In the Linux kernel, the following vulnerability has been resolved: net: stmmac: fix rx Scatter-Gather support When a received frame is larger than dma_buf_sz, the DMA scatters it across multiple RX descriptors (rx Scatter-Gather). The secondary RX buffer (sec_page) was only allocated and programmed when split-header (SPH) was active, so for regular frames buffer2 was neither allocated nor backed by a valid mapping. As soon as an incoming frame overflowed buffer1, the DMA wrote the overflow into the unmapped secondary-buffer address, triggering an SMMU translation fault on IOMMU-based platforms: arm-smmu 15000000.iommu: Unhandled context fault: fsr=0x402, iova=0x00000000, fsynr=0x7f0011, cbfrsynra=0x1c90, cb=11 arm-smmu 15000000.iommu: FSR = 00000402 [Format=2 TF], SID=0x1c90 arm-smmu 15000000.iommu: FSYNR0 = 007f0011 [S1CBNDX=127 WNR PLVL=1] Enable scatter-gather for non-SPH frames on cores that can program an independent secondary RX buffer (GMAC4/XGMAC): allocate and mark buffer2 as valid in stmmac_init_rx_buffers() and stmmac_rx_refill(), and account for it in the buffer length computation. Legacy cores have no set_sec_addr op, so they keep buffer2 disabled. Since buffer2 is handed to the DMA at page offset 0, the page pool sync window is widened to cover both buffers in every mode: offset is set to 0 and max_len to dma_buf_sz + stmmac_rx_offset(). The FCS can straddle the buffer1/buffer2 boundary and a descriptor boundary, so it is now stripped from the tail of the assembled frame with pskb_trim() instead of from a single buffer, avoiding an unsigned underflow for frames that overflow a buffer by 1..3 bytes. For single-buffer frames the XDP program must not see the FCS, so it is removed from the XDP buffer before the program runs. Native XDP currently only supports single-buffer (linear) frames. An oversized frame accepted by the MAC (jumbo enabled) is received via buffer2; the XDP program only sees buffer1, so on a TX/REDIRECT verdict the frame is forwarded truncated. XDP multi-buffer support to handle this case is planned as a follow-up. AF_XDP zero-copy RX is not covered by this change: a ZC queue still programs buffer2 at DMA address 0 (and XGMAC has no buffer2-valid bit), so an oversized frame overflowing buffer1 can still trigger the same SMMU translation fault. Handling is planned as a follow-up.
In the Linux kernel, the following vulnerability has been resolved: net: sparx5: skip ptp deinit if init was skipped sparx5_ptp_init() returns early on the base lan969x variants because they do not have the SPX5_FEATURE_PTP flag. It also returns early when no "ptp" interrupt is described. Unbinding the driver then causes a NULL pointer dereference when cleaning up uninitialized tx_skbs queues: Unable to handle kernel NULL pointer dereference at virtual address 0000000000000008 Call trace: skb_queue_purge_reason+0x68/0x120 (P) sparx5_ptp_deinit+0x64/0xc0 mchp_sparx5_remove+0x38/0x70 platform_remove+0x20/0x30 device_remove+0x4c/0x80 Before per-port tx_skbs were added, a NULL dereference would have happened when calling ptp_clock_unregister() on the never-registered PTP clocks. Fix by returning early in sparx5_ptp_deinit() if PTP is not used.
In the Linux kernel, the following vulnerability has been resolved: rust_binder: cancel deferred work items in thread exit If there are deferred work items on the thread todo list, then they are not cleaned up in the Thread::release() method. Thus, update the code to clean up the work items even if they are deferred. This can happen if the thread dies while it has an active outgoing transaction.
In the Linux kernel, the following vulnerability has been resolved: x86/mm: Don't apply va_align to hugetlb mappings on AMD F15h get_align_mask() returns huge_page_mask_align() for hugetlbfs, but get_align_bits() adds va_align.bits regardless, so vm_unmapped_area() returns an address off the huge page boundary and __unmap_hugepage_range() hits BUG_ON(start & ~huge_page_mask(h)) at teardown. This can be triggered on Carrizo and FX-8370E, both hstates. Pass the file to get_align_bits() and skip the randomisation for hugetlbfs. [ bp: Massage commit message. ]
In the Linux kernel, the following vulnerability has been resolved: thunderbolt: Tear down inactive DP tunnels when the domain is stopped tb_stop() only tears down DMA tunnels so a DP tunnel that is still waiting for dprx_work to complete keeps that work queued while the routers are removed and the control channel is stopped. The work only stops once the DPRX timeout has passed and because it requeues itself until then the flush_workqueue() in tb_domain_remove() won't wait for its final run. The callback then runs against a domain that is already torn down. A reference to that domain is kept so the completion waiting for that domain to disappear in unbind will block until the timeout is eventually reached. Tear down DP tunnels that are not active yet as well which also cancels that work. Tunnels for displays that are already alive are untouched and keep working.
In the Linux kernel, the following vulnerability has been resolved: thunderbolt: Fix domain reference leak when DPRX read is canceled tb_tunnel_one_dp() takes a domain reference which is only dropped once tb_dp_tunnel_active() has run on the work queue. If that work is cancelled that reference is leaked. Since commit f5cc545f5969 ("thunderbolt: Wait for tb_domain_release() to complete when driver is removed") instead of just leaking memory this now also blocks in the completion wait forever when unbinding the driver. This can be triggered whenever a DP tunnel is torn down before the DPRX read has completed, e.g. by unplugging within the timeout, and then unbinding the driver. That reference only exists to keep the domain around while the DPRX work is scheduled so let the work itself own it: take it in tb_dp_dprx_start() and drop it in both places that end the work. Get/put are then paired inside the same file and it doesn't matter anymore if the callback ever runs.
In the Linux kernel, the following vulnerability has been resolved: thunderbolt: Don't access a DP tunnel after its DPRX read was canceled tb_dp_dprx_work() checks ->dprx_canceled before it takes tb->lock so it misses a tb_dp_dprx_stop() that could not cancel the already running work. It then polls the DPRX capabilities and runs the callback for a tunnel that has already been torn down while the domain is suspending or going away. This can be hit by cancelling the DPRX read from outside the ordered tb->wq: During suspend tb_disconnect_and_release_dp() does just this and with a later patch tb_stop() will do it as well. The latter in combination with the Apple NHI where the DPRX read never completed is how I hit this. Check the flag with tb->lock held instead and check it again in tb_dp_tunnel_active() because the callback runs after the lock has been dropped again. Also clear the flag in tb_dp_dprx_start() so that it only ever describes the work that is currently in flight.
