Linux Audio

Check our new training course

Loading...
v5.9
  1// SPDX-License-Identifier: GPL-2.0-or-later
  2/* AFS volume management
  3 *
  4 * Copyright (C) 2002, 2007 Red Hat, Inc. All Rights Reserved.
  5 * Written by David Howells (dhowells@redhat.com)
 
 
 
 
 
  6 */
  7
  8#include <linux/kernel.h>
 
 
  9#include <linux/slab.h>
 
 
 
 10#include "internal.h"
 11
 12unsigned __read_mostly afs_volume_gc_delay = 10;
 13unsigned __read_mostly afs_volume_record_life = 60 * 60;
 14
 15/*
 16 * Insert a volume into a cell.  If there's an existing volume record, that is
 17 * returned instead with a ref held.
 18 */
 19static struct afs_volume *afs_insert_volume_into_cell(struct afs_cell *cell,
 20						      struct afs_volume *volume)
 21{
 22	struct afs_volume *p;
 23	struct rb_node *parent = NULL, **pp;
 24
 25	write_seqlock(&cell->volume_lock);
 26
 27	pp = &cell->volumes.rb_node;
 28	while (*pp) {
 29		parent = *pp;
 30		p = rb_entry(parent, struct afs_volume, cell_node);
 31		if (p->vid < volume->vid) {
 32			pp = &(*pp)->rb_left;
 33		} else if (p->vid > volume->vid) {
 34			pp = &(*pp)->rb_right;
 35		} else {
 36			volume = afs_get_volume(p, afs_volume_trace_get_cell_insert);
 37			goto found;
 38		}
 39	}
 40
 41	rb_link_node_rcu(&volume->cell_node, parent, pp);
 42	rb_insert_color(&volume->cell_node, &cell->volumes);
 43	hlist_add_head_rcu(&volume->proc_link, &cell->proc_volumes);
 44
 45found:
 46	write_sequnlock(&cell->volume_lock);
 47	return volume;
 48
 49}
 50
 51static void afs_remove_volume_from_cell(struct afs_volume *volume)
 52{
 53	struct afs_cell *cell = volume->cell;
 54
 55	if (!hlist_unhashed(&volume->proc_link)) {
 56		trace_afs_volume(volume->vid, atomic_read(&volume->usage),
 57				 afs_volume_trace_remove);
 58		write_seqlock(&cell->volume_lock);
 59		hlist_del_rcu(&volume->proc_link);
 60		rb_erase(&volume->cell_node, &cell->volumes);
 61		write_sequnlock(&cell->volume_lock);
 62	}
 63}
 64
 65/*
 66 * Allocate a volume record and load it up from a vldb record.
 67 */
 68static struct afs_volume *afs_alloc_volume(struct afs_fs_context *params,
 69					   struct afs_vldb_entry *vldb,
 70					   unsigned long type_mask)
 71{
 72	struct afs_server_list *slist;
 73	struct afs_volume *volume;
 74	int ret = -ENOMEM, nr_servers = 0, i;
 75
 76	for (i = 0; i < vldb->nr_servers; i++)
 77		if (vldb->fs_mask[i] & type_mask)
 78			nr_servers++;
 79
 80	volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
 81	if (!volume)
 82		goto error_0;
 83
 84	volume->vid		= vldb->vid[params->type];
 85	volume->update_at	= ktime_get_real_seconds() + afs_volume_record_life;
 86	volume->cell		= afs_get_cell(params->cell);
 87	volume->type		= params->type;
 88	volume->type_force	= params->force;
 89	volume->name_len	= vldb->name_len;
 90
 91	atomic_set(&volume->usage, 1);
 92	INIT_HLIST_NODE(&volume->proc_link);
 93	rwlock_init(&volume->servers_lock);
 94	rwlock_init(&volume->cb_v_break_lock);
 95	memcpy(volume->name, vldb->name, vldb->name_len + 1);
 96
