Add a key-value interface to pvbus(4) that allows to get or set values

in the underlying information store of the host from the OpenBSD-VM's
userspace.  OpenBSD did not provide access to these stores before,
mostly because we did not want to add a custom tool and interface for
each hypervisor.  The pvbus(4) interface provides backends for
xen(4)'s XenStore and vmt(4)'s VMware Tools "guestinfo".  These
information stores are fairly different, XenStore is a "filesystem"
while vmt is a RPC, and the key-value abstraction limits them a bit
but provides the most wanted functionality.

Discussed with many
OK mikeb@

9 files changed, 318 insertions, 10 deletions

diff --git a/sys/arch/amd64/amd64/conf.c b/sys/arch/amd64/amd64/conf.c
index bca710e7059..d69887b3061 100644
--- a/sys/arch/amd64/amd64/conf.c
+++ b/sys/arch/amd64/amd64/conf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: conf.c,v 1.54 2016/01/08 11:20:58 reyk Exp $	*/
+/*	$OpenBSD: conf.c,v 1.55 2016/01/27 09:04:19 reyk Exp $	*/
 
 /*
  * Copyright (c) 1994, 1995 Charles M. Hannum.  All rights reserved.
@@ -182,6 +182,7 @@ cdev_decl(pci);
 #include "vscsi.h"
 #include "pppx.h"
 #include "fuse.h"
+#include "pvbus.h"
 
 struct cdevsw	cdevsw[] =
 {
@@ -292,6 +293,7 @@ struct cdevsw	cdevsw[] =
 	cdev_fuse_init(NFUSE,fuse),	/* 92: fuse */
 	cdev_tun_init(NTUN,tap),	/* 93: Ethernet network tunnel */
 	cdev_tty_init(NVIOCON,viocon),  /* 94: virtio console */
+	cdev_pvbus_init(NPVBUS,pvbus),	/* 95: pvbus(4) control interface */
 };
 int	nchrdev = nitems(cdevsw);
 
diff --git a/sys/arch/i386/i386/conf.c b/sys/arch/i386/i386/conf.c
index ebd2cbb48fe..668e13724a5 100644
--- a/sys/arch/i386/i386/conf.c
+++ b/sys/arch/i386/i386/conf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: conf.c,v 1.151 2015/10/23 15:10:52 claudio Exp $	*/
+/*	$OpenBSD: conf.c,v 1.152 2016/01/27 09:04:19 reyk Exp $	*/
 /*	$NetBSD: conf.c,v 1.75 1996/05/03 19:40:20 christos Exp $	*/
 
 /*
@@ -176,6 +176,7 @@ cdev_decl(pci);
 #include "vscsi.h"
 #include "pppx.h"
 #include "fuse.h"
+#include "pvbus.h"
 
 struct cdevsw	cdevsw[] =
 {
@@ -286,6 +287,7 @@ struct cdevsw	cdevsw[] =
 	cdev_pppx_init(NPPPX,pppx),     /* 92: pppx */
 	cdev_fuse_init(NFUSE,fuse),	/* 93: fuse */ 
 	cdev_tun_init(NTUN,tap),	/* 94: Ethernet network tunnel */
+	cdev_pvbus_init(NPVBUS,pvbus),	/* 95: pvbus(4) control interface */
 };
 int	nchrdev = nitems(cdevsw);
 
diff --git a/sys/dev/pv/pvbus.c b/sys/dev/pv/pvbus.c
index 91c7c622140..49bbcb33d92 100644
--- a/sys/dev/pv/pvbus.c
+++ b/sys/dev/pv/pvbus.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pvbus.c,v 1.10 2015/12/12 12:47:49 reyk Exp $	*/
+/*	$OpenBSD: pvbus.c,v 1.11 2016/01/27 09:04:19 reyk Exp $	*/
 
