kernel/include, branch linux-6.14.y

NFS: Avoid flushing data while holding directory locks in nfs_rename()

2025-06-04T12:45:10Z

[ Upstream commit dcd21b609d4abc7303f8683bce4f35d78d7d6830 ] The Linux client assumes that all filehandles are non-volatile for renames within the same directory (otherwise sillyrename cannot work). However, the existence of the Linux 'subtree_check' export option has meant that nfs_rename() has always assumed it needs to flush writes before attempting to rename. Since NFSv4 does allow the client to query whether or not the server exhibits this behaviour, and since knfsd does actually set the appropriate flag when 'subtree_check' is enabled on an export, it should be OK to optimise away the write flushing behaviour in the cases where it is clearly not needed. Signed-off-by: Trond Myklebust Reviewed-by: Jeff Layton Signed-off-by: Sasha Levin

coredump: hand a pidfd to the usermode coredump helper

2025-06-04T12:45:09Z

commit b5325b2a270fcaf7b2a9a0f23d422ca8a5a8bdea upstream. Give userspace a way to instruct the kernel to install a pidfd into the usermode helper process. This makes coredump handling a lot more reliable for userspace. In parallel with this commit we already have systemd adding support for this in [1]. We create a pidfs file for the coredumping process when we process the corename pattern. When the usermode helper process is forked we then install the pidfs file as file descriptor three into the usermode helpers file descriptor table so it's available to the exec'd program. Since usermode helpers are either children of the system_unbound_wq workqueue or kthreadd we know that the file descriptor table is empty and can thus always use three as the file descriptor number. Note, that we'll install a pidfd for the thread-group leader even if a subthread is calling do_coredump(). We know that task linkage hasn't been removed due to delay_group_leader() and even if this @current isn't the actual thread-group leader we know that the thread-group leader cannot be reaped until @current has exited. [brauner: This is a backport for the v6.12 series. The upstream kernel has changed pidfs_alloc_file() to set O_RDWR implicitly instead of forcing callers to set it. Let's minimize the churn and just let the coredump umh handler raise O_RDWR.] Link: https://github.com/systemd/systemd/pull/37125 [1] Link: https://lore.kernel.org/20250414-work-coredump-v2-3-685bf231f828@kernel.org Tested-by: Luca Boccassi Reviewed-by: Oleg Nesterov Signed-off-by: Christian Brauner Signed-off-by: Greg Kroah-Hartman

iommu: Handle yet another race around registration

2025-06-04T12:45:09Z

commit da33e87bd2bfc63531cf7448a3cd7a3d42182f08 upstream. Next up on our list of race windows to close is another one during iommu_device_register() - it's now OK again for multiple instances to run their bus_iommu_probe() in parallel, but an iommu_probe_device() can still also race against a running bus_iommu_probe(). As Johan has managed to prove, this has now become a lot more visible on DT platforms wth driver_async_probe where a client driver is attempting to probe in parallel with its IOMMU driver - although commit b46064a18810 ("iommu: Handle race with default domain setup") resolves this from the client driver's point of view, this isn't before of_iommu_configure() has had the chance to attempt to "replay" a probe that the bus walk hasn't even tried yet, and so still cause the out-of-order group allocation behaviour that we're trying to clean up (and now warning about). The most reliable thing to do here is to explicitly keep track of the "iommu_device_register() is still running" state, so we can then special-case the ops lookup for the replay path (based on dev->iommu again) to let that think it's still waiting for the IOMMU driver to appear at all. This still leaves the longstanding theoretical case of iommu_bus_notifier() being triggered during bus_iommu_probe(), but it's not so simple to defer a notifier, and nobody's ever reported that being a visible issue, so let's quietly kick that can down the road for now... Reported-by: Johan Hovold Fixes: bcb81ac6ae3c ("iommu: Get DT/ACPI parsing into the proper probe path") Signed-off-by: Robin Murphy Link: https://lore.kernel.org/r/88d54c1b48fed8279aa47d30f3d75173685bb26a.1745516488.git.robin.murphy@arm.com Signed-off-by: Joerg Roedel Signed-off-by: Greg Kroah-Hartman

err.h: move IOMEM_ERR_PTR() to err.h

2025-05-29T09:14:08Z

commit 18311a766c587fc69b1806f1d5943305903b7e6e upstream. Since IOMEM_ERR_PTR() macro deals with an error pointer, a better place for it is err.h. This helps avoid dependency on io.h for the users that don't need it. Suggested-by: Andy Shevchenko Signed-off-by: Raag Jadav Acked-by: Arnd Bergmann Signed-off-by: Andy Shevchenko Signed-off-by: Greg Kroah-Hartman

spi: use container_of_cont() for to_spi_device()

