1 /* AFS volume location management
3 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
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.
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/init.h>
15 #include <linux/sched.h>
18 static unsigned afs_vlocation_timeout = 10; /* volume location timeout in seconds */
19 static unsigned afs_vlocation_update_timeout = 10 * 60;
21 static void afs_vlocation_reaper(struct work_struct *);
22 static void afs_vlocation_updater(struct work_struct *);
24 static LIST_HEAD(afs_vlocation_updates);
25 static LIST_HEAD(afs_vlocation_graveyard);
26 static DEFINE_SPINLOCK(afs_vlocation_updates_lock);
27 static DEFINE_SPINLOCK(afs_vlocation_graveyard_lock);
28 static DECLARE_DELAYED_WORK(afs_vlocation_reap, afs_vlocation_reaper);
29 static DECLARE_DELAYED_WORK(afs_vlocation_update, afs_vlocation_updater);
30 static struct workqueue_struct *afs_vlocation_update_worker;
33 * iterate through the VL servers in a cell until one of them admits knowing
34 * about the volume in question
36 static int afs_vlocation_access_vl_by_name(struct afs_vlocation *vl,
38 struct afs_cache_vlocation *vldb)
40 struct afs_cell *cell = vl->cell;
44 _enter("%s,%s", cell->name, vl->vldb.name);
46 down_write(&vl->cell->vl_sem);
48 for (count = cell->vl_naddrs; count > 0; count--) {
49 addr = cell->vl_addrs[cell->vl_curr_svix];
51 _debug("CellServ[%hu]: %08x", cell->vl_curr_svix, addr.s_addr);
53 /* attempt to access the VL server */
54 ret = afs_vl_get_entry_by_name(&addr, key, vl->vldb.name, vldb,
64 if (ret == -ENOMEM || ret == -ENONET)
74 /* rotate the server records upon lookup failure */
77 cell->vl_curr_svix %= cell->vl_naddrs;
81 up_write(&vl->cell->vl_sem);
87 * iterate through the VL servers in a cell until one of them admits knowing
88 * about the volume in question
90 static int afs_vlocation_access_vl_by_id(struct afs_vlocation *vl,
93 afs_voltype_t voltype,
94 struct afs_cache_vlocation *vldb)
96 struct afs_cell *cell = vl->cell;
100 _enter("%s,%x,%d,", cell->name, volid, voltype);
102 down_write(&vl->cell->vl_sem);
104 for (count = cell->vl_naddrs; count > 0; count--) {
105 addr = cell->vl_addrs[cell->vl_curr_svix];
107 _debug("CellServ[%hu]: %08x", cell->vl_curr_svix, addr.s_addr);
109 /* attempt to access the VL server */
110 ret = afs_vl_get_entry_by_id(&addr, key, volid, voltype, vldb,
120 if (ret == -ENOMEM || ret == -ENONET)
125 if (vl->upd_busy_cnt <= 3) {
126 if (vl->upd_busy_cnt > 1) {
127 /* second+ BUSY - sleep a little bit */
128 set_current_state(TASK_UNINTERRUPTIBLE);
130 __set_current_state(TASK_RUNNING);
143 /* rotate the server records upon lookup failure */
145 cell->vl_curr_svix++;
146 cell->vl_curr_svix %= cell->vl_naddrs;
147 vl->upd_busy_cnt = 0;
151 if (ret < 0 && vl->upd_rej_cnt > 0) {
152 printk(KERN_NOTICE "kAFS:"
153 " Active volume no longer valid '%s'\n",
159 up_write(&vl->cell->vl_sem);
160 _leave(" = %d", ret);
165 * allocate a volume location record
167 static struct afs_vlocation *afs_vlocation_alloc(struct afs_cell *cell,
