err.no Git - linux-2.6/blob - net/sunrpc/cache.c

   1 /*
   2  * net/sunrpc/cache.c
   3  *
   4  * Generic code for various authentication-related caches
   5  * used by sunrpc clients and servers.
   6  *
   7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   8  *
   9  * Released under terms in GPL version 2.  See COPYING.
  10  *
  11  */
  12
  13 #include <linux/types.h>
  14 #include <linux/fs.h>
  15 #include <linux/file.h>
  16 #include <linux/slab.h>
  17 #include <linux/signal.h>
  18 #include <linux/sched.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/module.h>
  22 #include <linux/ctype.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/poll.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/net.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/mutex.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
  35 #define  RPCDBG_FACILITY RPCDBG_CACHE
  36
  37 static int cache_defer_req(struct cache_req *req, struct cache_head *item);
  38 static void cache_revisit_request(struct cache_head *item);
  39
  40 static void cache_init(struct cache_head *h)
  41 {
  42         time_t now = get_seconds();
  43         h->next = NULL;
  44         h->flags = 0;
  45         kref_init(&h->ref);
  46         h->expiry_time = now + CACHE_NEW_EXPIRY;
  47         h->last_refresh = now;
  48 }
  49
  50 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  51                                        struct cache_head *key, int hash)
  52 {
  53         struct cache_head **head,  **hp;
  54         struct cache_head *new = NULL;
  55
  56         head = &detail->hash_table[hash];
  57
  58         read_lock(&detail->hash_lock);
  59
  60         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  61                 struct cache_head *tmp = *hp;
  62                 if (detail->match(tmp, key)) {
  63                         cache_get(tmp);
  64                         read_unlock(&detail->hash_lock);
  65                         return tmp;
  66                 }
  67         }
  68         read_unlock(&detail->hash_lock);
  69         /* Didn't find anything, insert an empty entry */
  70
  71         new = detail->alloc();
  72         if (!new)
  73                 return NULL;
  74         /* must fully initialise 'new', else
  75          * we might get lose if we need to
  76          * cache_put it soon.
  77          */
  78         cache_init(new);
  79         detail->init(new, key);
  80
  81         write_lock(&detail->hash_lock);
  82
  83         /* check if entry appeared while we slept */
  84         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  85                 struct cache_head *tmp = *hp;
  86                 if (detail->match(tmp, key)) {
  87                         cache_get(tmp);
  88                         write_unlock(&detail->hash_lock);
  89                         cache_put(new, detail);
  90                         return tmp;
  91                 }
  92         }
  93         new->next = *head;
  94         *head = new;
  95         detail->entries++;
  96         cache_get(new);
  97         write_unlock(&detail->hash_lock);
  98
  99         return new;
 100 }
 101 EXPORT_SYMBOL(sunrpc_cache_lookup);
 102
 103
 104 static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
 105
 106 static int cache_fresh_locked(struct cache_head *head, time_t expiry)
 107 {
 108         head->expiry_time = expiry;
 109         head->last_refresh = get_seconds();
 110         return !test_and_set_bit(CACHE_VALID, &head->flags);
 111 }
 112
 113 static void cache_fresh_unlocked(struct cache_head *head,
 114                         struct cache_detail *detail, int new)
 115 {
 116         if (new)
 117                 cache_revisit_request(head);
 118         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 119                 cache_revisit_request(head);
 120                 queue_loose(detail, head);
 121         }
 122 }
 123
 124 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 125                                        struct cache_head *new, struct cache_head *old, int hash)
 126 {
 127         /* The 'old' entry is to be replaced by 'new'.
 128          * If 'old' is not VALID, we update it directly,
 129          * otherwise we need to replace it
 130          */
 131         struct cache_head **head;
 132         struct cache_head *tmp;
 133         int is_new;
 134
 135         if (!test_bit(CACHE_VALID, &old->flags)) {
 136                 write_lock(&detail->hash_lock);
 137                 if (!test_bit(CACHE_VALID, &old->flags)) {
 138                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 139                                 set_bit(CACHE_NEGATIVE, &old->flags);
 140                         else
 141                                 detail->update(old, new);
 142                         is_new = cache_fresh_locked(old, new->expiry_time);
 143                         write_unlock(&detail->hash_lock);
 144                         cache_fresh_unlocked(old, detail, is_new);
 145                         return old;
 146                 }
 147                 write_unlock(&detail->hash_lock);
 148         }
 149         /* We need to insert a new entry */
 150         tmp = detail->alloc();
 151         if (!tmp) {
 152                 cache_put(old, detail);
 153                 return NULL;
 154         }
 155         cache_init(tmp);
 156         detail->init(tmp, old);
 157         head = &detail->hash_table[hash];
 158
 159         write_lock(&detail->hash_lock);
 160         if (test_bit(CACHE_NEGATIVE, &new->flags))
 161                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 162         else
 163                 detail->update(tmp, new);
 164         tmp->next = *head;
 165         *head = tmp;
 166         detail->entries++;
 167         cache_get(tmp);
 168         is_new = cache_fresh_locked(tmp, new->expiry_time);
 169         cache_fresh_locked(old, 0);
 170         write_unlock(&detail->hash_lock);
 171         cache_fresh_unlocked(tmp, detail, is_new);
 172         cache_fresh_unlocked(old, detail, 0);
 173         cache_put(old, detail);
 174         return tmp;
 175 }
 176 EXPORT_SYMBOL(sunrpc_cache_update);
 177
 178 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h);
 179 /*
 180  * This is the generic cache management routine for all
 181  * the authentication caches.
