ibcs-us-4.2.1 2023-09-28 Russell Stuart * new: Revert single inode number namespace for all devices * bug: fix short_inode_map() infinite loop. ibcs-us-4.2.0 2023-09-28 Russell Stuart * new: Single inode number namespace for all devices ibcs-us-4.1.8 2023-09-28 Russell Stuart * bug: fix short_inode_map() infinite loop. ibcs-us-4.1.7 2023-09-19 Russell Stuart * new: Port to Debian bookworm. ibcs-us-4.1.6 2021-08-25 Russell Stuart * new: Improve error messae when mmap fails. * new: Port to Debian bullseye ibcs-us-4.1.5 2019-10-17 Russell Stuart * bug: Fix usage error for unrecognised flag. * bug: Fix signal handling. * bug: Fix debug source paths. * new: Port to gcc 8 ibcs-us-4.1.4 2019-06-21 Russell Stuart * new: It seems to work, so probbaly won't be touched again for a long time, if ever. Make final additions to comments to job my memory in case I do have to touch it again. ibcs-us-4.1.3 2019-06-08 Russell Stuart * bug: Fix spelling errors in man page. ibcs-us-4.1.2 2019-06-07 Russell Stuart * bug: usage message needed some love. * bug: fix signal handling. * bug: don't decrement usage count on Linux close. * new: move trace closer to the old code, so diff's are easier. * new: Improve error messages. ibcs-us-4.1.1 2019-06-04 Russell Stuart * bug: move to semver versioning. ibcs-us-4.1 2019-06-03 Russell Stuart * bug: wasn't statically linked, so had un-needed libc-i386 dependency. ibcs-us-4.0 2019-06-03 Russell Stuart * Change to user space implmentatio, rename ibcs-is. ibcs64-3.9.5.1 2014-05-29 Russell Stuart * Fix debian packaging - make uscan use the sf redirector. ibcs64-3.9.5 2014-05-29 Russell Stuart * New home is on sourceforge. * Update README.txt. 2011-03-08 Russell Stuart * supports kernel 2.6.32. - xen support removed, as required symbols not exported by 2.6.32. * now works on amd64. - rewrote entry_64.S, altering how LDT entries are used. - renamed all _NR_syscall constants to _NR_IBCS_syscall - introduced SYS_NATIVE so native AMD64 syscall's can be made. - replaced "[unsigned] long" type with [u]32 in lots of places. * added ABI_NOOPT to turn off free compiling to make debugging easier. 2010-01-01 Alex * support for Kernel 2.6.29 added * support for Kernel 2.6.30 added * use ia32_sys_call_table instead of sys_call_table for 64-bit 2009-09-18 John Morris * support for Kernel 2.6.31 added 2009-01-31 Alex ibcs-3.8 * support for Kernel 2.6.28 added * created local "uts_sem" to overcome missing Kernel export 2008-10-31 John Morris * support for Kernel 2.6.27 added 2008-10-08 Christian Lademann * short inode mapping and "/proc" support 2008-07-13 Alex ibcs-3.7 * support for Kernel 2.6.26 added * New parameter ABI_ELFMARK - allocate LDT only for marked ELF * Fixed mmap(0) security fault when loading COFF/XOUT binaries * New Execution Feature ABI_Y2K_BUG - should be set for programs that missed 29/02/2000. * Xen 3.x support - dom0(production) and domU(experimental) * 64-bit compilation 2008-07-09 Lutz Sammer * "read_dir" bug fixes * fixed Xenix locking for newly created files 2008-04-27 Alex ibcs-3.6 * support for Kernel 2.6.25 added * Xenix-286 emulation is compiled by default * "lcall.c" was changed to use "modify_ldt" system call instead of "Load_LDT" macro 2008-03-15 Alex ibcs-3.5 * support for Kernel 2.6.24 added 2007-10-11 Alex ibcs-3.4 * support for Kernel 2.6.23 added 2007-06-19 Mauro Tassinari * support for Kernel 2.6.22 added 2007-05-03 Alex * support for Kernel 2.6.21 added 2007-04-08 Alex ibcs-3.3 * in iBCS-3.2 "late ABI detection" of ELF binaries on 2.6.20+ did not work without "binfmt_coff.ko" loaded * it is now possible to define personality on individual basis using "PERFLAGS" file. Reasons for the change - "COFF clues" or "ELF marks" are sometimes insufficient to determine correct personality; "ELF" executables called from a non-zero Domain will incorrectly remain in that Domain - UW7 module will execute in Domain 3 (instead of 14) if called from an OSR5 application. * new Configuration Parameter "ABI_SIGRETURN" - if set to "Y" (default in iBCS-3.3+) an extra code segment will be allocated per thread in order to restore application context after returning form all signal handling routines. Reason for the change - default "vsyscall sigreturn" fails for COFF modules in kernels 2.6.18+. * on RHEL5 and FC6 first "sys_brk" call from an ELF module compiled on SVR4 or OSR5 would fail and was usually followed by "Segmentation fault" - this has been fixed. * OSR5 "Rewind Directory" is now recognised - on previous versions of iBCS/ABI SCO "getcwd" library call would occasionally return ENOSYS. * new Inode and Device mapping functionality was implemented - details are in "INODEMAP" and "DEVICEMAP" files 2007-03-04 Alex ibcs-3.2 * support for Kernel 2.6.20 added * fixed compilation on Red Hat line of products * fixed "sys_ipc" parameter passing * replaced "svr4_getdents" with "sys_getdents" in "sco/sysent.c" 2007-02-17 Alex ibcs-3.1 * support for kernels from 2.6.5 to 2.6.19 2007-02-05 Alex ibcs-3.0 * pre-relese of iBCS3 based on abi patch for Suse Linux 10.1 * fixed "socket operation on non-socket" error 2002-01-03 Christoph Hellwig linux-abi­2.4.17.0 * rework/cleanup lcall7 syscall handling * make sure x.out sections aren't loaded into illegal areas * fix SVR4 mmap() to always assume post-SunOS4 behaviour 2001-11-24 Christoph Hellwig linux-abi-2.4.15.0 * simplify wyse socket handling, use native syscalls directly where possibles else use normal calling conventions * fix compilation for non-i386 machines (Olaf Hering) * fix alignment checks in binfmt_xout (based on patch by Joerg Ahrens) * rewrite abi_brk: fix (theoretical) races, check for page alignment * don't try to take a non-existant lock in xnx_nap, cleanup xnx_nap. * fix value reported by sysi86(SI86MEM). 2001-11-18 Christoph Hellwig linux-abi 2.4.14.0 * small binfmt_xout bugfixes (Joerg Ahrens) * move isc into it's own personality * make all personalities independant of abi-ibcs * merge socket handling into abi-wyse 2001-10-22 Christoph Hellwig linux-abi 2.4.12.0 * big mmap rewrite | correctly check for flags in svr4_mmap | use svr4_mmap for the UnixWare personality as well | OpenServer has it's own implementation now as it supports rather different flags than SVR4 & derivates * more header cleanups - all helper routines are now in abi/util/*.h 2001-10-04 Christoph Hellwig linux-abi 2.4.10.0 * add MODULE_DESCRIPTION/MODULE_AUTHOR/MODULE_LICENSE tags * don't use alternate prefix for directories (David Mosberger) * fix non-modular buil * cleanup stat handling | use Alexander Viro's kstat concept | all emulations now have their own xstat implementations | fixed overflows * add full-fledged types.h headers for all peronalities 2001-08-06 Christoph Hellwig linux-abi 2.4.9.0 * fix missing solaris socket exports * move Xenix support from ibcs to sco module, make unconditional * UnixWare LFS fixes * make the actual code work on the Alan Cox tree 2001-08-06 Christoph Hellwig linux-abi 2.4.7.0 * some final polishing for segmented 286 x.out binaries (Joerg Ahrens) * fix SPX handling (again) * fix setpgrp and cleanup procids while we're at it * update SCO syscall table to match OSR506a 2001-07-19 Christoph Hellwig linux-abi 2.4.6.1 * rewritten /dev/socksys handling * fix more non-x86 breakage (Olaf Hering) * port solaris native sockets from sparc64 * add sysctl support for runtime-tweaking * add cxenix to the ibcs sysent table for Xenix/386 (Jürgen Günther) * (re)add support for segmented x.out binaries (Joerg Ahrens) | I probably broke it while rebasing against -CURRENT * reject IRIX binaries on mips32 (David Woodhouse) 2001-07-09 Christoph Hellwig linux-abi 2.4.6.0 * svr4/signal.c typo fix * make cxenix debug printk's less verbose (Christian Lademann) * misc small bugfixes (Christian Lademann) * compile fixes for !CONFIG_ABI on non-x86 architectures 2001-06-18 Christoph Hellwig linux-abi 2.4.5.1 * rewrite of abi_sigsuspend() to make it work as expected * fix cxenix syscall table * actually initialize socksys 2001-06-09 Christoph Hellwig linux-abi 2.4.5.0 * rewrite of kernel exec_domain support * make faking of SCO utsname information configurable * get rid of SYS() and sysfunc_p (Arjan van de Ven) * fix socket families (Stephan Springl) * fix SCO signal mapping (Stephan Springl) * fix SCO (and Wyse) error mapping * continued source tree restructuring 2001-04-31 Christoph Hellwig linux-abi 2.4.4.0 * disable tracing by default * rewrite of the SYSV IPC code * add support for SCO OpenServer 5 ELF binaries * fix an error in binfmt_coff that does not clear the return value * provide generic template macros for stat and friends * start of source tree reorganization 2001-03-30 Christoph Hellwig linux-abi 2.4.3.0 * fix shared library support for COFF * x.out is now supported (again) * redo setting of personality based on ELF headers * get rid of CONFIG_ABI_TRACE (and a lot of nasty ifdefs) * added documentation, mostly from iBCS Thu Nov 5 21:45:41 GMT 1998 * Fixed the bug that lets reads and writes on /dev/socksys take out the kernel. -- Mike * Added a kludge around for Sybase isql which appears to have initialised a struct sockaddr using: sa.sa_family = htons(AF_INET); I guess this works on SCO? -- Mike Sat Oct 31 14:15:44 GMT 1998 * Documentation update. * If we are doing a T_BIND_REQ with a null address we can ignore the connection backlog. Listening makes no sense but some software seems to set it 1 for the hell of it. -- Mike Tue Oct 20 20:34:20 BST 1998 * Changed the debug messages printed when exec tracing sees a pointer error on an arg or env entry. It may not be a "real" error. It could be that we just cannot get a free page or perhaps the entry is too long for getname(). Whatever, we do not really care at this point anyway. -- Mike Sat Oct 17 20:24:59 BST 1998 * Documentation update for today's release. -- Mike Thu Oct 15 21:39:16 BST 1998 * When checking personality only look at the base personality and ignore the flags. There is at least one case I can think of, PER_SHORT_INODE, where the personality should be considered the same even when the flags are different. -- Mike * Added the location of the kernel source as an option in the CONFIG. Hopefully this will make it a