Sharing Unix File Systems q NFS Network File

Sharing Unix File Systems q NFS (Network File System ) allows you to share file systems q Reasons for sharing file system m Transparent to user • User can keep use their familiar commands • Access the same file from multiple nodes. m m m To provide disk space to diskless clients To prevent duplication To provide centrally supported programs and data To share data among users Simplifies central support tasks • Such as backup. • Space usage monitor NFS 1

NFS q Candidates could be: m Home dirs, m Web page, m /usr/local and other common utilities m Documentations m Temporary huge space request m Source code repository m …… NFS 2

NFS protocol versions q NFS was introduced by Sun Microsystems in 1985 q The original public release of NFS was version 2 q In 1990 s, version 3 m Increases performance • Makes writes safely asynchronous m Better support for large files q version 4 m No ancillary protocols – integrated locking and mount m Compound operations – bundle multiple RPC together in a single exchange m Strong security – uses RPCSSEC_GSS API m Require use of transport protocols that offer congestion control – hence NFS v 4 will not support UDP transport. NFS 3

More about NFS q NFS run on top of Sun’s RPC (Remote Procedure Call) protocol m RPC provides a system-independent way for processes to communicate in a client-server fashion over a network. q Some NFS versions run on top of UDP and TCP m Why? q Web. NFS m Promote the use of NFS over the Internet m A extention of NFS protocol that allow easier access to servers and clients through Internet firewalls m A Web. NFS-enhanced web browser can use an NFS URL to access data directory from the server – nfs: /www. Your. Company. com/ NFS 4

Security and NFS q NFS protocol was originally designed with essentially no concerns for security q Tight control over shared filesystems m Don’t export any filesystems if anyone that you don’t trust has root access on a client host m If you have a firewall, block port 2049, block Sun. RPC portmap deamon port 111 q Kerberos to authenticate to ensure that remote users really are who they say they are. q Ipsec to encrypt the data NFS 5

Network File System q NFS consists of a number of components m Mounting protocol m Mount server m Daemons that coordinate basic file service m Several diagnostic utilities q Basically two parts: m Client • Mount a directory m Server • Export a directory NFS 6

Server-side NFS q Mount a filesystem m Daemon: mountd q Access files m Daemon: nfsd q Both daemon started at boottime q Mountd and nfsd use the same database ( (xtab on most systems, sharetab on Solaris) q There are commands (exportfs, share) to add/remove the entries in the database m Exportfs reads /etc/exports file • Run exportfs –a • To remove entries, run exportfs –u NFS 7

Server-side NFS m On Solaris, /etc/dfstab is a shell script • Run shareall command, which equivalent to sh /etc/dfstab m m To remove entries, run unshare Solaris also provides command “exports”, which translates export options to share/unshare commands q Any directory can be exported m Client can mount a subdirectories of an exported directory • Example: On HP-UX 11 i You can do above, but you may not to mount another sub dir from the same exported directory. q Each device to be exported separately m For example • /user is a separate partition, / can be exported without exporting /users NFS 8

NFS Daemons on Solaris m The following daemons are running on server side • nfsd [nservers] • mountd • nfslogd m The following daemons are running on both server side and client side • rquotad • lockd • statd m How to start the daemons: • Daemons are under /usr/lib/nfs • Daemons are started at boot time by /etc/init. d/nfs. client and nfs. server • nfs. client starts statd and lockd • nfs. server starts nfsd (16 copies default), mountd, nfslogd • inetd start rquotad. NFS 9

Configure NFS on Solaris q The share command to export a dir share –F nfs [-o options] pathname q Options are: m rw • Export read-write to the entire world m rw=accesslist • Export read-only with access only by listed hosts m m m ro ro=accesslist root=list • Lists hosts permitted to access this filesystem as root • Otherwise, root access from a client is equivalent to access by “nobody” (UID – 2) NFS 10

Configure NFS on Solaris m anon=uid • m defines the uid used for users who do not provide a valid user ID. nosub • Forbids clients to mount subdirectories of the exported directory m nosuid • Prevents setuid and setgid files from being created via NFS q Access list format m Separated by : m Identify computers by • • individual hostnames/IP Domain, . domain NIS netgroup, groupname Network, @network[/prefix] NFS 11

