Network configuration: Difference between revisions
(→More about it: Wikipedia Kerneltrap entry: it is down for years. I wrote the 1st wayback machine I found instead. To my understanding, wayback archived that for more than one date. I didn't check the differences to their other dates records.) |
(→Local hostname resolution: add localhost to /etc/hosts for software that read the file directly and since it's also in Installation guide#Network configuration) |
||
Line 280: | Line 280: | ||
Some clients may however still rely on {{ic|/etc/hosts}}, see [https://lists.debian.org/debian-devel/2013/07/msg00809.html] [https://bugzilla.mozilla.org/show_bug.cgi?id=87717#c55] for examples. | Some clients may however still rely on {{ic|/etc/hosts}}, see [https://lists.debian.org/debian-devel/2013/07/msg00809.html] [https://bugzilla.mozilla.org/show_bug.cgi?id=87717#c55] for examples. | ||
To configure the hosts file, add the following | To configure the hosts file, add the following lines to {{ic|/etc/hosts}}: | ||
127.0.1.1 ''myhostname''.localdomain ''myhostname'' | 127.0.0.1 localhost.localdomain localhost | ||
::1 localhost.localdomain localhost | |||
127.0.1.1 ''myhostname''.localdomain ''myhostname'' | |||
{{Note|The order of hostnames/aliases that follow the IP address in {{ic|/etc/hosts}} is significant. The first string is considered the canonical hostname and may be appended with parent domains, where domain components are separated by a dot (ie. {{ic|.localdomain}} above). All following strings on the same line are considered aliases. See {{man|5|hosts}} for more info.}} | {{Note|The order of hostnames/aliases that follow the IP address in {{ic|/etc/hosts}} is significant. The first string is considered the canonical hostname and may be appended with parent domains, where domain components are separated by a dot (ie. {{ic|.localdomain}} above). All following strings on the same line are considered aliases. See {{man|5|hosts}} for more info.}} | ||
Line 289: | Line 291: | ||
{{hc|$ getent hosts| | {{hc|$ getent hosts| | ||
127.0.0.1 localhost | 127.0.0.1 localhost.localdomain localhost | ||
127.0.1.1 ''myhostname''.localdomain ''myhostname'' | 127.0.0.1 localhost.localdomain localhost | ||
127.0.1.1 ''myhostname''.localdomain ''myhostname'' | |||
}} | }} | ||
Revision as of 14:35, 25 March 2019
This article explains how to configure a network connection.
Check the connection
To troubleshoot a network connection, go through the following conditions and ensure that you meet them:
- Your network interface is listed and enabled.
- You are connected to the network. The cable is plugged in or you are connected to the wireless LAN.
- Your network interface has an IP address.
- Your routing table is correctly set up.
- You can ping a local IP address (e.g. your default gateway).
- You can ping a public IP address (e.g.
8.8.8.8
). - Check if you can resolve domain names (e.g.
archlinux.org
).
8.8.8.8
is a Google DNS server and is a convenient address to test with.Ping
ping is used to test if you can reach a host.
$ ping www.example.com
PING www.example.com (93.184.216.34): 56(84) data bytes 64 bytes from 93.184.216.34: icmp_seq=0 ttl=56 time=11.632 ms 64 bytes from 93.184.216.34: icmp_seq=1 ttl=56 time=11.726 ms 64 bytes from 93.184.216.34: icmp_seq=2 ttl=56 time=10.683 ms ...
For every reply you receive, the ping utility will print a line like the above. For more information see the ping(8) manual. Note that computers can be configured not to respond to ICMP echo requests. [1]
If you receive no reply, this may be related to your default gateway or your Internet Service Provider (ISP). You can run a traceroute to further diagnose the route to the host.
ping: icmp open socket: Operation not permitted
when executing ping, try to re-install the iputils package.Device driver
Check the status
udev should detect your network interface controller (NIC) and automatically load the necessary kernel module at startup. Check the "Ethernet controller" entry (or similar) from the lspci -v
output. It should tell you which kernel module contains the driver for your network device. For example:
$ lspci -v
02:00.0 Ethernet controller: Attansic Technology Corp. L1 Gigabit Ethernet Adapter (rev b0) ... Kernel driver in use: atl1 Kernel modules: atl1
Next, check that the driver was loaded via dmesg | grep module_name
. For example:
$ dmesg | grep atl1
... atl1 0000:02:00.0: eth0 link is up 100 Mbps full duplex
Skip the next section if the driver was loaded successfully. Otherwise, you will need to know which module is needed for your particular model.
