]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/afs/rotate.c
1 /* Handle fileserver selection and rotation.
3 * Copyright (C) 2017 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 Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
12 #include <linux/kernel.h>
13 #include <linux/slab.h>
15 #include <linux/sched.h>
16 #include <linux/delay.h>
17 #include <linux/sched/signal.h>
22 * Begin an operation on the fileserver.
24 * Fileserver operations are serialised on the server by vnode, so we serialise
25 * them here also using the io_lock.
27 bool afs_begin_vnode_operation(struct afs_fs_cursor
*fc
, struct afs_vnode
*vnode
,
30 memset(fc
, 0, sizeof(*fc
));
33 fc
->ac
.error
= SHRT_MAX
;
34 fc
->error
= -EDESTADDRREQ
;
36 if (mutex_lock_interruptible(&vnode
->io_lock
) < 0) {
38 fc
->flags
|= AFS_FS_CURSOR_STOP
;
42 if (vnode
->lock_state
!= AFS_VNODE_LOCK_NONE
)
43 fc
->flags
|= AFS_FS_CURSOR_CUR_ONLY
;
48 * Begin iteration through a server list, starting with the vnode's last used
49 * server if possible, or the last recorded good server if not.
51 static bool afs_start_fs_iteration(struct afs_fs_cursor
*fc
,
52 struct afs_vnode
*vnode
)
54 struct afs_cb_interest
*cbi
;
57 read_lock(&vnode
->volume
->servers_lock
);
58 fc
->server_list
= afs_get_serverlist(vnode
->volume
->servers
);
59 read_unlock(&vnode
->volume
->servers_lock
);
61 fc
->untried
= (1UL << fc
->server_list
->nr_servers
) - 1;
62 fc
->index
= READ_ONCE(fc
->server_list
->preferred
);
64 cbi
= vnode
->cb_interest
;
66 /* See if the vnode's preferred record is still available */
67 for (i
= 0; i
< fc
->server_list
->nr_servers
; i
++) {
68 if (fc
->server_list
->servers
[i
].cb_interest
== cbi
) {
74 /* If we have a lock outstanding on a server that's no longer
75 * serving this vnode, then we can't switch to another server
76 * and have to return an error.
78 if (fc
->flags
& AFS_FS_CURSOR_CUR_ONLY
) {
83 /* Note that the callback promise is effectively broken */
84 write_seqlock(&vnode
->cb_lock
);
85 ASSERTCMP(cbi
, ==, vnode
->cb_interest
);
86 vnode
->cb_interest
= NULL
;
87 if (test_and_clear_bit(AFS_VNODE_CB_PROMISED
, &vnode
->flags
))
89 write_sequnlock(&vnode
->cb_lock
);
91 afs_put_cb_interest(afs_v2net(vnode
), cbi
);
100 * Post volume busy note.
102 static void afs_busy(struct afs_volume
*volume
, u32 abort_code
)
106 switch (abort_code
) {
107 case VOFFLINE
: m
= "offline"; break;
108 case VRESTARTING
: m
= "restarting"; break;
109 case VSALVAGING
: m
= "being salvaged"; break;
110 default: m
= "busy"; break;
113 pr_notice("kAFS: Volume %llu '%s' is %s\n", volume
->vid
, volume
->name
, m
);
117 * Sleep and retry the operation to the same fileserver.
119 static bool afs_sleep_and_retry(struct afs_fs_cursor
*fc
)
121 msleep_interruptible(1000);
122 if (signal_pending(current
)) {
123 fc
->error
= -ERESTARTSYS
;
131 * Select the fileserver to use. May be called multiple times to rotate
132 * through the fileservers.
134 bool afs_select_fileserver(struct afs_fs_cursor
*fc
)
136 struct afs_addr_list
*alist
;
137 struct afs_server
*server
;
138 struct afs_vnode
*vnode
= fc
->vnode
;
141 int error
= fc
->ac
.error
, i
;
143 _enter("%lx[%d],%lx[%d],%d,%d",
144 fc
->untried
, fc
->index
,
145 fc
->ac
.tried
, fc
->ac
.index
,
146 error
, fc
->ac
.abort_code
);
148 if (fc
->flags
& AFS_FS_CURSOR_STOP
) {
149 _leave(" = f [stopped]");
155 /* Evaluate the result of the previous operation, if there was one. */
162 /* Success or local failure. Stop. */
164 fc
->flags
|= AFS_FS_CURSOR_STOP
;
165 _leave(" = f [okay/local %d]", error
);
169 /* The far side rejected the operation on some grounds. This
170 * might involve the server being busy or the volume having been moved.
