[PATCH] knfsd: Prepare knfsd for support of rsize/wsize of up to 1MB, over TCP
[linux-2.6] / fs / afs / cell.c
1 /* cell.c: AFS cell and server record management
2  *
3  * Copyright (C) 2002 Red Hat, Inc. All Rights Reserved.
4  * Written by David Howells (dhowells@redhat.com)
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  */
11
12 #include <linux/module.h>
13 #include <linux/sched.h>
14 #include <linux/slab.h>
15 #include <rxrpc/peer.h>
16 #include <rxrpc/connection.h>
17 #include "volume.h"
18 #include "cell.h"
19 #include "server.h"
20 #include "transport.h"
21 #include "vlclient.h"
22 #include "kafstimod.h"
23 #include "super.h"
24 #include "internal.h"
25
26 DECLARE_RWSEM(afs_proc_cells_sem);
27 LIST_HEAD(afs_proc_cells);
28
29 static struct list_head afs_cells = LIST_HEAD_INIT(afs_cells);
30 static DEFINE_RWLOCK(afs_cells_lock);
31 static DECLARE_RWSEM(afs_cells_sem); /* add/remove serialisation */
32 static struct afs_cell *afs_cell_root;
33
34 #ifdef AFS_CACHING_SUPPORT
35 static cachefs_match_val_t afs_cell_cache_match(void *target,
36                                                 const void *entry);
37 static void afs_cell_cache_update(void *source, void *entry);
38
39 struct cachefs_index_def afs_cache_cell_index_def = {
40         .name                   = "cell_ix",
41         .data_size              = sizeof(struct afs_cache_cell),
42         .keys[0]                = { CACHEFS_INDEX_KEYS_ASCIIZ, 64 },
43         .match                  = afs_cell_cache_match,
44         .update                 = afs_cell_cache_update,
45 };
46 #endif
47
48 /*****************************************************************************/
49 /*
50  * create a cell record
51  * - "name" is the name of the cell
52  * - "vllist" is a colon separated list of IP addresses in "a.b.c.d" format
53  */
54 int afs_cell_create(const char *name, char *vllist, struct afs_cell **_cell)
55 {
56         struct afs_cell *cell;
57         char *next;
58         int ret;
59
60         _enter("%s", name);
61
62         BUG_ON(!name); /* TODO: want to look up "this cell" in the cache */
63
64         /* allocate and initialise a cell record */
65         cell = kmalloc(sizeof(struct afs_cell) + strlen(name) + 1, GFP_KERNEL);
66         if (!cell) {
67                 _leave(" = -ENOMEM");
68                 return -ENOMEM;
69         }
70
71         down_write(&afs_cells_sem);
72
73         memset(cell, 0, sizeof(struct afs_cell));
74         atomic_set(&cell->usage, 0);
75
76         INIT_LIST_HEAD(&cell->link);
77
78         rwlock_init(&cell->sv_lock);
79         INIT_LIST_HEAD(&cell->sv_list);
80         INIT_LIST_HEAD(&cell->sv_graveyard);
81         spin_lock_init(&cell->sv_gylock);
82
83         init_rwsem(&cell->vl_sem);
84         INIT_LIST_HEAD(&cell->vl_list);
85         INIT_LIST_HEAD(&cell->vl_graveyard);
86         spin_lock_init(&cell->vl_gylock);
87
88         strcpy(cell->name,name);
89
90         /* fill in the VL server list from the rest of the string */
91         ret = -EINVAL;
92         do {
93                 unsigned a, b, c, d;
94
95                 next = strchr(vllist, ':');
96                 if (next)
97                         *next++ = 0;
98
99                 if (sscanf(vllist, "%u.%u.%u.%u", &a, &b, &c, &d) != 4)
100                         goto badaddr;
101
102                 if (a > 255 || b > 255 || c > 255 || d > 255)
103                         goto badaddr;
104
105                 cell->vl_addrs[cell->vl_naddrs++].s_addr =
106                         htonl((a << 24) | (b << 16) | (c << 8) | d);
107
108                 if (cell->vl_naddrs >= AFS_CELL_MAX_ADDRS)
109                         break;
110
111         } while(vllist = next, vllist);
112
113         /* add a proc dir for this cell */
114         ret = afs_proc_cell_setup(cell);
115         if (ret < 0)
116                 goto error;
117
118 #ifdef AFS_CACHING_SUPPORT
119         /* put it up for caching */
120         cachefs_acquire_cookie(afs_cache_netfs.primary_index,
121                                &afs_vlocation_cache_index_def,
122                                cell,
123                                &cell->cache);
124 #endif
125
126         /* add to the cell lists */
127         write_lock(&afs_cells_lock);
128         list_add_tail(&cell->link, &afs_cells);
129         write_unlock(&afs_cells_lock);
130
131         down_write(&afs_proc_cells_sem);
132         list_add_tail(&cell->proc_link, &afs_proc_cells);
133         up_write(&afs_proc_cells_sem);
134
135         *_cell = cell;
136         up_write(&afs_cells_sem);
137
138         _leave(" = 0 (%p)", cell);
139         return 0;
140
141  badaddr:
142         printk(KERN_ERR "kAFS: bad VL server IP address: '%s'\n", vllist);
143  error:
144         up_write(&afs_cells_sem);
145         kfree(cell);
146         _leave(" = %d", ret);
147         return ret;
148 } /* end afs_cell_create() */
149
150 /*****************************************************************************/
151 /*
152  * initialise the cell database from module parameters
153  */
154 int afs_cell_init(char *rootcell)
155 {
156         struct afs_cell *old_root, *new_root;
157         char *cp;
158         int ret;
159
160         _enter("");
161
162         if (!rootcell) {
163                 /* module is loaded with no parameters, or built statically.
