Alex Franco : Fix mode +st on TMPDIR when creating it
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 8 lines · On Tue, Sep 01, 2015 at 10:45:22AM +0000, [email protected] wrote:
Module: openembedded-core.git Branch: master Commit: 8236d57439640a185c0226312cd4903a3ce2f53b URL: http://git.openembedded.org/?p=openembedded-core.git&a=commit;h=8236d57439640a185c0226312cd4903a3ce2f53bAuthor: Alex Franco <[email protected]> Date: Fri Aug 28 17:34:04 2015 -0500
Quoted 7 lines
Fix mode +st on TMPDIR when creating it
A sanity check fails when TMPDIR has setuid, setgid set. It was proposed to fix this on TMPDIR creation instead of failing with a sanity test only. This patch handles removal of those special bits (and additonally, the sticky bit) from BUILDDIR and TMPDIR when these directories are created.
Can you explain why it's needed to remove them?
I have tmp-glibc mounted as tmpfs:
drwxrwxrwt 4 root jenkins 80 Sep 1 19:02 tmp-glibc
and since this change all bitbake executions are failing with:
NOTE: Started PRServer with DBfile: /home/jenkins/oe/world/shr-core/cache/prserv.sqlite3, IP: 127.0.0.1, PORT: 43102, PID: 9220
ERROR: Execution of event handler 'check_sanity_eventhandler' failed
Traceback (most recent call last):
File "check_sanity_eventhandler(e)", line 6, in check_sanity_eventhandler(e=<bb.event.SanityCheck object at 0x3c76ad0>)
File "sanity.bbclass", line 34, in check_sanity(sanity_data=<bb.data_smart.DataSmart object at 0x3c83890>)
File "sanity.bbclass", line 157, in check_sanity_everybuild(status=<SanityStatus object at 0x3c83850>, d=<bb.data_smart.DataSmart object at 0x3c83890>)
OSError: [Errno 1] Operation not permitted: '/home/jenkins/oe/world/shr-core/tmp-glibc'
I can change the permissions manually, but I would like to know why is this needed and checked every time the bitbake is executed.
Regards,
Quoted 53 lines
[YOCTO #7669]
Signed-off-by Alex Franco <[email protected]>Signed-off-by Richard Purdie <[email protected]>---
meta/classes/sanity.bbclass | 5 +++++ scripts/oe-setup-builddir | 1 + 2 files changed, 6 insertions(+)meta/classes/sanity.bbclass +5
-0@@ -686,6 +686,7 @@ def check_sanity_version_change(status, d): status.addresult(check_not_nfs(tmpdir, "TMPDIR")) def check_sanity_everybuild(status, d): + import os, stat # Sanity tests which test the users environment so need to run at each build (or are so cheap # it makes sense to always run them.@@ -839,6 +840,10 @@ def check_sanity_everybuild(status, d): status.addresult("Error, TMPDIR has changed location. You need to either move it back to %s or rebuild\n" % saved_tmpdir) else: bb.utils.mkdirhier(tmpdir) + # Remove setuid, setgid and sticky bits from TMPDIR + os.chmod(tmpdir, os.stat(tmpdir).st_mode & ~ stat.S_ISUID) + os.chmod(tmpdir, os.stat(tmpdir).st_mode & ~ stat.S_ISGID) + os.chmod(tmpdir, os.stat(tmpdir).st_mode & ~ stat.S_ISVTX) with open(checkfile, "w") as f: f.write(tmpdir)scripts/oe-setup-builddir +1
-0@@ -24,6 +24,7 @@ if [ -z "$BUILDDIR" ]; then fi mkdir -p "$BUILDDIR/conf" +chmod -R -st "$BUILDDIR" if [ ! -d "$BUILDDIR" ]; then echo >&2 "Error: The builddir ($BUILDDIR) does not exist!"Signature · 5 lines
-- _______________________________________________ Openembedded-commits mailing list [email protected] http://lists.openembedded.org/mailman/listinfo/openembedded-commits
Signature · 1 line
Martin 'JaMa' Jansa jabber: [email protected]
Hi Martin,
I gather the original reason (for the sanity test, as well as these changes) is that it is better for everything under BUILDDIR - and certainly TMPDIR to be owned by the user running bitbake and have standard permissions, the sticky bit itself would prevent a user running bitbake, even with +rwx permissions, from making changes to items it might need to manipulate, if those are not owned by this user. Similarly setuid and setgid might propagate user/group ownerships different from the user bitbake is running as, into BUILDDIR and TMPDIR (and items therein).