171 struct afs_vlocation *vl;
173 vl = kzalloc(sizeof(struct afs_vlocation), GFP_KERNEL);
176 vl->state = AFS_VL_NEW;
177 atomic_set(&vl->usage, 1);
178 INIT_LIST_HEAD(&vl->link);
179 INIT_LIST_HEAD(&vl->grave);
180 INIT_LIST_HEAD(&vl->update);
181 init_waitqueue_head(&vl->waitq);
182 spin_lock_init(&vl->lock);
183 memcpy(vl->vldb.name, name, namesz);
191 * update record if we found it in the cache
193 static int afs_vlocation_update_record(struct afs_vlocation *vl,
195 struct afs_cache_vlocation *vldb)
197 afs_voltype_t voltype;
201 /* try to look up a cached volume in the cell VL databases by ID */
202 _debug("Locally Cached: %s %02x { %08x(%x) %08x(%x) %08x(%x) }",
205 ntohl(vl->vldb.servers[0].s_addr),
206 vl->vldb.srvtmask[0],
207 ntohl(vl->vldb.servers[1].s_addr),
208 vl->vldb.srvtmask[1],
209 ntohl(vl->vldb.servers[2].s_addr),
210 vl->vldb.srvtmask[2]);
212 _debug("Vids: %08x %08x %08x",
217 if (vl->vldb.vidmask & AFS_VOL_VTM_RW) {
218 vid = vl->vldb.vid[0];
219 voltype = AFSVL_RWVOL;
220 } else if (vl->vldb.vidmask & AFS_VOL_VTM_RO) {
221 vid = vl->vldb.vid[1];
222 voltype = AFSVL_ROVOL;
223 } else if (vl->vldb.vidmask & AFS_VOL_VTM_BAK) {
224 vid = vl->vldb.vid[2];
225 voltype = AFSVL_BACKVOL;
232 /* contact the server to make sure the volume is still available
233 * - TODO: need to handle disconnected operation here
235 ret = afs_vlocation_access_vl_by_id(vl, key, vid, voltype, vldb);
239 printk(KERN_WARNING "kAFS:"
240 " failed to update volume '%s' (%x) up in '%s': %d\n",
241 vl->vldb.name, vid, vl->cell->name, ret);
242 _leave(" = %d", ret);
245 /* pulled from local cache into memory */
250 /* uh oh... looks like the volume got deleted */
252 printk(KERN_ERR "kAFS:"
253 " volume '%s' (%x) does not exist '%s'\n",
254 vl->vldb.name, vid, vl->cell->name);
256 /* TODO: make existing record unavailable */
257 _leave(" = %d", ret);
263 * apply the update to a VL record
265 static void afs_vlocation_apply_update(struct afs_vlocation *vl,
266 struct afs_cache_vlocation *vldb)
268 _debug("Done VL Lookup: %s %02x { %08x(%x) %08x(%x) %08x(%x) }",
269 vldb->name, vldb->vidmask,
270 ntohl(vldb->servers[0].s_addr), vldb->srvtmask[0],
271 ntohl(vldb->servers[1].s_addr), vldb->srvtmask[1],
272 ntohl(vldb->servers[2].s_addr), vldb->srvtmask[2]);
274 _debug("Vids: %08x %08x %08x",
275 vldb->vid[0], vldb->vid[1], vldb->vid[2]);
277 if (strcmp(vldb->name, vl->vldb.name) != 0)
278 printk(KERN_NOTICE "kAFS:"
279 " name of volume '%s' changed to '%s' on server\n",
280 vl->vldb.name, vldb->name);
284 #ifdef CONFIG_AFS_FSCACHE
285 fscache_update_cookie(vl->cache);
290 * fill in a volume location record, consulting the cache and the VL server
293 static int afs_vlocation_fill_in_record(struct afs_vlocation *vl,
296 struct afs_cache_vlocation vldb;
301 ASSERTCMP(vl->valid, ==, 0);
303 memset(&vldb, 0, sizeof(vldb));
305 /* see if we have an in-cache copy (will set vl->valid if there is) */
306 #ifdef CONFIG_AFS_FSCACHE
307 vl->cache = fscache_acquire_cookie(vl->cell->cache,
308 &afs_vlocation_cache_index_def, vl);
