xref: /illumos-gate/usr/src/cmd/cmd-inet/usr.lib/in.mpathd/mpd_main.c (revision bbf21555)

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 *
 * Copyright 2021 Tintri by DDN, Inc. All rights reserved.
 */

#include "mpd_defs.h"
#include "mpd_tables.h"

int debug = 0;				/* Debug flag */
static int pollfd_num = 0;		/* Num. of poll descriptors */
static struct pollfd *pollfds = NULL;	/* Array of poll descriptors */
					/* All times below in ms */
int	user_failure_detection_time;	/* user specified failure detection */
					/* time (fdt) */
int	user_probe_interval;		/* derived from user specified fdt */

/*
 * Structure to store mib2 information returned by the kernel.
 * This is used to process routing table information.
 */
typedef struct mib_item_s {
	struct mib_item_s	*mi_next;
	struct opthdr		mi_opthdr;
	void			*mi_valp;
} mib_item_t;

static int	rtsock_v4;		/* AF_INET routing socket */
static int	rtsock_v6;		/* AF_INET6 routing socket */
int	ifsock_v4 = -1;			/* IPv4 socket for ioctls  */
int	ifsock_v6 = -1;			/* IPv6 socket for ioctls  */
static int	lsock_v4;		/* Listen socket to detect mpathd */
static int	lsock_v6;		/* Listen socket to detect mpathd */
static int	mibfd = -1;		/* fd to get mib info */
static boolean_t force_mcast = _B_FALSE; /* Only for test purposes */

static uint_t	last_initifs_time;	/* Time when initifs was last run */
static	char **argv0;			/* Saved for re-exec on SIGHUP */
boolean_t handle_link_notifications = _B_TRUE;
static int	ipRouteEntrySize;	/* Size of IPv4 route entry */
static int	ipv6RouteEntrySize;	/* Size of IPv6 route entry */

static void	initlog(void);
static void	run_timeouts(void);
static void	initifs(void);
static void	check_if_removed(struct phyint_instance *pii);
static void	select_test_ifs(void);
static void	update_router_list(mib_item_t *item);
static void	mib_get_constants(mib_item_t *item);
static int	mibwalk(void (*proc)(mib_item_t *));
static void	ire_process_v4(mib2_ipRouteEntry_t *buf, size_t len);
static void	ire_process_v6(mib2_ipv6RouteEntry_t *buf, size_t len);
static void	router_add_common(int af, char *ifname,
    struct in6_addr nexthop);
static void	init_router_targets();
static void	cleanup(void);
static int	setup_listener(int af);
static void	check_config(void);
static void	check_testconfig(void);
static void	check_addr_unique(struct phyint_instance *,
    struct sockaddr_storage *);
static void	init_host_targets(void);
static void	dup_host_targets(struct phyint_instance *desired_pii);
static void	loopback_cmd(int sock, int family);
static boolean_t daemonize(void);
static int	closefunc(void *, int);
static unsigned int process_cmd(int newfd, union mi_commands *mpi);
static unsigned int process_query(int fd, mi_query_t *miq);
static unsigned int send_addrinfo(int fd, ipmp_addrinfo_t *adinfop);
static unsigned int send_groupinfo(int fd, ipmp_groupinfo_t *grinfop);
static unsigned int send_grouplist(int fd, ipmp_grouplist_t *grlistp);
static unsigned int send_ifinfo(int fd, ipmp_ifinfo_t *ifinfop);
static unsigned int send_result(int fd, unsigned int error, int syserror);

addrlist_t *localaddrs;

/*
 * Return the current time in milliseconds (from an arbitrary reference)
 * truncated to fit into an int. Truncation is ok since we are interested
 * only in differences and not the absolute values.
 */
uint_t
getcurrenttime(void)
{
	uint_t	cur_time;	/* In ms */

	/*
	 * Use of a non-user-adjustable source of time is
	 * required. However millisecond precision is sufficient.
	 * divide by 10^6
	 */
	cur_time = (uint_t)(gethrtime() / 1000000LL);
	return (cur_time);
}

uint64_t
getcurrentsec(void)
{
	return (gethrtime() / NANOSEC);
}

/*
 * Add fd to the set being polled. Returns 0 if ok; -1 if failed.
 */
int
poll_add(int fd)
{
	int i;
	int new_num;
	struct pollfd *newfds;
retry:
	/* Check if already present */
	for (i = 0; i < pollfd_num; i++) {
		if (pollfds[i].fd == fd)
			return (0);
	}
	/* Check for empty spot already present */
	for (i = 0; i < pollfd_num; i++) {
		if (pollfds[i].fd == -1) {
			pollfds[i].fd = fd;
			return (0);
		}
	}