Configure NFS on Solaris q Examples: on Colossus /etc/dfstab share -F nfs -o root=aa. aa rw=keeper. csl. mtu. edu: csl: scifi: language: pizza: icu: cs: fishlab: cec /home/csdept share -F nfs -o root=xx. xx: yy. yyy rw=csl: language: scifi: pizza: icu: cs: fishlab: physpecial: cec: cs. mtu. edu /export/major NFS 12

Configuring NFS on Linux q /etc/exports m The NFS server configuration file m Controls which files/dirs are exported m Access control q The general format of entries: directory [host(option)] m Host can be • • • m empty, means every host individual host, name or IP address Domains, *wrotethebook. com Networks, IP address/mask(prefix length) Netgroups, @group 1 Option can be • ro • rw NFS 13

LINUX NFS Permission Control q NFS server trusts local authentication q USE GID/UID to do regular Unix permission control. m m Mismatch uid/gid will be a unexpected problem What about root? • Do you want root users on NFS client to act like root users on NFS servers? Probably Not. • NFS prevents by default setting root_squash – Mapping root to nobody UID/GID • Set no_root_squash to allow it. m Option all_squash • Map all uids and gids to the anonymous user. Useful for NFS -exportd public FTP directories, news spool directories, etc. NFS 14

LINUX NFS q Assign anonymous a UID/GID using option anonuid=XXX, anongid=XXX This option is primarily useful for PC/NFS clients, where you might want all requests appear to be from one user. q UID/GID mapping file using option map_static=filename q Command exportfs, m build /var/lib/nfs/xtab m export all in /etc/exports file using “exportfs –a “ m Sync /etc/exports and /var/bin/nfs/xtab “exportfs –r” m Temporarily export a filesystem • exportfs fox: /usr/local –o rw m m remove from export list “exportfs –u fox: /usr/local” remove all export list “exportfs –ua “ NFS 15

Client-side NFS q Mounting remote filesystems m Use mount command to establish temporary network mounts m List in /etc/fstab or /etc/vfstabb m Automatic mounting services such as automount q what are exported or where m Ask administrator m Find out yourself: • what have been exported from a host: showmount –e hostname NFS 16

Mounting Remote Filesystems q Using mount command Basic format: mount hostname: remote-dir local_dir • • hostname is the NFS server local_dir must exist already q Using umount command umount local_dir[remote_dir] NFS 17

Mounting NFS q On Solaris, entries in /etc/vfstab are mounted by mountall during system startup q On Linux, entries in /etc/fstab are mounted via mount –a in startup files. q NFS mount flags ( page 502) m Options: • • rw ro bg hard – cause the operations that try to access it to block until the server comes backup • Soft – What about a job that has been running 18 hours and will be done in an hour and aborted? NFS 18

Mounting NFS m More mounting options • intr – Allow user to interrupt blocked operations • nointr • retrans=n – Specifies the number of times to repeat a request before returning an error on a soft-mountd filesystem • timeo=n – Set the timeout period for requests • rsize – 8 K for the same network • wsize • vers • tcp NFS 19

Common mount options on Linux Option Function async auto When –a is used dev Allow character and block special files on this file system exec noauto Don’t mount even with -a noexec nosuid Don’t allow setuid, setgid nouser Only root can mount remount ro rw suid Allow programs to run setuid or setgid sync user Ordinary users to mount the file system atime Update inode noatime defaults Rw, suid, dev, exec, auto, nouser, and async NFS 20

Dedicated NFS file Servers q Fast, reliable file service is one of the most important elements q Dedicated NFS servers m Optimized for file service m Storage scale smoothly m Reliable, simplified software m Redundant hardware m Support different file formats for different platforms, such as NFS, CIFS m Easier to administer than UNIX file servers m Good backup and checkpoint facilities NFS 21

NFS Automounter q Automatically mounts NFS when needed m Why automount? • Maintaining /etc/fstab can be tedious • Minimizes the number of active mount points to reduce/Avoid chaos when servers crashes m How? • Mount a virtual filesystem driver on the directories for automatic mount to occur – Kernel-resident filesystem driver called autfs is used • When the user references a directory withing the virtual filesystem, the automounter intercepts the reference and mount the actual filesystem the user is trying to reach. q NFS 22