little more obvious that iBCS is *dependant* on the kernel. -- Mike * Set SO_BSDCOMPAT on sockets. SYSV has BSD handling of ICMP errors on UDP sockets rather than RFC conformance. I think... -- Mike Wed Oct 14 22:50:48 BST 1998 * When using some user stack as temporary work space we only leave 1k between the work space and the real user stack. This is because Informix Dynamic Server uses threads with limited stack space and the idea of dropping 16k below what would normally be touched on a SCO system worries me a bit. -- Mike Sun Oct 11 11:58:58 BST 1998 * Changed the Tools Makefile to install mkmnttab in /usr/bin. Informix Dynamic Server _requires_ a valid /etc/mnttab. -- Mike Sun Oct 11 11:44:58 BST 1998 * When doing an I_PEEK or I_NREAD STREAMS ioctl we have to do a probe on sockets in case we need to generate a control message (e.g. T_CONN_IND) which should then be offered. This also allows T_CONN_IND messages to be returned in several MORECTL divided pieces should it ever be necessary. This is required by Informix Dynamic Server which does a poll then I_PEEK on a listening socket before actually picking up the connection. Dynamic Server is system call hell :-). -- Mike * When we do a timod operation via an ioctl there seems to be no way to cope if the returned message is larger than the original request. We can't expect the app to come back for the extra and returning MORECTL to the app seems to confuse things. Therefore we just discard the excess. This case is required when an app (e.g. Informix Dynamic Server) tries to set SO_LINGER using an int instead of a struct linger via an ioctl. -- Mike * Added some debug to the poll handler so we can see what is being polled. -- Mike * More debug message changes. -- Mike * Wrap SO_LINGER handling so we can survive if an int is passed instead of s struct linger. It seems that other systems do not do such robust error checking as Linux. Note that if an int is passed we probably do not do what other systems would but at least we don't give unexpected errors to broken applications. -- Mike Sat Oct 10 15:58:29 BST 1998 * Added an entry to Doc/HINTS noting the the Informix Dynamic Server install requires an /etc/mnttab. -- Mike * Wrong option length being passed through to setsockopt() from the TLI/XTI optmgmt routine. -- Mike * When stating files only map "ext2" to "HTFS" for SCO programs - and then only if the filesystem name really was "ext2". -- Mike * SCO has a different value for the _SC_PAGESIZE argument to sysconf() than SVR4. Informix wants it to work. -- Mike * Hmmm, Informix Dynamic Server may or may not be looking at the major number on a socket. (I _think_ it does sometimes). It definitely checks that sockets have S_IFCHR set in their file modes though... -- Mike * Changed some debug messages to make it easier to read logs when more than one iBCS handled process is running. -- Mike * If we get a STREAMS ioctl on a file descriptor only initialize the socksys handler on it if it is not already a socket but does belong to the socksys device. -- Mike Thu Oct 8 21:20:43 BST 1998 * When punting a pseudo device transport provider to a socket copy the device numbers between the inodes. This is because Informix Dynamic Server stats the pseudo device node (/dev/socksys) then stats the socket it gets and compares the major numbers. Sheesh... -- Mike * If socksys_close gets something that is not a socket it is probably the original pseudo device which was opened just for an ioctl or two. This is not an error. -- Mike * Some programs, notably the Informix Dynamic Server install program compare the filesystem name from a stat or sysfs against their own hard coded list of known "real" filesystems. Hence we call "ext2" "HTFS" instead for SCO processes. -- Mike * Informix DS also checks the release part of the utsname data to make sure it says "3.2v5.0.0". So now the utsname struct looks even more like SCO. -- Mike * Only dump the ctl part of messages in timod if tracing is enabled. -- Mike * SCO has statvfs in the same place as SVR4 but they have added a field slap bang in the middle of the struct to allow some inodes to be reserved for root in much the same way that some blocks may be reserved. -- Mike Thu Oct 8 20:48:46 BST 1998 * On timod bind and connect the size of the address is allowed to be the _used_ bytes whereas the underlying socket functions require the _total_ bytes. This is a problem for Sybase. -- Mike Sun Aug 30 21:49:46 BST 1998 * Changed socksys.c to handle new file_operations struct in kernel 2.1.119 and later. -- Mike Sat Aug 22 19:57:01 BST 1998 * Fixes and additions to user pointer checking in ipc.c -- Mike * Changed some more KERN_ERRs to KERN_DEBUGs in the STREAMS ioctl support. -- Mike Fri Aug 21 20:24:32 BST 1998 * Somehow fchdir had been missed from the main iBCS call map even though it had been put in the BSD map. Oops. -- Mike Wed Aug 19 23:12:47 BST 1998 * ALPHA 2.1 release number 2... -- Mike Wed Aug 19 20:43:09 BST 1998 * Avoid calling rt_sigaction if the mapped signal number is -1. It is not strictly necessary but why bother? -- Mike * Added David Bruce's updated x286emul. -- Mike Mon Aug 17 21:29:53 BST 1998 * Avoid calling fput with a NULL file pointer. -- Mike Sun Aug 16 17:32:20 BST 1998 * Fix to copy_{to,from}_user return checking. In many cases we do not care because they are behind verify_area checks but in some we do - and copy_{to,from}_user returns the number of bytes _uncopied_. -- Mike Sat Aug 15 23:39:42 BST 1998 * Change the flock handling to bounce the flock struct through kernel space rather than using an in place modification. The assumptions we were making about segment registers seem to have become invalid at some point. We could go to put_user(get_user()) but moving the whole struct is probably more robust. Which is faster? I dunno... -- Mike Sat Aug 15 22:30:41 BST 1998 * Changes to signal functions. Previous code confused old and new sigsets and could have lead to memory corruption and was likely the cause of signals getting blocked unexpectedly. Now we use the rt_sig* and new sigset interface to the kernel exclusively (in a few years the old stuff may be removed). This does means that a non-Linux program may now operate on signals >32 which may not have existed (and therefore would have errored) in the native environment. This may be considered a bug or a feature as appropriate... -- Mike Mon Jul 20 22:23:14 BST 1998 * ALPHA 2.1 release... -- Mike Mon Jul 20 20:55:47 BST 1998 * Added support for I_FDINSERT in the STREAMS/TLI message protocol handling. This allows TLI server code to work using a _real_ libnsl_s (tested briefly) but will break server code using my replacement libnsl_s until I update it. (It should probably use an I_FDINSERT simply so it can still run under SYSV if ever necessary.) -- Mike * Use TIOCINQ instead of FIONREAD in the STREAMS ioctl handler. FIONREAD isn't actually supported on sockets. The socksys code already had this right. -- Mike * Do the MOD_INC, MOD_DEC calls for sockets in the inherit and release handling only to reduce the opportunity for, ah, "oversights". -- Mike Thu Jul 16 22:11:48 BST 1998 * Finally got round to adding a GNU CopyLeft! -- Mike * Set up an XTI/TLI private state struct on all sockets at the same time we plug our own file operations in. This costs a small amount of extra memory per socket. However this is strictly correct (although unlikely to be actually _required_ by anything). It also allows us to set up TEPs created by XTI/TLI listen/accept easily so they have a chance of working. -- Mike * XTI/TLI fixes: Do not auto-listen in getmsg() - listen() after connect() appears to crash some 2.1 kernels(?). Change the probe for activity in getmsg() to use the new poll handler directly. Handle listening TEPs slightly better. -- Mike * Changed a stray KERN_ERR to KERN_DEBUG in ioctl.c -- Mike Fri Jul 3 23:20:37 BST 1998 * Fixed the initial set up of the argument and environment lists which were broken during the change to 2.1 style user memory access. -- Mike Mon Jun 29 22:50:11 BST 1998 * Put kernel_version definition back in emulate.c -- Mike * Change loaders to set start_text, end_text, start_data, end_data, start_brk, brk fully. -- Mike Thu Jun 25 21:18:11 BST 1998 * Added support for the "fat" IPC structures which use 32 bit uids, gids etc. -- Mike Wed Jun 24 21:30:09 BST 1998 * Change suser() checks to use the new capable() calls. -- Mike * Changed the CONFIG default to NOT support Xenix 286. It is an ever increasing build problem. It needs a volunteer to rewrite it. -- Mike * Add locking around system_utsname data using the uts_sem. -- Mike * Changed to use dentry stuff. Use fget()/fput() to get file pointers rather than accessing task fd tables directly. Use poll interface for SYSV poll and timod blocking checks. Use lock_kernel()/unlock_kernel() around stat routines and other namei stuff. Use dput() to release dentries obtained with namei(). Other incidental fixes to what went before as I noticed them. -- Mike Tue Jun 9 20:02:56 BST 1998 * More changes for new module conventions. -- Mike * Changed signal stuff to new conventions. -- Mike * Changed getname() calling convention. -- Mike * Changed fs type to mm_segment_t. -- Mike * Changed user VM accesses to use new conventions. -- Mike Sat Jun 6 20:30:00 BST 1998 * Changed to new module conventions. -- Mike Thu May 28 22:30:45 BST 1998 * Removed VM_STACK_FLAGS tests. -- Mike * Remove VM_GROWSDOWN tests. -- Mike * Removed MAP_EXECUTABLE tests. -- Mike * Removed MAP_DENYWRITE tests. -- Mike * Removed STACK_TOP tests. -- Mike * Removed FORKNOEXEC tests. -- Mike * Removed INIT_MM tests. -- Mike Mon May 18 22:49:56 BST 1998 * Fixed xnx_rdchk() return value. -- Mike * The serial number returned in the SCO utsname struct can now be set with sco_serial="..." as a module option. Some program apparently want to find a real, believable serial number there. This should be documented somewhere other than here I guess... -- Mike Mon May 18 22:03:49 BST 1998 * A class 't' ioctl with no arg type stuff in the top word is either an SVR4 tape ioctl a Coherent BSD-ish termios ioctl. -- Mike Mon Feb 2 22:57:45 GMT 1998 * Other relatively minor changes to allow