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 <linux/pagemap.h>
  31 #include <asm/ioctls.h>
  32 #include <linux/sunrpc/types.h>
  33 #include <linux/sunrpc/cache.h>
  34 #include <linux/sunrpc/stats.h>
  35
  36 #define  RPCDBG_FACILITY RPCDBG_CACHE
  37
  38 static int cache_defer_req(struct cache_req *req, struct cache_head *item);
  39 static void cache_revisit_request(struct cache_head *item);
  40
  41 static void cache_init(struct cache_head *h)
  42 {
  43         time_t now = get_seconds();
  44         h->next = NULL;
  45         h->flags = 0;
  46         kref_init(&h->ref);
  47         h->expiry_time = now + CACHE_NEW_EXPIRY;
  48         h->last_refresh = now;
  49 }
  50
  51 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  52                                        struct cache_head *key, int hash)
  53 {
  54         struct cache_head **head,  **hp;
  55         struct cache_head *new = NULL;
  56
  57         head = &detail->hash_table[hash];
  58
  59         read_lock(&detail->hash_lock);
  60
  61         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  62                 struct cache_head *tmp = *hp;
  63                 if (detail->match(tmp, key)) {
  64                         cache_get(tmp);
  65                         read_unlock(&detail->hash_lock);
  66                         return tmp;
  67                 }
  68         }
  69         read_unlock(&detail->hash_lock);
  70         /* Didn't find anything, insert an empty entry */
  71
  72         new = detail->alloc();
  73         if (!new)
  74                 return NULL;
  75         /* must fully initialise 'new', else
  76          * we might get lose if we need to
  77          * cache_put it soon.
  78          */
  79         cache_init(new);
  80         detail->init(new, key);
  81
  82         write_lock(&detail->hash_lock);
  83
  84         /* check if entry appeared while we slept */
  85         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  86                 struct cache_head *tmp = *hp;
  87                 if (detail->match(tmp, key)) {
  88                         cache_get(tmp);
  89                         write_unlock(&detail->hash_lock);
  90                         cache_put(new, detail);
  91                         return tmp;
  92                 }
  93         }
  94         new->next = *head;
  95         *head = new;
  96         detail->entries++;
  97         cache_get(new);
  98         write_unlock(&detail->hash_lock);
  99
 100         return new;
 101 }
 102 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
 103
 104
 105 static void queue_loose(struct cache_detail *detail, struct cache_head *ch);
 106
 107 static int cache_fresh_locked(struct cache_head *head, time_t expiry)
 108 {
 109         head->expiry_time = expiry;
 110         head->last_refresh = get_seconds();
 111         return !test_and_set_bit(CACHE_VALID, &head->flags);
 112 }
 113
 114 static void cache_fresh_unlocked(struct cache_head *head,
 115                         struct cache_detail *detail, int new)
 116 {
 117         if (new)
 118                 cache_revisit_request(head);
 119         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 120                 cache_revisit_request(head);
 121                 queue_loose(detail, head);
 122         }
 123 }
 124
 125 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 126                                        struct cache_head *new, struct cache_head *old, int hash)
 127 {
 128         /* The 'old' entry is to be replaced by 'new'.
 129          * If 'old' is not VALID, we update it directly,
 130          * otherwise we need to replace it
 131          */
 132         struct cache_head **head;
 133         struct cache_head *tmp;
 134         int is_new;
 135
 136         if (!test_bit(CACHE_VALID, &old->flags)) {
 137                 write_lock(&detail->hash_lock);
 138                 if (!test_bit(CACHE_VALID, &old->flags)) {
 139                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 140                                 set_bit(CACHE_NEGATIVE, &old->flags);
 141                         else
 142                                 detail->update(old, new);
 143                         is_new = cache_fresh_locked(old, new->expiry_time);
 144                         write_unlock(&detail->hash_lock);
 145                         cache_fresh_unlocked(old, detail, is_new);
 146                         return old;
 147                 }
 148                 write_unlock(&detail->hash_lock);
 149         }
 150         /* We need to insert a new entry */
 151         tmp = detail->alloc();
 152         if (!tmp) {
 153                 cache_put(old, detail);
 154                 return NULL;
 155         }
 156         cache_init(tmp);
 157         detail->init(tmp, old);
 158         head = &detail->hash_table[hash];
 159
 160         write_lock(&detail->hash_lock);
 161         if (test_bit(CACHE_NEGATIVE, &new->flags))
 162                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 163         else
 164                 detail->update(tmp, new);
 165         tmp->next = *head;
 166         *head = tmp;
 167         detail->entries++;
 168         cache_get(tmp);
 169         is_new = cache_fresh_locked(tmp, new->expiry_time);
 170         cache_fresh_locked(old, 0);
 171         write_unlock(&detail->hash_lock);
 172         cache_fresh_unlocked(tmp, detail, is_new);
 173         cache_fresh_unlocked(old, detail, 0);
 174         cache_put(old, detail);
 175         return tmp;
 176 }
 177 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 178
 179 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 180 {
 181         if (!cd->cache_upcall)
 182                 return -EINVAL;
 183         return cd->cache_upcall(cd, h);
 184 }
 185
 186 /*
 187  * This is the generic cache management routine for all
 188  * the authentication caches.
 189  * It checks the currency of a cache item and will (later)
 190  * initiate an upcall to fill it if needed.
