net/sunrpc/cache.c

   1 /*
   2  * net/sunrpc/cache.c
   3  *
   4  * Generic code for various authentication-related caches
   5  * used by sunrpc clients and servers.
   6  *
   7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   8  *
   9  * Released under terms in GPL version 2.  See COPYING.
  10  *
  11  */
  12
  13 #include <linux/types.h>
  14 #include <linux/fs.h>
  15 #include <linux/file.h>
  16 #include <linux/slab.h>
  17 #include <linux/signal.h>
  18 #include <linux/sched.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/module.h>
  22 #include <linux/ctype.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/poll.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/net.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/mutex.h>
  30 #include <asm/ioctls.h>
  31 #include <linux/sunrpc/types.h>
  32 #include <linux/sunrpc/cache.h>
  33 #include <linux/sunrpc/stats.h>
  34
  35 #define  RPCDBG_FACILITY RPCDBG_CACHE
  36
  37 static int cache_defer_req(struct cache_req *req, struct cache_head *item);
  38 static void cache_revisit_request(struct cache_head *item);
  39
  40 static void cache_init(struct cache_head *h)
  41 {
  42         time_t now = get_seconds();
  43         h->next = NULL;
  44         h->flags = 0;
  45         kref_init(&h->ref);
  46         h->expiry_time = now + CACHE_NEW_EXPIRY;
  47         h->last_refresh = now;
  48 }
  49
  50 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  51                                        struct cache_head *key, int hash)
  52 {
  53         struct cache_head **head,  **hp;
  54         struct cache_head *new = NULL;
  55
  56         head = &detail->hash_table[hash];
  57
  58         read_lock(&detail->hash_lock);
  59
  60         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  61                 struct cache_head *tmp = *hp;
  62                 if (detail->match(tmp, key)) {
  63                         cache_get(tmp);
  64                         read_unlock(&detail->hash_lock);
  65                         return tmp;
  66                 }
  67         }
  68         read_unlock(&detail->hash_lock);
  69         /* Didn't find anything, insert an empty entry */
  70
  71         new = detail->alloc();
  72         if (!new)
  73                 return NULL;
  74         /* must fully initialise 'new', else
  75          * we might get lose if we need to
  76          * cache_put it soon.
  77          */
  78         cache_init(new);
  79         detail->init(new, key);
  80
  81         write_lock(&detail->hash_lock);
  82
  83         /* check if entry appeared while we slept */
  84         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  85                 struct cache_head *tmp = *hp;
  86                 if (detail->match(tmp, key)) {
  87                         cache_get(tmp);
  88                         write_unlock(&detail->hash_lock);
  89                         cache_put(new, detail);
  90                         return tmp;
  91                 }
  92         }
  93         new->next = *head;
  94         *head = new;
  95         detail->entries++;
  96         cache_get(new);
  97         write_unlock(&detail->hash_lock);
  98
  99         return new;
 100 }
 101 EXPORT_SYMBOL(sunrpc_cache_lookup);
 102
 103
 104 static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
 105
 106 static int cache_fresh_locked(struct cache_head *head, time_t expiry)
 107 {
 108         head->expiry_time = expiry;
 109         head->last_refresh = get_seconds();
 110         return !test_and_set_bit(CACHE_VALID, &head->flags);
 111 }
 112
 113 static void cache_fresh_unlocked(struct cache_head *head,
 114                         struct cache_detail *detail, int new)
 115 {
 116         if (new)
 117                 cache_revisit_request(head);
 118         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 119                 cache_revisit_request(head);
 120                 queue_loose(detail, head);
 121         }
 122 }
 123
 124 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 125                                        struct cache_head *new, struct cache_head *old, int hash)
 126 {
 127         /* The 'old' entry is to be replaced by 'new'.
 128          * If 'old' is not VALID, we update it directly,
 129          * otherwise we need to replace it
 130          */
 131         struct cache_head **head;
 132         struct cache_head *tmp;
 133         int is_new;
 134
 135         if (!test_bit(CACHE_VALID, &old->flags)) {
 136                 write_lock(&detail->hash_lock);
 137                 if (!test_bit(CACHE_VALID, &old->flags)) {
 138                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 139                                 set_bit(CACHE_NEGATIVE, &old->flags);
 140                         else
 141                                 detail->update(old, new);
 142                         is_new = cache_fresh_locked(old, new->expiry_time);
 143                         write_unlock(&detail->hash_lock);
 144                         cache_fresh_unlocked(old, detail, is_new);
 145                         return old;
 146                 }
 147                 write_unlock(&detail->hash_lock);
 148         }
 149         /* We need to insert a new entry */
 150         tmp = detail->alloc();
 151         if (!tmp) {
 152                 cache_put(old, detail);
 153                 return NULL;
 154         }
 155         cache_init(tmp);
 156         detail->init(tmp, old);
 157         head = &detail->hash_table[hash];
 158
 159         write_lock(&detail->hash_lock);
 160         if (test_bit(CACHE_NEGATIVE, &new->flags))
 161                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 162         else
 163                 detail->update(tmp, new);
 164         tmp->next = *head;
 165         *head = tmp;
 166         detail->entries++;
 167         cache_get(tmp);
 168         is_new = cache_fresh_locked(tmp, new->expiry_time);
 169         cache_fresh_locked(old, 0);
 170         write_unlock(&detail->hash_lock);
 171         cache_fresh_unlocked(tmp, detail, is_new);
 172         cache_fresh_unlocked(old, detail, 0);
 173         cache_put(old, detail);
 174         return tmp;
 175 }
 176 EXPORT_SYMBOL(sunrpc_cache_update);
 177
 178 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h);
 179 /*
 180  * This is the generic cache management routine for all
 181  * the authentication caches.
