Does this Oracle ZFS detail hold true for illumos?

Does this Oracle ZFS detail hold true for illumos?

[Eric J Bowman]

[-- Attachment #1: Type: text/plain, Size: 476 bytes --]

Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...



"

Solaris by default will create EFI label when presented with whole disk

for zfs pool. The exception is when it's going to be used for boot (rpool),

on which it will use SMI label.

"



...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.



-Eric



[1] https://github.com/openzfs/zfs/issues/94

[-- Attachment #2: Type: text/html, Size: 3088 bytes --]

Re: [discuss] Does this Oracle ZFS detail hold true for illumos?

[Toomas Soome]

[-- Attachment #1: Type: text/plain, Size: 1736 bytes --]



> On 21. May 2024, at 16:02, Eric J Bowman via illumos-discuss <discuss@lists.illumos.org> wrote:
> 
> Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...
> 
> "
> Solaris by default will create EFI label when presented with whole disk
> for zfs pool. The exception is when it's going to be used for boot (rpool),
> on which it will use SMI label.
> "
> 
> ...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.
> 
> -Eric
> 
> [1] https://github.com/openzfs/zfs/issues/94
> 

Oracle Solaris 11.4 does create EFI (GPT) label with slice 0 for whole disk data pool, EFI (GPT) label with slice 0 for ESP and slice 1 for boot pool (rpool) on UEFI system.  On BIOS system, on boot disk, it will create BIOS_boot partition + rpool partition, where BIOS_boot is used by grub2 as raw space.

On illumos, for boot disk, we do create EFI label with slice 0 for ESP and slice 1 for boot pool (rpool). On BIOS system, the ESP is created, but not used; the boot program is stored in zfs label boot program area. This allows us to switch between UEFI and BIOS configuration (where supported).

When user did opt to use SMI label, we will create single Solaris2 MBR partition, with SMI label in it and slice 0 is created for boot pool.

Thats about defaults. Then there are other options - we still can boot from ufs, but since boot block area with ufs is very small, we need to use boot slice (raw space), either in SMI label or in EFI label, but this is no-default because boot environments (beadm) will not with ufs and you need to set bootpath property with correct device name for rootfs mount to happen.

rgds,
toomas


[-- Attachment #2: Type: text/html, Size: 5010 bytes --]

Re: [discuss] Does this Oracle ZFS detail hold true for illumos?

[Eric J Bowman]

[-- Attachment #1: Type: text/plain, Size: 2995 bytes --]

>

> When user did opt to use SMI label, we will create single

> Solaris2 MBR partition, with SMI label in it and slice 0 is

> created for boot pool.

>



Thank you, Toomas! I think I understand. I'm installing rpool to a partition, not "whole disk", under SMI. But, if the ESP is on the "active" partition, once rpool boots, fdisk assumes it's a 100% EFI disk. This does not match what it says in VTOC, so I get that i/o error?



I need to leave the first 34 sectors in "no-man's land" though, not inside the UEFI partition. Hmmm... "PMBCylinder", where fdisk partition 1 = PC98, partition 2 = SOLARIS2, set partition 1 active and put ESP there, let partition 2 be SMI with my rpool/zonepool partitioning? I believe that's my way forward. 



-Eric






---- On Tue, 21 May 2024 06:29:22 -0700 Toomas Soome via illumos-discuss <discuss@lists.illumos.org> wrote ---







On 21. May 2024, at 16:02, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...



"

Solaris by default will create EFI label when presented with whole disk

for zfs pool. The exception is when it's going to be used for boot (rpool),

on which it will use SMI label.

"



...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.



-Eric



[1] https://github.com/openzfs/zfs/issues/94










Oracle Solaris 11.4 does create EFI (GPT) label with slice 0 for whole disk data pool, EFI (GPT) label with slice 0 for ESP and slice 1 for boot pool (rpool) on UEFI system.  On BIOS system, on boot disk, it will create BIOS_boot partition + rpool partition, where BIOS_boot is used by grub2 as raw space.



On illumos, for boot disk, we do create EFI label with slice 0 for ESP and slice 1 for boot pool (rpool). On BIOS system, the ESP is created, but not used; the boot program is stored in zfs label boot program area. This allows us to switch between UEFI and BIOS configuration (where supported).



