[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);
}

#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... 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;
	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 ???
		 */
		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);
			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
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
	126	#define DS_FAULTY 1
	127	#define DS_INSYNC 2
	128	#define DS_WRITE_MOSTLY 4
	129	#define DS_SPARE 8
	130	#define DS_REMOVE 1024
	131
	132	int read_dev_state(int fd)
	133	{
	134	char buf[60];
	135	int n = read_attr(buf, 60, fd);
	136	char *cp;
	137	int rv = 0;
	138
	139	if (n <= 0)
	140	return 0;
	141
	142	cp = buf;
	143	while (cp) {
144	if (attr_match("faulty", cp))
145	rv \|= DS_FAULTY;
146	if (attr_match("in_sync", cp))
147	rv \|= DS_INSYNC;
148	if (attr_match("write_mostly", cp))
149	rv \|= DS_WRITE_MOSTLY;
150	if (attr_match("spare", cp))
151	rv \|= DS_SPARE;
152	cp = strchr(cp, ',');
153	if (cp)
154	cp++;
155	}
156	return rv;
157	}
158
159
160	/* Monitor a set of active md arrays - all of which share the
161	* same metadata - and respond to events that require
162	* metadata update.
163	*
164	* New arrays are detected by another thread which allocates
165	* required memory and attaches the data structure to our list.
166	*
167	* Events:
168	* Array stops.
169	* This is detected by array_state going to 'clear' or 'inactive'.
170	* while we thought it was active.
171	* Response is to mark metadata as clean and 'clear' the array(??)
172	* write-pending
173	* array_state if 'write-pending'
174	* We mark metadata as 'dirty' then set array to 'active'.
175	* active_idle
176	* Either ignore, or mark clean, then mark metadata as clean.
177	*
178	* device fails
179	* detected by rd-N/state reporting "faulty"
180	* mark device as 'failed' in metadata, the remove device
181	* by writing 'remove' to rd/state.
182	*
183	* sync completes
184	* sync_action was 'resync' and becomes 'idle' and resync_start becomes
185	* MaxSector
186	* Notify metadata that sync is complete.
187	* "Deal with Degraded"
188	*
189	* recovery completes
190	* sync_action changes from 'recover' to 'idle'
191	* Check each device state and mark metadata if 'faulty' or 'in_sync'.
192	* "Deal with Degraded"
193	*
194	* deal with degraded array
195	* We only do this when first noticing the array is degraded.
196	* This can be when we first see the array, when sync completes or
197	* when recovery completes.
198	*
199	* Check if number of failed devices suggests recovery is needed, and
200	* skip if not.
201	* Ask metadata for a spare device
202	* Add device as not in_sync and give a role
203	* Update metadata.
204	* Start recovery.
205	*
206	* deal with resync
c052ba30 DW	207	* This only happens on finding a new array... mdadm will have set
	208	* 'resync_start' to the correct value. If 'resync_start' indicates that an
	209	* resync needs to occur set the array to the 'active' state rather than the
	210	* initial read-auto state.
549e9569 NB	211	*
	212	*
	213	*
	214	* We wait for a change (poll/select) on array_state, sync_action, and
	215	* each rd-X/state file.
	216	* When we get any change, we check everything. So read each state file,
	217	* then decide what to do.
	218	*
	219	* The core action is to write new metadata to all devices in the array.
	220	* This is done at most once on any wakeup.
	221	* After that we might:
	222	* - update the array_state
	223	* - set the role of some devices.
	224	* - request a sync_action
	225	*
	226	*/
	227
	228	static int read_and_act(struct active_array *a)
	229	{
	230	int check_degraded;
	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;
	240	mdi->curr_state = read_dev_state(mdi->state_fd);
	241	}
	242
	243	if (a->curr_state <= inactive &&
	244	a->prev_state > inactive) {
	245	/* array has been stopped */
	246	get_sync_pos(a);
	247	a->container->ss->mark_clean(a, a->sync_pos);
	248	a->next_state = clear;
	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) {
	286	a->container->ss->set_disk(a, mdi->disk.raid_disk);
	287	if (! (mdi->curr_state & DS_INSYNC))
	288	check_degraded = 1;
	289	}
	290	}
	291
	292
	293	for (mdi = a->info.devs ; mdi ; mdi = mdi->next) {
	294	if (mdi->curr_state & DS_FAULTY) {
	295	a->container->ss->set_disk(a, mdi->disk.raid_disk);
	296	check_degraded = 1;
	297	mdi->next_state = DS_REMOVE;
	298	}
	299	}
	300
	301	if (check_degraded) {
	302	// FIXME;
	303	}
	304
	305	a->container->ss->sync_metadata(a);
	306
	307	/* Effect state changes in the array */
	308	if (a->next_state != bad_word)
	309	write_attr(array_states[a->next_state], a->info.state_fd);
	310	if (a->next_action != bad_action)
	311	write_attr(sync_actions[a->next_action], a->action_fd);
	312	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	313	if (mdi->next_state == DS_REMOVE)
	314	write_attr("remove", mdi->state_fd);
	315	if (mdi->next_state & DS_INSYNC)
	316	write_attr("+in_sync", mdi->state_fd);
	317	}
	318
	319	/* move curr_ to prev_ */
	320	a->prev_state = a->curr_state;
	321
	322	a->prev_action = a->curr_action;
	323
	324	for (mdi = a->info.devs; mdi ; mdi = mdi->next) {
	325	mdi->prev_state = mdi->curr_state;
	326	mdi->next_state = 0;
	327	}
	328
	329	return 1;
	330	}
	331
	332	static int wait_and_act(struct active_array *aa, int pfd, int nowait)
	333	{
	334	fd_set rfds;
	335	int maxfd = 0;
	336	struct active_array *a;
	337	int rv;
	338
	339	FD_ZERO(&rfds);
	340
	341	add_fd(&rfds, &maxfd, pfd);
	342	for (a = aa ; a ; a = a->next) {
	343	struct mdinfo *mdi;
344
345	add_fd(&rfds, &maxfd, a->info.state_fd);
346	add_fd(&rfds, &maxfd, a->action_fd);
347	for (mdi = a->info.devs ; mdi ; mdi = mdi->next)
348	add_fd(&rfds, &maxfd, mdi->state_fd);
349	}
350
351	if (!nowait) {
352	rv = select(maxfd+1, &rfds, NULL, NULL, NULL);
353
354	if (rv <= 0)
355	return rv;
356
357	if (FD_ISSET(pfd, &rfds)) {
358	char buf[4];
359	read(pfd, buf, 4);
360	; // FIXME read from the pipe
361	}
362	}
363
364	for (a = aa; a ; a = a->next) {
365	if (a->replaces) {
366	struct active_array **ap;
367	for (ap = &a->next; ap && ap != a->replaces;
368	ap = & (*ap)->next)
369	;
370	if (*ap)
371	ap = (ap)->next;
372	discard_this = a->replaces;
373	a->replaces = NULL;
374	}
375	rv += read_and_act(a);
376	}
377	return rv;
378	}
379
380	void do_monitor(struct supertype *container)
381	{
382	int rv;
383	int first = 1;
384	do {
385	rv = wait_and_act(container->arrays, container->pipe[0], first);
386	first = 0;
387	} while (rv >= 0);
388	}