openssh: remove RRECOMMENDS to rng-tools for sshd package
PATCH 1/1 openssh: remove RRECOMMENDS to rng-tools for sshd package
It appears that rngd is not needed as of linux-5.6 and later[1] and should not be installed by default since the purpose of rngd is to provide additional trusted sources of entropy.
We did some testing on real hardware, the result seems to support that we no longer need rngd by default on kernel v5.6 and later.
Testing result as below:
1. observing the crng init stage.
the "random: crng init done" always available before fs being mounted.
2. generating random number without rngd.
testing command: dd if=/dev/random of=/dev/null status=progress
on Marvell CN96xx RDB board, speed almost 20.4 MB/s without block on NXP i.mx6q board, speed almost 31.9 MB/s without block on qemu x86-64, speed almost 2.6MB/s without block
3. using rngtest command without rngd
testing command: rngtest -c 1000 </dev/random
on Marvell CN96xx RDB board:
rngtest: input channel speed: (min=4.340; avg=135.364; max=146.719)Mibits/s
rngtest: FIPS tests speed: (min=8.197; avg=69.020; max=72.800)Mibits/s
rngtest: Program run time: 418771 microsecondson NXP i.mx6q board:
rngtest: input channel speed: (min=96.820; avg=326.769; max=340.598)Mibits/s
rngtest: FIPS tests speed: (min=15.090; avg=37.543; max=40.324)Mibits/s
rngtest: Program run time: 570229 microsecondson qemu x86-64:
rngtest: input channel speed: (min=37.769; avg=101.136; max=136.239)Mibits/s
rngtest: FIPS tests speed: (min=10.288; avg=30.682; max=40.155)Mibits/s
rngtest: Program run time: 836800 microseconds4. observing sshd service.
using "systemctl disable rng-tools" disable service and reboot system. system boot up normal, sshd service also start in normal time without block.
Reference:
[1] https://github.com/torvalds/linux/commit/30c08efec8884fb106b8e57094baa51bb4c44e32
--- meta/recipes-connectivity/openssh/openssh_9.1p1.bb | 9 +-------- 1 file changed, 1 insertion(+), 8 deletions(-)
meta/recipes-connectivity/openssh/openssh_9.1p1.bb +1 -8
@@ -52,15 +52,12 @@ SYSTEMD_SERVICE:${PN}-sshd = "sshd.socket" inherit autotools-brokensep ptest -PACKAGECONFIG ??= "rng-tools" +PACKAGECONFIG ??= "" PACKAGECONFIG[kerberos] = "--with-kerberos5,--without-kerberos5,krb5" PACKAGECONFIG[ldns] = "--with-ldns,--without-ldns,ldns" PACKAGECONFIG[libedit] = "--with-libedit,--without-libedit,libedit" PACKAGECONFIG[manpages] = "--with-mantype=man,--with-mantype=cat" -# Add RRECOMMENDS to rng-tools for sshd package -PACKAGECONFIG[rng-tools] = "" - EXTRA_AUTORECONF += "--exclude=aclocal" # login path is hardcoded in sshd
@@ -160,10 +157,6 @@ FILES:${PN}-keygen = "${bindir}/ssh-keygen" RDEPENDS:${PN} += "${PN}-scp ${PN}-ssh ${PN}-sshd ${PN}-keygen ${PN}-sftp-server" RDEPENDS:${PN}-sshd += "${PN}-keygen ${@bb.utils.contains('DISTRO_FEATURES', 'pam', 'pam-plugin-keyinit pam-plugin-loginuid', '', d)}" -RRECOMMENDS:${PN}-sshd:append:class-target = "\ - ${@bb.utils.filter('PACKAGECONFIG', 'rng-tools', d)} \ -" - # gdb would make attach-ptrace test pass rather than skip but not worth the build dependencies RDEPENDS:${PN}-ptest += "${PN}-sftp ${PN}-misc ${PN}-sftp-server make sed sudo coreutils"
Signature · 1 line
2.34.1
Quoted 4 lines · On Fri., Dec. 9, 2022, 02:05 Xiangyu Chen, <[email protected]>…
It appears that rngd is not needed as of linux-5.6 and later[1] and should not be installed by default since the purpose of rngd is to provide additional trusted sources of entropy.
