SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/sunrpc/cache.c

   1 /*
   2  * net/sunrpc/cache.c
   3  *
   4  * Generic code for various authentication-related caches
   5  * used by sunrpc clients and servers.
   6  *
   7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   8  *
   9  * Released under terms in GPL version 2.  See COPYING.
  10  *
  11  */
  12
  13 #include <linux/types.h>
  14 #include <linux/fs.h>
  15 #include <linux/file.h>
  16 #include <linux/slab.h>
  17 #include <linux/signal.h>
  18 #include <linux/sched.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/module.h>
  22 #include <linux/ctype.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/poll.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/net.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/mutex.h>
  30 #include <asm/ioctls.h>
  31 #include <linux/sunrpc/types.h>
  32 #include <linux/sunrpc/cache.h>
  33 #include <linux/sunrpc/stats.h>
  34
  35 #define  RPCDBG_FACILITY RPCDBG_CACHE
  36
  37 static 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 #ifdef CONFIG_PROC_FS
 308 static int create_cache_proc_entries(struct cache_detail *cd)
 309 {
 310         struct proc_dir_entry *p;
 311
 312         cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
 313         if (cd->proc_ent == NULL)
 314                 goto out_nomem;
 315         cd->proc_ent->owner = cd->owner;
 316         cd->channel_ent = cd->content_ent = NULL;
 317
 318         p = create_proc_entry("flush", S_IFREG|S_IRUSR|S_IWUSR, cd->proc_ent);
 319         cd->flush_ent = p;
 320         if (p == NULL)
 321                 goto out_nomem;
 322         p->proc_fops = &cache_flush_operations;
 323         p->owner = cd->owner;
 324         p->data = cd;
 325
 326         if (cd->cache_request || cd->cache_parse) {
 327                 p = create_proc_entry("channel", S_IFREG|S_IRUSR|S_IWUSR,
 328                                       cd->proc_ent);
 329                 cd->channel_ent = p;
 330                 if (p == NULL)
 331                         goto out_nomem;
 332                 p->proc_fops = &cache_file_operations;
 333                 p->owner = cd->owner;
 334                 p->data = cd;
 335         }
 336         if (cd->cache_show) {
 337                 p = create_proc_entry("content", S_IFREG|S_IRUSR|S_IWUSR,
 338                                       cd->proc_ent);
 339                 cd->content_ent = p;
 340                 if (p == NULL)
 341                         goto out_nomem;
 342                 p->proc_fops = &content_file_operations;
 343                 p->owner = cd->owner;
 344                 p->data = cd;
 345         }
 346         return 0;
 347 out_nomem:
 348         remove_cache_proc_entries(cd);
 349         return -ENOMEM;
 350 }
 351 #else /* CONFIG_PROC_FS */
 352 static int create_cache_proc_entries(struct cache_detail *cd)
 353 {
 354         return 0;
 355 }
 356 #endif
 357
 358 int cache_register(struct cache_detail *cd)
 359 {
 360         int ret;
 361
 362         ret = create_cache_proc_entries(cd);
 363         if (ret)
 364                 return ret;
 365         rwlock_init(&cd->hash_lock);
 366         INIT_LIST_HEAD(&cd->queue);
 367         spin_lock(&cache_list_lock);
 368         cd->nextcheck = 0;
 369         cd->entries = 0;
 370         atomic_set(&cd->readers, 0);
 371         cd->last_close = 0;
 372         cd->last_warn = -1;
 373         list_add(&cd->others, &cache_list);
 374         spin_unlock(&cache_list_lock);
 375
 376         /* start the cleaning process */
 377         schedule_delayed_work(&cache_cleaner, 0);
 378         return 0;
 379 }
 380
 381 void cache_unregister(struct cache_detail *cd)
 382 {
 383         cache_purge(cd);
 384         spin_lock(&cache_list_lock);
 385         write_lock(&cd->hash_lock);
 386         if (cd->entries || atomic_read(&cd->inuse)) {
 387                 write_unlock(&cd->hash_lock);
 388                 spin_unlock(&cache_list_lock);
 389                 goto out;
 390         }
 391         if (current_detail == cd)
 392                 current_detail = NULL;
 393         list_del_init(&cd->others);
 394         write_unlock(&cd->hash_lock);
 395         spin_unlock(&cache_list_lock);
 396         remove_cache_proc_entries(cd);
 397         if (list_empty(&cache_list)) {
 398                 /* module must be being unloaded so its safe to kill the worker */
 399                 cancel_delayed_work_sync(&cache_cleaner);
 400         }
 401         return;
 402 out:
 403         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 404 }
 405
 406 /* clean cache tries to find something to clean
 407  * and cleans it.
