net/sunrpc/cache.c

   1 /*
   2  * net/sunrpc/cache.c
   3  *
   4  * Generic code for various authentication-related caches
   5  * used by sunrpc clients and servers.
   6  *
   7  * Copyright (C) 2002 Neil Brown <neilb@cse.unsw.edu.au>
   8  *
   9  * Released under terms in GPL version 2.  See COPYING.
  10  *
  11  */
  12
  13 #include <linux/types.h>
  14 #include <linux/fs.h>
  15 #include <linux/file.h>
  16 #include <linux/slab.h>
  17 #include <linux/signal.h>
  18 #include <linux/sched.h>
  19 #include <linux/kmod.h>
  20 #include <linux/list.h>
  21 #include <linux/module.h>
  22 #include <linux/ctype.h>
  23 #include <asm/uaccess.h>
  24 #include <linux/poll.h>
  25 #include <linux/seq_file.h>
  26 #include <linux/proc_fs.h>
  27 #include <linux/net.h>
  28 #include <linux/workqueue.h>
  29 #include <linux/mutex.h>
  30 #include <linux/pagemap.h>
  31 #include <linux/smp_lock.h>
  32 #include <asm/ioctls.h>
  33 #include <linux/sunrpc/types.h>
  34 #include <linux/sunrpc/cache.h>
  35 #include <linux/sunrpc/stats.h>
  36 #include <linux/sunrpc/rpc_pipe_fs.h>
  37
  38 #define  RPCDBG_FACILITY RPCDBG_CACHE
  39
  40 static int cache_defer_req(struct cache_req *req, struct cache_head *item);
  41 static void cache_revisit_request(struct cache_head *item);
  42
  43 static void cache_init(struct cache_head *h)
  44 {
  45         time_t now = get_seconds();
  46         h->next = NULL;
  47         h->flags = 0;
  48         kref_init(&h->ref);
  49         h->expiry_time = now + CACHE_NEW_EXPIRY;
  50         h->last_refresh = now;
  51 }
  52
  53 static inline int cache_is_expired(struct cache_detail *detail, struct cache_head *h)
  54 {
  55         return  (h->expiry_time < get_seconds()) ||
  56                 (detail->flush_time > h->last_refresh);
  57 }
  58
  59 struct cache_head *sunrpc_cache_lookup(struct cache_detail *detail,
  60                                        struct cache_head *key, int hash)
  61 {
  62         struct cache_head **head,  **hp;
  63         struct cache_head *new = NULL, *freeme = NULL;
  64
  65         head = &detail->hash_table[hash];
  66
  67         read_lock(&detail->hash_lock);
  68
  69         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  70                 struct cache_head *tmp = *hp;
  71                 if (detail->match(tmp, key)) {
  72                         if (cache_is_expired(detail, tmp))
  73                                 /* This entry is expired, we will discard it. */
  74                                 break;
  75                         cache_get(tmp);
  76                         read_unlock(&detail->hash_lock);
  77                         return tmp;
  78                 }
  79         }
  80         read_unlock(&detail->hash_lock);
  81         /* Didn't find anything, insert an empty entry */
  82
  83         new = detail->alloc();
  84         if (!new)
  85                 return NULL;
  86         /* must fully initialise 'new', else
  87          * we might get lose if we need to
  88          * cache_put it soon.
  89          */
  90         cache_init(new);
  91         detail->init(new, key);
  92
  93         write_lock(&detail->hash_lock);
  94
  95         /* check if entry appeared while we slept */
  96         for (hp=head; *hp != NULL ; hp = &(*hp)->next) {
  97                 struct cache_head *tmp = *hp;
  98                 if (detail->match(tmp, key)) {
  99                         if (cache_is_expired(detail, tmp)) {
 100                                 *hp = tmp->next;
 101                                 tmp->next = NULL;
 102                                 detail->entries --;
 103                                 freeme = tmp;
 104                                 break;
 105                         }
 106                         cache_get(tmp);
 107                         write_unlock(&detail->hash_lock);
 108                         cache_put(new, detail);
 109                         return tmp;
 110                 }
 111         }
 112         new->next = *head;
 113         *head = new;
 114         detail->entries++;
 115         cache_get(new);
 116         write_unlock(&detail->hash_lock);
 117
 118         if (freeme)
 119                 cache_put(freeme, detail);
 120         return new;
 121 }
 122 EXPORT_SYMBOL_GPL(sunrpc_cache_lookup);
 123
 124
 125 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch);
 126
 127 static void cache_fresh_locked(struct cache_head *head, time_t expiry)
 128 {
 129         head->expiry_time = expiry;
 130         head->last_refresh = get_seconds();
 131         set_bit(CACHE_VALID, &head->flags);
 132 }
 133
 134 static void cache_fresh_unlocked(struct cache_head *head,
 135                                  struct cache_detail *detail)
 136 {
 137         if (test_and_clear_bit(CACHE_PENDING, &head->flags)) {
 138                 cache_revisit_request(head);
 139                 cache_dequeue(detail, head);
 140         }
 141 }
 142
 143 struct cache_head *sunrpc_cache_update(struct cache_detail *detail,
 144                                        struct cache_head *new, struct cache_head *old, int hash)
 145 {
 146         /* The 'old' entry is to be replaced by 'new'.
 147          * If 'old' is not VALID, we update it directly,
 148          * otherwise we need to replace it
 149          */
 150         struct cache_head **head;
 151         struct cache_head *tmp;
 152
 153         if (!test_bit(CACHE_VALID, &old->flags)) {
 154                 write_lock(&detail->hash_lock);
 155                 if (!test_bit(CACHE_VALID, &old->flags)) {
 156                         if (test_bit(CACHE_NEGATIVE, &new->flags))
 157                                 set_bit(CACHE_NEGATIVE, &old->flags);
 158                         else
 159                                 detail->update(old, new);
 160                         cache_fresh_locked(old, new->expiry_time);
 161                         write_unlock(&detail->hash_lock);
 162                         cache_fresh_unlocked(old, detail);
 163                         return old;
 164                 }
 165                 write_unlock(&detail->hash_lock);
 166         }
 167         /* We need to insert a new entry */
 168         tmp = detail->alloc();
 169         if (!tmp) {
 170                 cache_put(old, detail);
 171                 return NULL;
 172         }
 173         cache_init(tmp);
 174         detail->init(tmp, old);
 175         head = &detail->hash_table[hash];
 176
 177         write_lock(&detail->hash_lock);
 178         if (test_bit(CACHE_NEGATIVE, &new->flags))
 179                 set_bit(CACHE_NEGATIVE, &tmp->flags);
 180         else
 181                 detail->update(tmp, new);
 182         tmp->next = *head;
 183         *head = tmp;
 184         detail->entries++;
 185         cache_get(tmp);
 186         cache_fresh_locked(tmp, new->expiry_time);
 187         cache_fresh_locked(old, 0);
 188         write_unlock(&detail->hash_lock);
 189         cache_fresh_unlocked(tmp, detail);
 190         cache_fresh_unlocked(old, detail);
 191         cache_put(old, detail);
 192         return tmp;
 193 }
 194 EXPORT_SYMBOL_GPL(sunrpc_cache_update);
 195
 196 static int cache_make_upcall(struct cache_detail *cd, struct cache_head *h)
 197 {
 198         if (!cd->cache_upcall)
 199                 return -EINVAL;
 200         return cd->cache_upcall(cd, h);
 201 }
 202
 203 static inline int cache_is_valid(struct cache_detail *detail, struct cache_head *h)
 204 {
 205         if (!test_bit(CACHE_VALID, &h->flags))
 206                 return -EAGAIN;
 207         else {
 208                 /* entry is valid */
 209                 if (test_bit(CACHE_NEGATIVE, &h->flags))
 210                         return -ENOENT;
 211                 else
 212                         return 0;
 213         }
 214 }
 215
 216 /*
 217  * This is the generic cache management routine for all
 218  * the authentication caches.
