1 <?xml version="1.0" encoding="UTF-8"?>
2 <!DOCTYPE article PUBLIC "-//OASIS//DTD DocBook XML V4.3//EN"
3 "http://www.oasis-open.org/docbook/xml/4.3/docbookx.dtd">
6 <title>Boot Manager Technical Documentation</title>
9 <firstname>Aaron</firstname>
11 <surname>Klingaman</surname>
13 <email>alk@cs.princeton.edu</email>
17 <orgname>Princeton University</orgname>
22 <revnumber>1.0</revnumber>
24 <date>March 15, 2005</date>
26 <authorinitials>AK</authorinitials>
29 <para>Initial draft.</para>
34 <revnumber>1.1</revnumber>
36 <date>May 31, 2005</date>
38 <authorinitials>AK</authorinitials>
41 <para>Updated post implementation and deployment.</para>
46 <revnumber>1.2</revnumber>
48 <date>November 16, 2005</date>
50 <authorinitials>AK</authorinitials>
53 <para>Add section on where source code is, and other updates to make
54 it consistent with implementation.</para>
61 <title>Components</title>
63 <para>The entire Boot Manager system consists of several components that
64 are designed to work together to provide the functionality outline in the
65 Boot Manager PDN <citation>1</citation>. These consist of:</para>
69 <para>A set of API calls available at PlanetLab Central</para>
73 <para>An API authentication mechanism used exclusively by the
74 BootManager, for the above API calls</para>
78 <para>A package to be run in the boot cd environment on nodes
79 containing core logic</para>
83 <para>A user interface allowing authorized users to add and manage
84 nodes and create node/BootManager configuration files</para>
88 <para>The previous implementation of the software responsible for
89 installing and booting nodes consisted of a set of boot scripts that the
90 boot cd would download and run, depending on the node's current boot
91 state. Only the necessary script for the current state would be
92 downloaded, and the logic behind which script the node was sent to the
93 node existed on the boot server in the form of PHP scripts. However, the
94 intention with the new BootManager system is to send the same script back
95 for all nodes (consisting of the core BootManager code), in all boot
96 states, each time the node starts. Then, the boot manager will run and
97 detiremine which operations to perform on the node, based on the current
98 boot state. All state based logic for the node boot, install, debug, and
99 reconfigure operations are contained in one place; there is no longer any
100 boot state specific logic at PLC.</para>
104 <title>Soure Code</title>
106 <para>All BootManager source code is located in the repository
107 'bootmanager' on the PlanetLab CVS system. For information on how to
108 access CVS, consult the PlanetLab website. Unless otherwise noted, all
109 file references refer to this repository.</para>
113 <title>API Calls</title>
115 <para>Most of the API calls available as part of the PlanetLab Central API
116 are intended to be run by users, and thus authentication for these calls
117 is done with the user's email address and password. However, the API calls
118 described below will be run by the nodes themselves, so a new
119 authentication mechanism is required.</para>
122 <title>Authentication</title>
124 <para>As is done with other PLC API calls, the first parameter to all
125 BootManager related calls will be an authentication structure,
126 consisting of these named fields:</para>
130 <para>AuthMethod</para>
132 <para>The authentication method, only 'hmac' is currently
139 <para>The node id, contained on the configuration file.</para>
145 <para>The node's primary IP address. This will be checked with the
146 node_id against PLC records.</para>
152 <para>The authentication string, depending on method. For the 'hmac'
153 method, a hash for the call using the HMAC algorithm, made from the
154 parameters of the call the key contained on the configuration file.
