Re: git: 8a62a2a5659d - main - zfs: merge openzfs/zfs@f8e5af53e
Date: Sat, 21 Mar 2026 18:40:27 UTC
Warner Losh wrote: > > > On Sat, Mar 21, 2026 at 9:23 AM Shawn Webb <shawn.webb@hardenedbsd.org > <mailto:shawn.webb@hardenedbsd.org>> wrote: > > On Sat, Mar 21, 2026 at 09:18:20AM -0600, Warner Losh wrote: > > On Sat, Mar 21, 2026, 9:08 AM Shawn Webb > <shawn.webb@hardenedbsd.org <mailto:shawn.webb@hardenedbsd.org>> wrote: > > > > > On Sat, Mar 21, 2026 at 09:04:17AM +0100, A FreeBSD User wrote: > > > > Am Tage des Herren Fri, 20 Mar 2026 23:27:20 -0400 > > > > Charlie Li <vishwin@freebsd.org <mailto:vishwin@freebsd.org>> > schrieb: > > > > > > > > > Shawn Webb wrote: > > > > > > On Tue, Mar 17, 2026 at 04:52:16PM +0000, Shawn Webb wrote: > > > > > >> On Tue, Mar 17, 2026 at 10:44:59AM -0600, Warner Losh wrote: > > > > > >>> On Tue, Mar 17, 2026 at 10:36 AM Shawn Webb < > > > shawn.webb@hardenedbsd.org <mailto:shawn.webb@hardenedbsd.org>> > > > > > >>> wrote: > > > > > >>> > > > > > >>>> Hey Martin, > > > > > >>>> > > > > > >>>> On Sat, Mar 14, 2026 at 01:26:23PM +0000, Martin > Matuska wrote: > > > > > >>>>> The branch main has been updated by mm: > > > > > >>>>> > > > > > >>>>> URL: > > > > > >>>> > > > https://cgit.FreeBSD.org/src/commit/? > id=8a62a2a5659d1839d8799b4274c04469d7f17c78 <https:// > cgit.FreeBSD.org/src/commit/? > id=8a62a2a5659d1839d8799b4274c04469d7f17c78> > > > > > > > > >>>>> > > > > > >>>>> commit 8a62a2a5659d1839d8799b4274c04469d7f17c78 > > > > > >>>>> Merge: f91464171d61 f8e5af53e92f > > > > > >>>>> Author: Martin Matuska <mm@FreeBSD.org> > > > > > >>>>> AuthorDate: 2026-03-14 12:14:56 +0000 > > > > > >>>>> Commit: Martin Matuska <mm@FreeBSD.org> > > > > > >>>>> CommitDate: 2026-03-14 12:14:56 +0000 > > > > > >>>>> > > > > > >>>>> [snip for brevity] > > > > > >>>>> > > > > > >>>>> Obtained from: OpenZFS > > > > > >>>>> OpenZFS commit: > f8e5af53e92fa7c03393fbd4922cb9c1d0c15920 > > > > > >>>> > > > > > >>>> This commit seems to cause issues when building boot > loader > > > related > > > > > >>>> code: > > > > > >>>> > > > > > >>>> ==== BEGIN LOG ==== > > > > > >>>> 114232 bytes available > > > > > >>>> btxld -v -f aout -e 0x200000 -o loader_simp -l > > > > > >>>> /usr/obj/usr/src/amd64.amd64/stand/i386/btx/btxldr/ > btxldr -b > > > > > >>>> /usr/obj/usr/src/amd64.amd64/stand/i386/btx/btx/btx > > > loader_simp.bin > > > > > >>>> kernel: ver=1.02 size=690 load=9000 entry=9010 map=16M > pgctl=0:58 > > > > > >>>> client: fmt=elf size=5e2e8 text=57930 data=514c > bss=7470 entry=0 > > > > > >>>> output: fmt=aout size=61000 text=1000 data=5f000 > org=200000 > > > entry=200000 > > > > > >>>> ===> stand/i386/pxeldr (all) > > > > > >>>> -560 bytes available > > > > > >>>> *** Error code 1 > > > > > >>>> > > > > > >>> > > > > > >>> What all do you have enabled? The defaults aren't even > close to > > > running out > > > > > >>> of space (though I've not looked at this). > > > > > >> > > > > > >> Hey Warner, > > > > > >> > > > > > >> Thanks for reaching out! I've uploaded `make showconfig` > here: > > > > > >> https://hardenedbsd.org/~shawn/2026-03-17_srcconf- > r01.txt <https://hardenedbsd.org/~shawn/2026-03-17_srcconf-r01.txt> > > > > > >> > > > > > >> The following options are specific to HardenedBSD (in no > particular > > > > > >> order): > > > > > >> > > > > > >> 1. MK_HBSD_UPDATE > > > > > >> 2. MK_HBSDCONTROL > > > > > >> 3. MK_PIE > > > > > >> 4. MK_RELRO > > > > > >> 5. MK_SHLIBRANDOM > > > > > >> 6. MK_ZERO_REGS > > > > > >> 7. MK_SPECTREV1_FIX > > > > > >> 8. MK_SAFESTACK > > > > > >> 9. MK_RETPOLINE > > > > > >> 10. MK_LTOLIB > > > > > >> 11. MK_CFI > > > > > > > > > > > > MK_RETPOLINE was the culprit. Something about this ZFS > commit causes > > > > > > LLVM to emit more retpoline entries than before--too many > for a > > > little > > > > > > bootloader. That might be something to investigate later, > but only to > > > > > > satisfy a curious mind, not to actuall fix anything > (since nothing's > > > > > > actually broken.) > > > > > > > > > > > > Since it doesn't really make sense to apply speculative > execution > > > > > > mitigations to a bootloader, I disabled retpoline for a > components > > > > > > in stand/. > > > > > > > > > > > > Good to go. > > > > > > > > > > > Also just got bit by this, albeit during the lua loader, > since I have > > > > > WITH_RETPOLINE in my src.conf. > > > > > > > > > > > > > Hello, > > > > > > > > I do not have WITH_RETPOLINE in my /etc/src.conf, but since I > got this > > > mysterious error about > > > > not enough bytes left, I use WITHOUT_LOADER_PXEBOOT= YES (due > to issues > > > with WITH_BEARSSL=YES > > > > also used). > > > > Despite not using any WITH_RETPOLINE I also catch the error ... > > > > > > Something about this ZFS commit causes the boot laoder to be > too big. > > > I guess the first sign of trouble was with retpolines, but > there seem > > > to now be additional signs. > > > > > > What's the process for filing a bug report against OpenzFS for > > > something like this? (Not asking you directly, just a general > question > > > for the thread.) > > > > > > > That's a good question. We are rapidly aporoaching the day we > will have to > > freeze the set of zfs feature that we can boot with the BIOS > path. There's > > only so much space. > > > > Right now, there's little to no bootloader testing, let alone > analysis > > done and that will need to change. > > Another question might also be: is it worth the time and effort? Are > people really using a ZFS-enabled traditional MBR BIOS bootloader on > FreeBSD 16-CURRENT (or eventual 16.0-RELEASE)? From my rather naive > (and ignorant, since I don't know FreeBSD's userbase well), it seems > like folks use either loader.efi or u-boot (or both!) > > > Nobody uses MBR BIOS bootloader. It's totally unsupported. They all use the > GPT one, which is used for all kinds of things, most recently several > light-weight > VM systems that don't have UEFI. BIOS is what imposes the limit, not MBR. > > But there's a huge number of systems that are already in the field that need > to be upgradeable, but we will eventually need to draw a line in the > sand for > that. > The machine where I hit this error on building the lua loader is a BIOS-GPT system. -- Charlie Li ...nope, still don't have an exit line.