 219  * It checks the currency of a cache item and will (later)
 220  * initiate an upcall to fill it if needed.
 221  *
 222  *
 223  * Returns 0 if the cache_head can be used, or cache_puts it and returns
 224  * -EAGAIN if upcall is pending and request has been queued
 225  * -ETIMEDOUT if upcall failed or request could not be queue or
 226  *           upcall completed but item is still invalid (implying that
 227  *           the cache item has been replaced with a newer one).
 228  * -ENOENT if cache entry was negative
 229  */
 230 int cache_check(struct cache_detail *detail,
 231                     struct cache_head *h, struct cache_req *rqstp)
 232 {
 233         int rv;
 234         long refresh_age, age;
 235
 236         /* First decide return status as best we can */
 237         rv = cache_is_valid(detail, h);
 238
 239         /* now see if we want to start an upcall */
 240         refresh_age = (h->expiry_time - h->last_refresh);
 241         age = get_seconds() - h->last_refresh;
 242
 243         if (rqstp == NULL) {
 244                 if (rv == -EAGAIN)
 245                         rv = -ENOENT;
 246         } else if (rv == -EAGAIN || age > refresh_age/2) {
 247                 dprintk("RPC:       Want update, refage=%ld, age=%ld\n",
 248                                 refresh_age, age);
 249                 if (!test_and_set_bit(CACHE_PENDING, &h->flags)) {
 250                         switch (cache_make_upcall(detail, h)) {
 251                         case -EINVAL:
 252                                 clear_bit(CACHE_PENDING, &h->flags);
 253                                 cache_revisit_request(h);
 254                                 if (rv == -EAGAIN) {
 255                                         set_bit(CACHE_NEGATIVE, &h->flags);
 256                                         cache_fresh_locked(h, get_seconds()+CACHE_NEW_EXPIRY);
 257                                         cache_fresh_unlocked(h, detail);
 258                                         rv = -ENOENT;
 259                                 }
 260                                 break;
 261
 262                         case -EAGAIN:
 263                                 clear_bit(CACHE_PENDING, &h->flags);
 264                                 cache_revisit_request(h);
 265                                 break;
 266                         }
 267                 }
 268         }
 269
 270         if (rv == -EAGAIN) {
 271                 if (cache_defer_req(rqstp, h) < 0) {
 272                         /* Request is not deferred */
 273                         rv = cache_is_valid(detail, h);
 274                         if (rv == -EAGAIN)
 275                                 rv = -ETIMEDOUT;
 276                 }
 277         }
 278         if (rv)
 279                 cache_put(h, detail);
 280         return rv;
 281 }
 282 EXPORT_SYMBOL_GPL(cache_check);
 283
 284 /*
 285  * caches need to be periodically cleaned.
 286  * For this we maintain a list of cache_detail and
 287  * a current pointer into that list and into the table
 288  * for that entry.
 289  *
 290  * Each time clean_cache is called it finds the next non-empty entry
 291  * in the current table and walks the list in that entry
 292  * looking for entries that can be removed.
 293  *
 294  * An entry gets removed if:
 295  * - The expiry is before current time
 296  * - The last_refresh time is before the flush_time for that cache
 297  *
 298  * later we might drop old entries with non-NEVER expiry if that table
 299  * is getting 'full' for some definition of 'full'
 300  *
 301  * The question of "how often to scan a table" is an interesting one
 302  * and is answered in part by the use of the "nextcheck" field in the
 303  * cache_detail.
 304  * When a scan of a table begins, the nextcheck field is set to a time
 305  * that is well into the future.
 306  * While scanning, if an expiry time is found that is earlier than the
 307  * current nextcheck time, nextcheck is set to that expiry time.
 308  * If the flush_time is ever set to a time earlier than the nextcheck
 309  * time, the nextcheck time is then set to that flush_time.
 310  *
 311  * A table is then only scanned if the current time is at least
 312  * the nextcheck time.
 313  *
 314  */
 315
 316 static LIST_HEAD(cache_list);
 317 static DEFINE_SPINLOCK(cache_list_lock);
 318 static struct cache_detail *current_detail;
 319 static int current_index;
 320
 321 static void do_cache_clean(struct work_struct *work);
 322 static DECLARE_DELAYED_WORK(cache_cleaner, do_cache_clean);
 323
 324 static void sunrpc_init_cache_detail(struct cache_detail *cd)
 325 {
 326         rwlock_init(&cd->hash_lock);
 327         INIT_LIST_HEAD(&cd->queue);
 328         spin_lock(&cache_list_lock);
 329         cd->nextcheck = 0;
 330         cd->entries = 0;
 331         atomic_set(&cd->readers, 0);
 332         cd->last_close = 0;
 333         cd->last_warn = -1;
 334         list_add(&cd->others, &cache_list);
 335         spin_unlock(&cache_list_lock);
 336
 337         /* start the cleaning process */
 338         schedule_delayed_work(&cache_cleaner, 0);
 339 }
 340
 341 static void sunrpc_destroy_cache_detail(struct cache_detail *cd)
 342 {
 343         cache_purge(cd);
 344         spin_lock(&cache_list_lock);
 345         write_lock(&cd->hash_lock);
 346         if (cd->entries || atomic_read(&cd->inuse)) {
 347                 write_unlock(&cd->hash_lock);
 348                 spin_unlock(&cache_list_lock);
 349                 goto out;
 350         }
 351         if (current_detail == cd)
 352                 current_detail = NULL;
 353         list_del_init(&cd->others);
 354         write_unlock(&cd->hash_lock);
 355         spin_unlock(&cache_list_lock);
 356         if (list_empty(&cache_list)) {
 357                 /* module must be being unloaded so its safe to kill the worker */
 358                 cancel_delayed_work_sync(&cache_cleaner);
 359         }
 360         return;
 361 out:
 362         printk(KERN_ERR "nfsd: failed to unregister %s cache\n", cd->name);
 363 }
 364
 365 /* clean cache tries to find something to clean
 366  * and cleans it.
 367  * It returns 1 if it cleaned something,
 368  *            0 if it didn't find anything this time
 369  *           -1 if it fell off the end of the list.
