CVE Database

In the Linux kernel, the following vulnerability has been resolved: net: qrtr: ns: Free the node during ctrl_cmd_bye() A node sends the BYE packet when it is about to go down. So the nameserver should advertise the removal of the node to all remote and local observers and free the node finally. But currently, the nameserver doesn't free the node memory even after processing the BYE packet. This causes the node memory to leak. Hence, remove the node from Xarray list and free the node memory during both success and failure case of ctrl_cmd_bye().

In the Linux kernel, the following vulnerability has been resolved: mm/page_alloc: return NULL early from alloc_frozen_pages_nolock() in NMI on UP On UP kernels (!CONFIG_SMP), spin_trylock() is a no-op that unconditionally succeeds even when the lock is already held. As a result, alloc_frozen_pages_nolock() called from NMI context can re-enter rmqueue() and acquire the zone lock that the interrupted context is already holding, corrupting the freelists. With CONFIG_DEBUG_SPINLOCK on UP, the following BUG is triggered with the slub_kunit test module: BUG: spinlock trylock failure on UP on CPU#0, kunit_try_catch/243 [...] Call Trace: <NMI> dump_stack_lvl+0x3f/0x60 do_raw_spin_trylock+0x41/0x50 _raw_spin_trylock+0x24/0x50 rmqueue.isra.0+0x2a9/0xa70 get_page_from_freelist+0xeb/0x450 alloc_frozen_pages_nolock_noprof+0x111/0x1e0 allocate_slab+0x42a/0x500 ___slab_alloc+0xa7/0x4c0 kmalloc_nolock_noprof+0x164/0x310 [...] </NMI> Fix this by returning NULL early when invoked from NMI on a UP kernel.

In the Linux kernel, the following vulnerability has been resolved: vfio/cdx: Fix NULL pointer dereference in interrupt trigger path Add validation to ensure MSI is configured before accessing cdx_irqs array in vfio_cdx_set_msi_trigger(). Without this check, userspace can trigger a NULL pointer dereference by calling VFIO_DEVICE_SET_IRQS with VFIO_IRQ_SET_DATA_BOOL or VFIO_IRQ_SET_DATA_NONE flags before ever setting up interrupts via VFIO_IRQ_SET_DATA_EVENTFD. The vfio_cdx_msi_enable() function allocates the cdx_irqs array and sets config_msi to 1 only when called through the EVENTFD path. The trigger loop (for DATA_BOOL/DATA_NONE) assumed this had already been done, but there was no enforcement of this call ordering. This matches the protection used in the PCI VFIO driver where vfio_pci_set_msi_trigger() checks irq_is() before the trigger loop.

In the Linux kernel, the following vulnerability has been resolved: KVM: nSVM: Triple fault if restore host CR3 fails on nested #VMEXIT If loading L1's CR3 fails on a nested #VMEXIT, nested_svm_vmexit() returns an error code that is ignored by most callers, and continues to run L1 with corrupted state. A sane recovery is not possible in this case, and HW behavior is to cause a shutdown. Inject a triple fault instead, and do not return early from nested_svm_vmexit(). Continue cleaning up the vCPU state (e.g. clear pending exceptions), to handle the failure as gracefully as possible. From the APM: Upon #VMEXIT, the processor performs the following actions in order to return to the host execution context: ... if (illegal host state loaded, or exception while loading host state) shutdown else execute first host instruction following the VMRUN Remove the return value of nested_svm_vmexit(), which is mostly unchecked anyway.

In the Linux kernel, the following vulnerability has been resolved: EDAC/versalnet: Fix device_node leak in mc_probe() of_parse_phandle() returns a device_node reference that must be released with of_node_put(). The original code never freed r5_core_node on any exit path, causing a memory leak. Fix this by using the automatic cleanup attribute __free(device_node) which ensures of_node_put() is called when the variable goes out of scope.

In the Linux kernel, the following vulnerability has been resolved: crypto: algif_aead - snapshot IV for async AEAD requests AF_ALG AEAD AIO requests currently use the socket-wide IV buffer during request processing. For async requests, later socket activity can update that shared state before the original request has fully completed, which can lead to inconsistent IV handling. Snapshot the IV into per-request storage when preparing the AEAD request, so in-flight operations no longer depend on mutable socket state.