155 For specifics on how this is created, see below.</para>
159 <para>Authentication is succesful if PLC is able to create the same hash
160 from the values usings its own copy of the node key. If the hash values
161 to not match, then either the keys do not match or the values of the
162 call were modified in transmision and the node cannot be
163 authenticated.</para>
165 <para>Both the BootManager and the authentication software at PLC must
166 agree on a method for creating the hash values for each call. This hash
167 is essentially a finger print of the method call, and is created by this
172 <para>Take the value of every part of each parameter, except the
173 authentication structure, and convert them to strings. For arrays,
174 each element is used. For dictionaries, not only is the value of all
175 the items used, but the keys themselves. Embedded types (arrays or
176 dictionaries inside arrays or dictionaries, etc), also have all
177 values extracted.</para>
181 <para>Alphabetically sort all the parameters.</para>
185 <para>Concatenate them into a single string.</para>
189 <para>Prepend the string with the method name and [, and append
194 <para>The implementation of this algorithm is in the function
195 serialize_params in the file source/BootAPI.py. The same algorithm is
196 located in the 'plc_api' repository, in the function serialize_params in
197 the file PLC/Auth.py.</para>
199 <para>The resultant string is fed into the HMAC algorithm with the node
200 key, and the resultant hash value is used in the authentication
203 <para>This authentication method makes a number of assumptions, detailed
208 <para>All calls made to PLC are done over SSL, so the details of the
209 authentication structure cannot be viewed by 3rd parties. If, in the
210 future, non-SSL based calls are desired, a sequence number or some
211 other value making each call unique will would be required to
212 prevent replay attacks. In fact, the current use of SSL negates the
213 need to create and send hashes across - technically, the key itself
214 could be sent directly to PLC, assuming the connection is made to an
215 HTTPS server with a third party signed SSL certificate.</para>
219 <para>Athough calls are done over SSL, they use the Python class
220 libary xmlrpclib, which does not do SSL certificate
227 <title>PLC API Calls</title>
229 <para>Full, up to date technical documentation of these functions can be
230 found in the PlanetLab Central API documentation. They are listed here
231 for completeness.</para>
235 <para>BootUpdateNode( authentication, update_values )</para>
237 <para>Update a node record, including its boot state, primary
238 network, or ssh host key.</para>
242 <para>BootCheckAuthentication( authentication )</para>
244 <para>Simply check to see if the node is recognized by the system
245 and is authorized.</para>
249 <para>BootGetNodeDetails( authentication )</para>
251 <para>Return details about a node, including its state, what
252 networks the PLC database has configured for the node, and what the
253 model of the node is.</para>
257 <para>BootNotifyOwners( authentication, message, include_pi,
258 include_tech, include_support )</para>
260 <para>Notify someone about an event that happened on the machine,
261 and optionally include the site PIs, technical contacts, and
262 PlanetLab Support.</para>
269 <title>Core Package</title>
271 <para>The Boot Manager core package, which is run on the nodes and
272 contacts the Boot API as necessary, is responsible for the following major
273 functional units:</para>
277 <para>Configuring node hardware and installing the PlanetLab operating
282 <para>Putting a node into a debug state so administrators can track
287 <para>Reconfiguring an already installed node to reflect new hardware,
288 or changed network settings</para>
292 <para>Booting an already installed node into the PlanetLab operating
298 <title>Boot States</title>
300 <para>Each node always has one of four possible boot states.</para>
306 <para>Install. The boot state cooresponds to a new node that has not
307 yet been installed, but record of it does exist. When the boot
308 manager starts, and the node is in this state, the user is prompted
309 to continue with the installation. The intention here is to prevent
310 a non-PlanetLab machine (like a user's desktop machine) from
