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 (test_bit(AFS_VNODE_READLOCKED, &vnode->flags) ||
50 test_bit(AFS_VNODE_WRITELOCKED, &vnode->flags))
51 fc->flags |= AFS_FS_CURSOR_CUR_ONLY;
56 * Begin iteration through a server list, starting with the vnode's last used
57 * server if possible, or the last recorded good server if not.
59 static bool afs_start_fs_iteration(struct afs_fs_cursor *fc,
60 struct afs_vnode *vnode)
62 struct afs_cb_interest *cbi;
65 read_lock(&vnode->volume->servers_lock);
66 fc->server_list = afs_get_serverlist(vnode->volume->servers);
67 read_unlock(&vnode->volume->servers_lock);
69 cbi = vnode->cb_interest;
71 /* See if the vnode's preferred record is still available */
72 for (i = 0; i < fc->server_list->nr_servers; i++) {
73 if (fc->server_list->servers[i].cb_interest == cbi) {
79 /* If we have a lock outstanding on a server that's no longer
80 * serving this vnode, then we can't switch to another server
81 * and have to return an error.
83 if (fc->flags & AFS_FS_CURSOR_CUR_ONLY) {
84 fc->ac.error = -ESTALE;
88 /* Note that the callback promise is effectively broken */
89 write_seqlock(&vnode->cb_lock);
90 ASSERTCMP(cbi, ==, vnode->cb_interest);
91 vnode->cb_interest = NULL;
92 if (test_and_clear_bit(AFS_VNODE_CB_PROMISED, &vnode->flags))
94 write_sequnlock(&vnode->cb_lock);
96 afs_put_cb_interest(afs_v2net(vnode), cbi);
99 fc->start = READ_ONCE(fc->server_list->index);
103 fc->index = fc->start;
108 * Post volume busy note.
110 static void afs_busy(struct afs_volume *volume, u32 abort_code)
114 switch (abort_code) {
115 case VOFFLINE: m = "offline"; break;
116 case VRESTARTING: m = "restarting"; break;
117 case VSALVAGING: m = "being salvaged"; break;
118 default: m = "busy"; break;
121 pr_notice("kAFS: Volume %u '%s' is %s\n", volume->vid, volume->name, m);
125 * Sleep and retry the operation to the same fileserver.
127 static bool afs_sleep_and_retry(struct afs_fs_cursor *fc)
129 msleep_interruptible(1000);
130 if (signal_pending(current)) {
131 fc->ac.error = -ERESTARTSYS;
139 * Select the fileserver to use. May be called multiple times to rotate
140 * through the fileservers.
142 bool afs_select_fileserver(struct afs_fs_cursor *fc)
144 struct afs_addr_list *alist;
145 struct afs_server *server;
146 struct afs_vnode *vnode = fc->vnode;
148 _enter("%u/%u,%u/%u,%d,%d",
149 fc->index, fc->start,
150 fc->ac.index, fc->ac.start,
151 fc->ac.error, fc->ac.abort_code);
153 if (fc->flags & AFS_FS_CURSOR_STOP) {
154 _leave(" = f [stopped]");
158 /* Evaluate the result of the previous operation, if there was one. */
159 switch (fc->ac.error) {
165 /* Success or local failure. Stop. */
166 fc->flags |= AFS_FS_CURSOR_STOP;
167 _leave(" = f [okay/local %d]", fc->ac.error);
171 /* The far side rejected the operation on some grounds. This
172 * might involve the server being busy or the volume having been moved.
174 switch (fc->ac.abort_code) {
176 /* This fileserver doesn't know about the volume.
177 * - May indicate that the VL is wrong - retry once and compare
179 * - May indicate that the fileserver couldn't attach to the vol.
181 if (fc->flags & AFS_FS_CURSOR_VNOVOL) {
182 fc->ac.error = -EREMOTEIO;
186 write_lock(&vnode->volume->servers_lock);
187 fc->server_list->vnovol_mask |= 1 << fc->index;
188 write_unlock(&vnode->volume->servers_lock);
190 set_bit(AFS_VOLUME_NEEDS_UPDATE, &vnode->volume->flags);
191 fc->ac.error = afs_check_volume_status(vnode->volume, fc->key);
192 if (fc->ac.error < 0)
195 if (test_bit(AFS_VOLUME_DELETED, &vnode->volume->flags)) {
196 fc->ac.error = -ENOMEDIUM;
200 /* If the server list didn't change, then assume that
201 * it's the fileserver having trouble.
