[thirdparty/mdadm.git] / managemon.c


/*
 * The management thread for monitoring active md arrays.
 * This thread does things which might block such as memory
 * allocation.
 * In particular:
 *
 * - Find out about new arrays in this container.
 *   Allocate the data structures and open the files.
 *
 *   For this we watch /proc/mdstat and find new arrays with
 *   metadata type that confirms sharing. e.g. "md4"
 *   When we find a new array we slip it into the list of
 *   arrays and signal 'monitor' by writing to a pipe.
 *
 * - Respond to reshape requests by allocating new data structures
 *   and opening new files.
 *
 *   These come as a change to raid_disks.  We allocate a new
 *   version of the data structures and slip it into the list.
 *   'monitor' will notice and release the old version.
 *   Changes to level, chunksize, layout.. do not need re-allocation.
 *   Reductions in raid_disks don't really either, but we handle
 *   them the same way for consistency.
 *
 * - When a device is added to the container, we add it to the metadata
 *   as a spare.
 *
 * - assist with activating spares by opening relevant sysfs file.
 *
 * - Pass on metadata updates from external programs such as
 *   mdadm creating a new array.
 *
 *   This is most-messy.
 *   It might involve adding a new array or changing the status of
 *   a spare, or any reconfig that the kernel doesn't get involved in.
 *
 *   The required updates are received via a named pipe.  There will
 *   be one named pipe for each container. Each message contains a
 *   sync marker: 0x5a5aa5a5, A byte count, and the message.  This is
 *   passed to the metadata handler which will interpret and process it.
 *   For 'DDF' messages are internal data blocks with the leading
 *   'magic number' signifying what sort of data it is.
 *
 */

/*
 * We select on /proc/mdstat and the named pipe.
 * We create new arrays or updated version of arrays and slip
 * them into the head of the list, then signal 'monitor' via a pipe write.
 * 'monitor' will notice and place the old array on a return list.
 * Metadata updates are placed on a queue just like they arrive
 * from the named pipe.
 *
 * When new arrays are found based on correct metadata string, we
 * need to identify them with an entry in the metadata.  Maybe we require
 * the metadata to be mdX/NN  when NN is the index into an appropriate table.
 *
 */

/*
 * List of tasks:
 * - Watch for spares to be added to the container, and write updated
 *   metadata to them.
 * - Watch for new arrays using this container, confirm they match metadata
 *   and if so, start monitoring them
 * - Watch for spares being added to monitored arrays.  This shouldn't
 *   happen, as we should do all the adding.  Just remove them.
 * - Watch for change in raid-disks, chunk-size, etc.  Update metadata and
 *   start a reshape.
 */
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
#include	"mdadm.h"
#include	"mdmon.h"
#include	<sys/socket.h>
#include	<signal.h>

static void close_aa(struct active_array *aa)
{
	struct mdinfo *d;

	for (d = aa->info.devs; d; d = d->next)
		close(d->state_fd);

	close(aa->action_fd);
	close(aa->info.state_fd);
	close(aa->resync_start_fd);
}

static void free_aa(struct active_array *aa)
{
	/* Note that this doesn't close fds if they are being used
	 * by a clone.  ->container will be set for a clone
	 */
	if (!aa->container)
		close_aa(aa);
	while (aa->info.devs) {
		struct mdinfo *d = aa->info.devs;
		aa->info.devs = d->next;
		free(d);
	}
	free(aa);
}

static void write_wakeup(struct supertype *c)
{
	static struct md_generic_cmd cmd = { .action = md_action_ping_monitor };
	int err;

	active_cmd = &cmd;