In the Linux kernel, the following vulnerability has been resolved: smb: client: flush and commit data before querying allocated ranges The mode 0 / FALLOC_FL_KEEP_SIZE fallocate emulation for small interior ranges in smb3_simple_fallocate_range() issues FSCTL_QUERY_ALLOCATED_RANGES and zero-fills any sub-range the server reports as an unallocated hole, to force block allocation without changing file contents. This is unsafe against Windows servers. Per the Win32 documentation, FSCTL_QUERY_ALLOCATED_RANGES only reports ranges that *may* contain nonzero data and is explicitly not coherent with recently written data: "a call to FSCTL_QUERY_ALLOCATED_RANGES [after writing to a network file] would not necessarily return a correct list of allocated regions. To ensure coherency [...] flush the data to the file" fsx (generic/363) reproduces the resulting corruption against Windows Server 2022: a range is written, an uncached read returns the data, yet the next FSCTL_QUERY_ALLOCATED_RANGES on that range reports it as a whole hole, so the emulation overwrites live data with zeroes. A network trace confirmed the WRITE, the READ returning data and the query returning an empty range list within microseconds. Samba does not exhibit this; its query is coherent with the written data. Write back the dirty pages, drain any outstanding I/O and issue an SMB2 FLUSH to force the server to commit the data before querying the allocated ranges, so the query reflects the data actually present. With this, generic/363 passes against both Windows Server 2022 and Samba over 100000 fsx operations.
In the Linux kernel, the following vulnerability has been resolved: smb: client: flush dirty data before zeroing a range smb3_zero_range() emulates FALLOC_FL_ZERO_RANGE by invalidating the page cache over the target range with truncate_pagecache_range() and then issuing FSCTL_SET_ZERO_DATA to the server. Dirty data was only flushed conditionally, when the range reached or extended EOF. For a purely interior zero range that does not reach EOF, no flush happened, so a dirty folio overlapping the range could be written back to the server after the FSCTL and refill the range that was just zeroed with stale data. Fix this by unconditionally flushing and waiting for dirty data in the range before invalidating the page cache and issuing FSCTL_SET_ZERO_DATA, exactly as was done for smb3_punch_hole() in commit d7d2adcd022b ("smb/client: flush dirty data before punching a hole"); the two paths are structurally identical here. This is observed as generic/363 randomly reading stale data where a zeroed range is expected against Windows Server.
In the Linux kernel, the following vulnerability has been resolved: iio: adc: ad4030: fix invalid oversampling_ratio validation In ad4030_set_avg_frame_len(), the logarithm is calculated before input validation. Passing zero or negative values leads to an undefined result from ilog2(). Validate that the input is strictly positive prior to computing its logarithm.
In the Linux kernel, the following vulnerability has been resolved: iio: adc: ade9000: handle Phase C dip events in IRQ1 handler ADE9000_ST1_CROSSING_DEPTH is the exclusive upper bound for scanning the STATUS1 event bits, so a value of 25 left out the highest event bit, ADE9000_ST1_DIPC_BIT (bit 25). Phase C dip events were never reported. Bump the value to 26 so bit 25 is scanned.
In the Linux kernel, the following vulnerability has been resolved: iio: adc: adi-axi-adc: Initialize state mutex The AXI ADC register access paths serialize transactions with st->lock, but probe does not initialize it. Initialize the mutex before registering the backend.
In the Linux kernel, the following vulnerability has been resolved: s390/uv: Prevent potential out-of-bounds read When the system has more than 85 secrets, the uv_secret_list struct array only holds up to 85 items per page, resulting in an out of bounds read in find_secret_in_page if the targeted secret is in the next page or not stored at all. Fix this by looping over the number of stored secrets which is the per sub-list count of stored secrets and not the overall count.
In the Linux kernel, the following vulnerability has been resolved: KVM: riscv: Fix NACL hfence entry update order The SBI v3.0 specification (section 15.1.2) requires a nested HFENCE entry to be populated as follows: 1) find an unused entry with Config.Pending == 0 2) update the Page_Number and Page_Count words 3) update the Config word with Config.Pending set __kvm_riscv_nacl_hfence() writes the Config word first, so the SBI implementation (or NACL hardware) can observe a pending entry with pnum/pcount values left over from the previous use of that entry, resulting in incorrect TLB flush ranges. Write pnum and pcount first and the Config word last. Since the consumer is an external agent on coherent shared memory, use WRITE_ONCE() to stop the compiler from reordering the stores and smp_wmb() to make the parameter words globally visible before the Pending bit is set.
In the Linux kernel, the following vulnerability has been resolved: KVM: arm64: Validate the SVE vector length in pkvm_vcpu_init_sve() pkvm_vcpu_init_sve() clamps only the upper bound of the host-provided sve_max_vl, so an invalid vector length reaches sve_state_size_from_vl() and the WARN_ON() there, which is fatal at EL2. The existing !sve_state_size test rejects such a length, but only after the macro has run. Check sve_vl_valid() before deriving the state size. A valid length cannot yield a zero size, so the !sve_state_size test goes with it.
In the Linux kernel, the following vulnerability has been resolved: KVM: arm64: Do not clear VM-wide SVE feature on vCPU init failure pkvm_vcpu_init_sve() clears KVM_ARM_VCPU_SVE in kvm->arch.vcpu_features when it fails, but vcpu_has_sve() tests KVM_ARCH_FLAG_GUEST_HAS_SVE, which is left set. Later vCPUs on that VM then skip the SVE setup and register with a NULL sve_state, which the guest's first FP access hands to sve_load_state(). Return the error without touching vcpu_features.
In the Linux kernel, the following vulnerability has been resolved: Bluetooth: btnxpuart: Fix skb leak in nxp_process_fw_dump() When CONFIG_DEV_COREDUMP=n, hci_devcd_append() returns -EOPNOTSUPP without freeing its skb argument. This leaks the cloned skb and also prevents nxp_set_ind_reset() from being called to perform recovery. Fix by guarding the hci_devcd_append(hdev, skb_clone(skb, GFP_ATOMIC)) call with IS_ENABLED(CONFIG_DEV_COREDUMP).
In the Linux kernel, the following vulnerability has been resolved: eth: fbnic: reset num_napi when the napi vectors are freed fbn->num_napi is the count of live napi vectors, each of which owns an IRQ. The PM path had freed them without clearing the count. fbnic_pm_suspend() tears the datapath down via ndo_stop() and frees the IRQs, but leaves netif_running() true so resume knows to re-open. Resume rebuilds the datapath in __fbnic_pm_resume() and fbnic_reset_queues() sets num_napi and __fbnic_open() re-allocates the vectors. When the datapath is torn down but never rebuilt, num_napi is left pointing at freed vectors under 2 different scenarios: - a PCIe error recovery that fails (fbnic_err_slot_reset() -> __fbnic_pm_resume() returns an error -> PCI_ERS_RESULT_DISCONNECT), so .resume never runs; or - an __fbnic_open() that fails partway on resume and unwinds, freeing the vectors after fbnic_reset_queues() has already set num_napi. The netdev is then running with num_napi > 0 but napi[] freed, and the eventual remove/unbind close re-enters fbnic_down() -> fbnic_dbg_down() and dereferences the freed vectors: BUG: kernel NULL pointer dereference, address: 0000000000000210 RIP: fbnic_dbg_down+0x28 Clear num_napi when the vectors are freed: in the suspend teardown (a good resume re-establishes it before __fbnic_open()) and on the resume open failure. A redundant ndo_stop() then walks an empty napi[]. The normal ndo_stop() down/up cycle is untouched and keeps num_napi for the next ndo_open().
