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         list_add(&cd->others, &cache_list);
 218         spin_unlock(&cache_list_lock);
 219
 220         /* start the cleaning process */
 221         schedule_work(&cache_cleaner);
 222 }
 223
 224 int cache_unregister(struct cache_detail *cd)
 225 {
 226         cache_purge(cd);
 227         spin_lock(&cache_list_lock);
 228         write_lock(&cd->hash_lock);
 229         if (cd->entries || atomic_read(&cd->inuse)) {
 230                 write_unlock(&cd->hash_lock);
 231                 spin_unlock(&cache_list_lock);
 232                 return -EBUSY;
 233         }
 234         if (current_detail == cd)
 235                 current_detail = NULL;
 236         list_del_init(&cd->others);
 237         write_unlock(&cd->hash_lock);
 238         spin_unlock(&cache_list_lock);
 239         if (cd->proc_ent) {
 240                 if (cd->flush_ent)
 241                         remove_proc_entry("flush", cd->proc_ent);
 242                 if (cd->channel_ent)
 243                         remove_proc_entry("channel", cd->proc_ent);
 244                 if (cd->content_ent)
 245                         remove_proc_entry("content", cd->proc_ent);
 246
 247                 cd->proc_ent = NULL;
 248                 remove_proc_entry(cd->name, proc_net_rpc);
 249         }
 250         if (list_empty(&cache_list)) {
 251                 /* module must be being unloaded so its safe to kill the worker */
 252                 cancel_delayed_work(&cache_cleaner);
 253                 flush_scheduled_work();
 254         }
 255         return 0;
 256 }
 257
 258 struct cache_detail *cache_find(char *name)
 259 {
 260         struct list_head *l;
 261
 262         spin_lock(&cache_list_lock);
 263         list_for_each(l, &cache_list) {
 264                 struct cache_detail *cd = list_entry(l, struct cache_detail, others);
 265
 266                 if (strcmp(cd->name, name)==0) {
 267                         atomic_inc(&cd->inuse);
 268                         spin_unlock(&cache_list_lock);
 269                         return cd;
 270                 }
 271         }
 272         spin_unlock(&cache_list_lock);
 273         return NULL;
 274 }
 275
 276 /* cache_drop must be called on any cache returned by
 277  * cache_find, after it has been used
 278  */
 279 void cache_drop(struct cache_detail *detail)
 280 {
 281         atomic_dec(&detail->inuse);
 282 }
 283
 284 /* clean cache tries to find something to clean
 285  * and cleans it.
 286  * It returns 1 if it cleaned something,
 287  *            0 if it didn't find anything this time
 288  *           -1 if it fell off the end of the list.
 289  */
 290 int cache_clean(void)
 291 {
 292         int rv = 0;
 293         struct list_head *next;
 294
 295         spin_lock(&cache_list_lock);
 296
 297         /* find a suitable table if we don't already have one */
 298         while (current_detail == NULL ||
 299             current_index >= current_detail->hash_size) {
 300                 if (current_detail)
 301                         next = current_detail->others.next;
 302                 else
 303                         next = cache_list.next;
 304                 if (next == &cache_list) {
 305                         current_detail = NULL;
 306                         spin_unlock(&cache_list_lock);
 307                         return -1;
 308                 }
 309                 current_detail = list_entry(next, struct cache_detail, others);
 310                 if (current_detail->nextcheck > get_seconds())
 311                         current_index = current_detail->hash_size;
 312                 else {
