nes@prcrs.UUCP (Nancy Shoemaker) (07/26/88)
Here's * the README from a patch tape dated 6/8 (from Massachusetts CSC) * a list of diffs from the 4/26 patch tape * the checksums from the 6/15 tape ====================================================================== README Run the install_patch script to install these patches. The script checks the sccs id module and updates the binaries for anyone found out of rev. You MUST be root to run this script. uipc_msg.o msg.h ---------------- panic: bad rmfree Can occur when processes using message queues swap. File system problems -------------------- gfs/{gfs_bio.o gfs_gnodeops.o gfs_syscalls.o} nfs/{nfs_server.o nfs_vnodeops.o vnodeops_gfs.o nfs_subr.o nfs_gfsops.o} ufs/{ufs_gnodeops.o ufs_gnode.o ufs_syscalls.o} sys/{kern_sig.o kern_acct.o} These files do not map one-to-one with bugs. The bugs that these files fix are: 1. Change to file generation number increments on NFS server. Would cause "NFS server: stale fhandle" message on server (should have only caused problems when accessing through multiple links to same file via NFS, and one link removed). 2. Fix for mode of named pipe created over NFS. 3. Fix for NFS client gnode race condition. Can cause gfs_unlock: unlocked gnode panic. 4. Better tempname algorithm when unlinking open files over NFS. Extremely unlikely to be seen; results in stale fhandle error. 5. Various fs panics when accounting changes state. (km_free: bad dealloc when shutting system down) 6. bhalt() changes. It could conceivably be writing out read-ahead blocks to disk at shutdown and panics now. 7. Access is not allowed over NFS for a filesystem that has not been explicitly exported. (Not true without these patches). However, changing the exports file while a system is running is not enough to prohibit access from a filesystem that is currently exported. The filesystem must also be unmounted and remounted on the server (tough for root). After that, any attempt to access that filesystem via NFS will be refused with ESTALE and an "unexported" message is printed on the server. 8. A df on a remote NFS filesystem now has the same effect as doing a df on the server - not true before, a df on the server was necessary for the stats, as seen by client, to be updated. 9. The utimes() system call now has the proper semantics over NFS - the owner of a file can change the times even if he does not have write permission. 10. Better NFS server fhandle error messages. tmscp.o tmscp.h tmscpreg.h -------------------------- Fixes all kinds of panics, panic: bad rmfree in the tmscp driver panic: bad rmfree in other drivers when freeing uba maps. The old driver could allow the writing of K mem in improper areas. uerf.bin -------- The file uerf.bin contains erroneuos data. UERF uses this file to do bit-to-text translations. Error log reports can contain bad data. Some of the bad information is obvious, such as VERY large or negative numbers for MSCP disk cylinder or head addresses. fsck ---- This replaces the fsck command that was distributed with ULTRIX-32 V2.2. It fixes panic: free: freeing free block. kern_synch.o ------------ Processes were running slowly because the process priority never decayed if the process had taken a signal. gfs_descrip.o ------------- Systems crashed periodically during execution of closef() routine with segmentation fault. dlogind ------- Fixes wtmp corruption problem. Two processes could open the wtmp file for writing at the same time. This would cause corruption of wtmp. * This is only if you have DECnet installed. lat_scl1.o lat_slot.o lat_vc.o ------------------------------ System would panic with segmentation fault when picking up an empty slot. Also, users would be dropped from LAT. s5make ------ Fixed to handle archives properly. ss.o ---- Fixes problem with MicroVAX 2000 console port and port 2. If a modem was attached to port 2 and you logged off the console, port 2 would be disconnected. (See V2.2 Release Notes, p. 2-15 for more information.) dmb.o ----- Fixes line hangs ... uucico ------ Fixes problem with uucico hanging. ====================================================================== Files that differ: README install_patch kern_acct.o msg.h -- sccs header only difference tmscp.h -- sccs header only difference tmscpreg.h -- sccs header only difference Only in new version: dlogind dmb.o gfs_descrip.o lat_scl1.o lat_slot.o lat_vc.o s5make ss.o uucico ====================================================================== 36622 5 README 20287 63 dlogind 61033 13 dmb.o 12738 69 fsck 39670 8 gfs_bio.o 07811 6 gfs_descrip.o 44916 8 gfs_gnodeops.o 34330 15 gfs_syscalls.o 20789 19 install_patch 53611 4 kern_acct.o 42261 9 kern_sig.o 30686 7 kern_synch.o 56705 6 lat_scl1.o 54454 4 lat_slot.o 53865 7 lat_vc.o 19108 5 msg.h 02419 9 nfs_gfsops.o 28292 13 nfs_server.o 51267 9 nfs_subr.o 18513 14 nfs_vnodeops.o 03470 42 s5make 42788 16 ss.o 21862 12 tmscp.h 38230 28 tmscp.o 10244 5 tmscpreg.h 39948 139 uerf.bin 36638 7 ufs_gnode.o 06472 6 ufs_gnodeops.o 09817 6 ufs_syscalls.o 17696 6 uipc_msg.o 38949 125 uucico 20147 7 vnodeops_gfs.o
animal@ernie.UUCP (Alan R. Silverman) (08/17/88)
Warning! I received my Ultrix 2.3 update and was warned by DEC Cust. Support to NOT install it if I installed the 2.2 bugfix. They couldn't reboot and fsck was helpless. We will load back to 2.2, then upgrade to 2.3 . We have been having incredible NFS stale fhandle messages and sometimes the VAX8250 gets so locked up with this problem, we have to physically halt the diskless node causing the problem. ( We run a VAX8250 exporting a large filesystem for 35 home directoriesw to 12 Suns ) Anyone else having problems like this? Anyone running 2.3 Ultrix?
davids@iscuva.ISCS.COM (David Schmidt) (08/26/88)
Reply-to: davids@iscuva.ISCS.COM (David Schmidt) In article <54080@felix.UUCP> animal@ernie.UUCP (Alan R. Silverman) writes: >Warning! I received my Ultrix 2.3 update and was warned by DEC Cust. >Support to NOT install it if I installed the 2.2 bugfix. > >[...] > >We will load back to 2.2, then upgrade to 2.3 . After seeing a warning here about installing 2.3 on a patched 2.2 system I called DEC Customer Support and heard the same thing. However, I remembered looking at the install script on the patch tape and seeing that it saved the old version 2.2 files. I was able to move the old files (named filename.org0) to their original names and rebuild my 2.2 kernel. I then moved the two other programs (fsck and uerf?) to their original locations, loaded the patch tape and everything worked fine. There is no need to load a backup tape of your original 2.2 system. Hope that helps! -- David Schmidt UUCP: davids@iscuva.ISCS.COM ISC Systems Corporation (uunet!iscuva!davids) East 22425 Appleway Phone: +1 509 927-5479 Liberty Lake, WA 99019