NFS Automounter q Daemons m automountd and automount are started by /etc/init. d/autofs • Automount reads the configuration file, setup autofs mounts • References to automounted filesystems are handled by a separate daemon automountd m On solaris 10, you can start automounter • #svcadm enable system/filesystem/autofs • The actual script is under /lib/svc/method/svc-autofs NFS 23

automount q automount m Basic configuration files • Master map • Direct maps • Indirect maps m Direct and indirect maps provide information about filesystems that are to be automounted m A master map lists direct and indirect maps that automount should pay attention to. • Only one master map that can be active • The default one is /etc/auto_master NFS 24

automount m Indirect maps • • Automount several filesystems underneath a common directory The path is specified in the master maps, not in the indirect map itself The name indirect map will be used as subdirectory to install the mount – Example, indirect map Info –ro chim. L/chim/info Using ls a directory full of indirect mounts can be confusing » automount does not show the subdirectories until their contents have been accessed » What does ls do? NFS 25

automount m Direct maps • List the maps that do not share a common prefix – Such as: /cs/tools –ro anchor: /cs/tools /usr/src chim: /usr/src • Each implemented with a separate autofs mount • Require slightly overhead • But mount point and directory structure always accessible by commands such as ls NFS 26

automount m master map • format: mount-point map-name option – List the direct and indirect map – For indirect map, it specify root directory used by the mounts defined in the map. – /- is for direct maps – Option set the defaults for all mounts within the map [ruihong@cslserver nis]$ more /etc/auto_master # cs and csl don't use auto_home and auto yet. #/net -hosts -nosuid, nobrowse #/home auto_home -nobrowse #/local auto_local -nobrowse /fisher auto_fisher /auto_direct NFS 27

automatic automount q Have automount to figure it out itself m Query mountd running on a remote server to find out what filesystems the server exports m Use –host as a map name, automount will map remote hosts’exports into the specified automount directory • Example The serer chim exported /usr/share/man The client’s master map has a line /net -hosts –nosuid, soft The mount point will be /net/chim/usr/share/man • It does enumerate all possible hosts • It waits for individual subdirectory names to be references, then runs off and mounts the exported filesystem NFS 28

automatic automount m Key substitutions • Impress a degree of regularity on the automounter maps • The ampersand (&) expands to the matched key value in a map • The asterisk(*) is a default case. • Example: a indirect map: usr 1 -rw thud: /export/home/usr 1 usr 2 -rw thud: /export/home/usr 2 usr 3 -rw thud: /export/home/usr 3 usr 4 -rw thud: /export/home/usr 4 Can be rewritten as only one line * -rw thud: /export/home/& NFS 29

NFS Automounter q Automount can use maps from NIS server. q Other type maps, like auto_home mount # Master map for automounter # #+auto_master #/xfn -xfn /net -hosts -nosuid /home auto_home /- auto_direct # Home directory map for automounter # #+auto_home craig almond: /export/home/craig * pecan: /export/home/& NFS 30

Practice In Lab 4, we created a dir called /research. As a group, let’s try the following tasks q Export /research, so m m some hosts in your group as rw some hosts in your group as ro one extra host have root mapping Access the exported from /net/hostname/home q Mount /research from some hosts in your group q Automount the exported /research to /research from some hosts in your group NFS 31

Network Information Service q NIS m Was originally called the Sun Yellow Pages m Is an administrative database • What information does database store? m Provides central control and automatic dissemination of important files. NFS 32

Network Information Service q The unit of sharing in NIS is the record, not the file. m A record is usually corresponds to one line q Mast server maintains the authoritative copies of system files m m Are kept in their original locations and formats Edited with a text editor before q A server process makes the contents of the file available over the network. q A server and its clients constitute an NIS “domain” m m associated with A set of maps Different from DNS domain NFS 33

Network Information Service q On the master server m Data file are preprocessed into database files (called maps) by a hashing library • One key associated with each entry • A system file may have to be translated into several NIS maps – Example: » /etc/passwd is translated into two different NIS Maps passwd. byname passwd. byuid • The record order is not preserved. NFS 34

Network Information Service q On the slave server m Relieve load on the master and keep clients functioning when master is down. m Replicate the network maps from master server q On the client m Does not distinguish between the master server and slaves m Traditional • Use IP broadcasting to locate servers • Place at least one NIS server on every physical network m Some implementation allows to specify the server name instead of broadcasting NFS 35

