Vulnerabilities

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> cgroup: fix race between task migration and iteration<br /> <br /> When a task is migrated out of a css_set, cgroup_migrate_add_task()<br /> first moves it from cset-&gt;tasks to cset-&gt;mg_tasks via:<br /> <br /> list_move_tail(&amp;task-&gt;cg_list, &amp;cset-&gt;mg_tasks);<br /> <br /> If a css_task_iter currently has it-&gt;task_pos pointing to this task,<br /> css_set_move_task() calls css_task_iter_skip() to keep the iterator<br /> valid. However, since the task has already been moved to -&gt;mg_tasks,<br /> the iterator is advanced relative to the mg_tasks list instead of the<br /> original tasks list. As a result, remaining tasks on cset-&gt;tasks, as<br /> well as tasks queued on cset-&gt;mg_tasks, can be skipped by iteration.<br /> <br /> Fix this by calling css_set_skip_task_iters() before unlinking<br /> task-&gt;cg_list from cset-&gt;tasks. This advances all active iterators to<br /> the next task on cset-&gt;tasks, so iteration continues correctly even<br /> when a task is concurrently being migrated.<br /> <br /> This race is hard to hit in practice without instrumentation, but it<br /> can be reproduced by artificially slowing down cgroup_procs_show().<br /> For example, on an Android device a temporary<br /> /sys/kernel/cgroup/cgroup_test knob can be added to inject a delay<br /> into cgroup_procs_show(), and then:<br /> <br /> 1) Spawn three long-running tasks (PIDs 101, 102, 103).<br /> 2) Create a test cgroup and move the tasks into it.<br /> 3) Enable a large delay via /sys/kernel/cgroup/cgroup_test.<br /> 4) In one shell, read cgroup.procs from the test cgroup.<br /> 5) Within the delay window, in another shell migrate PID 102 by<br /> writing it to a different cgroup.procs file.<br /> <br /> Under this setup, cgroup.procs can intermittently show only PID 101<br /> while skipping PID 103. Once the migration completes, reading the<br /> file again shows all tasks as expected.<br /> <br /> Note that this change does not allow removing the existing<br /> css_set_skip_task_iters() call in css_set_move_task(). The new call<br /> in cgroup_migrate_add_task() only handles iterators that are racing<br /> with migration while the task is still on cset-&gt;tasks. Iterators may<br /> also start after the task has been moved to cset-&gt;mg_tasks. If we<br /> dropped css_set_skip_task_iters() from css_set_move_task(), such<br /> iterators could keep task_pos pointing to a migrating task, causing<br /> css_task_iter_advance() to malfunction on the destination css_set,<br /> up to and including crashes or infinite loops.<br /> <br /> The race window between migration and iteration is very small, and<br /> css_task_iter is not on a hot path. In the worst case, when an<br /> iterator is positioned on the first thread of the migrating process,<br /> cgroup_migrate_add_task() may have to skip multiple tasks via<br /> css_set_skip_task_iters(). However, this only happens when migration<br /> and iteration actually race, so the performance impact is negligible<br /> compared to the correctness fix provided here.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> net/mana: Null service_wq on setup error to prevent double destroy<br /> <br /> In mana_gd_setup() error path, set gc-&gt;service_wq to NULL after<br /> destroy_workqueue() to match the cleanup in mana_gd_cleanup().<br /> This prevents a use-after-free if the workqueue pointer is checked<br /> after a failed setup.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> net: bonding: Fix nd_tbl NULL dereference when IPv6 is disabled<br /> <br /> When booting with the &amp;#39;ipv6.disable=1&amp;#39; parameter, the nd_tbl is never<br /> initialized because inet6_init() exits before ndisc_init() is called<br /> which initializes it. If bonding ARP/NS validation is enabled, an IPv6<br /> NS/NA packet received on a slave can reach bond_validate_na(), which<br /> calls bond_has_this_ip6(). That path calls ipv6_chk_addr() and can<br /> crash in __ipv6_chk_addr_and_flags().