Vulnerabilities

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> drm/plane: Fix create_in_format_blob() return value<br /> <br /> create_in_format_blob() is either supposed to return a valid<br /> pointer or an error, but never NULL. The caller will dereference<br /> the blob when it is not an error, and thus will oops if NULL<br /> returned. Return proper error values in the failure cases.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> cifs: fix memory leak in smb3_fs_context_parse_param error path<br /> <br /> Add proper cleanup of ctx-&gt;source and fc-&gt;source to the<br /> cifs_parse_mount_err error handler. This ensures that memory allocated<br /> for the source strings is correctly freed on all error paths, matching<br /> the cleanup already performed in the success path by<br /> smb3_cleanup_fs_context_contents().<br /> Pointers are also set to NULL after freeing to prevent potential<br /> double-free issues.<br /> <br /> This change fixes a memory leak originally detected by syzbot. The<br /> leak occurred when processing Opt_source mount options if an error<br /> happened after ctx-&gt;source and fc-&gt;source were successfully<br /> allocated but before the function completed.<br /> <br /> The specific leak sequence was:<br /> 1. ctx-&gt;source = smb3_fs_context_fullpath(ctx, &amp;#39;/&amp;#39;) allocates memory<br /> 2. fc-&gt;source = kstrdup(ctx-&gt;source, GFP_KERNEL) allocates more memory<br /> 3. A subsequent error jumps to cifs_parse_mount_err<br /> 4. The old error handler freed passwords but not the source strings,<br /> causing the memory to leak.<br /> <br /> This issue was not addressed by commit e8c73eb7db0a ("cifs: client:<br /> fix memory leak in smb3_fs_context_parse_param"), which only fixed<br /> leaks from repeated fsconfig() calls but not this error path.<br /> <br /> Patch updated with minor change suggested by kernel test robot

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> net: ethernet: ti: netcp: Standardize knav_dma_open_channel to return NULL on error<br /> <br /> Make knav_dma_open_channel consistently return NULL on error instead<br /> of ERR_PTR. Currently the header include/linux/soc/ti/knav_dma.h<br /> returns NULL when the driver is disabled, but the driver<br /> implementation does not even return NULL or ERR_PTR on failure,<br /> causing inconsistency in the users. This results in a crash in<br /> netcp_free_navigator_resources as followed (trimmed):<br /> <br /> Unhandled fault: alignment exception (0x221) at 0xfffffff2<br /> [fffffff2] *pgd=80000800207003, *pmd=82ffda003, *pte=00000000<br /> Internal error: : 221 [#1] SMP ARM<br /> Modules linked in:<br /> CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.17.0-rc7 #1 NONE<br /> Hardware name: Keystone<br /> PC is at knav_dma_close_channel+0x30/0x19c<br /> LR is at netcp_free_navigator_resources+0x2c/0x28c<br /> <br /> [... TRIM...]<br /> <br /> Call trace:<br /> knav_dma_close_channel from netcp_free_navigator_resources+0x2c/0x28c<br /> netcp_free_navigator_resources from netcp_ndo_open+0x430/0x46c<br /> netcp_ndo_open from __dev_open+0x114/0x29c<br /> __dev_open from __dev_change_flags+0x190/0x208<br /> __dev_change_flags from netif_change_flags+0x1c/0x58<br /> netif_change_flags from dev_change_flags+0x38/0xa0<br /> dev_change_flags from ip_auto_config+0x2c4/0x11f0<br /> ip_auto_config from do_one_initcall+0x58/0x200<br /> do_one_initcall from kernel_init_freeable+0x1cc/0x238<br /> kernel_init_freeable from kernel_init+0x1c/0x12c<br /> kernel_init from ret_from_fork+0x14/0x38<br /> [... TRIM...]<br /> <br /> Standardize the error handling by making the function return NULL on<br /> all error conditions. The API is used in just the netcp_core.c so the<br /> impact is limited.