 182  * It checks the currency of a cache item and will (later)
 183  * initiate an upcall to fill it if needed.
 184  *
 185  *
 186  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 187  * -EAGAIN if upcall is pending,
 188  * -ETIMEDOUT if upcall failed and should be retried,
 189  * -ENOENT if cache entry was negative
 190  */
 191 int cache_check(struct cache_detail *detail,
 192                     struct cache_head *h, struct cache_req *rqstp)
 193 {
 194         int rv;
 195         long refresh_age, age;
 196
 197         /* First decide return status as best we can */
 198         if (!test_bit(CACHE_VALID, &h->flags) ||
 199             h->expiry_time < get_seconds())
 200                 rv = -EAGAIN;
 201         else if (detail->flush_time > h->last_refresh)
 202                 rv = -EAGAIN;
 203         else {
 204                 /* entry is valid */
 205                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 206                         rv = -ENOENT;
 207                 else rv = 0;
 208         }
 209
 210         /* now see if we want to start an upcall */
 211         refresh_age = (h->expiry_time - h->last_refresh);
 212         age = get_seconds() - h->last_refresh;
 213
 214         if (rqstp == NULL) {
 215                 if (rv == -EAGAIN)
 216                         rv = -ENOENT;
 217         } else if (rv == -EAGAIN || age > refresh_age/2) {
 218                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 219                                 refresh_age, age);
 220                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 221                         switch (cache_make_upcall(detail, h)) {
 222                         case -EINVAL:
 223                                 clear_bit(CACHE_PENDING, &h->flags);
 224                                 if (rv == -EAGAIN) {
 225                                         set_bit(CACHE_NEGATIVE, &h->flags);
 226                                         cache_fresh_unlocked(h, detail,
 227                                              cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY));
 228                                         rv = -ENOENT;
 229                                 }
 230                                 break;
 231
 232                         case -EAGAIN:
 233                                 clear_bit(CACHE_PENDING, &h->flags);
 234                                 cache_revisit_request(h);
 235                                 break;
 236                         }
 237                 }
 238         }
 239
 240         if (rv == -EAGAIN)
 241                 if (cache_defer_req(rqstp, h) != 0)
 242                         rv = -ETIMEDOUT;
 243
 244         if (rv)
 245                 cache_put(h, detail);
 246         return rv;
 247 }
 248 EXPORT_SYMBOL(cache_check);
 249
 250 /*
 251  * caches need to be periodically cleaned.
 252  * For this we maintain a list of cache_detail and
 253  * a current pointer into that list and into the table
 254  * for that entry.
 255  *
 256  * Each time clean_cache is called it finds the next non-empty entry
 257  * in the current table and walks the list in that entry
 258  * looking for entries that can be removed.
 259  *
 260  * An entry gets removed if:
 261  * - The expiry is before current time
 262  * - The last_refresh time is before the flush_time for that cache
 263  *
 264  * later we might drop old entries with non-NEVER expiry if that table
 265  * is getting 'full' for some definition of 'full'
 266  *
 267  * The question of "how often to scan a table" is an interesting one
 268  * and is answered in part by the use of the "nextcheck" field in the
 269  * cache_detail.
 270  * When a scan of a table begins, the nextcheck field is set to a time
 271  * that is well into the future.
 272  * While scanning, if an expiry time is found that is earlier than the
 273  * current nextcheck time, nextcheck is set to that expiry time.
 274  * If the flush_time is ever set to a time earlier than the nextcheck
 275  * time, the nextcheck time is then set to that flush_time.
 276  *
 277  * A table is then only scanned if the current time is at least
 278  * the nextcheck time.