	/* send the monitor thread a pointer to the ping action */
	write(c->mgr_pipe[1], &err, 1);
	read(c->mon_pipe[0], &err, 1);
}

static void remove_old(void)
{
	if (discard_this) {
		discard_this->next = NULL;
		free_aa(discard_this);
		if (pending_discard == discard_this)
			pending_discard = NULL;
		discard_this = NULL;
	}
}

static void replace_array(struct supertype *container,
			  struct active_array *old,
			  struct active_array *new)
{
	/* To replace an array, we add it to the top of the list
	 * marked with ->replaces to point to the original.
	 * 'monitor' will take the original out of the list
	 * and put it on 'discard_this'.  We take it from there
	 * and discard it.
	 */
	remove_old();
	while (pending_discard) {
		write_wakeup(container);
		while (discard_this == NULL)
			sleep(1);
		remove_old();
	}
	pending_discard = old;
	new->replaces = old;
	new->next = container->arrays;
	container->arrays = new;
	write_wakeup(container);
}

struct metadata_update *update_queue = NULL;
struct metadata_update *update_queue_handled = NULL;
struct metadata_update *update_queue_pending = NULL;

void check_update_queue(struct supertype *container)
{
	while (update_queue_handled) {
		struct metadata_update *this = update_queue_handled;
		update_queue_handled = this->next;
		free(this->buf);
		free(this);
	}
	if (update_queue == NULL &&
	    update_queue_pending) {
		update_queue = update_queue_pending;
		update_queue_pending = NULL;
		write_wakeup(container);
	}
}

void queue_metadata_update(struct metadata_update *mu)
{
	struct metadata_update **qp;

	qp = &update_queue_pending;
	while (*qp)
		qp = & ((*qp)->next);
	*qp = mu;
}

void wait_update_handled(void)
{
	/* Wait for any pending update to be handled by monitor.
	 * i.e. wait until update_queue is NULL
	 */
	while (update_queue)
		usleep(100 * 1000);
}

static void manage_container(struct mdstat_ent *mdstat,
			     struct supertype *container)
{
	/* The only thing of interest here is if a new device
	 * has been added to the container.  We add it to the
	 * array ignoring any metadata on it.
	 * FIXME should we look for compatible metadata and take hints
	 * about spare assignment.... probably not.
	 *
	 */
	if (mdstat->devcnt != container->devcnt) {
		/* read /sys/block/NAME/md/dev-??/block/dev to find out
		 * what is there, and compare with container->info.devs
		 * To see what is removed and what is added.
		 * These need to be remove from, or added to, the array
		 */
		// FIXME
		container->devcnt = mdstat->devcnt;
	}
}

static void manage_member(struct mdstat_ent *mdstat,
			  struct active_array *a)
{
	/* Compare mdstat info with known state of member array.
	 * We do not need to look for device state changes here, that
	 * is dealt with by the monitor.
	 *
	 * We just look for changes which suggest that a reshape is
	 * being requested.
	 * Unfortunately decreases in raid_disks don't show up in
	 * mdstat until the reshape completes FIXME.
	 */
	// FIXME
	a->info.array.raid_disks = mdstat->raid_disks;
	a->info.array.chunk_size = mdstat->chunk_size;
	// MORE

}

static void manage_new(struct mdstat_ent *mdstat,
		       struct supertype *container,
		       struct active_array *victim)
{
	/* A new array has appeared in this container.
	 * Hopefully it is already recorded in the metadata.
	 * Check, then create the new array to report it to
	 * the monitor.
	 */

	struct active_array *new;
	struct mdinfo *mdi, *di;
	char *inst;
	int i;

	new = malloc(sizeof(*new));

	memset(new, 0, sizeof(*new));

	new->devnum = mdstat->devnum;

	new->prev_state = new->curr_state = new->next_state = inactive;
	new->prev_action= new->curr_action= new->next_action= idle;

	new->container = container;

	inst = &mdstat->metadata_version[10+strlen(container->devname)+1];

	mdi = sysfs_read(-1, new->devnum,
			 GET_LEVEL|GET_CHUNK|GET_DISKS|
			 GET_DEVS|GET_OFFSET|GET_SIZE|GET_STATE);
	if (!mdi) {
		/* Eeek. Cannot monitor this array.
		 * Mark it to be ignored by setting container to NULL
		 */
		new->container = NULL;
		replace_array(container, victim, new);
		return;
	}

	new->info.array = mdi->array;

	for (i = 0; i < new->info.array.raid_disks; i++) {
		struct mdinfo *newd = malloc(sizeof(*newd));

		for (di = mdi->devs; di; di = di->next)
			if (i == di->disk.raid_disk)
				break;

		if (di) {
			memcpy(newd, di, sizeof(*newd));

			sprintf(newd->sys_name, "rd%d", i);