164                  * - in the future we might initialize cell DB here.
165                  */
166                 _leave(" = 0 (but no root)");
167                 return 0;
168         }
169
170         cp = strchr(rootcell, ':');
171         if (!cp) {
172                 printk(KERN_ERR "kAFS: no VL server IP addresses specified\n");
173                 _leave(" = %d (no colon)", -EINVAL);
174                 return -EINVAL;
175         }
176
177         /* allocate a cell record for the root cell */
178         *cp++ = 0;
179         ret = afs_cell_create(rootcell, cp, &new_root);
180         if (ret < 0) {
181                 _leave(" = %d", ret);
182                 return ret;
183         }
184
185         /* as afs_put_cell() takes locks by itself, we have to do
186          * a little gymnastics to be race-free.
187          */
188         afs_get_cell(new_root);
189
190         write_lock(&afs_cells_lock);
191         while (afs_cell_root) {
192                 old_root = afs_cell_root;
193                 afs_cell_root = NULL;
194                 write_unlock(&afs_cells_lock);
195                 afs_put_cell(old_root);
196                 write_lock(&afs_cells_lock);
197         }
198         afs_cell_root = new_root;
199         write_unlock(&afs_cells_lock);
200
201         _leave(" = %d", ret);
202         return ret;
203
204 } /* end afs_cell_init() */
205
206 /*****************************************************************************/
207 /*
208  * lookup a cell record
209  */
210 int afs_cell_lookup(const char *name, unsigned namesz, struct afs_cell **_cell)
211 {
212         struct afs_cell *cell;
213         int ret;
214
215         _enter("\"%*.*s\",", namesz, namesz, name ? name : "");
216
217         *_cell = NULL;
218
219         if (name) {
220                 /* if the cell was named, look for it in the cell record list */
221                 ret = -ENOENT;
222                 cell = NULL;
223                 read_lock(&afs_cells_lock);
224
225                 list_for_each_entry(cell, &afs_cells, link) {
226                         if (strncmp(cell->name, name, namesz) == 0) {
227                                 afs_get_cell(cell);
228                                 goto found;
229                         }
230                 }
231                 cell = NULL;
232         found:
233
234                 read_unlock(&afs_cells_lock);
235
236                 if (cell)
237                         ret = 0;
238         }
239         else {
240                 read_lock(&afs_cells_lock);
241
242                 cell = afs_cell_root;
243                 if (!cell) {
244                         /* this should not happen unless user tries to mount
245                          * when root cell is not set. Return an impossibly
246                          * bizzare errno to alert the user. Things like
247                          * ENOENT might be "more appropriate" but they happen
248                          * for other reasons.