 279  *
 280  */
 281
 282 static LIST_HEAD(cache_list);
 283 static DEFINE_SPINLOCK(cache_list_lock);
 284 static struct cache_detail *current_detail;
 285 static int current_index;
 286
 287 static const struct file_operations cache_file_operations;
 288 static const struct file_operations content_file_operations;
 289 static const struct file_operations cache_flush_operations;
 290
 291 static void do_cache_clean(struct work_struct *work);
 292 static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
 293
 294 static void remove_cache_proc_entries(struct cache_detail *cd)
 295 {
 296         if (cd->proc_ent == NULL)
 297                 return;
 298         if (cd->flush_ent)
 299                 remove_proc_entry("flush", cd->proc_ent);
 300         if (cd->channel_ent)
 301                 remove_proc_entry("channel", cd->proc_ent);
 302         if (cd->content_ent)
 303                 remove_proc_entry("content", cd->proc_ent);
 304         cd->proc_ent = NULL;
 305         remove_proc_entry(cd->name, proc_net_rpc);
 306 }
 307
 308 #ifdef CONFIG_PROC_FS
 309 static int create_cache_proc_entries(struct cache_detail *cd)
 310 {
 311         struct proc_dir_entry *p;
 312
 313         cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
 314         if (cd->proc_ent == NULL)
 315                 goto out_nomem;
 316         cd->proc_ent->owner = cd->owner;
 317         cd->channel_ent = cd->content_ent = NULL;
 318
 319         p = proc_create("flush", S_IFREG|S_IRUSR|S_IWUSR,
 320                         cd->proc_ent, &cache_flush_operations);
 321         cd->flush_ent = p;
 322         if (p == NULL)
 323                 goto out_nomem;
 324         p->owner = cd->owner;
 325         p->data = cd;
 326
 327         if (cd->cache_request || cd->cache_parse) {
 328                 p = proc_create("channel", S_IFREG|S_IRUSR|S_IWUSR,
 329                                 cd->proc_ent, &cache_file_operations);
 330                 cd->channel_ent = p;
 331                 if (p == NULL)
 332                         goto out_nomem;
 333                 p->owner = cd->owner;
 334                 p->data = cd;
 335         }
 336         if (cd->cache_show) {
 337                 p = proc_create("content", S_IFREG|S_IRUSR|S_IWUSR,
 338                                 cd->proc_ent, &content_file_operations);
 339                 cd->content_ent = p;
 340                 if (p == NULL)
 341                         goto out_nomem;
 342                 p->owner = cd->owner;
 343                 p->data = cd;
 344         }
 345         return 0;
 346 out_nomem:
 347         remove_cache_proc_entries(cd);
 348         return -ENOMEM;
 349 }
 350 #else /* CONFIG_PROC_FS */
 351 static int create_cache_proc_entries(struct cache_detail *cd)
 352 {
 353         return 0;
 354 }
 355 #endif
 356
 357 int cache_register(struct cache_detail *cd)
 358 {
 359         int ret;
 360
 361         ret = create_cache_proc_entries(cd);
 362         if (ret)
 363                 return ret;
 364         rwlock_init(&cd->hash_lock);
 365         INIT_LIST_HEAD(&cd->queue);
 366         spin_lock(&cache_list_lock);
 367         cd->nextcheck = 0;
 368         cd->entries = 0;
 369         atomic_set(&cd->readers, 0);
 370         cd->last_close = 0;
 371         cd->last_warn = -1;
 372         list_add(&cd->others, &cache_list);
 373         spin_unlock(&cache_list_lock);
 374
 375         /* start the cleaning process */
 376         schedule_delayed_work(&cache_cleaner, 0);
 377         return 0;
 378 }
 379 EXPORT_SYMBOL(cache_register);
 380
 381 void cache_unregister(struct cache_detail *cd)
 382 {
 383         cache_purge(cd);
 384         spin_lock(&cache_list_lock);
 385         write_lock(&cd->hash_lock);
 386         if (cd->entries || atomic_read(&cd->inuse)) {
 387                 write_unlock(&cd->hash_lock);
 388                 spin_unlock(&cache_list_lock);
 389                 goto out;
 390         }
 391         if (current_detail == cd)
 392                 current_detail = NULL;
 393         list_del_init(&cd->others);
 394         write_unlock(&cd->hash_lock);
 395         spin_unlock(&cache_list_lock);
 396         remove_cache_proc_entries(cd);
 397         if (list_empty(&cache_list)) {
 398                 /* module must be being unloaded so its safe to kill the worker */
 399                 cancel_delayed_work_sync(&cache_cleaner);
 400         }
 401         return;
 402 out:
 403         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 404 }
 405 EXPORT_SYMBOL(cache_unregister);
 406
 407 /* clean cache tries to find something to clean
 408  * and cleans it.
 409  * It returns 1 if it cleaned something,
 410  *            0 if it didn't find anything this time
 411  *           -1 if it fell off the end of the list.
 412  */
 413 static int cache_clean(void)
 414 {
 415         int rv = 0;
 416         struct list_head *next;
 417
 418         spin_lock(&cache_list_lock);
 419
 420         /* find a suitable table if we don't already have one */
 421         while (current_detail == NULL ||
 422             current_index >= current_detail->hash_size) {
 423                 if (current_detail)
 424                         next = current_detail->others.next;
 425                 else
 426                         next = cache_list.next;
 427                 if (next == &cache_list) {
 428                         current_detail = NULL;
 429                         spin_unlock(&cache_list_lock);
 430                         return -1;
 431                 }
 432                 current_detail = list_entry(next, struct cache_detail, others);
 433                 if (current_detail->nextcheck > get_seconds())
 434                         current_index = current_detail->hash_size;
 435                 else {
 436                         current_index = 0;
 437                         current_detail->nextcheck = get_seconds()+30*60;
 438                 }
 439         }
 440
 441         /* find a non-empty bucket in the table */
 442         while (current_detail &&
 443                current_index < current_detail->hash_size &&
 444                current_detail->hash_table[current_index] == NULL)
 445                 current_index++;
 446
 447         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 448
 449         if (current_detail && current_index < current_detail->hash_size) {
 450                 struct cache_head *ch, **cp;
 451                 struct cache_detail *d;
 452
 453                 write_lock(&current_detail->hash_lock);
 454
 455                 /* Ok, now to clean this strand */
 456
 457                 cp = & current_detail->hash_table[current_index];
 458                 ch = *cp;
 459                 for (; ch; cp= & ch->next, ch= *cp) {
 460                         if (current_detail->nextcheck > ch->expiry_time)
 461                                 current_detail->nextcheck = ch->expiry_time+1;
 462                         if (ch->expiry_time >= get_seconds()
 463                             && ch->last_refresh >= current_detail->flush_time
 464                                 )
 465                                 continue;
 466                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 467                                 queue_loose(current_detail, ch);
 468
 469                         if (atomic_read(&ch->ref.refcount) == 1)
 470                                 break;
 471                 }
 472                 if (ch) {
 473                         *cp = ch->next;
 474                         ch->next = NULL;
 475                         current_detail->entries--;
 476                         rv = 1;
 477                 }
 478                 write_unlock(&current_detail->hash_lock);
 479                 d = current_detail;
 480                 if (!ch)
 481                         current_index ++;
 482                 spin_unlock(&cache_list_lock);
 483                 if (ch)
 484                         cache_put(ch, d);
 485         } else
 486                 spin_unlock(&cache_list_lock);
 487
 488         return rv;
 489 }
 490
 491 /*
 492  * We want to regularly clean the cache, so we need to schedule some work ...
