Date: Wed, 17 Apr 2013 15:56:03 -0700 From: Jeremy Chadwick <jdc@koitsu.org> To: Brooks Davis <brooks@FreeBSD.org> Cc: svn-src-stable@FreeBSD.org, svn-src-all@FreeBSD.org, src-committers@FreeBSD.org, svn-src-stable-9@FreeBSD.org Subject: Re: svn commit: r249549 - in stable/9/sys: amd64/conf i386/conf Message-ID: <20130417225603.GA13720@icarus.home.lan> In-Reply-To: <20130417194706.GA30583@lor.one-eyed-alien.net> References: <201304161609.r3GG9SID009937@svn.freebsd.org> <20130416161919.GA80626@icarus.home.lan> <20130417125433.GC30222@caravan.chchile.org> <20130417193538.GB9331@icarus.home.lan> <20130417194706.GA30583@lor.one-eyed-alien.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Apr 17, 2013 at 02:47:06PM -0500, Brooks Davis wrote: > On Wed, Apr 17, 2013 at 12:35:38PM -0700, Jeremy Chadwick wrote: > > On Wed, Apr 17, 2013 at 02:54:33PM +0200, Jeremie Le Hen wrote: > > > Hi Jeremy, > > > > > > On Tue, Apr 16, 2013 at 09:19:19AM -0700, Jeremy Chadwick wrote: > > > > > > > > Now that this has been enabled by default, I should warn folks of a > > > > caveat that I found in the buildworld/buildkernel framework. It's > > > > easiest to explain like this: > > > > > > > > 1. Install FreeBSD 9.x, svn checkout of stable/9, etc... > > > > 2. Add WITHOUT_CDDL=true to /etc/src.conf > > > > 3. Rebuild + install kernel/world per src/Makefile procedure > > > > 4. Remove WITHOUT_CDDL=true from /etc/src.conf > > > > 5. rm -fr /usr/obj/* > > > > 6. Rebuild world > > > > 7. Rebuild kernel -- fails, stating "ctfconvert: not found". > > > > > > > > For whatever reason the buildkernel bits make the assumption that > > > > ctfconvert exists on the system (presumably in $PATH or possibly a > > > > hard-coded), when ideally it should try to use the recently-built > > > > version in /usr/obj first. > > > > > > I've tested this is a freshly installed 9.1-RELEASE jail and I haven't > > > been biten by the bug you describe. > > > > > > ctfconvert(1) seems to be installed by default in 9.1-RELEASE, this is > > > probably there problem didn't occur. I can easily verify this in the > > > jail: > > > > > > % root@test9:/usr/src # ls -l /usr/bin/ctfconvert /usr/bin/vi /usr/bin/tail > > > % -r-xr-xr-x 1 root wheel 371536 Dec 4 09:33 /usr/bin/ctfconvert > > > % -r-xr-xr-x 1 root wheel 19848 Apr 17 06:28 /usr/bin/tail > > > % -r-xr-xr-x 6 root wheel 346432 Apr 17 06:28 /usr/bin/vi > > > > > > > > > Do you have a theory about why you've got the problem while I haven't? > > > FYI, it seems 9.0-RELEASE also has ctfconvert(1): > > > http://svnweb.freebsd.org/base/release/9.0.0/cddl/usr.bin/ctfconvert/ > > > > > > My guess is tha this might happen if you don't have /usr/bin/ctfconvert. > > > I've just removed it and trying to build kernel again. > > > > I will spend some time to figure out exactly how to reproduce this. > > > > Going from recent memory (~2 weeks ago), I encountered it on my VPS box > > (which does run ntpd, just FYI): > > > > 1. Initially installed with 9.1-RELEASE, > > 2. Upgraded to stable/9 (using svn), > > 3. WITHOUT_CDDL=true and WITHOUT_ZFS=true added to /etc/src.conf > > 4. world/kernel rebuilt/reinstalled/etc. (this includes make delete-old, > > as per instructions in src/Makefile -- which would delete > > /usr/bin/ctfconvert) > > 5. Fast forward many months > > 6. Removed WITHOUT_CDDL=true from src.conf > > 7. Encountered the above issue ("ctfconvert: not found") during > > buildkernel > > 8. Rebuilt kernel again -- same error > > 9. Removed WITHOUT_ZFS=true from src.conf > > 10. Rebuilt kernel again -- worked > > > > This could mean WITHOUT_ZFS=true has some bearing on this situation, but > > I don't see how/why, as WITHOUT_CDDL is supposed to be the "trigger" for > > ctf* utilities. I did poke around the Makefiles and framework a bit > > but didn't have any epiphanies. > > > > Like I said -- I'll try to reproduce the exact scenario. > > Looking at Makefile.inc1 around line 1164 (on HEAD), ctfconvert and > cftmerge are bootstrap tools only when they don't exist at all on the > host. The code there should be expanded to bootstrap for cases where > the installed ones are known to be broken (virtually all prior versions > given recent fixes) as well as when they aren't present on the host > system. I'm able to reproduce the issue I speak of with 100% reliability. Jeremie, I'm not sure why you're not able to reproduce this, because I can do so reliably/consistently. Below are my notes. This was done on a VMware Workstation VM instance, and I took VM snapshots along the way, so I can "roll back" to almost any phase/step listed below (in case someone wants me to verify what's on the filesystem or use "script" to save a log somewhere or run "make -DA" or something along those lines). 