[thirdparty/mdadm.git] / monitor.c


#include "mdadm.h"
#include "mdmon.h"

#include <sys/select.h>


static char *array_states[] = {
	"clear", "inactive", "suspended", "readonly", "read-auto",
	"clean", "active", "write-pending", "active-idle", NULL };
static char *sync_actions[] = {
	"idle", "reshape", "resync", "recover", "check", "repair", NULL
};

static int write_attr(char *attr, int fd)
{
	return write(fd, attr, strlen(attr));
}

static void add_fd(fd_set *fds, int *maxfd, int fd)
{
	if (fd < 0)
		return;
	if (fd > *maxfd)
		*maxfd = fd;
	FD_SET(fd, fds);
}

static int read_attr(char *buf, int len, int fd)
{
	int n;

	if (fd < 0) {
		buf[0] = 0;
		return 0;
	}
	lseek(fd, 0, 0);
	n = read(fd, buf, len - 1);

	if (n <= 0) {
		buf[0] = 0;
		return 0;
	}
	buf[n] = 0;
	if (buf[n-1] == '\n')
		buf[n-1] = 0;
	return n;
}

static int get_sync_pos(struct active_array *a)
{
	char buf[30];
	int n;

	n = read_attr(buf, 30, a->sync_pos_fd);
	if (n <= 0)
		return n;

	if (strncmp(buf, "max", 3) == 0) {
		a->sync_pos = ~(unsigned long long)0;
		return 1;
	}
	a->sync_pos = strtoull(buf, NULL, 10);
	return 1;
}

static int get_resync_start(struct active_array *a)
{
	char buf[30];
	int n;

	n = read_attr(buf, 30, a->resync_start_fd);
	if (n <= 0)
		return n;

	a->resync_start = strtoull(buf, NULL, 10);

	return 1;
}

static int attr_match(const char *attr, const char *str)
{
	/* See if attr, read from a sysfs file, matches
	 * str.  They must either be the same, or attr can
	 * have a trailing newline or comma
	 */
	while (*attr && *str && *attr == *str) {
		attr++;
		str++;
	}

	if (*str || (*attr && *attr != ',' && *attr != '\n'))
		return 0;
	return 1;
}

static int match_word(const char *word, char **list)
{
	int n;
	for (n=0; list[n]; n++)
		if (attr_match(word, list[n]))
			break;
	return n;
}

static enum array_state read_state(int fd)
{
	char buf[20];
	int n = read_attr(buf, 20, fd);

	if (n <= 0)
		return bad_word;
	return (enum array_state) match_word(buf, array_states);
}

static enum sync_action read_action( int fd)
{
	char buf[20];
	int n = read_attr(buf, 20, fd);

	if (n <= 0)
		return bad_action;
	return (enum sync_action) match_word(buf, sync_actions);
}

int read_dev_state(int fd)
{
	char buf[60];
	int n = read_attr(buf, 60, fd);
	char *cp;
	int rv = 0;

	if (n <= 0)
		return 0;

	cp = buf;
	while (cp) {
		if (attr_match(cp, "faulty"))
			rv |= DS_FAULTY;
		if (attr_match(cp, "in_sync"))
			rv |= DS_INSYNC;
		if (attr_match(cp, "write_mostly"))
			rv |= DS_WRITE_MOSTLY;
		if (attr_match(cp, "spare"))
			rv |= DS_SPARE;
		if (attr_match(cp, "blocked"))
			rv |= DS_BLOCKED;
		cp = strchr(cp, ',');
		if (cp)
			cp++;
	}
	return rv;
}


