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