 408  * It returns 1 if it cleaned something,
 409  *            0 if it didn't find anything this time
 410  *           -1 if it fell off the end of the list.
 411  */
 412 static int cache_clean(void)
 413 {
 414         int rv = 0;
 415         struct list_head *next;
 416
 417         spin_lock(&cache_list_lock);
 418
 419         /* find a suitable table if we don't already have one */
 420         while (current_detail == NULL ||
 421             current_index >= current_detail->hash_size) {
 422                 if (current_detail)
 423                         next = current_detail->others.next;
 424                 else
 425                         next = cache_list.next;
 426                 if (next == &cache_list) {
 427                         current_detail = NULL;
 428                         spin_unlock(&cache_list_lock);
 429                         return -1;
 430                 }
 431                 current_detail = list_entry(next, struct cache_detail, others);
 432                 if (current_detail->nextcheck > get_seconds())
 433                         current_index = current_detail->hash_size;
 434                 else {
 435                         current_index = 0;
 436                         current_detail->nextcheck = get_seconds()+30*60;
 437                 }
 438         }
 439
 440         /* find a non-empty bucket in the table */
 441         while (current_detail &&
 442                current_index < current_detail->hash_size &&
 443                current_detail->hash_table[current_index] == NULL)
 444                 current_index++;
 445
 446         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 447
 448         if (current_detail && current_index < current_detail->hash_size) {
 449                 struct cache_head *ch, **cp;
 450                 struct cache_detail *d;
 451
 452                 write_lock(&current_detail->hash_lock);
 453
 454                 /* Ok, now to clean this strand */
 455
 456                 cp = & current_detail->hash_table[current_index];
 457                 ch = *cp;
 458                 for (; ch; cp= & ch->next, ch= *cp) {
 459                         if (current_detail->nextcheck > ch->expiry_time)
 460                                 current_detail->nextcheck = ch->expiry_time+1;
 461                         if (ch->expiry_time >= get_seconds()
 462                             && ch->last_refresh >= current_detail->flush_time
 463                                 )
 464                                 continue;
 465                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 466                                 queue_loose(current_detail, ch);
 467
 468                         if (atomic_read(&ch->ref.refcount) == 1)
 469                                 break;
 470                 }
 471                 if (ch) {
 472                         *cp = ch->next;
 473                         ch->next = NULL;
 474                         current_detail->entries--;
 475                         rv = 1;
 476                 }
 477                 write_unlock(&current_detail->hash_lock);
 478                 d = current_detail;
 479                 if (!ch)
 480                         current_index ++;
 481                 spin_unlock(&cache_list_lock);
 482                 if (ch)
 483                         cache_put(ch, d);
 484         } else
 485                 spin_unlock(&cache_list_lock);
 486
 487         return rv;
 488 }
 489
 490 /*
 491  * We want to regularly clean the cache, so we need to schedule some work ...
 492  */
 493 static void do_cache_clean(struct work_struct *work)
 494 {
 495         int delay = 5;
 496         if (cache_clean() == -1)
 497                 delay = 30*HZ;
 498
 499         if (list_empty(&cache_list))
 500                 delay = 0;
 501
 502         if (delay)
 503                 schedule_delayed_work(&cache_cleaner, delay);
 504 }
 505
 506
 507 /*
 508  * Clean all caches promptly.  This just calls cache_clean
 509  * repeatedly until we are sure that every cache has had a chance to
 510  * be fully cleaned
 511  */
 512 void cache_flush(void)
 513 {
 514         while (cache_clean() != -1)
 515                 cond_resched();
 516         while (cache_clean() != -1)
 517                 cond_resched();
 518 }
 519
 520 void cache_purge(struct cache_detail *detail)
 521 {
 522         detail->flush_time = LONG_MAX;
 523         detail->nextcheck = get_seconds();
 524         cache_flush();
 525         detail->flush_time = 1;
 526 }
 527
 528
 529
 530 /*
 531  * Deferral and Revisiting of Requests.