/* Monitor a set of active md arrays - all of which share the
 * same metadata - and respond to events that require
 * metadata update.
 *
 * New arrays are detected by another thread which allocates
 * required memory and attaches the data structure to our list.
 *
 * Events:
 *  Array stops.
 *    This is detected by array_state going to 'clear' or 'inactive'.
 *    while we thought it was active.
 *    Response is to mark metadata as clean and 'clear' the array(??)
 *  write-pending
 *    array_state if 'write-pending'
 *    We mark metadata as 'dirty' then set array to 'active'.
 *  active_idle
 *    Either ignore, or mark clean, then mark metadata as clean.
 *
 *  device fails
 *    detected by rd-N/state reporting "faulty"
 *    mark device as 'failed' in metadata, let the kernel release the
 *    device by writing '-blocked' to rd/state, and finally write 'remove' to
 *    rd/state.  Before a disk can be replaced it must be failed and removed
 *    from all container members, this will be preemptive for the other
 *    arrays... safe?
 *
 *  sync completes
 *    sync_action was 'resync' and becomes 'idle' and resync_start becomes
 *    MaxSector
 *    Notify metadata that sync is complete.
 *    "Deal with Degraded"
 *
 *  recovery completes
 *    sync_action changes from 'recover' to 'idle'
 *    Check each device state and mark metadata if 'faulty' or 'in_sync'.
 *    "Deal with Degraded"
 *
 *  deal with degraded array
 *    We only do this when first noticing the array is degraded.
 *    This can be when we first see the array, when sync completes or
 *    when recovery completes.
 *
 *    Check if number of failed devices suggests recovery is needed, and
 *    skip if not.
 *    Ask metadata for a spare device
 *    Add device as not in_sync and give a role
 *    Update metadata.
 *    Start recovery.
 *
 *  deal with resync
 *    This only happens on finding a new array... mdadm will have set
 *    'resync_start' to the correct value.  If 'resync_start' indicates that an
 *    resync needs to occur set the array to the 'active' state rather than the
 *    initial read-auto state.
 *
 *
 *
 * We wait for a change (poll/select) on array_state, sync_action, and
 * each rd-X/state file.
 * When we get any change, we check everything.  So read each state file,
 * then decide what to do.
 *
 * The core action is to write new metadata to all devices in the array.
 * This is done at most once on any wakeup.
 * After that we might:
 *   - update the array_state
 *   - set the role of some devices.
 *   - request a sync_action
 *
 */

static int read_and_act(struct active_array *a)
{
	int check_degraded;
	int deactivate = 0;
	struct mdinfo *mdi;

	a->next_state = bad_word;
	a->next_action = bad_action;

	a->curr_state = read_state(a->info.state_fd);
	a->curr_action = read_action(a->action_fd);
	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
		mdi->next_state = 0;
		if (mdi->state_fd > 0)
			mdi->curr_state = read_dev_state(mdi->state_fd);
	}

	if (a->curr_state <= inactive &&
	    a->prev_state > inactive) {
		/* array has been stopped */
		get_sync_pos(a);
		a->container->ss->mark_clean(a, a->sync_pos);
		a->next_state = clear;
		deactivate = 1;
	}
	if (a->curr_state == write_pending) {
		a->container->ss->mark_clean(a, 0);
		a->next_state = active;
	}
	if (a->curr_state == active_idle) {
		/* Set array to 'clean' FIRST, then
		 * a->ss->mark_clean(a, ~0ULL);
		 * just ignore for now.
		 */
	}

	if (a->curr_state == readonly) {
		/* Well, I'm ready to handle things, so
		 * read-auto is OK. FIXME what if we really want
		 * readonly ???
		 */
		get_resync_start(a);
		if (a->resync_start == ~0ULL)
			a->next_state = read_auto; /* array is clean */
		else {
			a->container->ss->mark_clean(a, 0);
			a->next_state = active;
		}
	}

	if (a->curr_action == idle &&
	    a->prev_action == resync) {
		/* A resync has finished.  The endpoint is recorded in
		 * 'sync_start'.  We don't update the metadata
		 * until the array goes inactive or readonly though.
		 * Just check if we need to fiddle spares.
		 */
		check_degraded = 1;
	}