compilation under glibc (specifically 2.0.6 but it shouldn't matter). -- Mike * Added an explicit -I flag to search /usr/src/linux/include since glibc doesn't necessarily require the kernel includes to be symlinked under /usr/include. -- Mike * The personality defaults in binfmt_coff should probably select short inodes on the assumption that newer stuff is likely to be built as ELF anyway. I hope. -- Mike Wed Nov 19 19:28:07 GMT/BST 1997 * Attempt to work around problems with sign extension when programs believe that the fs magic in a sysfs() call is short rather than long. -- Mike * Added IPPROTO_IP/SOL_IP option name mapping for {get,set}sockopt. This is correct for SCO OS5 and hopefully for other SYSV derived Unices. Note that SCO OS5 ping still doesn't work because it uses an unsigned char for the IP_TTL value whereas Linux wants an int - and it objects to getting an error return. Apparently POSIX 1003g says IP_TTL should be an unsigned char so OS5, or at least its ping, is at fault. This probably isn't worth hacking around unless someone _really_ needs it. -- Mike Wed Nov 12 22:28:19 GMT/BST 1997 * Added the SCO-ish syscall shortcut to the socksys networking code. -- Mike Fri Nov 7 20:13:05 GMT/BST 1997 * Oops, I installed the SCO signal tables in the old personality mapping but forgot to create a specific SCO personality using them. -- Mike Thu Nov 6 08:04:37 GMT/BST 1997 * No, it really should be USR1 for URG under SCO, not USR2. This is according to the SCO Network Programmer's documentation. The previous problem was because confusion between the SVR4 and SCO/SVR3 requirements prevented a SCO program from registering a handler for the right signal, I think. -- Mike Wed Nov 5 14:23:22 GMT/BST 1997 * Created a new signal mapping for SCO and mapped Linux SIGURG to SCO SIGUSR2. Also changed the plain ibcs mapping for SIGURG back to IBCS_SIGURG. Previously I had SIGURG mapped to SIGUSR1 for what my subconscious says was a good reason but I can't figure out where I got it from. The USR2 mapping is according to Peter Brueckner - and if it works for him... -- Mike Wed Nov 5 09:14:27 GMT/BST 1997 * Repair the STREAMS based socket code. The new socket hashing in 2.0.30 and beyond means the old code no longer works. We now delay replacing the fd with a socket until the first read, write or ioctl. Since the open has completed at this stage and the fd is fully initialized we can then use dup() to do the fd switch, which should always work. As a side effect this allows us to auto-connect an SPX pipe to X :0 if the first operation is other than a single character write - which should allow us to handle v. old SVR3 X programs side by side with their more modern, and common, multi-SPX pipe descendants. This does mean that some error returns from an open of a TLI special file might be delayed until the first operation, however a) these are unlikely to occur (they are things like no such protocol) and b) most opens of such files are hidden in functions like t_open() anyway so if there is a problem a simple fix to libnsl to check the first ioctl is all that is needed. -- Mike * sysfs() used to enumerate available filesystems seems to be 0 based on Linux but 1 based on SYSV. -- Mike Sun Oct 12 00:18:33 GMT/BST 1997 * Ioctls on the NFS pseudo device need to go via the emulation code rather than the native ioctl handler. -- Mike Sat Aug 16 14:56:24 GMT/BST 1997 * Changed the use of errno in a prototype in ibcs.h to avoid possible conflict with an SMP errno definition which could get inherited from an include file if we aren't careful. -- Mike Sat Jul 12 01:00:00 GMT/BST 1997 * Added Xenix locking() modes 5, 6, 7 as per patch from David Bruce (there were others too). I rewrote it to be a bit cleaner but I haven't tested it (it's late...). Some one will tell me I guess. -- Mike Fri Jul 11 22:27:13 GMT/BST 1997 * Added more console ioctl traps to discourage programs from trying to do funky stuff with the console without logging unsupported ioctl messages. -- Mike * sysfs() can, apparently, map an fs magic number to an fs name. We do this outside the kernel sysfs() because there seems no clean way to do it. We simply have a list of known magic numbers hard coded :-(. -- Mike * Implemented sysconfig(). This is appears to be the SVR4 forerunner to the POSIX sysconf(). -- Mike Tue May 13 20:52:05 GMT/BST 1997 * Hand off xmknod calls via ibcs_mknod so that we can create a directory if that is what the flags imply. I have not tested to see if xmknod is allowed to create directories - I just noticed it in passing. -- Mike * Added SCO's F_GETHFDO (get highest fd open) fcntl. This is used by /bin/csh on SCO OS5 to save a few syscalls. -- Mike Tue May 13 00:10:09 GMT/BST 1997 * More changes to handle long/short inode environments plus more explanation in the code. Are we having fun yet? -- Mike Sat May 10 15:19:39 GMT/BST 1997 * Added socketpair() to the socksys emulation. This is as implemented by SCO OpenServer 5. -- Mike * Change binfmt_coff.c to recognise programs compiled for SCO OpenServer 5, programs compiled for SCO 3.2.4 and programs compiled under SCO OpenServer 5 but with compatibility for ODT3.0. Why? Well, sometimes we should be giving long inodes in returns from getdents, sometimes short inodes. We don't just want to mask off the high part because that can lead to some things becoming invisible. We don't want that. -- Mike Fri May 9 23:29:37 GMT/BST 1997 * Added some more tape ioctls. -- Mike Wed Apr 16 23:12:37 GMT/BST 1997 * Fix memory leaks in poll.c and change ENOMEM return to EAGAIN. -- Mike Tue Mar 11 21:29:15 GMT/BST 1997 * Add {get,set}rlimit to the SYSV personalities. -- Mike Fri Mar 7 21:04:24 GMT/BST 1997 * Only build x286emul if EMU_X286 is enabled in CONFIG. There are many insallations which appear to either not have the a.out compiler support or have it installed incorrectly. -- Mike Tue Feb 18 22:04:39 GMT/BST 1997 * Removed unnecessary and incorrect includes. -- Mike Wed Feb 12 22:03:13 GMT/BST 1997 * Documentation updates for release. -- Mike Mon Feb 10 22:36:27 GMT/BST 1997 * Dammit! The inode folding in stat() and getdents() was out of step _again_ :-(. -- Mike Sun Jan 5 17:20:20 GMT/BST 1997 * sysconf(_SC_CLK_TCK) should return the value of HZ. -- Mike * Small clarification to README concerning shared libraries. -- Mike * Removed the bit about UnixWare X binaries not working from Doc/HINTS. Now we have STREAMS/TLI emulation this is no longer true (and hasn't been for a while). -- Mike Sat Jan 4 19:31:26 GMT/BST 1997 * If we open() something that is really a socket we close it and reopen it using socket(), connect(). This allows a Unix domain socket to look like a named pipe which enables more SYSV X clients to connect to a local X server using the local method (with suitable symlinks to map the pathnames). -- Mike Fri Jan 3 22:39:15 GMT/BST 1997 * Added a (simplistic) implementation of the I_CANPUT SVR4 STREAMS ioctl. -- Mike * Changed CONFIG.i386 to note that a correct setting of SMP is now necessary. -- Mike Wed Oct 2 16:28:39 GMT/BST 1996 * Intercept mknod() and allow it to create directories. Linux doesn't allow this but SYSV does. (What about BSD?) -- Mike Wed Oct 2 15:56:57 GMT/BST 1996 * Separated out the spx connect code so we can support old, single connection spx implementations once we know how to recognise them. -- Mike Wed Oct 2 15:54:45 GMT/BST 1996 * At some stage I add some binfmt_elf.c changes that were made to the kernel ELF loader in some 2.0.x patch. I forget which now but no one should be using the iBCS ELF loader currently anyway. -- Mike Fri Aug 23 15:42:04 GMT/BST 1996 * Moved svr4sig to the right place. -- Mike * Moved error message strings to a separate file so they can be shared by emulate.c and solaris.c. -- Mike * CONFIG files now define ARCH to be the CPU architecture. This is used to ensure that architecture specific files are only used when necessary. -- Mike * Changed the getdents routine in open.c to use the same rules for folding 32 bit inodes to 16 as stat and read. -- Mike Mon Aug 19 13:33:42 GMT/BST 1996 * Correct IPC problem introduced with Sparc changes. -- Mike Fri Aug 9 13:27:49 GMT/BST 1996 * Fix the inode swapping for /dev/spx and XTI/TLI transports. -- Mike * If a COFF binary doesn't have a .comment section we have no way to tell what personality we should be using. Switch to SCO anyway - odds on it is. -- Mike Wed Aug 7 14:22:11 GMT/BST 1996 * On SCO at least lseek on a character or block device returns 0 not -ESPIPE. -- C.A. Lademann * Some problems with /dev/spx and TLI end point handling that could leak descriptors, memory and leave the lists of file locks in an inconsistent state. Not to mention the fact that the socket data in an inode also contains a pointer back to the process' file structure. I _think_ I have this sorted out now... -- Mike * Sparc changes broke select. While I was at it I changed it to use the newselect Linux call as well. If we build without tracing we'll save a few more bytes of code now too. -- Mike Wed Jul 31 14:16:38 GMT/BST 1996 * Map EAGAIN to EWOULDBLOCK for recv*() and send*() syscalls. Linux seems to return EAGAIN. X/Open allows either EAGAIN or EWOULDBLOCK. SVR4 and Wyse V/386 specify EWOULDBLOCK in their man pages. SCO doesn't admit to non-blocking possibilities in their man pages but some code seems to expect only EWOULDBLOCK. -- Mike Mon Jul 29 16:58:11 GMT/BST 1996 * Added a CONFIG option for SMP. This is enabled by default. I don't think this is a problem for non-SMP systems? -- Mike * Minor change to the SCO error map. -- Mike Fri Jul 26 09:13:43 GMT/BST 1996 * Updated README and RELEASE -- Mike * Merged Sparc patches from: Miguel de Icaza -- Mike Thu Jul 4 12:24:06 GMT/BST 1996 * The default is now *not* to build a versioned module. I got too many questions about why it wouldn't compile. -- Mike * Fix to binfmt_xout.c when Xenix 286 is emulation is not configured. -- Mike Fri Jun 14 13:36:18 GMT/BST 1996 * Added code to explicitly fail the SCO LD?MAP ioctls used for channel mapping. -- Mike Thu Jun 6 17:21:00 GMT/BST 1996 * Cleaned up Stephans SCO tape ioctls and added the corresponding SVR4 versions (untested). -- Mike Wed Jun 5 10:47:24 GMT/BST 