In subsys/debug/coredump/coredumpshell.c, printcoredumphdr() used the 16-bit tgtcode field of a stored Zephyr coredump header directly as an index into coredumptargetcode2str[], a fixed 7-element array of string pointers, with no bounds check.
A stored coredump whose tgtcode is >= 7 causes an out-of-bounds read of a char* up to ~64K entries past the array; that value is passed as the %s argument to shellprint, which dereferences and walks it as a string. The result is either disclosure of device memory contents to the shell user or a crash when the out-of-bounds pointer is unmapped.
The defect is reached via the coredump print shell command (cmdcoredumpprintstoreddump -> prettyprintcoredump -> parseandprintcoredump -> printcoredumphdr). The tgtcode field is device-generated and in-range during normal crash handling, so triggering requires local shell access plus the ability to stage or corrupt the stored coredump in the flash/in-memory backend.
Introduced in v4.2.0 (commit 13abd7fe730) and present through v4.4.0; fixed by clamping out-of-range codes to the 'unknown' (index 0) entry.
{
"cna_assigner": "zephyr",
"osv_generated_from": "https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/10xxx/CVE-2026-10684.json",
"cwe_ids": [
"CWE-125"
]
}