	if (a->curr_action == idle &&
	    a->prev_action == recover) {
		for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
			a->container->ss->set_disk(a, mdi->disk.raid_disk,
						   mdi->curr_state);
			if (! (mdi->curr_state & DS_INSYNC))
				check_degraded = 1;
		}
	}


	for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
		if (mdi->curr_state & DS_FAULTY) {
			a->container->ss->set_disk(a, mdi->disk.raid_disk,
						   mdi->curr_state);
			check_degraded = 1;
			mdi->next_state = DS_REMOVE;
		}
	}

	if (check_degraded) {
		// FIXME;
	}

	a->container->ss->sync_metadata(a);

	/* Effect state changes in the array */
	if (a->next_state != bad_word)
		write_attr(array_states[a->next_state], a->info.state_fd);
	if (a->next_action != bad_action)
		write_attr(sync_actions[a->next_action], a->action_fd);
	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
		if (mdi->next_state == DS_REMOVE && mdi->state_fd > 0) {
			int remove_err;

			write_attr("-blocked", mdi->state_fd);
			/* the kernel may not be able to immediately remove the
			 * disk, we can simply wait until the next event to try
			 * again.
			 */
			remove_err = write_attr("remove", mdi->state_fd);
			if (!remove_err) {
				close(mdi->state_fd);
				mdi->state_fd = -1;
			}
		}
		if (mdi->next_state & DS_INSYNC)
			write_attr("+in_sync", mdi->state_fd);
	}

	/* move curr_ to prev_ */
	a->prev_state = a->curr_state;

	a->prev_action = a->curr_action;

	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
		mdi->prev_state = mdi->curr_state;
		mdi->next_state = 0;
	}

	if (deactivate)
		a->container = NULL;

	return 1;
}

static struct mdinfo *
find_device(struct active_array *a, int major, int minor)
{
	struct mdinfo *mdi;

	for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
		if (mdi->disk.major == major && mdi->disk.minor == minor)
			return mdi;

	return NULL;
}

static void reconcile_failed(struct active_array *aa, struct mdinfo *failed)
{
	struct active_array *a;
	struct mdinfo *victim;

	for (a = aa; a; a = a->next) {
		if (!a->container)
			continue;
		victim = find_device(a, failed->disk.major, failed->disk.minor);
		if (!victim)
			continue;

		if (!(victim->curr_state & DS_FAULTY))
			write_attr("faulty", victim->state_fd);
	}
}

static int handle_remove_device(struct md_remove_device_cmd *cmd, struct active_array *aa)
{
	struct active_array *a;
	struct mdinfo *victim;
	int rv;

	/* scan all arrays for the given device, if ->state_fd is closed (-1)
	 * in all cases then mark the disk as removed in the metadata.
	 * Otherwise reply that it is busy.
	 */

	/* pass1 check that it is not in use anywhere */
	/* note: we are safe from re-adds as long as the device exists in the
	 * container
	 */
	for (a = aa; a; a = a->next) {
		if (!a->container)
			continue;
		victim = find_device(a, major(cmd->rdev), minor(cmd->rdev));
		if (!victim)
			continue;
		if (victim->state_fd > 0)
			return -EBUSY;
	}

	/* pass2 schedule and process removal per array */
	for (a = aa; a; a = a->next) {
		if (!a->container)
			continue;
		victim = find_device(a, major(cmd->rdev), minor(cmd->rdev));
		if (!victim)
			continue;
		victim->curr_state |= DS_REMOVE;
		rv = read_and_act(a);
		if (rv < 0)
			return rv;
	}

	return 0;
}

static int handle_pipe(struct md_generic_cmd *cmd, struct active_array *aa)
{
	switch (cmd->action) {
	case md_action_ping_monitor:
		return 0;
	case md_action_remove_device:
		return handle_remove_device((void *) cmd, aa);
	}

	return -1;
}

static int wait_and_act(struct active_array *aa, int pfd, int monfd, int nowait)
{
	fd_set rfds;
	int maxfd = 0;
	struct active_array *a;
	int rv;
	struct mdinfo *mdi;