 97	slist = afs_alloc_server_list(params->cell, params->key, vldb, type_mask);
 98	if (IS_ERR(slist)) {
 99		ret = PTR_ERR(slist);
100		goto error_1;
101	}
102
103	refcount_set(&slist->usage, 1);
104	rcu_assign_pointer(volume->servers, slist);
105	trace_afs_volume(volume->vid, 1, afs_volume_trace_alloc);
106	return volume;
107
108error_1:
109	afs_put_cell(params->net, volume->cell);
110	kfree(volume);
111error_0:
112	return ERR_PTR(ret);
113}
114
115/*
116 * Look up or allocate a volume record.
117 */
118static struct afs_volume *afs_lookup_volume(struct afs_fs_context *params,
119					    struct afs_vldb_entry *vldb,
120					    unsigned long type_mask)
121{
122	struct afs_volume *candidate, *volume;
123
124	candidate = afs_alloc_volume(params, vldb, type_mask);
125	if (IS_ERR(candidate))
126		return candidate;
127
128	volume = afs_insert_volume_into_cell(params->cell, candidate);
129	if (volume != candidate)
130		afs_put_volume(params->net, candidate, afs_volume_trace_put_cell_dup);
131	return volume;
132}
133
134/*
135 * Look up a VLDB record for a volume.
136 */
137static struct afs_vldb_entry *afs_vl_lookup_vldb(struct afs_cell *cell,
138						 struct key *key,
139						 const char *volname,
140						 size_t volnamesz)
141{
142	struct afs_vldb_entry *vldb = ERR_PTR(-EDESTADDRREQ);
143	struct afs_vl_cursor vc;
144	int ret;
145
146	if (!afs_begin_vlserver_operation(&vc, cell, key))
147		return ERR_PTR(-ERESTARTSYS);
148
149	while (afs_select_vlserver(&vc)) {
150		vldb = afs_vl_get_entry_by_name_u(&vc, volname, volnamesz);
151	}
152
153	ret = afs_end_vlserver_operation(&vc);
154	return ret < 0 ? ERR_PTR(ret) : vldb;
155}
156
157/*
158 * Look up a volume in the VL server and create a candidate volume record for
159 * it.
160 *
161 * The volume name can be one of the following:
162 *	"%[cell:]volume[.]"		R/W volume
163 *	"#[cell:]volume[.]"		R/O or R/W volume (rwparent=0),
164 *					 or R/W (rwparent=1) volume
165 *	"%[cell:]volume.readonly"	R/O volume
166 *	"#[cell:]volume.readonly"	R/O volume
167 *	"%[cell:]volume.backup"		Backup volume
168 *	"#[cell:]volume.backup"		Backup volume
169 *
170 * The cell name is optional, and defaults to the current cell.
171 *
172 * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
173 * Guide
174 * - Rule 1: Explicit type suffix forces access of that type or nothing
175 *           (no suffix, then use Rule 2 & 3)
176 * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
177 *           if not available
178 * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
179 *           explicitly told otherwise
180 */
181struct afs_volume *afs_create_volume(struct afs_fs_context *params)
182{
183	struct afs_vldb_entry *vldb;
184	struct afs_volume *volume;
185	unsigned long type_mask = 1UL << params->type;
186
187	vldb = afs_vl_lookup_vldb(params->cell, params->key,
188				  params->volname, params->volnamesz);
189	if (IS_ERR(vldb))
190		return ERR_CAST(vldb);
191
192	if (test_bit(AFS_VLDB_QUERY_ERROR, &vldb->flags)) {
193		volume = ERR_PTR(vldb->error);
 
