Total
398628 CVE
| CVE | Vendors | Products | Updated | CVSS v2 | CVSS v3 |
|---|---|---|---|---|---|
| CVE-2026-72139 | 2026-08-17 | N/A | 9.8 CRITICAL | ||
| In the Linux kernel, the following vulnerability has been resolved: tcp: defer md5sig_info kfree past RCU grace period in tcp_connect The md5+ao reconciliation in tcp_connect() (net/ipv4/tcp_output.c) has two symmetric branches: if (needs_md5) { tcp_ao_destroy_sock(sk, false); } else if (needs_ao) { tcp_clear_md5_list(sk); kfree(rcu_replace_pointer(tp->md5sig_info, NULL, ...)); } Both branches free a per-socket auth-info object while the socket is in TCP_SYN_SENT and is already on the inet ehash (inserted by inet_hash_connect() in tcp_v4_connect()). Both branches are reachable by softirq RX-path readers that load the corresponding info pointer via implicit RCU before bh_lock_sock_nested() is taken. The needs_md5 branch is fixed in the prior patch by re-introducing the call_rcu() free in tcp_ao_destroy_sock(): the equivalent per-key loop runs inside tcp_ao_info_free_rcu(), the RCU callback, so by the time it frees each tcp_ao_key all softirq readers that captured the container have already completed rcu_read_unlock(). The needs_ao branch is not symmetric in the same way. The container free can be deferred via kfree_rcu(md5sig, rcu) -- struct tcp_md5sig_info already has the required rcu member (include/net/tcp.h:1999-2002), and the rest of the tree already does this in the tcp_md5sig_info_add() rollback paths (net/ipv4/tcp_ipv4.c:1410, 1436). But the per-key teardown is done by tcp_clear_md5_list() in process context BEFORE the container's RCU grace period: it walks &md5sig->head and frees each tcp_md5sig_key with bare hlist_del + kfree. A concurrent softirq reader in __tcp_md5_do_lookup() / __tcp_md5_do_lookup_exact() (tcp_ipv4.c:1253, 1298) walks the same list via hlist_for_each_entry_rcu() and races with that bare kfree on the keys themselves -- a per-key slab use-after-free of the same class as the TCP-AO bug, on the same race window. Fix this in two halves: 1. Convert the bare kfree() in tcp_connect() to kfree_rcu() so the md5sig_info container joins the rest of the md5sig lifecycle. The local-variable lift is mechanical and required because kfree_rcu() is a macro that expects an lvalue. 2. Make tcp_clear_md5_list() RCU-safe by replacing hlist_del + kfree(key) with hlist_del_rcu + kfree_rcu(key, rcu). struct tcp_md5sig_key already carries the rcu member (include/net/tcp.h:1995) and tcp_md5_do_del() (net/ipv4/tcp_ipv4.c:1456) already uses kfree_rcu, so this restores the lifecycle invariant the rest of the file follows rather than introducing a one-off. The other caller of tcp_clear_md5_list() is tcp_md5_destruct_sock() (net/ipv4/tcp.c:412), which runs from the sock destructor when the socket is already unhashed and unreachable; the extra grace period there is unnecessary but harmless. Making the helper unconditionally RCU-safe is the cleaner contract. The needs_ao branch is not reachable by the userns reproducer used to demonstrate the AO-side splat (the repro installs both keys but ends up in the needs_md5 branch because the connect peer matches the MD5 key, not the AO key); however the symmetric race exists and a maintainer touching this code should not have to think about which branch escapes RCU and which one does not. [also credits to Qihang, who found that this races with tcp-diag] | |||||
| CVE-2026-72138 | 2026-08-17 | N/A | N/A | ||
| In the Linux kernel, the following vulnerability has been resolved: xen/gntdev: fix error handling in ioctl When gntdev_ioctl_map_grant_ref() fails to copy the operation result back to userspace after successfully adding the mapping to the list, the error path returns -EFAULT without releasing the reference acquired by gntdev_alloc_map(). The mapping remains in priv->maps with a refcount of 1, causing a memory leak and a dangling list entry. Additionally, gntdev_add_map() may modify map->index to avoid overlap with existing mappings. Therefore, the index returned to userspace must be obtained after gntdev_add_map() completes. Fix this by holding the mutex across gntdev_add_map(), retrieving the correct index, and copy_to_user(). If copy_to_user() fails, remove the mapping from the list and release the reference while still holding the lock. Fix these issues by properly handling all error cases. | |||||
