]>
Commit | Line | Data |
---|---|---|
f7dd881f DW |
1 | /* |
2 | * Copyright (C) 2008 Intel Corporation | |
3 | * | |
4 | * mdmon socket / message handling | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms and conditions of the GNU General Public License, | |
8 | * version 2, as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along with | |
16 | * this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA. | |
18 | */ | |
19 | #ifndef _GNU_SOURCE | |
20 | #define _GNU_SOURCE | |
21 | #endif | |
22 | #include <unistd.h> | |
23 | #include <stdio.h> | |
24 | #include <string.h> | |
25 | #include <stdlib.h> | |
26 | #include <errno.h> | |
27 | #include <fcntl.h> | |
28 | #include <sys/types.h> | |
29 | #include <sys/socket.h> | |
30 | #include <sys/un.h> | |
31 | #include "mdadm.h" | |
bfa44e2e | 32 | #include "mdmon.h" |
f7dd881f | 33 | |
bfa44e2e NB |
34 | static const __u32 start_magic = 0x5a5aa5a5; |
35 | static const __u32 end_magic = 0xa5a55a5a; | |
36 | ||
37 | static int send_buf(int fd, const void* buf, int len, int tmo) | |
f7dd881f | 38 | { |
bfa44e2e NB |
39 | fd_set set; |
40 | int rv; | |
41 | struct timeval timeout = {tmo, 0}; | |
42 | struct timeval *ptmo = tmo ? &timeout : NULL; | |
43 | ||
44 | while (len) { | |
45 | FD_ZERO(&set); | |
46 | FD_SET(fd, &set); | |
47 | rv = select(fd+1, NULL, &set, NULL, ptmo); | |
48 | if (rv <= 0) | |
49 | return -1; | |
50 | rv = write(fd, buf, len); | |
51 | if (rv <= 0) | |
52 | return -1; | |
53 | len -= rv; | |
54 | buf += rv; | |
55 | } | |
56 | return 0; | |
57 | } | |
f7dd881f | 58 | |
bfa44e2e NB |
59 | static int recv_buf(int fd, void* buf, int len, int tmo) |
60 | { | |
61 | fd_set set; | |
62 | int rv; | |
63 | struct timeval timeout = {tmo, 0}; | |
64 | struct timeval *ptmo = tmo ? &timeout : NULL; | |
65 | ||
66 | while (len) { | |
67 | FD_ZERO(&set); | |
68 | FD_SET(fd, &set); | |
69 | rv = select(fd+1, &set, NULL, NULL, ptmo); | |
70 | if (rv <= 0) | |
71 | return -1; | |
72 | rv = read(fd, buf, len); | |
73 | if (rv <= 0) | |
74 | return -1; | |
75 | len -= rv; | |
76 | buf += rv; | |
77 | } | |
78 | return 0; | |
f7dd881f DW |
79 | } |
80 | ||
81 | ||
bfa44e2e | 82 | int send_message(int fd, struct metadata_update *msg, int tmo) |
f7dd881f | 83 | { |
313a4a82 | 84 | __s32 len = msg->len; |
bfa44e2e NB |
85 | int rv; |
86 | ||
87 | rv = send_buf(fd, &start_magic, 4, tmo); | |
88 | rv = rv ?: send_buf(fd, &len, 4, tmo); | |
313a4a82 | 89 | if (len > 0) |
bfa44e2e NB |
90 | rv = rv ?: send_buf(fd, msg->buf, msg->len, tmo); |
91 | rv = send_buf(fd, &end_magic, 4, tmo); | |
92 | ||
93 | return rv; | |
f7dd881f DW |
94 | } |
95 | ||
bfa44e2e | 96 | int receive_message(int fd, struct metadata_update *msg, int tmo) |
f7dd881f | 97 | { |
bfa44e2e | 98 | __u32 magic; |
313a4a82 | 99 | __s32 len; |
bfa44e2e NB |
100 | int rv; |
101 | ||
102 | rv = recv_buf(fd, &magic, 4, tmo); | |
103 | if (rv < 0 || magic != start_magic) | |
104 | return -1; | |
105 | rv = recv_buf(fd, &len, 4, tmo); | |
106 | if (rv < 0 || len > MSG_MAX_LEN) | |
