[illumos-gate merge]

commit 3ee592424ed4bb7b850d9adccb9f3c493ce7534b
    3729 getifaddrs must learn to stop worrying and love the other address families

7 files changed, 358 insertions, 37 deletions

diff --git a/usr/src/cmd/fs.d/nfs/statd/sm_proc.c b/usr/src/cmd/fs.d/nfs/statd/sm_proc.c
index f3e050df8c..7c1f434ab7 100644
--- a/usr/src/cmd/fs.d/nfs/statd/sm_proc.c
+++ b/usr/src/cmd/fs.d/nfs/statd/sm_proc.c
@@ -1353,6 +1353,9 @@ merge_ips(void)
 			break;
 		}
 
+		case AF_LINK:
+			continue;
+
 		default:
 			syslog(LOG_WARNING, "Unknown address family %d for "
 			    "interface %s", sa->sa_family, cifap->ifa_name);
diff --git a/usr/src/head/ifaddrs.h b/usr/src/head/ifaddrs.h
index 4202b60998..815883948b 100644
--- a/usr/src/head/ifaddrs.h
+++ b/usr/src/head/ifaddrs.h
@@ -20,6 +20,7 @@
  */
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
+ * Copyright 2022 Sebastian Wiedenroth
  */
 #ifndef _IFADDRS_H
 #define	_IFADDRS_H
@@ -65,6 +66,13 @@ struct ifaddrs {
 };
 #endif
 
+#ifdef __PRAGMA_REDEFINE_EXTNAME
+#pragma redefine_extname getifaddrs __getifaddrs
+#else
+extern int __getifaddrs(struct ifaddrs **);
+#define	getifaddrs __getifaddrs
+#endif
+
 /*
  * Create a linked list of `struct ifaddrs' structures, one for each
  * network interface on the host machine.  If successful, store the
diff --git a/usr/src/lib/libipadm/common/ipadm_addr.c b/usr/src/lib/libipadm/common/ipadm_addr.c
index 46bb609bdb..64a6f09485 100644
--- a/usr/src/lib/libipadm/common/ipadm_addr.c
+++ b/usr/src/lib/libipadm/common/ipadm_addr.c
@@ -415,6 +415,9 @@ retry:
 	for (ifap = ifa; ifap != NULL; ifap = ifap->ifa_next) {
 		struct sockaddr_storage data;
 
+		if (ifap->ifa_addr->sa_family == AF_LINK)
+			continue;
+
 		(void) strlcpy(cifname, ifap->ifa_name, sizeof (cifname));
 		lnum = 0;
 		if ((sep = strrchr(cifname, ':')) != NULL) {
diff --git a/usr/src/lib/libipadm/common/ipadm_if.c b/usr/src/lib/libipadm/common/ipadm_if.c
index 7699937c4a..8bf65721f1 100644
--- a/usr/src/lib/libipadm/common/ipadm_if.c
+++ b/usr/src/lib/libipadm/common/ipadm_if.c
@@ -414,6 +414,8 @@ retry:
 			 * to find the interface state.
 			 */
 			for (ifap = ifa; ifap != NULL; ifap = ifap->ifa_next) {
+				if (ifap->ifa_addr->sa_family == AF_LINK)
+					continue;
 				if (strcmp(ifap->ifa_name, aifp->ifi_name) == 0)
 					break;
 			}
diff --git a/usr/src/lib/libsocket/common/mapfile-vers b/usr/src/lib/libsocket/common/mapfile-vers
index 614094c3f1..3329f3bfd3 100644
--- a/usr/src/lib/libsocket/common/mapfile-vers
+++ b/usr/src/lib/libsocket/common/mapfile-vers
@@ -41,6 +41,11 @@
 
 $mapfile_version 2
 
+SYMBOL_VERSION ILLUMOS_0.3 {	# AF_LINK workaround
+    global:
+	__getifaddrs;
+} ILLUMOS_0.2;
+
 SYMBOL_VERSION ILLUMOS_0.2 {	# reentrant ethers(3SOCKET)
     global:
 	ether_aton_r;
diff --git a/usr/src/lib/libsocket/inet/getifaddrs.c b/usr/src/lib/libsocket/inet/getifaddrs.c
index 66dfedbd10..2b379904de 100644
--- a/usr/src/lib/libsocket/inet/getifaddrs.c
+++ b/usr/src/lib/libsocket/inet/getifaddrs.c
@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright 2017 RackTop Systems.
+ * Copyright 2022 Sebastian Wiedenroth
  */
 
