[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 = malloc(len);
		if (msg->buf == NULL)
			return -1;
		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, &addr, sizeof(addr)) < 0) {
		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;
}

/* ping monitor using device number */
int ping_monitor_by_id(int devnum)
{
	int err = -1;
	char *container = devnum2devname(devnum);

	if (container) {
		err = ping_monitor(container);
		free(container);
	}

	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;
	int devnum = devname2devnum(container);

	if (!mdmon_running(devnum)) {
		/* 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->devnum, 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->devnum, 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->devnum, GET_VERSION);
			if (unblock_subarray(sra, freeze))
				pr_err("Failed to unfreeze %s\n", e2->dev);
		}

		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->devnum, 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->dev);
	}
	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	*
	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	}
f7dd881f DW	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	}
f7dd881f DW	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	}
f7dd881f DW	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;
f7dd881f DW	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);
b36c8935 JS	210	int err;
	211
	212	if (sfd >= 0) {
	213	err = fping_monitor(sfd);
	214	close(sfd);
	215	} else
	216	err = -1;
9f1da824	217
f7dd881f DW	218	return err;
f7dd881f DW	219	}
313a4a82	220
983fff45 AK	221	/* ping monitor using device number */
	222	int ping_monitor_by_id(int devnum)
	223	{
	224	int err = -1;
	225	char *container = devnum2devname(devnum);
	226
	227	if (container) {
	228	err = ping_monitor(container);
	229	free(container);
	230	}
	231
	232	return err;
	233	}
	234
bc77ed53 DW	235	static char ping_monitor_version(char devname)
	236	{
	237	int sfd = connect_monitor(devname);
	238	struct metadata_update msg;
	239	int err = 0;
	240
	241	if (sfd < 0)
	242	return NULL;
	243
	244	if (ack(sfd, 20) != 0)
	245	err = -1;
	246
	247	if (!err && receive_message(sfd, &msg, 20) != 0)
	248	err = -1;
	249
	250	close(sfd);
	251
	252	if (err \|\| !msg.len \|\| !msg.buf)
	253	return NULL;
	254	return msg.buf;
	255	}
	256
2a093697	257	int unblock_subarray(struct mdinfo *sra, const int unfreeze)
bc77ed53 DW	258	{
	259	char buf[64];
	260	int rc = 0;
	261
	262	if (sra) {
	263	sprintf(buf, "external:%s\n", sra->text_version);
	264	buf[9] = '/';
	265	} else
	266	buf[9] = '-';
	267
	268	if (buf[9] == '-' \|\|
	269	sysfs_set_str(sra, NULL, "metadata_version", buf) \|\|
	270	(unfreeze &&
	271	sysfs_attribute_available(sra, NULL, "sync_action") &&
	272	sysfs_set_str(sra, NULL, "sync_action", "idle")))
	273	rc = -1;
	274	return rc;
	275	}
	276
2a093697 AK	277	int block_subarray(struct mdinfo *sra)
	278	{
	279	char buf[64];
	280	int rc = 0;
	281
	282	sprintf(buf, "external:%s\n", sra->text_version);
	283	buf[9] = '-';
	284	if (sysfs_set_str(sra, NULL, "metadata_version", buf))
	285	rc = -1;
	286
	287	return rc;
	288	}
7aa437c2 AK	289
	290	/* check mdmon version if it supports
	291	* array blocking mechanism
	292	*/
	293	int check_mdmon_version(char *container)
	294	{
	295	char *version = NULL;
	296	int devnum = devname2devnum(container);
	297
	298	if (!mdmon_running(devnum)) {
	299	/* if mdmon is not active we assume that any instance that is
	300	* later started will match the current mdadm version, if this
	301	* assumption is violated we may inadvertantly rebuild an array
	302	* that was meant for reshape, or start rebuild on a spare that
	303	* was to be moved to another container
	304	*/
	305	/* pass */;
	306	} else {
	307	int ver;
	308
	309	version = ping_monitor_version(container);
	310	ver = version ? mdadm_version(version) : -1;
	311	free(version);
	312	if (ver < 3002000) {
e7b84f9d N	313	pr_err("mdmon instance for %s cannot be disabled\n",
e7b84f9d N	314	container);
7aa437c2 AK	315	return -1;
	316	}
	317	}
	318
	319	return 0;
	320	}
	321
bc77ed53 DW	322	/**
	323	* block_monitor - prevent mdmon spare assignment
	324	* @container - container to block
	325	* @freeze - flag to additionally freeze sync_action
	326	*
	327	* This is used by the reshape code to freeze the container, and the
746a6567 N	328	* auto-rebuild implementation to atomically move spares.
	329	* In both cases we need to stop mdmon from assigning spares to replace
	330	* failed devices as we might have other plans for the spare.
	331	* For the reshape case we also need to 'freeze' sync_action so that
	332	* no recovery happens until we have fully prepared for the reshape.
	333	*
	334	* We tell mdmon that the array is frozen by marking the 'metadata' name
	335	* with a leading '-'. The previously told mdmon "Don't make this array