f7dd881f | 107 | return -1; |
313a4a82 | 108 | if (len > 0) { |
bfa44e2e NB |
109 | msg->buf = malloc(len); |
110 | if (msg->buf == NULL) | |
111 | return -1; | |
112 | rv = recv_buf(fd, msg->buf, len, tmo); | |
113 | if (rv < 0) { | |
114 | free(msg->buf); | |
115 | return -1; | |
116 | } | |
117 | } else | |
118 | msg->buf = NULL; | |
119 | rv = recv_buf(fd, &magic, 4, tmo); | |
120 | if (rv < 0 || magic != end_magic) { | |
121 | free(msg->buf); | |
122 | return -1; | |
123 | } | |
124 | msg->len = len; | |
125 | return 0; | |
f7dd881f DW |
126 | } |
127 | ||
bfa44e2e | 128 | int ack(int fd, int tmo) |
f7dd881f | 129 | { |
bfa44e2e | 130 | struct metadata_update msg = { .len = 0 }; |
f7dd881f DW |
131 | |
132 | return send_message(fd, &msg, tmo); | |
133 | } | |
134 | ||
bfa44e2e | 135 | int wait_reply(int fd, int tmo) |
f7dd881f | 136 | { |
bfa44e2e | 137 | struct metadata_update msg; |
bc77ed53 DW |
138 | int err = receive_message(fd, &msg, tmo); |
139 | ||
140 | /* mdmon sent extra data, but caller only cares that we got a | |
141 | * successful reply | |
142 | */ | |
143 | if (err == 0 && msg.len > 0) | |
144 | free(msg.buf); | |
145 | ||
146 | return err; | |
f7dd881f DW |
147 | } |
148 | ||
f7dd881f DW |
149 | int connect_monitor(char *devname) |
150 | { | |
151 | char path[100]; | |
152 | int sfd; | |
153 | long fl; | |
154 | struct sockaddr_un addr; | |
c94709e8 DW |
155 | int pos; |
156 | char *c; | |
157 | ||
753cf905 | 158 | pos = sprintf(path, "%s/", MDMON_DIR); |
c94709e8 DW |
159 | if (is_subarray(devname)) { |
160 | devname++; | |
161 | c = strchr(devname, '/'); | |
162 | if (!c) | |
163 | return -1; | |
164 | snprintf(&path[pos], c - devname + 1, "%s", devname); | |
165 | pos += c - devname; | |
166 | } else | |
167 | pos += sprintf(&path[pos], "%s", devname); | |
168 | sprintf(&path[pos], ".sock"); | |
f7dd881f | 169 | |
f7dd881f DW |
170 | sfd = socket(PF_LOCAL, SOCK_STREAM, 0); |
171 | if (sfd < 0) | |
172 | return -1; | |
173 | ||
174 | addr.sun_family = PF_LOCAL; | |
175 | strcpy(addr.sun_path, path); | |
176 | if (connect(sfd, &addr, sizeof(addr)) < 0) { | |
177 | close(sfd); | |
178 | return -1; | |
179 | } | |
180 | ||
181 | fl = fcntl(sfd, F_GETFL, 0); | |
182 | fl |= O_NONBLOCK; | |
183 | fcntl(sfd, F_SETFL, fl); | |
184 | ||
185 | return sfd; | |
186 | } | |
187 | ||
9f1da824 | 188 | int fping_monitor(int sfd) |
f7dd881f | 189 | { |
f7dd881f DW |
190 | int err = 0; |
191 | ||
192 | if (sfd < 0) | |
193 | return sfd; | |
194 | ||
195 | /* try to ping existing socket */ | |
eb2c876f | 196 | if (ack(sfd, 20) != 0) |
f7dd881f DW |
197 | err = -1; |
198 | ||
199 | /* check the reply */ | |
eb2c876f | 200 | if (!err && wait_reply(sfd, 20) != 0) |
f7dd881f DW |
201 | err = -1; |
202 | ||
9f1da824 DW |
203 | return err; |
204 | } | |
205 | ||
9f1da824 DW |
206 | /* give the monitor a chance to update the metadata */ |
207 | int ping_monitor(char *devname) | |
208 | { | |
209 | int sfd = connect_monitor(devname); | |
210 | int err = fping_monitor(sfd); | |
