executor: move fixed pci devlink handle into network namespace

In case there is a fixed pci devlink handle "pci/pci/0000:00:10.0"
on the system (initial network namespace), it is moved to a working
network namespace.

Signed-off-by: Jiri Pirko <jiri@mellanox.com>

4 files changed, 124 insertions, 14 deletions

diff --git a/pkg/csource/common.go b/pkg/csource/common.go
index e537ed724..67d31ad81 100644
--- a/pkg/csource/common.go
+++ b/pkg/csource/common.go
@@ -95,6 +95,7 @@ func defineList(p, mmapProg *prog.Prog, opts Options) (defines []string) {
 		"SYZ_ENABLE_BINFMT_MISC":            opts.EnableBinfmtMisc,
 		"SYZ_ENABLE_CLOSE_FDS":              opts.EnableCloseFds,
 		"SYZ_ENABLE_KCSAN":                  opts.EnableKCSAN,
+		"SYZ_ENABLE_DEVLINK_PCI":            opts.EnableDevlinkPCI,
 		"SYZ_USE_TMP_DIR":                   opts.UseTmpDir,
 		"SYZ_HANDLE_SEGV":                   opts.HandleSegv,
 		"SYZ_REPRO":                         opts.Repro,
diff --git a/pkg/csource/generated.go b/pkg/csource/generated.go
index bcfc0a01f..2266f86d6 100644
--- a/pkg/csource/generated.go
+++ b/pkg/csource/generated.go
@@ -1130,7 +1130,7 @@ static bool write_file(const char* file, const char* what, ...)
 }
 #endif
 
-#if SYZ_EXECUTOR || SYZ_ENABLE_NETDEV || SYZ_TUN_ENABLE
+#if SYZ_EXECUTOR || SYZ_ENABLE_NETDEV || SYZ_TUN_ENABLE || SYZ_ENABLE_DEVLINK_PCI
 #include <arpa/inet.h>
 #include <net/if.h>
 #include <netinet/in.h>
@@ -1189,7 +1189,7 @@ static void netlink_done(void)
 }
 #endif
 
-static int netlink_send(int sock)
+static int netlink_send_ext(int sock, uint16 reply_type, int* reply_len)
 {
 	if (nlmsg.pos > nlmsg.buf + sizeof(nlmsg.buf) || nlmsg.nesting)
 		fail("nlmsg overflow/bad nesting");
@@ -1202,6 +1202,12 @@ static int netlink_send(int sock)
 	if (n != hdr->nlmsg_len)
 		fail("short netlink write: %d/%d", n, hdr->nlmsg_len);
 	n = recv(sock, nlmsg.buf, sizeof(nlmsg.buf), 0);
+	if (n < sizeof(struct nlmsghdr))
+		fail("short netlink read: %d", n);
+	if (reply_len && hdr->nlmsg_type == reply_type) {
+		*reply_len = n;
+		return 0;
+	}
 	if (n < sizeof(struct nlmsghdr) + sizeof(struct nlmsgerr))
 		fail("short netlink read: %d", n);
 	if (hdr->nlmsg_type != NLMSG_ERROR)
@@ -1209,6 +1215,11 @@ static int netlink_send(int sock)
 	return -((struct nlmsgerr*)(hdr + 1))->error;
 }
 
+static int netlink_send(int sock)
+{
+	return netlink_send_ext(sock, 0, NULL);
+}
+
 #if SYZ_EXECUTOR || SYZ_ENABLE_NETDEV
 static void netlink_add_device_impl(const char* type, const char* name)
 {
@@ -1262,6 +1273,7 @@ static void netlink_add_hsr(int sock, const char* name, const char* slave1, cons
 }
 #endif
 
+#if SYZ_EXECUTOR || SYZ_ENABLE_NETDEV || SYZ_TUN_ENABLE
 static void netlink_device_change(int sock, const char* name, bool up,
 				  const char* master, const void* mac, int macsize)
 {
@@ -1313,6 +1325,7 @@ static void netlink_add_addr6(int sock, const char* dev, const char* addr)
 	debug("netlink: add addr %s dev %s: %s\n", addr, dev, strerror(err));
 	(void)err;
 }
+#endif
 