 /*
  * Copyright (c) 2015 Reyk Floeter <reyk@openbsd.org>
@@ -29,9 +29,12 @@
 #include <sys/syslog.h>
 #include <sys/proc.h>
 #include <sys/socket.h>
+#include <sys/ioctl.h>
+#include <sys/fcntl.h>
 
 #include <machine/specialreg.h>
 #include <machine/cpu.h>
+#include <machine/conf.h>
 #ifdef __amd64__
 #include <machine/vmmvar.h>
 #endif
@@ -59,6 +62,9 @@ void	 pvbus_hyperv_print(struct pvbus_hv *);
 void	 pvbus_xen(struct pvbus_hv *);
 void	 pvbus_xen_print(struct pvbus_hv *);
 
+int	 pvbus_minor(struct pvbus_softc *, dev_t);
+int	 pvbusgetstr(size_t, const char *, char **);
+
 struct cfattach pvbus_ca = {
 	sizeof(struct pvbus_softc),
 	pvbus_match,
@@ -90,6 +96,7 @@ struct pvbus_type {
 };
 
 struct pvbus_hv pvbus_hv[PVBUS_MAX];
+struct pvbus_softc *pvbus_softc;
 
 int
 pvbus_probe(void)
@@ -114,6 +121,7 @@ pvbus_attach(struct device *parent, struct device *self, void *aux)
 	int i, cnt;
 
 	sc->pvbus_hv = pvbus_hv;
+	pvbus_softc = sc;
 