			newd->state_fd = sysfs_open(new->devnum,
						    newd->sys_name,
						    "state");

			newd->prev_state = read_dev_state(newd->state_fd);
			newd->curr_state = newd->curr_state;
		} else {
			newd->state_fd = -1;
		}
		newd->next = new->info.devs;
		new->info.devs = newd;
	}
	new->action_fd = sysfs_open(new->devnum, NULL, "sync_action");
	new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state");
	new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start");
	new->resync_start = 0;

	sysfs_free(mdi);
	// finds and compares.
	if (container->ss->open_new(container, new, inst) < 0) {
		// FIXME close all those files
		new->container = NULL;
		replace_array(container, victim, new);
		return;
	}
	replace_array(container, victim, new);
	return;
}

void manage(struct mdstat_ent *mdstat, struct supertype *container)
{
	/* We have just read mdstat and need to compare it with
	 * the known active arrays.
	 * Arrays with the wrong metadata are ignored.
	 */

	for ( ; mdstat ; mdstat = mdstat->next) {
		struct active_array *a;
		if (mdstat->devnum == container->devnum) {
			manage_container(mdstat, container);
			continue;
		}
		if (mdstat->metadata_version == NULL ||
		    strncmp(mdstat->metadata_version, "external:/", 10) != 0 ||
		    strncmp(mdstat->metadata_version+10, container->devname,
			    strlen(container->devname)) != 0 ||
		    mdstat->metadata_version[10+strlen(container->devname)]
		      != '/')
			/* Not for this array */
			continue;
		/* Looks like a member of this container */
		for (a = container->arrays; a; a = a->next) {
			if (mdstat->devnum == a->devnum) {
				if (a->container)
					manage_member(mdstat, a);
				break;
			}
		}
		if (a == NULL || !a->container)
			manage_new(mdstat, container, a);
	}
}

static int handle_message(struct supertype *container, struct md_message *msg)
{
	int err;
	struct md_generic_cmd *cmd = msg->buf;

	if (!cmd)
		return 0;

	switch (cmd->action) {
	case md_action_remove_device:

		/* forward to the monitor */
		active_cmd = cmd;
		write(container->mgr_pipe[1], &err, 1);
		read(container->mon_pipe[0], &err, 1);
		return err;

	default:
		return -1;
	}
}

void read_sock(struct supertype *container)
{
	int fd;
	struct md_message msg;
	int terminate = 0;
	long fl;
	int tmo = 3; /* 3 second timeout before hanging up the socket */

	fd = accept(container->sock, NULL, NULL);
	if (fd < 0)
		return;

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

	do {
		int err;

		msg.buf = NULL;

		/* read and validate the message */
		if (receive_message(fd, &msg, tmo) == 0) {
			err = handle_message(container, &msg);
			if (!err)
				ack(fd, msg.seq, tmo);
			else
				nack(fd, err, tmo);
		} else {
			terminate = 1;
			nack(fd, -1, tmo);
		}

		if (msg.buf)
			free(msg.buf);
	} while (!terminate);

	close(fd);
}

static int woke = 0;
void wake_me(int sig)
{
	woke = 1;
}

int exit_now = 0;
int manager_ready = 0;
void do_manager(struct supertype *container)
{
	struct mdstat_ent *mdstat;
	sigset_t block, orig;

	sigemptyset(&block);
	sigaddset(&block, SIGUSR1);

	signal(SIGUSR1, wake_me);

	do {
		woke = 0;

		if (exit_now)
			exit(0);

		mdstat = mdstat_read(1, 0);

		manage(mdstat, container);

		read_sock(container);

		free_mdstat(mdstat);

		remove_old();

		check_update_queue(container);

