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, 0);
  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 = get_seconds()+1;
 404         detail->nextcheck = get_seconds();
 405         cache_flush();
 406 }
 407
 408
 409
 410 /*
 411  * Deferral and Revisiting of Requests.
 412  *
 413  * If a cache lookup finds a pending entry, we
 414  * need to defer the request and revisit it later.
 415  * All deferred requests are stored in a hash table,
 416  * indexed by "struct cache_head *".
 417  * As it may be wasteful to store a whole request
 418  * structure, we allow the request to provide a
 419  * deferred form, which must contain a
 420  * 'struct cache_deferred_req'
 421  * This cache_deferred_req contains a method to allow
 422  * it to be revisited when cache info is available
 423  */
 424
 425 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 426 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 427
 428 #define DFR_MAX 300     /* ??? */
 429
 430 spinlock_t cache_defer_lock = SPIN_LOCK_UNLOCKED;
 431 static LIST_HEAD(cache_defer_list);
 432 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 433 static int cache_defer_cnt;
 434
 435 void cache_defer_req(struct cache_req *req, struct cache_head *item)
 436 {
 437         struct cache_deferred_req *dreq;
 438         int hash = DFR_HASH(item);
 439
 440         dreq = req->defer(req);
 441         if (dreq == NULL)
 442                 return;
 443
 444         dreq->item = item;
 445         dreq->recv_time = get_seconds();
 446
 447         spin_lock(&cache_defer_lock);
 448
 449         list_add(&dreq->recent, &cache_defer_list);
 450
 451         if (cache_defer_hash[hash].next == NULL)
 452                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 453         list_add(&dreq->hash, &cache_defer_hash[hash]);
 454
 455         /* it is in, now maybe clean up */
 456         dreq = NULL;
 457         if (++cache_defer_cnt > DFR_MAX) {
 458                 /* too much in the cache, randomly drop
 459                  * first or last
 460                  */
 461                 if (net_random()&1)
 462                         dreq = list_entry(cache_defer_list.next,
 463                                           struct cache_deferred_req,
 464                                           recent);
 465                 else
 466                         dreq = list_entry(cache_defer_list.prev,
 467                                           struct cache_deferred_req,
 468                                           recent);
 469                 list_del(&dreq->recent);
 470                 list_del(&dreq->hash);
 471                 cache_defer_cnt--;
 472         }
 473         spin_unlock(&cache_defer_lock);
 474
 475         if (dreq) {
 476                 /* there was one too many */
 477                 dreq->revisit(dreq, 1);
 478         }
 479         if (test_bit(CACHE_VALID, &item->flags)) {
 480                 /* must have just been validated... */
 481                 cache_revisit_request(item);
 482         }
 483 }
 484
 485 void cache_revisit_request(struct cache_head *item)
 486 {
 487         struct cache_deferred_req *dreq;
 488         struct list_head pending;
 489
 490         struct list_head *lp;
 491         int hash = DFR_HASH(item);
 492
 493         INIT_LIST_HEAD(&pending);
 494         spin_lock(&cache_defer_lock);
 495
 496         lp = cache_defer_hash[hash].next;
 497         if (lp) {
 498                 while (lp != &cache_defer_hash[hash]) {
 499                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 500                         lp = lp->next;
 501                         if (dreq->item == item) {
 502                                 list_del(&dreq->hash);
 503                                 list_move(&dreq->recent, &pending);
 504                                 cache_defer_cnt--;
 505                         }
 506                 }
 507         }
 508         spin_unlock(&cache_defer_lock);
 509
 510         while (!list_empty(&pending)) {
 511                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 512                 list_del_init(&dreq->recent);
 513                 dreq->revisit(dreq, 0);
 514         }
 515 }
 516
 517 void cache_clean_deferred(void *owner)
 518 {
 519         struct cache_deferred_req *dreq, *tmp;
 520         struct list_head pending;
 521
 522
 523         INIT_LIST_HEAD(&pending);
 524         spin_lock(&cache_defer_lock);
 525
 526         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 527                 if (dreq->owner == owner) {
 528                         list_del(&dreq->hash);
 529                         list_move(&dreq->recent, &pending);
 530                         cache_defer_cnt--;
 531                 }
 532         }
 533         spin_unlock(&cache_defer_lock);
 534
 535         while (!list_empty(&pending)) {
 536                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 537                 list_del_init(&dreq->recent);
 538                 dreq->revisit(dreq, 1);
 539         }
 540 }
 541
 542 /*
 543  * communicate with user-space
 544  *
 545  * We have a magic /proc file - /proc/sunrpc/cache
 546  * On read, you get a full request, or block
 547  * On write, an update request is processed
 548  * Poll works if anything to read, and always allows write
 549  *
 550  * Implemented by linked list of requests.  Each open file has
 551  * a ->private that also exists in this list.  New request are added
 552  * to the end and may wakeup and preceding readers.
