Unix Toolbox IT工作者

BOoRFGOnZ

http://cb.vu/unixtoolbox.xhtml


System

Hardware | Statistics | Users | Limits | Runlevels | root password | Compile kernel | Repair grub

Running kernel and system information

# uname -a                           # Get the kernel version (and BSD version)
# lsb_release -a                     # Full release info of any LSB distribution
# cat /etc/SuSE-release              # Get SuSE version
# cat /etc/debian_version            # Get Debian version


Use /etc/DISTR-release with DISTR= lsb (Ubuntu), redhat, gentoo, mandrake, sun (Solaris), and so on. See also /etc/issue.

# uptime                             # Show how long the system has been running + load
# hostname                           # system's host name
# hostname -i                        # Display the IP address of the host. (Linux only)
# man hier                           # Description of the file system hierarchy
# last reboot                        # Show system reboot history


Hardware Informations

Kernel detected hardware

# dmesg                              # Detected hardware and boot messages
# lsdev                              # information about installed hardware
# dd if=/dev/mem bs=1k skip=768 count=256 2>/dev/null | strings -n 8 # Read BIOS


Linux

# cat /proc/cpuinfo                  # CPU model
# cat /proc/meminfo                  # Hardware memory
# grep MemTotal /proc/meminfo        # Display the physical memory
# watch -n1 'cat /proc/interrupts'   # Watch changeable interrupts continuously
# free -m                            # Used and free memory (-m for MB)
# cat /proc/devices                  # Configured devices
# lspci -tv                          # Show PCI devices
# lsusb -tv                          # Show USB devices
# lshal                              # Show a list of all devices with their properties
# dmidecode                          # Show DMI/SMBIOS: hw info from the BIOS


FreeBSD

# sysctl hw.model                    # CPU model
# sysctl hw                          # Gives a lot of hardware information
# sysctl vm                          # Memory usage
# dmesg | grep "real mem"            # Hardware memory
# sysctl -a | grep mem               # Kernel memory settings and info
# sysctl dev                         # Configured devices
# pciconf -l -cv                     # Show PCI devices
# usbdevs -v                         # Show USB devices
# atacontrol list                    # Show ATA devices
# camcontrol devlist -v              # Show SCSI devices


Load, statistics and messages

The following commands are useful to find out what is going on on the system.

# top                                # display and update the top cpu processes
# mpstat 1                           # display processors related statistics
# vmstat 2                           # display virtual memory statistics
# iostat 2                           # display I/O statistics (2 s intervals)
# systat -vmstat 1                   # BSD summary of system statistics (1 s intervals)
# systat -tcp 1                      # BSD tcp connections (try also -ip)
# systat -netstat 1                  # BSD active network connections
# systat -ifstat 1                   # BSD network traffic through active interfaces
# systat -iostat 1                   # BSD CPU and and disk throughput
# tail -n 500 /var/log/messages      # Last 500 kernel/syslog messages
# tail /var/log/warn                 # System warnings messages see syslog.conf


Users

# id                                 # Show the active user id with login and group
# last                               # Show last logins on the system
# who                                # Show who is logged on the system
# groupadd admin                     # Add group "admin" and user colin (Linux/Solaris)
# useradd -c "Colin Barschel" -g admin -m colin
# usermod -a -G <group> <user>       # Add existing user to group (Debian)
# groupmod -A <user> <group>         # Add existing user to group (SuSE)
# userdel colin                      # Delete user colin (Linux/Solaris)
# adduser joe                        # FreeBSD add user joe (interactive)
# rmuser joe                         # FreeBSD delete user joe (interactive)
# pw groupadd admin                  # Use pw on FreeBSD
# pw groupmod admin -m newmember     # Add a new member to a group
# pw useradd colin -c "Colin Barschel" -g admin -m -s /bin/tcsh
# pw userdel colin; pw groupdel admin


Encrypted passwords are stored in /etc/shadow for Linux and Solaris and /etc/master.passwd on FreeBSD. If the master.passwd is modified manually (say to delete a password), run # pwd_mkdb -p master.passwd to rebuild the database.


To temporarily prevent logins system wide (for all users but root) use nologin. The message in nologin will be displayed (might not work with ssh pre-shared keys).

