第 31 章 Installation and Booting

Under certain circumstances, when installing the system from the boot DVD or ISO image, not all assigned IP addresses are shown in the network spoke once network connectivity is configured and enabled. To work around this problem, leave the network spoke and enter it again. After re-entering, all assigned addresses are shown correctly.

Network devices are not automatically enabled during installation unless the installation method requires network connectivity. As a consequence, a traceback error can occur during Kickstart installation due to inactive network devices. To work around this problem, set the ksdevice=link option on boot or add the --device=link option to the ks.cfg file to enable network devices with active links during Kickstart installation.

An interface with IPv6-only configuration does not bring up the network interface after manual graphical installation from an IPv6 source. Consequently, the system boots with the interface set to ONBOOT=no, and consequently the network connection does not work. Select the Automatically connect to network check box if available, or use kickstart with a command as follows:

network --noipv4 --bootproto=dhcp --activate

In both cases IPv6 will be configured to be active on system start.

If the network interface is set to IPv4 and IPv6 configuration, and is installed from an IPv6 address, after installation it will be configured to be active on system start (ONBOOT=yes).

The anaconda installer does not correctly handle adding of FCoE disks. As a consequence, adding FCoE disks on the anaconda advance storage page fails with the following error message:

No Fibre Channel Forwarders or VN2VN Responders Found

To work around this problem, simply repeat the steps to add the FCoE disks; the configuration process produces the correct outcome when repeated. Alternatively, run the lldpad -d command in the anaconda shell before adding the FCoE disks in the anaconda user interface to avoid the described problem.

The source code does not handle booting on a bnx2i iSCI driver correctly. As a consequence, when installing Red Hat Enterprise Linux 7.1, the server does not reboot automatically after the installation is completed. No workaround is currently available.

When booting in rescue mode on IBM System z architecture, the second and third rescue screens in the rescue shell are incomplete and not displayed properly.

When the /boot partition is not separated and the boot= parameter is specified on the kernel command line, an attempt to boot the system in the FIPS mode fails. To work around this issue, remove the boot= parameter from the kernel command line.

When the user inserts a space character anywhere between nameservers while configuring the nameservers in the Network Configuration dialog during a text-mode installation, the installer terminates unexpectedly. To work around this problem, if you want to configure multiple nameservers during the Network Configuration step of the installation, enter them in a comma-separated list without spaces between the nameservers. For example, while entering 1.1.1.1, 2.1.2.1 with a space in this situation causes the installer to crash, entering 1.1.1.1,2.1.2.1 without a space ensures the installer handles configuring multiple nameservers correctly and does not crash.

If the installation system has multiple iSCSI storage targets connected over separate active physical network interfaces, the installer will hang when starting iSCSI target discovery in the Installation Destination screen.

The same issue also appears with an iSCSI multipath target accessible over two different networks, and happens no matter whether the Bind targets to network interfaces option is selected.

To work around this problem, make sure only one active physical network interface has an available iSCSI target, and attach any additional targets on other interfaces after the installation.

When using a screen resolution of less than 1024x768 (such as 800x600) during a manual installation, some of the controls in the Manual Partitioning screen become unreachable. This problem commonly appears when connecting to the installation system using a VNC viewer, because by default the VNC server is set to 800x600.

To work around this issue, set the resolution to 1024x768 or higher using a boot option. For example:

linux inst.vnc inst.resolution=1024x768

For information about Anaconda boot options, see the Red Hat Enterprise Linux 7.1 Installation Guide.

A system booting with iSCSI using IPv6 times out while trying to connect to the iSCSI server after about 15 minutes, but then connects successfully and boots as expected.

When installing Red Hat Enterprise Linux 7 on UEFI-based systems, the Anaconda installer terminates unexpectedly with the following error:

BootLoaderError: failed to remove old efi boot entry

To work around this problem, edit the Install Red Hat Enterprise Linux 7 option in the boot menu by pressing the e key and append the efi_no_storage_paranoia kernel parameter to the end of the line that begins with linuxefi. Then press the F10 key to boot the modified option and start installation.

Due to the conversion of the iprutils package to use systemd instead of legacy init scripts, the sg driver is no longer loaded during system boot. Consequently, if the sg driver is not loaded, the /dev/sg* devices will not be present.

To work around this issue, manually issue modprobe sg or add it to an init script. Once the sg driver is loaded, the /dev/sg* devices will be present and the sg driver may be used to access SCSI devices.

It is not possible to use read-only disks as hard drive installation repository sources. When specifying the inst.repo=hd:device:path option ensure that device is writable.

Various platforms include BIOS or UEFI-assisted software RAID provided by LSI. This hardware requires the closed-source megasr driver, which is not included in Red Hat Enterprise Linux. Thus, platforms and adapters that depend on megasr are not supported by Red Hat. Also, the use of certain open-source RAID alternatives, such as the dmraid Disk Data Format 1 (DDF1) capability, is not currently supported on these systems.

However, on certain systems, such as IBM System x servers with the ServeRAID adapter, it is possible to disable the BIOS RAID function. To do this, enter the UEFI menu and navigate through the System Settings and Devices and I/O Ports submenus to the Configure the onboard SCU submenu. Then change the SCU setting from RAID to nonRAID. Save your changes and reboot the system. In this mode, the storage is configured using an open-source non-RAID LSI driver shipped with Red Hat Enterprise Linux, such as mptsas, mpt2sas, or mpt3sas.