312 /* try to update a known volume in the cell VL databases by
313 * ID as the name may have changed */
314 _debug("found in cache");
315 ret = afs_vlocation_update_record(vl, key, &vldb);
317 /* try to look up an unknown volume in the cell VL databases by
319 ret = afs_vlocation_access_vl_by_name(vl, key, &vldb);
321 printk("kAFS: failed to locate '%s' in cell '%s'\n",
322 vl->vldb.name, vl->cell->name);
327 afs_vlocation_apply_update(vl, &vldb);
333 * queue a vlocation record for updates
335 static void afs_vlocation_queue_for_updates(struct afs_vlocation *vl)
337 struct afs_vlocation *xvl;
339 /* wait at least 10 minutes before updating... */
340 vl->update_at = get_seconds() + afs_vlocation_update_timeout;
342 spin_lock(&afs_vlocation_updates_lock);
344 if (!list_empty(&afs_vlocation_updates)) {
345 /* ... but wait at least 1 second more than the newest record
346 * already queued so that we don't spam the VL server suddenly
347 * with lots of requests
349 xvl = list_entry(afs_vlocation_updates.prev,
350 struct afs_vlocation, update);
351 if (vl->update_at <= xvl->update_at)
352 vl->update_at = xvl->update_at + 1;
354 queue_delayed_work(afs_vlocation_update_worker,
355 &afs_vlocation_update,
356 afs_vlocation_update_timeout * HZ);
359 list_add_tail(&vl->update, &afs_vlocation_updates);
360 spin_unlock(&afs_vlocation_updates_lock);
364 * lookup volume location
365 * - iterate through the VL servers in a cell until one of them admits knowing
366 * about the volume in question
367 * - lookup in the local cache if not able to find on the VL server
368 * - insert/update in the local cache if did get a VL response
370 struct afs_vlocation *afs_vlocation_lookup(struct afs_cell *cell,
375 struct afs_vlocation *vl;
378 _enter("{%s},{%x},%*.*s,%zu",
379 cell->name, key_serial(key),
380 (int) namesz, (int) namesz, name, namesz);
382 if (namesz >= sizeof(vl->vldb.name)) {
383 _leave(" = -ENAMETOOLONG");
384 return ERR_PTR(-ENAMETOOLONG);
387 /* see if we have an in-memory copy first */
388 down_write(&cell->vl_sem);
389 spin_lock(&cell->vl_lock);
390 list_for_each_entry(vl, &cell->vl_list, link) {
391 if (vl->vldb.name[namesz] != '\0')
393 if (memcmp(vl->vldb.name, name, namesz) == 0)
394 goto found_in_memory;
396 spin_unlock(&cell->vl_lock);
398 /* not in the cell's in-memory lists - create a new record */
399 vl = afs_vlocation_alloc(cell, name, namesz);
401 up_write(&cell->vl_sem);
402 return ERR_PTR(-ENOMEM);
407 list_add_tail(&vl->link, &cell->vl_list);
408 vl->state = AFS_VL_CREATING;
409 up_write(&cell->vl_sem);
412 ret = afs_vlocation_fill_in_record(vl, key);
415 spin_lock(&vl->lock);
416 vl->state = AFS_VL_VALID;
417 spin_unlock(&vl->lock);
420 /* update volume entry in local cache */
421 #ifdef CONFIG_AFS_FSCACHE
422 fscache_update_cookie(vl->cache);
425 /* schedule for regular updates */
426 afs_vlocation_queue_for_updates(vl);
430 /* found in memory */
431 _debug("found in memory");
432 atomic_inc(&vl->usage);
433 spin_unlock(&cell->vl_lock);
