| Message ID | 1452008181.13361.328.camel@citrix.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Tue, Jan 05, 2016 at 03:36:21PM +0000, Ian Campbell wrote: > On Tue, 2016-01-05 at 14:37 +0000, Ian Campbell wrote: > > > > which on the basis of this discussion I wasn't expecting. I didn't see this > > new file on i686 or ARM*. > > > > My baseline is from the last time I committed, which would be last year, so > > maybe something other than my current batch of patches has caused this. > > > > I'm going to drop this one for now and (hopefully) get the rest of the > > batch squared away. Afterwards I'll take another look (with a new baseline > > filelist), but if someone can explain it in the meantime that would be > > super. > > So with a fresh basline I still see: > > --- ../FILE_LIST.BASE.staging.x86_64 2016-01-05 14:50:32.000000000 +0000 > +++ ../FILE_LIST.staging.x86_64 2016-01-05 15:11:15.000000000 +0000 > @@ -6,6 +6,7 @@ > dist/install/boot/xen-4.7-unstable.gz > dist/install/boot/xen-4.gz > dist/install/boot/xen.gz > +dist/install/boot/xenpolicy-4.7-unstable > dist/install/etc > dist/install/etc/bash_completion.d > dist/install/etc/bash_completion.d/xl.sh > @@ -386,6 +387,12 @@ > dist/install/usr/local/lib/xen/libexec > dist/install/usr/local/lib/xen/libexec/qemu-bridge-helper > dist/install/usr/local/sbin > +dist/install/usr/local/sbin/flask-get-bool > +dist/install/usr/local/sbin/flask-getenforce > +dist/install/usr/local/sbin/flask-label-pci > +dist/install/usr/local/sbin/flask-loadpolicy > +dist/install/usr/local/sbin/flask-set-bool > +dist/install/usr/local/sbin/flask-setenforce > dist/install/usr/local/sbin/gdbsx > dist/install/usr/local/sbin/gtracestat > dist/install/usr/local/sbin/gtraceview > *** FILES DIFFER *** > > On i686 and ARM* I only see the (expected) second hunk. > > I think the i686 case is explainable by the lack of a hypervisor build > there, but I'm unsure why ARM* and x86_64 should differ in this regard. > > config/Tools.mk is y only on x86_64, not on the others, which obviously > explains things, but the question is why only on x86_64 (I presume this has > always been the case and it was previously masked, but I've not checked). > > Ah, OK, I misread > > AX_ARG_DEFAULT_ENABLE([xsmpolicy], [Disable XSM policy compilation]) > > as being default disable, actually the default is "enabled iff checkpolicy > is installed" and it happens to be that it is only installed in my x86_64 > build env. > > So, in the end I think Wei was correct and this change will now, in some > circumstances, end up installing a /boot/xenpolicy-*. > I don't think it is related to this patch. I see an xenpoilcy file without this patch applied. As you said it only depends on availability of checkpolicy (part of generic SELinux utils, not the ones we build). That said, let me try to answer the following question. > So the question is do we mind that? > We might or might not. See below. I once submitted a patch to grub that look into /boot and generate XSM entries if there is policy file. The patch is not yet merged though. Since there is no way at the moment to tell if xen.gz has flask enabled, my not yet upstreamed patch only matches the version number of xen.gz and xenpolicy. Installing xenpolicy when xen.gz is not flaks-capable will make grub generate an XSM entry nonetheless, which makes no sense. Of course all the above is based on the theory that my grub patch is going to be upstreamed. Things have changed since I first submitted that patch. Doug's Kconfig work is good. With .config installed in suitable location we can make grub grep for flask information in config, hence avoiding generating wrong entries. I think this is better solution as we don't need to use version number to match xen.gz and xenpolicy. If we go down this route we don't mind having random xenpolicy lying around in /boot. We just need to reach an agreement on how to proceed. I would vote for the second solution. Wei. > Ian.
