From: Digestifier To: Linux-Development@senator-bedfellow.mit.edu Reply-To: Linux-Development@senator-bedfellow.mit.edu Date: Sat, 15 Oct 94 19:13:06 EDT Subject: Linux-Development Digest #313 Linux-Development Digest #313, Volume #2 Sat, 15 Oct 94 19:13:06 EDT Contents: Re: Shared Libs: working toward a permanent solution? (Eric Youngdale) Re: ext2fs vs. Berkeley FFS (Peter Mutsaers) Re: Anyone working on improving NFS? (Bill Heiser) Re: 1.1.53 Wont compile AHA152x.c (H.J. Lu) serial line problem (Joerg Wedeck) Re: ypbind didn't work (Dominik Kubla) Re: Kernel Patch 1.1.54: Make dep failed: File missing (Mark Lord) Drivers for DPT SCSI available ? (Boon Chong Benjamin Low) Re: insmod drv_hello.o gives "___moddi3 undefined" (Christopher Gori) Re: 8-bit colour ANSI and ncurses (Rasmus Lerdorf) Re: LINUX Logical volumes (Jay Ashworth) Re: Single host firewalling (Jay Ashworth) Re: PGP for Linux?? (Jon Freivald) Re: Linux Mud (Dirk Weigenand) Re: We a FAQ: Linux vs. *BSD!!! (Jordan K. Hubbard) fsync doesn't work on umsdos file systems (Marty Leisner 25733) Re: Serious Bug In The Networking Code (Daniel Tran) Re: A badly missed feature in gcc (Eric V. Smith) Re: ext2fs vs. Berkeley FFS (H. Peter Anvin) ---------------------------------------------------------------------------- From: eric@aib.com (Eric Youngdale) Subject: Re: Shared Libs: working toward a permanent solution? Date: Wed, 12 Oct 1994 22:09:41 GMT In article , H. Peter Anvin wrote: >Actually, the selector could be prepared by the kernel at library load >time (at the same time as the __libbase variable is initialized), and >all that needs to be done at entry time is loading the selector into >the appropriate register (9 cycles on 486, 19 on 386). In theory this could be done, I suppose. It would be a horrible kluge, I think. I would like to keep the implementation close to what is used on other i386 Unix platforms, and segment selectors are not in the picture there. For one thing, it would require patches to gcc to generate the new code, and it would require that we invent new relocation types and correctly handle these in gas and ld. Finally, the dynamic linker would have to initialize segment descriptors in the ldt for each shared image that is use. As it is with ELF, we are leveraging off of the efforts of people who would like to use gcc/gas/ld on SVr4 systems. >How does the current PIC code find out where the entrypoint is, >incidentally (i.e. how does the user code call the library)? All of the symbols are resolved at runtime. When the binary starts up, control is passed to the dynamic linker whose job it is to locate and load all of the shared libraries that are required. Then it performs relocations, which means that it essentially connects the application to the shared library. The dynamic loader has the overall picture, since it knows the VMA where each shared library was loaded into memory, and it picks apart the .hash, .dynsym and .rel.* sections and does the right thing. Incidentally, the default is to use lazy symbol binding which means that .plt entries are not initialized at startup time, so that once you call a function for the first time the dynamic linker is called instead. The dynamic linker looks up the actual address of the requested function, fixes the .plt entry and then calls the function so that the original invocation works correctly. The idea is that you do not waste a lot of time initializing a lot of .plt entries when you may only use a fraction of them. Again, if it can be demonstrated that the performance hit is unacceptable, then I believe that it is possible to implement something much more within the framework of PIC ELF which would not require that we give up the %ebx register. -Eric ------------------------------ From: plm@atcmp.nl (Peter Mutsaers) Subject: Re: ext2fs vs. Berkeley FFS Date: Mon, 10 Oct 1994 22:26:52 GMT >> On 10 Oct 1994 19:04:18 GMT, hstrong@eng1.uconn.edu (Hugh Strong) said: HS> : So the only way to go is to create a directory with files in it