211 | ||
f7dd881f DW |
212 | close(sfd); |
213 | return err; | |
214 | } | |
313a4a82 | 215 | |
983fff45 AK |
216 | /* ping monitor using device number */ |
217 | int ping_monitor_by_id(int devnum) | |
218 | { | |
219 | int err = -1; | |
220 | char *container = devnum2devname(devnum); | |
221 | ||
222 | if (container) { | |
223 | err = ping_monitor(container); | |
224 | free(container); | |
225 | } | |
226 | ||
227 | return err; | |
228 | } | |
229 | ||
bc77ed53 DW |
230 | static char *ping_monitor_version(char *devname) |
231 | { | |
232 | int sfd = connect_monitor(devname); | |
233 | struct metadata_update msg; | |
234 | int err = 0; | |
235 | ||
236 | if (sfd < 0) | |
237 | return NULL; | |
238 | ||
239 | if (ack(sfd, 20) != 0) | |
240 | err = -1; | |
241 | ||
242 | if (!err && receive_message(sfd, &msg, 20) != 0) | |
243 | err = -1; | |
244 | ||
245 | close(sfd); | |
246 | ||
247 | if (err || !msg.len || !msg.buf) | |
248 | return NULL; | |
249 | return msg.buf; | |
250 | } | |
251 | ||
2a093697 | 252 | int unblock_subarray(struct mdinfo *sra, const int unfreeze) |
bc77ed53 DW |
253 | { |
254 | char buf[64]; | |
255 | int rc = 0; | |
256 | ||
257 | if (sra) { | |
258 | sprintf(buf, "external:%s\n", sra->text_version); | |
259 | buf[9] = '/'; | |
260 | } else | |
261 | buf[9] = '-'; | |
262 | ||
263 | if (buf[9] == '-' || | |
264 | sysfs_set_str(sra, NULL, "metadata_version", buf) || | |
265 | (unfreeze && | |
266 | sysfs_attribute_available(sra, NULL, "sync_action") && | |
267 | sysfs_set_str(sra, NULL, "sync_action", "idle"))) | |
268 | rc = -1; | |
269 | return rc; | |
270 | } | |
271 | ||
2a093697 AK |
272 | int block_subarray(struct mdinfo *sra) |
273 | { | |
274 | char buf[64]; | |
275 | int rc = 0; | |
276 | ||
277 | sprintf(buf, "external:%s\n", sra->text_version); | |
278 | buf[9] = '-'; | |
279 | if (sysfs_set_str(sra, NULL, "metadata_version", buf)) | |
280 | rc = -1; | |
281 | ||
282 | return rc; | |
283 | } | |
7aa437c2 AK |
284 | |
285 | /* check mdmon version if it supports | |
286 | * array blocking mechanism | |
287 | */ | |
288 | int check_mdmon_version(char *container) | |
289 | { | |
290 | char *version = NULL; | |
291 | int devnum = devname2devnum(container); | |
292 | ||
293 | if (!mdmon_running(devnum)) { | |
294 | /* if mdmon is not active we assume that any instance that is | |
295 | * later started will match the current mdadm version, if this | |
296 | * assumption is violated we may inadvertantly rebuild an array | |
297 | * that was meant for reshape, or start rebuild on a spare that | |
298 | * was to be moved to another container | |
299 | */ | |
300 | /* pass */; | |
301 | } else { | |
302 | int ver; | |
303 | ||
304 | version = ping_monitor_version(container); | |
305 | ver = version ? mdadm_version(version) : -1; | |
306 | free(version); | |
307 | if (ver < 3002000) { | |
308 | fprintf(stderr, Name | |
309 | ": mdmon instance for %s cannot be disabled\n", | |
310 | container); | |
311 | return -1; | |
312 | } | |
313 | } | |