On Tue, 2016-01-05 at 16:13 +0000, Wei Liu wrote: > On Tue, Jan 05, 2016 at 03:36:21PM +0000, Ian Campbell wrote: > > On Tue, 2016-01-05 at 14:37 +0000, Ian Campbell wrote: > > > > > > which on the basis of this discussion I wasn't expecting. I didn't > > > see this > > > new file on i686 or ARM*. > > > > > > My baseline is from the last time I committed, which would be last > > > year, so > > > maybe something other than my current batch of patches has caused > > > this. > > > > > > I'm going to drop this one for now and (hopefully) get the rest of > > > the > > > batch squared away. Afterwards I'll take another look (with a new > > > baseline > > > filelist), but if someone can explain it in the meantime that would > > > be > > > super. > > > > So with a fresh basline I still see: > > > > --- ../FILE_LIST.BASE.staging.x86_64 2016-01-05 14:50:32.000000000 > > +0000 > > +++ ../FILE_LIST.staging.x86_64 2016-01-05 15:11:15.000000000 +0000 > > @@ -6,6 +6,7 @@ > > dist/install/boot/xen-4.7-unstable.gz > > dist/install/boot/xen-4.gz > > dist/install/boot/xen.gz > > +dist/install/boot/xenpolicy-4.7-unstable > > dist/install/etc > > dist/install/etc/bash_completion.d > > dist/install/etc/bash_completion.d/xl.sh > > @@ -386,6 +387,12 @@ > > dist/install/usr/local/lib/xen/libexec > > dist/install/usr/local/lib/xen/libexec/qemu-bridge-helper > > dist/install/usr/local/sbin > > +dist/install/usr/local/sbin/flask-get-bool > > +dist/install/usr/local/sbin/flask-getenforce > > +dist/install/usr/local/sbin/flask-label-pci > > +dist/install/usr/local/sbin/flask-loadpolicy > > +dist/install/usr/local/sbin/flask-set-bool > > +dist/install/usr/local/sbin/flask-setenforce > > dist/install/usr/local/sbin/gdbsx > > dist/install/usr/local/sbin/gtracestat > > dist/install/usr/local/sbin/gtraceview > > *** FILES DIFFER *** > > > > On i686 and ARM* I only see the (expected) second hunk. > > > > I think the i686 case is explainable by the lack of a hypervisor build > > there, but I'm unsure why ARM* and x86_64 should differ in this regard. > > > > config/Tools.mk is y only on x86_64, not on the others, which obviously > > explains things, but the question is why only on x86_64 (I presume this > > has > > always been the case and it was previously masked, but I've not > > checked). > > > > Ah, OK, I misread > > > > AX_ARG_DEFAULT_ENABLE([xsmpolicy], [Disable XSM policy compilation]) > > > > as being default disable, actually the default is "enabled iff > > checkpolicy > > is installed" and it happens to be that it is only installed in my > > x86_64 > > build env. > > > > So, in the end I think Wei was correct and this change will now, in > > some > > circumstances, end up installing a /boot/xenpolicy-*. > > > > I don't think it is related to this patch. I see an xenpoilcy file > without this patch applied. With XSM disabled? > As you said it only depends on availability > of checkpolicy (part of generic SELinux utils, not the ones we build). But then why does this file only show up for me with this patch applied? You initially objected to this patch because you thought it would add this file, but it seems like you have always had it. Is the answer just that you only just found that you always had it? > > That said, let me try to answer the following question. > > > So the question is do we mind that? > > > > We might or might not. See below. > > I once submitted a patch to grub that look into /boot and generate XSM > entries if there is policy file. The patch is not yet merged though. > > Since there is no way at the moment to tell if xen.gz has flask enabled, > my not yet upstreamed patch only matches the version number of xen.gz and > xenpolicy. Installing xenpolicy when xen.gz is not flaks-capable will > make grub generate an XSM entry nonetheless, which makes no sense. Indeed. > Of course all the above is based on the theory that my grub patch is > going to be upstreamed. > > Things have changed since I first submitted that patch. Doug's Kconfig > work is good. With .config installed in suitable location we can make > grub grep for flask information in config, hence avoiding generating > wrong entries. I think this is better solution as we don't need to use > version number to match xen.gz and xenpolicy. If we go down this route > we don't mind having random xenpolicy lying around in /boot. > We just need to reach an agreement on how to proceed. I would vote for > the second solution. Which is what? This patch as is? (and what is the first proposition?) Ian.