 191  *
 192  *
 193  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 194  * -EAGAIN if upcall is pending,
 195  * -ETIMEDOUT if upcall failed and should be retried,
 196  * -ENOENT if cache entry was negative
 197  */
 198 int cache_check(struct cache_detail *detail,
 199                     struct cache_head *h, struct cache_req *rqstp)
 200 {
 201         int rv;
 202         long refresh_age, age;
 203
 204         /* First decide return status as best we can */
 205         if (!test_bit(CACHE_VALID, &h->flags) ||
 206             h->expiry_time < get_seconds())
 207                 rv = -EAGAIN;
 208         else if (detail->flush_time > h->last_refresh)
 209                 rv = -EAGAIN;
 210         else {
 211                 /* entry is valid */
 212                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 213                         rv = -ENOENT;
 214                 else rv = 0;
 215         }
 216
 217         /* now see if we want to start an upcall */
 218         refresh_age = (h->expiry_time - h->last_refresh);
 219         age = get_seconds() - h->last_refresh;
 220
 221         if (rqstp == NULL) {
 222                 if (rv == -EAGAIN)
 223                         rv = -ENOENT;
 224         } else if (rv == -EAGAIN || age > refresh_age/2) {
 225                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 226                                 refresh_age, age);
 227                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 228                         switch (cache_make_upcall(detail, h)) {
 229                         case -EINVAL:
 230                                 clear_bit(CACHE_PENDING, &h->flags);
 231                                 if (rv == -EAGAIN) {
 232                                         set_bit(CACHE_NEGATIVE, &h->flags);
 233                                         cache_fresh_unlocked(h, detail,
 234                                              cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY));
 235                                         rv = -ENOENT;
 236                                 }
 237                                 break;
 238
 239                         case -EAGAIN:
 240                                 clear_bit(CACHE_PENDING, &h->flags);
 241                                 cache_revisit_request(h);
 242                                 break;
 243                         }
 244                 }
 245         }
 246
 247         if (rv == -EAGAIN)
 248                 if (cache_defer_req(rqstp, h) != 0)
 249                         rv = -ETIMEDOUT;
 250
 251         if (rv)
 252                 cache_put(h, detail);
 253         return rv;
 254 }
 255 EXPORT_SYMBOL_GPL(cache_check);
 256
 257 /*
 258  * caches need to be periodically cleaned.
 259  * For this we maintain a list of cache_detail and
 260  * a current pointer into that list and into the table
 261  * for that entry.
 262  *
 263  * Each time clean_cache is called it finds the next non-empty entry
 264  * in the current table and walks the list in that entry
 265  * looking for entries that can be removed.
 266  *
 267  * An entry gets removed if:
 268  * - The expiry is before current time
 269  * - The last_refresh time is before the flush_time for that cache
 270  *
 271  * later we might drop old entries with non-NEVER expiry if that table
 272  * is getting 'full' for some definition of 'full'
 273  *
 274  * The question of "how often to scan a table" is an interesting one
 275  * and is answered in part by the use of the "nextcheck" field in the
 276  * cache_detail.
 277  * When a scan of a table begins, the nextcheck field is set to a time
 278  * that is well into the future.
 279  * While scanning, if an expiry time is found that is earlier than the
 280  * current nextcheck time, nextcheck is set to that expiry time.
 281  * If the flush_time is ever set to a time earlier than the nextcheck
 282  * time, the nextcheck time is then set to that flush_time.
 283  *
 284  * A table is then only scanned if the current time is at least
 285  * the nextcheck time.
 286  *
 287  */
 288
 289 static LIST_HEAD(cache_list);
 290 static DEFINE_SPINLOCK(cache_list_lock);
 291 static struct cache_detail *current_detail;
 292 static int current_index;
 293
 294 static const struct file_operations cache_file_operations;
 295 static const struct file_operations content_file_operations;
 296 static const struct file_operations cache_flush_operations;
 297
 298 static void do_cache_clean(struct work_struct *work);
 299 static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
 300
 301 static void remove_cache_proc_entries(struct cache_detail *cd)
 302 {
 303         if (cd->proc_ent == NULL)
 304                 return;
 305         if (cd->flush_ent)
 306                 remove_proc_entry("flush", cd->proc_ent);
 307         if (cd->channel_ent)
 308                 remove_proc_entry("channel", cd->proc_ent);
 309         if (cd->content_ent)
 310                 remove_proc_entry("content", cd->proc_ent);
 311         cd->proc_ent = NULL;
 312         remove_proc_entry(cd->name, proc_net_rpc);
 313 }
 314
 315 #ifdef CONFIG_PROC_FS
 316 static int create_cache_proc_entries(struct cache_detail *cd)
 317 {
 318         struct proc_dir_entry *p;
 319
 320         cd->proc_ent = proc_mkdir(cd->name, proc_net_rpc);
 321         if (cd->proc_ent == NULL)
 322                 goto out_nomem;
 323         cd->channel_ent = cd->content_ent = NULL;
 324
 325         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
 326                              cd->proc_ent, &cache_flush_operations, cd);
 327         cd->flush_ent = p;
 328         if (p == NULL)
 329                 goto out_nomem;
 330
 331         if (cd->cache_upcall || cd->cache_parse) {
 332                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
 333                                      cd->proc_ent, &cache_file_operations, cd);
 334                 cd->channel_ent = p;
 335                 if (p == NULL)
 336                         goto out_nomem;
 337         }
 338         if (cd->cache_show) {
 339                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
