Skip to content

L
linux
Switch branch/tag
  1. 23 Jun, 2023 36 commits
  3. 22 Jun, 2023 4 commits
    • Paolo Bonzini's avatar
      Merge tag 'kvmarm-fixes-6.4-4' of... · 2623b3dc
      Paolo Bonzini authored 
      Merge tag 'kvmarm-fixes-6.4-4' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmarm into HEAD

KVM/arm64 fixes for 6.4, take #4

- Correctly save/restore PMUSERNR_EL0 when host userspace is using
  PMU counters directly

- Fix GICv2 emulation on GICv3 after the locking rework

- Don't use smp_processor_id() in kvm_pmu_probe_armpmu(), and
  document why...
      2623b3dc
    • Gavin Shan's avatar
      KVM: Avoid illegal stage2 mapping on invalid memory slot · 2230f9e1
      Gavin Shan authored 
      We run into guest hang in edk2 firmware when KSM is kept as running on
the host. The edk2 firmware is waiting for status 0x80 from QEMU's pflash
device (TYPE_PFLASH_CFI01) during the operation of sector erasing or
buffered write. The status is returned by reading the memory region of
the pflash device and the read request should have been forwarded to QEMU
and emulated by it. Unfortunately, the read request is covered by an
illegal stage2 mapping when the guest hang issue occurs. The read request
is completed with QEMU bypassed and wrong status is fetched. The edk2
firmware runs into an infinite loop with the wrong status.

The illegal stage2 mapping is populated due to same page sharing by KSM
at (C) even the associated memory slot has been marked as invalid at (B)
when the memory slot is requested to be deleted. It's notable that the
active and inactive memory slots can't be swapped when we're in the middle
of kvm_mmu_notifier_change_pte() because kvm->mn_active_invalidate_count
is elevated, and kvm_swap_active_memslots() will busy loop until it reaches
to zero again. Besides, the swapping from the active to the inactive memory
slots is also avoided by holding &kvm->srcu in __kvm_handle_hva_range(),
corresponding to synchronize_srcu_expedited() in kvm_swap_active_memslots().

  CPU-A                    CPU-B
  -----                    -----
                           ioctl(kvm_fd, KVM_SET_USER_MEMORY_REGION)
                           kvm_vm_ioctl_set_memory_region
                           kvm_set_memory_region
                           __kvm_set_memory_region
                           kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE)
                             kvm_invalidate_memslot
                               kvm_copy_memslot
                               kvm_replace_memslot
                               kvm_swap_active_memslots        (A)
                               kvm_arch_flush_shadow_memslot   (B)
  same page sharing by KSM
  kvm_mmu_notifier_invalidate_range_start
        :
  kvm_mmu_notifier_change_pte
    kvm_handle_hva_range
    __kvm_handle_hva_range
    kvm_set_spte_gfn            (C)
        :
  kvm_mmu_notifier_invalidate_range_end

Fix the issue by skipping the invalid memory slot at (C) to avoid the
illegal stage2 mapping so that the read request for the pflash's status
is forwarded to QEMU and emulated by it. In this way, the correct pflash's
status can be returned from QEMU to break the infinite loop in the edk2
firmware.

We tried a git-bisect and the first problematic commit is cd4c7183 ("
KVM: arm64: Convert to the gfn-based MMU notifier callbacks"). With this,
clean_dcache_guest_page() is called after the memory slots are iterated
in kvm_mmu_notifier_change_pte(). clean_dcache_guest_page() is called
before the iteration on the memory slots before this commit. This change
literally enlarges the racy window between kvm_mmu_notifier_change_pte()
and memory slot removal so that we're able to reproduce the issue in a
practical test case. However, the issue exists since commit d5d8184d
("KVM: ARM: Memory virtualization setup").

Cc: stable@vger.kernel.org # v3.9+
Fixes: d5d8184d ("KVM: ARM: Memory virtualization setup")
Reported-by: default avatarShuai Hu <hshuai@redhat.com>
Reported-by: default avatarZhenyu Zhang <zhenyzha@redhat.com>
Signed-off-by: default avatarGavin Shan <gshan@redhat.com>
Reviewed-by: default avatarDavid Hildenbrand <david@redhat.com>
Reviewed-by: default avatarOliver Upton <oliver.upton@linux.dev>
Reviewed-by: default avatarPeter Xu <peterx@redhat.com>
Reviewed-by: default avatarSean Christopherson <seanjc@google.com>
Reviewed-by: default avatarShaoqin Huang <shahuang@redhat.com>
Message-Id: <20230615054259.14911-1-gshan@redhat.com>
Signed-off-by: default avatarPaolo Bonzini <pbonzini@redhat.com>
      2230f9e1
    • Paolo Abeni's avatar
      Merge tag 'nf-23-06-21' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf · 2ba7e7eb
      Paolo Abeni authored 
      Pablo Neira Ayuso says:

====================
Netfilter/IPVS fixes for net

This is v3, including a crash fix for patch 01/14.

The following patchset contains Netfilter/IPVS fixes for net:

1) Fix UDP segmentation with IPVS tunneled traffic, from Terin Stock.