On 1/5/16 10:13 AM, Wei Liu wrote: > On Tue, Jan 05, 2016 at 03:36:21PM +0000, Ian Campbell wrote: >> On Tue, 2016-01-05 at 14:37 +0000, Ian Campbell wrote: >>> >>> which on the basis of this discussion I wasn't expecting. I didn't see this >>> new file on i686 or ARM*. >>> >>> My baseline is from the last time I committed, which would be last year, so >>> maybe something other than my current batch of patches has caused this. >>> >>> I'm going to drop this one for now and (hopefully) get the rest of the >>> batch squared away. Afterwards I'll take another look (with a new baseline >>> filelist), but if someone can explain it in the meantime that would be >>> super. >> >> So with a fresh basline I still see: >> >> --- ../FILE_LIST.BASE.staging.x86_64 2016-01-05 14:50:32.000000000 +0000 >> +++ ../FILE_LIST.staging.x86_64 2016-01-05 15:11:15.000000000 +0000 >> @@ -6,6 +6,7 @@ >> dist/install/boot/xen-4.7-unstable.gz >> dist/install/boot/xen-4.gz >> dist/install/boot/xen.gz >> +dist/install/boot/xenpolicy-4.7-unstable >> dist/install/etc >> dist/install/etc/bash_completion.d >> dist/install/etc/bash_completion.d/xl.sh >> @@ -386,6 +387,12 @@ >> dist/install/usr/local/lib/xen/libexec >> dist/install/usr/local/lib/xen/libexec/qemu-bridge-helper >> dist/install/usr/local/sbin >> +dist/install/usr/local/sbin/flask-get-bool >> +dist/install/usr/local/sbin/flask-getenforce >> +dist/install/usr/local/sbin/flask-label-pci >> +dist/install/usr/local/sbin/flask-loadpolicy >> +dist/install/usr/local/sbin/flask-set-bool >> +dist/install/usr/local/sbin/flask-setenforce >> dist/install/usr/local/sbin/gdbsx >> dist/install/usr/local/sbin/gtracestat >> dist/install/usr/local/sbin/gtraceview >> *** FILES DIFFER *** >> >> On i686 and ARM* I only see the (expected) second hunk. >> >> I think the i686 case is explainable by the lack of a hypervisor build >> there, but I'm unsure why ARM* and x86_64 should differ in this regard. >> >> config/Tools.mk is y only on x86_64, not on the others, which obviously >> explains things, but the question is why only on x86_64 (I presume this has >> always been the case and it was previously masked, but I've not checked). >> >> Ah, OK, I misread >> >> AX_ARG_DEFAULT_ENABLE([xsmpolicy], [Disable XSM policy compilation]) >> >> as being default disable, actually the default is "enabled iff checkpolicy >> is installed" and it happens to be that it is only installed in my x86_64 >> build env. >> >> So, in the end I think Wei was correct and this change will now, in some >> circumstances, end up installing a /boot/xenpolicy-*. >> > > I don't think it is related to this patch. I see an xenpoilcy file > without this patch applied. As you said it only depends on availability > of checkpolicy (part of generic SELinux utils, not the ones we build). > > That said, let me try to answer the following question. > >> So the question is do we mind that? >> > > We might or might not. See below. > > I once submitted a patch to grub that look into /boot and generate XSM > entries if there is policy file. The patch is not yet merged though. > > Since there is no way at the moment to tell if xen.gz has flask enabled, > my not yet upstreamed patch only matches the version number of xen.gz and > xenpolicy. Installing xenpolicy when xen.gz is not flaks-capable will > make grub generate an XSM entry nonetheless, which makes no sense. > > Of course all the above is based on the theory that my grub patch is > going to be upstreamed. > > Things have changed since I first submitted that patch. Doug's Kconfig > work is good. With .config installed in suitable location we can make > grub grep for flask information in config, hence avoiding generating > wrong entries. I think this is better solution as we don't need to use > version number to match xen.gz and xenpolicy. If we go down this route > we don't mind having random xenpolicy lying around in /boot. So I submitted a patch to put the .config in /boot to have the ability to do this. I figured it needed to be in /boot because that's how the existing 20_linux_xen but there's some disagreement on the location. If we can resolve that I'll happily update the 20_linux_xen patch as well. http://lists.xenproject.org/archives/html/xen-devel/2015-12/msg02369.html > > We just need to reach an agreement on how to proceed. I would vote for > the second solution. > > Wei. > > >> Ian.