203 if (vnode->volume->servers == fc->server_list) {
204 fc->ac.error = -EREMOTEIO;
209 fc->flags |= AFS_FS_CURSOR_VNOVOL;
210 _leave(" = t [vnovol]");
213 case VSALVAGE: /* TODO: Should this return an error or iterate? */
219 fc->ac.error = afs_abort_to_error(fc->ac.abort_code);
223 if (!test_and_set_bit(AFS_VOLUME_OFFLINE, &vnode->volume->flags)) {
224 afs_busy(vnode->volume, fc->ac.abort_code);
225 clear_bit(AFS_VOLUME_BUSY, &vnode->volume->flags);
227 if (fc->flags & AFS_FS_CURSOR_NO_VSLEEP) {
228 fc->ac.error = -EADV;
231 if (fc->flags & AFS_FS_CURSOR_CUR_ONLY) {
232 fc->ac.error = -ESTALE;
240 /* Retry after going round all the servers unless we
241 * have a file lock we need to maintain.
243 if (fc->flags & AFS_FS_CURSOR_NO_VSLEEP) {
244 fc->ac.error = -EBUSY;
247 if (!test_and_set_bit(AFS_VOLUME_BUSY, &vnode->volume->flags)) {
248 afs_busy(vnode->volume, fc->ac.abort_code);
249 clear_bit(AFS_VOLUME_OFFLINE, &vnode->volume->flags);
252 if (fc->flags & AFS_FS_CURSOR_CUR_ONLY) {
253 if (!afs_sleep_and_retry(fc))
256 /* Retry with same server & address */
257 _leave(" = t [vbusy]");
261 fc->flags |= AFS_FS_CURSOR_VBUSY;
265 /* The volume migrated to another server. We consider
266 * consider all locks and callbacks broken and request
267 * an update from the VLDB.
269 * We also limit the number of VMOVED hops we will
270 * honour, just in case someone sets up a loop.
272 if (fc->flags & AFS_FS_CURSOR_VMOVED) {
273 fc->ac.error = -EREMOTEIO;
276 fc->flags |= AFS_FS_CURSOR_VMOVED;
278 set_bit(AFS_VOLUME_WAIT, &vnode->volume->flags);
279 set_bit(AFS_VOLUME_NEEDS_UPDATE, &vnode->volume->flags);
280 fc->ac.error = afs_check_volume_status(vnode->volume, fc->key);
281 if (fc->ac.error < 0)
284 /* If the server list didn't change, then the VLDB is
285 * out of sync with the fileservers. This is hopefully
286 * a temporary condition, however, so we don't want to
287 * permanently block access to the file.
289 * TODO: Try other fileservers if we can.
291 * TODO: Retry a few times with sleeps.
293 if (vnode->volume->servers == fc->server_list) {
294 fc->ac.error = -ENOMEDIUM;
298 goto restart_from_beginning;
301 clear_bit(AFS_VOLUME_OFFLINE, &vnode->volume->flags);
302 clear_bit(AFS_VOLUME_BUSY, &vnode->volume->flags);
303 fc->ac.error = afs_abort_to_error(fc->ac.abort_code);
313 goto iterate_address;
316 restart_from_beginning:
318 afs_end_cursor(&fc->ac);
319 afs_put_cb_interest(afs_v2net(vnode), fc->cbi);
321 afs_put_serverlist(afs_v2net(vnode), fc->server_list);
322 fc->server_list = NULL;
325 /* See if we need to do an update of the volume record. Note that the
326 * volume may have moved or even have been deleted.
328 fc->ac.error = afs_check_volume_status(vnode->volume, fc->key);
329 if (fc->ac.error < 0)
332 if (!afs_start_fs_iteration(fc, vnode))
338 afs_put_cb_interest(afs_v2net(vnode), fc->cbi);
341 if (fc->index >= fc->server_list->nr_servers)
343 if (fc->index != fc->start)
346 /* That's all the servers poked to no good effect. Try again if some
349 if (fc->flags & AFS_FS_CURSOR_VBUSY)
350 goto restart_from_beginning;
352 fc->ac.error = -EDESTADDRREQ;
357 /* We're starting on a different fileserver from the list. We need to
358 * check it, create a callback intercept, find its address list and
359 * probe its capabilities before we use it.