# echo "Sorry no login now" > /etc/nologin       # (Linux)
# echo "Sorry no login now" > /var/run/nologin   # (FreeBSD)



Limits

Some application require higher limits on open files and sockets (like a proxy
web server, database). The default limits are usually too low.
Linux

Per shell/script

The shell limits are governed by ulimit. The status is checked
with ulimit -a. For example to change the open files limit from
1024 to 10240 do:

# ulimit -n 10240                    # This is only valid within the shell


The ulimit command can be used in a script to change the limits for the script only.

Per user/process

Login users and applications can be configured in /etc/security/limits.conf. For example:

# cat /etc/security/limits.conf
*   hard    nproc   250              # Limit user processes
asterisk hard nofile 409600          # Limit application open files


System wide

Kernel limits are set with sysctl. Permanent limits are set in /etc/sysctl.conf.

# sysctl -a                          # View all system limits
# sysctl fs.file-max                 # View max open files limit
# sysctl fs.file-max=102400          # Change max open files limit
# echo "1024 50000" > /proc/sys/net/ipv4/ip_local_port_range  # port range
# cat /etc/sysctl.conf
fs.file-max=102400                   # Permanent entry in sysctl.conf
# cat /proc/sys/fs/file-nr           # How many file descriptors are in use



FreeBSD

Per shell/script

Use the command limits in csh or tcsh or as in Linux, use ulimit in an sh or bash shell.
Per user/process

The default limits on login are set in /etc/login.conf. An unlimited value is still limited by the system maximal value.
System wide

Kernel limits are also set with sysctl. Permanent limits are set in /etc/sysctl.conf or /boot/loader.conf. The syntax is the same as Linux but the keys are different.

# sysctl -a                          # View all system limits
# sysctl kern.maxfiles=XXXX          # maximum number of file descriptors
kern.ipc.nmbclusters=32768           # Permanent entry in /etc/sysctl.conf
kern.maxfiles=65536                  # Typical values for Squid
kern.maxfilesperproc=32768
kern.ipc.somaxconn=8192              # TCP queue. Better for apache/sendmail
# sysctl kern.openfiles              # How many file descriptors are in use
# sysctl kern.ipc.numopensockets     # How many open sockets are in use
# sysctl -w net.inet.ip.portrange.last=50000 # Default is 1024-5000
# netstat -m                         # network memory buffers statistics


See The FreeBSD handbook Chapter 11http://www.freebsd.org/handbook/configtuning-kernel-limits.html for details.

Solaris

The following values in /etc/system will increase the maximum file descriptors per proc:

set rlim_fd_max = 4096               # Hard limit on file descriptors for a single proc
set rlim_fd_cur = 1024               # Soft limit on file descriptors for a single proc



Runlevels

Linux

Once booted, the kernel starts init which then starts rc which starts all scripts belonging to a runlevel. The scripts are stored in /etc/init.d and are linked into /etc/rc.d/rcN.d with N the runlevel number.

The default runlevel is configured in /etc/inittab. It is usually 3 or 5:

# grep default: /etc/inittab                                         
id:3:initdefault:


The actual runlevel can be changed with init. For example to go from 3 to 5:

# init 5                             # Enters runlevel 5



    *   0       Shutdown and halt

    *   1       Single-User mode (also S)

    *   2       Multi-user without network

    *   3       Multi-user with network

    *   5       Multi-user with X

    *   6       Reboot


Use chkconfig to configure the programs that will be started at boot in a runlevel.

# chkconfig --list                   # List all init scripts
# chkconfig --list sshd              # Report the status of sshd
# chkconfig sshd --level 35 on       # Configure sshd for levels 3 and 5
# chkconfig sshd off                 # Disable sshd for all runlevels


Debian and Debian based distributions like Ubuntu or Knoppix use the command update-rc.d to manage the runlevels scripts. Default is to start in 2,3,4 and 5 and shutdown in 0,1 and 6.

# update-rc.d sshd defaults          # Activate sshd with the default runlevels
# update-rc.d sshd start 20 2 3 4 5 . stop 20 0 1 6 .  # With explicit arguments
# update-rc.d -f sshd remove         # Disable sshd for all runlevels
# shutdown -h now (or # poweroff)    # Shutdown and halt the system



FreeBSD

The BSD boot approach is different from the SysV, there are no runlevels. The final boot state (single user, with or without X) is configured in /etc/ttys. All OS scripts are located in /etc/rc.d/ and in /usr/local/etc/rc.d/ for third-party applications. The activation of the service is configured in /etc/rc.conf and /etc/rc.conf.local. The default behavior is configured in /etc/defaults/rc.conf. The scripts responds at least to start|stop|status.

