Vulnerabilities

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> vmw_balloon: indicate success when effectively deflating during migration<br /> <br /> When migrating a balloon page, we first deflate the old page to then<br /> inflate the new page.<br /> <br /> However, if inflating the new page succeeded, we effectively deflated the<br /> old page, reducing the balloon size.<br /> <br /> In that case, the migration actually worked: similar to migrating+<br /> immediately deflating the new page. The old page will be freed back to<br /> the buddy.<br /> <br /> Right now, the core will leave the page be marked as isolated (as we<br /> returned an error). When later trying to putback that page, we will run<br /> into the WARN_ON_ONCE() in balloon_page_putback().<br /> <br /> That handling was changed in commit 3544c4faccb8 ("mm/balloon_compaction:<br /> stop using __ClearPageMovable()"); before that change, we would have<br /> tolerated that way of handling it.<br /> <br /> To fix it, let&amp;#39;s just return 0 in that case, making the core effectively<br /> just clear the "isolated" flag + freeing it back to the buddy as if the<br /> migration succeeded. Note that the new page will also get freed when the<br /> core puts the last reference.<br /> <br /> Note that this also makes it all be more consistent: we will no longer<br /> unisolate the page in the balloon driver while keeping it marked as being<br /> isolated in migration core.<br /> <br /> This was found by code inspection.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> most: usb: hdm_probe: Fix calling put_device() before device initialization<br /> <br /> The early error path in hdm_probe() can jump to err_free_mdev before<br /> &amp;mdev-&gt;dev has been initialized with device_initialize(). Calling<br /> put_device(&amp;mdev-&gt;dev) there triggers a device core WARN and ends up<br /> invoking kref_put(&amp;kobj-&gt;kref, kobject_release) on an uninitialized<br /> kobject.<br /> <br /> In this path the private struct was only kmalloc&amp;#39;ed and the intended<br /> release is effectively kfree(mdev) anyway, so free it directly instead<br /> of calling put_device() on an uninitialized device.<br /> <br /> This removes the WARNING and fixes the pre-initialization error path.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> hung_task: fix warnings caused by unaligned lock pointers<br /> <br /> The blocker tracking mechanism assumes that lock pointers are at least<br /> 4-byte aligned to use their lower bits for type encoding.<br /> <br /> However, as reported by Eero Tamminen, some architectures like m68k<br /> only guarantee 2-byte alignment of 32-bit values. This breaks the<br /> assumption and causes two related WARN_ON_ONCE checks to trigger.<br /> <br /> To fix this, the runtime checks are adjusted to silently ignore any lock<br /> that is not 4-byte aligned, effectively disabling the feature in such<br /> cases and avoiding the related warnings.<br /> <br /> Thanks to Geert Uytterhoeven for bisecting!

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> erofs: avoid infinite loops due to corrupted subpage compact indexes<br /> <br /> Robert reported an infinite loop observed by two crafted images.<br /> <br /> The root cause is that `clusterofs` can be larger than `lclustersize`<br /> for !NONHEAD `lclusters` in corrupted subpage compact indexes, e.g.:<br /> <br /> blocksize = lclustersize = 512 lcn = 6 clusterofs = 515<br /> <br /> Move the corresponding check for full compress indexes to<br /> `z_erofs_load_lcluster_from_disk()` to also cover subpage compact<br /> compress indexes.<br /> <br /> It also fixes the position of `m-&gt;type &gt;= Z_EROFS_LCLUSTER_TYPE_MAX`<br /> check, since it should be placed right after<br /> `z_erofs_load_{compact,full}_lcluster()`.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> misc: fastrpc: Fix dma_buf object leak in fastrpc_map_lookup<br /> <br /> In fastrpc_map_lookup, dma_buf_get is called to obtain a reference to<br /> the dma_buf for comparison purposes. However, this reference is never<br /> released when the function returns, leading to a dma_buf memory leak.<br /> <br /> Fix this by adding dma_buf_put before returning from the function,<br /> ensuring that the temporarily acquired reference is properly released<br /> regardless of whether a matching map is found.<br /> <br /> Rule: add

