Darrick J. Wong [Mon, 12 Jul 2021 19:58:48 +0000 (12:58 -0700)]
xfs: improve FSGROWFSRT precondition checking
Improve the checking at the start of a realtime grow operation so that
we avoid accidentally set a new extent size that is too large and avoid
adding an rt volume to a filesystem with rmap or reflink because we
don't support rt rmap or reflink yet.
While we're at it, separate the checks so that we're only testing one
aspect at a time.
Darrick J. Wong [Mon, 12 Jul 2021 19:58:51 +0000 (12:58 -0700)]
xfs: don't expose misaligned extszinherit hints to userspace
Commit 603f000b15f2 changed xfs_ioctl_setattr_check_extsize to reject an
attempt to set an EXTSZINHERIT extent size hint on a directory with
RTINHERIT set if the hint isn't a multiple of the realtime extent size.
However, I have recently discovered that it is possible to change the
realtime extent size when adding a rt device to a filesystem, which
means that the existence of directories with misaligned inherited hints
is not an accident.
As a result, it's possible that someone could have set a valid hint and
added an rt volume with a different rt extent size, which invalidates
the ondisk hints. After such a sequence, FSGETXATTR will report a
misaligned hint, which FSSETXATTR will trip over, causing confusion if
the user was doing the usual GET/SET sequence to change some other
attribute. Change xfs_fill_fsxattr to omit the hint if it isn't aligned
properly.
Fixes: 603f000b15f2 ("xfs: validate extsz hints against rt extent size when rtinherit is set") Signed-off-by: Darrick J. Wong <[email protected]> Reviewed-by: Christoph Hellwig <[email protected]>
Darrick J. Wong [Mon, 12 Jul 2021 19:58:50 +0000 (12:58 -0700)]
xfs: correct the narrative around misaligned rtinherit/extszinherit dirs
While auditing the realtime growfs code, I realized that the GROWFSRT
ioctl (and by extension xfs_growfs) has always allowed sysadmins to
change the realtime extent size when adding a realtime section to the
filesystem. Since we also have always allowed sysadmins to set
RTINHERIT and EXTSZINHERIT on directories even if there is no realtime
device, this invalidates the premise laid out in the comments added in
commit 603f000b15f2.
In other words, this is not a case of inadequate metadata validation.
This is a case of nearly forgotten (and apparently untested) but
supported functionality. Update the comments to reflect what we've
learned, and remove the log message about correcting the misalignment.
Fixes: 603f000b15f2 ("xfs: validate extsz hints against rt extent size when rtinherit is set") Signed-off-by: Darrick J. Wong <[email protected]> Reviewed-by: Carlos Maiolino <[email protected]> Reviewed-by: Christoph Hellwig <[email protected]>
Darrick J. Wong [Mon, 12 Jul 2021 19:58:48 +0000 (12:58 -0700)]
xfs: reset child dir '..' entry when unlinking child
While running xfs/168, I noticed a second source of post-shrink
corruption errors causing shutdowns.
Let's say that directory B has a low inode number and is a child of
directory A, which has a high number. If B is empty but open, and
unlinked from A, B's dotdot link continues to point to A. If A is then
unlinked and the filesystem shrunk so that A is no longer a valid inode,
a subsequent AIL push of B will trip the inode verifiers because the
dotdot entry points outside of the filesystem.
To avoid this problem, reset B's dotdot entry to the root directory when
unlinking directories, since the root directory cannot be removed.
Darrick J. Wong [Mon, 12 Jul 2021 19:58:47 +0000 (12:58 -0700)]
xfs: check for sparse inode clusters that cross new EOAG when shrinking
While running xfs/168, I noticed occasional write verifier shutdowns
involving inodes at the very end of the filesystem. Existing inode
btree validation code checks that all inode clusters are fully contained
within the filesystem.
However, due to inadequate checking in the fs shrink code, it's possible
that there could be a sparse inode cluster at the end of the filesystem
where the upper inodes of the cluster are marked as holes and the
corresponding blocks are free. In this case, the last blocks in the AG
are listed in the bnobt. This enables the shrink to proceed but results
in a filesystem that trips the inode verifiers. Fix this by disallowing
the shrink.
iomap: Don't create iomap_page objects for inline files
In iomap_readpage_actor, don't create iop objects for inline inodes.
Otherwise, iomap_read_inline_data will set PageUptodate without setting
iop->uptodate, and iomap_page_release will eventually complain.
To prevent this kind of bug from occurring in the future, make sure the
page doesn't have private data attached in iomap_read_inline_data.
iomap: Permit pages without an iop to enter writeback
Create an iop in the writeback path if one doesn't exist. This allows us
to avoid creating the iop in some cases. We'll initially do that for pages
with inline data, but it can be extended to pages which are entirely within
an extent. It also allows for an iop to be removed from pages in the
future (eg page split).
iomap: remove the length variable in iomap_seek_hole
The length variable is rather pointless given that it can be trivially
deduced from offset and size. Also the initial calculation can lead
to KASAN warnings.
iomap: remove the length variable in iomap_seek_data
The length variable is rather pointless given that it can be trivially
deduced from offset and size. Also the initial calculation can lead
to KASAN warnings.
Mark Rutland [Thu, 15 Jul 2021 12:30:49 +0000 (13:30 +0100)]
arm64: entry: fix KCOV suppression
We suppress KCOV for entry.o rather than entry-common.o. As entry.o is
built from entry.S, this is pointless, and permits instrumentation of
entry-common.o, which is built from entry-common.c.
Fix the Makefile to suppress KCOV for entry-common.o, as we had intended
to begin with. I've verified with objdump that this is working as
expected.
Mark Rutland [Wed, 14 Jul 2021 17:28:01 +0000 (18:28 +0100)]
arm64: entry: add missing noinstr
We intend that all the early exception handling code is marked as
`noinstr`, but we forgot this for __el0_error_handler_common(), which is
called before we have completed entry from user mode. If it were
instrumented, we could run into problems with RCU, lockdep, etc.