 	printf(":");
 	for (i = 0, cnt = 0; i < PVBUS_MAX; i++) {
@@ -276,3 +284,140 @@ pvbus_xen_print(struct pvbus_hv *hv)
 {
 	printf(" %u.%u", hv->hv_major, hv->hv_minor);
 }
+
+int
+pvbus_minor(struct pvbus_softc *sc, dev_t dev)
+{
+	int hvid, cnt;
+	struct pvbus_hv *hv;
+
+	for (hvid = 0, cnt = 0; hvid < PVBUS_MAX; hvid++) {
+		hv = &sc->pvbus_hv[hvid];
+		if (hv->hv_base == 0 || hv->hv_kvop == NULL)
+			continue;
+		if (minor(dev) == cnt++)
+			return (hvid);
+	}
+
+	return (-1);
+}
+
+int
+pvbusopen(dev_t dev, int flags, int mode, struct proc *p)
+{
+	if (pvbus_softc == NULL)
+		return (ENODEV);
+	if (pvbus_minor(pvbus_softc, dev) == -1)
+		return (ENXIO);
+	return (0);
+}
+
+int
+pvbusclose(dev_t dev, int flags, int mode, struct proc *p)
+{
+	if (pvbus_softc == NULL)
+		return (ENODEV);
+	if (pvbus_minor(pvbus_softc, dev) == -1)
+		return (ENXIO);
+	return (0);
+}
+
+int
+pvbusgetstr(size_t srclen, const char *src, char **dstp)
+{
+	int error = 0;
+	char *dst;
+
+	/*
+	 * Reject size that is too short or obviously too long:
+	 * - at least one byte for the nul terminator.
+	 * - PAGE_SIZE is an arbitrary value, but known pv backends seem
+	 *   to have a hard (PAGE_SIZE - x) limit in their messaging.
+	 */
+	if (srclen < 1)
+		return (EINVAL);
+	else if (srclen > PAGE_SIZE)
+		return (ENAMETOOLONG);
+
+	*dstp = dst = malloc(srclen + 1, M_TEMP|M_ZERO, M_WAITOK);
+	if (src != NULL) {
+		error = copyin(src, dst, srclen);
+		dst[srclen] = '\0';
+	}
+
+	return (error);
+}
+
+int
+pvbusioctl(dev_t dev, u_long cmd, caddr_t data, int flags, struct proc *p)
+{
+	struct pvbus_req *pvr = (struct pvbus_req *)data;
+	struct pvbus_softc *sc = pvbus_softc;
+	char *value = NULL, *key = NULL;
+	const char *str = NULL;
+	size_t valuelen = 0, keylen = 0, sz;
+	int hvid, error = 0, op;
+	struct pvbus_hv *hv;
+
+	if (sc == NULL)
+		return (ENODEV);
+	if ((hvid = pvbus_minor(sc, dev)) == -1)
+		return (ENXIO);
+
+	switch (cmd) {
+	case PVBUSIOC_KVWRITE:
+		if ((flags & FWRITE) == 0)
+			return (EPERM);
+	case PVBUSIOC_KVREAD:
+		hv = &sc->pvbus_hv[hvid];
+		if (hv->hv_base == 0 || hv->hv_kvop == NULL)
+			return (ENXIO);
+		break;
+	case PVBUSIOC_TYPE:
+		str = pvbus_types[hvid].name;
+		sz = strlen(str) + 1;
+		if (sz > pvr->pvr_keylen)
+			return (ENOMEM);
+		error = copyout(str, pvr->pvr_key, sz);
+		return (error);
+	default:
+		return (ENOTTY);
+	}
+
+	str = NULL;
+	op = PVBUS_KVREAD;
+
+	switch (cmd) {
+	case PVBUSIOC_KVWRITE:
+		str = pvr->pvr_value;
+		op = PVBUS_KVWRITE;
+
+		/* FALLTHROUGH */
+	case PVBUSIOC_KVREAD:
+		keylen = pvr->pvr_keylen;
+		if ((error = pvbusgetstr(keylen, pvr->pvr_key, &key)) != 0)
+			break;
+
+		valuelen = pvr->pvr_valuelen;
+		if ((error = pvbusgetstr(valuelen, str, &value)) != 0)
+			break;
+
+		/* Call driver-specific callback */
+		if ((error = (hv->hv_kvop)(hv->hv_arg, op,
+		    key, value, valuelen)) != 0)
+			break;
+
+		sz = strlen(value) + 1;
+		if ((error = copyout(value, pvr->pvr_value, sz)) != 0)
+			break;
+		break;
+	default:
+		error = ENOTTY;
+		break;
+	}
+
+	free(key, M_TEMP, keylen);
+	free(value, M_TEMP, valuelen);
+
+	return (error);
+}
diff --git a/sys/dev/pv/pvvar.h b/sys/dev/pv/pvvar.h
index cd15cad63b1..4091c196130 100644
--- a/sys/dev/pv/pvvar.h
+++ b/sys/dev/pv/pvvar.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pvvar.h,v 1.6 2015/12/12 12:33:49 reyk Exp $	*/
+/*	$OpenBSD: pvvar.h,v 1.7 2016/01/27 09:04:19 reyk Exp $	*/
 
 /*
  * Copyright (c) 2015 Reyk Floeter <reyk@openbsd.org>
@@ -19,6 +19,18 @@
 #ifndef _DEV_PV_PVVAR_H_
 #define _DEV_PV_PVVAR_H_
 
+struct pvbus_req {
+	size_t			 pvr_keylen;
+	char			*pvr_key;
+	size_t			 pvr_valuelen;
+	char			*pvr_value;
+};
+
+#define PVBUSIOC_KVREAD		_IOWR('V', 1, struct pvbus_req)
+#define PVBUSIOC_KVWRITE	_IOWR('V', 2, struct pvbus_req)
+#define PVBUSIOC_TYPE		_IOWR('V', 3, struct pvbus_req)
+
+#ifdef _KERNEL
 enum {
 	PVBUS_KVM,
 	PVBUS_HYPERV,
@@ -30,11 +42,20 @@ enum {
 	PVBUS_MAX
 };
 