Quoted 8 lines
We did some testing on real hardware, the result seems to support that we no longer need rngd by default on kernel v5.6 and later.
Testing result as below:
1. observing the crng init stage.
the "random: crng init done" always available before fs being mounted.
Super.
Quoted 26 lines
2. generating random number without rngd.
testing command: dd if=/dev/random of=/dev/null status=progress
on Marvell CN96xx RDB board, speed almost 20.4 MB/s without block on NXP i.mx6q board, speed almost 31.9 MB/s without block on qemu x86-64, speed almost 2.6MB/s without block3. using rngtest command without rngd
testing command: rngtest -c 1000 </dev/random
on Marvell CN96xx RDB board:rngtest: input channel speed: (min=4.340; avg=135.364; max=146.719)Mibits/s rngtest: FIPS tests speed: (min=8.197; avg=69.020;max=72.800)Mibits/s
rngtest: Program run time: 418771 microsecondson NXP i.mx6q board:
rngtest: input channel speed: (min=96.820; avg=326.769; max=340.598)Mibits/s rngtest: FIPS tests speed: (min=15.090; avg=37.543;max=40.324)Mibits/s
rngtest: Program run time: 570229 microsecondson qemu x86-64: rngtest: input channel speed: (min=37.769; avg=101.136; max=136.239)Mibits/s rngtest: FIPS tests speed: (min=10.288; avg=30.682;
Quoted 8 lines
max=40.155)Mibits/s
rngtest: Program run time: 836800 microseconds4. observing sshd service.
using "systemctl disable rng-tools" disable service and reboot system. system boot up normal, sshd service also start in normal time without block.
Nice.
Does openssh have more quantitative tests that can be run?
For the YP autobuilder tests that run in qemu, do we need, or do we already have, virtio-rng or is that also obsolete ( here's an old link from 2015:
https://www.redhat.com/en/blog/red-hat-enterprise-linux-virtual-machines-access-random-numbers-made-easy
).
Maybe the thing to do is to just try this change in master-next as usual and see it we get any timeouts from the YP test suite.
Thanks,
../Randy
Quoted 19 lines
Reference:
[1]
https://github.com/torvalds/linux/commit/30c08efec8884fb106b8e57094baa51bb4c44e32Signed-off-by Xiangyu Chen <[email protected]>--- meta/recipes-connectivity/openssh/openssh_9.1p1.bb | 9 +-------- 1 file changed, 1 insertion(+), 8 deletions(-)meta/recipes-connectivity/openssh/openssh_9.1p1.bb +0
-1@@ -52,15 +52,12 @@ SYSTEMD_SERVICE:${PN}-sshd = "sshd.socket" inherit autotools-brokensep ptest -PACKAGECONFIG ??= "rng-tools"
meta/recipes-connectivity/openssh/openssh_9.1p1.bb SYSTEMD_SERVICE:${PN}-sshd = "sshd.socket" · line 61
Quoted 10 lines
meta/recipes-connectivity/openssh/openssh_9.1p1.bb +1
-3+PACKAGECONFIG ??= "" PACKAGECONFIG[kerberos] = "--with-kerberos5,--without-kerberos5,krb5" PACKAGECONFIG[ldns] = "--with-ldns,--without-ldns,ldns" PACKAGECONFIG[libedit] = "--with-libedit,--without-libedit,libedit" PACKAGECONFIG[manpages] = "--with-mantype=man,--with-mantype=cat" -# Add RRECOMMENDS to rng-tools for sshd package -PACKAGECONFIG[rng-tools] = "" -
