]>
Commit | Line | Data |
---|---|---|
9cc6fc50 DH |
1 | /* Handle fileserver selection and rotation. |
2 | * | |
3 | * Copyright (C) 2017 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 Licence | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the Licence, or (at your option) any later version. | |
10 | */ | |
11 | ||
12 | #include <linux/kernel.h> | |
13 | #include <linux/slab.h> | |
d2ddc776 DH |
14 | #include <linux/fs.h> |
15 | #include <linux/sched.h> | |
16 | #include <linux/delay.h> | |
17 | #include <linux/sched/signal.h> | |
9cc6fc50 | 18 | #include "internal.h" |
d2ddc776 | 19 | #include "afs_fs.h" |
9cc6fc50 DH |
20 | |
21 | /* | |
22 | * Initialise a filesystem server cursor for iterating over FS servers. | |
23 | */ | |
24 | void afs_init_fs_cursor(struct afs_fs_cursor *fc, struct afs_vnode *vnode) | |
25 | { | |
26 | memset(fc, 0, sizeof(*fc)); | |
27 | } | |
28 | ||
d2ddc776 DH |
29 | /* |
30 | * Begin an operation on the fileserver. | |
31 | * | |
32 | * Fileserver operations are serialised on the server by vnode, so we serialise | |
33 | * them here also using the io_lock. | |
34 | */ | |
35 | bool afs_begin_vnode_operation(struct afs_fs_cursor *fc, struct afs_vnode *vnode, | |
36 | struct key *key) | |
37 | { | |
38 | afs_init_fs_cursor(fc, vnode); | |
39 | fc->vnode = vnode; | |
40 | fc->key = key; | |
41 | fc->ac.error = SHRT_MAX; | |
42 | ||
43 | if (mutex_lock_interruptible(&vnode->io_lock) < 0) { | |
44 | fc->ac.error = -EINTR; | |
45 | fc->flags |= AFS_FS_CURSOR_STOP; | |
46 | return false; | |
47 | } | |
48 | ||
0fafdc9f | 49 | if (vnode->lock_state != AFS_VNODE_LOCK_NONE) |
d2ddc776 DH |
50 | fc->flags |= AFS_FS_CURSOR_CUR_ONLY; |
51 | return true; | |
52 | } | |
53 | ||
54 | /* | |
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. | |
57 | */ | |
58 | static bool afs_start_fs_iteration(struct afs_fs_cursor *fc, | |
59 | struct afs_vnode *vnode) | |
60 | { | |
61 | struct afs_cb_interest *cbi; | |
62 | int i; | |
63 | ||
64 | read_lock(&vnode->volume->servers_lock); | |
65 | fc->server_list = afs_get_serverlist(vnode->volume->servers); | |
66 | read_unlock(&vnode->volume->servers_lock); | |
67 | ||
68 | cbi = vnode->cb_interest; | |
69 | if (cbi) { | |
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) { | |
73 | fc->start = i; | |
74 | goto found_interest; | |
75 | } | |
76 | } | |
77 | ||
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. | |
81 | */ | |
82 | if (fc->flags & AFS_FS_CURSOR_CUR_ONLY) { | |
83 | fc->ac.error = -ESTALE; | |
84 | return false; | |
85 | } | |
86 | ||
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)) | |
92 | vnode->cb_break++; | |
93 | write_sequnlock(&vnode->cb_lock); | |
94 | ||
95 | afs_put_cb_interest(afs_v2net(vnode), cbi); | |
96 | cbi = NULL; | |
97 | } else { | |
98 | fc->start = READ_ONCE(fc->server_list->index); | |
99 | } | |
100 | ||
101 | found_interest: | |
102 | fc->index = fc->start; | |
103 | return true; | |
104 | } | |
105 | ||
106 | /* | |
107 | * Post volume busy note. | |
108 | */ | |
109 | static void afs_busy(struct afs_volume *volume, u32 abort_code) | |
110 | { | |
111 | const char *m; | |
112 | ||
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; | |
118 | } | |
0fafdc9f | 119 | |
d2ddc776 DH |
120 | pr_notice("kAFS: Volume %u '%s' is %s\n", volume->vid, volume->name, m); |
121 | } | |
122 | ||
123 | /* | |
124 | * Sleep and retry the operation to the same fileserver. | |
125 | */ | |
