[thirdparty/mdadm.git] / msg.c

/*
 * Copyright (C) 2008 Intel Corporation
 *
 *	mdmon socket / message handling
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms and conditions of the GNU General Public License,
 * version 2, as published by the Free Software Foundation.
 *
 * This program is distributed in the hope it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * You should have received a copy of the GNU General Public License along with
 * this program; if not, write to the Free Software Foundation, Inc.,
 * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
 */
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
#include <unistd.h>
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/un.h>
#include "mdadm.h"
#include "mdmon.h"

static const __u32 start_magic = 0x5a5aa5a5;
static const __u32 end_magic = 0xa5a55a5a;

static int send_buf(int fd, const void* buf, int len, int tmo)
{
	fd_set set;
	int rv;
	struct timeval timeout = {tmo, 0};
	struct timeval *ptmo = tmo ? &timeout : NULL;

	while (len) {
		FD_ZERO(&set);
		FD_SET(fd, &set);
		rv = select(fd+1, NULL, &set, NULL, ptmo);
		if (rv <= 0)
			return -1;
		rv = write(fd, buf, len);
		if (rv <= 0)
			return -1;
		len -= rv;
		buf += rv;
	}
	return 0;
}

static int recv_buf(int fd, void* buf, int len, int tmo)
{
	fd_set set;
	int rv;
	struct timeval timeout = {tmo, 0};
	struct timeval *ptmo = tmo ? &timeout : NULL;

	while (len) {
		FD_ZERO(&set);
		FD_SET(fd, &set);
		rv = select(fd+1, &set, NULL, NULL, ptmo);
		if (rv <= 0)
			return -1;
		rv = read(fd, buf, len);
		if (rv <= 0)
			return -1;
		len -= rv;
		buf += rv;
	}
	return 0;
}

int send_message(int fd, struct metadata_update *msg, int tmo)
{
	__s32 len = msg->len;
	int rv;

	rv = send_buf(fd, &start_magic, 4, tmo);
	rv = rv ?: send_buf(fd, &len, 4, tmo);
	if (len > 0)
		rv = rv ?: send_buf(fd, msg->buf, msg->len, tmo);
	rv = send_buf(fd, &end_magic, 4, tmo);

	return rv;
}

int receive_message(int fd, struct metadata_update *msg, int tmo)
{
	__u32 magic;
	__s32 len;
	int rv;

	rv = recv_buf(fd, &magic, 4, tmo);
	if (rv < 0 || magic != start_magic)
		return -1;
	rv = recv_buf(fd, &len, 4, tmo);
	if (rv < 0 || len > MSG_MAX_LEN)
		return -1;
	if (len > 0) {
		msg->buf = xmalloc(len);
		rv = recv_buf(fd, msg->buf, len, tmo);
		if (rv < 0) {
			free(msg->buf);
			return -1;
		}
	} else
		msg->buf = NULL;
	rv = recv_buf(fd, &magic, 4, tmo);
	if (rv < 0 || magic != end_magic) {
		free(msg->buf);
		return -1;
	}
	msg->len = len;
	return 0;
}

int ack(int fd, int tmo)
{
	struct metadata_update msg = { .len = 0 };

	return send_message(fd, &msg, tmo);
}

int wait_reply(int fd, int tmo)
{
	struct metadata_update msg;
	int err = receive_message(fd, &msg, tmo);

	/* mdmon sent extra data, but caller only cares that we got a
	 * successful reply
	 */
	if (err == 0 && msg.len > 0)
		free(msg.buf);

	return err;
}

int connect_monitor(char *devname)
{
	char path[100];
	int sfd;
	long fl;
	struct sockaddr_un addr;
	int pos;
	char *c;

	pos = sprintf(path, "%s/", MDMON_DIR);
	if (is_subarray(devname)) {
		devname++;
		c = strchr(devname, '/');
		if (!c)
			return -1;
		snprintf(&path[pos], c - devname + 1, "%s", devname);
		pos += c - devname;
	} else
		pos += sprintf(&path[pos], "%s", devname);
	sprintf(&path[pos], ".sock");

	sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
	if (sfd < 0)
		return -1;

	addr.sun_family = PF_LOCAL;
	strcpy(addr.sun_path, path);
	if (connect(sfd, (struct sockaddr*)&addr, sizeof(addr)) < 0) {
		pr_err("Error connecting monitor with %s: %s\n",
			addr.sun_path, strerror(errno));
		close(sfd);
		return -1;
	}

	fl = fcntl(sfd, F_GETFL, 0);
	fl |= O_NONBLOCK;
	fcntl(sfd, F_SETFL, fl);

	return sfd;
}

int fping_monitor(int sfd)
{
	int err = 0;

	if (sfd < 0)
		return sfd;

	/* try to ping existing socket */
	if (ack(sfd, 20) != 0)
		err = -1;

	/* check the reply */
	if (!err && wait_reply(sfd, 20) != 0)
		err = -1;

	return err;
}

/* give the monitor a chance to update the metadata */
int ping_monitor(char *devname)
{
	int sfd = connect_monitor(devname);
	int err;

	if (sfd >= 0) {
		err = fping_monitor(sfd);
		close(sfd);
	} else
		err = -1;

	return err;
}

static char *ping_monitor_version(char *devname)
{
	int sfd = connect_monitor(devname);
	struct metadata_update msg;
	int err = 0;

	if (sfd < 0)
		return NULL;

	if (ack(sfd, 20) != 0)
		err = -1;

	if (!err && receive_message(sfd, &msg, 20) != 0)
		err = -1;

	close(sfd);

	if (err || !msg.len || !msg.buf)
		return NULL;
	return msg.buf;
}

int unblock_subarray(struct mdinfo *sra, const int unfreeze)
{
	char buf[64];
	int rc = 0;

	if (sra) {
		sprintf(buf, "external:%s\n", sra->text_version);
		buf[9] = '/';
	} else
		buf[9] = '-';

	if (buf[9] == '-' ||
	    sysfs_set_str(sra, NULL, "metadata_version", buf) ||
	    (unfreeze &&
	     sysfs_attribute_available(sra, NULL, "sync_action") &&
	     sysfs_set_str(sra, NULL, "sync_action", "idle")))
		rc = -1;
	return rc;
}

int block_subarray(struct mdinfo *sra)
{
	char buf[64];
	int rc = 0;

	sprintf(buf, "external:%s\n", sra->text_version);
	buf[9] = '-';
	if (sysfs_set_str(sra, NULL, "metadata_version", buf))
		rc = -1;

	return rc;
}

/* check mdmon version if it supports
 * array blocking mechanism
 */
int check_mdmon_version(char *container)
{
	char *version = NULL;

	if (!mdmon_running(container)) {
		/* if mdmon is not active we assume that any instance that is
		 * later started will match the current mdadm version, if this
		 * assumption is violated we may inadvertantly rebuild an array
		 * that was meant for reshape, or start rebuild on a spare that
		 * was to be moved to another container
		 */
		/* pass */;
	} else {
		int ver;

		version = ping_monitor_version(container);
		ver = version ? mdadm_version(version) : -1;
		free(version);
		if (ver < 3002000) {
			pr_err("mdmon instance for %s cannot be disabled\n",
			       container);
			return -1;
		}
	}

	return 0;
}

/**
 * block_monitor - prevent mdmon spare assignment
 * @container - container to block
 * @freeze - flag to additionally freeze sync_action
 *
 * This is used by the reshape code to freeze the container, and the
 * auto-rebuild implementation to atomically move spares.
 * In both cases we need to stop mdmon from assigning spares to replace
 * failed devices as we might have other plans for the spare.
 * For the reshape case we also need to 'freeze' sync_action so that
 * no recovery happens until we have fully prepared for the reshape.
 *
 * We tell mdmon that the array is frozen by marking the 'metadata' name
 * with a leading '-'.  The previously told mdmon "Don't make this array
 * read/write, leave it readonly".  Now it means a more general "Don't
 * reconfigure this array at all".
 * As older versions of mdmon (which might run from initrd) don't understand
 * this, we first check that the running mdmon is new enough.
 */
int block_monitor(char *container, const int freeze)
{
	struct mdstat_ent *ent, *e, *e2;
	struct mdinfo *sra = NULL;
	char buf[64];
	int rv = 0;