For more info take a look at: https://bugzilla.yoctoproject.org/show_bug.cgi?id=7669
Perhaps there are more pressing reasons, perhaps Christopher Larson can comment a bit on this.
Alex Franco
Quoted 88 lines · Martin Jansa
Quoted 8 lines · On Tue, Sep 01, 2015 at 10:45:22AM +0000, [email protected] wrote:
Module: openembedded-core.git Branch: master Commit: 8236d57439640a185c0226312cd4903a3ce2f53b URL: http://git.openembedded.org/?p=openembedded-core.git&a=commit;h=8236d57439640a185c0226312cd4903a3ce2f53bAuthor: Alex Franco <[email protected]> Date: Fri Aug 28 17:34:04 2015 -0500Quoted 7 lines
Fix mode +st on TMPDIR when creating it
A sanity check fails when TMPDIR has setuid, setgid set. It was proposed to fix this on TMPDIR creation instead of failing with a sanity test only. This patch handles removal of those special bits (and additonally, the sticky bit) from BUILDDIR and TMPDIR when these directories are created.
Can you explain why it's needed to remove them?
I have tmp-glibc mounted as tmpfs:
drwxrwxrwt 4 root jenkins 80 Sep 1 19:02 tmp-glibcand since this change all bitbake executions are failing with:
NOTE: Started PRServer with DBfile:
/home/jenkins/oe/world/shr-core/cache/prserv.sqlite3, IP: 127.0.0.1, PORT: 43102, PID: 9220
ERROR: Execution of event handler 'check_sanity_eventhandler' failed
Traceback (most recent call last):File "check_sanity_eventhandler(e)", line 6, in check_sanity_eventhandler(e=<bb.event.SanityCheck object at 0x3c76ad0>) File "sanity.bbclass", line 34, in check_sanity(sanity_data=<bb.data_smart.DataSmart object at 0x3c83890>) File "sanity.bbclass", line 157, in check_sanity_everybuild(status=<SanityStatus object at 0x3c83850>, d=<bb.data_smart.DataSmart object at 0x3c83890>)OSError: [Errno 1] Operation not permitted: '
/home/jenkins/oe/world/shr-core/tmp-glibc'I can change the permissions manually, but I would like to know why is this needed and checked every time the bitbake is executed.
Regards,
Quoted 52 lines
[YOCTO #7669]
Signed-off-by Alex Franco <[email protected]>Signed-off-by Richard Purdie <[email protected]>---
meta/classes/sanity.bbclass | 5 +++++ scripts/oe-setup-builddir | 1 + 2 files changed, 6 insertions(+)meta/classes/sanity.bbclass +5
-0@@ -686,6 +686,7 @@ def check_sanity_version_change(status, d): status.addresult(check_not_nfs(tmpdir, "TMPDIR")) def check_sanity_everybuild(status, d): + import os, stat # Sanity tests which test the users environment so need to run at each build (or are so cheap # it makes sense to always run them.@@ -839,6 +840,10 @@ def check_sanity_everybuild(status, d): status.addresult("Error, TMPDIR has changed location. You need to either move it back to %s or rebuild\n" % saved_tmpdir) else: bb.utils.mkdirhier(tmpdir) + # Remove setuid, setgid and sticky bits from TMPDIR + os.chmod(tmpdir, os.stat(tmpdir).st_mode & ~ stat.S_ISUID) + os.chmod(tmpdir, os.stat(tmpdir).st_mode & ~ stat.S_ISGID) + os.chmod(tmpdir, os.stat(tmpdir).st_mode & ~ stat.S_ISVTX) with open(checkfile, "w") as f: f.write(tmpdir)scripts/oe-setup-builddir +1
-0@@ -24,6 +24,7 @@ if [ -z "$BUILDDIR" ]; then fi mkdir -p "$BUILDDIR/conf" +chmod -R -st "$BUILDDIR" if [ ! -d "$BUILDDIR" ]; then echo >&2 "Error: The builddir ($BUILDDIR) does not exist!"Signature · 5 lines
-- _______________________________________________ Openembedded-commits mailing list [email protected] http://lists.openembedded.org/mailman/listinfo/openembedded-commits
Quoted 9 lines · Alex Franco
I gather the original reason (for the sanity test, as well as these changes) is that it is better for everything under BUILDDIR - and certainly TMPDIR to be owned by the user running bitbake and have standard permissions, the sticky bit itself would prevent a user running bitbake, even with +rwx permissions, from making changes to items it might need to manipulate, if those are not owned by this user. Similarly setuid and setgid might propagate user/group ownerships different from the user bitbake is running as, into BUILDDIR and TMPDIR (and items therein).