EXTRA_AUTORECONF += "--exclude=aclocal"
Quoted 33 lines
meta/recipes-connectivity/openssh/openssh_9.1p1.bb +0
-0# login path is hardcoded in sshd@@ -160,10 +157,6 @@ FILES:${PN}-keygen = "${bindir}/ssh-keygen" RDEPENDS:${PN} += "${PN}-scp ${PN}-ssh ${PN}-sshd ${PN}-keygen${PN}-sftp-server"
RDEPENDS:${PN}-sshd += "${PN}-keygen
${@bb.utils.contains('DISTRO_FEATURES', 'pam', 'pam-plugin-keyinit pam-plugin-loginuid', '', d)}"meta/recipes-connectivity/openssh/openssh_9.1p1.bb +0
-4-RRECOMMENDS:${PN}-sshd:append:class-target = "\ - ${@bb.utils.filter('PACKAGECONFIG', 'rng-tools', d)} \ -" - # gdb would make attach-ptrace test pass rather than skip but not worththe build dependencies
RDEPENDS:${PN}-ptest += "${PN}-sftp ${PN}-misc ${PN}-sftp-server make sed
sudo coreutils"--
2.34.1-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group. View/Reply Online (#174434):
https://lists.openembedded.org/g/openembedded-core/message/174434
Mute This Topic: https://lists.openembedded.org/mt/95556189/953399
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [ [email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
would be good to know some numbers on non-arm/non-x86 systems too.
Quoted 93 lines · Xiangyu Chen
It appears that rngd is not needed as of linux-5.6 and later[1] and should not be installed by default since the purpose of rngd is to provide additional trusted sources of entropy.
We did some testing on real hardware, the result seems to support that we no longer need rngd by default on kernel v5.6 and later.
Testing result as below:
1. observing the crng init stage.
the "random: crng init done" always available before fs being mounted.2. generating random number without rngd.
testing command: dd if=/dev/random of=/dev/null status=progress
on Marvell CN96xx RDB board, speed almost 20.4 MB/s without block on NXP i.mx6q board, speed almost 31.9 MB/s without block on qemu x86-64, speed almost 2.6MB/s without block3. using rngtest command without rngd
testing command: rngtest -c 1000 </dev/random
on Marvell CN96xx RDB board:rngtest: input channel speed: (min=4.340; avg=135.364; max=146.719)Mibits/s rngtest: FIPS tests speed: (min=8.197; avg=69.020; max=72.800)Mibits/s rngtest: Program run time: 418771 microsecondson NXP i.mx6q board: rngtest: input channel speed: (min=96.820; avg=326.769; max=340.598)Mibits/s rngtest: FIPS tests speed: (min=15.090; avg=37.543; max=40.324)Mibits/s rngtest: Program run time: 570229 microseconds on qemu x86-64: rngtest: input channel speed: (min=37.769; avg=101.136; max=136.239)Mibits/s rngtest: FIPS tests speed: (min=10.288; avg=30.682; max=40.155)Mibits/s rngtest: Program run time: 836800 microseconds4. observing sshd service.