 313                         current_index = 0;
 314                         current_detail->nextcheck = get_seconds()+30*60;
 315                 }
 316         }
 317
 318         /* find a non-empty bucket in the table */
 319         while (current_detail &&
 320                current_index < current_detail->hash_size &&
 321                current_detail->hash_table[current_index] == NULL)
 322                 current_index++;
 323
 324         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 325
 326         if (current_detail && current_index < current_detail->hash_size) {
 327                 struct cache_head *ch, **cp;
 328                 struct cache_detail *d;
 329
 330                 write_lock(&current_detail->hash_lock);
 331
 332                 /* Ok, now to clean this strand */
 333
 334                 cp = & current_detail->hash_table[current_index];
 335                 ch = *cp;
 336                 for (; ch; cp= & ch->next, ch= *cp) {
 337                         if (current_detail->nextcheck > ch->expiry_time)
 338                                 current_detail->nextcheck = ch->expiry_time+1;
 339                         if (ch->expiry_time >= get_seconds()
 340                             && ch->last_refresh >= current_detail->flush_time
 341                                 )
 342                                 continue;
 343                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 344                                 queue_loose(current_detail, ch);
 345
 346                         if (!atomic_read(&ch->refcnt))
 347                                 break;
 348                 }
 349                 if (ch) {
 350                         cache_get(ch);
 351                         clear_bit(CACHE_HASHED, &ch->flags);
 352                         *cp = ch->next;
 353                         ch->next = NULL;
 354                         current_detail->entries--;
 355                         rv = 1;
 356                 }
 357                 write_unlock(&current_detail->hash_lock);
 358                 d = current_detail;
 359                 if (!ch)
 360                         current_index ++;
 361                 spin_unlock(&cache_list_lock);
 362                 if (ch)
 363                         d->cache_put(ch, d);
 364         } else
 365                 spin_unlock(&cache_list_lock);
 366
 367         return rv;
 368 }
 369
 370 /*
 371  * We want to regularly clean the cache, so we need to schedule some work ...
 372  */
 373 static void do_cache_clean(void *data)
 374 {
 375         int delay = 5;
 376         if (cache_clean() == -1)
 377                 delay = 30*HZ;
 378
 379         if (list_empty(&cache_list))
 380                 delay = 0;
 381
 382         if (delay)
 383                 schedule_delayed_work(&cache_cleaner, delay);
 384 }
 385
 386
 387 /*
 388  * Clean all caches promptly.  This just calls cache_clean
 389  * repeatedly until we are sure that every cache has had a chance to
 390  * be fully cleaned
 391  */
 392 void cache_flush(void)
 393 {
 394         while (cache_clean() != -1)
 395                 cond_resched();
 396         while (cache_clean() != -1)
 397                 cond_resched();
 398 }
 399
 400 void cache_purge(struct cache_detail *detail)
 401 {
 402         detail->flush_time = get_seconds()+1;
 403         detail->nextcheck = get_seconds();
 404         cache_flush();
 405 }
 406
 407
 408
 409 /*
 410  * Deferral and Revisiting of Requests.
 411  *
 412  * If a cache lookup finds a pending entry, we
 413  * need to defer the request and revisit it later.
 414  * All deferred requests are stored in a hash table,
 415  * indexed by "struct cache_head *".