Network Information Service q Databases are called NIS maps m /etc/passwd m /etc/group m /etc/netgroup m /etc/hosts m /etc/networks m /etc/protocols m /etc/services m /etc/aliases m /etc/auto_master m… … NFS 36

Network Information Service q Netgroups m Sets of users, machines, and nets for easy reference in other system files m The format of netgroup entry is Groupname list-of-members m Members are separated by whitespace. A member is • Either a netgroup name • Or a tripnet of the form (hostname, username, nisdomainname) a empty value in a field matching any value in that field m Netgroup can be used in several system files that define permissions • Configuring NFS export ( /etc/exports file) • Restrict remote logins and shell access ( /etc/hosts. equiv file) • Local login access ( /etc/passwd file) NFS 37

Network Information Service q Exmaples m /etc/netgrouplab (goofy. csl. mtu. edu, , ) (mickey. csl. mtu. edu, , ) (minnie. csl. mtu. edu, , ) (donald. csl. mtu. edu, , ) (simba. csl. mtu. edu, , ) (nala. csl. mtu. edu, , ) (tramp. csl. mtu. edu, , ) noncs 1121 d (, msbenson, ) (, behart, ) (, adrogers, ) (, agwalthe, ) (, adpeters, ) (, ammurrel, ) (, aeblechi, ) (, jaeul, ) (, srkelley, ) (, daperry, ) (, anlevend, ) (, frtuntla, ) (, cwminnic, ) (, rrmilkov, ) (, iatrifon, ) (, ksrivast, ) (, talange, ) (, mtnielse, ) (, csdummy, ) m /etc/dfstab share -F nfs -o root=aa. aa rw=keeper. csl. mtu. edu: csl: scifi: language: pizza: icu: cs: fishlab: cec /home/csdept m /etc/passwd +@nonenrolled: : : /usr/local/shells/gone +@cs 1121 nons: : : /usr/local/shells/limitaccess NFS 38

Network Information Service q How NIS works m NIS’s data files are stored in one directory • Usually /var/yp m Each NIS map is stored in a hashed format in a subdirectory named for the NIS domain • Exact Map files names depends on the hashing library being used. • For example: – On cslserver, under /var/yp/csl. mtu. edu, there are ndbm files: password. byname. dir password. byname. pag auto. direct. dir auto. direct. pag NFS 39

Network Information Service m There is a makefile under /var/yp • Which calls makedbm to generate NIS maps from flat files. • After you modify a system file, cd to NIS dir which is /var/yp, and run make. – Make checks the modification date and rebuild some maps. m m Slave server run ypxfr command regularly as cron to pull the map copies Server runs yppush to instruct each slave server to execute ypxfr. • Yppush is used in makefile after rebuilding a updated map. • a special map called ypservers – no correspond flat file – constructed automatically NFS 40

Network Information Service q After the initial configuration m Deamon ypserv running only on the servers • Accept queries from clients and answers them by looking up information in the hashed map files m Deamon ypbind running on every machine, include servers • C library contacts the local ypbind to answer a query – provided /etc/nsswitch. conf says to check with nis • Ypbind locates a ypserv and returns its identify to the C library • C library contacts the server directly NFS 41

NIS commands and daemons (page 527) m m m m m ypserv ypbind domainname ypxfrd yppush makedbm ypmake ypinit ypset ypwhich ypcat ypmatch yppasswd ypchfn ypchsh yppassdd NIS server daemon NIS client daemon Sets the NIS domain a machine is in Downloads current version of a map from master Serves request from ypxfr (runs on master server) Makes slave servers update their versions of a map builds a hashed map from a flat file rebuild map Configure the host as a master or slave server Makes ypbind connect to a particular server find out which server the current host is using Prints the values contained in an NIS map prints map entries for a specified key Changes a password on the NIS master server Changes GECOS info on the NIS master server Changes a login shell on NIS master server Server for yppasswd, ypchsh, and ypchfn NFS 42

Set UP NIS q Procedure m Prepare the source file m Initialize master server • Run ypinit -m • Starting NIS service – Run ypserv m Initialize slave server • Set domainname • Run ypinit command with –c and master serve ip. m Setup NIS clients • • Setup default domainname /etc/nsswitch. conf Run ypinit –c Start ypbind daemon – Run ypbind NFS 43