 #include <netdb.h>
@@ -29,15 +30,35 @@
 #include <netinet/in.h>
 #include <sys/socket.h>
 #include <string.h>
+#include <strings.h>
 #include <stdio.h>
 #include <sys/sockio.h>
 #include <sys/types.h>
 #include <stdlib.h>
 #include <net/if.h>
+#include <door.h>
+#include <fcntl.h>
+#include <sys/mman.h>
+#include <sys/dld_ioc.h>
+#include <sys/dld.h>
+#include <sys/dls_mgmt.h>
+#include <sys/mac.h>
+#include <sys/dlpi.h>
+#include <net/if_types.h>
 #include <ifaddrs.h>
 #include <libsocket_priv.h>
 
 /*
+ * <ifaddrs.h> directs folks towards our internal symbol, __getifaddrs. This
+ * means we cannot name the original symbol 'getifaddrs' here or it will be
+ * renamed. Instead, we use another redefine_extname to take care of this. Note,
+ * the extern declaration is required as gcc and others will only apply this for
+ * things they see an extern declaration for.
+ */
+#pragma redefine_extname getifaddrs_old getifaddrs
+extern int getifaddrs_old(struct ifaddrs **);
+
+/*
  * Create a linked list of `struct ifaddrs' structures, one for each
  * address that is UP. If successful, store the list in *ifap and
  * return 0.  On errors, return -1 and set `errno'.
@@ -46,7 +67,7 @@
  * only be properly freed by passing it to `freeifaddrs'.
  */
 int
-getifaddrs(struct ifaddrs **ifap)
+_getifaddrs(struct ifaddrs **ifap, boolean_t can_handle_links)
 {
 	int		err;
 	char		*cp;
@@ -57,14 +78,57 @@ getifaddrs(struct ifaddrs **ifap)
 		return (-1);
 	}
 	*ifap = NULL;
-	err = getallifaddrs(AF_UNSPEC, ifap, LIFC_ENABLED);
-	if (err == 0) {
-		for (curr = *ifap; curr != NULL; curr = curr->ifa_next) {
-			if ((cp = strchr(curr->ifa_name, ':')) != NULL)
-				*cp = '\0';
+
+	if (can_handle_links) {
+		err = getallifaddrs(AF_UNSPEC, ifap, LIFC_ENABLED);
+	} else {
+		err = getallifaddrs(AF_INET, ifap, LIFC_ENABLED);
+		if (err != 0)
+			return (err);
+
+		/* Find end of the list to append to */
+		curr = *ifap;
+		while (curr && curr->ifa_next) {
+			curr = curr->ifa_next;
 		}
+
+		err = getallifaddrs(AF_INET6, curr ? &curr->ifa_next : ifap,
+		    LIFC_ENABLED);
+	}
+
+	if (err != 0)
+		return (err);
+
+	for (curr = *ifap; curr != NULL; curr = curr->ifa_next) {
+		if ((cp = strchr(curr->ifa_name, ':')) != NULL)
+			*cp = '\0';
 	}
-	return (err);
+
+	return (0);
+}
+
+/*
+ * Legacy symbol
+ * For a long time getifaddrs() only returned AF_INET and AF_INET6 entries.
+ * Some consumers came to expect that no other address family may be returned.
+ * To prevent existing binaries that can't handle AF_LINK entries from breaking
+ * this symbol is kept around. Consumers that want the fixed behaviour need to
+ * recompile and link to the fixed symbol.
+ */
+int
+getifaddrs_old(struct ifaddrs **ifap)
+{
+	return (_getifaddrs(ifap, B_FALSE));
+}
+
+/*
+ * Current symbol
+ * May return AF_INET, AF_INET6 and AF_LINK entries
+ */
+int
+__getifaddrs(struct ifaddrs **ifap)
+{
+	return (_getifaddrs(ifap, B_TRUE));
 }
 
 void
@@ -79,10 +143,140 @@ freeifaddrs(struct ifaddrs *ifa)
 		free(curr->ifa_addr);
 		free(curr->ifa_netmask);
 		free(curr->ifa_dstaddr);
+		free(curr->ifa_data);
 		free(curr);
 	}
 }
 