249                          */
250                         ret = -EDESTADDRREQ;
251                 }
252                 else {
253                         afs_get_cell(cell);
254                         ret = 0;
255                 }
256
257                 read_unlock(&afs_cells_lock);
258         }
259
260         *_cell = cell;
261         _leave(" = %d (%p)", ret, cell);
262         return ret;
263
264 } /* end afs_cell_lookup() */
265
266 /*****************************************************************************/
267 /*
268  * try and get a cell record
269  */
270 struct afs_cell *afs_get_cell_maybe(struct afs_cell **_cell)
271 {
272         struct afs_cell *cell;
273
274         write_lock(&afs_cells_lock);
275
276         cell = *_cell;
277         if (cell && !list_empty(&cell->link))
278                 afs_get_cell(cell);
279         else
280                 cell = NULL;
281
282         write_unlock(&afs_cells_lock);
283
284         return cell;
285 } /* end afs_get_cell_maybe() */
286
287 /*****************************************************************************/
288 /*
289  * destroy a cell record
290  */
291 void afs_put_cell(struct afs_cell *cell)
292 {
293         if (!cell)
294                 return;
295
296         _enter("%p{%d,%s}", cell, atomic_read(&cell->usage), cell->name);
297
298         /* sanity check */
299         BUG_ON(atomic_read(&cell->usage) <= 0);
300
301         /* to prevent a race, the decrement and the dequeue must be effectively
302          * atomic */
303         write_lock(&afs_cells_lock);
304
305         if (likely(!atomic_dec_and_test(&cell->usage))) {
306                 write_unlock(&afs_cells_lock);
307                 _leave("");
308                 return;
309         }
310
311         write_unlock(&afs_cells_lock);
312
313         BUG_ON(!list_empty(&cell->sv_list));
314         BUG_ON(!list_empty(&cell->sv_graveyard));
315         BUG_ON(!list_empty(&cell->vl_list));
316         BUG_ON(!list_empty(&cell->vl_graveyard));
317
318         _leave(" [unused]");
319 } /* end afs_put_cell() */
320
321 /*****************************************************************************/
322 /*
323  * destroy a cell record
324  */
325 static void afs_cell_destroy(struct afs_cell *cell)
326 {
327         _enter("%p{%d,%s}", cell, atomic_read(&cell->usage), cell->name);
328
329         /* to prevent a race, the decrement and the dequeue must be effectively
330          * atomic */
331         write_lock(&afs_cells_lock);
332
333         /* sanity check */
334         BUG_ON(atomic_read(&cell->usage) != 0);
335
336         list_del_init(&cell->link);
337
338         write_unlock(&afs_cells_lock);
339
340         down_write(&afs_cells_sem);
341
342         afs_proc_cell_remove(cell);
343
344         down_write(&afs_proc_cells_sem);
345         list_del_init(&cell->proc_link);
346         up_write(&afs_proc_cells_sem);
347
348 #ifdef AFS_CACHING_SUPPORT
349         cachefs_relinquish_cookie(cell->cache, 0);
350 #endif
351
352         up_write(&afs_cells_sem);
353
354         BUG_ON(!list_empty(&cell->sv_list));
355         BUG_ON(!list_empty(&cell->sv_graveyard));
356         BUG_ON(!list_empty(&cell->vl_list));
357         BUG_ON(!list_empty(&cell->vl_graveyard));
358
359         /* finish cleaning up the cell */
360         kfree(cell);
361
362         _leave(" [destroyed]");
363 } /* end afs_cell_destroy() */
364
365 /*****************************************************************************/
366 /*
367  * lookup the server record corresponding to an Rx RPC peer
368  */
369 int afs_server_find_by_peer(const struct rxrpc_peer *peer,
370                             struct afs_server **_server)
371 {
372         struct afs_server *server;
373         struct afs_cell *cell;
374
375         _enter("%p{a=%08x},", peer, ntohl(peer->addr.s_addr));
376
377         /* search the cell list */
378         read_lock(&afs_cells_lock);
379
380         list_for_each_entry(cell, &afs_cells, link) {
381
382                 _debug("? cell %s",cell->name);
383
384                 write_lock(&cell->sv_lock);
385
386                 /* check the active list */
387                 list_for_each_entry(server, &cell->sv_list, link) {
388                         _debug("?? server %08x", ntohl(server->addr.s_addr));
389
390                         if (memcmp(&server->addr, &peer->addr,
391                                    sizeof(struct in_addr)) == 0)
392                                 goto found_server;
393                 }
394
395                 /* check the inactive list */
396                 spin_lock(&cell->sv_gylock);
397                 list_for_each_entry(server, &cell->sv_graveyard, link) {
398                         _debug("?? dead server %08x",
399                                ntohl(server->addr.s_addr));
400
401                         if (memcmp(&server->addr, &peer->addr,
402                                    sizeof(struct in_addr)) == 0)
403                                 goto found_dead_server;
404                 }
405                 spin_unlock(&cell->sv_gylock);
406
407                 write_unlock(&cell->sv_lock);
408         }
409         read_unlock(&afs_cells_lock);
410
411         _leave(" = -ENOENT");
412         return -ENOENT;
413
414         /* we found it in the graveyard - resurrect it */
415  found_dead_server:
416         list_move_tail(&server->link, &cell->sv_list);
417         afs_get_server(server);
418         afs_kafstimod_del_timer(&server->timeout);