 370  */
 371 static int cache_clean(void)
 372 {
 373         int rv = 0;
 374         struct list_head *next;
 375
 376         spin_lock(&cache_list_lock);
 377
 378         /* find a suitable table if we don't already have one */
 379         while (current_detail == NULL ||
 380             current_index >= current_detail->hash_size) {
 381                 if (current_detail)
 382                         next = current_detail->others.next;
 383                 else
 384                         next = cache_list.next;
 385                 if (next == &cache_list) {
 386                         current_detail = NULL;
 387                         spin_unlock(&cache_list_lock);
 388                         return -1;
 389                 }
 390                 current_detail = list_entry(next, struct cache_detail, others);
 391                 if (current_detail->nextcheck > get_seconds())
 392                         current_index = current_detail->hash_size;
 393                 else {
 394                         current_index = 0;
 395                         current_detail->nextcheck = get_seconds()+30*60;
 396                 }
 397         }
 398
 399         /* find a non-empty bucket in the table */
 400         while (current_detail &&
 401                current_index < current_detail->hash_size &&
 402                current_detail->hash_table[current_index] == NULL)
 403                 current_index++;
 404
 405         /* find a cleanable entry in the bucket and clean it, or set to next bucket */
 406
 407         if (current_detail && current_index < current_detail->hash_size) {
 408                 struct cache_head *ch, **cp;
 409                 struct cache_detail *d;
 410
 411                 write_lock(&current_detail->hash_lock);
 412
 413                 /* Ok, now to clean this strand */
 414
 415                 cp = & current_detail->hash_table[current_index];
 416                 for (ch = *cp ; ch ; cp = & ch->next, ch = *cp) {
 417                         if (current_detail->nextcheck > ch->expiry_time)
 418                                 current_detail->nextcheck = ch->expiry_time+1;
 419                         if (!cache_is_expired(current_detail, ch))
 420                                 continue;
 421
 422                         *cp = ch->next;
 423                         ch->next = NULL;
 424                         current_detail->entries--;
 425                         rv = 1;
 426                         break;
 427                 }
 428
 429                 write_unlock(&current_detail->hash_lock);
 430                 d = current_detail;
 431                 if (!ch)
 432                         current_index ++;
 433                 spin_unlock(&cache_list_lock);
 434                 if (ch) {
 435                         if (test_and_clear_bit(CACHE_PENDING, &ch->flags))
 436                                 cache_dequeue(current_detail, ch);
 437                         cache_revisit_request(ch);
 438                         cache_put(ch, d);
 439                 }
 440         } else
 441                 spin_unlock(&cache_list_lock);
 442
 443         return rv;
 444 }
 445
 446 /*
 447  * We want to regularly clean the cache, so we need to schedule some work ...
 448  */
 449 static void do_cache_clean(struct work_struct *work)
 450 {
 451         int delay = 5;
 452         if (cache_clean() == -1)
 453                 delay = round_jiffies_relative(30*HZ);
 454
 455         if (list_empty(&cache_list))
 456                 delay = 0;
 457
 458         if (delay)
 459                 schedule_delayed_work(&cache_cleaner, delay);
 460 }
 461
 462
 463 /*
 464  * Clean all caches promptly.  This just calls cache_clean
 465  * repeatedly until we are sure that every cache has had a chance to
 466  * be fully cleaned
 467  */
 468 void cache_flush(void)
 469 {
 470         while (cache_clean() != -1)
 471                 cond_resched();
 472         while (cache_clean() != -1)
 473                 cond_resched();
 474 }
 475 EXPORT_SYMBOL_GPL(cache_flush);
 476
 477 void cache_purge(struct cache_detail *detail)
 478 {
 479         detail->flush_time = LONG_MAX;
 480         detail->nextcheck = get_seconds();
 481         cache_flush();
 482         detail->flush_time = 1;
 483 }
 484 EXPORT_SYMBOL_GPL(cache_purge);
 485
 486
 487 /*
 488  * Deferral and Revisiting of Requests.
 489  *
 490  * If a cache lookup finds a pending entry, we
 491  * need to defer the request and revisit it later.
 492  * All deferred requests are stored in a hash table,
 493  * indexed by "struct cache_head *".
 494  * As it may be wasteful to store a whole request
 495  * structure, we allow the request to provide a
 496  * deferred form, which must contain a
 497  * 'struct cache_deferred_req'
 498  * This cache_deferred_req contains a method to allow
 499  * it to be revisited when cache info is available
 500  */
 501
 502 #define DFR_HASHSIZE    (PAGE_SIZE/sizeof(struct list_head))
 503 #define DFR_HASH(item)  ((((long)item)>>4 ^ (((long)item)>>13)) % DFR_HASHSIZE)
 504
 505 #define DFR_MAX 300     /* ??? */
 506
 507 static DEFINE_SPINLOCK(cache_defer_lock);
 508 static LIST_HEAD(cache_defer_list);
 509 static struct list_head cache_defer_hash[DFR_HASHSIZE];
 510 static int cache_defer_cnt;
 511
 512 static int cache_defer_req(struct cache_req *req, struct cache_head *item)
 513 {
 514         struct cache_deferred_req *dreq, *discard;
 515         int hash = DFR_HASH(item);
 516
 517         if (cache_defer_cnt >= DFR_MAX) {
 518                 /* too much in the cache, randomly drop this one,
 519                  * or continue and drop the oldest below
 520                  */
 521                 if (net_random()&1)
 522                         return -ENOMEM;
 523         }
 524         dreq = req->defer(req);
 525         if (dreq == NULL)
 526                 return -ENOMEM;
 527
 528         dreq->item = item;
 529
 530         spin_lock(&cache_defer_lock);
 531
 532         list_add(&dreq->recent, &cache_defer_list);
 533
 534         if (cache_defer_hash[hash].next == NULL)
 535                 INIT_LIST_HEAD(&cache_defer_hash[hash]);
 536         list_add(&dreq->hash, &cache_defer_hash[hash]);
 537
 538         /* it is in, now maybe clean up */
 539         discard = NULL;
 540         if (++cache_defer_cnt > DFR_MAX) {
 541                 discard = list_entry(cache_defer_list.prev,
 542                                      struct cache_deferred_req, recent);
 543                 list_del_init(&discard->recent);
 544                 list_del_init(&discard->hash);
 545                 cache_defer_cnt--;
 546         }
 547         spin_unlock(&cache_defer_lock);
 548
 549         if (discard)
 550                 /* there was one too many */
 551                 discard->revisit(discard, 1);
 552
 553         if (!test_bit(CACHE_PENDING, &item->flags)) {
 554                 /* must have just been validated... */
 555                 cache_revisit_request(item);
 556                 return -EAGAIN;
 557         }
 558         return 0;
 559 }
 560
 561 static void cache_revisit_request(struct cache_head *item)
 562 {
 563         struct cache_deferred_req *dreq;
 564         struct list_head pending;
 565
 566         struct list_head *lp;
 567         int hash = DFR_HASH(item);
 568
 569         INIT_LIST_HEAD(&pending);
 570         spin_lock(&cache_defer_lock);
 571
 572         lp = cache_defer_hash[hash].next;
 573         if (lp) {
 574                 while (lp != &cache_defer_hash[hash]) {
 575                         dreq = list_entry(lp, struct cache_deferred_req, hash);
 576                         lp = lp->next;
 577                         if (dreq->item == item) {
 578                                 list_del_init(&dreq->hash);
 579                                 list_move(&dreq->recent, &pending);
 580                                 cache_defer_cnt--;
 581                         }
 582                 }
 583         }
 584         spin_unlock(&cache_defer_lock);
 585
 586         while (!list_empty(&pending)) {
 587                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 588                 list_del_init(&dreq->recent);
 589                 dreq->revisit(dreq, 0);
 590         }
 591 }
 592
 593 void cache_clean_deferred(void *owner)
 594 {
 595         struct cache_deferred_req *dreq, *tmp;
 596         struct list_head pending;
 597
 598
 599         INIT_LIST_HEAD(&pending);
 600         spin_lock(&cache_defer_lock);
 601
 602         list_for_each_entry_safe(dreq, tmp, &cache_defer_list, recent) {
 603                 if (dreq->owner == owner) {
 604                         list_del_init(&dreq->hash);
 605                         list_move(&dreq->recent, &pending);
 606                         cache_defer_cnt--;
 607                 }
 608         }
 609         spin_unlock(&cache_defer_lock);
 610
 611         while (!list_empty(&pending)) {
 612                 dreq = list_entry(pending.next, struct cache_deferred_req, recent);
 613                 list_del_init(&dreq->recent);
 614                 dreq->revisit(dreq, 1);
 615         }
 616 }
 617
 618 /*
 619  * communicate with user-space
 620  *
 621  * We have a magic /proc file - /proc/sunrpc/<cachename>/channel.