 340                                 cd->proc_ent, &content_file_operations, cd);
 341                 cd->content_ent = p;
 342                 if (p == NULL)
 343                         goto out_nomem;
 344         }
 345         return 0;
 346 out_nomem:
 347         remove_cache_proc_entries(cd);
 348         return -ENOMEM;
 349 }
 350 #else /* CONFIG_PROC_FS */
 351 static int create_cache_proc_entries(struct cache_detail *cd)
 352 {
 353         return 0;
 354 }
 355 #endif
 356
 357 static void sunrpc_init_cache_detail(struct cache_detail *cd)
 358 {
 359         rwlock_init(&cd->hash_lock);
 360         INIT_LIST_HEAD(&cd->queue);
 361         spin_lock(&cache_list_lock);
 362         cd->nextcheck = 0;
 363         cd->entries = 0;
 364         atomic_set(&cd->readers, 0);
 365         cd->last_close = 0;
 366         cd->last_warn = -1;
 367         list_add(&cd->others, &cache_list);
 368         spin_unlock(&cache_list_lock);
 369
 370         /* start the cleaning process */
 371         schedule_delayed_work(&cache_cleaner, 0);
 372 }
 373
 374 static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 375 {
 376         cache_purge(cd);
 377         spin_lock(&cache_list_lock);
 378         write_lock(&cd->hash_lock);
 379         if (cd->entries || atomic_read(&cd->inuse)) {
 380                 write_unlock(&cd->hash_lock);
 381                 spin_unlock(&cache_list_lock);
 382                 goto out;
 383         }
 384         if (current_detail == cd)
 385                 current_detail = NULL;
 386         list_del_init(&cd->others);
 387         write_unlock(&cd->hash_lock);
 388         spin_unlock(&cache_list_lock);
 389         if (list_empty(&cache_list)) {
 390                 /* module must be being unloaded so its safe to kill the worker */
 391                 cancel_delayed_work_sync(&cache_cleaner);
 392         }
 393         return;
 394 out:
 395         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 396 }
 397
 398 int cache_register(struct cache_detail *cd)
 399 {
 400         int ret;
 401
 402         sunrpc_init_cache_detail(cd);
 403         ret = create_cache_proc_entries(cd);
 404         if (ret)
 405                 sunrpc_destroy_cache_detail(cd);
 406         return ret;
 407 }
 408 EXPORT_SYMBOL_GPL(cache_register);
 409
 410 void cache_unregister(struct cache_detail *cd)
 411 {
 412         remove_cache_proc_entries(cd);
 413         sunrpc_destroy_cache_detail(cd);
 414 }
 415 EXPORT_SYMBOL_GPL(cache_unregister);
 416
 417 /* clean cache tries to find something to clean
 418  * and cleans it.
 419  * It returns 1 if it cleaned something,
 420  *            0 if it didn't find anything this time
 421  *           -1 if it fell off the end of the list.
 422  */
 423 static int cache_clean(void)
 424 {
 425         int rv = 0;
 426         struct list_head *next;
 427
 428         spin_lock(&cache_list_lock);
 429
 430         /* find a suitable table if we don't already have one */
 431         while (current_detail == NULL ||
 432             current_index >= current_detail->hash_size) {
 433                 if (current_detail)
 434                         next = current_detail->others.next;
 435                 else
 436                         next = cache_list.next;
 437                 if (next == &cache_list) {
 438                         current_detail = NULL;
 439                         spin_unlock(&cache_list_lock);
 440                         return -1;
 441                 }
 442                 current_detail = list_entry(next, struct cache_detail, others);
 443                 if (current_detail->nextcheck > get_seconds())
 444                         current_index = current_detail->hash_size;
 445                 else {
 446                         current_index = 0;
 447                         current_detail->nextcheck = get_seconds()+30*60;
 448                 }
 449         }
 450
 451         /* find a non-empty bucket in the table */
 452         while (current_detail &&
 453                current_index < current_detail->hash_size &&
 454                current_detail->hash_table[current_index] == NULL)
 455                 current_index++;
 456
 457         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 458
 459         if (current_detail && current_index < current_detail->hash_size) {
 460                 struct cache_head *ch, **cp;
 461                 struct cache_detail *d;
 462
 463                 write_lock(&current_detail->hash_lock);
 464
 465                 /* Ok, now to clean this strand */
 466
 467                 cp = & current_detail->hash_table[current_index];
 468                 ch = *cp;
 469                 for (; ch; cp= & ch->next, ch= *cp) {
 470                         if (current_detail->nextcheck > ch->expiry_time)
 471                                 current_detail->nextcheck = ch->expiry_time+1;
 472                         if (ch->expiry_time >= get_seconds()
 473                             && ch->last_refresh >= current_detail->flush_time
 474                                 )
 475                                 continue;
 476                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 477                                 queue_loose(current_detail, ch);
 478
 479                         if (atomic_read(&ch->ref.refcount) == 1)
 480                                 break;
 481                 }
 482                 if (ch) {
 483                         *cp = ch->next;
 484                         ch->next = NULL;
 485                         current_detail->entries--;
 486                         rv = 1;
 487                 }
 488                 write_unlock(&current_detail->hash_lock);
 489                 d = current_detail;
 490                 if (!ch)
 491                         current_index ++;
 492                 spin_unlock(&cache_list_lock);
 493                 if (ch)
 494                         cache_put(ch, d);
 495         } else
 496                 spin_unlock(&cache_list_lock);
 497
 498         return rv;
 499 }
 500
 501 /*
 502  * We want to regularly clean the cache, so we need to schedule some work ...
