/boot can safely have FS type of ext4

Boot kernel params are added in /etc/default/grub.  See my file "grub2.txt".

lsblk is a nice command to list block devices and mounts.
Seems that it doesn't understand btrfs.

To label at least ext* type partitions like: e2label /dev/sdb1 MY_LABEL
Or when mkfs-ing with -L just like for btrfs.
Labels can be seen with "ls /dev/disk/by-label.
The contents are updated dynamically (like after a label or re-label), and
does not require mounting, but sometimes it is pretty slow to update.
To set uuid of ext filesystems use tune2fs -u|-U clear|random|time|'val' /dev/X.
To get uuid of ext filesystems use tune2fs -l /dev/X | grep UUID

To label a swap partition:  swaplabel -L LABEL /dev/sdaX

GENERIC set label along with mkfs.* or mkswap:  Use -L switch.

Mount via fstab labels with device spec like:  LABEL=<label>
Mount via uuids with device spec like:  UUID=<label>

ALT-terminal switching.  In text mode, just ALT+Fx.  From graphical mode,
CTRL+ALT+Fx.

Probably better than adding name servers to /etc/resolv.conf is adding them
to /etc/sysconfig/network-scripts/ifcfg-eth* with vars DNS1, DNS2.

Default router normally goes into /etc/sysconfig/network (file) (with a
GATEWAYDEV (I wonder if DNS1, DNS2 can go here), but can alternatively go into
/etc/sysconfig/network-scripts/ifcfg-eth* with vars GATEWAY and DEFROUTER=yes.  

parted: Whole/Entire disk range is 0c to 100%.  UPDATE!  Lately this often gives warnings.  Getter:
    0% to 100%
This will actually start the start sector after the header stuff; and end
before the end of the disk.
Start and End values in parted ARE INCLUSIVE, so don't overlap them.
kB, MB, GB, TB are all 0-based.  Set units to GiB, etc. if that is what you want.
parted now has great new command that lets you move slice end:  resizepart
  For me it is throwing core when attempt to do this with mounted VMware slice.
Allowed filesystem types (contrary to docs):  linux-swap, ext4, btrfs
For "other" (for Linux-related purposes) I use "ext2"
GPT does not support overlapping partitions (could be parted with GPT).

NFS default is "ro".  I don't know why Suse specifies "ro", but need to
specify "rw" if that is what you want.

Standard partition table type is "msdos", where max of
1 extended allowed;  max of 4 primary + extended allowed; and logicals reside
within the extended.
gpt lacks these restrictions and has no primary/extended/logical distinction,
just a partition name.
Fucking parted somehow converts when you specify cylinders, so that if you
say:  'p1 X 100' and 'p2 101 Y', it may show 'p2 100...'.  WTF!

Neither parted nor fdisk let you resize a mounted partition nor move an
extended partition even if nothing is mounted.
Must re-create extended partition.
Parted does let you resize a mounted btrfs parition at least (asks for confirm).

Can re-create partitions with same boundaries without clobbering anything, but
fdisk and parted won't let you do it while participating partitions are mounted.
Can also extend partitions in this way.
Of course can not shrink nor extend into other used disk areas.
Basically to be non-destructive can only extend into unused area of disk.

Determine UUID of disks:
    ls -l /dev/disk/by-uuid/*
OR
    blkid /dev/sdaX

List all packages from a specific repository.  Example
    rpm -qa --queryformat "%{Name}:%{Vendor}\n" | grep -i vlc
    yum list installed | grep -i fusion
Details and limitations:  http://raftaman.net/?p=1300

df and du units:  Single letters like K, G are *ibibyte 1024 style.
*B 2-letters like KB, GB are 1000 style.
-k == --block-size=1K is the default.  Why the switch???
-m == --block-size=1M
No -g switch so must use:  --block-size=1G

List motherboard:  mdidecode -t 2

FQDN Host Names
They don't say it outright, but the man page for "hostname" implies that one
should always set hostname to the short host name, and let FQDNs be  resolved
dynamically using domain resolution systems.
That seems like a very bad strategy since, as they admit on that man page,
"it may either have multiple FQDNs/domain names or none at all".
Therefore, only way to nail down a persistent unique hostname seems to be to set FQDN with "hostname".