		manager_ready = 1;
		sigprocmask(SIG_SETMASK, &block, &orig);
		if (woke == 0)
			mdstat_wait_fd(container->sock, &orig);
		sigprocmask(SIG_SETMASK, &orig, NULL);
	} while(1);
}
Commit	Line	Data
549e9569 NB	1
	2	/*
	3	* The management thread for monitoring active md arrays.
	4	* This thread does things which might block such as memory
	5	* allocation.
	6	* In particular:
	7	*
	8	* - Find out about new arrays in this container.
	9	* Allocate the data structures and open the files.
	10	*
	11	* For this we watch /proc/mdstat and find new arrays with
	12	* metadata type that confirms sharing. e.g. "md4"
	13	* When we find a new array we slip it into the list of
	14	* arrays and signal 'monitor' by writing to a pipe.
	15	*
	16	* - Respond to reshape requests by allocating new data structures
	17	* and opening new files.
	18	*
	19	* These come as a change to raid_disks. We allocate a new
	20	* version of the data structures and slip it into the list.
	21	* 'monitor' will notice and release the old version.
	22	* Changes to level, chunksize, layout.. do not need re-allocation.
	23	* Reductions in raid_disks don't really either, but we handle
	24	* them the same way for consistency.
	25	*
	26	* - When a device is added to the container, we add it to the metadata
	27	* as a spare.
	28	*
	29	* - assist with activating spares by opening relevant sysfs file.
	30	*
	31	* - Pass on metadata updates from external programs such as
	32	* mdadm creating a new array.
	33	*
	34	* This is most-messy.
	35	* It might involve adding a new array or changing the status of
	36	* a spare, or any reconfig that the kernel doesn't get involved in.
	37	*
	38	* The required updates are received via a named pipe. There will
	39	* be one named pipe for each container. Each message contains a
	40	* sync marker: 0x5a5aa5a5, A byte count, and the message. This is
	41	* passed to the metadata handler which will interpret and process it.
	42	* For 'DDF' messages are internal data blocks with the leading
	43	* 'magic number' signifying what sort of data it is.
	44	*
	45	*/
	46
	47	/*
	48	* We select on /proc/mdstat and the named pipe.
	49	* We create new arrays or updated version of arrays and slip
	50	* them into the head of the list, then signal 'monitor' via a pipe write.
	51	* 'monitor' will notice and place the old array on a return list.
	52	* Metadata updates are placed on a queue just like they arrive
	53	* from the named pipe.
	54	*
	55	* When new arrays are found based on correct metadata string, we
	56	* need to identify them with an entry in the metadata. Maybe we require
	57	* the metadata to be mdX/NN when NN is the index into an appropriate table.
	58	*
	59	*/
	60
	61	/*
	62	* List of tasks:
	63	* - Watch for spares to be added to the container, and write updated
	64	* metadata to them.
65	* - Watch for new arrays using this container, confirm they match metadata
66	* and if so, start monitoring them
67	* - Watch for spares being added to monitored arrays. This shouldn't
68	* happen, as we should do all the adding. Just remove them.
69	* - Watch for change in raid-disks, chunk-size, etc. Update metadata and
70	* start a reshape.
71	*/
72	#ifndef _GNU_SOURCE
73	#define _GNU_SOURCE
74	#endif
75	#include "mdadm.h"
76	#include "mdmon.h"
77	#include <sys/socket.h>
1ed3f387	78	#include <signal.h>
549e9569	79
2a0bb19e DW	80	static void close_aa(struct active_array *aa)
	81	{
	82	struct mdinfo *d;
	83
	84	for (d = aa->info.devs; d; d = d->next)
	85	close(d->state_fd);
	86
	87	close(aa->action_fd);
	88	close(aa->info.state_fd);
