pending);
 747         spin_lock(&cache_defer_lock);
 748
 749         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 750                 if (dreq->owner == owner) {
 751                         __unhash_deferred_req(dreq);
 752                         list_add(&dreq->recent, &pending);
 753                 }
 754         }
 755         spin_unlock(&cache_defer_lock);
 756
 757         while (!list_empty(&pending)) {
 758                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 759                 list_del_init(&dreq->recent);
 760                 dreq->revisit(dreq, 1);
 761         }
 762 }
 763
 764 /*
 765  * communicate with user-space
 766  *
 767  * We have a magic /proc file - /proc/net/rpc/<cachename>/channel.
 768  * On read, you get a full request, or block.
 769  * On write, an update request is processed.
 770  * Poll works if anything to read, and always allows write.
 771  *
 772  * Implemented by linked list of requests.  Each open file has
 773  * a ->private that also exists in this list.  New requests are added
 774  * to the end and may wakeup and preceding readers.
 775  * New readers are added to the head.  If, on read, an item is found with
 776  * CACHE_UPCALLING clear, we free it from the list.
 777  *
 778  */
 779
 780 static DEFINE_SPINLOCK(queue_lock);
 781
 782 struct cache_queue {
 783         struct list_head        list;
 784         int                     reader; /* if 0, then request */
 785 };
 786 struct cache_request {
 787         struct cache_queue      q;
 788         struct cache_head       *item;
 789         char                    * buf;
 790         int                     len;
 791         int                     readers;
 792 };
 793 struct cache_reader {
 794         struct cache_queue      q;
 795         int                     offset; /* if non-0, we have a refcnt on next request */
 796 };
 797
 798 static int cache_request(struct cache_detail *detail,
 799                                struct cache_request *crq)
 800 {
 801         char *bp = crq->buf;
 802         int len = PAGE_SIZE;
 803
 804         detail->cache_request(detail, crq->item, &bp, &len);
 805         if (len < 0)
 806                 return -E2BIG;
 807         return PAGE_SIZE - len;
 808 }
 809
 810 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 811                           loff_t *ppos, struct cache_detail *cd)
 812 {
 813         struct cache_reader *rp = filp->private_data;
 814         struct cache_request *rq;
 815         struct inode *inode = file_inode(filp);
 816         int err;
 817
 818         if (count == 0)
 819                 return 0;
 820
 821         inode_lock(inode); /* protect against multiple concurrent
 822                               * readers on this file */
 823  again:
 824         spin_lock(&queue_lock);
 825         /* need to find next request */
 826         while (rp->q.list.next != &cd->queue &&
 827                list_entry(rp->q.list.next, struct cache_queue, list)
 828                ->reader) {
 829                 struct list_head *next = rp->q.list.next;
 830                 list_move(&rp->q.list, next);
 831         }
 832         if (rp->q.list.next == &cd->queue) {
 833                 spin_unlock(&queue_lock);
 834                 inode_unlock(inode);
 835                 WARN_ON_ONCE(rp->offset);
 836                 return 0;
 837         }
 838         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 839         WARN_ON_ONCE(rq->q.reader);
 840         if (rp->offset == 0)
 841                 rq->readers++;
 842         spin_unlock(&queue_lock);
 843
 844         if (rq->len == 0) {
 845                 err = cache_request(cd, rq);
 846                 if (err < 0)
 847                         goto out;
 848                 rq->len = err;
 849         }
 850
 851         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 852                 err = -EAGAIN;
 853                 spin_lock(&queue_lock);
 854                 list_move(&rp->q.list, &rq->q.list);
 855                 spin_unlock(&queue_lock);
 856         } else {
 857                 if (rp->offset + count > rq->len)
 858                         count = rq->len - rp->offset;
 859                 err = -EFAULT;
 860                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 861                         goto out;
 862                 rp->offset += count;
 863                 if (rp->offset >= rq->len) {
 864                         rp->offset = 0;
 865                         spin_lock(&queue_lock);
 866                         list_move(&rp->q.list, &rq->q.list);
 867                         spin_unlock(&queue_lock);
 868                 }
 869                 err = 0;
 870         }
 871  out:
 872         if (rp->offset == 0) {
 873                 /* need to release rq */
 874                 spin_lock(&queue_lock);
 875                 rq->readers--;
 876                 if (rq->readers == 0 &&
 877                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 878                         list_del(&rq->q.list);
 879                         spin_unlock(&queue_lock);
 880                         cache_put(rq->item, cd);
 881                         kfree(rq->buf);
 882                         kfree(rq);