In the Linux kernel, the following vulnerability has been resolved: eth: fbnic: Avoid rounding zero ring sizes roundup_pow_of_two() is undefined for zero. ethtool permits a zero ring size to reach the driver, where the minimum-size check should reject it. Leave zero unchanged while rounding nonzero ring sizes. The minimum-size check then rejects zero deterministically without changing the established behavior for other values.
In the Linux kernel, the following vulnerability has been resolved: s390/pci: Fix missing device lock in zpci_report_status() When pdev is non-NULL, zpci_report_status() accesses the device's driver. To get a consistent state matching the recovery, the device lock needs to be held. Do so by expanding the existing device lock critical section. The lock only needs to be held when the pdev is non-NULL, so extract the pdev-specific reporting into a helper function which also adds a lockdep assertion to detect calls without the device lock held.
In the Linux kernel, the following vulnerability has been resolved: KVM: SEV: Free have_run_cpus during VM destruction even if VM is no longer SEV Unconditionally free SEV's "have run CPUs" cpumask in the VM destroy path, i.e. even for what appear to be non-SEV VMs, as an SEV VM becomes a non-SEV VM if its state is intra-host migrated. Alternatively, the mask could be freed in sev_migrate_from() when "converting" the source VM, but that gets annoying because ideally KVM would nullify the mask to guard against UAF, and nullifying the mask would need be conditioned on CPUMASK_OFFSTACK=y. Freeing the mask during sev_migrate_from() is also not robust against other KVM bugs, though that's kind of a moot point since any such bugs would show up even if sev->active is never set. I.e. KVM must get that side of things correct. But, that's not a great reason to add more code just to make things marginally less robust.
In the Linux kernel, the following vulnerability has been resolved: KVM: arm64: nv: Fix null ptr deref on nested wp/unmap, teardown race Commit 7270cc9157f4 ("KVM: arm64: nv: Handle VNCR_EL2 invalidation from MMU notifiers") introduced VNCR_EL2 invalidation in both kvm_nested_s2_unmap() and kvm_nested_s2_wp(). However at the point of this being performed concurrent stage 2 teardown of a nested guest can cause kvm->arch.mmu.pgt to be set to NULL. This happens in kvm_flush_shadow_all() -> kvm_arch_flush_shadow_all() -> kvm_free_stage2_pgd() and is performed under the kvm->mmu_lock. Commit ec14c272408a ("KVM: arm64: nv: Unmap/flush shadow stage 2 page tables") introduced the teardown of the entire nested MMU range, which then invokes stage2_apply_range() with resched=true: mmu_notifier_invalidate_range_start() -> ... -> kvm_mmu_notifier_invalidate_range_start() -> kvm_mmu_unmap_gfn_range() -> kvm_unmap_gfn_range() -> kvm_nested_s2_unmap() -> kvm_stage2_unmap_range() -> __unmap_stage2_range() -> stage2_apply_range() This means that stage2_apply_range() can drop the kvm->mmu_lock and thus concurrent progress can be made in lockstep with kvm_arch_flush_shadow_all(). If kvm_arch_flush_shadow_all() advances ahead of stage2_apply_range() and completes its operation it guarantees a NULL pointer deref. Since kvm_free_stage2_pgd() is performed under the kvm->mmu_lock this will either be observed NULL or not and serialised against kvm_free_stage2_pgd(). Resolve the issue by abstracting the invalidation to a new function, kvm_invalidate_vncr_ipa_all(), and check that the pgt is non-NULL before dereferencing it.
In the Linux kernel, the following vulnerability has been resolved: xfs: fix wild memcpy access when formatting ondisk rtrefcount btree roots LOLLM noticed that the inode btree root formatting methods copy too many bytes -- there's only one set of keys in node blocks, not two. This causes memory corruption of whatever's beyond the buffers.
In the Linux kernel, the following vulnerability has been resolved: KVM: PPC: Book3S HV: fix secure device page leak on uv_page_in() failure In kvmppc_svm_page_in(), if uv_page_in() fails after kvmppc_uvmem_get_page() has succeeded, the secure device page is never released. kvmppc_uvmem_get_page() sets a bit in kvmppc_uvmem_bitmap, allocates a kvmppc_uvmem_page_pvt struct, marks the GFN as KVMPPC_GFN_UVMEM_PFN, and calls zone_device_page_init() which sets refcount=1 and locks the page. The subsequent goto out_finalize skips the *mig.dst assignment, so migrate_vma_finalize() is a no-op for the page, and none of those resources are ever reclaimed. Each occurrence permanently consumes one entry from the firmware-bounded secure memory pool (kvmppc_uvmem_bitmap), leaks pvt, and leaves the GFN marked as secure - making it unusable for the lifetime of the VM. The twin __kvmppc_svm_page_out() already handles the analogous uv_page_out() failure correctly with unlock_page(dpage); __free_page(dpage). Apply the same pattern here: unlock_page() followed by put_page(), which chains through free_zone_device_folio() into kvmppc_uvmem_folio_free() to clear the bitmap bit, free pvt, and reset the GFN state. Reachable whenever uv_page_in() returns an error (e.g. UV pool exhaustion) on any POWER9/10 + Ultravisor/PEF system.
