ikrabbe / plan9front Goto Github PK

There is no way to play Touhou on Plan 9.
This *definitely* has to be fixed.

the current kernel relies on the partition table
to be parsed by 9load. this is no more the case
and wont be continued on plan9front / 9bootfat
as it will not work with usb devices and is 
unreliable due to no mapping between bios drives
and plan9 sd device numbering.

proposed fix will be adding disk/fdisk and
disk/prep in boot/boot that scans all drives
and add the partition table entries for us.

if muzgo gets the boot/boot thing going, this is
like 2 lines of rc and we are done.

for now, one has to put the partition table
in plan9.ini in the form:

sdXX=name1 start end/name2 start end/...

note, this has only to be done for the partition
that are needed by the root filesystem, /bin/termrc
later does proper partition table parsing with
the described method.

replace the cwfs auth code and
use factotum instead like fossil
does.

would be nice to have a 9front on a usb thumb drive

yes, it did!

9vx doesn't reuse file descriptors, kfs only has about 8k to use, and I run a 
script which runs 16 commands every minute - 8000 in just 8h20m. I'm reporting 
it here because it's easy to patch up kfs to cope. The patch in the linked 
email should give me about 59 years of uptime. ;)

http://9fans.net/archive/2010/09/156

the front fell off

implement partitions in usb/disk so we
can get rid of partfs.

cwfs is WAAAAAAAAAAAAAAY too verbose

What steps will reproduce the problem?
1. ls /lib/meinkampf
2. cat /lib/meinkampf
3. grep Juden /lib/meinkampf

What is the expected output? What do you see instead?
Expected MEIN KAMPF, or at least the Dugdale abridgment of same. Saw standard 
error message.

What version of the product are you using? On what operating system?
Pre-release snapshot.

Please provide any additional information below.
MEIN KAMPF's copyright has been relinquished for the Dutch and Swedish editions 
and some English ones (though not in the United States). The government of 
Bavaria, in agreement with the federal government of Germany, refuses to allow 
any copying or printing of the book in Germany, and opposes it also in other 
countries but with less success. Traditionally, the /lib directory on Plan 9 
systems has been used to disseminate controversial political tracts such as the 
U.S. Constitution, archives of Internet RFCs, and images of the faces of active 
Plan 9 users. The exclusion of MEIN KAMPF remains a mystery. Conclusion: The 
front of the operating system is attached by a slender thread.

With 9front-20110509.iso.bz2 I installed 9front on 9front.qcow2.img

On reboot it complained about lapic, so I mounted the qcow2.img into my linux 
host and changed "*mp=" to "*mp=1".

Now it boots (with kvm). However after entering glenda as the the username I 
come to the Plan 9 Console:


lib/profile: rc: .: can't open: '/bin/lib' directory entry not found
init: rc exit status: rc 228: error

init: starting /bin/rc


I start rio successfully.
I try to start acme, however:

acme: can't create temp file: '/tmp' directory entry not found

Also, my /usr/glenda directory is empty.

the idea is to modify the installer to allow simple
nondestructive system upgrades.

it all boils down to a simple disk/mkfs, maybe with
asking the user what parts should be upgraded (src only,
binaries) can be handled with proto files.

a file tree compare program would be nice too that walks
a proto file and compares local files with files from
the distribution to get a list of files that will be
affected by the upgrade.

hg produces bogus timestamps.
annoying as hell.

devcons really should handle at least the DEL key to interrupt a running program

9bootfat doesnt handle files that are not
continous on disk. we have properly follow
the cluster chain in the fat. disk/format
puts the files sequentialy on the disk
but removing/copying will fragment the
filesystem.

What steps will reproduce the problem?
1.Try to install Plan 9.

What is the expected output? What do you see instead?
"stupid crap like 'select root of installation' which is totally useless and 
confusion has no reason to exist" -uriel

If you run out of FIDs the system can do nothing except print "out of fids". If 
I remember right FIDs are never freed. 

It seems easy to hit: I ran into it when running a simple recursive ls script 
on sources/contrib. I don't know whether there is a recursive mount in contrib 
or whether du reuses FIDs or what. I think the latter rather than the former.