 416  * As it may be wasteful to store a whole request
 417  * structure, we allow the request to provide a
 418  * deferred form, which must contain a
 419  * 'struct cache_deferred_req'
 420  * This cache_deferred_req contains a method to allow
 421  * it to be revisited when cache info is available
 422  */
 423
 424 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 425 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 426
 427 #define DFR_MAX 300     /* ??? */
 428
 429 spinlock_t cache_defer_lock = SPIN_LOCK_UNLOCKED;
 430 static LIST_HEAD(cache_defer_list);
 431 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 432 static int cache_defer_cnt;
 433
 434 void cache_defer_req(struct cache_req *req, struct cache_head *item)
 435 {
 436         struct cache_deferred_req *dreq;
 437         int hash = DFR_HASH(item);
 438
 439         dreq = req->defer(req);
 440         if (dreq == NULL)
 441                 return;
 442
 443         dreq->item = item;
 444         dreq->recv_time = get_seconds();
 445
 446         spin_lock(&cache_defer_lock);
 447
 448         list_add(&dreq->recent, &cache_defer_list);
 449
 450         if (cache_defer_hash[hash].next == NULL)
 451                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 452         list_add(&dreq->hash, &cache_defer_hash[hash]);
 453
 454         /* it is in, now maybe clean up */
 455         dreq = NULL;
 456         if (++cache_defer_cnt > DFR_MAX) {
 457                 /* too much in the cache, randomly drop
 458                  * first or last
 459                  */
 460                 if (net_random()&1)
 461                         dreq = list_entry(cache_defer_list.next,
 462                                           struct cache_deferred_req,
 463                                           recent);
 464                 else
 465                         dreq = list_entry(cache_defer_list.prev,
 466                                           struct cache_deferred_req,
 467                                           recent);
 468                 list_del(&dreq->recent);
 469                 list_del(&dreq->hash);
 470                 cache_defer_cnt--;
 471         }
 472         spin_unlock(&cache_defer_lock);
 473
 474         if (dreq) {
 475                 /* there was one too many */
 476                 dreq->revisit(dreq, 1);
 477         }
 478         if (test_bit(CACHE_VALID, &item->flags)) {
 479                 /* must have just been validated... */
 480                 cache_revisit_request(item);
 481         }
 482 }
 483
 484 void cache_revisit_request(struct cache_head *item)
 485 {
 486         struct cache_deferred_req *dreq;
 487         struct list_head pending;
 488
 489         struct list_head *lp;
 490         int hash = DFR_HASH(item);
 491
 492         INIT_LIST_HEAD(&pending);
 493         spin_lock(&cache_defer_lock);
 494
 495         lp = cache_defer_hash[hash].next;
 496         if (lp) {
 497                 while (lp != &cache_defer_hash[hash]) {
 498                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 499                         lp = lp->next;
 500                         if (dreq->item == item) {
 501                                 list_del(&dreq->hash);
 502                                 list_move(&dreq->recent, &pending);
 503                                 cache_defer_cnt--;
 504                         }
 505                 }
 506         }
 507         spin_unlock(&cache_defer_lock);
 508
 509         while (!list_empty(&pending)) {
 510                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 511                 list_del_init(&dreq->recent);
 512                 dreq->revisit(dreq, 0);
 513         }
 514 }
 515
 516 void cache_clean_deferred(void *owner)
 517 {
 518         struct cache_deferred_req *dreq, *tmp;
 519         struct list_head pending;
 520
 521
 522         INIT_LIST_HEAD(&pending);
 523         spin_lock(&cache_defer_lock);
 524
 525         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 526                 if (dreq->owner == owner) {
 527                         list_del(&dreq->hash);
 528                         list_move(&dreq->recent, &pending);
 529                         cache_defer_cnt--;
 530                 }
 531         }
 532         spin_unlock(&cache_defer_lock);
 533
 534         while (!list_empty(&pending)) {
 535                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 536                 list_del_init(&dreq->recent);
 537                 dreq->revisit(dreq, 1);
 538         }
 539 }
 540
 541 /*
 542  * communicate with user-space
 543  *
 544  * We have a magic /proc file - /proc/sunrpc/cache
 545  * On read, you get a full request, or block
 546  * On write, an update request is processed
 547  * Poll works if anything to read, and always allows write
 548  *
 549  * Implemented by linked list of requests.  Each open file has
 550  * a ->private that also exists in this list.  New request are added
 551  * to the end and may wakeup and preceding readers.
 552  * New readers are added to the head.  If, on read, an item is found with
 553  * CACHE_UPCALLING clear, we free it from the list.
