2 * Copyright (C) 2008 Intel Corporation
4 * mdmon socket / message handling
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.
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
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.
28 #include <sys/types.h>
29 #include <sys/socket.h>
34 static const __u32 start_magic = 0x5a5aa5a5;
35 static const __u32 end_magic = 0xa5a55a5a;
37 static int send_buf(int fd, const void* buf, int len, int tmo)
41 struct timeval timeout = {tmo, 0};
42 struct timeval *ptmo = tmo ? &timeout : NULL;
47 rv = select(fd+1, NULL, &set, NULL, ptmo);
50 rv = write(fd, buf, len);
59 static int recv_buf(int fd, void* buf, int len, int tmo)
63 struct timeval timeout = {tmo, 0};
64 struct timeval *ptmo = tmo ? &timeout : NULL;
69 rv = select(fd+1, &set, NULL, NULL, ptmo);
72 rv = read(fd, buf, len);
81 int send_message(int fd, struct metadata_update *msg, int tmo)
86 rv = send_buf(fd, &start_magic, 4, tmo);
87 rv = rv ?: send_buf(fd, &len, 4, tmo);
89 rv = rv ?: send_buf(fd, msg->buf, msg->len, tmo);
90 rv = send_buf(fd, &end_magic, 4, tmo);
95 int receive_message(int fd, struct metadata_update *msg, int tmo)
101 rv = recv_buf(fd, &magic, 4, tmo);
102 if (rv < 0 || magic != start_magic)
104 rv = recv_buf(fd, &len, 4, tmo);
105 if (rv < 0 || len > MSG_MAX_LEN)
108 msg->buf = xmalloc(len);
109 rv = recv_buf(fd, msg->buf, len, tmo);
116 rv = recv_buf(fd, &magic, 4, tmo);
117 if (rv < 0 || magic != end_magic) {
125 int ack(int fd, int tmo)
127 struct metadata_update msg = { .len = 0 };
129 return send_message(fd, &msg, tmo);
132 int wait_reply(int fd, int tmo)
134 struct metadata_update msg;
135 int err = receive_message(fd, &msg, tmo);
137 /* mdmon sent extra data, but caller only cares that we got a
140 if (err == 0 && msg.len > 0)
146 int connect_monitor(char *devname)
151 struct sockaddr_un addr;
155 pos = sprintf(path, "%s/", MDMON_DIR);
156 if (is_subarray(devname)) {
158 c = strchr(devname, '/');
161 snprintf(&path[pos], c - devname + 1, "%s", devname);
164 pos += sprintf(&path[pos], "%s", devname);
165 sprintf(&path[pos], ".sock");
167 sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
171 addr.sun_family = PF_LOCAL;
172 strcpy(addr.sun_path, path);
173 if (connect(sfd, (struct sockaddr*)&addr, sizeof(addr)) < 0) {
178 fl = fcntl(sfd, F_GETFL, 0);
180 fcntl(sfd, F_SETFL, fl);
185 int fping_monitor(int sfd)
192 /* try to ping existing socket */
193 if (ack(sfd, 20) != 0)
196 /* check the reply */
197 if (!err && wait_reply(sfd, 20) != 0)
203 /* give the monitor a chance to update the metadata */
204 int ping_monitor(char *devname)
206 int sfd = connect_monitor(devname);
210 err = fping_monitor(sfd);
218 static char *ping_monitor_version(char *devname)
220 int sfd = connect_monitor(devname);
221 struct metadata_update msg;
227 if (ack(sfd, 20) != 0)
230 if (!err && receive_message(sfd, &msg, 20) != 0)
235 if (err || !msg.len || !msg.buf)
240 int unblock_subarray(struct mdinfo *sra, const int unfreeze)
246 sprintf(buf, "external:%s\n", sra->text_version);
252 sysfs_set_str(sra, NULL, "metadata_version", buf) ||
254 sysfs_attribute_available(sra, NULL, "sync_action") &&
255 sysfs_set_str(sra, NULL, "sync_action", "idle")))
260 int block_subarray(struct mdinfo *sra)
265 sprintf(buf, "external:%s\n", sra->text_version);
267 if (sysfs_set_str(sra, NULL, "metadata_version", buf))
273 /* check mdmon version if it supports
274 * array blocking mechanism
276 int check_mdmon_version(char *container)
278 char *version = NULL;
280 if (!mdmon_running(container)) {
281 /* if mdmon is not active we assume that any instance that is
282 * later started will match the current mdadm version, if this
283 * assumption is violated we may inadvertantly rebuild an array
284 * that was meant for reshape, or start rebuild on a spare that
285 * was to be moved to another container
291 version = ping_monitor_version(container);
292 ver = version ? mdadm_version(version) : -1;
295 pr_err("mdmon instance for %s cannot be disabled\n",
305 * block_monitor - prevent mdmon spare assignment
306 * @container - container to block
307 * @freeze - flag to additionally freeze sync_action
309 * This is used by the reshape code to freeze the container, and the
310 * auto-rebuild implementation to atomically move spares.