In the Linux kernel, the following vulnerability has been resolved: mm/slab: do not wake up kswapd in __kfree_rcu_sheaf() Since kfree_rcu() can be called under pi_lock (a raw spinlock in the scheduler), kfree_rcu() itself should never allocate memory with __GFP_KSWAPD_RECLAIM as waking up kswapd ends up acquiring pi_lock, which leads to a deadlock. Reproducing the issue even intentionally was not straightforward. The set_cpus_allowed_force() path that is called under pi_lock is exercised very rarely, and competing tasks that allocate memory will most likely wake kswapd up. Therefore the existence of the deadlock was verified with a modified kernel that has a lockdep map for waking up kswapd: ====================================================== WARNING: possible circular locking dependency detected 7.2.0-rc1-slab-for-next+ #10 Not tainted ------------------------------------------------------ git/6660 is trying to acquire lock: ffff8e524533c550 (&p->pi_lock){-.-.}-{2:2}, at: _raw_spin_lock_irqsave+0x12/0x20 but task is already holding lock: ffff8e57bffff1a0 (&pgdat->kswapd_wait){....}-{3:3}, at: _raw_spin_lock_irqsave+0x12/0x20 which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #2 (&pgdat->kswapd_wait){....}-{3:3}: __lock_acquire+0x5a4/0xc50 lock_acquire.part.0+0xb7/0x240 lock_acquire+0x70/0x170 __raw_spin_lock_irqsave+0x44/0x80 _raw_spin_lock_irqsave+0x12/0x20 __wake_up_common_lock+0x31/0xa0 __wake_up+0x20/0x40 kswapd_wakeup_wake+0x7e/0x110 wakeup_kswapd+0x295/0x320 wake_all_kswapds+0xac/0x1a0 __alloc_pages_slowpath.constprop.0+0x282/0xf80 __alloc_frozen_pages_noprof+0x32a/0x360 alloc_slab_page+0x2e/0x160 allocate_slab+0x82/0x420 new_slab+0x52/0xb0 refill_objects+0x13d/0x190 refill_sheaf+0x5d/0xd0 __pcs_replace_empty_main+0x230/0xb10 [...] -> #1 (kswapd_wakeup){-.-.}-{0:0}: __lock_acquire+0x5a4/0xc50 lock_sync.part.0+0x75/0x100 lock_sync+0x36/0x70 might_wakeup_kswapd+0x61/0xa0 __kfree_rcu_sheaf+0x33/0xd90 kvfree_call_rcu+0x1d4/0x3b0 set_cpus_allowed_force+0x163/0x220 cpuset_cpus_allowed_fallback+0x18b/0x240 select_fallback_rq+0x1e6/0x250 [...] -> #0 (&p->pi_lock){-.-.}-{2:2}: check_prev_add+0xe6/0xe00 validate_chain+0x51e/0x6e0 __lock_acquire+0x5a4/0xc50 lock_acquire.part.0+0xb7/0x240 lock_acquire+0x70/0x170 __raw_spin_lock_irqsave+0x44/0x80 _raw_spin_lock_irqsave+0x12/0x20 try_to_wake_up+0x77/0xa90 default_wake_function+0x27/0x60 autoremove_wake_function+0x23/0xb0 __wake_up_common+0xb8/0x170 __wake_up_common_lock+0x51/0xa0 __wake_up+0x20/0x40 kswapd_wakeup_wake+0x7e/0x110 wakeup_kswapd+0x295/0x320 wake_all_kswapds+0xac/0x1a0 __alloc_pages_slowpath.constprop.0+0x282/0xf80 __alloc_frozen_pages_noprof+0x32a/0x360 alloc_slab_page+0x2e/0x160 allocate_slab+0x82/0x420 new_slab+0x52/0xb0 refill_objects+0x13d/0x190 refill_sheaf+0x5d/0xd0 __pcs_replace_empty_main+0x230/0xb10 [...] other info that might help us debug this: Chain exists of: &p->pi_lock --> kswapd_wakeup --> &pgdat->kswapd_wait Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&pgdat->kswapd_wait); lock(kswapd_wakeup); lock(&pgdat->kswapd_wait); lock(&p->pi_lock); *** DEADLOCK *** 3 locks held by git/6660: #0: ffff8e519f0e9e20 (&type->i_mutex_dir_key#6){++++}-{4:4}, at: lookup_slow+0x2d/0x60 #1: ffffffff8a31c460 (kswapd_wakeup){-.-.}- ---truncated---
In the Linux kernel, the following vulnerability has been resolved: bpf: Fix overflow of jump offset in constant blinding bpf_jit_blind_insn() rewrites 'if rX op imm goto off' into: AX = imm ^ rnd AX ^= rnd if rX op AX goto off and subtracts 2 from off when the jump is backward. off is s16, so -32767 and -32768 wrap and the jump goes 32767 insns forward, past the end of the prog. The jump is inside the patch, hence bpf_adj_branches() doesn't check it. The kernel crashes when such prog runs with bpf_jit_harden=2. Return -ERANGE in such case. bpf_jit_blind_constants() fails the same way when blinding of other insns moves a jump out of range.
In the Linux kernel, the following vulnerability has been resolved: nvme: fix command effects log lifetime for multipath heads KASAN reported a use-after-free when an I/O passthrough command was sent through a multipath namespace head after the controller path that first created the head had been removed: BUG: KASAN: slab-use-after-free in nvme_command_effects+0x192/0x200 [nvme_core] Read of size 4 at addr ffff888141b14400 by task nvme/19811 nvme_command_effects+0x192/0x200 [nvme_core] nvme_cmd_allowed+0x7e/0x1b0 [nvme_core] nvme_user_cmd.constprop.0+0x1b5/0x450 [nvme_core] nvme_ns_head_chr_ioctl+0xf4/0x2a0 [nvme_core] Move the log cache to the subsystem, with one entry per command set. Both controllers and namespace heads hold subsystem references, keeping their log pointers valid until subsystem release.
In the Linux kernel, the following vulnerability has been resolved: wifi: mac80211: fix slab-out-of-bounds read in ieee80211_monitor_select_queue() ieee80211_monitor_select_queue() validates the skb length using skb->len before dereferencing pointers into skb->data to access the 802.11 header. However, skb->len includes data in both the linear head buffer and non-linear fragments/pages. When AF_PACKET sends a packet large enough to become non-linear, the 802.11 header at skb->data + len_rthdr may extend past the linear head buffer even though skb->len appears sufficient. This leads to a slab-out-of-bounds read when accessing hdr->frame_control, as the kernel reads beyond the allocated skb head buffer: BUG: KASAN: slab-out-of-bounds in ieee80211_monitor_select_queue+0x1ed/0x220 syzbot has hit the same issues. Fix this by checking skb_headlen(skb) (which gives the length of the linear data region) instead of skb->len before dereferencing skb->data pointers. This ensures the required bytes are actually present in the linear portion of the skb. Apply the same fix to ieee80211_validate_radiotap_len() which has the same class of bug: it uses skb->len to validate accesses to skb->data, and to the corresponding checks in ieee80211_monitor_start_xmit(): for drivers advertising NETIF_F_SG the skb is not linearized before ndo_start_xmit, so the 802.11 header can be read past the linear buffer there as well.