	if (check_mdmon_version(container))
		return -1;

	ent = mdstat_read(0, 0);
	if (!ent) {
		pr_err("failed to read /proc/mdstat while disabling mdmon\n");
		return -1;
	}

	/* freeze container contents */
	for (e = ent; e; e = e->next) {
		if (!is_container_member(e, container))
			continue;
		sysfs_free(sra);
		sra = sysfs_read(-1, e->devnm, GET_VERSION);
		if (!sra) {
			pr_err("failed to read sysfs for subarray%s\n",
			       to_subarray(e, container));
			break;
		}
		/* can't reshape an array that we can't monitor */
		if (sra->text_version[0] == '-')
			break;

		if (freeze && sysfs_freeze_array(sra) < 1)
			break;
		/* flag this array to not be modified by mdmon (close race with
		 * takeover in reshape case and spare reassignment in the
		 * auto-rebuild case)
		 */
		if (block_subarray(sra))
			break;
		ping_monitor(container);

		/* check that we did not race with recovery */
		if ((freeze &&
		     !sysfs_attribute_available(sra, NULL, "sync_action")) ||
		    (freeze &&
		     sysfs_attribute_available(sra, NULL, "sync_action") &&
		     sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 &&
		     strcmp(buf, "frozen\n") == 0))
			/* pass */;
		else {
			unblock_subarray(sra, 0);
			break;
		}
		/* Double check against races - there should be no spares
		 * or part-spares
		 */
		sysfs_free(sra);
		sra = sysfs_read(-1, e->devnm, GET_DEVS | GET_STATE);
		if (sra && sra->array.spare_disks > 0) {
			unblock_subarray(sra, freeze);
			break;
		}
	}

	if (e) {
		pr_err("failed to freeze subarray%s\n",
			to_subarray(e, container));

		/* thaw the partially frozen container */
		for (e2 = ent; e2 && e2 != e; e2 = e2->next) {
			if (!is_container_member(e2, container))
				continue;
			sysfs_free(sra);
			sra = sysfs_read(-1, e2->devnm, GET_VERSION);
			if (unblock_subarray(sra, freeze))
				pr_err("Failed to unfreeze %s\n", e2->devnm);
		}

		ping_monitor(container); /* cleared frozen */
		rv = -1;
	}

	sysfs_free(sra);
	free_mdstat(ent);

	return rv;
}

void unblock_monitor(char *container, const int unfreeze)
{
	struct mdstat_ent *ent, *e;
	struct mdinfo *sra = NULL;
	int to_ping = 0;

	ent = mdstat_read(0, 0);
	if (!ent) {
		pr_err("failed to read /proc/mdstat while unblocking container\n");
		return;
	}

	/* unfreeze container contents */
	for (e = ent; e; e = e->next) {
		if (!is_container_member(e, container))
			continue;
		sysfs_free(sra);
		sra = sysfs_read(-1, e->devnm, GET_VERSION|GET_LEVEL);
		if (!sra)
			continue;
		if (sra->array.level > 0)
			to_ping++;
		if (unblock_subarray(sra, unfreeze))
			pr_err("Failed to unfreeze %s\n", e->devnm);
	}
	if (to_ping)
		ping_monitor(container);

	sysfs_free(sra);
	free_mdstat(ent);
}

/* give the manager a chance to view the updated container state.  This
 * would naturally happen due to the manager noticing a change in
 * /proc/mdstat; however, pinging encourages this detection to happen
 * while an exclusive open() on the container is active
 */
int ping_manager(char *devname)
{
	int sfd = connect_monitor(devname);
	struct metadata_update msg = { .len = -1 };
	int err = 0;

	if (sfd < 0)
		return sfd;

	err = send_message(sfd, &msg, 20);