 554  *
 555  */
 556
 557 static spinlock_t queue_lock = SPIN_LOCK_UNLOCKED;
 558 static DECLARE_MUTEX(queue_io_sem);
 559
 560 struct cache_queue {
 561         struct list_head        list;
 562         int                     reader; /* if 0, then request */
 563 };
 564 struct cache_request {
 565         struct cache_queue      q;
 566         struct cache_head       *item;
 567         char                    * buf;
 568         int                     len;
 569         int                     readers;
 570 };
 571 struct cache_reader {
 572         struct cache_queue      q;
 573         int                     offset; /* if non-0, we have a refcnt on next request */
 574 };
 575
 576 static ssize_t
 577 cache_read(struct file *filp, char *buf, size_t count, loff_t *ppos)
 578 {
 579         struct cache_reader *rp = filp->private_data;
 580         struct cache_request *rq;
 581         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
 582         int err;
 583
 584         if (ppos != &filp->f_pos)
 585                 return -ESPIPE;
 586
 587         if (count == 0)
 588                 return 0;
 589
 590         down(&queue_io_sem); /* protect against multiple concurrent
 591                               * readers on this file */
 592  again:
 593         spin_lock(&queue_lock);
 594         /* need to find next request */
 595         while (rp->q.list.next != &cd->queue &&
 596                list_entry(rp->q.list.next, struct cache_queue, list)
 597                ->reader) {
 598                 struct list_head *next = rp->q.list.next;
 599                 list_move(&rp->q.list, next);
 600         }
 601         if (rp->q.list.next == &cd->queue) {
 602                 spin_unlock(&queue_lock);
 603                 up(&queue_io_sem);
 604                 if (rp->offset)
 605                         BUG();
 606                 return 0;
 607         }
 608         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 609         if (rq->q.reader) BUG();
 610         if (rp->offset == 0)
 611                 rq->readers++;
 612         spin_unlock(&queue_lock);
 613
 614         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 615                 err = -EAGAIN;
 616                 spin_lock(&queue_lock);
 617                 list_move(&rp->q.list, &rq->q.list);
 618                 spin_unlock(&queue_lock);
 619         } else {
 620                 if (rp->offset + count > rq->len)
 621                         count = rq->len - rp->offset;
 622                 err = -EFAULT;
 623                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 624                         goto out;
 625                 rp->offset += count;
 626                 if (rp->offset >= rq->len) {
 627                         rp->offset = 0;
 628                         spin_lock(&queue_lock);
 629                         list_move(&rp->q.list, &rq->q.list);
 630                         spin_unlock(&queue_lock);
 631                 }
 632                 err = 0;
 633         }
 634  out:
 635         if (rp->offset == 0) {
 636                 /* need to release rq */
 637                 spin_lock(&queue_lock);
 638                 rq->readers--;
 639                 if (rq->readers == 0 &&
 640                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 641                         list_del(&rq->q.list);
 642                         spin_unlock(&queue_lock);
 643                         cd->cache_put(rq->item, cd);
 644                         kfree(rq->buf);
 645                         kfree(rq);
 646                 } else
 647                         spin_unlock(&queue_lock);
 648         }
 649         if (err == -EAGAIN)
 650                 goto again;
 651         up(&queue_io_sem);
 652         return err ? err :  count;
 653 }
 654
 655 static ssize_t
 656 cache_write(struct file *filp, const char *buf, size_t count,
 657             loff_t *ppos)
 658 {
 659         int err;
 660         char *page;
 661         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
 662
 663         if (ppos != &filp->f_pos)
 664                 return -ESPIPE;
 665
 666         if (count == 0)
 667                 return 0;
 668         if (count > PAGE_SIZE)
 669                 return -EINVAL;
 670
 671         down(&queue_io_sem);
 672
 673         page = kmalloc(PAGE_SIZE, GFP_KERNEL);
 674         if (page == NULL) {
 675                 up(&queue_io_sem);
 676                 return -ENOMEM;
 677         }
 678
 679         if (copy_from_user(page, buf, count)) {
 680                 up(&queue_io_sem);
 681                 kfree(page);
 682                 return -EFAULT;
 683         }