	/* Allocate space for 32 more fds and initialize to -1 */
	new_num = pollfd_num + 32;
	newfds = realloc(pollfds, new_num * sizeof (struct pollfd));
	if (newfds == NULL) {
		logperror("poll_add: realloc");
		return (-1);
	}
	for (i = pollfd_num; i < new_num; i++) {
		newfds[i].fd = -1;
		newfds[i].events = POLLIN;
	}
	pollfd_num = new_num;
	pollfds = newfds;
	goto retry;
}

/*
 * Remove fd from the set being polled. Returns 0 if ok; -1 if failed.
 */
int
poll_remove(int fd)
{
	int i;

	/* Check if already present */
	for (i = 0; i < pollfd_num; i++) {
		if (pollfds[i].fd == fd) {
			pollfds[i].fd = -1;
			return (0);
		}
	}
	return (-1);
}

/*
 * Extract information about the phyint instance. If the phyint instance still
 * exists in the kernel then set pii_in_use, else clear it. check_if_removed()
 * will use it to detect phyint instances that don't exist any longer and
 * remove them, from our database of phyint instances.
 * Return value:
 *	returns true if the phyint instance exists in the kernel,
 *	returns false otherwise
 */
static boolean_t
pii_process(int af, char *name, struct phyint_instance **pii_p)
{
	int err;
	struct phyint_instance *pii;
	struct phyint_instance *pii_other;

	if (debug & D_PHYINT)
		logdebug("pii_process(%s %s)\n", AF_STR(af), name);

	pii = phyint_inst_lookup(af, name);
	if (pii == NULL) {
		/*
		 * Phyint instance does not exist in our tables,
		 * create new phyint instance
		 */
		pii = phyint_inst_init_from_k(af, name);
	} else {
		/* Phyint exists in our tables */
		err = phyint_inst_update_from_k(pii);

		switch (err) {
		case PI_IOCTL_ERROR:
			/* Some ioctl error. don't change anything */
			pii->pii_in_use = 1;
			break;

		case PI_GROUP_CHANGED:
		case PI_IFINDEX_CHANGED:
			/*
			 * Interface index or group membership has changed.
			 * Delete the old state and recreate based on the new
			 * state (it may no longer be in a group).
			 */
			pii_other = phyint_inst_other(pii);
			if (pii_other != NULL)
				phyint_inst_delete(pii_other);
			phyint_inst_delete(pii);
			pii = phyint_inst_init_from_k(af, name);
			break;

		case PI_DELETED:
			/* Phyint instance has disappeared from kernel */
			pii->pii_in_use = 0;
			break;

		case PI_OK:
			/* Phyint instance exists and is fine */
			pii->pii_in_use = 1;
			break;

		default:
			/* Unknown status */
			logerr("pii_process: Unknown status %d\n", err);
			break;
		}
	}

	*pii_p = pii;
	if (pii != NULL)
		return (pii->pii_in_use ? _B_TRUE : _B_FALSE);
	else
		return (_B_FALSE);
}

/*
 * Scan all interfaces to detect changes as well as new and deleted interfaces
 */
static void
initifs()
{
	int	i, nlifr;
	int	af;
	char	*cp;
	char	*buf;
	int	sockfd;
	uint64_t	flags;
	struct lifnum	lifn;
	struct lifconf	lifc;
	struct lifreq	lifreq;
	struct lifreq	*lifr;
	struct logint	*li;
	struct phyint_instance *pii;
	struct phyint_instance *next_pii;
	struct phyint_group *pg, *next_pg;
	char		pi_name[LIFNAMSIZ + 1];

	if (debug & D_PHYINT)
		logdebug("initifs: Scanning interfaces\n");

	last_initifs_time = getcurrenttime();

	/*
	 * Free the existing local address list; we'll build a new list below.
	 */
	addrlist_free(&localaddrs);