 #if SYZ_EXECUTOR || SYZ_TUN_ENABLE
 static void netlink_add_neigh(int sock, const char* name,
@@ -1432,6 +1445,83 @@ static void initialize_tun(void)
 }
 #endif
 
+#if SYZ_EXECUTOR || __NR_syz_init_net_socket || SYZ_ENABLE_DEVLINK_PCI
+const int kInitNetNsFd = 239;
+#endif
+
+#if SYZ_EXECUTOR || SYZ_ENABLE_DEVLINK_PCI
+
+#include <linux/genetlink.h>
+
+#define DEVLINK_FAMILY_NAME "devlink"
+
+#define DEVLINK_CMD_RELOAD 37
+#define DEVLINK_ATTR_BUS_NAME 1
+#define DEVLINK_ATTR_DEV_NAME 2
+#define DEVLINK_ATTR_NETNS_FD 137
+
+static void netlink_devlink_netns_move(const char* bus_name, const char* dev_name, int netns_fd)
+{
+	struct genlmsghdr genlhdr;
+	struct nlattr* attr;
+	int sock, err, n;
+	uint16 id = 0;
+
+	sock = socket(AF_NETLINK, SOCK_RAW, NETLINK_GENERIC);
+	if (sock == -1)
+		fail("socket(AF_NETLINK) failed\n");
+	memset(&genlhdr, 0, sizeof(genlhdr));
+	genlhdr.cmd = CTRL_CMD_GETFAMILY;
+	netlink_init(GENL_ID_CTRL, 0, &genlhdr, sizeof(genlhdr));
+	netlink_attr(CTRL_ATTR_FAMILY_NAME, DEVLINK_FAMILY_NAME, strlen(DEVLINK_FAMILY_NAME) + 1);
+	err = netlink_send_ext(sock, GENL_ID_CTRL, &n);
+	if (err) {
+		debug("netlink: failed to get devlink family id: %s\n", strerror(err));
+		goto error;
+	}
+	attr = (struct nlattr*)(nlmsg.buf + NLMSG_HDRLEN + NLMSG_ALIGN(sizeof(genlhdr)));
+	for (; (char*)attr < nlmsg.buf + n; attr = (struct nlattr*)((char*)attr + NLMSG_ALIGN(attr->nla_len))) {
+		if (attr->nla_type == CTRL_ATTR_FAMILY_ID) {
+			id = *(uint16*)(attr + 1);
+			break;
+		}
+	}
+	if (!id) {
+		debug("netlink: failed to parse message for devlink family id\n");
+		goto error;
+	}
+	recv(sock, nlmsg.buf, sizeof(nlmsg.buf), 0); /* recv ack */
+	memset(&genlhdr, 0, sizeof(genlhdr));
+	genlhdr.cmd = DEVLINK_CMD_RELOAD;
+	netlink_init(id, 0, &genlhdr, sizeof(genlhdr));
+	netlink_attr(DEVLINK_ATTR_BUS_NAME, bus_name, strlen(bus_name) + 1);
+	netlink_attr(DEVLINK_ATTR_DEV_NAME, dev_name, strlen(dev_name) + 1);
+	netlink_attr(DEVLINK_ATTR_NETNS_FD, &netns_fd, sizeof(netns_fd));
+	netlink_send(sock);
+error:
+	close(sock);
+}
+
+static void initialize_devlink_pci(void)
+{
+#if SYZ_EXECUTOR
+	if (!flag_enable_devlink_pci)
+		return;
+#endif
+	int netns = open("/proc/self/ns/net", O_RDONLY);
+	if (netns == -1)
+		fail("open(/proc/self/ns/net) failed");
+	int ret = setns(kInitNetNsFd, 0);
+	if (ret == -1)
+		fail("set_ns(init_netns_fd) failed");
+	netlink_devlink_netns_move("pci", "0000:00:10.0", netns);
+	ret = setns(netns, 0);
+	if (ret == -1)
+		fail("set_ns(this_netns_fd) failed");
+	close(netns);
+}
+#endif
+
 #if SYZ_EXECUTOR || SYZ_ENABLE_NETDEV
 #include <arpa/inet.h>
 #include <errno.h>
@@ -2671,8 +2761,6 @@ static long syz_open_pts(volatile long a0, volatile long a1)
 #include <sys/stat.h>
 #include <sys/types.h>
 #include <unistd.h>
-
-const int kInitNetNsFd = 239;
 static long syz_init_net_socket(volatile long domain, volatile long type, volatile long proto)
 {
 	int netns = open("/proc/self/ns/net", O_RDONLY);
@@ -4571,7 +4659,7 @@ static void sandbox_common()
 	setpgrp();
 	setsid();
 