On Tue, 2016-01-05 at 10:34 -0600, Doug Goldstein wrote: > On 1/5/16 10:13 AM, Wei Liu wrote: > > On Tue, Jan 05, 2016 at 03:36:21PM +0000, Ian Campbell wrote: > > > On Tue, 2016-01-05 at 14:37 +0000, Ian Campbell wrote: > > > > > > > > which on the basis of this discussion I wasn't expecting. I didn't > > > > see this > > > > new file on i686 or ARM*. > > > > > > > > My baseline is from the last time I committed, which would be last > > > > year, so > > > > maybe something other than my current batch of patches has caused > > > > this. > > > > > > > > I'm going to drop this one for now and (hopefully) get the rest of > > > > the > > > > batch squared away. Afterwards I'll take another look (with a new > > > > baseline > > > > filelist), but if someone can explain it in the meantime that would > > > > be > > > > super. > > > > > > So with a fresh basline I still see: > > > > > > --- ../FILE_LIST.BASE.staging.x86_64 2016-01-05 14:50:32.000000000 > > > +0000 > > > +++ ../FILE_LIST.staging.x86_64 2016-01-05 15:11:15.000000000 +0000 > > > @@ -6,6 +6,7 @@ > > > dist/install/boot/xen-4.7-unstable.gz > > > dist/install/boot/xen-4.gz > > > dist/install/boot/xen.gz > > > +dist/install/boot/xenpolicy-4.7-unstable > > > dist/install/etc > > > dist/install/etc/bash_completion.d > > > dist/install/etc/bash_completion.d/xl.sh > > > @@ -386,6 +387,12 @@ > > > dist/install/usr/local/lib/xen/libexec > > > dist/install/usr/local/lib/xen/libexec/qemu-bridge-helper > > > dist/install/usr/local/sbin > > > +dist/install/usr/local/sbin/flask-get-bool > > > +dist/install/usr/local/sbin/flask-getenforce > > > +dist/install/usr/local/sbin/flask-label-pci > > > +dist/install/usr/local/sbin/flask-loadpolicy > > > +dist/install/usr/local/sbin/flask-set-bool > > > +dist/install/usr/local/sbin/flask-setenforce > > > dist/install/usr/local/sbin/gdbsx > > > dist/install/usr/local/sbin/gtracestat > > > dist/install/usr/local/sbin/gtraceview > > > *** FILES DIFFER *** > > > > > > On i686 and ARM* I only see the (expected) second hunk. > > > > > > I think the i686 case is explainable by the lack of a hypervisor > > > build > > > there, but I'm unsure why ARM* and x86_64 should differ in this > > > regard. > > > > > > config/Tools.mk is y only on x86_64, not on the others, which > > > obviously > > > explains things, but the question is why only on x86_64 (I presume > > > this has > > > always been the case and it was previously masked, but I've not > > > checked). > > > > > > Ah, OK, I misread > > > > > > AX_ARG_DEFAULT_ENABLE([xsmpolicy], [Disable XSM policy compilation]) > > > > > > as being default disable, actually the default is "enabled iff > > > checkpolicy > > > is installed" and it happens to be that it is only installed in my > > > x86_64 > > > build env. > > > > > > So, in the end I think Wei was correct and this change will now, in > > > some > > > circumstances, end up installing a /boot/xenpolicy-*. > > > > > > > I don't think it is related to this patch. I see an xenpoilcy file > > without this patch applied. As you said it only depends on availability > > of checkpolicy (part of generic SELinux utils, not the ones we build). > > > > That said, let me try to answer the following question. > > > > > So the question is do we mind that? > > > > > > > We might or might not. See below. > > > > I once submitted a patch to grub that look into /boot and generate XSM > > entries if there is policy file. The patch is not yet merged though. > > > > Since there is no way at the moment to tell if xen.gz has flask > > enabled, > > my not yet upstreamed patch only matches the version number of xen.gz > > and > > xenpolicy. Installing xenpolicy when xen.gz is not flaks-capable will > > make grub generate an XSM entry nonetheless, which makes no sense. > > > > Of course all the above is based on the theory that my grub patch is > > going to be upstreamed. > > > > Things have changed since I first submitted that patch. Doug's Kconfig > > work is good. With .config installed in suitable location we can make > > grub grep for flask information in config, hence avoiding generating > > wrong entries. I think this is better solution as we don't need to use > > version number to match xen.gz and xenpolicy. If we go down this route > > we don't mind having random xenpolicy lying around in /boot. > > So I submitted a patch to put the .config in /boot to have the ability > to do this. I figured it needed to be in /boot because that's how the > existing 20_linux_xen but there's some disagreement on the location. If > we can resolve that I'll happily update the 20_linux_xen patch as well. > > http://lists.xenproject.org/archives/html/xen-devel/2015-12/msg02369.html It would be worth referring that thread back to this one too. Ian.
