[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 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);
}

#define DS_FAULTY	1
#define	DS_INSYNC	2
#define	DS_WRITE_MOSTLY	4
#define	DS_SPARE	8
#define	DS_REMOVE	1024

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("faulty", cp))
			rv |= DS_FAULTY;
		if (attr_match("in_sync", cp))
			rv |= DS_INSYNC;
		if (attr_match("write_mostly", cp))
			rv |= DS_WRITE_MOSTLY;
		if (attr_match("spare", cp))
			rv |= DS_SPARE;
		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, the remove device
 *    by writing '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....
 *      Maybe this is done by mdadm before passing the array to us?
 *
 *    If array is 'clean' but metadata is 'dirty', start a resync
 *    and mark array as 'dirty'.
 *
 *
 *
 *
 * 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;
	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;
		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;
	}
	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 ???
		 */
		a->next_state = read_auto;
	}

	if (a->curr_action == idle &&
	    a->prev_action == resync) {
		/* check resync_start to see if it is 'max'.
		 * Do I open here, or have it open the whole time?
		 */
		get_sync_pos(a);
		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);
			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);
			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)
			write_attr("remove", mdi->state_fd);
		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;
	}

	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;

		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) {
			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;
		}
		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
	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
	67
	68	static int attr_match(const char attr, const char str)
	69	{
	70	/* See if attr, read from a sysfs file, matches
	71	* str. They must either be the same, or attr can
	72	* have a trailing newline or comma
	73	*/
	74	while (attr && str && attr == str) {
	75	attr++;
	76	str++;
	77	}
	78
	79	if (str \|\| (attr && attr != ',' && attr != '\n'))
	80	return 0;
	81	return 1;
	82	}
	83
	84	static int match_word(const char word, char *list)
	85	{
	86	int n;
	87	for (n=0; list[n]; n++)
	88	if (attr_match(word, list[n]))
	89	break;
	90	return n;
	91	}
	92
	93	static enum array_state read_state(int fd)
	94	{
	95	char buf[20];
	96	int n = read_attr(buf, 20, fd);
	97
	98	if (n <= 0)
	99	return bad_word;
	100	return (enum array_state) match_word(buf, array_states);
	101	}
	102
	103	static enum sync_action read_action( int fd)
	104	{
	105	char buf[20];
	106	int n = read_attr(buf, 20, fd);
	107
	108	if (n <= 0)
	109	return bad_action;
	110	return (enum sync_action) match_word(buf, sync_actions);
	111	}
	112
	113	#define DS_FAULTY 1
	114	#define DS_INSYNC 2
	115	#define DS_WRITE_MOSTLY 4
	116	#define DS_SPARE 8
	117	#define DS_REMOVE 1024
	118
	119	int read_dev_state(int fd)
	120	{
	121	char buf[60];
	122	int n = read_attr(buf, 60, fd);
	123	char *cp;
	124	int rv = 0;
	125
	126	if (n <= 0)
	127	return 0;
	128
	129	cp = buf;
	130	while (cp) {
	131	if (attr_match("faulty", cp))
	132	rv \|= DS_FAULTY;
	133	if (attr_match("in_sync", cp))
	134	rv \|= DS_INSYNC;
	135	if (attr_match("write_mostly", cp))
	136	rv \|= DS_WRITE_MOSTLY;
	137	if (attr_match("spare", cp))
	138	rv \|= DS_SPARE;
	139	cp = strchr(cp, ',');
	140	if (cp)
	141	cp++;
	142	}
	143	return rv;
	144	}
	145
	146
	147	/* Monitor a set of active md arrays - all of which share the
	148	* same metadata - and respond to events that require
	149	* metadata update.
	150	*
	151	* New arrays are detected by another thread which allocates
	152	* required memory and attaches the data structure to our list.
	153	*
	154	* Events:
	155	* Array stops.
	156	* This is detected by array_state going to 'clear' or 'inactive'.
	157	* while we thought it was active.
	158	* Response is to mark metadata as clean and 'clear' the array(??)
	159	* write-pending
	160	* array_state if 'write-pending'
	161	* We mark metadata as 'dirty' then set array to 'active'.
	162	* active_idle
	163	* Either ignore, or mark clean, then mark metadata as clean.
	164	*
	165	* device fails
	166	* detected by rd-N/state reporting "faulty"
	167	* mark device as 'failed' in metadata, the remove device
	168	* by writing 'remove' to rd/state.
	169	*
	170	* sync completes
	171	* sync_action was 'resync' and becomes 'idle' and resync_start becomes
	172	* MaxSector
	173	* Notify metadata that sync is complete.
	174	* "Deal with Degraded"
	175	*
	176	* recovery completes
	177	* sync_action changes from 'recover' to 'idle'
	178	* Check each device state and mark metadata if 'faulty' or 'in_sync'.
	179	* "Deal with Degraded"
	180	*
	181	* deal with degraded array
	182	* We only do this when first noticing the array is degraded.
	183	* This can be when we first see the array, when sync completes or
	184	* when recovery completes.
	185	*