In the Linux kernel, the following vulnerability has been resolved: net: qrtr: ns: Limit the maximum number of lookups Current code does no bound checking on the number of lookups a client can perform. Though the code restricts the lookups to local clients, there is still a possibility of a malicious local client sending a flood of NEW_LOOKUP messages over the same socket. Fix this issue by limiting the maximum number of lookups to 64 globally. Since the nameserver allows only atmost one local observer, this global lookup count will ensure that the lookups stay within the limit. Note that, limit of 64 is chosen based on the current platform requirements. If requirement changes in the future, this limit can be increased.

In the Linux kernel, the following vulnerability has been resolved: mm/damon/core: fix damon_call() vs kdamond_fn() exit race Patch series "mm/damon/core: fix damon_call()/damos_walk() vs kdmond exit race". damon_call() and damos_walk() can leak memory and/or deadlock when they race with kdamond terminations. Fix those. This patch (of 2); When kdamond_fn() main loop is finished, the function cancels all remaining damon_call() requests and unset the damon_ctx->kdamond so that API callers and API functions themselves can know the context is terminated. damon_call() adds the caller's request to the queue first. After that, it shows if the kdamond of the damon_ctx is still running (damon_ctx->kdamond is set). Only if the kdamond is running, damon_call() starts waiting for the kdamond's handling of the newly added request. The damon_call() requests registration and damon_ctx->kdamond unset are protected by different mutexes, though. Hence, damon_call() could race with damon_ctx->kdamond unset, and result in deadlocks. For example, let's suppose kdamond successfully finished the damon_call() requests cancelling. Right after that, damon_call() is called for the context. It registers the new request, and shows the context is still running, because damon_ctx->kdamond unset is not yet done. Hence the damon_call() caller starts waiting for the handling of the request. However, the kdamond is already on the termination steps, so it never handles the new request. As a result, the damon_call() caller threads infinitely waits. Fix this by introducing another damon_ctx field, namely call_controls_obsolete. It is protected by the damon_ctx->call_controls_lock, which protects damon_call() requests registration. Initialize (unset) it in kdamond_fn() before letting damon_start() returns and set it just before the cancelling of remaining damon_call() requests is executed. damon_call() reads the obsolete field under the lock and avoids adding a new request. After this change, only requests that are guaranteed to be handled or cancelled are registered. Hence the after-registration DAMON context termination check is no longer needed. Remove it together. Note that the deadlock will not happen when damon_call() is called for repeat mode request. In tis case, damon_call() returns instead of waiting for the handling when the request registration succeeds and it shows the kdamond is running. However, if the request also has dealloc_on_cancel, the request memory would be leaked. The issue is found by sashiko [1].

In the Linux kernel, the following vulnerability has been resolved: dm mirror: fix integer overflow in create_dirty_log() The argument count calculation in create_dirty_log() performs `*args_used = 2 + param_count` before validating against argc. When a user provides a param_count close to UINT_MAX via the device mapper table string, this unsigned addition wraps around to a small value, causing the subsequent `argc < *args_used` check to be bypassed. The overflowed param_count is then passed as argc to dm_dirty_log_create(), where it can cause out-of-bounds reads on the argv array. Fix by comparing param_count against argc - 2 before performing the addition, following the same pattern used by parse_features() in the same file. Since argc >= 2 is already guaranteed, the subtraction is safe.

In the Linux kernel, the following vulnerability has been resolved: thermal: core: Fix thermal zone governor cleanup issues If thermal_zone_device_register_with_trips() fails after adding a thermal governor to the thermal zone being registered, the governor is not removed from it as appropriate which may lead to a memory leak. In turn, thermal_zone_device_unregister() calls thermal_set_governor() without acquiring the thermal zone lock beforehand which may race with a governor update via sysfs and may lead to a use-after-free in that case. Address these issues by adding two thermal_set_governor() calls, one to thermal_release() to remove the governor from the given thermal zone, and one to the thermal zone registration error path to cover failures preceding the thermal zone device registration.

In the Linux kernel, the following vulnerability has been resolved: crypto: atmel-aes - Fix 3-page memory leak in atmel_aes_buff_cleanup atmel_aes_buff_init() allocates 4 pages using __get_free_pages() with ATMEL_AES_BUFFER_ORDER, but atmel_aes_buff_cleanup() frees only the first page using free_page(), leaking the remaining 3 pages. Use free_pages() with ATMEL_AES_BUFFER_ORDER to fix the memory leak.

