Fix network configuration changes through collective

Some may find switching VTs to be more intuitive
than tmux session management. Provide one extra
on tty2 for such a scenario.

confluent_client/bin/nodemedia

@@ -187,5 +187,7 @@ def main():
         argparser.print_help()
         sys.exit(1)
     handler(noderange, media)
+
 if __name__ == '__main__':
     main()
+    sys.exit(exitcode)

confluent_client/bin/nodesetboot

@@ -32,7 +32,7 @@ if path.startswith('/opt'):
 import confluent.client as client
 
 argparser = optparse.OptionParser(
-    usage='Usage: %prog [options] noderange [default|cd|network|setup|hd|floppy]')
+    usage='Usage: %prog [options] noderange [default|cd|network|setup|hd|usb|floppy]')
 argparser.add_option('-b', '--bios', dest='biosmode',
                      action='store_true', default=False,
                      help='Request BIOS style boot (rather than UEFI)')

confluent_client/confluent_env.sh

@@ -91,7 +91,7 @@ _confluent_nodeidentify_completion()
 
 _confluent_nodesetboot_completion()
 {
-    COMP_CANDIDATES=("default,cd,network,setup,hd,floppy -h -b -p")
+    COMP_CANDIDATES=("default,cd,network,setup,hd,floppy,usb -h -b -p")
     _confluent_generic_completion
 }
 

confluent_client/doc/man/nodesetboot.ronn

@@ -4,7 +4,7 @@ nodesetboot(8) -- Check or set next boot device for noderange
 ## SYNOPSIS
 
 `nodesetboot <noderange>`  
-`nodesetboot [options] <noderange> [default|cd|network|setup|hd|floppy]`
+`nodesetboot [options] <noderange> [default|cd|network|setup|hd|usb|floppy]`
 
 ## DESCRIPTION
 
@@ -46,6 +46,10 @@ control.
   Request boot from floppy.  Generally speaking firmware uses this to mean a USB
   flash drive or similar (whether virtual or physical).
 
+* `usb`:
+  Request boot from usb.  Generally speaking firmware uses this to mean a USB
+  flash drive or similar (whether virtual or physical).
+
 * `network`:
   Request boot to network
   

confluent_osdeploy/el7/profiles/default/scripts/infiniband/mofed.post

@@ -11,7 +11,7 @@ if lspci -d 15b3:: -n |grep 15b3 > /dev/null; then
     #mkdir MLNX_OFED
     #mount -o loop ofed.iso MLNX_OFED
     fetch_remote infiniband/mofed.tgz
-    tar xf mofed.tgz
+    tar xf infiniband/mofed.tgz
     # The rest is common between tar and iso
     cd MLNX_OFED*
     mount -o loop ofed

confluent_osdeploy/el8/initramfs/usr/lib/dracut/hooks/initqueue/01-confluent.sh

@@ -109,5 +109,13 @@ done < /etc/confluent/confluent.deploycfg
 if [ -e /lib/nm-lib.sh ]; then
     . /lib/nm-lib.sh
     nm_generate_connections
+    if [[ "$ifname" == ib* ]]; then
+        sed -i s/type=ethernet/type=infiniband/ /run/NetworkManager/system-connections/$ifname.nmconnection
+        if ! grep '\[infiniband\]' /run/NetworkManager/system-connections/$ifname.nmconnection > /dev/null; then
+            echo >> /run/NetworkManager/system-connections/$ifname.nmconnection
+            echo '[infiniband]' >> /run/NetworkManager/system-connections/$ifname.nmconnection
+            echo transport-mode=datagram >> /run/NetworkManager/system-connections/$ifname.nmconnection
+        fi
+    fi
 fi
 

confluent_osdeploy/el8/profiles/default/scripts/infiniband/mofed.post

@@ -11,7 +11,7 @@ if lspci -d 15b3:: -n |grep 15b3 > /dev/null; then
     #mkdir MLNX_OFED
     #mount -o loop ofed.iso MLNX_OFED
     fetch_remote infiniband/mofed.tgz
-    tar xf mofed.tgz
+    tar xf infiniband/mofed.tgz
     # The rest is common between tar and iso
     cd MLNX_OFED*
     mount -o loop ofed

