Repository navigation
Description
The weekly audit found 16 Edera commit(s) on edera/mainline (7ad1695c933c) that edera/6.18-lts (c20f0c9b497c) does not have. Audit run.
| Mainline commit | Subject | Backport status |
|---|---|---|
3f7ca89e3d70 |
xen/gntdev: don't copy to user space under priv->lock (#79) | none |
1dc86dd35ab6 |
x86/xen: clear the user pgd pointer when freeing it | #82 |
e79780b9d1b6 |
xen/gntdev: reset the copy batch's writeable flag (#81) | none |
8f315d39b605 |
xen/balloon: let the OOM notifier grow an empty domU balloon (#84) | none |
ec5588ab8435 |
x86: fault: emulate trampolines that start with endbr (#87) | none |
7d91f39740ed |
xen-pciback: free a device's config fields before its dev_data (#88) | none |
bc97ba75c220 |
mm: never give the brk area execute permission under OpenPaX MPROTECT | none |
77c9976b13e0 |
x86/mm: emulate OpenPaX trampolines only when the fault would be SIGSEGV | none |
93dbfdcf5437 |
openpax: do not enable trampoline emulation for every program by default | none |
a4543cd54a6c |
iommu/xen: keep reserving the PCI host-bridge windows in dom0 | #110 |
8cd845532e3c |
xen/pcifront: do not promise pcifront_machine_sbdf() to built-in code when modular | #110 |
25ec4b49a0e6 |
9p/xen: only let an errno come back from a control request's Rlerror | #111 |
ce62f82d2094 |
9p/xen: do not dereference priv->client while replaying a saved session | #111 |
78700e1ae0db |
9p/xen: keep a saved session across a suspend that has nothing to save | #111 |
85e1da02b838 |
9p/xen: end foreign access on the grant refs we granted, not the intf page's | #111 |
7ad1695c933c |
9p/xen: order the control reply buffer before the pending flag | #111 |
Each one is resolved by landing its backport on edera/6.18-lts, or, if it must never go there, by labelling its mainline pull request no-backport or listing its subject in .automation/backport-skip on edera/6.18-lts. A commit that was ported under another subject or in another form also goes in the skip file. This issue is updated weekly and closes itself once nothing is missing.
Claude's triage
Of the 16 commits listed, 4 are really missing from edera/6.18-lts and have no open backport pull request: one gntdev fix whose mainline PR #81 is backport-waiting, and three OpenPaX hardening fixes that have no backport at all. The other 12 are in flight in open LTS pull requests #82, #83, #85, #91, #92, #110 and #111. None needs a maintainer to skip it. (Open PR #113 carries the LTS series as a rebase. It does not carry any of the listed subjects.)
3f7ca89e3d70 xen/gntdev: don't copy to user space under priv->lock (#79)
In flight (#83)
Open LTS pull request #83 carries this change. Review and merge it.
1dc86dd35ab6 x86/xen: clear the user pgd pointer when freeing it
In flight (#82)
Open LTS pull request #82 carries this change. Review and merge it.
e79780b9d1b6 xen/gntdev: reset the copy batch's writeable flag (#81)
Missing
This makes gntdev_ioctl_grant_copy() reset batch->writeable together with nr_ops/nr_pages, so that a newly kmalloc'd batch never reaches gntdev_put_pages() with an uninitialised bool (a UBSAN invalid load). On LTS_TIP the function, at drivers/xen/gntdev.c about line 1111, resets only nr_ops and nr_pages. The only places that reset writeable are gntdev_put_pages() and the source-virt branch of gntdev_grant_copy_seg(), and neither covers a batch whose first segment fails. It is a one-line change with identical context on 6.18, so it should apply as is. No other listed commit depends on it. Backport it: mainline PR #81 is backport-waiting.
8f315d39b605 xen/balloon: let the OOM notifier grow an empty domU balloon (#84)
In flight (#85)
Open LTS pull request #85 carries this change. Review and merge it.
ec5588ab8435 x86: fault: emulate trampolines that start with endbr (#87)
In flight (#91)
Open LTS pull request #91 carries this change. Review and merge it. Its hunks change only the bodies of openpax_emulate_trampoline_32/64(), so they do not overlap with 77c9976b13e0 below.
7d91f39740ed xen-pciback: free a device's config fields before its dev_data (#88)
In flight (#92)
Open LTS pull request #92 carries this change. Review and merge it.
bc97ba75c220 mm: never give the brk area execute permission under OpenPaX MPROTECT
Missing
This strips VM_EXEC/VM_MAYEXEC in do_brk_flags() when the mm has PAXF_MPROTECT. Without it, personality(READ_IMPLIES_EXEC) plus brk(), or an executable segment's bss, still produce an RWX mapping. LTS_TIP's do_brk_flags() in mm/vma.c has no such check, and no LTS Edera commit touches brk under MPROTECT. The prerequisites, mm->pax_flags and PAXF_MPROTECT, are present on LTS. It needs adapting: mainline uses the new vma_flags_* API (vma_flags_clear(&vma_flags, VMA_EXEC_BIT, VMA_MAYEXEC_BIT)), but 6.18's do_brk_flags() still takes a vm_flags_t. The 6.18 version is vm_flags &= ~(VM_EXEC | VM_MAYEXEC), placed after vm_flags |= VM_DATA_DEFAULT_FLAGS | .... No later listed commit depends on it. Backport it. It is a security fix and no backport has been proposed yet.
77c9976b13e0 x86/mm: emulate OpenPaX trampolines only when the fault would be SIGSEGV
Missing
This moves the EMUTRAMP hook out of the top of do_user_addr_fault() and into __bad_area_nosemaphore(), after local_irq_enable(). That way only fetches that would otherwise end in SIGSEGV get emulated, and legitimate demand-paged text no longer does. On LTS_TIP the hook is still in do_user_addr_fault() (arch/x86/mm/fault.c:1533), and __bad_area_nosemaphore() has no call to it. The context at both sites matches mainline, so it should apply as is. It does not depend textually on ec5588ab8435 (#91), whose hunks are disjoint from it. Backport it. No backport has been proposed yet.
93dbfdcf5437 openpax: do not enable trampoline emulation for every program by default
Missing
This changes OPENPAX_EMUTRAMP_DEFAULT to default n and rewrites its help text. On LTS_TIP, security/Kconfig.openpax still has default y, and the file is identical to mainline's just before this commit. It should apply as is. No later listed commit depends on it. Backport it. No backport has been proposed yet. Also check whether the Edera LTS host config sets this option explicitly, because a Kconfig default change does not affect a config that sets the option itself.
a4543cd54a6c iommu/xen: keep reserving the PCI host-bridge windows in dom0
In flight (#110)
Open LTS pull request #110 carries this change. Review and merge it.
8cd845532e3c xen/pcifront: do not promise pcifront_machine_sbdf() to built-in code when modular
In flight (#110)
Open LTS pull request #110 carries this change. Review and merge it.
25ec4b49a0e6 9p/xen: only let an errno come back from a control request's Rlerror
In flight (#111)
Open LTS pull request #111 carries this change. Review and merge it.
ce62f82d2094 9p/xen: do not dereference priv->client while replaying a saved session
In flight (#111)
Open LTS pull request #111 carries this change. Review and merge it.
78700e1ae0db 9p/xen: keep a saved session across a suspend that has nothing to save
In flight (#111)
Open LTS pull request #111 carries this change. Review and merge it.
85e1da02b838 9p/xen: end foreign access on the grant refs we granted, not the intf page's
In flight (#111)
Open LTS pull request #111 carries this change. Review and merge it.
7ad1695c933c 9p/xen: order the control reply buffer before the pending flag
In flight (#111)
Open LTS pull request #111 carries this change. Review and merge it.
Activity
github-actions commentedon Nov 16, 2026
The set of missing commits changed; the description is updated. Audit run.