* [sonic-buildimage] Changes to make network specific sysctl
common for both host and docker namespace (in multi-npu).
This change is triggered with issue found in multi-npu platforms
where in docker namespace
net.ipv6.conf.all.forwarding was 0 (should be 1) because of
which RS/RA message were triggered and link-local router were learnt.
Beside this there were some other sysctl.net.ipv6* params whose value
in docker namespace is not same as host namespace.
So to make we are always in sync in host and docker namespace
created common file that list all sysctl.net.* params and used
both by host and docker namespace. Any change will get applied
to both namespace.
Signed-off-by: Abhishek Dosi <abdosi@microsoft.com>
* Address Review Comments and made sure to invoke augtool
only one and do string concatenation of all set commands
* Address Review Comments.
* [systemd-generator]: Fix the code to make sure that dependencies
of host services are generated correctly for multi-asic platforms.
Add code to make sure that systemd timer files are also modified
to add the correct service dependency for multi-asic platforms.
Signed-off-by: SuvarnaMeenakshi <sumeenak@microsoft.com>
* [systemd-generator]: Minor fix, remove debug code and
remove unused variable.
* Changes to make default route programming
correct in multi-asic platform where frr is not running
in host namespace. Change is to set correct administrative distance.
Also make NAMESPACE* enviroment variable available for all dockers
so that it can be used when needed.
Signed-off-by: Abhishek Dosi <abdosi@microsoft.com>
* Fix review comments
* Review comment to check to add default route
only if default route exist and delete is successful.
Add changes for syslog support for containers running in namespaces on multi ASIC platforms.
On Multi ASIC platforms
Rsyslog service is only running on the host. There is no rsyslog service running in each namespace.
On multi ASIC platforms the rsyslog service on the host will be listening on the docker0 ip address instead of loopback address.
The rsyslog.conf on the containers is modified to have omfwd target ip to be docker0 ipaddress instead of loopback ip
Signed-off-by: Arvindsrinivasan Lakshmi Narasimhan <arlakshm@microsoft.com>
While testing reboot case for 201911 facing error:
supervisor-proc-exit-listener FATAL command at '/usr/bin/supervisor-proc-exit-listener' is not executable
Signed-off-by: Roman Savchuk <romanx.savchuk@intel.com>
When building the SONiC image, used systemd to mask all services which are set to "disabled" in init_cfg.json.
This PR depends on https://github.com/Azure/sonic-utilities/pull/944, otherwise `config load_minigraph will fail when trying to restart disabled services.
- Ensure all features (services) are in the configured state when hostcfgd starts
- Better functionalization of code
- Also replace calls to deprecated `has_key()` method in `tacacs_server_handler()` and `tacacs_global_handler()` with `in` keyword.
This PR depends on https://github.com/Azure/sonic-utilities/pull/944, otherwise `config load_minigraph` will fail when trying to restart disabled services.
While migrating to SONiC 20181130, identified a couple of issues:
1. union-mount needs /host/machine.conf parameters for vendor specific checks : however, in case of migration, the /host/machine.conf is extracted from ONIE only in https://github.com/Azure/sonic-buildimage/blob/master/files/image_config/platform/rc.local#L127.
2. Since grub.cfg is updated to have net.ifnames=0 biosdevname=0, 70-persistent-net.rules changes are no longer required.
Don't limit iptables connection tracking to TCP protocol; allow connection tracking for all protocols. This allows services like NTP, which is UDP-based, to receive replies from an NTP server even if the port is blocked, as long as it is in reply to a request sent from the device itself.
* Fix the Build on 201911 (Stretch) where the directory
/usr/lib/systemd/system/ does not exist so creating
manually. Change should not harm Master (buster) where
the directory is created by Linux
* Fix as per review comments
**- Why I did it**
When I tested auto-restart feature of swss container by manually killing one of critical processes in it, swss will be stopped. Then syncd container as the peer container should also be
stopped as expected. However, I found sometimes syncd container can be stopped, sometimes
it can not be stopped. The reason why syncd container can not be stopped is the process
(/usr/local/bin/syncd.sh stop) to execute the stop() function will be stuck between the lines 164 –167. Systemd will wait for 90 seconds and then kill this process.