When user did opt to use SMI label, we will create single Solaris2 MBR partition, with SMI label in it and slice 0 is created for boot pool.



Thats about defaults. Then there are other options - we still can boot from ufs, but since boot block area with ufs is very small, we need to use boot slice (raw space), either in SMI label or in EFI label, but this is no-default because boot environments (beadm) will not with ufs and you need to set bootpath property with correct device name for rootfs mount to happen.



rgds,

toomas



https://illumos.topicbox.com/latest / illumos-discuss / see https://illumos.topicbox.com/groups/discuss + https://illumos.topicbox.com/groups/discuss/members + https://illumos.topicbox.com/groups/discuss/subscription https://illumos.topicbox.com/groups/discuss/T8c9973e131d44516-M3cec6c3b2dc4bfd44ab3c37f

[-- Attachment #2: Type: text/html, Size: 7063 bytes --]

Re: [discuss] Does this Oracle ZFS detail hold true for illumos?

[Eric J Bowman]

[-- Attachment #1: Type: text/plain, Size: 3445 bytes --]

Jakey needs a new PDB before I can resume testing, but I already know I can boot from USB to tribblix-sparc.iso on s0 (+34b). Question is, from there, can I chainload into my rpool on s5, where altroot=SPARC? I know one rpool can have alternate BE's, can they be different platforms?


-Eric






---- On Tue, 21 May 2024 07:39:40 -0700 Eric J Bowman via illumos-discuss <discuss@lists.illumos.org> wrote ---



>

> When user did opt to use SMI label, we will create single

> Solaris2 MBR partition, with SMI label in it and slice 0 is

> created for boot pool.

>



Thank you, Toomas! I think I understand. I'm installing rpool to a partition, not "whole disk", under SMI. But, if the ESP is on the "active" partition, once rpool boots, fdisk assumes it's a 100% EFI disk. This does not match what it says in VTOC, so I get that i/o error?



I need to leave the first 34 sectors in "no-man's land" though, not inside the UEFI partition. Hmmm... "PMBCylinder", where fdisk partition 1 = PC98, partition 2 = SOLARIS2, set partition 1 active and put ESP there, let partition 2 be SMI with my rpool/zonepool partitioning? I believe that's my way forward. 



-Eric







---- On Tue, 21 May 2024 06:29:22 -0700 Toomas Soome via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---







On 21. May 2024, at 16:02, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...



"

Solaris by default will create EFI label when presented with whole disk

for zfs pool. The exception is when it's going to be used for boot (rpool),

on which it will use SMI label.

"



...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.



-Eric



[1] https://github.com/openzfs/zfs/issues/94










Oracle Solaris 11.4 does create EFI (GPT) label with slice 0 for whole disk data pool, EFI (GPT) label with slice 0 for ESP and slice 1 for boot pool (rpool) on UEFI system.  On BIOS system, on boot disk, it will create BIOS_boot partition + rpool partition, where BIOS_boot is used by grub2 as raw space.



On illumos, for boot disk, we do create EFI label with slice 0 for ESP and slice 1 for boot pool (rpool). On BIOS system, the ESP is created, but not used; the boot program is stored in zfs label boot program area. This allows us to switch between UEFI and BIOS configuration (where supported).



When user did opt to use SMI label, we will create single Solaris2 MBR partition, with SMI label in it and slice 0 is created for boot pool.



Thats about defaults. Then there are other options - we still can boot from ufs, but since boot block area with ufs is very small, we need to use boot slice (raw space), either in SMI label or in EFI label, but this is no-default because boot environments (beadm) will not with ufs and you need to set bootpath property with correct device name for rootfs mount to happen.



rgds,

toomas










https://illumos.topicbox.com/latest / illumos-discuss / see https://illumos.topicbox.com/groups/discuss + https://illumos.topicbox.com/groups/discuss/members + https://illumos.topicbox.com/groups/discuss/subscription https://illumos.topicbox.com/groups/discuss/T8c9973e131d44516-M6108086d4faa207146f406de

[-- Attachment #2: Type: text/html, Size: 8057 bytes --]

Re: [discuss] Does this Oracle ZFS detail hold true for illumos?

