[9fans] plan9port: flock on sunos5.8?

[9fans] plan9port: flock on sunos5.8?

[Axel Belinfante]

yet another instance of 'spending some time on plan9port on sunos 5.8'.

lib9/create.c needs/uses bsd-style flock which is not available by default,
but accessable using /usr/usb/cc
(this is what I figure from the source and  sun's flock man page)
LOCK_SH and LOCK_EX can be found using /usr/ucbinclude/sys/file.h
and there is a ucb compatibility library.

Any good idea for a way around?

I tried a straightforward addition of -I/sys/ucbinclude in 9c
and -L/sys/ucblib -lucb to 9lbiut that is obviously not good enough.
it works for some files breaks breaks on some others :-(

Yet another thing I don't know much about...

Axel.

Re: [9fans] plan9port: flock on sunos5.8?

[boyd, rounin]

look at all the lock calls, write down what they're doing,
avoid /usr/ucb/cc and code a better algorithm.

all that lunix locking stuff is hideous.

i've only ever seen it done right once.

Re: [9fans] plan9port: flock on sunos5.8?

[Bengt Kleberg]

Axel Belinfante wrote:
> yet another instance of 'spending some time on plan9port on sunos 5.8'.
> 
> lib9/create.c needs/uses bsd-style flock which is not available by default,
> but accessable using /usr/usb/cc

the sun man page mentioned things like:

      Use of these interfaces should be restricted to only  appli-
      cations  written  on BSD platforms.  Use of these interfaces
      with any of the system libraries or in multi-thread applica-
      tions is unsupported.

i think that it is very difficult to use the sun flock(), without 
breaking many things.

what is the intended effect? perhaps another lock function (lockf(), 
fcntl(), etc) can be used?
my plan9port is plan9-20040517.tar.gz and does not have flock() in 
plan9/src/lib9/create.c
perhaps the benefits of adding flock() are not worth the problems it 
creates? then flock() could be removed.


bengt

Re: [9fans] plan9port: flock on sunos5.8?

[William Josephson]

On Thu, Jun 24, 2004 at 09:03:08AM +0000, Bengt Kleberg wrote:
> my plan9port is plan9-20040517.tar.gz and does not have flock() in 
> plan9/src/lib9/create.c
> perhaps the benefits of adding flock() are not worth the problems it 
> creates? then flock() could be removed.

It is needed for OLOCK emulation (we have something that
uses the Plan 9 ports stuff and requires a lockfile for
correct operation).

Re: [9fans] plan9port: flock on sunos5.8?

[William Josephson]

On Thu, Jun 24, 2004 at 09:20:02AM -0400, William Josephson wrote:
> On Thu, Jun 24, 2004 at 09:03:08AM +0000, Bengt Kleberg wrote:
> > my plan9port is plan9-20040517.tar.gz and does not have flock() in 
> > plan9/src/lib9/create.c
> > perhaps the benefits of adding flock() are not worth the problems it 
> > creates? then flock() could be removed.
> 
> It is needed for OLOCK emulation (we have something that

Er. DMEXCL. Sorry.  In any case, I don't think the answer
is to make exclusive use impossible.

Re: [9fans] plan9port (: flock) on sunos5.8?

[Axel Belinfante]

> > lib9/create.c needs/uses bsd-style flock which is not available by default,
> > but accessable using /usr/usb/cc

> i think that it is very difficult to use the sun flock(), without 
> breaking many things.
> 
> what is the intended effect? perhaps another lock function (lockf(), 
> fcntl(), etc) can be used?

> my plan9port is plan9-20040517.tar.gz ...

I'm using the cvs access, and updated yesterday or so.


Actually, the problem is rather limited: there are only two calls.
(I posted here mainly because, as I wrote, I knew nothing about it).
they are used to implement the OLOCK mode bit for create and open,
in lib9/^(create open)^.c

according to the sun man page, the compatibility flock builds on fcntl
so I decided to just use fcntl directly.

I added a #include <fcntl.h> to the list of includes and a
struct flock lockdata variable to the routine (create or open),
and then I replaced

	if(flock(fd, (rdwr==OREAD) ? LOCK_SH : LOCK_EX) < 0){
by
	lockdata.l_type = (rdwr==OREAD) ? F_RDLCK : F_WRLCK;
	lockdata.l_whence = SEEK_SET;
	lockdata.l_start = 0;
	lockdata.l_len = 0;
	if(fcntl(fd, F_SETLK, &lockdata) < 0){

which at least compiles ok, and as far as I could figure from
the man pages should have the right effect.


One other thing needs attention in this plan9port version:
libip/eipfmt.c has the following definition that sun cc does
not grok

uchar prefixvals[256] = {
[0x00] 0 | Isprefix,
[0x80] 1 | Isprefix,
[0xC0] 2 | Isprefix,
[0xE0] 3 | Isprefix,
[0xF0] 4 | Isprefix,
[0xF8] 5 | Isprefix,
[0xFC] 6 | Isprefix,
[0xFE] 7 | Isprefix,
[0xFF] 8 | Isprefix,
};

I moved the initialization inside the eipfmt routine.


for the rest of this plan9port version...
compilation when fine, apart from some warnings:

in libdraw/x11-init.c in XParseGeometry
signed/unsigned int warning for &width and &height

in cmd/9660/dump9660.c in convertnames a couple of (void*)
casts where sun cc prefers  (char* (*)(char*,char*))
in cmd/9660/unix.c a superflous `;' at the end of dirtoxdir

furthermore:
gview.c:173: warning: initializer does not fit or is out of range: -5592321
gview.c:174: warning: initializer does not fit or is out of range: -872414977
gview.c:175: warning: initializer does not fit or is out of range: -5635841
gview.c:177: warning: initializer does not fit or is out of range: -1431699201
gview.c:182: warning: initializer does not fit or is out of range: -1431633921
gview.c:184: warning: initializer does not fit or is out of range: -1728013825
gview.c:185: warning: initializer does not fit or is out of range: -1145324545


I do have some own hacks that I did not remove when updating from cvs,
and I did not test (use) much yet, after compiling, but what I have
tried works fine.

Axel.

Re: [9fans] plan9port (: flock) on sunos5.8?

[boyd, rounin]

i found some lunix exclusive use lock code here:

    http://www.insultant.net/tmp/mace/mail.c

sysVile and BSD.

Re: [9fans] plan9port (: flock) on sunos5.8?

[William Josephson]

On Thu, Jun 24, 2004 at 03:43:16PM +0200, Axel Belinfante wrote:
> I added a #include <fcntl.h> to the list of includes and a
> struct flock lockdata variable to the routine (create or open),
> and then I replaced

Send me a patch and maybe once I get the SunRay on my
desk working again I can try rolling the changes into
the tree..