]>
git.proxmox.com Git - mirror_ubuntu-bionic-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 * Initialise a filesystem server cursor for iterating over FS servers.
24 void afs_init_fs_cursor(struct afs_fs_cursor
*fc
, struct afs_vnode
*vnode
)
26 memset(fc
, 0, sizeof(*fc
));
30 * Begin an operation on the fileserver.
32 * Fileserver operations are serialised on the server by vnode, so we serialise
33 * them here also using the io_lock.
35 bool afs_begin_vnode_operation(struct afs_fs_cursor
*fc
, struct afs_vnode
*vnode
,
38 afs_init_fs_cursor(fc
, vnode
);
41 fc
->ac
.error
= SHRT_MAX
;
43 if (mutex_lock_interruptible(&vnode
->io_lock
) < 0) {
44 fc
->ac
.error
= -EINTR
;
45 fc
->flags
|= AFS_FS_CURSOR_STOP
;
49 if (vnode
->lock_state
!= AFS_VNODE_LOCK_NONE
)
50 fc
->flags
|= AFS_FS_CURSOR_CUR_ONLY
;
55 * Begin iteration through a server list, starting with the vnode's last used
56 * server if possible, or the last recorded good server if not.
58 static bool afs_start_fs_iteration(struct afs_fs_cursor
*fc
,
59 struct afs_vnode
*vnode
)
61 struct afs_cb_interest
*cbi
;
64 read_lock(&vnode
->volume
->servers_lock
);
65 fc
->server_list
= afs_get_serverlist(vnode
->volume
->servers
);
66 read_unlock(&vnode
->volume
->servers_lock
);
68 cbi
= vnode
->cb_interest
;
70 /* See if the vnode's preferred record is still available */
71 for (i
= 0; i
< fc
->server_list
->nr_servers
; i
++) {
72 if (fc
->server_list
->servers
[i
].cb_interest
== cbi
) {
78 /* If we have a lock outstanding on a server that's no longer
79 * serving this vnode, then we can't switch to another server
80 * and have to return an error.
82 if (fc
->flags
& AFS_FS_CURSOR_CUR_ONLY
) {
83 fc
->ac
.error
= -ESTALE
;
87 /* Note that the callback promise is effectively broken */
88 write_seqlock(&vnode
->cb_lock
);
89 ASSERTCMP(cbi
, ==, vnode
->cb_interest
);
90 vnode
->cb_interest
= NULL
;
91 if (test_and_clear_bit(AFS_VNODE_CB_PROMISED
, &vnode
->flags
))
93 write_sequnlock(&vnode
->cb_lock
);
95 afs_put_cb_interest(afs_v2net(vnode
), cbi
);
98 fc
->start
= READ_ONCE(fc
->server_list
->index
);
102 fc
->index
= fc
->start
;
107 * Post volume busy note.
109 static void afs_busy(struct afs_volume
*volume
, u32 abort_code
)
113 switch (abort_code
) {
114 case VOFFLINE
: m
= "offline"; break;
115 case VRESTARTING
: m
= "restarting"; break;
116 case VSALVAGING
: m
= "being salvaged"; break;
117 default: m
= "busy"; break;
120 pr_notice("kAFS: Volume %u '%s' is %s\n", volume
->vid
, volume
->name
, m
);
124 * Sleep and retry the operation to the same fileserver.
126 static bool afs_sleep_and_retry(struct afs_fs_cursor
*fc
)
128 msleep_interruptible(1000);
129 if (signal_pending(current
)) {
130 fc
->ac
.error
= -ERESTARTSYS
;
138 * Select the fileserver to use. May be called multiple times to rotate
139 * through the fileservers.
141 bool afs_select_fileserver(struct afs_fs_cursor
*fc
)
143 struct afs_addr_list
*alist
;
144 struct afs_server
*server
;
145 struct afs_vnode
*vnode
= fc
->vnode
;
147 _enter("%u/%u,%u/%u,%d,%d",
148 fc
->index
, fc
->start
,
149 fc
->ac
.index
, fc
->ac
.start
,
150 fc
->ac
.error
, fc
->ac
.abort_code
);
152 if (fc
->flags
& AFS_FS_CURSOR_STOP
) {
153 _leave(" = f [stopped]");
157 /* Evaluate the result of the previous operation, if there was one. */
158 switch (fc
->ac
.error
) {
164 /* Success or local failure. Stop. */
165 fc
->flags
|= AFS_FS_CURSOR_STOP
;
166 _leave(" = f [okay/local %d]", fc
->ac
.error
);
170 /* The far side rejected the operation on some grounds. This
171 * might involve the server being busy or the volume having been moved.
