windows-xp/Source/XPSP1/NT/base/subsys/posix/status.psx

The POSIX subsystem is modeled closely on the OS2 subsystem - I've re-arranged
and renamed the files so that it will be easier to compare functionality.
There is a problem with POSIX running in a window when a process forks
that is probably due to the fact that keeping track of mulitple sessions
is not done properly. Look at session.c in os2 and compare that code wiht
posix. 

The arguments to exec are not getting through properly.

It would be a good idea to check build.log after a complete build for warnings.
I haven't done that in a while.

The posix header files are not complete. They should be fleshed out (look
at the xenix headers on you mail machine) so that the NIST tests will compile
and so th languages guys know where conflicts may arise when they do the
merge of the posix headers and c runtimes into their main source tree.

On exit, smcomplete session has been returning c0000008 for a long time 
and I haven't taken time to track that down.

See note 6 below re Console IO. I have no test (we need a simple shell)
that does stdin as yet so that is not tested.

Good luck!!


POSIX STATUS

			Nov 15, 1990		Feb 15, 1991

Total functions:	97
Done			44 			    60
Optional (debatable)	12
			--
Unimplemented		41			    25

Process Primitives

	fork	    /
	execl	    /
	execv	    /
	execle	    /
	execve	    /
	execlp      convert file to path using PATH env variable
	execvp  	"

	wait 	    /
	waitpid	    /
	_exit	    /
	kill	    /
	sigemptyset /	NOTE 1
	sigfillset  /	
	sigdelset   /
	sigismember /
	sigaction   /
	sigprocmask /
	sigpending
	sigsuspend  /
	alarm	    /
	pause	    /
	sleep	    /
	
Process Environment

	getpid	    /
	getppid	    /
	getuid	    /
	geteuid	    /
	getgid	    /
	getegid	    /
	setuid
	setgid
	getgroups
	getlogin
	cuserid

	getpgrp	    /
	setsid	    /
	setpgid	    /
	uname	    /
	time 	    /
	times	    /
	getenv
	ctermid
	ttyname
	isatty
	sysconf	    /

Files and Directories

	opendir	    /	NOTE 2 re file times updating
	readdir	    /
	rewinddir   /
	closedir    /
	chdir	    /
	getcwd	    /
	open	    /
	creat	    /
	umask	    /
	link		NOTE 3
	mkdir	    / 	NOTE 4
	mkfifo	    /  	NOTE 4
	unlink 		NOTE 3
	rmdir 	    /   
	rename		NOTE 3 - and how does this relate to C runtime version??
	
	stat	    /	Always fail? - see comment fdapi.c, NT security chapter
	fstat	    /	NOTES 2, 3, 4
	access	    /   Needs RtlMakePosixAcl from JimK
	chmod	    /   Needs RtlIntrepretPosix and NtSetSecurityObject (JimK) 
	chown 	    /   Some open issues - su priv, security routines work?
	utime	    /   Need NT to POSIX uid/gid translations
	pathconf    /   Done for now - need NtFsControlFile for configurable
	fpathconf   /
	

Input and Output Primitives
	
	pipe	    /
	dup	    /
	dup2	    /
	close	    /
	read	    /
	write	    /
	fcntl
	lseek	    /
	
Device and Class Specific Functions

	cfgetispeed	( All optional ???)
	cfgetospeed	( tonye may be source for tty support - he's doing)
	cfsetispeed	( serial drivers)
	cfsetospeed
	tcgetattr
	tcsetattr
	tcsendbreak
	tcdrain
	tcflush
	tcflow
	tcgetpgrp
	tcsetpgrp
	
Language-Specific Services for the C Programming Language

	setlocale
	fileno	    /
	fdopen
	sigsetjmp
	siglongjmp
	tzset

System Databases

	getgrgid
	getgrnam
	getpwuid
	getpwnam
		
-----------------------------
NOTE 1
	Re: signals
	In order to make NT exception handling available to PSX processes, 
	check for exceptions in apiloop using exception port and convert 
	to signals.

NOTE 2
	Re: file times updating. 
	Implementing this for all relevant file systems functions will 
	need to be postponed until NTFS. GaryKi has been given a list 
	of PSX requirements for times updating. For now POSIX uses the
        times from NtQueryInformationFile for all but local pipes. Local
        pipe times are updated in the subsystem.
	See filereqs.psx in POSIX directory.

NOTE 3
	Re: link and unlink and rename - 
	Implementing this will need to be postponed until NTFS. The Pinball
	FS does not implement links right now since OS2 doesn't use them.
	The POSIX file system will be NTFS. GaryKi and TomM have been 
	contacted about this requirement. Check changes to 1.003a re
	symbolic links.

NOTE 4
	Re: security
	When security code is complete - these functions need to check
	the ACL and/or UID GID SID info appropriately. See Security section
	of workbook - chapter 14. Mike Massa (mikemas) is working on
    	security as he ports streams over to NT and POSIX. He was planning
	to write conversion functions (cracking ACLs to get POSIX rwx access)
	that should be used by the subsystem also.

