Merge branch 'bpf-offload-report-dev'

Jakub Kicinski says:

====================
This series is a redo of reporting offload device information to
user space after the first attempt did not take into account name
spaces.  As requested by Kirill offloads are now protected by an
r/w sem.  This allows us to remove the workqueue and free the
offload state fully when device is removed (suggested by Alexei).

Net namespace is reported with a device/inode pair.

The accompanying bpftool support is placed in common code because
maps will have very similar info.  Note that the UAPI information
can't be nicely encapsulated into a struct, because in case we
need to grow the device information the new fields will have to
be added at the end of struct bpf_prog_info, we can't grow
structures in the middle of bpf_prog_info.

v3:
 - use dev_get_by_index();
 - redo ns code (new patch 6).
v2:
 - rework the locking in patch 1 (use RCU instead of locking
   dependencies);
 - grab RTNL for a short time in patch 6;
 - minor update to the test in patch 8.
====================

Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>

5 files changed, 161 insertions, 11 deletions

diff --git a/tools/bpf/bpftool/common.c b/tools/bpf/bpftool/common.c
index b62c94e3997a..6601c95a9258 100644
--- a/tools/bpf/bpftool/common.c
+++ b/tools/bpf/bpftool/common.c
@@ -44,7 +44,9 @@
 #include <unistd.h>
 #include <linux/limits.h>
 #include <linux/magic.h>
+#include <net/if.h>
 #include <sys/mount.h>
+#include <sys/stat.h>
 #include <sys/types.h>
 #include <sys/vfs.h>
 
@@ -412,3 +414,53 @@ void delete_pinned_obj_table(struct pinned_obj_table *tab)
                 free(obj);
         }
 }
+
+static char *
+ifindex_to_name_ns(__u32 ifindex, __u32 ns_dev, __u32 ns_ino, char *buf)
+{
+        struct stat st;
+        int err;
+
+        err = stat("/proc/self/ns/net", &st);
+        if (err) {
+                p_err("Can't stat /proc/self: %s", strerror(errno));
+                return NULL;
+        }
+
+        if (st.st_dev != ns_dev || st.st_ino != ns_ino)
+                return NULL;
+
+        return if_indextoname(ifindex, buf);
+}
+
+void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
+{
+        char name[IF_NAMESIZE];
+
+        if (!ifindex)
+                return;
+
+        printf(" dev ");
+        if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
+                printf("%s", name);
+        else
+                printf("ifindex %u ns_dev %llu ns_ino %llu",
+                       ifindex, ns_dev, ns_inode);
+}
+
+void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode)
+{
+        char name[IF_NAMESIZE];
+
+        if (!ifindex)
+                return;
+
+        jsonw_name(json_wtr, "dev");
+        jsonw_start_object(json_wtr);
+        jsonw_uint_field(json_wtr, "ifindex", ifindex);
+        jsonw_uint_field(json_wtr, "ns_dev", ns_dev);
+        jsonw_uint_field(json_wtr, "ns_inode", ns_inode);
+        if (ifindex_to_name_ns(ifindex, ns_dev, ns_inode, name))
+                jsonw_string_field(json_wtr, "ifname", name);
+        jsonw_end_object(json_wtr);
+}
diff --git a/tools/bpf/bpftool/main.h b/tools/bpf/bpftool/main.h
index 8f6d3cac0347..65b526fe6e7e 100644
--- a/tools/bpf/bpftool/main.h
+++ b/tools/bpf/bpftool/main.h
@@ -96,6 +96,8 @@ struct pinned_obj {
 int build_pinned_obj_table(struct pinned_obj_table *table,
                            enum bpf_obj_type type);
 void delete_pinned_obj_table(struct pinned_obj_table *tab);
+void print_dev_plain(__u32 ifindex, __u64 ns_dev, __u64 ns_inode);
+void print_dev_json(__u32 ifindex, __u64 ns_dev, __u64 ns_inode);
 
 struct cmd {
         const char *cmd;
diff --git a/tools/bpf/bpftool/prog.c b/tools/bpf/bpftool/prog.c
index fd0873178503..98f871ed53d6 100644
--- a/tools/bpf/bpftool/prog.c
+++ b/tools/bpf/bpftool/prog.c
@@ -230,6 +230,8 @@ static void print_prog_json(struct bpf_prog_info *info, int fd)
                      info->tag[0], info->tag[1], info->tag[2], info->tag[3],
                      info->tag[4], info->tag[5], info->tag[6], info->tag[7]);
 
+        print_dev_json(info->ifindex, info->netns_dev, info->netns_ino);
+
         if (info->load_time) {
                 char buf[32];
 