	FD_ZERO(&rfds);

	add_fd(&rfds, &maxfd, pfd);
	for (a = aa ; a ; a = a->next) {
		/* once an array has been deactivated only the manager
		 * thread can make us care about it again
		 */
		if (!a->container)
			continue;

		add_fd(&rfds, &maxfd, a->info.state_fd);
		add_fd(&rfds, &maxfd, a->action_fd);
		for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
			add_fd(&rfds, &maxfd, mdi->state_fd);
	}

	if (!nowait) {
		rv = select(maxfd+1, &rfds, NULL, NULL, NULL);

		if (rv <= 0)
			return rv;

		if (FD_ISSET(pfd, &rfds)) {
			int err = -1;

			if (read(pfd, &err, 1) > 0)
				err = handle_pipe(active_cmd, aa);
			write(monfd, &err, 1);
		}
	}

	for (a = aa; a ; a = a->next) {
		if (a->replaces && !discard_this) {
			struct active_array **ap;
			for (ap = &a->next; *ap && *ap != a->replaces;
			     ap = & (*ap)->next)
				;
			if (*ap)
				*ap = (*ap)->next;
			discard_this = a->replaces;
			a->replaces = NULL;
		}
		if (a->container)
			rv += read_and_act(a);
	}

	/* propagate failures across container members */
	for (a = aa; a ; a = a->next) {
		if (!a->container)
			continue;
		for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
			if (mdi->curr_state & DS_FAULTY)
				reconcile_failed(aa, mdi);
	}