MBR
MBR  is first 446 bytes on disk
After that are 64 byte partition table and 2 bytes signature.
I don't know what the signature is, but it's apparently not a checksum since can zero just the mbr.

RDP clients:  FreeRPD.  [previously used rdesktop, krdc]
If need to see Linux windows at same time, then:
    rdesktop -g 1280x1024 -d adprod -u bsimpson w-bsimpson-31
Otherwise:
    1. Switch to a dedicated Linux WM virtual desktop
    2. rdesktop -f -d adprod -u bsimpson w-bsimpson-31
    3. Move Windows control panel to top of screen
    4. CTRL+ALT+ENTER to toggle off full-screen
    5. CTRL+ALT+DELETE is passed through directly.
(Regular Windows RDP rules apply, so see win.txt file about that).


SAR
For pacman it comes in the sysstat package
Use crontab to set up the maintenance job.
  Sample cron job at http://sebastien.godard.pagesperso-orange.fr/tutorial.html
  GOTCHA:  Same specifies -d switch to .../sa1, but it must be -D!
           And remove the 'root' specifier- that's not regular cron.
  sa1 script writes (perhaps updates?) binary files /var/log/sa/sa<DATE>
"kb" to sar means "kilobytes" (more commonly specified by kB or KB)
sar ARGS -f /path/to/sa13      ARGS defaults to "-u"
ARGS:
    Device I/O:    -d
    Network:       -n
    Indiv. CPU Ut. -P ALL       # %idle = 100-user-nice-system
    RAM (phys+vir) -r           # %memused  N.b. does not account for kernel mem
    CompreCPU Ut.  -u           # %idle = 100-user-nice-system
    Swap:          -S           # kbswpused
Less useful args:
    I/O and xfer:  -b
    swap paging:   -B
    Processors:    -P {cpu|ALL}
    Queue:         -q
    phys RAM:      -R
    Swap:          -W

* kbcommit SHOULD list total Mem *needed* in kB (according to man page), but unfortunately this demands ridiculously large amounts.  E.g. on system using only 3.2 GiB it is demanding 6.5 GiB.


Recovering partition tables and partitions.
Use "gpart" or "gdisk".
gdisk is hard to get.  As of today not available for 13.1, but check
http://www.rodsbooks.com/gdisk/download.html