@@ -287,6 +289,7 @@ static void print_prog_plain(struct bpf_prog_info *info, int fd)
 
         printf("tag ");
         fprint_hex(stdout, info->tag, BPF_TAG_SIZE, "");
+        print_dev_plain(info->ifindex, info->netns_dev, info->netns_ino);
         printf("\n");
 
         if (info->load_time) {
diff --git a/tools/include/uapi/linux/bpf.h b/tools/include/uapi/linux/bpf.h
index db1b0923a308..4e8c60acfa32 100644
--- a/tools/include/uapi/linux/bpf.h
+++ b/tools/include/uapi/linux/bpf.h
@@ -921,6 +921,9 @@ struct bpf_prog_info {
         __u32 nr_map_ids;
         __aligned_u64 map_ids;
         char name[BPF_OBJ_NAME_LEN];
+        __u32 ifindex;
+        __u64 netns_dev;
+        __u64 netns_ino;
 } __attribute__((aligned(8)));
 
 struct bpf_map_info {
diff --git a/tools/testing/selftests/bpf/test_offload.py b/tools/testing/selftests/bpf/test_offload.py
index c940505c2978..e3c750f17cb8 100755
--- a/tools/testing/selftests/bpf/test_offload.py
+++ b/tools/testing/selftests/bpf/test_offload.py
@@ -18,6 +18,8 @@ import argparse
 import json
 import os
 import pprint
+import random
+import string
 import subprocess
 import time
 
@@ -27,6 +29,7 @@ bpf_test_dir = os.path.dirname(os.path.realpath(__file__))
 pp = pprint.PrettyPrinter()
 devs = [] # devices we created for clean up
 files = [] # files to be removed
+netns = [] # net namespaces to be removed
 
 def log_get_sec(level=0):
     return "*" * (log_level + level)
@@ -128,22 +131,25 @@ def rm(f):
     if f in files:
         files.remove(f)
 
-def tool(name, args, flags, JSON=True, fail=True):
+def tool(name, args, flags, JSON=True, ns="", fail=True):
     params = ""
     if JSON:
         params += "%s " % (flags["json"])
 
-    ret, out = cmd(name + " " + params + args, fail=fail)
+    if ns != "":
+        ns = "ip netns exec %s " % (ns)
+
+    ret, out = cmd(ns + name + " " + params + args, fail=fail)
     if JSON and len(out.strip()) != 0:
         return ret, json.loads(out)
     else:
         return ret, out
 
-def bpftool(args, JSON=True, fail=True):
-    return tool("bpftool", args, {"json":"-p"}, JSON=JSON, fail=fail)
+def bpftool(args, JSON=True, ns="", fail=True):
+    return tool("bpftool", args, {"json":"-p"}, JSON=JSON, ns=ns, fail=fail)
 
-def bpftool_prog_list(expected=None):
-    _, progs = bpftool("prog show", JSON=True, fail=True)
+def bpftool_prog_list(expected=None, ns=""):
+    _, progs = bpftool("prog show", JSON=True, ns=ns, fail=True)
     if expected is not None:
         if len(progs) != expected:
             fail(True, "%d BPF programs loaded, expected %d" %
@@ -158,13 +164,13 @@ def bpftool_prog_list_wait(expected=0, n_retry=20):
         time.sleep(0.05)
     raise Exception("Time out waiting for program counts to stabilize want %d, have %d" % (expected, nprogs))
 
-def ip(args, force=False, JSON=True, fail=True):
+def ip(args, force=False, JSON=True, ns="", fail=True):
     if force:
         args = "-force " + args
-    return tool("ip", args, {"json":"-j"}, JSON=JSON, fail=fail)
+    return tool("ip", args, {"json":"-j"}, JSON=JSON, ns=ns, fail=fail)
 
-def tc(args, JSON=True, fail=True):
-    return tool("tc", args, {"json":"-p"}, JSON=JSON, fail=fail)
+def tc(args, JSON=True, ns="", fail=True):
+    return tool("tc", args, {"json":"-p"}, JSON=JSON, ns=ns, fail=fail)
 
 def ethtool(dev, opt, args, fail=True):
     return cmd("ethtool %s %s %s" % (opt, dev["ifname"], args), fail=fail)
@@ -178,6 +184,15 @@ def bpf_pinned(name):
 def bpf_bytecode(bytecode):
     return "bytecode \"%s\"" % (bytecode)
 
+def mknetns(n_retry=10):
+    for i in range(n_retry):
+        name = ''.join([random.choice(string.ascii_letters) for i in range(8)])
+        ret, _ = ip("netns add %s" % (name), fail=False)
+        if ret == 0:
+            netns.append(name)
+            return name
+    return None
+
 class DebugfsDir:
     """
     Class for accessing DebugFS directories as a dictionary.
@@ -237,6 +252,8 @@ class NetdevSim:
         self.dev = self._netdevsim_create()
         devs.append(self)
 
+        self.ns = ""
+
         self.dfs_dir = '/sys/kernel/debug/netdevsim/%s' % (self.dev['ifname'])
         self.dfs_refresh()
 
@@ -257,7 +274,7 @@ class NetdevSim:
 
     def remove(self):
         devs.remove(self)
-        ip("link del dev %s" % (self.dev["ifname"]))
+        ip("link del dev %s" % (self.dev["ifname"]), ns=self.ns)
 
     def dfs_refresh(self):
         self.dfs = DebugfsDir(self.dfs_dir)
@@ -285,6 +302,11 @@ class NetdevSim:
             time.sleep(0.05)
         raise Exception("Time out waiting for program counts to stabilize want %d/%d, have %d bound, %d loaded" % (bound, total, nbound, nprogs))
 