 182  * It checks the currency of a cache item and will (later)
 183  * initiate an upcall to fill it if needed.
 184  *
 185  *
 186  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 187  * -EAGAIN if upcall is pending,
 188  * -ETIMEDOUT if upcall failed and should be retried,
 189  * -ENOENT if cache entry was negative
 190  */
 191 int cache_check(struct cache_detail *detail,
 192                     struct cache_head *h, struct cache_req *rqstp)
 193 {
 194         int rv;
 195         long refresh_age, age;
 196
 197         /* First decide return status as best we can */
 198         if (!test_bit(CACHE_VALID, &h->flags) ||
 199             h->expiry_time < get_seconds())
 200                 rv = -EAGAIN;
 201         else if (detail->flush_time > h->last_refresh)
 202                 rv = -EAGAIN;
 203         else {
 204                 /* entry is valid */
 205                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 206                         rv = -ENOENT;
 207                 else rv = 0;
 208         }
 209
 210         /* now see if we want to start an upcall */
 211         refresh_age = (h->expiry_time - h->last_refresh);
 212         age = get_seconds() - h->last_refresh;
 213
 214         if (rqstp == NULL) {
 215                 if (rv == -EAGAIN)
 216                         rv = -ENOENT;
 217         } else if (rv == -EAGAIN || age > refresh_age/2) {
 218                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 219                                 refresh_age, age);
 220                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 221                         switch (cache_make_upcall(detail, h)) {
 222                         case -EINVAL:
 223                                 clear_bit(CACHE_PENDING, &h->flags);
 224                                 if (rv == -EAGAIN) {
 225                                         set_bit(CACHE_NEGATIVE, &h->flags);
 226                                         cache_fresh_unlocked(h, detail,
 227                                              cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY));
 228                                         rv = -ENOENT;
 229                                 }
 230                                 break;
 231
 232                         case -EAGAIN:
 233                                 clear_bit(CACHE_PENDING, &h->flags);
 234                                 cache_revisit_request(h);
 235                                 break;
 236                         }
 237                 }
 238         }
 239
 240         if (rv == -EAGAIN)
 241                 if (cache_defer_req(rqstp, h) != 0)
 242                         rv = -ETIMEDOUT;
 243
 244         if (rv)
 245                 cache_put(h, detail);
 246         return rv;
 247 }
 248
 249 /*
 250  * caches need to be periodically cleaned.
 251  * For this we maintain a list of cache_detail and
 252  * a current pointer into that list and into the table
 253  * for that entry.
 254  *
 255  * Each time clean_cache is called it finds the next non-empty entry
 256  * in the current table and walks the list in that entry
 257  * looking for entries that can be removed.
 258  *
 259  * An entry gets removed if:
 260  * - The expiry is before current time
 261  * - The last_refresh time is before the flush_time for that cache
 262  *
 263  * later we might drop old entries with non-NEVER expiry if that table
 264  * is getting 'full' for some definition of 'full'
 265  *
 266  * The question of "how often to scan a table" is an interesting one
 267  * and is answered in part by the use of the "nextcheck" field in the
 268  * cache_detail.
 269  * When a scan of a table begins, the nextcheck field is set to a time
 270  * that is well into the future.
 271  * While scanning, if an expiry time is found that is earlier than the
 272  * current nextcheck time, nextcheck is set to that expiry time.
 273  * If the flush_time is ever set to a time earlier than the nextcheck
 274  * time, the nextcheck time is then set to that flush_time.
 275  *
 276  * A table is then only scanned if the current time is at least
 277  * the nextcheck time.