 493  */
 494 static void do_cache_clean(struct work_struct *work)
 495 {
 496         int delay = 5;
 497         if (cache_clean() == -1)
 498                 delay = 30*HZ;
 499
 500         if (list_empty(&cache_list))
 501                 delay = 0;
 502
 503         if (delay)
 504                 schedule_delayed_work(&cache_cleaner, delay);
 505 }
 506
 507
 508 /*
 509  * Clean all caches promptly.  This just calls cache_clean
 510  * repeatedly until we are sure that every cache has had a chance to
 511  * be fully cleaned
 512  */
 513 void cache_flush(void)
 514 {
 515         while (cache_clean() != -1)
 516                 cond_resched();
 517         while (cache_clean() != -1)
 518                 cond_resched();
 519 }
 520 EXPORT_SYMBOL(cache_flush);
 521
 522 void cache_purge(struct cache_detail *detail)
 523 {
 524         detail->flush_time = LONG_MAX;
 525         detail->nextcheck = get_seconds();
 526         cache_flush();
 527         detail->flush_time = 1;
 528 }
 529 EXPORT_SYMBOL(cache_purge);
 530
 531
 532 /*
 533  * Deferral and Revisiting of Requests.
 534  *
 535  * If a cache lookup finds a pending entry, we
 536  * need to defer the request and revisit it later.
 537  * All deferred requests are stored in a hash table,
 538  * indexed by "struct cache_head *".
 539  * As it may be wasteful to store a whole request
 540  * structure, we allow the request to provide a
 541  * deferred form, which must contain a
 542  * 'struct cache_deferred_req'
 543  * This cache_deferred_req contains a method to allow
 544  * it to be revisited when cache info is available
 545  */
 546
 547 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 548 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 549
 550 #define DFR_MAX 300     /* ??? */
 551
 552 static DEFINE_SPINLOCK(cache_defer_lock);
 553 static LIST_HEAD(cache_defer_list);
 554 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 555 static int cache_defer_cnt;
 556
 557 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 558 {
 559         struct cache_deferred_req *dreq;
 560         int hash = DFR_HASH(item);
 561
 562         if (cache_defer_cnt >= DFR_MAX) {
 563                 /* too much in the cache, randomly drop this one,
 564                  * or continue and drop the oldest below
 565                  */
 566                 if (net_random()&1)
 567                         return -ETIMEDOUT;
 568         }
 569         dreq = req->defer(req);
 570         if (dreq == NULL)
 571                 return -ETIMEDOUT;
 572
 573         dreq->item = item;
 574
 575         spin_lock(&cache_defer_lock);
 576
 577         list_add(&dreq->recent, &cache_defer_list);
 578
 579         if (cache_defer_hash[hash].next == NULL)
 580                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 581         list_add(&dreq->hash, &cache_defer_hash[hash]);
 582
 583         /* it is in, now maybe clean up */
 584         dreq = NULL;
 585         if (++cache_defer_cnt > DFR_MAX) {
 586                 dreq = list_entry(cache_defer_list.prev,
 587                                   struct cache_deferred_req, recent);
 588                 list_del(&dreq->recent);
 589                 list_del(&dreq->hash);
 590                 cache_defer_cnt--;
 591         }
 592         spin_unlock(&cache_defer_lock);
 593
 594         if (dreq) {
 595                 /* there was one too many */
 596                 dreq->revisit(dreq, 1);
 597         }
 598         if (!test_bit(CACHE_PENDING, &item->flags)) {
 599                 /* must have just been validated... */
 600                 cache_revisit_request(item);
 601         }
 602         return 0;
 603 }
 604
 605 static void cache_revisit_request(struct cache_head *item)
 606 {
 607         struct cache_deferred_req *dreq;
 608         struct list_head pending;
 609
 610         struct list_head *lp;
 611         int hash = DFR_HASH(item);
 612
 613         INIT_LIST_HEAD(&pending);
 614         spin_lock(&cache_defer_lock);
 615
 616         lp = cache_defer_hash[hash].next;
 617         if (lp) {
 618                 while (lp != &cache_defer_hash[hash]) {
 619                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 620                         lp = lp->next;
 621                         if (dreq->item == item) {
 622                                 list_del(&dreq->hash);
 623                                 list_move(&dreq->recent, &pending);
 624                                 cache_defer_cnt--;
 625                         }
 626                 }
 627         }
 628         spin_unlock(&cache_defer_lock);
 629
 630         while (!list_empty(&pending)) {
 631                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 632                 list_del_init(&dreq->recent);
 633                 dreq->revisit(dreq, 0);
 634         }
 635 }
 636
 637 void cache_clean_deferred(void *owner)
 638 {
 639         struct cache_deferred_req *dreq, *tmp;
 640         struct list_head pending;
 641
 642
 643         INIT_LIST_HEAD(&pending);
 644         spin_lock(&cache_defer_lock);
 645
 646         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 647                 if (dreq->owner == owner) {
 648                         list_del(&dreq->hash);
 649                         list_move(&dreq->recent, &pending);
 650                         cache_defer_cnt--;
 651                 }
 652         }
 653         spin_unlock(&cache_defer_lock);
 654
 655         while (!list_empty(&pending)) {
 656                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 657                 list_del_init(&dreq->recent);
 658                 dreq->revisit(dreq, 1);
 659         }
 660 }
 661
 662 /*
 663  * communicate with user-space
 664  *
 665  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 666  * On read, you get a full request, or block.