confluent_osdeploy/esxi7/profiles/hypervisor/kickstart

@@ -4,3 +4,8 @@ install --firstdisk --overwritevmfs
 %include /tmp/ksnet
 %include /tmp/rootpw
 reboot
+%post
+localcli network firewall unload
+STATUP=$(mktemp)
+echo '{"status": "complete"}' > $STATUP
+/opt/confluent/bin/apiclient /confluent-api/self/updatestatus $STATUP

confluent_osdeploy/genesis/initramfs/opt/confluent/bin/rungenesis

@@ -8,14 +8,16 @@ if ! grep console= /proc/cmdline >& /dev/null; then
     echo $autocons > /tmp/01-autocons.devnode
     if [ ! -z "$autocons" ]; then
         echo "Using $(cat /tmp/01-autocons.conf)"
-        tmux a <> $autocons >&0 2>&1 &
+        (while :; do tmux a <> $autocons >&0 2>&1; done) &
+        (while :; do TERM=linux tmux a <> /dev/tty1 >&0 2>&1; done) &
     else
-        tmux a <> /dev/console >&0 2>&1 &
+        (while :; do tmux a <> /dev/console >&0 2>&1; done) &
     fi
 else
-    tmux a <> /dev/console >&0 2>&1 &
-    tmux a <> /dev/tty1 >&0 2>&1 &
+    (while :; do tmux a <> /dev/console >&0 2>&1; done) &
+    (while :; do TERM=linux tmux a <> /dev/tty1 >&0 2>&1; done) &
 fi
+(while :; do TERM=linux tmux <> /dev/tty2 >&0 2>&1; done) &
 echo -n "udevd: "
 /usr/lib/systemd/systemd-udevd --daemon
 echo -n "Loading drivers..."

confluent_osdeploy/genesis/profiles/default/profile.yaml

@@ -1,2 +1,2 @@
 label: Genesis
-kernelarags: quiet
+kernelargs: quiet

confluent_osdeploy/genesis/profiles/default/scripts/configbmc

@@ -411,7 +411,7 @@ def main():
         dotwait()
     while awaitprefix and not check_subnet(s, bmccfg['prefixv4'], channel):
         dotwait()
-    while awaitprefix and not check_gateway(s, bmccfg['bmcgw'], channel):
+    while awaitgw and not check_gateway(s, bmccfg['bmcgw'], channel):
         dotwait()
     sys.stdout.write('done\n')
     sys.stdout.flush()

confluent_osdeploy/utils/autocons.c

@@ -1,3 +1,4 @@
+#include <errno.h>
 #include <termios.h>
 #include <sys/ioctl.h>
 #include <fcntl.h>
@@ -20,20 +21,29 @@
 