| CVE-2026-72137 | 2026-08-17 | N/A | 9.8 CRITICAL | ||
| In the Linux kernel, the following vulnerability has been resolved: xfrm: nat_keepalive: avoid double free on send error nat_keepalive_send() frees the keepalive skb whenever the IPv4 or IPv6 send helper reports an error. That cleanup is only correct before the skb is handed to the output path. Once ip_build_and_send_pkt() or ip6_xmit() takes ownership, the networking stack may already have consumed the skb before returning an error, so freeing it again is unsafe. Handle the pre-handoff failure cases inside nat_keepalive_send_ipv4() and nat_keepalive_send_ipv6(), where the caller still owns the skb, and keep nat_keepalive_send() responsible only for family dispatch and the unsupported-family cleanup path. | |||||
| CVE-2026-72136 | 2026-08-17 | N/A | 8.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: xfrm: xfrm_interface: require CAP_NET_ADMIN in the device netns for changelink xfrmi_changelink() operates on at most two netns, dev_net(dev) and the interface link netns xi->net. They differ once the device is created in or moved to a netns other than the one the request runs in. The rtnl changelink path checks CAP_NET_ADMIN only against dev_net(dev), so a caller privileged there but not in xi->net can rewrite an interface that lives in xi->net. Gate xfrmi_changelink() on rtnl_dev_link_net_capable() at its top, before any attribute is parsed. | |||||
| CVE-2026-72135 | 2026-08-17 | N/A | 7.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: tpm: Make the TPM character devices non-seekable The TPM character devices expose a sequential command/response interface, but their open handlers leave FMODE_PREAD and FMODE_PWRITE enabled. After a command leaves a response pending, pread(fd, buf, 16, 0x1400) passes 0x1400 as *off to tpm_common_read(). The transfer length is bounded by response_length, but the offset is used unchecked when forming data_buffer + *off. A sufficiently large offset therefore causes an out-of-bounds heap read through copy_to_user() and, if the copy succeeds, an out-of-bounds zero-write through the following memset(). Positional I/O does not provide coherent semantics for this interface. An arbitrary pread offset cannot represent how much of a response has been consumed sequentially. The write callback always stores a command at the start of data_buffer, while pwrite() does not update file->f_pos and can leave the sequential read cursor stale. Call nonseekable_open() from both open handlers. This removes FMODE_PREAD and FMODE_PWRITE, causing positional reads and writes to fail with -ESPIPE before reaching the TPM callbacks, and explicitly marks the files non-seekable. Normal read() and write() continue to use the existing sequential f_pos cursor, leaving the response state machine unchanged. Tested on Linux 6.12 with KASAN and a swtpm TPM2 device: - sequential partial reads returned the complete response - pread() and preadv() with offset 0x1400 returned -ESPIPE - pwrite() and pwritev() with offset zero returned -ESPIPE - the pending response remained intact after the rejected operations - a subsequent normal command/response cycle completed normally - no KASAN report was produced. | |||||
| CVE-2026-72134 | 2026-08-17 | N/A | 7.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: spi: imx: reconfigure for PIO when DMA cannot be started When spi_imx_can_dma() selects DMA, the ECSPI is configured for DMA: spi_imx_setupxfer() sets CTRL.SMC and clears dynamic_burst, and spi_imx_dma_transfer() programs the dynamic-burst BURST_LENGTH and the SDMA watermarks. If the DMA descriptor cannot be prepared (dmaengine_prep_slave_single() returns NULL), the transfer is failed with SPI_TRANS_FAIL_NO_START and falls back to PIO. The dynamic-burst DMA path uses its own bounce buffers instead of the SPI core's mapping, so xfer->{tx,rx}_sg_mapped are not set and the core's DMA->PIO retry is skipped; the driver falls back to PIO internally. But none of the DMA-mode configuration is undone, so the PIO transfer runs with CTRL.SMC set, the wrong burst length and dynamic_burst cleared, and the