[Toomas Soome]

[-- Attachment #1: Type: text/plain, Size: 4232 bytes --]



> On 21. May 2024, at 23:21, Eric J Bowman via illumos-discuss <discuss@lists.illumos.org> wrote:
> 
> Jakey needs a new PDB before I can resume testing, but I already know I can boot from USB to tribblix-sparc.iso on s0 (+34b). Question is, from there, can I chainload into my rpool on s5, where altroot=SPARC? I know one rpool can have alternate BE's, can they be different platforms?
> 
> -Eric


Yes and no. from point of view of BE itself, sure, no problem. BE is just another filesystem and it really does not matter which platform it will serve.

You need to be careful about different endians; I would rather suspect that sparc boot code does not check (and perform) the endian conversion; loader on x86 needs patch (from my queue;)  to make it happen. It means, once you have either platform done with pool update, the other or neither is able to boot… (ok, with loader the code is already ready, so we can fix it for x86).

rgds,
toomas


> 
> 
> 
> ---- On Tue, 21 May 2024 07:39:40 -0700 Eric J Bowman via illumos-discuss <discuss@lists.illumos.org> wrote ---
> 
> >
> > When user did opt to use SMI label, we will create single
> > Solaris2 MBR partition, with SMI label in it and slice 0 is
> > created for boot pool.
> >
> 
> Thank you, Toomas! I think I understand. I'm installing rpool to a partition, not "whole disk", under SMI. But, if the ESP is on the "active" partition, once rpool boots, fdisk assumes it's a 100% EFI disk. This does not match what it says in VTOC, so I get that i/o error?
> 
> I need to leave the first 34 sectors in "no-man's land" though, not inside the UEFI partition. Hmmm... "PMBCylinder", where fdisk partition 1 = PC98, partition 2 = SOLARIS2, set partition 1 active and put ESP there, let partition 2 be SMI with my rpool/zonepool partitioning? I believe that's my way forward. 
> 
> -Eric
> 
> 
> 
> ---- On Tue, 21 May 2024 06:29:22 -0700 Toomas Soome via illumos-discuss <discuss@lists.illumos.org <mailto:discuss@lists.illumos.org>> wrote ---
> 
> 
> 
> On 21. May 2024, at 16:02, Eric J Bowman via illumos-discuss <discuss@lists.illumos.org <mailto:discuss@lists.illumos.org>> wrote:
> 
> Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...
> 
> "
> Solaris by default will create EFI label when presented with whole disk
> for zfs pool. The exception is when it's going to be used for boot (rpool),
> on which it will use SMI label.
> "
> 
> ...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.
> 
> -Eric
> 
> [1] https://github.com/openzfs/zfs/issues/94
> 
> 
> Oracle Solaris 11.4 does create EFI (GPT) label with slice 0 for whole disk data pool, EFI (GPT) label with slice 0 for ESP and slice 1 for boot pool (rpool) on UEFI system.  On BIOS system, on boot disk, it will create BIOS_boot partition + rpool partition, where BIOS_boot is used by grub2 as raw space.
> 
> On illumos, for boot disk, we do create EFI label with slice 0 for ESP and slice 1 for boot pool (rpool). On BIOS system, the ESP is created, but not used; the boot program is stored in zfs label boot program area. This allows us to switch between UEFI and BIOS configuration (where supported).
> 
> When user did opt to use SMI label, we will create single Solaris2 MBR partition, with SMI label in it and slice 0 is created for boot pool.
> 
> Thats about defaults. Then there are other options - we still can boot from ufs, but since boot block area with ufs is very small, we need to use boot slice (raw space), either in SMI label or in EFI label, but this is no-default because boot environments (beadm) will not with ufs and you need to set bootpath property with correct device name for rootfs mount to happen.
> 
> rgds,
> toomas
> 
> 
> 
> 
> illumos <https://illumos.topicbox.com/latest> / illumos-discuss / see discussions <https://illumos.topicbox.com/groups/discuss> + participants <https://illumos.topicbox.com/groups/discuss/members> + delivery options <https://illumos.topicbox.com/groups/discuss/subscription>Permalink <https://illumos.topicbox.com/groups/discuss/T8c9973e131d44516-M346cdf9186395b0910d6e668>

[-- Attachment #2: Type: text/html, Size: 9181 bytes --]

Re: [discuss] Does this Oracle ZFS detail hold true for illumos?