 int main(int argc, char* argv[]) {
     struct termios tty;
+    struct termios tty2;
+    struct winsize ws;
+    unsigned width, height;
     int ttyf;
-    int spcr;
+    int tmpi;
     int currspeed;
+    int flags;
     speed_t cspeed;
     char buff[128];
+    int bufflen;
+    fd_set set;
+    struct timeval timeout;
     char* offset;
     uint64_t address;
-    spcr = open("/sys/firmware/acpi/tables/SPCR", O_RDONLY);
-    if (spcr < 0) {
+    bufflen = 0;
+    tmpi = open("/sys/firmware/acpi/tables/SPCR", O_RDONLY);
+    if (tmpi < 0) {
         exit(0);
     }
-    if (read(spcr, buff, 80) < 80) {
+    if (read(tmpi, buff, 80) < 80) {
         exit(0);
     }
+    close(tmpi);
     if (buff[8] != 2) exit(0); //revision 2
     if (buff[36] != 0) exit(0); //16550 only
     if (buff[40] != 1) exit(0); //IO only
@@ -69,12 +79,49 @@ int main(int argc, char* argv[]) {
     }
     tcgetattr(ttyf, &tty);
     if (cspeed) {
-        cfsetospeed(&tty, B115200);
-        cfsetispeed(&tty, B115200);
+        cfsetospeed(&tty, cspeed);
+        cfsetispeed(&tty, cspeed);
     }
+    printf("%s\n", buff);
+    tcgetattr(ttyf, &tty2);
+    cfmakeraw(&tty2);
+    tcsetattr(ttyf, TCSANOW, &tty2);
+    flags = fcntl(ttyf, F_GETFL, 0);
+    fcntl(ttyf, F_SETFL, flags | O_NONBLOCK);
+    while (read(ttyf, buff, 64) > 0) {
+        // Drain any pending reads
+    }
+    timeout.tv_sec = 0;
+    timeout.tv_usec = 500000;
+    FD_ZERO(&set);
+    FD_SET(ttyf, &set);
+    write(ttyf, "\0337\033[999;999H\033[6n\0338", 18);
+    while (select(ttyf + 1, &set, NULL, NULL, &timeout) > 0) {
+        if ((tmpi = read(ttyf, buff + bufflen, 127 - bufflen)) < 0) {
+            if (errno == EAGAIN || errno == EWOULDBLOCK) {
+                continue;
+            } else {
+                break;
+            }
+        }
+        bufflen += tmpi;
+        buff[bufflen] = 0;
+        if (strchr(buff, 'R')) {
+            break;
+        }
+    }
+    fcntl(ttyf, F_SETFL, flags);
+    ws.ws_xpixel = 0;
+    ws.ws_ypixel = 0;
+    if (sscanf(buff, "\033[%u;%uR", &height, &width) == 2) {
+        ws.ws_col = width;
+        ws.ws_row = height;
+    } else {
+        ws.ws_col = 100;
+        ws.ws_row = 31;
+    }
+    ioctl(ttyf, TIOCSWINSZ, &ws);
     tcsetattr(ttyf, TCSANOW, &tty);
     ioctl(ttyf, TIOCCONS, 0);
-    printf("%s\n", buff);
-
 }
 

confluent_server/bin/confluentdbutil

@@ -74,7 +74,7 @@ if args[0] == 'restore':
             for targdir in os.walk('/etc/confluent'):
                 os.chown(targdir[0], owner, group)
                 for f in targdir[2]:
-                    os.chown(os.patht.join(targdir[0], f), owner, group)
+                    os.chown(os.path.join(targdir[0], f), owner, group)
     except Exception as e:
         print(str(e))
         sys.exit(1)

confluent_server/confluent/discovery/core.py

@@ -811,13 +811,17 @@ def get_smm_neighbor_fingerprints(smmaddr, cv):
     if ':' in smmaddr:
         smmaddr = '[{0}]'.format(smmaddr)
     wc = webclient.SecureHTTPConnection(smmaddr, verifycallback=cv)
-    neighs = wc.grab_json_response('/scripts/neighdata.json')
+    try:
+        neighs = wc.grab_json_response('/scripts/neighdata.json')
+    except Exception:
+        log.log({'error': 'Failure getting LLDP information from {}'.format(smmaddr)})
+        return
     if not neighs:
         return
-    for idx in (4, 5):
-        if 'sha256' not in neighs[idx]:
+    for neigh in neighs:
+        if 'sha256' not in neigh:
             continue
-        yield 'sha256$' + b64tohex(neighs[idx]['sha256'])
+        yield 'sha256$' + b64tohex(neigh['sha256'])
 
 
 def get_nodename(cfg, handler, info):