transferred data is corrupted. This is easily hit on i.MX8MP boards that describe ECSPI DMA in the device tree but run SDMA on ROM firmware (no external sdma-imx7d.bin): every ECSPI DMA prepare fails. An Infineon SLB9670 TPM on ECSPI1 then returns shifted TPM2_GetCapability data, is flagged "field failure mode", /dev/tpmrm0 is never created. Set controller->fallback before re-running spi_imx_setupxfer() so the ECSPI is reconfigured exactly like a normal PIO transfer. With controller->fallback set, spi_imx_setupxfer() sees spi_imx_can_dma() return false, so it clears spi_imx->usedma and reprograms the controller (clears CTRL.SMC, restores dynamic_burst and the PIO burst length). No explicit spi_imx->usedma = false is needed: setupxfer() already updates it from the can_dma() result. | |||||
| CVE-2026-72133 | 2026-08-17 | N/A | 8.4 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: spi: uniphier: Fix completion initialization order before devm_request_irq() The driver calls devm_request_irq() before initializing the completion used by the interrupt handler. Because the interrupt may occur immediately after devm_request_irq(), the handler may execute before init_completion(). This may result in calling complete() on an uninitialized completion, causing undefined behavior. This has been observed with KASAN. Fix this by initializing the completion before registering the IRQ. | |||||
| CVE-2026-72132 | 2026-08-17 | N/A | N/A | ||
| In the Linux kernel, the following vulnerability has been resolved: NFS: Charge unstable writes by request size, not folio size nfs_folio_mark_unstable() and nfs_folio_clear_commit() charge and uncharge NR_WRITEBACK/WB_WRITEBACK by folio_nr_pages(folio) once per *request* added to or removed from a commit list. This is correct only when a folio has a single associated request. When pg_test splits a folio into N sub-folio requests (e.g. pNFS flexfiles striping with a stripe unit smaller than the folio size, or plain wsize-limited splitting), each of the N requests independently charges the whole folio's page count, inflating the accounting by a factor of N per folio. With large folios and small stripe units this reaches multiple orders of magnitude: a 2 MiB folio split into 512 4 KiB requests can charge up to 512x its real size, pushing global dirty+writeback accounting past the system's dirty threshold and forcing every buffered writer on the host into the hard-throttle path, including unrelated in-kernel NFS server threads sharing the box. Charge each request only for the pages it actually covers. | |||||
| CVE-2026-72130 | 2026-08-17 | N/A | 9.8 CRITICAL | ||
| In the Linux kernel, the following vulnerability has been resolved: nvmet-auth: reject short AUTH_RECEIVE buffers nvmet_execute_auth_receive() trusts the AUTH_RECEIVE allocation length after checking only that it is nonzero and matches the transfer length. In the SUCCESS1 and FAILURE1/default states, that lets a remote NVMe-oF initiator reach the fixed-size DH-HMAC-CHAP response builders with a kmalloc() buffer shorter than the response, so nvmet_auth_success1() and nvmet_auth_failure1() write past the allocation; both only WARN_ON the short length and then format the message anyway. Impact: A remote NVMe-oF initiator with access to an auth-enabled target can trigger a 16-byte heap out-of-bounds write via a one-byte AUTH_RECEIVE allocation length. Compute the minimum response length for the current DH-HMAC-CHAP step in nvmet_auth_receive_data_len() and report a zero data length when the host-supplied allocation length is shorter, so the existing zero-length check in nvmet_execute_auth_receive() rejects the command before any builder runs. The SUCCESS1 minimum is sizeof(struct nvmf_auth_dhchap_success1_data) plus the HMAC hash length, because the response hash is written into the rval[] flexible-array tail, so the minimum is state dependent rather than a flat sizeof. CHALLENGE keeps its existing variable-length guard in nvmet_auth_challenge(). This is reachable only when in-band DH-HMAC-CHAP authentication is configured on the target. | |||||
| CVE-2026-72129 | 2026-08-17 | N/A | 9.8 CRITICAL | ||
| In the Linux kernel, the following vulnerability has been resolved: nvmet-rdma: handle inline data with a nonzero offset nvmet_rdma_use_inline_sg() maps the host-controlled inline data offset into the per-command inline scatterlist. The bounds check admits any offset with off + len <= inline_data_size, but the mapping still assumes the data begins in the first inline page: sg->offset = off; sg->length = min_t(int, len, PAGE_SIZE - off); When a port is configured with inline_data_size > PAGE_SIZE (settable up to max(SZ_16K, PAGE_SIZE)), an offset in (PAGE_SIZE, inline_data_size] makes "PAGE_SIZE - off" underflow, so sg->length is set to ~4 GiB and the block backend reads far past the first inline page. num_pages(len) also ignores the offset, so an in-bounds offset whose [off, off+len) span crosses a page boundary under-counts the scatterlist. Map the offset properly: split it into a page index and an in-page offset, start the scatterlist at that page, and size the page count from page_off + len. Because the request scatterlist may now start at inline_sg[page_idx] rather than inline_sg[0], generalize the inline-SGL identity test in nvmet_rdma_release_rsp() to a range test; otherwise the persistent inline scatterlist is mistaken for an allocated one and nvmet_req_free_sgls() frees an inline page (and warns in free_large_kmalloc()). | |||||
| CVE-2026-72128 | 2026-08-17 | N/A | N/A | ||
| In the Linux kernel, the following vulnerability has been resolved: nvmet: fix refcount leak in nvmet_sq_create() In nvmet_sq_create(), a reference on the ctrl is taken via kref_get_unless_zero() before calling nvmet_check_sqid(). If nvmet_check_sqid() fails, the function returns the error directly without releasing the reference, leading to a leak. Fix this by jumping to the "ctrl_put" label, which already performs the necessary nvmet_ctrl_put(ctrl). This ensures the reference is properly released on this error path. | |||||
| CVE-2026-72127 | 2026-08-17 | N/A | N/A | ||
| In the Linux kernel, the following vulnerability has been resolved: netdev-genl: report NAPI thread PID in the caller's pid namespace netdev_nl_napi_fill_one() reports the NAPI kthread PID in NETDEV_A_NAPI_PID using task_pid_nr(), which returns the PID in the initial pid namespace. NETDEV_CMD_NAPI_GET does not have GENL_ADMIN_PERM and the netdev genl family is netnsok, so a caller in a child pid namespace can issue it. That caller then sees the kthread's global PID, even though the kthread is not visible in its pid namespace, where the value should be 0. Translate the PID through the caller's pid namespace, the same way commit 3799c2570982 ("io_uring/fdinfo: translate SqThread PID through caller's pid_ns") did for the io_uring SQPOLL thread. The doit and dumpit paths both run synchronously in the caller's context, so task_active_pid_ns(current) is the caller's pid namespace. | |||||
| CVE-2026-72126 | 2026-08-17 | N/A | 7.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: can: isotp: use unconditional synchronize_rcu() in isotp_release() isotp_notify() unregisters the (RCU) CAN filters via can_rx_unregister() and clears so->bound without waiting for a grace period. isotp_release() uses so->bound to decide whether it needs to call synchronize_rcu() before cancelling so->rxtimer, so when NETDEV_UNREGISTER runs first it skips that synchronize_rcu() and can cancel the timer while an in-flight isotp_rcv() is still executing and about to re-arm it via isotp_send_fc(), leading to a use-after-free timer callback on the freed socket. sakisho-bot remarked a problem with rtnl_lock held in isotp_notify(), therefore make isotp_release() always call synchronize_rcu() before cancelling the timers, regardless of so->bound. This still closes the original race (isotp_notify() clearing so->bound without waiting for in-flight isotp_rcv() callers before isotp_release() cancels the RX timer) without adding any RCU wait to the netdevice notifier path. | |||||