locate is (file-)caching implementations of path-matching capability of find.
*It does not have other powerful features of find, incl. no criteria other than file-path based.*
*But even unrestricted updatedb+locate runs are massively faster than a single find run.*
Doesn't do anything with FS boundaries OTHER THAN ignoring some file system types.
Sym-links are always treated as terminals (no following), except for
 locate existence determination with -e option.
 A consequence of this is, entries and locate outputs can only include sym-links
 in terminal position.
2 executables:  "/usr/bin/locate", "/usr/bin/updatedb".  [slocate for slocate]
GNU version is in findutils-locate package (available for SuSE).
Centos/RH have mlocate (with same 2 executables).
AIX has slocate which seems to be same except main executable is "slocate".
Newer locate package is "mlocate" but still contains just "locate" and "updatedb" as executables (+ slocate as just a sym link).
By default the default system db will be updated daily by "updatedb" and used by "locate".
This cronjob is provided by all locate packages.
/etc/sysconfig/locate defaults RUN_UPDATEDB_AS to nobody.
updatedb invocations:
  completely replace the target db. ???  man page says:
  Non-root invocations MUST use "-l 0" switch because updatedb attmpts to chgrp on a temp file.
  that "If the database already exists, it data is reused to avoid rereading directories that have not changed"
  Though -U may take a relative path, regardless of how invoked, DB will always contain only absolute paths.
  (old version requires --localpaths=... and --output=... instead of -U -o)
mlocate package, but not my older findutils-locate, provides /etc/updatedb.conf.
It looks like findutils-locate wouldn't even honor /etc/* if you added it.
Old default db is /var/lib/locatedb, new default db is /var/lib/mlocate/mlocate.db.
ArchLinux
    Comes with systemd "updatedb.service" (that simply runs "updatedb")
      and "updatedb.timer".
    Just by mlocate package install, the timer gets enabled but not started.
    /etc/updatedb.conf
        Concern: PRUNENAMES and PRUNEPATHS are space-delimited and don't allow any patterns,
          so how in fuck do you specify a node name containing a space?  \s???
        PRUNEFS are FS types
        PRUNENAMES are unqualified DIRECTORY names
        PRUNEPATHS are absolute DIRECTORY paths
        No patterns supported in this file :(
        Whenever a FS is pruned, the top level node does make it into db!
        As a consequence of db entries never including non-terminal sym-links,
         prune values in conf should never contain sym paths (terminal or no).
        According to doc hints, its optimally efficient on multiple runs, skipping
        directories that haven't changed.  I've verified that it does remove
        entries for removed file.  When you give a different -U, the DB gets
        entirely recreated.
    Non-default DBs:
        # '-l 0' says to not hide unreadable entries from user (the default 1)
        #   GOTCHA:  Without this a chown command is run on the DB, to control locate readability.
        # Only 1 -U db root supported?  There is a way to support multiple, maybe with
        # multiple runs appending.  My ServiceNow discovery scripts do this.
        updatedb -o /alt/db [-l 0] [-U /start/root]  \
        [--prune-bind-mounts {0|1}] [--prunefs 'val'] [--prunenames 'names']  \
        [--prunepaths 'paths']  # where -prune*s are overrides 
        locate -d alt/db
        # If -U dir is under/equals a pruned path then you will get nothing
        # (just the one node if it equals).
    locate:  You always match against absolute path (default) or filename (-b)
        By far most common usage is:  cd search/scope/root; locate "$PWD/*..."
        Therefore current directory never matters.  Nothing relative to $PWD.
          Matters only indirectly for useful idiom of: locate "$PWD/..."
        Glob * and regexp . wildcards do match / characters
         (this behavior is similar to my 'textfile' and 'rgrep' scripts)
        locate -b '\exactFileName'  # This is globbed with no-glob glob char
                                  # ! Obviously must match ENTIRE filename
        locate -b fileNameSubstr
        locate -b 'fileName*Glob' # Any glob char and get glob pattern matching
                                  # !! Must match ENTIRE filename
        locate '\/exact/file/path'  # This is globbed with no-glob glob char
        locate file/path/substr
        locate 'file/path*Glob' # Any glob char and get glob pattern matching
                                  # !! Must match ENTIRE filepath
        GOTCHA!:  LOCATE_PATH does not replace the default but ADDS TO IT at lower priority‥
                  If -d switch is used then does same to that dir instead of system default.
                  BUT this behavior is highly independent on the locate implementation,
                    varying between GNU, BSD vs. mlocate.
        General capabilities:
          locate [-A] 'PAT1' 'PAT2'  # by default OR; -A switches to AND
          -d db/path  # Leaves existing default or LOCATE_PATH, but this is highest priority
                      # DBs normally in /var/lib/mlocate/ and are usually not
                      # directly readable to users (locate must be setuid or...)
          -b    # match against filename instead of filepath.
                # Seems that if provide other pattern(s) IN ADDITION TO '-b pattern', the other
                # pattern is ignored!  Therefore if give -b then only give the -b pattern.
          -c    # just report match count
          -l 3  # limit
          -0    # like find's -print0, for xargs
          -V    # version
          --regex  # Uses extended regex matching instead of globbing
                # For some damned reason !! Match anything WITHIN filename/path
                # I.e., there's an implicit ".*<YOUR_PATTERN>.*"
                # Therefore, for an entire-matching regexp use '^...$'
          -r    # DO NOT USE!  Not portable.
PATTERNS:
    * matches any non-0 number of segments.  No reason to ever specify **.
    Unlike find, there is no forgiveness for spurious slashes.  ...//... will match nothing!
Locate on Git-for-windows
    You get a very inferior version that doesn't support any config file.
    Documented at http://www.gnu.org/software/findutils .  See section 8 for invocation syntaxes.
    Unlike good updatedb, the --prunepath entries are emacs style regexp patterns.
        Seems to conflict with emacs rules, but you must specify spaces with \s.
    Unlike good updatedb, multiple localpaths are supported.  Again space-delimited!