<br /> <br /> Note, this change, in effect reverts commit 5b6cb43b4d62 ("net:<br /> ethernet: ti: netcp_core: return error while dma channel open issue"),<br /> but provides a less error prone implementation.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> mptcp: fix address removal logic in mptcp_pm_nl_rm_addr<br /> <br /> Fix inverted WARN_ON_ONCE condition that prevented normal address<br /> removal counter updates. The current code only executes decrement<br /> logic when the counter is already 0 (abnormal state), while<br /> normal removals (counter &gt; 0) are ignored.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> pinctrl: s32cc: fix uninitialized memory in s32_pinctrl_desc<br /> <br /> s32_pinctrl_desc is allocated with devm_kmalloc(), but not all of its<br /> fields are initialized. Notably, num_custom_params is used in<br /> pinconf_generic_parse_dt_config(), resulting in intermittent allocation<br /> errors, such as the following splat when probing i2c-imx:<br /> <br /> WARNING: CPU: 0 PID: 176 at mm/page_alloc.c:4795 __alloc_pages_noprof+0x290/0x300<br /> [...]<br /> Hardware name: NXP S32G3 Reference Design Board 3 (S32G-VNP-RDB3) (DT)<br /> [...]<br /> Call trace:<br /> __alloc_pages_noprof+0x290/0x300 (P)<br /> ___kmalloc_large_node+0x84/0x168<br /> __kmalloc_large_node_noprof+0x34/0x120<br /> __kmalloc_noprof+0x2ac/0x378<br /> pinconf_generic_parse_dt_config+0x68/0x1a0<br /> s32_dt_node_to_map+0x104/0x248<br /> dt_to_map_one_config+0x154/0x1d8<br /> pinctrl_dt_to_map+0x12c/0x280<br /> create_pinctrl+0x6c/0x270<br /> pinctrl_get+0xc0/0x170<br /> devm_pinctrl_get+0x50/0xa0<br /> pinctrl_bind_pins+0x60/0x2a0<br /> really_probe+0x60/0x3a0<br /> [...]<br /> __platform_driver_register+0x2c/0x40<br /> i2c_adap_imx_init+0x28/0xff8 [i2c_imx]<br /> [...]<br /> <br /> This results in later parse failures that can cause issues in dependent<br /> drivers:<br /> <br /> s32g-siul2-pinctrl 4009c240.pinctrl: /soc@0/pinctrl@4009c240/i2c0-pins/i2c0-grp0: could not parse node property<br /> s32g-siul2-pinctrl 4009c240.pinctrl: /soc@0/pinctrl@4009c240/i2c0-pins/i2c0-grp0: could not parse node property<br /> [...]<br /> pca953x 0-0022: failed writing register: -6<br /> i2c i2c-0: IMX I2C adapter registered<br /> s32g-siul2-pinctrl 4009c240.pinctrl: /soc@0/pinctrl@4009c240/i2c2-pins/i2c2-grp0: could not parse node property<br /> s32g-siul2-pinctrl 4009c240.pinctrl: /soc@0/pinctrl@4009c240/i2c2-pins/i2c2-grp0: could not parse node property<br /> i2c i2c-1: IMX I2C adapter registered<br /> s32g-siul2-pinctrl 4009c240.pinctrl: /soc@0/pinctrl@4009c240/i2c4-pins/i2c4-grp0: could not parse node property<br /> s32g-siul2-pinctrl 4009c240.pinctrl: /soc@0/pinctrl@4009c240/i2c4-pins/i2c4-grp0: could not parse node property<br /> i2c i2c-2: IMX I2C adapter registered<br /> <br /> Fix this by initializing s32_pinctrl_desc with devm_kzalloc() instead of<br /> devm_kmalloc() in s32_pinctrl_probe(), which sets the previously<br /> uninitialized fields to zero.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> drm/radeon: delete radeon_fence_process in is_signaled, no deadlock<br /> <br /> Delete the attempt to progress the queue when checking if fence is<br /> signaled. This avoids deadlock.<br /> <br /> dma-fence_ops::signaled can be called with the fence lock in unknown<br /> state. For radeon, the fence lock is also the wait queue lock. This can<br /> cause a self deadlock when signaled() tries to make forward progress on<br /> the wait queue. But advancing the queue is unneeded because incorrectly<br /> returning false from signaled() is perfectly acceptable.<br /> <br /> (cherry picked from commit 527ba26e50ec2ca2be9c7c82f3ad42998a75d0db)