In the Linux kernel, the following vulnerability has been resolved: ALSA: usb-audio: stop parsing UAC2 rates at MAX_NR_RATES parse_uac2_sample_rate_range() caps the number of enumerated rates at MAX_NR_RATES, but it only breaks out of the current rate loop. A malformed UAC2 RANGE response with additional triplets continues parsing the remaining triplets and repeatedly prints "invalid uac2 rates" while probe still holds register_mutex. Stop the whole parse once the cap is reached and return the number of rates collected so far.

In the Linux kernel, the following vulnerability has been resolved: mm: fix deferred split queue races during migration migrate_folio_move() records the deferred split queue state from src and replays it on dst. Replaying it after remove_migration_ptes(src, dst, 0) makes dst visible before it is requeued, so a concurrent rmap-removal path can mark dst partially mapped and trip the WARN in deferred_split_folio(). Move the requeue before remove_migration_ptes() so dst is back on the deferred split queue before it becomes visible again. Because migration still holds dst locked at that point, teach deferred_split_scan() to requeue a folio when folio_trylock() fails. Otherwise a fully mapped underused folio can be dequeued by the shrinker and silently lost from split_queue. [ziy@nvidia.com: move the comment]

In the Linux kernel, the following vulnerability has been resolved: remoteproc: xlnx: Only access buffer information if IPI is buffered In the receive callback check if message is NULL to prevent possibility of crash by NULL pointer dereferencing.

In the Linux kernel, the following vulnerability has been resolved: KVM: SVM: Add missing save/restore handling of LBR MSRs MSR_IA32_DEBUGCTLMSR and LBR MSRs are currently not enumerated by KVM_GET_MSR_INDEX_LIST, and LBR MSRs cannot be set with KVM_SET_MSRS. So save/restore is completely broken. Fix it by adding the MSRs to msrs_to_save_base, and allowing writes to LBR MSRs from userspace only (as they are read-only MSRs) if LBR virtualization is enabled. Additionally, to correctly restore L1's LBRs while L2 is running, make sure the LBRs are copied from the captured VMCB01 save area in svm_copy_vmrun_state(). Note, for VMX, this also fixes a flaw where MSR_IA32_DEBUGCTLMSR isn't reported as an MSR to save/restore. Note #2, over-reporting MSR_IA32_LASTxxx on Intel is ok, as KVM already handles unsupported reads and writes thanks to commit b5e2fec0ebc3 ("KVM: Ignore DEBUGCTL MSRs with no effect") (kvm_do_msr_access() will morph the unsupported userspace write into a nop). [sean: guard with lbrv checks, massage changelog]

In the Linux kernel, the following vulnerability has been resolved: mm/memfd_luo: fix physical address conversion in put_folios cleanup In memfd_luo_retrieve_folios()'s put_folios cleanup path: 1. kho_restore_folio() expects a phys_addr_t (physical address) but receives a raw PFN (pfolio->pfn). This causes kho_restore_page() to check the wrong physical address (pfn << PAGE_SHIFT instead of the actual physical address). 2. This loop lacks the !pfolio->pfn check that exists in the main retrieval loop and memfd_luo_discard_folios(), which could incorrectly process sparse file holes where pfn=0. Fix by converting PFN to physical address with PFN_PHYS() and adding the !pfolio->pfn check, matching the pattern used elsewhere in this file. This issue was identified by the AI review. https://sashiko.dev/#/patchset/20260323110747.193569-1-duanchenghao@kylinos.cn

In the Linux kernel, the following vulnerability has been resolved: rxrpc: Fix memory leaks in rxkad_verify_response() Fix rxkad_verify_response() to free the ticket and the server key under all circumstances by initialising the ticket pointer to NULL and then making all paths through the function after the first allocation has been done go through a single common epilogue that just releases everything - where all the releases skip on a NULL pointer.

In the Linux kernel, the following vulnerability has been resolved: PCI: endpoint: pci-epf-ntb: Remove duplicate resource teardown epf_ntb_epc_destroy() duplicates the teardown that the caller is supposed to do later. This leads to an oops when .allow_link fails or when .drop_link is performed. Remove the helper. Also drop pci_epc_put(). EPC device refcounting is tied to configfs EPC group lifetime, and pci_epc_put() in the .drop_link path is sufficient.