+static uint_t
+dlpi_iftype(uint_t dlpitype)
+{
+	switch (dlpitype) {
+	case DL_ETHER:
+		return (IFT_ETHER);
+
+	case DL_ATM:
+		return (IFT_ATM);
+
+	case DL_CSMACD:
+		return (IFT_ISO88023);
+
+	case DL_TPB:
+		return (IFT_ISO88024);
+
+	case DL_TPR:
+		return (IFT_ISO88025);
+
+	case DL_FDDI:
+		return (IFT_FDDI);
+
+	case DL_IB:
+		return (IFT_IB);
+
+	case DL_OTHER:
+		return (IFT_OTHER);
+	}
+
+	return (IFT_OTHER);
+}
+
+/*
+ * Make a door call to dlmgmtd.
+ * If successful the result is stored in rbuf and 0 returned.
+ * On errors, return -1 and set `errno'.
+ */
+static int
+dl_door_call(int door_fd, void *arg, size_t asize, void *rbuf, size_t *rsizep)
+{
+	int err;
+	door_arg_t	darg;
+	darg.data_ptr	= arg;
+	darg.data_size	= asize;
+	darg.desc_ptr	= NULL;
+	darg.desc_num	= 0;
+	darg.rbuf	= rbuf;
+	darg.rsize	= *rsizep;
+
+	if (door_call(door_fd, &darg) == -1) {
+		return (-1);
+	}
+
+	if (darg.rbuf != rbuf) {
+		/*
+		 * The size of the input rbuf was not big enough so that
+		 * the door allocated the rbuf itself. In this case, return
+		 * the required size to the caller.
+		 */
+		err = errno;
+		(void) munmap(darg.rbuf, darg.rsize);
+		*rsizep = darg.rsize;
+		errno = err;
+		return (-1);
+	} else if (darg.rsize != *rsizep) {
+		return (-1);
+	}
+	return (0);
+}
+
+
+/*
+ * Get the name from dlmgmtd by linkid.
+ * If successful the result is stored in name_retval and 0 returned.
+ * On errors, return -1 and set `errno'.
+ */
+static int
+dl_get_name(int door_fd, datalink_id_t linkid,
+    dlmgmt_getname_retval_t *name_retval)
+{
+	size_t name_sz = sizeof (*name_retval);
+	dlmgmt_door_getname_t getname;
+	bzero(&getname, sizeof (dlmgmt_door_getname_t));
+	getname.ld_cmd = DLMGMT_CMD_GETNAME;
+	getname.ld_linkid = linkid;
+
+	if (dl_door_call(door_fd, &getname, sizeof (getname), name_retval,
+	    &name_sz) < 0) {
+		return (-1);
+	}
+	if (name_retval->lr_err != 0) {
+		errno = name_retval->lr_err;
+		return (-1);
+	}
+	return (0);
+}
+
+/*
+ * Get the next link from dlmgmtd.
+ * Start iterating by passing DATALINK_INVALID_LINKID as linkid.
+ * The end is marked by next_retval.lr_linkid set to DATALINK_INVALID_LINKID.
+ * If successful the result is stored in next_retval and 0 returned.
+ * On errors, return -1 and set `errno'.
+ */
+static int
+dl_get_next(int door_fd, datalink_id_t linkid, datalink_class_t class,
+    datalink_media_t dmedia, uint32_t flags,
+    dlmgmt_getnext_retval_t *next_retval)
+{
+	size_t next_sz = sizeof (*next_retval);
+	dlmgmt_door_getnext_t getnext;
+	bzero(&getnext, sizeof (dlmgmt_door_getnext_t));
+	getnext.ld_cmd = DLMGMT_CMD_GETNEXT;
+	getnext.ld_class = class;
+	getnext.ld_dmedia = dmedia;
+	getnext.ld_flags = flags;
+	getnext.ld_linkid = linkid;
+
+	if (dl_door_call(door_fd, &getnext, sizeof (getnext), next_retval,
+	    &next_sz) < 0) {
+		return (-1);
+	}
+	if (next_retval->lr_err != 0) {
+		errno = next_retval->lr_err;
+		return (-1);
+	}
+	return (0);
+}
+
 /*
  * Returns all addresses configured on the system. If flags contain
  * LIFC_ENABLED, only the addresses that are UP are returned.
@@ -98,9 +292,21 @@ getallifaddrs(sa_family_t af, struct ifaddrs **ifap, int64_t flags)
 	int ret;
 	int s, n, numifs;
 	struct ifaddrs *curr, *prev;
+	struct sockaddr_dl *ifa_addr = NULL;
+	if_data_t *ifa_data = NULL;
 	sa_family_t lifr_af;
-	int sock4;
-	int sock6;
+	datalink_id_t linkid;
+	dld_ioc_attr_t dia;
+	dld_macaddrinfo_t *dmip;
+	dld_ioc_macaddrget_t *iomp = NULL;
+	dlmgmt_getnext_retval_t next_retval;
+	dlmgmt_getname_retval_t	name_retval;
+	int bufsize;
+	int nmacaddr = 1024;
+	int sock4 = -1;
+	int sock6 = -1;
+	int door_fd = -1;
+	int dld_fd = -1;
 	int err;
 