 883                 } else
 884                         spin_unlock(&queue_lock);
 885         }
 886         if (err == -EAGAIN)
 887                 goto again;
 888         inode_unlock(inode);
 889         return err ? err :  count;
 890 }
 891
 892 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 893                                  size_t count, struct cache_detail *cd)
 894 {
 895         ssize_t ret;
 896
 897         if (count == 0)
 898                 return -EINVAL;
 899         if (copy_from_user(kaddr, buf, count))
 900                 return -EFAULT;
 901         kaddr[count] = '\0';
 902         ret = cd->cache_parse(cd, kaddr, count);
 903         if (!ret)
 904                 ret = count;
 905         return ret;
 906 }
 907
 908 static ssize_t cache_downcall(struct address_space *mapping,
 909                               const char __user *buf,
 910                               size_t count, struct cache_detail *cd)
 911 {
 912         char *write_buf;
 913         ssize_t ret = -ENOMEM;
 914
 915         if (count >= 32768) { /* 32k is max userland buffer, lets check anyway */
 916                 ret = -EINVAL;
 917                 goto out;
 918         }
 919
 920         write_buf = kvmalloc(count + 1, GFP_KERNEL);
 921         if (!write_buf)
 922                 goto out;
 923
 924         ret = cache_do_downcall(write_buf, buf, count, cd);
 925         kvfree(write_buf);
 926 out:
 927         return ret;
 928 }
 929
 930 static ssize_t cache_write(struct file *filp, const char __user *buf,
 931                            size_t count, loff_t *ppos,
 932                            struct cache_detail *cd)
 933 {
 934         struct address_space *mapping = filp->f_mapping;
 935         struct inode *inode = file_inode(filp);
 936         ssize_t ret = -EINVAL;
 937
 938         if (!cd->cache_parse)
 939                 goto out;
 940
 941         inode_lock(inode);
 942         ret = cache_downcall(mapping, buf, count, cd);
 943         inode_unlock(inode);
 944 out:
 945         return ret;
 946 }
 947
 948 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 949
 950 static __poll_t cache_poll(struct file *filp, poll_table *wait,
 951                                struct cache_detail *cd)
 952 {
 953         __poll_t mask;
 954         struct cache_reader *rp = filp->private_data;
 955         struct cache_queue *cq;
 956
 957         poll_wait(filp, &queue_wait, wait);
 958
 959         /* alway allow write */
 960         mask = EPOLLOUT | EPOLLWRNORM;
 961
 962         if (!rp)
 963                 return mask;
 964
 965         spin_lock(&queue_lock);
 966
 967         for (cq= &rp->q; &cq->list != &cd->queue;
 968              cq = list_entry(cq->list.next, struct cache_queue, list))
 969                 if (!cq->reader) {
 970                         mask |= EPOLLIN | EPOLLRDNORM;
 971                         break;
 972                 }
 973         spin_unlock(&queue_lock);
 974         return mask;
 975 }
 976
 977 static int cache_ioctl(struct inode *ino, struct file *filp,
 978                        unsigned int cmd, unsigned long arg,
 979                        struct cache_detail *cd)
 980 {
 981         int len = 0;
 982         struct cache_reader *rp = filp->private_data;
 983         struct cache_queue *cq;
 984
 985         if (cmd != FIONREAD || !rp)
 986                 return -EINVAL;
 987
 988         spin_lock(&queue_lock);
 989
 990         /* only find the length remaining in current request,
 991          * or the length of the next request
 992          */
 993         for (cq= &rp->q; &cq->list != &cd->queue;
 994              cq = list_entry(cq->list.next, struct cache_queue, list))
 995                 if (!cq->reader) {
 996                         struct cache_request *cr =
 997                                 container_of(cq, struct cache_request, q);
 998                         len = cr->len - rp->offset;
 999                         break;
1000                 }
1001         spin_unlock(&queue_lock);
1002
1003         return put_user(len, (int __user *)arg);
1004 }
1005
1006 static int cache_open(struct inode *inode, struct file *filp,
1007                       struct cache_detail *cd)
1008 {
1009         struct cache_reader *rp = NULL;
1010
1011         if (!cd || !try_module_get(cd->owner))
1012                 return -EACCES;
1013         nonseekable_open(inode, filp);
1014         if (filp->f_mode & FMODE_READ) {
1015                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
1016                 if (!rp) {
1017                         module_put(cd->owner);
1018                         return -ENOMEM;
1019                 }
1020                 rp->offset = 0;
1021                 rp->q.reader = 1;
1022
1023                 spin_lock(&queue_lock);
1024                 list_add(&rp->q.list, &cd->queue);
1025                 spin_unlock(&queue_lock);
1026         }
1027         if (filp->f_mode & FMODE_WRITE)
1028                 atomic_inc(&cd->writers);
1029         filp->private_data = rp;
1030         return 0;
1031 }
1032
1033 static int cache_release(struct inode *inode, struct file *filp,
1034                          struct cache_detail *cd)
1035 {
1036         struct cache_reader *rp = filp->private_data;