[Eric J Bowman]

[-- Attachment #1: Type: text/plain, Size: 4482 bytes --]

Just pondering, not arguing, but do we need loader? Isn't that so we can gptzfsboot? Jakey thinks there's a bootable rpool on an SMI partition.



-Eric







---- On Tue, 21 May 2024 13:48:13 -0700 Toomas Soome via illumos-discuss <discuss@lists.illumos.org> wrote ---







On 21. May 2024, at 23:21, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Jakey needs a new PDB before I can resume testing, but I already know I can boot from USB to tribblix-sparc.iso on s0 (+34b). Question is, from there, can I chainload into my rpool on s5, where altroot=SPARC? I know one rpool can have alternate BE's, can they be different platforms?



-Eric









Yes and no. from point of view of BE itself, sure, no problem. BE is just another filesystem and it really does not matter which platform it will serve.



You need to be careful about different endians; I would rather suspect that sparc boot code does not check (and perform) the endian conversion; loader on x86 needs patch (from my queue;)  to make it happen. It means, once you have either platform done with pool update, the other or neither is able to boot… (ok, with loader the code is already ready, so we can fix it for x86).



rgds,

toomas











---- On Tue, 21 May 2024 07:39:40 -0700 Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---



>

> When user did opt to use SMI label, we will create single

> Solaris2 MBR partition, with SMI label in it and slice 0 is

> created for boot pool.

>



Thank you, Toomas! I think I understand. I'm installing rpool to a partition, not "whole disk", under SMI. But, if the ESP is on the "active" partition, once rpool boots, fdisk assumes it's a 100% EFI disk. This does not match what it says in VTOC, so I get that i/o error?



I need to leave the first 34 sectors in "no-man's land" though, not inside the UEFI partition. Hmmm... "PMBCylinder", where fdisk partition 1 = PC98, partition 2 = SOLARIS2, set partition 1 active and put ESP there, let partition 2 be SMI with my rpool/zonepool partitioning? I believe that's my way forward. 



-Eric







---- On Tue, 21 May 2024 06:29:22 -0700 Toomas Soome via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---







On 21. May 2024, at 16:02, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...



"

Solaris by default will create EFI label when presented with whole disk

for zfs pool. The exception is when it's going to be used for boot (rpool),

on which it will use SMI label.

"



...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.



-Eric



[1] https://github.com/openzfs/zfs/issues/94










Oracle Solaris 11.4 does create EFI (GPT) label with slice 0 for whole disk data pool, EFI (GPT) label with slice 0 for ESP and slice 1 for boot pool (rpool) on UEFI system.  On BIOS system, on boot disk, it will create BIOS_boot partition + rpool partition, where BIOS_boot is used by grub2 as raw space.



On illumos, for boot disk, we do create EFI label with slice 0 for ESP and slice 1 for boot pool (rpool). On BIOS system, the ESP is created, but not used; the boot program is stored in zfs label boot program area. This allows us to switch between UEFI and BIOS configuration (where supported).



When user did opt to use SMI label, we will create single Solaris2 MBR partition, with SMI label in it and slice 0 is created for boot pool.



Thats about defaults. Then there are other options - we still can boot from ufs, but since boot block area with ufs is very small, we need to use boot slice (raw space), either in SMI label or in EFI label, but this is no-default because boot environments (beadm) will not with ufs and you need to set bootpath property with correct device name for rootfs mount to happen.



rgds,

toomas























https://illumos.topicbox.com/latest / illumos-discuss / see https://illumos.topicbox.com/groups/discuss + https://illumos.topicbox.com/groups/discuss/members + https://illumos.topicbox.com/groups/discuss/subscription https://illumos.topicbox.com/groups/discuss/T8c9973e131d44516-M35fe4843586d54f9fcd902b9

[-- Attachment #2: Type: text/html, Size: 10086 bytes --]

Re: [discuss] Does this Oracle ZFS detail hold true for illumos?

[Toomas Soome]

[-- Attachment #1: Type: text/plain, Size: 5063 bytes --]



> On 22. May 2024, at 00:00, Eric J Bowman via illumos-discuss <discuss@lists.illumos.org> wrote:
> 
> Just pondering, not arguing, but do we need loader? Isn't that so we can gptzfsboot? Jakey thinks there's a bootable rpool on an SMI partition.
> 
> -Eric

The zfs reader code in gptzfsboot and loader is the same shared code, gptzfsboot by itself in our case does only load and start loader, not OS kernel.

