Enter your search terms Submit search form
	Web pdp-10.trailing-edge.com

28-Feb-88 03:35:25-PST,1659;000000000000
Received: from XX.LCS.MIT.EDU by SRI-NIC.ARPA with TCP; Sun 28 Feb 88 03:34:50-PST
Date: Sat, 27 Feb 1988  19:31 EST
Message-ID: <[email protected]>
From: Rob Austein <[email protected]>
To:   [email protected]
cc:   [email protected]
Subject: redirected stdout destroys existing files on "quota exceeded"

It seems that, if you redirect stdout to a disk file in a directory
that is at (or over) quota, KCC will quietly proceed to write and
CLOSF% an zero-length file.  This is a Very Bad Thing.  If it can't
write the file it should flame and abort, rather than destroying the
old copy of the file.  printf() isn't even returning an error when
this happens.  I expect that the problem also applies to fopen()'d
files, but I haven't checked.

[PHOTO:  Recording initiated  Sat 27-Feb-88 7:09PM]

 MIT TOPS-20 Command Processor 5(312161)-2
XX>fxnntp <$nntp$.grep
soc.women 15048 *BBOARD:SOC_WOMEN.TXT.1
rec.music.folk 1130 *BBOARD:REC_MUSIC_FOLK.TXT.1
rec.arts.comics 6178 *BBOARD:REC_ARTS_COMICS.TXT.1
news.lists 369 *BBOARD:NEWS_LISTS.TXT.1
news.groups 2661 *BBOARD:NEWS_GROUPS.TXT.1
news.config 467 *BBOARD:NEWS_CONFIG.TXT.1
news.announce.conferences 202 *BBOARD:NEWS_ANNOUNCE_CONFERENCES.TXT.1
news.admin 1655 *BBOARD:NEWS_ADMIN.TXT.1
XX>fxnntp <$nntp$.grep >bb:$nntp$.status
XX>i disk bb:
 XX:<BBOARD>
 5000 Pages assigned
 5000 Working pages, 5000 Permanent pages allowed
 14129 Pages free on XX:, 365871 pages used.
XX>v bb:$nntp$.status.0

   XX:<BBOARD>
 $NNTP$.STATUS.161;P777777   0 0(7)      27-Feb-88 19:10:50 SRA       
XX>pop

[PHOTO:  Recording terminated  Sat 27-Feb-88 7:11PM]
28-Feb-88 19:00:03-PST,1771;000000000000
Mail-From: KLH created at 28-Feb-88 18:55:27
Date: Sun, 28 Feb 88 18:55:21 PST
From: Ken Harrenstien <[email protected]>
Subject: Re: redirected stdout destroys existing files on "quota exceeded"
To: [email protected], [email protected]
cc: [email protected]
In-Reply-To: <[email protected]>
Message-ID: <[email protected]>

Well, what SHOULD happen?  "flame and abort" is imprecise.  ITS does
this much better -- it generates an interrupt which DDT knows about,
and you can proceed once you have freed up enough stuff.  This removes
the need for every program in the world to have its own
"check-for-overquota-error-and-tell-user-and-wait" routines.

In other words, this seems like a TOPS-20 problem to me.  How could
OPENF and CLOSF destroy a file, anyway?  If you have your gen-ret-count
set to 1, that's your mistake.

I agree the stdio functions should return an error if the output write()
fails, and the write() should fail if the SOUT%/BOUT% fails.  That,
I can look into later.  Note that most programs, however, never bother
to check the return value of putc or printf or any other output function.
Usual Unix brain damage.

Note that .ICQTA is bound to SIGXFSZ.  It isn't being triggered presumably
because an ERJMP turns the error condition into a simple error return.
I could, however, cause it to be invoked, and this could either
(1) call a user-defined signal handler, if one was specified by signal(),
or (2) as a default action, "flame and abort" (whatever you want that to be).
It could presumably ask the user whether to continue or not, although it
may be hard to know whether that makes sense (there may not be a "user" and
the channel to use may not be obvious either).
-------
28-Feb-88 19:25:03-PST,3104;000000000000
Received: from XX.LCS.MIT.EDU by SRI-NIC.ARPA with TCP; Sun 28 Feb 88 19:21:11-PST
Date: Sun 28 Feb 88 22:20:25-EST
From: Rob Austein <[email protected]>
Subject: Re: redirected stdout destroys existing files on "quota exceeded"
To: [email protected]
cc: [email protected], [email protected]
In-Reply-To: <[email protected]>
Message-ID: <[email protected]>

    Date: Sun, 28 Feb 88 18:55:21 PST
    From: Ken Harrenstien <[email protected]>

    Well, what SHOULD happen?  "flame and abort" is imprecise.  ITS does
    this much better -- it generates an interrupt which DDT knows about,
    and you can proceed once you have freed up enough stuff.  This removes
    the need for every program in the world to have its own
    "check-for-overquota-error-and-tell-user-and-wait" routines.

    In other words, this seems like a TOPS-20 problem to me.  How could
    OPENF and CLOSF destroy a file, anyway?  If you have your gen-ret-count
    set to 1, that's your mistake.

I have the generation retention count set to six or something like
that.  But if a program runs under a batch job every few hours and
tells that batch job that it successfully completed and writes out a
new empty file each time, it will eventually cause the last good copy
of the file to disappear.  I suppose I could devote an entire disk
pack to the status files of this program so that I could keep a week's
worth around when I go on vacation, but that seems kind of excessive.

    I agree the stdio functions should return an error if the output write()
    fails, and the write() should fail if the SOUT%/BOUT% fails.  That,
    I can look into later.  Note that most programs, however, never bother
    to check the return value of putc or printf or any other output function.
    Usual Unix brain damage.

    Note that .ICQTA is bound to SIGXFSZ.  It isn't being triggered
    presumably because an ERJMP turns the error condition into a
    simple error return.  I could, however, cause it to be invoked,
    and this could either (1) call a user-defined signal handler, if
    one was specified by signal(), or (2) as a default action, "flame
    and abort" (whatever you want that to be).  It could presumably
    ask the user whether to continue or not, although it may be hard
    to know whether that makes sense (there may not be a "user" and
    the channel to use may not be obvious either).

Is it reasonable to say that if an output (write()) to a file failed,
the file should be closed with CZ%ABT?  Maybe not, this might screw
tape programs.

I'd settle for having the error status be reported by printf().  What
really got me was that there was nothing my program could have done
differently to detect the error so that it could abort().

Perhaps the right thing to do is call a handler routine if one was specified
by signal(), else just return an error code from write() back up through
printf().  I'm not familiar enough with the various unix signal() codes
to know if SIGXFSZ is the right one for this.
-------