In the Linux kernel, the following vulnerability has been resolved: mm/damon/core: fix damos_walk() vs kdamond_fn() exit race When kdamond_fn() main loop is finished, the function cancels remaining damos_walk() request and unset the damon_ctx->kdamond so that API callers and API functions themselves can show the context is terminated. damos_walk() adds the caller's request to the queue first. After that, it shows if the kdamond of the damon_ctx is still running (damon_ctx->kdamond is set). Only if the kdamond is running, damos_walk() starts waiting for the kdamond's handling of the newly added request. The damos_walk() requests registration and damon_ctx->kdamond unset are protected by different mutexes, though. Hence, damos_walk() could race with damon_ctx->kdamond unset, and result in deadlocks. For example, let's suppose kdamond successfully finished the damow_walk() request cancelling. Right after that, damos_walk() is called for the context. It registers the new request, and shows the context is still running, because damon_ctx->kdamond unset is not yet done. Hence the damos_walk() caller starts waiting for the handling of the request. However, the kdamond is already on the termination steps, so it never handles the new request. As a result, the damos_walk() caller thread infinitely waits. Fix this by introducing another damon_ctx field, namely walk_control_obsolete. It is protected by the damon_ctx->walk_control_lock, which protects damos_walk() request registration. Initialize (unset) it in kdamond_fn() before letting damon_start() returns and set it just before the cancelling of the remaining damos_walk() request is executed. damos_walk() reads the obsolete field under the lock and avoids adding a new request. After this change, only requests that are guaranteed to be handled or cancelled are registered. Hence the after-registration DAMON context termination check is no longer needed. Remove it together. The issue is found by sashiko [1].

In the Linux kernel, the following vulnerability has been resolved: hwmon: (powerz) Avoid cacheline sharing for DMA buffer Depending on the architecture the transfer buffer may share a cacheline with the following mutex. As the buffer may be used for DMA, that is problematic. Use the high-level DMA helpers to make sure that cacheline sharing can not happen. Also drop the comment, as the helpers are documentation enough. https://sashiko.dev/#/message/20260408175814.934BFC19421%40smtp.kernel.org

In the Linux kernel, the following vulnerability has been resolved: xfs: fix a resource leak in xfs_alloc_buftarg() In the error path, call fs_put_dax() to drop the DAX device reference.

In the Linux kernel, the following vulnerability has been resolved: net: qrtr: ns: Limit the total number of nodes Currently, the nameserver doesn't limit the number of nodes it handles. This can be an attack vector if a malicious client starts registering random nodes, leading to memory exhaustion. Hence, limit the maximum number of nodes to 64. Note that, limit of 64 is chosen based on the current platform requirements. If requirement changes in the future, this limit can be increased.

In the Linux kernel, the following vulnerability has been resolved: ext2: reject inodes with zero i_nlink and valid mode in ext2_iget() ext2_iget() already rejects inodes with i_nlink == 0 when i_mode is zero or i_dtime is set, treating them as deleted. However, the case of i_nlink == 0 with a non-zero mode and zero dtime slips through. Since ext2 has no orphan list, such a combination can only result from filesystem corruption - a legitimate inode deletion always sets either i_dtime or clears i_mode before freeing the inode. A crafted image can exploit this gap to present such an inode to the VFS, which then triggers WARN_ON inside drop_nlink() (fs/inode.c) via ext2_unlink(), ext2_rename() and ext2_rmdir(): WARNING: CPU: 3 PID: 609 at fs/inode.c:336 drop_nlink+0xad/0xd0 fs/inode.c:336 CPU: 3 UID: 0 PID: 609 Comm: syz-executor Not tainted 6.12.77+ #1 Call Trace: <TASK> inode_dec_link_count include/linux/fs.h:2518 [inline] ext2_unlink+0x26c/0x300 fs/ext2/namei.c:295 vfs_unlink+0x2fc/0x9b0 fs/namei.c:4477 do_unlinkat+0x53e/0x730 fs/namei.c:4541 __x64_sys_unlink+0xc6/0x110 fs/namei.c:4587 do_syscall_64+0xf5/0x220 arch/x86/entry/common.c:78 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> WARNING: CPU: 0 PID: 646 at fs/inode.c:336 drop_nlink+0xad/0xd0 fs/inode.c:336 CPU: 0 UID: 0 PID: 646 Comm: syz.0.17 Not tainted 6.12.77+ #1 Call Trace: <TASK> inode_dec_link_count include/linux/fs.h:2518 [inline] ext2_rename+0x35e/0x850 fs/ext2/namei.c:374 vfs_rename+0xf2f/0x2060 fs/namei.c:5021 do_renameat2+0xbe2/0xd50 fs/namei.c:5178 __x64_sys_rename+0x7e/0xa0 fs/namei.c:5223 do_syscall_64+0xf5/0x220 arch/x86/entry/common.c:78 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> WARNING: CPU: 0 PID: 634 at fs/inode.c:336 drop_nlink+0xad/0xd0 fs/inode.c:336 CPU: 0 UID: 0 PID: 634 Comm: syz-executor Not tainted 6.12.77+ #1 Call Trace: <TASK> inode_dec_link_count include/linux/fs.h:2518 [inline] ext2_rmdir+0xca/0x110 fs/ext2/namei.c:311 vfs_rmdir+0x204/0x690 fs/namei.c:4348 do_rmdir+0x372/0x3e0 fs/namei.c:4407 __x64_sys_unlinkat+0xf0/0x130 fs/namei.c:4577 do_syscall_64+0xf5/0x220 arch/x86/entry/common.c:78 entry_SYSCALL_64_after_hwframe+0x77/0x7f </TASK> Extend the existing i_nlink == 0 check to also catch this case, reporting the corruption via ext2_error() and returning -EFSCORRUPTED. This rejects the inode at load time and prevents it from reaching any of the namei.c paths. Found by Linux Verification Center (linuxtesting.org) with Syzkaller.