 667  * On write, an update request is processed.
 668  * Poll works if anything to read, and always allows write.
 669  *
 670  * Implemented by linked list of requests.  Each open file has
 671  * a ->private that also exists in this list.  New requests are added
 672  * to the end and may wakeup and preceding readers.
 673  * New readers are added to the head.  If, on read, an item is found with
 674  * CACHE_UPCALLING clear, we free it from the list.
 675  *
 676  */
 677
 678 static DEFINE_SPINLOCK(queue_lock);
 679 static DEFINE_MUTEX(queue_io_mutex);
 680
 681 struct cache_queue {
 682         struct list_head        list;
 683         int                     reader; /* if 0, then request */
 684 };
 685 struct cache_request {
 686         struct cache_queue      q;
 687         struct cache_head       *item;
 688         char                    * buf;
 689         int                     len;
 690         int                     readers;
 691 };
 692 struct cache_reader {
 693         struct cache_queue      q;
 694         int                     offset; /* if non-0, we have a refcnt on next request */
 695 };
 696
 697 static ssize_t
 698 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
 699 {
 700         struct cache_reader *rp = filp->private_data;
 701         struct cache_request *rq;
 702         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 703         int err;
 704
 705         if (count == 0)
 706                 return 0;
 707
 708         mutex_lock(&queue_io_mutex); /* protect against multiple concurrent
 709                               * readers on this file */
 710  again:
 711         spin_lock(&queue_lock);
 712         /* need to find next request */
 713         while (rp->q.list.next != &cd->queue &&
 714                list_entry(rp->q.list.next, struct cache_queue, list)
 715                ->reader) {
 716                 struct list_head *next = rp->q.list.next;
 717                 list_move(&rp->q.list, next);
 718         }
 719         if (rp->q.list.next == &cd->queue) {
 720                 spin_unlock(&queue_lock);
 721                 mutex_unlock(&queue_io_mutex);
 722                 BUG_ON(rp->offset);
 723                 return 0;
 724         }
 725         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 726         BUG_ON(rq->q.reader);
 727         if (rp->offset == 0)
 728                 rq->readers++;
 729         spin_unlock(&queue_lock);
 730
 731         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 732                 err = -EAGAIN;
 733                 spin_lock(&queue_lock);
 734                 list_move(&rp->q.list, &rq->q.list);
 735                 spin_unlock(&queue_lock);
 736         } else {
 737                 if (rp->offset + count > rq->len)
 738                         count = rq->len - rp->offset;
 739                 err = -EFAULT;
 740                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 741                         goto out;
 742                 rp->offset += count;
 743                 if (rp->offset >= rq->len) {
 744                         rp->offset = 0;
 745                         spin_lock(&queue_lock);
 746                         list_move(&rp->q.list, &rq->q.list);
 747                         spin_unlock(&queue_lock);
 748                 }
 749                 err = 0;
 750         }
 751  out:
 752         if (rp->offset == 0) {
 753                 /* need to release rq */
 754                 spin_lock(&queue_lock);
 755                 rq->readers--;
 756                 if (rq->readers == 0 &&
 757                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 758                         list_del(&rq->q.list);
 759                         spin_unlock(&queue_lock);
 760                         cache_put(rq->item, cd);
 761                         kfree(rq->buf);
 762                         kfree(rq);
 763                 } else
 764                         spin_unlock(&queue_lock);
 765         }
 766         if (err == -EAGAIN)
 767                 goto again;
 768         mutex_unlock(&queue_io_mutex);
 769         return err ? err :  count;
 770 }
 771
 772 static char write_buf[8192]; /* protected by queue_io_mutex */
 773
 774 static ssize_t
 775 cache_write(struct file *filp, const char __user *buf, size_t count,
 776             loff_t *ppos)
 777 {
 778         int err;
 779         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 780
 781         if (count == 0)
 782                 return 0;
 783         if (count >= sizeof(write_buf))
 784                 return -EINVAL;
 785
 786         mutex_lock(&queue_io_mutex);
 787
 788         if (copy_from_user(write_buf, buf, count)) {
 789                 mutex_unlock(&queue_io_mutex);
 790                 return -EFAULT;
 791         }
 792         write_buf[count] = '\0';
 793         if (cd->cache_parse)
 794                 err = cd->cache_parse(cd, write_buf, count);
 795         else
 796                 err = -EINVAL;
 797
 798         mutex_unlock(&queue_io_mutex);
 799         return err ? err : count;
 800 }
 801
 802 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 803
 804 static unsigned int
 805 cache_poll(struct file *filp, poll_table *wait)
 806 {
 807         unsigned int mask;
 808         struct cache_reader *rp = filp->private_data;
 809         struct cache_queue *cq;