using "systemctl disable rng-tools" disable service and reboot system. system boot up normal, sshd service also start in normal time without block.Reference:
[1] https://github.com/torvalds/linux/commit/30c08efec8884fb106b8e57094baa51bb4c44e32Signed-off-by Xiangyu Chen <[email protected]>--- meta/recipes-connectivity/openssh/openssh_9.1p1.bb | 9 +-------- 1 file changed, 1 insertion(+), 8 deletions(-)meta/recipes-connectivity/openssh/openssh_9.1p1.bb +1
-9@@ -52,15 +52,12 @@ SYSTEMD_SERVICE:${PN}-sshd = "sshd.socket" inherit autotools-brokensep ptest -PACKAGECONFIG ??= "rng-tools" +PACKAGECONFIG ??= "" PACKAGECONFIG[kerberos] = "--with-kerberos5,--without-kerberos5,krb5" PACKAGECONFIG[ldns] = "--with-ldns,--without-ldns,ldns" PACKAGECONFIG[libedit] = "--with-libedit,--without-libedit,libedit" PACKAGECONFIG[manpages] = "--with-mantype=man,--with-mantype=cat" -# Add RRECOMMENDS to rng-tools for sshd package -PACKAGECONFIG[rng-tools] = "" - EXTRA_AUTORECONF += "--exclude=aclocal" # login path is hardcoded in sshd@@ -160,10 +157,6 @@ FILES:${PN}-keygen = "${bindir}/ssh-keygen" RDEPENDS:${PN} += "${PN}-scp ${PN}-ssh ${PN}-sshd ${PN}-keygen ${PN}-sftp-server" RDEPENDS:${PN}-sshd += "${PN}-keygen ${@bb.utils.contains('DISTRO_FEATURES', 'pam', 'pam-plugin-keyinit pam-plugin-loginuid', '', d)}" -RRECOMMENDS:${PN}-sshd:append:class-target = "\ - ${@bb.utils.filter('PACKAGECONFIG', 'rng-tools', d)} \ -" - # gdb would make attach-ptrace test pass rather than skip but not worth the build dependencies RDEPENDS:${PN}-ptest += "${PN}-sftp ${PN}-misc ${PN}-sftp-server make sed sudo coreutils" --2.34.1
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#174434): https://lists.openembedded.org/g/openembedded-core/message/174434
Mute This Topic: https://lists.openembedded.org/mt/95556189/1997914
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
In the kernel is the jitter entropy system. It should work on all platforms that have high res timers available. (This is the same mechanism that haveged was using before as well. So no change in the RNG quality, just now built into the kernel itself.)
The only place we've observed an issue with the new approach is on a qemu machine that was moderately loaned. The jitter was apparently too large (our theory) and caused the crng init to happen VERY slowly. (This would not have been resolved using rngd.)
All-in-all, I thought this change had already been made, and think it should go in.
(We tested this on microblaze, and performance with and without haveged was the same. I don't have number, but I expect the same behaviors on any architecture.)
--Mark
Quoted 101 lines · Khem Raj
would be good to know some numbers on non-arm/non-x86 systems too.
Quoted 97 lines · Xiangyu Chen
It appears that rngd is not needed as of linux-5.6 and later[1] and should not be installed by default since the purpose of rngd is to provide additional trusted sources of entropy.
We did some testing on real hardware, the result seems to support that we no longer need rngd by default on kernel v5.6 and later.
Testing result as below:
1. observing the crng init stage.
the "random: crng init done" always available before fs being mounted.2. generating random number without rngd.
testing command: dd if=/dev/random of=/dev/null status=progress
on Marvell CN96xx RDB board, speed almost 20.4 MB/s without block on NXP i.mx6q board, speed almost 31.9 MB/s without block on qemu x86-64, speed almost 2.6MB/s without block3. using rngtest command without rngd
testing command: rngtest -c 1000 </dev/randomon Marvell CN96xx RDB board: rngtest: input channel speed: (min=4.340; avg=135.364; max=146.719)Mibits/s rngtest: FIPS tests speed: (min=8.197; avg=69.020; max=72.800)Mibits/s rngtest: Program run time: 418771 microseconds on NXP i.mx6q board: rngtest: input channel speed: (min=96.820; avg=326.769; max=340.598)Mibits/s rngtest: FIPS tests speed: (min=15.090; avg=37.543; max=40.324)Mibits/s rngtest: Program run time: 570229 microseconds on qemu x86-64: rngtest: input channel speed: (min=37.769; avg=101.136; max=136.239)Mibits/s rngtest: FIPS tests speed: (min=10.288; avg=30.682; max=40.155)Mibits/s rngtest: Program run time: 836800 microseconds4. observing sshd service.