 503  */
 504 static void do_cache_clean(struct work_struct *work)
 505 {
 506         int delay = 5;
 507         if (cache_clean() == -1)
 508                 delay = round_jiffies_relative(30*HZ);
 509
 510         if (list_empty(&cache_list))
 511                 delay = 0;
 512
 513         if (delay)
 514                 schedule_delayed_work(&cache_cleaner, delay);
 515 }
 516
 517
 518 /*
 519  * Clean all caches promptly.  This just calls cache_clean
 520  * repeatedly until we are sure that every cache has had a chance to
 521  * be fully cleaned
 522  */
 523 void cache_flush(void)
 524 {
 525         while (cache_clean() != -1)
 526                 cond_resched();
 527         while (cache_clean() != -1)
 528                 cond_resched();
 529 }
 530 EXPORT_SYMBOL_GPL(cache_flush);
 531
 532 void cache_purge(struct cache_detail *detail)
 533 {
 534         detail->flush_time = LONG_MAX;
 535         detail->nextcheck = get_seconds();
 536         cache_flush();
 537         detail->flush_time = 1;
 538 }
 539 EXPORT_SYMBOL_GPL(cache_purge);
 540
 541
 542 /*
 543  * Deferral and Revisiting of Requests.
 544  *
 545  * If a cache lookup finds a pending entry, we
 546  * need to defer the request and revisit it later.
 547  * All deferred requests are stored in a hash table,
 548  * indexed by "struct cache_head *".
 549  * As it may be wasteful to store a whole request
 550  * structure, we allow the request to provide a
 551  * deferred form, which must contain a
 552  * 'struct cache_deferred_req'
 553  * This cache_deferred_req contains a method to allow
 554  * it to be revisited when cache info is available
 555  */
 556
 557 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 558 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 559
 560 #define DFR_MAX 300     /* ??? */
 561
 562 static DEFINE_SPINLOCK(cache_defer_lock);
 563 static LIST_HEAD(cache_defer_list);
 564 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 565 static int cache_defer_cnt;
 566
 567 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 568 {
 569         struct cache_deferred_req *dreq;
 570         int hash = DFR_HASH(item);
 571
 572         if (cache_defer_cnt >= DFR_MAX) {
 573                 /* too much in the cache, randomly drop this one,
 574                  * or continue and drop the oldest below
 575                  */
 576                 if (net_random()&1)
 577                         return -ETIMEDOUT;
 578         }
 579         dreq = req->defer(req);
 580         if (dreq == NULL)
 581                 return -ETIMEDOUT;
 582
 583         dreq->item = item;
 584
 585         spin_lock(&cache_defer_lock);
 586
 587         list_add(&dreq->recent, &cache_defer_list);
 588
 589         if (cache_defer_hash[hash].next == NULL)
 590                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 591         list_add(&dreq->hash, &cache_defer_hash[hash]);
 592
 593         /* it is in, now maybe clean up */
 594         dreq = NULL;
 595         if (++cache_defer_cnt > DFR_MAX) {
 596                 dreq = list_entry(cache_defer_list.prev,
 597                                   struct cache_deferred_req, recent);
 598                 list_del(&dreq->recent);
 599                 list_del(&dreq->hash);
 600                 cache_defer_cnt--;
 601         }
 602         spin_unlock(&cache_defer_lock);
 603
 604         if (dreq) {
 605                 /* there was one too many */
 606                 dreq->revisit(dreq, 1);
 607         }
 608         if (!test_bit(CACHE_PENDING, &item->flags)) {
 609                 /* must have just been validated... */
 610                 cache_revisit_request(item);
 611         }
 612         return 0;
 613 }
 614
 615 static void cache_revisit_request(struct cache_head *item)
 616 {
 617         struct cache_deferred_req *dreq;
 618         struct list_head pending;
 619
 620         struct list_head *lp;
 621         int hash = DFR_HASH(item);
 622
 623         INIT_LIST_HEAD(&pending);
 624         spin_lock(&cache_defer_lock);
 625
 626         lp = cache_defer_hash[hash].next;
 627         if (lp) {
 628                 while (lp != &cache_defer_hash[hash]) {
 629                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 630                         lp = lp->next;
 631                         if (dreq->item == item) {
 632                                 list_del(&dreq->hash);
 633                                 list_move(&dreq->recent, &pending);
 634                                 cache_defer_cnt--;
 635                         }
 636                 }
 637         }
 638         spin_unlock(&cache_defer_lock);
 639
 640         while (!list_empty(&pending)) {
 641                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 642                 list_del_init(&dreq->recent);
 643                 dreq->revisit(dreq, 0);
 644         }
 645 }
 646
 647 void cache_clean_deferred(void *owner)
 648 {
 649         struct cache_deferred_req *dreq, *tmp;
 650         struct list_head pending;
 651
 652
 653         INIT_LIST_HEAD(&pending);
 654         spin_lock(&cache_defer_lock);
 655
 656         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 657                 if (dreq->owner == owner) {
 658                         list_del(&dreq->hash);
 659                         list_move(&dreq->recent, &pending);
 660                         cache_defer_cnt--;
 661                 }
 662         }
 663         spin_unlock(&cache_defer_lock);
 664
 665         while (!list_empty(&pending)) {
 666                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 667                 list_del_init(&dreq->recent);
 668                 dreq->revisit(dreq, 1);
 669         }
 670 }
 671
 672 /*
 673  * communicate with user-space
 674  *
 675  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 676  * On read, you get a full request, or block.