 278  *
 279  */
 280
 281 static LIST_HEAD(cache_list);
 282 static DEFINE_SPINLOCK(cache_list_lock);
 283 static struct cache_detail *current_detail;
 284 static int current_index;
 285
 286 static const struct file_operations cache_file_operations;
 287 static const struct file_operations content_file_operations;
 288 static const struct file_operations cache_flush_operations;
 289
 290 static void do_cache_clean(struct work_struct *work);
 291 static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
 292
 293 void cache_register(struct cache_detail *cd)
 294 {
 295         cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
 296         if (cd->proc_ent) {
 297                 struct proc_dir_entry *p;
 298                 cd->proc_ent->owner = cd->owner;
 299                 cd->channel_ent = cd->content_ent = NULL;
 300
 301                 p = create_proc_entry("flush", S_IFREG|S_IRUSR|S_IWUSR,
 302                                       cd->proc_ent);
 303                 cd->flush_ent =  p;
 304                 if (p) {
 305                         p->proc_fops = &cache_flush_operations;
 306                         p->owner = cd->owner;
 307                         p->data = cd;
 308                 }
 309
 310                 if (cd->cache_request || cd->cache_parse) {
 311                         p = create_proc_entry("channel", S_IFREG|S_IRUSR|S_IWUSR,
 312                                               cd->proc_ent);
 313                         cd->channel_ent = p;
 314                         if (p) {
 315                                 p->proc_fops = &cache_file_operations;
 316                                 p->owner = cd->owner;
 317                                 p->data = cd;
 318                         }
 319                 }
 320                 if (cd->cache_show) {
 321                         p = create_proc_entry("content", S_IFREG|S_IRUSR|S_IWUSR,
 322                                               cd->proc_ent);
 323                         cd->content_ent = p;
 324                         if (p) {
 325                                 p->proc_fops = &content_file_operations;
 326                                 p->owner = cd->owner;
 327                                 p->data = cd;
 328                         }
 329                 }
 330         }
 331         rwlock_init(&cd->hash_lock);
 332         INIT_LIST_HEAD(&cd->queue);
 333         spin_lock(&cache_list_lock);
 334         cd->nextcheck = 0;
 335         cd->entries = 0;
 336         atomic_set(&cd->readers, 0);
 337         cd->last_close = 0;
 338         cd->last_warn = -1;
 339         list_add(&cd->others, &cache_list);
 340         spin_unlock(&cache_list_lock);
 341
 342         /* start the cleaning process */
 343         schedule_delayed_work(&cache_cleaner, 0);
 344 }
 345
 346 int cache_unregister(struct cache_detail *cd)
 347 {
 348         cache_purge(cd);
 349         spin_lock(&cache_list_lock);
 350         write_lock(&cd->hash_lock);
 351         if (cd->entries || atomic_read(&cd->inuse)) {
 352                 write_unlock(&cd->hash_lock);
 353                 spin_unlock(&cache_list_lock);
 354                 return -EBUSY;
 355         }
 356         if (current_detail == cd)
 357                 current_detail = NULL;
 358         list_del_init(&cd->others);
 359         write_unlock(&cd->hash_lock);
 360         spin_unlock(&cache_list_lock);
 361         if (cd->proc_ent) {
 362                 if (cd->flush_ent)
 363                         remove_proc_entry("flush", cd->proc_ent);
 364                 if (cd->channel_ent)
 365                         remove_proc_entry("channel", cd->proc_ent);
 366                 if (cd->content_ent)
 367                         remove_proc_entry("content", cd->proc_ent);
 368
 369                 cd->proc_ent = NULL;
 370                 remove_proc_entry(cd->name, proc_net_rpc);
 371         }
 372         if (list_empty(&cache_list)) {
 373                 /* module must be being unloaded so its safe to kill the worker */
 374                 cancel_delayed_work(&cache_cleaner);
 375                 flush_scheduled_work();
 376         }
 377         return 0;
 378 }
 379
 380 /* clean cache tries to find something to clean
 381  * and cleans it.
 382  * It returns 1 if it cleaned something,
 383  *            0 if it didn't find anything this time
 384  *           -1 if it fell off the end of the list.
 385  */
 386 static int cache_clean(void)
 387 {
 388         int rv = 0;
 389         struct list_head *next;
 390
 391         spin_lock(&cache_list_lock);
 392
 393         /* find a suitable table if we don't already have one */
 394         while (current_detail == NULL ||
 395             current_index >= current_detail->hash_size) {
 396                 if (current_detail)
 397                         next = current_detail->others.next;
 398                 else
 399                         next = cache_list.next;
 400                 if (next == &cache_list) {
 401                         current_detail = NULL;
 402                         spin_unlock(&cache_list_lock);
 403                         return -1;
 404                 }
 405                 current_detail = list_entry(next, struct cache_detail, others);
 406                 if (current_detail->nextcheck > get_seconds())
 407                         current_index = current_detail->hash_size;
 408                 else {
 409                         current_index = 0;
 410                         current_detail->nextcheck = get_seconds()+30*60;
 411                 }
 412         }
 413
 414         /* find a non-empty bucket in the table */
 415         while (current_detail &&
 416                current_index < current_detail->hash_size &&
 417                current_detail->hash_table[current_index] == NULL)
 418                 current_index++;
 419
 420         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 421
 422         if (current_detail && current_index < current_detail->hash_size) {
 423                 struct cache_head *ch, **cp;
 424                 struct cache_detail *d;
 425
 426                 write_lock(&current_detail->hash_lock);
 427
 428                 /* Ok, now to clean this strand */
 429
 430                 cp = & current_detail->hash_table[current_index];
 431                 ch = *cp;
 432                 for (; ch; cp= & ch->next, ch= *cp) {
 433                         if (current_detail->nextcheck > ch->expiry_time)
 434                                 current_detail->nextcheck = ch->expiry_time+1;
 435                         if (ch->expiry_time >= get_seconds()
 436                             && ch->last_refresh >= current_detail->flush_time
 437                                 )
 438                                 continue;
 439                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 440                                 queue_loose(current_detail, ch);
 441
 442                         if (atomic_read(&ch->ref.refcount) == 1)
 443                                 break;
 444                 }
 445                 if (ch) {
 446                         *cp = ch->next;
 447                         ch->next = NULL;
 448                         current_detail->entries--;
 449                         rv = 1;
 450                 }
 451                 write_unlock(&current_detail->hash_lock);
 452                 d = current_detail;
 453                 if (!ch)
 454                         current_index ++;
 455                 spin_unlock(&cache_list_lock);
 456                 if (ch)
 457                         cache_put(ch, d);
 458         } else
 459                 spin_unlock(&cache_list_lock);
 460
 461         return rv;
 462 }
 463
 464 /*
 465  * We want to regularly clean the cache, so we need to schedule some work ...