In the Linux kernel, the following vulnerability has been resolved: smb: client: delete compound mids on send failure before unlock When sending a compound request fails, smb_send_rqst() kicks off a reconnect. compound_send_recv() still has those mids on pending_mid_q, but it unlocks the server without removing them first. During reconnect, cifs_abort_connection() walks pending_mid_q and runs each mid callback. With no response yet, those callbacks return credits and drop in_flight. Then compound_send_recv()'s send-error path returns the same credits again. in_flight ends up decremented twice and smb2_add_credits() WARNs. syzbot hits this during SMB2_negotiate when the socket send fails. cifs_call_async() already calls delete_mid() before unlock on send failure. Do the same for compound chains and set cancelled_mid[] so the out: path does not delete them again.
In the Linux kernel, the following vulnerability has been resolved: net: cap skb->queue_mapping when the tx queue is picked skbedit can set skb->queue_mapping and raise the per-CPU skip_txqueue flag so __dev_queue_xmit() honours the mapping. __dev_queue_xmit() cleared the flag before sch_handle_egress() and only read it afterwards, so the flag was not confined to the xmit that set it: a nested xmit (mirred redirect or mirror, or a drop after skbedit) could set the flag and the outer xmit would consume it for an skb that never went through skbedit. A forwarded packet still carries the ingress NIC's rx_queue + 1 in skb->queue_mapping, so the outer device then indexes its tx queue state with that stale value. Taprio's child array q->qdiscs[] is sized to the device's queue count, so taprio_enqueue() indexes past its allocation and dereferences the result as a struct Qdisc *. We (ab)use the skb->nf_skip_egress which means "skip netfilter egress for this packet" to tag to "am I in tc egress?". Despite the overload I dont see it as a conflict since the marker is set only around the single sch_handle_egress() call and ingress path is guarded by tc_at_ingress. I will send a followup(net-next) patch once this hits net-next to rename the skb->nf_skip_egress bit/flag to skb->skip_egress Arm the flag only from the egress classifier that can use it: raise skip_txqueue from tcf_skbedit_act() only when it runs inside sch_handle_egress(), thanks to skb->nf_skip_egress. An egress qdisc classifier runs in q->enqueue(), after the tx queue has been picked, so a mapping it sets cannot affect the current packet; arming the flag there only pollutes it for a later xmit. Then own the flag for the xmit frame the egress hook runs in: save the incoming value and clear it just before sch_handle_egress(), and restore it after the hook - on the consumed (drop) path, or, in the same call that reads it, on the surviving path. The save and the restores stay inside the egress_needed_key static branch, so a packet pays for them only when egress hooks are active (2f1e85b1aee4). Store the value netdev_cap_txqueue() selected back into skb->queue_mapping in netdev_tx_queue_mapping(), as netdev_core_pick_tx() already does, so the skip_txqueue path never hands a later reader on the xmit path a mapping the device cannot serve. A store made still later in the same frame, by a tc BPF program attached to a transmit qdisc, is outside this path and is not re-capped; a separate followup will resolve that path. netdev_xmit_skip_txqueue() returns the previous flag value so the save-and-clear is one call, and a no-op stub is provided when CONFIG_NET_EGRESS is disabled. skb->nf_skip_egress is compiled under CONFIG_NET_EGRESS rather than CONFIG_NETFILTER_SKIP_EGRESS, so skb_at_tc_egress() is valid whenever the egress path is built. A local user in a network namespace can redirect a packet from a device with more TX queues to one with fewer after setting a mapping valid only on the larger device. That reaches these reads and, under KASAN, faults with "slab-out-of-bounds in taprio_enqueue". Conditions to recreate the bug: the report's own trigger is a local user with CAP_NET_ADMIN in a network namespace, so no eBPF program is needed. With CONFIG_NET_SCH_TAPRIO=y, CONFIG_NET_ACT_SKBEDIT=y, CONFIG_NET_ACT_MIRRED=y, CONFIG_NET_CLS_MATCHALL=y, CONFIG_NET_SCH_PRIO=y and KASAN enabled, create qa (3 queues), qb (2 queues) and qc (1 queue) as dummy devices; put a taprio root on qb (num_tc 1, queues 2@0) and clsact on all three; then add an egress matchall filter on every device. On qa: "action skbedit queue_mapping 2 pipe action mirred egress redirect dev qb". On qb: "action mirred egress mirror dev qc". On qc: "action skbedit queue_mapping 0 pipe". Send one packet out qa. qc's skbedit sets the flag while qb's outer xmit is in flight; without the fix qb consumes it and reads its two-entry taprio child array with the forwarded packet's stale mapping. A qc whose skb ---truncated---
In the Linux kernel, the following vulnerability has been resolved: audit: fix exe mark UAF in kill_rules() kill_rules() removes mixed AUDIT_DIR and AUDIT_EXE rules when an audit tree is pruned. It drops entry->rule.exe before removing the rule from the RCU-visible filter lists. After a rule has been installed with AUDIT_ADD_RULE, which requires CAP_AUDIT_CONTROL, removing the watched directory can race with another task that is still evaluating the rule. In that case, fsnotify can free the executable mark before the reader reaches audit_mark_compare(), causing a use-after-free. KASAN reports: BUG: KASAN: slab-use-after-free in audit_mark_compare+0x8d/0xa0 Unlink the published rules from the RCU-visible lists and retain them on tree->rules for cleanup. If any rule has an executable mark, wait for a single RCU grace period before removing the marks and scheduling the entries for freeing. Otherwise, call_rcu() already provides the required deferred freeing without a synchronous wait.
In the Linux kernel, the following vulnerability has been resolved: HID: multitouch: stop the release timer from being rearmed on remove mt_remove() quiesces the sticky-finger timer before stopping the hardware: timer_delete_sync(&td->release_timer); sysfs_remove_group(&hdev->dev.kobj, &mt_attribute_group); hid_hw_stop(hdev); timer_delete_sync() waits for a running callback and dequeues the timer, but it does not stop the timer from being armed again. The transport is still delivering reports at that point, and the report path rearms it: if (app->quirks & MT_QUIRK_STICKY_FINGERS) { if (td->mt_io_flags & MT_IO_SLOTS_MASK) mod_timer(&td->release_timer, jiffies + msecs_to_jiffies(100)); A report that arrives after timer_delete_sync() has returned therefore leaves the timer queued. td is allocated with devm_kzalloc() against hdev->dev, so it is freed when the driver is unbound, after mt_remove() returns. When the timer fires afterwards, mt_expired_timeout() dereferences the freed td: struct mt_device *td = timer_container_of(td, t, release_timer); struct hid_device *hdev = td->hdev; if (test_and_set_bit_lock(MT_IO_FLAGS_RUNNING, &td->mt_io_flags)) Simply moving the teardown after hid_hw_stop() does not fix this on its own, because mt_expired_timeout() calls mt_release_contacts(), which walks hdev->inputs; the timer still has to be quiesced before hid_hw_stop() tears the input devices down. Use timer_shutdown_sync() instead, which additionally makes any later mod_timer() a no-op, so neither ordering constraint has to be traded off against the other. This is the final-teardown pattern the function was introduced for, and hid-wiimote already uses it for the same reason.