 532  *
 533  * If a cache lookup finds a pending entry, we
 534  * need to defer the request and revisit it later.
 535  * All deferred requests are stored in a hash table,
 536  * indexed by "struct cache_head *".
 537  * As it may be wasteful to store a whole request
 538  * structure, we allow the request to provide a
 539  * deferred form, which must contain a
 540  * 'struct cache_deferred_req'
 541  * This cache_deferred_req contains a method to allow
 542  * it to be revisited when cache info is available
 543  */
 544
 545 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 546 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 547
 548 #define DFR_MAX 300     /* ??? */
 549
 550 static DEFINE_SPINLOCK(cache_defer_lock);
 551 static LIST_HEAD(cache_defer_list);
 552 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 553 static int cache_defer_cnt;
 554
 555 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 556 {
 557         struct cache_deferred_req *dreq;
 558         int hash = DFR_HASH(item);
 559
 560         if (cache_defer_cnt >= DFR_MAX) {
 561                 /* too much in the cache, randomly drop this one,
 562                  * or continue and drop the oldest below
 563                  */
 564                 if (net_random()&1)
 565                         return -ETIMEDOUT;
 566         }
 567         dreq = req->defer(req);
 568         if (dreq == NULL)
 569                 return -ETIMEDOUT;
 570
 571         dreq->item = item;
 572         dreq->recv_time = get_seconds();
 573
 574         spin_lock(&cache_defer_lock);
 575
 576         list_add(&dreq->recent, &cache_defer_list);
 577
 578         if (cache_defer_hash[hash].next == NULL)
 579                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 580         list_add(&dreq->hash, &cache_defer_hash[hash]);
 581
 582         /* it is in, now maybe clean up */
 583         dreq = NULL;
 584         if (++cache_defer_cnt > DFR_MAX) {
 585                 dreq = list_entry(cache_defer_list.prev,
 586                                   struct cache_deferred_req, recent);
 587                 list_del(&dreq->recent);
 588                 list_del(&dreq->hash);
 589                 cache_defer_cnt--;
 590         }
 591         spin_unlock(&cache_defer_lock);
 592
 593         if (dreq) {
 594                 /* there was one too many */
 595                 dreq->revisit(dreq, 1);
 596         }
 597         if (!test_bit(CACHE_PENDING, &item->flags)) {
 598                 /* must have just been validated... */
 599                 cache_revisit_request(item);
 600         }
 601         return 0;
 602 }
 603
 604 static void cache_revisit_request(struct cache_head *item)
 605 {
 606         struct cache_deferred_req *dreq;
 607         struct list_head pending;
 608
 609         struct list_head *lp;
 610         int hash = DFR_HASH(item);
 611
 612         INIT_LIST_HEAD(&pending);
 613         spin_lock(&cache_defer_lock);
 614
 615         lp = cache_defer_hash[hash].next;
 616         if (lp) {
 617                 while (lp != &cache_defer_hash[hash]) {
 618                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 619                         lp = lp->next;
 620                         if (dreq->item == item) {
 621                                 list_del(&dreq->hash);
 622                                 list_move(&dreq->recent, &pending);
 623                                 cache_defer_cnt--;
 624                         }
 625                 }
 626         }
 627         spin_unlock(&cache_defer_lock);
 628
 629         while (!list_empty(&pending)) {
 630                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 631                 list_del_init(&dreq->recent);
 632                 dreq->revisit(dreq, 0);
 633         }
 634 }
 635
 636 void cache_clean_deferred(void *owner)
 637 {
 638         struct cache_deferred_req *dreq, *tmp;
 639         struct list_head pending;
 640
 641
 642         INIT_LIST_HEAD(&pending);
 643         spin_lock(&cache_defer_lock);
 644
 645         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 646                 if (dreq->owner == owner) {
 647                         list_del(&dreq->hash);
 648                         list_move(&dreq->recent, &pending);
 649                         cache_defer_cnt--;
 650                 }
 651         }
 652         spin_unlock(&cache_defer_lock);
 653
 654         while (!list_empty(&pending)) {
 655                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 656                 list_del_init(&dreq->recent);
 657                 dreq->revisit(dreq, 1);
 658         }
 659 }
 660
 661 /*
 662  * communicate with user-space
 663  *
 664  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 665  * On read, you get a full request, or block.