Determine filesystem, even if wrapped by fuse, etc.:
blkid /dev/sdd1
VFAT is just long-filename hack (via special files) on top of FAT* fs.
To determine FAT 12 vs. 16 vs. 32:
    FAT32 == $(dd status=progress if=/dev/sdb1 bs=1 skip=82 count=5)
    FAT16 == $(dd status=progress if=/dev/sdb1 bs=1 skip=54 count=5)  (may have other vals?)
    Otherwise FAT12?

To view dhcp leases.
File /var/lib/dhcp/[db/]dhcpd.leases

FIND
    Finding with funky filename characters.
    GNU-specific:  find ... -print0 | xargs -0...

    Limiting depth, GNU-specific:  find ... -maxdepth 0  # Would do NO recursion

    By default -criteria switches are ANDED together.  To make them OR you must
    use -or switches (and '(', ')' for grouping when that's needed)..

    -prune is portable.  Stops recursion.
        The pruned directories will not match.
        Always use this pattern:
          find <FINDROOTS> <PRUNE_DIR_CRIT> -prune -o <REGULAR_CRIT> -print

    Output value does not indicate match counts or -exec return values, etc.
    It only returns non-zero to indicate error with print itself.
    You must check the stdout/stderr for application-specific error-checking.

    -context takes a pattern with simple * wildcard characters which are
    matched against files' SELinux context strings.

EPSON PRINTERS
For some damned reason, you must search download.ebz.epson.net for
"XP-412" to find the downloads.
Get "epson-printer-utility-<ER>.x86_64.rpm"
http://download.ebz.epson.net/man/linux/utility.html


Root disk duplication at file level:
    cp -ax /src/mt/pt /dest/mt/pt
OR
    rsync -axXSAH src/mt/pt/ dest/mt/pt/
When done, enforce that new dir contains /dev/pts [SUSE ONLY!]
and /proc (both empty).
If dest disk is mounted on temp dir within src disk, then rmdir the tmp dir on
 the dest disk.

xfs get/set label:  xfs_admin -l  /  xfs_admin -L NEW_LBL /dev/X
xfs get/set uuid:   xfs_admin -u  /  xfs_admin -U LBL|nil|generate /dev/X

LOOP device setup
dd to create bin file
losetup /dev/loopX path/to/bin/file   # Creates /dev/* up to loopX

Hard-copy utilizing copy:
cp -al...


Disk partition alignment
http://h20564.www2.hpe.com/hpsc/doc/public/display?docId=emr_na-c03479326&DocLang=en&docLocale=en_US&jumpid=reg_r11944_uken_c-001_title_r0001

To make new disk devices visible/recognized without rebooting
    echo '- - -' >/sys/class/scsi_host/host<n>/scan
To make geometry changes visible to OS (and parted):
    partprobe
OR
    # THIS TACTIC WORKS FOR archlinux ON vmWare!
    cd /sys/class/scsi_device/*/device  # For desired device *
    echo 1 > rescan
    You need to re-run parted to "fix" the disk label (and resize last partition).

cron
loads NO env files
env vars set in the cron file are exported
Don't need to quote values in variable assignments
Can not nest assignments.  I.e. $REFERENCES in assignment values do not work.
    If you include $X in a value, it will expand as $X.
Linux allows blank lines.
Linux allows env var settings of the form:  NAME = tokenone  tokentwo
 (leading and trailing spaces of both name and values are stripped).
 Use quotes to include leading or trailing spaces.
 They are exported too!
Env vars must be set before they are used, but they effect ALL jobs after them.
UNLIKE BASH SHELL, ~ works only at beginning of non-ws tokens.
    This DOES NOT WORK: echo one=~morgan
Operates like +x option in effect.  Unset variables expand to nothing, just
 like plain Bash (not even a '' token), unless you quote.
Silently fails if you assign to cron variable HOME.
TODO:  Look into MAILTO capability to automatically email upon job failure.
cronie package has optional dependency packages to give emailing capability
 (probably needed for MAILTO feature):  smtp-server, smtp-forwarder
Enabling cronie on archlinux:  

SEE unix.txt file for Linux commands that are also available on UNIX.