+enum {
+	PVBUS_KVREAD,
+	PVBUS_KVWRITE,
+	PVBUS_KVLS
+};
+
 struct pvbus_hv {
 	uint32_t		 hv_base;
 	uint32_t		 hv_features;
 	uint16_t		 hv_major;
 	uint16_t		 hv_minor;
+
+	void			*hv_arg;
+	int			(*hv_kvop)(void *, int, char *, char *, size_t);
 };
 
 struct pvbus_softc {
@@ -54,4 +75,5 @@ struct pv_attach_args {
 void	 pvbus_identify(void);
 int	 pvbus_probe(void);
 
+#endif /* _KERNEL */
 #endif /* _DEV_PV_PVBUS_H_ */
diff --git a/sys/dev/pv/vmt.c b/sys/dev/pv/vmt.c
index 4333439bf92..a9b5dba82a8 100644
--- a/sys/dev/pv/vmt.c
+++ b/sys/dev/pv/vmt.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: vmt.c,v 1.7 2015/12/17 09:26:36 mpi Exp $ */
+/*	$OpenBSD: vmt.c,v 1.8 2016/01/27 09:04:19 reyk Exp $ */
 
 /*
  * Copyright (c) 2007 David Crawshaw <david@zentus.com>
@@ -35,6 +35,7 @@
 #include <sys/syslog.h>
 #include <sys/proc.h>
 #include <sys/socket.h>
+#include <sys/ioctl.h>
 
 #include <net/if.h>
 #include <net/if_var.h>
@@ -166,7 +167,7 @@ struct vmt_softc {
 	int			sc_rpc_error;
 	int			sc_tclo_ping;
 	int			sc_set_guest_os;
-#define VMT_RPC_BUFLEN			256
+#define VMT_RPC_BUFLEN		1024
 
 	struct timeout		sc_tick;
 	struct timeout		sc_tclo_tick;
@@ -199,6 +200,8 @@ int	 vm_rpc_send_rpci_tx(struct vmt_softc *, const char *, ...)
 	    __attribute__((__format__(__kprintf__,2,3)));
 int	 vm_rpci_response_successful(struct vmt_softc *);
 
+int	 vmt_kvop(void *, int, char *, char *, size_t);
+
 void	 vmt_probe_cmd(struct vm_backdoor *, uint16_t);
 void	 vmt_tclo_state_change_success(struct vmt_softc *, int, char);
 void	 vmt_do_reboot(struct vmt_softc *);
@@ -334,6 +337,8 @@ void
 vmt_attach(struct device *parent, struct device *self, void *aux)
 {
 	struct vmt_softc *sc = (struct vmt_softc *)self;
+	struct pv_attach_args	*pva = aux;
+	struct pvbus_hv		*hv = &pva->pva_hv[PVBUS_VMWARE];
 
 	printf("\n");
 	sc->sc_rpc_buf = malloc(VMT_RPC_BUFLEN, M_DEVBUF, M_NOWAIT);
@@ -371,12 +376,62 @@ vmt_attach(struct device *parent, struct device *self, void *aux)
 	timeout_add_sec(&sc->sc_tclo_tick, 1);
 	sc->sc_tclo_ping = 1;
 
+	/* pvbus(4) key/value interface */
+	hv->hv_kvop = vmt_kvop;
+	hv->hv_arg = sc;
+
 	return;
 