In the Linux kernel, the following vulnerability has been resolved: io_uring: fix task_work add use-after-free with SQPOLL For SQPOLL, the sqpoll thread runs task_work off its own list and doesn't need to be told about it, so it can pop and complete a request as soon as mpscq_push() has published it. If that was the last request holding the ring alive, io_ring_exit_work() then frees the ctx, stops the thread, and with it the tctx, while io_req_normal_work_add() is still looking at both after the push: BUG: KASAN: slab-use-after-free in io_req_normal_work_add+0x439/0x510 Read of size 4 at addr ff11000000ca0000 by task repro/55 (...) BUG: KASAN: slab-use-after-free in queue_work_on+0x25/0x70 Write of size 8 at addr ff11000000c3bd00 by task repro/55 Hold an RCU read lock across the add, like io_req_local_work_add() already does, and have io_ring_exit_work() wait for a grace period for SQPOLL rings as well before freeing the ctx.
In the Linux kernel, the following vulnerability has been resolved: of/overlay: put property on deadprops only after changeset add succeeds add_changeset_property() links a new property of a not-yet-live target node into the node's deadprops list before handing it to of_changeset_add_property(). If that fails, the error path frees the property but leaves the freed pointer linked in deadprops, and of_node_release() frees it a second time when the aborted overlay's node is released. Record the changeset entry first and link the property into deadprops only on success. of_changeset_add_property() never looks at the node's property lists, so the order of the two steps is otherwise immaterial, and the error path frees a property that nothing references.
In the Linux kernel, the following vulnerability has been resolved: net: extend IPv6 exthdr detection of tunneled packets Commit c4336a07eb6b ("net: correctly handle tunneled traffic on IPV6_CSUM GSO fallback") split skb_gso_has_extension_hdr() into mutually exclusive branches on skb->encapsulation. This did not yet address all paths: 1. With skb->encapsulation set, the outer header is not checked. IPv6 tunnels such as ip6_gre and ip6_tunnel add an outer Destination Options header by default (encap_limit). Their GSO packets skip software GSO, then skb_csum_hwoffload_help() sees the outer extension header and calls skb_checksum_help() on the GSO skb, which warns and drops it. 2. Tunnels over IPv6 without an outer transport header, such as ip6_tunnel, leave skb->transport_header at the inner transport header. skb_network_header_len() then spans the outer IPv6, tunnel and inner IP headers, a false positive. 3. UDP tunnels without an inner network header, such as SCTP-in-UDP or PSP, have no inner IPv6 header to check. Decide on the outer header alone. 4. Directly dereferencing inner_ip_hdr(skb)->version without skb_header_pointer() is unsafe. Instead, check ipv6_ext_hdr(nexthdr) on the outer IPv6 header and, if set, on the inner IPv6 header. Read the headers with skb_header_pointer(). Keep the skb_network_header_len() check when !skb->encapsulation to also catch encapsulated packets without skb->encapsulation (e.g., virtio). Use the same helper in skb_csum_hwoffload_help(). Its open coded test has the false positive of (2) and ignores the inner header. Background: checksum offload of tunneled packets invariants: Non-GSO skb: - If the inner packet is CHECKSUM_PARTIAL, Local Checksum Offload computes the outer checksum in software and the device offloads only the inner L4 checksum. - If the inner packet is CHECKSUM_NONE (e.g., SCTP-in-UDP, ESP-in-UDP, Remote Checksum Offload), the device offloads the outer UDP or GRE checksum instead. GSO skb: - A device with NETIF_F_GSO_UDP_TUNNEL_CSUM or NETIF_F_GSO_GRE_CSUM computes both inner and outer checksums per segment. The outer checksum is seeded with only the pseudo-header checksum. A NETIF_F_IPV6_CSUM device must parse through the outer headers to reach the inner ones.
In the Linux kernel, the following vulnerability has been resolved: netfilter: nft_flow_offload: drop flowtable reference on init error path nft_flow_offload_init() bumps the flowtable use count with nft_use_inc() before calling nf_ct_netns_get(). When the latter fails, the error is returned as-is and the reference is leaked. The upper layers do not balance it either: nf_tables_newexpr() clears expr->ops when the expression init callback fails, so the nft_expr_more() iteration in nft_rule_expr_deactivate() and nf_tables_rule_destroy() stops right before the failed expression and its ->destroy callback, which would drop the reference, never runs. Each failed rule addition therefore leaks one flowtable reference and the flowtable can no longer be removed: NFT_MSG_DELFLOWTABLE keeps reporting -EBUSY even though no rule references it. Save the nf_ct_netns_get() return value and undo the nft_use_inc() when it fails, restoring the inc/dec pairing within nft_flow_offload_init() itself.
In the Linux kernel, the following vulnerability has been resolved: net/packet: prevent TX_RING byte count overflow tpacket_snd() drains every SEND_REQUEST frame in a TX ring and adds each valid packet length to the signed int len_sum. Userspace can recycle completed frames concurrently, so a single blocking send call is not bounded by the ring size. After enough successful transmissions len_sum wraps into the negative range. In particular, 65536 65535-byte frames followed by a 65007-byte frame produce 0xfffffdef, or -EIOCBQUEUED. sock_sendmsg_nosec() treats that as an impossible return and triggers a BUG. On systems configured to panic on oops, a CAP_NET_RAW holder can panic the host kernel. Stop before the next frame would make the byte count exceed INT_MAX. The positive short count leaves that frame pending for a later send call. A two-thread TPACKET_V2 reproducer triggered the BUG on an arm64 build of Linux 7.3-rc4. With this change, the same reproducer returned the expected positive short count without an oops.
In the Linux kernel, the following vulnerability has been resolved: rtc: ac100: Assign .num before accessing .hws Commit f316cdff8d67 ("clk: Annotate struct clk_hw_onecell_data with __counted_by") annotated the hws member of 'struct clk_hw_onecell_data' with __counted_by, which informs the bounds sanitizer (UBSAN_BOUNDS) about the number of elements in .hws[], so that it can warn when .hws[] is accessed out of bounds. As noted in that change, the __counted_by member must be initialized with the number of elements before the first array access happens, otherwise there will be a warning from each access prior to the initialization because the number of elements is zero. This occurs in ac100_rtc_register_clks() due to .num being assigned only after every clkout clock has been stored in .hws[]. With CONFIG_UBSAN_BOUNDS and a compiler that implements __counted_by (GCC 15.1+ or Clang 20.1+), this triggers an array-index-out-of-bounds report during probe, and with CONFIG_UBSAN_TRAP the first store traps. Initialize .num with AC100_CLKOUT_NUM, the number of elements .hws[] was allocated with, right after the allocation. That is the value the loop counter ends up at on the success path anyway, so the provider's behaviour is unchanged.