confluent_server/confluent/discovery/protocols/slp.py

@@ -19,6 +19,7 @@ import confluent.util as util
 import confluent.log as log
 import os
 import random
+import eventlet.greenpool
 import eventlet.green.select as select
 import eventlet.green.socket as socket
 import struct
@@ -106,6 +107,8 @@ def _parse_slp_packet(packet, peer, rsps, xidmap):
     if '%' in addr:
         addr = addr[:addr.index('%')]
     mac = None
+    if addr not in neighutil.neightable:
+        neighutil.update_neigh()
     if addr in neighutil.neightable:
         identifier = neighutil.neightable[addr]
         mac = identifier
@@ -532,6 +535,8 @@ def active_scan(handler, protocol=None):
     for scanned in scan():
         for addr in scanned['addresses']:
             ip = addr[0].partition('%')[0]  # discard scope if present
+            if ip not in neighutil.neightable:
+                neighutil.update_neigh()
             if ip not in neighutil.neightable:
                 continue
             if addr in known_peers:
@@ -583,6 +588,8 @@ def scan(srvtypes=_slp_services, addresses=None, localonly=False):
         # reduced chance of many responses overwhelming receive buffer.
     _grab_rsps((net, net4), rsps, 1, xidmap)
     # now to analyze and flesh out the responses
+    handleids = set([])
+    gp = eventlet.greenpool.GreenPool(128)
     for id in rsps:
         for srvurl in rsps[id].get('urls', ()):
             if len(srvurl) > 4:
@@ -597,7 +604,10 @@ def scan(srvtypes=_slp_services, addresses=None, localonly=False):
                     break
             else:
                 continue
-        _add_attributes(rsps[id])
+        gp.spawn_n(_add_attributes, rsps[id])
+        handleids.add(id)
+    gp.waitall()
+    for id in handleids:
         if 'service:lighttpd' in rsps[id]['services']:
             currinf = rsps[id]
             curratt = currinf.get('attributes', {})

confluent_server/confluent/discovery/protocols/ssdp.py

@@ -62,6 +62,8 @@ def active_scan(handler, protocol=None):
     for scanned in scan(['urn:dmtf-org:service:redfish-rest:1']):
         for addr in scanned['addresses']:
             ip = addr[0].partition('%')[0]  # discard scope if present
+            if ip not in neighutil.neightable:
+                neighutil.update_neigh()
             if ip not in neighutil.neightable:
                 continue
             if addr in known_peers:
@@ -311,6 +313,8 @@ def _parse_ssdp(peer, rsp, peerdata):
     ip = peer[0].partition('%')[0]
     nid = ip
     mac = None
+    if ip not in neighutil.neightable:
+        neighutil.update_neigh()
     if ip in neighutil.neightable:
         nid = neighutil.neightable[ip]
         mac = nid

confluent_server/confluent/messages.py

@@ -1054,7 +1054,7 @@ class InputNetworkConfiguration(ConfluentInputMessage):
         if 'ipv4_gateway' not in inputdata:
             inputdata['ipv4_gateway'] = None
 
-        if 'ipv4_configuration' in inputdata:
+        if 'ipv4_configuration' in inputdata and inputdata['ipv4_configuration']:
             if inputdata['ipv4_configuration'].lower() not in ['dhcp','static']:
                 raise exc.InvalidArgumentException(
                                             'Unrecognized ipv4_configuration')
@@ -1147,6 +1147,7 @@ class BootDevice(ConfluentChoiceMessage):
         'default',
         'cd',
         'floppy',
+        'usb',
     ])
 
     valid_bootmodes = set([

confluent_server/confluent/netutil.py

@@ -82,8 +82,11 @@ _idxtoifnamemap = {}
 def _rebuildidxmap():
     _idxtoifnamemap.clear()
     for iname in os.listdir('/sys/class/net'):
-        ci = int(open('/sys/class/net/{0}/ifindex'.format(iname)).read())
-        _idxtoifnamemap[ci] = iname
+        try:
+            ci = int(open('/sys/class/net/{0}/ifindex'.format(iname)).read())
+            _idxtoifnamemap[ci] = iname
+        except Exception:  # there may be non interface in /sys/class/net
+            pass
 
 
 def myiptonets(svrip):