434 if (!list_empty(&vl->grave)) {
435 spin_lock(&afs_vlocation_graveyard_lock);
436 list_del_init(&vl->grave);
437 spin_unlock(&afs_vlocation_graveyard_lock);
439 up_write(&cell->vl_sem);
441 /* see if it was an abandoned record that we might try filling in */
442 spin_lock(&vl->lock);
443 while (vl->state != AFS_VL_VALID) {
444 afs_vlocation_state_t state = vl->state;
446 _debug("invalid [state %d]", state);
448 if (state == AFS_VL_NEW || state == AFS_VL_NO_VOLUME) {
449 vl->state = AFS_VL_CREATING;
450 spin_unlock(&vl->lock);
454 /* must now wait for creation or update by someone else to
458 spin_unlock(&vl->lock);
459 ret = wait_event_interruptible(vl->waitq,
460 vl->state == AFS_VL_NEW ||
461 vl->state == AFS_VL_VALID ||
462 vl->state == AFS_VL_NO_VOLUME);
465 spin_lock(&vl->lock);
467 spin_unlock(&vl->lock);
474 spin_lock(&vl->lock);
475 vl->state = AFS_VL_NEW;
476 spin_unlock(&vl->lock);
480 afs_put_vlocation(vl);
481 _leave(" = %d", ret);
486 * finish using a volume location record
488 void afs_put_vlocation(struct afs_vlocation *vl)
493 _enter("%s", vl->vldb.name);
495 ASSERTCMP(atomic_read(&vl->usage), >, 0);
497 if (likely(!atomic_dec_and_test(&vl->usage))) {
502 spin_lock(&afs_vlocation_graveyard_lock);
503 if (atomic_read(&vl->usage) == 0) {
505 list_move_tail(&vl->grave, &afs_vlocation_graveyard);
506 vl->time_of_death = get_seconds();
507 schedule_delayed_work(&afs_vlocation_reap,
508 afs_vlocation_timeout * HZ);
510 /* suspend updates on this record */
511 if (!list_empty(&vl->update)) {
512 spin_lock(&afs_vlocation_updates_lock);
513 list_del_init(&vl->update);
514 spin_unlock(&afs_vlocation_updates_lock);
517 spin_unlock(&afs_vlocation_graveyard_lock);
518 _leave(" [killed?]");
522 * destroy a dead volume location record
524 static void afs_vlocation_destroy(struct afs_vlocation *vl)
528 #ifdef CONFIG_AFS_FSCACHE
529 fscache_relinquish_cookie(vl->cache, 0);
531 afs_put_cell(vl->cell);
536 * reap dead volume location records
538 static void afs_vlocation_reaper(struct work_struct *work)
541 struct afs_vlocation *vl;
542 unsigned long delay, expiry;
548 spin_lock(&afs_vlocation_graveyard_lock);
550 while (!list_empty(&afs_vlocation_graveyard)) {
551 vl = list_entry(afs_vlocation_graveyard.next,
552 struct afs_vlocation, grave);
554 _debug("check %p", vl);
556 /* the queue is ordered most dead first */
557 expiry = vl->time_of_death + afs_vlocation_timeout;
559 delay = (expiry - now) * HZ;
560 _debug("delay %lu", delay);
561 if (!schedule_delayed_work(&afs_vlocation_reap,
563 cancel_delayed_work(&afs_vlocation_reap);
564 schedule_delayed_work(&afs_vlocation_reap,
570 spin_lock(&vl->cell->vl_lock);
571 if (atomic_read(&vl->usage) > 0) {
573 list_del_init(&vl->grave);
576 list_move_tail(&vl->grave, &corpses);
577 list_del_init(&vl->link);
579 spin_unlock(&vl->cell->vl_lock);
582 spin_unlock(&afs_vlocation_graveyard_lock);
584 /* now reap the corpses we've extracted */
585 while (!list_empty(&corpses)) {
586 vl = list_entry(corpses.next, struct afs_vlocation, grave);
587 list_del(&vl->grave);