 677  * On write, an update request is processed.
 678  * Poll works if anything to read, and always allows write.
 679  *
 680  * Implemented by linked list of requests.  Each open file has
 681  * a ->private that also exists in this list.  New requests are added
 682  * to the end and may wakeup and preceding readers.
 683  * New readers are added to the head.  If, on read, an item is found with
 684  * CACHE_UPCALLING clear, we free it from the list.
 685  *
 686  */
 687
 688 static DEFINE_SPINLOCK(queue_lock);
 689 static DEFINE_MUTEX(queue_io_mutex);
 690
 691 struct cache_queue {
 692         struct list_head        list;
 693         int                     reader; /* if 0, then request */
 694 };
 695 struct cache_request {
 696         struct cache_queue      q;
 697         struct cache_head       *item;
 698         char                    * buf;
 699         int                     len;
 700         int                     readers;
 701 };
 702 struct cache_reader {
 703         struct cache_queue      q;
 704         int                     offset; /* if non-0, we have a refcnt on next request */
 705 };
 706
 707 static ssize_t
 708 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
 709 {
 710         struct cache_reader *rp = filp->private_data;
 711         struct cache_request *rq;
 712         struct inode *inode = filp->f_path.dentry->d_inode;
 713         struct cache_detail *cd = PDE(inode)->data;
 714         int err;
 715
 716         if (count == 0)
 717                 return 0;
 718
 719         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 720                               * readers on this file */
 721  again:
 722         spin_lock(&queue_lock);
 723         /* need to find next request */
 724         while (rp->q.list.next != &cd->queue &&
 725                list_entry(rp->q.list.next, struct cache_queue, list)
 726                ->reader) {
 727                 struct list_head *next = rp->q.list.next;
 728                 list_move(&rp->q.list, next);
 729         }
 730         if (rp->q.list.next == &cd->queue) {
 731                 spin_unlock(&queue_lock);
 732                 mutex_unlock(&inode->i_mutex);
 733                 BUG_ON(rp->offset);
 734                 return 0;
 735         }
 736         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 737         BUG_ON(rq->q.reader);
 738         if (rp->offset == 0)
 739                 rq->readers++;
 740         spin_unlock(&queue_lock);
 741
 742         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 743                 err = -EAGAIN;
 744                 spin_lock(&queue_lock);
 745                 list_move(&rp->q.list, &rq->q.list);
 746                 spin_unlock(&queue_lock);
 747         } else {
 748                 if (rp->offset + count > rq->len)
 749                         count = rq->len - rp->offset;
 750                 err = -EFAULT;
 751                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 752                         goto out;
 753                 rp->offset += count;
 754                 if (rp->offset >= rq->len) {
 755                         rp->offset = 0;
 756                         spin_lock(&queue_lock);
 757                         list_move(&rp->q.list, &rq->q.list);
 758                         spin_unlock(&queue_lock);
 759                 }
 760                 err = 0;
 761         }
 762  out:
 763         if (rp->offset == 0) {
 764                 /* need to release rq */
 765                 spin_lock(&queue_lock);
 766                 rq->readers--;
 767                 if (rq->readers == 0 &&
 768                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 769                         list_del(&rq->q.list);
 770                         spin_unlock(&queue_lock);
 771                         cache_put(rq->item, cd);
 772                         kfree(rq->buf);
 773                         kfree(rq);
 774                 } else
 775                         spin_unlock(&queue_lock);
 776         }
 777         if (err == -EAGAIN)
 778                 goto again;
 779         mutex_unlock(&inode->i_mutex);
 780         return err ? err :  count;
 781 }
 782
 783 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 784                                  size_t count, struct cache_detail *cd)
 785 {
 786         ssize_t ret;
 787
 788         if (copy_from_user(kaddr, buf, count))
 789                 return -EFAULT;
 790         kaddr[count] = '\0';
 791         ret = cd->cache_parse(cd, kaddr, count);
 792         if (!ret)
 793                 ret = count;
 794         return ret;
 795 }
 796
 797 static ssize_t cache_slow_downcall(const char __user *buf,
 798                                    size_t count, struct cache_detail *cd)
 799 {
 800         static char write_buf[8192]; /* protected by queue_io_mutex */
 801         ssize_t ret = -EINVAL;
 802
 803         if (count >= sizeof(write_buf))
 804                 goto out;
 805         mutex_lock(&queue_io_mutex);
 806         ret = cache_do_downcall(write_buf, buf, count, cd);
 807         mutex_unlock(&queue_io_mutex);
 808 out:
 809         return ret;
 810 }
 811
 812 static ssize_t cache_downcall(struct address_space *mapping,
 813                               const char __user *buf,
 814                               size_t count, struct cache_detail *cd)
 815 {
 816         struct page *page;
 817         char *kaddr;
 818         ssize_t ret = -ENOMEM;
 819
 820         if (count >= PAGE_CACHE_SIZE)
 821                 goto out_slow;
 822
 823         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 824         if (!page)
 825                 goto out_slow;
 826
 827         kaddr = kmap(page);
 828         ret = cache_do_downcall(kaddr, buf, count, cd);
 829         kunmap(page);
 830         unlock_page(page);
 831         page_cache_release(page);
 832         return ret;
 833 out_slow:
 834         return cache_slow_downcall(buf, count, cd);
 835 }
 836
 837 static ssize_t