1037
1038         if (rp) {
1039                 spin_lock(&queue_lock);
1040                 if (rp->offset) {
1041                         struct cache_queue *cq;
1042                         for (cq= &rp->q; &cq->list != &cd->queue;
1043                              cq = list_entry(cq->list.next, struct cache_queue, list))
1044                                 if (!cq->reader) {
1045                                         container_of(cq, struct cache_request, q)
1046                                                 ->readers--;
1047                                         break;
1048                                 }
1049                         rp->offset = 0;
1050                 }
1051                 list_del(&rp->q.list);
1052                 spin_unlock(&queue_lock);
1053
1054                 filp->private_data = NULL;
1055                 kfree(rp);
1056
1057         }
1058         if (filp->f_mode & FMODE_WRITE) {
1059                 atomic_dec(&cd->writers);
1060                 cd->last_close = seconds_since_boot();
1061         }
1062         module_put(cd->owner);
1063         return 0;
1064 }
1065
1066
1067
1068 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
1069 {
1070         struct cache_queue *cq, *tmp;
1071         struct cache_request *cr;
1072         struct list_head dequeued;
1073
1074         INIT_LIST_HEAD(&dequeued);
1075         spin_lock(&queue_lock);
1076         list_for_each_entry_safe(cq, tmp, &detail->queue, list)
1077                 if (!cq->reader) {
1078                         cr = container_of(cq, struct cache_request, q);
1079                         if (cr->item != ch)
1080                                 continue;
1081                         if (test_bit(CACHE_PENDING, &ch->flags))
1082                                 /* Lost a race and it is pending again */
1083                                 break;
1084                         if (cr->readers != 0)
1085                                 continue;
1086                         list_move(&cr->q.list, &dequeued);
1087                 }
1088         spin_unlock(&queue_lock);
1089         while (!list_empty(&dequeued)) {
1090                 cr = list_entry(dequeued.next, struct cache_request, q.list);
1091                 list_del(&cr->q.list);
1092                 cache_put(cr->item, detail);
1093                 kfree(cr->buf);
1094                 kfree(cr);
1095         }
1096 }
1097
1098 /*
1099  * Support routines for text-based upcalls.
1100  * Fields are separated by spaces.
1101  * Fields are either mangled to quote space tab newline slosh with slosh
1102  * or a hexified with a leading \x
1103  * Record is terminated with newline.
1104  *
1105  */
1106
1107 void qword_add(char **bpp, int *lp, char *str)
1108 {
1109         char *bp = *bpp;
1110         int len = *lp;
1111         int ret;
1112
1113         if (len < 0) return;
1114
1115         ret = string_escape_str(str, bp, len, ESCAPE_OCTAL, "\\ \n\t");
1116         if (ret >= len) {
1117                 bp += len;
1118                 len = -1;
1119         } else {
1120                 bp += ret;
1121                 len -= ret;
1122                 *bp++ = ' ';
1123                 len--;
1124         }
1125         *bpp = bp;
1126         *lp = len;
1127 }
1128 EXPORT_SYMBOL_GPL(qword_add);
1129
1130 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1131 {
1132         char *bp = *bpp;
1133         int len = *lp;
1134
1135         if (len < 0) return;
1136
1137         if (len > 2) {
1138                 *bp++ = '\\';
1139                 *bp++ = 'x';
1140                 len -= 2;
1141                 while (blen && len >= 2) {
1142                         bp = hex_byte_pack(bp, *buf++);
1143                         len -= 2;
1144                         blen--;
1145                 }
1146         }
1147         if (blen || len<1) len = -1;
1148         else {
1149                 *bp++ = ' ';
1150                 len--;
1151         }
1152         *bpp = bp;
1153         *lp = len;
1154 }
1155 EXPORT_SYMBOL_GPL(qword_addhex);
1156
1157 static void warn_no_listener(struct cache_detail *detail)
1158 {
1159         if (detail->last_warn != detail->last_close) {
1160                 detail->last_warn = detail->last_close;
1161                 if (detail->warn_no_listener)
1162                         detail->warn_no_listener(detail, detail->last_close != 0);
1163         }
1164 }
1165
1166 static bool cache_listeners_exist(struct cache_detail *detail)
1167 {
1168         if (atomic_read(&detail->writers))
1169                 return true;
1170         if (detail->last_close == 0)
1171                 /* This cache was never opened */
1172                 return false;
1173         if (detail->last_close < seconds_since_boot() - 30)
1174                 /*
1175                  * We allow for the possibility that someone might
1176                  * restart a userspace daemon without restarting the
1177                  * server; but after 30 seconds, we give up.
1178                  */
1179                  return false;
1180         return true;
1181 }
1182
1183 /*
1184  * register an upcall request to user-space and queue it up for read() by the
1185  * upcall daemon.
1186  *
1187  * Each request is at most one page long.