	/* check the reply */
	if (!err && wait_reply(sfd, 20) != 0)
		err = -1;

	close(sfd);
	return err;
}

/* using takeover operation for grow purposes, mdadm has to be sure
 * that mdmon processes all updates, and if necessary it will be closed
 * at takeover to raid0 operation
  */
void flush_mdmon(char *container)
{
	ping_manager(container);
	ping_monitor(container);
}
Commit	Line	Data
f7dd881f DW	1	/*
	2	* Copyright (C) 2008 Intel Corporation
	3	*
1011e834	4	* mdmon socket / message handling
f7dd881f DW	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	}
f7dd881f DW	80
bfa44e2e	81	int send_message(int fd, struct metadata_update *msg, int tmo)
f7dd881f	82	{
313a4a82	83	__s32 len = msg->len;
bfa44e2e NB	84	int rv;
	85
	86	rv = send_buf(fd, &start_magic, 4, tmo);
	87	rv = rv ?: send_buf(fd, &len, 4, tmo);
313a4a82	88	if (len > 0)
bfa44e2e NB	89	rv = rv ?: send_buf(fd, msg->buf, msg->len, tmo);
	90	rv = send_buf(fd, &end_magic, 4, tmo);
	91
	92	return rv;
f7dd881f DW	93	}
f7dd881f DW	94
bfa44e2e	95	int receive_message(int fd, struct metadata_update *msg, int tmo)
f7dd881f	96	{
bfa44e2e	97	__u32 magic;
313a4a82	98	__s32 len;
bfa44e2e NB	99	int rv;
	100
	101	rv = recv_buf(fd, &magic, 4, tmo);
	102	if (rv < 0 \|\| magic != start_magic)
	103	return -1;
	104	rv = recv_buf(fd, &len, 4, tmo);
	105	if (rv < 0 \|\| len > MSG_MAX_LEN)
f7dd881f	106	return -1;
313a4a82	107	if (len > 0) {
503975b9	108	msg->buf = xmalloc(len);
bfa44e2e NB	109	rv = recv_buf(fd, msg->buf, len, tmo);
	110	if (rv < 0) {
	111	free(msg->buf);
	112	return -1;
	113	}
	114	} else
	115	msg->buf = NULL;
	116	rv = recv_buf(fd, &magic, 4, tmo);
	117	if (rv < 0 \|\| magic != end_magic) {
	118	free(msg->buf);
	119	return -1;
	120	}
	121	msg->len = len;
	122	return 0;
f7dd881f DW	123	}
f7dd881f DW	124
bfa44e2e	125	int ack(int fd, int tmo)
f7dd881f	126	{
bfa44e2e	127	struct metadata_update msg = { .len = 0 };
f7dd881f DW	128
	129	return send_message(fd, &msg, tmo);
	130	}
	131
bfa44e2e	132	int wait_reply(int fd, int tmo)
f7dd881f	133	{
bfa44e2e	134	struct metadata_update msg;
bc77ed53 DW	135	int err = receive_message(fd, &msg, tmo);
	136
	137	/* mdmon sent extra data, but caller only cares that we got a
	138	* successful reply
	139	*/
	140	if (err == 0 && msg.len > 0)
	141	free(msg.buf);
	142
	143	return err;
f7dd881f DW	144	}
f7dd881f DW	145
f7dd881f DW	146	int connect_monitor(char *devname)
	147	{
	148	char path[100];
	149	int sfd;
	150	long fl;
	151	struct sockaddr_un addr;
c94709e8 DW	152	int pos;
	153	char *c;
	154
753cf905	155	pos = sprintf(path, "%s/", MDMON_DIR);
c94709e8 DW	156	if (is_subarray(devname)) {
	157	devname++;
	158	c = strchr(devname, '/');
	159	if (!c)
	160	return -1;
	161	snprintf(&path[pos], c - devname + 1, "%s", devname);
	162	pos += c - devname;
	163	} else
	164	pos += sprintf(&path[pos], "%s", devname);
	165	sprintf(&path[pos], ".sock");
f7dd881f	166
f7dd881f DW	167	sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
	168	if (sfd < 0)
	169	return -1;
	170
