[PATCH nfs-utils 00/17] exportfs/exportd/mountd: allow them to use netlink for up/downcalls

[PATCH nfs-utils 00/17] exportfs/exportd/mountd: allow them to use netlink for up/downcalls

[Jeff Layton]

This adds support for the new netlink-based upcalls and downcalls in
mountd, exportd and exportfs. With this, mountd is no longer reliant on
/proc for sunrpc cache upcalls.

There are also a few bugfixes and cleanups for existing code and
documentation in here too.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
Jeff Layton (17):
      nfsdctl: move *_netlink.h to support/include/
      support/export: remove unnecessary static variables in nfsd_fh expkey lookup
      exportfs: remove obsolete legacy mode documentation from manpage
      support/include: update netlink headers for all cache upcalls
      build: add libnl3 and netlink header support for exportd and mountd
      xlog: claim D_FAC3 as D_NETLINK
      exportd/mountd: add netlink support for svc_export cache
      exportd/mountd: add netlink support for the nfsd.fh cache
      exportd/mountd: add netlink support for the auth.unix.ip cache
      exportd/mountd: add netlink support for the auth.unix.gid cache
      mountd/exportd: only use /proc interfaces if netlink setup fails
      support/export: check for pending requests after opening netlink sockets
      exportd/mountd: use cache type from notifications to target scanning
      nfsd/sunrpc: add cache flush command and attribute enums
      exportfs: add netlink support for cache flush with /proc fallback
      exportfs: use netlink to probe kernel support, skip export_test
      mountd/exportd/exportfs: add --no-netlink option to disable netlink

 configure.ac                                       |   32 +-
 support/export/Makefile.am                         |    5 +-
 support/export/cache.c                             | 1892 +++++++++++++++++---
 support/export/cache_flush.c                       |  166 ++
 support/include/Makefile.am                        |    7 +-
 {utils/nfsdctl => support/include}/lockd_netlink.h |    0
 support/include/nfsd_netlink.h                     |  232 +++
 support/include/sunrpc_netlink.h                   |   85 +
 support/include/xlog.h                             |    2 +-
 support/nfs/cacheio.c                              |   49 +-
 utils/exportd/Makefile.am                          |    2 +-
 utils/exportd/exportd.c                            |   10 +-
 utils/exportd/exportd.man                          |   12 +-
 utils/exportfs/Makefile.am                         |    6 +-
 utils/exportfs/exportfs.c                          |   55 +-
 utils/exportfs/exportfs.man                        |   79 +-
 utils/mountd/Makefile.am                           |    2 +-
 utils/mountd/mountd.c                              |    9 +-
 utils/mountd/mountd.man                            |    9 +
 utils/nfsdctl/nfsd_netlink.h                       |   98 -
 20 files changed, 2343 insertions(+), 409 deletions(-)
---
base-commit: b9c9eea7c3afcfc77e241f71302a792dd15c8ea8
change-id: 20260316-exportd-netlink-a53bf66ae034

Best regards,
-- 
Jeff Layton <jlayton@kernel.org>

[PATCH nfs-utils 01/17] nfsdctl: move *_netlink.h to support/include/

[Jeff Layton]

Move lockd_netlink.h and nfsd_netlink.h from utils/nfsdctl/ to
support/include/ so they are available to other components.

The support/include/ directory is already in the default include path
(via AC_CONFIG_HEADERS), so no source-level #include changes are
needed.
---
 support/include/Makefile.am                        | 6 ++++--
 {utils/nfsdctl => support/include}/lockd_netlink.h | 0
 {utils/nfsdctl => support/include}/nfsd_netlink.h  | 0
 3 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/support/include/Makefile.am b/support/include/Makefile.am
index 631a84f808ae9f2a13d9539c70c09670c5ef0ff2..2b45e56b044c35c0e7a11aec5f583d600fcae6a0 100644
--- a/support/include/Makefile.am
+++ b/support/include/Makefile.am
@@ -4,13 +4,16 @@ SUBDIRS = nfs rpcsvc sys
 
 noinst_HEADERS = \
 	cld.h \
+	conffile.h \
 	exportfs.h \
 	ha-callout.h \
 	junction.h \
+	lockd_netlink.h \
 	misc.h \
 	nfs_mntent.h \
 	nfs_paths.h \
 	nfs_ucred.h \
+	nfsd_netlink.h \
 	nfsd_path.h \
 	nfslib.h \
 	nfsrpc.h \
@@ -26,7 +29,6 @@ noinst_HEADERS = \
 	xlog.h \
 	xmalloc.h \
 	xcommon.h \
-	xstat.h \
-	conffile.h
+	xstat.h
 
 MAINTAINERCLEANFILES = Makefile.in
diff --git a/utils/nfsdctl/lockd_netlink.h b/support/include/lockd_netlink.h
similarity index 100%
rename from utils/nfsdctl/lockd_netlink.h
rename to support/include/lockd_netlink.h
diff --git a/utils/nfsdctl/nfsd_netlink.h b/support/include/nfsd_netlink.h
similarity index 100%
rename from utils/nfsdctl/nfsd_netlink.h
rename to support/include/nfsd_netlink.h

-- 
2.53.0

[PATCH nfs-utils 02/17] support/export: remove unnecessary static variables in nfsd_fh expkey lookup

[Jeff Layton]

The prev and mnt variables in the CROSSMOUNT submount iteration loop
only need to persist across iterations of the inner for loop, not
across calls to the function. Convert them from static to local
variables scoped to the outer for loop.
---
 support/export/cache.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/support/export/cache.c b/support/export/cache.c
index 6859a55b7ae5c132255b756052c6ddd1734173f0..c052455889481f04631e4ef0e16cd48bc6c11964 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -810,7 +810,10 @@ static int nfsd_handle_fh(int f, char *bp, int blen)
 
 	/* Now determine export point for this fsid/domain */
 	for (i=0 ; i < MCL_MAXTYPES; i++) {
+		nfs_export *prev = NULL;
 		nfs_export *next_exp;
+		void *mnt = NULL;
+
 		for (exp = exportlist[i].p_head; exp; exp = next_exp) {
 			char *path;
 
@@ -820,9 +823,6 @@ static int nfsd_handle_fh(int f, char *bp, int blen)
 			}
 
 			if (exp->m_export.e_flags & NFSEXP_CROSSMOUNT) {
-				static nfs_export *prev = NULL;
-				static void *mnt = NULL;
-				
 				if (prev == exp) {
 					/* try a submount */
 					path = next_mnt(&mnt, exp->m_export.e_path);

-- 
2.53.0

[PATCH nfs-utils 03/17] exportfs: remove obsolete legacy mode documentation from manpage

[Jeff Layton]

The exportfs manpage described a "legacy mode" for 2.4 and earlier
kernels that used the nfsservctl() syscall to inject exports directly
into the kernel. This syscall was removed from the kernel in Linux 3.1
and the corresponding nfs-utils code was dropped long ago. Remove the
outdated text and simplify the description.
---
 utils/exportfs/exportfs.man | 29 ++---------------------------
 1 file changed, 2 insertions(+), 27 deletions(-)

diff --git a/utils/exportfs/exportfs.man b/utils/exportfs/exportfs.man
index 6d417a700340f050c0af5c8af848ebe8403f8379..af0e5571cef83d4f3de6915608b4871690a8853a 100644
--- a/utils/exportfs/exportfs.man
+++ b/utils/exportfs/exportfs.man
@@ -53,39 +53,14 @@ by using the
 command.
 .PP
 .B exportfs
-and its partner program
-.B rpc.mountd
-work in one of two modes: a legacy mode which applies to 2.4 and
-earlier versions of the Linux kernel, and a new mode which applies to
-2.6 and later versions, providing the
-.B nfsd
-virtual filesystem has been mounted at
-.I /proc/fs/nfsd
-or
-.IR /proc/fs/nfs .
-On 2.6 kernels, if this filesystem is not mounted, the legacy mode is used.
-.PP
-In the new mode,
-.B exportfs
-does not give any information to the kernel, but provides it only to
+does not give any information to the kernel directly, but provides it
+only to
 .B rpc.mountd
 through the
 .I /var/lib/nfs/etab
 file.
 .B rpc.mountd
 then manages kernel requests for information about exports, as needed.
-.PP
-In the legacy mode,
-exports which identify a specific host, rather than a subnet or netgroup,
-are entered directly into the kernel's export table,
-as well as being written to
-.IR /var/lib/nfs/etab .
-Further, exports listed in
-.I /var/lib/nfs/rmtab
-which match a non host-specific export request will cause an
-appropriate export entry for the host given in
-.I rmtab
-to be added to the kernel's export table.
 .SH OPTIONS
 .TP
 .B \-d kind " or " \-\-debug kind

-- 
2.53.0

[PATCH nfs-utils 04/17] support/include: update netlink headers for all cache upcalls

[Jeff Layton]

Sync the bundled nfsd_netlink.h with the current kernel UAPI header.
This brings in:
 - NFSD_CACHE_TYPE_EXPKEY in the cache-type enum
 - NFSD_A_SVC_EXPORT_FLAGS and NFSD_A_SVC_EXPORT_FSID attributes
 - NFSD_A_EXPKEY_* and NFSD_A_EXPKEY_REQS_* attribute enums
 - NFSD_CMD_EXPKEY_GET_REQS and NFSD_CMD_EXPKEY_SET_REQS commands

Add a new bundled sunrpc_netlink.h header for the sunrpc genl family,
which handles auth.unix.ip and auth.unix.gid cache upcalls via netlink.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/include/Makefile.am      |   1 +
 support/include/nfsd_netlink.h   | 126 +++++++++++++++++++++++++++++++++++++++
 support/include/sunrpc_netlink.h |  77 ++++++++++++++++++++++++
 3 files changed, 204 insertions(+)

diff --git a/support/include/Makefile.am b/support/include/Makefile.am
index 2b45e56b044c35c0e7a11aec5f583d600fcae6a0..9737c476a6bbe9e7a2938213c3fc5042da2d3579 100644
--- a/support/include/Makefile.am
+++ b/support/include/Makefile.am
@@ -15,6 +15,7 @@ noinst_HEADERS = \
 	nfs_ucred.h \
 	nfsd_netlink.h \
 	nfsd_path.h \
+	sunrpc_netlink.h \
 	nfslib.h \
 	nfsrpc.h \
 	nls.h \
diff --git a/support/include/nfsd_netlink.h b/support/include/nfsd_netlink.h
index e9efbc9e63d83ed25fcd790b7a877c0023638f15..e96cc1d23366bce13624084fd476dda65aef140a 100644
--- a/support/include/nfsd_netlink.h
+++ b/support/include/nfsd_netlink.h
@@ -10,6 +10,45 @@
 #define NFSD_FAMILY_NAME	"nfsd"
 #define NFSD_FAMILY_VERSION	1
 
+enum nfsd_cache_type {
+	NFSD_CACHE_TYPE_SVC_EXPORT = 1,
+	NFSD_CACHE_TYPE_EXPKEY = 2,
+};
+
+enum nfsd_export_flags {
+	NFSD_EXPORT_FLAGS_READONLY = 1,
+	NFSD_EXPORT_FLAGS_INSECURE_PORT = 2,
+	NFSD_EXPORT_FLAGS_ROOTSQUASH = 4,
+	NFSD_EXPORT_FLAGS_ALLSQUASH = 8,
+	NFSD_EXPORT_FLAGS_ASYNC = 16,
+	NFSD_EXPORT_FLAGS_GATHERED_WRITES = 32,
+	NFSD_EXPORT_FLAGS_NOREADDIRPLUS = 64,
+	NFSD_EXPORT_FLAGS_SECURITY_LABEL = 128,
+	NFSD_EXPORT_FLAGS_SIGN_FH = 256,
+	NFSD_EXPORT_FLAGS_NOHIDE = 512,
+	NFSD_EXPORT_FLAGS_NOSUBTREECHECK = 1024,
+	NFSD_EXPORT_FLAGS_NOAUTHNLM = 2048,
+	NFSD_EXPORT_FLAGS_MSNFS = 4096,
+	NFSD_EXPORT_FLAGS_FSID = 8192,
+	NFSD_EXPORT_FLAGS_CROSSMOUNT = 16384,
+	NFSD_EXPORT_FLAGS_NOACL = 32768,
+	NFSD_EXPORT_FLAGS_V4ROOT = 65536,
+	NFSD_EXPORT_FLAGS_PNFS = 131072,
+};
+
+enum nfsd_xprtsec_mode {
+	NFSD_XPRTSEC_MODE_NONE = 1,
+	NFSD_XPRTSEC_MODE_TLS = 2,
+	NFSD_XPRTSEC_MODE_MTLS = 4,
+};
+
+enum {
+	NFSD_A_CACHE_NOTIFY_CACHE_TYPE = 1,
+
+	__NFSD_A_CACHE_NOTIFY_MAX,
+	NFSD_A_CACHE_NOTIFY_MAX = (__NFSD_A_CACHE_NOTIFY_MAX - 1)
+};
+
 enum {
 	NFSD_A_RPC_STATUS_XID = 1,
 	NFSD_A_RPC_STATUS_FLAGS,
@@ -36,6 +75,7 @@ enum {
 	NFSD_A_SERVER_LEASETIME,
 	NFSD_A_SERVER_SCOPE,
 	NFSD_A_SERVER_MIN_THREADS,
+	NFSD_A_SERVER_FH_KEY,
 
 	__NFSD_A_SERVER_MAX,
 	NFSD_A_SERVER_MAX = (__NFSD_A_SERVER_MAX - 1)
@@ -80,6 +120,84 @@ enum {
 	NFSD_A_POOL_MODE_MAX = (__NFSD_A_POOL_MODE_MAX - 1)
 };
 
+enum {
+	NFSD_A_FSLOCATION_HOST = 1,
+	NFSD_A_FSLOCATION_PATH,
+
+	__NFSD_A_FSLOCATION_MAX,
+	NFSD_A_FSLOCATION_MAX = (__NFSD_A_FSLOCATION_MAX - 1)
+};
+
+enum {
+	NFSD_A_FSLOCATIONS_LOCATION = 1,
+
+	__NFSD_A_FSLOCATIONS_MAX,
+	NFSD_A_FSLOCATIONS_MAX = (__NFSD_A_FSLOCATIONS_MAX - 1)
+};
+
+enum {
+	NFSD_A_AUTH_FLAVOR_PSEUDOFLAVOR = 1,
+	NFSD_A_AUTH_FLAVOR_FLAGS,
+
+	__NFSD_A_AUTH_FLAVOR_MAX,
+	NFSD_A_AUTH_FLAVOR_MAX = (__NFSD_A_AUTH_FLAVOR_MAX - 1)
+};
+
+enum {
+	NFSD_A_SVC_EXPORT_REQ_SEQNO = 1,
+	NFSD_A_SVC_EXPORT_REQ_CLIENT,
+	NFSD_A_SVC_EXPORT_REQ_PATH,
+
+	__NFSD_A_SVC_EXPORT_REQ_MAX,
+	NFSD_A_SVC_EXPORT_REQ_MAX = (__NFSD_A_SVC_EXPORT_REQ_MAX - 1)
+};
+
+enum {
+	NFSD_A_SVC_EXPORT_SEQNO = 1,
+	NFSD_A_SVC_EXPORT_CLIENT,
+	NFSD_A_SVC_EXPORT_PATH,
+	NFSD_A_SVC_EXPORT_NEGATIVE,
+	NFSD_A_SVC_EXPORT_EXPIRY,
+	NFSD_A_SVC_EXPORT_ANON_UID,
+	NFSD_A_SVC_EXPORT_ANON_GID,
+	NFSD_A_SVC_EXPORT_FSLOCATIONS,
+	NFSD_A_SVC_EXPORT_UUID,
+	NFSD_A_SVC_EXPORT_SECINFO,
+	NFSD_A_SVC_EXPORT_XPRTSEC,
+	NFSD_A_SVC_EXPORT_FLAGS,
+	NFSD_A_SVC_EXPORT_FSID,
+
+	__NFSD_A_SVC_EXPORT_MAX,
+	NFSD_A_SVC_EXPORT_MAX = (__NFSD_A_SVC_EXPORT_MAX - 1)
+};
+
+enum {
+	NFSD_A_SVC_EXPORT_REQS_REQUESTS = 1,
+
+	__NFSD_A_SVC_EXPORT_REQS_MAX,
+	NFSD_A_SVC_EXPORT_REQS_MAX = (__NFSD_A_SVC_EXPORT_REQS_MAX - 1)
+};
+
+enum {
+	NFSD_A_EXPKEY_SEQNO = 1,
+	NFSD_A_EXPKEY_CLIENT,
+	NFSD_A_EXPKEY_FSIDTYPE,
+	NFSD_A_EXPKEY_FSID,
+	NFSD_A_EXPKEY_NEGATIVE,
+	NFSD_A_EXPKEY_EXPIRY,
+	NFSD_A_EXPKEY_PATH,
+
+	__NFSD_A_EXPKEY_MAX,
+	NFSD_A_EXPKEY_MAX = (__NFSD_A_EXPKEY_MAX - 1)
+};
+
+enum {
+	NFSD_A_EXPKEY_REQS_REQUESTS = 1,
+
+	__NFSD_A_EXPKEY_REQS_MAX,
+	NFSD_A_EXPKEY_REQS_MAX = (__NFSD_A_EXPKEY_REQS_MAX - 1)
+};
+
 enum {
 	NFSD_CMD_RPC_STATUS_GET = 1,
 	NFSD_CMD_THREADS_SET,
@@ -90,9 +208,17 @@ enum {
 	NFSD_CMD_LISTENER_GET,
 	NFSD_CMD_POOL_MODE_SET,
 	NFSD_CMD_POOL_MODE_GET,
+	NFSD_CMD_CACHE_NOTIFY,
+	NFSD_CMD_SVC_EXPORT_GET_REQS,
+	NFSD_CMD_SVC_EXPORT_SET_REQS,
+	NFSD_CMD_EXPKEY_GET_REQS,
+	NFSD_CMD_EXPKEY_SET_REQS,
 