126 | static bool afs_sleep_and_retry(struct afs_fs_cursor *fc) | |
127 | { | |
128 | msleep_interruptible(1000); | |
129 | if (signal_pending(current)) { | |
130 | fc->ac.error = -ERESTARTSYS; | |
131 | return false; | |
132 | } | |
133 | ||
134 | return true; | |
135 | } | |
136 | ||
137 | /* | |
138 | * Select the fileserver to use. May be called multiple times to rotate | |
139 | * through the fileservers. | |
140 | */ | |
141 | bool afs_select_fileserver(struct afs_fs_cursor *fc) | |
142 | { | |
143 | struct afs_addr_list *alist; | |
144 | struct afs_server *server; | |
145 | struct afs_vnode *vnode = fc->vnode; | |
146 | ||
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); | |
151 | ||
152 | if (fc->flags & AFS_FS_CURSOR_STOP) { | |
153 | _leave(" = f [stopped]"); | |
154 | return false; | |
155 | } | |
156 | ||
157 | /* Evaluate the result of the previous operation, if there was one. */ | |
158 | switch (fc->ac.error) { | |
159 | case SHRT_MAX: | |
160 | goto start; | |
161 | ||
162 | case 0: | |
163 | default: | |
164 | /* Success or local failure. Stop. */ | |
165 | fc->flags |= AFS_FS_CURSOR_STOP; | |
166 | _leave(" = f [okay/local %d]", fc->ac.error); | |
167 | return false; | |
168 | ||
169 | case -ECONNABORTED: | |
170 | /* The far side rejected the operation on some grounds. This | |
171 | * might involve the server being busy or the volume having been moved. | |
172 | */ | |
173 | switch (fc->ac.abort_code) { | |
174 | case VNOVOL: | |
175 | /* This fileserver doesn't know about the volume. | |
176 | * - May indicate that the VL is wrong - retry once and compare | |
177 | * the results. | |
178 | * - May indicate that the fileserver couldn't attach to the vol. | |
179 | */ | |
180 | if (fc->flags & AFS_FS_CURSOR_VNOVOL) { | |
181 | fc->ac.error = -EREMOTEIO; | |
182 | goto failed; | |
183 | } | |
184 | ||
185 | write_lock(&vnode->volume->servers_lock); | |
186 | fc->server_list->vnovol_mask |= 1 << fc->index; | |
187 | write_unlock(&vnode->volume->servers_lock); | |
188 | ||
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) | |
192 | goto failed; | |
193 | ||
194 | if (test_bit(AFS_VOLUME_DELETED, &vnode->volume->flags)) { | |
195 | fc->ac.error = -ENOMEDIUM; | |
196 | goto failed; | |
197 | } | |
198 | ||
199 | /* If the server list didn't change, then assume that | |
200 | * it's the fileserver having trouble. | |
201 | */ | |
202 | if (vnode->volume->servers == fc->server_list) { | |
203 | fc->ac.error = -EREMOTEIO; | |
204 | goto failed; | |
205 | } | |
206 | ||
207 | /* Try again */ | |
208 | fc->flags |= AFS_FS_CURSOR_VNOVOL; | |
209 | _leave(" = t [vnovol]"); | |
210 | return true; | |
211 | ||
212 | case VSALVAGE: /* TODO: Should this return an error or iterate? */ | |
213 | case VVOLEXISTS: | |
214 | case VNOSERVICE: | |
215 | case VONLINE: | |
216 | case VDISKFULL: | |
217 | case VOVERQUOTA: | |
218 | fc->ac.error = afs_abort_to_error(fc->ac.abort_code); | |
219 | goto next_server; | |
220 | ||
221 | case VOFFLINE: | |
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); | |
225 | } | |
226 | if (fc->flags & AFS_FS_CURSOR_NO_VSLEEP) { | |
227 | fc->ac.error = -EADV; | |
228 | goto failed; | |
229 | } | |
230 | if (fc->flags & AFS_FS_CURSOR_CUR_ONLY) { | |
231 | fc->ac.error = -ESTALE; | |
232 | goto failed; | |
233 | } | |
234 | goto busy; | |
235 | ||
236 | case VSALVAGING: | |
237 | case VRESTARTING: | |
238 | case VBUSY: | |
239 | /* Retry after going round all the servers unless we | |