Mark it as `noinstr` to prevent this.
The few other functions in entry-common.c which do not have `noinstr` are
called once we've completed entry, and are safe to instrument.
Mark Rutland [Wed, 14 Jul 2021 14:38:41 +0000 (15:38 +0100)]
arm64: mte: fix restoration of GCR_EL1 from suspend
Since commit:
bad1e1c663e0a72f ("arm64: mte: switch GCR_EL1 in kernel entry and exit")
we saved/restored the user GCR_EL1 value at exception boundaries, and
update_gcr_el1_excl() is no longer used for this. However it is used to
restore the kernel's GCR_EL1 value when returning from a suspend state.
Thus, the comment is misleading (and an ISB is necessary).
When restoring the kernel's GCR value, we need an ISB to ensure this is
used by subsequent instructions. We don't necessarily get an ISB by
other means (e.g. if the kernel is built without support for pointer
authentication). As __cpu_setup() initialised GCR_EL1.Exclude to 0xffff,
until a context synchronization event, allocation tag 0 may be used
rather than the desired set of tags.
This patch drops the misleading comment, adds the missing ISB, and for
clarity folds update_gcr_el1_excl() into its only user.
Robin Murphy [Mon, 12 Jul 2021 14:27:46 +0000 (15:27 +0100)]
arm64: Avoid premature usercopy failure
Al reminds us that the usercopy API must only return complete failure
if absolutely nothing could be copied. Currently, if userspace does
something silly like giving us an unaligned pointer to Device memory,
or a size which overruns MTE tag bounds, we may fail to honour that
requirement when faulting on a multi-byte access even though a smaller
access could have succeeded.
Add a mitigation to the fixup routines to fall back to a single-byte
copy if we faulted on a larger access before anything has been written
to the destination, to guarantee making *some* forward progress. We
needn't be too concerned about the overall performance since this should
only occur when callers are doing something a bit dodgy in the first
place. Particularly broken userspace might still be able to trick
generic_perform_write() into an infinite loop by targeting write() at
an mmap() of some read-only device register where the fault-in load
succeeds but any store synchronously aborts such that copy_to_user() is
genuinely unable to make progress, but, well, don't do that...
xen-blkfront has a weird protocol where close message from the remote
side can be delayed, and where hot removals are treated somewhat
differently from regular removals, all leading to potential NULL
pointer removals, and a del_gendisk from the block device release
method, which will deadlock. Fix this by just performing normal hot
removals even when the device is opened like all other Linux block
drivers.
Merge tag 'nvme-5.14-2021-07-15' of git://git.infradead.org/nvme into block-5.14
Pull NVMe fixes from Christoph:
"nvme fixes for Linux 5.14
- fix various races in nvme-pci when shutting down just after probing
(Casey Chen)
- fix a net_device leak in nvme-tcp (Prabhakar Kushwaha)"
* tag 'nvme-5.14-2021-07-15' of git://git.infradead.org/nvme:
nvme-pci: do not call nvme_dev_remove_admin from nvme_remove
nvme-pci: fix multiple races in nvme_setup_io_queues
nvme-tcp: use __dev_get_by_name instead dev_get_by_name for OPT_HOST_IFACE
Rob Herring [Tue, 13 Jul 2021 19:34:53 +0000 (13:34 -0600)]
dt-bindings: More dropping redundant minItems/maxItems
Another round of removing redundant minItems/maxItems from new schema in
the recent merge window.
If a property has an 'items' list, then a 'minItems' or 'maxItems' with the
same size as the list is redundant and can be dropped. Note that is DT
schema specific behavior and not standard json-schema behavior. The tooling
will fixup the final schema adding any unspecified minItems/maxItems.
This condition is partially checked with the meta-schema already, but
only if both 'minItems' and 'maxItems' are equal to the 'items' length.
An improved meta-schema is pending.
Vitaly Kuznetsov [Mon, 28 Jun 2021 10:44:25 +0000 (12:44 +0200)]
KVM: selftests: smm_test: Test SMM enter from L2
Two additional tests are added:
- SMM triggered from L2 does not currupt L1 host state.
- Save/restore during SMM triggered from L2 does not corrupt guest/host
state.
Vitaly Kuznetsov [Mon, 28 Jun 2021 10:44:24 +0000 (12:44 +0200)]
KVM: nSVM: Restore nested control upon leaving SMM
If the VM was migrated while in SMM, no nested state was saved/restored,
and therefore svm_leave_smm has to load both save and control area
of the vmcb12. Save area is already loaded from HSAVE area,
so now load the control area as well from the vmcb12.
Vitaly Kuznetsov [Mon, 28 Jun 2021 10:44:23 +0000 (12:44 +0200)]
KVM: nSVM: Fix L1 state corruption upon return from SMM
VMCB split commit 4995a3685f1b ("KVM: SVM: Use a separate vmcb for the
nested L2 guest") broke return from SMM when we entered there from guest
(L2) mode. Gen2 WS2016/Hyper-V is known to do this on boot. The problem
manifests itself like this:
Note: return to L2 succeeded but upon first exit to L1 its RIP points to
'RSM' instruction but we're not in SMM.
The problem appears to be that VMCB01 gets irreversibly destroyed during
SMM execution. Previously, we used to have 'hsave' VMCB where regular
(pre-SMM) L1's state was saved upon nested_svm_vmexit() but now we just
switch to VMCB01 from VMCB02.
Pre-split (working) flow looked like:
- SMM is triggered during L2's execution
- L2's state is pushed to SMRAM
- nested_svm_vmexit() restores L1's state from 'hsave'
- SMM -> RSM
- enter_svm_guest_mode() switches to L2 but keeps 'hsave' intact so we have
pre-SMM (and pre L2 VMRUN) L1's state there
- L2's state is restored from SMRAM
- upon first exit L1's state is restored from L1.
This was always broken with regards to svm_get_nested_state()/
svm_set_nested_state(): 'hsave' was never a part of what's being
save and restored so migration happening during SMM triggered from L2 would
never restore L1's state correctly.