	89	close(aa->resync_start_fd);
2a0bb19e DW	90	}
2a0bb19e DW	91
549e9569 NB	92	static void free_aa(struct active_array *aa)
549e9569 NB	93	{
2a0bb19e DW	94	/* Note that this doesn't close fds if they are being used
2a0bb19e DW	95	* by a clone. ->container will be set for a clone
549e9569	96	*/
2a0bb19e DW	97	if (!aa->container)
2a0bb19e DW	98	close_aa(aa);
549e9569 NB	99	while (aa->info.devs) {
	100	struct mdinfo *d = aa->info.devs;
	101	aa->info.devs = d->next;
	102	free(d);
	103	}
	104	free(aa);
	105	}
	106
2a0bb19e DW	107	static void write_wakeup(struct supertype *c)
2a0bb19e DW	108	{
3e70c845 DW	109	static struct md_generic_cmd cmd = { .action = md_action_ping_monitor };
	110	int err;
	111
	112	active_cmd = &cmd;
	113
	114	/* send the monitor thread a pointer to the ping action */
	115	write(c->mgr_pipe[1], &err, 1);
	116	read(c->mon_pipe[0], &err, 1);
2a0bb19e DW	117	}
2a0bb19e DW	118
1ed3f387 NB	119	static void remove_old(void)
	120	{
	121	if (discard_this) {
	122	discard_this->next = NULL;
	123	free_aa(discard_this);
	124	if (pending_discard == discard_this)
	125	pending_discard = NULL;
	126	discard_this = NULL;
	127	}
	128	}
	129
549e9569 NB	130	static void replace_array(struct supertype *container,
	131	struct active_array *old,
	132	struct active_array *new)
	133	{
	134	/* To replace an array, we add it to the top of the list
	135	* marked with ->replaces to point to the original.
	136	* 'monitor' will take the original out of the list
	137	* and put it on 'discard_this'. We take it from there
	138	* and discard it.
	139	*/
1ed3f387	140	remove_old();
549e9569	141	while (pending_discard) {
1ed3f387	142	write_wakeup(container);
549e9569 NB	143	while (discard_this == NULL)
549e9569 NB	144	sleep(1);
1ed3f387	145	remove_old();
549e9569 NB	146	}
	147	pending_discard = old;
	148	new->replaces = old;
	149	new->next = container->arrays;
	150	container->arrays = new;
2a0bb19e	151	write_wakeup(container);
549e9569 NB	152	}
549e9569 NB	153
2e735d19 NB	154	struct metadata_update *update_queue = NULL;
	155	struct metadata_update *update_queue_handled = NULL;
	156	struct metadata_update *update_queue_pending = NULL;
	157
	158	void check_update_queue(struct supertype *container)
	159	{
	160	while (update_queue_handled) {
	161	struct metadata_update *this = update_queue_handled;
	162	update_queue_handled = this->next;
	163	free(this->buf);
	164	free(this);
	165	}
	166	if (update_queue == NULL &&
	167	update_queue_pending) {
	168	update_queue = update_queue_pending;
	169	update_queue_pending = NULL;
	170	write_wakeup(container);
	171	}
	172	}
	173
	174	void queue_metadata_update(struct metadata_update *mu)
	175	{
	176	struct metadata_update **qp;
	177
	178	qp = &update_queue_pending;
	179	while (*qp)
	180	qp = & ((*qp)->next);
	181	*qp = mu;
	182	}
	183
	184	void wait_update_handled(void)
	185	{
	186	/* Wait for any pending update to be handled by monitor.
	187	* i.e. wait until update_queue is NULL
	188	*/
	189	while (update_queue)
	190	usleep(100 * 1000);
	191	}
	192
549e9569 NB	193	static void manage_container(struct mdstat_ent *mdstat,
	194	struct supertype *container)
	195	{
	196	/* The only thing of interest here is if a new device
	197	* has been added to the container. We add it to the
	198	* array ignoring any metadata on it.
	199	* FIXME should we look for compatible metadata and take hints
	200	* about spare assignment.... probably not.
	201	*
	202	*/
	203	if (mdstat->devcnt != container->devcnt) {
	204	/* read /sys/block/NAME/md/dev-??/block/dev to find out
	205	* what is there, and compare with container->info.devs