 553  * New readers are added to the head.  If, on read, an item is found with
 554  * CACHE_UPCALLING clear, we free it from the list.
 555  *
 556  */
 557
 558 static spinlock_t queue_lock = SPIN_LOCK_UNLOCKED;
 559 static DECLARE_MUTEX(queue_io_sem);
 560
 561 struct cache_queue {
 562         struct list_head        list;
 563         int                     reader; /* if 0, then request */
 564 };
 565 struct cache_request {
 566         struct cache_queue      q;
 567         struct cache_head       *item;
 568         char                    * buf;
 569         int                     len;
 570         int                     readers;
 571 };
 572 struct cache_reader {
 573         struct cache_queue      q;
 574         int                     offset; /* if non-0, we have a refcnt on next request */
 575 };
 576
 577 static ssize_t
 578 cache_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
 579 {
 580         struct cache_reader *rp = filp->private_data;
 581         struct cache_request *rq;
 582         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
 583         int err;
 584
 585         if (ppos != &filp->f_pos)
 586                 return -ESPIPE;
 587
 588         if (count == 0)
 589                 return 0;
 590
 591         down(&queue_io_sem); /* protect against multiple concurrent
 592                               * readers on this file */
 593  again:
 594         spin_lock(&queue_lock);
 595         /* need to find next request */
 596         while (rp->q.list.next != &cd->queue &&
 597                list_entry(rp->q.list.next, struct cache_queue, list)
 598                ->reader) {
 599                 struct list_head *next = rp->q.list.next;
 600                 list_move(&rp->q.list, next);
 601         }
 602         if (rp->q.list.next == &cd->queue) {
 603                 spin_unlock(&queue_lock);
 604                 up(&queue_io_sem);
 605                 if (rp->offset)
 606                         BUG();
 607                 return 0;
 608         }
 609         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 610         if (rq->q.reader) BUG();
 611         if (rp->offset == 0)
 612                 rq->readers++;
 613         spin_unlock(&queue_lock);
 614
 615         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 616                 err = -EAGAIN;
 617                 spin_lock(&queue_lock);
 618                 list_move(&rp->q.list, &rq->q.list);
 619                 spin_unlock(&queue_lock);
 620         } else {
 621                 if (rp->offset + count > rq->len)
 622                         count = rq->len - rp->offset;
 623                 err = -EFAULT;
 624                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 625                         goto out;
 626                 rp->offset += count;
 627                 if (rp->offset >= rq->len) {
 628                         rp->offset = 0;
 629                         spin_lock(&queue_lock);
 630                         list_move(&rp->q.list, &rq->q.list);
 631                         spin_unlock(&queue_lock);
 632                 }
 633                 err = 0;
 634         }
 635  out:
 636         if (rp->offset == 0) {
 637                 /* need to release rq */
 638                 spin_lock(&queue_lock);
 639                 rq->readers--;
 640                 if (rq->readers == 0 &&
 641                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 642                         list_del(&rq->q.list);
 643                         spin_unlock(&queue_lock);
 644                         cd->cache_put(rq->item, cd);
 645                         kfree(rq->buf);
 646                         kfree(rq);
 647                 } else
 648                         spin_unlock(&queue_lock);
 649         }
 650         if (err == -EAGAIN)
 651                 goto again;
 652         up(&queue_io_sem);
 653         return err ? err :  count;
 654 }
 655
 656 static char write_buf[8192]; /* protected by queue_io_sem */
 657
 658 static ssize_t
 659 cache_write(struct file *filp, const char __user *buf, size_t count,
 660             loff_t *ppos)
 661 {
 662         int err;
 663         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
 664
 665         if (ppos != &filp->f_pos)
 666                 return -ESPIPE;