 466  */
 467 static void do_cache_clean(struct work_struct *work)
 468 {
 469         int delay = 5;
 470         if (cache_clean() == -1)
 471                 delay = 30*HZ;
 472
 473         if (list_empty(&cache_list))
 474                 delay = 0;
 475
 476         if (delay)
 477                 schedule_delayed_work(&cache_cleaner, delay);
 478 }
 479
 480
 481 /*
 482  * Clean all caches promptly.  This just calls cache_clean
 483  * repeatedly until we are sure that every cache has had a chance to
 484  * be fully cleaned
 485  */
 486 void cache_flush(void)
 487 {
 488         while (cache_clean() != -1)
 489                 cond_resched();
 490         while (cache_clean() != -1)
 491                 cond_resched();
 492 }
 493
 494 void cache_purge(struct cache_detail *detail)
 495 {
 496         detail->flush_time = LONG_MAX;
 497         detail->nextcheck = get_seconds();
 498         cache_flush();
 499         detail->flush_time = 1;
 500 }
 501
 502
 503
 504 /*
 505  * Deferral and Revisiting of Requests.
 506  *
 507  * If a cache lookup finds a pending entry, we
 508  * need to defer the request and revisit it later.
 509  * All deferred requests are stored in a hash table,
 510  * indexed by "struct cache_head *".
 511  * As it may be wasteful to store a whole request
 512  * structure, we allow the request to provide a
 513  * deferred form, which must contain a
 514  * 'struct cache_deferred_req'
 515  * This cache_deferred_req contains a method to allow
 516  * it to be revisited when cache info is available
 517  */
 518
 519 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 520 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 521
 522 #define DFR_MAX 300     /* ??? */
 523
 524 static DEFINE_SPINLOCK(cache_defer_lock);
 525 static LIST_HEAD(cache_defer_list);
 526 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 527 static int cache_defer_cnt;
 528
 529 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 530 {
 531         struct cache_deferred_req *dreq;
 532         int hash = DFR_HASH(item);
 533
 534         if (cache_defer_cnt >= DFR_MAX) {
 535                 /* too much in the cache, randomly drop this one,
 536                  * or continue and drop the oldest below
 537                  */
 538                 if (net_random()&1)
 539                         return -ETIMEDOUT;
 540         }
 541         dreq = req->defer(req);
 542         if (dreq == NULL)
 543                 return -ETIMEDOUT;
 544
 545         dreq->item = item;
 546         dreq->recv_time = get_seconds();
 547
 548         spin_lock(&cache_defer_lock);
 549
 550         list_add(&dreq->recent, &cache_defer_list);
 551
 552         if (cache_defer_hash[hash].next == NULL)
 553                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 554         list_add(&dreq->hash, &cache_defer_hash[hash]);
 555
 556         /* it is in, now maybe clean up */
 557         dreq = NULL;
 558         if (++cache_defer_cnt > DFR_MAX) {
 559                 dreq = list_entry(cache_defer_list.prev,
 560                                   struct cache_deferred_req, recent);
 561                 list_del(&dreq->recent);
 562                 list_del(&dreq->hash);
 563                 cache_defer_cnt--;
 564         }
 565         spin_unlock(&cache_defer_lock);
 566
 567         if (dreq) {
 568                 /* there was one too many */
 569                 dreq->revisit(dreq, 1);
 570         }
 571         if (!test_bit(CACHE_PENDING, &item->flags)) {
 572                 /* must have just been validated... */
 573                 cache_revisit_request(item);
 574         }
 575         return 0;
 576 }
 577
 578 static void cache_revisit_request(struct cache_head *item)
 579 {
 580         struct cache_deferred_req *dreq;
 581         struct list_head pending;
 582
 583         struct list_head *lp;
 584         int hash = DFR_HASH(item);
 585
 586         INIT_LIST_HEAD(&pending);
 587         spin_lock(&cache_defer_lock);
 588
 589         lp = cache_defer_hash[hash].next;
 590         if (lp) {
 591                 while (lp != &cache_defer_hash[hash]) {
 592                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 593                         lp = lp->next;
 594                         if (dreq->item == item) {
 595                                 list_del(&dreq->hash);
 596                                 list_move(&dreq->recent, &pending);
 597                                 cache_defer_cnt--;
 598                         }
 599                 }
 600         }
 601         spin_unlock(&cache_defer_lock);
 602
 603         while (!list_empty(&pending)) {
 604                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 605                 list_del_init(&dreq->recent);
 606                 dreq->revisit(dreq, 0);
 607         }
 608 }
 609
 610 void cache_clean_deferred(void *owner)
 611 {
 612         struct cache_deferred_req *dreq, *tmp;
 613         struct list_head pending;
 614
 615
 616         INIT_LIST_HEAD(&pending);
 617         spin_lock(&cache_defer_lock);
 618
 619         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 620                 if (dreq->owner == owner) {
 621                         list_del(&dreq->hash);
 622                         list_move(&dreq->recent, &pending);
 623                         cache_defer_cnt--;
 624                 }
 625         }
 626         spin_unlock(&cache_defer_lock);
 627
 628         while (!list_empty(&pending)) {
 629                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 630                 list_del_init(&dreq->recent);
 631                 dreq->revisit(dreq, 1);
 632         }
 633 }
 634
 635 /*
 636  * communicate with user-space
 637  *
 638  * We have a magic /proc file - /proc/sunrpc/cache
 639  * On read, you get a full request, or block
 640  * On write, an update request is processed
 641  * Poll works if anything to read, and always allows write
 642  *
 643  * Implemented by linked list of requests.  Each open file has
 644  * a ->private that also exists in this list.  New request are added
 645  * to the end and may wakeup and preceding readers.