 	__NFSD_CMD_MAX,
 	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
 };
 
+#define NFSD_MCGRP_NONE		"none"
+#define NFSD_MCGRP_EXPORTD	"exportd"
+
 #endif /* _UAPI_LINUX_NFSD_NETLINK_H */
diff --git a/support/include/sunrpc_netlink.h b/support/include/sunrpc_netlink.h
new file mode 100644
index 0000000000000000000000000000000000000000..52c9aa3f19b6a218ec7d0518b58062beff11c774
--- /dev/null
+++ b/support/include/sunrpc_netlink.h
@@ -0,0 +1,77 @@
+/* SPDX-License-Identifier: ((GPL-2.0 WITH Linux-syscall-note) OR BSD-3-Clause) */
+/* Do not edit directly, auto-generated from: */
+/*	Documentation/netlink/specs/sunrpc_cache.yaml */
+/* YNL-GEN uapi header */
+/* To regenerate run: tools/net/ynl/ynl-regen.sh */
+
+#ifndef _UAPI_LINUX_SUNRPC_NETLINK_H
+#define _UAPI_LINUX_SUNRPC_NETLINK_H
+
+#define SUNRPC_FAMILY_NAME	"sunrpc"
+#define SUNRPC_FAMILY_VERSION	1
+
+enum sunrpc_cache_type {
+
+	SUNRPC_CACHE_TYPE_IP_MAP = 1,
+	SUNRPC_CACHE_TYPE_UNIX_GID = 2,
+};
+
+enum {
+	SUNRPC_A_CACHE_NOTIFY_CACHE_TYPE = 1,
+
+	__SUNRPC_A_CACHE_NOTIFY_MAX,
+	SUNRPC_A_CACHE_NOTIFY_MAX = (__SUNRPC_A_CACHE_NOTIFY_MAX - 1)
+};
+
+enum {
+	SUNRPC_A_IP_MAP_SEQNO = 1,
+	SUNRPC_A_IP_MAP_CLASS,
+	SUNRPC_A_IP_MAP_ADDR,
+	SUNRPC_A_IP_MAP_DOMAIN,
+	SUNRPC_A_IP_MAP_NEGATIVE,
+	SUNRPC_A_IP_MAP_EXPIRY,
+
+	__SUNRPC_A_IP_MAP_MAX,
+	SUNRPC_A_IP_MAP_MAX = (__SUNRPC_A_IP_MAP_MAX - 1)
+};
+
+enum {
+	SUNRPC_A_IP_MAP_REQS_REQUESTS = 1,
+
+	__SUNRPC_A_IP_MAP_REQS_MAX,
+	SUNRPC_A_IP_MAP_REQS_MAX = (__SUNRPC_A_IP_MAP_REQS_MAX - 1)
+};
+
+enum {
+	SUNRPC_A_UNIX_GID_SEQNO = 1,
+	SUNRPC_A_UNIX_GID_UID,
+	SUNRPC_A_UNIX_GID_GIDS,
+	SUNRPC_A_UNIX_GID_NEGATIVE,
+	SUNRPC_A_UNIX_GID_EXPIRY,
+
+	__SUNRPC_A_UNIX_GID_MAX,
+	SUNRPC_A_UNIX_GID_MAX = (__SUNRPC_A_UNIX_GID_MAX - 1)
+};
+
+enum {
+	SUNRPC_A_UNIX_GID_REQS_REQUESTS = 1,
+
+	__SUNRPC_A_UNIX_GID_REQS_MAX,
+	SUNRPC_A_UNIX_GID_REQS_MAX = (__SUNRPC_A_UNIX_GID_REQS_MAX - 1)
+};
+
+enum {
+	SUNRPC_CMD_CACHE_NOTIFY = 1,
+	SUNRPC_CMD_IP_MAP_GET_REQS,
+	SUNRPC_CMD_IP_MAP_SET_REQS,
+	SUNRPC_CMD_UNIX_GID_GET_REQS,
+	SUNRPC_CMD_UNIX_GID_SET_REQS,
+
+	__SUNRPC_CMD_MAX,
+	SUNRPC_CMD_MAX = (__SUNRPC_CMD_MAX - 1)
+};
+
+#define SUNRPC_MCGRP_NONE	"none"
+#define SUNRPC_MCGRP_EXPORTD	"exportd"
+
+#endif /* _UAPI_LINUX_SUNRPC_NETLINK_H */

-- 
2.53.0

[PATCH nfs-utils 05/17] build: add libnl3 and netlink header support for exportd and mountd

[Jeff Layton]

Add libnl3 dependency and netlink header detection for exportd and
mountd builds. Update the nfsd_netlink.h sentinel to check for
NFSD_A_EXPKEY_SEQNO to ensure the system header includes expkey
support. Add AC_COMPILE_IFELSE check for sunrpc_netlink.h using
SUNRPC_CMD_CACHE_NOTIFY as sentinel.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 configure.ac               | 32 +++++++++++++++++++-------------
 support/export/Makefile.am |  3 ++-
 utils/exportd/Makefile.am  |  2 +-
 utils/mountd/Makefile.am   |  2 +-
 4 files changed, 23 insertions(+), 16 deletions(-)

diff --git a/configure.ac b/configure.ac
index b134c98c7c6e1ba56c2b2a0b4e8d614d58093f35..f8c04fbfeaaafa63c07f7552dd7c71ef234eab10 100644
--- a/configure.ac
+++ b/configure.ac
@@ -252,26 +252,32 @@ AC_ARG_ENABLE(nfsdcltrack,
 	enable_nfsdcltrack=$enableval,
 	enable_nfsdcltrack="no")
 
+PKG_CHECK_MODULES(LIBNL3, libnl-3.0 >= 3.1)
+PKG_CHECK_MODULES(LIBNLGENL3, libnl-genl-3.0 >= 3.1)
+
+AC_CHECK_HEADERS(linux/nfsd_netlink.h)
+
+# ensure we have the expkey attributes
+AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[#include <linux/nfsd_netlink.h>]],
+			                   [[int foo = NFSD_A_EXPKEY_SEQNO;]])],
+			   [AC_DEFINE([USE_SYSTEM_NFSD_NETLINK_H], 1,
+				      ["Use system's linux/nfsd_netlink.h"])])
+AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[#include <linux/lockd_netlink.h>]],
+			                   [[int foo = LOCKD_CMD_SERVER_GET;]])],
+			   [AC_DEFINE([USE_SYSTEM_LOCKD_NETLINK_H], 1,
+				      ["Use system's linux/lockd_netlink.h"])])
+AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[#include <linux/sunrpc_netlink.h>]],
+			                   [[int foo = SUNRPC_CMD_CACHE_NOTIFY;]])],
+			   [AC_DEFINE([USE_SYSTEM_SUNRPC_NETLINK_H], 1,
+				      ["Use system's linux/sunrpc_netlink.h"])])
+
 AC_ARG_ENABLE(nfsdctl,
 	[AS_HELP_STRING([--disable-nfsdctl],[disable nfsdctl program for controlling nfsd@<:@default=no@:>@])],
 	enable_nfsdctl=$enableval,
 	enable_nfsdctl="yes")
 	AM_CONDITIONAL(CONFIG_NFSDCTL, [test "$enable_nfsdctl" = "yes" ])
 	if test "$enable_nfsdctl" = yes; then
-		PKG_CHECK_MODULES(LIBNL3, libnl-3.0 >= 3.1)
-		PKG_CHECK_MODULES(LIBNLGENL3, libnl-genl-3.0 >= 3.1)
 		PKG_CHECK_MODULES(LIBREADLINE, readline)
-		AC_CHECK_HEADERS(linux/nfsd_netlink.h)
-
-		# ensure we have the MIN_THREADS attribute
-		AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[#include <linux/nfsd_netlink.h>]],
-				                   [[int foo = NFSD_A_SERVER_MIN_THREADS;]])],
-				   [AC_DEFINE([USE_SYSTEM_NFSD_NETLINK_H], 1,
-					      ["Use system's linux/nfsd_netlink.h"])])
-		AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[#include <linux/lockd_netlink.h>]],
-				                   [[int foo = LOCKD_CMD_SERVER_GET;]])],
-				   [AC_DEFINE([USE_SYSTEM_LOCKD_NETLINK_H], 1,
-					      ["Use system's linux/lockd_netlink.h"])])
 	fi
 
 AC_ARG_ENABLE(nfsv4server,
diff --git a/support/export/Makefile.am b/support/export/Makefile.am
index 7338e1c7e719b2684745e0d71810c9f52d08a3c9..ae7ace44112b889f1c461c5473fb1bd42a42f182 100644
--- a/support/export/Makefile.am
+++ b/support/export/Makefile.am
@@ -14,7 +14,8 @@ libexport_a_SOURCES = client.c export.c hostname.c \
 		      xtab.c mount_clnt.c mount_xdr.c \
 		      cache.c auth.c v4root.c fsloc.c \
 		      v4clients.c
-libexport_a_CPPFLAGS = $(AM_CPPFLAGS) $(CPPFLAGS) -I$(top_srcdir)/support/reexport
+libexport_a_CPPFLAGS = $(AM_CPPFLAGS) $(CPPFLAGS) -I$(top_srcdir)/support/reexport \
+		       $(LIBNL3_CFLAGS) $(LIBNLGENL3_CFLAGS)
 
 BUILT_SOURCES 	= $(GENFILES)
 
diff --git a/utils/exportd/Makefile.am b/utils/exportd/Makefile.am
index 26078c9b11b54d2af7bce0273f1cac577f0d8097..e25166b1aa1871492d11a83decd63e76a1528f9b 100644
--- a/utils/exportd/Makefile.am
+++ b/utils/exportd/Makefile.am
@@ -18,7 +18,7 @@ exportd_LDADD = ../../support/export/libexport.a \
 			../../support/misc/libmisc.a \
 			../../support/reexport/libreexport.a \
 			$(OPTLIBS) $(LIBBLKID) $(LIBPTHREAD) \
-			-luuid
+			-luuid $(LIBNL3_LIBS) $(LIBNLGENL3_LIBS)
 
 exportd_CPPFLAGS = $(AM_CPPFLAGS) $(CPPFLAGS) \
 		-I$(top_srcdir)/support/export
diff --git a/utils/mountd/Makefile.am b/utils/mountd/Makefile.am
index 197ef29b4fe4a855da3b26c701fa0b41916ea59c..808baf318fe54ccee13cf60b4cd71ade0444508b 100644
--- a/utils/mountd/Makefile.am
+++ b/utils/mountd/Makefile.am
@@ -20,7 +20,7 @@ mountd_LDADD = ../../support/export/libexport.a \
 	       ../../support/reexport/libreexport.a \
 	       $(OPTLIBS) \
 	       $(LIBBSD) $(LIBWRAP) $(LIBNSL) $(LIBBLKID) -luuid $(LIBTIRPC) \
-	       $(LIBPTHREAD)
+	       $(LIBPTHREAD) $(LIBNL3_LIBS) $(LIBNLGENL3_LIBS)
 
 mountd_CPPFLAGS = $(AM_CPPFLAGS) $(CPPFLAGS) \
 		  -I$(top_builddir)/support/include \

-- 
2.53.0

[PATCH nfs-utils 06/17] xlog: claim D_FAC3 as D_NETLINK

[Jeff Layton]

Nothing uses the D_FAC* debugging bits, so claim D_FAC3 as D_NETLINK
for logging about netlink activity.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/include/xlog.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/support/include/xlog.h b/support/include/xlog.h
index 69cdf61eaf370bf70a0da57c422f65f2f49ee06b..755105e0dc9a0298a17be6a3f63f0b471c53f8a4 100644
--- a/support/include/xlog.h
+++ b/support/include/xlog.h
@@ -25,7 +25,7 @@
 #define D_GENERAL	0x0001		/* general debug info */
 #define D_CALL		0x0002
 #define D_AUTH		0x0004
-#define D_FAC3		0x0008
+#define D_NETLINK	0x0008
 #define D_FAC4		0x0010
 #define D_FAC5		0x0020
 #define D_PARSE		0x0040

-- 
2.53.0

[PATCH nfs-utils 07/17] exportd/mountd: add netlink support for svc_export cache

[Jeff Layton]

Add netlink-based svc_export cache handling alongside the existing
procfs channel mechanism. When the kernel sends an
NFSD_CMD_SVC_EXPORT_NOTIFY multicast notification on the "exportd"
genetlink group, userspace responds by issuing
NFSD_CMD_SVC_EXPORT_GET_REQS to retrieve all pending export requests,
resolving each one via lookup_export(), and sending back the results
with NFSD_CMD_SVC_EXPORT_SET_REQS.

Two netlink sockets are used: a non-blocking notification socket that
joins the multicast group and is integrated into the existing select()
loop, and a blocking command socket for the GET_REQS dump and SET_REQS
response. The notification socket is added to the fd_set in
cache_set_fds() and checked in cache_process_req().

If the kernel does not support the nfsd genetlink family or the
"exportd" multicast group (older kernels), the netlink setup silently
fails and the daemon falls back to procfs-only operation. Both
mechanisms can coexist safely since the kernel's CACHE_PENDING check
prevents duplicate processing.

Include the NFSD_A_SVC_EXPORT_FLAGS and NFSD_A_SVC_EXPORT_FSID
attributes in nl_add_export() responses to the kernel.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/export/cache.c | 560 ++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 558 insertions(+), 2 deletions(-)

diff --git a/support/export/cache.c b/support/export/cache.c
index c052455889481f04631e4ef0e16cd48bc6c11964..090ea0a6cc6da7e9c2ce255601103d2442526067 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -36,6 +36,18 @@
 #include "reexport.h"
 #include "fsloc.h"
 
+#include <netlink/genl/genl.h>
+#include <netlink/genl/ctrl.h>
+#include <netlink/msg.h>
+#include <netlink/attr.h>
+#include <linux/netlink.h>
+
+#ifdef USE_SYSTEM_NFSD_NETLINK_H
+#include <linux/nfsd_netlink.h>
+#else
+#include "nfsd_netlink.h"
+#endif
+
 #ifdef USE_BLKID
 #include "blkid/blkid.h"
 #endif
@@ -1052,6 +1064,541 @@ static void write_xprtsec(char **bp, int *blen, struct exportent *ep)
 		qword_addint(bp, blen, p->info->number);
 }
 