 667
 668         if (count == 0)
 669                 return 0;
 670         if (count >= sizeof(write_buf))
 671                 return -EINVAL;
 672
 673         down(&queue_io_sem);
 674
 675         if (copy_from_user(write_buf, buf, count)) {
 676                 up(&queue_io_sem);
 677                 return -EFAULT;
 678         }
 679         write_buf[count] = '\0';
 680         if (cd->cache_parse)
 681                 err = cd->cache_parse(cd, write_buf, count);
 682         else
 683                 err = -EINVAL;
 684
 685         up(&queue_io_sem);
 686         return err ? err : count;
 687 }
 688
 689 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 690
 691 static unsigned int
 692 cache_poll(struct file *filp, poll_table *wait)
 693 {
 694         unsigned int mask;
 695         struct cache_reader *rp = filp->private_data;
 696         struct cache_queue *cq;
 697         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
 698
 699         poll_wait(filp, &queue_wait, wait);
 700
 701         /* alway allow write */
 702         mask = POLL_OUT | POLLWRNORM;
 703
 704         if (!rp)
 705                 return mask;
 706
 707         spin_lock(&queue_lock);
 708
 709         for (cq= &rp->q; &cq->list != &cd->queue;
 710              cq = list_entry(cq->list.next, struct cache_queue, list))
 711                 if (!cq->reader) {
 712                         mask |= POLLIN | POLLRDNORM;
 713                         break;
 714                 }
 715         spin_unlock(&queue_lock);
 716         return mask;
 717 }
 718
 719 static int
 720 cache_ioctl(struct inode *ino, struct file *filp,
 721             unsigned int cmd, unsigned long arg)
 722 {
 723         int len = 0;
 724         struct cache_reader *rp = filp->private_data;
 725         struct cache_queue *cq;
 726         struct cache_detail *cd = PDE(ino)->data;
 727
 728         if (cmd != FIONREAD || !rp)
 729                 return -EINVAL;
 730
 731         spin_lock(&queue_lock);
 732
 733         /* only find the length remaining in current request,
 734          * or the length of the next request
 735          */
 736         for (cq= &rp->q; &cq->list != &cd->queue;
 737              cq = list_entry(cq->list.next, struct cache_queue, list))
 738                 if (!cq->reader) {
 739                         struct cache_request *cr =
 740                                 container_of(cq, struct cache_request, q);
 741                         len = cr->len - rp->offset;
 742                         break;
 743                 }
 744         spin_unlock(&queue_lock);
 745
 746         return put_user(len, (int __user *)arg);
 747 }
 748
 749 static int
 750 cache_open(struct inode *inode, struct file *filp)
 751 {
 752         struct cache_reader *rp = NULL;
 753
 754         if (filp->f_mode & FMODE_READ) {
 755                 struct cache_detail *cd = PDE(inode)->data;
 756
 757                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 758                 if (!rp)
 759                         return -ENOMEM;
 760                 rp->offset = 0;
 761                 rp->q.reader = 1;
 762                 atomic_inc(&cd->readers);
 763                 spin_lock(&queue_lock);
 764                 list_add(&rp->q.list, &cd->queue);
 765                 spin_unlock(&queue_lock);
 766         }
 767         filp->private_data = rp;
 768         return 0;
 769 }
 770
 771 static int
 772 cache_release(struct inode *inode, struct file *filp)
 773 {
 774         struct cache_reader *rp = filp->private_data;
 775         struct cache_detail *cd = PDE(inode)->data;
 776
 777         if (rp) {
 778                 spin_lock(&queue_lock);
 779                 if (rp->offset) {
 780                         struct cache_queue *cq;
 781                         for (cq= &rp->q; &cq->list != &cd->queue;
 782                              cq = list_entry(cq->list.next, struct cache_queue, list))
 783                                 if (!cq->reader) {
 784                                         container_of(cq, struct cache_request, q)
 785                                                 ->readers--;
 786                                         break;
 787                                 }
 788                         rp->offset = 0;
 789                 }