 646  * New readers are added to the head.  If, on read, an item is found with
 647  * CACHE_UPCALLING clear, we free it from the list.
 648  *
 649  */
 650
 651 static DEFINE_SPINLOCK(queue_lock);
 652 static DEFINE_MUTEX(queue_io_mutex);
 653
 654 struct cache_queue {
 655         struct list_head        list;
 656         int                     reader; /* if 0, then request */
 657 };
 658 struct cache_request {
 659         struct cache_queue      q;
 660         struct cache_head       *item;
 661         char                    * buf;
 662         int                     len;
 663         int                     readers;
 664 };
 665 struct cache_reader {
 666         struct cache_queue      q;
 667         int                     offset; /* if non-0, we have a refcnt on next request */
 668 };
 669
 670 static ssize_t
 671 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
 672 {
 673         struct cache_reader *rp = filp->private_data;
 674         struct cache_request *rq;
 675         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 676         int err;
 677
 678         if (count == 0)
 679                 return 0;
 680
 681         mutex_lock(&queue_io_mutex); /* protect against multiple concurrent
 682                               * readers on this file */
 683  again:
 684         spin_lock(&queue_lock);
 685         /* need to find next request */
 686         while (rp->q.list.next != &cd->queue &&
 687                list_entry(rp->q.list.next, struct cache_queue, list)
 688                ->reader) {
 689                 struct list_head *next = rp->q.list.next;
 690                 list_move(&rp->q.list, next);
 691         }
 692         if (rp->q.list.next == &cd->queue) {
 693                 spin_unlock(&queue_lock);
 694                 mutex_unlock(&queue_io_mutex);
 695                 BUG_ON(rp->offset);
 696                 return 0;
 697         }
 698         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 699         BUG_ON(rq->q.reader);
 700         if (rp->offset == 0)
 701                 rq->readers++;
 702         spin_unlock(&queue_lock);
 703
 704         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 705                 err = -EAGAIN;
 706                 spin_lock(&queue_lock);
 707                 list_move(&rp->q.list, &rq->q.list);
 708                 spin_unlock(&queue_lock);
 709         } else {
 710                 if (rp->offset + count > rq->len)
 711                         count = rq->len - rp->offset;
 712                 err = -EFAULT;
 713                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 714                         goto out;
 715                 rp->offset += count;
 716                 if (rp->offset >= rq->len) {
 717                         rp->offset = 0;
 718                         spin_lock(&queue_lock);
 719                         list_move(&rp->q.list, &rq->q.list);
 720                         spin_unlock(&queue_lock);
 721                 }
 722                 err = 0;
 723         }
 724  out:
 725         if (rp->offset == 0) {
 726                 /* need to release rq */
 727                 spin_lock(&queue_lock);
 728                 rq->readers--;
 729                 if (rq->readers == 0 &&
 730                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 731                         list_del(&rq->q.list);
 732                         spin_unlock(&queue_lock);
 733                         cache_put(rq->item, cd);
 734                         kfree(rq->buf);
 735                         kfree(rq);
 736                 } else
 737                         spin_unlock(&queue_lock);
 738         }
 739         if (err == -EAGAIN)
 740                 goto again;
 741         mutex_unlock(&queue_io_mutex);
 742         return err ? err :  count;
 743 }
 744
 745 static char write_buf[8192]; /* protected by queue_io_mutex */
 746
 747 static ssize_t
 748 cache_write(struct file *filp, const char __user *buf, size_t count,
 749             loff_t *ppos)
 750 {
 751         int err;
 752         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 753
 754         if (count == 0)
 755                 return 0;
 756         if (count >= sizeof(write_buf))
 757                 return -EINVAL;
 758
 759         mutex_lock(&queue_io_mutex);
 760
 761         if (copy_from_user(write_buf, buf, count)) {
 762                 mutex_unlock(&queue_io_mutex);
 763                 return -EFAULT;
 764         }
 765         write_buf[count] = '\0';
 766         if (cd->cache_parse)
 767                 err = cd->cache_parse(cd, write_buf, count);
 768         else
 769                 err = -EINVAL;
 770
 771         mutex_unlock(&queue_io_mutex);
 772         return err ? err : count;
 773 }
 774
 775 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 776
 777 static unsigned int
 778 cache_poll(struct file *filp, poll_table *wait)
 779 {
 780         unsigned int mask;
 781         struct cache_reader *rp = filp->private_data;
 782         struct cache_queue *cq;
 783         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 784
 785         poll_wait(filp, &queue_wait, wait);
 786
 787         /* alway allow write */