1188  */
1189 static int cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1190 {
1191         char *buf;
1192         struct cache_request *crq;
1193         int ret = 0;
1194
1195         if (test_bit(CACHE_CLEANED, &h->flags))
1196                 /* Too late to make an upcall */
1197                 return -EAGAIN;
1198
1199         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1200         if (!buf)
1201                 return -EAGAIN;
1202
1203         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1204         if (!crq) {
1205                 kfree(buf);
1206                 return -EAGAIN;
1207         }
1208
1209         crq->q.reader = 0;
1210         crq->buf = buf;
1211         crq->len = 0;
1212         crq->readers = 0;
1213         spin_lock(&queue_lock);
1214         if (test_bit(CACHE_PENDING, &h->flags)) {
1215                 crq->item = cache_get(h);
1216                 list_add_tail(&crq->q.list, &detail->queue);
1217                 trace_cache_entry_upcall(detail, h);
1218         } else
1219                 /* Lost a race, no longer PENDING, so don't enqueue */
1220                 ret = -EAGAIN;
1221         spin_unlock(&queue_lock);
1222         wake_up(&queue_wait);
1223         if (ret == -EAGAIN) {
1224                 kfree(buf);
1225                 kfree(crq);
1226         }
1227         return ret;
1228 }
1229
1230 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h)
1231 {
1232         if (test_and_set_bit(CACHE_PENDING, &h->flags))
1233                 return 0;
1234         return cache_pipe_upcall(detail, h);
1235 }
1236 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1237
1238 int sunrpc_cache_pipe_upcall_timeout(struct cache_detail *detail,
1239                                      struct cache_head *h)
1240 {
1241         if (!cache_listeners_exist(detail)) {
1242                 warn_no_listener(detail);
1243                 trace_cache_entry_no_listener(detail, h);
1244                 return -EINVAL;
1245         }
1246         return sunrpc_cache_pipe_upcall(detail, h);
1247 }
1248 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall_timeout);
1249
1250 /*
1251  * parse a message from user-space and pass it
1252  * to an appropriate cache
1253  * Messages are, like requests, separated into fields by
1254  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1255  *
1256  * Message is
1257  *   reply cachename expiry key ... content....
1258  *
1259  * key and content are both parsed by cache
1260  */
1261
1262 int qword_get(char **bpp, char *dest, int bufsize)
1263 {
1264         /* return bytes copied, or -1 on error */
1265         char *bp = *bpp;
1266         int len = 0;
1267
1268         while (*bp == ' ') bp++;
1269
1270         if (bp[0] == '\\' && bp[1] == 'x') {
1271                 /* HEX STRING */
1272                 bp += 2;
1273                 while (len < bufsize - 1) {
1274                         int h, l;
1275
1276                         h = hex_to_bin(bp[0]);
1277                         if (h < 0)
1278                                 break;
1279
1280                         l = hex_to_bin(bp[1]);
1281                         if (l < 0)
1282                                 break;
1283
1284                         *dest++ = (h << 4) | l;
1285                         bp += 2;
1286                         len++;
1287                 }
1288         } else {
1289                 /* text with \nnn octal quoting */
1290                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1291                         if (*bp == '\\' &&
1292                             isodigit(bp[1]) && (bp[1] <= '3') &&
1293                             isodigit(bp[2]) &&
1294                             isodigit(bp[3])) {
1295                                 int byte = (*++bp -'0');
1296                                 bp++;
1297                                 byte = (byte << 3) | (*bp++ - '0');
1298                                 byte = (byte << 3) | (*bp++ - '0');
1299                                 *dest++ = byte;
1300                                 len++;
1301                         } else {
1302                                 *dest++ = *bp++;
1303                                 len++;
1304                         }
1305                 }
1306         }
1307
1308         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1309                 return -1;
1310         while (*bp == ' ') bp++;
1311         *bpp = bp;
1312         *dest = '\0';
1313         return len;
1314 }
1315 EXPORT_SYMBOL_GPL(qword_get);
1316
1317
1318 /*
1319  * support /proc/net/rpc/$CACHENAME/content
1320  * as a seqfile.