 
 
 
194		goto error;
195	}
196
197	/* Make the final decision on the type we want */
198	volume = ERR_PTR(-ENOMEDIUM);
 
 
 
 
 
 
 
199	if (params->force) {
200		if (!(vldb->flags & type_mask))
201			goto error;
202	} else if (test_bit(AFS_VLDB_HAS_RO, &vldb->flags)) {
203		params->type = AFSVL_ROVOL;
204	} else if (test_bit(AFS_VLDB_HAS_RW, &vldb->flags)) {
205		params->type = AFSVL_RWVOL;
206	} else {
207		goto error;
208	}
209
210	type_mask = 1UL << params->type;
211	volume = afs_lookup_volume(params, vldb, type_mask);
212
213error:
214	kfree(vldb);
215	return volume;
216}
217
218/*
219 * Destroy a volume record
220 */
221static void afs_destroy_volume(struct afs_net *net, struct afs_volume *volume)
222{
223	_enter("%p", volume);
224
225#ifdef CONFIG_AFS_FSCACHE
226	ASSERTCMP(volume->cache, ==, NULL);
227#endif
228
229	afs_remove_volume_from_cell(volume);
230	afs_put_serverlist(net, rcu_access_pointer(volume->servers));
231	afs_put_cell(net, volume->cell);
232	trace_afs_volume(volume->vid, atomic_read(&volume->usage),
233			 afs_volume_trace_free);
234	kfree_rcu(volume, rcu);
235
236	_leave(" [destroyed]");
237}
 
 
238
239/*
240 * Get a reference on a volume record.
241 */
242struct afs_volume *afs_get_volume(struct afs_volume *volume,
243				  enum afs_volume_trace reason)
244{
245	if (volume) {
246		int u = atomic_inc_return(&volume->usage);
247		trace_afs_volume(volume->vid, u, reason);
248	}
249	return volume;
250}
251
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
252
253/*
254 * Drop a reference on a volume record.
255 */
256void afs_put_volume(struct afs_net *net, struct afs_volume *volume,
257		    enum afs_volume_trace reason)
258{
259	if (volume) {
260		afs_volid_t vid = volume->vid;
261		int u = atomic_dec_return(&volume->usage);
262		trace_afs_volume(vid, u, reason);
263		if (u == 0)
264			afs_destroy_volume(net, volume);
265	}
266}
267
268/*
269 * Activate a volume.
270 */
271void afs_activate_volume(struct afs_volume *volume)
272{
273#ifdef CONFIG_AFS_FSCACHE
274	volume->cache = fscache_acquire_cookie(volume->cell->cache,
275					       &afs_volume_cache_index_def,
276					       &volume->vid, sizeof(volume->vid),
277					       NULL, 0,
278					       volume, 0, true);
279#endif
280}
 
281
282/*
283 * Deactivate a volume.
284 */
285void afs_deactivate_volume(struct afs_volume *volume)
286{
287	_enter("%s", volume->name);
288
289#ifdef CONFIG_AFS_FSCACHE
290	fscache_relinquish_cookie(volume->cache, NULL,
291				  test_bit(AFS_VOLUME_DELETED, &volume->flags));
292	volume->cache = NULL;
293#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
294
295	_leave("");
 
 
 
 
296}
297
298/*
299 * Query the VL service to update the volume status.
300 */
301static int afs_update_volume_status(struct afs_volume *volume, struct key *key)
302{
303	struct afs_server_list *new, *old, *discard;
304	struct afs_vldb_entry *vldb;
305	char idbuf[16];
306	int ret, idsz;
307
308	_enter("");
 
309
310	/* We look up an ID by passing it as a decimal string in the
311	 * operation's name parameter.
312	 */
313	idsz = sprintf(idbuf, "%llu", volume->vid);
314
315	vldb = afs_vl_lookup_vldb(volume->cell, key, idbuf, idsz);
316	if (IS_ERR(vldb)) {
317		ret = PTR_ERR(vldb);
318		goto error;
319	}
320
321	/* See if the volume got renamed. */
322	if (vldb->name_len != volume->name_len ||
323	    memcmp(vldb->name, volume->name, vldb->name_len) != 0) {
324		/* TODO: Use RCU'd string. */
325		memcpy(volume->name, vldb->name, AFS_MAXVOLNAME);
326		volume->name_len = vldb->name_len;
327	}
328
329	/* See if the volume's server list got updated. */
330	new = afs_alloc_server_list(volume->cell, key,
331				    vldb, (1 << volume->type));
332	if (IS_ERR(new)) {
333		ret = PTR_ERR(new);
334		goto error_vldb;
335	}
336
337	write_lock(&volume->servers_lock);
338
339	discard = new;
340	old = rcu_dereference_protected(volume->servers,
341					lockdep_is_held(&volume->servers_lock));
342	if (afs_annotate_server_list(new, old)) {
343		new->seq = volume->servers_seq + 1;
344		rcu_assign_pointer(volume->servers, new);
345		smp_wmb();
346		volume->servers_seq++;
347		discard = old;
348	}
349
350	volume->update_at = ktime_get_real_seconds() + afs_volume_record_life;
351	write_unlock(&volume->servers_lock);
352	ret = 0;
 
