[1664] in testers
[jik@cats.UCSC.EDU: delete changes]
daemon@ATHENA.MIT.EDU (daemon@ATHENA.MIT.EDU)
Wed Jul 31 15:02:47 1991
Date: Wed, 31 Jul 91 15:02:48 -0400
From: Mark Rosenstein <mar@MIT.EDU>
To: testers@MIT.EDU
This should take care of our concerns for delete in AFS lockers.
-Mark
Date: Fri, 26 Jul 91 15:37:40 -0700
From: "Jonathan I. Kamens" <jik@cats.UCSC.EDU>
Sender: jik@cats.UCSC.EDU
To: mar@MIT.EDU
Cc: mar@MIT.EDU
In-Reply-To: Mark Rosenstein's message of Fri, 26 Jul 91 16:50:04 -0400 <9107262050.AA01049@Toto.MIT.EDU>
Subject: delete changes
Two things:
1. I'd rather only do the utimes() when AFS_MOUNTPOINTS is defined, so
that people who are compiling delete on non-AFS systems won't lose
the modification time unnecessarily.
However, I don't know if (a) we compile with AFS_MOUNTPOINTS
defined on systems that are accessing AFS through the AFS->NFS
translator (and I doubt it), or (b) if renaming a file through the
AFS->NFS translator will update its ctime.
If the answers to (a) and (b) are both "no," then I guess we'll
have to do the utimes() all the time, or, at the very least, put
another symbol in the Imakefile so that people can toggle the
behavior.
Please look into it and let me know.
2. The man page should really be updated.
Below is a patch that only enables the utimes() call it
AFS_MOUNTPOINTS is defined, and updates the man page as well.
jik
*** /tmp/,RCSt1002991 Fri Jul 26 18:37:52 1991
--- delete.c Fri Jul 26 18:26:19 1991
***************
*** 15,20 ****
--- 15,23 ----
#endif
#include <sys/types.h>
+ #ifdef AFS_MOUNTPOINTS
+ #include <sys/time.h>
+ #endif
#include <stdio.h>
#ifdef POSIX
#include <dirent.h>
***************
*** 78,83 ****
--- 81,89 ----
int force, interactive, recursive, noop, verbose, filesonly, directoriesonly;
int emulate_rm, linked_to_rm, linked_to_rmdir;
extern int errno;
+ #ifdef AFS_MOUNTPOINTS
+ struct timeval tvp[2];
+ #endif
main(argc, argv)
int argc;
***************
*** 90,96 ****
whoami = lastpart(argv[0]);
initialize_del_error_table();
!
force = interactive = recursive = noop = verbose = filesonly =
directoriesonly = emulate_rm = linked_to_rm = linked_to_rmdir = 0;
--- 96,107 ----
whoami = lastpart(argv[0]);
initialize_del_error_table();
!
! #ifdef AFS_MOUNTPOINTS
! gettimeofday(&tvp[0], (struct timezone *)0);
! bcopy(&tvp[0], &tvp[1], sizeof(struct timeval));
! #endif
!
force = interactive = recursive = noop = verbose = filesonly =
directoriesonly = emulate_rm = linked_to_rm = linked_to_rmdir = 0;
***************
*** 503,508 ****
--- 514,536 ----
else {
if (verbose)
fprintf(stderr, "%s: %s removed\n", whoami, filename);
+ #ifdef AFS_MOUNTPOINTS
+ /*
+ * Normally, expunge uses the ctime to determine how long
+ * ago a file was deleted (since the ctime is normally
+ * updated when a file is renamed). However, in AFS,
+ * renaming a file does not change the ctime, mtime OR
+ * atime, so we have to use utimes to force the change.
+ * This unfortunately causes the loss of the real mtime, but
+ * there's nothing we can do about that, if we want expunge
+ * to be able to do the right thing.
+ *
+ * Don't bother checking for errors, because we can't do
+ * anything about them anyway, and in any case, this isn't a
+ * *really* important operation.
+ */
+ utimes(buf, tvp);
+ #endif
return 0;
}
}
*** /tmp/,RCSt1002996 Fri Jul 26 18:38:03 1991
--- man1/delete.1 Fri Jul 26 18:34:29 1991
***************
*** 131,136 ****
--- 131,147 ----
file names, even if they start with a dash.
.SH "SEE ALSO"
undelete(1), purge(1), expunge(1), lsdel(1), rm(1), rmdir(1)
+ .SH BUGS
+ .I Delete
+ may change the access and modification times on a file when it is
+ deleted. This is necessary in order to guarantee that
+ .I expunge
+ will work correctly in
+ .I AFS
+ filesystems. You might say that this is a bug in
+ .IR AFS ,
+ rather than a bug in
+ .IR delete .
.SH AUTHOR
Jonathan I. Kamens (MIT-Project Athena)
.SH COPYRIGHT