+/*
+ * Netlink-based svc_export cache support.
+ *
+ * The kernel can notify userspace of pending export cache requests via
+ * the "exportd" genetlink multicast group. We listen for
+ * NFSD_CMD_SVC_EXPORT_NOTIFY, then issue NFSD_CMD_SVC_EXPORT_GET_REQS
+ * to retrieve pending requests, resolve them, and respond with
+ * NFSD_CMD_SVC_EXPORT_SET_REQS.
+ */
+static nfs_export *lookup_export(char *dom, char *path, struct addrinfo *ai);
+static struct nl_msg *cache_nl_new_msg(int family, int cmd, int flags);
+
+static struct nl_sock *nfsd_nl_notify_sock;	/* multicast notifications */
+static struct nl_sock *nfsd_nl_cmd_sock;	/* GET_REQS / SET_REQS commands */
+static int nfsd_nl_family;
+
+#define NL_BUFFER_SIZE	65536
+
+struct export_req {
+	char	*client;
+	char	*path;
+};
+
+static struct nl_sock *nl_sock_setup(void)
+{
+	struct nl_sock *sock;
+	int val = 1;
+
+	sock = nl_socket_alloc();
+	if (!sock)
+		return NULL;
+
+	if (genl_connect(sock)) {
+		nl_socket_free(sock);
+		return NULL;
+	}
+
+	nl_socket_set_buffer_size(sock, NL_BUFFER_SIZE, NL_BUFFER_SIZE);
+	setsockopt(nl_socket_get_fd(sock), SOL_NETLINK, NETLINK_EXT_ACK,
+		   &val, sizeof(val));
+
+	return sock;
+}
+
+static int cache_nfsd_nl_open(void)
+{
+	int grp;
+
+	nfsd_nl_family = 0;
+
+	nfsd_nl_cmd_sock = nl_sock_setup();
+	if (!nfsd_nl_cmd_sock) {
+		xlog(D_NETLINK, "cache_nfsd_nl_open: failed to allocate command socket");
+		return -ENOMEM;
+	}
+
+	nfsd_nl_family = genl_ctrl_resolve(nfsd_nl_cmd_sock, NFSD_FAMILY_NAME);
+	if (nfsd_nl_family < 0) {
+		xlog(D_NETLINK, "cache_nfsd_nl_open: nfsd netlink family not found");
+		goto out_free_cmd;
+	}
+
+	grp = genl_ctrl_resolve_grp(nfsd_nl_cmd_sock, NFSD_FAMILY_NAME,
+				    NFSD_MCGRP_EXPORTD);
+	if (grp < 0) {
+		xlog(D_NETLINK, "cache_nfsd_nl_open: exportd multicast group not found");
+		goto out_free_cmd;
+	}
+
+	nfsd_nl_notify_sock = nl_sock_setup();
+	if (!nfsd_nl_notify_sock) {
+		xlog(D_NETLINK, "cache_nfsd_nl_open: failed to allocate notify socket");
+		goto out_free_cmd;
+	}
+
+	nl_socket_disable_seq_check(nfsd_nl_notify_sock);
+
+	if (nl_socket_add_membership(nfsd_nl_notify_sock, grp)) {
+		xlog(L_WARNING, "cache_nfsd_nl_open: failed to join exportd multicast group");
+		goto out_free_notify;
+	}
+
+	nl_socket_set_nonblocking(nfsd_nl_notify_sock);
+	xlog(D_NETLINK, "cache_nfsd_nl_open: listening for export notifications");
+	return 0;
+
+out_free_notify:
+	nl_socket_free(nfsd_nl_notify_sock);
+	nfsd_nl_notify_sock = NULL;
+out_free_cmd:
+	nl_socket_free(nfsd_nl_cmd_sock);
+	nfsd_nl_cmd_sock = NULL;
+	nfsd_nl_family = 0;
+	return -ENOENT;
+}
+
+static int nfsd_nl_notify_handler(struct nl_msg *UNUSED(msg), void *UNUSED(arg))
+{
+	return NL_OK;
+}
+
+static void cache_nfsd_nl_drain(void)
+{
+	struct nl_cb *cb;
+
+	cb = nl_cb_alloc(NL_CB_DEFAULT);
+	if (!cb)
+		return;
+
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, nfsd_nl_notify_handler, NULL);
+	nl_recvmsgs(nfsd_nl_notify_sock, cb);
+	nl_cb_put(cb);
+}
+
+struct get_export_reqs_data {
+	struct export_req	*reqs;
+	int			nreqs;
+	int			maxreqs;
+	int			err;
+};
+
+static int get_export_reqs_cb(struct nl_msg *msg, void *arg)
+{
+	struct get_export_reqs_data *data = arg;
+	struct genlmsghdr *gnlh = nlmsg_data(nlmsg_hdr(msg));
+	struct nlattr *attr;
+	int rem;
+
+	nla_for_each_attr(attr, genlmsg_attrdata(gnlh, 0),
+			  genlmsg_attrlen(gnlh, 0), rem) {
+		struct nlattr *tb[NFSD_A_SVC_EXPORT_PATH + 1];
+		struct export_req *req;
+
+		if (nla_type(attr) != NFSD_A_SVC_EXPORT_REQS_REQUESTS)
+			continue;
+
+		if (nla_parse_nested(tb, NFSD_A_SVC_EXPORT_PATH, attr,
+				     NULL))
+			continue;
+
+		if (!tb[NFSD_A_SVC_EXPORT_CLIENT] ||
+		    !tb[NFSD_A_SVC_EXPORT_PATH])
+			continue;
+
+		if (data->nreqs >= data->maxreqs) {
+			int newmax = data->maxreqs ? data->maxreqs * 2 : 16;
+			struct export_req *tmp;
+
+			tmp = realloc(data->reqs,
+				      newmax * sizeof(*tmp));
+			if (!tmp) {
+				data->err = -ENOMEM;
+				return NL_STOP;
+			}
+			data->reqs = tmp;
+			data->maxreqs = newmax;
+		}
+
+		req = &data->reqs[data->nreqs++];
+		req->client = strdup(nla_get_string(tb[NFSD_A_SVC_EXPORT_CLIENT]));
+		req->path = strdup(nla_get_string(tb[NFSD_A_SVC_EXPORT_PATH]));
+
+		if (!req->client || !req->path) {
+			data->err = -ENOMEM;
+			return NL_STOP;
+		}
+	}
+
+	return NL_OK;
+}
+
+static int nl_finish_cb(struct nl_msg *UNUSED(msg), void *arg)
+{
+	int *done = arg;
+
+	*done = 1;
+	return NL_STOP;
+}
+
+static int nl_error_cb(struct sockaddr_nl *UNUSED(nla),
+		       struct nlmsgerr *UNUSED(nlerr), void *arg)
+{
+	int *done = arg;
+
+	*done = 1;
+	return NL_STOP;
+}
+
+static int cache_nl_get_export_reqs(struct export_req **reqs_out, int *nreqs_out)
+{
+	struct get_export_reqs_data data = { };
+	struct nl_msg *msg;
+	struct nl_cb *cb;
+	int done = 0;
+	int ret;
+
+	msg = cache_nl_new_msg(nfsd_nl_family,
+			       NFSD_CMD_SVC_EXPORT_GET_REQS, NLM_F_DUMP);
+	if (!msg)
+		return -ENOMEM;
+
+	cb = nl_cb_alloc(NL_CB_DEFAULT);
+	if (!cb) {
+		nlmsg_free(msg);
+		return -ENOMEM;
+	}
+
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, get_export_reqs_cb, &data);
+	nl_cb_set(cb, NL_CB_FINISH, NL_CB_CUSTOM, nl_finish_cb, &done);
+	nl_cb_err(cb, NL_CB_CUSTOM, nl_error_cb, &done);
+
+	ret = nl_send_auto(nfsd_nl_cmd_sock, msg);
+	nlmsg_free(msg);
+	if (ret < 0) {
+		nl_cb_put(cb);
+		return ret;
+	}
+
+	while (!done) {
+		ret = nl_recvmsgs(nfsd_nl_cmd_sock, cb);
+		if (ret < 0)
+			break;
+	}
+
+	nl_cb_put(cb);
+
+	if (data.err) {
+		int i;
+		for (i = 0; i < data.nreqs; i++) {
+			free(data.reqs[i].client);
+			free(data.reqs[i].path);
+		}
+		free(data.reqs);
+		return data.err;
+	}
+
+	*reqs_out = data.reqs;
+	*nreqs_out = data.nreqs;
+	return 0;
+}
+
+static int nfsd_nl_add_fsloc(struct nl_msg *msg, struct exportent *ep)
+{
+	struct servers *servers;
+	struct nlattr *fslocs;
+	int i;
+
+	if (ep->e_fslocmethod == FSLOC_NONE)
+		return 0;
+
+	servers = replicas_lookup(ep->e_fslocmethod, ep->e_fslocdata);
+	if (!servers)
+		return 0;
+
+	if (servers->h_num < 0) {
+		release_replicas(servers);
+		return 0;
+	}
+
+	fslocs = nla_nest_start(msg, NFSD_A_SVC_EXPORT_FSLOCATIONS);
+	if (!fslocs) {
+		release_replicas(servers);
+		return -1;
+	}
+
+	for (i = 0; i < servers->h_num; i++) {
+		struct nlattr *loc;
+
+		loc = nla_nest_start(msg, NFSD_A_FSLOCATIONS_LOCATION);
+		if (!loc) {
+			release_replicas(servers);
+			return -1;
+		}
+
+		if (nla_put_string(msg, NFSD_A_FSLOCATION_HOST,
+				   servers->h_mp[i]->h_host) < 0 ||
+		    nla_put_string(msg, NFSD_A_FSLOCATION_PATH,
+				   servers->h_mp[i]->h_path) < 0) {
+			nla_nest_cancel(msg, fslocs);
+			release_replicas(servers);
+			return -1;
+		}
+		nla_nest_end(msg, loc);
+	}
+
+	nla_nest_end(msg, fslocs);
+	release_replicas(servers);
+	return 0;
+}
+
+static int nfsd_nl_add_secinfo(struct nl_msg *msg, struct exportent *ep)
+{
+	struct sec_entry *p;
+
+	for (p = ep->e_secinfo; p->flav; p++)
+		; /* count */
+	if (p == ep->e_secinfo)
+		return 0;
+
+	fix_pseudoflavor_flags(ep);
+	for (p = ep->e_secinfo; p->flav; p++) {
+		struct nlattr *sec;
+
+		sec = nla_nest_start(msg, NFSD_A_SVC_EXPORT_SECINFO);
+		if (!sec)
+			return -1;
+
+		if (nla_put_u32(msg, NFSD_A_AUTH_FLAVOR_PSEUDOFLAVOR,
+				p->flav->fnum) < 0 ||
+		    nla_put_u32(msg, NFSD_A_AUTH_FLAVOR_FLAGS,
+				p->flags) < 0)
+			return -1;
+		nla_nest_end(msg, sec);
+	}
+
+	return 0;
+}
+
+static int nfsd_nl_add_xprtsec(struct nl_msg *msg, struct exportent *ep)
+{
+	struct xprtsec_entry *p;
+
+	for (p = ep->e_xprtsec; p->info; p++)
+		;
+	if (p == ep->e_xprtsec)
+		return 0;
+
+	for (p = ep->e_xprtsec; p->info; p++)
+		if (nla_put_u32(msg, NFSD_A_SVC_EXPORT_XPRTSEC,
+				p->info->number) < 0)
+			return -1;
+
+	return 0;
+}
+
+static int nfsd_nl_add_export(struct nl_msg *msg, char *domain, char *path,
+			 struct exportent *exp, int ttl)
+{
+	struct nlattr *nest;
+	time_t now = time(0);
+	char u[16];
+
+	if (ttl <= 1)
+		ttl = default_ttl;
+
+	nest = nla_nest_start(msg, NFSD_A_SVC_EXPORT_REQS_REQUESTS);
+	if (!nest)
+		return -1;
+
+	if (nla_put_string(msg, NFSD_A_SVC_EXPORT_CLIENT, domain) < 0 ||
+	    nla_put_string(msg, NFSD_A_SVC_EXPORT_PATH, path) < 0 ||
+	    nla_put_u64(msg, NFSD_A_SVC_EXPORT_EXPIRY, now + ttl) < 0)
+		goto nla_failure;
+
+	if (!exp) {
+		if (nla_put_flag(msg, NFSD_A_SVC_EXPORT_NEGATIVE) < 0)
+			goto nla_failure;
+	} else {
+		if (nla_put_u32(msg, NFSD_A_SVC_EXPORT_ANON_UID,
+				exp->e_anonuid) < 0 ||
+		    nla_put_u32(msg, NFSD_A_SVC_EXPORT_ANON_GID,
+				exp->e_anongid) < 0 ||
+		    nla_put_u32(msg, NFSD_A_SVC_EXPORT_FLAGS,
+				exp->e_flags) < 0 ||
+		    nla_put_s32(msg, NFSD_A_SVC_EXPORT_FSID,
+				exp->e_fsid) < 0)
+			goto nla_failure;
+
+		if (nfsd_nl_add_fsloc(msg, exp))
+			goto nla_failure;
+
+		if (exp->e_uuid) {
+			get_uuid(exp->e_uuid, 16, u);
+			if (nla_put(msg, NFSD_A_SVC_EXPORT_UUID,
+				    16, u) < 0)
+				goto nla_failure;
+		} else if (uuid_by_path(path, 0, 16, u)) {
+			if (nla_put(msg, NFSD_A_SVC_EXPORT_UUID,
+				    16, u) < 0)
+				goto nla_failure;
+		}
+
+		if (nfsd_nl_add_secinfo(msg, exp))
+			goto nla_failure;
+
+		if (nfsd_nl_add_xprtsec(msg, exp))
+			goto nla_failure;
+	}
+
+	nla_nest_end(msg, nest);
+	return 0;
+
+nla_failure:
+	nla_nest_cancel(msg, nest);
+	return -1;
+}
+
+static struct nl_msg *cache_nl_new_msg(int family, int cmd, int flags)
+{
+	struct nl_msg *msg;
+
+	msg = nlmsg_alloc();
+	if (!msg)
+		return NULL;
+
+	if (!genlmsg_put(msg, NL_AUTO_PID, NL_AUTO_SEQ, family,
+			 0, flags, cmd, 0)) {
+		nlmsg_free(msg);
+		return NULL;
+	}
+	return msg;
+}
+
+static int cache_nl_set_reqs(struct nl_msg *msg)
+{
+	struct nl_cb *cb;
+	int done = 0;
+	int ret;
+
+	cb = nl_cb_alloc(NL_CB_DEFAULT);
+	if (!cb)
+		return -ENOMEM;
+
+	nl_cb_set(cb, NL_CB_ACK, NL_CB_CUSTOM, nl_finish_cb, &done);
+	nl_cb_err(cb, NL_CB_CUSTOM, nl_error_cb, &done);
+
+	ret = nl_send_auto(nfsd_nl_cmd_sock, msg);
+	if (ret < 0) {
+		nl_cb_put(cb);
+		return ret;
+	}
+
+	while (!done) {
+		ret = nl_recvmsgs(nfsd_nl_cmd_sock, cb);
+		if (ret < 0)
+			break;
+	}
+
+	nl_cb_put(cb);
+	if (ret < 0)
+		xlog(L_WARNING, "%s: SET_REQS failed: %d", __func__, ret);
+	return ret;
+}
+
+static void cache_nl_process_export(void)
+{
+	struct export_req *reqs = NULL;
+	int nreqs = 0;
+	struct nl_msg *msg;
+	int i;
+
+	/* Drain pending notifications */
+	cache_nfsd_nl_drain();
+
+	/* Fetch all pending requests from the kernel */
+	if (cache_nl_get_export_reqs(&reqs, &nreqs)) {
+		xlog(L_WARNING, "cache_nl_process_export: failed to get export requests");
+		return;
+	}
+
+	if (!nreqs)
+		return;
+
+	xlog(D_CALL, "cache_nl_process_export: %d pending export requests", nreqs);
+
+	/* Build the SET_REQS response */
+	msg = cache_nl_new_msg(nfsd_nl_family,
+			       NFSD_CMD_SVC_EXPORT_SET_REQS, 0);
+	if (!msg)
+		goto out_free;
+
+	auth_reload();
+
+	for (i = 0; i < nreqs; i++) {
+		char *dom = reqs[i].client;
+		char *path = reqs[i].path;
+		struct addrinfo *ai = NULL;
+		nfs_export *found = NULL;
+		struct exportent *epp = NULL;
+		int ttl = 0;
+
+		if (is_ipaddr_client(dom)) {
+			ai = lookup_client_addr(dom);
+			if (!ai)
+				xlog(D_AUTH, "cache_nl_process_export: "
+				     "failed to resolve client %s", dom);
+		}
+
+		if (ai || !is_ipaddr_client(dom))
+			found = lookup_export(dom, path, ai);
+
+		if (found) {
+			char *mp = found->m_export.e_mountpoint;
+
+			if (mp && !*mp)
+				mp = found->m_export.e_path;
+			if (mp && !is_mountpoint(mp)) {
+				xlog(L_WARNING,
+				     "Cannot export path '%s': not a mountpoint",
+				     path);
+				ttl = 60;
+			} else {
+				epp = &found->m_export;
+			}
+		}
+
+		if (nfsd_nl_add_export(msg, dom, path, epp, ttl) < 0) {
+			cache_nl_set_reqs(msg);
+			nlmsg_free(msg);
+			msg = cache_nl_new_msg(nfsd_nl_family,
+					       NFSD_CMD_SVC_EXPORT_SET_REQS, 0);
+			if (!msg) {
+				nfs_freeaddrinfo(ai);
+				goto out_free;
+			}
+			if (nfsd_nl_add_export(msg, dom, path,
+					       epp, ttl) < 0)
+				xlog(L_WARNING, "%s: skipping oversized "
+				     "entry for %s", __func__, path);
+		}
+
+		nfs_freeaddrinfo(ai);
+	}
+
+	cache_nl_set_reqs(msg);
+	nlmsg_free(msg);
+
+out_free:
+	for (i = 0; i < nreqs; i++) {
+		free(reqs[i].client);
+		free(reqs[i].path);
+	}
+	free(reqs);
+}
+
 static int can_reexport_via_fsidnum(struct exportent *exp, struct statfs *st)
 {
 	if (st->f_type != 0x6969 /* NFS_SUPER_MAGIC */)
@@ -1612,7 +2159,7 @@ extern int manage_gids;
  * cache_open - prepare communications channels with kernel RPC caches
  *
  */
-void cache_open(void) 
+void cache_open(void)
 {
 	int i;
 
@@ -1623,6 +2170,7 @@ void cache_open(void)
 		sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
 		cachelist[i].f = open(path, O_RDWR);
 	}
+	cache_nfsd_nl_open();
 }
 
 /**
@@ -1636,13 +2184,15 @@ void cache_set_fds(fd_set *fdset)
 		if (cachelist[i].f >= 0)
 			FD_SET(cachelist[i].f, fdset);
 	}
+	if (nfsd_nl_notify_sock)
+		FD_SET(nl_socket_get_fd(nfsd_nl_notify_sock), fdset);
 }
 
 /**
  * cache_process_req - process any active cache file descriptors during service loop iteration
  * @fdset: pointer to fd_set to examine for activity
  */
-int cache_process_req(fd_set *readfds) 
+int cache_process_req(fd_set *readfds)
 {
 	int i;
 	int cnt = 0;
@@ -1654,6 +2204,12 @@ int cache_process_req(fd_set *readfds)
 			FD_CLR(cachelist[i].f, readfds);
 		}
 	}
+	if (nfsd_nl_notify_sock &&
+	    FD_ISSET(nl_socket_get_fd(nfsd_nl_notify_sock), readfds)) {
+		cnt++;
+		cache_nl_process_export();
+		FD_CLR(nl_socket_get_fd(nfsd_nl_notify_sock), readfds);
+	}
 	return cnt;
 }
 

-- 
2.53.0

[PATCH nfs-utils 08/17] exportd/mountd: add netlink support for the nfsd.fh cache

[Jeff Layton]

Refactor cache_nl_set_reqs() to take a socket parameter so it can be
reused for multiple genl families.

Rename cache_nl_process() to cache_nl_process_export() and factor out
the notification drain and auth_reload() into a new
cache_nfsd_nl_process() wrapper.

Add the expkey (nfsd.fh) netlink cache handler, which uses the same
nfsd genl family as svc_export. The handler resolves fsid-based
lookups including CROSSMOUNT submount iteration and responds via
NFSD_CMD_EXPKEY_SET_REQS.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/export/cache.c | 323 +++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 314 insertions(+), 9 deletions(-)

diff --git a/support/export/cache.c b/support/export/cache.c
index 090ea0a6cc6da7e9c2ce255601103d2442526067..c350662fd97c33c40c1d59297b9638141a67befb 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -1477,7 +1477,7 @@ static struct nl_msg *cache_nl_new_msg(int family, int cmd, int flags)
 	return msg;
 }
 
-static int cache_nl_set_reqs(struct nl_msg *msg)
+static int cache_nl_set_reqs(struct nl_sock *sock, struct nl_msg *msg)
 {
 	struct nl_cb *cb;
 	int done = 0;
@@ -1490,14 +1490,14 @@ static int cache_nl_set_reqs(struct nl_msg *msg)
 	nl_cb_set(cb, NL_CB_ACK, NL_CB_CUSTOM, nl_finish_cb, &done);
 	nl_cb_err(cb, NL_CB_CUSTOM, nl_error_cb, &done);
 
-	ret = nl_send_auto(nfsd_nl_cmd_sock, msg);
+	ret = nl_send_auto(sock, msg);
 	if (ret < 0) {
 		nl_cb_put(cb);
 		return ret;
 	}
 
 	while (!done) {
-		ret = nl_recvmsgs(nfsd_nl_cmd_sock, cb);
+		ret = nl_recvmsgs(sock, cb);
 		if (ret < 0)
 			break;
 	}
@@ -1515,9 +1515,6 @@ static void cache_nl_process_export(void)
 	struct nl_msg *msg;
 	int i;
 
-	/* Drain pending notifications */
-	cache_nfsd_nl_drain();
-
 	/* Fetch all pending requests from the kernel */
 	if (cache_nl_get_export_reqs(&reqs, &nreqs)) {
 		xlog(L_WARNING, "cache_nl_process_export: failed to get export requests");
@@ -1571,7 +1568,7 @@ static void cache_nl_process_export(void)
 		}
 
 		if (nfsd_nl_add_export(msg, dom, path, epp, ttl) < 0) {
-			cache_nl_set_reqs(msg);
+			cache_nl_set_reqs(nfsd_nl_cmd_sock, msg);
 			nlmsg_free(msg);
 			msg = cache_nl_new_msg(nfsd_nl_family,
 					       NFSD_CMD_SVC_EXPORT_SET_REQS, 0);
@@ -1588,7 +1585,7 @@ static void cache_nl_process_export(void)
 		nfs_freeaddrinfo(ai);
 	}
 