 790                 list_del(&rp->q.list);
 791                 spin_unlock(&queue_lock);
 792
 793                 filp->private_data = NULL;
 794                 kfree(rp);
 795
 796                 cd->last_close = get_seconds();
 797                 atomic_dec(&cd->readers);
 798         }
 799         return 0;
 800 }
 801
 802
 803
 804 static struct file_operations cache_file_operations = {
 805         .owner          = THIS_MODULE,
 806         .llseek         = no_llseek,
 807         .read           = cache_read,
 808         .write          = cache_write,
 809         .poll           = cache_poll,
 810         .ioctl          = cache_ioctl, /* for FIONREAD */
 811         .open           = cache_open,
 812         .release        = cache_release,
 813 };
 814
 815
 816 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
 817 {
 818         struct cache_queue *cq;
 819         spin_lock(&queue_lock);
 820         list_for_each_entry(cq, &detail->queue, list)
 821                 if (!cq->reader) {
 822                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 823                         if (cr->item != ch)
 824                                 continue;
 825                         if (cr->readers != 0)
 826                                 break;
 827                         list_del(&cr->q.list);
 828                         spin_unlock(&queue_lock);
 829                         detail->cache_put(cr->item, detail);
 830                         kfree(cr->buf);
 831                         kfree(cr);
 832                         return;
 833                 }
 834         spin_unlock(&queue_lock);
 835 }
 836
 837 /*
 838  * Support routines for text-based upcalls.
 839  * Fields are separated by spaces.
 840  * Fields are either mangled to quote space tab newline slosh with slosh
 841  * or a hexified with a leading \x
 842  * Record is terminated with newline.
 843  *
 844  */
 845
 846 void qword_add(char **bpp, int *lp, char *str)
 847 {
 848         char *bp = *bpp;
 849         int len = *lp;
 850         char c;
 851
 852         if (len < 0) return;
 853
 854         while ((c=*str++) && len)
 855                 switch(c) {
 856                 case ' ':
 857                 case '\t':
 858                 case '\n':
 859                 case '\\':
 860                         if (len >= 4) {
 861                                 *bp++ = '\\';
 862                                 *bp++ = '0' + ((c & 0300)>>6);
 863                                 *bp++ = '0' + ((c & 0070)>>3);
 864                                 *bp++ = '0' + ((c & 0007)>>0);
 865                         }
 866                         len -= 4;
 867                         break;
 868                 default:
 869                         *bp++ = c;
 870                         len--;
 871                 }
 872         if (c || len <1) len = -1;
 873         else {
 874                 *bp++ = ' ';
 875                 len--;
 876         }
 877         *bpp = bp;
 878         *lp = len;
 879 }
 880
 881 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
 882 {
 883         char *bp = *bpp;
 884         int len = *lp;
 885
 886         if (len < 0) return;
 887
 888         if (len > 2) {
 889                 *bp++ = '\\';
 890                 *bp++ = 'x';
 891                 len -= 2;
 892                 while (blen && len >= 2) {
 893                         unsigned char c = *buf++;
 894                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
 895                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
 896                         len -= 2;
 897                         blen--;
 898                 }
 899         }
 900         if (blen || len<1) len = -1;
 901         else {
 902                 *bp++ = ' ';
 903                 len--;
 904         }
 905         *bpp = bp;
 906         *lp = len;
 907 }
 908
 909 void warn_no_listener(struct cache_detail *detail)
 910 {
 911         if (detail->last_warn != detail->last_close) {
 912                 detail->last_warn = detail->last_close;
 913                 if (detail->warn_no_listener)
 914                         detail->warn_no_listener(detail);
 915         }
 916 }
 917
 918 /*
 919  * register an upcall request to user-space.
 920  * Each request is at most one page long.
