1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* Handle vlserver selection and rotation.
4 * Copyright (C) 2018 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #include <linux/kernel.h>
9 #include <linux/sched.h>
10 #include <linux/sched/signal.h>
15 * Begin an operation on a volume location server.
17 bool afs_begin_vlserver_operation(struct afs_vl_cursor
*vc
, struct afs_cell
*cell
,
20 memset(vc
, 0, sizeof(*vc
));
23 vc
->error
= -EDESTADDRREQ
;
24 vc
->ac
.error
= SHRT_MAX
;
26 if (signal_pending(current
)) {
28 vc
->flags
|= AFS_VL_CURSOR_STOP
;
36 * Begin iteration through a server list, starting with the last used server if
37 * possible, or the last recorded good server if not.
39 static bool afs_start_vl_iteration(struct afs_vl_cursor
*vc
)
41 struct afs_cell
*cell
= vc
->cell
;
42 unsigned int dns_lookup_count
;
44 if (cell
->dns_source
== DNS_RECORD_UNAVAILABLE
||
45 cell
->dns_expiry
<= ktime_get_real_seconds()) {
46 dns_lookup_count
= smp_load_acquire(&cell
->dns_lookup_count
);
47 set_bit(AFS_CELL_FL_DO_LOOKUP
, &cell
->flags
);
48 afs_queue_cell(cell
, afs_cell_trace_get_queue_dns
);
50 if (cell
->dns_source
== DNS_RECORD_UNAVAILABLE
) {
51 if (wait_var_event_interruptible(
52 &cell
->dns_lookup_count
,
53 smp_load_acquire(&cell
->dns_lookup_count
)
54 != dns_lookup_count
) < 0) {
55 vc
->error
= -ERESTARTSYS
;
60 /* Status load is ordered after lookup counter load */
61 if (cell
->dns_source
== DNS_RECORD_UNAVAILABLE
) {
62 vc
->error
= -EDESTADDRREQ
;
67 read_lock(&cell
->vl_servers_lock
);
68 vc
->server_list
= afs_get_vlserverlist(
69 rcu_dereference_protected(cell
->vl_servers
,
70 lockdep_is_held(&cell
->vl_servers_lock
)));
71 read_unlock(&cell
->vl_servers_lock
);
72 if (!vc
->server_list
->nr_servers
)
75 vc
->untried
= (1UL << vc
->server_list
->nr_servers
) - 1;
81 * Select the vlserver to use. May be called multiple times to rotate
82 * through the vlservers.
84 bool afs_select_vlserver(struct afs_vl_cursor
*vc
)
86 struct afs_addr_list
*alist
;
87 struct afs_vlserver
*vlserver
;
90 int error
= vc
->ac
.error
, i
;
92 _enter("%lx[%d],%lx[%d],%d,%d",
93 vc
->untried
, vc
->index
,
94 vc
->ac
.tried
, vc
->ac
.index
,
95 error
, vc
->ac
.abort_code
);
97 if (vc
->flags
& AFS_VL_CURSOR_STOP
) {
98 _leave(" = f [stopped]");
104 /* Evaluate the result of the previous operation, if there was one. */
111 /* Success or local failure. Stop. */
113 vc
->flags
|= AFS_VL_CURSOR_STOP
;
114 _leave(" = f [okay/local %d]", vc
->ac
.error
);
118 /* The far side rejected the operation on some grounds. This
119 * might involve the server being busy or the volume having been moved.
121 switch (vc
->ac
.abort_code
) {
123 case AFSVL_BADVOLOPER
:
125 /* The server went weird. */
126 vc
->error
= -EREMOTEIO
;
127 //write_lock(&vc->cell->vl_servers_lock);
128 //vc->server_list->weird_mask |= 1 << vc->index;
129 //write_unlock(&vc->cell->vl_servers_lock);
133 vc
->error
= afs_abort_to_error(vc
->ac
.abort_code
);
145 _debug("no conn %d", error
);
147 goto iterate_address
;
150 _debug("call reset");
152 vc
->flags
|= AFS_VL_CURSOR_RETRY
;
160 restart_from_beginning
:
162 afs_end_cursor(&vc
->ac
);
163 afs_put_vlserverlist(vc
->cell
->net
, vc
->server_list
);
164 vc
->server_list
= NULL
;
165 if (vc
->flags
& AFS_VL_CURSOR_RETRIED
)
167 vc
->flags
|= AFS_VL_CURSOR_RETRIED
;
171 if (!afs_start_vl_iteration(vc
))
174 error
= afs_send_vl_probes(vc
->cell
->net
, vc
->key
, vc
->server_list
);
176 goto failed_set_error
;
179 _debug("pick [%lx]", vc
->untried
);
181 error
= afs_wait_for_vl_probes(vc
->server_list
, vc
->untried
);
183 goto failed_set_error
;
185 /* Pick the untried server with the lowest RTT. */
186 vc
->index
= vc
->server_list
->preferred
;
187 if (test_bit(vc
->index
, &vc
->untried
))
188 goto selected_server
;
192 for (i
= 0; i
< vc
->server_list
->nr_servers
; i
++) {
193 struct afs_vlserver
*s
= vc
->server_list
->servers
[i
].server
;
195 if (!test_bit(i
, &vc
->untried
) ||
196 !test_bit(AFS_VLSERVER_FL_RESPONDING
, &s
->flags
))
198 if (s
->probe
.rtt
< rtt
) {
205 goto no_more_servers
;
208 _debug("use %d", vc
->index
);
209 __clear_bit(vc
->index
, &vc
->untried
);
211 /* We're starting on a different vlserver from the list. We need to
212 * check it, find its address list and probe its capabilities before we
215 ASSERTCMP(vc
->ac
.alist
, ==, NULL
);
216 vlserver
= vc
->server_list
->servers
[vc
->index
].server
;
217 vc
->server
= vlserver
;
219 _debug("USING VLSERVER: %s", vlserver
->name
);
221 read_lock(&vlserver
->lock
);
222 alist
= rcu_dereference_protected(vlserver
->addresses
,
223 lockdep_is_held(&vlserver
->lock
));
224 afs_get_addrlist(alist
);
225 read_unlock(&vlserver
->lock
);
227 memset(&vc
->ac
, 0, sizeof(vc
->ac
));
230 vc
->ac
.alist
= alist
;
232 afs_put_addrlist(alist
);
237 ASSERT(vc
->ac
.alist
);
238 /* Iterate over the current server's address list to try and find an
239 * address on which it will respond to us.