-	cache_nl_set_reqs(msg);
+	cache_nl_set_reqs(nfsd_nl_cmd_sock, msg);
 	nlmsg_free(msg);
 
 out_free:
@@ -1599,6 +1596,314 @@ out_free:
 	free(reqs);
 }
 
+/*
+ * Netlink-based expkey (nfsd.fh) cache support.
+ *
+ * Uses the same nfsd genl family as svc_export. The kernel sends
+ * NFSD_CMD_CACHE_NOTIFY with NFSD_CACHE_TYPE_EXPKEY to signal
+ * pending expkey cache requests.
+ */
+struct expkey_req {
+	char	*client;
+	int	fsidtype;
+	char	*fsid;
+	int	fsidlen;
+};
+
+struct get_expkey_reqs_data {
+	struct expkey_req	*reqs;
+	int			nreqs;
+	int			maxreqs;
+	int			err;
+};
+
+static int get_expkey_reqs_cb(struct nl_msg *msg, void *arg)
+{
+	struct get_expkey_reqs_data *data = arg;
+	struct genlmsghdr *gnlh = nlmsg_data(nlmsg_hdr(msg));
+	struct nlattr *attr;
+	int rem;
+
+	nla_for_each_attr(attr, genlmsg_attrdata(gnlh, 0),
+			  genlmsg_attrlen(gnlh, 0), rem) {
+		struct nlattr *tb[NFSD_A_EXPKEY_PATH + 1];
+		struct expkey_req *req;
+
+		if (nla_type(attr) != NFSD_A_EXPKEY_REQS_REQUESTS)
+			continue;
+
+		if (nla_parse_nested(tb, NFSD_A_EXPKEY_PATH, attr, NULL))
+			continue;
+
+		if (!tb[NFSD_A_EXPKEY_CLIENT] ||
+		    !tb[NFSD_A_EXPKEY_FSIDTYPE] ||
+		    !tb[NFSD_A_EXPKEY_FSID])
+			continue;
+
+		if (data->nreqs >= data->maxreqs) {
+			int newmax = data->maxreqs ? data->maxreqs * 2 : 16;
+			struct expkey_req *tmp;
+
+			tmp = realloc(data->reqs, newmax * sizeof(*tmp));
+			if (!tmp) {
+				data->err = -ENOMEM;
+				return NL_STOP;
+			}
+			data->reqs = tmp;
+			data->maxreqs = newmax;
+		}
+
+		req = &data->reqs[data->nreqs++];
+		req->client = strdup(nla_get_string(tb[NFSD_A_EXPKEY_CLIENT]));
+		req->fsidtype = nla_get_u8(tb[NFSD_A_EXPKEY_FSIDTYPE]);
+		req->fsidlen = nla_len(tb[NFSD_A_EXPKEY_FSID]);
+		req->fsid = malloc(req->fsidlen);
+
+		if (!req->client || !req->fsid) {
+			data->err = -ENOMEM;
+			return NL_STOP;
+		}
+		memcpy(req->fsid, nla_data(tb[NFSD_A_EXPKEY_FSID]),
+		       req->fsidlen);
+	}
+
+	return NL_OK;
+}
+
+static int cache_nl_get_expkey_reqs(struct expkey_req **reqs_out,
+				    int *nreqs_out)
+{
+	struct get_expkey_reqs_data data = { };
+	struct nl_msg *msg;
+	struct nl_cb *cb;
+	int done = 0;
+	int ret;
+
+	msg = cache_nl_new_msg(nfsd_nl_family,
+			       NFSD_CMD_EXPKEY_GET_REQS, NLM_F_DUMP);
+	if (!msg)
+		return -ENOMEM;
+
+	cb = nl_cb_alloc(NL_CB_DEFAULT);
+	if (!cb) {
+		nlmsg_free(msg);
+		return -ENOMEM;
+	}
+
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, get_expkey_reqs_cb, &data);
+	nl_cb_set(cb, NL_CB_FINISH, NL_CB_CUSTOM, nl_finish_cb, &done);
+	nl_cb_err(cb, NL_CB_CUSTOM, nl_error_cb, &done);
+
+	ret = nl_send_auto(nfsd_nl_cmd_sock, msg);
+	nlmsg_free(msg);
+	if (ret < 0) {
+		nl_cb_put(cb);
+		return ret;
+	}
+
+	while (!done) {
+		ret = nl_recvmsgs(nfsd_nl_cmd_sock, cb);
+		if (ret < 0)
+			break;
+	}
+
+	nl_cb_put(cb);
+
+	if (data.err) {
+		int i;
+		for (i = 0; i < data.nreqs; i++) {
+			free(data.reqs[i].client);
+			free(data.reqs[i].fsid);
+		}
+		free(data.reqs);
+		return data.err;
+	}
+
+	*reqs_out = data.reqs;
+	*nreqs_out = data.nreqs;
+	return 0;
+}
+
+static int nfsd_nl_add_expkey(struct nl_msg *msg, char *dom, int fsidtype,
+			 char *fsid, int fsidlen, char *found_path)
+{
+	struct nlattr *nest;
+
+	nest = nla_nest_start(msg, NFSD_A_EXPKEY_REQS_REQUESTS);
+	if (!nest)
+		return -1;
+
+	if (nla_put_string(msg, NFSD_A_EXPKEY_CLIENT, dom) < 0 ||
+	    nla_put_u8(msg, NFSD_A_EXPKEY_FSIDTYPE, fsidtype) < 0 ||
+	    nla_put(msg, NFSD_A_EXPKEY_FSID, fsidlen, fsid) < 0 ||
+	    nla_put_u64(msg, NFSD_A_EXPKEY_EXPIRY, 0x7fffffff) < 0)
+		goto nla_failure;
+
+	if (found_path) {
+		if (nla_put_string(msg, NFSD_A_EXPKEY_PATH,
+				   found_path) < 0)
+			goto nla_failure;
+	} else {
+		if (nla_put_flag(msg, NFSD_A_EXPKEY_NEGATIVE) < 0)
+			goto nla_failure;
+	}
+
+	nla_nest_end(msg, nest);
+	return 0;
+
+nla_failure:
+	nla_nest_cancel(msg, nest);
+	return -1;
+}
+
+static void cache_nl_process_expkey(void)
+{
+	struct expkey_req *reqs = NULL;
+	int nreqs = 0;
+	struct nl_msg *msg;
+	int i;
+
+	if (cache_nl_get_expkey_reqs(&reqs, &nreqs)) {
+		xlog(L_WARNING, "cache_nl_process_expkey: failed to get expkey requests");
+		return;
+	}
+
+	if (!nreqs)
+		return;
+
+	xlog(D_CALL, "cache_nl_process_expkey: %d pending expkey requests", nreqs);
+
+	msg = cache_nl_new_msg(nfsd_nl_family, NFSD_CMD_EXPKEY_SET_REQS, 0);
+	if (!msg)
+		goto out_free;
+
+	for (i = 0; i < nreqs; i++) {
+		char *dom = reqs[i].client;
+		int fsidtype = reqs[i].fsidtype;
+		char *fsid = reqs[i].fsid;
+		int fsidlen = reqs[i].fsidlen;
+		struct parsed_fsid parsed;
+		struct addrinfo *ai = NULL;
+		struct exportent *found = NULL;
+		char *found_path = NULL;
+		nfs_export *exp;
+		int j;
+
+		if (parse_fsid(fsidtype, fsidlen, fsid, &parsed))
+			goto do_add_expkey;
+
+		if (is_ipaddr_client(dom)) {
+			ai = lookup_client_addr(dom);
+			if (!ai)
+				goto do_add_expkey;
+		}
+
+		for (j = 0; j < MCL_MAXTYPES; j++) {
+			nfs_export *prev = NULL;
+			nfs_export *next_exp;
+			void *mnt = NULL;
+
+			for (exp = exportlist[j].p_head; exp;
+			     exp = next_exp) {
+				char *path;
+
+				if (exp->m_export.e_flags &
+				    NFSEXP_CROSSMOUNT) {
+					if (prev == exp) {
+						path = next_mnt(&mnt,
+							exp->m_export.e_path);
+						if (!path) {
+							next_exp = exp->m_next;
+							prev = NULL;
+							continue;
+						}
+						next_exp = exp;
+					} else {
+						prev = exp;
+						mnt = NULL;
+						path = exp->m_export.e_path;
+						next_exp = exp;
+					}
+				} else {
+					path = exp->m_export.e_path;
+					next_exp = exp->m_next;
+				}
+
+				if (!is_ipaddr_client(dom) &&
+				    !namelist_client_matches(exp, dom))
+					continue;
+
+				switch (match_fsid(&parsed, exp, path)) {
+				case 0:
+					continue;
+				case -1:
+					continue;
+				}
+
+				if (is_ipaddr_client(dom) &&
+				    !ipaddr_client_matches(exp, ai))
+					continue;
+
+				if (!found ||
+				    subexport(&exp->m_export, found)) {
+					found = &exp->m_export;
+					free(found_path);
+					found_path = strdup(path);
+					if (!found_path)
+						goto do_add_expkey;
+				}
+			}
+		}
+
+do_add_expkey:
+		if (nfsd_nl_add_expkey(msg, dom, fsidtype, fsid,
+				       fsidlen, found_path) < 0) {
+			cache_nl_set_reqs(nfsd_nl_cmd_sock, msg);
+			nlmsg_free(msg);
+			msg = cache_nl_new_msg(nfsd_nl_family,
+					       NFSD_CMD_EXPKEY_SET_REQS, 0);
+			if (!msg) {
+				free(found_path);
+				nfs_freeaddrinfo(ai);
+				goto out_free;
+			}
+			if (nfsd_nl_add_expkey(msg, dom, fsidtype, fsid,
+					       fsidlen, found_path) < 0)
+				xlog(L_WARNING, "%s: skipping oversized "
+				     "entry", __func__);
+		}
+		if (!found)
+			xlog(D_AUTH, "denied access to %s",
+			     *dom == '$' ? dom + 1 : dom);
+		free(found_path);
+		nfs_freeaddrinfo(ai);
+	}
+
+	cache_nl_set_reqs(nfsd_nl_cmd_sock, msg);
+	nlmsg_free(msg);
+
+out_free:
+	for (i = 0; i < nreqs; i++) {
+		free(reqs[i].client);
+		free(reqs[i].fsid);
+	}
+	free(reqs);
+}
+
+static void cache_nfsd_nl_process(void)
+{
+	/* Drain pending nfsd notifications */
+	cache_nfsd_nl_drain();
+
+	auth_reload();
+
+	/* Handle any pending svc_export requests */
+	cache_nl_process_export();
+
+	/* Handle any pending expkey requests */
+	cache_nl_process_expkey();
+}
+
 static int can_reexport_via_fsidnum(struct exportent *exp, struct statfs *st)
 {
 	if (st->f_type != 0x6969 /* NFS_SUPER_MAGIC */)
@@ -2207,7 +2512,7 @@ int cache_process_req(fd_set *readfds)
 	if (nfsd_nl_notify_sock &&
 	    FD_ISSET(nl_socket_get_fd(nfsd_nl_notify_sock), readfds)) {
 		cnt++;
-		cache_nl_process_export();
+		cache_nfsd_nl_process();
 		FD_CLR(nl_socket_get_fd(nfsd_nl_notify_sock), readfds);
 	}
 	return cnt;

-- 
2.53.0

[PATCH nfs-utils 09/17] exportd/mountd: add netlink support for the auth.unix.ip cache

[Jeff Layton]

Refactor cache_nfsd_nl_open() into a generic cache_genl_open() helper
that takes the family name, multicast group name, and output pointers
for the command socket, notification socket, and resolved family ID.
Convert cache_nfsd_nl_open() to use it.

Add the sunrpc genl family socket setup for handling auth.unix.ip and
auth.unix.gid cache upcalls. The sunrpc family is resolved at startup
and silently falls back if the kernel doesn't support it.

Add the ip_map (auth.unix.ip) netlink cache handler. For each pending
request, the handler resolves the IP address using client_resolve()
and responds via SUNRPC_CMD_IP_MAP_SET_REQS with the domain name or a
negative entry.

Wire the sunrpc notification socket into cache_open(), cache_set_fds(),
and cache_process_req().

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/export/cache.c | 366 +++++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 342 insertions(+), 24 deletions(-)

diff --git a/support/export/cache.c b/support/export/cache.c
index c350662fd97c33c40c1d59297b9638141a67befb..50c2de08c504da1a05631938ee51251d82c52377 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -48,6 +48,12 @@
 #include "nfsd_netlink.h"
 #endif
 
+#ifdef USE_SYSTEM_SUNRPC_NETLINK_H
+#include <linux/sunrpc_netlink.h>
+#else
+#include "sunrpc_netlink.h"
+#endif
+
 #ifdef USE_BLKID
 #include "blkid/blkid.h"
 #endif
@@ -1108,58 +1114,71 @@ static struct nl_sock *nl_sock_setup(void)
 	return sock;
 }
 
-static int cache_nfsd_nl_open(void)
+static int cache_genl_open(const char *family_name, const char *mcgrp_name,
+			   struct nl_sock **cmd_sock,
+			   struct nl_sock **notify_sock, int *family_out)
 {
 	int grp;
 
-	nfsd_nl_family = 0;
+	*family_out = 0;
 
-	nfsd_nl_cmd_sock = nl_sock_setup();
-	if (!nfsd_nl_cmd_sock) {
-		xlog(D_NETLINK, "cache_nfsd_nl_open: failed to allocate command socket");
+	*cmd_sock = nl_sock_setup();
+	if (!*cmd_sock) {
+		xlog(D_NETLINK, "%s: failed to allocate command socket",
+		     family_name);
 		return -ENOMEM;
 	}
 
-	nfsd_nl_family = genl_ctrl_resolve(nfsd_nl_cmd_sock, NFSD_FAMILY_NAME);
-	if (nfsd_nl_family < 0) {
-		xlog(D_NETLINK, "cache_nfsd_nl_open: nfsd netlink family not found");
+	*family_out = genl_ctrl_resolve(*cmd_sock, family_name);
+	if (*family_out < 0) {
+		xlog(D_NETLINK, "%s: netlink family not found", family_name);
 		goto out_free_cmd;
 	}
 
-	grp = genl_ctrl_resolve_grp(nfsd_nl_cmd_sock, NFSD_FAMILY_NAME,
-				    NFSD_MCGRP_EXPORTD);
+	grp = genl_ctrl_resolve_grp(*cmd_sock, family_name, mcgrp_name);
 	if (grp < 0) {
-		xlog(D_NETLINK, "cache_nfsd_nl_open: exportd multicast group not found");
+		xlog(D_NETLINK, "%s: %s multicast group not found",
+		     family_name, mcgrp_name);
 		goto out_free_cmd;
 	}
 
-	nfsd_nl_notify_sock = nl_sock_setup();
-	if (!nfsd_nl_notify_sock) {
-		xlog(D_NETLINK, "cache_nfsd_nl_open: failed to allocate notify socket");
+	*notify_sock = nl_sock_setup();
+	if (!*notify_sock) {
+		xlog(D_NETLINK, "%s: failed to allocate notify socket",
+		     family_name);
 		goto out_free_cmd;
 	}
 
-	nl_socket_disable_seq_check(nfsd_nl_notify_sock);
+	nl_socket_disable_seq_check(*notify_sock);
 
-	if (nl_socket_add_membership(nfsd_nl_notify_sock, grp)) {
-		xlog(L_WARNING, "cache_nfsd_nl_open: failed to join exportd multicast group");
+	if (nl_socket_add_membership(*notify_sock, grp)) {
+		xlog(L_WARNING, "%s: failed to join %s multicast group",
+		     family_name, mcgrp_name);
 		goto out_free_notify;
 	}
 
-	nl_socket_set_nonblocking(nfsd_nl_notify_sock);
-	xlog(D_NETLINK, "cache_nfsd_nl_open: listening for export notifications");
+	nl_socket_set_nonblocking(*notify_sock);
+	xlog(D_NETLINK, "%s: listening for %s notifications",
+	     family_name, mcgrp_name);
 	return 0;
 
 out_free_notify:
-	nl_socket_free(nfsd_nl_notify_sock);
-	nfsd_nl_notify_sock = NULL;
+	nl_socket_free(*notify_sock);
+	*notify_sock = NULL;
 out_free_cmd:
-	nl_socket_free(nfsd_nl_cmd_sock);
-	nfsd_nl_cmd_sock = NULL;
-	nfsd_nl_family = 0;
+	nl_socket_free(*cmd_sock);
+	*cmd_sock = NULL;
+	*family_out = 0;
 	return -ENOENT;
 }
 
+static int cache_nfsd_nl_open(void)
+{
+	return cache_genl_open(NFSD_FAMILY_NAME, NFSD_MCGRP_EXPORTD,
+			       &nfsd_nl_cmd_sock, &nfsd_nl_notify_sock,
+			       &nfsd_nl_family);
+}
+
 static int nfsd_nl_notify_handler(struct nl_msg *UNUSED(msg), void *UNUSED(arg))
 {
 	return NL_OK;
@@ -1904,6 +1923,296 @@ static void cache_nfsd_nl_process(void)
 	cache_nl_process_expkey();
 }
 