 622  * On read, you get a full request, or block.
 623  * On write, an update request is processed.
 624  * Poll works if anything to read, and always allows write.
 625  *
 626  * Implemented by linked list of requests.  Each open file has
 627  * a ->private that also exists in this list.  New requests are added
 628  * to the end and may wakeup and preceding readers.
 629  * New readers are added to the head.  If, on read, an item is found with
 630  * CACHE_UPCALLING clear, we free it from the list.
 631  *
 632  */
 633
 634 static DEFINE_SPINLOCK(queue_lock);
 635 static DEFINE_MUTEX(queue_io_mutex);
 636
 637 struct cache_queue {
 638         struct list_head        list;
 639         int                     reader; /* if 0, then request */
 640 };
 641 struct cache_request {
 642         struct cache_queue      q;
 643         struct cache_head       *item;
 644         char                    * buf;
 645         int                     len;
 646         int                     readers;
 647 };
 648 struct cache_reader {
 649         struct cache_queue      q;
 650         int                     offset; /* if non-0, we have a refcnt on next request */
 651 };
 652
 653 static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
 654                           loff_t *ppos, struct cache_detail *cd)
 655 {
 656         struct cache_reader *rp = filp->private_data;
 657         struct cache_request *rq;
 658         struct inode *inode = filp->f_path.dentry->d_inode;
 659         int err;
 660
 661         if (count == 0)
 662                 return 0;
 663
 664         mutex_lock(&inode->i_mutex); /* protect against multiple concurrent
 665                               * readers on this file */
 666  again:
 667         spin_lock(&queue_lock);
 668         /* need to find next request */
 669         while (rp->q.list.next != &cd->queue &&
 670                list_entry(rp->q.list.next, struct cache_queue, list)
 671                ->reader) {
 672                 struct list_head *next = rp->q.list.next;
 673                 list_move(&rp->q.list, next);
 674         }
 675         if (rp->q.list.next == &cd->queue) {
 676                 spin_unlock(&queue_lock);
 677                 mutex_unlock(&inode->i_mutex);
 678                 BUG_ON(rp->offset);
 679                 return 0;
 680         }
 681         rq = container_of(rp->q.list.next, struct cache_request, q.list);
 682         BUG_ON(rq->q.reader);
 683         if (rp->offset == 0)
 684                 rq->readers++;
 685         spin_unlock(&queue_lock);
 686
 687         if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
 688                 err = -EAGAIN;
 689                 spin_lock(&queue_lock);
 690                 list_move(&rp->q.list, &rq->q.list);
 691                 spin_unlock(&queue_lock);
 692         } else {
 693                 if (rp->offset + count > rq->len)
 694                         count = rq->len - rp->offset;
 695                 err = -EFAULT;
 696                 if (copy_to_user(buf, rq->buf + rp->offset, count))
 697                         goto out;
 698                 rp->offset += count;
 699                 if (rp->offset >= rq->len) {
 700                         rp->offset = 0;
 701                         spin_lock(&queue_lock);
 702                         list_move(&rp->q.list, &rq->q.list);
 703                         spin_unlock(&queue_lock);
 704                 }
 705                 err = 0;
 706         }
 707  out:
 708         if (rp->offset == 0) {
 709                 /* need to release rq */
 710                 spin_lock(&queue_lock);
 711                 rq->readers--;
 712                 if (rq->readers == 0 &&
 713                     !test_bit(CACHE_PENDING, &rq->item->flags)) {
 714                         list_del(&rq->q.list);
 715                         spin_unlock(&queue_lock);
 716                         cache_put(rq->item, cd);
 717                         kfree(rq->buf);
 718                         kfree(rq);
 719                 } else
 720                         spin_unlock(&queue_lock);
 721         }
 722         if (err == -EAGAIN)
 723                 goto again;
 724         mutex_unlock(&inode->i_mutex);
 725         return err ? err :  count;
 726 }
 727
 728 static ssize_t cache_do_downcall(char *kaddr, const char __user *buf,
 729                                  size_t count, struct cache_detail *cd)
 730 {
 731         ssize_t ret;
 732
 733         if (copy_from_user(kaddr, buf, count))
 734                 return -EFAULT;
 735         kaddr[count] = '\0';
 736         ret = cd->cache_parse(cd, kaddr, count);
 737         if (!ret)
 738                 ret = count;
 739         return ret;
 740 }
 741
 742 static ssize_t cache_slow_downcall(const char __user *buf,
 743                                    size_t count, struct cache_detail *cd)
 744 {
 745         static char write_buf[8192]; /* protected by queue_io_mutex */
 746         ssize_t ret = -EINVAL;
 747
 748         if (count >= sizeof(write_buf))
 749                 goto out;
 750         mutex_lock(&queue_io_mutex);
 751         ret = cache_do_downcall(write_buf, buf, count, cd);
 752         mutex_unlock(&queue_io_mutex);
 753 out:
 754         return ret;
 755 }
 756
 757 static ssize_t cache_downcall(struct address_space *mapping,
 758                               const char __user *buf,
 759                               size_t count, struct cache_detail *cd)
 760 {
 761         struct page *page;
 762         char *kaddr;
 763         ssize_t ret = -ENOMEM;
 764
 765         if (count >= PAGE_CACHE_SIZE)
 766                 goto out_slow;
 767
 768         page = find_or_create_page(mapping, 0, GFP_KERNEL);
 769         if (!page)
 770                 goto out_slow;
 771
 772         kaddr = kmap(page);
 773         ret = cache_do_downcall(kaddr, buf, count, cd);
 774         kunmap(page);
 775         unlock_page(page);
 776         page_cache_release(page);
 777         return ret;
 778 out_slow:
 779         return cache_slow_downcall(buf, count, cd);
 780 }
 781
 782 static ssize_t cache_write(struct file *filp, const char __user *buf,
 783                            size_t count, loff_t *ppos,
 784                            struct cache_detail *cd)
 785 {
 786         struct address_space *mapping = filp->f_mapping;
 787         struct inode *inode = filp->f_path.dentry->d_inode;
 788         ssize_t ret = -EINVAL;
 789
 790         if (!cd->cache_parse)
 791                 goto out;
 792
 793         mutex_lock(&inode->i_mutex);
 794         ret = cache_downcall(mapping, buf, count, cd);
 795         mutex_unlock(&inode->i_mutex);
 796 out:
 797         return ret;
 798 }
 799
 800 static DECLARE_WAIT_QUEUE_HEAD(queue_wait);
 801
 802 static unsigned int cache_poll(struct file *filp, poll_table *wait,
 803                                struct cache_detail *cd)
 804 {
 805         unsigned int mask;
 806         struct cache_reader *rp = filp->private_data;
 807         struct cache_queue *cq;
 808
 809         poll_wait(filp, &queue_wait, wait);
 810
 811         /* alway allow write */
 812         mask = POLL_OUT | POLLWRNORM;
 813
 814         if (!rp)
 815                 return mask;
 816
 817         spin_lock(&queue_lock);
 818
 819         for (cq= &rp->q; &cq->list != &cd->queue;