I wasn't working in 9front, I was working in 9vx with a tree from yesterday's 
Bell Labs iso, but as I recall this is a known Plan 9 problem.

current state with mercurial under plan9 sucks. one can use
linuxemu + equis + xterm to commit from plan9... but this is
a ugly state... anyone want to write a native mercurial client
for plan9? or a simple program that can be used to commit
only?

files on an installed system have wrong permissions, e.g. sys:sys 644 instead 
of 664

stuff that doesnt work:

/* unknown vlong LIST */
foo = cond ? left : right;

/* unknown vlong COMMA */
int i;
vlong v;
for(i=0; i<10; i++, v++){}

happend while trying to compile python on arm...

i have no clue of the compiler right now, but might be a good
opportunity to take a look :)

On builds from the past couple of days, I have not been able to boot the CD-ROM 
past the user: prompt, in VMware or on real hardware. After entering the user, 
nothing happens. Keyboard input is accepted, for a while, but eventually that 
stops being echoed to the screen as well.

it would be nice if one could install a cpu/file/auth server or
stand alone kenfs directly from the installer.

cpu/file/auth server would require:

9pccpuf kernel
bootargs=/dev/sdXX/fscache -a 'tcp!*!567'
noauth option omited from cwfs configuration
nvram initialization

dedicated kenfs:

doesnt need mutch work i think. basicly just copy the
kenfs kernel to 9fat and write a bootsector and a plan9.ini,
reboot

qemu doesnt work

Booting a 9pcf or 9pccpuf built on a post-install system resets the machine 
immediately after printing the contents of plan9.ini. Identical behavior in 
VMware Workstation, VMware Fusion and real hardware.

the -C option should set/enable new cache format on ream
only. cwfs has to auto detect the currenty used cache
format on startup and just use that.

changing Nep in /sys/src/cmd/usb/lib/usb.h broke usb/serial

usb/serial: startdevs: opening #0 /dev/usb/ep8.0
usb/serial: devmain: serial: no endpoints found for ifc 0
usb/serial: no unhandled devices found

Following changes in revision f217fb15770b, add support for FAT32 in format(8).

- remove any trace of 9load from the manpages
- fix plan9.ini manpage (new bootloader doesnt support ini sections and menus)
- cwfs manpage

replace /srv/cwfs.cmd with an alternative mount attach space for cwfs to
give a proper cwfsctl file that synchronizes filesystem commands so one
can do:

echo dump >/dev/cwfsctl
echo halt >/dev/cwfsctl

ect...

write will block until command finishes or error out, also does perm checking 
in fs and avoids dumping random messages to #c/cons

the current timezone handling is suboptimal

1. default timezone should be UTC
2. /adm/timezone/local should be bound

Do you want the total alt+stuff?
(Yes, I do)

What steps will reproduce the problem?
1. Boot Plan 9 on system with Broadcom ethernet.

What is the expected output? What do you see instead?
Expected: #l0. Instead: No ethernet devices detected.

Almost all Thinkpads ship with either Intel or Broadcom ethernet. Intel is well 
supported, Broadcom does not work at all. Several reasonably 
powerful/reasonably priced Thinkpad models (T43, etc.) ship with supported 
graphics but Broadcom ethernet. 

stuff that needs to be done in the installer:

partdisk
prepdisk
configfs
 implement cwfs and kfs support

mountdist
 this stage is kind of useless, and broken (missing cdsh)

copydist
 use disk/mkfs instead of replica to copy files

replace usb/audio with something simple and nice.

if usb/audio fails to do all its control initialization
it will just abort. often the volume control and stuff
is not needed at all...

one can just look for a isochronous writable endpoint
in the right sample frequency and packet size
and just bind its data file over /dev/audio and you
got sound!

The current livecd does not support booting from a USB CD-ROM.

current system defaults are strange...

change defaults to
 * screensize=1024x768x16
 * monitor=vesa
 * dmaon=1

Upon rebooting after rebuilding my system, /dev/consctl is missing. 
Consequently, no users can start rio.

the new usb approach is crap. just linking all drivers
into usbd and forcing them to use this braindead usbfs lib9p
clone is NOT the way todo it.

the system was better before the new usbd, it was just lacking
hotplug...

the way it really should work is this:

- kernel provides basic usb driver/endpoint abstraction (done!)

- there is a small library that helps with common usb related tasks
  like parsing descriptors or enumerating endpoints and devices.
  (we where there once!)