	171	addr.sun_family = PF_LOCAL;
	172	strcpy(addr.sun_path, path);
50d72ed4	173	if (connect(sfd, (struct sockaddr*)&addr, sizeof(addr)) < 0) {
4515fb28 XN	174	pr_err("Error connecting monitor with %s: %s\n",
4515fb28 XN	175	addr.sun_path, strerror(errno));
f7dd881f DW	176	close(sfd);
	177	return -1;
	178	}
	179
	180	fl = fcntl(sfd, F_GETFL, 0);
	181	fl \|= O_NONBLOCK;
	182	fcntl(sfd, F_SETFL, fl);
	183
	184	return sfd;
	185	}
	186
9f1da824	187	int fping_monitor(int sfd)
f7dd881f	188	{
f7dd881f DW	189	int err = 0;
	190
	191	if (sfd < 0)
	192	return sfd;
	193
	194	/* try to ping existing socket */
eb2c876f	195	if (ack(sfd, 20) != 0)
f7dd881f DW	196	err = -1;
	197
	198	/* check the reply */
eb2c876f	199	if (!err && wait_reply(sfd, 20) != 0)
f7dd881f DW	200	err = -1;
f7dd881f DW	201
9f1da824 DW	202	return err;
	203	}
	204
9f1da824 DW	205	/* give the monitor a chance to update the metadata */
	206	int ping_monitor(char *devname)
	207	{
	208	int sfd = connect_monitor(devname);
b36c8935 JS	209	int err;
	210
	211	if (sfd >= 0) {
	212	err = fping_monitor(sfd);
	213	close(sfd);
	214	} else
	215	err = -1;
9f1da824	216
f7dd881f DW	217	return err;
f7dd881f DW	218	}
313a4a82	219
bc77ed53 DW	220	static char ping_monitor_version(char devname)
	221	{
	222	int sfd = connect_monitor(devname);
	223	struct metadata_update msg;
	224	int err = 0;
	225
	226	if (sfd < 0)
	227	return NULL;
	228
	229	if (ack(sfd, 20) != 0)
	230	err = -1;
	231
	232	if (!err && receive_message(sfd, &msg, 20) != 0)
	233	err = -1;
	234
	235	close(sfd);
	236
	237	if (err \|\| !msg.len \|\| !msg.buf)
	238	return NULL;
	239	return msg.buf;
	240	}
	241
2a093697	242	int unblock_subarray(struct mdinfo *sra, const int unfreeze)
bc77ed53 DW	243	{
	244	char buf[64];
	245	int rc = 0;
	246
	247	if (sra) {
	248	sprintf(buf, "external:%s\n", sra->text_version);
	249	buf[9] = '/';
	250	} else
	251	buf[9] = '-';
	252
	253	if (buf[9] == '-' \|\|
	254	sysfs_set_str(sra, NULL, "metadata_version", buf) \|\|
	255	(unfreeze &&
	256	sysfs_attribute_available(sra, NULL, "sync_action") &&
	257	sysfs_set_str(sra, NULL, "sync_action", "idle")))
	258	rc = -1;
	259	return rc;
	260	}
	261
2a093697 AK	262	int block_subarray(struct mdinfo *sra)
	263	{
	264	char buf[64];
	265	int rc = 0;
	266
	267	sprintf(buf, "external:%s\n", sra->text_version);
	268	buf[9] = '-';
	269	if (sysfs_set_str(sra, NULL, "metadata_version", buf))
	270	rc = -1;
	271
	272	return rc;
	273	}
7aa437c2 AK	274
	275	/* check mdmon version if it supports
	276	* array blocking mechanism
	277	*/
	278	int check_mdmon_version(char *container)
	279	{
	280	char *version = NULL;
7aa437c2	281
4dd2df09	282	if (!mdmon_running(container)) {
7aa437c2 AK	283	/* if mdmon is not active we assume that any instance that is
	284	* later started will match the current mdadm version, if this
	285	* assumption is violated we may inadvertantly rebuild an array
	286	* that was meant for reshape, or start rebuild on a spare that
	287	* was to be moved to another container
	288	*/
	289	/* pass */;