 820              cq = list_entry(cq->list.next, struct cache_queue, list))
 821                 if (!cq->reader) {
 822                         mask |= POLLIN | POLLRDNORM;
 823                         break;
 824                 }
 825         spin_unlock(&queue_lock);
 826         return mask;
 827 }
 828
 829 static int cache_ioctl(struct inode *ino, struct file *filp,
 830                        unsigned int cmd, unsigned long arg,
 831                        struct cache_detail *cd)
 832 {
 833         int len = 0;
 834         struct cache_reader *rp = filp->private_data;
 835         struct cache_queue *cq;
 836
 837         if (cmd != FIONREAD || !rp)
 838                 return -EINVAL;
 839
 840         spin_lock(&queue_lock);
 841
 842         /* only find the length remaining in current request,
 843          * or the length of the next request
 844          */
 845         for (cq= &rp->q; &cq->list != &cd->queue;
 846              cq = list_entry(cq->list.next, struct cache_queue, list))
 847                 if (!cq->reader) {
 848                         struct cache_request *cr =
 849                                 container_of(cq, struct cache_request, q);
 850                         len = cr->len - rp->offset;
 851                         break;
 852                 }
 853         spin_unlock(&queue_lock);
 854
 855         return put_user(len, (int __user *)arg);
 856 }
 857
 858 static int cache_open(struct inode *inode, struct file *filp,
 859                       struct cache_detail *cd)
 860 {
 861         struct cache_reader *rp = NULL;
 862
 863         if (!cd || !try_module_get(cd->owner))
 864                 return -EACCES;
 865         nonseekable_open(inode, filp);
 866         if (filp->f_mode & FMODE_READ) {
 867                 rp = kmalloc(sizeof(*rp), GFP_KERNEL);
 868                 if (!rp)
 869                         return -ENOMEM;
 870                 rp->offset = 0;
 871                 rp->q.reader = 1;
 872                 atomic_inc(&cd->readers);
 873                 spin_lock(&queue_lock);
 874                 list_add(&rp->q.list, &cd->queue);
 875                 spin_unlock(&queue_lock);
 876         }
 877         filp->private_data = rp;
 878         return 0;
 879 }
 880
 881 static int cache_release(struct inode *inode, struct file *filp,
 882                          struct cache_detail *cd)
 883 {
 884         struct cache_reader *rp = filp->private_data;
 885
 886         if (rp) {
 887                 spin_lock(&queue_lock);
 888                 if (rp->offset) {
 889                         struct cache_queue *cq;
 890                         for (cq= &rp->q; &cq->list != &cd->queue;
 891                              cq = list_entry(cq->list.next, struct cache_queue, list))
 892                                 if (!cq->reader) {
 893                                         container_of(cq, struct cache_request, q)
 894                                                 ->readers--;
 895                                         break;
 896                                 }
 897                         rp->offset = 0;
 898                 }
 899                 list_del(&rp->q.list);
 900                 spin_unlock(&queue_lock);
 901
 902                 filp->private_data = NULL;
 903                 kfree(rp);
 904
 905                 cd->last_close = get_seconds();
 906                 atomic_dec(&cd->readers);
 907         }
 908         module_put(cd->owner);
 909         return 0;
 910 }
 911
 912
 913
 914 static void cache_dequeue(struct cache_detail *detail, struct cache_head *ch)
 915 {
 916         struct cache_queue *cq;
 917         spin_lock(&queue_lock);
 918         list_for_each_entry(cq, &detail->queue, list)
 919                 if (!cq->reader) {
 920                         struct cache_request *cr = container_of(cq, struct cache_request, q);
 921                         if (cr->item != ch)
 922                                 continue;
 923                         if (cr->readers != 0)
 924                                 continue;
 925                         list_del(&cr->q.list);
 926                         spin_unlock(&queue_lock);
 927                         cache_put(cr->item, detail);
 928                         kfree(cr->buf);
 929                         kfree(cr);
 930                         return;
 931                 }
 932         spin_unlock(&queue_lock);
 933 }
 934
 935 /*
 936  * Support routines for text-based upcalls.
 937  * Fields are separated by spaces.
 938  * Fields are either mangled to quote space tab newline slosh with slosh
 939  * or a hexified with a leading \x
 940  * Record is terminated with newline.
 941  *
 942  */
 943
 944 void qword_add(char **bpp, int *lp, char *str)
 945 {
 946         char *bp = *bpp;
 947         int len = *lp;
 948         char c;
 949
 950         if (len < 0) return;
 951
 952         while ((c=*str++) && len)
 953                 switch(c) {
 954                 case ' ':
 955                 case '\t':
 956                 case '\n':
 957                 case '\\':
 958                         if (len >= 4) {
 959                                 *bp++ = '\\';
 960                                 *bp++ = '0' + ((c & 0300)>>6);
 961                                 *bp++ = '0' + ((c & 0070)>>3);
 962                                 *bp++ = '0' + ((c & 0007)>>0);
 963                         }
 964                         len -= 4;
 965                         break;
 966                 default:
 967                         *bp++ = c;
 968                         len--;
 969                 }
 970         if (c || len <1) len = -1;
 971         else {
 972                 *bp++ = ' ';
 973                 len--;
 974         }
 975         *bpp = bp;
 976         *lp = len;
 977 }
 978 EXPORT_SYMBOL_GPL(qword_add);
 979
 980 void qword_addhex(char **bpp, int *lp, char *buf, int blen)
 981 {
 982         char *bp = *bpp;
 983         int len = *lp;
 984
 985         if (len < 0) return;
 986
 987         if (len > 2) {
 988                 *bp++ = '\\';
 989                 *bp++ = 'x';
 990                 len -= 2;
 991                 while (blen && len >= 2) {
 992                         unsigned char c = *buf++;
 993                         *bp++ = '0' + ((c&0xf0)>>4) + (c>=0xa0)*('a'-'9'-1);
 994                         *bp++ = '0' + (c&0x0f) + ((c&0x0f)>=0x0a)*('a'-'9'-1);
 995                         len -= 2;
 996                         blen--;
 997                 }
 998         }
 999         if (blen || len<1) len = -1;
1000         else {
1001                 *bp++ = ' ';
1002                 len--;
1003         }
1004         *bpp = bp;
1005         *lp = len;
1006 }
1007 EXPORT_SYMBOL_GPL(qword_addhex);
1008
1009 static void warn_no_listener(struct cache_detail *detail)
1010 {
1011         if (detail->last_warn != detail->last_close) {
1012                 detail->last_warn = detail->last_close;
1013                 if (detail->warn_no_listener)
1014                         detail->warn_no_listener(detail, detail->last_close != 0);
1015         }
1016 }
1017
1018 /*
1019  * register an upcall request to user-space and queue it up for read() by the
1020  * upcall daemon.
1021  *
1022  * Each request is at most one page long.