On Tue, Jan 05, 2016 at 04:24:19PM +0000, Ian Campbell wrote: > On Tue, 2016-01-05 at 16:13 +0000, Wei Liu wrote: > > On Tue, Jan 05, 2016 at 03:36:21PM +0000, Ian Campbell wrote: > > > On Tue, 2016-01-05 at 14:37 +0000, Ian Campbell wrote: > > > > > > > > which on the basis of this discussion I wasn't expecting. I didn't > > > > see this > > > > new file on i686 or ARM*. > > > > > > > > My baseline is from the last time I committed, which would be last > > > > year, so > > > > maybe something other than my current batch of patches has caused > > > > this. > > > > > > > > I'm going to drop this one for now and (hopefully) get the rest of > > > > the > > > > batch squared away. Afterwards I'll take another look (with a new > > > > baseline > > > > filelist), but if someone can explain it in the meantime that would > > > > be > > > > super. > > > > > > So with a fresh basline I still see: > > > > > > --- ../FILE_LIST.BASE.staging.x86_64 2016-01-05 14:50:32.000000000 > > > +0000 > > > +++ ../FILE_LIST.staging.x86_64 2016-01-05 15:11:15.000000000 +0000 > > > @@ -6,6 +6,7 @@ > > > dist/install/boot/xen-4.7-unstable.gz > > > dist/install/boot/xen-4.gz > > > dist/install/boot/xen.gz > > > +dist/install/boot/xenpolicy-4.7-unstable > > > dist/install/etc > > > dist/install/etc/bash_completion.d > > > dist/install/etc/bash_completion.d/xl.sh > > > @@ -386,6 +387,12 @@ > > > dist/install/usr/local/lib/xen/libexec > > > dist/install/usr/local/lib/xen/libexec/qemu-bridge-helper > > > dist/install/usr/local/sbin > > > +dist/install/usr/local/sbin/flask-get-bool > > > +dist/install/usr/local/sbin/flask-getenforce > > > +dist/install/usr/local/sbin/flask-label-pci > > > +dist/install/usr/local/sbin/flask-loadpolicy > > > +dist/install/usr/local/sbin/flask-set-bool > > > +dist/install/usr/local/sbin/flask-setenforce > > > dist/install/usr/local/sbin/gdbsx > > > dist/install/usr/local/sbin/gtracestat > > > dist/install/usr/local/sbin/gtraceview > > > *** FILES DIFFER *** > > > > > > On i686 and ARM* I only see the (expected) second hunk. > > > > > > I think the i686 case is explainable by the lack of a hypervisor build > > > there, but I'm unsure why ARM* and x86_64 should differ in this regard. > > > > > > config/Tools.mk is y only on x86_64, not on the others, which obviously > > > explains things, but the question is why only on x86_64 (I presume this > > > has > > > always been the case and it was previously masked, but I've not > > > checked). > > > > > > Ah, OK, I misread > > > > > > AX_ARG_DEFAULT_ENABLE([xsmpolicy], [Disable XSM policy compilation]) > > > > > > as being default disable, actually the default is "enabled iff > > > checkpolicy > > > is installed" and it happens to be that it is only installed in my > > > x86_64 > > > build env. > > > > > > So, in the end I think Wei was correct and this change will now, in > > > some > > > circumstances, end up installing a /boot/xenpolicy-*. > > > > > > > I don't think it is related to this patch. I see an xenpoilcy file > > without this patch applied. > > With XSM disabled? > > > As you said it only depends on availability > > of checkpolicy (part of generic SELinux utils, not the ones we build). > > But then why does this file only show up for me with this patch applied? > > You initially objected to this patch because you thought it would add this > file, but it seems like you have always had it. Is the answer just that you > only just found that you always had it? > Hmm... After I make distclean, things changed. So to be clear: without this patch applied, I don't have xenpolicy file even if checkpolicy is available. This patch does alter the behaviour somehow. I'm in the middle of rebasing one patch series, so I haven't looked into all the details. > > > > That said, let me try to answer the following question. > > > > > So the question is do we mind that? > > > > > > > We might or might not. See below. > > > > I once submitted a patch to grub that look into /boot and generate XSM > > entries if there is policy file. The patch is not yet merged though. > > > > Since there is no way at the moment to tell if xen.gz has flask enabled, > > my not yet upstreamed patch only matches the version number of xen.gz and > > xenpolicy. Installing xenpolicy when xen.gz is not flaks-capable will > > make grub generate an XSM entry nonetheless, which makes no sense. > > Indeed. > > > Of course all the above is based on the theory that my grub patch is > > going to be upstreamed. > > > > Things have changed since I first submitted that patch. Doug's Kconfig > > work is good. With .config installed in suitable location we can make > > grub grep for flask information in config, hence avoiding generating > > wrong entries. I think this is better solution as we don't need to use > > version number to match xen.gz and xenpolicy. If we go down this route > > we don't mind having random xenpolicy lying around in /boot. > > > We just need to reach an agreement on how to proceed. I would vote for > > the second solution. > > Which is what? This patch as is? (and what is the first proposition?) > That was referring to grub generating XSM entries. First solution is my not yet upstream patch; second is to make gurb grep .config for flask information. Wei. > Ian.