240 | * have a file lock we need to maintain. | |
241 | */ | |
242 | if (fc->flags & AFS_FS_CURSOR_NO_VSLEEP) { | |
243 | fc->ac.error = -EBUSY; | |
244 | goto failed; | |
245 | } | |
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); | |
249 | } | |
250 | busy: | |
251 | if (fc->flags & AFS_FS_CURSOR_CUR_ONLY) { | |
252 | if (!afs_sleep_and_retry(fc)) | |
253 | goto failed; | |
254 | ||
255 | /* Retry with same server & address */ | |
256 | _leave(" = t [vbusy]"); | |
257 | return true; | |
258 | } | |
259 | ||
260 | fc->flags |= AFS_FS_CURSOR_VBUSY; | |
261 | goto next_server; | |
262 | ||
263 | case VMOVED: | |
264 | /* The volume migrated to another server. We consider | |
265 | * consider all locks and callbacks broken and request | |
266 | * an update from the VLDB. | |
267 | * | |
268 | * We also limit the number of VMOVED hops we will | |
269 | * honour, just in case someone sets up a loop. | |
270 | */ | |
271 | if (fc->flags & AFS_FS_CURSOR_VMOVED) { | |
272 | fc->ac.error = -EREMOTEIO; | |
273 | goto failed; | |
274 | } | |
275 | fc->flags |= AFS_FS_CURSOR_VMOVED; | |
276 | ||
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) | |
281 | goto failed; | |
282 | ||
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. | |
287 | * | |
288 | * TODO: Try other fileservers if we can. | |
289 | * | |
290 | * TODO: Retry a few times with sleeps. | |
291 | */ | |
292 | if (vnode->volume->servers == fc->server_list) { | |
293 | fc->ac.error = -ENOMEDIUM; | |
294 | goto failed; | |
295 | } | |
296 | ||
297 | goto restart_from_beginning; | |
298 | ||
299 | default: | |
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); | |
303 | goto failed; | |
304 | } | |
305 | ||
306 | case -ENETUNREACH: | |
307 | case -EHOSTUNREACH: | |
308 | case -ECONNREFUSED: | |
309 | case -ETIMEDOUT: | |
310 | case -ETIME: | |
311 | _debug("no conn"); | |
312 | goto iterate_address; | |
313 | } | |
314 | ||
315 | restart_from_beginning: | |
316 | _debug("restart"); | |
317 | afs_end_cursor(&fc->ac); | |
318 | afs_put_cb_interest(afs_v2net(vnode), fc->cbi); | |
319 | fc->cbi = NULL; | |
320 | afs_put_serverlist(afs_v2net(vnode), fc->server_list); | |
321 | fc->server_list = NULL; | |
322 | start: | |
323 | _debug("start"); | |
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. | |
326 | */ | |
327 | fc->ac.error = afs_check_volume_status(vnode->volume, fc->key); | |
328 | if (fc->ac.error < 0) | |
329 | goto failed; | |
330 | ||
331 | if (!afs_start_fs_iteration(fc, vnode)) | |
332 | goto failed; | |
333 | goto use_server; | |
334 | ||
335 | next_server: | |
336 | _debug("next"); | |
337 | afs_put_cb_interest(afs_v2net(vnode), fc->cbi); | |
338 | fc->cbi = NULL; | |
339 | fc->index++; | |
340 | if (fc->index >= fc->server_list->nr_servers) | |
341 | fc->index = 0; | |
342 | if (fc->index != fc->start) | |
343 | goto use_server; | |
344 | ||
345 | /* That's all the servers poked to no good effect. Try again if some | |
346 | * of them were busy. | |
347 | */ | |
348 | if (fc->flags & AFS_FS_CURSOR_VBUSY) | |
349 | goto restart_from_beginning; | |
350 | ||
351 | fc->ac.error = -EDESTADDRREQ; | |
352 | goto failed; | |
353 | ||
354 | use_server: | |
355 | _debug("use"); | |
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. | |
359 | */ | |
360 | ASSERTCMP(fc->ac.alist, ==, NULL); | |
361 | server = fc->server_list->servers[fc->index].server; | |
362 | ||
363 | if (!afs_check_server_record(fc, server)) | |
364 | goto failed; | |
365 | ||