1023  */
1024 int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
1025                 void (*cache_request)(struct cache_detail *,
1026                                       struct cache_head *,
1027                                       char **,
1028                                       int *))
1029 {
1030
1031         char *buf;
1032         struct cache_request *crq;
1033         char *bp;
1034         int len;
1035
1036         if (atomic_read(&detail->readers) == 0 &&
1037             detail->last_close < get_seconds() - 30) {
1038                         warn_no_listener(detail);
1039                         return -EINVAL;
1040         }
1041
1042         buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
1043         if (!buf)
1044                 return -EAGAIN;
1045
1046         crq = kmalloc(sizeof (*crq), GFP_KERNEL);
1047         if (!crq) {
1048                 kfree(buf);
1049                 return -EAGAIN;
1050         }
1051
1052         bp = buf; len = PAGE_SIZE;
1053
1054         cache_request(detail, h, &bp, &len);
1055
1056         if (len < 0) {
1057                 kfree(buf);
1058                 kfree(crq);
1059                 return -EAGAIN;
1060         }
1061         crq->q.reader = 0;
1062         crq->item = cache_get(h);
1063         crq->buf = buf;
1064         crq->len = PAGE_SIZE - len;
1065         crq->readers = 0;
1066         spin_lock(&queue_lock);
1067         list_add_tail(&crq->q.list, &detail->queue);
1068         spin_unlock(&queue_lock);
1069         wake_up(&queue_wait);
1070         return 0;
1071 }
1072 EXPORT_SYMBOL_GPL(sunrpc_cache_pipe_upcall);
1073
1074 /*
1075  * parse a message from user-space and pass it
1076  * to an appropriate cache
1077  * Messages are, like requests, separated into fields by
1078  * spaces and dequotes as \xHEXSTRING or embedded \nnn octal
1079  *
1080  * Message is
1081  *   reply cachename expiry key ... content....
1082  *
1083  * key and content are both parsed by cache
1084  */
1085
1086 #define isodigit(c) (isdigit(c) && c <= '7')
1087 int qword_get(char **bpp, char *dest, int bufsize)
1088 {
1089         /* return bytes copied, or -1 on error */
1090         char *bp = *bpp;
1091         int len = 0;
1092
1093         while (*bp == ' ') bp++;
1094
1095         if (bp[0] == '\\' && bp[1] == 'x') {
1096                 /* HEX STRING */
1097                 bp += 2;
1098                 while (isxdigit(bp[0]) && isxdigit(bp[1]) && len < bufsize) {
1099                         int byte = isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1100                         bp++;
1101                         byte <<= 4;
1102                         byte |= isdigit(*bp) ? *bp-'0' : toupper(*bp)-'A'+10;
1103                         *dest++ = byte;
1104                         bp++;
1105                         len++;
1106                 }
1107         } else {
1108                 /* text with \nnn octal quoting */
1109                 while (*bp != ' ' && *bp != '\n' && *bp && len < bufsize-1) {
1110                         if (*bp == '\\' &&
1111                             isodigit(bp[1]) && (bp[1] <= '3') &&
1112                             isodigit(bp[2]) &&
1113                             isodigit(bp[3])) {
1114                                 int byte = (*++bp -'0');
1115                                 bp++;
1116                                 byte = (byte << 3) | (*bp++ - '0');
1117                                 byte = (byte << 3) | (*bp++ - '0');
1118                                 *dest++ = byte;
1119                                 len++;
1120                         } else {
1121                                 *dest++ = *bp++;
1122                                 len++;
1123                         }
1124                 }
1125         }
1126
1127         if (*bp != ' ' && *bp != '\n' && *bp != '\0')
1128                 return -1;
1129         while (*bp == ' ') bp++;
1130         *bpp = bp;
1131         *dest = '\0';
1132         return len;
1133 }
1134 EXPORT_SYMBOL_GPL(qword_get);
1135
1136
1137 /*
1138  * support /proc/sunrpc/cache/$CACHENAME/content
1139  * as a seqfile.
1140  * We call ->cache_show passing NULL for the item to
1141  * get a header, then pass each real item in the cache
1142  */
1143
1144 struct handle {
1145         struct cache_detail *cd;
1146 };
1147
1148 static void *c_start(struct seq_file *m, loff_t *pos)
1149         __acquires(cd->hash_lock)
1150 {
1151         loff_t n = *pos;
1152         unsigned hash, entry;
1153         struct cache_head *ch;
1154         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1155
1156
1157         read_lock(&cd->hash_lock);
1158         if (!n--)
1159                 return SEQ_START_TOKEN;
1160         hash = n >> 32;
1161         entry = n & ((1LL<<32) - 1);
1162
1163         for (ch=cd->hash_table[hash]; ch; ch=ch->next)
1164                 if (!entry--)
1165                         return ch;
1166         n &= ~((1LL<<32) - 1);
1167         do {
1168                 hash++;
1169                 n += 1LL<<32;
1170         } while(hash < cd->hash_size &&
1171                 cd->hash_table[hash]==NULL);
1172         if (hash >= cd->hash_size)
1173                 return NULL;
1174         *pos = n+1;
1175         return cd->hash_table[hash];
1176 }
1177
1178 static void *c_next(struct seq_file *m, void *p, loff_t *pos)
1179 {
1180         struct cache_head *ch = p;
1181         int hash = (*pos >> 32);
1182         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1183
1184         if (p == SEQ_START_TOKEN)
1185                 hash = 0;
1186         else if (ch->next == NULL) {
1187                 hash++;
1188                 *pos += 1LL<<32;
1189         } else {
1190                 ++*pos;
1191                 return ch->next;
1192         }
1193         *pos &= ~((1LL<<32) - 1);
1194         while (hash < cd->hash_size &&
1195                cd->hash_table[hash] == NULL) {
1196                 hash++;
1197                 *pos += 1LL<<32;
1198         }
1199         if (hash >= cd->hash_size)
1200                 return NULL;
1201         ++*pos;
1202         return cd->hash_table[hash];
1203 }
1204
1205 static void c_stop(struct seq_file *m, void *p)
1206         __releases(cd->hash_lock)
1207 {
1208         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1209         read_unlock(&cd->hash_lock);
1210 }
1211
1212 static int c_show(struct seq_file *m, void *p)
1213 {
1214         struct cache_head *cp = p;
1215         struct cache_detail *cd = ((struct handle*)m->private)->cd;
1216
1217         if (p == SEQ_START_TOKEN)
1218                 return cd->cache_show(m, cd, NULL);
1219
1220         ifdebug(CACHE)
1221                 seq_printf(m, "# expiry=%ld refcnt=%d flags=%lx\n",
1222                            cp->expiry_time, atomic_read(&cp->ref.refcount), cp->flags);
1223         cache_get(cp);
1224         if (cache_check(cd, cp, NULL))
1225                 /* cache_check does a cache_put on failure */
1226                 seq_printf(m, "# ");
1227         else
1228                 cache_put(cp, cd);
1229
1230         return cd->cache_show(m, cd, cp);
1231 }
1232
1233 static const struct seq_operations cache_content_op = {
1234         .start  = c_start,
1235         .next   = c_next,
1236         .stop   = c_stop,
1237         .show   = c_show,
1238 };
1239
1240 static int content_open(struct inode *inode, struct file *file,
