CVE-2026-68160

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> ceph: fix pre-auth out-of-bounds read on snaptrace in ceph_handle_caps()<br /> <br /> ceph_handle_caps() reads snap_trace_len from the wire-format<br /> ceph_mds_caps header and uses it unconditionally to build a fake<br /> end pointer (snaptrace + snaptrace_len) that is later handed to<br /> ceph_update_snap_trace() in the CEPH_CAP_OP_IMPORT case:<br /> <br /> snaptrace = h + 1;<br /> snaptrace_len = le32_to_cpu(h-&gt;snap_trace_len);<br /> p = snaptrace + snaptrace_len;<br /> ...<br /> case CEPH_CAP_OP_IMPORT:<br /> if (snaptrace_len) {<br /> ...<br /> if (ceph_update_snap_trace(mdsc, snaptrace,<br /> snaptrace + snaptrace_len,<br /> false, &amp;realm)) { ... }<br /> <br /> ceph_update_snap_trace() then decodes a struct ceph_mds_snap_realm<br /> from snaptrace using ceph_decode_need(&amp;p, e, sizeof(*ri), bad)<br /> with the attacker-supplied fake end e == snaptrace + snaptrace_len.<br /> With snaptrace_len == 0xFFFFFFFF the bound check is trivially<br /> satisfied, ri = p reads sizeof(struct ceph_mds_snap_realm) past<br /> the legitimate msg-&gt;front buffer, and ri-&gt;num_snaps /<br /> ri-&gt;num_prior_parent_snaps then drive further out-of-bounds<br /> reads of the encoded snap arrays.<br /> <br /> The eleven msg_version &gt;= 2 .. msg_version &gt;= 12 decoder blocks<br /> above the op switch each catch this OOB through their<br /> ceph_decode_*_safe() / ceph_decode_need() helpers, but they sit<br /> behind a hdr.version-gated if, so a malicious or compromised<br /> MDS that sets msg-&gt;hdr.version = 1 reaches the IMPORT path with<br /> no version-gated decoder having validated snap_trace_len. The<br /> shape has been present since ceph_handle_caps() was introduced.<br /> <br /> Validate snap_trace_len against the message front buffer before<br /> consuming it, using the canonical ceph_decode_need() / ceph_has_room()<br /> helper. The helper bounds the length with subtraction (n = p) rather than pointer addition, so it is wrap-safe<br /> for the attacker-controlled u32 length on 32-bit builds where<br /> p + snap_trace_len could overflow the address space. This matches the<br /> rest of the ceph decode path (e.g. the pool_ns_len check a few lines<br /> below), and the existing goto bad cleanup already covers this exit<br /> path.