366 | _debug("USING SERVER: %pU", &server->uuid); | |
367 | ||
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. | |
372 | */ | |
373 | fc->ac.error = afs_register_server_cb_interest( | |
374 | vnode, &fc->server_list->servers[fc->index]); | |
375 | if (fc->ac.error < 0) | |
376 | goto failed; | |
377 | ||
378 | fc->cbi = afs_get_cb_interest(vnode->cb_interest); | |
379 | ||
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); | |
385 | ||
386 | ||
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); | |
390 | ||
391 | if (!afs_probe_fileserver(fc)) | |
392 | goto failed; | |
393 | } | |
394 | ||
395 | if (!fc->ac.alist) | |
396 | fc->ac.alist = alist; | |
397 | else | |
398 | afs_put_addrlist(alist); | |
399 | ||
400 | fc->ac.addr = NULL; | |
401 | fc->ac.start = READ_ONCE(alist->index); | |
402 | fc->ac.index = fc->ac.start; | |
403 | fc->ac.error = 0; | |
404 | fc->ac.begun = false; | |
405 | goto iterate_address; | |
406 | ||
407 | 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. | |
412 | */ | |
413 | if (afs_iterate_addresses(&fc->ac)) { | |
414 | _leave(" = t"); | |
415 | return true; | |
416 | } | |
417 | ||
418 | afs_end_cursor(&fc->ac); | |
419 | goto next_server; | |
420 | ||
421 | failed: | |
422 | fc->flags |= AFS_FS_CURSOR_STOP; | |
423 | _leave(" = f [failed %d]", fc->ac.error); | |
424 | return false; | |
425 | } | |
426 | ||
427 | /* | |
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. | |
431 | */ | |
432 | bool afs_select_current_fileserver(struct afs_fs_cursor *fc) | |
433 | { | |
434 | struct afs_vnode *vnode = fc->vnode; | |
435 | struct afs_cb_interest *cbi = vnode->cb_interest; | |
436 | struct afs_addr_list *alist; | |
437 | ||
438 | _enter(""); | |
439 | ||
0fafdc9f DH |
440 | switch (fc->ac.error) { |
441 | case SHRT_MAX: | |
442 | if (!cbi) { | |
443 | fc->ac.error = -ESTALE; | |
444 | fc->flags |= AFS_FS_CURSOR_STOP; | |
445 | return false; | |
446 | } | |
447 | ||
448 | fc->cbi = afs_get_cb_interest(vnode->cb_interest); | |
449 | ||
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); | |
455 | if (!alist) { | |
456 | fc->ac.error = -ESTALE; | |
457 | fc->flags |= AFS_FS_CURSOR_STOP; | |
458 | return false; | |
459 | } | |
460 | ||
461 | fc->ac.alist = alist; | |
462 | fc->ac.addr = NULL; | |
463 | fc->ac.start = READ_ONCE(alist->index); | |
464 | fc->ac.index = fc->ac.start; | |
465 | fc->ac.error = 0; | |
466 | fc->ac.begun = false; | |
467 | goto iterate_address; | |
468 | ||
469 | case 0: | |
470 | default: | |
471 | /* Success or local failure. Stop. */ | |
d2ddc776 | 472 | fc->flags |= AFS_FS_CURSOR_STOP; |
0fafdc9f | 473 | _leave(" = f [okay/local %d]", fc->ac.error); |
d2ddc776 | 474 | return false; |
d2ddc776 | 475 | |
0fafdc9f | 476 | case -ECONNABORTED: |
d2ddc776 | 477 | fc->flags |= AFS_FS_CURSOR_STOP; |
0fafdc9f | 478 | _leave(" = f [abort]"); |
d2ddc776 | 479 | return false; |
0fafdc9f DH |
480 | |
481 | case -ENETUNREACH: | |
482 | case -EHOSTUNREACH: | |
483 | case -ECONNREFUSED: | |
484 | case -ETIMEDOUT: | |
485 | case -ETIME: | |
486 | _debug("no conn"); | |
487 | goto iterate_address; | |
d2ddc776 DH |
488 | } |
489 | ||
0fafdc9f DH |
490 | 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. | |
493 | */ | |
494 | if (afs_iterate_addresses(&fc->ac)) { | |
495 | _leave(" = t"); | |
496 | return true; | |
497 | } | |
498 | ||
499 | afs_end_cursor(&fc->ac); | |
500 | return false; | |
d2ddc776 DH |
501 | } |
502 | ||
503 | /* | |