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> mm: don&amp;#39;t spin in add_stack_record when gfp flags don&amp;#39;t allow<br /> <br /> syzbot was able to find the following path:<br /> add_stack_record_to_list mm/page_owner.c:182 [inline]<br /> inc_stack_record_count mm/page_owner.c:214 [inline]<br /> __set_page_owner+0x2c3/0x4a0 mm/page_owner.c:333<br /> set_page_owner include/linux/page_owner.h:32 [inline]<br /> post_alloc_hook+0x240/0x2a0 mm/page_alloc.c:1851<br /> prep_new_page mm/page_alloc.c:1859 [inline]<br /> get_page_from_freelist+0x21e4/0x22c0 mm/page_alloc.c:3858<br /> alloc_pages_nolock_noprof+0x94/0x120 mm/page_alloc.c:7554<br /> <br /> Don&amp;#39;t spin in add_stack_record_to_list() when it is called<br /> from *_nolock() context.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> staging: rtl8723bs: fix out-of-bounds read in OnBeacon ESR IE parsing<br /> <br /> The Extended Supported Rates (ESR) IE handling in OnBeacon accessed<br /> *(p + 1 + ielen) and *(p + 2 + ielen) without verifying that these<br /> offsets lie within the received frame buffer. A malformed beacon with<br /> an ESR IE positioned at the end of the buffer could cause an<br /> out-of-bounds read, potentially triggering a kernel panic.<br /> <br /> Add a boundary check to ensure that the ESR IE body and the subsequent<br /> bytes are within the limits of the frame before attempting to access<br /> them.<br /> <br /> This prevents OOB reads caused by malformed beacon frames.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> staging: rtl8723bs: fix stack buffer overflow in OnAssocReq IE parsing<br /> <br /> The Supported Rates IE length from an incoming Association Request frame<br /> was used directly as the memcpy() length when copying into a fixed-size<br /> 16-byte stack buffer (supportRate). A malicious station can advertise an<br /> IE length larger than 16 bytes, causing a stack buffer overflow.<br /> <br /> Clamp ie_len to the buffer size before copying the Supported Rates IE,<br /> and correct the bounds check when merging Extended Supported Rates to<br /> prevent a second potential overflow.<br /> <br /> This prevents kernel stack corruption triggered by malformed association<br /> requests.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> binfmt_misc: restore write access before closing files opened by open_exec()<br /> <br /> bm_register_write() opens an executable file using open_exec(), which<br /> internally calls do_open_execat() and denies write access on the file to<br /> avoid modification while it is being executed.<br /> <br /> However, when an error occurs, bm_register_write() closes the file using<br /> filp_close() directly. This does not restore the write permission, which<br /> may cause subsequent write operations on the same file to fail.<br /> <br /> Fix this by calling exe_file_allow_write_access() before filp_close() to<br /> restore the write permission properly.

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> nilfs2: avoid having an active sc_timer before freeing sci<br /> <br /> Because kthread_stop did not stop sc_task properly and returned -EINTR,<br /> the sc_timer was not properly closed, ultimately causing the problem [1]<br /> reported by syzbot when freeing sci due to the sc_timer not being closed.<br /> <br /> Because the thread sc_task main function nilfs_segctor_thread() returns 0<br /> when it succeeds, when the return value of kthread_stop() is not 0 in<br /> nilfs_segctor_destroy(), we believe that it has not properly closed<br /> sc_timer.<br /> <br /> We use timer_shutdown_sync() to sync wait for sc_timer to shutdown, and<br /> set the value of sc_task to NULL under the protection of lock<br /> sc_state_lock, so as to avoid the issue caused by sc_timer not being<br /> properly shutdowned.<br /> <br /> [1]<br /> ODEBUG: free active (active state 0) object: 00000000dacb411a object type: timer_list hint: nilfs_construction_timeout<br /> Call trace:<br /> nilfs_segctor_destroy fs/nilfs2/segment.c:2811 [inline]<br /> nilfs_detach_log_writer+0x668/0x8cc fs/nilfs2/segment.c:2877<br /> nilfs_put_super+0x4c/0x12c fs/nilfs2/super.c:509