 788         mask = POLL_OUT | POLLWRNORM;
 789
 790         if (!rp)
 791                 return mask;
 792
 793         spin_lock(&queue_lock);
 794
 795         for (cq= &rp->q; &cq->list != &cd->queue;
 796              cq = list_entry(cq->list.next, struct cache_queue, list))
 797                 if (!cq->reader) {
 798                         mask |= POLLIN | POLLRDNORM;
 799                         break;
 800                 }
 801         spin_unlock(&queue_lock);
 802         return mask;
 803 }
 804
 805 static int
 806 cache_ioctl(struct inode *ino, struct file *filp,
 807             unsigned int cmd, unsigned long arg)
 808 {
 809         int len = 0;
 810         struct cache_reader *rp = filp->private_data;
 811         struct cache_queue *cq;
 812         struct cache_detail *cd = PDE(ino)->data;
 813
 814         if (cmd != FIONREAD || !rp)
 815                 return -EINVAL;
 816
 817         spin_lock(&queue_lock);
 818
 819         /* only find the length remaining in current request,
 820          * or the length of the next request
 821          */
 822         for (cq= &rp->q; &cq->list != &cd->queue;
 823              cq = list_entry(cq->list.next, struct cache_queue, list))
 824                 if (!cq->reader) {
 825                         struct cache_request *cr =
 826                                 container_of(cq, struct cache_request, q);
 827                         len = cr->len - rp->offset;
 828                         break;
 829                 }
 830         spin_unlock(&queue_lock);
 831
 832         return put_user(len, (int __user *)arg);
 833 }
 834
 835 static int
 836 cache_open(struct inode *inode, struct file *filp)
 837 {
 838         struct cache_reader *rp = NULL;
 839
 840         nonseekable_open(inode, filp);
 841         if (filp->f_mode & FMODE_READ) {
 842                 struct cache_detail *cd = PDE(inode)->data;
 843
 844                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 845                 if (!rp)
 846                         return -ENOMEM;
 847                 rp->offset = 0;
 848                 rp->q.reader = 1;
 849                 atomic_inc(&cd->readers);
 850                 spin_lock(&queue_lock);
 851                 list_add(&rp->q.list, &cd->queue);
 852                 spin_unlock(&queue_lock);
 853         }
 854         filp->private_data = rp;
 855         return 0;
 856 }
 857
 858 static int
 859 cache_release(struct inode *inode, struct file *filp)
 860 {
 861         struct cache_reader *rp = filp->private_data;
 862         struct cache_detail *cd = PDE(inode)->data;
 863
 864         if (rp) {
 865                 spin_lock(&queue_lock);
 866                 if (rp->offset) {
 867                         struct cache_queue *cq;
 868                         for (cq= &rp->q; &cq->list != &cd->queue;
 869                              cq = list_entry(cq->list.next, struct cache_queue, list))
 870                                 if (!cq->reader) {
 871                                         container_of(cq, struct cache_request, q)
 872                                                 ->readers--;
 873                                         break;
 874                                 }
 875                         rp->offset = 0;
 876                 }
 877                 list_del(&rp->q.list);
 878                 spin_unlock(&queue_lock);
 879
 880                 filp->private_data = NULL;
 881                 kfree(rp);
 882
 883                 cd->last_close = get_seconds();
 884                 atomic_dec(&cd->readers);
 885         }
 886         return 0;
 887 }
 888
 889
 890
 891 static const struct file_operations cache_file_operations = {
 892         .owner          = THIS_MODULE,
 893         .llseek         = no_llseek,
 894         .read           = cache_read,
 895         .write          = cache_write,
 896         .poll           = cache_poll,
 897         .ioctl          = cache_ioctl, /* for FIONREAD */
 898         .open           = cache_open,
 899         .release        = cache_release,
 900 };
 901
 902
 903 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
 904 {
 905         struct cache_queue *cq;
 906         spin_lock(&queue_lock);
 907         list_for_each_entry(cq, &detail->queue, list)
 908                 if (!cq->reader) {
 909                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 910                         if (cr->item != ch)
 911                                 continue;
 912                         if (cr->readers != 0)
 913                                 continue;
 914                         list_del(&cr->q.list);
 915                         spin_unlock(&queue_lock);
 916                         cache_put(cr->item, detail);
 917                         kfree(cr->buf);
 918                         kfree(cr);
 919                         return;
 920                 }
 921         spin_unlock(&queue_lock);
 922 }
 923
 924 /*
 925  * Support routines for text-based upcalls.
 926  * Fields are separated by spaces.
 927  * Fields are either mangled to quote space tab newline slosh with slosh
 928  * or a hexified with a leading \x
 929  * Record is terminated with newline.