Post-split flow (broken) looks like:
- SMM is triggered during L2's execution
- L2's state is pushed to SMRAM
- nested_svm_vmexit() switches to VMCB01 from VMCB02
- SMM -> RSM
- enter_svm_guest_mode() switches from VMCB01 to VMCB02 but pre-SMM VMCB01
is already lost.
- L2's state is restored from SMRAM
- upon first exit L1's state is restored from VMCB01 but it is corrupted
(reflects the state during 'RSM' execution).
VMX doesn't have this problem because unlike VMCB, VMCS keeps both guest
and host state so when we switch back to VMCS02 L1's state is intact there.
To resolve the issue we need to save L1's state somewhere. We could've
created a third VMCB for SMM but that would require us to modify saved
state format. L1's architectural HSAVE area (pointed by MSR_VM_HSAVE_PA)
seems appropriate: L0 is free to save any (or none) of L1's state there.
Currently, KVM does 'none'.
Note, for nested state migration to succeed, both source and destination
hypervisors must have the fix. We, however, don't need to create a new
flag indicating the fact that HSAVE area is now populated as migration
during SMM triggered from L2 was always broken.
Fixes: 4995a3685f1b ("KVM: SVM: Use a separate vmcb for the nested L2 guest") Signed-off-by: Vitaly Kuznetsov <[email protected]> Signed-off-by: Paolo Bonzini <[email protected]>
Vitaly Kuznetsov [Mon, 28 Jun 2021 10:44:22 +0000 (12:44 +0200)]
KVM: nSVM: Introduce svm_copy_vmrun_state()
Separate the code setting non-VMLOAD-VMSAVE state from
svm_set_nested_state() into its own function. This is going to be
re-used from svm_enter_smm()/svm_leave_smm().
Vitaly Kuznetsov [Mon, 28 Jun 2021 10:44:21 +0000 (12:44 +0200)]
KVM: nSVM: Check that VM_HSAVE_PA MSR was set before VMRUN
APM states that "The address written to the VM_HSAVE_PA MSR, which holds
the address of the page used to save the host state on a VMRUN, must point
to a hypervisor-owned page. If this check fails, the WRMSR will fail with
a #GP(0) exception. Note that a value of 0 is not considered valid for the
VM_HSAVE_PA MSR and a VMRUN that is attempted while the HSAVE_PA is 0 will
fail with a #GP(0) exception."
svm_set_msr() already checks that the supplied address is valid, so only
check for '0' is missing. Add it to nested_svm_vmrun().
Vitaly Kuznetsov [Mon, 28 Jun 2021 10:44:20 +0000 (12:44 +0200)]
KVM: nSVM: Check the value written to MSR_VM_HSAVE_PA
APM states that #GP is raised upon write to MSR_VM_HSAVE_PA when
the supplied address is not page-aligned or is outside of "maximum
supported physical address for this implementation".
page_address_valid() check seems suitable. Also, forcefully page-align
the address when it's written from VMM.
KVM: SVM: Fix sev_pin_memory() error checks in SEV migration utilities
Use IS_ERR() instead of checking for a NULL pointer when querying for
sev_pin_memory() failures. sev_pin_memory() always returns an error code
cast to a pointer, or a valid pointer; it never returns NULL.
KVM: SVM: add module param to control the #SMI interception
In theory there are no side effects of not intercepting #SMI,
because then #SMI becomes transparent to the OS and the KVM.
Plus an observation on recent Zen2 CPUs reveals that these
CPUs ignore #SMI interception and never deliver #SMI VMexits.
This is also useful to test nested KVM to see that L1
handles #SMIs correctly in case when L1 doesn't intercept #SMI.
Finally the default remains the same, the SMI are intercepted
by default thus this patch doesn't have any effect unless
non default module param value is used.
Kernel never sends real INIT even to CPUs, other than on boot.
Thus INIT interception is an error which should be caught
by a check for an unknown VMexit reason.
On top of that, the current INIT VM exit handler skips
the current instruction which is wrong.
That was added in commit 5ff3a351f687 ("KVM: x86: Move trivial
instruction-based exit handlers to common code").
KVM: SVM: #SMI interception must not skip the instruction
Commit 5ff3a351f687 ("KVM: x86: Move trivial instruction-based
exit handlers to common code"), unfortunately made a mistake of
treating nop_on_interception and nop_interception in the same way.
Former does truly nothing while the latter skips the instruction.
SMI VM exit handler should do nothing.
(SMI itself is handled by the host when we do STGI)
vmx_msr_index was used to record the list of MSRs which can be lazily
restored when kvm returns to userspace. It is now reimplemented as
kvm_uret_msrs_list, a common x86 list which is only used inside x86.c.
So just remove the obsolete declaration in vmx.h.
Lai Jiangshan [Mon, 28 Jun 2021 17:26:32 +0000 (01:26 +0800)]
KVM: X86: Disable hardware breakpoints unconditionally before kvm_x86->run()
When the host is using debug registers but the guest is not using them
nor is the guest in guest-debug state, the kvm code does not reset
the host debug registers before kvm_x86->run(). Rather, it relies on
the hardware vmentry instruction to automatically reset the dr7 registers
which ensures that the host breakpoints do not affect the guest.
This however violates the non-instrumentable nature around VM entry
and exit; for example, when a host breakpoint is set on vcpu->arch.cr2,
Another issue is consistency. When the guest debug registers are active,
the host breakpoints are reset before kvm_x86->run(). But when the
guest debug registers are inactive, the host breakpoints are delayed to
be disabled. The host tracing tools may see different results depending
on what the guest is doing.
To fix the problems, we clear %db7 unconditionally before kvm_x86->run()
if the host has set any breakpoints, no matter if the guest is using
them or not.
KVM: selftests: Address extra memslot parameters in vm_vaddr_alloc
Commit a75a895e6457 ("KVM: selftests: Unconditionally use memslot 0 for
vaddr allocations") removed the memslot parameters from vm_vaddr_alloc.