311 becoming inadvertantly wiped and installed with the PlanetLab node
318 <para>Reinstall. In this state, a node will reinstall the node
319 software, erasing anything that might have been on the disk
326 <para>Boot. This state cooresponds with nodes that have sucessfully
327 installed, and can be chain booted to the runtime node
334 <para>Debug. Regardless of whether or not a machine has been
335 installed, this state sets up a node to be debugged by
336 administrators.</para>
342 <title>Flow Chart</title>
344 <para>Below is a high level flow chart of the boot manager, from the
345 time it is executed to when it exits. This core state machine is located
346 in source/BootManager.py.</para>
349 <title>Boot Manager Flow Chart</title>
353 <imagedata align="left" fileref="boot-manager-flowchart.png"
361 <title>Boot CD Environment</title>
363 <para>The boot manager needs to be able to operate under all currently
364 supported boot cds. The new 3.0 cd contains software the current 2.x cds
365 do not contain, including the Logical Volume Manager (LVM) client tools,
366 RPM, and YUM, among other packages. Given this requirement, the boot cd
367 will need to download as necessary the extra support files it needs to
368 run. Depending on the size of these files, they may only be downloaded
369 by specific steps in the flow chart in figure 1, and thus are not
372 <para>See the PlanetLab BootCD Documentation for more information about
373 the current, 3.x boot cds, how they are build, and what they provide to
374 the BootManager.</para>
378 <title>Node Configuration Files</title>
380 <para>To remain compatible with 2.x boot cds, the format and existing
381 contents of the configuration files for the nodes will not change. There
382 will be, however, the addition of three fields:</para>
386 <para>NET_DEVICE</para>
388 <para>If present, use the device with the specified mac address to
389 contact PLC. The network on this device will be setup. If not
390 present, the device represented by 'eth0' will be used.</para>
394 <para>NODE_KEY</para>
396 <para>The unique, per-node key to be used during authentication and
397 identity verification. This is a fixed length, random value that is
398 only known to the node and PLC.</para>
404 <para>The PLC assigned node identifier.</para>
408 <para>An example of a configuration file for a dhcp networked
411 <programlisting>IP_METHOD="dhcp"
412 HOST_NAME="planetlab-1"
413 DOMAIN_NAME="cs.princeton.edu"
414 NET_DEVICE="00:06:5B:EC:33:BB"
415 NODE_KEY="79efbe871722771675de604a227db8386bc6ef482a4b74"
416 NODE_ID="121"</programlisting>
418 <para>An example of a configuration file for the same machine, only with
419 a statically assigned network address:</para>
421 <programlisting>IP_METHOD="static"
422 IP_ADDRESS="128.112.139.71"
423 IP_GATEWAY="128.112.139.65"
424 IP_NETMASK="255.255.255.192"
425 IP_NETADDR="128.112.139.127"
426 IP_BROADCASTADDR="128.112.139.127"
427 IP_DNS1="128.112.136.10"
428 IP_DNS2="128.112.136.12"
429 HOST_NAME="planetlab-1"
430 DOMAIN_NAME="cs.princeton.edu"
431 NET_DEVICE="00:06:5B:EC:33:BB"
432 NODE_KEY="79efbe871722771675de604a227db8386bc6ef482a4b74"
433 NODE_ID="121"</programlisting>
435 <para>Existing 2.x boot cds will look for the configuration files only
436 on a floppy disk, and the file must be named 'planet.cnf'. The new 3.x
437 boot cds, however, will initially look for a file named 'plnode.txt' on
438 either a floppy disk, or burned onto the cd itself. Alternatively, it
439 will fall back to looking for the original file name, 'planet.cnf'. This
440 initial file reading is performed by the boot cd itself to bring the
441 nodes network online, so it can download and execute the Boot
444 <para>However, the Boot Manager will also need to identify the location
445 of and read in the file, so it can get the extra fields not initially
446 used to bring the network online (primarily node_key and node_id). Below
447 is the search order that the BootManager will use to locate a
450 <para>Configuration file location search order:<informaltable>
454 <entry>File name</entry>
456 <entry>Floppy drive</entry>
458 <entry>Flash devices</entry>
460 <entry>CDRom, in /usr/boot</entry>
462 <entry>CDRom, in /usr</entry>
466 <entry>plode.txt</entry>
478 <entry>planet.cnf</entry>
490 </informaltable></para>
495 <title>User Interface for Node Management</title>
498 <title>Adding Nodes</title>