that HS> : belong together. It has been that way since the beginning. What is HS> : wrong with that. HS> Everything, if you wish to give files attributes. HS> UNIX programs read and write *files*, not directories. HS> This *would* break existing code, if we were to suddenly start HS> using separate directories to store little bits of fluff associated HS> with each file. Forks are for storing information that pertains HS> to a particular file, not for keeping separate but related HS> data sets together in one large file. Whatever notation you choose, whether directories, or file:ICON etc., your programs must be adapted to use it. Except, when you use directories, existing probrams such as 'cp' can still be used to access the seperate parts of a 'fork'. HS> Really, keeping everything the way it was in My Father's Time for HS> the pure simple joy of it is not a good idea. The suggestion given before is no more than 'syntactic sugar' I think. You have a file on a different filesystem that is composite somehow. There are several ways to map that into the Unix filesystem. I would choose the mapping that requires no extension in the current kernel. User programs can always put a layer above this and shield the details for higher levels. HS> : It is basic in the Unix philosophy that files are untyped and that the HS> : kernel does not care what is in the file. Adding such things is HS> : completely against the Unix way of thinking. HS> Forks are untyped and unstructured as well. Nothing should prevent HS> you from writing any arbitrary chunk of data into a fork, except HS> lack of priveledges. You have one sequential stream of bytes, a file. If you have more streams of bytes, (so called forks apparently) you have more files. Either in a directory, or with different extensions (like the *.pag an *.dir files from dbm). It seems completely useless to me to extend the existing semantics of the system calls for this. -- Peter Mutsaers | AT Computing bv, P.O. Box 1428, plm@atcmp.nl | 6501 BK Nijmegen, The Netherlands tel. work: +31 (0)80 527248 | tel. home: +31 (0)3405 71093 | "... En..., doet ie het al?" ------------------------------ From: bill@bhhome.ci.net (Bill Heiser) Subject: Re: Anyone working on improving NFS? Date: 15 Oct 1994 14:44:19 GMT Amrik Thethi (at@setanta.demon.co.uk) wrote: : In article stock@cs.buffalo.edu (Matthew D Stock) writes: : >Hi. Is anyone currently working on improving NFS under Linux? If I : The latest kernels (> 1.1.50 I think) have a simple static 5 block cache : in the NFS client. This improves average performance significantly (~50%). I found that by increasing the wsize and rsize to 8192, performance is made quite reasonable even with older kernels (i.e. 1.1.4x). However my biggest problem with LINUX NFS is that there is no File Locking done (i.e. rpc.lockd in SunOS). So, for example, I can't NFS mount my mail spool directory from another machine - if I do it gets corrupted. -- Bill Heiser: bill@bhhome.ci.net, heiser@world.std.com ------------------------------ From: hjl@nynexst.com (H.J. Lu) Subject: Re: 1.1.53 Wont compile AHA152x.c Date: 15 Oct 1994 11:48:24 GMT Earl A. Stutes (estutes@netcom20.netcom.com) wrote: : >> Subject says it all. I get a GCC internal error on this file now. : >> Gcc2.6.0 : This has been going on for several pathc levels. I had the problem : back when the patch level was about 1.1.40 or so. every now and then, : I try to build the whole kernel with 2.6.0 and it breaks on aha152x.c. : I switch my gcc symbolic link to 2.5.8, delete the .o file and finish : the build. So, I am running 1.52 compiled mostly on 2.6.0 with a few : modules 2.5.8. It has been working fine for me. I could not see : anything obvious in aha152x.c that should be causing the problem. That is a gcc bug. I have said many times that gcc 2.6.0 is too buggy to use. If you really want to use gcc 2.6.x, please join the GCC channel. Gcc 2.6.0-941008 compiles 1.1.54 just fine. H.J. ------------------------------ From: jw@peanuts.informatik.uni-tuebingen.de (Joerg Wedeck) Subject: serial line problem Date: 15 Oct 1994 14:38:54 GMT Hi all, it appears to me that i have a serial line problem with kernel 1.1.54 on a system with an i468 dx2/66, asus sp3g motherboard, IDE disk + SCSI disk, both using the onboard controller, a spea v7 mercury lite pci graphics adater and a serial mouse. when running without dialing out via the second serial everything works fine. when dialing out, the system completely hangs after a while, i suspect ther is a deadlock somewhere. This especially happens when switching from X to a virtual console or switching back. Does anybody have a solution in mind ? Joerg -- ==================================+====================================== Joerg Wedeck | E-Mail: jw@peanuts.informatik\ Universitaet Tuebingen | .uni-tuebingen.de Lehrstuhl fuer Techn. Informatik | Sand 13 | 72076 Tuebingen | Germany | ==================================+====================================== Deswegen bin ich ja Optio, und Du nur Legionaer :-) ========================================================================+ ------------------------------ From: kubla@Uni-Mainz.DE (Dominik Kubla) Subject: Re: ypbind didn't work Date: 14 Oct 1994 06:19:06 GMT Ari Widodo writes: First, I would like to say thanks for all whom gave me information about NIS. And I got the yp-client packages and install that. But I canN't run ypbind. --- /usr/sbin/ypbind Undefined C library functions: 1. light C shared image (Use the real one instead.) --- Should I recompile my C library ? Can somebody give me some explanation for this error ? No. Each libc binary distribution consists of TWO libc shared images, by default called /lib/libc.so... /lib/libc-lite.so.< major>.. The difference between the full-featured and the lite version is that in the latter yp, locale and other things. are missing. This is done to fit the library on the boot/rootdisk combos. The error message simply states that either the lite library was installed under the name of the augmented version or that the symbolic link /lib/libc.so. points to the lite version. In the first case you will just need to install the latest binary release (or compile it yourself if you like). In the second case simply run 'ldconfig -v'. Cheers, Dominik -- =========================================================================== eMail: Dominik.Kubla@Uni-Mainz.DE sMail: Dominik Kubla, Lannerstrasse 53 55270 Ober-Olm, F.R. of Germany >>> Save the environment NOW! <<< ****** European Union ****** ------------------------------ From: mlord@bnr.ca (Mark Lord) Subject: Re: Kernel Patch 1.1.54: Make dep failed: File missing Date: 15 Oct 1994 15:03:22 GMT In article bart@dunedin.es.co.nz writes: I am trying to use the module utilities with the following set up: > > kernel version: 1.1.51 > cc version: 2.5.8 > modules.tar.gz: dated June 27, 1994 Same as you but 1.1.53. > >When attempting to load the sample module dvr_hello.o, however, >I get the following error: > >___moddi3 undefined > Exact same error. Why? >Is there something special I need to do when compiling module >object files? > >Here is a list of the symbols in drv_hello.o: [nm output exactly same as mine deleted] Any help would be appreciated. Chris Gori cgori@isengard.stanford.edu ------------------------------ From: rasmus@io.org (Rasmus Lerdorf) Subject: Re: 8-bit colour ANSI and ncurses Date: 13 Oct 1994 06:42:39 -0400 hpa@ahab.eecs.nwu.edu (H. Peter Anvin) writes: >Output: > ) U I looked through the console driver source and noticed that ]11m does what I want. ie. show the chars for ascii codes <32 and also switches to the first alternate character set. From within ncurses it appears as though attrset(A_ALTCHARSET) does the trick. ]11m along with 10, 12, 21,22,24 and 27 are mode codes I haven't seen before. They are not in the ANSI spec, and they don't look like VT100 sequences to me. *Rasmus* ------------------------------ From: jra@zeus.IntNet.net (Jay Ashworth) Subject: Re: LINUX Logical volumes Date: 10 Oct 1994 20:23:22 -0400 killianr@beldin.sun.ac.za (Richelo Killian) writes: > Is it posible to create logigal volumes across drives and/or > partitions and then mount a single filesystem on that volume? I know it > can be done on HP-UX, but I want to do it on my LINUX box? To clarify, what I believe you're asking about is the ability to create a logical filesystem/volume which spans physical volumes, i.e. a 3GB filesystem spanning 3 1GB drives. No, I don't believe anyone's done this yet. Other commercial Unices which allow this include Data General's DG-UX, where I first saw the feature. It _is_ nice... but it would be murder to hack into the kernel, I suspect. Not a problem of implementation, but of design. Cheers, -- jra -- Jay R. Ashworth High Technology Systems Consulting Ashworth Designer Linux: The Choice of a GNU Generation & Associates ka1fjx/4 +1 813 790 7592 jra@baylink.com "Hey! Do any of you guys know how to Madison?" NIC: jra3 ------------------------------ From: jra@zeus.IntNet.net (Jay Ashworth) Subject: Re: Single host firewalling Date: 10 Oct 1994 20:28:20 -0400 riku.saikkonen@compart.fi (Riku Saikkonen) writes: [ about securing your machine from a network ] >>If you're talking about MS-Windows, you probably have little to worry >>about, provided you don't leave any listener processes up. Now, >In Linux, the lazy way out is to kill inetd... (right?) Almost. Most network server process on a Unix box are started by inet, but not all... noticeable exceptions tend to include your mailer-daemon, and let's not forget that big, lovable, X server sitting on port 6000. Kill inetd, and then do a netstat, and see who's still listening. Cheers, -- jra -- Jay R. Ashworth High Technology Systems Consulting Ashworth Designer Linux: The Choice of a GNU Generation & Associates ka1fjx/4 +1 813 790 7592 jra@baylink.com "Hey! Do any of you guys know how to Madison?" NIC: jra3 ------------------------------ From: jaf@pine.liii.com (Jon Freivald) Subject: Re: PGP for Linux?? Date: Sat, 15 Oct 1994 02:26:06 GMT mwe@dfw.net wrote: : Zack T. Smith (zack@netcom.com) wrote: : : Can anyone tell me whether PGP (the encyption utility) been ported Linux? : : I haven't been able to find it in the archives... : "since about forever" PGP compiles out-of-the-box under Linux. : grab a copy of the source code and type 'make'. Actually, type 'make linux'. ------------------------------ From: weigo@asi09.rz.uni-leipzig.de (Dirk Weigenand) Subject: Re: Linux Mud Date: 15 Oct 1994 15:33:48 GMT Try Amylaars driver available from mud.uni-muenster.de. It's a driver for LP-Muds capable running in compat and native mode. There is also a mudlib available at that adress. This driver is developed on linux and hence runs without a flaw. You should also install byacc since yacc/bison complain about illformed rules in the grammar file used. Have fun, Dirk ------------------------------ From: jkh@freefall.cdrom.com (Jordan K. Hubbard) Crossposted-To: comp.os.386bsd.development Subject: Re: We a FAQ: Linux vs. *BSD!!! Date: 15 Oct 1994 05:12:01 GMT In article <37mflh$f6s@unix1.cc.uop.edu> hughes@napa.eng.uop.edu (Ken Hughes) writes: Normally I would agree, but in this case I doubt that ignoring the questions will stop them from being asked. New people come into these Read what I said more carefully, please. I didn't propose to stop the questions, I simply proposed to stop the answers. The questions will keep coming in, but will "quench" themselves far more quickly if people simply ignore them. We're always going to have this question, there's no doubt, so the best we can do is try to minimize the amount of damage it does. 386BSD-1.0", so please don't start THAT thread again either). IMHO, the only chance for resolving this issue would be for someone to sit down and compile a list of features that shows which OS has what and which doesn't. I don't agree. As you seem to concur, the FAQ will be out of date and essentially useless far too rapidly to make the effort of writing it anything but a wasted one. Jordan ------------------------------ From: leisner@batman (Marty Leisner 25733) Subject: fsync doesn't work on umsdos file systems Reply-To: leisner@sdsp.mc.xerox.com Date: Fri, 14 Oct 1994 21:21:59 GMT I'm running Keith Bostic's nvi (1.34) on linux... I have kernel 1.1.52 It doesn't an fsync when it writes the file...but fsync doesn't work on umsdos file systems (but it works on msdos file systems...) I gets an error of "invalid argument"... -- marty leisner@sdsp.mc.xerox.com Member of the League for Programming Freedom C++ is to C as lung cancer is to lung. ------------------------------ From: dtran@emelnitz.ucla.edu (Daniel Tran) Crossposted-To: comp.os.linux.admin,comp.os.linux.help Subject: Re: Serious Bug In The Networking Code Date: Mon, 10 Oct 1994 17:51:46 GMT In article <37a37b$a2d@Venus.mcs.com> munster@MCS.COM (Jerry Ablan) writes: >: There appears to be a serious bug in some of the networking code >: supplied with linux/slackware, that causes the computer to get >: 'network unreachable' after approximately 3 minutes of perfect >: functioning. I have no idea what the problem might be, and if >: somebody tell me where to look, I can try to figure out what versions >: my drivers etc. are. Here are the configurations I ve gotten this >: problem with: >I've noticed that this occurs when you run routed. Do not run routed and see >if it still happens. >-- Jerry Would running running routed with -q parameter help. Daniel Tran - dtran@emelnitz.ucla.edu ------------------------------ From: esmith@access1.digex.net (Eric V. Smith) Subject: Re: A badly missed feature in gcc Date: 14 Oct 1994 03:20:54 -0400 I too will suspend my good judgement and jump in on this one. In article <37julm$59t@nz12.rz.uni-karlsruhe.de>, Thomas Koenig wrote: >Microsoft et.al. are trying to pull the same tricks on the users that >IBM got away with a long time. When people with this kind of mindset >collide with the portability fanatics from UNIX systems, you get a >culture clash ;-) > I believe "portability fanatics from UNIX systems" to be an oxymoron. In general, I find the Unix crowd to be about the least concerned with (or aware of) portability of any group I've seen. From "integers are pointers" to "pointers to data can hold pointers to functions" to "ints are 32 bits", this group of programmers has written some of the _least_ portable code I've seen. My, $0.02, and I apologize in advance for giving into my temptation to follow up on this tired thread. Eric. ------------------------------ From: hpa@ahab.eecs.nwu.edu (H. Peter Anvin) Subject: Re: ext2fs vs. Berkeley FFS Reply-To: hpa@nwu.edu (H. Peter Anvin) Date: Mon, 10 Oct 1994 17:51:12 GMT Followup to: <37bpse$ue@newsy.ifm.liu.se> By author: peter@ifm.liu.se (Peter Eriksson) In newsgroup: comp.os.linux.development > > >The icon (for a window manager) for the file could be > >accessed by the following call. > > > fd1 = open("MyDataFile:ICON",O_RDONLY); > > If one were to implement something like this, then it would be much > better to use the "/" character to separate the filename and the subforks... > > Just a little thought. > And, incidentally, it works without kernel mods!! (A multifork file is *exactly* the same as a directory!) /hpa -- INTERNET: hpa@nwu.edu --- Allah'u'abha --- IBM MAIL: I0050052 at IBMMAIL HAM RADIO: N9ITP or SM4TKN FIDONET: 1:115/511 or 1:115/512 STORMNET: 181:294/1 or 181:294/101 "NT is not a bad thing if I don't have to use it..." -- xmsb@borland.com ------------------------------ ** FOR YOUR REFERENCE ** The service address, to which questions about the list itself and requests to be added to or deleted from it should be directed, is: Internet: Linux-Development-Request@NEWS-DIGESTS.MIT.EDU You can send mail to the entire list (and comp.os.linux.development) via: Internet: Linux-Development@NEWS-DIGESTS.MIT.EDU Linux may be obtained via one of these FTP sites: nic.funet.fi pub/OS/Linux tsx-11.mit.edu pub/linux sunsite.unc.edu pub/Linux End of Linux-Development Digest ****************************** .