314 | ||
315 | return 0; | |
316 | } | |
317 | ||
bc77ed53 DW |
318 | /** |
319 | * block_monitor - prevent mdmon spare assignment | |
320 | * @container - container to block | |
321 | * @freeze - flag to additionally freeze sync_action | |
322 | * | |
323 | * This is used by the reshape code to freeze the container, and the | |
746a6567 N |
324 | * auto-rebuild implementation to atomically move spares. |
325 | * In both cases we need to stop mdmon from assigning spares to replace | |
326 | * failed devices as we might have other plans for the spare. | |
327 | * For the reshape case we also need to 'freeze' sync_action so that | |
328 | * no recovery happens until we have fully prepared for the reshape. | |
329 | * | |
330 | * We tell mdmon that the array is frozen by marking the 'metadata' name | |
331 | * with a leading '-'. The previously told mdmon "Don't make this array | |
332 | * read/write, leave it readonly". Now it means a more general "Don't | |
333 | * reconfigure this array at all". | |
334 | * As older versions of mdmon (which might run from initrd) don't understand | |
335 | * this, we first check that the running mdmon is new enough. | |
bc77ed53 DW |
336 | */ |
337 | int block_monitor(char *container, const int freeze) | |
338 | { | |
bc77ed53 DW |
339 | struct mdstat_ent *ent, *e, *e2; |
340 | struct mdinfo *sra = NULL; | |
bc77ed53 DW |
341 | char buf[64]; |
342 | int rv = 0; | |
343 | ||
7aa437c2 AK |
344 | if (check_mdmon_version(container)) |
345 | return -1; | |
bc77ed53 DW |
346 | |
347 | ent = mdstat_read(0, 0); | |
348 | if (!ent) { | |
349 | fprintf(stderr, Name | |
350 | ": failed to read /proc/mdstat while disabling mdmon\n"); | |
351 | return -1; | |
352 | } | |
353 | ||
354 | /* freeze container contents */ | |
355 | for (e = ent; e; e = e->next) { | |
356 | if (!is_container_member(e, container)) | |
357 | continue; | |
358 | sysfs_free(sra); | |
359 | sra = sysfs_read(-1, e->devnum, GET_VERSION); | |
360 | if (!sra) { | |
361 | fprintf(stderr, Name | |
362 | ": failed to read sysfs for subarray%s\n", | |
363 | to_subarray(e, container)); | |
364 | break; | |
365 | } | |
366 | /* can't reshape an array that we can't monitor */ | |
367 | if (sra->text_version[0] == '-') | |
368 | break; | |
369 | ||
370 | if (freeze && sysfs_freeze_array(sra) < 1) | |
371 | break; | |
372 | /* flag this array to not be modified by mdmon (close race with | |
373 | * takeover in reshape case and spare reassignment in the | |
374 | * auto-rebuild case) | |
375 | */ | |
2a093697 | 376 | if (block_subarray(sra)) |
bc77ed53 DW |
377 | break; |
378 | ping_monitor(container); | |
379 | ||
380 | /* check that we did not race with recovery */ | |
381 | if ((freeze && | |
382 | !sysfs_attribute_available(sra, NULL, "sync_action")) || | |
383 | (freeze && | |
384 | sysfs_attribute_available(sra, NULL, "sync_action") && | |
385 | sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 && | |
386 | strcmp(buf, "frozen\n") == 0)) | |
387 | /* pass */; | |
885f9845 N |