	290	} else {
	291	int ver;
	292
	293	version = ping_monitor_version(container);
	294	ver = version ? mdadm_version(version) : -1;
	295	free(version);
	296	if (ver < 3002000) {
e7b84f9d N	297	pr_err("mdmon instance for %s cannot be disabled\n",
e7b84f9d N	298	container);
7aa437c2 AK	299	return -1;
	300	}
	301	}
	302
	303	return 0;
	304	}
	305
bc77ed53 DW	306	/**
	307	* block_monitor - prevent mdmon spare assignment
	308	* @container - container to block
	309	* @freeze - flag to additionally freeze sync_action
	310	*
	311	* This is used by the reshape code to freeze the container, and the
746a6567 N	312	* auto-rebuild implementation to atomically move spares.
	313	* In both cases we need to stop mdmon from assigning spares to replace
	314	* failed devices as we might have other plans for the spare.
	315	* For the reshape case we also need to 'freeze' sync_action so that
	316	* no recovery happens until we have fully prepared for the reshape.
	317	*
	318	* We tell mdmon that the array is frozen by marking the 'metadata' name
	319	* with a leading '-'. The previously told mdmon "Don't make this array
	320	* read/write, leave it readonly". Now it means a more general "Don't
	321	* reconfigure this array at all".
	322	* As older versions of mdmon (which might run from initrd) don't understand
	323	* this, we first check that the running mdmon is new enough.
bc77ed53 DW	324	*/
	325	int block_monitor(char *container, const int freeze)
	326	{
bc77ed53 DW	327	struct mdstat_ent ent, e, *e2;
bc77ed53 DW	328	struct mdinfo *sra = NULL;
bc77ed53 DW	329	char buf[64];
	330	int rv = 0;
	331
7aa437c2 AK	332	if (check_mdmon_version(container))
7aa437c2 AK	333	return -1;
bc77ed53 DW	334
	335	ent = mdstat_read(0, 0);
	336	if (!ent) {
e7b84f9d	337	pr_err("failed to read /proc/mdstat while disabling mdmon\n");
bc77ed53 DW	338	return -1;
	339	}
	340
	341	/* freeze container contents */
	342	for (e = ent; e; e = e->next) {
	343	if (!is_container_member(e, container))
	344	continue;
	345	sysfs_free(sra);
4dd2df09	346	sra = sysfs_read(-1, e->devnm, GET_VERSION);
bc77ed53	347	if (!sra) {
e7b84f9d N	348	pr_err("failed to read sysfs for subarray%s\n",
e7b84f9d N	349	to_subarray(e, container));
bc77ed53 DW	350	break;
	351	}
	352	/* can't reshape an array that we can't monitor */
	353	if (sra->text_version[0] == '-')
	354	break;
	355
	356	if (freeze && sysfs_freeze_array(sra) < 1)
	357	break;
	358	/* flag this array to not be modified by mdmon (close race with
	359	* takeover in reshape case and spare reassignment in the
	360	* auto-rebuild case)
	361	*/
2a093697	362	if (block_subarray(sra))
bc77ed53 DW	363	break;
	364	ping_monitor(container);
	365
	366	/* check that we did not race with recovery */
	367	if ((freeze &&
	368	!sysfs_attribute_available(sra, NULL, "sync_action")) \|\|
	369	(freeze &&
	370	sysfs_attribute_available(sra, NULL, "sync_action") &&
	371	sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 &&
	372	strcmp(buf, "frozen\n") == 0))
	373	/* pass */;
885f9845 N	374	else {