 666  * On write, an update request is processed.
 667  * Poll works if anything to read, and always allows write.
 668  *
 669  * Implemented by linked list of requests.  Each open file has
 670  * a ->private that also exists in this list.  New requests are added
 671  * to the end and may wakeup and preceding readers.
 672  * New readers are added to the head.  If, on read, an item is found with
 673  * CACHE_UPCALLING clear, we free it from the list.
 674  *
 675  */
 676
 677 static DEFINE_SPINLOCK(queue_lock);
 678 static DEFINE_MUTEX(queue_io_mutex);
 679
 680 struct cache_queue {
 681         struct list_head        list;
 682         int                     reader; /* if 0, then request */
 683 };
 684 struct cache_request {
 685         struct cache_queue      q;
 686         struct cache_head       *item;
 687         char                    * buf;
 688         int                     len;
 689         int                     readers;
 690 };
 691 struct cache_reader {
 692         struct cache_queue      q;
 693         int                     offset; /* if non-0, we have a refcnt on next request */
 694 };
 695
 696 static ssize_t
 697 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
 698 {
 699         struct cache_reader *rp = filp->private_data;
 700         struct cache_request *rq;
 701         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 702         int err;
 703
 704         if (count == 0)
 705                 return 0;
 706
 707         mutex_lock(&queue_io_mutex); /* protect against multiple concurrent
 708                               * readers on this file */
 709  again:
 710         spin_lock(&queue_lock);
 711         /* need to find next request */
 712         while (rp->q.list.next != &cd->queue &&
 713                list_entry(rp->q.list.next, struct cache_queue, list)
 714                ->reader) {
 715                 struct list_head *next = rp->q.list.next;
 716                 list_move(&rp->q.list, next);
 717         }
 718         if (rp->q.list.next == &cd->queue) {
 719                 spin_unlock(&queue_lock);
 720                 mutex_unlock(&queue_io_mutex);
 721                 BUG_ON(rp->offset);
 722                 return 0;
 723         }
 724         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 725         BUG_ON(rq->q.reader);
 726         if (rp->offset == 0)
 727                 rq->readers++;
 728         spin_unlock(&queue_lock);
 729
 730         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 731                 err = -EAGAIN;
 732                 spin_lock(&queue_lock);
 733                 list_move(&rp->q.list, &rq->q.list);
 734                 spin_unlock(&queue_lock);
 735         } else {
 736                 if (rp->offset + count > rq->len)
 737                         count = rq->len - rp->offset;
 738                 err = -EFAULT;
 739                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 740                         goto out;
 741                 rp->offset += count;
 742                 if (rp->offset >= rq->len) {
 743                         rp->offset = 0;
 744                         spin_lock(&queue_lock);
 745                         list_move(&rp->q.list, &rq->q.list);
 746                         spin_unlock(&queue_lock);
 747                 }
 748                 err = 0;
 749         }
 750  out:
 751         if (rp->offset == 0) {
 752                 /* need to release rq */
 753                 spin_lock(&queue_lock);
 754                 rq->readers--;
 755                 if (rq->readers == 0 &&
 756                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 757                         list_del(&rq->q.list);
 758                         spin_unlock(&queue_lock);
 759                         cache_put(rq->item, cd);
 760                         kfree(rq->buf);
 761                         kfree(rq);
 762                 } else
 763                         spin_unlock(&queue_lock);
 764         }
 765         if (err == -EAGAIN)
 766                 goto again;
 767         mutex_unlock(&queue_io_mutex);
 768         return err ? err :  count;
 769 }
 770
 771 static char write_buf[8192]; /* protected by queue_io_mutex */
 772
 773 static ssize_t
 774 cache_write(struct file *filp, const char __user *buf, size_t count,
 775             loff_t *ppos)
 776 {
 777         int err;