 free:
 	free(sc->sc_rpc_buf, M_DEVBUF, VMT_RPC_BUFLEN);
 }
 
+int
+vmt_kvop(void *arg, int op, char *key, char *value, size_t valuelen)
+{
+	struct vmt_softc *sc = arg;
+	char buf[VMT_RPC_BUFLEN], *ptr;
+
+	switch (op) {
+	case PVBUS_KVWRITE:
+		if ((size_t)snprintf(buf, sizeof(buf), "info-set %s %s",
+		    key, value) >= sizeof(buf)) {
+			DPRINTF("%s: write command too long", DEVNAME(sc));
+			return (EINVAL);
+		}
+		break;
+	case PVBUS_KVREAD:
+		if ((size_t)snprintf(buf, sizeof(buf), "info-get %s",
+		    key) >= sizeof(buf)) {
+			DPRINTF("%s: read command too long", DEVNAME(sc));
+			return (EINVAL);
+		}
+		break;
+	default:
+		return (EOPNOTSUPP);
+	}
+
+	if (vm_rpc_send_rpci_tx(sc, buf) != 0) {
+		DPRINTF("%s: error sending command: %s\n", DEVNAME(sc), buf);
+		sc->sc_rpc_error = 1;
+		return (EIO);
+	}
+
+	if (vm_rpci_response_successful(sc) == 0) {
+		DPRINTF("%s: host rejected command: %s\n", DEVNAME(sc), buf);
+		return (EINVAL);
+	}
+
+	/* skip response that was tested in vm_rpci_response_successful() */
+	ptr = sc->sc_rpc_buf + 2;
+
+	/* might truncat, copy anyway but return error */
+	if (strlcpy(value, ptr, valuelen) >= valuelen)
+		return (ENOMEM);
+
+	return (0);
+}
+
 void
 vmt_resume(void)
 {
diff --git a/sys/dev/pv/xen.c b/sys/dev/pv/xen.c
index 138e5e36c55..0ac3186bc69 100644
--- a/sys/dev/pv/xen.c
+++ b/sys/dev/pv/xen.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: xen.c,v 1.39 2016/01/26 15:51:07 mikeb Exp $	*/
+/*	$OpenBSD: xen.c,v 1.40 2016/01/27 09:04:19 reyk Exp $	*/
 
 /*
  * Copyright (c) 2015 Mike Belopuhov
@@ -156,6 +156,10 @@ xen_attach(struct device *parent, struct device *self, void *aux)
 
 	xen_probe_devices(sc);
 
+	/* pvbus(4) key/value interface */
+	hv->hv_kvop = xs_kvop;
+	hv->hv_arg = sc;
+
 	xen_disable_emulated_devices(sc);
 
 	config_mountroot(self, xen_deferred);
diff --git a/sys/dev/pv/xenstore.c b/sys/dev/pv/xenstore.c
index 021d6260f99..adfc52f1185 100644
--- a/sys/dev/pv/xenstore.c
+++ b/sys/dev/pv/xenstore.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: xenstore.c,v 1.21 2016/01/25 15:22:56 mikeb Exp $	*/
+/*	$OpenBSD: xenstore.c,v 1.22 2016/01/27 09:04:19 reyk Exp $	*/
 
 /*
  * Copyright (c) 2015 Mike Belopuhov
@@ -23,11 +23,13 @@
 #include <sys/malloc.h>
 #include <sys/device.h>
 #include <sys/mutex.h>
+#include <sys/ioctl.h>
 
 #include <machine/bus.h>
 
 #include <uvm/uvm_extern.h>
 
+#include <dev/pv/pvvar.h>
 #include <dev/pv/xenreg.h>
 #include <dev/pv/xenvar.h>
 