Load the module
Search in the Internet for the right module/driver for the chipset. Some common modules are 8139too
for cards with a Realtek chipset, or sis900
for cards with a SiS chipset. Once you know which module to use, try to load it manually. If you get an error saying that the module was not found, it is possible that the driver is not included in Arch kernel. You may search the AUR for the module name.
If udev is not detecting and loading the proper module automatically during bootup, see Kernel module#Automatic module loading with systemd.
Network management
To set up a network connection, go through the following steps:
- Ensure your network interface is listed and enabled.
- Connect to the network. Plug in the Ethernet cable or connect to the wireless LAN.
- Configure your network connection:
- static IP address
- dynamic IP address: use DHCP
dhcpcd@interface.service
) for wired network devices on boot.net-tools
Arch Linux has deprecated net-tools in favor of iproute2.[2]
Deprecated command | Replacement commands |
---|---|
arp | ip neighbor |
ifconfig | ip address, ip link |
netstat | ss |
route | ip route |
For a more complete rundown, see this blog post.
iproute2
iproute2 is part of the base group and provides the ip(8) command-line interface, used to manage network interfaces, IP addresses and the routing table. Be aware that configuration made using ip
will be lost after a reboot. For persistent configuration, you can use a network manager or automate ip commands using scripts and systemd units. Also note that ip
commands can generally be abbreviated, for clarity they are however spelled out in this article.
Network interfaces
By default udev assigns names to your network interfaces using Predictable Network Interface Names, which prefixes interfaces names with en
(wired/Ethernet), wl
(wireless/WLAN), or ww
(WWAN).
Listing network interfaces
Both wired and wireless interface names can be found via ls /sys/class/net
or ip link
. Note that lo
is the loop device and not used in making network connections.
Wireless device names can also be retrieved using iw dev
. See also Wireless network configuration#Get the name of the interface.
If your network interface is not listed, make sure your device driver was loaded successfully.
Enabling and disabling network interfaces
Network interfaces can be enabled / disabled using ip link set interface up|down
, see ip-link(8).
To check the status of the interface eth0
:
$ ip link show dev eth0
2: eth0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master br0 state DOWN mode DEFAULT qlen 1000 ...
The UP
in <BROADCAST,MULTICAST,UP,LOWER_UP>
is what indicates the interface is up, not the later state DOWN
.
eth0
, taking it down will also remove the route, and bringing it back up will not automatically re-establish the default route. See #Routing table for re-establishing it.Static IP address
A static IP address can be configured with most standard network managers and also dhcpcd.
To manually configure a static IP address, add an IP address as described in #IP addresses, set up your routing table and configure your DNS servers.
IP addresses
IP addresses are managed using ip-address(8).
List IP addresses:
$ ip address show
Add an IP address to an interface:
# ip address add address/prefix_len broadcast + dev interface
- Note that:
- the address is given in CIDR notation to also supply a subnet mask
+
is a special symbol that makesip
derive the broadcast address from the IP address and the subnet mask
- Note: Make sure manually assigned IP addresses do not conflict with DHCP assigned ones.
Delete an IP address from an interface:
$ ip address del address/prefix_len dev interface
Delete all addresses matching a criteria, e.g. of a specific interface:
$ ip address flush dev interface
Routing table
The routing table is used to determine if you can reach an IP address directly or what gateway (router) you should use. If no other route matches the IP address, the default gateway is used.
The routing table is managed using ip-route(8).
PREFIX is either a CIDR notation or default
for the default gateway.
List routes:
$ ip route show
Add a route:
# ip route add PREFIX via address dev interface
Delete a route:
# ip route del PREFIX via address dev interface
DHCP
A Dynamic Host Configuration Protocol (DHCP) server provides clients with a dynamic IP address, the subnet mask, the default gateway IP address and optionally also with DNS name servers.
To use DHCP you need a DHCP server in your network and a DHCP client:
Client | Package | Archiso | Note | Systemd units |
---|---|---|---|---|
dhcpcd | dhcpcd | Yes | DHCP, DHCPv6, ZeroConf, static IP | dhcpcd.service , dhcpcd@interface.service
|
ISC dhclient | dhclient | Yes | DHCP, DHCPv6, BOOTP, static IP | dhclient@interface.service
|
Note that instead of directly using a DHCP client you can also use a network manager.