1321  * We call ->cache_show passing NULL for the item to
1322  * get a header, then pass each real item in the cache
1323  */
1324
1325 static void *__cache_seq_start(struct seq_file *m, loff_t *pos)
1326 {
1327         loff_t n = *pos;
1328         unsigned int hash, entry;
1329         struct cache_head *ch;
1330         struct cache_detail *cd = m->private;
1331
1332         if (!n--)
1333                 return SEQ_START_TOKEN;
1334         hash = n >> 32;
1335         entry = n & ((1LL<<32) - 1);
1336
1337         hlist_for_each_entry_rcu(ch, &cd->hash_table[hash], cache_list)
1338                 if (!entry--)
1339                         return ch;
1340         n &= ~((1LL<<32) - 1);
1341         do {
1342                 hash++;
1343                 n += 1LL<<32;
1344         } while(hash < cd->hash_size &&
1345                 hlist_empty(&cd->hash_table[hash]));
1346         if (hash >= cd->hash_size)
1347                 return NULL;
1348         *pos = n+1;
1349         return hlist_entry_safe(rcu_dereference_raw(
1350                                 hlist_first_rcu(&cd->hash_table[hash])),
1351                                 struct cache_head, cache_list);
1352 }
1353
1354 static void *cache_seq_next(struct seq_file *m, void *p, loff_t *pos)
1355 {
1356         struct cache_head *ch = p;
1357         int hash = (*pos >> 32);
1358         struct cache_detail *cd = m->private;
1359
1360         if (p == SEQ_START_TOKEN)
1361                 hash = 0;
1362         else if (ch->cache_list.next == NULL) {
1363                 hash++;
1364                 *pos += 1LL<<32;
1365         } else {
1366                 ++*pos;
1367                 return hlist_entry_safe(rcu_dereference_raw(
1368                                         hlist_next_rcu(&ch->cache_list)),
1369                                         struct cache_head, cache_list);
1370         }
1371         *pos &= ~((1LL<<32) - 1);
1372         while (hash < cd->hash_size &&
1373                hlist_empty(&cd->hash_table[hash])) {
1374                 hash++;
1375                 *pos += 1LL<<32;
1376         }
1377         if (hash >= cd->hash_size)
1378                 return NULL;
1379         ++*pos;
1380         return hlist_entry_safe(rcu_dereference_raw(
1381                                 hlist_first_rcu(&cd->hash_table[hash])),
1382                                 struct cache_head, cache_list);
1383 }
1384
1385 void *cache_seq_start_rcu(struct seq_file *m, loff_t *pos)
1386         __acquires(RCU)
1387 {
1388         rcu_read_lock();
1389         return __cache_seq_start(m, pos);
1390 }
1391 EXPORT_SYMBOL_GPL(cache_seq_start_rcu);
1392
1393 void *cache_seq_next_rcu(struct seq_file *file, void *p, loff_t *pos)
1394 {
1395         return cache_seq_next(file, p, pos);
1396 }
1397 EXPORT_SYMBOL_GPL(cache_seq_next_rcu);
1398
1399 void cache_seq_stop_rcu(struct seq_file *m, void *p)
1400         __releases(RCU)
1401 {
1402         rcu_read_unlock();
1403 }
1404 EXPORT_SYMBOL_GPL(cache_seq_stop_rcu);
1405
1406 static int c_show(struct seq_file *m, void *p)
1407 {
1408         struct cache_head *cp = p;
1409         struct cache_detail *cd = m->private;
1410
1411         if (p == SEQ_START_TOKEN)
1412                 return cd->cache_show(m, cd, NULL);
1413
1414         ifdebug(CACHE)
1415                 seq_printf(m, "# expiry=%lld refcnt=%d flags=%lx\n",
1416                            convert_to_wallclock(cp->expiry_time),
1417                            kref_read(&cp->ref), cp->flags);
1418         cache_get(cp);
1419         if (cache_check(cd, cp, NULL))
1420                 /* cache_check does a cache_put on failure */
1421                 seq_puts(m, "# ");
1422         else {
1423                 if (cache_is_expired(cd, cp))
1424                         seq_puts(m, "# ");
1425                 cache_put(cp, cd);
1426         }
1427
1428         return cd->cache_show(m, cd, cp);
1429 }
1430
1431 static const struct seq_operations cache_content_op = {
1432         .start  = cache_seq_start_rcu,
1433         .next   = cache_seq_next_rcu,
1434         .stop   = cache_seq_stop_rcu,
1435         .show   = c_show,
1436 };
1437
1438 static int content_open(struct inode *inode, struct file *file,
1439                         struct cache_detail *cd)
1440 {
1441         struct seq_file *seq;
1442         int err;
1443
1444         if (!cd || !try_module_get(cd->owner))
1445                 return -EACCES;
1446
1447         err = seq_open(file, &cache_content_op);
1448         if (err) {
1449                 module_put(cd->owner);
1450                 return err;
1451         }
1452
1453         seq = file->private_data;
1454         seq->private = cd;
1455         return 0;
1456 }
1457
1458 static int content_release(struct inode *inode, struct file *file,
1459                 struct cache_detail *cd)
1460 {
1461         int ret = seq_release(inode, file);
1462         module_put(cd->owner);
1463         return ret;
1464 }
1465
1466 static int open_flush(struct inode *inode, struct file *file,
1467                         struct cache_detail *cd)
1468 {
1469         if (!cd || !try_module_get(cd->owner))
1470                 return -EACCES;
1471         return nonseekable_open(inode, file);
1472 }
1473
1474 static int release_flush(struct inode *inode, struct file *file,
1475                         struct cache_detail *cd)
1476 {
1477         module_put(cd->owner);
1478         return 0;
1479 }
1480
1481 static ssize_t read_flush(struct file *file, char __user *buf,
1482                           size_t count, loff_t *ppos,
1483                           struct cache_detail *cd)
1484 {
1485         char tbuf[22];
1486         size_t len;
1487
1488         len = snprintf(tbuf, sizeof(tbuf), "%llu\n",
1489                         convert_to_wallclock(cd->flush_time));
1490         return simple_read_from_buffer(buf, count, ppos, tbuf, len);
1491 }
1492
1493 static ssize_t write_flush(struct file *file, const char __user *buf,
1494                            size_t count, loff_t *ppos,
1495                            struct cache_detail *cd)
1496 {
1497         char tbuf[20];
1498         char *ep;
1499         time64_t now;
1500
1501         if (*ppos || count > sizeof(tbuf)-1)
1502                 return -EINVAL;
1503         if (copy_from_user(tbuf, buf, count))
1504                 return -EFAULT;
1505         tbuf[count] = 0;
1506         simple_strtoul(tbuf, &ep, 0);
1507         if (*ep && *ep != '\n')
1508                 return -EINVAL;
1509         /* Note that while we check that 'buf' holds a valid number,
1510          * we always ignore the value and just flush everything.