To obtain the megasr driver for IBM systems, refer to the IBM support page.

Certain Cisco Unified Computing System (UCS) platforms are also impacted by this restriction. However, it is not possible to disable the BIOS RAID function on these systems. To obtain the megasr driver, refer to the Cisco support page.

During CPU hot plugging, the kernel can sometimes issue the following warning message:

WARNING: at block/blk-mq.c:701__blk_mq_run_hw_queue+0x31d/0x330()

The message is harmless, and you can ignore it.

The Linux kernel Non-Uniform Memory Access (NUMA) balancing does not always work correctly. As a consequence, when the numa_balancing parameter is set, some of the memory can move to an arbitrary non-destination node before moving to the constrained nodes, and the memory on the destination node also decreases under certain circumstances. There is currently no known workaround available.

An attempt to remove the bnx2x module while the bnx2fc driver is processing a corrupted frame causes a kernel panic. To work around this problem, shut down any active FCoE interfaces before executing the modprobe -r bnx2x command.

The QLogic 1G iSCSI Adapter present in the system can cause a call trace error when the qla4xx driver is sharing the interrupt line with the USB sub-system. This error has no impact on the system functionality. The error can be found in the kernel log messages located in the /var/log/messages file. To prevent the call trace from logging into the kernel log messages, add the nousb kernel parameter when the system is booting.

When using the bnx2x driver with a BCM57711 device and sending traffic over Virtual Extensible LAN (VXLAN), the transmitted packets have bad checksums. Consequently, communication fails, and UDP: bad checksum messages are displayed in the kernel log on the receiving side. To work around this problem, disable checksum offload on the bnx2x device using the ethtool utility.

If you change certain parameters while the Network Interface Card (NIC) is set to down, the system can become unresponsive if you are using a qlge driver. This problem occurs due to a race condition between the New API (NAPI) registration and unregistration. There is no workaround currently available.

In the Kernel Dump Configuration window, selecting the Raw device option in the Target settings tab does not work. To work around this problem, edit the kdump.conf file manually.

Due to a bug in the yaboot boot loader, upgrading from Red Hat Enterprise Linux 6 to Red Hat Enterprise Linux 7 can fail on the IBM Power Systems with an Unknown or corrupt filesystem error.

The anaconda installer cannot handle disks with labels from the IBM AIX operating systems correctly. As a consequence, an attempt to install Red Hat Enterprise Linux on such a disk fails. Users are advised to not use disks with AIX labels in order prevent the installation failures.

If you attempt to perform an in-place upgrade from Red Hat Enterprise Linux 6.6 running on IBM System z architecture to Red Hat Enterprise Linux 7.1 and have the kernel-kdump package installed on Red Hat Enterprise Linux 6.6, the kdump boot record is not removed. Consequently, the upgrade fails when the zipl utility is called. To work around this problem, remove the kdump boot record from the /etc/zipl.conf file before performing the upgrade.

Setting only full name and no user name for a new user in text installation does not require root password to be set. As a consequence, when such a user is configured and no root password is set, the user is not able to log in either, and neither is root. There is also no straightforward way to create a user or set the root password after such an installation since initial-setup crashes due to this bug. To work around this problem, set the root password during installation or set the user name for the user during text installation.

The installer terminates unexpectedly if you set up partitioning before adding an iSCSI disk and then set up partitioning again. As a consequence, it is impossible to successfully complete the installation in this situation. To work around this problem, reset storage or reboot before adding iSCSI or FCoE disks during installation.

The anaconda installer expects a ks.cfg file if booting with the inst.ks=cdrom:/ks.cfg parameter, and enters the emergency mode if the ks.cfg file is not provided within several minutes. With some enterprise servers that take a long time to boot, Anaconda does not wait long enough to enable the user to provide the ks.cfg file in time.

To work around this problem, add the rd.retry boot parameter and use a large value. For example, using rd.retry=86400 causes a time-out after 24 hours, and using rd.retry=1<<15 should, in theory, time out after about 34 years, which provides the user with sufficient time in all known scenarios.

The Back button used in the firstboot utility is not working properly. It is often disabled, and if it is enabled, pressing it has no effect. Consequently, during Subscription Management Registration, clicking Back does not return you to the previous panel. If you want to go back, enter an invalid server or invalid credentials and click Done. After this, either an Unable to reach the server dialog or an Unable to register the system dialog appears at the top of the initial firstboot panel. Dismiss the error dialog, and choose the No, I prefer to register at a later time option.

The GRUB2 bootloader supports network booting over the Hypertext Transfer Protocol (HTTP) and the Trivial File Transfer Protocol (TFTP). However, under heavy network traffic, network boot over HTTP is very slow and may cause timeout failures. If this problem occurs, use TFTP to load the kernel and initrd images. To do so, put the boot files in the TFTP server directory and add the following to the grub.cfg file where 1.1.1.1 is the address of the TFTP server:

insmod tftp
set root=tftp,1.1.1.1

The Driver Update Disk loader does not reconfigure network devices if they have already been configured. Consequently, installations that use a Driver Update Disk to replace an existing, functional network driver with a different version will not be able to use the network to fetch the installer runtime image.

To work around this problem, use the provided network driver during the installation process and update the network driver after the installation.