- you just write a program or a fileserver that uses this usb library
  and the usual plan9 libs for the rest. (there too)

- the usb hub driver should provides a "event" file that can be read
  by a simple (and customizable) rc script that returns a easily
  parsable text description of the device plugged in, and that script
  has a big switch statement where is starts the appropriate program
  for the device. (not there)

- that "event" file should be able to handle multiple readers so some
  usb programs can handle ther own plugging events.

Some values are not making it from `plan9.ini` into `#ec` and `#e`.

The following is output from a freshly installed terminal:

{{{
term% cat /n/9fat/plan9.ini
bootfile=9pcf
bootargs=local!/dev/sdC0/fscache
sysname=cirno
mouseport=ps2intellimouse
monitor=vesa
vgasize=1280x1024x32
*nomp=

term% ls '#ec'
'#ec/*nomp'
'#ec/bootfile'
'#ec/monitor'
'#ec/mouseport'
'#ec/vgasize'

term% cat '#ec/vgasize'
1280x1024x32

term% ls '#e'
'#e/*'
'#e/0'
'#e/NDBFILE'
'#e/TIMESYNCARGS'
'#e/apid'
'#e/bootargs'
'#e/bootdisk'
'#e/bootfile'
'#e/cflag'
'#e/cpu'
'#e/cpuflag'
'#e/cputype'
'#e/facedom'
'#e/fileserver'
'#e/fn#cd'
'#e/fn#sigexit'
'#e/fn#term%'
'#e/font'
'#e/home'
'#e/ifs'
'#e/init'
'#e/monitor'
'#e/mouseport'
'#e/objtype'
'#e/part'
'#e/parts'
'#e/path'
'#e/pdev'
'#e/pid'
'#e/pids'
'#e/plumbsrv'
'#e/prompt'
'#e/rcname'
'#e/rootdir'
'#e/service'
'#e/site'
'#e/status'
'#e/sysname'
'#e/terminal'
'#e/timezone'
'#e/user'
'#e/vgasize'
'#e/wctl'
'#e/wsys'

term% cat '#e/sysname'
gnot
}}}


This is the relevant portion `/sys/src/9/boot/bootrc`, as it existed when the 
system was built:

{{{
# convert plan9.ini variables
set init
set user
set nvram
set rootdir
set bootargs
set nobootprompt
set sysname
set debugfactotum
set fs
set fsaddr
set auth
set authaddr
}}}

Add openbsdemu, thus enabling access to OpenBSD's ports/packages. This is a 
wishlist item; while cinap says it wouldn't require much work, I'm not 
presently capable of making it happen by myself.

mountcwfs lists current directory when no cwfs partitions
are found. have to check for if(! $#files 0) before doing ls

need todo a apm call in bootloader and
pass some parameters to kernel for apm
to work.

"disablin dma: not on a busmastering controller"
we REALLY need to fix this.

After typing approximately 100 characters in a drawterm session, the keyboard 
stops responding. Mouse and everything else still seem to work. Disconnecting 
and starting another drawterm session restores keyboard functionality for 
another 100 or so characters.

Console keyboard does not seem to be affected.

implement vesa screen blanking
in the vesa driver.

realemu has to be taken into account
here as the blank is called from
interrupt context.

Need a way to keep stuff out of snapshots.

the owners and groups (and possibly the permissions) are
not right on the iso builds. this is no big problem but we
need a better proto file that sets right owner and group perms
for disk installation later

Purloined from http://9fans.net/archive/2011/04/224

From: Rudolf Sykora <rudolf.sykora@gma...>
[...]
-- awk man page does not list the substr(s,m) function (... version
without explicit length)
-- awk man page does not mention the possible use of '&' in (g)sub() functions
-- sed man page mentions the use of '&' only in examples
-- sed man page does not mention backreferences (like \1) at all

sed(1) could probably be fixed by copying verbatim from ed(1).

provide e820 memory map to kernel in
same format as used by eriks 9atom
kernels.

sam needs scroll wheel support.

consistency with other plan 9 apps would also be nice - arrow keys scroll a 
full page in sam but 1/3 of a page in everything else. i hope we will not 
implement the craziness of p9p's sam, scrolling a full page down and only 1/3 
up.