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ipv4: route: Prevent rt_bind_exception() from rebinding stale fnhe<br /> <br /> The sit driver&amp;#39;s packet transmission path calls: sit_tunnel_xmit() -&gt;<br /> update_or_create_fnhe(), which lead to fnhe_remove_oldest() being called<br /> to delete entries exceeding FNHE_RECLAIM_DEPTH+random.<br /> <br /> The race window is between fnhe_remove_oldest() selecting fnheX for<br /> deletion and the subsequent kfree_rcu(). During this time, the<br /> concurrent path&amp;#39;s __mkroute_output() -&gt; find_exception() can fetch the<br /> soon-to-be-deleted fnheX, and rt_bind_exception() then binds it with a<br /> new dst using a dst_hold(). When the original fnheX is freed via RCU,<br /> the dst reference remains permanently leaked.<br /> <br /> CPU 0 CPU 1<br /> __mkroute_output()<br /> find_exception() [fnheX]<br /> update_or_create_fnhe()<br /> fnhe_remove_oldest() [fnheX]<br /> rt_bind_exception() [bind dst]<br /> RCU callback [fnheX freed, dst leak]<br /> <br /> This issue manifests as a device reference count leak and a warning in<br /> dmesg when unregistering the net device:<br /> <br /> unregister_netdevice: waiting for sitX to become free. Usage count = N<br /> <br /> Ido Schimmel provided the simple test validation method [1].<br /> <br /> The fix clears &amp;#39;oldest-&gt;fnhe_daddr&amp;#39; before calling fnhe_flush_routes().<br /> Since rt_bind_exception() checks this field, setting it to zero prevents<br /> the stale fnhe from being reused and bound to a new dst just before it<br /> is freed.<br /> <br /> [1]<br /> ip netns add ns1<br /> ip -n ns1 link set dev lo up<br /> ip -n ns1 address add 192.0.2.1/32 dev lo<br /> ip -n ns1 link add name dummy1 up type dummy<br /> ip -n ns1 route add 192.0.2.2/32 dev dummy1<br /> ip -n ns1 link add name gretap1 up arp off type gretap \<br /> local 192.0.2.1 remote 192.0.2.2<br /> ip -n ns1 route add 198.51.0.0/16 dev gretap1<br /> taskset -c 0 ip netns exec ns1 mausezahn gretap1 \<br /> -A 198.51.100.1 -B 198.51.0.0/16 -t udp -p 1000 -c 0 -q &amp;<br /> taskset -c 2 ip netns exec ns1 mausezahn gretap1 \<br /> -A 198.51.100.1 -B 198.51.0.0/16 -t udp -p 1000 -c 0 -q &amp;<br /> sleep 10<br /> ip netns pids ns1 | xargs kill<br /> ip netns del ns1

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> NFS: Fix LTP test failures when timestamps are delegated<br /> <br /> The utimes01 and utime06 tests fail when delegated timestamps are<br /> enabled, specifically in subtests that modify the atime and mtime<br /> fields using the &amp;#39;nobody&amp;#39; user ID.<br /> <br /> The problem can be reproduced as follow:<br /> <br /> # echo "/media *(rw,no_root_squash,sync)" &gt;&gt; /etc/exports<br /> # export -ra<br /> # mount -o rw,nfsvers=4.2 127.0.0.1:/media /tmpdir<br /> # cd /opt/ltp<br /> # ./runltp -d /tmpdir -s utimes01<br /> # ./runltp -d /tmpdir -s utime06<br /> <br /> This issue occurs because nfs_setattr does not verify the inode&amp;#39;s<br /> UID against the caller&amp;#39;s fsuid when delegated timestamps are<br /> permitted for the inode.<br /> <br /> This patch adds the UID check and if it does not match then the<br /> request is sent to the server for permission checking.