1996 * Rewritten the Xenix locking() syscall. The previous one was crap. -- Mike * Change the read on directory emulation to fold long inodes to shorts in the same way as stat and getdents. This may help old programs that use stat and read to do a getcwd rather than a stat and getdents. -- Mike * Use __get_free_page instead of get_free_page in getdents since we don't need the buffer cleared initially (I think). -- Mike Sat Jun 1 09:50:30 MET DST 1996 * Added some tape ioctrls for SCO to iBCSemul/ioctl.c. -- Stephan Fri May 31 08:44:51 GMT/BST 1996 * Move bsdioctl.c to the main group of source files from than the BSD specifics. Just about everything else has BSD style sgtty stuff and some seemingly modern code actually depends on it (e.g. ISC SVR4 telnet)! -- Mike * Add CONFIG option to build a versioned module. -- Mike * Install the modules in /lib/modules/`uname -r`/misc for compatibility with the modules tools. -- Mike * If the requested connection indications in a bind request is greater than zero do a listen() if the bind is successful. We still also do a listen() if the program selects() on a bound but unconnected stream. This may help some broken programs. It may also break some broken programs. It's debatable whether this should now be in or out. -- Mike * The bit vector mapping used for converting vectors of signal flags had an off by one error. The signal maps themselves were also one entry short. -- Mike * At some stage I changed the I_SETOWN STREAMS ioctl but never committed it? -- Mike Thu May 9 12:51:10 GMT/BST 1996 * Change to install in /lib/modules/`uname -r`/misc instead of /usr/lib/modules for compatibility with the way the modules tools have gone. -- Mike Thu Apr 25 12:34:06 GMT/BST 1996 * Use the CONIND_number in a T_BIND_REQ to decide whether or not we should listen() on the socket. -- Mike Mon Apr 22 15:42:47 GMT/BST 1996 * Added a simple loader that will attempt to pass scripts that start with ":" and things that look like simple text to /bin/sh. Hopefully this isn't going to conflict with magic for other needed formats... -- Mike Fri Mar 29 17:11:35 GMT/BST 1996 * Committing the all new TLI/XTI options handling. Note that TLI and XTI option handling is mutually exclusive at the moment. The default is to enable TLI options handling since that is what existing SVR3/4 systems use. I haven't found one that actually handles X/Open format (they use the TNOTSUPPORT cop out) so I don't know how the stack is told to use XTI format instead of TLI. Note that only SOL_SOCKET/* and SOL_TCP/TCP_NDELAY are known to work to any extent at the moment. Others may (or may not need mapping) but I can't find the motivation to wade through the headers and create yet another table of magic numbers. Hopefully everyone just nicked the BSD reference code... -- Mike * Some more 1.3.7x changes related to process accounting (fork but no exec etc.). -- Mike Wed Mar 20 13:36:07 GMT/BST 1996 * I seem to have forgotten to add a comment about merging changes from mid 1.3.7x kernels. -- Mike Wed Feb 28 14:53:00 GMT/BST 1996 * Fix to shared library loader in binfmt_aout.c from kernel patch 1.3.69. -- Mike Wed Jan 24 09:58:34 GMT/BST 1996 * Implemented I_PEEK. -- Mike * Changed I_NREAD to understand that control messages constitute queued data. -- Mike Fri Jan 19 11:57:20 GMT/BST 1996 * Make the socket level connect in timod.c happen synchronously always. Allowing the async possibility is a little harder. -- Mike Thu Jan 18 16:06:13 GMT/BST 1996 * Added (some of) the intial framework for option management. This is documented by X/Open but many XTI stacks don't seem to implement it. Certainly the man page for t_optmgmt in SCO OS5 says it isn't yet implemented. There do seem to be programs out there that make options requests though. Unfortunately I don't have one right now so this is currently disabled and MAY CAUSE KERNEL FAULTS. I don't know... -- Mike Tue Jan 16 12:35:21 GMT/BST 1996 * Added a slight change to mmap.c. The SVR4 headers describe a slight change in functionality which is specified with a special flag that is supposedly set by the C library interface. I haven't actually seen this flag set by the SVR4 dynamic loader (which is the only mmapper I've seen) and things seem to work with or without this change. It may not be necessary at all? -- Mike Fri Jan 12 14:56:38 GMT/BST 1996 * Remove trace flag from context() in the callmaps. SVR4's /bin/sh seems to use it but doesn't care if it fails. -- Mike * Added the SCO OS5 SI86GETFEATURES request to sysi86(). OS5 uses this to establish what "features" are available when a program starts (i.e. in crt*.o). Currently we just return the same vector as OS5 itself. I haven't found what the flags mean - but one of them indicates xstat extensions which I have added to iBCS. -- Mike * Change .comment parsers in binfmt_coff.c and binfmt_elf.c so we only grab a single page (asking for two contiguous pages is antisocial) and then roll through the comment section in stages. -- Mike * Fixes to binfmt_elf.c and binfmt_aout.c so they compile and load with 1.3 kernels. -- Mike Tue Jan 9 14:29:49 GMT/BST 1996 * Add a kludge to the TLI connect code. For some reason, under ISC SVR4, telnet seems to gve the port number in network byte order ok but the X programs give it in host byte order. One of them is wrong but presumably both work on ISC (I don't have the hardware to actually install it). We kludge around it here by checking if we are SVR4, AF_INET, port 6000 (host order) and, if so, change it to port 6000 (network order). This will break anything that wants to make an outgoing call to port 28695. There are probably other things that expect incorrect port numbers to work but there seems to be no easy solution here. -- Mike Fri Jan 5 13:34:43 GMT/BST 1996 * First draft of code to handle select/poll on transport end points correctly. This isn't well tested but should be good enough to support normal timod usage. -- Mike Thu Jan 4 13:52:25 GMT/BST 1996 * TLI changes to get error returns from connects correctly. -- Mike Wed Jan 3 17:06:07 GMT/BST 1996 * Added a manual page in the Doc directory. This is not installed by default. -- Mike * Fixed a problem in the .comment parsers in binfmt_elf.c and binfmt_coff.c where a number of pages was passed to __get_free_pages() instead of a binary power. This has been in release versions of binfmt_coff.c for a long time but exactly one person has reported anything that might be attributable to it. Strange, since the bug should have manifested itself as insidious memory leakage and corruption... -- Mike Wed Jan 3 12:16:47 GMT/BST 1996 * Removed all kernel patches in the Patches directory. None should be need for kernels 1.3.50 (or earlier?) and onwards. None should be essential for 1.2.13 - although some capability is lost notably SCO OpenServer 5 ELF binaries (if the kernel ELF loader is present) and BSD a.out binaries. -- Mike * Another fix to termio[s] ioctls to get the control character settings right for SVR4. Previously this was just copied from the SCO mapping and just never got changed. -- Mike Thu Dec 14 10:41:36 GMT 1995 * Added basic getpmsg/putpmsg support. It doesn't do anything with bands but we don't have anyway to use them. -- Mike Tue Dec 12 09:38:01 GMT 1995 * First commit of the major TLI rewrite for Eric Sixt. This should be enough for normal TCP & UDP clients. It may be sufficient for servers as well but this is untested so far. The client stuff has been tested with SCO cu, ISC telnet and custom test code down to the getmsg/putmsg level. A shared libnsl_s is also included. This has the functions needed for clients but lacks some needed for servers currently. It has been tested on a real SCO system. It could be done better (and should be) but for now we are interested in making existing programs work rather than handling all the error conditions by the book. The library uses SVR3 shared library tools and needs to be built on an SVR3 system (I used SCO). Included is a rewrite of the /dev/spx kludging (the old didn't work if TLI was enabled). This now allows connections to other than /tmp/.X11-unix/X0 (if configured) so you can run multiple X servers on the VCs and display SCO X programs on all of them. The major/minor numbers used for the emulation devices have (necessarily) changed. The top level Makefile will (re)create them on a "make install" or "make devices". Oh yeah, I built stubs for other libraries as well (specifically SCO/SecureWare protlib_s) but don't expect it to have any real support. If programs _really_ need stuff from there they are like to have reduced or completely broken functionality on Linux! The script I used to generate the skeleton APIs for the shared libraries is libs/mkltab if anyone feels the urge to look at other libraries... -- Mike * Removed the old quota compile option. It isn't compatible with the quota stuff in 1.3.46 and would only cause confusion. This means you can't use iBCS if you have the old quota patches in a pre-1.3.46 kernel. Tough :-). -- Mike Mon Dec 11 15:14:46 GMT 1995 * Map I_SETSIG/I_GETSIG to FIOSETOWN/FIOGETOWN and pray... -- Mike * Fixed possible memory leak in COFF & ELF personality recognition when the offset/size of the .comments section is invalid. This could leak when the .comments section exists but is zero bytes long I think. -- Mike Wed Dec 6 11:31:27 GMT 1995 * A stat of a file must give an inode that matches what we get from a read of the directory since code that figures out cwd needs the inodes to match. Mind you, if there are several inode numbers greater than 65534 in the directory we are going to get some strange paths. I don't know if this is fixable properly at all? -- Mike * Yes it is. We just mask off the high word to get the short inode number for the stat structure. -- Mike Thu Nov 30 16:21:32 GMT 1995 * Fix the SVR4 additions to the V7 terminal ioctls. Add handling of TIOC[GS]ETD. Have TIOCL[GS]ET trivially succeed (I can't find the documentation for the arguments). -- Mike Wed Nov 29 12:57:42 GMT 1995 * Changed the guesswork svr4_waitsys to implement the correct waitid syscall. -- Mike * Enable the i486-linuxaout option in x286emul/Makefile by default. There seem to be significant numbers of people using ELF compilers now. This may be a mistake... -- Mike * Fixes to sigsuspend and sigpause in signal.c. The previous invocations of the Linux sigsuspend syscall were wrong and liable to lead to random freeze ups of programs which used sigsuspend() or sigpause(). -- Mike Fri Nov 24 11:03:01 GMT 1995 * Interactive SVR4 seems to be overloading the BSD termio ioctls with its own. -- Mike * The SVR4 procid subcodes don't match the SCO ones. Or, at least, I seem to remember I got the orignal set from SCO and UnixWare and Interactive SVR4 are different (but match each other). -- Mike Thu Nov 23 17:21:56 GMT 1995 * Interactive SVR4's /bin/sh calls access(..., 011) but Linux returns EINVAL if the access mode has any other bits than 007 set. So we have to wrap it and mask it :-(. -- Mike Wed Nov 22 10:11:49 GMT 1995 * Change to binfmt_elf.c to set total_vm value introduced in kernel 1.3.43. -- Mike Thu Nov 16 15:02:58 GMT 1995 * Added support for SCO OpenServer 5 binaries using .comment section parsing as with COFF. Built a BSD capable a.out loader as part of iBCS. The iBCS ELF loader is only used for recognised (using .comments) binaries if the standard ELF loader is configured in the kernel iBCS is built against. If the ELF loader is not configured in the kernel iBCS is built against the iBCS ELF loader is used for all ELF binaries. This prevents the iBCS module from becoming un-unloadable on ELF based Linux systems. Similarly the iBCS a.out loader tries to avoid dealing with anything other than BSD binaries for the same reasons. This requires a kernel 1.3.42 or greater to allow iBCS to insert its loaders into the list before the standard kernel ones. Kernels 1.3.39(~) to 1.3.41 have some support but there are bugs which will likely prevent you running *anything* as soon as iBCS is loaded. Mea culpa. Tough. -- Mike Thu Nov 16 11:00:08 GMT 1995 * Minor clean up in binfmt_coff.c and fix a minor bug in parse_comments that caused it to miss checking a string if there was a zero length string in among. -- Mike Fri Nov 10 12:22:05 GMT 1995 * Changes for latest module bogosities (~1.3.38). We need to define __NO_VERSION__ to avoid linux/module.h putting a kernel_version in *every* damn file. -- Mike Tue Nov 7 10:55:05 GMT 1995 * When stealing temp space from the stack we have to actually move esp down and then restore it as the 1.3.x kernels add a check to trap out of bounds stack accesses. This is not tested but I think it only affects the TLI emulation code which is disabled by default. -- Mike Mon Oct 9 11:22:29 BST 1995 * Use kernel readv/writev if available. -- Mike Thu Sep 14 12:21:48 BST 1995 * Changed references to current->sigaction to allow for the changes in kernel patch 1.3.26 which allow sharing of signal state between clone processes. -- Mike Mon Sep 4 10:04:22 BST 1995 Originally: Wed Aug 2 09:57:56 GMT 1995 * Removed Xenix interrupt bug, created wait() workaround for INFORMIX-SQL and built the locking() to fcntl() mapping. -- Peter Wed Aug 30 09:19:54 BST 1995 * Merged kernel patch for binfmt_elf.c from 1.3.21 which sets EDX to zero on process start up. See the comment in the code for reasons. -- Mike Tue Aug 29 08:44:50 BST 1995 * Workaround. Local X won't work because a putmsg occurs after we have already swapped the open descriptor for a Unix domain socket. We either need to peek at socket internals or actually implement the messages for passing file descriptors. This quick fix enables local X connections if we build without XTI support. -- Mike Tue Aug 8 11:36:41 BST 1995 * Fix streams kludging so local X works again (should do anyway - it's untested as yet due to hard disk collapse. Hopefully someone will tell me if it doesn't work.) -- Mike Mon Jul 31 13:25:58 BST 1995 * Changes to allow compilation in the presence of a kernel built with the quota patches. -- Dimitrios Mpougoulias Tue Jul 18 09:01:53 BST 1995 * Fixed timod getinfo ioctl. This functionality should move into the message handling as bind has. -- Mike Mon Jul 17 10:16:43 BST 1995 * Added handling of BSD-ish sgtty ioctls. For some reason the SVR4 network programs such as telnet have been modified to pushd the ttold STREAMS modules and use sgtty calls instead of simply using termio[s]. Don't blame me! :-) -- Mike * Restructuring of timod.c complete. Timod ioctls now use message routines to avoid duplicating code. -- Mike Wed Jul 12 14:44:30 BST 1995 * Made debug output in binfmt_coff.c less verbose. We only really care about matched comment strings. If we dump them all here we overrun the kernel message buffer and miss interesting stuff when the program actually starts running. -- Mike * Changed __get_free_pages in binfmt_coff.c to give the extra argument required in 1.3 kernels. -- Mike Tue Jul 4 11:48:27 BST 1995 * Restructure timod.c ready to rewrite timod_ioctl to use putmsg/getmsg rather than reimplementing the same message protocol again. -- Mike Mon Jul 3 13:41:49 BST 1995 * Initial TLI support for outgoing TCP and UDP. Merged BSD/Wyse socket ioctl handling with the socksys handler. Fixed (some) bugs in poll() emulation. Devices in /dev/inet have changed to support TLI access. See README for details. -- Mike Tue Jun 27 09:00:02 BST 1995 * Don't export symbols from iBCS. I'm not sure when register_symtab was introduced so we only drop our symbol table in 1.3 and later. -- Mike * Added missing brackets in binfmt_elf.c as per kernel patch 1.3.4. -- Mike Thu Jun 22 13:09:49 BST 1995 * Big step forward with TLI. I now seem to understand what is happening when and have enough to do the initial open and get info requests. This may one day actually work... -- Mike * Trap EUC ioctls and return EINVAL. We don't do EUC :-). -- Mike * Changes for the 1.3 development kernel. This compiles but is as yet untested. It still seems ok with 1.2.10. -- Mike Wed Jun 14 09:15:39 BST 1995 * Added Eric's protection mapping fixes to binfmt_elf.c from 1.2.10. -- Mike Fri Jun 9 12:31:53 BST 1995 * Linux can't stat unmounted filesystems but SCO can (by specifying the pathname/fd of a block device and the filesystem type. Linux will just stat the filesystem that the device node is on (i.e. the root filesystem). There isn't much we can do about it. I just lie and claim there is 100MB free of 1GB. I hope this won't cause too much grief... -- Mike Thu May 18 12:06:50 BST 1995 * When trying to get an IP domain name don't try the utsname domainname if it is "(none)". If we get that far we have to give in and return a blank domain name. -- Mike Wed May 17 10:15:42 BST 1995 * Recheck the socket functions when accept returns a new socket. This is pedantic at this stage since we must have had a socket initially and the current kernel code isn't likely to handle removal and reloading of in use modules. iBCS can handle this happening to it though :-). -- Mike * Fix timod faking to correctly return an error if given an invalid file descriptor rather than trying to dereference a null pointer in kernel mode. -- Mike Tue Apr 25 11:35:43 BST 1995 * If nap() is given a zero timeout return immediately rather than blocking indefinitely. Ensure that SIGALRM is not ignored during the pause or we never get woken up by the timeout. -- Mike Mon Apr 24 09:21:30 BST 1995 * Various documentation updates from Eric. -- Mike Fri Apr 21 14:34:25 BST 1995 * Fixes to IPC. Now it really *should* work... -- Mike Thu Apr 13 14:03:45 BST 1995 * Tidy up documentation ready for new release. -- Mike Wed Apr 12 11:07:52 BST 1995 * Moved to an ELF development system with a GCC 2.6.4 snapshot. This pointed out a few signed/unsigned mismatches so I fixed them. We also have to ensure that x286emul is built as a QMAGIC a.out because it won't work any other way. It isn't likely to work any other way unless someone pays for for it to be done - it isn't what you might call a "sexy" project... -- Mike Wed Apr 12 08:53:22 BST 1995 * Added the kernel patch to handle statically linked SVR4 ELF binaries to the Patches directory since it doesn't seem to have made any of the 1.2.x patches yet. -- Mike Tue Mar 28 09:55:38 BST 1995 * Made the ISC specific stuff optional via CONFIG. -- Mike * 1. INTERACTIVE UNIX signal numbers are different from IBCS2. I added new signalmaps and exec_domains. 2. setpgrp does not deattach the controlling terminal. Try the setpgrp test program included here after the patch on real SCO or xxxx and Linux/IBCS2. 3. sigset behavior is incorrect. Test program also included. Short description: signal handlers registered whith sigset should run with the signal blocked, and after return the handler must be restored, so sa_flags = 0 is the correct setting. Calling sigset should remove the signal from the blocked set. -- Remete Gabor Fri Mar 24 10:20:57 GMT 1995 * Set the fragment size to zero for [f]statfs() just to be pedantic (SCO does this). -- Mike Tue Mar 21 10:24:14 GMT 1995 * Fixes to 286 overlay and x.out loader by Don Camp . -- Mike * New code for SVR4 {get,set}groups since SVR4 uses longs for gids whereas Linux uses shorts. -- Mike Mon Mar 20 17:06:23 GMT 1995 * Added code for waitsys(). This is guesswork at this stage but appears sufficient for ISC 4.0 ksh to do something reasonable in the simplest case. -- Mike Tue Mar 14 09:44:13 GMT 1995 * Initial implementation of ptrace. This allows programs (SCO and Wyse V/386 programs) to access a process' registers but actual tracing semantics will be difficult to make functional. -- Mike * Alter emulate return to allow return of numbers > 2^^31 which don't fall in the error code space. -- Mike * Added signal mapping to wait() syscalls. -- Mike * Updated the main README with the vger mailing list details and a few other tidy ups. -- Mike Thu Mar 9 10:10:53 GMT 1995 * Added a load more system calls to the SVR4 tables (these match UnixWare at least I think). Some of these may work, some are just marked to shut up run time warnings. -- Mike * Increased the size of the buffer used for parsing comment sections in binfmt_coff.c to 8k (from 1k). There are programs out there that need it. -- Mike Tue Mar 7 16:12:36 GMT 1995 * More XLI/TLI changes - mostly fixes. -- Mike * Added basic handling of the termiox extension used by SVR4 to give access to hardware flow control. Only