<br /> <br /> BUG: kernel NULL pointer dereference, address: 00000000000005d8<br /> Oops: Oops: 0000 [#1] SMP NOPTI<br /> RIP: 0010:__ipv6_chk_addr_and_flags+0x69/0x170<br /> Call Trace:<br /> <br /> ipv6_chk_addr+0x1f/0x30<br /> bond_validate_na+0x12e/0x1d0 [bonding]<br /> ? __pfx_bond_handle_frame+0x10/0x10 [bonding]<br /> bond_rcv_validate+0x1a0/0x450 [bonding]<br /> bond_handle_frame+0x5e/0x290 [bonding]<br /> ? srso_alias_return_thunk+0x5/0xfbef5<br /> __netif_receive_skb_core.constprop.0+0x3e8/0xe50<br /> ? srso_alias_return_thunk+0x5/0xfbef5<br /> ? update_cfs_rq_load_avg+0x1a/0x240<br /> ? srso_alias_return_thunk+0x5/0xfbef5<br /> ? __enqueue_entity+0x5e/0x240<br /> __netif_receive_skb_one_core+0x39/0xa0<br /> process_backlog+0x9c/0x150<br /> __napi_poll+0x30/0x200<br /> ? srso_alias_return_thunk+0x5/0xfbef5<br /> net_rx_action+0x338/0x3b0<br /> handle_softirqs+0xc9/0x2a0<br /> do_softirq+0x42/0x60<br /> <br /> <br /> __local_bh_enable_ip+0x62/0x70<br /> __dev_queue_xmit+0x2d3/0x1000<br /> ? srso_alias_return_thunk+0x5/0xfbef5<br /> ? srso_alias_return_thunk+0x5/0xfbef5<br /> ? packet_parse_headers+0x10a/0x1a0<br /> packet_sendmsg+0x10da/0x1700<br /> ? kick_pool+0x5f/0x140<br /> ? srso_alias_return_thunk+0x5/0xfbef5<br /> ? __queue_work+0x12d/0x4f0<br /> __sys_sendto+0x1f3/0x220<br /> __x64_sys_sendto+0x24/0x30<br /> do_syscall_64+0x101/0xf80<br /> ? exc_page_fault+0x6e/0x170<br /> ? srso_alias_return_thunk+0x5/0xfbef5<br /> entry_SYSCALL_64_after_hwframe+0x77/0x7f<br /> <br /> <br /> Fix this by checking ipv6_mod_enabled() before dispatching IPv6 packets to<br /> bond_na_rcv(). If IPv6 is disabled, return early from bond_rcv_validate()<br /> and avoid the path to ipv6_chk_addr().

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> io_uring: fix physical SQE bounds check for SQE_MIXED 128-byte ops<br /> <br /> When IORING_SETUP_SQE_MIXED is used without IORING_SETUP_NO_SQARRAY,<br /> the boundary check for 128-byte SQE operations in io_init_req()<br /> validated the logical SQ head position rather than the physical SQE<br /> index.<br /> <br /> The existing check:<br /> <br /> !(ctx-&gt;cached_sq_head &amp; (ctx-&gt;sq_entries - 1))<br /> <br /> ensures the logical position isn&amp;#39;t at the end of the ring, which is<br /> correct for NO_SQARRAY rings where physical == logical. However, when<br /> sq_array is present, an unprivileged user can remap any logical<br /> position to an arbitrary physical index via sq_array. Setting<br /> sq_array[N] = sq_entries - 1 places a 128-byte operation at the last<br /> physical SQE slot, causing the 128-byte memcpy in<br /> io_uring_cmd_sqe_copy() to read 64 bytes past the end of the SQE<br /> array.<br /> <br /> Replace the cached_sq_head alignment check with a direct validation<br /> of the physical SQE index, which correctly handles both sq_array and<br /> NO_SQARRAY cases.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ASoC: amd: acp-mach-common: Add missing error check for clock acquisition<br /> <br /> The acp_card_rt5682_init() and acp_card_rt5682s_init() functions did not<br /> check the return values of clk_get(). This could lead to a kernel crash<br /> when the invalid pointers are later dereferenced by clock core<br /> functions.<br /> <br /> Fix this by:<br /> 1. Changing clk_get() to the device-managed devm_clk_get().<br /> 2. Adding IS_ERR() checks immediately after each clock acquisition.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> drm/amdkfd: Unreserve bo if queue update failed<br /> <br /> Error handling path should unreserve bo then return failed.<br /> <br /> (cherry picked from commit c24afed7de9ecce341825d8ab55a43a254348b33)

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> e1000/e1000e: Fix leak in DMA error cleanup<br /> <br /> If an error is encountered while mapping TX buffers, the driver should<br /> unmap any buffers already mapped for that skb.<br /> <br /> Because count is incremented after a successful mapping, it will always<br /> match the correct number of unmappings needed when dma_error is reached.<br /> Decrementing count before the while loop in dma_error causes an<br /> off-by-one error. If any mapping was successful before an unsuccessful<br /> mapping, exactly one DMA mapping would leak.<br /> <br /> In these commits, a faulty while condition caused an infinite loop in<br /> dma_error:<br /> Commit 03b1320dfcee ("e1000e: remove use of skb_dma_map from e1000e<br /> driver")<br /> Commit 602c0554d7b0 ("e1000: remove use of skb_dma_map from e1000 driver")<br /> <br /> Commit c1fa347f20f1 ("e1000/e1000e/igb/igbvf/ixgb/ixgbe: Fix tests of<br /> unsigned in *_tx_map()") fixed the infinite loop, but introduced the<br /> off-by-one error.