	return rv;
}

void do_monitor(struct supertype *container)
{
	int rv;
	int first = 1;
	do {
		rv = wait_and_act(container->arrays, container->mgr_pipe[0],
				  container->mon_pipe[1], first);
		first = 0;
	} while (rv >= 0);
}
Commit	Line	Data
549e9569 NB	1
	2	#include "mdadm.h"
	3	#include "mdmon.h"
	4
	5	#include <sys/select.h>
	6
	7
	8	static char *array_states[] = {
	9	"clear", "inactive", "suspended", "readonly", "read-auto",
	10	"clean", "active", "write-pending", "active-idle", NULL };
	11	static char *sync_actions[] = {
	12	"idle", "reshape", "resync", "recover", "check", "repair", NULL
	13	};
	14
	15	static int write_attr(char *attr, int fd)
	16	{
	17	return write(fd, attr, strlen(attr));
	18	}
	19
	20	static void add_fd(fd_set fds, int maxfd, int fd)
	21	{
	22	if (fd < 0)
	23	return;
	24	if (fd > *maxfd)
	25	*maxfd = fd;
	26	FD_SET(fd, fds);
	27	}
	28
	29	static int read_attr(char *buf, int len, int fd)
	30	{
	31	int n;
	32
	33	if (fd < 0) {
	34	buf[0] = 0;
	35	return 0;
	36	}
	37	lseek(fd, 0, 0);
	38	n = read(fd, buf, len - 1);
	39
	40	if (n <= 0) {
	41	buf[0] = 0;
	42	return 0;
	43	}
	44	buf[n] = 0;
	45	if (buf[n-1] == '\n')
	46	buf[n-1] = 0;
	47	return n;
	48	}
	49
	50	static int get_sync_pos(struct active_array *a)
	51	{
	52	char buf[30];
	53	int n;
	54
	55	n = read_attr(buf, 30, a->sync_pos_fd);
	56	if (n <= 0)
	57	return n;
	58
	59	if (strncmp(buf, "max", 3) == 0) {
	60	a->sync_pos = ~(unsigned long long)0;
	61	return 1;
	62	}
	63	a->sync_pos = strtoull(buf, NULL, 10);
	64	return 1;
65	}
66
c052ba30 DW	67	static int get_resync_start(struct active_array *a)
	68	{
	69	char buf[30];
	70	int n;
	71
	72	n = read_attr(buf, 30, a->resync_start_fd);
	73	if (n <= 0)
	74	return n;
	75
	76	a->resync_start = strtoull(buf, NULL, 10);
	77
	78	return 1;
	79	}
549e9569 NB	80
	81	static int attr_match(const char attr, const char str)
	82	{
	83	/* See if attr, read from a sysfs file, matches
	84	* str. They must either be the same, or attr can
	85	* have a trailing newline or comma
	86	*/
	87	while (attr && str && attr == str) {
	88	attr++;
	89	str++;
	90	}
	91
	92	if (str \|\| (attr && attr != ',' && attr != '\n'))
	93	return 0;
	94	return 1;
	95	}
	96
	97	static int match_word(const char word, char *list)
	98	{
	99	int n;
	100	for (n=0; list[n]; n++)
	101	if (attr_match(word, list[n]))
	102	break;
	103	return n;
	104	}
	105
	106	static enum array_state read_state(int fd)
	107	{
	108	char buf[20];
	109	int n = read_attr(buf, 20, fd);
	110
	111	if (n <= 0)
	112	return bad_word;
	113	return (enum array_state) match_word(buf, array_states);
	114	}
	115
	116	static enum sync_action read_action( int fd)
	117	{
	118	char buf[20];
	119	int n = read_attr(buf, 20, fd);
	120
	121	if (n <= 0)
	122	return bad_action;
	123	return (enum sync_action) match_word(buf, sync_actions);
	124	}
	125
549e9569 NB	126	int read_dev_state(int fd)
	127	{
	128	char buf[60];
	129	int n = read_attr(buf, 60, fd);
	130	char *cp;
	131	int rv = 0;
	132
	133	if (n <= 0)
	134	return 0;
	135
	136	cp = buf;
	137	while (cp) {
8d45d196	138	if (attr_match(cp, "faulty"))
549e9569	139	rv \|= DS_FAULTY;
8d45d196	140	if (attr_match(cp, "in_sync"))
549e9569	141	rv \|= DS_INSYNC;
8d45d196	142	if (attr_match(cp, "write_mostly"))
549e9569	143	rv \|= DS_WRITE_MOSTLY;
8d45d196	144	if (attr_match(cp, "spare"))
549e9569	145	rv \|= DS_SPARE;
8d45d196 DW	146	if (attr_match(cp, "blocked"))
8d45d196 DW	147	rv \|= DS_BLOCKED;
549e9569 NB	148	cp = strchr(cp, ',');
	149	if (cp)
	150	cp++;
	151	}
	152	return rv;
	153	}
	154
	155
	156	/* Monitor a set of active md arrays - all of which share the