By “loader”, I usually refer to boot loader as whole - UEFI and BIOS version (and BIOS version is consisting of files implementing different stages).

rgds,
toomas


> 
> 
> 
> ---- On Tue, 21 May 2024 13:48:13 -0700 Toomas Soome via illumos-discuss <discuss@lists.illumos.org> wrote ---
> 
> 
> 
> On 21. May 2024, at 23:21, Eric J Bowman via illumos-discuss <discuss@lists.illumos.org <mailto:discuss@lists.illumos.org>> wrote:
> 
> Jakey needs a new PDB before I can resume testing, but I already know I can boot from USB to tribblix-sparc.iso on s0 (+34b). Question is, from there, can I chainload into my rpool on s5, where altroot=SPARC? I know one rpool can have alternate BE's, can they be different platforms?
> 
> -Eric
> 
> 
> Yes and no. from point of view of BE itself, sure, no problem. BE is just another filesystem and it really does not matter which platform it will serve.
> 
> You need to be careful about different endians; I would rather suspect that sparc boot code does not check (and perform) the endian conversion; loader on x86 needs patch (from my queue;)  to make it happen. It means, once you have either platform done with pool update, the other or neither is able to boot… (ok, with loader the code is already ready, so we can fix it for x86).
> 
> rgds,
> toomas
> 
> 
> 
> 
> 
> ---- On Tue, 21 May 2024 07:39:40 -0700 Eric J Bowman via illumos-discuss <discuss@lists.illumos.org <mailto:discuss@lists.illumos.org>> wrote ---
> 
> >
> > When user did opt to use SMI label, we will create single
> > Solaris2 MBR partition, with SMI label in it and slice 0 is
> > created for boot pool.
> >
> 
> Thank you, Toomas! I think I understand. I'm installing rpool to a partition, not "whole disk", under SMI. But, if the ESP is on the "active" partition, once rpool boots, fdisk assumes it's a 100% EFI disk. This does not match what it says in VTOC, so I get that i/o error?
> 
> I need to leave the first 34 sectors in "no-man's land" though, not inside the UEFI partition. Hmmm... "PMBCylinder", where fdisk partition 1 = PC98, partition 2 = SOLARIS2, set partition 1 active and put ESP there, let partition 2 be SMI with my rpool/zonepool partitioning? I believe that's my way forward. 
> 
> -Eric
> 
> 
> 
> ---- On Tue, 21 May 2024 06:29:22 -0700 Toomas Soome via illumos-discuss <discuss@lists.illumos.org <mailto:discuss@lists.illumos.org>> wrote ---
> 
> 
> 
> On 21. May 2024, at 16:02, Eric J Bowman via illumos-discuss <discuss@lists.illumos.org <mailto:discuss@lists.illumos.org>> wrote:
> 
> Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...
> 
> "
> Solaris by default will create EFI label when presented with whole disk
> for zfs pool. The exception is when it's going to be used for boot (rpool),
> on which it will use SMI label.
> "
> 
> ...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.
> 
> -Eric
> 
> [1] https://github.com/openzfs/zfs/issues/94
> 
> 
> Oracle Solaris 11.4 does create EFI (GPT) label with slice 0 for whole disk data pool, EFI (GPT) label with slice 0 for ESP and slice 1 for boot pool (rpool) on UEFI system.  On BIOS system, on boot disk, it will create BIOS_boot partition + rpool partition, where BIOS_boot is used by grub2 as raw space.
> 
> On illumos, for boot disk, we do create EFI label with slice 0 for ESP and slice 1 for boot pool (rpool). On BIOS system, the ESP is created, but not used; the boot program is stored in zfs label boot program area. This allows us to switch between UEFI and BIOS configuration (where supported).
> 
> When user did opt to use SMI label, we will create single Solaris2 MBR partition, with SMI label in it and slice 0 is created for boot pool.
> 
> Thats about defaults. Then there are other options - we still can boot from ufs, but since boot block area with ufs is very small, we need to use boot slice (raw space), either in SMI label or in EFI label, but this is no-default because boot environments (beadm) will not with ufs and you need to set bootpath property with correct device name for rootfs mount to happen.
> 
> rgds,
> toomas
> 
> 
> 
> 
> 
> 
> illumos <https://illumos.topicbox.com/latest> / illumos-discuss / see discussions <https://illumos.topicbox.com/groups/discuss> + participants <https://illumos.topicbox.com/groups/discuss/members> + delivery options <https://illumos.topicbox.com/groups/discuss/subscription>Permalink <https://illumos.topicbox.com/groups/discuss/T8c9973e131d44516-Ma3b96d288e16bc851d08228b>