Network managers
A network manager lets you manage network connection settings in so called network profiles to facilitate switching networks.
Network manager | GUI | Archiso [3] | CLI tools | PPP support (e.g. 3G modem) |
DHCP client | Systemd units |
---|---|---|---|---|---|---|
ConnMan | 8 unofficial | No | connmanctl(1) | Yes | internal | connman.service
|
netctl | 2 unofficial | Yes (base) | netctl(1), wifi-menu | Yes | dhcpcd or dhclient | netctl-ifplugd@interface.service , netctl-auto@interface.service
|
NetworkManager | Yes | No | nmcli(1), nmtui(1) | Yes | internal, dhcpcd or dhclient | NetworkManager.service
|
systemd-networkd | No | Yes (base) | networkctl(1) | No | internal | systemd-networkd.service , systemd-resolved.service
|
Wicd | Yes | No | wicd-cli(8), wicd-curses(8) | No | dhcpcd | wicd.service
|
There also is Wifi Radar, a GUI application to manage WiFi networks with wireless_tools, it however does not handle wired connections.
See also List of applications#Network managers.
Set the hostname
A hostname is a unique name created to identify a machine on a network, configured in /etc/hostname
—see hostname(5) and hostname(7) for details. The file can contain the system's domain name, if any. To set the hostname, edit /etc/hostname
to include a single line with myhostname
:
/etc/hostname
myhostname
Alternatively, using hostnamectl(1):
# hostnamectl set-hostname myhostname
To temporarily set the hostname (until reboot), use hostname(1) from inetutils:
# hostname myhostname
To set the "pretty" hostname and other machine metadata, see machine-info(5) § https://www.freedesktop.org/software/systemd/man/machine-info.html.
Local hostname resolution
The myhostname
Name Service Switch (NSS) module of systemd provides local hostname resolution without having to edit /etc/hosts
(hosts(5)). It is enabled by default.
Some clients may however still rely on /etc/hosts
, see [4] [5] for examples.
To configure the hosts file, add the following lines to /etc/hosts
:
127.0.0.1 localhost.localdomain localhost ::1 localhost.localdomain localhost 127.0.1.1 myhostname.localdomain myhostname
/etc/hosts
is significant. The first string is considered the canonical hostname and may be appended with parent domains, where domain components are separated by a dot (ie. .localdomain
above). All following strings on the same line are considered aliases. See hosts(5) for more info.As a result the system resolves to both entries:
$ getent hosts
127.0.0.1 localhost.localdomain localhost 127.0.0.1 localhost.localdomain localhost 127.0.1.1 myhostname.localdomain myhostname
For a system with a permanent IP address, that permanent IP address should be used instead of 127.0.1.1
.
Local network hostname resolution
To make your machine accessible in your LAN via its hostname you can:
- edit the
/etc/hosts
file for every device in your LAN, see hosts(5) - set up a DNS server to resolve your hostname and make the LAN devices use it (e.g. via #DHCP)
- or the easy way: use a Zero-configuration networking service:
- Samba provides hostname resolution via Microsoft's NetBIOS. It only requires installation of samba and enabling of the
nmb.service
service. Computers running Windows, macOS, or Linux withnmb
running, will be able to find your machine. - Avahi provides hostname resolution via zeroconf, also known as Avahi or Bonjour. It requires slightly more complex configuration than Samba: see Avahi#Hostname resolution for details. Computers running macOS, or Linux with an Avahi daemon running, will be able to find your machine. Windows does not have a built-in Avahi client or daemon.
- Samba provides hostname resolution via Microsoft's NetBIOS. It only requires installation of samba and enabling of the
Tips and tricks
Change interface name
You can change the device name by defining the name manually with an udev-rule. For example:
/etc/udev/rules.d/10-network.rules
SUBSYSTEM=="net", ACTION=="add", ATTR{address}=="aa:bb:cc:dd:ee:ff", NAME="net1" SUBSYSTEM=="net", ACTION=="add", ATTR{address}=="ff:ee:dd:cc:bb:aa", NAME="net0"
These rules will be applied automatically at boot.
A couple of things to note:
- To get the MAC address of each card, use this command:
cat /sys/class/net/device_name/address
- Make sure to use the lower-case hex values in your udev rules. It does not like upper-case.