241 if (!afs_iterate_addresses(&vc
->ac
))
244 _debug("VL address %d/%d", vc
->ac
.index
, vc
->ac
.alist
->nr_addrs
);
246 _leave(" = t %pISpc", &vc
->ac
.alist
->addrs
[vc
->ac
.index
].transport
);
251 afs_end_cursor(&vc
->ac
);
255 /* That's all the servers poked to no good effect. Try again if some
258 if (vc
->flags
& AFS_VL_CURSOR_RETRY
)
259 goto restart_from_beginning
;
261 e
.error
= -EDESTADDRREQ
;
263 for (i
= 0; i
< vc
->server_list
->nr_servers
; i
++) {
264 struct afs_vlserver
*s
= vc
->server_list
->servers
[i
].server
;
266 if (test_bit(AFS_VLSERVER_FL_RESPONDING
, &s
->flags
))
268 afs_prioritise_error(&e
, READ_ONCE(s
->probe
.error
),
269 s
->probe
.abort_code
);
277 vc
->flags
|= AFS_VL_CURSOR_STOP
;
278 afs_end_cursor(&vc
->ac
);
279 _leave(" = f [failed %d]", vc
->error
);
284 * Dump cursor state in the case of the error being EDESTADDRREQ.
286 static void afs_vl_dump_edestaddrreq(const struct afs_vl_cursor
*vc
)
291 if (!IS_ENABLED(CONFIG_AFS_DEBUG_CURSOR
) || count
> 3)
296 pr_notice("EDESTADDR occurred\n");
297 pr_notice("VC: ut=%lx ix=%u ni=%hu fl=%hx err=%hd\n",
298 vc
->untried
, vc
->index
, vc
->nr_iterations
, vc
->flags
, vc
->error
);
300 if (vc
->server_list
) {
301 const struct afs_vlserver_list
*sl
= vc
->server_list
;
302 pr_notice("VC: SL nr=%u ix=%u\n",
303 sl
->nr_servers
, sl
->index
);
304 for (i
= 0; i
< sl
->nr_servers
; i
++) {
305 const struct afs_vlserver
*s
= sl
->servers
[i
].server
;
306 pr_notice("VC: server %s+%hu fl=%lx E=%hd\n",
307 s
->name
, s
->port
, s
->flags
, s
->probe
.error
);
309 const struct afs_addr_list
*a
=
310 rcu_dereference(s
->addresses
);
311 pr_notice("VC: - nr=%u/%u/%u pf=%u\n",
312 a
->nr_ipv4
, a
->nr_addrs
, a
->max_addrs
,
314 pr_notice("VC: - R=%lx F=%lx\n",
315 a
->responded
, a
->failed
);
316 if (a
== vc
->ac
.alist
)
317 pr_notice("VC: - current\n");
322 pr_notice("AC: t=%lx ax=%u ac=%d er=%d r=%u ni=%u\n",
323 vc
->ac
.tried
, vc
->ac
.index
, vc
->ac
.abort_code
, vc
->ac
.error
,
324 vc
->ac
.responded
, vc
->ac
.nr_iterations
);
329 * Tidy up a volume location server cursor and unlock the vnode.
331 int afs_end_vlserver_operation(struct afs_vl_cursor
*vc
)
333 struct afs_net
*net
= vc
->cell
->net
;
335 if (vc
->error
== -EDESTADDRREQ
||
336 vc
->error
== -EADDRNOTAVAIL
||
337 vc
->error
== -ENETUNREACH
||
338 vc
->error
== -EHOSTUNREACH
)
339 afs_vl_dump_edestaddrreq(vc
);
341 afs_end_cursor(&vc
->ac
);
342 afs_put_vlserverlist(net
, vc
->server_list
);
344 if (vc
->error
== -ECONNABORTED
)
345 vc
->error
= afs_abort_to_error(vc
->ac
.abort_code
);