1241                         struct cache_detail *cd)
1242 {
1243         struct handle *han;
1244
1245         if (!cd || !try_module_get(cd->owner))
1246                 return -EACCES;
1247         han = __seq_open_private(file, &cache_content_op, sizeof(*han));
1248         if (han == NULL) {
1249                 module_put(cd->owner);
1250                 return -ENOMEM;
1251         }
1252
1253         han->cd = cd;
1254         return 0;
1255 }
1256
1257 static int content_release(struct inode *inode, struct file *file,
1258                 struct cache_detail *cd)
1259 {
1260         int ret = seq_release_private(inode, file);
1261         module_put(cd->owner);
1262         return ret;
1263 }
1264
1265 static int open_flush(struct inode *inode, struct file *file,
1266                         struct cache_detail *cd)
1267 {
1268         if (!cd || !try_module_get(cd->owner))
1269                 return -EACCES;
1270         return nonseekable_open(inode, file);
1271 }
1272
1273 static int release_flush(struct inode *inode, struct file *file,
1274                         struct cache_detail *cd)
1275 {
1276         module_put(cd->owner);
1277         return 0;
1278 }
1279
1280 static ssize_t read_flush(struct file *file, char __user *buf,
1281                           size_t count, loff_t *ppos,
1282                           struct cache_detail *cd)
1283 {
1284         char tbuf[20];
1285         unsigned long p = *ppos;
1286         size_t len;
1287
1288         sprintf(tbuf, "%lu\n", cd->flush_time);
1289         len = strlen(tbuf);
1290         if (p >= len)
1291                 return 0;
1292         len -= p;
1293         if (len > count)
1294                 len = count;
1295         if (copy_to_user(buf, (void*)(tbuf+p), len))
1296                 return -EFAULT;
1297         *ppos += len;
1298         return len;
1299 }
1300
1301 static ssize_t write_flush(struct file *file, const char __user *buf,
1302                            size_t count, loff_t *ppos,
1303                            struct cache_detail *cd)
1304 {
1305         char tbuf[20];
1306         char *ep;
1307         long flushtime;
1308         if (*ppos || count > sizeof(tbuf)-1)
1309                 return -EINVAL;
1310         if (copy_from_user(tbuf, buf, count))
1311                 return -EFAULT;
1312         tbuf[count] = 0;
1313         flushtime = simple_strtoul(tbuf, &ep, 0);
1314         if (*ep && *ep != '\n')
1315                 return -EINVAL;
1316
1317         cd->flush_time = flushtime;
1318         cd->nextcheck = get_seconds();
1319         cache_flush();
1320
1321         *ppos += count;
1322         return count;
1323 }
1324
1325 static ssize_t cache_read_procfs(struct file *filp, char __user *buf,
1326                                  size_t count, loff_t *ppos)
1327 {
1328         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1329
1330         return cache_read(filp, buf, count, ppos, cd);
1331 }
1332
1333 static ssize_t cache_write_procfs(struct file *filp, const char __user *buf,
1334                                   size_t count, loff_t *ppos)
1335 {
1336         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1337
1338         return cache_write(filp, buf, count, ppos, cd);
1339 }
1340
1341 static unsigned int cache_poll_procfs(struct file *filp, poll_table *wait)
1342 {
1343         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1344
1345         return cache_poll(filp, wait, cd);
1346 }
1347
1348 static long cache_ioctl_procfs(struct file *filp,
1349                                unsigned int cmd, unsigned long arg)
1350 {
1351         long ret;
1352         struct inode *inode = filp->f_path.dentry->d_inode;
1353         struct cache_detail *cd = PDE(inode)->data;
1354
1355         lock_kernel();
1356         ret = cache_ioctl(inode, filp, cmd, arg, cd);
1357         unlock_kernel();
1358
1359         return ret;
1360 }
1361
1362 static int cache_open_procfs(struct inode *inode, struct file *filp)
1363 {
1364         struct cache_detail *cd = PDE(inode)->data;
1365
1366         return cache_open(inode, filp, cd);
1367 }
1368
1369 static int cache_release_procfs(struct inode *inode, struct file *filp)
1370 {
1371         struct cache_detail *cd = PDE(inode)->data;
1372
1373         return cache_release(inode, filp, cd);
1374 }
1375
1376 static const struct file_operations cache_file_operations_procfs = {
1377         .owner          = THIS_MODULE,
1378         .llseek         = no_llseek,
1379         .read           = cache_read_procfs,
1380         .write          = cache_write_procfs,
1381         .poll           = cache_poll_procfs,
1382         .unlocked_ioctl = cache_ioctl_procfs, /* for FIONREAD */
1383         .open           = cache_open_procfs,
1384         .release        = cache_release_procfs,
1385 };
1386
1387 static int content_open_procfs(struct inode *inode, struct file *filp)
1388 {
1389         struct cache_detail *cd = PDE(inode)->data;
1390
1391         return content_open(inode, filp, cd);
1392 }
1393
1394 static int content_release_procfs(struct inode *inode, struct file *filp)
1395 {
1396         struct cache_detail *cd = PDE(inode)->data;
1397
1398         return content_release(inode, filp, cd);
1399 }
1400
1401 static const struct file_operations content_file_operations_procfs = {
1402         .open           = content_open_procfs,
1403         .read           = seq_read,
1404         .llseek         = seq_lseek,
1405         .release        = content_release_procfs,
1406 };
1407
1408 static int open_flush_procfs(struct inode *inode, struct file *filp)
1409 {
1410         struct cache_detail *cd = PDE(inode)->data;
1411
1412         return open_flush(inode, filp, cd);
1413 }
1414
1415 static int release_flush_procfs(struct inode *inode, struct file *filp)
1416 {
1417         struct cache_detail *cd = PDE(inode)->data;
1418
1419         return release_flush(inode, filp, cd);
1420 }
1421
1422 static ssize_t read_flush_procfs(struct file *filp, char __user *buf,
1423                             size_t count, loff_t *ppos)
1424 {
1425         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1426
1427         return read_flush(filp, buf, count, ppos, cd);
1428 }
1429
1430 static ssize_t write_flush_procfs(struct file *filp,
1431                                   const char __user *buf,
1432                                   size_t count, loff_t *ppos)
1433 {
1434         struct cache_detail *cd = PDE(filp->f_path.dentry->d_inode)->data;
1435
1436         return write_flush(filp, buf, count, ppos, cd);
1437 }
1438
1439 static const struct file_operations cache_flush_operations_procfs = {
1440         .open           = open_flush_procfs,
1441         .read           = read_flush_procfs,
1442         .write          = write_flush_procfs,
1443         .release        = release_flush_procfs,
1444 };
1445
1446 static void remove_cache_proc_entries(struct cache_detail *cd)
1447 {
1448         if (cd->u.procfs.proc_ent == NULL)
1449                 return;
1450         if (cd->u.procfs.flush_ent)
1451                 remove_proc_entry("flush", cd->u.procfs.proc_ent);
1452         if (cd->u.procfs.channel_ent)
1453                 remove_proc_entry("channel", cd->u.procfs.proc_ent);
1454         if (cd->u.procfs.content_ent)
1455                 remove_proc_entry("content", cd->u.procfs.proc_ent);
1456         cd->u.procfs.proc_ent = NULL;
1457         remove_proc_entry(cd->name, proc_net_rpc);
1458 }
1459