[-- Attachment #2: Type: text/html, Size: 10962 bytes --]

Re: [discuss] Does this Oracle ZFS detail hold true for illumos?

[Eric J Bowman]

[-- Attachment #1: Type: text/plain, Size: 5989 bytes --]

OK... x86 bootloader goes in 1st fdisk partition, PC98 "active" ESP, p1s2.

SPARC bootloader goes in 2nd fdisk partition, SMI, p2s3...?



So long as p2s3 isn't "active" illumos won't think it's a GPT partition?

Nothing else would think to boot from p2s3.



Could always use separate partitions for rpool-sparc and rpool-x86, too.



This setup only works if the iso9660 s0 is overwritten with a platform-

specific .iso; which, curiously, COW controllers don't care about.

They only care if you try to delete an iso9660-formatted partition.



Obviously, if my package system "packages" are detached-zone tarballs

I'll have to make them platform-specific. Installing one may call zap

to add some TRIB packages to the global zone, but this is done with 

Tribblix sparse-root zone templates.



Thanks for the help!


-Eric







---- On Tue, 21 May 2024 14:05:48 -0700 Toomas Soome via illumos-discuss <discuss@lists.illumos.org> wrote ---







On 22. May 2024, at 00:00, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Just pondering, not arguing, but do we need loader? Isn't that so we can gptzfsboot? Jakey thinks there's a bootable rpool on an SMI partition.



-Eric







The zfs reader code in gptzfsboot and loader is the same shared code, gptzfsboot by itself in our case does only load and start loader, not OS kernel.



By “loader”, I usually refer to boot loader as whole - UEFI and BIOS version (and BIOS version is consisting of files implementing different stages).



rgds,

toomas











---- On Tue, 21 May 2024 13:48:13 -0700 Toomas Soome via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---







On 21. May 2024, at 23:21, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Jakey needs a new PDB before I can resume testing, but I already know I can boot from USB to tribblix-sparc.iso on s0 (+34b). Question is, from there, can I chainload into my rpool on s5, where altroot=SPARC? I know one rpool can have alternate BE's, can they be different platforms?



-Eric









Yes and no. from point of view of BE itself, sure, no problem. BE is just another filesystem and it really does not matter which platform it will serve.



You need to be careful about different endians; I would rather suspect that sparc boot code does not check (and perform) the endian conversion; loader on x86 needs patch (from my queue;)  to make it happen. It means, once you have either platform done with pool update, the other or neither is able to boot… (ok, with loader the code is already ready, so we can fix it for x86).



rgds,

toomas











---- On Tue, 21 May 2024 07:39:40 -0700 Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---



>

> When user did opt to use SMI label, we will create single

> Solaris2 MBR partition, with SMI label in it and slice 0 is

> created for boot pool.

>



Thank you, Toomas! I think I understand. I'm installing rpool to a partition, not "whole disk", under SMI. But, if the ESP is on the "active" partition, once rpool boots, fdisk assumes it's a 100% EFI disk. This does not match what it says in VTOC, so I get that i/o error?



I need to leave the first 34 sectors in "no-man's land" though, not inside the UEFI partition. Hmmm... "PMBCylinder", where fdisk partition 1 = PC98, partition 2 = SOLARIS2, set partition 1 active and put ESP there, let partition 2 be SMI with my rpool/zonepool partitioning? I believe that's my way forward. 



-Eric







---- On Tue, 21 May 2024 06:29:22 -0700 Toomas Soome via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---







On 21. May 2024, at 16:02, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...



"

Solaris by default will create EFI label when presented with whole disk

for zfs pool. The exception is when it's going to be used for boot (rpool),

on which it will use SMI label.

"



...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.