 810         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 811
 812         poll_wait(filp, &queue_wait, wait);
 813
 814         /* alway allow write */
 815         mask = POLL_OUT | POLLWRNORM;
 816
 817         if (!rp)
 818                 return mask;
 819
 820         spin_lock(&queue_lock);
 821
 822         for (cq= &rp->q; &cq->list != &cd->queue;
 823              cq = list_entry(cq->list.next, struct cache_queue, list))
 824                 if (!cq->reader) {
 825                         mask |= POLLIN | POLLRDNORM;
 826                         break;
 827                 }
 828         spin_unlock(&queue_lock);
 829         return mask;
 830 }
 831
 832 static int
 833 cache_ioctl(struct inode *ino, struct file *filp,
 834             unsigned int cmd, unsigned long arg)
 835 {
 836         int len = 0;
 837         struct cache_reader *rp = filp->private_data;
 838         struct cache_queue *cq;
 839         struct cache_detail *cd = PDE(ino)->data;
 840
 841         if (cmd != FIONREAD || !rp)
 842                 return -EINVAL;
 843
 844         spin_lock(&queue_lock);
 845
 846         /* only find the length remaining in current request,
 847          * or the length of the next request
 848          */
 849         for (cq= &rp->q; &cq->list != &cd->queue;
 850              cq = list_entry(cq->list.next, struct cache_queue, list))
 851                 if (!cq->reader) {
 852                         struct cache_request *cr =
 853                                 container_of(cq, struct cache_request, q);
 854                         len = cr->len - rp->offset;
 855                         break;
 856                 }
 857         spin_unlock(&queue_lock);
 858
 859         return put_user(len, (int __user *)arg);
 860 }
 861
 862 static int
 863 cache_open(struct inode *inode, struct file *filp)
 864 {
 865         struct cache_reader *rp = NULL;
 866
 867         nonseekable_open(inode, filp);
 868         if (filp->f_mode & FMODE_READ) {
 869                 struct cache_detail *cd = PDE(inode)->data;
 870
 871                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 872                 if (!rp)
 873                         return -ENOMEM;
 874                 rp->offset = 0;
 875                 rp->q.reader = 1;
 876                 atomic_inc(&cd->readers);
 877                 spin_lock(&queue_lock);
 878                 list_add(&rp->q.list, &cd->queue);
 879                 spin_unlock(&queue_lock);
 880         }
 881         filp->private_data = rp;
 882         return 0;
 883 }
 884
 885 static int
 886 cache_release(struct inode *inode, struct file *filp)
 887 {
 888         struct cache_reader *rp = filp->private_data;
 889         struct cache_detail *cd = PDE(inode)->data;
 890
 891         if (rp) {
 892                 spin_lock(&queue_lock);
 893                 if (rp->offset) {
 894                         struct cache_queue *cq;
 895                         for (cq= &rp->q; &cq->list != &cd->queue;
 896                              cq = list_entry(cq->list.next, struct cache_queue, list))
 897                                 if (!cq->reader) {
 898                                         container_of(cq, struct cache_request, q)
 899                                                 ->readers--;
 900                                         break;
 901                                 }
 902                         rp->offset = 0;
 903                 }
 904                 list_del(&rp->q.list);
 905                 spin_unlock(&queue_lock);
 906
 907                 filp->private_data = NULL;
 908                 kfree(rp);
 909
 910                 cd->last_close = get_seconds();
 911                 atomic_dec(&cd->readers);
 912         }
 913         return 0;
 914 }
 915
 916
 917
 918 static const struct file_operations cache_file_operations = {
 919         .owner          = THIS_MODULE,
 920         .llseek         = no_llseek,
 921         .read           = cache_read,
 922         .write          = cache_write,
 923         .poll           = cache_poll,
 924         .ioctl          = cache_ioctl, /* for FIONREAD */
 925         .open           = cache_open,
 926         .release        = cache_release,
 927 };
 928
 929
 930 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
 931 {
 932         struct cache_queue *cq;
 933         spin_lock(&queue_lock);
 934         list_for_each_entry(cq, &detail->queue, list)
 935                 if (!cq->reader) {
 936                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 937                         if (cr->item != ch)
 938                                 continue;
 939                         if (cr->readers != 0)
 940                                 continue;
 941                         list_del(&cr->q.list);
 942                         spin_unlock(&queue_lock);
 943                         cache_put(cr->item, detail);
 944                         kfree(cr->buf);
 945                         kfree(cr);
 946                         return;
 947                 }
 948         spin_unlock(&queue_lock);
 949 }
 950
 951 /*
 952  * Support routines for text-based upcalls.
 953  * Fields are separated by spaces.
 954  * Fields are either mangled to quote space tab newline slosh with slosh
 955  * or a hexified with a leading \x
 956  * Record is terminated with newline.