173 switch (fc
->ac
.abort_code
) {
175 /* This fileserver doesn't know about the volume.
176 * - May indicate that the VL is wrong - retry once and compare
178 * - May indicate that the fileserver couldn't attach to the vol.
180 if (fc
->flags
& AFS_FS_CURSOR_VNOVOL
) {
181 fc
->ac
.error
= -EREMOTEIO
;
185 write_lock(&vnode
->volume
->servers_lock
);
186 fc
->server_list
->vnovol_mask
|= 1 << fc
->index
;
187 write_unlock(&vnode
->volume
->servers_lock
);
189 set_bit(AFS_VOLUME_NEEDS_UPDATE
, &vnode
->volume
->flags
);
190 fc
->ac
.error
= afs_check_volume_status(vnode
->volume
, fc
->key
);
191 if (fc
->ac
.error
< 0)
194 if (test_bit(AFS_VOLUME_DELETED
, &vnode
->volume
->flags
)) {
195 fc
->ac
.error
= -ENOMEDIUM
;
199 /* If the server list didn't change, then assume that
200 * it's the fileserver having trouble.
202 if (vnode
->volume
->servers
== fc
->server_list
) {
203 fc
->ac
.error
= -EREMOTEIO
;
208 fc
->flags
|= AFS_FS_CURSOR_VNOVOL
;
209 _leave(" = t [vnovol]");
212 case VSALVAGE
: /* TODO: Should this return an error or iterate? */
218 fc
->ac
.error
= afs_abort_to_error(fc
->ac
.abort_code
);
222 if (!test_and_set_bit(AFS_VOLUME_OFFLINE
, &vnode
->volume
->flags
)) {
223 afs_busy(vnode
->volume
, fc
->ac
.abort_code
);
224 clear_bit(AFS_VOLUME_BUSY
, &vnode
->volume
->flags
);
226 if (fc
->flags
& AFS_FS_CURSOR_NO_VSLEEP
) {
227 fc
->ac
.error
= -EADV
;
230 if (fc
->flags
& AFS_FS_CURSOR_CUR_ONLY
) {
231 fc
->ac
.error
= -ESTALE
;
239 /* Retry after going round all the servers unless we
240 * have a file lock we need to maintain.
242 if (fc
->flags
& AFS_FS_CURSOR_NO_VSLEEP
) {
243 fc
->ac
.error
= -EBUSY
;
246 if (!test_and_set_bit(AFS_VOLUME_BUSY
, &vnode
->volume
->flags
)) {
247 afs_busy(vnode
->volume
, fc
->ac
.abort_code
);
248 clear_bit(AFS_VOLUME_OFFLINE
, &vnode
->volume
->flags
);
251 if (fc
->flags
& AFS_FS_CURSOR_CUR_ONLY
) {
252 if (!afs_sleep_and_retry(fc
))
255 /* Retry with same server & address */
256 _leave(" = t [vbusy]");
260 fc
->flags
|= AFS_FS_CURSOR_VBUSY
;
264 /* The volume migrated to another server. We consider
265 * consider all locks and callbacks broken and request
266 * an update from the VLDB.
268 * We also limit the number of VMOVED hops we will
269 * honour, just in case someone sets up a loop.
271 if (fc
->flags
& AFS_FS_CURSOR_VMOVED
) {
272 fc
->ac
.error
= -EREMOTEIO
;
275 fc
->flags
|= AFS_FS_CURSOR_VMOVED
;
277 set_bit(AFS_VOLUME_WAIT
, &vnode
->volume
->flags
);
278 set_bit(AFS_VOLUME_NEEDS_UPDATE
, &vnode
->volume
->flags
);
279 fc
->ac
.error
= afs_check_volume_status(vnode
->volume
, fc
->key
);
280 if (fc
->ac
.error
< 0)
283 /* If the server list didn't change, then the VLDB is
284 * out of sync with the fileservers. This is hopefully
285 * a temporary condition, however, so we don't want to
286 * permanently block access to the file.