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> scsi: core: Fix a regression triggered by scsi_host_busy()<br /> <br /> Commit 995412e23bb2 ("blk-mq: Replace tags-&gt;lock with SRCU for tag<br /> iterators") introduced the following regression:<br /> <br /> Call trace:<br /> __srcu_read_lock+0x30/0x80 (P)<br /> blk_mq_tagset_busy_iter+0x44/0x300<br /> scsi_host_busy+0x38/0x70<br /> ufshcd_print_host_state+0x34/0x1bc<br /> ufshcd_link_startup.constprop.0+0xe4/0x2e0<br /> ufshcd_init+0x944/0xf80<br /> ufshcd_pltfrm_init+0x504/0x820<br /> ufs_rockchip_probe+0x2c/0x88<br /> platform_probe+0x5c/0xa4<br /> really_probe+0xc0/0x38c<br /> __driver_probe_device+0x7c/0x150<br /> driver_probe_device+0x40/0x120<br /> __driver_attach+0xc8/0x1e0<br /> bus_for_each_dev+0x7c/0xdc<br /> driver_attach+0x24/0x30<br /> bus_add_driver+0x110/0x230<br /> driver_register+0x68/0x130<br /> __platform_driver_register+0x20/0x2c<br /> ufs_rockchip_pltform_init+0x1c/0x28<br /> do_one_initcall+0x60/0x1e0<br /> kernel_init_freeable+0x248/0x2c4<br /> kernel_init+0x20/0x140<br /> ret_from_fork+0x10/0x20<br /> <br /> Fix this regression by making scsi_host_busy() check whether the SCSI<br /> host tag set has already been initialized. tag_set-&gt;ops is set by<br /> scsi_mq_setup_tags() just before blk_mq_alloc_tag_set() is called. This<br /> fix is based on the assumption that scsi_host_busy() and<br /> scsi_mq_setup_tags() calls are serialized. This is the case in the UFS<br /> driver.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> lib/test_kho: check if KHO is enabled<br /> <br /> We must check whether KHO is enabled prior to issuing KHO commands,<br /> otherwise KHO internal data structures are not initialized.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ksm: use range-walk function to jump over holes in scan_get_next_rmap_item<br /> <br /> Currently, scan_get_next_rmap_item() walks every page address in a VMA to<br /> locate mergeable pages. This becomes highly inefficient when scanning<br /> large virtual memory areas that contain mostly unmapped regions, causing<br /> ksmd to use large amount of cpu without deduplicating much pages.<br /> <br /> This patch replaces the per-address lookup with a range walk using<br /> walk_page_range(). The range walker allows KSM to skip over entire<br /> unmapped holes in a VMA, avoiding unnecessary lookups. This problem was<br /> previously discussed in [1].<br /> <br /> Consider the following test program which creates a 32 TiB mapping in the<br /> virtual address space but only populates a single page:<br /> <br /> #include <br /> #include <br /> #include <br /> <br /> /* 32 TiB */<br /> const size_t size = 32ul * 1024 * 1024 * 1024 * 1024;<br /> <br /> int main() {<br /> char *area = mmap(NULL, size, PROT_READ | PROT_WRITE,<br /> MAP_NORESERVE | MAP_PRIVATE | MAP_ANON, -1, 0);<br /> <br /> if (area == MAP_FAILED) {<br /> perror("mmap() failed\n");<br /> return -1;<br /> }<br /> <br /> /* Populate a single page such that we get an anon_vma. */<br /> *area = 0;<br /> <br /> /* Enable KSM. */<br /> madvise(area, size, MADV_MERGEABLE);<br /> pause();<br /> return 0;<br /> }<br /> <br /> $ ./ksm-sparse &amp;<br /> $ echo 1 &gt; /sys/kernel/mm/ksm/run <br /> <br /> Without this patch ksmd uses 100% of the cpu for a long time (more then 1<br /> hour in my test machine) scanning all the 32 TiB virtual address space<br /> that contain only one mapped page. This makes ksmd essentially deadlocked<br /> not able to deduplicate anything of value. With this patch ksmd walks<br /> only the one mapped page and skips the rest of the 32 TiB virtual address<br /> space, making the scan fast using little cpu.