361 ASSERTCMP(fc->ac.alist, ==, NULL);
362 server = fc->server_list->servers[fc->index].server;
364 if (!afs_check_server_record(fc, server))
367 _debug("USING SERVER: %pU", &server->uuid);
369 /* Make sure we've got a callback interest record for this server. We
370 * have to link it in before we send the request as we can be sent a
371 * break request before we've finished decoding the reply and
372 * installing the vnode.
374 fc->ac.error = afs_register_server_cb_interest(
375 vnode, &fc->server_list->servers[fc->index]);
376 if (fc->ac.error < 0)
379 fc->cbi = afs_get_cb_interest(vnode->cb_interest);
381 read_lock(&server->fs_lock);
382 alist = rcu_dereference_protected(server->addresses,
383 lockdep_is_held(&server->fs_lock));
384 afs_get_addrlist(alist);
385 read_unlock(&server->fs_lock);
388 /* Probe the current fileserver if we haven't done so yet. */
389 if (!test_bit(AFS_SERVER_FL_PROBED, &server->flags)) {
390 fc->ac.alist = afs_get_addrlist(alist);
392 if (!afs_probe_fileserver(fc))
397 fc->ac.alist = alist;
399 afs_put_addrlist(alist);
402 fc->ac.start = READ_ONCE(alist->index);
403 fc->ac.index = fc->ac.start;
405 fc->ac.begun = false;
406 goto iterate_address;
409 ASSERT(fc->ac.alist);
410 _debug("iterate %d/%d", fc->ac.index, fc->ac.alist->nr_addrs);
411 /* Iterate over the current server's address list to try and find an
412 * address on which it will respond to us.
414 if (afs_iterate_addresses(&fc->ac)) {
419 afs_end_cursor(&fc->ac);
423 fc->flags |= AFS_FS_CURSOR_STOP;
424 _leave(" = f [failed %d]", fc->ac.error);
429 * Select the same fileserver we used for a vnode before and only that
430 * fileserver. We use this when we have a lock on that file, which is backed
431 * only by the fileserver we obtained it from.
433 bool afs_select_current_fileserver(struct afs_fs_cursor *fc)
435 struct afs_vnode *vnode = fc->vnode;
436 struct afs_cb_interest *cbi = vnode->cb_interest;
437 struct afs_addr_list *alist;
442 fc->ac.error = -ESTALE;
443 fc->flags |= AFS_FS_CURSOR_STOP;
447 read_lock(&cbi->server->fs_lock);
448 alist = afs_get_addrlist(cbi->server->addresses);
449 read_unlock(&cbi->server->fs_lock);
451 fc->ac.error = -ESTALE;
452 fc->flags |= AFS_FS_CURSOR_STOP;
456 fc->ac.alist = alist;
462 * Tidy up a filesystem cursor and unlock the vnode.
464 int afs_end_vnode_operation(struct afs_fs_cursor *fc)
466 struct afs_net *net = afs_v2net(fc->vnode);
469 mutex_unlock(&fc->vnode->io_lock);
471 afs_end_cursor(&fc->ac);
472 afs_put_cb_interest(net, fc->cbi);
473 afs_put_serverlist(net, fc->server_list);
476 if (ret == -ECONNABORTED)
477 afs_abort_to_error(fc->ac.abort_code);
484 * Set a filesystem server cursor for using a specific FS server.