288 * TODO: Try other fileservers if we can.
290 * TODO: Retry a few times with sleeps.
292 if (vnode
->volume
->servers
== fc
->server_list
) {
293 fc
->ac
.error
= -ENOMEDIUM
;
297 goto restart_from_beginning
;
300 clear_bit(AFS_VOLUME_OFFLINE
, &vnode
->volume
->flags
);
301 clear_bit(AFS_VOLUME_BUSY
, &vnode
->volume
->flags
);
302 fc
->ac
.error
= afs_abort_to_error(fc
->ac
.abort_code
);
312 goto iterate_address
;
315 restart_from_beginning
:
317 afs_end_cursor(&fc
->ac
);
318 afs_put_cb_interest(afs_v2net(vnode
), fc
->cbi
);
320 afs_put_serverlist(afs_v2net(vnode
), fc
->server_list
);
321 fc
->server_list
= NULL
;
324 /* See if we need to do an update of the volume record. Note that the
325 * volume may have moved or even have been deleted.
327 fc
->ac
.error
= afs_check_volume_status(vnode
->volume
, fc
->key
);
328 if (fc
->ac
.error
< 0)
331 if (!afs_start_fs_iteration(fc
, vnode
))
337 afs_put_cb_interest(afs_v2net(vnode
), fc
->cbi
);
340 if (fc
->index
>= fc
->server_list
->nr_servers
)
342 if (fc
->index
!= fc
->start
)
345 /* That's all the servers poked to no good effect. Try again if some
348 if (fc
->flags
& AFS_FS_CURSOR_VBUSY
)
349 goto restart_from_beginning
;
351 fc
->ac
.error
= -EDESTADDRREQ
;
356 /* We're starting on a different fileserver from the list. We need to
357 * check it, create a callback intercept, find its address list and
358 * probe its capabilities before we use it.
360 ASSERTCMP(fc
->ac
.alist
, ==, NULL
);
361 server
= fc
->server_list
->servers
[fc
->index
].server
;
363 if (!afs_check_server_record(fc
, server
))
366 _debug("USING SERVER: %pU", &server
->uuid
);
368 /* Make sure we've got a callback interest record for this server. We
369 * have to link it in before we send the request as we can be sent a
370 * break request before we've finished decoding the reply and
371 * installing the vnode.
373 fc
->ac
.error
= afs_register_server_cb_interest(
374 vnode
, &fc
->server_list
->servers
[fc
->index
]);
375 if (fc
->ac
.error
< 0)
378 fc
->cbi
= afs_get_cb_interest(vnode
->cb_interest
);
380 read_lock(&server
->fs_lock
);
381 alist
= rcu_dereference_protected(server
->addresses
,
382 lockdep_is_held(&server
->fs_lock
));
383 afs_get_addrlist(alist
);
384 read_unlock(&server
->fs_lock
);
387 /* Probe the current fileserver if we haven't done so yet. */
388 if (!test_bit(AFS_SERVER_FL_PROBED
, &server
->flags
)) {
389 fc
->ac
.alist
= afs_get_addrlist(alist
);
391 if (!afs_probe_fileserver(fc
))
396 fc
->ac
.alist
= alist
;
398 afs_put_addrlist(alist
);
401 fc
->ac
.start
= READ_ONCE(alist
->index
);
402 fc
->ac
.index
= fc
->ac
.start
;
404 fc
->ac
.begun
= false;
405 goto iterate_address
;
408 ASSERT(fc
->ac
.alist
);
409 _debug("iterate %d/%d", fc
->ac
.index
, fc
->ac
.alist
->nr_addrs
);
410 /* Iterate over the current server's address list to try and find an
411 * address on which it will respond to us.
413 if (afs_iterate_addresses(&fc
->ac
)) {
418 afs_end_cursor(&fc
->ac
);
422 fc
->flags
|= AFS_FS_CURSOR_STOP
;
423 _leave(" = f [failed %d]", fc
->ac
.error
);
428 * Select the same fileserver we used for a vnode before and only that
429 * fileserver. We use this when we have a lock on that file, which is backed
430 * only by the fileserver we obtained it from.