In the Linux kernel, the following vulnerability has been resolved: rxrpc: Fix conn-level packet handling to unshare RESPONSE packets The security operations that verify the RESPONSE packets decrypt bits of it in place - however, the sk_buff may be shared with a packet sniffer, which would lead to the sniffer seeing an apparently corrupt packet (actually decrypted). Fix this by handing a copy of the packet off to the specific security handler if the packet was cloned.

In the Linux kernel, the following vulnerability has been resolved: scsi: sd: fix missing put_disk() when device_add(&disk_dev) fails If device_add(&sdkp->disk_dev) fails, put_device() runs scsi_disk_release(), which frees the scsi_disk but leaves the gendisk referenced. The device_add_disk() error path in sd_probe() calls put_disk(gd); call put_disk(gd) here to mirror that cleanup.

In the Linux kernel, the following vulnerability has been resolved: LoongArch: Add spectre boundry for syscall dispatch table The LoongArch syscall number is directly controlled by userspace, but does not have a array_index_nospec() boundry to prevent access past the syscall function pointer tables.

In the Linux kernel, the following vulnerability has been resolved: slub: fix data loss and overflow in krealloc() Commit 2cd8231796b5 ("mm/slub: allow to set node and align in k[v]realloc") introduced the ability to force a reallocation if the original object does not satisfy new alignment or NUMA node, even when the object is being shrunk. This introduced two bugs in the reallocation fallback path: 1. Data loss during NUMA migration: The jump to 'alloc_new' happens before 'ks' and 'orig_size' are initialized. As a result, the memcpy() in the 'alloc_new' block would copy 0 bytes into the new allocation. 2. Buffer overflow during shrinking: When shrinking an object while forcing a new alignment, 'new_size' is smaller than the old size. However, the memcpy() used the old size ('orig_size ?: ks'), leading to an out-of-bounds write. The same overflow bug exists in the kvrealloc() fallback path, where the old bucket size ksize(p) is copied into the new buffer without being bounded by the new size. A simple reproducer: // e.g. add to lkdtm as KREALLOC_SHRINK_OVERFLOW while (1) { void *p = kmalloc(128, GFP_KERNEL); p = krealloc_node_align(p, 64, 256, GFP_KERNEL, NUMA_NO_NODE); kfree(p); } demonstrates the issue: ================================================================== BUG: KFENCE: out-of-bounds write in memcpy_orig+0x68/0x130 Out-of-bounds write at 0xffff8883ad757038 (120B right of kfence-#47): memcpy_orig+0x68/0x130 krealloc_node_align_noprof+0x1c8/0x340 lkdtm_KREALLOC_SHRINK_OVERFLOW+0x8c/0xc0 [lkdtm] lkdtm_do_action+0x3a/0x60 [lkdtm] ... kfence-#47: 0xffff8883ad756fc0-0xffff8883ad756fff, size=64, cache=kmalloc-64 allocated by task 316 on cpu 7 at 97.680481s (0.021813s ago): krealloc_node_align_noprof+0x19c/0x340 lkdtm_KREALLOC_SHRINK_OVERFLOW+0x8c/0xc0 [lkdtm] lkdtm_do_action+0x3a/0x60 [lkdtm] ... ================================================================== Fix it by moving the old size calculation to the top of __do_krealloc() and bounding all copy lengths by the new allocation size.