 778         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 779
 780         if (count == 0)
 781                 return 0;
 782         if (count >= sizeof(write_buf))
 783                 return -EINVAL;
 784
 785         mutex_lock(&queue_io_mutex);
 786
 787         if (copy_from_user(write_buf, buf, count)) {
 788                 mutex_unlock(&queue_io_mutex);
 789                 return -EFAULT;
 790         }
 791         write_buf[count] = '\0';
 792         if (cd->cache_parse)
 793                 err = cd->cache_parse(cd, write_buf, count);
 794         else
 795                 err = -EINVAL;
 796
 797         mutex_unlock(&queue_io_mutex);
 798         return err ? err : count;
 799 }
 800
 801 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 802
 803 static unsigned int
 804 cache_poll(struct file *filp, poll_table *wait)
 805 {
 806         unsigned int mask;
 807         struct cache_reader *rp = filp->private_data;
 808         struct cache_queue *cq;
 809         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 810
 811         poll_wait(filp, &queue_wait, wait);
 812
 813         /* alway allow write */
 814         mask = POLL_OUT | POLLWRNORM;
 815
 816         if (!rp)
 817                 return mask;
 818
 819         spin_lock(&queue_lock);
 820
 821         for (cq= &rp->q; &cq->list != &cd->queue;
 822              cq = list_entry(cq->list.next, struct cache_queue, list))
 823                 if (!cq->reader) {
 824                         mask |= POLLIN | POLLRDNORM;
 825                         break;
 826                 }
 827         spin_unlock(&queue_lock);
 828         return mask;
 829 }
 830
 831 static int
 832 cache_ioctl(struct inode *ino, struct file *filp,
 833             unsigned int cmd, unsigned long arg)
 834 {
 835         int len = 0;
 836         struct cache_reader *rp = filp->private_data;
 837         struct cache_queue *cq;
 838         struct cache_detail *cd = PDE(ino)->data;
 839
 840         if (cmd != FIONREAD || !rp)
 841                 return -EINVAL;
 842
 843         spin_lock(&queue_lock);
 844
 845         /* only find the length remaining in current request,
 846          * or the length of the next request
 847          */
 848         for (cq= &rp->q; &cq->list != &cd->queue;
 849              cq = list_entry(cq->list.next, struct cache_queue, list))
 850                 if (!cq->reader) {
 851                         struct cache_request *cr =
 852                                 container_of(cq, struct cache_request, q);
 853                         len = cr->len - rp->offset;
 854                         break;
 855                 }
 856         spin_unlock(&queue_lock);
 857
 858         return put_user(len, (int __user *)arg);
 859 }
 860
 861 static int
 862 cache_open(struct inode *inode, struct file *filp)
 863 {
 864         struct cache_reader *rp = NULL;
 865
 866         nonseekable_open(inode, filp);
 867         if (filp->f_mode & FMODE_READ) {
 868                 struct cache_detail *cd = PDE(inode)->data;
 869
 870                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 871                 if (!rp)
 872                         return -ENOMEM;
 873                 rp->offset = 0;
 874                 rp->q.reader = 1;
 875                 atomic_inc(&cd->readers);
 876                 spin_lock(&queue_lock);
 877                 list_add(&rp->q.list, &cd->queue);
 878                 spin_unlock(&queue_lock);
 879         }
 880         filp->private_data = rp;
 881         return 0;
 882 }
 883
 884 static int
 885 cache_release(struct inode *inode, struct file *filp)
 886 {
 887         struct cache_reader *rp = filp->private_data;
 888         struct cache_detail *cd = PDE(inode)->data;
 889
 890         if (rp) {
 891                 spin_lock(&queue_lock);
 892                 if (rp->offset) {
 893                         struct cache_queue *cq;
 894                         for (cq= &rp->q; &cq->list != &cd->queue;
 895                              cq = list_entry(cq->list.next, struct cache_queue, list))
 896                                 if (!cq->reader) {
 897                                         container_of(cq, struct cache_request, q)
 898                                                 ->readers--;
 899                                         break;
 900                                 }