2) Fix chain binding transaction logic, add a bound flag to rule
   transactions. Remove incorrect logic in nft_data_hold() and
   nft_data_release().

3) Add a NFT_TRANS_PREPARE_ERROR deactivate state to deal with releasing
   the set/chain as a follow up to 1240eb93 ("netfilter: nf_tables:
   incorrect error path handling with NFT_MSG_NEWRULE")

4) Drop map element references from preparation phase instead of
   set destroy path, otherwise bogus EBUSY with transactions such as:

        flush chain ip x y
        delete chain ip x w

   where chain ip x y contains jump/goto from set elements.

5) Pipapo set type does not regard generation mask from the walk
   iteration.

6) Fix reference count underflow in set element reference to
   stateful object.

7) Several patches to tighten the nf_tables API:
   - disallow set element updates of bound anonymous set
   - disallow unbound anonymous set/chain at the end of transaction.
   - disallow updates of anonymous set.
   - disallow timeout configuration for anonymous sets.

8) Fix module reference leak in chain updates.

9) Fix nfnetlink_osf module autoload.

10) Fix deletion of basechain when NFTA_CHAIN_HOOK is specified as
    in iptables-nft.

This Netfilter batch is larger than usual at this stage, I am aware we
are fairly late in the -rc cycle, if you prefer to route them through
net-next, please let me know.

netfilter pull request 23-06-21

* tag 'nf-23-06-21' of git://git.kernel.org/pub/scm/linux/kernel/git/netfilter/nf:
  netfilter: nf_tables: Fix for deleting base chains with payload
  netfilter: nfnetlink_osf: fix module autoload
  netfilter: nf_tables: drop module reference after updating chain
  netfilter: nf_tables: disallow timeout for anonymous sets
  netfilter: nf_tables: disallow updates of anonymous sets
  netfilter: nf_tables: reject unbound chain set before commit phase
  netfilter: nf_tables: reject unbound anonymous set before commit phase
  netfilter: nf_tables: disallow element updates of bound anonymous sets
  netfilter: nf_tables: fix underflow in object reference counter
  netfilter: nft_set_pipapo: .walk does not deal with generations
  netfilter: nf_tables: drop map element references from preparation phase
  netfilter: nf_tables: add NFT_TRANS_PREPARE_ERROR to deal with bound set/chain
  netfilter: nf_tables: fix chain binding transaction logic
  ipvs: align inner_mac_header for encapsulation
====================

Link: https://lore.kernel.org/r/20230621100731.68068-1-pablo@netfilter.orgSigned-off-by: default avatarPaolo Abeni <pabeni@redhat.com>
      2ba7e7eb
    • Maciej Żenczykowski's avatar
      revert "net: align SO_RCVMARK required privileges with SO_MARK" · a9628e88
      Maciej Żenczykowski authored 
      This reverts commit 1f86123b ("net: align SO_RCVMARK required
privileges with SO_MARK") because the reasoning in the commit message
is not really correct:
  SO_RCVMARK is used for 'reading' incoming skb mark (via cmsg), as such
  it is more equivalent to 'getsockopt(SO_MARK)' which has no priv check
  and retrieves the socket mark, rather than 'setsockopt(SO_MARK) which
  sets the socket mark and does require privs.

  Additionally incoming skb->mark may already be visible if
  sysctl_fwmark_reflect and/or sysctl_tcp_fwmark_accept are enabled.

  Furthermore, it is easier to block the getsockopt via bpf
  (either cgroup setsockopt hook, or via syscall filters)
  then to unblock it if it requires CAP_NET_RAW/ADMIN.

On Android the socket mark is (among other things) used to store
the network identifier a socket is bound to.  Setting it is privileged,
but retrieving it is not.  We'd like unprivileged userspace to be able
to read the network id of incoming packets (where mark is set via
iptables [to be moved to bpf])...

An alternative would be to add another sysctl to control whether
setting SO_RCVMARK is privilged or not.
(or even a MASK of which bits in the mark can be exposed)
But this seems like over-engineering...

Note: This is a non-trivial revert, due to later merged commit e42c7bee
("bpf: net: Consider has_current_bpf_ctx() when testing capable() in sk_setsockopt()")
which changed both 'ns_capable' into 'sockopt_ns_capable' calls.

Fixes: 1f86123b ("net: align SO_RCVMARK required privileges with SO_MARK")
Cc: Larysa Zaremba <larysa.zaremba@intel.com>
Cc: Simon Horman <simon.horman@corigine.com>
Cc: Paolo Abeni <pabeni@redhat.com>
Cc: Eyal Birger <eyal.birger@gmail.com>
Cc: Jakub Kicinski <kuba@kernel.org>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Patrick Rohr <prohr@google.com>
Signed-off-by: default avatarMaciej Żenczykowski <maze@google.com>
Reviewed-by: default avatarSimon Horman <simon.horman@corigine.com>
Reviewed-by: default avatarKuniyuki Iwashima <kuniyu@amazon.com>
Link: https://lore.kernel.org/r/20230618103130.51628-1-maze@google.comSigned-off-by: default avatarPaolo Abeni <pabeni@redhat.com>
      a9628e88
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7