Bugzilla – Bug 110
pkgadd broken in Preview 1
Last modified: 2008-02-13 09:49:15 UTC
You need to log in before you can comment on or make changes to this bug.
pkgadd/pkginstall fails It's looking like a pkgadd bug, or more accurately, a bug in /usr/sadm/install/bin/pkginstall which is run as a pkgadd subprocess. It gets a SIGSEGV just after it has checked for the existence of the /opt/tarantella directory: 5627: getuid() = 0 [0] 5627: door_info(4, 0x08045AA0) = 0 5627: door_call(4, 0x08045B08) = 0 5627: statvfs64("/opt/tarantella", 0xFEF8F4A8) = 0 5627: Incurred fault #6, FLTBOUNDS %pc = 0xFEF59C4C 5627: siginfo: SIGSEGV SEGV_MAPERR addr=0x00000098 5627: Received signal #11, SIGSEGV [default] 5627: siginfo: SIGSEGV SEGV_MAPERR addr=0x00000098 The same thing happens regardless of whether or not /opt/tarantella exists prior to running pkgadd. Sun Secure Global Desktp install successfully on other builds (75a, 75b etc.)
Here's another backtrace when pkgadd was run. core 'core' of 731: /usr/sadm/install/bin/pkginstall -G -O parent-zone-name=global -O pare fef59c4c putcvfpfile (a0ab428, 80a2008) + 8f0 08069f62 output (80a2008, a0ab428, 80a5dd0) + 42 08069751 do_new_ent (80a2008, a0ab428, 0) + 1dd 08068f9f pkgdbmerg (80a1fb0, 80a2008, 80a55c8, 3c) + 22f 080629b3 sortmap (809c7ec, 80a5510, 80a1fb0, 80a2008, 0) + 137 08060860 main (a, 8047afc, 8047b28) + 1ab8 080586a2 _start (a, 8047c0c, 8047c2d, 8047c30, 8047c33, 8047c4b) + 7a
We'd like to get a fix for this for next week for the first repo update; this is preventing people from adding drivers they need. Is this a bug in the INSTALL/latest/SUNWpkgcmds* packages? - Bart
Hmm, a couple of possibilities: - one of the late changes in libtransfer has left us with a stray or missing file or link in the installed root (Moinak doesn't seem to yet have a bugzilla account so I'll send this to him separately) - there was a putback to the install gate late Tuesday which fixed lack of full 64-bit support in all the pkg* tools; those were sync'ed into the Slim gate at some point. But according to the timestamp on that putback (10:10 PM Pacific) it was after the packages on indiana-build.central:/export/slim_stage/INSTALL/latest were built (9:09 PM Mountain), and the slim gate logs show that they were brought over into it a few hours later (Wed Oct 31 08:12:02 2007 GMT). So that seems double unlikely. - we're actually running into a problem due to lack of full 64-bit support and the putback above not being included. ZFS pool sizes have been the impetus for these fixes in the packaging tools, but I doubt that's the case here. Really, though, I don't think we actually needed to deliver these packages out of the Slim gate, as the project hasn't changed the packaging code that I can think of. One simple (though somewhat time-consuming) test would be to build a repo then CD image using the 75a WOS versions of SUNWpkgcmds*. If we still get the behavior, then that eliminates the second possibility and libtransfer issues seem much more likely. It's my guess that's where the problem lies.
*** Bug 119 has been marked as a duplicate of this bug. ***
It appears that point #2 from Dave's comment holds. As a quick check I downloaded the packaging tools sources from http://www.opensolaris.org/os/downloads/install/ and built them. I was successfully able to perform a pkgadd after copying the new libpkg.so.1 to my Indiana test laptop. However now there is a new problem. pkgrm complains about missing libwsreg.so.1. I will try to replace bits on my test machine from the build 75a SUNWpkgcmds* packages and try out once more.
I've verified that the WOS 75a pkgadd works in a freshly built and installed image, so the issue is with the packages, not the transfer service. pkgrm, however, fails as Moinak noted with the missing SUNWwsr2, which is not included in the image. Reassigning back to myself to deal with updating the cluster definition to include it. We need to test with WOS 76 versions of the packages to ensure something hasn't regressed independently.
Created an attachment (id=28) [details] Patch for cluster update
Pushed to pkg/gate.
*** Bug 152 has been marked as a duplicate of this bug. ***
Fixed in Indiana Preview 2.