 921  */
 922 static int cache_make_upcall(struct cache_detail *detail, struct cache_head *h)
 923 {
 924
 925         char *buf;
 926         struct cache_request *crq;
 927         char *bp;
 928         int len;
 929
 930         if (detail->cache_request == NULL)
 931                 return -EINVAL;
 932
 933         if (atomic_read(&detail->readers) == 0 &&
 934             detail->last_close < get_seconds() - 30) {
 935                         warn_no_listener(detail);
 936                         return -EINVAL;
 937         }
 938
 939         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 940         if (!buf)
 941                 return -EAGAIN;
 942
 943         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
 944         if (!crq) {
 945                 kfree(buf);
 946                 return -EAGAIN;
 947         }
 948
 949         bp = buf; len = PAGE_SIZE;
 950
 951         detail->cache_request(detail, h, &bp, &len);
 952
 953         if (len < 0) {
 954                 kfree(buf);
 955                 kfree(crq);
 956                 return -EAGAIN;
 957         }
 958         crq->q.reader = 0;
 959         crq->item = cache_get(h);
 960         crq->buf = buf;
 961         crq->len = PAGE_SIZE - len;
 962         crq->readers = 0;
 963         spin_lock(&queue_lock);
 964         list_add_tail(&crq->q.list, &detail->queue);
 965         spin_unlock(&queue_lock);
 966         wake_up(&queue_wait);
 967         return 0;
 968 }
 969
 970 /*
 971  * parse a message from user-space and pass it
 972  * to an appropriate cache
 973  * Messages are, like requests, separated into fields by
 974  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
 975  *
 976  * Message is
 977  *   reply cachename expiry key ... content....
 978  *
 979  * key and content are both parsed by cache
 980  */
 981
 982 #define isodigit(c) (isdigit(c) && c <= '7')
 983 int qword_get(char **bpp, char *dest, int bufsize)
 984 {
 985         /* return bytes copied, or -1 on error */
 986         char *bp = *bpp;
 987         int len = 0;
 988
 989         while (*bp == ' ') bp++;
 990
 991         if (bp[0] == '\\' && bp[1] == 'x') {
 992                 /* HEX STRING */
 993                 bp += 2;
 994                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
 995                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
 996                         bp++;
 997                         byte <<= 4;
 998                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
 999                         *dest++ = byte;
1000                         bp++;
1001                         len++;
1002                 }
1003         } else {
1004                 /* text with \nnn octal quoting */
1005                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1006                         if (*bp == '\\' &&
1007                             isodigit(bp[1]) && (bp[1] <= '3') &&
1008                             isodigit(bp[2]) &&
1009                             isodigit(bp[3])) {
1010                                 int byte = (*++bp -'0');
1011                                 bp++;
1012                                 byte = (byte << 3) | (*bp++ - '0');
1013                                 byte = (byte << 3) | (*bp++ - '0');
1014                                 *dest++ = byte;
1015                                 len++;
1016                         } else {
1017                                 *dest++ = *bp++;
1018                                 len++;
1019                         }
1020                 }
1021         }
1022
1023         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1024                 return -1;
1025         while (*bp == ' ') bp++;
1026         *bpp = bp;
1027         *dest = '\0';
1028         return len;
1029 }
1030
1031
1032 /*
1033  * support /proc/sunrpc/cache/$CACHENAME/content
1034  * as a seqfile.
1035  * We call ->cache_show passing NULL for the item to
1036  * get a header, then pass each real item in the cache
1037  */
1038
1039 struct handle {
1040         struct cache_detail *cd;
1041 };
1042
1043 static void *c_start(struct seq_file *m, loff_t *pos)
1044 {
1045         loff_t n = *pos;
1046         unsigned hash, entry;
1047         struct cache_head *ch;
1048         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1049
1050
1051         read_lock(&cd->hash_lock);
1052         if (!n--)
1053                 return SEQ_START_TOKEN;
1054         hash = n >> 32;
1055         entry = n & ((1LL<<32) - 1);
1056
1057         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1058                 if (!entry--)
1059                         return ch;
1060         n &= ~((1LL<<32) - 1);
1061         do {
1062                 hash++;
1063                 n += 1LL<<32;
1064         } while(hash < cd->hash_size &&
1065                 cd->hash_table[hash]==NULL);
1066         if (hash >= cd->hash_size)
