net/sunrpc/cache.c

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