It addressed all callers except one under lib/aarch64/, due to a race
with commit e3db7579ef35 ("KVM: selftests: Add exception handling
support for aarch64")
Fix the vm_vaddr_alloc call in lib/aarch64/processor.c.
Pavel Skripkin [Thu, 1 Jul 2021 19:55:00 +0000 (22:55 +0300)]
kvm: debugfs: fix memory leak in kvm_create_vm_debugfs
In commit bc9e9e672df9 ("KVM: debugfs: Reuse binary stats descriptors")
loop for filling debugfs_stat_data was copy-pasted 2 times, but
in the second loop pointers are saved over pointers allocated
in the first loop. All this causes is a memory leak, fix it.
LICENSES/dual/CC-BY-4.0: Git rid of "smart quotes"
A couple of exotic quote characters came in with this license text; they
can confuse software that is not expecting non-ASCII text. Switch to
normal quotes here, with no changes to the actual license text.
James Clark [Thu, 24 Jun 2021 16:43:02 +0000 (17:43 +0100)]
perf cs-etm: Split Coresight decode by aux records
Populate the auxtrace queues using AUX records rather than whole
auxtrace buffers so that the decoder is reset between each aux record.
This is similar to the auxtrace_queues__process_index() ->
auxtrace_queues__add_indexed_event() flow where
perf_session__peek_event() is used to read AUXTRACE events out of random
positions in the file based on the auxtrace index.
But now we loop over all PERF_RECORD_AUX events instead of AUXTRACE
buffers. For each PERF_RECORD_AUX event, we find the corresponding
AUXTRACE buffer using the index, and add a fragment of that buffer to
the auxtrace queues.
No other changes to decoding were made, apart from populating the
auxtrace queues. The result of decoding is identical to before, except
in cases where decoding failed completely, due to not resetting the
decoder.
The reason for this change is because AUX records are emitted any time
tracing is disabled, for example when the process is scheduled out.
Because ETM was disabled and enabled again, the decoder also needs to be
reset to force the search for a sync packet. Otherwise there would be
fatal decoding errors.
Testing
=======
Testing was done with the following script, to diff the decoding results
between the patched and un-patched versions of perf:
* Per-cpu explicit threads:
perf record -e cs_etm/@tmc_etr0/u --pid 853,854
* System-wide (per-cpu):
perf record -e cs_etm/@tmc_etr0/u -a
* No data collected (no aux buffers)
Can happen with any command when run for a short period
* Containing truncated records
Can happen with any command
* Containing aux records with 0 size
Can happen with any command
* Snapshot mode (various files with and without buffer wrap)
perf record -e cs_etm/@tmc_etr0/u -a --snapshot
Some differences were observed in the following scenario:
* Snapshot mode (with duplicate buffers)
perf record -e cs_etm/@tmc_etr0/u -a --snapshot
Fewer samples are generated in snapshot mode if duplicate buffers
were gathered because buffers with the same offset are now only added
once. This gives different, but more correct results and no duplicate
data is decoded any more.
tools headers: Remove broken definition of __LITTLE_ENDIAN
The linux/kconfig.h file was copied from the kernel but the line where
with the generated/autoconf.h include from where the CONFIG_ entries
would come from was deleted, as tools/ build system don't create that
file, so we ended up always defining just __LITTLE_ENDIAN as
CONFIG_CPU_BIG_ENDIAN was nowhere to be found.
This in turn ended up breaking the build in some systems where
__LITTLE_ENDIAN was already defined, such as the androind NDK.
So just ditch that block that depends on the CONFIG_CPU_BIG_ENDIAN
define.
The kconfig.h file was copied just to get IS_ENABLED() and a
'make -C tools/all' doesn't breaks with this removal.
- ipv6:
- allocate enough headroom in ip6_finish_output2() in case
iptables TEE is used
- tcp: drop silly ICMPv6 packet too big messages to avoid
expensive and pointless lookups (which may serve as a DDOS
vector)
- make sure fwmark is copied in SYNACK packets
- fix 'disable_policy' for forwarded packets (align with IPv4)
- netfilter: conntrack:
- do not renew entry stuck in tcp SYN_SENT state
- do not mark RST in the reply direction coming after SYN packet
for an out-of-sync entry
- mptcp: cleanly handle error conditions with MP_JOIN and syncookies
- mptcp: fix double free when rejecting a join due to port mismatch
- validate lwtstate->data before returning from skb_tunnel_info()
- tcp: call sk_wmem_schedule before sk_mem_charge in zerocopy path
- mt76: mt7921: continue to probe driver when fw already downloaded
- bonding: fix multiple issues with offloading IPsec to (thru?) bond
- stmmac: ptp: fix issues around Qbv support and setting time back
- bcmgenet: always clear wake-up based on energy detection
Misc:
- sctp: move 198 addresses from unusable to private scope
- ptp: support virtual clocks and timestamping
- openvswitch: optimize operation for key comparison"
* tag 'net-5.14-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (158 commits)
net: dsa: properly check for the bridge_leave methods in dsa_switch_bridge_leave()
sfc: add logs explaining XDP_TX/REDIRECT is not available
sfc: ensure correct number of XDP queues
sfc: fix lack of XDP TX queues - error XDP TX failed (-22)
net: fddi: fix UAF in fza_probe
net: dsa: sja1105: fix address learning getting disabled on the CPU port
net: ocelot: fix switchdev objects synced for wrong netdev with LAG offload
net: Use nlmsg_unicast() instead of netlink_unicast()
octeontx2-pf: Fix uninitialized boolean variable pps
ipv6: allocate enough headroom in ip6_finish_output2()
net: hdlc: rename 'mod_init' & 'mod_exit' functions to be module-specific
net: bridge: multicast: fix MRD advertisement router port marking race
net: bridge: multicast: fix PIM hello router port marking race
net: phy: marvell10g: fix differentiation of 88X3310 from 88X3340
dsa: fix for_each_child.cocci warnings
virtio_net: check virtqueue_add_sgs() return value
mptcp: properly account bulk freed memory
selftests: mptcp: fix case multiple subflows limited by server
mptcp: avoid processing packet if a subflow reset
mptcp: fix syncookie process if mptcp can not_accept new subflow
...