	/*
	 * Mark the interfaces so that we can find phyints and logints
	 * which have disappeared from the kernel. pii_process() and
	 * logint_init_from_k() will set {pii,li}_in_use when they find
	 * the interface in the kernel. Also, clear dupaddr bit on probe
	 * logint. check_addr_unique() will set the dupaddr bit on the
	 * probe logint, if the testaddress is not unique.
	 */
	for (pii = phyint_instances; pii != NULL; pii = pii->pii_next) {
		pii->pii_in_use = 0;
		for (li = pii->pii_logint; li != NULL; li = li->li_next) {
			li->li_in_use = 0;
			if (pii->pii_probe_logint == li)
				li->li_dupaddr = 0;
		}
	}

	/*
	 * As above, mark groups so that we can detect IPMP interfaces which
	 * have been removed from the kernel.  Also, delete the group address
	 * list since we'll iteratively recreate it below.
	 */
	for (pg = phyint_groups; pg != NULL; pg = pg->pg_next) {
		pg->pg_in_use = _B_FALSE;
		addrlist_free(&pg->pg_addrs);
	}

	lifn.lifn_family = AF_UNSPEC;
	lifn.lifn_flags = LIFC_ALLZONES | LIFC_UNDER_IPMP;
again:
	if (ioctl(ifsock_v4, SIOCGLIFNUM, (char *)&lifn) < 0) {
		logperror("initifs: ioctl (get interface count)");
		return;
	}
	/*
	 * Pad the interface count to detect when additional interfaces have
	 * been configured between SIOCGLIFNUM and SIOCGLIFCONF.
	 */
	lifn.lifn_count += 4;

	if ((buf = calloc(lifn.lifn_count, sizeof (struct lifreq))) == NULL) {
		logperror("initifs: calloc");
		return;
	}

	lifc.lifc_family = AF_UNSPEC;
	lifc.lifc_flags = LIFC_ALLZONES | LIFC_UNDER_IPMP;
	lifc.lifc_len = lifn.lifn_count * sizeof (struct lifreq);
	lifc.lifc_buf = buf;

	if (ioctl(ifsock_v4, SIOCGLIFCONF, (char *)&lifc) < 0) {
		logperror("initifs: ioctl (get interface configuration)");
		free(buf);
		return;
	}

	/*
	 * If every lifr_req slot is taken, then additional interfaces must
	 * have been plumbed between the SIOCGLIFNUM and the SIOCGLIFCONF.
	 * Recalculate to make sure we didn't miss any interfaces.
	 */
	nlifr = lifc.lifc_len / sizeof (struct lifreq);
	if (nlifr >= lifn.lifn_count) {
		free(buf);
		goto again;
	}

	/*
	 * Walk through the lifreqs returned by SIOGGLIFCONF, and refresh the
	 * global list of addresses, phyint groups, phyints, and logints.
	 */
	for (lifr = lifc.lifc_req, i = 0; i < nlifr; i++, lifr++) {
		af = lifr->lifr_addr.ss_family;
		sockfd = (af == AF_INET) ? ifsock_v4 : ifsock_v6;
		(void) strlcpy(lifreq.lifr_name, lifr->lifr_name, LIFNAMSIZ);

		if (ioctl(sockfd, SIOCGLIFFLAGS, &lifreq) == -1) {
			if (errno != ENXIO)
				logperror("initifs: ioctl (SIOCGLIFFLAGS)");
			continue;
		}
		flags = lifreq.lifr_flags;

		/*
		 * If the address is IFF_UP, add it to the local address list.
		 * (We ignore addresses that aren't IFF_UP since another node
		 * might legitimately have that address IFF_UP.)
		 */
		if (flags & IFF_UP) {
			(void) addrlist_add(&localaddrs, lifr->lifr_name, flags,
			    &lifr->lifr_addr);
		}

		/*
		 * If this address is on an IPMP meta-interface, update our
		 * phyint_group information (either by recording that group
		 * still exists or creating a new group), and track what
		 * group the address is part of.
		 */
		if (flags & IFF_IPMP) {
			if (ioctl(sockfd, SIOCGLIFGROUPNAME, &lifreq) == -1) {
				if (errno != ENXIO)
					logperror("initifs: ioctl "
					    "(SIOCGLIFGROUPNAME)");
				continue;
			}

			pg = phyint_group_lookup(lifreq.lifr_groupname);
			if (pg == NULL) {
				pg = phyint_group_create(lifreq.lifr_groupname);
				if (pg == NULL) {
					logerr("initifs: cannot create group "