@@ -820,3 +822,69 @@ xs_setprop(struct xen_attach_args *xa, const char *property, char *value,
 	}
 	return (error);
 }
+
+int
+xs_kvop(void *arg, int op, char *key, char *value, size_t valuelen)
+{
+	struct xen_softc *sc = arg;
+	struct xs_transaction xst;
+	struct iovec iov, *iovp = &iov;
+	int error = 0, iov_cnt, cmd, i;
+
+	switch (op) {
+	case PVBUS_KVWRITE:
+		cmd = XS_WRITE;
+		iov.iov_base = value;
+		iov.iov_len = strlen(value);
+		iov_cnt = 1;
+		break;
+	case PVBUS_KVREAD:
+		cmd = XS_READ;
+		break;
+	case PVBUS_KVLS:
+		cmd = XS_LIST;
+		break;
+	default:
+		return (EOPNOTSUPP);
+	}
+
+	memset(&xst, 0, sizeof(xst));
+	xst.xst_id = 0;
+	xst.xst_sc = sc->sc_xs;
+
+	if ((error = xs_cmd(&xst, cmd, key, &iovp, &iov_cnt)) != 0)
+		return (error);
+
+	memset(value, 0, valuelen);
+
+	switch (cmd) {
+	case XS_READ:
+		if (iov_cnt == 1 && iovp[0].iov_len == 1) {
+			xs_resfree(&xst, iovp, iov_cnt);
+
+			/*
+			 * We cannot distinguish if the returned value is
+			 * a directory or a file in the xenstore.  The only
+			 * indication is that the read value of a directory
+			 * returns an empty string (single nul byte),
+			 * so try to get the directory list in this case.
+			 */
+			return (xs_kvop(arg, PVBUS_KVLS, key, value, valuelen));
+		}
+		/* FALLTHROUGH */
+	case XS_LIST:
+		for (i = 0; i < iov_cnt; i++) {
+			if (i && strlcat(value, "\n", valuelen) >= valuelen)
+				break;
+			if (strlcat(value, iovp[i].iov_base,
+			    valuelen) >= valuelen)
+				break;
+		}
+		xs_resfree(&xst, iovp, iov_cnt);
+		break;
+	default:
+		break;
+	}
+
+	return (0);
+}
diff --git a/sys/dev/pv/xenvar.h b/sys/dev/pv/xenvar.h
index bf2587e1550..3042e5b00e9 100644
--- a/sys/dev/pv/xenvar.h
+++ b/sys/dev/pv/xenvar.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: xenvar.h,v 1.23 2016/01/25 15:22:56 mikeb Exp $	*/
+/*	$OpenBSD: xenvar.h,v 1.24 2016/01/27 09:04:19 reyk Exp $	*/
 
 /*
  * Copyright (c) 2015 Mike Belopuhov
@@ -150,5 +150,6 @@ int	xs_cmd(struct xs_transaction *, int, const char *, struct iovec **,
 int	xs_getprop(struct xen_attach_args *, const char *, char *, int);
 int	xs_setprop(struct xen_attach_args *, const char *, char *, int);
 void	xs_resfree(struct xs_transaction *, struct iovec *, int);
+int	xs_kvop(void *, int, char *, char *, size_t);
 
 #endif	/* _XENVAR_H_ */
diff --git a/sys/sys/conf.h b/sys/sys/conf.h
index 9be9b67750e..95212cb7569 100644
--- a/sys/sys/conf.h
+++ b/sys/sys/conf.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: conf.h,v 1.136 2016/01/06 09:09:16 kettenis Exp $	*/
+/*	$OpenBSD: conf.h,v 1.137 2016/01/27 09:04:19 reyk Exp $	*/
 /*	$NetBSD: conf.h,v 1.33 1996/05/03 20:03:32 christos Exp $	*/
 
 /*-
@@ -465,6 +465,14 @@ extern struct cdevsw cdevsw[];
 	(dev_type_stop((*))) enodev, 0, dev_init(c,n,poll), \
 	(dev_type_mmap((*))) enodev, 0, D_CLONE, dev_init(c,n,kqfilter) }
 
+#define cdev_pvbus_init(c,n) { \
+	dev_init(c,n,open), dev_init(c,n,close), \
+	(dev_type_read((*))) enodev, \
+	(dev_type_write((*))) enodev, \
+	 dev_init(c,n,ioctl), \
+	(dev_type_stop((*))) enodev, 0, selfalse, \
+	(dev_type_mmap((*))) enodev }
+
 #endif
 
 /*
@@ -597,6 +605,7 @@ cdev_decl(hotplug);
 cdev_decl(gpio);
 cdev_decl(amdmsr);
 cdev_decl(fuse);
+cdev_decl(pvbus);
 
 #endif