 838 cache_write(struct file *filp, const char __user *buf, size_t count,
 839             loff_t *ppos)
 840 {
 841         struct address_space *mapping = filp->f_mapping;
 842         struct inode *inode = filp->f_path.dentry->d_inode;
 843         struct cache_detail *cd = PDE(inode)->data;
 844         ssize_t ret = -EINVAL;
 845
 846         if (!cd->cache_parse)
 847                 goto out;
 848
 849         mutex_lock(&inode->i_mutex);
 850         ret = cache_downcall(mapping, buf, count, cd);
 851         mutex_unlock(&inode->i_mutex);
 852 out:
 853         return ret;
 854 }
 855
 856 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 857
 858 static unsigned int
 859 cache_poll(struct file *filp, poll_table *wait)
 860 {
 861         unsigned int mask;
 862         struct cache_reader *rp = filp->private_data;
 863         struct cache_queue *cq;
 864         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
 865
 866         poll_wait(filp, &queue_wait, wait);
 867
 868         /* alway allow write */
 869         mask = POLL_OUT | POLLWRNORM;
 870
 871         if (!rp)
 872                 return mask;
 873
 874         spin_lock(&queue_lock);
 875
 876         for (cq= &rp->q; &cq->list != &cd->queue;
 877              cq = list_entry(cq->list.next, struct cache_queue, list))
 878                 if (!cq->reader) {
 879                         mask |= POLLIN | POLLRDNORM;
 880                         break;
 881                 }
 882         spin_unlock(&queue_lock);
 883         return mask;
 884 }
 885
 886 static int
 887 cache_ioctl(struct inode *ino, struct file *filp,
 888             unsigned int cmd, unsigned long arg)
 889 {
 890         int len = 0;
 891         struct cache_reader *rp = filp->private_data;
 892         struct cache_queue *cq;
 893         struct cache_detail *cd = PDE(ino)->data;
 894
 895         if (cmd != FIONREAD || !rp)
 896                 return -EINVAL;
 897
 898         spin_lock(&queue_lock);
 899
 900         /* only find the length remaining in current request,
 901          * or the length of the next request
 902          */
 903         for (cq= &rp->q; &cq->list != &cd->queue;
 904              cq = list_entry(cq->list.next, struct cache_queue, list))
 905                 if (!cq->reader) {
 906                         struct cache_request *cr =
 907                                 container_of(cq, struct cache_request, q);
 908                         len = cr->len - rp->offset;
 909                         break;
 910                 }
 911         spin_unlock(&queue_lock);
 912
 913         return put_user(len, (int __user *)arg);
 914 }
 915
 916 static int
 917 cache_open(struct inode *inode, struct file *filp)
 918 {
 919         struct cache_reader *rp = NULL;
 920
 921         nonseekable_open(inode, filp);
 922         if (filp->f_mode & FMODE_READ) {
 923                 struct cache_detail *cd = PDE(inode)->data;
 924
 925                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 926                 if (!rp)
 927                         return -ENOMEM;
 928                 rp->offset = 0;
 929                 rp->q.reader = 1;
 930                 atomic_inc(&cd->readers);
 931                 spin_lock(&queue_lock);
 932                 list_add(&rp->q.list, &cd->queue);
 933                 spin_unlock(&queue_lock);
 934         }
 935         filp->private_data = rp;
 936         return 0;
 937 }
 938
 939 static int
 940 cache_release(struct inode *inode, struct file *filp)
 941 {
 942         struct cache_reader *rp = filp->private_data;
 943         struct cache_detail *cd = PDE(inode)->data;
 944
 945         if (rp) {
 946                 spin_lock(&queue_lock);
 947                 if (rp->offset) {
 948                         struct cache_queue *cq;
 949                         for (cq= &rp->q; &cq->list != &cd->queue;
 950                              cq = list_entry(cq->list.next, struct cache_queue, list))
 951                                 if (!cq->reader) {
 952                                         container_of(cq, struct cache_request, q)
 953                                                 ->readers--;
 954                                         break;
 955                                 }
 956                         rp->offset = 0;
 957                 }
 958                 list_del(&rp->q.list);
 959                 spin_unlock(&queue_lock);
 960
 961                 filp->private_data = NULL;
 962                 kfree(rp);
 963
 964                 cd->last_close = get_seconds();
 965                 atomic_dec(&cd->readers);
 966         }
 967         return 0;
 968 }
 969
 970
 971
 972 static const struct file_operations cache_file_operations = {
 973         .owner          = THIS_MODULE,
 974         .llseek         = no_llseek,
 975         .read           = cache_read,
 976         .write          = cache_write,
 977         .poll           = cache_poll,
 978         .ioctl          = cache_ioctl, /* for FIONREAD */
 979         .open           = cache_open,
 980         .release        = cache_release,
 981 };
 982
 983
 984 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
 985 {
 986         struct cache_queue *cq;
 987         spin_lock(&queue_lock);
 988         list_for_each_entry(cq, &detail->queue, list)
 989                 if (!cq->reader) {
 990                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 991                         if (cr->item != ch)
 992                                 continue;
 993                         if (cr->readers != 0)
 994                                 continue;
 995                         list_del(&cr->q.list);
 996                         spin_unlock(&queue_lock);
 997                         cache_put(cr->item, detail);
 998                         kfree(cr->buf);
 999                         kfree(cr);
1000                         return;
1001                 }
1002         spin_unlock(&queue_lock);
1003 }
1004
1005 /*
1006  * Support routines for text-based upcalls.
1007  * Fields are separated by spaces.