using "systemctl disable rng-tools" disable service and reboot system. system boot up normal, sshd service also start in normal time without block.Reference:
[1] https://github.com/torvalds/linux/commit/30c08efec8884fb106b8e57094baa51bb4c44e32Signed-off-by Xiangyu Chen <[email protected]>--- meta/recipes-connectivity/openssh/openssh_9.1p1.bb | 9 +-------- 1 file changed, 1 insertion(+), 8 deletions(-)meta/recipes-connectivity/openssh/openssh_9.1p1.bb +1
-9@@ -52,15 +52,12 @@ SYSTEMD_SERVICE:${PN}-sshd = "sshd.socket" inherit autotools-brokensep ptest -PACKAGECONFIG ??= "rng-tools" +PACKAGECONFIG ??= "" PACKAGECONFIG[kerberos] = "--with-kerberos5,--without-kerberos5,krb5" PACKAGECONFIG[ldns] = "--with-ldns,--without-ldns,ldns" PACKAGECONFIG[libedit] = "--with-libedit,--without-libedit,libedit" PACKAGECONFIG[manpages] = "--with-mantype=man,--with-mantype=cat" -# Add RRECOMMENDS to rng-tools for sshd package -PACKAGECONFIG[rng-tools] = "" - EXTRA_AUTORECONF += "--exclude=aclocal" # login path is hardcoded in sshd@@ -160,10 +157,6 @@ FILES:${PN}-keygen = "${bindir}/ssh-keygen" RDEPENDS:${PN} += "${PN}-scp ${PN}-ssh ${PN}-sshd ${PN}-keygen ${PN}-sftp-server" RDEPENDS:${PN}-sshd += "${PN}-keygen ${@bb.utils.contains('DISTRO_FEATURES', 'pam', 'pam-plugin-keyinit pam-plugin-loginuid', '', d)}" -RRECOMMENDS:${PN}-sshd:append:class-target = "\ - ${@bb.utils.filter('PACKAGECONFIG', 'rng-tools', d)} \ -" - # gdb would make attach-ptrace test pass rather than skip but not worth the build dependencies RDEPENDS:${PN}-ptest += "${PN}-sftp ${PN}-misc ${PN}-sftp-server make sed sudo coreutils" --2.34.1
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#174469): https://lists.openembedded.org/g/openembedded-core/message/174469
Mute This Topic: https://lists.openembedded.org/mt/95556189/3616948
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-
Quoted 13 lines · Mark Hatle
CAUTION: This email comes from a non Wind River email account!
Do not click links or open attachments unless you recognize the sender and know the content is safe.In the kernel is the jitter entropy system. It should work on all platforms that have high res timers available. (This is the same mechanism that haveged was using before as well. So no change in the RNG quality, just now built into the kernel itself.)
The only place we've observed an issue with the new approach is on a qemu machine that was moderately loaned. The jitter was apparently too large (our theory) and caused the crng init to happen VERY slowly. (This would not have been resolved using rngd.)
Quoted 57 lines
All-in-all, I thought this change had already been made, and think it should go in.
(We tested this on microblaze, and performance with and without haveged was the same. I don't have number, but I expect the same behaviors on any architecture.)
--Mark
Quoted 49 lines · Khem Raj
would be good to know some numbers on non-arm/non-x86 systems too.
Quoted 45 lines · Xiangyu Chen
It appears that rngd is not needed as of linux-5.6 and later[1] and should not be installed by default since the purpose of rngd is to provide additional trusted sources of entropy.
We did some testing on real hardware, the result seems to support that we no longer need rngd by default on kernel v5.6 and later.
Testing result as below:
1. observing the crng init stage.
the "random: crng init done" always available before fs being mounted.2. generating random number without rngd.
testing command: dd if=/dev/random of=/dev/null status=progress on Marvell CN96xx RDB board, speed almost 20.4 MB/s without block on NXP i.mx6q board, speed almost 31.9 MB/s without block on qemu x86-64, speed almost 2.6MB/s without block3. using rngtest command without rngd
testing command: rngtest -c 1000 </dev/random
on Marvell CN96xx RDB board:
rngtest: input channel speed: (min=4.340; avg=135.364; max=146.719)Mibits/s
rngtest: FIPS tests speed: (min=8.197; avg=69.020; max=72.800)Mibits/s
rngtest: Program run time: 418771 microsecondson NXP i.mx6q board:
rngtest: input channel speed: (min=96.820; avg=326.769; max=340.598)Mibits/s
rngtest: FIPS tests speed: (min=15.090; avg=37.543; max=40.324)Mibits/s
rngtest: Program run time: 570229 microsecondson qemu x86-64:
rngtest: input channel speed: (min=37.769; avg=101.136; max=136.239)Mibits/s
rngtest: FIPS tests speed: (min=10.288; avg=30.682; max=40.155)Mibits/s
rngtest: Program run time: 836800 microseconds4. observing sshd service.
