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