1008  * Fields are either mangled to quote space tab newline slosh with slosh
1009  * or a hexified with a leading \x
1010  * Record is terminated with newline.
1011  *
1012  */
1013
1014 void qword_add(char **bpp, int *lp, char *str)
1015 {
1016         char *bp = *bpp;
1017         int len = *lp;
1018         char c;
1019
1020         if (len < 0) return;
1021
1022         while ((c=*str++) && len)
1023                 switch(c) {
1024                 case ' ':
1025                 case '\t':
1026                 case '\n':
1027                 case '\\':
1028                         if (len >= 4) {
1029                                 *bp++ = '\\';
1030                                 *bp++ = '0' + ((c & 0300)>>6);
1031                                 *bp++ = '0' + ((c & 0070)>>3);
1032                                 *bp++ = '0' + ((c & 0007)>>0);
1033                         }
1034                         len -= 4;
1035                         break;
1036                 default:
1037                         *bp++ = c;
1038                         len--;
1039                 }
1040         if (c || len <1) len = -1;
1041         else {
1042                 *bp++ = ' ';
1043                 len--;
1044         }
1045         *bpp = bp;
1046         *lp = len;
1047 }
1048 EXPORT_SYMBOL_GPL(qword_add);
1049
1050 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
1051 {
1052         char *bp = *bpp;
1053         int len = *lp;
1054
1055         if (len < 0) return;
1056
1057         if (len > 2) {
1058                 *bp++ = '\\';
1059                 *bp++ = 'x';
1060                 len -= 2;
1061                 while (blen && len >= 2) {
1062                         unsigned char c = *buf++;
1063                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
1064                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
1065                         len -= 2;
1066                         blen--;
1067                 }
1068         }
1069         if (blen || len<1) len = -1;
1070         else {
1071                 *bp++ = ' ';
1072                 len--;
1073         }
1074         *bpp = bp;
1075         *lp = len;
1076 }
1077 EXPORT_SYMBOL_GPL(qword_addhex);
1078
1079 static void warn_no_listener(struct cache_detail *detail)
1080 {
1081         if (detail->last_warn != detail->last_close) {
1082                 detail->last_warn = detail->last_close;
1083                 if (detail->warn_no_listener)
1084                         detail->warn_no_listener(detail, detail->last_close != 0);
1085         }
1086 }
1087
1088 /*
1089  * register an upcall request to user-space and queue it up for read() by the
1090  * upcall daemon.
1091  *
1092  * Each request is at most one page long.
1093  */
1094 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1095                 void (*cache_request)(struct cache_detail *,
1096                                       struct cache_head *,
1097                                       char **,
1098                                       int *))
1099 {
1100
1101         char *buf;
1102         struct cache_request *crq;
1103         char *bp;
1104         int len;
1105
1106         if (atomic_read(&detail->readers) == 0 &&
1107             detail->last_close < get_seconds() - 30) {
1108                         warn_no_listener(detail);
1109                         return -EINVAL;
1110         }
1111
1112         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1113         if (!buf)
1114                 return -EAGAIN;
1115
1116         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1117         if (!crq) {
1118                 kfree(buf);
1119                 return -EAGAIN;
1120         }
1121
1122         bp = buf; len = PAGE_SIZE;
1123
1124         cache_request(detail, h, &bp, &len);
1125
1126         if (len < 0) {
1127                 kfree(buf);
1128                 kfree(crq);
1129                 return -EAGAIN;
1130         }
1131         crq->q.reader = 0;
1132         crq->item = cache_get(h);
1133         crq->buf = buf;
1134         crq->len = PAGE_SIZE - len;
1135         crq->readers = 0;
1136         spin_lock(&queue_lock);
1137         list_add_tail(&crq->q.list, &detail->queue);
1138         spin_unlock(&queue_lock);
1139         wake_up(&queue_wait);
1140         return 0;
1141 }
1142 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1143
1144 /*
1145  * parse a message from user-space and pass it
1146  * to an appropriate cache
1147  * Messages are, like requests, separated into fields by
1148  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1149  *
1150  * Message is
1151  *   reply cachename expiry key ... content....
1152  *
1153  * key and content are both parsed by cache
1154  */
1155
1156 #define isodigit(c) (isdigit(c) && c <= '7')
1157 int qword_get(char **bpp, char *dest, int bufsize)
1158 {
1159         /* return bytes copied, or -1 on error */
1160         char *bp = *bpp;
1161         int len = 0;
1162
1163         while (*bp == ' ') bp++;
1164
1165         if (bp[0] == '\\' && bp[1] == 'x') {
1166                 /* HEX STRING */
1167                 bp += 2;
1168                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1169                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1170                         bp++;
1171                         byte <<= 4;
1172                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1173                         *dest++ = byte;
1174                         bp++;
1175                         len++;
1176                 }
1177         } else {
1178                 /* text with \nnn octal quoting */
1179                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1180                         if (*bp == '\\' &&
1181                             isodigit(bp[1]) && (bp[1] <= '3') &&
1182                             isodigit(bp[2]) &&
1183                             isodigit(bp[3])) {
1184                                 int byte = (*++bp -'0');
1185                                 bp++;
1186                                 byte = (byte << 3) | (*bp++ - '0');
1187                                 byte = (byte << 3) | (*bp++ - '0');
1188                                 *dest++ = byte;
1189                                 len++;
1190                         } else {
1191                                 *dest++ = *bp++;
1192                                 len++;
1193                         }
1194                 }
1195         }
1196
1197         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1198                 return -1;
1199         while (*bp == ' ') bp++;
1200         *bpp = bp;
1201         *dest = '\0';
1202         return len;
1203 }
1204 EXPORT_SYMBOL_GPL(qword_get);
1205
1206
1207 /*
1208  * support /proc/sunrpc/cache/$CACHENAME/content
1209  * as a seqfile.