On 1/5/16 10:42 AM, Wei Liu wrote: > On Tue, Jan 05, 2016 at 04:24:19PM +0000, Ian Campbell wrote: >> On Tue, 2016-01-05 at 16:13 +0000, Wei Liu wrote: >>> On Tue, Jan 05, 2016 at 03:36:21PM +0000, Ian Campbell wrote: >>>> On Tue, 2016-01-05 at 14:37 +0000, Ian Campbell wrote: >>>>> >>>>> which on the basis of this discussion I wasn't expecting. I didn't >>>>> see this >>>>> new file on i686 or ARM*. >>>>> >>>>> My baseline is from the last time I committed, which would be last >>>>> year, so >>>>> maybe something other than my current batch of patches has caused >>>>> this. >>>>> >>>>> I'm going to drop this one for now and (hopefully) get the rest of >>>>> the >>>>> batch squared away. Afterwards I'll take another look (with a new >>>>> baseline >>>>> filelist), but if someone can explain it in the meantime that would >>>>> be >>>>> super. >>>> >>>> So with a fresh basline I still see: >>>> >>>> --- ../FILE_LIST.BASE.staging.x86_64 2016-01-05 14:50:32.000000000 >>>> +0000 >>>> +++ ../FILE_LIST.staging.x86_64 2016-01-05 15:11:15.000000000 +0000 >>>> @@ -6,6 +6,7 @@ >>>> dist/install/boot/xen-4.7-unstable.gz >>>> dist/install/boot/xen-4.gz >>>> dist/install/boot/xen.gz >>>> +dist/install/boot/xenpolicy-4.7-unstable >>>> dist/install/etc >>>> dist/install/etc/bash_completion.d >>>> dist/install/etc/bash_completion.d/xl.sh >>>> @@ -386,6 +387,12 @@ >>>> dist/install/usr/local/lib/xen/libexec >>>> dist/install/usr/local/lib/xen/libexec/qemu-bridge-helper >>>> dist/install/usr/local/sbin >>>> +dist/install/usr/local/sbin/flask-get-bool >>>> +dist/install/usr/local/sbin/flask-getenforce >>>> +dist/install/usr/local/sbin/flask-label-pci >>>> +dist/install/usr/local/sbin/flask-loadpolicy >>>> +dist/install/usr/local/sbin/flask-set-bool >>>> +dist/install/usr/local/sbin/flask-setenforce >>>> dist/install/usr/local/sbin/gdbsx >>>> dist/install/usr/local/sbin/gtracestat >>>> dist/install/usr/local/sbin/gtraceview >>>> *** FILES DIFFER *** >>>> >>>> On i686 and ARM* I only see the (expected) second hunk. >>>> >>>> I think the i686 case is explainable by the lack of a hypervisor build >>>> there, but I'm unsure why ARM* and x86_64 should differ in this regard. >>>> >>>> config/Tools.mk is y only on x86_64, not on the others, which obviously >>>> explains things, but the question is why only on x86_64 (I presume this >>>> has >>>> always been the case and it was previously masked, but I've not >>>> checked). >>>> >>>> Ah, OK, I misread >>>> >>>> AX_ARG_DEFAULT_ENABLE([xsmpolicy], [Disable XSM policy compilation]) >>>> >>>> as being default disable, actually the default is "enabled iff >>>> checkpolicy >>>> is installed" and it happens to be that it is only installed in my >>>> x86_64 >>>> build env. >>>> >>>> So, in the end I think Wei was correct and this change will now, in >>>> some >>>> circumstances, end up installing a /boot/xenpolicy-*. >>>> >>> >>> I don't think it is related to this patch. I see an xenpoilcy file >>> without this patch applied. >> >> With XSM disabled? >> >>> As you said it only depends on availability >>> of checkpolicy (part of generic SELinux utils, not the ones we build). >> >> But then why does this file only show up for me with this patch applied? >> >> You initially objected to this patch because you thought it would add this >> file, but it seems like you have always had it. Is the answer just that you >> only just found that you always had it? >> > > Hmm... After I make distclean, things changed. > > So to be clear: without this patch applied, I don't have xenpolicy file > even if checkpolicy is available. This patch does alter the behaviour > somehow. > > I'm in the middle of rebasing one patch series, so I haven't looked > into all the details. > >>> >>> That said, let me try to answer the following question. >>> >>>> So the question is do we mind that? >>>> >>> >>> We might or might not. See below. >>> >>> I once submitted a patch to grub that look into /boot and generate XSM >>> entries if there is policy file. The patch is not yet merged though. >>> >>> Since there is no way at the moment to tell if xen.gz has flask enabled, >>> my not yet upstreamed patch only matches the version number of xen.gz and >>> xenpolicy. Installing xenpolicy when xen.gz is not flaks-capable will >>> make grub generate an XSM entry nonetheless, which