2025-05-29T09:14:07Z

[ Upstream commit 1007ae0d464ceb55a3740634790521d3543aaab9 ] Some places in the spi core pass in a const pointer to a device and the default container_of() casts that away, which is not a good idea. Preserve the proper const attribute by using container_of_const() for to_spi_device() instead, which is what it was designed for. Note, this removes the NULL check for a device pointer in the call, but no one was ever checking for that return value, and a device pointer should never be NULL overall anyway, so this should be a safe change. Cc: Mark Brown Fixes: d69d80484598 ("driver core: have match() callback in struct bus_type take a const *") Signed-off-by: Greg Kroah-Hartman Link: https://patch.msgid.link/2025052230-fidgeting-stooge-66f5@gregkh Signed-off-by: Mark Brown Signed-off-by: Sasha Levin

mm: fix VM_UFFD_MINOR == VM_SHADOW_STACK on USERFAULTFD=y && ARM64_GCS=y

2025-05-29T09:14:06Z

commit 0f518255bde881d2a2605bbc080b438b532b6ab2 upstream. On configs with CONFIG_ARM64_GCS=y, VM_SHADOW_STACK is bit 38. On configs with CONFIG_HAVE_ARCH_USERFAULTFD_MINOR=y (selected by CONFIG_ARM64 when CONFIG_USERFAULTFD=y), VM_UFFD_MINOR is _also_ bit 38. This bit being shared by two different VMA flags could lead to all sorts of unintended behaviors. Presumably, a process could maybe call into userfaultfd in a way that disables the shadow stack vma flag. I can't think of any attack where this would help (presumably, if an attacker tries to disable shadow stacks, they are trying to hijack control flow so can't arbitrarily call into userfaultfd yet anyway) but this still feels somewhat scary. Link: https://lkml.kernel.org/r/20250507131000.1204175-2-revest@chromium.org Fixes: ae80e1629aea ("mm: Define VM_SHADOW_STACK for arm64 when we support GCS") Signed-off-by: Florent Revest Reviewed-by: Mark Brown Cc: Borislav Betkov Cc: Brendan Jackman Cc: Catalin Marinas Cc: Florent Revest Cc: "H. Peter Anvin" Cc: Ingo Molnar Cc: Thiago Jung Bauermann Cc: Thomas Gleinxer Cc: Will Deacon Cc: Signed-off-by: Andrew Morton Signed-off-by: Greg Kroah-Hartman

mm: mmap: map MAP_STACK to VM_NOHUGEPAGE only if THP is enabled

2025-05-29T09:14:06Z

commit 7190b3c8bd2b0cde483bd440cf91ba1c518b4261 upstream. commit c4608d1bf7c6 ("mm: mmap: map MAP_STACK to VM_NOHUGEPAGE") maps the mmap option MAP_STACK to VM_NOHUGEPAGE. This is also done if CONFIG_TRANSPARENT_HUGEPAGE is not defined. But in that case, the VM_NOHUGEPAGE does not make sense. I discovered this issue when trying to use the tool CRIU to checkpoint and restore a container. Our running kernel is compiled without CONFIG_TRANSPARENT_HUGEPAGE. CRIU parses the output of /proc//smaps and saves the "nh" flag. When trying to restore the container, CRIU fails to restore the "nh" mappings, since madvise() MADV_NOHUGEPAGE always returns an error because CONFIG_TRANSPARENT_HUGEPAGE is not defined. Link: https://lkml.kernel.org/r/20250507-map-map_stack-to-vm_nohugepage-only-if-thp-is-enabled-v5-1-c6c38cfefd6e@kuka.com Fixes: c4608d1bf7c6 ("mm: mmap: map MAP_STACK to VM_NOHUGEPAGE") Signed-off-by: Ignacio Moreno Gonzalez Acked-by: David Hildenbrand Reviewed-by: Lorenzo Stoakes Reviewed-by: Yang Shi Reviewed-by: Liam R. Howlett Cc: Signed-off-by: Andrew Morton Signed-off-by: Greg Kroah-Hartman