NIS q Case study: m Let’s take a look at where the startup files for starting NIS server and NIS client. • /etc/init. d/rpc • /usr/lib/netsvc/yp/ypstart (ypstop) m Maps • /var/yp/csl. mtu. edu m Let’s take a look at the make file NFS 44

NIS q Work with DNS m Hostname can be converted to IP address by DNS, NIS, and the host file. q /etc/nsswitch. conf m Define the order in which the sources are searched. Choices are: • Dns • Nis • Files For example: hosts: dns nis files networks: nis [NOTFOUND=return] files services: nis files protocols: nis files NFS 45

NIS+ q NIS+ is a completely new software product and structured entirely differently from NIS m m m Sun’s new administrative database released in early 90 s Provides all the functionality of NIS Improved security • • m m NIS does not authenticate servers or clients NIS+ use secure DES NIS provide same access to everyone in the domain NIS+ provide access levels to different users NIS+ is a hierarchical decentralized architecture. NIS+ has enhanced data structures. • Multiple columns table and relation query between tables m NIS+ is complex and not as popular as NIS NFS 46

NIS practice Refer to the Sun document m “System Administration Guide: Naming and Directory Service “ • section “ NIS Setup and Administration” q As a group m Set up a NIS server and a slave server m On the NIS server • create a user account named visitor 1 with home dir /research/visitor 1 m Set up a NIS client • Modify /etc/nsswitch. conf – To have passwd line to include nis NFS 47

Copying files around q Not elegant solution, but it works m Example: • A change to a startup script • A new version of software q File copying systems can use “push” model or “pull” model m m m rdist: push files rsync: push files more securely expect: pull files NFS 48

Copying files around q Rdist m The easiest way to distribute files from a central server. • Sort of like make – A text specification of the files to be distributed – Rdist copies file when they are out of date. • Preserves owner, group, mode and modification time of files. • When updating an existing file, it first deletes the old version before installing the new – Makes rdist suitable for transferring executables that might be in use during the update. NFS 49

Copying files around m It runs on rsh and relies on rsh-style authentication to gain access to remote system • The protocol can also be changed from command line rdist -P /usr/local/bin/ssh -f my. Distfile m Control file has the statements of the form Label: pathnames -> destination commands Where: » label: associate a name with the statement » pathname: lists of files to be copied » destination: list of hosts to copy files to » commands: modify the behavior ( default is to copy files to the equivalent paths) NFS 50

Copying files around EXAMPLE: SYS_FILES = (/etc/passwd /etc/group /etc/mail/aliases) GET_ALL = (chim lollopop barkadon) GET_SOME = (whammo spiff) all: ${SYS_FILES} -> ${GET_ALL} notify barb; special /etc/mail/aliases “/usr/bin/newaliases” some: ${SYS_FILES} -> $ {GET_SOME} except /etc/mail/aliases; notify eddie@spiff; NFS 51

Copying files around q Rsync: push files m Flexiblie and faster replacement of rcp • Only transmit the differences • Support for copying links, devices, owners, groups and permissions • Can use any transparent remote shell rsh, ssh • Can use rsync daemon mode m m m No controlfile Must run repeatedly to transfer a set of files to multiple hosts. Example: • #rsync –gopt –password-file=/etc/rsync. pwd /etc/passwd lollipop: : sysfiles • #rsync –gopt –e ssh /etc/passwd /etc/shadow lollipop: /etc NFS 52

Copying files around q Expect: pull files m Expect is a TCL extension scripting language • Allows to you write control scripts for interactive programs – The output produced by each operation is examined to determine what input should be send next m Use expect to retrieve and install them • Fundamental expect commands are: – spawn : start up a subprocess to control – send: feed input to a subprocess – expect: take action depending on a subprocess’s output m Example: Make system files available via FTP from central server NFS 53

Copying files around Example: ftps the /etc/passwd file from the machine netserver spawn /usr/bin/ftp netserver while 1 { expect { “Name *: ” {send “netclientr”} “Password: ” {send “netclientpasswordr”} “ftp> “ {break} “failed” {send_user “Can’t log in. r”; exit 1} timeout {send_user ‘Timeout problem. r”; exit 2} }} send “lcd /etcr” expect “ftp>” {send “cd pub/sysfilesr”} expect “ftp>” {send “get passwdr”} expect “ftp>” {send “quitr”, send_user “r”} exit 0 NFS 54

Summary q NFS q NIS q Copying file around NFS 55