	186	* Check if number of failed devices suggests recovery is needed, and
	187	* skip if not.
	188	* Ask metadata for a spare device
	189	* Add device as not in_sync and give a role
	190	* Update metadata.
	191	* Start recovery.
	192	*
	193	* deal with resync
	194	* This only happens on finding a new array....
	195	* Maybe this is done by mdadm before passing the array to us?
	196	*
	197	* If array is 'clean' but metadata is 'dirty', start a resync
	198	* and mark array as 'dirty'.
	199	*
	200	*
	201	*
	202	*
	203	* We wait for a change (poll/select) on array_state, sync_action, and
	204	* each rd-X/state file.
	205	* When we get any change, we check everything. So read each state file,
	206	* then decide what to do.
	207	*
	208	* The core action is to write new metadata to all devices in the array.
	209	* This is done at most once on any wakeup.
	210	* After that we might:
	211	* - update the array_state
	212	* - set the role of some devices.
	213	* - request a sync_action
	214	*
	215	*/
	216
	217	static int read_and_act(struct active_array *a)
	218	{
	219	int check_degraded;
	220	struct mdinfo *mdi;
	221
	222	a->next_state = bad_word;
	223	a->next_action = bad_action;
	224
	225	a->curr_state = read_state(a->info.state_fd);
	226	a->curr_action = read_action(a->action_fd);
	227	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	228	mdi->next_state = 0;
	229	mdi->curr_state = read_dev_state(mdi->state_fd);
	230	}
	231
	232	if (a->curr_state <= inactive &&
	233	a->prev_state > inactive) {
	234	/* array has been stopped */
	235	get_sync_pos(a);
	236	a->container->ss->mark_clean(a, a->sync_pos);
	237	a->next_state = clear;
	238	}
	239	if (a->curr_state == write_pending) {
	240	a->container->ss->mark_dirty(a);
	241	a->next_state = active;
	242	}
	243	if (a->curr_state == active_idle) {
	244	/* Set array to 'clean' FIRST, then
	245	* a->ss->mark_clean(a);
	246	* just ignore for now.
	247	*/
	248	}
	249
	250	if (a->curr_state == readonly) {
	251	/* Well, I'm ready to handle things, so
	252	* read-auto is OK. FIXME what if we really want
	253	* readonly ???
	254	*/
	255	a->next_state = read_auto;
	256	}
	257
	258	if (a->curr_action == idle &&
	259	a->prev_action == resync) {
	260	/* check resync_start to see if it is 'max'.
	261	* Do I open here, or have it open the whole time?
	262	*/
	263	get_sync_pos(a);
	264	check_degraded = 1;
	265	}
	266
	267	if (a->curr_action == idle &&
	268	a->prev_action == recover) {
	269	for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
	270	a->container->ss->set_disk(a, mdi->disk.raid_disk);
	271	if (! (mdi->curr_state & DS_INSYNC))
	272	check_degraded = 1;
	273	}
	274	}
	275
	276
	277	for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
	278	if (mdi->curr_state & DS_FAULTY) {
	279	a->container->ss->set_disk(a, mdi->disk.raid_disk);
	280	check_degraded = 1;
	281	mdi->next_state = DS_REMOVE;
	282	}
	283	}
	284
	285	if (check_degraded) {
	286	// FIXME;
	287	}
	288
	289	a->container->ss->sync_metadata(a);
	290
	291	/* Effect state changes in the array */
	292	if (a->next_state != bad_word)
	293	write_attr(array_states[a->next_state], a->info.state_fd);
	294	if (a->next_action != bad_action)
	295	write_attr(sync_actions[a->next_action], a->action_fd);
	296	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	297	if (mdi->next_state == DS_REMOVE)
	298	write_attr("remove", mdi->state_fd);
	299	if (mdi->next_state & DS_INSYNC)
	300	write_attr("+in_sync", mdi->state_fd);
	301	}
	302
	303	/* move curr_ to prev_ */
	304	a->prev_state = a->curr_state;
	305
	306	a->prev_action = a->curr_action;
	307
	308	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	309	mdi->prev_state = mdi->curr_state;
	310	mdi->next_state = 0;
	311	}
	312
	313	return 1;
	314	}
	315
	316	static int wait_and_act(struct active_array *aa, int pfd, int nowait)
	317	{
	318	fd_set rfds;
	319	int maxfd = 0;
	320	struct active_array *a;
	321	int rv;
	322
	323	FD_ZERO(&rfds);
	324
	325	add_fd(&rfds, &maxfd, pfd);
	326	for (a = aa ; a ; a = a->next) {
	327	struct mdinfo *mdi;
	328
	329	add_fd(&rfds, &maxfd, a->info.state_fd);
	330	add_fd(&rfds, &maxfd, a->action_fd);
	331	for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
	332	add_fd(&rfds, &maxfd, mdi->state_fd);
	333	}
	334
	335	if (!nowait) {
	336	rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
	337
	338	if (rv <= 0)
	339	return rv;
	340
	341	if (FD_ISSET(pfd, &rfds)) {
	342	char buf[4];
	343	read(pfd, buf, 4);
	344	; // FIXME read from the pipe
	345	}
	346	}
	347
	348	for (a = aa; a ; a = a->next) {
	349	if (a->replaces) {
	350	struct active_array **ap;
	351	for (ap = &a->next; ap && ap != a->replaces;
	352	ap = & (*ap)->next)
	353	;
	354	if (*ap)
	355	ap = (ap)->next;
	356	discard_this = a->replaces;
	357	a->replaces = NULL;
	358	}
	359	rv += read_and_act(a);
	360	}
	361	return rv;
	362	}
	363
	364	void do_monitor(struct supertype *container)
	365	{
	366	int rv;
	367	int first = 1;
	368	do {
	369	rv = wait_and_act(container->arrays, container->pipe[0], first);
	370	first = 0;
	371	} while (rv >= 0);
	372	}