taskstats: fix struct taskstats breaks backward compatibility since version 15

2025-05-29T09:14:05Z

commit 0bf2d838de1ffb6d0bb6f8d18a6ccc59b7d9a705 upstream. Problem ======== commit 658eb5ab916d ("delayacct: add delay max to record delay peak") - adding more fields commit f65c64f311ee ("delayacct: add delay min to record delay peak") - adding more fields commit b016d0873777 ("taskstats: modify taskstats version") - version bump to 15 Since version 15 (TASKSTATS_VERSION=15) the new layout of the structure adds fields in the middle of the structure, rendering all old software incompatible with newer kernels and software compiled against the new kernel headers incompatible with older kernels. Solution ========= move delay max and delay min to the end of taskstat, and bump the version to 16 after the change [wang.yaxin@zte.com.cn: adjust indentation] Link: https://lkml.kernel.org/r/202505192131489882NSciXV4EGd8zzjLuwoOK@zte.com.cn Link: https://lkml.kernel.org/r/20250510155413259V4JNRXxukdDgzsaL0Fo6a@zte.com.cn Fixes: f65c64f311ee ("delayacct: add delay min to record delay peak") Signed-off-by: Wang Yaxin Signed-off-by: xu xin Signed-off-by: Kun Jiang Reviewed-by: Yang Yang Cc: Balbir Singh Cc: Signed-off-by: Andrew Morton Signed-off-by: Greg Kroah-Hartman

highmem: add folio_test_partial_kmap()

2025-05-29T09:14:04Z

commit 97dfbbd135cb5e4426f37ca53a8fa87eaaa4e376 upstream. In commit c749d9b7ebbc ("iov_iter: fix copy_page_from_iter_atomic() if KMAP_LOCAL_FORCE_MAP"), Hugh correctly noted that if KMAP_LOCAL_FORCE_MAP is enabled, we must limit ourselves to PAGE_SIZE bytes per call to kmap_local(). The same problem exists in memcpy_from_folio(), memcpy_to_folio(), folio_zero_tail(), folio_fill_tail() and memcpy_from_file_folio(), so add folio_test_partial_kmap() to do this more succinctly. Link: https://lkml.kernel.org/r/20250514170607.3000994-2-willy@infradead.org Fixes: 00cdf76012ab ("mm: add memcpy_from_file_folio()") Signed-off-by: Matthew Wilcox (Oracle) Cc: Al Viro Cc: Hugh Dickins Cc: Signed-off-by: Andrew Morton Signed-off-by: Greg Kroah-Hartman

alloc_tag: allocate percpu counters for module tags dynamically

2025-05-29T09:14:04Z

commit 12ca42c237756182aad8ab04654c952765cb9061 upstream. When a module gets unloaded it checks whether any of its tags are still in use and if so, we keep the memory containing module's allocation tags alive until all tags are unused. However percpu counters referenced by the tags are freed by free_module(). This will lead to UAF if the memory allocated by a module is accessed after module was unloaded. To fix this we allocate percpu counters for module allocation tags dynamically and we keep it alive for tags which are still in use after module unloading. This also removes the requirement of a larger PERCPU_MODULE_RESERVE when memory allocation profiling is enabled because percpu memory for counters does not need to be reserved anymore. Link: https://lkml.kernel.org/r/20250517000739.5930-1-surenb@google.com Fixes: 0db6f8d7820a ("alloc_tag: load module tags into separate contiguous memory") Signed-off-by: Suren Baghdasaryan Reported-by: David Wang <00107082@163.com> Closes: https://lore.kernel.org/all/20250516131246.6244-1-00107082@163.com/ Tested-by: David Wang <00107082@163.com> Cc: Christoph Lameter (Ampere) Cc: Dennis Zhou Cc: Kent Overstreet Cc: Pasha Tatashin Cc: Tejun Heo Cc: Signed-off-by: Andrew Morton Signed-off-by: Greg Kroah-Hartman