164 # wait until syncd quit gracefully
165 while docker top syncd$DEV | grep -q /usr/bin/syncd; do
166 sleep 0.1
167 done
The first thing I did is to profile how long this while loop will spin if syncd container can be
normally stopped after swss container is stopped. The result is 5 seconds or 6 seconds. If syncd
container can be normally stopped, two messages will be written into syslog:
str-a7050-acs-3 NOTICE syncd#dsserve: child /usr/bin/syncd exited status: 134
str-a7050-acs-3 INFO syncd#supervisord: syncd [5] child /usr/bin/syncd exited status: 134
The second thing I did was to add a timer in the condition of while loop to ensure this while loop will be forced to exit after 20 seconds:
After that, the testing result is that syncd container can be normally stopped if swss is stopped
first. One more thing I want to mention is that if syncd container is stopped during 5 seconds or 6 seconds, then the two log messages can be still seen in syslog. However, if the execution
time of while loop is longer than 20 seconds and is forced to exit, although syncd container can be stopped, I did not see these two messages in syslog. Further, although I observed the auto-restart feature of swss container can work correctly right now, I can not make sure the issue which syncd container can not stopped will occur in future.
**- How I did it**
I added a timer around the while loop in stop() function. This while loop will exit after spinning
20 seconds.
Signed-off-by: Yong Zhao <yozhao@microsoft.com>
**- Why I did it**
To ensure telemetry service is enabled by default after installing a fresh SONiC image
**- How I did it**
Set telemetry feature status to "enabled" when generating init_cfg.json file
Found another syncd timing issue related to clock going backwards.
To be safe disable the ntp long jump.
Signed-off-by: Ying Xie <ying.xie@microsoft.com>
Fix hostcfgd so that changes to the "FEATURE" table in ConfigDB are properly handled. Three changes here:
1. Fix indenting such that the handling of each key actually occurs in the for key in status_data.keys(): loop
2. Add calls to sudo systemctl mask and sudo systemctl unmask as appropriate to ensure changes persist across reboots
3. Substitute returns with continues so that even if one service fails, we still try to handle the others
Note that the masking is persistent, even if the configuration is not saved. We may want to consider only calling systemctl enable/disable in hostcfgd when the DB table changes, and only call systemctl mask/unmask upon calling config save.
* Run fsck filesystem check support prior mounting filesystem
If the filesystem become non clean ("dirty"), SONiC does not run fsck to
repair and mark it as clean again.
This patch adds the functionality to run fsck on each boot, prior to the
filesystem being mounted. This allows the filesystem to be repaired if
needed.
Note that if the filesystem is maked as clean, fsck does nothing and simply
return so this is perfectly fine to call fsck every time prior to mount the
filesystem.
How to verify this patch (using bash):
Using an image without this patch:
Make the filesystem "dirty" (not clean)
[we are making the assumption that filesystem is stored in /dev/sda3 - Please adjust depending of the platform]
[do this only on a test platform!]
dd if=/dev/sda3 of=superblock bs=1 count=2048
printf "$(printf '\\x%02X' 2)" | dd of="superblock" bs=1 seek=1082 count=1 conv=notrunc &> /dev/null
dd of=/dev/sda3 if=superblock bs=1 count=2048
Verify that filesystem is not clean
tune2fs -l /dev/sda3 | grep "Filesystem state:"
reboot and verify that the filesystem is still not clean
Redo the same test with an image with this patch, and verify that at next reboot the filesystem is repaired and becomes clean.
fsck log is stored on syslog, using the string FSCK as markup.
I found that with IPv4Network types, calling list(ip_ntwrk.hosts()) is reliable. However, when doing the same with an IPv6Network, I found that the conversion to a list can hang indefinitely. This appears to me to be a bug in the ipaddress.IPv6Network implementation. However, I could not find any other reports on the web.
This patch changes the behavior to call next() on the ip_ntwrk.hosts() generator instead, which returns the IP address of the first host.
Since the introduction of VRF, interface-related tables in ConfigDB will have multiple entries, one of which only contains the interface name and no IP prefix. Thus, when iterating over the keys in the tables, we need to ignore the entries which do not contain IP prefixes.