 901                         rp->offset = 0;
 902                 }
 903                 list_del(&rp->q.list);
 904                 spin_unlock(&queue_lock);
 905
 906                 filp->private_data = NULL;
 907                 kfree(rp);
 908
 909                 cd->last_close = get_seconds();
 910                 atomic_dec(&cd->readers);
 911         }
 912         return 0;
 913 }
 914
 915
 916
 917 static const struct file_operations cache_file_operations = {
 918         .owner          = THIS_MODULE,
 919         .llseek         = no_llseek,
 920         .read           = cache_read,
 921         .write          = cache_write,
 922         .poll           = cache_poll,
 923         .ioctl          = cache_ioctl, /* for FIONREAD */
 924         .open           = cache_open,
 925         .release        = cache_release,
 926 };
 927
 928
 929 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
 930 {
 931         struct cache_queue *cq;
 932         spin_lock(&queue_lock);
 933         list_for_each_entry(cq, &detail->queue, list)
 934                 if (!cq->reader) {
 935                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 936                         if (cr->item != ch)
 937                                 continue;
 938                         if (cr->readers != 0)
 939                                 continue;
 940                         list_del(&cr->q.list);
 941                         spin_unlock(&queue_lock);
 942                         cache_put(cr->item, detail);
 943                         kfree(cr->buf);
 944                         kfree(cr);
 945                         return;
 946                 }
 947         spin_unlock(&queue_lock);
 948 }
 949
 950 /*
 951  * Support routines for text-based upcalls.
 952  * Fields are separated by spaces.
 953  * Fields are either mangled to quote space tab newline slosh with slosh
 954  * or a hexified with a leading \x
 955  * Record is terminated with newline.
 956  *
 957  */
 958
 959 void qword_add(char **bpp, int *lp, char *str)
 960 {
 961         char *bp = *bpp;
 962         int len = *lp;
 963         char c;
 964
 965         if (len < 0) return;
 966
 967         while ((c=*str++) && len)
 968                 switch(c) {
 969                 case ' ':
 970                 case '\t':
 971                 case '\n':
 972                 case '\\':
 973                         if (len >= 4) {
 974                                 *bp++ = '\\';
 975                                 *bp++ = '0' + ((c & 0300)>>6);
 976                                 *bp++ = '0' + ((c & 0070)>>3);
 977                                 *bp++ = '0' + ((c & 0007)>>0);
 978                         }
 979                         len -= 4;
 980                         break;
 981                 default:
 982                         *bp++ = c;
 983                         len--;
 984                 }
 985         if (c || len <1) len = -1;
 986         else {
 987                 *bp++ = ' ';
 988                 len--;
 989         }
 990         *bpp = bp;
 991         *lp = len;
 992 }
 993
 994 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
 995 {
 996         char *bp = *bpp;
 997         int len = *lp;
 998
 999         if (len < 0) return;
1000
1001         if (len > 2) {
1002                 *bp++ = '\\';
1003                 *bp++ = 'x';
1004                 len -= 2;
1005                 while (blen && len >= 2) {
1006                         unsigned char c = *buf++;
1007                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1008                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1009                         len -= 2;
1010                         blen--;
1011                 }
1012         }
1013         if (blen || len<1) len = -1;
1014         else {
1015                 *bp++ = ' ';
1016                 len--;
1017         }
1018         *bpp = bp;
1019         *lp = len;
1020 }
1021
1022 static void warn_no_listener(struct cache_detail *detail)
1023 {
1024         if (detail->last_warn != detail->last_close) {
1025                 detail->last_warn = detail->last_close;
1026                 if (detail->warn_no_listener)
1027                         detail->warn_no_listener(detail);
1028         }
1029 }
1030
1031 /*
1032  * register an upcall request to user-space.
1033  * Each request is at most one page long.
