[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
 *
 *  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_dirty(a);
		a->next_state = active;
	}
	if (a->curr_state == active_idle) {
		/* Set array to 'clean' FIRST, then
		 * a->ss->mark_clean(a);
		 * 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_dirty(a);
			a->next_state = active;
		}
	}

	if (a->curr_action == idle &&
	    a->prev_action == resync) {
		/* check resync_start to see if it is 'max' */
		get_resync_start(a);
		a->container->ss->mark_sync(a, a->resync_start);
		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 int wait_and_act(struct active_array *aa, int pfd, int nowait)
{
	fd_set rfds;
	int maxfd = 0;
	struct active_array *a;
	int rv;

	FD_ZERO(&rfds);

	add_fd(&rfds, &maxfd, pfd);
	for (a = aa ; a ; a = a->next) {
		struct mdinfo *mdi;

		/* 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)) {
			char buf[4];
			read(pfd, buf, 4);
			; // FIXME read from the pipe
		}
	}

	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);
	}
	return rv;
}

void do_monitor(struct supertype *container)
{
	int rv;
	int first = 1;
	do {
		rv = wait_and_act(container->arrays, container->pipe[0], 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
	177	* device by writing '-blocked' to rd/state, and finally write 'remove' to
	178	* rd/state
549e9569 NB	179	*
	180	* sync completes
	181	* sync_action was 'resync' and becomes 'idle' and resync_start becomes
	182	* MaxSector
	183	* Notify metadata that sync is complete.
	184	* "Deal with Degraded"
	185	*
	186	* recovery completes
	187	* sync_action changes from 'recover' to 'idle'
	188	* Check each device state and mark metadata if 'faulty' or 'in_sync'.
	189	* "Deal with Degraded"
	190	*
	191	* deal with degraded array
	192	* We only do this when first noticing the array is degraded.
	193	* This can be when we first see the array, when sync completes or
	194	* when recovery completes.
	195	*
	196	* Check if number of failed devices suggests recovery is needed, and
	197	* skip if not.
	198	* Ask metadata for a spare device
	199	* Add device as not in_sync and give a role
	200	* Update metadata.
	201	* Start recovery.
	202	*
	203	* deal with resync
c052ba30 DW	204	* This only happens on finding a new array... mdadm will have set
	205	* 'resync_start' to the correct value. If 'resync_start' indicates that an
	206	* resync needs to occur set the array to the 'active' state rather than the
	207	* initial read-auto state.
549e9569 NB	208	*
	209	*
	210	*
	211	* We wait for a change (poll/select) on array_state, sync_action, and
	212	* each rd-X/state file.
	213	* When we get any change, we check everything. So read each state file,
	214	* then decide what to do.
	215	*
	216	* The core action is to write new metadata to all devices in the array.
	217	* This is done at most once on any wakeup.
	218	* After that we might:
	219	* - update the array_state
	220	* - set the role of some devices.
	221	* - request a sync_action
	222	*
	223	*/
	224
	225	static int read_and_act(struct active_array *a)
	226	{
	227	int check_degraded;
2a0bb19e	228	int deactivate = 0;
549e9569 NB	229	struct mdinfo *mdi;
	230
	231	a->next_state = bad_word;
	232	a->next_action = bad_action;
	233
	234	a->curr_state = read_state(a->info.state_fd);
	235	a->curr_action = read_action(a->action_fd);
	236	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	237	mdi->next_state = 0;
8d45d196 DW	238	if (mdi->state_fd > 0)
8d45d196 DW	239	mdi->curr_state = read_dev_state(mdi->state_fd);
549e9569 NB	240	}
	241
	242	if (a->curr_state <= inactive &&
	243	a->prev_state > inactive) {
	244	/* array has been stopped */
	245	get_sync_pos(a);
	246	a->container->ss->mark_clean(a, a->sync_pos);
	247	a->next_state = clear;
2a0bb19e	248	deactivate = 1;
549e9569 NB	249	}
	250	if (a->curr_state == write_pending) {
	251	a->container->ss->mark_dirty(a);
	252	a->next_state = active;
	253	}
	254	if (a->curr_state == active_idle) {
	255	/* Set array to 'clean' FIRST, then
	256	* a->ss->mark_clean(a);
	257	* just ignore for now.
	258	*/
	259	}
	260
	261	if (a->curr_state == readonly) {