 684         if (count < PAGE_SIZE)
 685                 page[count] = '\0';
 686         if (cd->cache_parse)
 687                 err = cd->cache_parse(cd, page, count);
 688         else
 689                 err = -EINVAL;
 690
 691         up(&queue_io_sem);
 692         kfree(page);
 693         return err ? err : count;
 694 }
 695
 696 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 697
 698 static unsigned int
 699 cache_poll(struct file *filp, poll_table *wait)
 700 {
 701         unsigned int mask;
 702         struct cache_reader *rp = filp->private_data;
 703         struct cache_queue *cq;
 704         struct cache_detail *cd = PDE(filp->f_dentry->d_inode)->data;
 705
 706         poll_wait(filp, &queue_wait, wait);
 707
 708         /* alway allow write */
 709         mask = POLL_OUT | POLLWRNORM;
 710
 711         if (!rp)
 712                 return mask;
 713
 714         spin_lock(&queue_lock);
 715
 716         for (cq= &rp->q; &cq->list != &cd->queue;
 717              cq = list_entry(cq->list.next, struct cache_queue, list))
 718                 if (!cq->reader) {
 719                         mask |= POLLIN | POLLRDNORM;
 720                         break;
 721                 }
 722         spin_unlock(&queue_lock);
 723         return mask;
 724 }
 725
 726 static int
 727 cache_ioctl(struct inode *ino, struct file *filp,
 728             unsigned int cmd, unsigned long arg)
 729 {
 730         int len = 0;
 731         struct cache_reader *rp = filp->private_data;
 732         struct cache_queue *cq;
 733         struct cache_detail *cd = PDE(ino)->data;
 734
 735         if (cmd != FIONREAD || !rp)
 736                 return -EINVAL;
 737
 738         spin_lock(&queue_lock);
 739
 740         /* only find the length remaining in current request,
 741          * or the length of the next request
 742          */
 743         for (cq= &rp->q; &cq->list != &cd->queue;
 744              cq = list_entry(cq->list.next, struct cache_queue, list))
 745                 if (!cq->reader) {
 746                         struct cache_request *cr =
 747                                 container_of(cq, struct cache_request, q);
 748                         len = cr->len - rp->offset;
 749                         break;
 750                 }
 751         spin_unlock(&queue_lock);
 752
 753         return put_user(len, (int *)arg);
 754 }
 755
 756 static int
 757 cache_open(struct inode *inode, struct file *filp)
 758 {
 759         struct cache_reader *rp = NULL;
 760
 761         if (filp->f_mode & FMODE_READ) {
 762                 struct cache_detail *cd = PDE(inode)->data;
 763
 764                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 765                 if (!rp)
 766                         return -ENOMEM;
 767                 rp->offset = 0;
 768                 rp->q.reader = 1;
 769                 atomic_inc(&cd->readers);
 770                 spin_lock(&queue_lock);
 771                 list_add(&rp->q.list, &cd->queue);
 772                 spin_unlock(&queue_lock);
 773         }
 774         filp->private_data = rp;
 775         return 0;
 776 }
 777
 778 static int
 779 cache_release(struct inode *inode, struct file *filp)
 780 {
 781         struct cache_reader *rp = filp->private_data;
 782         struct cache_detail *cd = PDE(inode)->data;
 783
 784         if (rp) {
 785                 spin_lock(&queue_lock);
 786                 if (rp->offset) {
 787                         struct cache_queue *cq;
 788                         for (cq= &rp->q; &cq->list != &cd->queue;
 789                              cq = list_entry(cq->list.next, struct cache_queue, list))
 790                                 if (!cq->reader) {
 791                                         container_of(cq, struct cache_request, q)
 792                                                 ->readers--;
 793                                         break;
 794                                 }
 795                         rp->offset = 0;
 796                 }
 797                 list_del(&rp->q.list);
 798                 spin_unlock(&queue_lock);
 799
 800                 filp->private_data = NULL;
 801                 kfree(rp);
 802
 803                 cd->last_close = get_seconds();
 804                 atomic_dec(&cd->readers);
 805         }
 806         return 0;
 807 }
 808
 809
 810
 811 static struct file_operations cache_file_operations = {
 812         .owner          = THIS_MODULE,
 813         .llseek         = no_llseek,
 814         .read           = cache_read,
 815         .write          = cache_write,
 816         .poll           = cache_poll,