 	/*
@@ -111,14 +317,16 @@ getallifaddrs(sa_family_t af, struct ifaddrs **ifap, int64_t flags)
 		return (EINVAL);
 	*ifap = NULL;
 
-	if ((sock4 = socket(AF_INET, SOCK_DGRAM, 0)) < 0)
-		return (-1);
-	if ((sock6 = socket(AF_INET6, SOCK_DGRAM, 0)) < 0) {
-		err = errno;
-		close(sock4);
-		errno = err;
-		return (-1);
-	}
+	if ((sock4 = socket(AF_INET, SOCK_DGRAM, 0)) < 0 ||
+	    (sock6 = socket(AF_INET6, SOCK_DGRAM, 0)) < 0 ||
+	    (door_fd = open(DLMGMT_DOOR, O_RDONLY)) < 0 ||
+	    (dld_fd = open(DLD_CONTROL_DEV, O_RDWR)) < 0)
+		goto fail;
+
+	bufsize = sizeof (dld_ioc_macaddrget_t) + nmacaddr *
+	    sizeof (dld_macaddrinfo_t);
+	if ((iomp = calloc(1, bufsize)) == NULL)
+		goto fail;
 
 retry:
 	/* Get all interfaces from SIOCGLIFCONF */
@@ -205,19 +413,118 @@ retry:
 		}
 
 	}
+
+	/* add AF_LINK entries */
+	if (af == AF_UNSPEC || af == AF_LINK) {
+
+		linkid = DATALINK_INVALID_LINKID;
+		for (;;) {
+			if (dl_get_next(door_fd, linkid, DATALINK_CLASS_ALL,
+			    DATALINK_ANY_MEDIATYPE, DLMGMT_ACTIVE,
+			    &next_retval) != 0) {
+				break;
+			}
+
+			linkid = next_retval.lr_linkid;
+			if (linkid == DATALINK_INVALID_LINKID)
+				break;
+
+			/* get mac addr */
+			iomp->dig_size = nmacaddr * sizeof (dld_macaddrinfo_t);
+			iomp->dig_linkid = linkid;
+
+			if (ioctl(dld_fd, DLDIOC_MACADDRGET, iomp) < 0)
+				continue;
+
+			dmip = (dld_macaddrinfo_t *)(iomp + 1);
+
+			/* get name */
+			if (dl_get_name(door_fd, linkid, &name_retval) != 0)
+				continue;
+
+			/* get MTU */
+			dia.dia_linkid = linkid;
+			if (ioctl(dld_fd, DLDIOC_ATTR, &dia) < 0)
+				continue;
+
+			curr = calloc(1, sizeof (struct ifaddrs));
+			if (curr == NULL)
+				goto fail;
+
+			if (prev != NULL) {
+				prev->ifa_next = curr;
+			} else {
+				/* First node in the linked list */
+				*ifap = curr;
+			}
+			prev = curr;
+
+			if ((curr->ifa_name = strdup(name_retval.lr_link)) ==
+			    NULL)
+				goto fail;
+
+			curr->ifa_addr =
+			    calloc(1, sizeof (struct sockaddr_storage));
+			if (curr->ifa_addr == NULL)
+				goto fail;
+
+			curr->ifa_data = calloc(1, sizeof (if_data_t));
+			if (curr->ifa_data == NULL)
+				goto fail;
+
+			curr->ifa_addr->sa_family = AF_LINK;
+			ifa_addr = (struct sockaddr_dl *)curr->ifa_addr;
+			ifa_data = curr->ifa_data;
+
+			(void) memcpy(ifa_addr->sdl_data, dmip->dmi_addr,
+			    dmip->dmi_addrlen);
+			ifa_addr->sdl_alen = dmip->dmi_addrlen;
+
+			ifa_data->ifi_mtu = dia.dia_max_sdu;
+			ifa_data->ifi_type = dlpi_iftype(next_retval.lr_media);
+
+			/*
+			 * get interface index
+			 * This is only possible if the link has been plumbed.
+			 */
+			if (strlcpy(lifrl.lifr_name, name_retval.lr_link,
+			    sizeof (lifrl.lifr_name)) >=
+			    sizeof (lifrl.lifr_name))
+				continue;
+
+			if (ioctl(sock4, SIOCGLIFINDEX, (caddr_t)&lifrl) >= 0) {
+				ifa_addr->sdl_index = lifrl.lifr_index;
+			} else if (ioctl(sock6, SIOCGLIFINDEX,
+			    (caddr_t)&lifrl) >= 0) {
+				/* retry for IPv6 */
+				ifa_addr->sdl_index = lifrl.lifr_index;
+			}
+		}
+	}
 	free(buf);
-	close(sock4);
-	close(sock6);
+	free(iomp);
+	(void) close(sock4);
+	(void) close(sock6);
+	(void) close(door_fd);
+	(void) close(dld_fd);
 	return (0);
 fail:
 	err = errno;
 	free(buf);
+	free(iomp);
 	freeifaddrs(*ifap);
 	*ifap = NULL;
 	if (err == ENXIO)
 		goto retry;
-	close(sock4);
-	close(sock6);
+
+	if (sock4 != -1)
+		(void) close(sock4);
+	if (sock6 != -1)
+		(void) close(sock6);
+	if (door_fd != -1)
+		(void) close(door_fd);
+	if (dld_fd != -1)
+		(void) close(dld_fd);
 	errno = err;
 	return (-1);
 }
diff --git a/usr/src/man/man3socket/getifaddrs.3socket b/usr/src/man/man3socket/getifaddrs.3socket
index 274ae8f720..9a9c2ad13d 100644
--- a/usr/src/man/man3socket/getifaddrs.3socket
+++ b/usr/src/man/man3socket/getifaddrs.3socket
@@ -12,11 +12,10 @@
 .\"
 .\" Copyright (c) 2013, Joyent, Inc. All rights reserved.
 .\"
-.TH GETIFADDRS 3SOCKET "Apr 18, 2013"
+.TH GETIFADDRS 3SOCKET "Feb 1, 2022"
 .SH NAME
 getifaddrs, freeifaddrs \- get interface addresses
 .SH SYNOPSIS
-.LP
 .nf
 \fBcc\fR [ \fIflag\fR ... ] \fIfile\fR ... \fB-lsocket\fR \fB-lnsl\fR \
 [ \fIlibrary\fR ... ]
@@ -36,7 +35,6 @@ getifaddrs, freeifaddrs \- get interface addresses
 .fi
 