+/*
+ * Netlink-based sunrpc cache support.
+ *
+ * The sunrpc genl family handles auth.unix.ip and auth.unix.gid caches.
+ * A SUNRPC_CMD_CACHE_NOTIFY on the "exportd" multicast group signals
+ * pending cache requests.
+ */
+static struct nl_sock *sunrpc_nl_notify_sock;
+static struct nl_sock *sunrpc_nl_cmd_sock;
+static int sunrpc_nl_family;
+
+static int cache_sunrpc_nl_open(void)
+{
+	return cache_genl_open(SUNRPC_FAMILY_NAME, SUNRPC_MCGRP_EXPORTD,
+			       &sunrpc_nl_cmd_sock, &sunrpc_nl_notify_sock,
+			       &sunrpc_nl_family);
+}
+
+static void cache_sunrpc_nl_drain(void)
+{
+	struct nl_cb *cb;
+
+	cb = nl_cb_alloc(NL_CB_DEFAULT);
+	if (!cb)
+		return;
+
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, nfsd_nl_notify_handler, NULL);
+	nl_recvmsgs(sunrpc_nl_notify_sock, cb);
+	nl_cb_put(cb);
+}
+
+/*
+ * ip_map (auth.unix.ip) netlink handler
+ */
+struct ip_map_req {
+	char	*class;
+	char	*addr;
+};
+
+struct get_ip_map_reqs_data {
+	struct ip_map_req	*reqs;
+	int			nreqs;
+	int			maxreqs;
+	int			err;
+};
+
+static int get_ip_map_reqs_cb(struct nl_msg *msg, void *arg)
+{
+	struct get_ip_map_reqs_data *data = arg;
+	struct genlmsghdr *gnlh = nlmsg_data(nlmsg_hdr(msg));
+	struct nlattr *attr;
+	int rem;
+
+	nla_for_each_attr(attr, genlmsg_attrdata(gnlh, 0),
+			  genlmsg_attrlen(gnlh, 0), rem) {
+		struct nlattr *tb[SUNRPC_A_IP_MAP_EXPIRY + 1];
+		struct ip_map_req *req;
+
+		if (nla_type(attr) != SUNRPC_A_IP_MAP_REQS_REQUESTS)
+			continue;
+
+		if (nla_parse_nested(tb, SUNRPC_A_IP_MAP_EXPIRY, attr,
+				     NULL))
+			continue;
+
+		if (!tb[SUNRPC_A_IP_MAP_CLASS] ||
+		    !tb[SUNRPC_A_IP_MAP_ADDR])
+			continue;
+
+		if (data->nreqs >= data->maxreqs) {
+			int newmax = data->maxreqs ? data->maxreqs * 2 : 16;
+			struct ip_map_req *tmp;
+
+			tmp = realloc(data->reqs, newmax * sizeof(*tmp));
+			if (!tmp) {
+				data->err = -ENOMEM;
+				return NL_STOP;
+			}
+			data->reqs = tmp;
+			data->maxreqs = newmax;
+		}
+
+		req = &data->reqs[data->nreqs++];
+		req->class = strdup(nla_get_string(tb[SUNRPC_A_IP_MAP_CLASS]));
+		req->addr = strdup(nla_get_string(tb[SUNRPC_A_IP_MAP_ADDR]));
+
+		if (!req->class || !req->addr) {
+			data->err = -ENOMEM;
+			return NL_STOP;
+		}
+	}
+
+	return NL_OK;
+}
+
+static int cache_nl_get_ip_map_reqs(struct ip_map_req **reqs_out,
+				    int *nreqs_out)
+{
+	struct get_ip_map_reqs_data data = { };
+	struct nl_msg *msg;
+	struct nl_cb *cb;
+	int done = 0;
+	int ret;
+
+	msg = cache_nl_new_msg(sunrpc_nl_family,
+			       SUNRPC_CMD_IP_MAP_GET_REQS, NLM_F_DUMP);
+	if (!msg)
+		return -ENOMEM;
+
+	cb = nl_cb_alloc(NL_CB_DEFAULT);
+	if (!cb) {
+		nlmsg_free(msg);
+		return -ENOMEM;
+	}
+
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, get_ip_map_reqs_cb, &data);
+	nl_cb_set(cb, NL_CB_FINISH, NL_CB_CUSTOM, nl_finish_cb, &done);
+	nl_cb_err(cb, NL_CB_CUSTOM, nl_error_cb, &done);
+
+	ret = nl_send_auto(sunrpc_nl_cmd_sock, msg);
+	nlmsg_free(msg);
+	if (ret < 0) {
+		nl_cb_put(cb);
+		return ret;
+	}
+
+	while (!done) {
+		ret = nl_recvmsgs(sunrpc_nl_cmd_sock, cb);
+		if (ret < 0)
+			break;
+	}
+
+	nl_cb_put(cb);
+
+	if (data.err) {
+		int i;
+		for (i = 0; i < data.nreqs; i++) {
+			free(data.reqs[i].class);
+			free(data.reqs[i].addr);
+		}
+		free(data.reqs);
+		return data.err;
+	}
+
+	*reqs_out = data.reqs;
+	*nreqs_out = data.nreqs;
+	return 0;
+}
+
+static int nl_add_ip_map(struct nl_msg *msg, char *class, char *addr,
+			 char *domain)
+{
+	struct nlattr *nest;
+	time_t now = time(0);
+
+	nest = nla_nest_start(msg, SUNRPC_A_IP_MAP_REQS_REQUESTS);
+	if (!nest)
+		return -1;
+
+	if (nla_put_string(msg, SUNRPC_A_IP_MAP_CLASS, class) < 0 ||
+	    nla_put_string(msg, SUNRPC_A_IP_MAP_ADDR, addr) < 0 ||
+	    nla_put_u64(msg, SUNRPC_A_IP_MAP_EXPIRY,
+			now + default_ttl) < 0)
+		goto nla_failure;
+
+	if (domain) {
+		if (nla_put_string(msg, SUNRPC_A_IP_MAP_DOMAIN,
+				   domain) < 0)
+			goto nla_failure;
+	} else {
+		if (nla_put_flag(msg, SUNRPC_A_IP_MAP_NEGATIVE) < 0)
+			goto nla_failure;
+	}
+
+	nla_nest_end(msg, nest);
+	return 0;
+
+nla_failure:
+	nla_nest_cancel(msg, nest);
+	return -1;
+}
+
+static void cache_nl_process_ip_map(void)
+{
+	struct ip_map_req *reqs = NULL;
+	int nreqs = 0;
+	struct nl_msg *msg;
+	int i;
+
+	if (cache_nl_get_ip_map_reqs(&reqs, &nreqs)) {
+		xlog(L_WARNING, "cache_nl_process_ip_map: failed to get ip_map requests");
+		return;
+	}
+
+	if (!nreqs)
+		return;
+
+	xlog(D_CALL, "cache_nl_process_ip_map: %d pending ip_map requests",
+	     nreqs);
+
+	msg = cache_nl_new_msg(sunrpc_nl_family,
+			       SUNRPC_CMD_IP_MAP_SET_REQS, 0);
+	if (!msg)
+		goto out_free;
+
+	for (i = 0; i < nreqs; i++) {
+		char *class = reqs[i].class;
+		char *ipaddr = reqs[i].addr;
+		char *client = NULL;
+		char *domain = NULL;
+		char *dom_alloc = NULL;
+		struct addrinfo *tmp = NULL;
+
+		if (strcmp(class, "nfsd") == 0) {
+			tmp = host_pton(ipaddr);
+			if (tmp) {
+				struct addrinfo *ai;
+
+				ai = client_resolve(tmp->ai_addr);
+				if (ai) {
+					client = client_compose(ai);
+					nfs_freeaddrinfo(ai);
+				}
+			}
+
+			if (use_ipaddr && client) {
+				dom_alloc = malloc(strlen(ipaddr) + 2);
+				if (dom_alloc) {
+					dom_alloc[0] = '$';
+					strcpy(dom_alloc + 1, ipaddr);
+					domain = dom_alloc;
+				}
+			} else if (client) {
+				domain = *client ? client : "DEFAULT";
+			}
+		}
+
+		if (nl_add_ip_map(msg, class, ipaddr, domain) < 0) {
+			cache_nl_set_reqs(sunrpc_nl_cmd_sock, msg);
+			nlmsg_free(msg);
+			msg = cache_nl_new_msg(sunrpc_nl_family,
+					       SUNRPC_CMD_IP_MAP_SET_REQS, 0);
+			if (!msg) {
+				free(dom_alloc);
+				free(client);
+				nfs_freeaddrinfo(tmp);
+				goto out_free;
+			}
+			if (nl_add_ip_map(msg, class, ipaddr, domain) < 0)
+				xlog(L_WARNING, "%s: skipping oversized "
+				     "entry for %s", __func__, ipaddr);
+		}
+
+		if (tmp && !client)
+			xlog(D_AUTH, "failed authentication for IP %s",
+			     ipaddr);
+		else if (client && !use_ipaddr)
+			xlog(D_AUTH, "successful authentication for IP %s as %s",
+			     ipaddr, *client ? client : "DEFAULT");
+		else if (client)
+			xlog(D_AUTH, "successful authentication for IP %s",
+			     ipaddr);
+
+		free(dom_alloc);
+		free(client);
+		nfs_freeaddrinfo(tmp);
+	}
+
+	cache_nl_set_reqs(sunrpc_nl_cmd_sock, msg);
+	nlmsg_free(msg);
+
+out_free:
+	for (i = 0; i < nreqs; i++) {
+		free(reqs[i].class);
+		free(reqs[i].addr);
+	}
+	free(reqs);
+}
+
+static void cache_sunrpc_nl_process(void)
+{
+	/* Drain pending sunrpc notifications */
+	cache_sunrpc_nl_drain();
+
+	auth_reload();
+
+	/* Handle any pending ip_map requests */
+	cache_nl_process_ip_map();
+}
+
 static int can_reexport_via_fsidnum(struct exportent *exp, struct statfs *st)
 {
 	if (st->f_type != 0x6969 /* NFS_SUPER_MAGIC */)
@@ -2476,6 +2785,7 @@ void cache_open(void)
 		cachelist[i].f = open(path, O_RDWR);
 	}
 	cache_nfsd_nl_open();
+	cache_sunrpc_nl_open();
 }
 
 /**
@@ -2491,6 +2801,8 @@ void cache_set_fds(fd_set *fdset)
 	}
 	if (nfsd_nl_notify_sock)
 		FD_SET(nl_socket_get_fd(nfsd_nl_notify_sock), fdset);
+	if (sunrpc_nl_notify_sock)
+		FD_SET(nl_socket_get_fd(sunrpc_nl_notify_sock), fdset);
 }
 
 /**
@@ -2515,6 +2827,12 @@ int cache_process_req(fd_set *readfds)
 		cache_nfsd_nl_process();
 		FD_CLR(nl_socket_get_fd(nfsd_nl_notify_sock), readfds);
 	}
+	if (sunrpc_nl_notify_sock &&
+	    FD_ISSET(nl_socket_get_fd(sunrpc_nl_notify_sock), readfds)) {
+		cnt++;
+		cache_sunrpc_nl_process();
+		FD_CLR(nl_socket_get_fd(sunrpc_nl_notify_sock), readfds);
+	}
 	return cnt;
 }
 

-- 
2.53.0

[PATCH nfs-utils 10/17] exportd/mountd: add netlink support for the auth.unix.gid cache

[Jeff Layton]

Add the unix_gid (auth.unix.gid) netlink cache handler. For each
pending request, the handler resolves the UID to a group list using
getpwuid() and getgrouplist(), and responds via
SUNRPC_CMD_UNIX_GID_SET_REQS.

The handler is only active when manage_gids is set, matching the
behavior of the existing procfs handler.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/export/cache.c | 236 +++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 236 insertions(+)

diff --git a/support/export/cache.c b/support/export/cache.c
index 50c2de08c504da1a05631938ee51251d82c52377..43cb16079da867e6633b9cc6436689ab0e156e44 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -110,6 +110,7 @@ static bool path_lookup_error(int err)
 #define INITIAL_MANAGED_GROUPS 100
 
 extern int use_ipaddr;
+extern int manage_gids;
 
 static void auth_unix_ip(int f)
 {
@@ -2202,6 +2203,237 @@ out_free:
 	free(reqs);
 }
 
+/*
+ * unix_gid (auth.unix.gid) netlink handler
+ */
+struct unix_gid_req {
+	uid_t	uid;
+};
+
+struct get_unix_gid_reqs_data {
+	struct unix_gid_req	*reqs;
+	int			nreqs;
+	int			maxreqs;
+	int			err;
+};
+
+static int get_unix_gid_reqs_cb(struct nl_msg *msg, void *arg)
+{
+	struct get_unix_gid_reqs_data *data = arg;
+	struct genlmsghdr *gnlh = nlmsg_data(nlmsg_hdr(msg));
+	struct nlattr *attr;
+	int rem;
+
+	nla_for_each_attr(attr, genlmsg_attrdata(gnlh, 0),
+			  genlmsg_attrlen(gnlh, 0), rem) {
+		struct nlattr *tb[SUNRPC_A_UNIX_GID_EXPIRY + 1];
+		struct unix_gid_req *req;
+
+		if (nla_type(attr) != SUNRPC_A_UNIX_GID_REQS_REQUESTS)
+			continue;
+
+		if (nla_parse_nested(tb, SUNRPC_A_UNIX_GID_EXPIRY, attr,
+				     NULL))
+			continue;
+
+		if (!tb[SUNRPC_A_UNIX_GID_UID])
+			continue;
+
+		if (data->nreqs >= data->maxreqs) {
+			int newmax = data->maxreqs ? data->maxreqs * 2 : 16;
+			struct unix_gid_req *tmp;
+
+			tmp = realloc(data->reqs, newmax * sizeof(*tmp));
+			if (!tmp) {
+				data->err = -ENOMEM;
+				return NL_STOP;
+			}
+			data->reqs = tmp;
+			data->maxreqs = newmax;
+		}
+
+		req = &data->reqs[data->nreqs++];
+		req->uid = nla_get_u32(tb[SUNRPC_A_UNIX_GID_UID]);
+	}
+
+	return NL_OK;
+}
+
+static int cache_nl_get_unix_gid_reqs(struct unix_gid_req **reqs_out,
+				      int *nreqs_out)
+{
+	struct get_unix_gid_reqs_data data = { };
+	struct nl_msg *msg;
+	struct nl_cb *cb;
+	int done = 0;
+	int ret;
+
+	msg = cache_nl_new_msg(sunrpc_nl_family,
+			       SUNRPC_CMD_UNIX_GID_GET_REQS, NLM_F_DUMP);
+	if (!msg)
+		return -ENOMEM;
+
+	cb = nl_cb_alloc(NL_CB_DEFAULT);
+	if (!cb) {
+		nlmsg_free(msg);
+		return -ENOMEM;
+	}
+
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, get_unix_gid_reqs_cb,
+		  &data);
+	nl_cb_set(cb, NL_CB_FINISH, NL_CB_CUSTOM, nl_finish_cb, &done);
+	nl_cb_err(cb, NL_CB_CUSTOM, nl_error_cb, &done);
+
+	ret = nl_send_auto(sunrpc_nl_cmd_sock, msg);
+	nlmsg_free(msg);
+	if (ret < 0) {
+		nl_cb_put(cb);
+		return ret;
+	}
+
+	while (!done) {
+		ret = nl_recvmsgs(sunrpc_nl_cmd_sock, cb);
+		if (ret < 0)
+			break;
+	}
+
+	nl_cb_put(cb);
+
+	if (data.err) {
+		free(data.reqs);
+		return data.err;
+	}
+
+	*reqs_out = data.reqs;
+	*nreqs_out = data.nreqs;
+	return 0;
+}
+
+static int nl_add_unix_gid(struct nl_msg *msg, uid_t uid, gid_t *groups,
+			   int ngroups)
+{
+	struct nlattr *nest;
+	time_t now = time(0);
+	int i;
+
+	nest = nla_nest_start(msg, SUNRPC_A_UNIX_GID_REQS_REQUESTS);
+	if (!nest)
+		return -1;
+
+	if (nla_put_u32(msg, SUNRPC_A_UNIX_GID_UID, uid) < 0 ||
+	    nla_put_u64(msg, SUNRPC_A_UNIX_GID_EXPIRY, now + default_ttl) < 0)
+		goto nla_failure;
+
+	if (ngroups >= 0) {
+		for (i = 0; i < ngroups; i++)
+			if (nla_put_u32(msg, SUNRPC_A_UNIX_GID_GIDS, groups[i]) < 0)
+				goto nla_failure;
+	} else {
+		if (nla_put_flag(msg, SUNRPC_A_UNIX_GID_NEGATIVE) < 0)
+			goto nla_failure;
+	}
+
+	nla_nest_end(msg, nest);
+	return 0;
+nla_failure:
+	nla_nest_cancel(msg, nest);
+	return -1;
+}
+
+static void cache_nl_process_unix_gid(void)
+{
+	struct unix_gid_req *reqs = NULL;
+	int nreqs = 0;
+	struct nl_msg *msg;
+	static gid_t *groups = NULL;
+	static int groups_len = 0;
+	int i;
+
+	if (cache_nl_get_unix_gid_reqs(&reqs, &nreqs)) {
+		xlog(L_WARNING, "cache_nl_process_unix_gid: failed to get unix_gid requests");
+		return;
+	}
+
+	if (!nreqs)
+		return;
+
+	xlog(D_CALL, "cache_nl_process_unix_gid: %d pending unix_gid requests",
+	     nreqs);
+
+	if (groups_len == 0) {
+		groups = malloc(sizeof(gid_t) * INITIAL_MANAGED_GROUPS);
+		if (!groups)
+			goto out_free;
+		groups_len = INITIAL_MANAGED_GROUPS;
+	}
+
+	msg = cache_nl_new_msg(sunrpc_nl_family,
+			       SUNRPC_CMD_UNIX_GID_SET_REQS, 0);
+	if (!msg)
+		goto out_free;
+
+	for (i = 0; i < nreqs; i++) {
+		uid_t uid = reqs[i].uid;
+		struct passwd *pw;
+		int ngroups;
+		int rv;
+		int ret;
+
+		ngroups = groups_len;
+		pw = getpwuid(uid);
+		if (!pw) {
+			rv = -1;
+		} else {
+			rv = getgrouplist(pw->pw_name, pw->pw_gid,
+					  groups, &ngroups);
+			if (rv == -1 && ngroups >= groups_len) {
+				gid_t *more_groups;
+
+				more_groups = realloc(groups,
+						      sizeof(gid_t) * ngroups);
+				if (!more_groups) {
+					rv = -1;
+				} else {
+					groups = more_groups;
+					groups_len = ngroups;
+					rv = getgrouplist(pw->pw_name,
+							  pw->pw_gid,
+							  groups, &ngroups);
+				}
+			}
+		}
+
+		if (rv >= 0)
+			ret = nl_add_unix_gid(msg, uid, groups, ngroups);
+		else
+			ret = nl_add_unix_gid(msg, uid, NULL, -1);
+
+		if (ret < 0) {
+			/* Flush current message and retry with a fresh one */
+			cache_nl_set_reqs(sunrpc_nl_cmd_sock, msg);
+			nlmsg_free(msg);
+			msg = cache_nl_new_msg(sunrpc_nl_family,
+					       SUNRPC_CMD_UNIX_GID_SET_REQS, 0);
+			if (!msg)
+				goto out_free;
+
+			if (rv >= 0)
+				ret = nl_add_unix_gid(msg, uid, groups, ngroups);
+			else
+				ret = nl_add_unix_gid(msg, uid, NULL, -1);
+			if (ret < 0)
+				xlog(L_WARNING, "%s: skipping oversized entry for uid %u",
+				     __func__, uid);
+		}
+	}
+
+	cache_nl_set_reqs(sunrpc_nl_cmd_sock, msg);
+	nlmsg_free(msg);
+
+out_free:
+	free(reqs);
+}
+
 static void cache_sunrpc_nl_process(void)
 {
 	/* Drain pending sunrpc notifications */
@@ -2211,6 +2443,10 @@ static void cache_sunrpc_nl_process(void)
 
 	/* Handle any pending ip_map requests */
 	cache_nl_process_ip_map();
+
+	/* Handle any pending unix_gid requests */
+	if (manage_gids)
+		cache_nl_process_unix_gid();
 }
 
 static int can_reexport_via_fsidnum(struct exportent *exp, struct statfs *st)

-- 
2.53.0

[PATCH nfs-utils 11/17] mountd/exportd: only use /proc interfaces if netlink setup fails

[Jeff Layton]

Now that all of the exportd upcalls are converted to use netlink,
don't bother opening /proc channel files if that succeeds.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/export/cache.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/support/export/cache.c b/support/export/cache.c
index 43cb16079da867e6633b9cc6436689ab0e156e44..19cfbf6594b0a51d85814460f3153add89aa3a8a 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -3013,6 +3013,16 @@ void cache_open(void)
 {
 	int i;
 
+	if (cache_nfsd_nl_open() == 0) {
+		if (cache_sunrpc_nl_open() == 0)
+			return;
+		xlog(L_NOTICE, "sunrpc netlink family unavailable, falling back to /proc");
+		nl_socket_free(nfsd_nl_notify_sock);
+		nfsd_nl_notify_sock = NULL;
+		nl_socket_free(nfsd_nl_cmd_sock);
+		nfsd_nl_cmd_sock = NULL;
+	}
+
 	for (i=0; cachelist[i].cache_name; i++ ) {
 		char path[100];
 		if (!manage_gids && cachelist[i].cache_handle == auth_unix_gid)
@@ -3020,8 +3030,6 @@ void cache_open(void)
 		sprintf(path, "/proc/net/rpc/%s/channel", cachelist[i].cache_name);
 		cachelist[i].f = open(path, O_RDWR);
 	}
-	cache_nfsd_nl_open();
-	cache_sunrpc_nl_open();
 }
 
 /**

-- 
2.53.0

[PATCH nfs-utils 12/17] support/export: check for pending requests after opening netlink sockets

[Jeff Layton]

It's possible for there to be requests queued before mountd comes
online. Always check for already-pending requests after opening netlink
sockets.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/export/cache.c | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/support/export/cache.c b/support/export/cache.c
index 19cfbf6594b0a51d85814460f3153add89aa3a8a..4261e1861215ea53183dd8ce14890b0195d841e8 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -3014,8 +3014,19 @@ void cache_open(void)
 	int i;
 
 	if (cache_nfsd_nl_open() == 0) {
-		if (cache_sunrpc_nl_open() == 0)
+		if (cache_sunrpc_nl_open() == 0) {
+			/*
+			 * Check for pending requests, in case any
+			 * were queued before we opened the socket.
+			 */
+			auth_reload();
+			cache_nl_process_export();
+			cache_nl_process_expkey();
+			cache_nl_process_ip_map();
+			if (manage_gids)
+				cache_nl_process_unix_gid();
 			return;
+		}
 		xlog(L_NOTICE, "sunrpc netlink family unavailable, falling back to /proc");
 		nl_socket_free(nfsd_nl_notify_sock);
 		nfsd_nl_notify_sock = NULL;

-- 
2.53.0

[PATCH nfs-utils 13/17] exportd/mountd: use cache type from notifications to target scanning

[Jeff Layton]

Parse the NFSD_A_CACHE_NOTIFY_CACHE_TYPE and
SUNRPC_A_CACHE_NOTIFY_CACHE_TYPE attributes from netlink multicast
notifications to determine which specific caches have pending
requests. Only issue GET_REQS for the caches that actually need
servicing, avoiding unnecessary netlink round-trips.