Determine motherboard.  "Base Board" section of demidecode output.

xfreerdp is RDP client that unlike rdesktop supports modern Windows
(package is freerdp).
  xfreerdp {file.rdp|/v:ip|/v:hostname} /audio-mode:output  \
    /p:'pwd' '/u:dom\user' /cert-ignore {/f|/dynamic-resolution} &
Beware docs say can specify domain as user@dom instead of 'dom\user', but
  beware @ only works for domain names!  (Not ., hostname, or netbios name).
    Dot for local account works great with either '/u:.\user' or '-u user -d .'
I have no idea what /admin switch does.  Docs do not clarify.
Copy & Paste works awesome.  Can copy files from Dolphin into File Explorer.

7-Zip variants:
    7zr:  if using 7z format.  Smallest.
    7za:  supports many formats.  Small and reliable.
    7z:   supports most formats.  Large and less reliable.
GOTCHA: Always specify -spf switch to 7z or it will sometimes wipe directory paths!

Necessary on my workstation to prevent screen tearing for KDE with Plasma:
export KWIN_TRIPLE_BUFFER=1

pgrep -{a|l} [-fx] [-gGuU grpOrUser] extendedPattern
-a: List full command (== ps 'args', 'cmd', 'command') (in addition to pid)
-l: List process name 'comm' (in addition to pid)
-x: exact complete (not substr) match instead of extended regex.
-f: match against full args/cmd/command instead of just process 'comm'
GOTCHA:  'comm' does not necessarily begin 'args/cmd/command'.
  Often args/cmd/command begins with path when comm is a bare word.

VNC:  tigervnc seems best for many years now.  vncviewer bundled with server
  rpm for some damned reason.
See "tech/archLinux-ec2.txt" for setup details.
Can't get krdc to work.
Not sure but it looks like regular setup method is for headless, while the
  other methods are to share an existing physical X11 system.
Big gotcha for regular setup is that by default the daemon dies after one
  client connection (both native VNC and SSH-tunneled).  Just use my systemd
  unit override file to fix that.
SSH Tunneling
  vncviewer's "-via" switch for SSH tunneling doesn't work as directed by ArchLinux wiki.
  So I use other tactic specified there.
  Extremely easy to set up.
    Server side is to use setting 'localhost at /etc/tigervnc/*mandatory.
    Client:
    ssh -fL 9901:localhost:5901 10.1.10.2 sleep 10; vncviewer localhost:9901

If you just mkfs a whole disk device then parted shows a "loop" type label
with a partition 1, but this isn't a real slice and is extremely different
from a gpt label with 1 slice.  The virtual loop isn't real and can't be
mounted whereas the gpt slice 1 must be mounted.

jq:  great JSON transformer
by default it does identity transform . and just pretty-prints output.
GOTCHA:  To specify input files rather than using stdin, you MUST specify a transformation.
  So to just pretty-print file(s), do: jq . file1...
To display only certain attr keys:  jq [-r] .attrName
      jq [-r] .[].attr  # for top-level list
      jq [-r] .products[].cpe.cpeName,.products[].cpe.refs  # for nested lists
      For multiple, delimit specs with commas.
   works if the value is a nested JSON item.
   Walk with usual operators: jq 'login.uris[0],id'
   Remove attrs:  jq 'del(.name, .type)'
   Filter out all but specified attrs:  jq '{key1, key2}' # No work for nested
   Filter out all but specified attrs, with key renaming:
     jq '{n:.name, t:.type}'  # works for nested input attrs
       I think that specified output keys must be at top level.
   Display multiple attributes/arrays:  jq '.trades[] | {a,b}'
-r switch has it output unquoted strings instead of JSON.  Great for using jq from shell scripts.

xclip
NOTHING works to copy files so that can paste across xfreewin into FileManager.
[-i default, loads contents of specified file, if no file specified then stdin]
-o clipboard to stdout
-selection:  pimary (default), secondary, clipboard, buffer-cut?
xclip-copyfile [-p]  # -p preserves path
  and corrolaries: xclip-cutfile xclip-pastefile

hunspell << word
* means work is directly in dictionary
+ means it's a valid alternative, reporting base word
& means bad, with suggestions

Excellent program for testing:  seq

We prefer expect's 'mkpasswd-expect' over 'pwgen'.  WTF: that's a different purpose.

import:  screen-shot of targeted window

ps memory report units.  VSZ KiB, RSS KB

robotty text from scripts:
pacman -S espeakup speech-dispatcher
then check that voices installed:  spd-say -L
and say anything:  spd-say anything