311 * In both cases we need to stop mdmon from assigning spares to replace
312 * failed devices as we might have other plans for the spare.
313 * For the reshape case we also need to 'freeze' sync_action so that
314 * no recovery happens until we have fully prepared for the reshape.
316 * We tell mdmon that the array is frozen by marking the 'metadata' name
317 * with a leading '-'. The previously told mdmon "Don't make this array
318 * read/write, leave it readonly". Now it means a more general "Don't
319 * reconfigure this array at all".
320 * As older versions of mdmon (which might run from initrd) don't understand
321 * this, we first check that the running mdmon is new enough.
323 int block_monitor(char *container, const int freeze)
325 struct mdstat_ent *ent, *e, *e2;
326 struct mdinfo *sra = NULL;
330 if (check_mdmon_version(container))
333 ent = mdstat_read(0, 0);
335 pr_err("failed to read /proc/mdstat while disabling mdmon\n");
339 /* freeze container contents */
340 for (e = ent; e; e = e->next) {
341 if (!is_container_member(e, container))
344 sra = sysfs_read(-1, e->devnm, GET_VERSION);
346 pr_err("failed to read sysfs for subarray%s\n",
347 to_subarray(e, container));
350 /* can't reshape an array that we can't monitor */
351 if (sra->text_version[0] == '-')
354 if (freeze && sysfs_freeze_array(sra) < 1)
356 /* flag this array to not be modified by mdmon (close race with
357 * takeover in reshape case and spare reassignment in the
360 if (block_subarray(sra))
362 ping_monitor(container);
364 /* check that we did not race with recovery */
366 !sysfs_attribute_available(sra, NULL, "sync_action")) ||
368 sysfs_attribute_available(sra, NULL, "sync_action") &&
369 sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 &&
370 strcmp(buf, "frozen\n") == 0))
373 unblock_subarray(sra, 0);
376 /* Double check against races - there should be no spares
380 sra = sysfs_read(-1, e->devnm, GET_DEVS | GET_STATE);
381 if (sra && sra->array.spare_disks > 0) {
382 unblock_subarray(sra, freeze);
388 pr_err("failed to freeze subarray%s\n",
389 to_subarray(e, container));
391 /* thaw the partially frozen container */
392 for (e2 = ent; e2 && e2 != e; e2 = e2->next) {
393 if (!is_container_member(e2, container))
396 sra = sysfs_read(-1, e2->devnm, GET_VERSION);
397 if (unblock_subarray(sra, freeze))
398 pr_err("Failed to unfreeze %s\n", e2->devnm);
401 ping_monitor(container); /* cleared frozen */
411 void unblock_monitor(char *container, const int unfreeze)
413 struct mdstat_ent *ent, *e;
414 struct mdinfo *sra = NULL;
417 ent = mdstat_read(0, 0);
419 pr_err("failed to read /proc/mdstat while unblocking container\n");
423 /* unfreeze container contents */
424 for (e = ent; e; e = e->next) {
425 if (!is_container_member(e, container))
428 sra = sysfs_read(-1, e->devnm, GET_VERSION|GET_LEVEL);
431 if (sra->array.level > 0)
433 if (unblock_subarray(sra, unfreeze))
434 pr_err("Failed to unfreeze %s\n", e->devnm);
437 ping_monitor(container);
443 /* give the manager a chance to view the updated container state. This
444 * would naturally happen due to the manager noticing a change in
445 * /proc/mdstat; however, pinging encourages this detection to happen
446 * while an exclusive open() on the container is active
448 int ping_manager(char *devname)
450 int sfd = connect_monitor(devname);
451 struct metadata_update msg = { .len = -1 };
457 err = send_message(sfd, &msg, 20);
459 /* check the reply */
460 if (!err && wait_reply(sfd, 20) != 0)
467 /* using takeover operation for grow purposes, mdadm has to be sure
468 * that mdmon processes all updates, and if necessary it will be closed
469 * at takeover to raid0 operation
471 void flush_mdmon(char *container)
473 ping_manager(container);
474 ping_monitor(container);