419         spin_unlock(&cell->sv_gylock);
420         goto success;
421
422         /* we found it - increment its ref count and return it */
423  found_server:
424         afs_get_server(server);
425
426  success:
427         write_unlock(&cell->sv_lock);
428         read_unlock(&afs_cells_lock);
429
430         *_server = server;
431         _leave(" = 0 (s=%p c=%p)", server, cell);
432         return 0;
433
434 } /* end afs_server_find_by_peer() */
435
436 /*****************************************************************************/
437 /*
438  * purge in-memory cell database on module unload or afs_init() failure
439  * - the timeout daemon is stopped before calling this
440  */
441 void afs_cell_purge(void)
442 {
443         struct afs_vlocation *vlocation;
444         struct afs_cell *cell;
445
446         _enter("");
447
448         afs_put_cell(afs_cell_root);
449
450         while (!list_empty(&afs_cells)) {
451                 cell = NULL;
452
453                 /* remove the next cell from the front of the list */
454                 write_lock(&afs_cells_lock);
455
456                 if (!list_empty(&afs_cells)) {
457                         cell = list_entry(afs_cells.next,
458                                           struct afs_cell, link);
459                         list_del_init(&cell->link);
460                 }
461
462                 write_unlock(&afs_cells_lock);
463
464                 if (cell) {
465                         _debug("PURGING CELL %s (%d)",
466                                cell->name, atomic_read(&cell->usage));
467
468                         BUG_ON(!list_empty(&cell->sv_list));
469                         BUG_ON(!list_empty(&cell->vl_list));
470
471                         /* purge the cell's VL graveyard list */
472                         _debug(" - clearing VL graveyard");
473
474                         spin_lock(&cell->vl_gylock);
475
476                         while (!list_empty(&cell->vl_graveyard)) {
477                                 vlocation = list_entry(cell->vl_graveyard.next,
478                                                        struct afs_vlocation,
479                                                        link);
480                                 list_del_init(&vlocation->link);
481
482                                 afs_kafstimod_del_timer(&vlocation->timeout);
483
484                                 spin_unlock(&cell->vl_gylock);
485
486                                 afs_vlocation_do_timeout(vlocation);
487                                 /* TODO: race if move to use krxtimod instead
488                                  * of kafstimod */
489
490                                 spin_lock(&cell->vl_gylock);
491                         }
492
493                         spin_unlock(&cell->vl_gylock);
494
495                         /* purge the cell's server graveyard list */
496                         _debug(" - clearing server graveyard");
497
498                         spin_lock(&cell->sv_gylock);
499
500                         while (!list_empty(&cell->sv_graveyard)) {
501                                 struct afs_server *server;
502
503                                 server = list_entry(cell->sv_graveyard.next,
504                                                     struct afs_server, link);
505                                 list_del_init(&server->link);
506
507                                 afs_kafstimod_del_timer(&server->timeout);
508
509                                 spin_unlock(&cell->sv_gylock);
510
511                                 afs_server_do_timeout(server);
512
513                                 spin_lock(&cell->sv_gylock);
514                         }
515
516                         spin_unlock(&cell->sv_gylock);
517
518                         /* now the cell should be left with no references */
519                         afs_cell_destroy(cell);
520                 }
521         }
522
523         _leave("");
524 } /* end afs_cell_purge() */
525
526 /*****************************************************************************/
527 /*
528  * match a cell record obtained from the cache
529  */
530 #ifdef AFS_CACHING_SUPPORT
531 static cachefs_match_val_t afs_cell_cache_match(void *target,
532                                                 const void *entry)
533 {
534         const struct afs_cache_cell *ccell = entry;
535         struct afs_cell *cell = target;
536
537         _enter("{%s},{%s}", ccell->name, cell->name);
538
539         if (strncmp(ccell->name, cell->name, sizeof(ccell->name)) == 0) {
540                 _leave(" = SUCCESS");
541                 return CACHEFS_MATCH_SUCCESS;
542         }
543
544         _leave(" = FAILED");
545         return CACHEFS_MATCH_FAILED;
546 } /* end afs_cell_cache_match() */
547 #endif
548
549 /*****************************************************************************/
550 /*
551  * update a cell record in the cache
552  */
553 #ifdef AFS_CACHING_SUPPORT
554 static void afs_cell_cache_update(void *source, void *entry)
555 {
556         struct afs_cache_cell *ccell = entry;
557         struct afs_cell *cell = source;
558
559         _enter("%p,%p", source, entry);
560
561         strncpy(ccell->name, cell->name, sizeof(ccell->name));
562
563         memcpy(ccell->vl_servers,
564                cell->vl_addrs,
565                min(sizeof(ccell->vl_servers), sizeof(cell->vl_addrs)));
566
567 } /* end afs_cell_cache_update() */
568 #endif