	157	* same metadata - and respond to events that require
	158	* metadata update.
	159	*
	160	* New arrays are detected by another thread which allocates
	161	* required memory and attaches the data structure to our list.
	162	*
	163	* Events:
	164	* Array stops.
	165	* This is detected by array_state going to 'clear' or 'inactive'.
	166	* while we thought it was active.
	167	* Response is to mark metadata as clean and 'clear' the array(??)
	168	* write-pending
	169	* array_state if 'write-pending'
	170	* We mark metadata as 'dirty' then set array to 'active'.
	171	* active_idle
	172	* Either ignore, or mark clean, then mark metadata as clean.
	173	*
	174	* device fails
	175	* detected by rd-N/state reporting "faulty"
8d45d196 DW	176	* mark device as 'failed' in metadata, let the kernel release the
8d45d196 DW	177	* device by writing '-blocked' to rd/state, and finally write 'remove' to
0af73f61 DW	178	* rd/state. Before a disk can be replaced it must be failed and removed
	179	* from all container members, this will be preemptive for the other
	180	* arrays... safe?
549e9569 NB	181	*
	182	* sync completes
	183	* sync_action was 'resync' and becomes 'idle' and resync_start becomes
	184	* MaxSector
	185	* Notify metadata that sync is complete.
	186	* "Deal with Degraded"
	187	*
	188	* recovery completes
	189	* sync_action changes from 'recover' to 'idle'
	190	* Check each device state and mark metadata if 'faulty' or 'in_sync'.
	191	* "Deal with Degraded"
	192	*
	193	* deal with degraded array
	194	* We only do this when first noticing the array is degraded.
	195	* This can be when we first see the array, when sync completes or
	196	* when recovery completes.
	197	*
	198	* Check if number of failed devices suggests recovery is needed, and
	199	* skip if not.
	200	* Ask metadata for a spare device
	201	* Add device as not in_sync and give a role
	202	* Update metadata.
	203	* Start recovery.
	204	*
	205	* deal with resync
c052ba30 DW	206	* This only happens on finding a new array... mdadm will have set
	207	* 'resync_start' to the correct value. If 'resync_start' indicates that an
	208	* resync needs to occur set the array to the 'active' state rather than the
	209	* initial read-auto state.
549e9569 NB	210	*
	211	*
	212	*
	213	* We wait for a change (poll/select) on array_state, sync_action, and
	214	* each rd-X/state file.
	215	* When we get any change, we check everything. So read each state file,
	216	* then decide what to do.
	217	*
	218	* The core action is to write new metadata to all devices in the array.
	219	* This is done at most once on any wakeup.
	220	* After that we might:
	221	* - update the array_state
	222	* - set the role of some devices.
	223	* - request a sync_action
	224	*
	225	*/
	226
	227	static int read_and_act(struct active_array *a)
	228	{
	229	int check_degraded;
2a0bb19e	230	int deactivate = 0;
549e9569 NB	231	struct mdinfo *mdi;
	232
	233	a->next_state = bad_word;
	234	a->next_action = bad_action;
	235
	236	a->curr_state = read_state(a->info.state_fd);
	237	a->curr_action = read_action(a->action_fd);
	238	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	239	mdi->next_state = 0;
8d45d196 DW	240	if (mdi->state_fd > 0)
8d45d196 DW	241	mdi->curr_state = read_dev_state(mdi->state_fd);
549e9569 NB	242	}
	243
	244	if (a->curr_state <= inactive &&
	245	a->prev_state > inactive) {
	246	/* array has been stopped */
	247	get_sync_pos(a);
	248	a->container->ss->mark_clean(a, a->sync_pos);
	249	a->next_state = clear;
2a0bb19e	250	deactivate = 1;
549e9569 NB	251	}
549e9569 NB	252	if (a->curr_state == write_pending) {