# /etc/rc.d/sshd status
sshd is running as pid 552.
# shutdown now                       # Go into single-user mode
# exit                               # Go back to multi-user mode
# shutdown -p now                    # Shutdown and halt the system
# shutdown -r now                    # Reboot


The process init can also be used to reach one of the following states level. For example # init 6 for reboot.


    *   0       Halt and turn the power off (signal USR2)

    *   1       Go to single-user mode (signal TERM)

    *   6       Reboot the machine (signal INT)

    *   c       Block further logins (signal TSTP)

    *   q       Rescan the ttys(5) file (signal HUP)



Reset root password

Linux method 1

At the boot loader (lilo or grub), enter the following boot option:

init=/bin/sh


The kernel will mount the root partition and init will start the bourne shell
instead of rc and then a runlevel. Use the command passwd at the prompt to change the password and then reboot. Forget the single user mode as you need the password for that.

If, after booting, the root partition is mounted read only, remount it rw:

# mount -o remount,rw /
# passwd                             # or delete the root password (/etc/shadow)
# sync; mount -o remount,ro /        # sync before to remount read only
# reboot


FreeBSD method 1

On FreeBSD, boot in single user mode, remount / rw and use passwd. You can select the single user mode on the boot menu (option 4) which is displayed for 10 seconds at startup. The single user mode will give you a root shell on the / partition.

# mount -u /; mount -a               # will mount / rw
# passwd
# reboot



Unixes and FreeBSD and Linux method 2

Other Unixes might not let you go away with the simple init trick. The solution is to mount the root partition from an other OS (like a rescue CD) and change the password on the disk.


    *   Boot a live CD or installation CD into a rescue mode which will give you a shell.

    *   Find the root partition with fdisk e.g. fdisk /dev/sda

    *   Mount it and use chroot:


# mount -o rw /dev/ad4s3a /mnt
# chroot /mnt                        # chroot into /mnt
# passwd
# reboot



Kernel modules

Linux

# lsmod                              # List all modules loaded in the kernel
# modprobe isdn                      # To load a module (here isdn)


FreeBSD

# kldstat                            # List all modules loaded in the kernel
# kldload crypto                     # To load a module (here crypto)


Compile Kernel

Linux

# cd /usr/src/linux
# make mrproper                      # Clean everything, including config files
# make oldconfig                     # Reuse the old .config if existent
# make menuconfig                    # or xconfig (Qt) or gconfig (GTK)
# make                               # Create a compressed kernel image
# make modules                       # Compile the modules
# make modules_install               # Install the modules
# make install                       # Install the kernel
# reboot


FreeBSD

Optionally update the source tree (in /usr/src) with csup (as of FreeBSD 6.2 or later):

# csup <supfile>


I use the following supfile:

*default host=cvsup5.FreeBSD.org  # www.freebsd.org/handbook/cvsup.html#CVSUP-MIRRORS
*default prefix=/usr
*default base=/var/db
*default release=cvs delete tag=RELENG_7
src-all


To modify and rebuild the kernel, copy the generic configuration file to a new name and edit it as needed (you can also edit the file GENERIC directly). To restart the build after an interruption, add the option NO_CLEAN=YES to the make command to avoid cleaning the objects already build.

# cd /usr/src/sys/i386/conf/
# cp GENERIC MYKERNEL
# cd /usr/src
# make buildkernel KERNCONF=MYKERNEL
# make installkernel KERNCONF=MYKERNEL


To rebuild the full OS:

# make buildworld                    # Build the full OS but not the kernel
# make buildkernel                   # Use KERNCONF as above if appropriate
# make installkernel
# reboot
# mergemaster -p                     # Compares only files known to be essential
# make installworld
# mergemaster -i -U                  # Update all configurations and other files
# reboot


For small changes in the source you can use NO_CLEAN=yes to avoid rebuilding the whole tree.

