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