 .SH DESCRIPTION
-.LP
 The \fBgetifaddrs\fR() function is used to obtain the list of network
 interfaces on the local machine.  A reference to a linked list of \fBifaddrs\fR
 structures, as defined in \fB<ifaddrs.h>\fR, is stored in the memory referenced
@@ -97,7 +95,9 @@ for more information.
 
 .sp
 .LP
-The \fIifa_data\fR member is presently unused.
+The \fIifa_data\fR member is specific to the address family. It is currently
+only available for AF_LINK entries where it contains a pointer to the
+\fBstruct if_data\fR (as defined in \fBif.h\fR(3HEAD)).
 
 .sp
 .LP
@@ -105,12 +105,10 @@ The memory used by \fBgetifaddrs\fR() to back the list is dynamically allocated.
 It should be freed using \fBfreeifaddrs\fR().
 
 .SH RETURN VALUES
-.LP
 If successful, \fBgetifaddrs\fR() returns the value \fB0\fR; otherwise it
 returns \fB\(mi1\fR and sets \fIerrno\fR to indicate the error.
 
 .SH ERRORS
-.LP
 The \fBgetifaddrs\fR() function may fail and set \fIerrno\fR for any of the
 errors specified for the library routines \fBioctl\fR(2),
 \fBsocket\fR(3SOCKET), and \fBmalloc\fR(3C).
@@ -128,7 +126,6 @@ MT-Level	MT-Safe
 .TE
 
 .SH SEE ALSO
-.LP
 .BR ioctl (2),
 .BR malloc (3C),
 .BR socket.h (3HEAD),
@@ -140,12 +137,8 @@ MT-Level	MT-Safe
 .BR ipadm (8)
 
 .SH NOTES
-.LP
-On an illumos system, this function lists only interfaces with the \fBIFF_UP\fR
-flag set; see \fBif_tcp\fR(4P) and \fBifconfig\fR(8) for more information.
-
-.SH BUGS
-.LP
-At present, this function only lists addresses from the \fBAF_INET\fR and
-\fBAF_INET6\fR families.  Other families, such as \fBAF_LINK\fR, are not
-included.
+This function lists interfaces of type AF_INET, AF_INET6, and AF_LINK.
+For AF_INET and AF_INET6 only interfaces with the \fBIFF_UP\fR
+flag set are listed; see \fBif_tcp\fR(7P) and \fBifconfig\fR(1M) for more
+information. For AF_LINK entries the interface index is only available when the
+link is plumbed.