CVE-2024-42305

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ext4: check dot and dotdot of dx_root before making dir indexed<br /> <br /> Syzbot reports a issue as follows:<br /> ============================================<br /> BUG: unable to handle page fault for address: ffffed11022e24fe<br /> PGD 23ffee067 P4D 23ffee067 PUD 0<br /> Oops: Oops: 0000 [#1] PREEMPT SMP KASAN PTI<br /> CPU: 0 PID: 5079 Comm: syz-executor306 Not tainted 6.10.0-rc5-g55027e689933 #0<br /> Call Trace:<br /> <br /> make_indexed_dir+0xdaf/0x13c0 fs/ext4/namei.c:2341<br /> ext4_add_entry+0x222a/0x25d0 fs/ext4/namei.c:2451<br /> ext4_rename fs/ext4/namei.c:3936 [inline]<br /> ext4_rename2+0x26e5/0x4370 fs/ext4/namei.c:4214<br /> [...]<br /> ============================================<br /> <br /> The immediate cause of this problem is that there is only one valid dentry<br /> for the block to be split during do_split, so split==0 results in out of<br /> bounds accesses to the map triggering the issue.<br /> <br /> do_split<br /> unsigned split<br /> dx_make_map<br /> count = 1<br /> split = count/2 = 0;<br /> continued = hash2 == map[split - 1].hash;<br /> ---&gt; map[4294967295]<br /> <br /> The maximum length of a filename is 255 and the minimum block size is 1024,<br /> so it is always guaranteed that the number of entries is greater than or<br /> equal to 2 when do_split() is called.<br /> <br /> But syzbot&amp;#39;s crafted image has no dot and dotdot in dir, and the dentry<br /> distribution in dirblock is as follows:<br /> <br /> bus dentry1 hole dentry2 free<br /> |xx--|xx-------------|...............|xx-------------|...............|<br /> 0 12 (8+248)=256 268 256 524 (8+256)=264 788 236 1024<br /> <br /> So when renaming dentry1 increases its name_len length by 1, neither hole<br /> nor free is sufficient to hold the new dentry, and make_indexed_dir() is<br /> called.<br /> <br /> In make_indexed_dir() it is assumed that the first two entries of the<br /> dirblock must be dot and dotdot, so bus and dentry1 are left in dx_root<br /> because they are treated as dot and dotdot, and only dentry2 is moved<br /> to the new leaf block. That&amp;#39;s why count is equal to 1.<br /> <br /> Therefore add the ext4_check_dx_root() helper function to add more sanity<br /> checks to dot and dotdot before starting the conversion to avoid the above<br /> issue.