 930  *
 931  */
 932
 933 void qword_add(char **bpp, int *lp, char *str)
 934 {
 935         char *bp = *bpp;
 936         int len = *lp;
 937         char c;
 938
 939         if (len < 0) return;
 940
 941         while ((c=*str++) && len)
 942                 switch(c) {
 943                 case ' ':
 944                 case '\t':
 945                 case '\n':
 946                 case '\\':
 947                         if (len >= 4) {
 948                                 *bp++ = '\\';
 949                                 *bp++ = '0' + ((c & 0300)>>6);
 950                                 *bp++ = '0' + ((c & 0070)>>3);
 951                                 *bp++ = '0' + ((c & 0007)>>0);
 952                         }
 953                         len -= 4;
 954                         break;
 955                 default:
 956                         *bp++ = c;
 957                         len--;
 958                 }
 959         if (c || len <1) len = -1;
 960         else {
 961                 *bp++ = ' ';
 962                 len--;
 963         }
 964         *bpp = bp;
 965         *lp = len;
 966 }
 967
 968 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
 969 {
 970         char *bp = *bpp;
 971         int len = *lp;
 972
 973         if (len < 0) return;
 974
 975         if (len > 2) {
 976                 *bp++ = '\\';
 977                 *bp++ = 'x';
 978                 len -= 2;
 979                 while (blen && len >= 2) {
 980                         unsigned char c = *buf++;
 981                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
 982                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
 983                         len -= 2;
 984                         blen--;
 985                 }
 986         }
 987         if (blen || len<1) len = -1;
 988         else {
 989                 *bp++ = ' ';
 990                 len--;
 991         }
 992         *bpp = bp;
 993         *lp = len;
 994 }
 995
 996 static void warn_no_listener(struct cache_detail *detail)
 997 {
 998         if (detail->last_warn != detail->last_close) {
 999                 detail->last_warn = detail->last_close;
1000                 if (detail->warn_no_listener)
1001                         detail->warn_no_listener(detail);
1002         }
1003 }
1004
1005 /*
1006  * register an upcall request to user-space.
1007  * Each request is at most one page long.
1008  */
1009 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
1010 {
1011
1012         char *buf;
1013         struct cache_request *crq;
1014         char *bp;
1015         int len;
1016
1017         if (detail->cache_request == NULL)
1018                 return -EINVAL;
1019
1020         if (atomic_read(&detail->readers) == 0 &&
1021             detail->last_close < get_seconds() - 30) {
1022                         warn_no_listener(detail);
1023                         return -EINVAL;
1024         }
1025
1026         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1027         if (!buf)
1028                 return -EAGAIN;
1029
1030         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1031         if (!crq) {
1032                 kfree(buf);
1033                 return -EAGAIN;
1034         }
1035
1036         bp = buf; len = PAGE_SIZE;
1037
1038         detail->cache_request(detail, h, &bp, &len);
1039
1040         if (len < 0) {
1041                 kfree(buf);
1042                 kfree(crq);
1043                 return -EAGAIN;
1044         }
1045         crq->q.reader = 0;
1046         crq->item = cache_get(h);
1047         crq->buf = buf;
1048         crq->len = PAGE_SIZE - len;
1049         crq->readers = 0;
1050         spin_lock(&queue_lock);
1051         list_add_tail(&crq->q.list, &detail->queue);
1052         spin_unlock(&queue_lock);
1053         wake_up(&queue_wait);
1054         return 0;
1055 }
1056
1057 /*
1058  * parse a message from user-space and pass it
1059  * to an appropriate cache
1060  * Messages are, like requests, separated into fields by
1061  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1062  *
1063  * Message is
1064  *   reply cachename expiry key ... content....
1065  *
1066  * key and content are both parsed by cache
1067  */
1068
1069 #define isodigit(c) (isdigit(c) && c <= '7')
1070 int qword_get(char **bpp, char *dest, int bufsize)
1071 {
1072         /* return bytes copied, or -1 on error */
1073         char *bp = *bpp;
1074         int len = 0;
1075
1076         while (*bp == ' ') bp++;
1077
1078         if (bp[0] == '\\' && bp[1] == 'x') {
1079                 /* HEX STRING */
1080                 bp += 2;
1081                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1082                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1083                         bp++;
1084                         byte <<= 4;
1085                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1086                         *dest++ = byte;
1087                         bp++;
1088                         len++;
1089                 }
1090         } else {
1091                 /* text with \nnn octal quoting */
1092                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1093                         if (*bp == '\\' &&
1094                             isodigit(bp[1]) && (bp[1] <= '3') &&
1095                             isodigit(bp[2]) &&
1096                             isodigit(bp[3])) {
1097                                 int byte = (*++bp -'0');
1098                                 bp++;
1099                                 byte = (byte << 3) | (*bp++ - '0');
1100                                 byte = (byte << 3) | (*bp++ - '0');
1101                                 *dest++ = byte;
1102                                 len++;
1103                         } else {
1104                                 *dest++ = *bp++;
1105                                 len++;
1106                         }
1107                 }
1108         }
1109
1110         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1111                 return -1;
1112         while (*bp == ' ') bp++;
1113         *bpp = bp;
1114         *dest = '\0';
1115         return len;
1116 }
1117
1118
1119 /*
1120  * support /proc/sunrpc/cache/$CACHENAME/content
1121  * as a seqfile.