	336	* read/write, leave it readonly". Now it means a more general "Don't
	337	* reconfigure this array at all".
	338	* As older versions of mdmon (which might run from initrd) don't understand
	339	* this, we first check that the running mdmon is new enough.
bc77ed53 DW	340	*/
	341	int block_monitor(char *container, const int freeze)
	342	{
bc77ed53 DW	343	struct mdstat_ent ent, e, *e2;
bc77ed53 DW	344	struct mdinfo *sra = NULL;
bc77ed53 DW	345	char buf[64];
	346	int rv = 0;
	347
7aa437c2 AK	348	if (check_mdmon_version(container))
7aa437c2 AK	349	return -1;
bc77ed53 DW	350
	351	ent = mdstat_read(0, 0);
	352	if (!ent) {
e7b84f9d	353	pr_err("failed to read /proc/mdstat while disabling mdmon\n");
bc77ed53 DW	354	return -1;
	355	}
	356
	357	/* freeze container contents */
	358	for (e = ent; e; e = e->next) {
	359	if (!is_container_member(e, container))
	360	continue;
	361	sysfs_free(sra);
	362	sra = sysfs_read(-1, e->devnum, GET_VERSION);
	363	if (!sra) {
e7b84f9d N	364	pr_err("failed to read sysfs for subarray%s\n",
e7b84f9d N	365	to_subarray(e, container));
bc77ed53 DW	366	break;
	367	}
	368	/* can't reshape an array that we can't monitor */
	369	if (sra->text_version[0] == '-')
	370	break;
	371
	372	if (freeze && sysfs_freeze_array(sra) < 1)
	373	break;
	374	/* flag this array to not be modified by mdmon (close race with
	375	* takeover in reshape case and spare reassignment in the
	376	* auto-rebuild case)
	377	*/
2a093697	378	if (block_subarray(sra))
bc77ed53 DW	379	break;
	380	ping_monitor(container);
	381
	382	/* check that we did not race with recovery */
	383	if ((freeze &&
	384	!sysfs_attribute_available(sra, NULL, "sync_action")) \|\|
	385	(freeze &&
	386	sysfs_attribute_available(sra, NULL, "sync_action") &&
	387	sysfs_get_str(sra, NULL, "sync_action", buf, 20) > 0 &&
	388	strcmp(buf, "frozen\n") == 0))
	389	/* pass */;
885f9845 N	390	else {
885f9845 N	391	unblock_subarray(sra, 0);
bc77ed53	392	break;
885f9845 N	393	}
	394	/* Double check against races - there should be no spares
	395	* or part-spares
	396	*/
	397	sysfs_free(sra);
	398	sra = sysfs_read(-1, e->devnum, GET_DEVS \| GET_STATE);
	399	if (sra && sra->array.spare_disks > 0) {
	400	unblock_subarray(sra, freeze);
	401	break;
	402	}
bc77ed53 DW	403	}
	404
	405	if (e) {
e7b84f9d	406	pr_err("failed to freeze subarray%s\n",
bc77ed53 DW	407	to_subarray(e, container));
	408
	409	/* thaw the partially frozen container */
	410	for (e2 = ent; e2 && e2 != e; e2 = e2->next) {
	411	if (!is_container_member(e2, container))
	412	continue;
	413	sysfs_free(sra);
	414	sra = sysfs_read(-1, e2->devnum, GET_VERSION);
	415	if (unblock_subarray(sra, freeze))
e7b84f9d	416	pr_err("Failed to unfreeze %s\n", e2->dev);
bc77ed53 DW	417	}
	418
	419	ping_monitor(container); /* cleared frozen */
	420	rv = -1;
	421	}
	422
	423	sysfs_free(sra);
	424	free_mdstat(ent);
bc77ed53 DW	425
	426	return rv;
	427	}
	428
	429	void unblock_monitor(char *container, const int unfreeze)
	430	{
	431	struct mdstat_ent ent, e;
	432	struct mdinfo *sra = NULL;
5158aef1	433	int to_ping = 0;
bc77ed53 DW	434
	435	ent = mdstat_read(0, 0);
	436	if (!ent) {
e7b84f9d	437	pr_err("failed to read /proc/mdstat while unblocking container\n");
bc77ed53 DW	438	return;
	439	}
	440
	441	/* unfreeze container contents */
	442	for (e = ent; e; e = e->next) {
	443	if (!is_container_member(e, container))
	444	continue;
	445	sysfs_free(sra);
5158aef1	446	sra = sysfs_read(-1, e->devnum, GET_VERSION\|GET_LEVEL);
29b59ca5 JS	447	if (!sra)
29b59ca5 JS	448	continue;
5158aef1 AK	449	if (sra->array.level > 0)
5158aef1 AK	450	to_ping++;
bc77ed53	451	if (unblock_subarray(sra, unfreeze))
e7b84f9d	452	pr_err("Failed to unfreeze %s\n", e->dev);
bc77ed53	453	}
5158aef1 AK	454	if (to_ping)
5158aef1 AK	455	ping_monitor(container);
bc77ed53 DW	456
	457	sysfs_free(sra);
	458	free_mdstat(ent);
	459	}
	460
	461
	462
313a4a82 DW	463	/* give the manager a chance to view the updated container state. This
	464	* would naturally happen due to the manager noticing a change in
	465	* /proc/mdstat; however, pinging encourages this detection to happen
	466	* while an exclusive open() on the container is active
	467	*/
	468	int ping_manager(char *devname)
	469	{
	470	int sfd = connect_monitor(devname);
	471	struct metadata_update msg = { .len = -1 };
	472	int err = 0;
	473
	474	if (sfd < 0)
	475	return sfd;
	476
	477	err = send_message(sfd, &msg, 20);
	478
	479	/* check the reply */
	480	if (!err && wait_reply(sfd, 20) != 0)
	481	err = -1;
	482
	483	close(sfd);
	484	return err;
	485	}
78340e26 AK	486
	487	/* using takeover operation for grow purposes, mdadm has to be sure
	488	* that mdmon processes all updates, and if necessary it will be closed
	489	* at takeover to raid0 operation
	490	*/
	491	void flush_mdmon(char *container)
	492	{
	493	ping_manager(container);
	494	ping_monitor(container);
	495	}