 957  *
 958  */
 959
 960 void qword_add(char **bpp, int *lp, char *str)
 961 {
 962         char *bp = *bpp;
 963         int len = *lp;
 964         char c;
 965
 966         if (len < 0) return;
 967
 968         while ((c=*str++) && len)
 969                 switch(c) {
 970                 case ' ':
 971                 case '\t':
 972                 case '\n':
 973                 case '\\':
 974                         if (len >= 4) {
 975                                 *bp++ = '\\';
 976                                 *bp++ = '0' + ((c & 0300)>>6);
 977                                 *bp++ = '0' + ((c & 0070)>>3);
 978                                 *bp++ = '0' + ((c & 0007)>>0);
 979                         }
 980                         len -= 4;
 981                         break;
 982                 default:
 983                         *bp++ = c;
 984                         len--;
 985                 }
 986         if (c || len <1) len = -1;
 987         else {
 988                 *bp++ = ' ';
 989                 len--;
 990         }
 991         *bpp = bp;
 992         *lp = len;
 993 }
 994 EXPORT_SYMBOL(qword_add);
 995
 996 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
 997 {
 998         char *bp = *bpp;
 999         int len = *lp;
1000
1001         if (len < 0) return;
1002
1003         if (len > 2) {
1004                 *bp++ = '\\';
1005                 *bp++ = 'x';
1006                 len -= 2;
1007                 while (blen && len >= 2) {
1008                         unsigned char c = *buf++;
1009                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1010                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1011                         len -= 2;
1012                         blen--;
1013                 }
1014         }
1015         if (blen || len<1) len = -1;
1016         else {
1017                 *bp++ = ' ';
1018                 len--;
1019         }
1020         *bpp = bp;
1021         *lp = len;
1022 }
1023 EXPORT_SYMBOL(qword_addhex);
1024
1025 static void warn_no_listener(struct cache_detail *detail)
1026 {
1027         if (detail->last_warn != detail->last_close) {
1028                 detail->last_warn = detail->last_close;
1029                 if (detail->warn_no_listener)
1030                         detail->warn_no_listener(detail);
1031         }
1032 }
1033
1034 /*
1035  * register an upcall request to user-space.
1036  * Each request is at most one page long.
1037  */
1038 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
1039 {
1040
1041         char *buf;
1042         struct cache_request *crq;
1043         char *bp;
1044         int len;
1045
1046         if (detail->cache_request == NULL)
1047                 return -EINVAL;
1048
1049         if (atomic_read(&detail->readers) == 0 &&
1050             detail->last_close < get_seconds() - 30) {
1051                         warn_no_listener(detail);
1052                         return -EINVAL;
1053         }
1054
1055         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1056         if (!buf)
1057                 return -EAGAIN;
1058
1059         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1060         if (!crq) {
1061                 kfree(buf);
1062                 return -EAGAIN;
1063         }
1064
1065         bp = buf; len = PAGE_SIZE;
1066
1067         detail->cache_request(detail, h, &bp, &len);
1068
1069         if (len < 0) {
1070                 kfree(buf);
1071                 kfree(crq);
1072                 return -EAGAIN;
1073         }
1074         crq->q.reader = 0;
1075         crq->item = cache_get(h);
1076         crq->buf = buf;
1077         crq->len = PAGE_SIZE - len;
1078         crq->readers = 0;
1079         spin_lock(&queue_lock);
1080         list_add_tail(&crq->q.list, &detail->queue);
1081         spin_unlock(&queue_lock);
1082         wake_up(&queue_wait);
1083         return 0;
1084 }
1085
1086 /*
1087  * parse a message from user-space and pass it
1088  * to an appropriate cache
1089  * Messages are, like requests, separated into fields by
1090  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1091  *
1092  * Message is
1093  *   reply cachename expiry key ... content....
1094  *
1095  * key and content are both parsed by cache
1096  */
1097
1098 #define isodigit(c) (isdigit(c) && c <= '7')
1099 int qword_get(char **bpp, char *dest, int bufsize)
1100 {
1101         /* return bytes copied, or -1 on error */
1102         char *bp = *bpp;
1103         int len = 0;
1104
1105         while (*bp == ' ') bp++;
1106
1107         if (bp[0] == '\\' && bp[1] == 'x') {
1108                 /* HEX STRING */
1109                 bp += 2;
1110                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1111                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1112                         bp++;
1113                         byte <<= 4;
1114                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1115                         *dest++ = byte;
1116                         bp++;
1117                         len++;
1118                 }
1119         } else {
1120                 /* text with \nnn octal quoting */
1121                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1122                         if (*bp == '\\' &&
1123                             isodigit(bp[1]) && (bp[1] <= '3') &&
1124                             isodigit(bp[2]) &&
1125                             isodigit(bp[3])) {
1126                                 int byte = (*++bp -'0');
1127                                 bp++;
1128                                 byte = (byte << 3) | (*bp++ - '0');
1129                                 byte = (byte << 3) | (*bp++ - '0');
1130                                 *dest++ = byte;
1131                                 len++;
1132                         } else {
1133                                 *dest++ = *bp++;
1134                                 len++;
1135                         }
1136                 }
1137         }
1138
1139         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1140                 return -1;
1141         while (*bp == ' ') bp++;
1142         *bpp = bp;
1143         *dest = '\0';
1144         return len;
1145 }
1146 EXPORT_SYMBOL(qword_get);
1147
1148
1149 /*
1150  * support /proc/sunrpc/cache/$CACHENAME/content
1151  * as a seqfile.