 
353
354	afs_put_serverlist(volume->cell->net, discard);
355error_vldb:
356	kfree(vldb);
357error:
358	_leave(" = %d", ret);
359	return ret;
 
360}
361
362/*
363 * Make sure the volume record is up to date.
 
364 */
365int afs_check_volume_status(struct afs_volume *volume, struct afs_operation *op)
366{
367	int ret, retries = 0;
 
 
368
369	_enter("");
370
371retry:
372	if (test_bit(AFS_VOLUME_WAIT, &volume->flags))
373		goto wait;
374	if (volume->update_at <= ktime_get_real_seconds() ||
375	    test_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags))
376		goto update;
377	_leave(" = 0");
378	return 0;
379
380update:
381	if (!test_and_set_bit_lock(AFS_VOLUME_UPDATING, &volume->flags)) {
382		clear_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
383		ret = afs_update_volume_status(volume, op->key);
384		if (ret < 0)
385			set_bit(AFS_VOLUME_NEEDS_UPDATE, &volume->flags);
386		clear_bit_unlock(AFS_VOLUME_WAIT, &volume->flags);
387		clear_bit_unlock(AFS_VOLUME_UPDATING, &volume->flags);
388		wake_up_bit(&volume->flags, AFS_VOLUME_WAIT);
389		_leave(" = %d", ret);
390		return ret;
391	}
392
393wait:
394	if (!test_bit(AFS_VOLUME_WAIT, &volume->flags)) {
395		_leave(" = 0 [no wait]");
396		return 0;
397	}
398
399	ret = wait_on_bit(&volume->flags, AFS_VOLUME_WAIT,
400			  (op->flags & AFS_OPERATION_UNINTR) ?
401			  TASK_UNINTERRUPTIBLE : TASK_INTERRUPTIBLE);
402	if (ret == -ERESTARTSYS) {
403		_leave(" = %d", ret);
404		return ret;
405	}
406
407	retries++;
408	if (retries == 4) {
409		_leave(" = -ESTALE");
410		return -ESTALE;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
411	}
412	goto retry;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
413}
v4.10.11
 
  1/* AFS volume management
  2 *
  3 * Copyright (C) 2002, 2007 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/kernel.h>
 13#include <linux/module.h>
 14#include <linux/init.h>
 15#include <linux/slab.h>
 16#include <linux/fs.h>
 17#include <linux/pagemap.h>
 18#include <linux/sched.h>
 19#include "internal.h"
 20
 21static const char *afs_voltypes[] = { "R/W", "R/O", "BAK" };
 