-Eric



[1] https://github.com/openzfs/zfs/issues/94










Oracle Solaris 11.4 does create EFI (GPT) label with slice 0 for whole disk data pool, EFI (GPT) label with slice 0 for ESP and slice 1 for boot pool (rpool) on UEFI system.  On BIOS system, on boot disk, it will create BIOS_boot partition + rpool partition, where BIOS_boot is used by grub2 as raw space.



On illumos, for boot disk, we do create EFI label with slice 0 for ESP and slice 1 for boot pool (rpool). On BIOS system, the ESP is created, but not used; the boot program is stored in zfs label boot program area. This allows us to switch between UEFI and BIOS configuration (where supported).



When user did opt to use SMI label, we will create single Solaris2 MBR partition, with SMI label in it and slice 0 is created for boot pool.



Thats about defaults. Then there are other options - we still can boot from ufs, but since boot block area with ufs is very small, we need to use boot slice (raw space), either in SMI label or in EFI label, but this is no-default because boot environments (beadm) will not with ufs and you need to set bootpath property with correct device name for rootfs mount to happen.



rgds,

toomas



































https://illumos.topicbox.com/latest / illumos-discuss / see https://illumos.topicbox.com/groups/discuss + https://illumos.topicbox.com/groups/discuss/members + https://illumos.topicbox.com/groups/discuss/subscription https://illumos.topicbox.com/groups/discuss/T8c9973e131d44516-M2115d39f40e9b7ef16f44d53

[-- Attachment #2: Type: text/html, Size: 12847 bytes --]

Re: [discuss] Does this Oracle ZFS detail hold true for illumos?

[Eric J Bowman]

[-- Attachment #1: Type: text/plain, Size: 7770 bytes --]

From my desktop, it isn't "zlogin -C Browser" to use the Browser, it's click the launcher icon and ssh -X pulls the browser instance into a Moksha (ETK) window, after booting (or just resuming) the browser zone, which is halted when the Moksha window is closed.



For a package like "autopsy" we *could* install the Java and LibreOffice TRIB-package dependencies in the global zone, but we don't want our other LibreOffice activities to interfere with the autopsy integration, nor do we want that LibreOffice, or Java, "changed out from under us", so they're included in the Autopsy Zone, irregardless of their presence/absence in the global zone. They all update "atomically" as a coherent package.



CAINE is overwhelming, because all those duplicate copies of apps are actually different prerequisites for integrating various other packages, with the overall result that CAINE takes twice as long to boot as, with only half the stability of, Windows. However, it provides an excellent "live-installed" reference when configuring your own autopsy package.



Bisonix 0.9 from '09 is history, "v1" was a global-zone hypervisor; "v2" is a desktop... configuration more than a distro, so I'll call it "v2.44" as it'll be Q4 before I'm ready to share "Solaris Qubes." Oh, if you want to install the CAINE package, the detached-zone tarball is installed as an Lx zone, the installer workflow is the same except this time, the launcher does "zlogin -C CAINE" and it's up to you how to script the reboot/poweroff behavior if you don't like my default reboot=persist, poweroff=erase ZFS snapshotting.



-Eric 







---- On Tue, 21 May 2024 14:23:15 -0700 Eric J Bowman via illumos-discuss <discuss@lists.illumos.org> wrote ---



OK... x86 bootloader goes in 1st fdisk partition, PC98 "active" ESP, p1s2.

SPARC bootloader goes in 2nd fdisk partition, SMI, p2s3...?



So long as p2s3 isn't "active" illumos won't think it's a GPT partition?

Nothing else would think to boot from p2s3.



Could always use separate partitions for rpool-sparc and rpool-x86, too.



This setup only works if the iso9660 s0 is overwritten with a platform-

specific .iso; which, curiously, COW controllers don't care about.

They only care if you try to delete an iso9660-formatted partition.



Obviously, if my package system "packages" are detached-zone tarballs

I'll have to make them platform-specific. Installing one may call zap

to add some TRIB packages to the global zone, but this is done with 

Tribblix sparse-root zone templates.



Thanks for the help!



-Eric








---- On Tue, 21 May 2024 14:05:48 -0700 Toomas Soome via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---







On 22. May 2024, at 00:00, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Just pondering, not arguing, but do we need loader? Isn't that so we can gptzfsboot? Jakey thinks there's a bootable rpool on an SMI partition.



