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