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