Quoted 6 lines
For more info take a look at:
https://bugzilla.yoctoproject.org/show_bug.cgi?id=7669Perhaps there are more pressing reasons, perhaps Christopher Larson can comment a bit on this.
Indeed, the g+s can flow down into the files put into the packages, and then into the rootfs. I ran into a this problem quite a bit, since my project areas are in a g+s area to keep the files accessible to the group.
Signature · 5 lines
Christopher Larson clarson at kergoth dot com Founder - BitBake, OpenEmbedded, OpenZaurus Maintainer - Tslib Senior Software Engineer, Mentor Graphics
Quoted 17 lines · Christopher Larson
On Tue, Sep 1, 2015 at 4:11 PM, Alex Franco <[email protected]> wrote:
I gather the original reason (for the sanity test, as well as these changes) is that it is better for everything under BUILDDIR - and certainly TMPDIR to be owned by the user running bitbake and have standard permissions, the sticky bit itself would prevent a user running bitbake, even with +rwx permissions, from making changes to items it might need to manipulate, if those are not owned by this user. Similarly setuid and setgid might propagate user/group ownerships different from the user bitbake is running as, into BUILDDIR and TMPDIR (and items therein).For more info take a look at:
https://bugzilla.yoctoproject.org/show_bug.cgi?id=7669
Quoted 7 lines
Perhaps there are more pressing reasons, perhaps Christopher Larson can comment a bit on this.
Indeed, the g+s can flow down into the files put into the packages, and then into the rootfs. I ran into a this problem quite a bit, since my project areas are in a g+s area to keep the files accessible to the group.
There is actually another problem here. The -R means if you do this to an existing build directory, it takes around 10 mins on the machine I just rebooted with a cold cache. We need to lose the -R there.
Cheers,
Richard
Thanks, I have sent a patch after removing the -R and testing a bit
Alex Franco
Quoted 34 lines · Richard Purdie
Quoted 16 lines · Christopher Larson
On Tue, Sep 1, 2015 at 4:11 PM, Alex Franco <[email protected]> wrote:
I gather the original reason (for the sanity test, as well as these changes) is that it is better for everything under BUILDDIR - and certainly TMPDIR to be owned by the user running bitbake and have standard permissions, the sticky bit itself would prevent a user running bitbake, even with +rwx permissions, from making changes to items it might need to manipulate, if those are not owned by this user. Similarly setuid and setgid might propagate user/group ownerships different from the user bitbake is running as, into BUILDDIR and TMPDIR (and items therein).For more info take a look at:
https://bugzilla.yoctoproject.org/show_bug.cgi?id=7669Quoted 7 lines
Perhaps there are more pressing reasons, perhaps Christopher Larson can comment a bit on this.
Indeed, the g+s can flow down into the files put into the packages, and then into the rootfs. I ran into a this problem quite a bit, since my project areas are in a g+s area to keep the files accessible to the group.
There is actually another problem here. The -R means if you do this to an existing build directory, it takes around 10 mins on the machine I just rebooted with a cold cache. We need to lose the -R there.
Cheers,
Richard
Quoted 11 lines · Martin Jansa
Quoted 10 lines · On Tue, Sep 01, 2015 at 10:45:22AM +0000, [email protected] wrote:
Module: openembedded-core.git Branch: master Commit: 8236d57439640a185c0226312cd4903a3ce2f53b URL: http://git.openembedded.org/?p=openembedded-core.git&a=commit;h=8236d57439640a185c0226312cd4903a3ce2f53bAuthor: Alex Franco <[email protected]> Date: Fri Aug 28 17:34:04 2015 -0500Fix mode +st on TMPDIR when creating it
Quoted 7 lines
Quoted 5 lines
A sanity check fails when TMPDIR has setuid, setgid set. It was proposed to fix this on TMPDIR creation instead of failing with a sanity test only. This patch handles removal of those special bits (and additonally, the sticky bit) from BUILDDIR and TMPDIR when these directories are created.
Can you explain why it's needed to remove them?
At least in theory the sticky bit (t) can end up influencing the target package file ownership. In the interests of build determinism we really need TMPDIR to start in a consistent way with a consistent set of permissions.
Put another way, I don't really want to have to add new sanity tests that test if builds work the same if TMPDIR is +t compared to a dir that isn't set that way.
So if you could tweak your autobuilder setup to lose the sticky bit on that directory, I think that would be preferable to adding more corner cases to the build environment.
If everyone had setups like this, that would change my view but it doesn't seem to eb the ase.
Cheers,
Richard