500 <para>New nodes are added to the system explicitly by either a PI or a
501 tech contact, either directly through the API calls, or by using the
502 appropriate interfaces on the website. As nodes are added, their
503 hostname, network configuration method (dhcp or static), and any static
504 settings are required to be entered. Regardless of network configuration
505 method, IP address is required. When the node is brought online, the
506 records at PLC will be updated with any remaining information.</para>
508 <para>After a node is added, the user has the option of creating a
509 configuration file for that node. Once the node is added, the contents
510 of the file are created automatically, and the user is prompted to
511 download and save the file. This file contains only the primary network
512 interface information (necessary to contact PLC), the node id, and the
515 <para>The default boot state of a new node is 'inst', which requires the
516 user to confirm the installation at the node, by typing yes on the
517 console. If this is not desired, as is the case with nodes in a
518 co-location site, or for a large number of nodes being setup at the same
519 time, the administrator can change the node state, after the entry is in
520 the PLC records, from 'inst' to 'reinstall'. This will bypass the
521 confirmation screen, and proceed directly to reinstall the machine (even
522 if it already had a node installation on it).</para>
526 <title>Updating Node Network Settings</title>
528 <para>If the primary node network address must be updated, if the node
529 is moved to a new network for example, then two steps must be performed
530 to successfully complete the move:</para>
534 <para>The node network will need to be updated at PLC, either
535 through the API directly or via the website.</para>
539 <para>Either the floppy file regenerated and put into the machine,
540 or, update the existing floppy to match the new settings.</para>
544 <para>If the node ip address on the floppy does not match the record at
545 PLC, then the node will not boot until they do match, as authentication
546 will fail. The intention here is to prevent a malicious user from taking
547 the floppy disk, altering the network settings, and trying to bring up a
548 new machine with the new settings.</para>
550 <para>On the other hand, if a non-primary network address needs to be
551 updated, then simply updating the record in the configuration file will
552 suffice. The boot manager, at next restart, will reconfigure the
553 machine, and update the PLC records to match the configuration
558 <title>Removing Nodes</title>
560 <para>Nodes are removed from the system by:</para>
564 <para>Deleting the record of the node at PLC</para>
568 <para>Shutting down the machine.</para>
572 <para>Once this is done, even if the machine attempts to come back
573 online, it cannot be authorized with PLC and will not boot.</para>
578 <title>BootManager Configuration</title>
580 <para>All run time configuration options for the BootManager exist in a
581 single file located at source/configuration. These values are described
586 <para><literal>VERSION</literal></para>
588 <para>The current BootManager version. During install, written out to
589 /etc/planetlab/install_version</para>
593 <para><literal>TEMP_PATH</literal></para>
595 <para>A writable path on the boot cd we can use for temporary storage
600 <para><literal>SYSIMG_PATH</literal></para>
602 <para>The path were we will mount the node logical volumes during any
603 step that requires access to the disks.</para>
607 <para>CACERT_PATH</para>
609 <para>Variable not used anymore.</para>
613 <para><literal>NONCE_FILE</literal></para>
615 <para>Variable not used anymore.</para>
619 <para><literal>PLCONF_DIR</literal></para>
621 <para>The path that PlanetLab node configuration files will be created
622 in during install. This should not be changed from /etc/planetlab, as
623 this path is assumed in other PlanetLab components.</para>
627 <para><literal>SUPPORT_FILE_DIR</literal></para>
629 <para>A path on the boot server where per-step additional files may be
630 located. For example, the packages that include the tools to allow
631 older 2.x version boot cds to partition disks with LVM.</para>
635 <para><literal>ROOT_SIZE</literal></para>
637 <para>During install, this sets the size of the node root partition.
