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