| CVE-2026-72122 | 2026-08-17 | N/A | 7.3 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: can: bcm: fix lockless bound/ifindex race and silent RX_SETUP failure bcm_sendmsg() reads bo->ifindex and checks bo->bound before taking lock_sock(), while bcm_notify(), bcm_connect() and bcm_release() all mutate both fields under that same lock. Because the lockless reads and the locked writes are unordered with respect to each other, a racing bcm_notify() (device unregister) or bcm_connect() (concurrent bind on another thread sharing the socket) can make bcm_sendmsg() observe an inconsistent combination, e.g. a stale bound=1 together with the now-cleared ifindex=0, silently turning a socket bound to a specific CAN interface into one that also matches "any" interface. Keep the lockless bo->bound check purely as a fast-path reject, and move the ifindex read (and a bo->bound re-check) into the locked section, where every writer already serializes. This removes the possibility of observing the two fields torn against each other, rather than trying to fix it with more READ_ONCE()/WRITE_ONCE() pairs on two independently updated fields. Annotate the now-purely-lockless bo->bound accesses consistently across all its write sites. Also fix bcm_rx_setup() silently returning success when the target device disappears concurrently instead of reporting -ENODEV, so a broken RX op is no longer left registered as if it had succeeded. | |||||
| CVE-2026-72120 | 2026-08-17 | N/A | 7.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: can: bcm: add missing rcu list annotations and operations sashiko-bot remarked the missing use of list_add_rcu() in bcm_[rx|tx]_setup() to have a proper initialized bcm_op structure when bcm_proc_show() traverses the bcm_op's under rcu_read_lock(). To cover all initial settings of the bcm_op's the list_add_rcu() calls are moved to the end of the setup code. While at it, also fix the mirroring removal side: bcm_release() called bcm_remove_op() - which frees the op via call_rcu() - on ops that were still linked in bo->tx_ops/bo->rx_ops, without list_del_rcu() first. Unlink each op with list_del_rcu() before handing it to bcm_remove_op(), matching the existing pattern in bcm_delete_tx_op()/bcm_delete_rx_op(). | |||||
| CVE-2026-72112 | 2026-08-17 | N/A | 7.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: io_uring/bpf-ops: reject re-registration of an already-bound ops io_install_bpf() only rejects a second registration on the ctx side (ctx->bpf_ops) and sets the per-map back-pointer ops->priv unconditionally. The struct_ops link path never advances a map past BPF_STRUCT_OPS_STATE_READY, so the same io_uring_bpf_ops map can be registered more than once, and bpf_io_reg() re-resolves the target ring via fget(ops->ring_fd) on every call. A caller can therefore point the same ring_fd at a different io_ring_ctx between two BPF_LINK_CREATE calls. The second registration passes the ctx->bpf_ops check (the new ctx has none) and overwrites ops->priv, orphaning the first ctx. Teardown (io_eject_bpf()/bpf_io_unreg()) only reaches a ctx through ops->priv, so the orphaned ctx is never torn down: its ctx->loop_step keeps pointing into the struct_ops trampoline, which is freed once the map is gone. A later io_uring_enter() on the orphaned ring then calls the dangling ctx->loop_step from io_run_loop() -- a use-after-free of freed executable memory, reachable by a task with CAP_BPF + CAP_PERFMON. Reject registration when ops->priv is already set, as hid_bpf_reg() does for its struct_ops. | |||||
| CVE-2026-72111 | 2026-08-17 | N/A | 8.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: bpf: Reset register bounds before narrowing retval range in check_mem_access() When the BPF verifier processes a context load of an LSM hook return value, it calls __mark_reg_s32_range() to narrow the register to the hook's valid range. However, __mark_reg_s32_range() intersects the new range with the register's existing bounds using max_t()/min_t() rather than replacing them. If the destination register carries stale bounds from a prior instruction (e.g. BPF_MOV64_IMM), the intersection can produce a range narrower than reality. The verifier then believes it knows the register's exact value, while at runtime the actual hook return value is loaded, creating a verifier/runtime mismatch that can be used to bypass BPF memory safety checks. The else branch already calls mark_reg_unknown() to reset register state before any narrowing. Apply the same reset in the is_retval path so stale bounds are cleared before __mark_reg_s32_range() intersects. | |||||
| CVE-2026-72109 | 2026-08-17 | N/A | 7.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: net: sparx5: unregister blocking notifier on init failure sparx5_register_notifier_blocks() registers the switchdev blocking notifier before allocating the ordered workqueue. If the workqueue allocation fails, the error path unregisters the switchdev and netdevice notifiers, but leaves the blocking notifier registered. Add a separate error label for the workqueue allocation failure path and unregister the switchdev blocking notifier there. | |||||
| CVE-2026-72108 | 2026-08-17 | N/A | 7.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: dm thin metadata: fix metadata snapshot consistency on commit failure __reserve_metadata_snap() and __release_metadata_snap() modify the superblock's held_root directly in the block_manager's buffer. If the subsequent metadata commit fails, the held_root gets flushed to disk through the abort_transaction path, resulting in inconsistent metadata. Reproducer 1: __reserve_metadata_snap() 1. Create a 2 MiB metadata device and make the region after the 14th block inaccessible, to trigger metadata commit failure in the subsequent reserve_metadata_snap operation. The 14th block will be the shadow destination for the index block. dmsetup create tmeta --table "0 112 linear /dev/sdc 0 112 3984 error" 2. Create a 16 MiB thin-pool dmsetup create tdata --table "0 32768 zero" dd if=/dev/zero of=/dev/mapper/tmeta bs=4k count=1 dmsetup create tpool --table "0 32768 thin-pool /dev/mapper/tmeta \ /dev/mapper/tdata 128 0 1 skip_block_zeroing" 3. Take a metadata snapshot to trigger metadata commit failure and transaction abort. However, the held_root is written to disk, breaking metadata consistency. dmsetup message tpool 0 "reserve_metadata_snap" thin_check v1.2.2 result: Bad reference count for metadata block 6. Expected 2, but space map contains 1. Bad reference count for metadata block 7. Expected 2, but space map contains 1. Bad reference count for metadata block 13. Expected 1, but space map contains 0. Reproducer 2: __release_metadata_snap() 1. Create a 2 MiB metadata device and make the region after the 16th block inaccessible, to trigger metadata commit failure in the subsequent release_metadata_snap operation. The 16th block will be the shadow destination for the index block. dmsetup create tmeta --table "0 128 linear /dev/sdc 0 128 3968 error" 2. Create a 16 MiB thin-pool dmsetup create tdata --table "0 32768 zero" dd if=/dev/zero of=/dev/mapper/tmeta bs=4k count=1 dmsetup create tpool --table "0 32768 thin-pool /dev/mapper/tmeta \ /dev/mapper/tdata 128 0 1 skip_block_zeroing" 3. Reserve then release the metadata snapshot, to trigger metadata commit failure and transaction abort. The held_root gets removed from the on-disk superblock, causing inconsistent metadata. dmsetup message tpool 0 "reserve_metadata_snap" dmsetup message tpool 0 "release_metadata_snap" thin_check v1.2.2 result: Bad reference count for metadata block 6. Expected 1, but space map contains 2. Bad reference count for metadata block 7. Expected 1, but space map contains 2. 1 metadata blocks have leaked. Fix by deferring the held_root update to commit time. Additionally, move the existing-snapshot check in __reserve_metadata_snap before the shadow operation to avoid unnecessary work. In __release_metadata_snap, clear pmd->held_root before btree deletion so partial failure leaks blocks rather than leaving a stale reference, and unlock the snapshot block before decrementing its refcount. | |||||
| CVE-2026-72107 | 2026-08-17 | N/A | 8.8 HIGH | ||
| In the Linux kernel, the following vulnerability has been resolved: dm era: fix out-of-bounds memory access for non-zero start sector dm-era tracks writes in target-relative blocks, but era_map() calculates the writeset block before applying the target offset. Tables with a non-zero start sector can therefore pass an absolute mapped-device block to metadata_current_marked(). If the absolute block is beyond the current writeset size, writeset_marked() tests past the end of the in-core bitset. KASAN reports this as a vmalloc-out-of-bounds access. Apply the target offset before calculating the era block so writeset lookups use the target-relative block number. | |||||