Add a simple helper that filesystems can use in their parameter parser
to parse the "source" parameter. A few places open-coded this function
and that already caused a bug in the cgroup v1 parser that we fixed.
Let's make it harder to get this wrong by introducing a helper which
performs all necessary checks.
The following sequence can be used to trigger a UAF:
int fscontext_fd = fsopen("cgroup");
int fd_null = open("/dev/null, O_RDONLY);
int fsconfig(fscontext_fd, FSCONFIG_SET_FD, "source", fd_null);
close_range(3, ~0U, 0);
The cgroup v1 specific fs parser expects a string for the "source"
parameter. However, it is perfectly legitimate to e.g. specify a file
descriptor for the "source" parameter. The fs parser doesn't know what
a filesystem allows there. So it's a bug to assume that "source" is
always of type fs_value_is_string when it can reasonably also be
fs_value_is_file.
This assumption in the cgroup code causes a UAF because struct
fs_parameter uses a union for the actual value. Access to that union is
guarded by the param->type member. Since the cgroup paramter parser
didn't check param->type but unconditionally moved param->string into
fc->source a close on the fscontext_fd would trigger a UAF during
put_fs_context() which frees fc->source thereby freeing the file stashed
in param->file causing a UAF during a close of the fd_null.
Fix this by verifying that param->type is actually a string and report
an error if not.
In follow up patches I'll add a new generic helper that can be used here
and by other filesystems instead of this error-prone copy-pasta fix.
But fixing it in here first makes backporting a it to stable a lot
easier.
Kefeng Wang [Sat, 26 Jun 2021 07:03:04 +0000 (15:03 +0800)]
KVM: mmio: Fix use-after-free Read in kvm_vm_ioctl_unregister_coalesced_mmio
BUG: KASAN: use-after-free in kvm_vm_ioctl_unregister_coalesced_mmio+0x7c/0x1ec arch/arm64/kvm/../../../virt/kvm/coalesced_mmio.c:183
Read of size 8 at addr ffff0000c03a2500 by task syz-executor083/4269
If kvm_io_bus_unregister_dev() return -ENOMEM, we already call kvm_iodevice_destructor()
inside this function to delete 'struct kvm_coalesced_mmio_dev *dev' from list
and free the dev, but kvm_iodevice_destructor() is called again, it will lead
the above issue.
Let's check the the return value of kvm_io_bus_unregister_dev(), only call
kvm_iodevice_destructor() if the return value is 0.
KVM: SVM: Revert clearing of C-bit on GPA in #NPF handler
Don't clear the C-bit in the #NPF handler, as it is a legal GPA bit for
non-SEV guests, and for SEV guests the C-bit is dropped before the GPA
hits the NPT in hardware. Clearing the bit for non-SEV guests causes KVM
to mishandle #NPFs with that collide with the host's C-bit.
Although the APM doesn't explicitly state that the C-bit is not reserved
for non-SEV, Tom Lendacky confirmed that the following snippet about the
effective reduction due to the C-bit does indeed apply only to SEV guests.
Note that because guest physical addresses are always translated
through the nested page tables, the size of the guest physical address
space is not impacted by any physical address space reduction indicated
in CPUID 8000_001F[EBX]. If the C-bit is a physical address bit however,
the guest physical address space is effectively reduced by 1 bit.
And for SEV guests, the APM clearly states that the bit is dropped before
walking the nested page tables.
If the C-bit is an address bit, this bit is masked from the guest
physical address when it is translated through the nested page tables.
Consequently, the hypervisor does not need to be aware of which pages
the guest has chosen to mark private.
Note, the bogus C-bit clearing was removed from legacy #PF handler in
commit 6d1b867d0456 ("KVM: SVM: Don't strip the C-bit from CR2 on #PF
interception").
KVM: x86/mmu: Do not apply HPA (memory encryption) mask to GPAs
Ignore "dynamic" host adjustments to the physical address mask when
generating the masks for guest PTEs, i.e. the guest PA masks. The host
physical address space and guest physical address space are two different
beasts, e.g. even though SEV's C-bit is the same bit location for both
host and guest, disabling SME in the host (which clears shadow_me_mask)
does not affect the guest PTE->GPA "translation".
For non-SEV guests, not dropping bits is the correct behavior. Assuming
KVM and userspace correctly enumerate/configure guest MAXPHYADDR, bits
that are lost as collateral damage from memory encryption are treated as
reserved bits, i.e. KVM will never get to the point where it attempts to
generate a gfn using the affected bits. And if userspace wants to create
a bogus vCPU, then userspace gets to deal with the fallout of hardware
doing odd things with bad GPAs.
For SEV guests, not dropping the C-bit is technically wrong, but it's a
moot point because KVM can't read SEV guest's page tables in any case
since they're always encrypted. Not to mention that the current KVM code
is also broken since sme_me_mask does not have to be non-zero for SEV to
be supported by KVM. The proper fix would be to teach all of KVM to
correctly handle guest private memory, but that's a task for the future.
KVM: x86: Use kernel's x86_phys_bits to handle reduced MAXPHYADDR
Use boot_cpu_data.x86_phys_bits instead of the raw CPUID information to
enumerate the MAXPHYADDR for KVM guests when TDP is disabled (the guest
version is only relevant to NPT/TDP).
When using shadow paging, any reductions to the host's MAXPHYADDR apply
to KVM and its guests as well, i.e. using the raw CPUID info will cause
KVM to misreport the number of PA bits available to the guest.
Unconditionally zero out the "Physical Address bit reduction" entry.
For !TDP, the adjustment is already done, and for TDP enumerating the
host's reduction is wrong as the reduction does not apply to GPAs.
KVM: x86: Use guest MAXPHYADDR from CPUID.0x8000_0008 iff TDP is enabled
Ignore the guest MAXPHYADDR reported by CPUID.0x8000_0008 if TDP, i.e.
NPT, is disabled, and instead use the host's MAXPHYADDR. Per AMD'S APM:
Maximum guest physical address size in bits. This number applies only
to guests using nested paging. When this field is zero, refer to the
PhysAddrSize field for the maximum guest physical address size.
Revert "KVM: x86: WARN and reject loading KVM if NX is supported but not enabled"
Let KVM load if EFER.NX=0 even if NX is supported, the analysis and
testing (or lack thereof) for the non-PAE host case was garbage.
If the kernel won't be using PAE paging, .Ldefault_entry in head_32.S
skips over the entire EFER sequence. Hopefully that can be changed in
the future to allow KVM to require EFER.NX, but the motivation behind
KVM's requirement isn't yet merged. Reverting and revisiting the mess
at a later date is by far the safest approach.
Commit b78f4a59669 ("KVM: selftests: Rename vm_handle_exception")
raced with a couple of new x86 tests, missing two vm_handle_exception
to vm_install_exception_handler conversions.
Help the two broken tests to catch up with the new world.
With the recent fixes for fallthrough warnings, it is now possible to
enable -Wimplicit-fallthrough for Clang.
It's important to mention that since we have adopted the use of the
pseudo-keyword macro fallthrough; we also want to avoid having more
/* fall through */ comments being introduced. Notice that contrary
to GCC, Clang doesn't recognize any comments as implicit fall-through
markings when the -Wimplicit-fallthrough option is enabled. So, in
order to avoid having more comments being introduced, we have to use
the option -Wimplicit-fallthrough=5 for GCC, which similar to Clang,
will cause a warning in case a code comment is intended to be used
as a fall-through marking.
cruid should only be used for the initial mount and after this we should use the current
users credentials.
Ignore the original cruid mount argument when creating a new context for a multiuser mount
following a DFS link.
Shyam Prasad N [Tue, 18 May 2021 15:05:50 +0000 (15:05 +0000)]
cifs: use the expiry output of dns_query to schedule next resolution
We recently fixed DNS resolution of the server hostname during reconnect.
However, server IP address may change, even when the old one continues
to server (although sub-optimally).
We should schedule the next DNS resolution based on the TTL of
the DNS record used for the last resolution. This way, we resolve the
server hostname again when a DNS record expires.
CC util/pfm.o
util/pfm.c: In function ‘parse_libpfm_events_option’:
util/pfm.c:102:30: error: ‘struct evsel’ has no member named ‘leader’
102 | evsel->leader = grp_leader;
| ^~
Committer notes:
There is this entry in 'make -C tools/perf build-test' to test the build
with libpfm:
That is the filter expression attached to the raw_syscalls:sys_{enter,exit}
tracepoints.
$ grep memfd_secret tools/perf/arch/x86/entry/syscalls/syscall_64.tbl
447 common memfd_secret sys_memfd_secret
$
This addresses these perf build warnings:
Warning: Kernel ABI header at 'tools/arch/arm64/include/uapi/asm/unistd.h' differs from latest version at 'arch/arm64/include/uapi/asm/unistd.h'
diff -u tools/arch/arm64/include/uapi/asm/unistd.h arch/arm64/include/uapi/asm/unistd.h
Warning: Kernel ABI header at 'tools/include/uapi/asm-generic/unistd.h' differs from latest version at 'include/uapi/asm-generic/unistd.h'
diff -u tools/include/uapi/asm-generic/unistd.h include/uapi/asm-generic/unistd.h
Warning: Kernel ABI header at 'tools/perf/arch/x86/entry/syscalls/syscall_64.tbl' differs from latest version at 'arch/x86/entry/syscalls/syscall_64.tbl'
diff -u tools/perf/arch/x86/entry/syscalls/syscall_64.tbl arch/x86/entry/syscalls/syscall_64.tbl
But for uncore events that's not a suitable method. Uncore has nothing
to do with hybrid. So for uncore events, we aggregate event counts from
all PMUs and report the counts without PMUs.
Before:
# perf stat -e arb/event=0x81,umask=0x1/,arb/event=0x84,umask=0x1/ -a true
Jin Yao [Thu, 8 Jul 2021 01:37:00 +0000 (09:37 +0800)]
perf tests: Fix 'Roundtrip evsel->name' on core-only system
If the atom CPUs are offlined, the 'cpu_atom' is not valid.
Perf will not create two events for one hw event, so the
evsel->idx doesn't need to be divided by 2 before comparing.
Jin Yao [Thu, 8 Jul 2021 01:36:58 +0000 (09:36 +0800)]
perf pmu: Skip invalid hybrid pmu
On hybrid platform, such as Alderlake, if atom CPUs are offlined,
the kernel still exports the sysfs path '/sys/devices/cpu_atom/' for
'cpu_atom' pmu but the file '/sys/devices/cpu_atom/cpus' is empty,
which indicates this is an invalid pmu.
Need to check and skip the invalid hybrid pmu.
Before:
# perf list
...
branch-instructions OR cpu_atom/branch-instructions/ [Kernel PMU event]
branch-instructions OR cpu_core/branch-instructions/ [Kernel PMU event]
branch-misses OR cpu_atom/branch-misses/ [Kernel PMU event]
branch-misses OR cpu_core/branch-misses/ [Kernel PMU event]
bus-cycles OR cpu_atom/bus-cycles/ [Kernel PMU event]
bus-cycles OR cpu_core/bus-cycles/ [Kernel PMU event]
...
The cpu_atom events are still displayed even if atom CPUs are offlined.
After:
# perf list
...
branch-instructions OR cpu_core/branch-instructions/ [Kernel PMU event]
branch-misses OR cpu_core/branch-misses/ [Kernel PMU event]
bus-cycles OR cpu_core/bus-cycles/ [Kernel PMU event]
...
Matthew Auld [Tue, 13 Jul 2021 13:04:31 +0000 (14:04 +0100)]
drm/i915/gtt: drop the page table optimisation
We skip filling out the pt with scratch entries if the va range covers
the entire pt, since we later have to fill it with the PTEs for the
object pages anyway. However this might leave open a small window where
the PTEs don't point to anything valid for the HW to consume.
When for example using 2M GTT pages this fill_px() showed up as being
quite significant in perf measurements, and ends up being completely
wasted since we ignore the pt and just use the pde directly.
Anyway, currently we have our PTE construction split between alloc and
insert, which is probably slightly iffy nowadays, since the alloc
doesn't actually allocate anything anymore, instead it just sets up the
page directories and points the PTEs at the scratch page. Later when we
do the insert step we re-program the PTEs again. Better might be to
squash the alloc and insert into a single step, then bringing back this
optimisation(along with some others) should be possible.
Lu Baolu [Mon, 12 Jul 2021 07:17:12 +0000 (15:17 +0800)]
iommu/vt-d: Fix clearing real DMA device's scalable-mode context entries
The commit 2b0140c69637e ("iommu/vt-d: Use pci_real_dma_dev() for mapping")
fixes an issue of "sub-device is removed where the context entry is cleared
for all aliases". But this commit didn't consider the PASID entry and PASID
table in VT-d scalable mode. This fix increases the coverage of scalable
mode.
iommu/vt-d: Global devTLB flush when present context entry changed
This fixes a bug in context cache clear operation. The code was not
following the correct invalidation flow. A global device TLB invalidation
should be added after the IOTLB invalidation. At the same time, it
uses the domain ID from the context entry. But in scalable mode, the
domain ID is in PASID table entry, not context entry.
iommu/qcom: Revert "iommu/arm: Cleanup resources in case of probe error path"
QCOM IOMMU driver calls bus_set_iommu() for every IOMMU device controller,
what fails for the second and latter IOMMU devices. This is intended and
must be not fatal to the driver registration process. Also the cleanup
path should take care of the runtime PM state, what is missing in the
current patch. Revert relevant changes to the QCOM IOMMU driver until
a proper fix is prepared.
Vladimir Oltean [Tue, 13 Jul 2021 09:40:21 +0000 (12:40 +0300)]
net: dsa: properly check for the bridge_leave methods in dsa_switch_bridge_leave()
This was not caught because there is no switch driver which implements
the .port_bridge_join but not .port_bridge_leave method, but it should
nonetheless be fixed, as in certain conditions (driver development) it
might lead to NULL pointer dereference.
Fixes: f66a6a69f97a ("net: dsa: permit cross-chip bridging between all trees in the system") Signed-off-by: Vladimir Oltean <[email protected]> Signed-off-by: David S. Miller <[email protected]>
arch/mips/include/asm/fpu.h:79:3: warning: fallthrough annotation in unreachable code [-Wimplicit-fallthrough]
Originally, the /* fallthrough */ comment was introduced by commit:
597ce1723e0f ("MIPS: Support for 64-bit FP with O32 binaries")
and it was wrongly replaced with fallthrough; by commit:
c9b029903466 ("MIPS: Use fallthrough for arch/mips")
As the original comment is actually useful, fix this issue by
removing unreachable fallthrough; statement and place the original
/* fallthrough */ comment back.
Ville Syrjälä [Wed, 30 Jun 2021 16:44:13 +0000 (19:44 +0300)]
drm/i915/gt: Fix -EDEADLK handling regression
The conversion to ww mutexes failed to address the fence code which
already returns -EDEADLK when we run out of fences. Ww mutexes on
the other hand treat -EDEADLK as an internal errno value indicating
a need to restart the operation due to a deadlock. So now when the
fence code returns -EDEADLK the higher level code erroneously
restarts everything instead of returning the error to userspace
as is expected.
To remedy this let's switch the fence code to use a different errno
value for this. -ENOBUFS seems like a semi-reasonable unique choice.
Apart from igt the only user of this I could find is sna, and even
there all we do is dump the current fence registers from debugfs
into the X server log. So no user visible functionality is affected.
If we really cared about preserving this we could of course convert
back to -EDEADLK higher up, but doesn't seem like that's worth
the hassle here.
Not quite sure which commit specifically broke this, but I'll
just attribute it to the general gem ww mutex work.
Merge tag 'vboxsf-v5.14-1' of git://git.kernel.org/pub/scm/linux/kernel/git/hansg/linux
Pull vboxsf fixes from Hans de Goede:
"This adds support for the atomic_open directory-inode op to vboxsf.
Note this is not just an enhancement this also fixes an actual issue
which users are hitting, see the commit message of the "boxsf: Add
support for the atomic_open directory-inode" patch"
* tag 'vboxsf-v5.14-1' of git://git.kernel.org/pub/scm/linux/kernel/git/hansg/linux:
vboxsf: Add support for the atomic_open directory-inode op
vboxsf: Add vboxsf_[create|release]_sf_handle() helpers
vboxsf: Make vboxsf_dir_create() return the handle for the created file
vboxsf: Honor excl flag to the dir-inode create op
Merge tag 'for-5.14-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux
Pull btrfs zoned mode fixes from David Sterba:
- fix deadlock when allocating system chunk
- fix wrong mutex unlock on an error path
- fix extent map splitting for append operation
- update and fix message reporting unusable chunk space
- don't block when background zone reclaim runs with balance in
parallel
* tag 'for-5.14-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
btrfs: zoned: fix wrong mutex unlock on failure to allocate log root tree
btrfs: don't block if we can't acquire the reclaim lock
btrfs: properly split extent_map for REQ_OP_ZONE_APPEND
btrfs: rework chunk allocation to avoid exhaustion of the system chunk array
btrfs: fix deadlock with concurrent chunk allocations involving system chunks
btrfs: zoned: print unusable percentage when reclaiming block groups
btrfs: zoned: fix types for u64 division in btrfs_reclaim_bgs_work
Commit 7fe1e79b59ba ("configfs: implement the .read_iter and .write_iter
methods") changed the simple_read_from_buffer() calls into copy_to_iter()
calls and the simple_write_to_buffer() calls into copy_from_iter() calls.
The simple*buffer() methods update the file offset (*ppos) but the read
and write iterators not yet. Make the read and write iterators update the
file offset (iocb->ki_pos).
This patch has been tested as follows:
# modprobe target_core_user
# dd if=/sys/kernel/config/target/dbroot bs=1
/var/target
12+0 records in
12+0 records out
12 bytes copied, 9.5539e-05 s, 126 kB/s
# cd /sys/kernel/config/acpi/table
# mkdir test
# cd test
# dmesg -c >/dev/null; printf 'SSDT\x8\0\0\0abcdefghijklmnopqrstuvwxyz' | dd of=aml bs=1; dmesg -c
34+0 records in
34+0 records out
34 bytes copied, 0.010627 s, 3.2 kB/s
[ 261.056551] ACPI configfs: invalid table length
David S. Miller [Tue, 13 Jul 2021 17:02:41 +0000 (10:02 -0700)]
Merge branch 'sfc-tx-queues'
Íñigo Huguet says:
====================
sfc: Fix lack of XDP TX queues
A change introduced in commit e26ca4b53582 ("sfc: reduce the number of
requested xdp ev queues") created a bug in XDP_TX and XDP_REDIRECT
because it unintentionally reduced the number of XDP TX queues, letting
not enough queues to have one per CPU, which leaded to errors if XDP
TX/REDIRECT was done from a high numbered CPU.
This patchs make the following changes:
- Fix the bug mentioned above
- Revert commit 99ba0ea616aa ("sfc: adjust efx->xdp_tx_queue_count with
the real number of initialized queues") which intended to fix a related
problem, created by mentioned bug, but it's no longer necessary
- Add a new error log message if there are not enough resources to make
XDP_TX/REDIRECT work
V1 -> V2: keep the calculation of how many tx queues can handle a single
event queue, but apply the "max. tx queues per channel" upper limit.
V2 -> V3: WARN_ON if the number of initialized XDP TXQs differs from the
expected.
====================
sfc: add logs explaining XDP_TX/REDIRECT is not available
If it's not possible to allocate enough channels for XDP, XDP_TX and
XDP_REDIRECT don't work. However, only a message saying that not enough
channels were available was shown, but not saying what are the
consequences in that case. The user didn't know if he/she can use XDP
or not, if the performance is reduced, or what.
Commit 99ba0ea616aa ("sfc: adjust efx->xdp_tx_queue_count with the real
number of initialized queues") intended to fix a problem caused by a
round up when calculating the number of XDP channels and queues.
However, this was not the real problem. The real problem was that the
number of XDP TX queues had been reduced to half in
commit e26ca4b53582 ("sfc: reduce the number of requested xdp ev queues"),
but the variable xdp_tx_queue_count had remained the same.
Once the correct number of XDP TX queues is created again in the
previous patch of this series, this also can be reverted since the error
doesn't actually exist.
Only in the case that there is a bug in the code we can have different
values in xdp_queue_number and efx->xdp_tx_queue_count. Because of this,
and per Edward Cree's suggestion, I add instead a WARN_ON to catch if it
happens again in the future.
Note that the number of allocated queues can be higher than the number
of used ones due to the round up, as explained in the existing comment
in the code. That's why we also have to stop increasing xdp_queue_number
beyond efx->xdp_tx_queue_count.
Fixes: e26ca4b53582 sfc: reduce the number of requested xdp ev queues
The buggy commit intended to allocate less channels for XDP in order to
be more unlikely to reach the limit of 32 channels of the driver.
The idea was to use each IRQ/eventqeue for more XDP TX queues than
before, calculating which is the maximum number of TX queues that one
event queue can handle. For example, in EF10 each event queue could
handle up to 8 queues, better than the 4 they were handling before the
change. This way, it would have to allocate half of channels than before
for XDP TX.
The problem is that the TX queues are also contained inside the channel
structs, and there are only 4 queues per channel. Reducing the number of
channels means also reducing the number of queues, resulting in not
having the desired number of 1 queue per CPU.
This leads to getting errors on XDP_TX and XDP_REDIRECT if they're
executed from a high numbered CPU, because there only exist queues for
the low half of CPUs, actually. If XDP_TX/REDIRECT is executed in a low
numbered CPU, the error doesn't happen. This is the error in the logs
(repeated many times, even rate limited):
sfc 0000:5e:00.0 ens3f0np0: XDP TX failed (-22)
This errors happens in function efx_xdp_tx_buffers, where it expects to
have a dedicated XDP TX queue per CPU.
Reverting the change makes again more likely to reach the limit of 32
channels in machines with many CPUs. If this happen, no XDP_TX/REDIRECT
will be possible at all, and we will have this log error messages:
At interface probe:
sfc 0000:5e:00.0: Insufficient resources for 12 XDP event queues (24 other channels, max 32)
At every subsequent XDP_TX/REDIRECT failure, rate limited:
sfc 0000:5e:00.0 ens3f0np0: XDP TX failed (-22)
However, without reverting the change, it makes the user to think that
everything is OK at probe time, but later it fails in an unpredictable
way, depending on the CPU that handles the packet.
It is better to restore the predictable behaviour. If the user sees the
error message at probe time, he/she can try to configure the best way it
fits his/her needs. At least, he/she will have 2 options:
- Accept that XDP_TX/REDIRECT is not available (he/she may not need it)
- Load sfc module with modparam 'rss_cpus' with a lower number, thus
creating less normal RX queues/channels, letting more free resources
for XDP, with some performance penalty.
Anyway, let the calculation of maximum TX queues that can be handled by
a single event queue, and use it only if it's less than the number of TX
queues per channel. This doesn't happen in practice, but could happen if
some constant values are tweaked in the future, such us
EFX_MAX_TXQ_PER_CHANNEL, EFX_MAX_EVQ_SIZE or EFX_MAX_DMAQ_SIZE.
Related mailing list thread:
https://lore.kernel.org/bpf/20201215104327.2be76156@carbon/
projects / linux.git / log