-#if SYZ_EXECUTOR || __NR_syz_init_net_socket
+#if SYZ_EXECUTOR || __NR_syz_init_net_socket || SYZ_ENABLE_DEVLINK_PCI
 	int netns = open("/proc/self/ns/net", O_RDONLY);
 	if (netns == -1)
 		fail("open(/proc/self/ns/net) failed");
@@ -4685,6 +4773,9 @@ static int do_sandbox_none(void)
 	if (unshare(CLONE_NEWNET)) {
 		debug("unshare(CLONE_NEWNET): %d\n", errno);
 	}
+#if SYZ_EXECUTOR || SYZ_ENABLE_DEVLINK_PCI
+	initialize_devlink_pci();
+#endif
 #if SYZ_EXECUTOR || SYZ_TUN_ENABLE
 	initialize_tun();
 #endif
@@ -4719,6 +4810,9 @@ static int do_sandbox_setuid(void)
 	if (unshare(CLONE_NEWNET)) {
 		debug("unshare(CLONE_NEWNET): %d\n", errno);
 	}
+#if SYZ_EXECUTOR || SYZ_ENABLE_DEVLINK_PCI
+	initialize_devlink_pci();
+#endif
 #if SYZ_EXECUTOR || SYZ_TUN_ENABLE
 	initialize_tun();
 #endif
@@ -4763,6 +4857,9 @@ static int namespace_sandbox_proc(void* arg)
 #endif
 	if (unshare(CLONE_NEWNET))
 		fail("unshare(CLONE_NEWNET)");
+#if SYZ_EXECUTOR || SYZ_ENABLE_DEVLINK_PCI
+	initialize_devlink_pci();
+#endif
 #if SYZ_EXECUTOR || SYZ_TUN_ENABLE
 	initialize_tun();
 #endif
diff --git a/pkg/csource/options.go b/pkg/csource/options.go
index 7ad8c1d9c..a4cc45d45 100644
--- a/pkg/csource/options.go
+++ b/pkg/csource/options.go
@@ -38,6 +38,7 @@ type Options struct {
 	EnableBinfmtMisc bool `json:"binfmt_misc,omitempty"`
 	EnableCloseFds   bool `json:"close_fds"`
 	EnableKCSAN      bool `json:"kcsan,omitempty"`
+	EnableDevlinkPCI bool `json:"devlinkpci,omitempty"`
 