If the notification can't be parsed, fall back to scanning all caches
for that family.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/export/cache.c | 81 ++++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 66 insertions(+), 15 deletions(-)

diff --git a/support/export/cache.c b/support/export/cache.c
index 4261e1861215ea53183dd8ce14890b0195d841e8..5a2c5760cb5410845971ba831a9ae779d17a6d87 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -1180,22 +1180,45 @@ static int cache_nfsd_nl_open(void)
 			       &nfsd_nl_family);
 }
 
-static int nfsd_nl_notify_handler(struct nl_msg *UNUSED(msg), void *UNUSED(arg))
+static int nl_seq_check_handler(struct nl_msg *UNUSED(msg), void *UNUSED(arg))
 {
 	return NL_OK;
 }
 
-static void cache_nfsd_nl_drain(void)
+static int nfsd_notify_handler(struct nl_msg *msg, void *arg)
 {
+	unsigned int *cache_mask = arg;
+	struct genlmsghdr *gnlh = nlmsg_data(nlmsg_hdr(msg));
+	struct nlattr *tb[NFSD_A_CACHE_NOTIFY_MAX + 1];
+
+	if (nla_parse(tb, NFSD_A_CACHE_NOTIFY_MAX,
+		      genlmsg_attrdata(gnlh, 0),
+		      genlmsg_attrlen(gnlh, 0), NULL) == 0 &&
+	    tb[NFSD_A_CACHE_NOTIFY_CACHE_TYPE])
+		*cache_mask |= nla_get_u32(tb[NFSD_A_CACHE_NOTIFY_CACHE_TYPE]);
+	else
+		*cache_mask = ~0U;
+
+	xlog(D_NETLINK, "nfsd_notify_handler: cache_mask=%x", *cache_mask);
+	return NL_OK;
+}
+
+static unsigned int cache_nfsd_nl_drain(void)
+{
+	unsigned int cache_mask = 0;
 	struct nl_cb *cb;
 
 	cb = nl_cb_alloc(NL_CB_DEFAULT);
 	if (!cb)
-		return;
+		return ~0U;
 
-	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, nfsd_nl_notify_handler, NULL);
+	nl_cb_set(cb, NL_CB_SEQ_CHECK, NL_CB_CUSTOM,
+		  nl_seq_check_handler, NULL);
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, nfsd_notify_handler,
+		  &cache_mask);
 	nl_recvmsgs(nfsd_nl_notify_sock, cb);
 	nl_cb_put(cb);
+	return cache_mask;
 }
 
 struct get_export_reqs_data {
@@ -1552,8 +1575,6 @@ static void cache_nl_process_export(void)
 	if (!msg)
 		goto out_free;
 
-	auth_reload();
-
 	for (i = 0; i < nreqs; i++) {
 		char *dom = reqs[i].client;
 		char *path = reqs[i].path;
@@ -1912,16 +1933,20 @@ out_free:
 
 static void cache_nfsd_nl_process(void)
 {
+	unsigned int cache_mask;
+
 	/* Drain pending nfsd notifications */
-	cache_nfsd_nl_drain();
+	cache_mask = cache_nfsd_nl_drain();
 
 	auth_reload();
 
 	/* Handle any pending svc_export requests */
-	cache_nl_process_export();
+	if (cache_mask & NFSD_CACHE_TYPE_SVC_EXPORT)
+		cache_nl_process_export();
 
 	/* Handle any pending expkey requests */
-	cache_nl_process_expkey();
+	if (cache_mask & NFSD_CACHE_TYPE_EXPKEY)
+		cache_nl_process_expkey();
 }
 
 /*
@@ -1942,17 +1967,40 @@ static int cache_sunrpc_nl_open(void)
 			       &sunrpc_nl_family);
 }
 
-static void cache_sunrpc_nl_drain(void)
+static int sunrpc_notify_handler(struct nl_msg *msg, void *arg)
 {
+	unsigned int *cache_mask = arg;
+	struct genlmsghdr *gnlh = nlmsg_data(nlmsg_hdr(msg));
+	struct nlattr *tb[SUNRPC_A_CACHE_NOTIFY_MAX + 1];
+
+	if (nla_parse(tb, SUNRPC_A_CACHE_NOTIFY_MAX,
+		      genlmsg_attrdata(gnlh, 0),
+		      genlmsg_attrlen(gnlh, 0), NULL) == 0 &&
+	    tb[SUNRPC_A_CACHE_NOTIFY_CACHE_TYPE])
+		*cache_mask |= nla_get_u32(tb[SUNRPC_A_CACHE_NOTIFY_CACHE_TYPE]);
+	else
+		*cache_mask = ~0U;
+
+	xlog(D_NETLINK, "sunrpc_notify_handler: cache_mask=%x", *cache_mask);
+	return NL_OK;
+}
+
+static unsigned int cache_sunrpc_nl_drain(void)
+{
+	unsigned int cache_mask = 0;
 	struct nl_cb *cb;
 
 	cb = nl_cb_alloc(NL_CB_DEFAULT);
 	if (!cb)
-		return;
+		return ~0U;
 
-	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, nfsd_nl_notify_handler, NULL);
+	nl_cb_set(cb, NL_CB_SEQ_CHECK, NL_CB_CUSTOM,
+		  nl_seq_check_handler, NULL);
+	nl_cb_set(cb, NL_CB_VALID, NL_CB_CUSTOM, sunrpc_notify_handler,
+		  &cache_mask);
 	nl_recvmsgs(sunrpc_nl_notify_sock, cb);
 	nl_cb_put(cb);
+	return cache_mask;
 }
 
 /*
@@ -2436,16 +2484,19 @@ out_free:
 
 static void cache_sunrpc_nl_process(void)
 {
+	unsigned int cache_mask;
+
 	/* Drain pending sunrpc notifications */
-	cache_sunrpc_nl_drain();
+	cache_mask = cache_sunrpc_nl_drain();
 
 	auth_reload();
 
 	/* Handle any pending ip_map requests */
-	cache_nl_process_ip_map();
+	if (cache_mask & SUNRPC_CACHE_TYPE_IP_MAP)
+		cache_nl_process_ip_map();
 
 	/* Handle any pending unix_gid requests */
-	if (manage_gids)
+	if (manage_gids && (cache_mask & SUNRPC_CACHE_TYPE_UNIX_GID))
 		cache_nl_process_unix_gid();
 }
 

-- 
2.53.0

[PATCH nfs-utils 14/17] nfsd/sunrpc: add cache flush command and attribute enums

[Jeff Layton]

Add NFSD_CMD_CACHE_FLUSH and SUNRPC_CMD_CACHE_FLUSH command enums,
along with their corresponding NFSD_A_CACHE_FLUSH_MASK and
SUNRPC_A_CACHE_FLUSH_MASK attribute enums, to the UAPI netlink headers.

These correspond to new kernel netlink commands that allow userspace to
flush nfsd and sunrpc caches via generic netlink instead of writing to
/proc/net/rpc/*/flush. The mask attribute allows selective per-cache
flushing; omitting it flushes all caches in the family.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/include/nfsd_netlink.h   | 8 ++++++++
 support/include/sunrpc_netlink.h | 8 ++++++++
 2 files changed, 16 insertions(+)

diff --git a/support/include/nfsd_netlink.h b/support/include/nfsd_netlink.h
index e96cc1d23366bce13624084fd476dda65aef140a..08b8a77179f88f596d482d7a2ec5c19ee98b2e77 100644
--- a/support/include/nfsd_netlink.h
+++ b/support/include/nfsd_netlink.h
@@ -198,6 +198,13 @@ enum {
 	NFSD_A_EXPKEY_REQS_MAX = (__NFSD_A_EXPKEY_REQS_MAX - 1)
 };
 
+enum {
+	NFSD_A_CACHE_FLUSH_MASK = 1,
+
+	__NFSD_A_CACHE_FLUSH_MAX,
+	NFSD_A_CACHE_FLUSH_MAX = (__NFSD_A_CACHE_FLUSH_MAX - 1)
+};
+
 enum {
 	NFSD_CMD_RPC_STATUS_GET = 1,
 	NFSD_CMD_THREADS_SET,
@@ -213,6 +220,7 @@ enum {
 	NFSD_CMD_SVC_EXPORT_SET_REQS,
 	NFSD_CMD_EXPKEY_GET_REQS,
 	NFSD_CMD_EXPKEY_SET_REQS,
+	NFSD_CMD_CACHE_FLUSH,
 
 	__NFSD_CMD_MAX,
 	NFSD_CMD_MAX = (__NFSD_CMD_MAX - 1)
diff --git a/support/include/sunrpc_netlink.h b/support/include/sunrpc_netlink.h
index 52c9aa3f19b6a218ec7d0518b58062beff11c774..8deef0fa11d4382a077ea2e8a4c55856376f1731 100644
--- a/support/include/sunrpc_netlink.h
+++ b/support/include/sunrpc_netlink.h
@@ -60,12 +60,20 @@ enum {
 	SUNRPC_A_UNIX_GID_REQS_MAX = (__SUNRPC_A_UNIX_GID_REQS_MAX - 1)
 };
 
+enum {
+	SUNRPC_A_CACHE_FLUSH_MASK = 1,
+
+	__SUNRPC_A_CACHE_FLUSH_MAX,
+	SUNRPC_A_CACHE_FLUSH_MAX = (__SUNRPC_A_CACHE_FLUSH_MAX - 1)
+};
+
 enum {
 	SUNRPC_CMD_CACHE_NOTIFY = 1,
 	SUNRPC_CMD_IP_MAP_GET_REQS,
 	SUNRPC_CMD_IP_MAP_SET_REQS,
 	SUNRPC_CMD_UNIX_GID_GET_REQS,
 	SUNRPC_CMD_UNIX_GID_SET_REQS,
+	SUNRPC_CMD_CACHE_FLUSH,
 
 	__SUNRPC_CMD_MAX,
 	SUNRPC_CMD_MAX = (__SUNRPC_CMD_MAX - 1)

-- 
2.53.0

[PATCH nfs-utils 15/17] exportfs: add netlink support for cache flush with /proc fallback

[Jeff Layton]

Move cache_flush() from support/nfs/cacheio.c to a new file
support/export/cache_flush.c, which has access to libnl.

The new implementation tries netlink first:
- Sends SUNRPC_CMD_CACHE_FLUSH to flush auth.unix.ip and auth.unix.gid
- Sends NFSD_CMD_CACHE_FLUSH to flush nfsd.fh and nfsd.export
- Flushes sunrpc caches before nfsd caches to maintain dependency order

If the sunrpc genl family cannot be resolved (older kernel), falls back
to the original /proc/net/rpc/*/flush write path.

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 support/export/Makefile.am   |   2 +-
 support/export/cache_flush.c | 164 +++++++++++++++++++++++++++++++++++++++++++
 support/nfs/cacheio.c        |  49 +------------
 3 files changed, 166 insertions(+), 49 deletions(-)

diff --git a/support/export/Makefile.am b/support/export/Makefile.am
index ae7ace44112b889f1c461c5473fb1bd42a42f182..33416176c41eb6eadd1880f7c7432b4ca2d8c973 100644
--- a/support/export/Makefile.am
+++ b/support/export/Makefile.am
@@ -12,7 +12,7 @@ EXTRA_DIST	= mount.x
 noinst_LIBRARIES = libexport.a
 libexport_a_SOURCES = client.c export.c hostname.c \
 		      xtab.c mount_clnt.c mount_xdr.c \
-		      cache.c auth.c v4root.c fsloc.c \
+		      cache.c cache_flush.c auth.c v4root.c fsloc.c \
 		      v4clients.c
 libexport_a_CPPFLAGS = $(AM_CPPFLAGS) $(CPPFLAGS) -I$(top_srcdir)/support/reexport \
 		       $(LIBNL3_CFLAGS) $(LIBNLGENL3_CFLAGS)
diff --git a/support/export/cache_flush.c b/support/export/cache_flush.c
new file mode 100644
index 0000000000000000000000000000000000000000..7d7f12b212967e5b3d1a2357de07bc3ba5f0b674
--- /dev/null
+++ b/support/export/cache_flush.c
@@ -0,0 +1,164 @@
+/*
+ * support/export/cache_flush.c
+ *
+ * Flush knfsd caches via netlink with /proc fallback.
+ */
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <errno.h>
+#include <time.h>
+#include <string.h>
+#include <inttypes.h>
+
+#include "nfslib.h"
+#include "xlog.h"
+
+#include <netlink/genl/genl.h>
+#include <netlink/genl/ctrl.h>
+#include <netlink/msg.h>
+
+#ifdef USE_SYSTEM_NFSD_NETLINK_H
+#include <linux/nfsd_netlink.h>
+#else
+#include "nfsd_netlink.h"
+#endif
+
+#ifdef USE_SYSTEM_SUNRPC_NETLINK_H
+#include <linux/sunrpc_netlink.h>
+#else
+#include "sunrpc_netlink.h"
+#endif
+
+static int nl_send_flush(struct nl_sock *sock, int family, int cmd)
+{
+	struct nl_msg *msg;
+	int ret;
+
+	msg = nlmsg_alloc();
+	if (!msg)
+		return -ENOMEM;
+
+	if (!genlmsg_put(msg, NL_AUTO_PID, NL_AUTO_SEQ, family,
+			 0, 0, cmd, 0)) {
+		nlmsg_free(msg);
+		return -ENOMEM;
+	}
+
+	/* No mask attribute = flush all caches in this family */
+	ret = nl_send_auto(sock, msg);
+	if (ret >= 0)
+		ret = nl_wait_for_ack(sock);
+	nlmsg_free(msg);
+	return ret;
+}
+
+static int cache_nl_flush(void)
+{
+	struct nl_sock *sock;
+	int family, ret, val = 1;
+
+	sock = nl_socket_alloc();
+	if (!sock)
+		return -1;
+
+	if (genl_connect(sock)) {
+		nl_socket_free(sock);
+		return -1;
+	}
+
+	setsockopt(nl_socket_get_fd(sock), SOL_NETLINK, NETLINK_EXT_ACK,
+		   &val, sizeof(val));
+
+	/* Flush sunrpc caches first (dependency order) */
+	family = genl_ctrl_resolve(sock, SUNRPC_FAMILY_NAME);
+	if (family < 0) {
+		xlog(D_NETLINK, "sunrpc genl family not found, "
+		     "skipping netlink flush");
+		nl_socket_free(sock);
+		return -1;
+	}
+
+	ret = nl_send_flush(sock, family, SUNRPC_CMD_CACHE_FLUSH);
+	if (ret < 0) {
+		xlog(D_NETLINK, "sunrpc cache flush failed: %d", ret);
+		nl_socket_free(sock);
+		return -1;
+	}
+
+	/* Flush nfsd caches */
+	family = genl_ctrl_resolve(sock, NFSD_FAMILY_NAME);
+	if (family < 0) {
+		xlog(D_NETLINK, "nfsd genl family not found, "
+		     "skipping nfsd cache flush");
+		nl_socket_free(sock);
+		return 0;
+	}
+
+	ret = nl_send_flush(sock, family, NFSD_CMD_CACHE_FLUSH);
+	nl_socket_free(sock);
+	if (ret < 0) {
+		xlog(D_NETLINK, "nfsd cache flush failed: %d", ret);
+		return -1;
+	}
+
+	return 0;
+}
+
+static void cache_proc_flush(void)
+{
+	int c;
+	char stime[32];
+	char path[200];
+	time_t now;
+	/* Note: the order of these caches is important.
+	 * They need to be flushed in dependency order. So
+	 * a cache that references items in another cache,
+	 * as nfsd.fh entries reference items in nfsd.export,
+	 * must be flushed before the cache that it references.
+	 */
+	static char *cachelist[] = {
+		"auth.unix.ip",
+		"auth.unix.gid",
+		"nfsd.fh",
+		"nfsd.export",
+		NULL
+	};
+	now = time(0);
+
+	/* Since v4.16-rc2-3-g3b68e6ee3cbd the timestamp written is ignored.
+	 * It is safest always to flush caches if there is any doubt.
+	 * For earlier kernels, writing the next second from now is
+	 * the best we can do.
+	 */
+	sprintf(stime, "%" PRId64 "\n", (int64_t)now+1);
+	for (c=0; cachelist[c]; c++) {
+		int fd;
+		sprintf(path, "/proc/net/rpc/%s/flush", cachelist[c]);
+		fd = open(path, O_RDWR);
+		if (fd >= 0) {
+			if (write(fd, stime, strlen(stime)) != (ssize_t)strlen(stime)) {
+				xlog_warn("Writing to '%s' failed: errno %d (%s)",
+				path, errno, strerror(errno));
+			}
+			close(fd);
+		}
+	}
+}
+
+void
+cache_flush(void)
+{
+	if (cache_nl_flush() == 0) {
+		xlog(D_NETLINK, "cache flush via netlink succeeded");
+		return;
+	}
+	/* Fallback: /proc path */
+	cache_proc_flush();
+}
diff --git a/support/nfs/cacheio.c b/support/nfs/cacheio.c
index bd4da0e5ad93f963bf8090062541f34ededd1e03..95eeabbdb978cd227da9877b81e5ed2d65dedaf9 100644
--- a/support/nfs/cacheio.c
+++ b/support/nfs/cacheio.c
@@ -203,51 +203,4 @@ int qword_get_uint(char **bpp, unsigned int *anint)
 	return 0;
 }
 
-/* flush the kNFSd caches.
- * Set the flush time to the mtime of the etab state file or
- * if force, to now.
- * the caches to flush are:
- *  auth.unix.ip nfsd.export nfsd.fh
- */
-
-void
-cache_flush(void)
-{
-	int c;
-	char stime[32];
-	char path[200];
-	time_t now;
-	/* Note: the order of these caches is important.
-	 * They need to be flushed in dependancy order. So
-	 * a cache that references items in another cache,
-	 * as nfsd.fh entries reference items in nfsd.export,
-	 * must be flushed before the cache that it references.
-	 */
-	static char *cachelist[] = {
-		"auth.unix.ip",
-		"auth.unix.gid",
-		"nfsd.fh",
-		"nfsd.export",
-		NULL
-	};
-	now = time(0);
-
-	/* Since v4.16-rc2-3-g3b68e6ee3cbd the timestamp written is ignored.
-	 * It is safest always to flush caches if there is any doubt.
-	 * For earlier kernels, writing the next second from now is
-	 * the best we can do.
-	 */
-	sprintf(stime, "%" PRId64 "\n", (int64_t)now+1);
-	for (c=0; cachelist[c]; c++) {
-		int fd;
-		sprintf(path, "/proc/net/rpc/%s/flush", cachelist[c]);
-		fd = open(path, O_RDWR);
-		if (fd >= 0) {
-			if (write(fd, stime, strlen(stime)) != (ssize_t)strlen(stime)) {
-				xlog_warn("Writing to '%s' failed: errno %d (%s)",
-				path, errno, strerror(errno));
-			}
-			close(fd);
-		}
-	}
-}
+/* cache_flush() has moved to support/export/cache.c for netlink support */

-- 
2.53.0

[PATCH nfs-utils 16/17] exportfs: use netlink to probe kernel support, skip export_test

[Jeff Layton]

Refactor can_test() to first try resolving the sunrpc genl family via
netlink. If successful, the kernel supports the new netlink-based cache
interfaces and we can skip the advisory export_test() probe -- mountd
and exportd will validate exports at cache-fill time via check_export().

can_test() now returns:
  2 = netlink available (skip export_test)
  1 = /proc available (keep export_test as before)
  0 = neither available

Update validate_export() to check this return value and skip the
export_test() calls when netlink is available.