Modified caclmgrd behavior to enhance control plane security as follows:
Upon starting or receiving notification of ACL table/rule changes in Config DB:
1. Add iptables/ip6tables commands to allow all incoming packets from established TCP sessions or new TCP sessions which are related to established TCP sessions
2. Add iptables/ip6tables commands to allow bidirectional ICMPv4 ping and traceroute
3. Add iptables/ip6tables commands to allow bidirectional ICMPv6 ping and traceroute
4. Add iptables/ip6tables commands to allow all incoming Neighbor Discovery Protocol (NDP) NS/NA/RS/RA messages
5. Add iptables/ip6tables commands to allow all incoming IPv4 DHCP packets
6. Add iptables/ip6tables commands to allow all incoming IPv6 DHCP packets
7. Add iptables/ip6tables commands to allow all incoming BGP traffic
8. Add iptables/ip6tables commands for all ACL rules for recognized services (currently SSH, SNMP, NTP)
9. For all services which we did not find configured ACL rules, add iptables/ip6tables commands to allow all incoming packets for those services (allows the device to accept SSH connections before the device is configured)
10. Add iptables rules to drop all packets destined for loopback interface IP addresses
11. Add iptables rules to drop all packets destined for management interface IP addresses
12. Add iptables rules to drop all packets destined for point-to-point interface IP addresses
13. Add iptables rules to drop all packets destined for our VLAN interface gateway IP addresses
14. Add iptables/ip6tables commands to allow all incoming packets with TTL of 0 or 1 (This allows the device to respond to tools like tcptraceroute)
15. If we found control plane ACLs in the configuration and applied them, we lastly add iptables/ip6tables commands to drop all other incoming packets
* [ntp] enable/disable NTP long jump according to reboot type
- Enable NTP long jump after cold reboot.
- Disable NTP long jump after warrm/fast reboot.
Signed-off-by: Ying Xie <ying.xie@microsoft.com>
* fix typo
* further refactoring
* use sonic-db-cli instead
* Changes to support config-setup service for multi-npu
platforms. For Multi-npu we are not supporting as of
now config initializtion and ZTP. It will support creating
config db from minigraph or using config db from previous
file system
Signed-off-by: Abhishek Dosi <abdosi@microsoft.com>
* Address Review Comments.
* Address Review comments
* Address Review Comments of using pyhton based config load_minigraph/
config save/config reload from shell scripts so that we don't duplicate
code. Also while running from shell we will skip stop/start services
done by those commands.
* Updated to use python command so no code duplication.
and then we load image and reboot even if there was existing
config_db.json we will look for DHCP Service. we should disbale
update_graph in such cases. This behaviour is silimar to what we have in
201811 image.
* Changes for LLDP for Multi NPU Platoforms:-
a) Enable LLDP for Host namespace for Management Port
b) Make sure Management IP is avaliable in per asic namespace
needed for LLDP Chassis configuration
c) Make sure chassis mac-address is correct in per asic namespace
d) Do not run lldp on eth0 of per asic namespace and avoid chassis
configuration for same
e) Use Linux hostname instead from Device Metadata for lldp chassis
configuration since in multi-npu platforms device metadata hostname
will be differnt
Signed-off-by: Abhishek Dosi <abdosi@microsoft.com>
* Address Review Comment with following changes:
a) Use Device Metadata hostname even in per namespace conatiner.
updated minigraph parsing for same to have hostname as system
hostname and add new key for asic name
b) Minigraph changes to have MGMT_INTERFACE Key in per asic/namespace
config also as needed for LLDP for setting chassis management IP.
Signed-off-by: Abhishek Dosi <abdosi@microsoft.com>
* Address Review Comments
* Multi DB with namespace support, Introducing the database_global.json file
for supporting accessing DB's in other namespaces for service running in
linux host
* Updates based on comments
* Adding the j2 templates for database_config and database_global files.
* Updating to retrieve the redis DIR's to be mounted from database_global.json file.
* Additional check to see if asic.conf file exists before sourcing it.
* Updates based on PR comments discussion.
* Review comments update
* Updates to the argument "-n" for namespace used in both context of parsing minigraph and multi DB access.
* Update with the attribute "persistence_for_warm_boot" that was added to database_config.json file earlier.
* Removing the database_config.json file to avioid confusion in future.
We use the database_config.json.j2 file to generate database_config.json files dynamically.
* Update the comments for sudo usage in docker_image_ctrl.j2
* Update with the new logic in PING PONG tests using sonic-db-cli. With this we wait till the
PONG response is received when redis server is up.
* Similar changes in swss and syncd scripts for the PING tests with sonic-db-cli
* Updated with a missing , in the database_config.json.j2 file, Do pip install of j2cli in docker-base-buster.
1. ebtables -t filter -A FORWARD -p 802_1Q --vlan-encap 0806 -j DROP
The ARP packet with vlan tag can't match the default rule.
Signed-off-by: wangshengjun <wangshengjun@asterfusion.com>
Dynamic threshold setting changed to 0 and WRED profile green min threshold set to 250000 for Tomahawk devices
Changed the dynamic threshold settings in pg_profile_lookup.ini
Added a macro for WRED profiles in qos.json.j2 for Tomahawk devices
Necessary changes made in qos.config.j2 to use the macro if present
Signed-off-by: Neetha John <nejo@microsoft.com>
The one big bgp configuration template was splitted into chunks.