If the network card has a dynamic MAC, you can use DEVPATH
, for example:
/etc/udev/rules.d/10-network.rules
SUBSYSTEM=="net", DEVPATH=="/devices/platform/wemac.*", NAME="int" SUBSYSTEM=="net", DEVPATH=="/devices/pci*/*1c.0/*/net/*", NAME="en"
The device path should match both the new and old device name, since the rule may be executed more than once on bootup. For example, in the second rule, "/devices/pci*/*1c.0/*/net/enp*"
would be wrong since it will stop matching once the name is changed to en
. Only the system-default rule will fire the second time around, causing the name to be changed back to e.g. enp1s0
.
If you're using a USB network device (e.g. Android phone tethering) that has a dynamic MAC address and you want to be able to use different USB ports, you could use a rule that matched depending on vendor and model ID instead:
/etc/udev/rules.d/10-network.rules
SUBSYSTEM=="net", ACTION="add", ENV{ID_VENDOR_ID}=="12ab", ENV{ID_MODEL_ID}=="3cd4", NAME="net2"
To test your rules, they can be triggered directly from userspace, e.g. with udevadm --debug test /sys/DEVPATH
. Remember to first take down the interface you are trying to rename (e.g. ip link set enp1s0 down
).
net0
, net1
, wifi0
, wifi1
. For further details please see the systemd documentation.Revert to traditional interface names
If you would prefer to retain traditional interface names such as eth0, Predictable Network Interface Names can be disabled by masking the udev rule:
# ln -s /dev/null /etc/udev/rules.d/80-net-setup-link.rules
Alternatively, add net.ifnames=0
to the kernel parameters.
Set device MTU and queue length
You can change the device MTU and queue length by defining manually with an udev-rule. For example:
/etc/udev/rules.d/10-network.rules
ACTION=="add", SUBSYSTEM=="net", KERNEL=="wl*", ATTR{mtu}="1500", ATTR{tx_queue_len}="2000"
mtu
: For PPPoE, the MTU should be no larger than 1492. You can also set MTU via systemd.netdev(5).tx_queue_len
: Small value for slower devices with a high latency like modem links and ISDN. High value is recommend for server connected over the high-speed Internet connections that perform large data transfers.
ifplugd for laptops
ifplugd is a daemon which will automatically configure your Ethernet device when a cable is plugged in and automatically unconfigure it if the cable is pulled. This is useful on laptops with onboard network adapters, since it will only configure the interface when a cable is really connected. Another use is when you just need to restart the network but do not want to restart the computer or do it from the shell.
By default it is configured to work for the eth0
device. This and other settings like delays can be configured in /etc/ifplugd/ifplugd.conf
.
netctl-ifplugd@.service
, otherwise you can use ifplugd@.service
from ifplugd package. For example, enable ifplugd@eth0.service
.Bonding or LAG
See netctl#Bonding or Wireless bonding.
IP address aliasing
IP aliasing is the process of adding more than one IP address to a network interface. With this, one node on a network can have multiple connections to a network, each serving a different purpose. Typical uses are virtual hosting of Web and FTP servers, or reorganizing servers without having to update any other machines (this is especially useful for nameservers).
Example
To manually set an alias, for some NIC, use iproute2 to execute
# ip addr add 192.168.2.101/24 dev eth0 label eth0:1
To remove a given alias execute
# ip addr del 192.168.2.101/24 dev eth0:1
Packets destined for a subnet will use the primary alias by default. If the destination IP is within a subnet of a secondary alias, then the source IP is set respectively. Consider the case where there is more than one NIC, the default routes can be listed with ip route
.
Promiscuous mode
Toggling promiscuous mode will make a (wireless) NIC forward all traffic it receives to the OS for further processing. This is opposite to "normal mode" where a NIC will drop frames it is not intended to receive. It is most often used for advanced network troubleshooting and packet sniffing.
/etc/systemd/system/promiscuous@.service
[Unit] Description=Set %i interface in promiscuous mode After=network.target [Service] Type=oneshot ExecStart=/usr/bin/ip link set dev %i promisc on RemainAfterExit=yes [Install] WantedBy=multi-user.target
If you want to enable promiscuous mode on interface eth0
run enable promiscuous@eth0.service
.
Investigate sockets
ss is a utility to investigate network ports and is part of the iproute2 package. It has a similar functionality to the deprecated netstat utility.