1210  * We call ->cache_show passing NULL for the item to
1211  * get a header, then pass each real item in the cache
1212  */
1213
1214 struct handle {
1215         struct cache_detail *cd;
1216 };
1217
1218 static void *c_start(struct seq_file *m, loff_t *pos)
1219         __acquires(cd->hash_lock)
1220 {
1221         loff_t n = *pos;
1222         unsigned hash, entry;
1223         struct cache_head *ch;
1224         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1225
1226
1227         read_lock(&cd->hash_lock);
1228         if (!n--)
1229                 return SEQ_START_TOKEN;
1230         hash = n >> 32;
1231         entry = n & ((1LL<<32) - 1);
1232
1233         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1234                 if (!entry--)
1235                         return ch;
1236         n &= ~((1LL<<32) - 1);
1237         do {
1238                 hash++;
1239                 n += 1LL<<32;
1240         } while(hash < cd->hash_size &&
1241                 cd->hash_table[hash]==NULL);
1242         if (hash >= cd->hash_size)
1243                 return NULL;
1244         *pos = n+1;
1245         return cd->hash_table[hash];
1246 }
1247
1248 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1249 {
1250         struct cache_head *ch = p;
1251         int hash = (*pos >> 32);
1252         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1253
1254         if (p == SEQ_START_TOKEN)
1255                 hash = 0;
1256         else if (ch->next == NULL) {
1257                 hash++;
1258                 *pos += 1LL<<32;
1259         } else {
1260                 ++*pos;
1261                 return ch->next;
1262         }
1263         *pos &= ~((1LL<<32) - 1);
1264         while (hash < cd->hash_size &&
1265                cd->hash_table[hash] == NULL) {
1266                 hash++;
1267                 *pos += 1LL<<32;
1268         }
1269         if (hash >= cd->hash_size)
1270                 return NULL;
1271         ++*pos;
1272         return cd->hash_table[hash];
1273 }
1274
1275 static void c_stop(struct seq_file *m, void *p)
1276         __releases(cd->hash_lock)
1277 {
1278         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1279         read_unlock(&cd->hash_lock);
1280 }
1281
1282 static int c_show(struct seq_file *m, void *p)
1283 {
1284         struct cache_head *cp = p;
1285         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1286
1287         if (p == SEQ_START_TOKEN)
1288                 return cd->cache_show(m, cd, NULL);
1289
1290         ifdebug(CACHE)
1291                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1292                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1293         cache_get(cp);
1294         if (cache_check(cd, cp, NULL))
1295                 /* cache_check does a cache_put on failure */
1296                 seq_printf(m, "# ");
1297         else
1298                 cache_put(cp, cd);
1299
1300         return cd->cache_show(m, cd, cp);
1301 }
1302
1303 static const struct seq_operations cache_content_op = {
1304         .start  = c_start,
1305         .next   = c_next,
1306         .stop   = c_stop,
1307         .show   = c_show,
1308 };
1309
1310 static int content_open(struct inode *inode, struct file *file)
1311 {
1312         struct handle *han;
1313         struct cache_detail *cd = PDE(inode)->data;
1314
1315         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1316         if (han == NULL)
1317                 return -ENOMEM;
1318
1319         han->cd = cd;
1320         return 0;
1321 }
1322
1323 static const struct file_operations content_file_operations = {
1324         .open           = content_open,
1325         .read           = seq_read,
1326         .llseek         = seq_lseek,
1327         .release        = seq_release_private,
1328 };
1329
1330 static ssize_t read_flush(struct file *file, char __user *buf,
1331                             size_t count, loff_t *ppos)
1332 {
1333         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1334         char tbuf[20];
1335         unsigned long p = *ppos;
1336         size_t len;
1337
1338         sprintf(tbuf, "%lu\n", cd->flush_time);
1339         len = strlen(tbuf);
1340         if (p >= len)
1341                 return 0;
1342         len -= p;
1343         if (len > count)
1344                 len = count;
1345         if (copy_to_user(buf, (void*)(tbuf+p), len))
1346                 return -EFAULT;
1347         *ppos += len;
1348         return len;
1349 }
1350
1351 static ssize_t write_flush(struct file * file, const char __user * buf,
1352                              size_t count, loff_t *ppos)
1353 {
1354         struct cache_detail *cd = PDE(file->f_path.dentry->d_inode)->data;
1355         char tbuf[20];
1356         char *ep;
1357         long flushtime;
1358         if (*ppos || count > sizeof(tbuf)-1)
1359                 return -EINVAL;
1360         if (copy_from_user(tbuf, buf, count))
1361                 return -EFAULT;
1362         tbuf[count] = 0;
1363         flushtime = simple_strtoul(tbuf, &ep, 0);
1364         if (*ep && *ep != '\n')
1365                 return -EINVAL;
1366
1367         cd->flush_time = flushtime;
1368         cd->nextcheck = get_seconds();
1369         cache_flush();
1370
1371         *ppos += count;
1372         return count;
1373 }
1374
1375 static const struct file_operations cache_flush_operations = {
1376         .open           = nonseekable_open,
1377         .read           = read_flush,
1378         .write          = write_flush,
1379 };