388 | else { |
389 | unblock_subarray(sra, 0); | |
bc77ed53 | 390 | break; |
885f9845 N |
391 | } |
392 | /* Double check against races - there should be no spares | |
393 | * or part-spares | |
394 | */ | |
395 | sysfs_free(sra); | |
396 | sra = sysfs_read(-1, e->devnum, GET_DEVS | GET_STATE); | |
397 | if (sra && sra->array.spare_disks > 0) { | |
398 | unblock_subarray(sra, freeze); | |
399 | break; | |
400 | } | |
bc77ed53 DW |
401 | } |
402 | ||
403 | if (e) { | |
404 | fprintf(stderr, Name ": failed to freeze subarray%s\n", | |
405 | to_subarray(e, container)); | |
406 | ||
407 | /* thaw the partially frozen container */ | |
408 | for (e2 = ent; e2 && e2 != e; e2 = e2->next) { | |
409 | if (!is_container_member(e2, container)) | |
410 | continue; | |
411 | sysfs_free(sra); | |
412 | sra = sysfs_read(-1, e2->devnum, GET_VERSION); | |
413 | if (unblock_subarray(sra, freeze)) | |
414 | fprintf(stderr, Name ": Failed to unfreeze %s\n", e2->dev); | |
415 | } | |
416 | ||
417 | ping_monitor(container); /* cleared frozen */ | |
418 | rv = -1; | |
419 | } | |
420 | ||
421 | sysfs_free(sra); | |
422 | free_mdstat(ent); | |
bc77ed53 DW |
423 | |
424 | return rv; | |
425 | } | |
426 | ||
427 | void unblock_monitor(char *container, const int unfreeze) | |
428 | { | |
429 | struct mdstat_ent *ent, *e; | |
430 | struct mdinfo *sra = NULL; | |
5158aef1 | 431 | int to_ping = 0; |
bc77ed53 DW |
432 | |
433 | ent = mdstat_read(0, 0); | |
434 | if (!ent) { | |
435 | fprintf(stderr, Name | |
436 | ": failed to read /proc/mdstat while unblocking container\n"); | |
437 | return; | |
438 | } | |
439 | ||
440 | /* unfreeze container contents */ | |
441 | for (e = ent; e; e = e->next) { | |
442 | if (!is_container_member(e, container)) | |
443 | continue; | |
444 | sysfs_free(sra); | |
5158aef1 AK |
445 | sra = sysfs_read(-1, e->devnum, GET_VERSION|GET_LEVEL); |
446 | if (sra->array.level > 0) | |
447 | to_ping++; | |
bc77ed53 DW |
448 | if (unblock_subarray(sra, unfreeze)) |
449 | fprintf(stderr, Name ": Failed to unfreeze %s\n", e->dev); | |
450 | } | |
5158aef1 AK |
451 | if (to_ping) |
452 | ping_monitor(container); | |
bc77ed53 DW |
453 | |
454 | sysfs_free(sra); | |
455 | free_mdstat(ent); | |
456 | } | |
457 | ||
458 | ||
459 | ||
313a4a82 DW |
460 | /* give the manager a chance to view the updated container state. This |
461 | * would naturally happen due to the manager noticing a change in | |
462 | * /proc/mdstat; however, pinging encourages this detection to happen | |
463 | * while an exclusive open() on the container is active | |
464 | */ | |
465 | int ping_manager(char *devname) | |
466 | { | |
467 | int sfd = connect_monitor(devname); | |
468 | struct metadata_update msg = { .len = -1 }; | |
469 | int err = 0; | |
470 | ||
471 | if (sfd < 0) | |
472 | return sfd; | |
473 | ||
474 | err = send_message(sfd, &msg, 20); | |
475 | ||
476 | /* check the reply */ | |
477 | if (!err && wait_reply(sfd, 20) != 0) | |
478 | err = -1; | |
479 | ||
480 | close(sfd); | |
481 | return err; | |
482 | } |