	206	* To see what is removed and what is added.
	207	* These need to be remove from, or added to, the array
	208	*/
	209	// FIXME
	210	container->devcnt = mdstat->devcnt;
	211	}
	212	}
	213
	214	static void manage_member(struct mdstat_ent *mdstat,
	215	struct active_array *a)
	216	{
	217	/* Compare mdstat info with known state of member array.
	218	* We do not need to look for device state changes here, that
	219	* is dealt with by the monitor.
	220	*
	221	* We just look for changes which suggest that a reshape is
	222	* being requested.
	223	* Unfortunately decreases in raid_disks don't show up in
	224	* mdstat until the reshape completes FIXME.
	225	*/
	226	// FIXME
	227	a->info.array.raid_disks = mdstat->raid_disks;
	228	a->info.array.chunk_size = mdstat->chunk_size;
	229	// MORE
	230
	231	}
	232
549e9569	233	static void manage_new(struct mdstat_ent *mdstat,
2a0bb19e DW	234	struct supertype *container,
2a0bb19e DW	235	struct active_array *victim)
549e9569 NB	236	{
	237	/* A new array has appeared in this container.
	238	* Hopefully it is already recorded in the metadata.
	239	* Check, then create the new array to report it to
	240	* the monitor.
	241	*/
	242
	243	struct active_array *new;
	244	struct mdinfo mdi, di;
cba0191b	245	char *inst;
549e9569 NB	246	int i;
	247
	248	new = malloc(sizeof(*new));
	249
d52690ac NB	250	memset(new, 0, sizeof(*new));
d52690ac NB	251
549e9569 NB	252	new->devnum = mdstat->devnum;
	253
	254	new->prev_state = new->curr_state = new->next_state = inactive;
	255	new->prev_action= new->curr_action= new->next_action= idle;
	256
	257	new->container = container;
	258
cba0191b	259	inst = &mdstat->metadata_version[10+strlen(container->devname)+1];
549e9569 NB	260
	261	mdi = sysfs_read(-1, new->devnum,
	262	GET_LEVEL\|GET_CHUNK\|GET_DISKS\|
	263	GET_DEVS\|GET_OFFSET\|GET_SIZE\|GET_STATE);
	264	if (!mdi) {
	265	/* Eeek. Cannot monitor this array.
	266	* Mark it to be ignored by setting container to NULL
	267	*/
	268	new->container = NULL;
2a0bb19e	269	replace_array(container, victim, new);
549e9569 NB	270	return;
	271	}
	272
	273	new->info.array = mdi->array;
	274
	275	for (i = 0; i < new->info.array.raid_disks; i++) {
	276	struct mdinfo newd = malloc(sizeof(newd));
	277
	278	for (di = mdi->devs; di; di = di->next)
	279	if (i == di->disk.raid_disk)
	280	break;
	281
	282	if (di) {
	283	memcpy(newd, di, sizeof(*newd));
	284
	285	sprintf(newd->sys_name, "rd%d", i);
	286
	287	newd->state_fd = sysfs_open(new->devnum,
	288	newd->sys_name,
	289	"state");
	290
	291	newd->prev_state = read_dev_state(newd->state_fd);
	292	newd->curr_state = newd->curr_state;
	293	} else {
	294	newd->state_fd = -1;
	295	}
	296	newd->next = new->info.devs;
	297	new->info.devs = newd;
	298	}
	299	new->action_fd = sysfs_open(new->devnum, NULL, "sync_action");
	300	new->info.state_fd = sysfs_open(new->devnum, NULL, "array_state");
c052ba30	301	new->resync_start_fd = sysfs_open(new->devnum, NULL, "resync_start");
77402e51	302	new->resync_start = 0;
549e9569	303
4fa5aef9	304	sysfs_free(mdi);
549e9569 NB	305	// finds and compares.
	306	if (container->ss->open_new(container, new, inst) < 0) {
	307	// FIXME close all those files
	308	new->container = NULL;
2a0bb19e	309	replace_array(container, victim, new);
549e9569 NB	310	return;
549e9569 NB	311	}
2a0bb19e	312	replace_array(container, victim, new);
549e9569 NB	313	return;
	314	}
	315
5d19760d	316	void manage(struct mdstat_ent mdstat, struct supertype container)
549e9569 NB	317	{
	318	/* We have just read mdstat and need to compare it with