1. Installed 9.1-RELEASE. 2. Added ntpdate_enable and ntpd_enable to /etc/rc.conf, and ran /etc/rc.d/ntpdate start ; /etc/rc.d/ntpd start. 3. Added following to /etc/src.conf: WITHOUT_CDDL=true WITHOUT_CLANG=true WITHOUT_INET6=true WITHOUT_IPFILTER=true WITHOUT_LIB32=true WITHOUT_KERBEROS=true WITHOUT_PAM_SUPPORT=true WITHOUT_SENDMAIL=true WITHOUT_ZFS=true WITH_OPENSSH_NONE_CIPHER=true 4. Installed subversion via ports/pkg_add -r and pulled down a fresh copy of stable/9 (src, r249561) and ports. 5. Verified that /usr/bin/ctfconvert and other CTF utilities exist on system (obviously part of 9.1-RELEASE). 6. Followed instructions in src/Makefile. Shown here: # 1. `cd /usr/src' (or to the directory containing your source tree). # 2. `make buildworld' # 3. `make buildkernel KERNCONF=YOUR_KERNEL_HERE' (default is GENERIC). # 4. `make installkernel KERNCONF=YOUR_KERNEL_HERE' (default is GENERIC). # [steps 3. & 4. can be combined by using the "kernel" target] # 5. `reboot' (in single user mode: boot -s from the loader prompt). # 6. `mergemaster -p' # 7. `make installworld' # 8. `make delete-old' # 9. `mergemaster' (you may wish to use -i, along with -U or -F). # 10. `reboot' # 11. `make delete-old-libs' (in case no 3rd party program uses them anymore) Had to do "mergemaster -p" prior to installkernel, but Brooks knows about this and it's not relevant to the issue. 7. Verified the "make delete-old" phase deleted /usr/bin/ctfconvert and other CTF utilities; this is normal. 8. Removed WITHOUT_CDDL=true from /etc/src.conf 9. rm -fr /usr/obj/* && cd /usr/src && make -j2 buildworld. 10. Verification of everything at this point -- ctf* utilities are clearly in /usr/obj, but are not in /usr/bin (because of delete-old): root@testbox:/usr/src # find /usr/obj -name "ctf*" -type f -perm 0755 -ls 1783448 728 -rwxr-xr-x 1 root wheel 371738 Apr 17 15:29 /usr/obj/usr/src/cddl/usr.bin/ctfconvert/ctfconvert 1783453 64 -rwxr-xr-x 1 root wheel 28706 Apr 17 15:29 /usr/obj/usr/src/cddl/usr.bin/ctfdump/ctfdump 1783472 168 -rwxr-xr-x 1 root wheel 82711 Apr 17 15:29 /usr/obj/usr/src/cddl/usr.bin/ctfmerge/ctfmerge root@testbox:/usr/src # find /usr/bin -name "ctf*" -type f -perm 0755 -ls root@testbox:/usr/src # 11. make -j2 buildkernel fails: *** [aac_disk.o] Error code 127 cc -c -O2 -frename-registers -pipe -fno-strict-aliasing -std=c99 -g -Wall -Wredundant-decls -Wnested-externs -Wstrict-prototypes -Wmissing-prototypes -Wpointer-arith -Winline -Wcast-qual -Wundef -Wno-pointer-sign -fformat-extensions -Wmissing-include-dirs -fdiagnostics-show-option -nostdinc -I. -I/usr/src/sys -I/usr/src/sys/contrib/altq -D_KERNEL -DHAVE_KERNEL_OPTION_HEADERS -include opt_global.h -fno-common -finline-limit=8000 --param inline-unit-growth=100 --param large-function-growth=1000 -fno-omit-frame-pointer -mcmodel=kernel -mno-red-zone -mno-mmx -mno-sse -msoft-float -fno-asynchronous-unwind-tables -ffreestanding -fstack-protector -Werror /usr/src/sys/cam/cam.c ctfconvert -L VERSION -g cam.o ctfconvert: not found *** [cam.o] Error code 127 ctfconvert -L VERSION -g aac_cam.o ctfconvert: not found *** [aac_cam.o] Error code 127 2 errors *** [all] Error code 2 1 error *** [modules-all] Error code 2 2 errors *** [buildkernel] Error code 2 1 error *** [buildkernel] Error code 2 1 error I'll be doing the following to see where exactly the failure happens since as we know parallel make causes confusing output sometimes, rm -fr /usr/obj/* && make -j2 buildworld && make buildkernel I doubt the parallelism has anything to do with the issue, however -- it seems very clear cut to me that the issue is that buildkernel assumes ctfconvert is in one's $PATH, which is not true if you have an active system with WITHOUT_CDDL=true which you're trying to move *to* have CDDL. -- | Jeremy Chadwick jdc@koitsu.org | | UNIX Systems Administrator http://jdc.koitsu.org/ | | Mountain View, CA, US | | Making life hard for others since 1977. PGP 4BD6C0CB |
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130417225603.GA13720>