504 | * Tidy up a filesystem cursor and unlock the vnode. | |
505 | */ | |
506 | int afs_end_vnode_operation(struct afs_fs_cursor *fc) | |
507 | { | |
508 | struct afs_net *net = afs_v2net(fc->vnode); | |
509 | int ret; | |
510 | ||
511 | mutex_unlock(&fc->vnode->io_lock); | |
512 | ||
513 | afs_end_cursor(&fc->ac); | |
514 | afs_put_cb_interest(net, fc->cbi); | |
515 | afs_put_serverlist(net, fc->server_list); | |
516 | ||
517 | ret = fc->ac.error; | |
518 | if (ret == -ECONNABORTED) | |
519 | afs_abort_to_error(fc->ac.abort_code); | |
520 | ||
521 | return fc->ac.error; | |
522 | } | |
523 | ||
524 | #if 0 | |
9cc6fc50 DH |
525 | /* |
526 | * Set a filesystem server cursor for using a specific FS server. | |
527 | */ | |
528 | int afs_set_fs_cursor(struct afs_fs_cursor *fc, struct afs_vnode *vnode) | |
529 | { | |
530 | afs_init_fs_cursor(fc, vnode); | |
531 | ||
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); | |
536 | else | |
537 | fc->ac.error = vnode->cb_interest->server->fs_state; | |
538 | } else { | |
539 | fc->ac.error = -ESTALE; | |
540 | } | |
541 | read_sequnlock_excl(&vnode->cb_lock); | |
542 | ||
543 | return fc->ac.error; | |
544 | } | |
545 | ||
546 | /* | |
547 | * pick a server to use to try accessing this volume | |
548 | * - returns with an elevated usage count on the server chosen | |
549 | */ | |
550 | bool afs_volume_pick_fileserver(struct afs_fs_cursor *fc, struct afs_vnode *vnode) | |
551 | { | |
552 | struct afs_volume *volume = vnode->volume; | |
553 | struct afs_server *server; | |
554 | int ret, state, loop; | |
555 | ||
556 | _enter("%s", volume->vlocation->vldb.name); | |
557 | ||
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); | |
561 | goto set_server; | |
562 | } | |
563 | ||
564 | down_read(&volume->server_sem); | |
565 | ||
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); | |
571 | return false; | |
572 | } | |
573 | ||
574 | /* basically, just search the list for the first live server and use | |
575 | * that */ | |
576 | ret = 0; | |
577 | for (loop = 0; loop < volume->nservers; loop++) { | |
578 | server = volume->servers[loop]; | |
579 | state = server->fs_state; | |
580 | ||
581 | _debug("consider %d [%d]", loop, state); | |
582 | ||
583 | switch (state) { | |
584 | case 0: | |
585 | goto picked_server; | |
586 | ||
587 | case -ENETUNREACH: | |
588 | if (ret == 0) | |
589 | ret = state; | |
590 | break; | |
591 | ||
592 | case -EHOSTUNREACH: | |
593 | if (ret == 0 || | |
594 | ret == -ENETUNREACH) | |
595 | ret = state; | |
596 | break; | |
597 | ||
598 | case -ECONNREFUSED: | |
599 | if (ret == 0 || | |
600 | ret == -ENETUNREACH || | |
601 | ret == -EHOSTUNREACH) | |
602 | ret = state; | |
603 | break; | |
604 | ||
605 | default: | |
606 | case -EREMOTEIO: | |
607 | if (ret == 0 || | |
608 | ret == -ENETUNREACH || | |
609 | ret == -EHOSTUNREACH || | |
610 | ret == -ECONNREFUSED) | |
611 | ret = state; | |
612 | break; | |
613 | } | |
614 | } | |
615 | ||
616 | error: | |
617 | fc->ac.error = ret; | |
618 | ||
619 | /* no available servers | |
620 | * - TODO: handle the no active servers case better | |
621 | */ | |
622 | up_read(&volume->server_sem); | |
623 | _leave(" = f [%d]", fc->ac.error); | |
624 | return false; | |
625 | ||
626 | picked_server: | |
627 | /* Found an apparently healthy server. We need to register an interest | |
628 | * in receiving callbacks before we talk to it. | |
629 | */ | |
630 | ret = afs_register_server_cb_interest(vnode, | |
631 | &volume->cb_interests[loop], server); | |