Common usage includes:
Display all TCP Sockets with service names:
$ ss -at
Display all TCP Sockets with port numbers:
$ ss -atn
Display all UDP Sockets:
$ ss -au
For more information see ss(8).
Troubleshooting
Swapping computers on the cable modem
Some cable ISPs (Vidéotron for example) have the cable modem configured to recognize only one client PC, by the MAC address of its network interface. Once the cable modem has learned the MAC address of the first PC or equipment that talks to it, it will not respond to another MAC address in any way. Thus if you swap one PC for another (or for a router), the new PC (or router) will not work with the cable modem, because the new PC (or router) has a MAC address different from the old one. To reset the cable modem so that it will recognise the new PC, you must power the cable modem off and on again. Once the cable modem has rebooted and gone fully online again (indicator lights settled down), reboot the newly connected PC so that it makes a DHCP request, or manually make it request a new DHCP lease.
If this method does not work, you will need to clone the MAC address of the original machine. See also MAC address spoofing.
The TCP window scaling problem
TCP packets contain a "window" value in their headers indicating how much data the other host may send in return. This value is represented with only 16 bits, hence the window size is at most 64Kb. TCP packets are cached for a while (they have to be reordered), and as memory is (or used to be) limited, one host could easily run out of it.
Back in 1992, as more and more memory became available, RFC 1323 was written to improve the situation: Window Scaling. The "window" value, provided in all packets, will be modified by a Scale Factor defined once, at the very beginning of the connection. That 8-bit Scale Factor allows the Window to be up to 32 times higher than the initial 64Kb.
It appears that some broken routers and firewalls on the Internet are rewriting the Scale Factor to 0 which causes misunderstandings between hosts. The Linux kernel 2.6.17 introduced a new calculation scheme generating higher Scale Factors, virtually making the aftermaths of the broken routers and firewalls more visible.
The resulting connection is at best very slow or broken.
How to diagnose the problem
First of all, let us make it clear: this problem is odd. In some cases, you will not be able to use TCP connections (HTTP, FTP, ...) at all and in others, you will be able to communicate with some hosts (very few).
When you have this problem, the dmesg
's output is OK, logs are clean and ip addr
will report normal status... and actually everything appears normal.
If you cannot browse any website, but you can ping some random hosts, chances are great that you are experiencing this problem: ping uses ICMP and is not affected by TCP problems.
You can try to use Wireshark. You might see successful UDP and ICMP communications but unsuccessful TCP communications (only to foreign hosts).
Ways of fixing it
Bad
To fix it the bad way, you can change the tcp_rmem
value, on which Scale Factor calculation is based. Although it should work for most hosts, it is not guaranteed, especially for very distant ones.
# echo "4096 87380 174760" > /proc/sys/net/ipv4/tcp_rmem
Good
Simply disable Window Scaling. Since Window Scaling is a nice TCP feature, it may be uncomfortable to disable it, especially if you cannot fix the broken router. There are several ways to disable Window Scaling, and it seems that the most bulletproof way (which will work with most kernels) is to add the following line to /etc/sysctl.d/99-disable_window_scaling.conf
(see also sysctl):
net.ipv4.tcp_window_scaling = 0
Best
This problem is caused by broken routers/firewalls, so let us change them. Some users have reported that the broken router was their very own DSL router.
More about it
This section is based on the LWN article TCP window scaling and broken routers and an archived Kernel Trap article: Window Scaling on the Internet.
There are also several relevant threads on the LKML.
Explicit Congestion Notification
Explicit Congestion Notification (ECN) may cause traffic problems with old/bad routers [6]. As of systemd 239, it is enabled for both ingoing and outgoing traffic.
To enable ECN only when requested by incoming connections (the reasonably safe, kernel default):
# sysctl net.ipv4.tcp_ecn=2
To disable ECN completely (to e.g. test whether ECN was causing problems):
# sysctl net.ipv4.tcp_ecn=0
See also the kernel documentation.
Realtek no link / WOL problem
Users with Realtek 8168 8169 8101 8111(C) based NICs (cards / and on-board) may notice a problem where the NIC seems to be disabled on boot and has no Link light. This can usually be found on a dual boot system where Windows is also installed. It seems that using the official Realtek drivers (dated anything after May 2007) under Windows is the cause. These newer drivers disable the Wake-On-LAN feature by disabling the NIC at Windows shutdown time, where it will remain disabled until the next time Windows boots. You will be able to notice if this problem is affecting you if the Link light remains off until Windows boots up; during Windows shutdown the Link light will switch off. Normal operation should be that the link light is always on as long as the system is on, even during POST. This problem will also affect other operating systems without newer drivers (eg. Live CDs). Here are a few fixes for this problem.