486 int afs_set_fs_cursor(struct afs_fs_cursor *fc, struct afs_vnode *vnode)
488 afs_init_fs_cursor(fc, vnode);
490 read_seqlock_excl(&vnode->cb_lock);
491 if (vnode->cb_interest) {
492 if (vnode->cb_interest->server->fs_state == 0)
493 fc->server = afs_get_server(vnode->cb_interest->server);
495 fc->ac.error = vnode->cb_interest->server->fs_state;
497 fc->ac.error = -ESTALE;
499 read_sequnlock_excl(&vnode->cb_lock);
505 * pick a server to use to try accessing this volume
506 * - returns with an elevated usage count on the server chosen
508 bool afs_volume_pick_fileserver(struct afs_fs_cursor *fc, struct afs_vnode *vnode)
510 struct afs_volume *volume = vnode->volume;
511 struct afs_server *server;
512 int ret, state, loop;
514 _enter("%s", volume->vlocation->vldb.name);
516 /* stick with the server we're already using if we can */
517 if (vnode->cb_interest && vnode->cb_interest->server->fs_state == 0) {
518 fc->server = afs_get_server(vnode->cb_interest->server);
522 down_read(&volume->server_sem);
524 /* handle the no-server case */
525 if (volume->nservers == 0) {
526 fc->ac.error = volume->rjservers ? -ENOMEDIUM : -ESTALE;
527 up_read(&volume->server_sem);
528 _leave(" = f [no servers %d]", fc->ac.error);
532 /* basically, just search the list for the first live server and use
535 for (loop = 0; loop < volume->nservers; loop++) {
536 server = volume->servers[loop];
537 state = server->fs_state;
539 _debug("consider %d [%d]", loop, state);
558 ret == -ENETUNREACH ||
559 ret == -EHOSTUNREACH)
566 ret == -ENETUNREACH ||
567 ret == -EHOSTUNREACH ||
568 ret == -ECONNREFUSED)
577 /* no available servers
578 * - TODO: handle the no active servers case better
580 up_read(&volume->server_sem);
581 _leave(" = f [%d]", fc->ac.error);
585 /* Found an apparently healthy server. We need to register an interest
586 * in receiving callbacks before we talk to it.
588 ret = afs_register_server_cb_interest(vnode,
589 &volume->cb_interests[loop], server);
593 fc->server = afs_get_server(server);
594 up_read(&volume->server_sem);
596 fc->ac.alist = afs_get_addrlist(fc->server->addrs);
597 fc->ac.addr = &fc->ac.alist->addrs[0];
598 _debug("USING SERVER: %pIS\n", &fc->ac.addr->transport);
599 _leave(" = t (picked %pIS)", &fc->ac.addr->transport);
604 * release a server after use
605 * - releases the ref on the server struct that was acquired by picking
606 * - records result of using a particular server to access a volume
607 * - return true to try again, false if okay or to issue error
608 * - the caller must release the server struct if result was false
610 bool afs_iterate_fs_cursor(struct afs_fs_cursor *fc,
611 struct afs_vnode *vnode)
613 struct afs_volume *volume = vnode->volume;
614 struct afs_server *server = fc->server;
618 volume->vlocation->vldb.name, &fc->ac.addr->transport,
621 switch (fc->ac.error) {
624 server->fs_state = 0;
628 /* the fileserver denied all knowledge of the volume */
630 down_write(&volume->server_sem);
632 /* firstly, find where the server is in the active list (if it
634 for (loop = 0; loop < volume->nservers; loop++)
635 if (volume->servers[loop] == server)
638 /* no longer there - may have been discarded by another op */
639 goto try_next_server_upw;
643 memmove(&volume->servers[loop],
644 &volume->servers[loop + 1],
645 sizeof(volume->servers[loop]) *
646 (volume->nservers - loop));
647 volume->servers[volume->nservers] = NULL;
648 afs_put_server(afs_v2net(vnode), server);
651 if (volume->nservers > 0)
652 /* another server might acknowledge its existence */
653 goto try_next_server_upw;
655 /* handle the case where all the fileservers have rejected the
657 * - TODO: try asking the fileservers for volume information
658 * - TODO: contact the VL server again to see if the volume is
659 * no longer registered
661 up_write(&volume->server_sem);
662 afs_put_server(afs_v2net(vnode), server);
664 _leave(" = f [completely rejected]");
667 /* problem reaching the server */
674 /* mark the server as dead
675 * TODO: vary dead timeout depending on error
677 spin_lock(&server->fs_lock);
678 if (!server->fs_state) {
679 server->fs_state = fc->ac.error;
680 printk("kAFS: SERVER DEAD state=%d\n", fc->ac.error);
682 spin_unlock(&server->fs_lock);
683 goto try_next_server;
685 /* miscellaneous error */
689 /* tell the caller to accept the result */
690 afs_put_server(afs_v2net(vnode), server);
692 _leave(" = f [local failure]");
696 /* tell the caller to loop around and try the next server */
698 up_write(&volume->server_sem);
700 afs_put_server(afs_v2net(vnode), server);
701 _leave(" = t [try next server]");
706 * Clean up a fileserver cursor.
708 int afs_end_fs_cursor(struct afs_fs_cursor *fc, struct afs_net *net)
710 afs_end_cursor(&fc->ac);
711 afs_put_server(net, fc->server);