432 bool afs_select_current_fileserver(struct afs_fs_cursor
*fc
)
434 struct afs_vnode
*vnode
= fc
->vnode
;
435 struct afs_cb_interest
*cbi
= vnode
->cb_interest
;
436 struct afs_addr_list
*alist
;
440 switch (fc
->ac
.error
) {
443 fc
->ac
.error
= -ESTALE
;
444 fc
->flags
|= AFS_FS_CURSOR_STOP
;
448 fc
->cbi
= afs_get_cb_interest(vnode
->cb_interest
);
450 read_lock(&cbi
->server
->fs_lock
);
451 alist
= rcu_dereference_protected(cbi
->server
->addresses
,
452 lockdep_is_held(&cbi
->server
->fs_lock
));
453 afs_get_addrlist(alist
);
454 read_unlock(&cbi
->server
->fs_lock
);
456 fc
->ac
.error
= -ESTALE
;
457 fc
->flags
|= AFS_FS_CURSOR_STOP
;
461 fc
->ac
.alist
= alist
;
463 fc
->ac
.start
= READ_ONCE(alist
->index
);
464 fc
->ac
.index
= fc
->ac
.start
;
466 fc
->ac
.begun
= false;
467 goto iterate_address
;
471 /* Success or local failure. Stop. */
472 fc
->flags
|= AFS_FS_CURSOR_STOP
;
473 _leave(" = f [okay/local %d]", fc
->ac
.error
);
477 fc
->flags
|= AFS_FS_CURSOR_STOP
;
478 _leave(" = f [abort]");
487 goto iterate_address
;
491 /* Iterate over the current server's address list to try and find an
492 * address on which it will respond to us.
494 if (afs_iterate_addresses(&fc
->ac
)) {
499 afs_end_cursor(&fc
->ac
);
504 * Tidy up a filesystem cursor and unlock the vnode.
506 int afs_end_vnode_operation(struct afs_fs_cursor
*fc
)
508 struct afs_net
*net
= afs_v2net(fc
->vnode
);
511 mutex_unlock(&fc
->vnode
->io_lock
);
513 afs_end_cursor(&fc
->ac
);
514 afs_put_cb_interest(net
, fc
->cbi
);
515 afs_put_serverlist(net
, fc
->server_list
);
518 if (ret
== -ECONNABORTED
)
519 afs_abort_to_error(fc
->ac
.abort_code
);
526 * Set a filesystem server cursor for using a specific FS server.
528 int afs_set_fs_cursor(struct afs_fs_cursor
*fc
, struct afs_vnode
*vnode
)
530 afs_init_fs_cursor(fc
, vnode
);
532 read_seqlock_excl(&vnode
->cb_lock
);
533 if (vnode
->cb_interest
) {
534 if (vnode
->cb_interest
->server
->fs_state
== 0)
535 fc
->server
= afs_get_server(vnode
->cb_interest
->server
);
537 fc
->ac
.error
= vnode
->cb_interest
->server
->fs_state
;
539 fc
->ac
.error
= -ESTALE
;
541 read_sequnlock_excl(&vnode
->cb_lock
);
547 * pick a server to use to try accessing this volume
548 * - returns with an elevated usage count on the server chosen
550 bool afs_volume_pick_fileserver(struct afs_fs_cursor
*fc
, struct afs_vnode
*vnode
)
552 struct afs_volume
*volume
= vnode
->volume
;
553 struct afs_server
*server
;
554 int ret
, state
, loop
;
556 _enter("%s", volume
->vlocation
->vldb
.name
);
558 /* stick with the server we're already using if we can */
559 if (vnode
->cb_interest
&& vnode
->cb_interest
->server
->fs_state
== 0) {
560 fc
->server
= afs_get_server(vnode
->cb_interest
->server
);
564 down_read(&volume
->server_sem
);
566 /* handle the no-server case */
567 if (volume
->nservers
== 0) {
568 fc
->ac
.error
= volume
->rjservers
? -ENOMEDIUM
: -ESTALE
;
569 up_read(&volume
->server_sem
);
570 _leave(" = f [no servers %d]", fc
->ac
.error
);
574 /* basically, just search the list for the first live server and use
577 for (loop
= 0; loop
< volume
->nservers
; loop
++) {
578 server
= volume
->servers
[loop
];
579 state
= server
->fs_state
;
581 _debug("consider %d [%d]", loop
, state
);
600 ret
== -ENETUNREACH
||
601 ret
== -EHOSTUNREACH
)
608 ret
== -ENETUNREACH
||
609 ret
== -EHOSTUNREACH
||
610 ret
== -ECONNREFUSED
)
619 /* no available servers
620 * - TODO: handle the no active servers case better
622 up_read(&volume
->server_sem
);
623 _leave(" = f [%d]", fc
->ac
.error
);
627 /* Found an apparently healthy server. We need to register an interest
628 * in receiving callbacks before we talk to it.