172 switch (fc
->ac
.abort_code
) {
174 /* This fileserver doesn't know about the volume.
175 * - May indicate that the VL is wrong - retry once and compare
177 * - May indicate that the fileserver couldn't attach to the vol.
179 if (fc
->flags
& AFS_FS_CURSOR_VNOVOL
) {
180 fc
->error
= -EREMOTEIO
;
184 write_lock(&vnode
->volume
->servers_lock
);
185 fc
->server_list
->vnovol_mask
|= 1 << fc
->index
;
186 write_unlock(&vnode
->volume
->servers_lock
);
188 set_bit(AFS_VOLUME_NEEDS_UPDATE
, &vnode
->volume
->flags
);
189 error
= afs_check_volume_status(vnode
->volume
, fc
->key
);
191 goto failed_set_error
;
193 if (test_bit(AFS_VOLUME_DELETED
, &vnode
->volume
->flags
)) {
194 fc
->error
= -ENOMEDIUM
;
198 /* If the server list didn't change, then assume that
199 * it's the fileserver having trouble.
201 if (vnode
->volume
->servers
== fc
->server_list
) {
202 fc
->error
= -EREMOTEIO
;
207 fc
->flags
|= AFS_FS_CURSOR_VNOVOL
;
208 _leave(" = t [vnovol]");
211 case VSALVAGE
: /* TODO: Should this return an error or iterate? */
217 fc
->error
= afs_abort_to_error(fc
->ac
.abort_code
);
221 if (!test_and_set_bit(AFS_VOLUME_OFFLINE
, &vnode
->volume
->flags
)) {
222 afs_busy(vnode
->volume
, fc
->ac
.abort_code
);
223 clear_bit(AFS_VOLUME_BUSY
, &vnode
->volume
->flags
);
225 if (fc
->flags
& AFS_FS_CURSOR_NO_VSLEEP
) {
229 if (fc
->flags
& AFS_FS_CURSOR_CUR_ONLY
) {
238 /* Retry after going round all the servers unless we
239 * have a file lock we need to maintain.
241 if (fc
->flags
& AFS_FS_CURSOR_NO_VSLEEP
) {
245 if (!test_and_set_bit(AFS_VOLUME_BUSY
, &vnode
->volume
->flags
)) {
246 afs_busy(vnode
->volume
, fc
->ac
.abort_code
);
247 clear_bit(AFS_VOLUME_OFFLINE
, &vnode
->volume
->flags
);
250 if (fc
->flags
& AFS_FS_CURSOR_CUR_ONLY
) {
251 if (!afs_sleep_and_retry(fc
))
254 /* Retry with same server & address */
255 _leave(" = t [vbusy]");
259 fc
->flags
|= AFS_FS_CURSOR_VBUSY
;
263 /* The volume migrated to another server. We consider
264 * consider all locks and callbacks broken and request
265 * an update from the VLDB.
267 * We also limit the number of VMOVED hops we will
268 * honour, just in case someone sets up a loop.
270 if (fc
->flags
& AFS_FS_CURSOR_VMOVED
) {
271 fc
->error
= -EREMOTEIO
;
274 fc
->flags
|= AFS_FS_CURSOR_VMOVED
;
276 set_bit(AFS_VOLUME_WAIT
, &vnode
->volume
->flags
);
277 set_bit(AFS_VOLUME_NEEDS_UPDATE
, &vnode
->volume
->flags
);
278 error
= afs_check_volume_status(vnode
->volume
, fc
->key
);
280 goto failed_set_error
;
282 /* If the server list didn't change, then the VLDB is
283 * out of sync with the fileservers. This is hopefully
284 * a temporary condition, however, so we don't want to
285 * permanently block access to the file.
287 * TODO: Try other fileservers if we can.
289 * TODO: Retry a few times with sleeps.
291 if (vnode
->volume
->servers
== fc
->server_list
) {
292 fc
->error
= -ENOMEDIUM
;
296 goto restart_from_beginning
;
299 clear_bit(AFS_VOLUME_OFFLINE
, &vnode
->volume
->flags
);
300 clear_bit(AFS_VOLUME_BUSY
, &vnode
->volume
->flags
);
301 fc
->error
= afs_abort_to_error(fc
->ac
.abort_code
);
307 if (fc
->error
!= -EDESTADDRREQ
)
308 goto iterate_address
;
318 goto iterate_address
;
321 _debug("call reset");
326 restart_from_beginning
:
328 afs_end_cursor(&fc
->ac
);
329 afs_put_cb_interest(afs_v2net(vnode
), fc
->cbi
);
331 afs_put_serverlist(afs_v2net(vnode
), fc
->server_list
);
332 fc
->server_list
= NULL
;
335 /* See if we need to do an update of the volume record. Note that the
336 * volume may have moved or even have been deleted.