# make buildworld NO_CLEAN=yes       # Don't delete the old objects
# make buildkernel KERNCONF=MYKERNEL NO_CLEAN=yes



Repair grub

So you broke grub? Boot from a live cd, [find your linux partition under /dev and use fdisk to find the linux partion] mount the linux partition, add /proc and /dev and use grub-install /dev/xyz. Suppose linux lies on /dev/sda6:

# mount /dev/sda6 /mnt               # mount the linux partition on /mnt
# mount --bind /proc /mnt/proc       # mount the proc subsystem into /mnt
# mount --bind /dev /mnt/dev         # mount the devices into /mnt
# chroot /mnt                        # change root to the linux partition
# grub-install /dev/sda              # reinstall grub with your old settings





Processes

Listing | Priority | Background/Foreground | Top | Kill

Listing and PIDs

Each process has a unique number, the PID. A list of all running process is retrieved with ps.

# ps -auxefw                         # Extensive list of all running process


However more typical usage is with a pipe or with pgrep:

# ps axww | grep cron
  586  ??  Is     0:01.48 /usr/sbin/cron -s
# ps axjf                            # All processes in a tree format (Linux)
# ps aux | grep 'ss[h]'              # Find all ssh pids without the grep pid
# pgrep -l sshd                      # Find the PIDs of processes by (part of) name
# echo $$                            # The PID of your shell
# fuser -va 22/tcp                   # List processes using port 22 (Linux)
# pmap PID                           # Memory map of process (hunt memory leaks) (Linux)
# fuser -va /home                    # List processes accessing the /home partition
# strace df                          # Trace system calls and signals
# truss df                           # same as above on FreeBSD/Solaris/Unixware



Priority

Change the priority of a running process with renice. Negative numbers have a higher priority, the lowest is -20 and "nice" have a positive value.

# renice -5 586                      # Stronger priority
586: old priority 0, new priority -5


Start the process with a defined priority with nice. Positive is "nice" or weak, negative is strong scheduling priority. Make sure you know if /usr/bin/nice or the shell built-in is used (check with # which nice).

# nice -n -5 top                     # Stronger priority (/usr/bin/nice)
# nice -n 5 top                      # Weaker priority (/usr/bin/nice)
# nice +5 top                        # tcsh builtin nice (same as above!)


While nice changes the CPU scheduler, an other useful command ionice will schedule the disk IO. This is very useful for intensive IO application (e.g. compiling). You can select a class (idle - best effort - real time), the man page is short and well explained.

# ionice c3 -p123                    # set idle class for pid 123 (Linux only)
# ionice -c2 -n0 firefox             # Run firefox with best effort and high priority
# ionice -c3 -p$$                    # Set the actual shell to idle priority


The last command is very useful to compile (or debug) a large project. Every command launched from this shell will have a lover priority. $$ is your shell pid (try echo $$).

FreeBSD uses idprio/rtprio (0 = max priority, 31 = most idle):

# idprio 31 make                     # compile in the lowest priority
# idprio 31 -1234                    # set PID 1234 with lowest priority
# idprio -t -1234                    # -t removes any real time/idle priority



Background/Foreground

When started from a shell, processes can be brought in the background and back to the foreground with [Ctrl]-[Z] (^Z), bg and fg. List the processes with jobs.

# ping cb.vu > ping.log
^Z                                   # ping is suspended (stopped) with [Ctrl]-[Z]
# bg                                 # put in background and continues running
# jobs -l                            # List processes in background
[1]  - 36232 Running                       ping cb.vu > ping.log
[2]  + 36233 Suspended (tty output)        top
# fg %2                              # Bring process 2 back in foreground


Use nohup to start a process which has to keep running when the shell is closed (immune to hangups).

# nohup ping -i 60 > ping.log &



Top

The program top displays running information of processes. See also the program htop from htop.sourceforge.net (a more powerful version of top) which runs on Linux and FreeBSD (ports/sysutils/htop/). While top is running press the key h for a help overview. Useful keys are:


    *   u [user name] To display only the processes belonging to the user. Use + or blank to see all users

    *   k [pid] Kill the process with pid.

    *   1 To display all processors statistics (Linux only)

    *    R Toggle normal/reverse sort.


Signals/Kill

Terminate or send a signal with kill or killall.

# ping -i 60 cb.vu > ping.log &
[1] 4712
# kill -s TERM 4712                  # same as kill -15 4712
# killall -1 httpd                   # Kill HUP processes by exact name
# pkill -9 http                      # Kill TERM processes by (part of) name
# pkill -TERM -u www                 # Kill TERM processes owned by www
# fuser -k -TERM -m /home            # Kill every process accessing /home (to umount)


Important signals are: