Re: PATCH: warn on "hard" hard link failure

Re: PATCH: warn on "hard" hard link failure

[Clint Adams]

On Tue, Mar 10, 2009 at 02:18:41PM +0000, Peter Stephenson wrote:
> This is reponse to Sourceforge issue 1968395: failures to create hard
> links aren't gracefully handled.  Failing to create a link because the
> file already exists is the normal condition being handled by the locking
> code; however, a failure for any other reason is presumably a hard
> failure worthy of reporting.  The same goes if stat() fails with
> anything other than ENOENT: in both cases we give up and should say so.
> 
> It's possible this can be improved further.

Apparently this has caused someone distress because of the hard failure
preventing the correction of the out-of-diskspace condition causing the
history failure.

Re: PATCH: warn on "hard" hard link failure

[François Revol]

> On Tue, Mar 10, 2009 at 02:18:41PM +0000, Peter Stephenson wrote:
> > This is reponse to Sourceforge issue 1968395: failures to create
> > hard
> > links aren't gracefully handled.  Failing to create a link because
> > the
> > file already exists is the normal condition being handled by the
> > locking
> > code; however, a failure for any other reason is presumably a hard
> > failure worthy of reporting.  The same goes if stat() fails with
> > anything other than ENOENT: in both cases we give up and should say
> > so.
> >
> > It's possible this can be improved further.
>
> Apparently this has caused someone distress because of the hard
> failure
> preventing the correction of the out-of-diskspace condition causing
> the
> history failure.

I don't know what you use hardlinks for, but beware, not every OS out
there supports them.
BeOS and Haiku don't for example.
Actually they don't forbid them at the VFS layer, but BFS doesn't
support hardlinks due to the way it is implemented to support indices
and queries.

So please do provide a fallback in case of failure.

I should probably post my zsh diff someday...

François.

PATCH: warn on "hard" hard link failure

[Peter Stephenson]

This is reponse to Sourceforge issue 1968395: failures to create hard
links aren't gracefully handled.  Failing to create a link because the
file already exists is the normal condition being handled by the locking
code; however, a failure for any other reason is presumably a hard
failure worthy of reporting.  The same goes if stat() fails with
anything other than ENOENT: in both cases we give up and should say so.

It's possible this can be improved further.

Index: Src/hist.c
===================================================================
RCS file: /cvsroot/zsh/zsh/Src/hist.c,v
retrieving revision 1.87
diff -u -r1.87 hist.c
--- Src/hist.c	25 Feb 2009 10:30:15 -0000	1.87
+++ Src/hist.c	10 Mar 2009 14:15:07 -0000
@@ -2433,13 +2433,16 @@
 	    } else
 		close(fd);
 	    while (link(tmpfile, lockfile) < 0) {
-		if (errno != EEXIST || !keep_trying)
+		if (errno != EEXIST)
+		    zerr("failed to create hard link as lock file %s: %e",
+			 lockfile, errno);
+		else if (!keep_trying)
 		    ;
 		else if (stat(lockfile, &sb) < 0) {
 		    if (errno == ENOENT)
 			continue;
-		}
-		else {
+		    zerr("failed to stat lock file %s: %e", lockfile, errno);
+		} else {
 		    if (time(NULL) - sb.st_mtime < 10)
 			sleep(1);
 		    else


-- 
Peter Stephenson <pws@csr.com>                  Software Engineer
CSR PLC, Churchill House, Cambridge Business Park, Cowley Road
Cambridge, CB4 0WZ, UK                          Tel: +44 (0)1223 692070