4e5528c6	253	a->container->ss->mark_clean(a, 0);
549e9569 NB	254	a->next_state = active;
	255	}
	256	if (a->curr_state == active_idle) {
	257	/* Set array to 'clean' FIRST, then
4e5528c6	258	* a->ss->mark_clean(a, ~0ULL);
549e9569 NB	259	* just ignore for now.
	260	*/
	261	}
	262
	263	if (a->curr_state == readonly) {
	264	/* Well, I'm ready to handle things, so
	265	* read-auto is OK. FIXME what if we really want
	266	* readonly ???
	267	*/
c052ba30 DW	268	get_resync_start(a);
	269	if (a->resync_start == ~0ULL)
	270	a->next_state = read_auto; /* array is clean */
	271	else {
4e5528c6	272	a->container->ss->mark_clean(a, 0);
c052ba30 DW	273	a->next_state = active;
c052ba30 DW	274	}
549e9569 NB	275	}
	276
	277	if (a->curr_action == idle &&
	278	a->prev_action == resync) {
4e5528c6 NB	279	/* A resync has finished. The endpoint is recorded in
	280	* 'sync_start'. We don't update the metadata
	281	* until the array goes inactive or readonly though.
	282	* Just check if we need to fiddle spares.
	283	*/
549e9569 NB	284	check_degraded = 1;
	285	}
	286
	287	if (a->curr_action == idle &&
	288	a->prev_action == recover) {
	289	for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
8d45d196 DW	290	a->container->ss->set_disk(a, mdi->disk.raid_disk,
8d45d196 DW	291	mdi->curr_state);
549e9569 NB	292	if (! (mdi->curr_state & DS_INSYNC))
	293	check_degraded = 1;
	294	}
	295	}
	296
	297
	298	for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
	299	if (mdi->curr_state & DS_FAULTY) {
8d45d196 DW	300	a->container->ss->set_disk(a, mdi->disk.raid_disk,
8d45d196 DW	301	mdi->curr_state);
549e9569 NB	302	check_degraded = 1;
	303	mdi->next_state = DS_REMOVE;
	304	}
	305	}
	306
	307	if (check_degraded) {
	308	// FIXME;
	309	}
	310
	311	a->container->ss->sync_metadata(a);
	312
	313	/* Effect state changes in the array */
	314	if (a->next_state != bad_word)
	315	write_attr(array_states[a->next_state], a->info.state_fd);
	316	if (a->next_action != bad_action)
	317	write_attr(sync_actions[a->next_action], a->action_fd);
	318	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
8d45d196 DW	319	if (mdi->next_state == DS_REMOVE && mdi->state_fd > 0) {
	320	int remove_err;
	321
	322	write_attr("-blocked", mdi->state_fd);
	323	/* the kernel may not be able to immediately remove the
	324	* disk, we can simply wait until the next event to try
	325	* again.
	326	*/
	327	remove_err = write_attr("remove", mdi->state_fd);
	328	if (!remove_err) {
	329	close(mdi->state_fd);
	330	mdi->state_fd = -1;
	331	}
	332	}
549e9569 NB	333	if (mdi->next_state & DS_INSYNC)
	334	write_attr("+in_sync", mdi->state_fd);
	335	}
	336
	337	/* move curr_ to prev_ */
	338	a->prev_state = a->curr_state;
	339
	340	a->prev_action = a->curr_action;
	341
	342	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	343	mdi->prev_state = mdi->curr_state;
	344	mdi->next_state = 0;
	345	}
	346
2a0bb19e DW	347	if (deactivate)
	348	a->container = NULL;
	349
549e9569 NB	350	return 1;
	351	}
	352
0af73f61 DW	353	static struct mdinfo *
	354	find_device(struct active_array *a, int major, int minor)
	355	{
	356	struct mdinfo *mdi;
	357
	358	for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
	359	if (mdi->disk.major == major && mdi->disk.minor == minor)
	360	return mdi;
	361
	362	return NULL;
	363	}
	364
	365	static void reconcile_failed(struct active_array aa, struct mdinfo failed)
	366	{
	367	struct active_array *a;
	368	struct mdinfo *victim;
	369
	370	for (a = aa; a; a = a->next) {
	371	if (!a->container)
	372	continue;
	373	victim = find_device(a, failed->disk.major, failed->disk.minor);
	374	if (!victim)