 817         .ioctl          = cache_ioctl, /* for FIONREAD */
 818         .open           = cache_open,
 819         .release        = cache_release,
 820 };
 821
 822
 823 static void queue_loose(struct cache_detail *detail, struct cache_head *ch)
 824 {
 825         struct cache_queue *cq;
 826         spin_lock(&queue_lock);
 827         list_for_each_entry(cq, &detail->queue, list)
 828                 if (!cq->reader) {
 829                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 830                         if (cr->item != ch)
 831                                 continue;
 832                         if (cr->readers != 0)
 833                                 break;
 834                         list_del(&cr->q.list);
 835                         spin_unlock(&queue_lock);
 836                         detail->cache_put(cr->item, detail);
 837                         kfree(cr->buf);
 838                         kfree(cr);
 839                         return;
 840                 }
 841         spin_unlock(&queue_lock);
 842 }
 843
 844 /*
 845  * Support routines for text-based upcalls.
 846  * Fields are separated by spaces.
 847  * Fields are either mangled to quote space tab newline slosh with slosh
 848  * or a hexified with a leading \x
 849  * Record is terminated with newline.
 850  *
 851  */
 852
 853 void qword_add(char **bpp, int *lp, char *str)
 854 {
 855         char *bp = *bpp;
 856         int len = *lp;
 857         char c;
 858
 859         if (len < 0) return;
 860
 861         while ((c=*str++) && len)
 862                 switch(c) {
 863                 case ' ':
 864                 case '\t':
 865                 case '\n':
 866                 case '\\':
 867                         if (len >= 4) {
 868                                 *bp++ = '\\';
 869                                 *bp++ = '0' + ((c & 0300)>>6);
 870                                 *bp++ = '0' + ((c & 0070)>>3);
 871                                 *bp++ = '0' + ((c & 0007)>>0);
 872                         }
 873                         len -= 4;
 874                         break;
 875                 default:
 876                         *bp++ = c;
 877                         len--;
 878                 }
 879         if (c || len <1) len = -1;
 880         else {
 881                 *bp++ = ' ';
 882                 len--;
 883         }
 884         *bpp = bp;
 885         *lp = len;
 886 }
 887
 888 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
 889 {
 890         char *bp = *bpp;
 891         int len = *lp;
 892
 893         if (len < 0) return;
 894
 895         if (len > 2) {
 896                 *bp++ = '\\';
 897                 *bp++ = 'x';
 898                 len -= 2;
 899                 while (blen && len >= 2) {
 900                         unsigned char c = *buf++;
 901                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
 902                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
 903                         len -= 2;
 904                         blen--;
 905                 }
 906         }
 907         if (blen || len<1) len = -1;
 908         else {
 909                 *bp++ = ' ';
 910                 len--;
 911         }
 912         *bpp = bp;
 913         *lp = len;
 914 }
 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() - 60)
 935                 /* nobody is listening */
 936                 return -EINVAL;
 937
 938         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
 939         if (!buf)
 940                 return -EAGAIN;
 941
 942         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
 943         if (!crq) {
 944                 kfree(buf);
 945                 return -EAGAIN;
 946         }
 947
 948         bp = buf; len = PAGE_SIZE;
 949
 950         detail->cache_request(detail, h, &bp, &len);
 951
 952         if (len < 0) {
 953                 kfree(buf);
 954                 kfree(crq);
 955                 return -EAGAIN;
 956         }
 957         crq->q.reader = 0;
 958         crq->item = cache_get(h);
 959         crq->buf = buf;
 960         crq->len = PAGE_SIZE - len;
 961         crq->readers = 0;
 962         spin_lock(&queue_lock);
 963         list_add_tail(&crq->q.list, &detail->queue);
 964         spin_unlock(&queue_lock);
 965         wake_up(&queue_wait);
 966         return 0;
 967 }
 968
 969 /*
 970  * parse a message from user-space and pass it
 971  * to an appropriate cache
 972  * Messages are, like requests, separated into fields by
 973  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
 974  *
 975  * Message is
 976  *   reply cachename expiry key ... content....