1034  */
1035 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
1036 {
1037
1038         char *buf;
1039         struct cache_request *crq;
1040         char *bp;
1041         int len;
1042
1043         if (detail->cache_request == NULL)
1044                 return -EINVAL;
1045
1046         if (atomic_read(&detail->readers) == 0 &&
1047             detail->last_close < get_seconds() - 30) {
1048                         warn_no_listener(detail);
1049                         return -EINVAL;
1050         }
1051
1052         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1053         if (!buf)
1054                 return -EAGAIN;
1055
1056         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1057         if (!crq) {
1058                 kfree(buf);
1059                 return -EAGAIN;
1060         }
1061
1062         bp = buf; len = PAGE_SIZE;
1063
1064         detail->cache_request(detail, h, &bp, &len);
1065
1066         if (len < 0) {
1067                 kfree(buf);
1068                 kfree(crq);
1069                 return -EAGAIN;
1070         }
1071         crq->q.reader = 0;
1072         crq->item = cache_get(h);
1073         crq->buf = buf;
1074         crq->len = PAGE_SIZE - len;
1075         crq->readers = 0;
1076         spin_lock(&queue_lock);
1077         list_add_tail(&crq->q.list, &detail->queue);
1078         spin_unlock(&queue_lock);
1079         wake_up(&queue_wait);
1080         return 0;
1081 }
1082
1083 /*
1084  * parse a message from user-space and pass it
1085  * to an appropriate cache
1086  * Messages are, like requests, separated into fields by
1087  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1088  *
1089  * Message is
1090  *   reply cachename expiry key ... content....
1091  *
1092  * key and content are both parsed by cache
1093  */
1094
1095 #define isodigit(c) (isdigit(c) && c <= '7')
1096 int qword_get(char **bpp, char *dest, int bufsize)
1097 {
1098         /* return bytes copied, or -1 on error */
1099         char *bp = *bpp;
1100         int len = 0;
1101
1102         while (*bp == ' ') bp++;
1103
1104         if (bp[0] == '\\' && bp[1] == 'x') {
1105                 /* HEX STRING */
1106                 bp += 2;
1107                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1108                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1109                         bp++;
1110                         byte <<= 4;
1111                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1112                         *dest++ = byte;
1113                         bp++;
1114                         len++;
1115                 }
1116         } else {
1117                 /* text with \nnn octal quoting */
1118                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1119                         if (*bp == '\\' &&
1120                             isodigit(bp[1]) && (bp[1] <= '3') &&
1121                             isodigit(bp[2]) &&
1122                             isodigit(bp[3])) {
1123                                 int byte = (*++bp -'0');
1124                                 bp++;
1125                                 byte = (byte << 3) | (*bp++ - '0');
1126                                 byte = (byte << 3) | (*bp++ - '0');
1127                                 *dest++ = byte;
1128                                 len++;
1129                         } else {
1130                                 *dest++ = *bp++;
1131                                 len++;
1132                         }
1133                 }
1134         }
1135
1136         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1137                 return -1;
1138         while (*bp == ' ') bp++;
1139         *bpp = bp;
1140         *dest = '\0';
1141         return len;
1142 }
1143
1144
1145 /*
1146  * support /proc/sunrpc/cache/$CACHENAME/content
1147  * as a seqfile.