<br /> <br /> This issue may still exist in the igbvf driver, but I did not address it<br /> in this patch.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> USB: usbtmc: Use usb_bulk_msg_killable() with user-specified timeouts<br /> <br /> The usbtmc driver accepts timeout values specified by the user in an<br /> ioctl command, and uses these timeouts for some usb_bulk_msg() calls.<br /> Since the user can specify arbitrarily long timeouts and<br /> usb_bulk_msg() uses unkillable waits, call usb_bulk_msg_killable()<br /> instead to avoid the possibility of the user hanging a kernel thread<br /> indefinitely.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> usb: yurex: fix race in probe<br /> <br /> The bbu member of the descriptor must be set to the value<br /> standing for uninitialized values before the URB whose<br /> completion handler sets bbu is submitted. Otherwise there is<br /> a window during which probing can overwrite already retrieved<br /> data.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> xhci: Fix NULL pointer dereference when reading portli debugfs files<br /> <br /> Michal reported and debgged a NULL pointer dereference bug in the<br /> recently added portli debugfs files<br /> <br /> Oops is caused when there are more port registers counted in<br /> xhci-&gt;max_ports than ports reported by Supported Protocol capabilities.<br /> This is possible if max_ports is more than maximum port number, or<br /> if there are gaps between ports of different speeds the &amp;#39;Supported<br /> Protocol&amp;#39; capabilities.<br /> <br /> In such cases port-&gt;rhub will be NULL so we can&amp;#39;t reach xhci behind it.<br /> Add an explicit NULL check for this case, and print portli in hex<br /> without dereferencing port-&gt;rhub.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> usb: xhci: Fix memory leak in xhci_disable_slot()<br /> <br /> xhci_alloc_command() allocates a command structure and, when the<br /> second argument is true, also allocates a completion structure.<br /> Currently, the error handling path in xhci_disable_slot() only frees<br /> the command structure using kfree(), causing the completion structure<br /> to leak.<br /> <br /> Use xhci_free_command() instead of kfree(). xhci_free_command() correctly<br /> frees both the command structure and the associated completion structure.<br /> Since the command structure is allocated with zero-initialization,<br /> command-&gt;in_ctx is NULL and will not be erroneously freed by<br /> xhci_free_command().<br /> <br /> This bug was found using an experimental static analysis tool we are<br /> developing. The tool is based on the LLVM framework and is specifically<br /> designed to detect memory management issues. It is currently under<br /> active development and not yet publicly available, but we plan to<br /> open-source it after our research is published.<br /> <br /> The bug was originally detected on v6.13-rc1 using our static analysis<br /> tool, and we have verified that the issue persists in the latest mainline<br /> kernel.<br /> <br /> We performed build testing on x86_64 with allyesconfig using GCC=11.4.0.<br /> Since triggering these error paths in xhci_disable_slot() requires specific<br /> hardware conditions or abnormal state, we were unable to construct a test<br /> case to reliably trigger these specific error paths at runtime.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> rust_binder: avoid reading the written value in offsets array<br /> <br /> When sending a transaction, its offsets array is first copied into the<br /> target proc&amp;#39;s vma, and then the values are read back from there. This is<br /> normally fine because the vma is a read-only mapping, so the target<br /> process cannot change the value under us.<br /> <br /> However, if the target process somehow gains the ability to write to its<br /> own vma, it could change the offset before it&amp;#39;s read back, causing the<br /> kernel to misinterpret what the sender meant. If the sender happens to<br /> send a payload with a specific shape, this could in the worst case lead<br /> to the receiver being able to privilege escalate into the sender.<br /> <br /> The intent is that gaining the ability to change the read-only vma of<br /> your own process should not be exploitable, so remove this TOCTOU read<br /> even though it&amp;#39;s unexploitable without another Binder bug.