 977  *
 978  * key and content are both parsed by cache
 979  */
 980
 981 #define isodigit(c) (isdigit(c) && c <= '7')
 982 int qword_get(char **bpp, char *dest, int bufsize)
 983 {
 984         /* return bytes copied, or -1 on error */
 985         char *bp = *bpp;
 986         int len = 0;
 987
 988         while (*bp == ' ') bp++;
 989
 990         if (bp[0] == '\\' && bp[1] == 'x') {
 991                 /* HEX STRING */
 992                 bp += 2;
 993                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
 994                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
 995                         bp++;
 996                         byte <<= 4;
 997                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
 998                         *dest++ = byte;
 999                         bp++;
1000                         len++;
1001                 }
1002         } else {
1003                 /* text with \nnn octal quoting */
1004                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1005                         if (*bp == '\\' &&
1006                             isodigit(bp[1]) && (bp[1] <= '3') &&
1007                             isodigit(bp[2]) &&
1008                             isodigit(bp[3])) {
1009                                 int byte = (*++bp -'0');
1010                                 bp++;
1011                                 byte = (byte << 3) | (*bp++ - '0');
1012                                 byte = (byte << 3) | (*bp++ - '0');
1013                                 *dest++ = byte;
1014                                 len++;
1015                         } else {
1016                                 *dest++ = *bp++;
1017                                 len++;
1018                         }
1019                 }
1020         }
1021
1022         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1023                 return -1;
1024         while (*bp == ' ') bp++;
1025         *bpp = bp;
1026         *dest = '\0';
1027         return len;
1028 }
1029
1030
1031 /*
1032  * support /proc/sunrpc/cache/$CACHENAME/content
1033  * as a seqfile.
1034  * We call ->cache_show passing NULL for the item to
1035  * get a header, then pass each real item in the cache
1036  */
1037
1038 struct handle {
1039         struct cache_detail *cd;
1040 };
1041
1042 static void *c_start(struct seq_file *m, loff_t *pos)
1043 {
1044         loff_t n = *pos;
1045         unsigned hash, entry;
1046         struct cache_head *ch;
1047         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1048
1049
1050         read_lock(&cd->hash_lock);
1051         if (!n--)
1052                 return SEQ_START_TOKEN;
1053         hash = n >> 32;
1054         entry = n & ((1LL<<32) - 1);
1055
1056         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1057                 if (!entry--)
1058                         return ch;
1059         n &= ~((1LL<<32) - 1);
1060         do {
1061                 hash++;
1062                 n += 1LL<<32;
1063         } while(hash < cd->hash_size &&
1064                 cd->hash_table[hash]==NULL);
1065         if (hash >= cd->hash_size)
1066                 return NULL;
1067         *pos = n+1;
1068         return cd->hash_table[hash];
1069 }
1070
1071 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1072 {
1073         struct cache_head *ch = p;
1074         int hash = (*pos >> 32);
1075         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1076
1077         if (p == SEQ_START_TOKEN)
1078                 hash = 0;
1079         else if (ch->next == NULL) {
1080                 hash++;
1081                 *pos += 1LL<<32;
1082         } else {
1083                 ++*pos;
1084                 return ch->next;
1085         }
1086         *pos &= ~((1LL<<32) - 1);
1087         while (hash < cd->hash_size &&
1088                cd->hash_table[hash] == NULL) {
1089                 hash++;
1090                 *pos += 1LL<<32;
1091         }
1092         if (hash >= cd->hash_size)