638 It must be large enough to house all the node operational software. It
639 does not store any user/slice files. Include 'G' suffix in this value,
640 indicating gigabytes.</para>
644 <para><literal>SWAP_SIZE</literal></para>
646 <para>How much swap to configure the node with during install. Include
647 'G' suffix in this value, indicating gigabytes.</para>
651 <para><literal>SKIP_HARDWARE_REQUIREMENT_CHECK</literal></para>
653 <para>Whether or not to skip any of the hardware requirement checks,
654 including total disk and memory size constraints.</para>
658 <para><literal>MINIMUM_MEMORY</literal></para>
660 <para>How much memory is required by a running PlanetLab node. If a
661 machine contains less physical memory than this value, the install
662 will not proceed.</para>
666 <para><literal>MINIMUM_DISK_SIZE</literal></para>
668 <para>The size of the small disk we are willing to attempt to use
669 during the install, in gigabytes. Do not include any suffixes.</para>
673 <para><literal>TOTAL_MINIMUM_DISK_SIZE</literal></para>
675 <para>The size of all usable disks must be at least this sizse, in
676 gigabytes. Do not include any suffixes.</para>
680 <para><literal>INSTALL_LANGS</literal></para>
682 <para>Which language support to install. This value is used by RPM,
683 and is used in writting /etc/rpm/macros before any RPMs are
688 <para><literal>NUM_AUTH_FAILURES_BEFORE_DEBUG</literal></para>
690 <para>How many authentication failures the BootManager is willing to
691 except for any set of calls, before stopping and putting the node into
698 <title>Installer Hardware Detection</title>
700 <para>When a node is being installed, the Boot Manager must identify which
701 hardware the machine has that is applicable to a running node, and
702 configure the node properly so it can boot properly post-install. The
703 general procedure for doing so is outline in this section. It is
704 implemented in the <filename>source/systeminfo.py</filename> file.</para>
706 <para>The process for identifying which kernel module needs to be load
711 <para>Create a lookup table of all modules, and which PCI ids
712 coorespond to this module.</para>
716 <para>For each PCI device on the system, lookup its module in the
721 <para>If a module is found, put in into one of two categories of
722 modules, either network module or scsi module, based on the PCI device
727 <para>For each network module, write out an 'eth<index>' entry
728 in the modprobe.conf configuration file.</para>
732 <para>For each scsi module, write out a
733 'scsi_hostadapter<index>' entry in the modprobe.conf
734 configuration file.</para>
738 <para>This process is fairly straight forward, and is simplified by the
739 fact that we currently do not need support for USB, sound, or video
740 devices when the node is fully running. The boot cd itself uses a similar
741 process, but includes USB devices. Consult the boot cd technical
742 documentation for more information.</para>
744 <para>The creation of the PCI id to kernel module table lookup uses three
745 different sources of information, and merges them together into a single
746 table for easier lookups. With these three sources of information, a
747 fairly comprehensive lookup table can be generated for the devices that
748 PlanetLab nodes need to have configured. They include:</para>
752 <para>The installed <filename>/usr/share/hwdata/pcitable
753 </filename>file</para>
755 <para>Created at the time the hwdata rpm was built, this file contains
756 mappings of PCI ids to devices for a large number of devices. It is
757 not necessarily complete, and doesn't take into account the modules
758 that are actually available by the built PlanetLab kernel, which is a
759 subset of the full set available (again, PlanetLab nodes do not have a
760 use for network or video drivers, and thus are not typically
765 <para>From the built kernel, the <filename>modules.pcimap</filename>
766 from the <filename>/lib/modules/<kernelversion>/</filename>
769 <para>This file is generated at the time the kernel is installed, and
770 pulls the PCI ids out of each module, for the modules list they
771 devices they support. Not all modules list all devices they sort, and
772 some contain wild cards (that match any device of a single
773 manufacturer).</para>
777 <para>From the built kernel, the <filename>modules.dep</filename> from
778 the <filename>/lib/modules/<kernelversion>/</filename>
781 <para>This file is also generated at the time the kernel is installed,
782 but lists the dependencies between various modules. It is used to
783 generate a list of modules that are actually available.</para>
787 <para>It should be noted here that SATA (Serial ATA) devices have been