885f9845 N	375	unblock_subarray(sra, 0);
bc77ed53	376	break;
885f9845 N	377	}
	378	/* Double check against races - there should be no spares
	379	* or part-spares
	380	*/
	381	sysfs_free(sra);
4dd2df09	382	sra = sysfs_read(-1, e->devnm, GET_DEVS \| GET_STATE);
885f9845 N	383	if (sra && sra->array.spare_disks > 0) {
	384	unblock_subarray(sra, freeze);
	385	break;
	386	}
bc77ed53 DW	387	}
	388
	389	if (e) {
e7b84f9d	390	pr_err("failed to freeze subarray%s\n",
bc77ed53 DW	391	to_subarray(e, container));
	392
	393	/* thaw the partially frozen container */
	394	for (e2 = ent; e2 && e2 != e; e2 = e2->next) {
	395	if (!is_container_member(e2, container))
	396	continue;
	397	sysfs_free(sra);
4dd2df09	398	sra = sysfs_read(-1, e2->devnm, GET_VERSION);
bc77ed53	399	if (unblock_subarray(sra, freeze))
9581efb1	400	pr_err("Failed to unfreeze %s\n", e2->devnm);
bc77ed53 DW	401	}
	402
	403	ping_monitor(container); /* cleared frozen */
	404	rv = -1;
	405	}
	406
	407	sysfs_free(sra);
	408	free_mdstat(ent);
bc77ed53 DW	409
	410	return rv;
	411	}
	412
	413	void unblock_monitor(char *container, const int unfreeze)
	414	{
	415	struct mdstat_ent ent, e;
	416	struct mdinfo *sra = NULL;
5158aef1	417	int to_ping = 0;
bc77ed53 DW	418
	419	ent = mdstat_read(0, 0);
	420	if (!ent) {
e7b84f9d	421	pr_err("failed to read /proc/mdstat while unblocking container\n");
bc77ed53 DW	422	return;
	423	}
	424
	425	/* unfreeze container contents */
	426	for (e = ent; e; e = e->next) {
	427	if (!is_container_member(e, container))
	428	continue;
	429	sysfs_free(sra);
4dd2df09	430	sra = sysfs_read(-1, e->devnm, GET_VERSION\|GET_LEVEL);
29b59ca5 JS	431	if (!sra)
29b59ca5 JS	432	continue;
5158aef1 AK	433	if (sra->array.level > 0)
5158aef1 AK	434	to_ping++;
bc77ed53	435	if (unblock_subarray(sra, unfreeze))
9581efb1	436	pr_err("Failed to unfreeze %s\n", e->devnm);
bc77ed53	437	}
5158aef1 AK	438	if (to_ping)
5158aef1 AK	439	ping_monitor(container);
bc77ed53 DW	440
	441	sysfs_free(sra);
	442	free_mdstat(ent);
	443	}
	444
313a4a82 DW	445	/* give the manager a chance to view the updated container state. This
	446	* would naturally happen due to the manager noticing a change in
	447	* /proc/mdstat; however, pinging encourages this detection to happen
	448	* while an exclusive open() on the container is active
	449	*/
	450	int ping_manager(char *devname)
	451	{
	452	int sfd = connect_monitor(devname);
	453	struct metadata_update msg = { .len = -1 };
	454	int err = 0;
	455
	456	if (sfd < 0)
	457	return sfd;
	458
	459	err = send_message(sfd, &msg, 20);
	460
	461	/* check the reply */
	462	if (!err && wait_reply(sfd, 20) != 0)
	463	err = -1;
	464
	465	close(sfd);
	466	return err;
	467	}
78340e26 AK	468
	469	/* using takeover operation for grow purposes, mdadm has to be sure
	470	* that mdmon processes all updates, and if necessary it will be closed
	471	* at takeover to raid0 operation
	472	*/
	473	void flush_mdmon(char *container)
	474	{
	475	ping_manager(container);
	476	ping_monitor(container);
	477	}