TODO

1. Details are missing in the pre-ellen functions (such as setting uid, 
   file times etc in file functions). Go over all functions carefully to 
   see that they are FULLY functional re: spec.

2. Operations on local pipes need to update the times in the ionode.
   Pipe() initializes these. Open, read and write (other??) need to 
   update manually.

3. When security is in place re: ACLs on files - all functions manipulating
   and/or accessing file modes (access, open, fstat, chmod) must translate 
   between POSIX file permissions and NT ACLs and update appropriately.

4. Many routines in fdapi have to open the file (NtOpen) to get a handle 
   for other routines. Check NtOpens actions on the file times of the file 
   vs what's supposed to happen to those times via the posix routine.

5. Check into all routines that say they must be fixed to work with
   view memory (grep for view).

6. (This may negate some comments above.) ALL I/O except OPEN, PIPES, and FILE
   LOCKING needs to be moved into the client and out of the server. read,
   write, stat, etc. NOTE: FileRead and FileWrite functions in psxfile.c use 
   NtReadVirtualMemory and NtWriteVirtualMemory to transfer the data
   to be read from or written to the file between the user buffer and the 
   server buffer. This is a STOPGAP ONLY measure required for POSIX
   to work now that NTWriteFile no longer accepts a parameter specifying that
   the write should be done to an alternate process' data. When this code is
   moved out into the client and done strictly in the client's space, this
   code is obviously unnecessary.

   Currently, Console IO is caught at the client level (see IS_CONSOLE_IO
   in dllio.c) and vectored off to the posix.exe to do the proper windows
   crt read and write.. This should be done as os2 does it (os2 has all io
   on the client side) using a vector of functions and not special-casing.
   (See IoVectorArray in os2\client\dllhandl.c).

7. There are lots of places in the client code where the user parameters
   (ptrs to) are taken at face value and not checked. These should be probed
   for read and write access before using. Esp uname times time wait waitpid etc
   in pdxproc.c).

8. The POSIX C Runtimes (crts) are incomplete and patched together and
   not merged into the regular source tree. Languages (jeffrob) are
   supposed to do this 'any time now'. Until this work is done (which
   includes supplying header files that do not conflict with posix header
   files), building apps and tests is iffy. NIST testing is virtually 
   impossible. Currently I have a patched together studio.h in inc\posix
   so that printf will work.
   When this work is done, mainCRTstartup() may change names. It will have
   to be changed in rtl/startup.c to match what is in the runtime.
   CONTACTS: jeffrob, gregf (jeff's boss) Get LOUP to follow up.

9. MIPS builds with only one warning - due to that fact that the MIPS
   version of the CONTEXT_TO_PROGRAM_COUNTER macro in ntmips.h is
   wrong (should be cast to a PVOID). Update - I haven't built for MIPS
   since March 1991.

TESTING issues

STAT - test 'real' code when NtQuerySecurityObject is fixed.

NIST tests - SteveSc has loaded them onto a system and will try to
build with the posix runtimes that mukund did last summer. THis may
be a frustrating experience until we have 'real'crts and headers from 
languages.
*************************************************************************
CRT INFO - this is mail from the student who fixed up a POSIX CRT for me.
This work needs to be merged into the languages CRT tree and built
regularly.

Here's a list of the POSIX related stuff that has to be done to assimilate the stuff back into the tree:

Heap, String, Convert, Misc: The code is the same as WIN32.

Stdio : Changes littered along the souce code. Could  probably be put in the source tree directly.

Time : An issue that will be resolved. The POSIX version is based on the C7 code, and uses floating point.

Startup: Code based on WIN32 version. Only stdenvp.c and crt0msg.c has any new code. 

Setjmp and Lngjmp: to be done.

Include files: Most of them could be put in the tree directly. There are a few POSIX specific include files
(dirent.h, grp.h, pwd.h, sys\wait.h, sys\utsname.h, sys\termios.h, sys\times.h, unistd.h) that have to be added. Additional macros are defined in the POSIX version of stat.h and the POSIX version of types.h defines some additonal types. 

The POSIX source files can be found on \\crt1\vangogh in a 
directory called posix. The posix-specific include files can
be found in \posix\psxinc.

I had a really good time working here this summer and I want to thank all of you for making it such a good experience.
			Mukund

The server is:

	\\vangogh\rdcrt1	password: QIEPCKE

The tree is: \posix.

-Gregf

*************************************************************************

RtlUnixPathToNtPath()

Consult with SteveWo to sort out what should be done here.
We need to be able to designate the posix file system(s) (ntfs only due
to links needing to be supported) and how to translate pathname 
(eg \posix) to filesystem (eg what used to be \\harddiskX\partitionY\posix)
for NT.

Call RtlPosixPathToNtPath in client wherever a path argument is passed in.

There is currently a RtlDosPathToNtPath() function to use as a guide.