 	UseTmpDir  bool `json:"tmpdir,omitempty"`
 	HandleSegv bool `json:"segv,omitempty"`
@@ -127,6 +128,9 @@ func (opts Options) checkLinuxOnly(OS string) error {
 	if opts.EnableKCSAN {
 		return fmt.Errorf("option EnableKCSAN is not supported on %v", OS)
 	}
+	if opts.EnableDevlinkPCI {
+		return fmt.Errorf("option EnableDevlinkPCI is not supported on %v", OS)
+	}
 	if opts.Sandbox == sandboxNamespace ||
 		(opts.Sandbox == sandboxSetuid && !(OS == openbsd || OS == freebsd || OS == netbsd)) ||
 		opts.Sandbox == sandboxAndroidUntrustedApp {
@@ -154,6 +158,7 @@ func DefaultOpts(cfg *mgrconfig.Config) Options {
 		EnableCgroups:    true,
 		EnableBinfmtMisc: true,
 		EnableCloseFds:   true,
+		EnableDevlinkPCI: true,
 		UseTmpDir:        true,
 		HandleSegv:       true,
 		Repro:            true,
@@ -165,6 +170,12 @@ func DefaultOpts(cfg *mgrconfig.Config) Options {
 		opts.EnableCgroups = false
 		opts.EnableBinfmtMisc = false
 		opts.EnableCloseFds = false
+		opts.EnableDevlinkPCI = false
+	}
+	if cfg.Sandbox == "" || cfg.Sandbox == "setuid" {
+		opts.EnableNetReset = false
+	}
+	if err := opts.Check(cfg.TargetOS); err != nil {
 	}
 	if cfg.Sandbox == "" || cfg.Sandbox == "setuid" {
 		opts.EnableNetReset = false
@@ -244,6 +255,7 @@ func defaultFeatures(value bool) Features {
 		"cgroups":     {"setup cgroups for testing", value},
 		"binfmt_misc": {"setup binfmt_misc for testing", value},
 		"close_fds":   {"close fds after each program", value},
+		"devlink_pci": {"setup devlink PCI device", value},
 	}
 }
 
diff --git a/pkg/csource/options_test.go b/pkg/csource/options_test.go
index 13afa84d3..34c64f9e2 100644
--- a/pkg/csource/options_test.go
+++ b/pkg/csource/options_test.go
@@ -213,31 +213,31 @@ func TestParseFeaturesFlags(t *testing.T) {
 		Features map[string]bool
 	}{
 		{"none", "none", true, map[string]bool{
-			"tun": true, "net_dev": true, "net_reset": true, "cgroups": true, "binfmt_misc": true, "close_fds": true,
+			"tun": true, "net_dev": true, "net_reset": true, "cgroups": true, "binfmt_misc": true, "close_fds": true, "devlink_pci": true,
 		}},
 		{"none", "none", false, map[string]bool{
-			"tun": false, "net_dev": false, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": false,
+			"tun": false, "net_dev": false, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": false, "devlink_pci": false,
 		}},
 		{"all", "none", true, map[string]bool{
-			"tun": true, "net_dev": true, "net_reset": true, "cgroups": true, "binfmt_misc": true, "close_fds": true,
+			"tun": true, "net_dev": true, "net_reset": true, "cgroups": true, "binfmt_misc": true, "close_fds": true, "devlink_pci": true,
 		}},
 		{"", "none", true, map[string]bool{
-			"tun": false, "net_dev": false, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": false,
+			"tun": false, "net_dev": false, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": false, "devlink_pci": false,
 		}},
 		{"none", "all", true, map[string]bool{
-			"tun": false, "net_dev": false, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": false,
+			"tun": false, "net_dev": false, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": false, "devlink_pci": false,
 		}},
 		{"none", "", true, map[string]bool{
-			"tun": true, "net_dev": true, "net_reset": true, "cgroups": true, "binfmt_misc": true, "close_fds": true,
+			"tun": true, "net_dev": true, "net_reset": true, "cgroups": true, "binfmt_misc": true, "close_fds": true, "devlink_pci": true,
 		}},
 		{"tun,net_dev", "none", true, map[string]bool{
-			"tun": true, "net_dev": true, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": false,
+			"tun": true, "net_dev": true, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": false, "devlink_pci": false,
 		}},
 		{"none", "cgroups,net_dev", true, map[string]bool{
-			"tun": true, "net_dev": false, "net_reset": true, "cgroups": false, "binfmt_misc": true, "close_fds": true,
+			"tun": true, "net_dev": false, "net_reset": true, "cgroups": false, "binfmt_misc": true, "close_fds": true, "devlink_pci": true,
 		}},
 		{"close_fds", "none", true, map[string]bool{
-			"tun": false, "net_dev": false, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": true,
+			"tun": false, "net_dev": false, "net_reset": false, "cgroups": false, "binfmt_misc": false, "close_fds": true, "devlink_pci": false,
 		}},
 	}
 	for i, test := range tests {