makes no sense. >> >> Indeed. >> >>> Of course all the above is based on the theory that my grub patch is >>> going to be upstreamed. >>> >>> Things have changed since I first submitted that patch. Doug's Kconfig >>> work is good. With .config installed in suitable location we can make >>> grub grep for flask information in config, hence avoiding generating >>> wrong entries. I think this is better solution as we don't need to use >>> version number to match xen.gz and xenpolicy. If we go down this route >>> we don't mind having random xenpolicy lying around in /boot. >> >>> We just need to reach an agreement on how to proceed. I would vote for >>> the second solution. >> >> Which is what? This patch as is? (and what is the first proposition?) >> > > That was referring to grub generating XSM entries. First solution is my > not yet upstream patch; second is to make gurb grep .config for flask > information. > > Wei. > >> Ian. Ok so I'm at a loss what steps I need to take. I've submitted patches to put the config in /boot so that this check can be made but there's a disagreement if that's even necessary or not. Do I need to supply a patch to make --disable-xsmpolicy the default so that this change doesn't generate the policy by default? The point of this patch is to compile the necessarily bits always which will help shake out bugs earlier. If we don't want the policy file to be installed then we should use the proper setting for that and not the fact that the utility isn't being compiled.
On Fri, Jan 08, 2016 at 12:49:07PM -0600, Doug Goldstein wrote: [...] > Ok so I'm at a loss what steps I need to take. I've submitted patches to > put the config in /boot so that this check can be made but there's a > disagreement if that's even necessary or not. > That's a bit unfortunate. :-( But if I'm not mistaken that's orthogonal to this problem, right? That's one more step down the road regarding grub integration. > Do I need to supply a patch to make --disable-xsmpolicy the default so > that this change doesn't generate the policy by default? The point of > this patch is to compile the necessarily bits always which will help > shake out bugs earlier. If we don't want the policy file to be installed > then we should use the proper setting for that and not the fact that the > utility isn't being compiled. > I think one solution would be to modify flask/Makefile to guard policy compilation against (FLASK_ENABLE && FLASK_POLICY). What do you think? Admittedly I haven't followed closely all the KConfig work so I might be talking nonsense. Ian and Ian? Wei. > -- > Doug Goldstein >
On 1/11/16 9:19 AM, Wei Liu wrote: > On Fri, Jan 08, 2016 at 12:49:07PM -0600, Doug Goldstein wrote: > [...] >> Ok so I'm at a loss what steps I need to take. I've submitted patches to >> put the config in /boot so that this check can be made but there's a >> disagreement if that's even necessary or not. >> > > That's a bit unfortunate. :-( > > But if I'm not mistaken that's orthogonal to this problem, right? That's > one more step down the road regarding grub integration. > >> Do I need to supply a patch to make --disable-xsmpolicy the default so >> that this change doesn't generate the policy by default? The point of >> this patch is to compile the necessarily bits always which will help >> shake out bugs earlier. If we don't want the policy file to be installed >> then we should use the proper setting for that and not the fact that the >> utility isn't being compiled. >> > > I think one solution would be to modify flask/Makefile to guard policy > compilation against (FLASK_ENABLE && FLASK_POLICY). > > What do you think? Admittedly I haven't followed closely all the KConfig > work so I might be talking nonsense. > > Ian and Ian? > > Wei. Wei (and Ian and Ian and Daniel), There's already a guard against compiling the policy in the tools/ directory's configure script called --{enable,disable}-xsmpolicy What I could do is disable it by default because it is currently enabled by default. I honestly think that would be an improvement because we would compile all the source code (causing us to shake bugs out earlier) but only generate the policy when the user explicitly requests it. Right now the policy is made whenever the utilities are compiled. Let me know if that sounds appealing to you. Thanks.
