This may be misleading:
cat xattrtest1.c
#include <stdlib.h>
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
int main(int ac, char *av[])
{
int dirfd;
int xattrfd;
dirfd = openat(AT_FDCWD, "foo", O_CREAT|O_RDONLY|O_NONBLOCK,0666);
xattrfd = openat(dirfd, ".", O_RDONLY|O_XATTR);
fchdir(xattrfd);
system("/bin/pwd");
chdir("..");
system("/bin/pwd");
return (EXIT_SUCCESS);
}
$ truss ./a.out 2>&1 | fgrep chdir
fchdir(4) = 0
chdir("..") Err#20 ENOTDIR
This makes sense. As the Exceed Windows NFS server people say, the
O_XATTR directories have a life time as long the directory is open,
and used, and afterwards its discarded, and rebuild anew from Window's
list of alternate stream data.
On Solaris its different, the O_XATTR directories are real
directories, but have no parent. The /sbin/sh, /bin/sh, /usr/bin/sh,
/usr/bin/csh, /usr/bin/bash etc shells have the convention that doing
a cd .. in a O_XATTR returns to the filesystem node the O_XATTR
attributes are for, but this is only a convention at shell level, and
not at C level.
This might be confusing, until you notice that Solaris utilities do
not use openat(fd, ...,O_XATTR), but instead use attropen():
NAME
attropen - open a file
SYNOPSIS
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
int attropen(const char *path, const char *attrpath, int oflag,
/* mode_t mode */...);
DESCRIPTION
The attropen() function is similar to the open(2) function
except that it takes a second path argument, attrpath, that
identifies an extended attribute file associated with the
first path argument. This function returns a file descriptor
for the extended attribute rather than the file named by the
initial argument.
The O_XATTR flag is set by default for attropen() and the
attrpath argument is always interpreted as a reference to an
extended attribute. Extended attributes must be referenced
with a relative path; providing an absolute path results in
a normal file reference.
RETURN VALUES
Refer to open(2).
But seriously, *STRONGLY* prefer to deal with this using cd -@, and
have cd .. use the parent node name to return me to the parent node,
as the original patch for cd -@ did.
That is much more *user* friendly.
For scripts it may be better to extend /dev/file/ to have /dev/file/xattr:
ksh -c 'mkdir -p foo2 ; redirect {d}<foo2/ ; redirect
{n}<>/dev/file/xattr:/dev/fd/$d/myxattr ; print -u$n "bla" ; cd -@
foo2 ; ls ; cat myxattr' myxattr SUNWattr_ro SUNWattr_rw
bla
Olga
On Fri, Aug 30, 2013 at 7:40 PM, Simon Toedt <[email protected]> wrote:
> On Fri, Aug 30, 2013 at 6:32 PM, Glenn Fowler <[email protected]> wrote:
>>
>> On Fri, 30 Aug 2013 10:00:35 -0400 Glenn Fowler wrote:
>>> On Fri, 30 Aug 2013 13:04:17 +0200 Simon Toedt wrote:
>>> > On Fri, Aug 30, 2013 at 7:11 AM, Glenn Fowler <[email protected]>
>>> > wrote:
>>> > >
>>> > > On Fri, 30 Aug 2013 04:53:04 +0200 Cedric Blancher wrote:
>>> > >> On 29 August 2013 19:45, Glenn Fowler <[email protected]> wrote:
>>> > >> >
>>> > >> > the AT&T Software Technology ast alpha 2013-08-29 source release
>>> > >> > has been posted to the download site
>>> > >> > http://www.research.att.com/sw/download/alpha/
>>> > >> > the package names and md5 checksums are
>>> > >> > INIT 132e0403af573fa1cb1e202267fedeb8
>>> > >> > ast-open 334615fb3a652575106194c281d27b5c
>>> > >> > ast-ksh ebcc56d9ab673aaafbb163d6eee1a93c
>>> > >> > the md5 sums should match the ones listed on the download page
>>> > >> >
>>> > >> > this is still a work in progress, but we are getting closer to a beta
>>> > >
>>> > >> No, you don't. This release if OFFICIALLY broken beyond usability and
>>> > >> USELESS.