	262	/* Well, I'm ready to handle things, so
	263	* read-auto is OK. FIXME what if we really want
	264	* readonly ???
	265	*/
c052ba30 DW	266	get_resync_start(a);
	267	if (a->resync_start == ~0ULL)
	268	a->next_state = read_auto; /* array is clean */
	269	else {
	270	a->container->ss->mark_dirty(a);
	271	a->next_state = active;
	272	}
549e9569 NB	273	}
	274
	275	if (a->curr_action == idle &&
	276	a->prev_action == resync) {
fd7cde1b DW	277	/* check resync_start to see if it is 'max' */
	278	get_resync_start(a);
	279	a->container->ss->mark_sync(a, a->resync_start);
549e9569 NB	280	check_degraded = 1;
	281	}
	282
	283	if (a->curr_action == idle &&
	284	a->prev_action == recover) {
	285	for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
8d45d196 DW	286	a->container->ss->set_disk(a, mdi->disk.raid_disk,
8d45d196 DW	287	mdi->curr_state);
549e9569 NB	288	if (! (mdi->curr_state & DS_INSYNC))
	289	check_degraded = 1;
	290	}
	291	}
	292
	293
	294	for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
	295	if (mdi->curr_state & DS_FAULTY) {
8d45d196 DW	296	a->container->ss->set_disk(a, mdi->disk.raid_disk,
8d45d196 DW	297	mdi->curr_state);
549e9569 NB	298	check_degraded = 1;
	299	mdi->next_state = DS_REMOVE;
	300	}
	301	}
	302
	303	if (check_degraded) {
	304	// FIXME;
	305	}
	306
	307	a->container->ss->sync_metadata(a);
	308
	309	/* Effect state changes in the array */
	310	if (a->next_state != bad_word)
	311	write_attr(array_states[a->next_state], a->info.state_fd);
	312	if (a->next_action != bad_action)
	313	write_attr(sync_actions[a->next_action], a->action_fd);
	314	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
8d45d196 DW	315	if (mdi->next_state == DS_REMOVE && mdi->state_fd > 0) {
	316	int remove_err;
	317
	318	write_attr("-blocked", mdi->state_fd);
	319	/* the kernel may not be able to immediately remove the
	320	* disk, we can simply wait until the next event to try
	321	* again.
	322	*/
	323	remove_err = write_attr("remove", mdi->state_fd);
	324	if (!remove_err) {
	325	close(mdi->state_fd);
	326	mdi->state_fd = -1;
	327	}
	328	}
549e9569 NB	329	if (mdi->next_state & DS_INSYNC)
	330	write_attr("+in_sync", mdi->state_fd);
	331	}
	332
	333	/* move curr_ to prev_ */
	334	a->prev_state = a->curr_state;
	335
	336	a->prev_action = a->curr_action;
	337
	338	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	339	mdi->prev_state = mdi->curr_state;
	340	mdi->next_state = 0;
	341	}
	342
2a0bb19e DW	343	if (deactivate)
	344	a->container = NULL;
	345
549e9569 NB	346	return 1;
	347	}
	348
	349	static int wait_and_act(struct active_array *aa, int pfd, int nowait)
	350	{
	351	fd_set rfds;
	352	int maxfd = 0;
	353	struct active_array *a;
	354	int rv;
	355
	356	FD_ZERO(&rfds);
	357
	358	add_fd(&rfds, &maxfd, pfd);
	359	for (a = aa ; a ; a = a->next) {
	360	struct mdinfo *mdi;
	361
2a0bb19e DW	362	/* once an array has been deactivated only the manager
	363	* thread can make us care about it again
	364	*/
	365	if (!a->container)
	366	continue;
	367
549e9569 NB	368	add_fd(&rfds, &maxfd, a->info.state_fd);
	369	add_fd(&rfds, &maxfd, a->action_fd);
	370	for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
	371	add_fd(&rfds, &maxfd, mdi->state_fd);
	372	}
	373
	374	if (!nowait) {
	375	rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
	376
	377	if (rv <= 0)
	378	return rv;
	379
	380	if (FD_ISSET(pfd, &rfds)) {
	381	char buf[4];
	382	read(pfd, buf, 4);
	383	; // FIXME read from the pipe
	384	}
	385	}
	386
	387	for (a = aa; a ; a = a->next) {
2a0bb19e	388	if (a->replaces && !discard_this) {
549e9569 NB	389	struct active_array **ap;
	390	for (ap = &a->next; ap && ap != a->replaces;
	391	ap = & (*ap)->next)
	392	;
	393	if (*ap)
	394	ap = (ap)->next;
	395	discard_this = a->replaces;
	396	a->replaces = NULL;
	397	}
2a0bb19e DW	398	if (a->container)
2a0bb19e DW	399	rv += read_and_act(a);
549e9569 NB	400	}
	401	return rv;
	402	}
	403
	404	void do_monitor(struct supertype *container)
	405	{
	406	int rv;
	407	int first = 1;
	408	do {
	409	rv = wait_and_act(container->arrays, container->pipe[0], first);
	410	first = 0;
	411	} while (rv >= 0);
	412	}