RTS & CTS or nothing is allowed because that's all Linux does. The same comments as below apply to the status of this. -- Mike * Rework ioctl handling for termios. SCO uses short in a termios structure, SVR4 uses longs. The bit shuffling used for SVR4 is the same as for SCO - there are known to be some slight differences but these are not believed to be critical for most things. A large amount of guesswork went in to this. One day I may be able to test it... -- Mike Mon Mar 6 12:31:05 GMT 1995 * Added a -DMODULE to iBCSemul/Makefile to make new style kernel modules work. -- Mike Fri Mar 3 15:04:14 GMT 1995 * Patches from Eric. (Use Eric's [f]statvfs instead of mine). -- Mike Thu Mar 2 11:46:03 GMT 1995 * Fixed a silly in the handling of SO_IMASOCKET and SO_PROTOTYPE plus added a check on the given buffer length. -- Mike * Fixed a silly in the sockopt.inc options mapping. -- Mike Wed Mar 1 14:20:06 GMT 1995 * Added SVR4 console keyboard mapping and termcap/terminfo options to the PROD.Patches directory. -- Mike Mon Feb 27 13:30:17 GMT 1995 * Added simpleminded implementation of [f]statvfs(). -- Mike * [f]statfs() now respects the length argument to allow for older/smartass code which uses a smaller buffer size. -- Mike Fri Feb 17 10:33:23 GMT 1995 * More tidy up, a little more work on TLI/XTI (still nowhere near usable), fixed the signal debugging messages so it makes more sense. -- Mike Thu Feb 2 12:45:25 GMT 1995 * Changed the BSD termio ioctl handler to recognise the fact that some provide V7 compatibility and may be invoked from non-BSD binaries - possibly without BSD size encoding. -- Mike * Changes for 1.1.88. More definitions moved around in the Linux header files :-). -- Mike Fri Jan 27 10:12:51 GMT 1995 * Also mapped new Linux SIGIO to iBCS SIGIO and changed Linux SIGURG to map to SIGUSR1 as documented in SCO's TCP/IP programmer's guide. Is SVR4 the same? How badly do we care? -- Mike * Had to add a new personality for Xenix binaries since they have a different number for SIGPOLL - sigh... -- Mike Mon Jan 23 15:34:01 GMT 1995 * Changes to includes for 1.1.84. -- Mike * Change to binfmt_elf.c from kernel patch 1.1.84. -- Mike Tue Jan 17 17:10:25 GMT 1995 * Added tracing of argument and environment vectors passed to exec(). -- Mike Mon Jan 16 11:34:58 GMT 1995 * Change socksys.c for 1.1.81. There doesn't seem to be a convenient define to test but I intended to roll out most of the "magic" define testing for 1.2 to clean things up. -- Mike * Ensure the segment registers are correctly set up when we exec a new program - the LDT we may have been using before no longer exists. -- Mike Fri Jan 6 11:32:52 GMT 1995 * Fixed a problem where the Xenix 286 emultor wasn't getting the correct return and error codes from lcall7s. -- Mike * Added support for exec() in the Xenix 286 emulator. -- Mike * Made the spin-before-launch trap in binfmt_xout.c settable via a trace option, "xout-block". This is used to attach gdb to a 286 process to debug the emulator. -- Mike * Fixed a problem with binfmt_xout.c setting the wrong intial brk value for impure segmented binaries. The checks for brk collisions with mapped space were failing brk changes. Before the brk checks were introduced I image we simply trashed text or data... -- Mike Thu Jan 5 11:21:51 GMT 1995 * Added some debug to the STREAMS I_PUSH and I_POP. Also made them succeed without actually doing anything. We may be able to work round some instances of STREAMS usage one day... -- Mike Wed Jan 4 11:17:14 GMT 1995 * Change yesterday's mkdir to use getname()/putname() to fetch the pathname from user space. I hadn't realised these were already in ksyms.c (although how long have they been there? This breaks compatibility with older versions of the kernel I guess). -- Mike * Implement the NIOC ioctls for getting file handles since Oracle's TCP server seems to be wanting to do this for some reason (why???). This is a simple implementation that simply fills in the device number of the filesystem and the inode number of the file. This seems to agree with what SCO is doing. I don't know what the "exported" fields are or should be. -- Mike Tue Jan 3 14:31:13 GMT 1995 * POSIX says that a pathname ending with a '/' means the current directory whereas SYSV drops the trailing slash. This is only a problem with mkdir() I think. -- Mike Fri Dec 16 16:25:44 GMT 1994 * Added basic support for Wyse V/386 MPX syslocal(). -- Mike Fri Dec 9 09:14:04 GMT 1994 * Changed eaccess() again. We should always set the uid/gid rather than fsuid/fsgid since the Linux access() call is getting fsuid/fsgid from the current uid/gid. -- Mike * Don't bother trying to read x.out segments which have no data in the executable. -- Mike Thu Dec 8 11:51:06 GMT 1994 * Only include if hasn't defined KERNEL_DS. The header in the linux subdirectory was obsoleted in 1.1.69. -- Mike Fri Dec 2 13:50:03 GMT 1994 * Force QMAGIC format when building x286emul. -- Mike * Patches for 1.1.69. -- Mike Thu Dec 1 13:50:37 GMT 1994 * Binfmt_xout.c now checks the trace flag to see if debug messages should be produced. Trace now has an "xout" option. Trace is now a Linux binary which uses a personality() syscall to enable emulation so the iBCS trace syscall becomes accessible. -- Mike * Changed binfmt_xout.c to use 0x1020 as the entry point for the x286emul kludge overlay. This is the expected value for a QMAGIC binary which is the default with the newest compiler. Really I think uselib() should return something useful but I haven't decided what. -- Mike * Made the schedule() trap (so gdb can be attached before the process starts running) in binfmt_xout.c a CONFIG option controlled by the XOUT_TRACE setting. -- Mike * Moved the configuration for the optional items out of Makefile and into CONFIG. -- Mike Wed Nov 30 17:08:05 GMT 1994 * Fixed a nasty bug in binfmt_coff.c where an extra page was being allocated in the bss causing brk() checks in 1.1.64 and later to fail _in_certain_circumstances_. -- Mike Wed Nov 30 13:58:46 GMT 1994 * Added support for unpacking ioctls passed via the STREAMS interface and forwarding them to the file descriptor. With a slight change to socksys so that devices with minor 2 get an anonymous dgram socket immediately they are opened this allows many more network programs to work. Thanks to an old piece of code from Chip Rosenthal for revealing how it all worked - it isn't in the documentation :-). -- Mike Wed Nov 30 11:27:56 GMT 1994 * Fixed handling of brk and BSD sbrk so error returns are handled correctly. Mon Nov 28 10:48:25 GMT 1994 * When doing an IPC_STAT on a semaphore force the semaphore number to be zero as the kernel code erroneously checks it. -- Mike Fri Nov 25 14:26:41 GMT 1994 * Massive rework of ipc.c. The previous one was buggy as hell. This one works with all the demo programs from the Wyse V/386 IPC documentation. Hopefully someone is going to test it with some *real* applications! -- Mike Tue Nov 22 09:11:46 GMT 1994 * Change the TTYDEVTYPE ioctl to say we on a pseudo terminal rather than a console. Anything calling this is likely to want to try fancy stuff like mapping the console memory and playing with I/O ports if it thinks we are on a console. -- Mike Tue Nov 22 09:07:04 GMT 1994 * Allow direct usage of 57600 and 115200 serial speeds from the BSD domain with kernels 1.1.65 and later. The mask-and-merge for iBCS<->Linux termios should be checked as we tend to try and preserve unmappable bits where there is no other conflict. In 99% of cases we won't see a problem though... -- Mike Mon Nov 21 10:05:19 GMT 1994 * Slight change to previous patch. Confusion over which struct size we should be using for the verify_area plus we need to do a VERIFY_WRITE as we will rewrite the data before returning. -- Mike * Changes to ipc.c for reading structures from user space. -- Roberto Bagnara Thu Nov 17 15:24:23 GMT 1994 * Some of the unused KD ioctls were removed from Linux 1.1.64 (or was it 63?). Changed vtkd.c accordingly. -- Mike Fri Nov 11 14:15:09 GMT 1994 * Moved the x286emul overlay to /usr/lib instead of /lib. There is no real need for this to be on the root filesystem. -- Mike Mon Nov 7 13:51:55 GMT 1994 * Added a version of the BSD exec.c patch for Linux 1.1.62. -- Mike * Extended SCO keyboard mapping for new kbd. -- Mike Tue Nov 1 10:57:18 GMT 1994 * Changed the personality test in the BSD kernel patch. Apparently FreeBSD uses a machtype of 134 so it's safer to test for Linux rather than BSD. -- Mike Fri Oct 28 11:25:43 GMT 1994 * General clean up of the configuration options in the Makefile and elsewhere. It is now possible to choose which loaders are included and whether or not BSD or Xenix 286 emulation is required. -- Mike * Added the x286emul overlay library. This is still very alpha but SCO's /bin/masm seems to do the things expected of it :-). -- Mike Wed Oct 5 17:00:13 BST 1994 * Fleshed out the F_CHSIZE and F_RDCHK fcntls. I still don't know if these are ever used from the user layer. -- Mike Tue Oct 4 13:17:32 BST 1994 * Fixed the BSD termios mappings. This _looks_ ok from the point of view of the NetBSD stty. -- Mike Fri Sep 23 11:08:31 BST 1994 * Add SCO ODT version of CorelDraw to COMPAT file. -- Doug Ledford Thu Sep 22 09:25:04 BST 1994 * Added the phone numbers of McGraw Hill's order desk to the README file. -- Mike * chsize() was causing a segment error. The callmap entry punted to a Linux system call but the argument count was not negative resulting in a call to a bad function pointer. -- Mike * Linux doesn't have the l_sysid field in struct flock so we need to slide the l_pid field down a word after doing a locking fcntl. -- Mike Tue Sep 20 10:31:01 BST 1994 * Added a simplistic implementation of writev since BSD seems so keen to use it as much as possible. -- Mike * Fixed the x.out loader (mostly). This requires a minor patch to the kernel. Expect this to be put in to the kernel itself sometime after 1.1.51. Segment support is still buggy and therefore disabled. If programs assume they know what segments exist it they are broken anyway! -- Mike Wed Sep 14 11:24:18 BST 1994 * Added extra fcntl code 14 (used by SVR4 for GETLCK). -- Mike Tue Sep 6 10:58:49 BST 1994 * Added setting of fsuid/fsgid