-Eric







The zfs reader code in gptzfsboot and loader is the same shared code, gptzfsboot by itself in our case does only load and start loader, not OS kernel.



By “loader”, I usually refer to boot loader as whole - UEFI and BIOS version (and BIOS version is consisting of files implementing different stages).



rgds,

toomas











---- On Tue, 21 May 2024 13:48:13 -0700 Toomas Soome via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---







On 21. May 2024, at 23:21, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Jakey needs a new PDB before I can resume testing, but I already know I can boot from USB to tribblix-sparc.iso on s0 (+34b). Question is, from there, can I chainload into my rpool on s5, where altroot=SPARC? I know one rpool can have alternate BE's, can they be different platforms?



-Eric









Yes and no. from point of view of BE itself, sure, no problem. BE is just another filesystem and it really does not matter which platform it will serve.



You need to be careful about different endians; I would rather suspect that sparc boot code does not check (and perform) the endian conversion; loader on x86 needs patch (from my queue;)  to make it happen. It means, once you have either platform done with pool update, the other or neither is able to boot… (ok, with loader the code is already ready, so we can fix it for x86).



rgds,

toomas











---- On Tue, 21 May 2024 07:39:40 -0700 Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---



>

> When user did opt to use SMI label, we will create single

> Solaris2 MBR partition, with SMI label in it and slice 0 is

> created for boot pool.

>



Thank you, Toomas! I think I understand. I'm installing rpool to a partition, not "whole disk", under SMI. But, if the ESP is on the "active" partition, once rpool boots, fdisk assumes it's a 100% EFI disk. This does not match what it says in VTOC, so I get that i/o error?



I need to leave the first 34 sectors in "no-man's land" though, not inside the UEFI partition. Hmmm... "PMBCylinder", where fdisk partition 1 = PC98, partition 2 = SOLARIS2, set partition 1 active and put ESP there, let partition 2 be SMI with my rpool/zonepool partitioning? I believe that's my way forward. 



-Eric







---- On Tue, 21 May 2024 06:29:22 -0700 Toomas Soome via illumos-discuss <mailto:discuss@lists.illumos.org> wrote ---







On 21. May 2024, at 16:02, Eric J Bowman via illumos-discuss <mailto:discuss@lists.illumos.org> wrote:



Changing my search from "multiboot interop" to "zfsboot interop" has proven fruitful. I just need to know if this answer...



"

Solaris by default will create EFI label when presented with whole disk

for zfs pool. The exception is when it's going to be used for boot (rpool),

on which it will use SMI label.

"



...is the *right* answer? It was written in 2011 in reference to Oracle ZFS.



-Eric



[1] https://github.com/openzfs/zfs/issues/94










Oracle Solaris 11.4 does create EFI (GPT) label with slice 0 for whole disk data pool, EFI (GPT) label with slice 0 for ESP and slice 1 for boot pool (rpool) on UEFI system.  On BIOS system, on boot disk, it will create BIOS_boot partition + rpool partition, where BIOS_boot is used by grub2 as raw space.



On illumos, for boot disk, we do create EFI label with slice 0 for ESP and slice 1 for boot pool (rpool). On BIOS system, the ESP is created, but not used; the boot program is stored in zfs label boot program area. This allows us to switch between UEFI and BIOS configuration (where supported).



When user did opt to use SMI label, we will create single Solaris2 MBR partition, with SMI label in it and slice 0 is created for boot pool.



Thats about defaults. Then there are other options - we still can boot from ufs, but since boot block area with ufs is very small, we need to use boot slice (raw space), either in SMI label or in EFI label, but this is no-default because boot environments (beadm) will not with ufs and you need to set bootpath property with correct device name for rootfs mount to happen.



rgds,

toomas









































https://illumos.topicbox.com/latest / illumos-discuss / see https://illumos.topicbox.com/groups/discuss + https://illumos.topicbox.com/groups/discuss/members + https://illumos.topicbox.com/groups/discuss/subscription https://illumos.topicbox.com/groups/discuss/T8c9973e131d44516-Mcc9e260ce0da032d66871c47

[-- Attachment #2: Type: text/html, Size: 15183 bytes --]