1511          * Making use of the number leads to races.
1512          */
1513
1514         now = seconds_since_boot();
1515         /* Always flush everything, so behave like cache_purge()
1516          * Do this by advancing flush_time to the current time,
1517          * or by one second if it has already reached the current time.
1518          * Newly added cache entries will always have ->last_refresh greater
1519          * that ->flush_time, so they don't get flushed prematurely.
1520          */
1521
1522         if (cd->flush_time >= now)
1523                 now = cd->flush_time + 1;
1524
1525         cd->flush_time = now;
1526         cd->nextcheck = now;
1527         cache_flush();
1528
1529         if (cd->flush)
1530                 cd->flush();
1531
1532         *ppos += count;
1533         return count;
1534 }
1535
1536 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1537                                  size_t count, loff_t *ppos)
1538 {
1539         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1540
1541         return cache_read(filp, buf, count, ppos, cd);
1542 }
1543
1544 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1545                                   size_t count, loff_t *ppos)
1546 {
1547         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1548
1549         return cache_write(filp, buf, count, ppos, cd);
1550 }
1551
1552 static __poll_t cache_poll_procfs(struct file *filp, poll_table *wait)
1553 {
1554         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1555
1556         return cache_poll(filp, wait, cd);
1557 }
1558
1559 static long cache_ioctl_procfs(struct file *filp,
1560                                unsigned int cmd, unsigned long arg)
1561 {
1562         struct inode *inode = file_inode(filp);
1563         struct cache_detail *cd = PDE_DATA(inode);
1564
1565         return cache_ioctl(inode, filp, cmd, arg, cd);
1566 }
1567
1568 static int cache_open_procfs(struct inode *inode, struct file *filp)
1569 {
1570         struct cache_detail *cd = PDE_DATA(inode);
1571
1572         return cache_open(inode, filp, cd);
1573 }
1574
1575 static int cache_release_procfs(struct inode *inode, struct file *filp)
1576 {
1577         struct cache_detail *cd = PDE_DATA(inode);
1578
1579         return cache_release(inode, filp, cd);
1580 }
1581
1582 static const struct proc_ops cache_channel_proc_ops = {
1583         .proc_lseek     = no_llseek,
1584         .proc_read      = cache_read_procfs,
1585         .proc_write     = cache_write_procfs,
1586         .proc_poll      = cache_poll_procfs,
1587         .proc_ioctl     = cache_ioctl_procfs, /* for FIONREAD */
1588         .proc_open      = cache_open_procfs,
1589         .proc_release   = cache_release_procfs,
1590 };
1591
1592 static int content_open_procfs(struct inode *inode, struct file *filp)
1593 {
1594         struct cache_detail *cd = PDE_DATA(inode);
1595
1596         return content_open(inode, filp, cd);
1597 }
1598
1599 static int content_release_procfs(struct inode *inode, struct file *filp)
1600 {
1601         struct cache_detail *cd = PDE_DATA(inode);
1602
1603         return content_release(inode, filp, cd);
1604 }
1605
1606 static const struct proc_ops content_proc_ops = {
1607         .proc_open      = content_open_procfs,
1608         .proc_read      = seq_read,
1609         .proc_lseek     = seq_lseek,
1610         .proc_release   = content_release_procfs,
1611 };
1612
1613 static int open_flush_procfs(struct inode *inode, struct file *filp)
1614 {
1615         struct cache_detail *cd = PDE_DATA(inode);
1616
1617         return open_flush(inode, filp, cd);
1618 }
1619
1620 static int release_flush_procfs(struct inode *inode, struct file *filp)
1621 {
1622         struct cache_detail *cd = PDE_DATA(inode);
1623
1624         return release_flush(inode, filp, cd);
1625 }
1626
1627 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1628                             size_t count, loff_t *ppos)
1629 {
1630         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1631
1632         return read_flush(filp, buf, count, ppos, cd);
1633 }
1634
1635 static ssize_t write_flush_procfs(struct file *filp,
1636                                   const char __user *buf,
1637                                   size_t count, loff_t *ppos)
1638 {
1639         struct cache_detail *cd = PDE_DATA(file_inode(filp));
1640
1641         return write_flush(filp, buf, count, ppos, cd);