1152  * We call ->cache_show passing NULL for the item to
1153  * get a header, then pass each real item in the cache
1154  */
1155
1156 struct handle {
1157         struct cache_detail *cd;
1158 };
1159
1160 static void *c_start(struct seq_file *m, loff_t *pos)
1161         __acquires(cd->hash_lock)
1162 {
1163         loff_t n = *pos;
1164         unsigned hash, entry;
1165         struct cache_head *ch;
1166         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1167
1168
1169         read_lock(&cd->hash_lock);
1170         if (!n--)
1171                 return SEQ_START_TOKEN;
1172         hash = n >> 32;
1173         entry = n & ((1LL<<32) - 1);
1174
1175         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1176                 if (!entry--)
1177                         return ch;
1178         n &= ~((1LL<<32) - 1);
1179         do {
1180                 hash++;
1181                 n += 1LL<<32;
1182         } while(hash < cd->hash_size &&
1183                 cd->hash_table[hash]==NULL);
1184         if (hash >= cd->hash_size)
1185                 return NULL;
1186         *pos = n+1;
1187         return cd->hash_table[hash];
1188 }
1189
1190 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1191 {
1192         struct cache_head *ch = p;
1193         int hash = (*pos >> 32);
1194         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1195
1196         if (p == SEQ_START_TOKEN)
1197                 hash = 0;
1198         else if (ch->next == NULL) {
1199                 hash++;
1200                 *pos += 1LL<<32;
1201         } else {
1202                 ++*pos;
1203                 return ch->next;
1204         }
1205         *pos &= ~((1LL<<32) - 1);
1206         while (hash < cd->hash_size &&
1207                cd->hash_table[hash] == NULL) {
1208                 hash++;
1209                 *pos += 1LL<<32;
1210         }
1211         if (hash >= cd->hash_size)
1212                 return NULL;
1213         ++*pos;
1214         return cd->hash_table[hash];
1215 }
1216
1217 static void c_stop(struct seq_file *m, void *p)
1218         __releases(cd->hash_lock)
1219 {
1220         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1221         read_unlock(&cd->hash_lock);
1222 }
1223
1224 static int c_show(struct seq_file *m, void *p)
1225 {
1226         struct cache_head *cp = p;
1227         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1228
1229         if (p == SEQ_START_TOKEN)
1230                 return cd->cache_show(m, cd, NULL);
1231
1232         ifdebug(CACHE)
1233                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1234                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1235         cache_get(cp);
1236         if (cache_check(cd, cp, NULL))
1237                 /* cache_check does a cache_put on failure */
1238                 seq_printf(m, "# ");
1239         else
1240                 cache_put(cp, cd);
1241
1242         return cd->cache_show(m, cd, cp);
1243 }
1244
1245 static const struct seq_operations cache_content_op = {
1246         .start  = c_start,
1247         .next   = c_next,
1248         .stop   = c_stop,
1249         .show   = c_show,
1250 };
1251
1252 static int content_open(struct inode *inode, struct file *file)
1253 {
1254         struct handle *han;
1255         struct cache_detail *cd = PDE(inode)->data;
1256
1257         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1258         if (han == NULL)
1259                 return -ENOMEM;
1260
1261         han->cd = cd;
1262         return 0;
1263 }
1264
1265 static const struct file_operations content_file_operations = {
1266         .open           = content_open,
1267         .read           = seq_read,
1268         .llseek         = seq_lseek,
1269         .release        = seq_release_private,
1270 };
1271
1272 static ssize_t read_flush(struct file *file, char __user *buf,
1273                             size_t count, loff_t *ppos)
1274 {
1275         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1276         char tbuf[20];
1277         unsigned long p = *ppos;
1278         size_t len;
1279
1280         sprintf(tbuf, "%lu\n", cd->flush_time);
1281         len = strlen(tbuf);
1282         if (p >= len)
1283                 return 0;
1284         len -= p;
1285         if (len > count)
1286                 len = count;
1287         if (copy_to_user(buf, (void*)(tbuf+p), len))
1288                 return -EFAULT;
1289         *ppos += len;
1290         return len;
1291 }
1292
1293 static ssize_t write_flush(struct file * file, const char __user * buf,
1294                              size_t count, loff_t *ppos)
1295 {
1296         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1297         char tbuf[20];
1298         char *ep;
1299         long flushtime;
1300         if (*ppos || count > sizeof(tbuf)-1)
1301                 return -EINVAL;
1302         if (copy_from_user(tbuf, buf, count))
1303                 return -EFAULT;
1304         tbuf[count] = 0;
1305         flushtime = simple_strtoul(tbuf, &ep, 0);
1306         if (*ep && *ep != '\n')
1307                 return -EINVAL;
1308
1309         cd->flush_time = flushtime;
1310         cd->nextcheck = get_seconds();
1311         cache_flush();
1312
1313         *ppos += count;
1314         return count;
1315 }
1316
1317 static const struct file_operations cache_flush_operations = {
1318         .open           = nonseekable_open,
1319         .read           = read_flush,
1320         .write          = write_flush,
1321 };