630 ret
= afs_register_server_cb_interest(vnode
,
631 &volume
->cb_interests
[loop
], server
);
635 fc
->server
= afs_get_server(server
);
636 up_read(&volume
->server_sem
);
638 fc
->ac
.alist
= afs_get_addrlist(fc
->server
->addrs
);
639 fc
->ac
.addr
= &fc
->ac
.alist
->addrs
[0];
640 _debug("USING SERVER: %pIS\n", &fc
->ac
.addr
->transport
);
641 _leave(" = t (picked %pIS)", &fc
->ac
.addr
->transport
);
646 * release a server after use
647 * - releases the ref on the server struct that was acquired by picking
648 * - records result of using a particular server to access a volume
649 * - return true to try again, false if okay or to issue error
650 * - the caller must release the server struct if result was false
652 bool afs_iterate_fs_cursor(struct afs_fs_cursor
*fc
,
653 struct afs_vnode
*vnode
)
655 struct afs_volume
*volume
= vnode
->volume
;
656 struct afs_server
*server
= fc
->server
;
660 volume
->vlocation
->vldb
.name
, &fc
->ac
.addr
->transport
,
663 switch (fc
->ac
.error
) {
666 server
->fs_state
= 0;
670 /* the fileserver denied all knowledge of the volume */
672 down_write(&volume
->server_sem
);
674 /* firstly, find where the server is in the active list (if it
676 for (loop
= 0; loop
< volume
->nservers
; loop
++)
677 if (volume
->servers
[loop
] == server
)
680 /* no longer there - may have been discarded by another op */
681 goto try_next_server_upw
;
685 memmove(&volume
->servers
[loop
],
686 &volume
->servers
[loop
+ 1],
687 sizeof(volume
->servers
[loop
]) *
688 (volume
->nservers
- loop
));
689 volume
->servers
[volume
->nservers
] = NULL
;
690 afs_put_server(afs_v2net(vnode
), server
);
693 if (volume
->nservers
> 0)
694 /* another server might acknowledge its existence */
695 goto try_next_server_upw
;
697 /* handle the case where all the fileservers have rejected the
699 * - TODO: try asking the fileservers for volume information
700 * - TODO: contact the VL server again to see if the volume is
701 * no longer registered
703 up_write(&volume
->server_sem
);
704 afs_put_server(afs_v2net(vnode
), server
);
706 _leave(" = f [completely rejected]");
709 /* problem reaching the server */
716 /* mark the server as dead
717 * TODO: vary dead timeout depending on error
719 spin_lock(&server
->fs_lock
);
720 if (!server
->fs_state
) {
721 server
->fs_state
= fc
->ac
.error
;
722 printk("kAFS: SERVER DEAD state=%d\n", fc
->ac
.error
);
724 spin_unlock(&server
->fs_lock
);
725 goto try_next_server
;
727 /* miscellaneous error */
731 /* tell the caller to accept the result */
732 afs_put_server(afs_v2net(vnode
), server
);
734 _leave(" = f [local failure]");
738 /* tell the caller to loop around and try the next server */
740 up_write(&volume
->server_sem
);
742 afs_put_server(afs_v2net(vnode
), server
);
743 _leave(" = t [try next server]");
748 * Clean up a fileserver cursor.
750 int afs_end_fs_cursor(struct afs_fs_cursor
*fc
, struct afs_net
*net
)
752 afs_end_cursor(&fc
->ac
);
753 afs_put_server(net
, fc
->server
);