CVE-2024-50265

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ocfs2: remove entry once instead of null-ptr-dereference in ocfs2_xa_remove()<br /> <br /> Syzkaller is able to provoke null-ptr-dereference in ocfs2_xa_remove():<br /> <br /> [ 57.319872] (a.out,1161,7):ocfs2_xa_remove:2028 ERROR: status = -12<br /> [ 57.320420] (a.out,1161,7):ocfs2_xa_cleanup_value_truncate:1999 ERROR: Partial truncate while removing xattr overlay.upper. Leaking 1 clusters and removing the entry<br /> [ 57.321727] BUG: kernel NULL pointer dereference, address: 0000000000000004<br /> [...]<br /> [ 57.325727] RIP: 0010:ocfs2_xa_block_wipe_namevalue+0x2a/0xc0<br /> [...]<br /> [ 57.331328] Call Trace:<br /> [ 57.331477] <br /> [...]<br /> [ 57.333511] ? do_user_addr_fault+0x3e5/0x740<br /> [ 57.333778] ? exc_page_fault+0x70/0x170<br /> [ 57.334016] ? asm_exc_page_fault+0x2b/0x30<br /> [ 57.334263] ? __pfx_ocfs2_xa_block_wipe_namevalue+0x10/0x10<br /> [ 57.334596] ? ocfs2_xa_block_wipe_namevalue+0x2a/0xc0<br /> [ 57.334913] ocfs2_xa_remove_entry+0x23/0xc0<br /> [ 57.335164] ocfs2_xa_set+0x704/0xcf0<br /> [ 57.335381] ? _raw_spin_unlock+0x1a/0x40<br /> [ 57.335620] ? ocfs2_inode_cache_unlock+0x16/0x20<br /> [ 57.335915] ? trace_preempt_on+0x1e/0x70<br /> [ 57.336153] ? start_this_handle+0x16c/0x500<br /> [ 57.336410] ? preempt_count_sub+0x50/0x80<br /> [ 57.336656] ? _raw_read_unlock+0x20/0x40<br /> [ 57.336906] ? start_this_handle+0x16c/0x500<br /> [ 57.337162] ocfs2_xattr_block_set+0xa6/0x1e0<br /> [ 57.337424] __ocfs2_xattr_set_handle+0x1fd/0x5d0<br /> [ 57.337706] ? ocfs2_start_trans+0x13d/0x290<br /> [ 57.337971] ocfs2_xattr_set+0xb13/0xfb0<br /> [ 57.338207] ? dput+0x46/0x1c0<br /> [ 57.338393] ocfs2_xattr_trusted_set+0x28/0x30<br /> [ 57.338665] ? ocfs2_xattr_trusted_set+0x28/0x30<br /> [ 57.338948] __vfs_removexattr+0x92/0xc0<br /> [ 57.339182] __vfs_removexattr_locked+0xd5/0x190<br /> [ 57.339456] ? preempt_count_sub+0x50/0x80<br /> [ 57.339705] vfs_removexattr+0x5f/0x100<br /> [...]<br /> <br /> Reproducer uses faultinject facility to fail ocfs2_xa_remove() -&gt;<br /> ocfs2_xa_value_truncate() with -ENOMEM.<br /> <br /> In this case the comment mentions that we can return 0 if<br /> ocfs2_xa_cleanup_value_truncate() is going to wipe the entry<br /> anyway. But the following &amp;#39;rc&amp;#39; check is wrong and execution flow do<br /> &amp;#39;ocfs2_xa_remove_entry(loc);&amp;#39; twice:<br /> * 1st: in ocfs2_xa_cleanup_value_truncate();<br /> * 2nd: returning back to ocfs2_xa_remove() instead of going to &amp;#39;out&amp;#39;.<br /> <br /> Fix this by skipping the 2nd removal of the same entry and making<br /> syzkaller repro happy.