In the Linux kernel, the following vulnerability has been resolved: KVM: nSVM: Sync interrupt shadow to cached vmcb12 after VMRUN of L2 After VMRUN in guest mode, nested_sync_control_from_vmcb02() syncs fields written by the CPU from vmcb02 to the cached vmcb12. This is because the cached vmcb12 is used as the authoritative copy of some of the controls, and is the payload when saving/restoring nested state. int_state is also written by the CPU, specifically bit 0 (i.e. SVM_INTERRUPT_SHADOW_MASK) for nested VMs, but it is not sync'd to cached vmcb12. This does not cause a problem if KVM_SET_NESTED_STATE preceeds KVM_SET_VCPU_EVENTS in the restore path, as an interrupt shadow would be correctly restored to vmcb02 (KVM_SET_VCPU_EVENTS overwrites what KVM_SET_NESTED_STATE restored in int_state). However, if KVM_SET_VCPU_EVENTS preceeds KVM_SET_NESTED_STATE, an interrupt shadow would be restored into vmcb01 instead of vmcb02. This would mostly be benign for L1 (delays an interrupt), but not for L2. For L2, the vCPU could hang (e.g. if a wakeup interrupt is delivered before a HLT that should have been in an interrupt shadow). Sync int_state to the cached vmcb12 in nested_sync_control_from_vmcb02() to avoid this problem. With that, KVM_SET_NESTED_STATE restores the correct interrupt shadow state, and if KVM_SET_VCPU_EVENTS follows it would overwrite it with the same value.

In the Linux kernel, the following vulnerability has been resolved: crypto: ccree - fix a memory leak in cc_mac_digest() Add cc_unmap_result() if cc_map_hash_request_final() fails to prevent potential memory leak.

In the Linux kernel, the following vulnerability has been resolved: ext4: don't set EXT4_GET_BLOCKS_CONVERT when splitting before submitting I/O When allocating blocks during within-EOF DIO and writeback with dioread_nolock enabled, EXT4_GET_BLOCKS_PRE_IO was set to split an existing large unwritten extent. However, EXT4_GET_BLOCKS_CONVERT was set when calling ext4_split_convert_extents(), which may potentially result in stale data issues. Assume we have an unwritten extent, and then DIO writes the second half. [UUUUUUUUUUUUUUUU] on-disk extent U: unwritten extent [UUUUUUUUUUUUUUUU] extent status tree |<- ->| ----> dio write this range First, ext4_iomap_alloc() call ext4_map_blocks() with EXT4_GET_BLOCKS_PRE_IO, EXT4_GET_BLOCKS_UNWRIT_EXT and EXT4_GET_BLOCKS_CREATE flags set. ext4_map_blocks() find this extent and call ext4_split_convert_extents() with EXT4_GET_BLOCKS_CONVERT and the above flags set. Then, ext4_split_convert_extents() calls ext4_split_extent() with EXT4_EXT_MAY_ZEROOUT, EXT4_EXT_MARK_UNWRIT2 and EXT4_EXT_DATA_VALID2 flags set, and it calls ext4_split_extent_at() to split the second half with EXT4_EXT_DATA_VALID2, EXT4_EXT_MARK_UNWRIT1, EXT4_EXT_MAY_ZEROOUT and EXT4_EXT_MARK_UNWRIT2 flags set. However, ext4_split_extent_at() failed to insert extent since a temporary lack -ENOSPC. It zeroes out the first half but convert the entire on-disk extent to written since the EXT4_EXT_DATA_VALID2 flag set, but left the second half as unwritten in the extent status tree. [0000000000SSSSSS] data S: stale data, 0: zeroed [WWWWWWWWWWWWWWWW] on-disk extent W: written extent [WWWWWWWWWWUUUUUU] extent status tree Finally, if the DIO failed to write data to the disk, the stale data in the second half will be exposed once the cached extent entry is gone. Fix this issue by not passing EXT4_GET_BLOCKS_CONVERT when splitting an unwritten extent before submitting I/O, and make ext4_split_convert_extents() to zero out the entire extent range to zero for this case, and also mark the extent in the extent status tree for consistency.