 22
 23/*
 24 * lookup a volume by name
 25 * - this can be one of the following:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 26 *	"%[cell:]volume[.]"		R/W volume
 27 *	"#[cell:]volume[.]"		R/O or R/W volume (rwparent=0),
 28 *					 or R/W (rwparent=1) volume
 29 *	"%[cell:]volume.readonly"	R/O volume
 30 *	"#[cell:]volume.readonly"	R/O volume
 31 *	"%[cell:]volume.backup"		Backup volume
 32 *	"#[cell:]volume.backup"		Backup volume
 33 *
 34 * The cell name is optional, and defaults to the current cell.
 35 *
 36 * See "The Rules of Mount Point Traversal" in Chapter 5 of the AFS SysAdmin
 37 * Guide
 38 * - Rule 1: Explicit type suffix forces access of that type or nothing
 39 *           (no suffix, then use Rule 2 & 3)
 40 * - Rule 2: If parent volume is R/O, then mount R/O volume by preference, R/W
 41 *           if not available
 42 * - Rule 3: If parent volume is R/W, then only mount R/W volume unless
 43 *           explicitly told otherwise
 44 */
 45struct afs_volume *afs_volume_lookup(struct afs_mount_params *params)
 46{
 47	struct afs_vlocation *vlocation = NULL;
 48	struct afs_volume *volume = NULL;
 49	struct afs_server *server = NULL;
 50	char srvtmask;
 51	int ret, loop;
 52
 53	_enter("{%*.*s,%d}",
 54	       params->volnamesz, params->volnamesz, params->volname, params->rwpath);
 55
 56	/* lookup the volume location record */
 57	vlocation = afs_vlocation_lookup(params->cell, params->key,
 58					 params->volname, params->volnamesz);
 59	if (IS_ERR(vlocation)) {
 60		ret = PTR_ERR(vlocation);
 61		vlocation = NULL;
 62		goto error;
 63	}
 64
 65	/* make the final decision on the type we want */
 66	ret = -ENOMEDIUM;
 67	if (params->force && !(vlocation->vldb.vidmask & (1 << params->type)))
 68		goto error;
 69
 70	srvtmask = 0;
 71	for (loop = 0; loop < vlocation->vldb.nservers; loop++)
 72		srvtmask |= vlocation->vldb.srvtmask[loop];
 73
 74	if (params->force) {
 75		if (!(srvtmask & (1 << params->type)))
 76			goto error;
 77	} else if (srvtmask & AFS_VOL_VTM_RO) {
 78		params->type = AFSVL_ROVOL;
 79	} else if (srvtmask & AFS_VOL_VTM_RW) {
 80		params->type = AFSVL_RWVOL;
 81	} else {
 82		goto error;
 83	}
 84
 85	down_write(&params->cell->vl_sem);
 
 86
 87	/* is the volume already active? */
 88	if (vlocation->vols[params->type]) {
 89		/* yes - re-use it */
 90		volume = vlocation->vols[params->type];
 91		afs_get_volume(volume);
 92		goto success;
 93	}
 
 
 
 
 
 
 
 
 94
 95	/* create a new volume record */
 96	_debug("creating new volume record");
 
 
 
 
 97
 98	ret = -ENOMEM;
 99	volume = kzalloc(sizeof(struct afs_volume), GFP_KERNEL);
100	if (!volume)
101		goto error_up;
102
103	atomic_set(&volume->usage, 1);
104	volume->type		= params->type;
105	volume->type_force	= params->force;
106	volume->cell		= params->cell;
107	volume->vid		= vlocation->vldb.vid[params->type];
 
 
 
 
 
 
 
108
109	ret = bdi_setup_and_register(&volume->bdi, "afs");
110	if (ret)
111		goto error_bdi;
112
113	init_rwsem(&volume->server_sem);
114
115	/* look up all the applicable server records */
116	for (loop = 0; loop < 8; loop++) {
117		if (vlocation->vldb.srvtmask[loop] & (1 << volume->type)) {
118			server = afs_lookup_server(
119			       volume->cell, &vlocation->vldb.servers[loop]);
120			if (IS_ERR(server)) {
121				ret = PTR_ERR(server);
122				goto error_discard;
123			}
124
125			volume->servers[volume->nservers] = server;
126			volume->nservers++;
127		}
 
 
 
 
 
 
 
 
 
128	}
 
129
130	/* attach the cache and volume location */
 
 
 
 
131#ifdef CONFIG_AFS_FSCACHE
132	volume->cache = fscache_acquire_cookie(vlocation->cache,
133					       &afs_volume_cache_index_def,
134					       volume, true);
 
 
135#endif
136	afs_get_vlocation(vlocation);
137	volume->vlocation = vlocation;
138
139	vlocation->vols[volume->type] = volume;
 
 
 