	375	continue;
	376
	377	if (!(victim->curr_state & DS_FAULTY))
	378	write_attr("faulty", victim->state_fd);
	379	}
	380	}
	381
3e70c845 DW	382	static int handle_remove_device(struct md_remove_device_cmd cmd, struct active_array aa)
	383	{
	384	struct active_array *a;
	385	struct mdinfo *victim;
	386	int rv;
	387
	388	/* scan all arrays for the given device, if ->state_fd is closed (-1)
	389	* in all cases then mark the disk as removed in the metadata.
	390	* Otherwise reply that it is busy.
	391	*/
	392
	393	/* pass1 check that it is not in use anywhere */
	394	/* note: we are safe from re-adds as long as the device exists in the
	395	* container
	396	*/
	397	for (a = aa; a; a = a->next) {
	398	if (!a->container)
	399	continue;
	400	victim = find_device(a, major(cmd->rdev), minor(cmd->rdev));
	401	if (!victim)
	402	continue;
	403	if (victim->state_fd > 0)
	404	return -EBUSY;
	405	}
	406
	407	/* pass2 schedule and process removal per array */
	408	for (a = aa; a; a = a->next) {
	409	if (!a->container)
	410	continue;
	411	victim = find_device(a, major(cmd->rdev), minor(cmd->rdev));
	412	if (!victim)
	413	continue;
	414	victim->curr_state \|= DS_REMOVE;
	415	rv = read_and_act(a);
	416	if (rv < 0)
	417	return rv;
	418	}
	419
	420	return 0;
	421	}
	422
	423	static int handle_pipe(struct md_generic_cmd cmd, struct active_array aa)
	424	{
	425	switch (cmd->action) {
	426	case md_action_ping_monitor:
	427	return 0;
	428	case md_action_remove_device:
	429	return handle_remove_device((void *) cmd, aa);
	430	}
	431
	432	return -1;
	433	}
	434
	435	static int wait_and_act(struct active_array *aa, int pfd, int monfd, int nowait)
549e9569 NB	436	{
	437	fd_set rfds;
	438	int maxfd = 0;
	439	struct active_array *a;
	440	int rv;
0af73f61	441	struct mdinfo *mdi;
549e9569 NB	442
	443	FD_ZERO(&rfds);
	444
	445	add_fd(&rfds, &maxfd, pfd);
	446	for (a = aa ; a ; a = a->next) {
2a0bb19e DW	447	/* once an array has been deactivated only the manager
	448	* thread can make us care about it again
	449	*/
	450	if (!a->container)
	451	continue;
	452
549e9569 NB	453	add_fd(&rfds, &maxfd, a->info.state_fd);
	454	add_fd(&rfds, &maxfd, a->action_fd);
	455	for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
	456	add_fd(&rfds, &maxfd, mdi->state_fd);
	457	}
	458
	459	if (!nowait) {
	460	rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
	461
	462	if (rv <= 0)
	463	return rv;
	464
	465	if (FD_ISSET(pfd, &rfds)) {
3e70c845 DW	466	int err = -1;
	467
	468	if (read(pfd, &err, 1) > 0)
	469	err = handle_pipe(active_cmd, aa);
	470	write(monfd, &err, 1);
549e9569 NB	471	}
	472	}
	473
	474	for (a = aa; a ; a = a->next) {
2a0bb19e	475	if (a->replaces && !discard_this) {
549e9569 NB	476	struct active_array **ap;
	477	for (ap = &a->next; ap && ap != a->replaces;
	478	ap = & (*ap)->next)
	479	;
	480	if (*ap)
	481	ap = (ap)->next;
	482	discard_this = a->replaces;
	483	a->replaces = NULL;
	484	}
2a0bb19e DW	485	if (a->container)
2a0bb19e DW	486	rv += read_and_act(a);
549e9569	487	}
0af73f61 DW	488
	489	/* propagate failures across container members */
	490	for (a = aa; a ; a = a->next) {
	491	if (!a->container)
	492	continue;
	493	for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
	494	if (mdi->curr_state & DS_FAULTY)
	495	reconcile_failed(aa, mdi);
	496	}
	497
549e9569 NB	498	return rv;
	499	}
	500
	501	void do_monitor(struct supertype *container)
	502	{
	503	int rv;
	504	int first = 1;
	505	do {
3e70c845 DW	506	rv = wait_and_act(container->arrays, container->mgr_pipe[0],
3e70c845 DW	507	container->mon_pipe[1], first);
549e9569 NB	508	first = 0;
	509	} while (rv >= 0);
	510	}