1642 }
1643
1644 static const struct proc_ops cache_flush_proc_ops = {
1645         .proc_open      = open_flush_procfs,
1646         .proc_read      = read_flush_procfs,
1647         .proc_write     = write_flush_procfs,
1648         .proc_release   = release_flush_procfs,
1649         .proc_lseek     = no_llseek,
1650 };
1651
1652 static void remove_cache_proc_entries(struct cache_detail *cd)
1653 {
1654         if (cd->procfs) {
1655                 proc_remove(cd->procfs);
1656                 cd->procfs = NULL;
1657         }
1658 }
1659
1660 #ifdef CONFIG_PROC_FS
1661 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1662 {
1663         struct proc_dir_entry *p;
1664         struct sunrpc_net *sn;
1665
1666         sn = net_generic(net, sunrpc_net_id);
1667         cd->procfs = proc_mkdir(cd->name, sn->proc_net_rpc);
1668         if (cd->procfs == NULL)
1669                 goto out_nomem;
1670
1671         p = proc_create_data("flush", S_IFREG | 0600,
1672                              cd->procfs, &cache_flush_proc_ops, cd);
1673         if (p == NULL)
1674                 goto out_nomem;
1675
1676         if (cd->cache_request || cd->cache_parse) {
1677                 p = proc_create_data("channel", S_IFREG | 0600, cd->procfs,
1678                                      &cache_channel_proc_ops, cd);
1679                 if (p == NULL)
1680                         goto out_nomem;
1681         }
1682         if (cd->cache_show) {
1683                 p = proc_create_data("content", S_IFREG | 0400, cd->procfs,
1684                                      &content_proc_ops, cd);
1685                 if (p == NULL)
1686                         goto out_nomem;
1687         }
1688         return 0;
1689 out_nomem:
1690         remove_cache_proc_entries(cd);
1691         return -ENOMEM;
1692 }
1693 #else /* CONFIG_PROC_FS */
1694 static int create_cache_proc_entries(struct cache_detail *cd, struct net *net)
1695 {
1696         return 0;
1697 }
1698 #endif
1699
1700 void __init cache_initialize(void)
1701 {
1702         INIT_DEFERRABLE_WORK(&cache_cleaner, do_cache_clean);
1703 }
1704
1705 int cache_register_net(struct cache_detail *cd, struct net *net)
1706 {
1707         int ret;
1708
1709         sunrpc_init_cache_detail(cd);
1710         ret = create_cache_proc_entries(cd, net);
1711         if (ret)
1712                 sunrpc_destroy_cache_detail(cd);
1713         return ret;
1714 }
1715 EXPORT_SYMBOL_GPL(cache_register_net);
1716
1717 void cache_unregister_net(struct cache_detail *cd, struct net *net)
1718 {
1719         remove_cache_proc_entries(cd);
1720         sunrpc_destroy_cache_detail(cd);
1721 }
1722 EXPORT_SYMBOL_GPL(cache_unregister_net);
1723
1724 struct cache_detail *cache_create_net(const struct cache_detail *tmpl, struct net *net)
1725 {
1726         struct cache_detail *cd;
1727         int i;
1728
1729         cd = kmemdup(tmpl, sizeof(struct cache_detail), GFP_KERNEL);
1730         if (cd == NULL)
1731                 return ERR_PTR(-ENOMEM);
1732
1733         cd->hash_table = kcalloc(cd->hash_size, sizeof(struct hlist_head),
1734                                  GFP_KERNEL);
1735         if (cd->hash_table == NULL) {
1736                 kfree(cd);
1737                 return ERR_PTR(-ENOMEM);
1738         }
1739
1740         for (i = 0; i < cd->hash_size; i++)
1741                 INIT_HLIST_HEAD(&cd->hash_table[i]);
1742         cd->net = net;
1743         return cd;
1744 }
1745 EXPORT_SYMBOL_GPL(cache_create_net);
1746
1747 void cache_destroy_net(struct cache_detail *cd, struct net *net)
1748 {
1749         kfree(cd->hash_table);
1750         kfree(cd);
1751 }
1752 EXPORT_SYMBOL_GPL(cache_destroy_net);
1753
1754 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1755                                  size_t count, loff_t *ppos)
1756 {
1757         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1758
1759         return cache_read(filp, buf, count, ppos, cd);
1760 }
1761
1762 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1763                                   size_t count, loff_t *ppos)
1764 {
1765         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1766
1767         return cache_write(filp, buf, count, ppos, cd);
1768 }
1769
1770 static __poll_t cache_poll_pipefs(struct file *filp, poll_table *wait)
1771 {
1772         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1773
1774         return cache_poll(filp, wait, cd);
1775 }