confluent_server/confluent/networking/lldp.py

@@ -139,7 +139,10 @@ def get_fingerprint(switch, port, configmanager, portmatch):
             continue
         if info.get('switch', None) != switch:
             continue
-        if portmatch(info.get('port'), port):
+        if portmatch(info.get('portid', None), port):
+            return ('sha256$' + b64tohex(info['peersha256fingerprint']),
+                    info.get('verified', False))
+        elif portmatch(info.get('port', None), port):
             return ('sha256$' + b64tohex(info['peersha256fingerprint']),
                     info.get('verified', False))
     return None, False
@@ -235,19 +238,23 @@ def _extract_neighbor_data_b(args):
         idx = oidindex[0][-1]
         idxtoifname[idx] = _lldpdesc_to_ifname(sid, idx, str(oidindex[1]))
     for remotedesc in conn.walk('1.0.8802.1.1.2.1.4.1.1.10'):
-        iname = idxtoifname[remotedesc[0][-2]]
+        iname = idxtoifname.get(remotedesc[0][-2],
+                                idxtoportid[remotedesc[0][-2]])
         _init_lldp(lldpdata, iname, remotedesc[0][-2], idxtoportid, switch)
         _extract_extended_desc(lldpdata[iname], remotedesc[1], user)
     for remotename in conn.walk('1.0.8802.1.1.2.1.4.1.1.9'):
-        iname = idxtoifname[remotename[0][-2]]
+        iname = idxtoifname.get(remotename[0][-2],
+                                idxtoportid[remotename[0][-2]])
         _init_lldp(lldpdata, iname, remotename[0][-2], idxtoportid, switch)
         lldpdata[iname]['peername'] = str(remotename[1])
     for remotename in conn.walk('1.0.8802.1.1.2.1.4.1.1.7'):
-        iname = idxtoifname[remotename[0][-2]]
+        iname = idxtoifname.get(remotename[0][-2],
+                                idxtoportid[remotename[0][-2]])
         _init_lldp(lldpdata, iname, remotename[0][-2], idxtoportid, switch)
         lldpdata[iname]['peerportid'] = sanitize(remotename[1])
     for remoteid in conn.walk('1.0.8802.1.1.2.1.4.1.1.5'):
-        iname = idxtoifname[remoteid[0][-2]]
+        iname = idxtoifname.get(remoteid[0][-2],
+                                idxtoportid[remoteid[0][-2]])
         _init_lldp(lldpdata, iname, remoteid[0][-2], idxtoportid, switch)
         lldpdata[iname]['peerchassisid'] = sanitize(remoteid[1])
     for entry in lldpdata:

confluent_server/confluent/networking/macmap.py

@@ -82,6 +82,8 @@ _blacklistnames = (
 
 
 def _namesmatch(switchdesc, userdesc):
+    if switchdesc is None:
+        return False
     if switchdesc == userdesc:
         return True
     try:
@@ -217,13 +219,18 @@ def _map_switch_backend(args):
                 *([int(x) for x in oid[-6:]])
             )
             mactobridge[macaddr] = int(bridgeport)
-    #ciscoiftovlanmap = {}
     vlanstocheck = set([])
-    for vb in conn.walk('.1.3.6.1.4.1.9.9.68.1.2.2.1.2'):
-        vlanstocheck.add(vb[1])
-    #ciscotrunktovlanmap = {}
-    for vb in conn.walk('.1.3.6.1.4.1.9.9.46.1.6.1.1.5'):
-        vlanstocheck.add(vb[1])
+    try:
+        #ciscoiftovlanmap = {}
+        for vb in conn.walk('.1.3.6.1.4.1.9.9.68.1.2.2.1.2'):
+            vlanstocheck.add(vb[1])
+        #ciscotrunktovlanmap = {}
+        for vb in conn.walk('.1.3.6.1.4.1.9.9.46.1.6.1.1.5'):
+            vlanstocheck.add(vb[1])
+    except Exception:
+        # We might have crashed snmp on a non-cisco switch
+        # in such a case, delay 8 seconds to allow recovery to complete
+        eventlet.sleep(8)
     if not vlanstocheck:
         vlanstocheck.add(None)
     bridgetoifmap = {}