On Mon, Jan 11, 2016 at 11:10:35AM -0600, Doug Goldstein wrote: > On 1/11/16 9:19 AM, Wei Liu wrote: > > On Fri, Jan 08, 2016 at 12:49:07PM -0600, Doug Goldstein wrote: > > [...] > >> Ok so I'm at a loss what steps I need to take. I've submitted patches to > >> put the config in /boot so that this check can be made but there's a > >> disagreement if that's even necessary or not. > >> > > > > That's a bit unfortunate. :-( > > > > But if I'm not mistaken that's orthogonal to this problem, right? That's > > one more step down the road regarding grub integration. > > > >> Do I need to supply a patch to make --disable-xsmpolicy the default so > >> that this change doesn't generate the policy by default? The point of > >> this patch is to compile the necessarily bits always which will help > >> shake out bugs earlier. If we don't want the policy file to be installed > >> then we should use the proper setting for that and not the fact that the > >> utility isn't being compiled. > >> > > > > I think one solution would be to modify flask/Makefile to guard policy > > compilation against (FLASK_ENABLE && FLASK_POLICY). > > > > What do you think? Admittedly I haven't followed closely all the KConfig > > work so I might be talking nonsense. > > > > Ian and Ian? > > > > Wei. > > Wei (and Ian and Ian and Daniel), > > There's already a guard against compiling the policy in the tools/ > directory's configure script called --{enable,disable}-xsmpolicy What I > could do is disable it by default because it is currently enabled by > default. > > I honestly think that would be an improvement because we would compile > all the source code (causing us to shake bugs out earlier) but only > generate the policy when the user explicitly requests it. Right now the > policy is made whenever the utilities are compiled. > > Let me know if that sounds appealing to you. > Fine by me. I don't really have a strong opinion at this point. My original concern that the installed xenpolicy file interferes with grub was based on the assumption that we only had version numbers as indicator to match hypervisor binary and xenpolicy file. But now since I think there is better way to generate grub entry I don't think my objection based on the (bad) assumption to this patch is relevant anymore. Wei. > Thanks. > -- > Doug Goldstein >
--- ../FILE_LIST.BASE.staging.x86_64 2016-01-05 14:50:32.000000000 +0000 +++ ../FILE_LIST.staging.x86_64 2016-01-05 15:11:15.000000000 +0000 @@ -6,6 +6,7 @@ dist/install/boot/xen-4.7-unstable.gz dist/install/boot/xen-4.gz dist/install/boot/xen.gz +dist/install/boot/xenpolicy-4.7-unstable dist/install/etc dist/install/etc/bash_completion.d dist/install/etc/bash_completion.d/xl.sh @@ -386,6 +387,12 @@ dist/install/usr/local/lib/xen/libexec dist/install/usr/local/lib/xen/libexec/qemu-bridge-helper dist/install/usr/local/sbin +dist/install/usr/local/sbin/flask-get-bool +dist/install/usr/local/sbin/flask-getenforce +dist/install/usr/local/sbin/flask-label-pci +dist/install/usr/local/sbin/flask-loadpolicy +dist/install/usr/local/sbin/flask-set-bool +dist/install/usr/local/sbin/flask-setenforce dist/install/usr/local/sbin/gdbsx dist/install/usr/local/sbin/gtracestat dist/install/usr/local/sbin/gtraceview