 
 
140
141success:
142	_debug("kAFS selected %s volume %08x",
143	       afs_voltypes[volume->type], volume->vid);
144	up_write(&params->cell->vl_sem);
145	afs_put_vlocation(vlocation);
146	_leave(" = %p", volume);
147	return volume;
148
149	/* clean up */
150error_up:
151	up_write(&params->cell->vl_sem);
152error:
153	afs_put_vlocation(vlocation);
154	_leave(" = %d", ret);
155	return ERR_PTR(ret);
156
157error_discard:
158	bdi_destroy(&volume->bdi);
159error_bdi:
160	up_write(&params->cell->vl_sem);
161
162	for (loop = volume->nservers - 1; loop >= 0; loop--)
163		afs_put_server(volume->servers[loop]);
164
165	kfree(volume);
166	goto error;
167}
168
169/*
170 * destroy a volume record
171 */
172void afs_put_volume(struct afs_volume *volume)
173{
174	struct afs_vlocation *vlocation;
175	int loop;
 
 
176
177	if (!volume)
178		return;
179
180	_enter("%p", volume);
 
 
 
181
182	ASSERTCMP(atomic_read(&volume->usage), >, 0);
 
 
 
 
183
184	vlocation = volume->vlocation;
185
186	/* to prevent a race, the decrement and the dequeue must be effectively
187	 * atomic */
188	down_write(&vlocation->cell->vl_sem);
 
 
189
190	if (likely(!atomic_dec_and_test(&volume->usage))) {
191		up_write(&vlocation->cell->vl_sem);
192		_leave("");
193		return;
 
 
194	}
195
196	vlocation->vols[volume->type] = NULL;
197
198	up_write(&vlocation->cell->vl_sem);
 
 
 
 
 
 
 
 
 
199
200	/* finish cleaning up the volume */
201#ifdef CONFIG_AFS_FSCACHE
202	fscache_relinquish_cookie(volume->cache, 0);
203#endif
204	afs_put_vlocation(vlocation);
205
206	for (loop = volume->nservers - 1; loop >= 0; loop--)
207		afs_put_server(volume->servers[loop]);
208
209	bdi_destroy(&volume->bdi);
210	kfree(volume);
211
212	_leave(" [destroyed]");
213}
214
215/*
216 * pick a server to use to try accessing this volume
217 * - returns with an elevated usage count on the server chosen
218 */
219struct afs_server *afs_volume_pick_fileserver(struct afs_vnode *vnode)
220{
221	struct afs_volume *volume = vnode->volume;
222	struct afs_server *server;
223	int ret, state, loop;
224
225	_enter("%s", volume->vlocation->vldb.name);
226
227	/* stick with the server we're already using if we can */
228	if (vnode->server && vnode->server->fs_state == 0) {
229		afs_get_server(vnode->server);
230		_leave(" = %p [current]", vnode->server);
231		return vnode->server;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
232	}
233
234	down_read(&volume->server_sem);
 
 
 