Enable the NIC directly in Linux
Follow #Enabling and disabling network interfaces to enable the interface.
Rollback/change Windows driver
You can roll back your Windows NIC driver to the Microsoft provided one (if available), or roll back/install an official Realtek driver pre-dating May 2007 (may be on the CD that came with your hardware).
Enable WOL in Windows driver
Probably the best and the fastest fix is to change this setting in the Windows driver. This way it should be fixed system-wide and not only under Arch (eg. live CDs, other operating systems). In Windows, under Device Manager, find your Realtek network adapter and double-click it. Under the "Advanced" tab, change "Wake-on-LAN after shutdown" to "Enable".
In Windows XP (example):
Right click my computer and choose "Properties" --> "Hardware" tab --> Device Manager --> Network Adapters --> "double click" Realtek ... --> Advanced tab --> Wake-On-Lan After Shutdown --> Enable
Disable
has no effect (you will notice the Link light still turns off upon Windows shutdown). One rather dirty workaround is to boot to Windows and just reset the system (perform an ungraceful restart/shutdown) thus not giving the Windows driver a chance to disable LAN. The Link light will remain on and the LAN adapter will remain accessible after POST - that is until you boot back to Windows and shut it down properly again.Newer Realtek Linux driver
Any newer driver for these Realtek cards can be found for Linux on the realtek site (untested but believed to also solve the problem).
Enable LAN Boot ROM in BIOS/CMOS
It appears that setting Integrated Peripherals > Onboard LAN Boot ROM > Enabled in BIOS/CMOS reactivates the Realtek LAN chip on system boot-up, despite the Windows driver disabling it on OS shutdown.
No interface with Atheros chipsets
Users of some Atheros ethernet chips are reporting it does not work out-of-the-box (with installation media of February 2014). The working solution for this is to install backports-patchedAUR.
Broadcom BCM57780
This Broadcom chipset sometimes does not behave well unless you specify the order of the modules to be loaded. The modules are broadcom
and tg3
, the former needing to be loaded first.
These steps should help if your computer has this chipset:
- Find your NIC in lspci output:
$ lspci | grep Ethernet
02:00.0 Ethernet controller: Broadcom Corporation NetLink BCM57780 Gigabit Ethernet PCIe (rev 01)
- If your wired networking is not functioning in some way or another, unplug your cable then do the following:
# modprobe -r tg3 # modprobe broadcom # modprobe tg3
- Plug your network cable back in and check whether the module succeeded with:
$ dmesg | greg tg3
- If this procedure solved the issue you can make it permanent by adding
broadcom
andtg3
(in this order) to theMODULES
array:
/etc/mkinitcpio.conf
MODULES=(.. broadcom tg3 ..)
- Regenerate the initramfs
- Alternatively, you can create an
/etc/modprobe.d/broadcom.conf
:
softdep tg3 pre: broadcom
Realtek RTL8111/8168B
# lspci | grep Ethernet
03:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL8111/8168B PCI Express Gigabit Ethernet controller (rev 02)
The adapter should be recognized by the r8169
module. However, with some chip revisions the connection may go off and on all the time. The alternative r8168 should be used for a reliable connection in this case. Blacklist r8169
, if r8168 is not automatically loaded by udev, see Kernel modules#Automatic module loading with systemd.
Another fault in the drivers for some revisions of this adapter is poor IPv6 support. IPv6#Disable functionality can be helpful if you encounter issues such as hanging webpages and slow speeds.
Gigabyte Motherboard with Realtek 8111/8168/8411
With motherboards such as the Gigabyte GA-990FXA-UD3, booting with IOMMU off (which can be the default) will cause the network interface to be unreliable, often failing to connect or connecting but allowing no throughput. This will apply to the onboard NIC and to any other pci-NIC in the box because the IOMMU setting affects the entire network interface on the board. Enabling IOMMU and booting with the install media will throw AMD I-10/xhci page faults for a second, but then boots normally, resulting in a fully functional onboard NIC (even with the r8169 module).
When configuring the boot process for your installation, add iommu=soft
as a kernel parameter to eliminate the error messages on boot and restore USB3.0 functionality.