	319	* the known active arrays.
	320	* Arrays with the wrong metadata are ignored.
	321	*/
	322
	323	for ( ; mdstat ; mdstat = mdstat->next) {
	324	struct active_array *a;
	325	if (mdstat->devnum == container->devnum) {
	326	manage_container(mdstat, container);
	327	continue;
	328	}
	329	if (mdstat->metadata_version == NULL \|\|
	330	strncmp(mdstat->metadata_version, "external:/", 10) != 0 \|\|
	331	strncmp(mdstat->metadata_version+10, container->devname,
	332	strlen(container->devname)) != 0 \|\|
	333	mdstat->metadata_version[10+strlen(container->devname)]
	334	!= '/')
	335	/* Not for this array */
	336	continue;
	337	/* Looks like a member of this container */
5d19760d	338	for (a = container->arrays; a; a = a->next) {
549e9569 NB	339	if (mdstat->devnum == a->devnum) {
	340	if (a->container)
	341	manage_member(mdstat, a);
	342	break;
	343	}
	344	}
2a0bb19e DW	345	if (a == NULL \|\| !a->container)
2a0bb19e DW	346	manage_new(mdstat, container, a);
549e9569 NB	347	}
	348	}
	349
3e70c845 DW	350	static int handle_message(struct supertype container, struct md_message msg)
	351	{
	352	int err;
	353	struct md_generic_cmd *cmd = msg->buf;
	354
	355	if (!cmd)
	356	return 0;
	357
	358	switch (cmd->action) {
	359	case md_action_remove_device:
	360
	361	/* forward to the monitor */
	362	active_cmd = cmd;
	363	write(container->mgr_pipe[1], &err, 1);
	364	read(container->mon_pipe[0], &err, 1);
	365	return err;
	366
	367	default:
	368	return -1;
	369	}
	370	}
	371
	372	void read_sock(struct supertype *container)
549e9569 NB	373	{
549e9569 NB	374	int fd;
b109d928 DW	375	struct md_message msg;
	376	int terminate = 0;
	377	long fl;
	378	int tmo = 3; /* 3 second timeout before hanging up the socket */
549e9569	379
3e70c845	380	fd = accept(container->sock, NULL, NULL);
549e9569 NB	381	if (fd < 0)
549e9569 NB	382	return;
b109d928 DW	383
	384	fl = fcntl(fd, F_GETFL, 0);
	385	fl \|= O_NONBLOCK;
	386	fcntl(fd, F_SETFL, fl);
	387
	388	do {
3e70c845 DW	389	int err;
3e70c845 DW	390
b109d928 DW	391	msg.buf = NULL;
	392
	393	/* read and validate the message */
	394	if (receive_message(fd, &msg, tmo) == 0) {
3e70c845 DW	395	err = handle_message(container, &msg);
	396	if (!err)
	397	ack(fd, msg.seq, tmo);
	398	else
	399	nack(fd, err, tmo);
b109d928 DW	400	} else {
	401	terminate = 1;
	402	nack(fd, -1, tmo);
	403	}
	404
	405	if (msg.buf)
	406	free(msg.buf);
	407	} while (!terminate);
	408
549e9569 NB	409	close(fd);
549e9569 NB	410	}
1ed3f387 NB	411
	412	static int woke = 0;
	413	void wake_me(int sig)
	414	{
	415	woke = 1;
	416	}
	417
e0d6609f NB	418	int exit_now = 0;
e0d6609f NB	419	int manager_ready = 0;
549e9569 NB	420	void do_manager(struct supertype *container)
	421	{
	422	struct mdstat_ent *mdstat;
1ed3f387 NB	423	sigset_t block, orig;
	424
	425	sigemptyset(&block);
	426	sigaddset(&block, SIGUSR1);
	427
	428	signal(SIGUSR1, wake_me);
549e9569 NB	429
549e9569 NB	430	do {
1ed3f387 NB	431	woke = 0;
1ed3f387 NB	432
e0d6609f NB	433	if (exit_now)
	434	exit(0);
	435
549e9569 NB	436	mdstat = mdstat_read(1, 0);
549e9569 NB	437
5d19760d	438	manage(mdstat, container);
549e9569	439
3e70c845	440	read_sock(container);
549e9569	441
4fa5aef9 DW	442	free_mdstat(mdstat);
4fa5aef9 DW	443
1ed3f387 NB	444	remove_old();
1ed3f387 NB	445
2e735d19 NB	446	check_update_queue(container);
2e735d19 NB	447
e0d6609f	448	manager_ready = 1;
1ed3f387 NB	449	sigprocmask(SIG_SETMASK, &block, &orig);
	450	if (woke == 0)
	451	mdstat_wait_fd(container->sock, &orig);
	452	sigprocmask(SIG_SETMASK, &orig, NULL);
549e9569 NB	453	} while(1);
549e9569 NB	454	}