 
235
236	/* handle the no-server case */
237	if (volume->nservers == 0) {
238		ret = volume->rjservers ? -ENOMEDIUM : -ESTALE;
239		up_read(&volume->server_sem);
240		_leave(" = %d [no servers]", ret);
241		return ERR_PTR(ret);
242	}
243
244	/* basically, just search the list for the first live server and use
245	 * that */
246	ret = 0;
247	for (loop = 0; loop < volume->nservers; loop++) {
248		server = volume->servers[loop];
249		state = server->fs_state;
250
251		_debug("consider %d [%d]", loop, state);
252
253		switch (state) {
254			/* found an apparently healthy server */
255		case 0:
256			afs_get_server(server);
257			up_read(&volume->server_sem);
258			_leave(" = %p (picked %08x)",
259			       server, ntohl(server->addr.s_addr));
260			return server;
261
262		case -ENETUNREACH:
263			if (ret == 0)
264				ret = state;
265			break;
266
267		case -EHOSTUNREACH:
268			if (ret == 0 ||
269			    ret == -ENETUNREACH)
270				ret = state;
271			break;
272
273		case -ECONNREFUSED:
274			if (ret == 0 ||
275			    ret == -ENETUNREACH ||
276			    ret == -EHOSTUNREACH)
277				ret = state;
278			break;
279
280		default:
281		case -EREMOTEIO:
282			if (ret == 0 ||
283			    ret == -ENETUNREACH ||
284			    ret == -EHOSTUNREACH ||
285			    ret == -ECONNREFUSED)
286				ret = state;
287			break;
288		}
289	}
290
291	/* no available servers
292	 * - TODO: handle the no active servers case better
293	 */
294	up_read(&volume->server_sem);
295	_leave(" = %d", ret);
296	return ERR_PTR(ret);
297}
298
299/*
300 * release a server after use
301 * - releases the ref on the server struct that was acquired by picking
302 * - records result of using a particular server to access a volume
303 * - return 0 to try again, 1 if okay or to issue error
304 * - the caller must release the server struct if result was 0
305 */
306int afs_volume_release_fileserver(struct afs_vnode *vnode,
307				  struct afs_server *server,
308				  int result)
309{
310	struct afs_volume *volume = vnode->volume;
311	unsigned loop;
312
313	_enter("%s,%08x,%d",
314	       volume->vlocation->vldb.name, ntohl(server->addr.s_addr),
315	       result);
316
317	switch (result) {
318		/* success */
319	case 0:
320		server->fs_act_jif = jiffies;
321		server->fs_state = 0;
322		_leave("");
323		return 1;
324
325		/* the fileserver denied all knowledge of the volume */
326	case -ENOMEDIUM:
327		server->fs_act_jif = jiffies;
328		down_write(&volume->server_sem);
329
330		/* firstly, find where the server is in the active list (if it
331		 * is) */
332		for (loop = 0; loop < volume->nservers; loop++)
333			if (volume->servers[loop] == server)
334				goto present;
335
336		/* no longer there - may have been discarded by another op */
337		goto try_next_server_upw;
338
339	present:
340		volume->nservers--;
341		memmove(&volume->servers[loop],
342			&volume->servers[loop + 1],
343			sizeof(volume->servers[loop]) *
344			(volume->nservers - loop));
345		volume->servers[volume->nservers] = NULL;
346		afs_put_server(server);
347		volume->rjservers++;
348
349		if (volume->nservers > 0)
350			/* another server might acknowledge its existence */
351			goto try_next_server_upw;
352
353		/* handle the case where all the fileservers have rejected the
354		 * volume
355		 * - TODO: try asking the fileservers for volume information
356		 * - TODO: contact the VL server again to see if the volume is
357		 *         no longer registered
358		 */
359		up_write(&volume->server_sem);
360		afs_put_server(server);
361		_leave(" [completely rejected]");
362		return 1;
363
364		/* problem reaching the server */
365	case -ENETUNREACH:
366	case -EHOSTUNREACH:
367	case -ECONNREFUSED:
368	case -ETIME:
369	case -ETIMEDOUT:
370	case -EREMOTEIO:
371		/* mark the server as dead
372		 * TODO: vary dead timeout depending on error
373		 */
374		spin_lock(&server->fs_lock);
375		if (!server->fs_state) {
376			server->fs_dead_jif = jiffies + HZ * 10;
377			server->fs_state = result;
378			printk("kAFS: SERVER DEAD state=%d\n", result);
379		}
380		spin_unlock(&server->fs_lock);
381		goto try_next_server;
382
383		/* miscellaneous error */
384	default:
385		server->fs_act_jif = jiffies;
386	case -ENOMEM:
387	case -ENONET:
388		/* tell the caller to accept the result */
389		afs_put_server(server);
390		_leave(" [local failure]");
391		return 1;
392	}
393
394	/* tell the caller to loop around and try the next server */
395try_next_server_upw:
396	up_write(&volume->server_sem);
397try_next_server:
398	afs_put_server(server);
399	_leave(" [try next server]");
400	return 0;
401}