>>> > >> You really had to try and tinker with cd -@ again, did you? I'm really
>>> > >> angry. cd .. no longer works within NFSv4 xattr directories, e.g.
>>> > >> echo "" >x ; cd -@ x ; touch xattr ; cd .. ; rm x now fails with a cd:
>>> > >> /home/ced/prod4/test19/x//@//..: [Not a directory]
>>> > >
>>> > > hey gsf, there's a problem with the way you did foo
>>> > > here's a sequence of commands / code that shows the problem
>>> > >
>>> > > its a bug
>>> > > patch is below
>>> > > with ksh builtins like wc you should be able to do things like
>>> > >
>>> > > wc /home/ced/prod4/test19/x//@//xattr
>>> > >
>>> > > and you get a pwd that can be passed on to other processes and used
>>> > > long after your process dies
>>> > >
>>> > > /home/ced/prod4/test19/x//@//
>>
>>> > Is this some late Aprils Fool joke or is this serious? How do you
>>> > prevent that someone clobbers his path together from variables or
>>> > through printf and accidentally creates a path like that?
>>
>>> > ksh -c 'typeset -a a=( /foo bar /baz/ /@ / /append/ /bum ) ; IFS="/" ;
>>> > printf "%s\n" "${a[*]}"'
>>> > /foo/bar//baz///@////append///bum
>>
>>> > That strikes me as extra risky strategy.
>>
>>> alright
>>> I got most of the feedback I needed to understand the (non-)namespace
>>> working on an update
>>
>>> I do have one more question from the first "cd -@" example
>>
>>> cd -@ t.c
>>> cd ..
>>
>>> if t.c is a regular file and O_XATTR(t.c) really hangs in a nameless part
>>> of the filesystem
>>> how can ".." inside the O_XATTR(t.c) xattr dir have any meaning other than
>>> ".. points to the parent object"
>>> indeed ls -ail inside O_XATTR(t.c) shows .. as a regular file
>>> and /bin/pwd (without hints from $PWD) chokes because .. is not a directory
>>
>>> so if the sequence
>>> cd -@ t.c
>>> cd ..
>>> really does get you back where you started in a solaris shell:
>>> how does the documentation rationalize that?
>>
>> I think I have the answer to my own question
>> when you "cd -@" you are not in a place named by a path but instead in a
>> place named by an fd
>> so all relative paths in that fd dir are relative to the fd
>> which then makes "cd .." from within "/dev/file/xattr@somepath" ambiguous
>> when viewed
>> as a path operation "/dev/file/xattr@somepath" + ".." =>
>> ""/dev/file/xattr@somepath/.." (ambiguous)
>>
>> I don't have a solaris with a shell with working "cd -@"
>> what is is the output of the sh pwd in a solaris /bin/sh:
>>
>> cd -@ t.c
>> cd ..
>> pwd
>>
>
> /sbin/sh -x -c 'echo "" >foo ; /bin/pwd ; cd -@ foo ; /bin/pwd ; cd ..
> ; /bin/pwd ; :'
> + echo ''
> + 1> foo
> + /bin/pwd
> /home/sim
> + cd -@ foo
> + /bin/pwd
> pwd: cannot determine current directory!
> + cd ..
> + /bin/pwd
> /home/sim
> + :
>
> /sbin/sh -x -c 'echo "" >foo ; pwd ; cd -@ foo ; cd .. ; pwd ; :'
> + echo ''
> + 1> foo
> + pwd
> /home/sim
> + cd -@ foo
> + cd ..
> + pwd
> /home/sim
> + :
>
> Simon
> _______________________________________________
> ast-developers mailing list
> [email protected]
> http://lists.research.att.com/mailman/listinfo/ast-developers
--
, _ _ ,
{ \/`o;====- Olga Kryzhanovska -====;o`\/ }
.----'-/`-/ [email protected] \-`\-'----.
`'-..-| / http://twitter.com/fleyta \ |-..-'`
/\/\ Solaris/BSD//C/C++ programmer /\/\
`--` `--`
_______________________________________________
ast-developers mailing list
[email protected]
http://lists.research.att.com/mailman/listinfo/ast-developers