632 | if (ret < 0) | |
633 | goto error; | |
634 | ||
635 | fc->server = afs_get_server(server); | |
636 | up_read(&volume->server_sem); | |
637 | set_server: | |
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); | |
642 | return true; | |
643 | } | |
644 | ||
645 | /* | |
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 | |
651 | */ | |
652 | bool afs_iterate_fs_cursor(struct afs_fs_cursor *fc, | |
653 | struct afs_vnode *vnode) | |
654 | { | |
655 | struct afs_volume *volume = vnode->volume; | |
656 | struct afs_server *server = fc->server; | |
657 | unsigned loop; | |
658 | ||
659 | _enter("%s,%pIS,%d", | |
660 | volume->vlocation->vldb.name, &fc->ac.addr->transport, | |
661 | fc->ac.error); | |
662 | ||
663 | switch (fc->ac.error) { | |
664 | /* success */ | |
665 | case 0: | |
666 | server->fs_state = 0; | |
667 | _leave(" = f"); | |
668 | return false; | |
669 | ||
670 | /* the fileserver denied all knowledge of the volume */ | |
671 | case -ENOMEDIUM: | |
672 | down_write(&volume->server_sem); | |
673 | ||
674 | /* firstly, find where the server is in the active list (if it | |
675 | * is) */ | |
676 | for (loop = 0; loop < volume->nservers; loop++) | |
677 | if (volume->servers[loop] == server) | |
678 | goto present; | |
679 | ||
680 | /* no longer there - may have been discarded by another op */ | |
681 | goto try_next_server_upw; | |
682 | ||
683 | present: | |
684 | volume->nservers--; | |
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); | |
691 | volume->rjservers++; | |
692 | ||
693 | if (volume->nservers > 0) | |
694 | /* another server might acknowledge its existence */ | |
695 | goto try_next_server_upw; | |
696 | ||
697 | /* handle the case where all the fileservers have rejected the | |
698 | * volume | |
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 | |
702 | */ | |
703 | up_write(&volume->server_sem); | |
704 | afs_put_server(afs_v2net(vnode), server); | |
705 | fc->server = NULL; | |
706 | _leave(" = f [completely rejected]"); | |
707 | return false; | |
708 | ||
709 | /* problem reaching the server */ | |
710 | case -ENETUNREACH: | |
711 | case -EHOSTUNREACH: | |
712 | case -ECONNREFUSED: | |
713 | case -ETIME: | |
714 | case -ETIMEDOUT: | |
715 | case -EREMOTEIO: | |
716 | /* mark the server as dead | |
717 | * TODO: vary dead timeout depending on error | |
718 | */ | |
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); | |
723 | } | |
724 | spin_unlock(&server->fs_lock); | |
725 | goto try_next_server; | |
726 | ||
727 | /* miscellaneous error */ | |
728 | default: | |
729 | case -ENOMEM: | |
730 | case -ENONET: | |
731 | /* tell the caller to accept the result */ | |
732 | afs_put_server(afs_v2net(vnode), server); | |
733 | fc->server = NULL; | |
734 | _leave(" = f [local failure]"); | |
735 | return false; | |
736 | } | |
737 | ||
738 | /* tell the caller to loop around and try the next server */ | |
739 | try_next_server_upw: | |
740 | up_write(&volume->server_sem); | |
741 | try_next_server: | |
742 | afs_put_server(afs_v2net(vnode), server); | |
743 | _leave(" = t [try next server]"); | |
744 | return true; | |
745 | } | |
746 | ||
747 | /* | |
748 | * Clean up a fileserver cursor. | |
749 | */ | |
750 | int afs_end_fs_cursor(struct afs_fs_cursor *fc, struct afs_net *net) | |
751 | { | |
752 | afs_end_cursor(&fc->ac); | |
753 | afs_put_server(net, fc->server); | |
754 | return fc->ac.error; | |
755 | } | |
d2ddc776 DH |
756 | |
757 | #endif |