788 known to exist with both a PCI SCSI device class, and with a PCI IDE
789 device class. Under linux 2.6 kernels, all SATA modules need to be listed
790 in modprobe.conf under 'scsi_hostadapter' lines. This case is handled in
791 the hardware loading scripts by making the assumption that if an IDE
792 device matches a loadable module, it should be put in the modprobe.conf
793 file, as 'real' IDE drivers are all currently built into the kernel, and
794 do not need to be loaded. SATA devices that have a PCI SCSI device class
795 are easily identified.</para>
797 <para>It is enssential that the modprobe.conf configuration file contain
798 the correct drivers for the disks on the system, if they are present, as
799 during kernel installation the creation of the initrd (initial ramdisk)
800 which is responsible for booting the system uses this file to identify
801 which drivers to include in it. A failure to do this typically results in
802 an kernel panic at boot with a 'no init found' message.</para>
806 <title>Backward Compatibility</title>
808 <para>Given the large number of nodes in PlanetLab, and the lack of direct
809 physical access to them, the process of updating all configuration files
810 to include the new node id and node key will take a fairly significant
811 amount of time. Rather than delay deployment of the Boot Manager until all
812 machines are updated, alternative methods for aquiring these values is
813 used for existing nodes.</para>
815 <para>First, the node id. For any machine already part of PlanetLab, there
816 exists a record of its IP address and MAC address in PlanetLab central. To
817 get the node_id value, if it is not located in the configuration file, the
818 BootManager uses a standard HTTP POST request to a known php page on the
819 boot server, sending the IP and MAC address of the node. This php page
820 queries the PLC database, and returns a node_Id if the node is part of
821 PlanetLab, -1 otherwise.</para>
823 <para>Second, the node key. All Boot CDs currently in use, at the time
824 they request a script from PLC to run, send in the request a randomly
825 generated value called a boot_nonce, usually 32 bytes or larger. During
826 normal BootManager operation, this value is ignored. However, in the
827 absense of a node key, we can use this value. Although it is not as secure
828 as a typical node key (because it is not distributed through external
829 mechanisms, but is generated by the node itself), it can be used if we
830 validate that the IP address of the node making the request matches the
831 PLC record. This means that nodes behind firewalls can no longer be
832 allowed in this situation.</para>
836 <title>Common Scenarios</title>
838 <para>Below are common scenarios that the BootManager might encounter that
839 would exist outside of the documented procedures for handling nodes. A
840 full description of how they will be handled by the BootManager follows
845 <para>A configuration file from previously installed and functioning
846 node is copied or moved to another machine, and the networks settings
847 are updated on it (but the key and node_id is left the same).</para>
849 <para>Since the authentication for a node consists of matching not
850 only the node id, but the primary node ip, this step will fail, and
851 the node will not allow the boot manager to be run. Instead, the new
852 node must be created at PLC first, and a network configuration file
853 for it must be generated, with its own node key.</para>
857 <para>After a node is installed and running, the administrators
858 mistakenly remove the cd and media containing the configuration
861 <para>The node installer clears all boot records from the disk, so the
862 node will not boot. Typically, the bios will report no operating
867 <para>A new network configuration file is generated on the website,
868 but is not put on the node.</para>
870 <para>Creating a new network configuration file through the PLC
871 interfaces will generate a new node key, effectively invalidating the
872 old configuration file (still in use by the machine). The next time
873 the node reboots and attempts to authentication with PLC, it will
874 fail. After two consecutive authentication failures, the node will
875 automatically put itself into debug mode. In this case, regardless of
876 the API function being called that was unable to authentication, the
877 software at PLC will automatically notify the PlanetLab
878 administrators, and the contacts at the site of the node was able to
879 be identified (usually through its IP address or node_id by searching
880 PLC records.).</para>
889 <title>The PlanetLab Boot Manager</title>
891 <date>January 14, 2005</date>
894 <firstname>Aaron</firstname>
896 <surname>Klingaman</surname>