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