1122  * We call ->cache_show passing NULL for the item to
1123  * get a header, then pass each real item in the cache
1124  */
1125
1126 struct handle {
1127         struct cache_detail *cd;
1128 };
1129
1130 static void *c_start(struct seq_file *m, loff_t *pos)
1131 {
1132         loff_t n = *pos;
1133         unsigned hash, entry;
1134         struct cache_head *ch;
1135         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1136
1137
1138         read_lock(&cd->hash_lock);
1139         if (!n--)
1140                 return SEQ_START_TOKEN;
1141         hash = n >> 32;
1142         entry = n & ((1LL<<32) - 1);
1143
1144         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1145                 if (!entry--)
1146                         return ch;
1147         n &= ~((1LL<<32) - 1);
1148         do {
1149                 hash++;
1150                 n += 1LL<<32;
1151         } while(hash < cd->hash_size &&
1152                 cd->hash_table[hash]==NULL);
1153         if (hash >= cd->hash_size)
1154                 return NULL;
1155         *pos = n+1;
1156         return cd->hash_table[hash];
1157 }
1158
1159 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1160 {
1161         struct cache_head *ch = p;
1162         int hash = (*pos >> 32);
1163         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1164
1165         if (p == SEQ_START_TOKEN)
1166                 hash = 0;
1167         else if (ch->next == NULL) {
1168                 hash++;
1169                 *pos += 1LL<<32;
1170         } else {
1171                 ++*pos;
1172                 return ch->next;
1173         }
1174         *pos &= ~((1LL<<32) - 1);
1175         while (hash < cd->hash_size &&
1176                cd->hash_table[hash] == NULL) {
1177                 hash++;
1178                 *pos += 1LL<<32;
1179         }
1180         if (hash >= cd->hash_size)
1181                 return NULL;
1182         ++*pos;
1183         return cd->hash_table[hash];
1184 }
1185
1186 static void c_stop(struct seq_file *m, void *p)
1187 {
1188         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1189         read_unlock(&cd->hash_lock);
1190 }
1191
1192 static int c_show(struct seq_file *m, void *p)
1193 {
1194         struct cache_head *cp = p;
1195         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1196
1197         if (p == SEQ_START_TOKEN)
1198                 return cd->cache_show(m, cd, NULL);
1199
1200         ifdebug(CACHE)
1201                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1202                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1203         cache_get(cp);
1204         if (cache_check(cd, cp, NULL))
1205                 /* cache_check does a cache_put on failure */
1206                 seq_printf(m, "# ");
1207         else
1208                 cache_put(cp, cd);
1209
1210         return cd->cache_show(m, cd, cp);
1211 }
1212
1213 static struct seq_operations cache_content_op = {
1214         .start  = c_start,
1215         .next   = c_next,
1216         .stop   = c_stop,
1217         .show   = c_show,
1218 };
1219
1220 static int content_open(struct inode *inode, struct file *file)
1221 {
1222         int res;
1223         struct handle *han;
1224         struct cache_detail *cd = PDE(inode)->data;
1225
1226         han = kmalloc(sizeof(*han), GFP_KERNEL);
1227         if (han == NULL)
1228                 return -ENOMEM;
1229
1230         han->cd = cd;
1231
1232         res = seq_open(file, &cache_content_op);
1233         if (res)
1234                 kfree(han);
1235         else
1236                 ((struct seq_file *)file->private_data)->private = han;
1237
1238         return res;
1239 }
1240 static int content_release(struct inode *inode, struct file *file)
1241 {
1242         struct seq_file *m = (struct seq_file *)file->private_data;
1243         struct handle *han = m->private;
1244         kfree(han);
1245         m->private = NULL;
1246         return seq_release(inode, file);
1247 }
1248
1249 static const struct file_operations content_file_operations = {
1250         .open           = content_open,
1251         .read           = seq_read,
1252         .llseek         = seq_lseek,
1253         .release        = content_release,
1254 };
1255
1256 static ssize_t read_flush(struct file *file, char __user *buf,
1257                             size_t count, loff_t *ppos)
1258 {
1259         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1260         char tbuf[20];
1261         unsigned long p = *ppos;
1262         int len;
1263
1264         sprintf(tbuf, "%lu\n", cd->flush_time);
1265         len = strlen(tbuf);
1266         if (p >= len)
1267                 return 0;
1268         len -= p;
1269         if (len > count) len = count;
1270         if (copy_to_user(buf, (void*)(tbuf+p), len))
1271                 len = -EFAULT;
1272         else
1273                 *ppos += len;
1274         return len;
1275 }
1276
1277 static ssize_t write_flush(struct file * file, const char __user * buf,
1278                              size_t count, loff_t *ppos)
1279 {
1280         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1281         char tbuf[20];
1282         char *ep;
1283         long flushtime;
1284         if (*ppos || count > sizeof(tbuf)-1)
1285                 return -EINVAL;
1286         if (copy_from_user(tbuf, buf, count))
1287                 return -EFAULT;
1288         tbuf[count] = 0;
1289         flushtime = simple_strtoul(tbuf, &ep, 0);
1290         if (*ep && *ep != '\n')
1291                 return -EINVAL;
1292
1293         cd->flush_time = flushtime;
1294         cd->nextcheck = get_seconds();
1295         cache_flush();
1296
1297         *ppos += count;
1298         return count;
1299 }
1300
1301 static const struct file_operations cache_flush_operations = {
1302         .open           = nonseekable_open,
1303         .read           = read_flush,
1304         .write          = write_flush,
1305 };