1776
1777 static long cache_ioctl_pipefs(struct file *filp,
1778                               unsigned int cmd, unsigned long arg)
1779 {
1780         struct inode *inode = file_inode(filp);
1781         struct cache_detail *cd = RPC_I(inode)->private;
1782
1783         return cache_ioctl(inode, filp, cmd, arg, cd);
1784 }
1785
1786 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1787 {
1788         struct cache_detail *cd = RPC_I(inode)->private;
1789
1790         return cache_open(inode, filp, cd);
1791 }
1792
1793 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1794 {
1795         struct cache_detail *cd = RPC_I(inode)->private;
1796
1797         return cache_release(inode, filp, cd);
1798 }
1799
1800 const struct file_operations cache_file_operations_pipefs = {
1801         .owner          = THIS_MODULE,
1802         .llseek         = no_llseek,
1803         .read           = cache_read_pipefs,
1804         .write          = cache_write_pipefs,
1805         .poll           = cache_poll_pipefs,
1806         .unlocked_ioctl = cache_ioctl_pipefs, /* for FIONREAD */
1807         .open           = cache_open_pipefs,
1808         .release        = cache_release_pipefs,
1809 };
1810
1811 static int content_open_pipefs(struct inode *inode, struct file *filp)
1812 {
1813         struct cache_detail *cd = RPC_I(inode)->private;
1814
1815         return content_open(inode, filp, cd);
1816 }
1817
1818 static int content_release_pipefs(struct inode *inode, struct file *filp)
1819 {
1820         struct cache_detail *cd = RPC_I(inode)->private;
1821
1822         return content_release(inode, filp, cd);
1823 }
1824
1825 const struct file_operations content_file_operations_pipefs = {
1826         .open           = content_open_pipefs,
1827         .read           = seq_read,
1828         .llseek         = seq_lseek,
1829         .release        = content_release_pipefs,
1830 };
1831
1832 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1833 {
1834         struct cache_detail *cd = RPC_I(inode)->private;
1835
1836         return open_flush(inode, filp, cd);
1837 }
1838
1839 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1840 {
1841         struct cache_detail *cd = RPC_I(inode)->private;
1842
1843         return release_flush(inode, filp, cd);
1844 }
1845
1846 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1847                             size_t count, loff_t *ppos)
1848 {
1849         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1850
1851         return read_flush(filp, buf, count, ppos, cd);
1852 }
1853
1854 static ssize_t write_flush_pipefs(struct file *filp,
1855                                   const char __user *buf,
1856                                   size_t count, loff_t *ppos)
1857 {
1858         struct cache_detail *cd = RPC_I(file_inode(filp))->private;
1859
1860         return write_flush(filp, buf, count, ppos, cd);
1861 }
1862
1863 const struct file_operations cache_flush_operations_pipefs = {
1864         .open           = open_flush_pipefs,
1865         .read           = read_flush_pipefs,
1866         .write          = write_flush_pipefs,
1867         .release        = release_flush_pipefs,
1868         .llseek         = no_llseek,
1869 };
1870
1871 int sunrpc_cache_register_pipefs(struct dentry *parent,
1872                                  const char *name, umode_t umode,
1873                                  struct cache_detail *cd)
1874 {
1875         struct dentry *dir = rpc_create_cache_dir(parent, name, umode, cd);
1876         if (IS_ERR(dir))
1877                 return PTR_ERR(dir);
1878         cd->pipefs = dir;
1879         return 0;
1880 }
1881 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1882
1883 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1884 {
1885         if (cd->pipefs) {
1886                 rpc_remove_cache_dir(cd->pipefs);
1887                 cd->pipefs = NULL;
1888         }
1889 }
1890 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1891
1892 void sunrpc_cache_unhash(struct cache_detail *cd, struct cache_head *h)
1893 {
1894         spin_lock(&cd->hash_lock);
1895         if (!hlist_unhashed(&h->cache_list)){
1896                 sunrpc_begin_cache_remove_entry(h, cd);
1897                 spin_unlock(&cd->hash_lock);
1898                 sunrpc_end_cache_remove_entry(h, cd);
1899         } else
1900                 spin_unlock(&cd->hash_lock);
1901 }
1902 EXPORT_SYMBOL_GPL(sunrpc_cache_unhash);