using "systemctl disable rng-tools" disable service and reboot system. system boot up normal, sshd service also start in normal time without block.Reference:
[1] https://github.com/torvalds/linux/commit/30c08efec8884fb106b8e57094baa51bb4c44e32Signed-off-by Xiangyu Chen <[email protected]>
Hello Steve,
According to oe release kernel version(https://wiki.yoctoproject.org/wiki/Linux_Yocto), the kirkstone and langdale is also using linux v5.6 later version kernel, so could you help to cherry pick this commit on master to kirkstone and langdale branch? Thanks!
The commit now on master:
commit: 868dfb46d96a27ec9041cb902fb769330277257d
https://git.openembedded.org/openembedded-core/commit/?id=868dfb46d96a27ec9041cb902fb769330277257d
Thanks,
Xiangyu
Quoted 55 lines
Quoted 36 lines
Quoted 36 lines
---
meta/recipes-connectivity/openssh/openssh_9.1p1.bb | 9 +--------1 file changed, 1 insertion(+), 8 deletions(-)meta/recipes-connectivity/openssh/openssh_9.1p1.bb +0
-0@@ -52,15 +52,12 @@ SYSTEMD_SERVICE:${PN}-sshd = "sshd.socket"inherit autotools-brokensep ptest
meta/recipes-connectivity/openssh/openssh_9.1p1.bb +1
-1-PACKAGECONFIG ??= "rng-tools" +PACKAGECONFIG ??= ""PACKAGECONFIG[kerberos] = "--with-kerberos5,--without-kerberos5,krb5" PACKAGECONFIG[ldns] = "--with-ldns,--without-ldns,ldns" PACKAGECONFIG[libedit] = "--with-libedit,--without-libedit,libedit" PACKAGECONFIG[manpages] = "--with-mantype=man,--with-mantype=cat"
meta/recipes-connectivity/openssh/openssh_9.1p1.bb +0
-3-# Add RRECOMMENDS to rng-tools for sshd package -PACKAGECONFIG[rng-tools] = "" -
EXTRA_AUTORECONF+= "--exclude=aclocal"# login path is hardcoded in sshdmeta/recipes-connectivity/openssh/openssh_9.1p1.bb +0
-0@@ -160,10 +157,6 @@ FILES:${PN}-keygen = "${bindir}/ssh-keygen"RDEPENDS:${PN} += "${PN}-scp ${PN}-ssh ${PN}-sshd ${PN}-keygen ${PN}-sftp-server"
RDEPENDS:${PN}-sshd += "${PN}-keygen ${@bb.utils.contains('DISTRO_FEATURES', 'pam', 'pam-plugin-keyinit pam-plugin-loginuid', '', d)}"meta/recipes-connectivity/openssh/openssh_9.1p1.bb +0
-4-RRECOMMENDS:${PN}-sshd:append:class-target = "\ - ${@bb.utils.filter('PACKAGECONFIG', 'rng-tools', d)} \ -" -# gdb would make attach-ptrace test pass rather than skip but not worth the build dependencies RDEPENDS:${PN}-ptest += "${PN}-sftp ${PN}-misc ${PN}-sftp-server make sed sudo coreutils"Quoted 10 lines
Quoted 10 lines
Signature · 1 line
2.34.1-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#174478): https://lists.openembedded.org/g/openembedded-core/message/174478
Mute This Topic: https://lists.openembedded.org/mt/95556189/7175143
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-