338 error
= afs_check_volume_status(vnode
->volume
, fc
->key
);
340 goto failed_set_error
;
342 if (!afs_start_fs_iteration(fc
, vnode
))
345 _debug("__ VOL %llx __", vnode
->volume
->vid
);
346 error
= afs_probe_fileservers(afs_v2net(vnode
), fc
->key
, fc
->server_list
);
348 goto failed_set_error
;
351 _debug("pick [%lx]", fc
->untried
);
353 error
= afs_wait_for_fs_probes(fc
->server_list
, fc
->untried
);
355 goto failed_set_error
;
357 /* Pick the untried server with the lowest RTT. If we have outstanding
358 * callbacks, we stick with the server we're already using if we can.
361 _debug("cbi %u", fc
->index
);
362 if (test_bit(fc
->index
, &fc
->untried
))
363 goto selected_server
;
364 afs_put_cb_interest(afs_v2net(vnode
), fc
->cbi
);
371 for (i
= 0; i
< fc
->server_list
->nr_servers
; i
++) {
372 struct afs_server
*s
= fc
->server_list
->servers
[i
].server
;
374 if (!test_bit(i
, &fc
->untried
) || !s
->probe
.responded
)
376 if (s
->probe
.rtt
< rtt
) {
383 goto no_more_servers
;
386 _debug("use %d", fc
->index
);
387 __clear_bit(fc
->index
, &fc
->untried
);
389 /* We're starting on a different fileserver from the list. We need to
390 * check it, create a callback intercept, find its address list and
391 * probe its capabilities before we use it.
393 ASSERTCMP(fc
->ac
.alist
, ==, NULL
);
394 server
= fc
->server_list
->servers
[fc
->index
].server
;
396 if (!afs_check_server_record(fc
, server
))
399 _debug("USING SERVER: %pU", &server
->uuid
);
401 /* Make sure we've got a callback interest record for this server. We
402 * have to link it in before we send the request as we can be sent a
403 * break request before we've finished decoding the reply and
404 * installing the vnode.
406 error
= afs_register_server_cb_interest(vnode
, fc
->server_list
,
409 goto failed_set_error
;
411 fc
->cbi
= afs_get_cb_interest(vnode
->cb_interest
);
413 read_lock(&server
->fs_lock
);
414 alist
= rcu_dereference_protected(server
->addresses
,
415 lockdep_is_held(&server
->fs_lock
));
416 afs_get_addrlist(alist
);
417 read_unlock(&server
->fs_lock
);
419 memset(&fc
->ac
, 0, sizeof(fc
->ac
));
422 fc
->ac
.alist
= alist
;
424 afs_put_addrlist(alist
);
429 ASSERT(fc
->ac
.alist
);
430 /* Iterate over the current server's address list to try and find an
431 * address on which it will respond to us.
433 if (!afs_iterate_addresses(&fc
->ac
))
436 _debug("address [%u] %u/%u", fc
->index
, fc
->ac
.index
, fc
->ac
.alist
->nr_addrs
);
443 afs_end_cursor(&fc
->ac
);
447 /* That's all the servers poked to no good effect. Try again if some
450 if (fc
->flags
& AFS_FS_CURSOR_VBUSY
)
451 goto restart_from_beginning
;
453 e
.error
= -EDESTADDRREQ
;
455 for (i
= 0; i
< fc
->server_list
->nr_servers
; i
++) {
456 struct afs_server
*s
= fc
->server_list
->servers
[i
].server
;
458 afs_prioritise_error(&e
, READ_ONCE(s
->probe
.error
),
459 s
->probe
.abort_code
);
465 fc
->flags
|= AFS_FS_CURSOR_STOP
;
466 afs_end_cursor(&fc
->ac
);
467 _leave(" = f [failed %d]", fc
->error
);
472 * Select the same fileserver we used for a vnode before and only that
473 * fileserver. We use this when we have a lock on that file, which is backed
474 * only by the fileserver we obtained it from.