+    def set_ns(self, ns):
+        name = "1" if ns == "" else ns
+        ip("link set dev %s netns %s" % (self.dev["ifname"], name), ns=self.ns)
+        self.ns = ns
+
     def set_mtu(self, mtu, fail=True):
         return ip("link set dev %s mtu %d" % (self.dev["ifname"], mtu),
                   fail=fail)
@@ -372,6 +394,8 @@ def clean_up():
         dev.remove()
     for f in files:
         cmd("rm -f %s" % (f))
+    for ns in netns:
+        cmd("ip netns delete %s" % (ns))
 
 def pin_prog(file_name, idx=0):
     progs = bpftool_prog_list(expected=(idx + 1))
@@ -381,6 +405,35 @@ def pin_prog(file_name, idx=0):
 
     return file_name, bpf_pinned(file_name)
 
+def check_dev_info(other_ns, ns, pin_file=None, removed=False):
+    if removed:
+        bpftool_prog_list(expected=0)
+        ret, err = bpftool("prog show pin %s" % (pin_file), fail=False)
+        fail(ret == 0, "Showing prog with removed device did not fail")
+        fail(err["error"].find("No such device") == -1,
+             "Showing prog with removed device expected ENODEV, error is %s" %
+             (err["error"]))
+        return
+    progs = bpftool_prog_list(expected=int(not removed), ns=ns)
+    prog = progs[0]
+
+    fail("dev" not in prog.keys(), "Device parameters not reported")
+    dev = prog["dev"]
+    fail("ifindex" not in dev.keys(), "Device parameters not reported")
+    fail("ns_dev" not in dev.keys(), "Device parameters not reported")
+    fail("ns_inode" not in dev.keys(), "Device parameters not reported")
+
+    if not removed and not other_ns:
+        fail("ifname" not in dev.keys(), "Ifname not reported")
+        fail(dev["ifname"] != sim["ifname"],
+             "Ifname incorrect %s vs %s" % (dev["ifname"], sim["ifname"]))
+    else:
+        fail("ifname" in dev.keys(), "Ifname is reported for other ns")
+        if removed:
+            fail(dev["ifindex"] != 0, "Device perameters not zero on removed")
+            fail(dev["ns_dev"] != 0, "Device perameters not zero on removed")
+            fail(dev["ns_inode"] != 0, "Device perameters not zero on removed")
+
 # Parse command line
 parser = argparse.ArgumentParser()
 parser.add_argument("--log", help="output verbose log to given file")
@@ -417,6 +470,12 @@ for s in samples:
     skip(ret != 0, "sample %s/%s not found, please compile it" %
          (bpf_test_dir, s))
 
+# Check if net namespaces seem to work
+ns = mknetns()
+skip(ns is None, "Could not create a net namespace")
+cmd("ip netns delete %s" % (ns))
+netns = []
+
 try:
     obj = bpf_obj("sample_ret0.o")
     bytecode = bpf_bytecode("1,6 0 0 4294967295,")
@@ -549,6 +608,8 @@ try:
     progs = bpftool_prog_list(expected=1)
     fail(ipl["xdp"]["prog"]["id"] != progs[0]["id"],
          "Loaded program has wrong ID")
+    fail("dev" in progs[0].keys(),
+         "Device parameters reported for non-offloaded program")
 
     start_test("Test XDP prog replace with bad flags...")
     ret, _ = sim.set_xdp(obj, "offload", force=True, fail=False)
@@ -673,6 +734,35 @@ try:
     fail(time_diff < delay_sec, "Removal process took %s, expected %s" %
          (time_diff, delay_sec))
 
+    # Remove all pinned files and reinstantiate the netdev
+    clean_up()
+    bpftool_prog_list_wait(expected=0)
+
+    sim = NetdevSim()
+    sim.set_ethtool_tc_offloads(True)
+    sim.set_xdp(obj, "offload")
+
+    start_test("Test bpftool bound info reporting (own ns)...")
+    check_dev_info(False, "")
+
+    start_test("Test bpftool bound info reporting (other ns)...")
+    ns = mknetns()
+    sim.set_ns(ns)
+    check_dev_info(True, "")
+
+    start_test("Test bpftool bound info reporting (remote ns)...")
+    check_dev_info(False, ns)
+
+    start_test("Test bpftool bound info reporting (back to own ns)...")
+    sim.set_ns("")
+    check_dev_info(False, "")
+
+    pin_file, _ = pin_prog("/sys/fs/bpf/tmp")
+    sim.remove()
+
+    start_test("Test bpftool bound info reporting (removed dev)...")
+    check_dev_info(True, "", pin_file=pin_file, removed=True)
+
     print("%s: OK" % (os.path.basename(__file__)))
 
 finally: