In the Linux kernel, the following vulnerability has been resolved:

scsi: core: Wake up the error handler when final completions race against each other

The fragile ordering between marking commands completed or failed so that the error handler only wakes when the last running command completes or times out has race conditions. These race conditions can cause the SCSI layer to fail to wake the error handler, leaving I/O through the SCSI host stuck as the error state cannot advance.

First, there is an memory ordering issue within scsidechostbusy(). The write which clears SCMDSTATEINFLIGHT may be reordered with reads counting in scsihostbusy(). While the local CPU will see its own write, reordering can allow other CPUs in scsidechostbusy() or scsiehinchostfailed() to see a raised busy count, causing no CPU to see a host busy equal to the hostfailed count.

This race condition can be prevented with a memory barrier on the error path to force the write to be visible before counting host busy commands.

Second, there is a general ordering issue with scsiehinchostfailed(). By counting busy commands before incrementing hostfailed, it can race with a final command in scsidechostbusy(), such that scsidechostbusy() does not see hostfailed incremented but scsiehinchostfailed() counts busy commands before SCMDSTATEINFLIGHT is cleared by scsidechostbusy(), resulting in neither waking the error handler task.

This needs the call to scsihostbusy() to be moved after hostfailed is incremented to close the race condition.

In the Linux kernel, the following vulnerability has been resolved:

procfs: avoid fetching build ID while holding VMA lock

Fix PROCMAPQUERY to fetch optional build ID only after dropping mmaplock or per-VMA lock, whichever was used to lock VMA under question, to avoid deadlock reported by syzbot:

-> #1 (&mm->mmaplock){++++}-{4:4}: mightfault+0xed/0x170 copytoiter+0x118/0x1720 copypagetoiter+0x12d/0x1e0 filemapread+0x720/0x10a0 blkdevreaditer+0x2b5/0x4e0 vfsread+0x7f4/0xae0 ksysread+0x12a/0x250 dosyscall64+0xcb/0xf80 entrySYSCALL64afterhwframe+0x77/0x7f

-> #0 (&sb->stype->imutexkey#8){++++}-{4:4}: lockacquire+0x1509/0x26d0 lockacquire+0x185/0x340 downread+0x98/0x490 blkdevreaditer+0x2a7/0x4e0 kernelread+0x39a/0xa90 freaderfetch+0x1d5/0xa80 buildidparse.isra.0+0xea/0x6a0 doprocmapquery+0xd75/0x1050 procfsprocmapioctl+0x7a/0xb0 x64sysioctl+0x18e/0x210 dosyscall64+0xcb/0xf80 entrySYSCALL64afterhwframe+0x77/0x7f

other info that might help us debug this:

Possible unsafe locking scenario:

CPU0 CPU1 ---- ---- rlock(&mm->mmaplock); lock(&sb->stype->imutexkey#8); lock(&mm->mmaplock); rlock(&sb->stype->imutexkey#8);

DEADLOCK

This seems to be exacerbated (as we haven't seen these syzbot reports before that) by the recent:

777a8560fd29 ("lib/buildid: use kernelread() for sleepable context")

To make this safe, we need to grab file refcount while VMA is still locked, but other than that everything is pretty straightforward. Internal buildidparse() API assumes VMA is passed, but it only needs the underlying file reference, so just add another variant buildidparsefile() that expects file passed directly.

[akpm: fix up kerneldoc]

In the Linux kernel, the following vulnerability has been resolved:

lib/buildid: use kernelread() for sleepable context

Prevent a "BUG: unable to handle kernel NULL pointer dereference in filemapreadfolio".

For the sleepable context, convert freader to use kernelread() instead of direct page cache access via readcachefolio(). This simplifies the faultable code path by using the standard kernel file reading interface which handles all the complexity of reading file data.

At the moment we are not changing the code for non-sleepable context which uses filemapgetfolio() and only succeeds if the target folios are already in memory and up-to-date. The reason is to keep the patch simple and easier to backport to stable kernels.

Syzbot repro does not crash the kernel anymore and the selftests run successfully.

In the follow up we will make kernelread() with IOCBNOWAIT work for non-sleepable contexts. In addition, I would like to replace the secretmem check with a more generic approach and will add fstest for the buildid code.

In the Linux kernel, the following vulnerability has been resolved:

crypto: afalg - zero initialize memory allocated via sockkmalloc

Several crypto user API contexts and requests allocated with sockkmalloc() were left uninitialized, relying on callers to set fields explicitly. This resulted in the use of uninitialized data in certain error paths or when new fields are added in the future.

The ACVP patches also contain two user-space interface files: algifkpp.c and algifakcipher.c. These too rely on proper initialization of their context structures.

A particular issue has been observed with the newly added 'inflight' variable introduced in afalgctx by commit:

67b164a871af ("crypto: afalg - Disallow multiple in-flight AIO requests")

Because the context is not memset to zero after allocation, the inflight variable has contained garbage values. As a result, afalgallocareq() has incorrectly returned -EBUSY randomly when the garbage value was interpreted as true:

https://github.com/gregkh/linux/blame/master/crypto/afalg.c#L1209

The check directly tests ctx->inflight without explicitly comparing against true/false. Since inflight is only ever set to true or false later, an uninitialized value has triggered -EBUSY failures. Zero-initializing memory allocated with sockkmalloc() ensures inflight and other fields start in a known state, removing random issues caused by uninitialized data.

In the Linux kernel, the following vulnerability has been resolved:

flexproportions: make fpropnewperiod() hardirq safe

Bernd has reported a lockdep splat from flexible proportions code that is essentially complaining about the following race:

<timer fires> runtimersoftirq - we are in softirq context calltimerfn writeoutperiod fpropnewperiod writeseqcountbegin(&p->sequence);

<hardirq is raised> ... blkmqendrequest() blkupdaterequest() ext4endbio() folioendwriteback() wbwriteoutadd() fpropaddpercpumax() if (unlikely(maxfrac < FPROPFRACBASE)) { fpropfractionpercpu() seq = readseqcountbegin(&p->sequence); - sees odd sequence so loops indefinitely

Note that a deadlock like this is only possible if the bdi has configured maximum fraction of writeout throughput which is very rare in general but frequent for example for FUSE bdis. To fix this problem we have to make sure write section of the sequence counter is irqsafe.

In the Linux kernel, the following vulnerability has been resolved:

iouring/io-wq: check IOWQBITEXIT inside work run loop

Currently this is checked before running the pending work. Normally this is quite fine, as work items either end up blocking (which will create a new worker for other items), or they complete fairly quickly. But syzbot reports an issue where io-wq takes seemingly forever to exit, and with a bit of debugging, this turns out to be because it queues a bunch of big (2GB - 4096b) reads with a /dev/msr file. Since this file type doesn't support ->readiter(), looprwiter() ends up handling them. Each read returns 16MB of data read, which takes 20 (!!) seconds. With a bunch of these pending, processing the whole chain can take a long time. Easily longer than the syzbot uninterruptible sleep timeout of 140 seconds. This then triggers a complaint off the io-wq exit path:

INFO: task syz.4.135:6326 blocked for more than 143 seconds. Not tainted syzkaller #0 Blocked by coredump. "echo 0 > /proc/sys/kernel/hungtasktimeoutsecs" disables this message. task:syz.4.135 state:D stack:26824 pid:6326 tgid:6324 ppid:5957 taskflags:0x400548 flags:0x00080000 Call Trace: <TASK> contextswitch kernel/sched/core.c:5256 [inline] schedule+0x1139/0x6150 kernel/sched/core.c:6863 scheduleloop kernel/sched/core.c:6945 [inline] schedule+0xe7/0x3a0 kernel/sched/core.c:6960 scheduletimeout+0x257/0x290 kernel/time/sleeptimeout.c:75 dowaitforcommon kernel/sched/completion.c:100 [inline] waitforcommon+0x2fc/0x4e0 kernel/sched/completion.c:121 iowqexitworkers iouring/io-wq.c:1328 [inline] iowqputandexit+0x271/0x8a0 iouring/io-wq.c:1356 iouringcleantctx+0x10d/0x190 iouring/tctx.c:203 iouringcancelgeneric+0x69c/0x9a0 iouring/cancel.c:651 iouringfilescancel include/linux/iouring.h:19 [inline] doexit+0x2ce/0x2bd0 kernel/exit.c:911 dogroupexit+0xd3/0x2a0 kernel/exit.c:1112 getsignal+0x2671/0x26d0 kernel/signal.c:3034 archdosignalorrestart+0x8f/0x7e0 arch/x86/kernel/signal.c:337 exittousermodeloop kernel/entry/common.c:41 [inline] exittousermodeloop+0x8c/0x540 kernel/entry/common.c:75 exittousermodeprepare include/linux/irq-entry-common.h:226 [inline] syscallexittousermodeprepare include/linux/irq-entry-common.h:256 [inline] syscallexittousermodework include/linux/entry-common.h:159 [inline] syscallexittousermode include/linux/entry-common.h:194 [inline] dosyscall64+0x4ee/0xf80 arch/x86/entry/syscall64.c:100 entrySYSCALL64afterhwframe+0x77/0x7f RIP: 0033:0x7fa02738f749 RSP: 002b:00007fa0281ae0e8 EFLAGS: 00000246 ORIGRAX: 00000000000000ca RAX: fffffffffffffe00 RBX: 00007fa0275e6098 RCX: 00007fa02738f749 RDX: 0000000000000000 RSI: 0000000000000080 RDI: 00007fa0275e6098 RBP: 00007fa0275e6090 R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 R13: 00007fa0275e6128 R14: 00007fff14e4fcb0 R15: 00007fff14e4fd98

There's really nothing wrong here, outside of processing these reads will take a LONG time. However, we can speed up the exit by checking the IOWQBITEXIT inside the ioworkerhandlework() loop, as syzbot will exit the ring after queueing up all of these reads. Then once the first item is processed, io-wq will simply cancel the rest. That should avoid syzbot running into this complaint again.

In the Linux kernel, the following vulnerability has been resolved:

net: fix segmentation of forwarding fraglist GRO

This patch enhances GSO segment handling by properly checking the SKBGSODODGY flag for fraglist GSO packets, addressing low throughput issues observed when a station accesses IPv4 servers via hotspots with an IPv6-only upstream interface.

Specifically, it fixes a bug in GSO segmentation when forwarding GRO packets containing a fraglist. The function skbsegmentlist cannot correctly process GRO skbs that have been converted by XLAT, since XLAT only translates the header of the head skb. Consequently, skbs in the fraglist may remain untranslated, resulting in protocol inconsistencies and reduced throughput.

To address this, the patch explicitly sets the SKBGSODODGY flag for GSO packets in XLAT's IPv4/IPv6 protocol translation helpers (bpfskbproto4to6 and bpfskbproto6to4). This marks GSO packets as potentially modified after protocol translation. As a result, GSO segmentation will avoid using skbsegmentlist and instead falls back to skbsegment for packets with the SKBGSODODGY flag. This ensures that only safe and fully translated fraglist packets are processed by skbsegmentlist, resolving protocol inconsistencies and improving throughput when forwarding GRO packets converted by XLAT.

In the Linux kernel, the following vulnerability has been resolved:

mm/hugetlb: fix hugetlbpmdshared()

Patch series "mm/hugetlb: fixes for PMD table sharing (incl. using mmugather)", v3.

One functional fix, one performance regression fix, and two related comment fixes.

I cleaned up my prototype I recently shared [1] for the performance fix, deferring most of the cleanups I had in the prototype to a later point. While doing that I identified the other things.

The goal of this patch set is to be backported to stable trees "fairly" easily. At least patch #1 and #4.

Patch #1 fixes hugetlbpmdshared() not detecting any sharing Patch #2 + #3 are simple comment fixes that patch #4 interacts with. Patch #4 is a fix for the reported performance regression due to excessive IPI broadcasts during fork()+exit().

The last patch is all about TLB flushes, IPIs and mmugather. Read: complicated

There are plenty of cleanups in the future to be had + one reasonable optimization on x86. But that's all out of scope for this series.

Runtime tested, with a focus on fixing the performance regression using the original reproducer [2] on x86.

This patch (of 4):

We switched from (wrongly) using the page count to an independent shared count. Now, shared page tables have a refcount of 1 (excluding speculative references) and instead use ptdesc->ptsharecount to identify sharing.

We didn't convert hugetlbpmdshared(), so right now, we would never detect a shared PMD table as such, because sharing/unsharing no longer touches the refcount of a PMD table.

Page migration, like mbind() or migratepages() would allow for migrating folios mapped into such shared PMD tables, even though the folios are not exclusive. In smaps we would account them as "private" although they are "shared", and we would be wrongly setting the PMMMAPEXCLUSIVE in the pagemap interface.

Fix it by properly using ptdescpmdisshared() in hugetlbpmdshared().

In the Linux kernel, the following vulnerability has been resolved:

scsi: xen: scsiback: Fix potential memory leak in scsibackremove()

Memory allocated for struct vscsiblkinfo in scsibackprobe() is not freed in scsibackremove() leading to potential memory leaks on remove, as well as in the scsibackprobe() error paths. Fix that by freeing it in scsibackremove().

In the Linux kernel, the following vulnerability has been resolved:

vsock/virtio: cap TX credit to local buffer size

The virtio transports derives its TX credit directly from peerbufalloc, which is set from the remote endpoint's SOVMSOCKETSBUFFERSIZE value.

On the host side this means that the amount of data we are willing to queue for a connection is scaled by a guest-chosen buffer size, rather than the host's own vsock configuration. A malicious guest can advertise a large buffer and read slowly, causing the host to allocate a correspondingly large amount of skbuff memory. The same thing would happen in the guest with a malicious host, since virtio transports share the same code base.

Introduce a small helper, virtiotransporttxbufsize(), that returns min(peerbufalloc, bufalloc), and use it wherever we consume peerbufalloc.

This ensures the effective TX window is bounded by both the peer's advertised buffer and our own bufalloc (already clamped to buffermaxsize via SOVMSOCKETSBUFFERMAXSIZE), so a remote peer cannot force the other to queue more data than allowed by its own vsock settings.

On an unpatched Ubuntu 22.04 host (~64 GiB RAM), running a PoC with 32 guest vsock connections advertising 2 GiB each and reading slowly drove Slab/SUnreclaim from ~0.5 GiB to ~57 GiB; the system only recovered after killing the QEMU process. That said, if QEMU memory is limited with cgroups, the maximum memory used will be limited.

With this patch applied:

Before: MemFree: ~61.6 GiB Slab: ~142 MiB SUnreclaim: ~117 MiB

After 32 high-credit connections: MemFree: ~61.5 GiB Slab: ~178 MiB SUnreclaim: ~152 MiB

Only ~35 MiB increase in Slab/SUnreclaim, no host OOM, and the guest remains responsive.

Compatibility with non-virtio transports:

- VMCI uses the AFVSOCK buffer knobs to size its queue pairs per socket based on the local vsk->buffer values; the remote side cannot enlarge those queues beyond what the local endpoint configured.

- Hyper-V's vsock transport uses fixed-size VMBus ring buffers and an MTU bound; there is no peer-controlled credit field comparable to peerbufalloc, and the remote endpoint cannot drive in-flight kernel memory above those ring sizes.

- The loopback path reuses virtiotransportcommon.c, so it naturally follows the same semantics as the virtio transport.

This change is limited to virtiotransportcommon.c and thus affects virtio-vsock, vhost-vsock, and loopback, bringing them in line with the "remote window intersected with local policy" behaviour that VMCI and Hyper-V already effectively have.

[Stefano: small adjustments after changing the previous patch] [Stefano: tweak the commit message]

In the Linux kernel, the following vulnerability has been resolved:

dmaengine: qcom: gpi: Fix memory leak in gpiperipheralconfig()

Fix a memory leak in gpiperipheralconfig() where the original memory pointed to by gchan->config could be lost if krealloc() fails.

The issue occurs when: 1. gchan->config points to previously allocated memory 2. krealloc() fails and returns NULL 3. The function directly assigns NULL to gchan->config, losing the reference to the original memory 4. The original memory becomes unreachable and cannot be freed

Fix this by using a temporary variable to hold the krealloc() result and only updating gchan->config when the allocation succeeds.

Found via static analysis and code review.

In the Linux kernel, the following vulnerability has been resolved:

crypto: authencesn - reject too-short AAD (assoclen<8) to match ESP/ESN spec

authencesn assumes an ESP/ESN-formatted AAD. When assoclen is shorter than the minimum expected length, cryptoauthencesndecrypt() can advance past the end of the destination scatterlist and trigger a NULL pointer dereference in scatterwalkmapandcopy(), leading to a kernel panic (DoS).

Add a minimum AAD length check to fail fast on invalid inputs.

In the Linux kernel, the following vulnerability has been resolved:

be2net: Fix NULL pointer dereference in becmdgetmacfromlist

When the parameter pmacidvalid argument of becmdgetmacfromlist() is set to false, the driver may request the PMACID from the firmware of the network card, and this function will store that PMACID at the provided address pmacid. This is the contract of this function.

However, there is a location within the driver where both pmacidvalid == false and pmacid == NULL are being passed. This could result in dereferencing a NULL pointer.

To resolve this issue, it is necessary to pass the address of a stub variable to the function.

In the Linux kernel, the following vulnerability has been resolved:

x86/fpu: Clear XSTATEBV[i] in guest XSAVE state whenever XFD[i]=1

When loading guest XSAVE state via KVMSETXSAVE, and when updating XFD in response to a guest WRMSR, clear XFD-disabled features in the saved (or to be restored) XSTATEBV to ensure KVM doesn't attempt to load state for features that are disabled via the guest's XFD. Because the kernel executes XRSTOR with the guest's XFD, saving XSTATEBV[i]=1 with XFD[i]=1 will cause XRSTOR to #NM and panic the kernel.

E.g. if fpuupdateguestxfd() sets XFD without clearing XSTATEBV:

------------[ cut here ]------------ WARNING: arch/x86/kernel/traps.c:1524 at excdevicenotavailable+0x101/0x110, CPU#29: amxtest/848 Modules linked in: kvmintel kvm irqbypass CPU: 29 UID: 1000 PID: 848 Comm: amxtest Not tainted 6.19.0-rc2-ffa07f7fd437-x86amxnmxfdnoninit-vm #171 NONE Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015 RIP: 0010:excdevicenotavailable+0x101/0x110 Call Trace: <TASK> asmexcdevicenotavailable+0x1a/0x20 RIP: 0010:restorefpregsfromfpstate+0x36/0x90 switchfpureturn+0x4a/0xb0 kvmarchvcpuioctlrun+0x1245/0x1e40 [kvm] kvmvcpuioctl+0x2c3/0x8f0 [kvm] x64sysioctl+0x8f/0xd0 dosyscall64+0x62/0x940 entrySYSCALL64afterhwframe+0x4b/0x53 </TASK> ---[ end trace 0000000000000000 ]---

This can happen if the guest executes WRMSR(MSRIA32XFD) to set XFD[18] = 1, and a host IRQ triggers kernelfpubegin() prior to the vmexit handler's call to fpuupdateguestxfd().

and if userspace stuffs XSTATEBV[i]=1 via KVMSETXSAVE:

------------[ cut here ]------------ WARNING: arch/x86/kernel/traps.c:1524 at excdevicenotavailable+0x101/0x110, CPU#14: amxtest/867 Modules linked in: kvmintel kvm irqbypass CPU: 14 UID: 1000 PID: 867 Comm: amxtest Not tainted 6.19.0-rc2-2dace9faccd6-x86amxnmxfdnoninit-vm #168 NONE Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 0.0.0 02/06/2015 RIP: 0010:excdevicenotavailable+0x101/0x110 Call Trace: <TASK> asmexcdevicenotavailable+0x1a/0x20 RIP: 0010:restorefpregsfromfpstate+0x36/0x90 fpuswapkvmfpstate+0x6b/0x120 kvmloadguestfpu+0x30/0x80 [kvm] kvmarchvcpuioctlrun+0x85/0x1e40 [kvm] kvmvcpuioctl+0x2c3/0x8f0 [kvm] x64sysioctl+0x8f/0xd0 dosyscall64+0x62/0x940 entrySYSCALL64afterhwframe+0x4b/0x53 </TASK> ---[ end trace 0000000000000000 ]---

The new behavior is consistent with the AMX architecture. Per Intel's SDM, XSAVE saves XSTATEBV as '0' for components that are disabled via XFD (and non-compacted XSAVE saves the initial configuration of the state component):

If XSAVE, XSAVEC, XSAVEOPT, or XSAVES is saving the state component i, the instruction does not generate #NM when XCR0[i] = IA32XFD[i] = 1; instead, it operates as if XINUSE[i] = 0 (and the state component was in its initial state): it saves bit i of XSTATEBV field of the XSAVE header as 0; in addition, XSAVE saves the initial configuration of the state component (the other instructions do not save state component i).

Alternatively, KVM could always do XRSTOR with XFD=0, e.g. by using a constant XFD based on the set of enabled features when XSAVEing for a struct fpuguest. However, having XSTATEBV[i]=1 for XFD-disabled features can only happen in the above interrupt case, or in similar scenarios involving preemption on preemptible kernels, because fpuswapkvmfpstate()'s call to savefpregstofpstate() saves the outgoing FPU state with the current XFD; and that is (on all but the first WRMSR to XFD) the guest XFD.

Therefore, XFD can only go out of sync with XSTATEBV in the above interrupt case, or in similar scenarios involving preemption on preemptible kernels, and it we can consider it (de facto) part of KVM ABI that KVMGETXSAVE returns XSTATEBV[i]=0 for XFD-disabled features.

[Move clea ---truncated---

In the Linux kernel, the following vulnerability has been resolved:

ipv4: ipgre: make ipgreheader() robust

Analog to commit db5b4e39c4e6 ("ip6gre: make ip6greheader() robust")

Over the years, syzbot found many ways to crash the kernel in ipgreheader() [1].

This involves team or bonding drivers ability to dynamically change their dev->neededheadroom and/or dev->hardheaderlen

In this particular crash mldnewpack() allocated an skb with a too small reserve/headroom, and by the time mldsendpack() was called, syzbot managed to attach an ipgre device.

[1] skbuff: skbunderpanic: text:ffffffff89ea3cb7 len:2030915468 put:2030915372 head:ffff888058b43000 data:ffff887fdfa6e194 tail:0x120 end:0x6c0 dev:team0 kernel BUG at net/core/skbuff.c:213 ! Oops: invalid opcode: 0000 [#1] SMP KASAN PTI CPU: 1 UID: 0 PID: 1322 Comm: kworker/1:9 Not tainted syzkaller #0 PREEMPT(full) Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 10/25/2025 Workqueue: mld mldifcwork RIP: 0010:skbpanic+0x157/0x160 net/core/skbuff.c:213 Call Trace: <TASK> skbunderpanic net/core/skbuff.c:223 [inline] skbpush+0xc3/0xe0 net/core/skbuff.c:2641 ipgreheader+0x67/0x290 net/ipv4/ipgre.c:897 devhardheader include/linux/netdevice.h:3436 [inline] neighconnectedoutput+0x286/0x460 net/core/neighbour.c:1618 NFHOOKCOND include/linux/netfilter.h:307 [inline] ip6output+0x340/0x550 net/ipv6/ip6output.c:247 NFHOOK+0x9e/0x380 include/linux/netfilter.h:318 mldsendpack+0x8d4/0xe60 net/ipv6/mcast.c:1855 mldsendcr net/ipv6/mcast.c:2154 [inline] mldifcwork+0x83e/0xd60 net/ipv6/mcast.c:2693 processonework kernel/workqueue.c:3257 [inline] processscheduledworks+0xad1/0x1770 kernel/workqueue.c:3340 workerthread+0x8a0/0xda0 kernel/workqueue.c:3421 kthread+0x711/0x8a0 kernel/kthread.c:463 retfromfork+0x510/0xa50 arch/x86/kernel/process.c:158 retfromforkasm+0x1a/0x30 arch/x86/entry/entry64.S:246

In the Linux kernel, the following vulnerability has been resolved:

net: marvell: prestera: fix NULL dereference on devlinkalloc() failure

devlinkalloc() may return NULL on allocation failure, but presteradevlinkalloc() unconditionally calls devlinkpriv() on the returned pointer.

This leads to a NULL pointer dereference if devlink allocation fails. Add a check for a NULL devlink pointer and return NULL early to avoid the crash.

In the Linux kernel, the following vulnerability has been resolved:

bpf: Fix reference count leak in bpfprogtestrunxdp()

syzbot is reporting

unregisternetdevice: waiting for sit0 to become free. Usage count = 2

problem. A debug printk() patch found that a refcount is obtained at xdpconvertmdtobuff() from bpfprogtestrunxdp().

According to commit ec94670fcb3b ("bpf: Support specifying ingress via xdpmd context in BPFPROGTESTRUN"), the refcount obtained by xdpconvertmdtobuff() will be released by xdpconvertbufftomd().

Therefore, we can consider that the error handling path introduced by commit 1c1949982524 ("bpf: introduce frags support to bpfprogtestrunxdp()") forgot to call xdpconvertbufftomd().

In the Linux kernel, the following vulnerability has been resolved:

dmaengine: athdmac: fix device leak on ofdmaxlate()

Make sure to drop the reference taken when looking up the DMA platform device during ofdmaxlate() when releasing channel resources.

Note that commit 3832b78b3ec2 ("dmaengine: athdmac: add missing putdevice() call in atdmaxlate()") fixed the leak in a couple of error paths but the reference is still leaking on successful allocation.

In the Linux kernel, the following vulnerability has been resolved:

dmaengine: stm32: dmamux: fix device leak on route allocation

Make sure to drop the reference taken when looking up the DMA mux platform device during route allocation.

Note that holding a reference to a device does not prevent its driver data from going away so there is no point in keeping the reference.

In the Linux kernel, the following vulnerability has been resolved:

dmaengine: dw: dmamux: fix OF node leak on route allocation failure

Make sure to drop the reference taken to the DMA master OF node also on late route allocation failures.

In the Linux kernel, the following vulnerability has been resolved:

dmaengine: bcm-sba-raid: fix device leak on probe

Make sure to drop the reference taken when looking up the mailbox device during probe on probe failures and on driver unbind.

dmaengine: idxd: fix device leaks on compat bind and unbind

In the Linux kernel, the following vulnerability has been resolved:

dmaengine: ti: dma-crossbar: fix device leak on am335x route allocation

Make sure to drop the reference taken when looking up the crossbar platform device during am335x route allocation.

In the Linux kernel, the following vulnerability has been resolved:

KVM: x86: Fix VM hard lockup after prolonged inactivity with periodic HV timer

When advancing the target expiration for the guest's APIC timer in periodic mode, set the expiration to "now" if the target expiration is in the past (similar to what is done in updatetargetexpiration()). Blindly adding the period to the previous target expiration can result in KVM generating a practically unbounded number of hrtimer IRQs due to programming an expired timer over and over. In extreme scenarios, e.g. if userspace pauses/suspends a VM for an extended duration, this can even cause hard lockups in the host.

Currently, the bug only affects Intel CPUs when using the hypervisor timer (HV timer), a.k.a. the VMX preemption timer. Unlike the software timer, a.k.a. hrtimer, which KVM keeps running even on exits to userspace, the HV timer only runs while the guest is active. As a result, if the vCPU does not run for an extended duration, there will be a huge gap between the target expiration and the current time the vCPU resumes running. Because the target expiration is incremented by only one period on each timer expiration, this leads to a series of timer expirations occurring rapidly after the vCPU/VM resumes.

More critically, when the vCPU first triggers a periodic HV timer expiration after resuming, advancing the expiration by only one period will result in a target expiration in the past. As a result, the delta may be calculated as a negative value. When the delta is converted into an absolute value (tscdeadline is an unsigned u64), the resulting value can overflow what the HV timer is capable of programming. I.e. the large value will exceed the VMX Preemption Timer's maximum bit width of cpupreemptiontimermulti + 32, and thus cause KVM to switch from the HV timer to the software timer (hrtimers).

After switching to the software timer, periodic timer expiration callbacks may be executed consecutively within a single clock interrupt handler, because hrtimers honors KVM's request for an expiration in the past and immediately re-invokes KVM's callback after reprogramming. And because the interrupt handler runs with IRQs disabled, restarting KVM's hrtimer over and over until the target expiration is advanced to "now" can result in a hard lockup.

E.g. the following hard lockup was triggered in the host when running a Windows VM (only relevant because it used the APIC timer in periodic mode) after resuming the VM from a long suspend (in the host).

NMI watchdog: Watchdog detected hard LOCKUP on cpu 45 ... RIP: 0010:advanceperiodictargetexpiration+0x4d/0x80 [kvm] ... RSP: 0018:ff4f88f5d98d8ef0 EFLAGS: 00000046 RAX: fff0103f91be678e RBX: fff0103f91be678e RCX: 00843a7d9e127bcc RDX: 0000000000000002 RSI: 0052ca4003697505 RDI: ff440d5bfbdbd500 RBP: ff440d5956f99200 R08: ff2ff2a42deb6a84 R09: 000000000002a6c0 R10: 0122d794016332b3 R11: 0000000000000000 R12: ff440db1af39cfc0 R13: ff440db1af39cfc0 R14: ffffffffc0d4a560 R15: ff440db1af39d0f8 FS: 00007f04a6ffd700(0000) GS:ff440db1af380000(0000) knlGS:000000e38a3b8000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 000000d5651feff8 CR3: 000000684e038002 CR4: 0000000000773ee0 PKRU: 55555554 Call Trace: <IRQ> apictimerfn+0x31/0x50 [kvm] hrtimerrunqueues+0x100/0x280 hrtimerinterrupt+0x100/0x210 ? ttwudowakeup+0x19/0x160 smpapictimerinterrupt+0x6a/0x130 apictimerinterrupt+0xf/0x20 </IRQ>

Moreover, if the suspend duration of the virtual machine is not long enough to trigger a hard lockup in this scenario, since commit 98c25ead5eda ("KVM: VMX: Move preemption timer <=> hrtimer dance to common x86"), KVM will continue using the software timer until the guest reprograms the APIC timer in some way. Since the periodic timer does not require frequent APIC timer register programming, the guest may continue to use the software timer in ---truncated---

In the Linux kernel, the following vulnerability has been resolved:

viawdt: fix critical boot hang due to unnamed resource allocation

The VIA watchdog driver uses allocateresource() to reserve a MMIO region for the watchdog control register. However, the allocated resource was not given a name, which causes the kernel resource tree to contain an entry marked as "<BAD>" under /proc/iomem on x86 platforms.

During boot, this unnamed resource can lead to a critical hang because subsequent resource lookups and conflict checks fail to handle the invalid entry properly.

In the Linux kernel, the following vulnerability has been resolved:

mptcp: fallback earlier on simult connection

Syzkaller reports a simult-connect race leading to inconsistent fallback status:

WARNING: CPU: 3 PID: 33 at net/mptcp/subflow.c:1515 subflowdataready+0x40b/0x7c0 net/mptcp/subflow.c:1515 Modules linked in: CPU: 3 UID: 0 PID: 33 Comm: ksoftirqd/3 Not tainted syzkaller #0 PREEMPT(full) Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014 RIP: 0010:subflowdataready+0x40b/0x7c0 net/mptcp/subflow.c:1515 Code: 89 ee e8 78 61 3c f6 40 84 ed 75 21 e8 8e 66 3c f6 44 89 fe bf 07 00 00 00 e8 c1 61 3c f6 41 83 ff 07 74 09 e8 76 66 3c f6 90 <0f> 0b 90 e8 6d 66 3c f6 48 89 df e8 e5 ad ff ff 31 ff 89 c5 89 c6 RSP: 0018:ffffc900006cf338 EFLAGS: 00010246 RAX: 0000000000000000 RBX: ffff888031acd100 RCX: ffffffff8b7f2abf RDX: ffff88801e6ea440 RSI: ffffffff8b7f2aca RDI: 0000000000000005 RBP: 0000000000000000 R08: 0000000000000005 R09: 0000000000000007 R10: 0000000000000004 R11: 0000000000002c10 R12: ffff88802ba69900 R13: 1ffff920000d9e67 R14: ffff888046f81800 R15: 0000000000000004 FS: 0000000000000000(0000) GS:ffff8880d69bc000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 CR2: 0000560fc0ca1670 CR3: 0000000032c3a000 CR4: 0000000000352ef0 Call Trace: <TASK> tcpdataqueue+0x13b0/0x4f90 net/ipv4/tcpinput.c:5197 tcprcvstateprocess+0xfdf/0x4ec0 net/ipv4/tcpinput.c:6922 tcpv6dorcv+0x492/0x1740 net/ipv6/tcpipv6.c:1672 tcpv6rcv+0x2976/0x41e0 net/ipv6/tcpipv6.c:1918 ip6protocoldeliverrcu+0x188/0x1520 net/ipv6/ip6input.c:438 ip6inputfinish+0x1e4/0x4b0 net/ipv6/ip6input.c:489 NFHOOK include/linux/netfilter.h:318 [inline] NFHOOK include/linux/netfilter.h:312 [inline] ip6input+0x105/0x2f0 net/ipv6/ip6input.c:500 dstinput include/net/dst.h:471 [inline] ip6rcvfinish net/ipv6/ip6input.c:79 [inline] NFHOOK include/linux/netfilter.h:318 [inline] NFHOOK include/linux/netfilter.h:312 [inline] ipv6rcv+0x264/0x650 net/ipv6/ip6input.c:311 netifreceiveskbonecore+0x12d/0x1e0 net/core/dev.c:5979 netifreceiveskb+0x1d/0x160 net/core/dev.c:6092 processbacklog+0x442/0x15e0 net/core/dev.c:6444 napipoll.constprop.0+0xba/0x550 net/core/dev.c:7494 napipoll net/core/dev.c:7557 [inline] netrxaction+0xa9f/0xfe0 net/core/dev.c:7684 handlesoftirqs+0x216/0x8e0 kernel/softirq.c:579 runksoftirqd kernel/softirq.c:968 [inline] runksoftirqd+0x3a/0x60 kernel/softirq.c:960 smpbootthreadfn+0x3f7/0xae0 kernel/smpboot.c:160 kthread+0x3c2/0x780 kernel/kthread.c:463 retfromfork+0x5d7/0x6f0 arch/x86/kernel/process.c:148 retfromforkasm+0x1a/0x30 arch/x86/entry/entry64.S:245 </TASK>

The TCP subflow can process the simult-connect syn-ack packet after transitioning to TCPFIN1 state, bypassing the MPTCP fallback check, as the skstatechange() callback is not invoked for -> FINWAIT1 transitions.