In the Linux kernel, the following vulnerability has been resolved: ipvs: bound LBLCR and LBLC cache growth ip_vs_lblcr_new() and ip_vs_lblc_new() create cache entries for every previously unseen destination address. The table max_size only tells the periodic collector to reclaim entries after the cache has already exceeded the limit. It does not reclaim entries that the attacker continues to use. Reject new cache entries once either table reaches max_size * 3 / 2. The extra headroom lets the periodic collector catch up while the existing scheduler fallback continues to use the selected destination when cache creation fails. New traffic therefore stays serviceable without growing the tables further.
In the Linux kernel, the following vulnerability has been resolved: of/overlay: only treat a positive changeset id as registered of_overlay_fdt_apply() stores the idr_alloc() return value in ovcs->id before checking it. On failure the stored id is negative, free_overlay_changeset()'s "if (ovcs->id)" check passes, idr_remove() is called with a negative id and list_del() runs on ovcs->ovcs_list, which is not initialized until after the id allocation. An allocation failure at that point dereferences NULL. Make free_overlay_changeset() treat only a strict-positive id as registered. The rest of the function already copes with a partially-initialized ovcs, so the error path stays a plain goto err_free_ovcs.
In the Linux kernel, the following vulnerability has been resolved: rtc: spear: initialize IRQ state before requesting alarm IRQ devm_request_irq() enables the interrupt before it returns, so the handler may run while probe is still initializing the device. Initialize the MMIO address and spinlock before requesting the alarm IRQ.
In the Linux kernel, the following vulnerability has been resolved: sctp: check RCV_SHUTDOWN after the sendmsg connect wait sctp_wait_for_connect() drops the socket lock while it sleeps. An out-of-the-blue ABORT can then be processed from the socket backlog and unlink the association. If a concurrent shutdown(fd, SHUT_RD) sets RCV_SHUTDOWN, the waiter breaks with err == 0 before checking asoc->base.dead. Its final sctp_association_put() can then free the association, leaving sctp_sendmsg_to_asoc() to continue with a dangling pointer. Check RCV_SHUTDOWN along with the wait error in sctp_sendmsg_to_asoc() before using the association again. The check only accesses the socket, so it needs no additional association reference. Return the existing -ESRCH so that sctp_sendmsg() skips freeing a new association that may already have been destroyed. Keep sctp_wait_for_connect() unchanged to preserve its behavior for the connect() caller.
In the Linux kernel, the following vulnerability has been resolved: tpm: Disable TPM on null key name mismatch The null key name check exists to protect against TPM reset attacks, so a mismatch should stop the device from serving further requests. Currently it does not disable the chip when it finds a mismatch. The mismatch is logged: tpm tpm0: null key integrity check failed but the chip keeps serving commands: / # tpm2_getcap -c properties-fixed TPM_PT_FAMILY_INDICATOR: as UINT32: 0x08322e3000 as string: "2.0" ... tpm2_load_null() where the null key name check is run sets the chip as disabled only if the rc is non zero. When the mismatch is seen, rc is zero at that point and it returns success. The other issue is that the caller expects the null key handle to be populated when the function returns 0 which it does here without writing the handle and proceeds assuming the null key handle is valid. During the test, I noticed that tpm2_start_auth_session() uses the uninitialized stack value as the key handle since tpm2_load_null() returns 0 despite the integrity failure and proceeds with TPM2_CC_START_AUTH_SESS with this value as salt key handle. Set rc to -ENODEV on the mismatch. The error path then disables the chip and returns the correct code to the caller. Tested in QEMU with swtpm and CONFIG_TCG_TPM2_HMAC=y by making TPM2_CC_CONTEXT_LOAD fail with TPM2_RC_INTEGRITY and changing the name of the re-created null key. The chip is now disabled on the mismatch.
In the Linux kernel, the following vulnerability has been resolved: netfs: clear post-EOF pagecache when extending a file via write Fix netfs to erase the contents of a hole created after the EOF by an ordinary write if dirty data has been previously left there by writes through an mmapped region. Neither the buffered nor the unbuffered/DIO write path clears that stale pagecache. Zero the tail of the folio straddling the EOF before an extending write. That is the only folio that can hold data written past the EOF through an mmap, as pages wholly beyond the EOF can't be faulted in. The folio is zeroed rather than dropped so a concurrent extending write can't lose data. Both write paths downgrade the i_rwsem to shared, so extending writes can run concurrently and the i_size read by the caller may be stale by the time the folio is locked. Re-read i_size under the folio lock and clamp the zeroed range up to it, so a racing write that already put data into the folio isn't clobbered. Wait for any writeback on the folio to finish before zeroing it so that the pagecache isn't modified while it may still be read by the transport during transmission. Honour IOCB_NOWAIT by returning -EAGAIN rather than blocking on the folio lock, on writeback, or in folio_mkclean()'s rmap walk when the folio is mapped. truncate_pagecache() can't be used here: it must be called with the i_rwsem held exclusively, but these write paths only hold it shared, and it would block unconditionally, breaking IOCB_NOWAIT. Callers that hold i_rwsem exclusively for the whole resize (truncate, setattr, fallocate, clone) exclude any genuine concurrent buffered writer, so staleness can instead be decided from the folio's dirty state, as pagecache_isize_extended() already does for filesystems that serialise writes against truncate/setattr via a single i_rwsem. Export netfs_clear_stale_post_isize() helper to handle such case. The helper is required by the CIFS client to fix generic/363.
In the Linux kernel, the following vulnerability has been resolved: EDAC/altera: Use parent device for devres in altr_portb_setup() Anchor the devres group and the devm-managed IRQ requests in altr_portb_setup() to the actual parent device (device->edac->dev) instead of the embedded struct device inside the copied per-port altr_edac_device_dev. This keeps devres_open_group(), devm_request_irq(), devres_remove_group() and devres_release_group() all referring to the same long-lived device so the group and the resources allocated inside it are torn down together.
In the Linux kernel, the following vulnerability has been resolved: mtd: core: avoid double-free of OTP NVMEM device If factory OTP setup fails after the user OTP NVMEM device has been registered, mtd_otp_nvmem_add() unregisters the user device but leaves mtd->otp_user_nvmem set. On an error, the caller unregisters it again. For -EOPNOTSUPP, registration continues and normal teardown unregisters it again. Clear mtd->otp_user_nvmem after unregistering it.