588 afs_vlocation_destroy(vl);
595 * initialise the VL update process
597 int __init afs_vlocation_update_init(void)
599 afs_vlocation_update_worker =
600 create_singlethread_workqueue("kafs_vlupdated");
601 return afs_vlocation_update_worker ? 0 : -ENOMEM;
605 * discard all the volume location records for rmmod
607 void afs_vlocation_purge(void)
609 afs_vlocation_timeout = 0;
611 spin_lock(&afs_vlocation_updates_lock);
612 list_del_init(&afs_vlocation_updates);
613 spin_unlock(&afs_vlocation_updates_lock);
614 cancel_delayed_work(&afs_vlocation_update);
615 queue_delayed_work(afs_vlocation_update_worker,
616 &afs_vlocation_update, 0);
617 destroy_workqueue(afs_vlocation_update_worker);
619 cancel_delayed_work(&afs_vlocation_reap);
620 schedule_delayed_work(&afs_vlocation_reap, 0);
624 * update a volume location
626 static void afs_vlocation_updater(struct work_struct *work)
628 struct afs_cache_vlocation vldb;
629 struct afs_vlocation *vl, *xvl;
638 /* find a record to update */
639 spin_lock(&afs_vlocation_updates_lock);
641 if (list_empty(&afs_vlocation_updates)) {
642 spin_unlock(&afs_vlocation_updates_lock);
643 _leave(" [nothing]");
647 vl = list_entry(afs_vlocation_updates.next,
648 struct afs_vlocation, update);
649 if (atomic_read(&vl->usage) > 0)
651 list_del_init(&vl->update);
654 timeout = vl->update_at - now;
656 queue_delayed_work(afs_vlocation_update_worker,
657 &afs_vlocation_update, timeout * HZ);
658 spin_unlock(&afs_vlocation_updates_lock);
659 _leave(" [nothing]");
663 list_del_init(&vl->update);
664 atomic_inc(&vl->usage);
665 spin_unlock(&afs_vlocation_updates_lock);
667 /* we can now perform the update */
668 _debug("update %s", vl->vldb.name);
669 vl->state = AFS_VL_UPDATING;
671 vl->upd_busy_cnt = 0;
673 ret = afs_vlocation_update_record(vl, NULL, &vldb);
674 spin_lock(&vl->lock);
677 afs_vlocation_apply_update(vl, &vldb);
678 vl->state = AFS_VL_VALID;
681 vl->state = AFS_VL_VOLUME_DELETED;
684 vl->state = AFS_VL_UNCERTAIN;
687 spin_unlock(&vl->lock);
690 /* and then reschedule */
691 _debug("reschedule");
692 vl->update_at = get_seconds() + afs_vlocation_update_timeout;
694 spin_lock(&afs_vlocation_updates_lock);
696 if (!list_empty(&afs_vlocation_updates)) {
697 /* next update in 10 minutes, but wait at least 1 second more
698 * than the newest record already queued so that we don't spam
699 * the VL server suddenly with lots of requests
701 xvl = list_entry(afs_vlocation_updates.prev,
702 struct afs_vlocation, update);
703 if (vl->update_at <= xvl->update_at)
704 vl->update_at = xvl->update_at + 1;
705 xvl = list_entry(afs_vlocation_updates.next,
706 struct afs_vlocation, update);
707 timeout = xvl->update_at - now;
711 timeout = afs_vlocation_update_timeout;
714 ASSERT(list_empty(&vl->update));
716 list_add_tail(&vl->update, &afs_vlocation_updates);
718 _debug("timeout %ld", timeout);
719 queue_delayed_work(afs_vlocation_update_worker,
720 &afs_vlocation_update, timeout * HZ);
721 spin_unlock(&afs_vlocation_updates_lock);
722 afs_put_vlocation(vl);