That will move the msk socket to an inconsistent status and the next incoming data will hit the reported splat.

Close the race moving the simult-fallback check at the earliest possible stage - that is at syn-ack generation time.

About the fixes tags: [2] was supposed to also fix this issue introduced by [3]. [1] is required as a dependence: it was not explicitly marked as a fix, but it is one and it has already been backported before [3]. In other words, this commit should be backported up to [3], including [2] and [1] if that's not already there.

In the Linux kernel, the following vulnerability has been resolved:

ipv4: Fix reference count leak when using error routes with nexthop objects

When a nexthop object is deleted, it is marked as dead and then fibtableflush() is called to flush all the routes that are using the dead nexthop.

The current logic in fibtableflush() is to only flush error routes (e.g., blackhole) when it is called as part of network namespace dismantle (i.e., with flushall=true). Therefore, error routes are not flushed when their nexthop object is deleted:

# ip link add name dummy1 up type dummy # ip nexthop add id 1 dev dummy1 # ip route add 198.51.100.1/32 nhid 1 # ip route add blackhole 198.51.100.2/32 nhid 1 # ip nexthop del id 1 # ip route show blackhole 198.51.100.2 nhid 1 dev dummy1

As such, they keep holding a reference on the nexthop object which in turn holds a reference on the nexthop device, resulting in a reference count leak:

# ip link del dev dummy1 [ 70.516258] unregisternetdevice: waiting for dummy1 to become free. Usage count = 2

Fix by flushing error routes when their nexthop is marked as dead.

IPv6 does not suffer from this problem.

In the Linux kernel, the following vulnerability has been resolved:

net: nfc: fix deadlock between nfcunregisterdevice and rfkillfopwrite

A deadlock can occur between nfcunregisterdevice() and rfkillfopwrite() due to lock ordering inversion between devicelock and rfkillglobalmutex.

The problematic lock order is:

Thread A (rfkillfopwrite): rfkillfopwrite() mutexlock(&rfkillglobalmutex) rfkillsetblock() nfcrfkillsetblock() nfcdevdown() devicelock(&dev->dev) <- waits for devicelock

Thread B (nfcunregisterdevice): nfcunregisterdevice() devicelock(&dev->dev) rfkillunregister() mutexlock(&rfkillglobalmutex) <- waits for rfkillglobalmutex

This creates a classic ABBA deadlock scenario.

Fix this by moving rfkillunregister() and rfkilldestroy() outside the devicelock critical section. Store the rfkill pointer in a local variable before releasing the lock, then call rfkillunregister() after releasing devicelock.

This change is safe because rfkillfopwrite() holds rfkillglobalmutex while calling the rfkill callbacks, and rfkillunregister() also acquires rfkillglobalmutex before cleanup. Therefore, rfkillunregister() will wait for any ongoing callback to complete before proceeding, and devicedel() is only called after rfkillunregister() returns, preventing any use-after-free.

The similar lock ordering in nfcregisterdevice() (devicelock -> rfkillglobalmutex via rfkillregister) is safe because during registration the device is not yet in rfkilllist, so no concurrent rfkill operations can occur on this device.

In the Linux kernel, the following vulnerability has been resolved:

mm/damon/sysfs: cleanup attrs subdirs on context dir setup failure

When a context DAMON sysfs directory setup is failed after setup of attrs/ directory, subdirectories of attrs/ directory are not cleaned up. As a result, DAMON sysfs interface is nearly broken until the system reboots, and the memory for the unremoved directory is leaked.

Cleanup the directories under such failures.

In the Linux kernel, the following vulnerability has been resolved:

migrate: correct lock ordering for hugetlb file folios

Syzbot has found a deadlock (analyzed by Lance Yang):

1) Task (5749): Holds foliolock, then tries to acquire immaprwsem(read lock). 2) Task (5754): Holds immaprwsem(write lock), then tries to acquire foliolock.

migratepages() -> migratehugetlbs() -> unmapandmovehugepage() <- Takes foliolock! -> removemigrationptes() -> rmapwalkfile() -> immaplockread() <- Waits for immaprwsem(read lock)!

hugetlbfsfallocate() -> hugetlbfspunchhole() <- Takes immaprwsem(write lock)! -> hugetlbfszeropartialpage() -> filemaplockhugetlbfolio() -> filemaplockfolio() -> filemapgetfolio <- Waits for foliolock!

The migration path is the one taking locks in the wrong order according to the documentation at the top of mm/rmap.c. So expand the scope of the existing immaplock to cover the calls to removemigrationptes() too.

This is (mostly) how it used to be after commit c0d0381ade79. That was removed by 336bf30eb765 for both file & anon hugetlb pages when it should only have been removed for anon hugetlb pages.