476 bool afs_select_current_fileserver(struct afs_fs_cursor
*fc
)
478 struct afs_vnode
*vnode
= fc
->vnode
;
479 struct afs_cb_interest
*cbi
= vnode
->cb_interest
;
480 struct afs_addr_list
*alist
;
481 int error
= fc
->ac
.error
;
489 fc
->flags
|= AFS_FS_CURSOR_STOP
;
493 fc
->cbi
= afs_get_cb_interest(vnode
->cb_interest
);
495 read_lock(&cbi
->server
->fs_lock
);
496 alist
= rcu_dereference_protected(cbi
->server
->addresses
,
497 lockdep_is_held(&cbi
->server
->fs_lock
));
498 afs_get_addrlist(alist
);
499 read_unlock(&cbi
->server
->fs_lock
);
502 fc
->flags
|= AFS_FS_CURSOR_STOP
;
506 memset(&fc
->ac
, 0, sizeof(fc
->ac
));
507 fc
->ac
.alist
= alist
;
509 goto iterate_address
;
513 /* Success or local failure. Stop. */
515 fc
->flags
|= AFS_FS_CURSOR_STOP
;
516 _leave(" = f [okay/local %d]", error
);
520 fc
->error
= afs_abort_to_error(fc
->ac
.abort_code
);
521 fc
->flags
|= AFS_FS_CURSOR_STOP
;
522 _leave(" = f [abort]");
535 goto iterate_address
;
539 /* Iterate over the current server's address list to try and find an
540 * address on which it will respond to us.
542 if (afs_iterate_addresses(&fc
->ac
)) {
547 afs_end_cursor(&fc
->ac
);
552 * Dump cursor state in the case of the error being EDESTADDRREQ.
554 static void afs_dump_edestaddrreq(const struct afs_fs_cursor
*fc
)
559 if (!IS_ENABLED(CONFIG_AFS_DEBUG_CURSOR
) || count
> 3)
565 pr_notice("EDESTADDR occurred\n");
566 pr_notice("FC: cbb=%x cbb2=%x fl=%hx err=%hd\n",
567 fc
->cb_break
, fc
->cb_break_2
, fc
->flags
, fc
->error
);
568 pr_notice("FC: ut=%lx ix=%d ni=%u\n",
569 fc
->untried
, fc
->index
, fc
->nr_iterations
);
571 if (fc
->server_list
) {
572 const struct afs_server_list
*sl
= fc
->server_list
;
573 pr_notice("FC: SL nr=%u pr=%u vnov=%hx\n",
574 sl
->nr_servers
, sl
->preferred
, sl
->vnovol_mask
);
575 for (i
= 0; i
< sl
->nr_servers
; i
++) {
576 const struct afs_server
*s
= sl
->servers
[i
].server
;
577 pr_notice("FC: server fl=%lx av=%u %pU\n",
578 s
->flags
, s
->addr_version
, &s
->uuid
);
580 const struct afs_addr_list
*a
=
581 rcu_dereference(s
->addresses
);
582 pr_notice("FC: - av=%u nr=%u/%u/%u pr=%u\n",
584 a
->nr_ipv4
, a
->nr_addrs
, a
->max_addrs
,
586 pr_notice("FC: - pr=%lx R=%lx F=%lx\n",
587 a
->probed
, a
->responded
, a
->failed
);
588 if (a
== fc
->ac
.alist
)
589 pr_notice("FC: - current\n");
594 pr_notice("AC: t=%lx ax=%u ac=%d er=%d r=%u ni=%u\n",
595 fc
->ac
.tried
, fc
->ac
.index
, fc
->ac
.abort_code
, fc
->ac
.error
,
596 fc
->ac
.responded
, fc
->ac
.nr_iterations
);
601 * Tidy up a filesystem cursor and unlock the vnode.
603 int afs_end_vnode_operation(struct afs_fs_cursor
*fc
)
605 struct afs_net
*net
= afs_v2net(fc
->vnode
);
607 if (fc
->error
== -EDESTADDRREQ
||
608 fc
->error
== -EADDRNOTAVAIL
||
609 fc
->error
== -ENETUNREACH
||
610 fc
->error
== -EHOSTUNREACH
)
611 afs_dump_edestaddrreq(fc
);
613 mutex_unlock(&fc
->vnode
->io_lock
);
615 afs_end_cursor(&fc
->ac
);
616 afs_put_cb_interest(net
, fc
->cbi
);
617 afs_put_serverlist(net
, fc
->server_list
);
619 if (fc
->error
== -ECONNABORTED
)
620 fc
->error
= afs_abort_to_error(fc
->ac
.abort_code
);