Link exportfs against libnl3/libnl-genl-3 (already a mandatory build
dependency for nfs-utils).

Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 utils/exportfs/Makefile.am |  6 ++++--
 utils/exportfs/exportfs.c  | 42 +++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 45 insertions(+), 3 deletions(-)

diff --git a/utils/exportfs/Makefile.am b/utils/exportfs/Makefile.am
index 7f8ce9faf2b469c8560f5fbee40b81e3443eab78..8db5fdae1741e1c79639bbdfd5d63f8571363e63 100644
--- a/utils/exportfs/Makefile.am
+++ b/utils/exportfs/Makefile.am
@@ -11,8 +11,10 @@ exportfs_LDADD = ../../support/export/libexport.a \
 	       	 ../../support/nfs/libnfs.la \
 		 ../../support/misc/libmisc.a \
 		 ../../support/reexport/libreexport.a \
-		 $(LIBWRAP) $(LIBNSL) $(LIBPTHREAD)
+		 $(LIBWRAP) $(LIBNSL) $(LIBPTHREAD) \
+		 $(LIBNL3_LIBS) $(LIBNLGENL3_LIBS)
 
-exportfs_CPPFLAGS = $(AM_CPPFLAGS) $(CPPFLAGS) -I$(top_srcdir)/support/reexport
+exportfs_CPPFLAGS = $(AM_CPPFLAGS) $(CPPFLAGS) -I$(top_srcdir)/support/reexport \
+		    $(LIBNL3_CFLAGS) $(LIBNLGENL3_CFLAGS)
 
 MAINTAINERCLEANFILES = Makefile.in
diff --git a/utils/exportfs/exportfs.c b/utils/exportfs/exportfs.c
index 748c38e3e966703c7a870a6aa9589d32708edcd2..04753fa169f97c6b07893613197739ff36e0d09b 100644
--- a/utils/exportfs/exportfs.c
+++ b/utils/exportfs/exportfs.c
@@ -40,6 +40,15 @@
 #include "conffile.h"
 #include "reexport.h"
 
+#include <netlink/genl/genl.h>
+#include <netlink/genl/ctrl.h>
+
+#ifdef USE_SYSTEM_SUNRPC_NETLINK_H
+#include <linux/sunrpc_netlink.h>
+#else
+#include "sunrpc_netlink.h"
+#endif
+
 static void	export_all(int verbose);
 static void	exportfs(char *arg, char *options, int verbose);
 static void	unexportfs(char *arg, int verbose);
@@ -476,13 +485,34 @@ unexportfs(char *arg, int verbose)
 		xlog(L_ERROR, "Invalid export syntax: %s", arg);
 }
 
+/* Return values:
+ *   2 = netlink available (skip export_test)
+ *   1 = /proc available (keep export_test)
+ *   0 = neither available
+ */
 static int can_test(void)
 {
+	struct nl_sock *sock;
+	int family;
 	char buf[1024] = { 0 };
 	int fd;
 	int n;
 	size_t bufsiz = sizeof(buf);
 
+	/* Try netlink first: resolve sunrpc genl family */
+	sock = nl_socket_alloc();
+	if (sock) {
+		if (genl_connect(sock) == 0) {
+			family = genl_ctrl_resolve(sock, SUNRPC_FAMILY_NAME);
+			nl_socket_free(sock);
+			if (family >= 0)
+				return 2;
+		} else {
+			nl_socket_free(sock);
+		}
+	}
+
+	/* Fallback: /proc probe */
 	fd = open("/proc/net/rpc/auth.unix.ip/channel", O_WRONLY);
 	if (fd < 0)
 		return 0;
@@ -522,6 +552,7 @@ validate_export(nfs_export *exp)
 	char *path = exportent_realpath(&exp->m_export);
 	struct statfs stf;
 	int fs_has_fsid = 0;
+	int test_result;
 
 	if (stat(path, &stb) < 0) {
 		xlog(L_ERROR, "Failed to stat %s: %m", path);
@@ -532,7 +563,16 @@ validate_export(nfs_export *exp)
 			"Remote access will fail", path);
 		return;
 	}
-	if (!can_test())
+
+	test_result = can_test();
+	if (!test_result)
+		return;
+
+	/*
+	 * When netlink is available, skip the export_test() probe.
+	 * mountd/exportd will validate exports at cache-fill time.
+	 */
+	if (test_result == 2)
 		return;
 
 	if (!statfs(path, &stf) &&

-- 
2.53.0

[PATCH nfs-utils 17/17] mountd/exportd/exportfs: add --no-netlink option to disable netlink

[Jeff Layton]

Add a --no-netlink command-line option (short: -L) and no-netlink=
config file setting that disables netlink support, forcing the use of
/proc interfaces instead.

For mountd and exportd, the setting lives in their respective [mountd]
and [exportd] sections of nfs.conf. For exportfs, it's in the
[exportfs] section.

This is useful for debugging, testing the /proc fallback path, or
working around kernel netlink issues.
---
 support/export/cache.c       |  3 ++-
 support/export/cache_flush.c |  4 ++-
 utils/exportd/exportd.c      | 10 ++++++--
 utils/exportd/exportd.man    | 12 +++++++--
 utils/exportfs/exportfs.c    | 13 +++++++++-
 utils/exportfs/exportfs.man  | 58 ++++++++++++++++++++++++++++++++++++++------
 utils/mountd/mountd.c        |  9 ++++++-
 utils/mountd/mountd.man      |  9 +++++++
 8 files changed, 102 insertions(+), 16 deletions(-)

diff --git a/support/export/cache.c b/support/export/cache.c
index 5a2c5760cb5410845971ba831a9ae779d17a6d87..2f128d7db7bd63d86530f0c4003af58327db2c70 100644
--- a/support/export/cache.c
+++ b/support/export/cache.c
@@ -111,6 +111,7 @@ static bool path_lookup_error(int err)
 
 extern int use_ipaddr;
 extern int manage_gids;
+extern int no_netlink;
 
 static void auth_unix_ip(int f)
 {
@@ -3064,7 +3065,7 @@ void cache_open(void)
 {
 	int i;
 
-	if (cache_nfsd_nl_open() == 0) {
+	if (!no_netlink && cache_nfsd_nl_open() == 0) {
 		if (cache_sunrpc_nl_open() == 0) {
 			/*
 			 * Check for pending requests, in case any
diff --git a/support/export/cache_flush.c b/support/export/cache_flush.c
index 7d7f12b212967e5b3d1a2357de07bc3ba5f0b674..ed7b964f9d5372f4accba21254ee9c5f40ffd44d 100644
--- a/support/export/cache_flush.c
+++ b/support/export/cache_flush.c
@@ -20,6 +20,8 @@
 #include "nfslib.h"
 #include "xlog.h"
 
+extern int no_netlink;
+
 #include <netlink/genl/genl.h>
 #include <netlink/genl/ctrl.h>
 #include <netlink/msg.h>
@@ -155,7 +157,7 @@ static void cache_proc_flush(void)
 void
 cache_flush(void)
 {
-	if (cache_nl_flush() == 0) {
+	if (!no_netlink && cache_nl_flush() == 0) {
 		xlog(D_NETLINK, "cache flush via netlink succeeded");
 		return;
 	}
diff --git a/utils/exportd/exportd.c b/utils/exportd/exportd.c
index a2e370ac506f56d0feab306bd252c32ef58ba009..a08aaaccbc2f2ec8504c53bbf07daf1ac2be0c32 100644
--- a/utils/exportd/exportd.c
+++ b/utils/exportd/exportd.c
@@ -32,6 +32,7 @@ static int num_threads = 1;
 #define MAX_THREADS 64
 
 int manage_gids;
+int no_netlink;
 int use_ipaddr = -1;
 
 static struct option longopts[] =
@@ -40,13 +41,14 @@ static struct option longopts[] =
 	{ "debug", 1, 0, 'd' },
 	{ "help", 0, 0, 'h' },
 	{ "manage-gids", 0, 0, 'g' },
+	{ "no-netlink", 0, 0, 'L' },
 	{ "num-threads", 1, 0, 't' },
 	{ "log-auth", 0, 0, 'l' },
 	{ "cache-use-ipaddr", 0, 0, 'i' },
 	{ "ttl", 0, 0, 'T' },
 	{ NULL, 0, 0, 0 }
 };
-static char shortopts[] = "d:fghs:t:liT:";
+static char shortopts[] = "d:fghs:t:liLT:";
 
 /*
  * Signal handlers.
@@ -109,7 +111,7 @@ usage(const char *prog, int n)
 		"Usage: %s [-f|--foreground] [-h|--help] [-d kind|--debug kind]\n"
 "	[-g|--manage-gids] [-l|--log-auth] [-i|--cache-use-ipaddr] [-T|--ttl ttl]\n"
 "	[-s|--state-directory-path path]\n"
-"	[-t num|--num-threads=num]\n", prog);
+"	[-t num|--num-threads=num] [-L|--no-netlink]\n", prog);
 	exit(n);
 }
 
@@ -124,6 +126,7 @@ read_exportd_conf(char *progname, char **argv)
 	xlog_set_debug(progname);
 
 	manage_gids = conf_get_bool("exportd", "manage-gids", manage_gids);
+	no_netlink = conf_get_bool("exportd", "no-netlink", no_netlink);
 	num_threads = conf_get_num("exportd", "threads", num_threads);
 	if (conf_get_bool("mountd", "cache-use-ipaddr", 0))
 		use_ipaddr = 2;
@@ -171,6 +174,9 @@ main(int argc, char **argv)
 		case 'g':
 			manage_gids = 1;
 			break;
+		case 'L':
+			no_netlink = 1;
+			break;
 		case 'h':
 			usage(progname, 0);
 			break;
diff --git a/utils/exportd/exportd.man b/utils/exportd/exportd.man
index fae434b5f03bfb5a252f1e5c6d7fc8fc2a3f5567..d024868c6471c60f6804f427317a2627cbddb0af 100644
--- a/utils/exportd/exportd.man
+++ b/utils/exportd/exportd.man
@@ -106,6 +106,13 @@ the server. Note that the 'primary' group id is not affected so a
 .B newgroup
 command on the client will still be effective.  This function requires
 a Linux Kernel with version at least 2.6.21.
+.TP
+.B \-L " or " \-\-no-netlink
+Disable the use of netlink for kernel communication and force the use
+of the legacy
+.I /proc/net/rpc
+interfaces instead.  This can be useful for debugging or working around
+kernel netlink issues.
 .SH CONFIGURATION FILE
 Many of the options that can be set on the command line can also be
 controlled through values set in the
@@ -120,8 +127,9 @@ Values recognized in the
 section include 
 .B cache\-use\-ipaddr ,
 .BR ttl ,
-.BR manage-gids ", and"
-.B debug 
+.BR manage-gids ,
+.BR no\-netlink ", and"
+.B debug
 which each have the same effect as the option with the same name.
 .SH FILES
 .TP 2.5i
diff --git a/utils/exportfs/exportfs.c b/utils/exportfs/exportfs.c
index 04753fa169f97c6b07893613197739ff36e0d09b..9eedb750e316a2ef42bb541d4df4925deeee4874 100644
--- a/utils/exportfs/exportfs.c
+++ b/utils/exportfs/exportfs.c
@@ -49,6 +49,8 @@
 #include "sunrpc_netlink.h"
 #endif
 
+int no_netlink;
+
 static void	export_all(int verbose);
 static void	exportfs(char *arg, char *options, int verbose);
 static void	unexportfs(char *arg, int verbose);
@@ -109,8 +111,11 @@ read_exportfs_conf(char **argv)
 	xlog_set_debug("exportfs");
 
 	/* NOTE: following uses "mountd" section of nfs.conf !!!! */
+	no_netlink = conf_get_bool("mountd", "no-netlink", no_netlink);
 	s = conf_get_str("mountd", "state-directory-path");
 	/* Also look in the exportd section */
+	if (!no_netlink)
+		no_netlink = conf_get_bool("exportd", "no-netlink", no_netlink);
 	if (s == NULL)
 		s = conf_get_str("exportd", "state-directory-path");
 	if (s && !state_setup_basedir(argv[0], s))
@@ -145,7 +150,7 @@ main(int argc, char **argv)
 
 	nfsd_path_init();
 
-	while ((c = getopt(argc, argv, "ad:fhio:ruvs")) != EOF) {
+	while ((c = getopt(argc, argv, "ad:fhiLo:ruvs")) != EOF) {
 		switch(c) {
 		case 'a':
 			f_all = 1;
@@ -162,6 +167,9 @@ main(int argc, char **argv)
 		case 'i':
 			f_ignore = 1;
 			break;
+		case 'L':
+			no_netlink = 1;
+			break;
 		case 'o':
 			options = optarg;
 			break;
@@ -500,6 +508,8 @@ static int can_test(void)
 	size_t bufsiz = sizeof(buf);
 
 	/* Try netlink first: resolve sunrpc genl family */
+	if (no_netlink)
+		goto try_proc;
 	sock = nl_socket_alloc();
 	if (sock) {
 		if (genl_connect(sock) == 0) {
@@ -513,6 +523,7 @@ static int can_test(void)
 	}
 
 	/* Fallback: /proc probe */
+try_proc:
 	fd = open("/proc/net/rpc/auth.unix.ip/channel", O_WRONLY);
 	if (fd < 0)
 		return 0;
diff --git a/utils/exportfs/exportfs.man b/utils/exportfs/exportfs.man
index af0e5571cef83d4f3de6915608b4871690a8853a..3737ee81ab275aa65e942ec1602f33a7abbfc80e 100644
--- a/utils/exportfs/exportfs.man
+++ b/utils/exportfs/exportfs.man
@@ -53,14 +53,41 @@ by using the
 command.
 .PP
 .B exportfs
-does not give any information to the kernel directly, but provides it
-only to
-.B rpc.mountd
-through the
+does not communicate with the kernel directly.
+It writes export information to
 .I /var/lib/nfs/etab
-file.
+and relies on its partner programs
+.B rpc.mountd
+and
+.B nfsv4.exportd
+to manage kernel communication.
+These daemons work in one of two modes: a netlink mode and a
+.I /proc
+mode.
+.PP
+In the netlink mode, available on sufficiently recent kernels,
 .B rpc.mountd
-then manages kernel requests for information about exports, as needed.
+(or
+.BR nfsv4.exportd )
+communicates with the kernel via generic netlink sockets.
+The kernel sends multicast notifications when cache entries need to be
+resolved, and the daemon responds with the appropriate export
+information.
+Cache flushing (via
+.BR "exportfs \-f" )
+is also performed over netlink.
+This mode can be disabled with the
+.B \-L
+option.
+.PP
+In the
+.I /proc
+mode, used when netlink is unavailable,
+.B rpc.mountd
+manages kernel requests for information about exports
+via the
+.I /proc/net/rpc
+channel files.
 .SH OPTIONS
 .TP
 .B \-d kind " or " \-\-debug kind
@@ -123,6 +150,12 @@ options.
 .TP
 .B -s
 Display the current export list suitable for /etc/exports.
+.TP
+.B -L
+Disable the use of netlink for kernel communication and force the use
+of the legacy
+.I /proc
+interfaces for cache flushing and export validation.
 