1067                 return NULL;
1068         *pos = n+1;
1069         return cd->hash_table[hash];
1070 }
1071
1072 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1073 {
1074         struct cache_head *ch = p;
1075         int hash = (*pos >> 32);
1076         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1077
1078         if (p == SEQ_START_TOKEN)
1079                 hash = 0;
1080         else if (ch->next == NULL) {
1081                 hash++;
1082                 *pos += 1LL<<32;
1083         } else {
1084                 ++*pos;
1085                 return ch->next;
1086         }
1087         *pos &= ~((1LL<<32) - 1);
1088         while (hash < cd->hash_size &&
1089                cd->hash_table[hash] == NULL) {
1090                 hash++;
1091                 *pos += 1LL<<32;
1092         }
1093         if (hash >= cd->hash_size)
1094                 return NULL;
1095         ++*pos;
1096         return cd->hash_table[hash];
1097 }
1098
1099 static void c_stop(struct seq_file *m, void *p)
1100 {
1101         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1102         read_unlock(&cd->hash_lock);
1103 }
1104
1105 static int c_show(struct seq_file *m, void *p)
1106 {
1107         struct cache_head *cp = p;
1108         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1109
1110         if (p == SEQ_START_TOKEN)
1111                 return cd->cache_show(m, cd, NULL);
1112
1113         ifdebug(CACHE)
1114                 seq_printf(m, "# expiry=%ld refcnt=%d\n",
1115                            cp->expiry_time, atomic_read(&cp->refcnt));
1116         cache_get(cp);
1117         if (cache_check(cd, cp, NULL))
1118                 /* cache_check does a cache_put on failure */
1119                 seq_printf(m, "# ");
1120         else
1121                 cache_put(cp, cd);
1122
1123         return cd->cache_show(m, cd, cp);
1124 }
1125
1126 struct seq_operations cache_content_op = {
1127         .start  = c_start,
1128         .next   = c_next,
1129         .stop   = c_stop,
1130         .show   = c_show,
1131 };
1132
1133 static int content_open(struct inode *inode, struct file *file)
1134 {
1135         int res;
1136         struct handle *han;
1137         struct cache_detail *cd = PDE(inode)->data;
1138
1139         han = kmalloc(sizeof(*han), GFP_KERNEL);
1140         if (han == NULL)
1141                 return -ENOMEM;
1142
1143         han->cd = cd;
1144
1145         res = seq_open(file, &cache_content_op);
1146         if (res)
1147                 kfree(han);
1148         else
1149                 ((struct seq_file *)file->private_data)->private = han;
1150
1151         return res;
1152 }
1153 static int content_release(struct inode *inode, struct file *file)
1154 {
1155         struct seq_file *m = (struct seq_file *)file->private_data;
1156         struct handle *han = m->private;
1157         kfree(han);
1158         m->private = NULL;
1159         return seq_release(inode, file);
1160 }
1161
1162 static struct file_operations content_file_operations = {
1163         .open           = content_open,
1164         .read           = seq_read,
1165         .llseek         = seq_lseek,
1166         .release        = content_release,
1167 };
1168
1169 static ssize_t read_flush(struct file *file, char __user *buf,
1170                             size_t count, loff_t *ppos)
1171 {
1172         struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
1173         char tbuf[20];
1174         unsigned long p = *ppos;
1175         int len;
1176
1177         sprintf(tbuf, "%lu\n", cd->flush_time);
1178         len = strlen(tbuf);
1179         if (p >= len)
1180                 return 0;
1181         len -= p;
1182         if (len > count) len = count;
1183         if (copy_to_user(buf, (void*)(tbuf+p), len))
1184                 len = -EFAULT;
1185         else
1186                 *ppos += len;
1187         return len;
1188 }
1189
1190 static ssize_t write_flush(struct file * file, const char __user * buf,
1191                              size_t count, loff_t *ppos)
1192 {
1193         struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
1194         char tbuf[20];
1195         char *ep;
1196         long flushtime;
1197         if (*ppos || count > sizeof(tbuf)-1)
1198                 return -EINVAL;
1199         if (copy_from_user(tbuf, buf, count))
1200                 return -EFAULT;
1201         tbuf[count] = 0;
1202         flushtime = simple_strtoul(tbuf, &ep, 0);
1203         if (*ep && *ep != '\n')
1204                 return -EINVAL;
1205
1206         cd->flush_time = flushtime;
1207         cd->nextcheck = get_seconds();
1208         cache_flush();
1209
1210         *ppos += count;
1211         return count;
1212 }
1213
1214 static struct file_operations cache_flush_operations = {
1215         .read           = read_flush,
1216         .write          = write_flush,
1217 };