to the loaders and changed the eaccess() function in xnx.c to flip fsuid/fsgid rather than uid/gid. If you were having problems with setuid/setgid iBCS binaries this is likely to help. -- Mike Mon Sep 5 15:07:06 BST 1994 * Fixed the bad initialisation of howcnv in signal.c. -- Mike Fri Sep 2 11:01:26 BST 1994 * Added a little program to the Tools subdirectory which will restamp old binaries with the Linux machine type rather than zero which is used by BSD. Once this is done you can reliably distinguish between BSD and Linux binaries (the current Linux linker will set the correct machine type on all new binaries). -- Mike * Updated the BSD patch. -- Mike * Changed binfmt_coff to avoid passing meaningless MAP_DENYWRITE and MAP_EXECUTABLE flags to mmap when we are not actually mapping from a file. It would probably never be a problem... -- Mike Tue Aug 23 17:21:45 BST 1994 * Added my device trace stub in case anyone needs to investigate and emulate some special device. -- Mike Wed Aug 17 14:06:34 BST 1994 * Added an extern definition of the formats list needed by the 1.0.x hooks. Presumably this went AWOL at some stage? -- Mike Fri Aug 12 09:52:38 BST 1994 * Slight change to the socksys major allocation. Removed redundant error message. -- Mike Wed Aug 10 08:57:32 BST 1994 * Added the spelling corrections from the 1.1.42 kernel patch. -- Mike Fri Aug 5 10:05:14 BST 1994 * Added Scott Michel's SCO multiscreen patches. -- Mike * More changes to loaders for latest kernel changes. -- Mike Wed Jul 27 10:59:14 BST 1994 * Changes for 1.1.36 vm/mprotect. -- Mike Tue Jul 26 14:20:27 BST 1994 * Tidied up the system call maps a little. ISC is now handled by the same maps as SCO. -- Mike Wed Jul 20 12:39:55 BST 1994 * Removed the lock.patch. This is in the latest kernels. -- Mike * Changed the socksys/SCO utsname stuff to do the right thing if we have the hostname set to the fqdn and the domainname set to the NIS domain. If hostname is not an fqdn we assume we are using the domainname == IP domain convention. -- Mike Wed Jun 29 13:34:34 BST 1994 * Revised the lock.patch to fix the unlock problem and allow locks on devices. This should go to Linus (and probably has). -- Mike * Removed the kernel 1.1.20 patch. This isn't needed as of 1.1.22. -- Mike Mon Jun 27 09:26:24 BST 1994 * Can't use select() for nap() since the Linux select system call expects to pull its arguments from the user stack. Rewritten nap() using itimers. -- Mike * More fixes from Eric for 1.0 compatibility. -- Mike Fri Jun 24 09:37:50 BST 1994 * Added a bit about time zones to the HINTS file. -- Mike * First draft of BSD termios mapping. Still buggy. Needs work to be useful. -- Mike * Fixed BSD getdirentries() to align dirents on long word boundaries. 'ls' now works and 'tcsh' finds and executes commands successfully. -- Mike * ibcs_exit should be listed in callmap.inc with arg count 1 not -1. -- Mike Thu Jun 23 09:48:43 BST 1994 * Lots more BSD-ish stuff. Commit now for release since there is confusion over versions at the moment. -- Mike * List Xess MOTIF spreadsheet in COMPAT file. -- Michael K.Johnson * When slurping in mis-aligned COFF binaries we shouldn't be aligning the file offset and virtual address to a page boundary! -- Mike * Merged Eric's latest patches for 1.0 compatibility and Linux ELF capability. This was done by hand. I think it's perfect... -- Mike Wed Jun 22 14:28:38 BST 1994 * Many sigsuspend entries in the callmap had argumetn counts of 1 instead of Spl. This caused kernel faults. -- Mike * Implemented the ioctls for the pseudo nfs device which under SCO is used to get/set the NIS domain. You need to link /dev/nfsd to /dev/socksys. -- Mike * The socksys getdomainname() was doing the same write null to kernel space instead of user space as the wysev386 version. -- Mike Tue Jun 21 08:54:34 BST 1994 * Use the map_bitvec() function (added as part of the BSD changes) in signal.c. Note that there were quite a few mapping bugs in the old code. I think I got them all but few would have been triggered by my SCO and Wyse stuff. -- Mike * Snapshot BSD stuff again ready for release. -- Mike * binfmt_coff.c and binfmt_xout.c should be using mmap(NULL,...) instead of zeromap_page_range as this will create the vm area descriptor as well as mapping the pages. This is needed for the verify_area() in 1.1.20 which has tighter checks than previous versions. -- Mike * Map stack region as a GROWSDOWN vm area in binfmt_elf.c -- Mike Fri Jun 17 16:42:59 BST 1994 * Major and on going work to add BSD support (optional - see the Makefile). I'm committing this now so I have a current tree ready for release when Linus goes to 1.1.20. -- Mike * Wyse get{host,domain}name were using direct assignment to add a trailing null instead of put_fs_byte(). Oops... -- Mike * Changes for execution domain support which will be in the main kernel as of 1.1.20. This allows easy registration of lcall7 handlers, binary loaders etc. for different personalities. -- Mike Fri Jun 10 10:12:55 BST 1994 * Added patch file for kernel locking from Niklas Hallqvist . This (or something very similar) should end up in the kernel at some stage. -- Mike * Merged Eric's changes for 1.0 compatibilty - or rather didn't. After playing with diff for quite a while I finally realised we'd both done exactly the same changes :-). However Eric has tested them... -- Mike * Added Brandon's iBCS trace binary from Eric's 1.0 compatibility release. -- Mike Thu Jun 9 10:22:56 BST 1994 * Mapped msgsys to ibcs_msgsys in callmap.inc. It existed, we just didn't admit it before :-). -- Mike Tue Jun 7 08:50:34 BST 1994 * Sorted out some more ioctl stuff and added a handler for the STREAMS ioctl set. This is needed for I_NREAD (number of characters readable). Some SYSV X libraries contain implicit assumptions that the X connection is via a STREAM and use I_NREAD in preference to BSD/socket FIONREAD. -- Mike * Oh, Jeez... The changes made for the official kernel patches completely shafted the 1.0 compatibility. I've remerged the previous code with the new code - there may still be some things that need pulling back though. At this point I don't have a 1.0 kernel tree around to test against. -- Mike * If a COFF program is not correctly aligned disable demand paging and resort to slurping the whole lot in at start up. -- Mike * Added missing -I../include to the Tools Makefile. -- Mike Fri Jun 3 11:53:21 BST 1994 * Added my virtual system tools. A virtualised SCO 3.2.4 system is usable for real work for me... -- Mike * Added the synchronous ioctl set. This conflicts with the SCO 3.2.x (x < 4) ioctl for getting the video map so it should be dependent on the file descriptor it is applied to but since we don't currently support either... -- Mike Thu Jun 2 17:02:26 BST 1994 * Added support for FIORDCHK ioctl. SCO says it's there for "backward compatibility" but the system programs still seem to use it (notably ksh). -- Mike Tue May 31 13:39:34 BST 1994 * Patches to sysfs.c for 1.0.x compatibility from Tor with some extra clean up. -- Mike Fri May 27 09:15:21 BST 1994 * Ensure we set up signal() handlers as SA_ONESHOT and sigaction() handlers as sticky - similar to patch from Remete Gabor. -- Mike * Added the SCO cxenix sigaction extensions. Barely tested... -- Mike * Added the cxenix {get,set}itimer calls discovered on SCO by Brandon. Currently these are directly mapped to the Linux syscalls. They are undocumented by SCO (as far as we know) and untested by us (for now). -- Mike Thu May 26 11:58:18 BST 1994 * Don't include the 0.0.0.0 to localhost address mapping if we are building for 1.1.15+. This is mapped in the kernel as of 1.1.6 and since we don't support 1.1 kernels before 15... -- Mike * Type of signal map vectors should be unsigned long not int. -- Mike * Allow tracing to be compiled in to the COFF loader and turned on and off with the Tools/trace program. -- Mike * Signal maps moved out to maps/signal.inc. The only special mapping that happens is for ISC which has a problem with Oracle sending SIGSTOP to its backend when SIGSTOP doesn't appear to be defined for ISC. We just map SIGSTOP to 0 so it appears to work but does nothing. -- Mike * Changes for 1.1.15. This will not work with kernels between 1.1.10 and 1.1.14 inclusive - guaranteed. It should still work with the 1.0.x series but has not been tested recently. It probably works against 1.1.0 to 1.1.13 inclusive but that is purely coincidence - you shouldn't be using old 1.1.x kernels as it's a development series. -- Mike Tue May 24 17:27:54 1994 Eric Youngdale (eric@tantalus) * binfmt_xout.c: Use linux_binfmt struct, register_binfmt, unregister_binfmt as in pl14. * binfmt_elf.c, binfmt_xout.c: Likewise. * signals.c: Make signals[] and lsignals[] global variables. * ibcs/include/ibcs.h: Add prototypes for signals[] and lsignals[]. Tue May 24 10:37:01 BST 1994 * Added map entries for the ISC personality to all the maps files. Must be more careful adding personalities. -- Mike * Added Oracle to the COMPAT file. -- Mike * Remember the problem with interrupting an Informix 4GL program causing it to go psycho because of I/O errors from the tty? Well, this doesn't happen with the new tty drivers in 1.1.13. I never found why it happened... -- Mike * Noted that the 1.1.11 kernel patches are still ok with 1.1.13. -- Mike Mon May 23 08:50:21 BST 1994 * Mapped ISC sysconf() to the existing Xenix sysconf(). This looks right but isn't checked at all. -- Mike * Added ISC TCSETPGRP and TCGETPGRP ioctl traps. -- Mike Thu May 19 09:26:34 BST 1994 * Added a bug hunting section to the README. -- Mike * Always access the socket layer by picking up the file operations from an open socket descriptor. This removes the need for kernel socket functions to be global and should be friendlier towards a loadable socket layer one day. -- Mike Mon May 16 10:20:38 BST 1994 * Always access system calls via the system call table regardless of whether we are using a loadable module or linking directly to the kernel. This allows other things to be loadable too. -- Mike Wed May 11 13:52:12 BST 1994 * Added a tip on how to fix X font problems to the HINTS file. Other "soft" changes welcome. Does anyone read the ChangeLog? -- Mike * Introduced the PROD.Patches directory and put the WP install fixes in there. Documented in COMPAT and README. -- Mike * Added some hints on incompatible shell behaviour. -- Mike Mon May 9 11:07:52 BST 1994 * Updated the 1.1 patch file for 1.1.11. -- Mike * Additions for Interactive 4.0 -- Karl Kiniger (ki@kretz.co.at) Fri May 6 12:03:00 BST 1994 * Merged old sysfs.c with the new sysfs.c. The code will now work against any kernel - 1.0, 1.1, 1.1.9, 1.1.10... -- Mike Thu May 5 10:39:52 BST 1994 * Reworked for compatibility with 1.1.10. Kernel patches are *not* updated yet. We expect the kernel changes to go in to the official distribution any time now... -- Mike Tue May 3 12:03:31 BST 1994 * Joe/Brad's IPC additions. Fri Apr 29 10:06:10 BST 1994 * Updated the README to explain the difference between sysfs.c and sysfs.c.old -- Mike Wed Apr 27 11:38:52 BST 1994 * Added mapping of chsize() to ftruncate(). Implemented nap() using select(). -- Mike * Further fix to termios mapping for iexten and tostop. -- Mike * Updated the 1.1.? patch. New ksyms.c with fewer symbols plus removed all the socket patches. Sockfunc.h is no longer required. The 1.0 patch isn't updated but the extra cruft that's in there now shouldn't be any problem. -- Mike Tue Apr 26 11:49:07 BST 1994 * Removed references to linux/sockfunc.h. We seem to have run out of internal socket functions to put in it finally. -- Mike * Removed use of socki_lookup(). This just wraps some checks around inode->i_socket. We assume the socket layer is stable by now... -- Mike * Access [gs]etsockopt via the socketcall interface. -- Mike * If we have to use our own CFLAGS look in the include directory above in case we have been dropped in to the kernel. -- Eric * Grab socket file operations when we create a socket rather than linking against them directly. Similar to Eric's changes but different... -- Mike Fri Apr 22 11:10:18 BST 1994 * The 'x' class of ioctls are back to being termios rather than termio. The problem was that the field size in iBCS termios and Linux termios is different. -- Mike * Added iBCS <-> Linux termios mapping. This is untested and probably imperfect at this stage but nothing crashes with this code... -- Mike * The iBCS time() syscall doesn't take an argument, the save, if any, is handled in the interface library. -- Mike * Updated the 1.1 kernel patch to be against 1.1.8. It won't work with anything less than 1.1.8 but then if you are playing with 1.1 you should be playing with the latest... This involves a change in the way sysfs() works. If you are using 1.0 and have applications that call sysfs() (practically none do) then you need to build with sysfs.c.old. -- Mike * The Linux kernel 1.1.8 replaces the static file_systems table with a linked list ready for loadable file systems. Naturally, I was ready for this... :-( -- Mike Thu Apr 21 10:34:34 BST 1994 * sysfs.c should be including linux/string.h to get the inline string functions rather than needing library functions. -- Mike * Added code to map between Linux termio structs and SCO/iBCS termio structs. There are slight differences. Mapping needs to be done for termios structs too. Anything that uses termios at this stage is likely to ge a nasty shock! -- Mike Tue Apr 19 22:56:55 1994 Eric Youngdale (eric@esp22) * ibcs.h: Put declaration of sys_call_table outside of ifdef. (So can be compiled into kernel instead as a module). Tue Apr 19 10:21:17 BST 1994 * It looks as though the 'x' class of ioctls should be termio ioctls rather than termios. At least this is the case as far as SCO 3.2.4 stty appears to be concerned. -- Mike * Avoid copying the entire lock structure backwards and forwards just to increment one field in open.c. There could be a problem here - SCO has a sysid field *before* the pid field but only if _XOPEN_SOURCE or _POSIX_SOURCE is defined at compile time. There appears to be no obvious way to know what is being used??? -- Mike Mon Apr 18 08:49:37 BST 1994 * There appear to odd little niceties involved with SYSV which depend on system configuration and the way users do things. Added a HINTS file with known problems/workarounds. -- Mike * Changed some annoying KERN_ERRs to KERN_DEBUGs in signal.c -- Mike * Watch out for (mis)use of INADDR_ANY instead of localhost (primarily by syslog() code?) and replace it with the localhost address. We should perhaps do this for sendto() as well? (Eric spotted this one on CorelDraw but it's a problem with the standard libsocket.c) -- Mike * Processes should be allowed to reduce their file size limit but only increase it if they have root privileges. -- Mike * Fixed the bug in [f]statfs where the first two arguments to memcpt_tofs() were exchanged (spotted by Eric). -- Mike * SCO seems to like utsys (v7_utsname) to return the same thing for sysname and nodename fields just like the SCO extended utsname structure has. The SVR4 (abi_utsname) return is unknown at this time. -- Mike * Removed comments on IPC from README. It *is* fixed? -- Mike * Fix IPC which was broken by syscall changes. Also fix structure mappings. -- Joseph L. Portman III Thu Apr 14 11:27:24 BST 1994 * At some stage SCO managed to build some of their binaries with the copyright comment string in quotes so we have a new recognition string in the COFF loader. (Seen in uname on a 3.2.2 system). -- Mike * If we have tracing compiled in then only give unsupported syscall messages if we are tracing the API or the syscall in question. This allows us to quieten unsupported but unimportant syscalls. -- Mike * Comment on the IPC limitations in the README file. -- Mike * Added implementation of xnx_pathconf(). Actually it doesn't look as if we really need it. The Microsoft C compiler seems to be misusing it anyway and doesn't even care that it gets an error. -- Mike Wed Apr 13 09:14:50 BST 1994 * Added handling of TIOC[GS]PGRP. -- Mike * Added implementation of xnx_eaccess(). This is currently simpleminded because Linux access() is. My eaccess() may even be wrong. -- Mike * When tracing signal delivery we shouldn't be clearing the signal flag after reporting it or the signal is never actually delivered. I copied the assembler from the kernel and forgot to deleted an opcode... -- Mike * Signal 0 should have been identity mapped so kill 0 works. -- Brandon S. Allbery (kf8nh@kf8nh.ampr.org) (bsa@kf8nh.wariat.org) Tue Apr 12 14:30:25 BST 1994 * Corrected file size passed to/from setrlimit/getrlimit -- Graham Adams (gadams@ddrive.demon.co.uk) Tue Apr 12 11:16:45 BST 1994 * Added support for an obscure SCO extension that is used to get SCO's extended utsname information. -- Mike * Changed ipc.c to always go via the ipc syscall rather than direct to internal functions. This should survive if ipc isn't in the kernel. It should even do sensible things if the ipc module is loaded/unloaded on us. -- Mike * Initial changes to access system calls via sys_call_table -- Mike 05 April 1994 * Eric's pre-release fixes. Wed Mar 30 22:35:28 1994 Eric Youngdale (eric@esp22) * ioctl.c: Change 'F' to 'f' for SVr4 FIONREAD ioctl. * Makefile: Add svr4.c. * svr4.c: New file (sort of - it got lost in shuffle before). * include/ibcs/ibcs.h: Include prototypes for abi_uname. * maps/callmap.inc: Insert abi_uname in syscall number 135. Add sys_rename in slot 134 (emacs v18 requires this). Tue Mar 29 23:32:35 1994 Eric Youngdale (eric@esp22) * Makefile: Fix so that we do not need uname. Use symbol from .config instead. * README: Update a bit. 28 March 1994 * Preparation for release: Tidy up documentation and create CREDITS file from the old ToDo list. -- Mike 27 March 1994 * Preparation for release: Move headers into a subdirectory. Move maps into a subdirectory. -- Mike 25 March 1994 * Changed the COFF loader so that if the filesystem doesn't support mmap we read in the whole lot initially and let it page to swap if necessary. This is already in the x.out loader, it should go in the ELF loader too at some point. -- Mike 24 March 1994 * Added a loader for x.out i386 small model binaries - i.e 386 Xenix programs. <=286, non-small model binaries are not supported and not likely to be in the near future. -- Mike Wed Mar 23 23:12:54 1994 Eric Youngdale (eric@esp22) * Add ioctl for 0x4004667f (FIONREAD) (used by SVr4). * map.h (map_segment.map): Make unsigned. * hrtsys.c: New file implementing parts of the hrtsys syscall. 02 March 1994 * Add socket value mappings. This meant a general move round to tidy things up and group map functions/tables. There is a new kernel patch in the Patches directory called net.patch which gives us access to the sock_*sockopts functions directly. -- Mike 28 February 1994 * Implementation of poll(). -- Eric 25 February 1994 * Pedantic change to call maps so that IBCS_function contains a void * instead of a function pointer - we cast it as necessary later in emulate.c. The warnings were annoying me. * Moved struct abi_sigaction from signal.c to abi4.h so it is available for prototype declarations. Changed prototype for abi_sigsuspend to correspond to implementation in signal.c. -- Mike * Reversed out Eric's earlier signal patches and added new ones. -- Mike * Updated trace code and trace control program to be useful. Control of tracing individual functions is still not there yet - perhaps another day. Default trace mode is now none (i.e. only functions with the trace flag set are traced). Use the trace program to change the trace level. -- Mike * File modes (open/fcntl flags) are different between Linux and iBCS so we need to map between them. Open also requires this so fcntl.c is now merged with open.c. Failure to set and reset non-blocking mode was what was breaking Unipox. -- Mike * Signal handling function changes to map to and from the Linux signal set and sigaction flags correctly. -- Eric 24 February 1994 * Added code to the emulate() function to let us know when we are about to process a signal on exit from the syscall. -- Mike * Implemented proctl() as a no-op. It's not really relevent under Linux. -- Mike * Added argument count and type for eaccess() -- Mike * Have emulate.c return -ENOSYS for unimplemented system calls rather than zero. -- Mike * Added Eric's patches to waitpid. * Created the ChangeLog!