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> fs: Fix uninitialized &amp;#39;offp&amp;#39; in statmount_string()<br /> <br /> In statmount_string(), most flags assign an output offset pointer (offp)<br /> which is later updated with the string offset. However, the<br /> STATMOUNT_MNT_UIDMAP and STATMOUNT_MNT_GIDMAP cases directly set the<br /> struct fields instead of using offp. This leaves offp uninitialized,<br /> leading to a possible uninitialized dereference when *offp is updated.<br /> <br /> Fix it by assigning offp for UIDMAP and GIDMAP as well, keeping the code<br /> path consistent.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> idpf: fix possible vport_config NULL pointer deref in remove<br /> <br /> Attempting to remove the driver will cause a crash in cases where<br /> the vport failed to initialize. Following trace is from an instance where<br /> the driver failed during an attempt to create a VF:<br /> [ 1661.543624] idpf 0000:84:00.7: Device HW Reset initiated<br /> [ 1722.923726] idpf 0000:84:00.7: Transaction timed-out (op:1 cookie:2900 vc_op:1 salt:29 timeout:60000ms)<br /> [ 1723.353263] BUG: kernel NULL pointer dereference, address: 0000000000000028<br /> ...<br /> [ 1723.358472] RIP: 0010:idpf_remove+0x11c/0x200 [idpf]<br /> ...<br /> [ 1723.364973] Call Trace:<br /> [ 1723.365475] <br /> [ 1723.365972] pci_device_remove+0x42/0xb0<br /> [ 1723.366481] device_release_driver_internal+0x1a9/0x210<br /> [ 1723.366987] pci_stop_bus_device+0x6d/0x90<br /> [ 1723.367488] pci_stop_and_remove_bus_device+0x12/0x20<br /> [ 1723.367971] pci_iov_remove_virtfn+0xbd/0x120<br /> [ 1723.368309] sriov_disable+0x34/0xe0<br /> [ 1723.368643] idpf_sriov_configure+0x58/0x140 [idpf]<br /> [ 1723.368982] sriov_numvfs_store+0xda/0x1c0<br /> <br /> Avoid the NULL pointer dereference by adding NULL pointer check for<br /> vport_config[i], before freeing user_config.q_coalesce.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> timers: Fix NULL function pointer race in timer_shutdown_sync()<br /> <br /> There is a race condition between timer_shutdown_sync() and timer<br /> expiration that can lead to hitting a WARN_ON in expire_timers().<br /> <br /> The issue occurs when timer_shutdown_sync() clears the timer function<br /> to NULL while the timer is still running on another CPU. The race<br /> scenario looks like this:<br /> <br /> CPU0 CPU1<br /> <br /> lock_timer_base()<br /> expire_timers()<br /> base-&gt;running_timer = timer;<br /> unlock_timer_base()<br /> [call_timer_fn enter]<br /> mod_timer()<br /> ...<br /> timer_shutdown_sync()<br /> lock_timer_base()<br /> // For now, will not detach the timer but only clear its function to NULL<br /> if (base-&gt;running_timer != timer)<br /> ret = detach_if_pending(timer, base, true);<br /> if (shutdown)<br /> timer-&gt;function = NULL;<br /> unlock_timer_base()<br /> [call_timer_fn exit]<br /> lock_timer_base()<br /> base-&gt;running_timer = NULL;<br /> unlock_timer_base()<br /> ...<br /> // Now timer is pending while its function set to NULL.<br /> // next timer trigger<br /> <br /> expire_timers()<br /> WARN_ON_ONCE(!fn) // hit<br /> ...<br /> lock_timer_base()<br /> // Now timer will detach<br /> if (base-&gt;running_timer != timer)<br /> ret = detach_if_pending(timer, base, true);<br /> if (shutdown)<br /> timer-&gt;function = NULL;<br /> unlock_timer_base()<br /> <br /> The problem is that timer_shutdown_sync() clears the timer function<br /> regardless of whether the timer is currently running. This can leave a<br /> pending timer with a NULL function pointer, which triggers the<br /> WARN_ON_ONCE(!fn) check in expire_timers().<br /> <br /> Fix this by only clearing the timer function when actually detaching the<br /> timer. If the timer is running, leave the function pointer intact, which is<br /> safe because the timer will be properly detached when it finishes running.