1148  * We call ->cache_show passing NULL for the item to
1149  * get a header, then pass each real item in the cache
1150  */
1151
1152 struct handle {
1153         struct cache_detail *cd;
1154 };
1155
1156 static void *c_start(struct seq_file *m, loff_t *pos)
1157         __acquires(cd->hash_lock)
1158 {
1159         loff_t n = *pos;
1160         unsigned hash, entry;
1161         struct cache_head *ch;
1162         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1163
1164
1165         read_lock(&cd->hash_lock);
1166         if (!n--)
1167                 return SEQ_START_TOKEN;
1168         hash = n >> 32;
1169         entry = n & ((1LL<<32) - 1);
1170
1171         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1172                 if (!entry--)
1173                         return ch;
1174         n &= ~((1LL<<32) - 1);
1175         do {
1176                 hash++;
1177                 n += 1LL<<32;
1178         } while(hash < cd->hash_size &&
1179                 cd->hash_table[hash]==NULL);
1180         if (hash >= cd->hash_size)
1181                 return NULL;
1182         *pos = n+1;
1183         return cd->hash_table[hash];
1184 }
1185
1186 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1187 {
1188         struct cache_head *ch = p;
1189         int hash = (*pos >> 32);
1190         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1191
1192         if (p == SEQ_START_TOKEN)
1193                 hash = 0;
1194         else if (ch->next == NULL) {
1195                 hash++;
1196                 *pos += 1LL<<32;
1197         } else {
1198                 ++*pos;
1199                 return ch->next;
1200         }
1201         *pos &= ~((1LL<<32) - 1);
1202         while (hash < cd->hash_size &&
1203                cd->hash_table[hash] == NULL) {
1204                 hash++;
1205                 *pos += 1LL<<32;
1206         }
1207         if (hash >= cd->hash_size)
1208                 return NULL;
1209         ++*pos;
1210         return cd->hash_table[hash];
1211 }
1212
1213 static void c_stop(struct seq_file *m, void *p)
1214         __releases(cd->hash_lock)
1215 {
1216         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1217         read_unlock(&cd->hash_lock);
1218 }
1219
1220 static int c_show(struct seq_file *m, void *p)
1221 {
1222         struct cache_head *cp = p;
1223         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1224
1225         if (p == SEQ_START_TOKEN)
1226                 return cd->cache_show(m, cd, NULL);
1227
1228         ifdebug(CACHE)
1229                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1230                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1231         cache_get(cp);
1232         if (cache_check(cd, cp, NULL))
1233                 /* cache_check does a cache_put on failure */
1234                 seq_printf(m, "# ");
1235         else
1236                 cache_put(cp, cd);
1237
1238         return cd->cache_show(m, cd, cp);
1239 }
1240
1241 static const struct seq_operations cache_content_op = {
1242         .start  = c_start,
1243         .next   = c_next,
1244         .stop   = c_stop,
1245         .show   = c_show,
1246 };
1247
1248 static int content_open(struct inode *inode, struct file *file)
1249 {
1250         struct handle *han;
1251         struct cache_detail *cd = PDE(inode)->data;
1252
1253         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1254         if (han == NULL)
1255                 return -ENOMEM;
1256
1257         han->cd = cd;
1258         return 0;
1259 }
1260
1261 static const struct file_operations content_file_operations = {
1262         .open           = content_open,
1263         .read           = seq_read,
1264         .llseek         = seq_lseek,
1265         .release        = seq_release_private,
1266 };
1267
1268 static ssize_t read_flush(struct file *file, char __user *buf,
1269                             size_t count, loff_t *ppos)
1270 {
1271         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1272         char tbuf[20];
1273         unsigned long p = *ppos;
1274         size_t len;
1275
1276         sprintf(tbuf, "%lu\n", cd->flush_time);
1277         len = strlen(tbuf);
1278         if (p >= len)
1279                 return 0;
1280         len -= p;
1281         if (len > count)
1282                 len = count;
1283         if (copy_to_user(buf, (void*)(tbuf+p), len))
1284                 return -EFAULT;
1285         *ppos += len;
1286         return len;
1287 }
1288
1289 static ssize_t write_flush(struct file * file, const char __user * buf,
1290                              size_t count, loff_t *ppos)
1291 {
1292         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1293         char tbuf[20];
1294         char *ep;
1295         long flushtime;
1296         if (*ppos || count > sizeof(tbuf)-1)
1297                 return -EINVAL;
1298         if (copy_from_user(tbuf, buf, count))
1299                 return -EFAULT;
1300         tbuf[count] = 0;
1301         flushtime = simple_strtoul(tbuf, &ep, 0);
1302         if (*ep && *ep != '\n')
1303                 return -EINVAL;
1304
1305         cd->flush_time = flushtime;
1306         cd->nextcheck = get_seconds();
1307         cache_flush();
1308
1309         *ppos += count;
1310         return count;
1311 }
1312
1313 static const struct file_operations cache_flush_operations = {
1314         .open           = nonseekable_open,
1315         .read           = read_flush,
1316         .write          = write_flush,
1317 };