1460 #ifdef CONFIG_PROC_FS
1461 static int create_cache_proc_entries(struct cache_detail *cd)
1462 {
1463         struct proc_dir_entry *p;
1464
1465         cd->u.procfs.proc_ent = proc_mkdir(cd->name, proc_net_rpc);
1466         if (cd->u.procfs.proc_ent == NULL)
1467                 goto out_nomem;
1468         cd->u.procfs.channel_ent = NULL;
1469         cd->u.procfs.content_ent = NULL;
1470
1471         p = proc_create_data("flush", S_IFREG|S_IRUSR|S_IWUSR,
1472                              cd->u.procfs.proc_ent,
1473                              &cache_flush_operations_procfs, cd);
1474         cd->u.procfs.flush_ent = p;
1475         if (p == NULL)
1476                 goto out_nomem;
1477
1478         if (cd->cache_upcall || cd->cache_parse) {
1479                 p = proc_create_data("channel", S_IFREG|S_IRUSR|S_IWUSR,
1480                                      cd->u.procfs.proc_ent,
1481                                      &cache_file_operations_procfs, cd);
1482                 cd->u.procfs.channel_ent = p;
1483                 if (p == NULL)
1484                         goto out_nomem;
1485         }
1486         if (cd->cache_show) {
1487                 p = proc_create_data("content", S_IFREG|S_IRUSR|S_IWUSR,
1488                                 cd->u.procfs.proc_ent,
1489                                 &content_file_operations_procfs, cd);
1490                 cd->u.procfs.content_ent = p;
1491                 if (p == NULL)
1492                         goto out_nomem;
1493         }
1494         return 0;
1495 out_nomem:
1496         remove_cache_proc_entries(cd);
1497         return -ENOMEM;
1498 }
1499 #else /* CONFIG_PROC_FS */
1500 static int create_cache_proc_entries(struct cache_detail *cd)
1501 {
1502         return 0;
1503 }
1504 #endif
1505
1506 int cache_register(struct cache_detail *cd)
1507 {
1508         int ret;
1509
1510         sunrpc_init_cache_detail(cd);
1511         ret = create_cache_proc_entries(cd);
1512         if (ret)
1513                 sunrpc_destroy_cache_detail(cd);
1514         return ret;
1515 }
1516 EXPORT_SYMBOL_GPL(cache_register);
1517
1518 void cache_unregister(struct cache_detail *cd)
1519 {
1520         remove_cache_proc_entries(cd);
1521         sunrpc_destroy_cache_detail(cd);
1522 }
1523 EXPORT_SYMBOL_GPL(cache_unregister);
1524
1525 static ssize_t cache_read_pipefs(struct file *filp, char __user *buf,
1526                                  size_t count, loff_t *ppos)
1527 {
1528         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1529
1530         return cache_read(filp, buf, count, ppos, cd);
1531 }
1532
1533 static ssize_t cache_write_pipefs(struct file *filp, const char __user *buf,
1534                                   size_t count, loff_t *ppos)
1535 {
1536         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1537
1538         return cache_write(filp, buf, count, ppos, cd);
1539 }
1540
1541 static unsigned int cache_poll_pipefs(struct file *filp, poll_table *wait)
1542 {
1543         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1544
1545         return cache_poll(filp, wait, cd);
1546 }
1547
1548 static int cache_ioctl_pipefs(struct inode *inode, struct file *filp,
1549                               unsigned int cmd, unsigned long arg)
1550 {
1551         struct cache_detail *cd = RPC_I(inode)->private;
1552
1553         return cache_ioctl(inode, filp, cmd, arg, cd);
1554 }
1555
1556 static int cache_open_pipefs(struct inode *inode, struct file *filp)
1557 {
1558         struct cache_detail *cd = RPC_I(inode)->private;
1559
1560         return cache_open(inode, filp, cd);
1561 }
1562
1563 static int cache_release_pipefs(struct inode *inode, struct file *filp)
1564 {
1565         struct cache_detail *cd = RPC_I(inode)->private;
1566
1567         return cache_release(inode, filp, cd);
1568 }
1569
1570 const struct file_operations cache_file_operations_pipefs = {
1571         .owner          = THIS_MODULE,
1572         .llseek         = no_llseek,
1573         .read           = cache_read_pipefs,
1574         .write          = cache_write_pipefs,
1575         .poll           = cache_poll_pipefs,
1576         .ioctl          = cache_ioctl_pipefs, /* for FIONREAD */
1577         .open           = cache_open_pipefs,
1578         .release        = cache_release_pipefs,
1579 };
1580
1581 static int content_open_pipefs(struct inode *inode, struct file *filp)
1582 {
1583         struct cache_detail *cd = RPC_I(inode)->private;
1584
1585         return content_open(inode, filp, cd);
1586 }
1587
1588 static int content_release_pipefs(struct inode *inode, struct file *filp)
1589 {
1590         struct cache_detail *cd = RPC_I(inode)->private;
1591
1592         return content_release(inode, filp, cd);
1593 }
1594
1595 const struct file_operations content_file_operations_pipefs = {
1596         .open           = content_open_pipefs,
1597         .read           = seq_read,
1598         .llseek         = seq_lseek,
1599         .release        = content_release_pipefs,
1600 };
1601
1602 static int open_flush_pipefs(struct inode *inode, struct file *filp)
1603 {
1604         struct cache_detail *cd = RPC_I(inode)->private;
1605
1606         return open_flush(inode, filp, cd);
1607 }
1608
1609 static int release_flush_pipefs(struct inode *inode, struct file *filp)
1610 {
1611         struct cache_detail *cd = RPC_I(inode)->private;
1612
1613         return release_flush(inode, filp, cd);
1614 }
1615
1616 static ssize_t read_flush_pipefs(struct file *filp, char __user *buf,
1617                             size_t count, loff_t *ppos)
1618 {
1619         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1620
1621         return read_flush(filp, buf, count, ppos, cd);
1622 }
1623
1624 static ssize_t write_flush_pipefs(struct file *filp,
1625                                   const char __user *buf,
1626                                   size_t count, loff_t *ppos)
1627 {
1628         struct cache_detail *cd = RPC_I(filp->f_path.dentry->d_inode)->private;
1629
1630         return write_flush(filp, buf, count, ppos, cd);
1631 }
1632
1633 const struct file_operations cache_flush_operations_pipefs = {
1634         .open           = open_flush_pipefs,
1635         .read           = read_flush_pipefs,
1636         .write          = write_flush_pipefs,
1637         .release        = release_flush_pipefs,
1638 };
1639
1640 int sunrpc_cache_register_pipefs(struct dentry *parent,
1641                                  const char *name, mode_t umode,
1642                                  struct cache_detail *cd)
1643 {
1644         struct qstr q;
1645         struct dentry *dir;
1646         int ret = 0;
1647
1648         sunrpc_init_cache_detail(cd);
1649         q.name = name;
1650         q.len = strlen(name);
1651         q.hash = full_name_hash(q.name, q.len);
1652         dir = rpc_create_cache_dir(parent, &q, umode, cd);
1653         if (!IS_ERR(dir))
1654                 cd->u.pipefs.dir = dir;
1655         else {
1656                 sunrpc_destroy_cache_detail(cd);
1657                 ret = PTR_ERR(dir);
1658         }
1659         return ret;
1660 }
1661 EXPORT_SYMBOL_GPL(sunrpc_cache_register_pipefs);
1662
1663 void sunrpc_cache_unregister_pipefs(struct cache_detail *cd)
1664 {
1665         rpc_remove_cache_dir(cd->u.pipefs.dir);
1666         cd->u.pipefs.dir = NULL;
1667         sunrpc_destroy_cache_detail(cd);
1668 }
1669 EXPORT_SYMBOL_GPL(sunrpc_cache_unregister_pipefs);
1670