1093                 return NULL;
1094         ++*pos;
1095         return cd->hash_table[hash];
1096 }
1097
1098 static void c_stop(struct seq_file *m, void *p)
1099 {
1100         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1101         read_unlock(&cd->hash_lock);
1102 }
1103
1104 static int c_show(struct seq_file *m, void *p)
1105 {
1106         struct cache_head *cp = p;
1107         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1108
1109         if (p == SEQ_START_TOKEN)
1110                 return cd->cache_show(m, cd, NULL);
1111
1112         ifdebug(CACHE)
1113                 seq_printf(m, "# expiry=%ld refcnt=%d\n",
1114                            cp->expiry_time, atomic_read(&cp->refcnt));
1115         cache_get(cp);
1116         if (cache_check(cd, cp, NULL))
1117                 /* cache_check does a cache_put on failure */
1118                 seq_printf(m, "# ");
1119         else
1120                 cache_put(cp, cd);
1121
1122         return cd->cache_show(m, cd, cp);
1123 }
1124
1125 struct seq_operations cache_content_op = {
1126         .start  = c_start,
1127         .next   = c_next,
1128         .stop   = c_stop,
1129         .show   = c_show,
1130 };
1131
1132 static int content_open(struct inode *inode, struct file *file)
1133 {
1134         int res;
1135         struct handle *han;
1136         struct cache_detail *cd = PDE(inode)->data;
1137
1138         han = kmalloc(sizeof(*han), GFP_KERNEL);
1139         if (han == NULL)
1140                 return -ENOMEM;
1141
1142         han->cd = cd;
1143
1144         res = seq_open(file, &cache_content_op);
1145         if (res)
1146                 kfree(han);
1147         else
1148                 ((struct seq_file *)file->private_data)->private = han;
1149
1150         return res;
1151 }
1152 static int content_release(struct inode *inode, struct file *file)
1153 {
1154         struct seq_file *m = (struct seq_file *)file->private_data;
1155         struct handle *han = m->private;
1156         kfree(han);
1157         m->private = NULL;
1158         return seq_release(inode, file);
1159 }
1160
1161 static struct file_operations content_file_operations = {
1162         .open           = content_open,
1163         .read           = seq_read,
1164         .llseek         = seq_lseek,
1165         .release        = content_release,
1166 };
1167
1168 static ssize_t read_flush(struct file *file, char *buf,
1169                             size_t count, loff_t *ppos)
1170 {
1171         struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
1172         char tbuf[20];
1173         unsigned long p = *ppos;
1174         int len;
1175
1176         sprintf(tbuf, "%lu\n", cd->flush_time);
1177         len = strlen(tbuf);
1178         if (p >= len)
1179                 return 0;
1180         len -= p;
1181         if (len > count) len = count;
1182         if (copy_to_user(buf, (void*)(tbuf+p), len))
1183                 len = -EFAULT;
1184         else
1185                 *ppos += len;
1186         return len;
1187 }
1188
1189 static ssize_t write_flush(struct file * file, const char * buf,
1190                              size_t count, loff_t *ppos)
1191 {
1192         struct cache_detail *cd = PDE(file->f_dentry->d_inode)->data;
1193         char tbuf[20];
1194         char *ep;
1195         long flushtime;
1196         if (*ppos || count > sizeof(tbuf)-1)
1197                 return -EINVAL;
1198         if (copy_from_user(tbuf, buf, count))
1199                 return -EFAULT;
1200         tbuf[count] = 0;
1201         flushtime = simple_strtoul(tbuf, &ep, 0);
1202         if (*ep && *ep != '\n')
1203                 return -EINVAL;
1204
1205         cd->flush_time = flushtime;
1206         cd->nextcheck = get_seconds();
1207         cache_flush();
1208
1209         *ppos += count;
1210         return count;
1211 }
1212
1213 static struct file_operations cache_flush_operations = {
1214         .read           = read_flush,
1215         .write          = write_flush,
1216 };