Currently we have three types of bgp neighbor peers:
general bgp peers. They are represented by CONFIG_DB::BGP_NEIGHBOR table entries
dynamic bgp peers. They are represented by CONFIG_DB::BGP_PEER_RANGE table entries
monitors bgp peers. They are represented by CONFIG_DB::BGP_MONITORS table entries
This PR introduces three templates for each peer type:
bgp policies: represent policieas that will be applied to the bgp peer-group (ip prefix-lists, route-maps, etc)
bgp peer-group: represent bgp peer group which has common configuration for the bgp peer type and uses bgp routing policy from the previous item
bgp peer-group instance: represent bgp configuration, which will be used to instatiate a bgp peer-group for the bgp peer-type. Usually this one is simple, consist of the referral to the bgp peer-group, bgp peer description and bgp peer ip address.
This PR redefined constant.yml file. Now this file has a setting for to use or don't use bgp_neighbor metadata. This file has more parameters for now, which are not used. They will be used in the next iteration of bgpcfgd.
Currently all tests have been disabled. I'm going to create next PR with the tests right after this PR is merged.
I'm going to introduce better bgpcfgd in a short time. It will include support of dynamic changes for the templates.
FIX:: #4231
Install kubeadm, which transparently installs kubelet & kubectl
As well download required Kubernetes images required to run as kubernetes node.
The kubelet service is intentionally kept in disabled state, as it would otherwise
continuously restart wasting resources, until join to master.
sonic-netns-exec fails to execute below command in swss.sh:
sonic-netns-exec "$NET_NS" sonic-db-cli $1 EVAL "
local tables = {$2}
for i = 1, table.getn(tables) do
local matches = redis.call('KEYS', tables[i])
for j,name in ipairs(matches) do
redis.call('DEL', name)
end
end" 0
This command fails with error " redis.exceptions.ResponseError: value is not an integer or out of range" .
Root cause:
When sonic-netns-exec executes the above function, argument passed to sonic-db-cli is NOT executed as a single script.
The argument is passed as separate keywords to sonic-db-cli, as below:
['EVAL', 'local', 'tables', '=', "{'PORT_TABLE*'}", 'for', 'i', '=', '1,', 'table.getn(tables)', 'do', 'local', 'matches', '=', "redis.call('KEYS',", 'tables[i])', 'for', 'j,name', 'in', 'ipairs(matches)', 'do', "redis.call('DEL',", 'name)', 'end', 'end', '0']
- How I did it
To make sure that the parameters are passed as they were set initially, fix sonic-netns-exec to use double quoted "$@", where "$@" is "$1" "$2" "$3" ... "${N}"
After fix, the argument passed to sonic-db-cli is as below:
Argument passed to sonic-db-cli:
['EVAL', "\n local tables = {'PORT_TABLE*'}\n for i = 1, table.getn(tables) do\n local matches = redis.call('KEYS', tables[i])\n for j,name in ipairs(matches) do\n redis.call('DEL', name)\n end\n end", '0']
Signed-off-by: SuvarnaMeenakshi <sumeenak@microsoft.com>
* Fix the CMD for the PROCESSSTATS entries so that
there is a space between the command name and the
arguments.
Signed-off-by: Garrick He <garrick_he@dell.com>
Instead of updating hostname manualy on Config DB hostname change,
simply share containers UTS namespace with host OS.
Ideally, instead of setting `--uts=host` for every container in SONiC,
this setting can be set per container if feature requires.
One behaviour change is introduced in this commit, when `--privileged`
or `--cap-add=CAP_SYS_ADMIN` and `--uts=host` are combined, container
has privilege to change host OS and every other container hostname.
Such privilege should be fixed by limiting containers capabilities.
Signed-off-by: Stepan Blyschak <stepanb@mellanox.com>
- What I did
Add configuration to avoid ntpd from panic and exit if the drift between new time and current system time is large.
- How I did it
Added "tinker panic 0" in ntp.conf file.
- How to verify it
[this assumes that there is a valid NTP server IP in config_db/ntp.conf]
Change the current system time to a bad time with a large drift from time in ntp server; drift should be greater than 1000s.
Reboot the device.
Before the fix:
3. upon reboot, ntp-config service comes up fine, ntp service goes to active(exited) state without any error message. This is because the offset between new time (from ntp server) and the current system time is very large, ntpd goes to panic mode and exits. The system continues to show the bad time.
After the fix:
3. Upon reboot, ntp-config comes up fine, ntp services comes up from and stays in active (running) state. The system clock gets synced with the ntp server time.