In the Linux kernel, the following vulnerability has been resolved: seg6: ensure packet data is writable before modifying SRH and IPv6 DA advance_nextseg() modifies the SRH Segments Left field and the IPv6 destination address without ensuring the packet data is writable. seg6_next_csid_advance_arg() has the same problem when it advances the NEXT-C-SID argument in the destination address. The skb may be cloned, for example by an AF_PACKET socket receiving on the ingress device. advance_nextseg() and seg6_next_csid_advance_arg() then write into the packet data shared with the clone. A read from that socket can return the modified packet instead of the received one. The simplified path below shows this for advance_nextseg(): __netif_receive_skb_one_core __netif_receive_skb_core deliver_skb [orig: users=2, cloned=0] packet_rcv skb_clone clone queued to the AF_PACKET socket consume_skb(orig) [orig: users=1, cloned=1] ipv6_rcv ip6_rcv_core skb_share_check: no-op [orig: users=1] [...] input_action_end_core advance_nextseg writes into the data shared with the clone Call skb_ensure_writable() in advance_nextseg() and in seg6_next_csid_advance_arg() before they modify the packet data. skb_ensure_writable() may reallocate skb->head, which invalidates the pointers into the packet data taken before the call. advance_nextseg() now returns a valid SRH pointer, or NULL if skb_ensure_writable() fails. seg6_next_csid_advance_arg() takes the pointer to the destination address from the skb after skb_ensure_writable(). On failure, the callers drop the packet with SKB_DROP_REASON_NOMEM.
In the Linux kernel, the following vulnerability has been resolved: bpf: Wait for an RCU tasks grace period before freeing trampoline progs When a prog is detached from a trampoline, it is freed after an RCU grace period, or an RCU tasks trace one if it is sleepable. This covers the tasks that are running the prog, since the prog's enter helper takes the matching read lock before the prog is called. On a preemptible kernel, it doesn't cover a task that was preempted in the trampoline just before the enter helper. That task holds no lock yet, and it calls the prog after it was freed: BUG: KASAN: vmalloc-out-of-bounds in __bpf_prog_enter_recur+0x3a5/0x3f0 Read of size 8 at addr ffffc90000055040 by task candidate/110 CPU: 1 UID: 0 PID: 110 Comm: candidate Not tainted 7.3.0-rc2-00014-g15071f2a1263-dirty #2 PREEMPT(full) Call Trace: <TASK> __bpf_prog_enter_recur+0x3a5/0x3f0 bpf_trampoline_6442509193+0x37/0xf1 __x64_sys_futex+0x9/0x410 do_syscall_64+0xb0/0x530 ... Wait for an RCU tasks grace period before the existing one when freeing a prog that was linked to a trampoline. An RCU tasks grace period only ends once the tasks that were preempted have run again, and bpf_tramp_image_put() already relies on it to free the image. It doesn't wait for tasks that sleep in the trampoline, the next commit takes care of those. Only progs that were linked to a trampoline can be called this way, so bpf_trampoline_add_prog() marks them and other progs are still freed as before.
In the Linux kernel, the following vulnerability has been resolved: bpf: Skip detached progs in trampoline images that are still in use bpf_tramp_image_put() makes sure a trampoline image is not freed while a task may still be running in it, but nothing similar is done for the progs called by that image. Detach drops the last prog reference right away and the prog is freed after grace periods, on the basis that a task still in the traced function skips the fexit progs once the nop at ip_after_call is patched to a jump. That leaves out a task sleeping in a sleepable prog that runs before the detached one, which no grace period waits for: CPU 0 CPU 1 in image I, sleeping in prog S detach P from I's trampoline -> new image, bpf_tramp_image_put(I) bpf_prog_put(P), last ref grace periods, P freed back from S __bpf_prog_enter(P) call P->bpf_func If S and P are fexit progs the task is already past the patched jump, and fentry only images don't have one. On x86 this is an int3 in poisoned bpf_prog_pack memory: Oops: int3: 0000 [#1] SMP NOPTI CPU: 18 UID: 0 PID: 94573 Comm: x169 Not tainted 6.18.44 #1 PREEMPT(lazy) RIP: 0010:0xffffffffc0601d8d Call Trace: <TASK> ? bpf_trampoline_6442515411+0x1a4/0x21b bpf_lsm_bprm_committed_creds+0x5/0x10 security_bprm_committed_creds+0x5f/0x70 begin_new_exec+0x2d6/0x410 ... We hit this in production when progs attached through trampolines got detached while their hooks were busy, and it was independently found with a fuzzer and KASAN. Have the JITs emit a patchable nop in front of each prog call sequence and record it in the image. When a prog is detached, patch its nop to a jump over the call sequence. Progs that stay attached keep running for the tasks that are in the image, and ip_after_call isn't needed anymore. The task can be in any image that isn't freed yet, not only in the current one. It sleeps in image I1 that calls S, P and Q, then P is detached and the trampoline moves to image I2, then Q is detached and its call is still in I1. So the trampoline keeps a list of its images until they are freed, and detaching a prog patches its nop in all of them. Images hold a reference on the trampoline for that long. On riscv and loongarch a jump of any range takes several instructions, and a task preempted in the middle of them could resume into half of the new sequence. The nop is a single instruction there, patched to a near branch through arch_bpf_trampoline_skip(). With the extra nops, BPF_MAX_TRAMP_LINKS progs no longer fit in a page on x86 and arm64 (and already didn't on powerpc), so lower the limit there like s390 does.
In the Linux kernel, the following vulnerability has been resolved: bpf: Fix objects stuck in free_by_rcu_ttrace do_call_rcu_ttrace() returns early when call_rcu_ttrace_in_progress is set and leaves the objects in free_by_rcu_ttrace. __free_rcu() frees waiting_for_gp_ttrace only and clears the flag. Hence the objects that free_bulk() or __free_by_rcu() added while RCU tasks trace GP was in flight stay in free_by_rcu_ttrace until free_bulk() or alloc_bulk() is called for the same bpf_mem_cache again, which may never happen. The number of such objects is not bounded. Turn call_rcu_ttrace_in_progress into three states: 0 - idle 1 - __free_rcu() is queued 2 - __free_rcu() is queued and free_by_rcu_ttrace got more objects since do_call_rcu_ttrace() sets 2. __free_rcu() does cmpxchg(1 -> 0) and starts the next GP when it fails. It cannot clear the flag first and check free_by_rcu_ttrace later, since bpf_mem_alloc_destroy() frees bpf_mem_cache without waiting for RCU callbacks when the flag is zero. Now __free_rcu() queues itself, so the one that didn't see 'draining' may do call_rcu_tasks_trace() after rcu_barrier_tasks_trace() in free_mem_alloc(). Queue it under rcu_read_lock() and do synchronize_rcu() before the barriers. Calling rcu_barrier_tasks_trace() twice works too, but creating and destroying hash maps in a loop on many cpus slows down to one free_mem_alloc() per GP and kworkers pile up.