 .SH CONFIGURATION FILE
 The
@@ -142,11 +175,20 @@ When a list is given, the members should be comma-separated.
 .B exportfs
 will also recognize the
 .B state-directory-path
-value from both the 
+and
+.B no\-netlink
+values from both the
 .B [mountd]
 section and the
 .B [exportd]
-section
+section.
+When
+.B no\-netlink
+is set,
+.B exportfs
+will skip the netlink probe and use the legacy
+.I /proc
+interfaces for cache flushing and export validation
 
 .SH DISCUSSION
 .SS Exporting Directories
diff --git a/utils/mountd/mountd.c b/utils/mountd/mountd.c
index 6e6777cd1daa0227f3ff81f826c1cbc8627b4a8a..92d8c4690efc48fcfa12d9618cac9172c4752f4f 100644
--- a/utils/mountd/mountd.c
+++ b/utils/mountd/mountd.c
@@ -41,6 +41,7 @@ static struct nfs_fh_len *get_rootfh(struct svc_req *, dirpath *, nfs_export **,
 
 int reverse_resolve = 0;
 int manage_gids;
+int no_netlink;
 int apply_root_cred;
 int use_ipaddr = -1;
 
@@ -72,6 +73,7 @@ static struct option longopts[] =
 	{ "num-threads", 1, 0, 't' },
 	{ "reverse-lookup", 0, 0, 'r' },
 	{ "manage-gids", 0, 0, 'g' },
+	{ "no-netlink", 0, 0, 'L' },
 	{ "no-udp", 0, 0, 'u' },
 	{ "log-auth", 0, 0, 'l'},
 	{ "cache-use-ipaddr", 0, 0, 'i'},
@@ -667,6 +669,7 @@ read_mountd_conf(char **argv)
 
 	xlog_set_debug("mountd");
 	manage_gids = conf_get_bool("mountd", "manage-gids", manage_gids);
+	no_netlink = conf_get_bool("mountd", "no-netlink", no_netlink);
 	descriptors = conf_get_num("mountd", "descriptors", descriptors);
 	port = conf_get_num("mountd", "port", port);
 	num_threads = conf_get_num("mountd", "threads", num_threads);
@@ -734,6 +737,9 @@ main(int argc, char **argv)
 		case 'g':
 			manage_gids = 1;
 			break;
+		case 'L':
+			no_netlink = 1;
+			break;
 		case 'o':
 			descriptors = atoi(optarg);
 			if (descriptors <= 0) {
@@ -951,6 +957,7 @@ usage(const char *prog, int n)
 "	[-N version|--no-nfs-version version] [-n|--no-tcp]\n"
 "	[-H prog |--ha-callout prog] [-r |--reverse-lookup]\n"
 "	[-s|--state-directory-path path] [-g|--manage-gids]\n"
-"	[-t num|--num-threads=num] [-u|--no-udp]\n", prog);
+"	[-t num|--num-threads=num] [-u|--no-udp]\n"
+"	[-L|--no-netlink]\n", prog);
 	exit(n);
 }
diff --git a/utils/mountd/mountd.man b/utils/mountd/mountd.man
index 2fa396c3288f37b1afa247b54a6166ca4f1b5e06..8bec38db131d9f70d1e04a000133023cca955fe1 100644
--- a/utils/mountd/mountd.man
+++ b/utils/mountd/mountd.man
@@ -284,6 +284,14 @@ the server. Note that the 'primary' group id is not affected so a
 command on the client will still be effective.  This function requires
 a Linux Kernel with version at least 2.6.21.
 
+.TP
+.B \-L " or " \-\-no-netlink
+Disable the use of netlink for kernel communication and force the use
+of the legacy
+.I /proc/net/rpc
+interfaces instead.  This can be useful for debugging or working around
+kernel netlink issues.
+
 .SH CONFIGURATION FILE
 Many of the options that can be set on the command line can also be
 controlled through values set in the
@@ -297,6 +305,7 @@ Values recognized in the
 .B [mountd]
 section include
 .BR manage-gids ,
+.BR no\-netlink ,
 .BR cache\-use\-ipaddr ,
 .BR descriptors ,
 .BR port ,

-- 
2.53.0

Re: [PATCH nfs-utils 17/17] mountd/exportd/exportfs: add --no-netlink option to disable netlink

[Jeff Layton]

On Mon, 2026-03-16 at 11:16 -0400, Jeff Layton wrote:
> Add a --no-netlink command-line option (short: -L) and no-netlink=
> config file setting that disables netlink support, forcing the use of
> /proc interfaces instead.
> 
> For mountd and exportd, the setting lives in their respective [mountd]
> and [exportd] sections of nfs.conf. For exportfs, it's in the
> [exportfs] section.

Correction: exportfs doesn't have a section. It will look for the
setting in the mountd and exportd sections. I'll fix up the commit log
in my tree.

> 
> This is useful for debugging, testing the /proc fallback path, or
> working around kernel netlink issues.
> ---
>  support/export/cache.c       |  3 ++-
>  support/export/cache_flush.c |  4 ++-
>  utils/exportd/exportd.c      | 10 ++++++--
>  utils/exportd/exportd.man    | 12 +++++++--
>  utils/exportfs/exportfs.c    | 13 +++++++++-
>  utils/exportfs/exportfs.man  | 58 ++++++++++++++++++++++++++++++++++++++------
>  utils/mountd/mountd.c        |  9 ++++++-
>  utils/mountd/mountd.man      |  9 +++++++
>  8 files changed, 102 insertions(+), 16 deletions(-)
> 
> diff --git a/support/export/cache.c b/support/export/cache.c
> index 5a2c5760cb5410845971ba831a9ae779d17a6d87..2f128d7db7bd63d86530f0c4003af58327db2c70 100644
> --- a/support/export/cache.c
> +++ b/support/export/cache.c
> @@ -111,6 +111,7 @@ static bool path_lookup_error(int err)
>  
>  extern int use_ipaddr;
>  extern int manage_gids;
> +extern int no_netlink;
>  
>  static void auth_unix_ip(int f)
>  {
> @@ -3064,7 +3065,7 @@ void cache_open(void)
>  {
>  	int i;
>  
> -	if (cache_nfsd_nl_open() == 0) {
> +	if (!no_netlink && cache_nfsd_nl_open() == 0) {
>  		if (cache_sunrpc_nl_open() == 0) {
>  			/*
>  			 * Check for pending requests, in case any
> diff --git a/support/export/cache_flush.c b/support/export/cache_flush.c
> index 7d7f12b212967e5b3d1a2357de07bc3ba5f0b674..ed7b964f9d5372f4accba21254ee9c5f40ffd44d 100644
> --- a/support/export/cache_flush.c
> +++ b/support/export/cache_flush.c
> @@ -20,6 +20,8 @@
>  #include "nfslib.h"
>  #include "xlog.h"
>  
> +extern int no_netlink;
> +
>  #include <netlink/genl/genl.h>
>  #include <netlink/genl/ctrl.h>
>  #include <netlink/msg.h>
> @@ -155,7 +157,7 @@ static void cache_proc_flush(void)
>  void
>  cache_flush(void)
>  {
> -	if (cache_nl_flush() == 0) {
> +	if (!no_netlink && cache_nl_flush() == 0) {
>  		xlog(D_NETLINK, "cache flush via netlink succeeded");
>  		return;
>  	}
> diff --git a/utils/exportd/exportd.c b/utils/exportd/exportd.c
> index a2e370ac506f56d0feab306bd252c32ef58ba009..a08aaaccbc2f2ec8504c53bbf07daf1ac2be0c32 100644
> --- a/utils/exportd/exportd.c
> +++ b/utils/exportd/exportd.c
> @@ -32,6 +32,7 @@ static int num_threads = 1;
>  #define MAX_THREADS 64
>  
>  int manage_gids;
> +int no_netlink;
>  int use_ipaddr = -1;
>  
>  static struct option longopts[] =
> @@ -40,13 +41,14 @@ static struct option longopts[] =
>  	{ "debug", 1, 0, 'd' },
>  	{ "help", 0, 0, 'h' },
>  	{ "manage-gids", 0, 0, 'g' },
> +	{ "no-netlink", 0, 0, 'L' },
>  	{ "num-threads", 1, 0, 't' },
>  	{ "log-auth", 0, 0, 'l' },
>  	{ "cache-use-ipaddr", 0, 0, 'i' },
>  	{ "ttl", 0, 0, 'T' },
>  	{ NULL, 0, 0, 0 }
>  };
> -static char shortopts[] = "d:fghs:t:liT:";
> +static char shortopts[] = "d:fghs:t:liLT:";
>  
>  /*
>   * Signal handlers.
> @@ -109,7 +111,7 @@ usage(const char *prog, int n)
>  		"Usage: %s [-f|--foreground] [-h|--help] [-d kind|--debug kind]\n"
>  "	[-g|--manage-gids] [-l|--log-auth] [-i|--cache-use-ipaddr] [-T|--ttl ttl]\n"
>  "	[-s|--state-directory-path path]\n"
> -"	[-t num|--num-threads=num]\n", prog);
> +"	[-t num|--num-threads=num] [-L|--no-netlink]\n", prog);
>  	exit(n);
>  }
>  
> @@ -124,6 +126,7 @@ read_exportd_conf(char *progname, char **argv)
>  	xlog_set_debug(progname);
>  
>  	manage_gids = conf_get_bool("exportd", "manage-gids", manage_gids);
> +	no_netlink = conf_get_bool("exportd", "no-netlink", no_netlink);
>  	num_threads = conf_get_num("exportd", "threads", num_threads);
>  	if (conf_get_bool("mountd", "cache-use-ipaddr", 0))
>  		use_ipaddr = 2;
> @@ -171,6 +174,9 @@ main(int argc, char **argv)
>  		case 'g':
>  			manage_gids = 1;
>  			break;
> +		case 'L':
> +			no_netlink = 1;
> +			break;
>  		case 'h':
>  			usage(progname, 0);
>  			break;
> diff --git a/utils/exportd/exportd.man b/utils/exportd/exportd.man
> index fae434b5f03bfb5a252f1e5c6d7fc8fc2a3f5567..d024868c6471c60f6804f427317a2627cbddb0af 100644
> --- a/utils/exportd/exportd.man
> +++ b/utils/exportd/exportd.man
> @@ -106,6 +106,13 @@ the server. Note that the 'primary' group id is not affected so a
>  .B newgroup
>  command on the client will still be effective.  This function requires
>  a Linux Kernel with version at least 2.6.21.
> +.TP
> +.B \-L " or " \-\-no-netlink
> +Disable the use of netlink for kernel communication and force the use
> +of the legacy
> +.I /proc/net/rpc
> +interfaces instead.  This can be useful for debugging or working around
> +kernel netlink issues.
>  .SH CONFIGURATION FILE
>  Many of the options that can be set on the command line can also be
>  controlled through values set in the
> @@ -120,8 +127,9 @@ Values recognized in the
>  section include 
>  .B cache\-use\-ipaddr ,
>  .BR ttl ,
> -.BR manage-gids ", and"
> -.B debug 
> +.BR manage-gids ,
> +.BR no\-netlink ", and"
> +.B debug
>  which each have the same effect as the option with the same name.
>  .SH FILES
>  .TP 2.5i
> diff --git a/utils/exportfs/exportfs.c b/utils/exportfs/exportfs.c
> index 04753fa169f97c6b07893613197739ff36e0d09b..9eedb750e316a2ef42bb541d4df4925deeee4874 100644
> --- a/utils/exportfs/exportfs.c
> +++ b/utils/exportfs/exportfs.c
> @@ -49,6 +49,8 @@
>  #include "sunrpc_netlink.h"
>  #endif
>  
> +int no_netlink;
> +
>  static void	export_all(int verbose);
>  static void	exportfs(char *arg, char *options, int verbose);
>  static void	unexportfs(char *arg, int verbose);
> @@ -109,8 +111,11 @@ read_exportfs_conf(char **argv)
>  	xlog_set_debug("exportfs");
>  
>  	/* NOTE: following uses "mountd" section of nfs.conf !!!! */
> +	no_netlink = conf_get_bool("mountd", "no-netlink", no_netlink);
>  	s = conf_get_str("mountd", "state-directory-path");
>  	/* Also look in the exportd section */
> +	if (!no_netlink)
> +		no_netlink = conf_get_bool("exportd", "no-netlink", no_netlink);
>  	if (s == NULL)
>  		s = conf_get_str("exportd", "state-directory-path");
>  	if (s && !state_setup_basedir(argv[0], s))
> @@ -145,7 +150,7 @@ main(int argc, char **argv)
>  
>  	nfsd_path_init();
>  
> -	while ((c = getopt(argc, argv, "ad:fhio:ruvs")) != EOF) {
> +	while ((c = getopt(argc, argv, "ad:fhiLo:ruvs")) != EOF) {
>  		switch(c) {
>  		case 'a':
>  			f_all = 1;
> @@ -162,6 +167,9 @@ main(int argc, char **argv)
>  		case 'i':
>  			f_ignore = 1;
>  			break;
> +		case 'L':
> +			no_netlink = 1;
> +			break;
>  		case 'o':
>  			options = optarg;
>  			break;
> @@ -500,6 +508,8 @@ static int can_test(void)
>  	size_t bufsiz = sizeof(buf);
>  
>  	/* Try netlink first: resolve sunrpc genl family */
> +	if (no_netlink)
> +		goto try_proc;
>  	sock = nl_socket_alloc();
>  	if (sock) {
>  		if (genl_connect(sock) == 0) {
> @@ -513,6 +523,7 @@ static int can_test(void)
>  	}
>  
>  	/* Fallback: /proc probe */
> +try_proc:
>  	fd = open("/proc/net/rpc/auth.unix.ip/channel", O_WRONLY);
>  	if (fd < 0)
>  		return 0;
> diff --git a/utils/exportfs/exportfs.man b/utils/exportfs/exportfs.man
> index af0e5571cef83d4f3de6915608b4871690a8853a..3737ee81ab275aa65e942ec1602f33a7abbfc80e 100644
> --- a/utils/exportfs/exportfs.man
> +++ b/utils/exportfs/exportfs.man
> @@ -53,14 +53,41 @@ by using the
>  command.
>  .PP
>  .B exportfs
> -does not give any information to the kernel directly, but provides it
> -only to
> -.B rpc.mountd
> -through the
> +does not communicate with the kernel directly.
> +It writes export information to
>  .I /var/lib/nfs/etab
> -file.
> +and relies on its partner programs
> +.B rpc.mountd
> +and
> +.B nfsv4.exportd
> +to manage kernel communication.
> +These daemons work in one of two modes: a netlink mode and a
> +.I /proc
> +mode.
> +.PP
> +In the netlink mode, available on sufficiently recent kernels,
>  .B rpc.mountd
> -then manages kernel requests for information about exports, as needed.
> +(or
> +.BR nfsv4.exportd )
> +communicates with the kernel via generic netlink sockets.
> +The kernel sends multicast notifications when cache entries need to be
> +resolved, and the daemon responds with the appropriate export
> +information.
> +Cache flushing (via
> +.BR "exportfs \-f" )
> +is also performed over netlink.
> +This mode can be disabled with the
> +.B \-L
> +option.
> +.PP
> +In the
> +.I /proc
> +mode, used when netlink is unavailable,
> +.B rpc.mountd
> +manages kernel requests for information about exports
> +via the
> +.I /proc/net/rpc
> +channel files.
>  .SH OPTIONS
>  .TP
>  .B \-d kind " or " \-\-debug kind
> @@ -123,6 +150,12 @@ options.
>  .TP
>  .B -s
>  Display the current export list suitable for /etc/exports.
> +.TP
> +.B -L
> +Disable the use of netlink for kernel communication and force the use
> +of the legacy
> +.I /proc
> +interfaces for cache flushing and export validation.
>  
>  .SH CONFIGURATION FILE
>  The
> @@ -142,11 +175,20 @@ When a list is given, the members should be comma-separated.
>  .B exportfs
>  will also recognize the
>  .B state-directory-path
> -value from both the 
> +and
> +.B no\-netlink
> +values from both the
>  .B [mountd]
>  section and the
>  .B [exportd]
> -section
> +section.
> +When
> +.B no\-netlink
> +is set,
> +.B exportfs
> +will skip the netlink probe and use the legacy
> +.I /proc
> +interfaces for cache flushing and export validation
>  
>  .SH DISCUSSION
>  .SS Exporting Directories
> diff --git a/utils/mountd/mountd.c b/utils/mountd/mountd.c
> index 6e6777cd1daa0227f3ff81f826c1cbc8627b4a8a..92d8c4690efc48fcfa12d9618cac9172c4752f4f 100644
> --- a/utils/mountd/mountd.c
> +++ b/utils/mountd/mountd.c
> @@ -41,6 +41,7 @@ static struct nfs_fh_len *get_rootfh(struct svc_req *, dirpath *, nfs_export **,
>  
>  int reverse_resolve = 0;
>  int manage_gids;
> +int no_netlink;
>  int apply_root_cred;
>  int use_ipaddr = -1;
>  
> @@ -72,6 +73,7 @@ static struct option longopts[] =
>  	{ "num-threads", 1, 0, 't' },
>  	{ "reverse-lookup", 0, 0, 'r' },
>  	{ "manage-gids", 0, 0, 'g' },
> +	{ "no-netlink", 0, 0, 'L' },
>  	{ "no-udp", 0, 0, 'u' },
>  	{ "log-auth", 0, 0, 'l'},
>  	{ "cache-use-ipaddr", 0, 0, 'i'},
> @@ -667,6 +669,7 @@ read_mountd_conf(char **argv)
>  
>  	xlog_set_debug("mountd");
>  	manage_gids = conf_get_bool("mountd", "manage-gids", manage_gids);
> +	no_netlink = conf_get_bool("mountd", "no-netlink", no_netlink);
>  	descriptors = conf_get_num("mountd", "descriptors", descriptors);
>  	port = conf_get_num("mountd", "port", port);
>  	num_threads = conf_get_num("mountd", "threads", num_threads);
> @@ -734,6 +737,9 @@ main(int argc, char **argv)
>  		case 'g':
>  			manage_gids = 1;
>  			break;
> +		case 'L':
> +			no_netlink = 1;
> +			break;
>  		case 'o':
>  			descriptors = atoi(optarg);
>  			if (descriptors <= 0) {
> @@ -951,6 +957,7 @@ usage(const char *prog, int n)
>  "	[-N version|--no-nfs-version version] [-n|--no-tcp]\n"
>  "	[-H prog |--ha-callout prog] [-r |--reverse-lookup]\n"
>  "	[-s|--state-directory-path path] [-g|--manage-gids]\n"
> -"	[-t num|--num-threads=num] [-u|--no-udp]\n", prog);
> +"	[-t num|--num-threads=num] [-u|--no-udp]\n"
> +"	[-L|--no-netlink]\n", prog);
>  	exit(n);
>  }
> diff --git a/utils/mountd/mountd.man b/utils/mountd/mountd.man
> index 2fa396c3288f37b1afa247b54a6166ca4f1b5e06..8bec38db131d9f70d1e04a000133023cca955fe1 100644
> --- a/utils/mountd/mountd.man
> +++ b/utils/mountd/mountd.man
> @@ -284,6 +284,14 @@ the server. Note that the 'primary' group id is not affected so a
>  command on the client will still be effective.  This function requires
>  a Linux Kernel with version at least 2.6.21.
>  
> +.TP
> +.B \-L " or " \-\-no-netlink
> +Disable the use of netlink for kernel communication and force the use
> +of the legacy
> +.I /proc/net/rpc
> +interfaces instead.  This can be useful for debugging or working around
> +kernel netlink issues.
> +
>  .SH CONFIGURATION FILE
>  Many of the options that can be set on the command line can also be
>  controlled through values set in the
> @@ -297,6 +305,7 @@ Values recognized in the
>  .B [mountd]
>  section include
>  .BR manage-gids ,
> +.BR no\-netlink ,
>  .BR cache\-use\-ipaddr ,
>  .BR descriptors ,
>  .BR port ,

-- 
Jeff Layton <jlayton@kernel.org>