{"id":"CVE-2026-99092","published":"2026-10-26T15:16:48.110","lastModified":"2026-10-26T15:16:48.110","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nriscv: patch: fix handling of bpf-jit execmem addresses\n\nCommit 8718e5a3090b (\"riscv: patch: skip fixmap mapping when kernel text\nis already writable\") gated the core_kernel_text() branch of patch_map()\non CONFIG_STRICT_KERNEL_RWX, and explicitly left the vmalloc branch\nunchanged. That branch has a separate problem.\n\nThat branch only creates a temporary writable fixmap alias when\nCONFIG_STRICT_MODULE_RWX is enabled, and otherwise assumes the target is\ndirectly writable and returns the address unchanged. That assumption no\nlonger holds: bpf_prog_pack_alloc calls set_memory_rox() on every pack\nunconditionally in alloc_new_pack(), independently of any\nCONFIG_STRICT_*_RWX option, so BPF text in the vmalloc area is read-only\nregardless of what CONFIG_STRICT_MODULE_RWX says.\n\nWith CONFIG_STRICT_MODULE_RWX=n, patch_map() returns the read-only\naddress directly, the subsequent copy_to_kernel_nofault() takes a\npage fault and returns -EFAULT. bpf_arch_text_copy() turns that into\n-EINVAL, which propagates through bpf_jit_binary_pack_finalize() to the\nWARN_ON() in bpf_int_jit_compile() and leaves the program un-JITed.\n\nNote that BPF cannot be fixed the way kprobes was in\ncommit bdc46e507b59 (\"riscv: mm: make EXECMEM_KPROBES writable without\nCONFIG_STRICT_MODULE_RWX\"). EXECMEM_BPF is already PAGE_KERNEL, i.e.\nwritable at allocation time; the read-only mapping is established\nafterwards by generic code in alloc_new_pack(), which arch code cannot\ninfluence. The only place this can be handled is patch_map().\n\nThis is the same problem that was fixed on arm64 by commit b1480ed230ac\n(\"arm64: patching: fix handling of execmem addresses\"), and the fix is\nthe same: CONFIG_EXECMEM is what actually tracks whether the vmalloc\narea can hold executable memory that needs a temporary alias to be\nwritten. CONFIG_BPF_JIT, CONFIG_KPROBES and CONFIG_MODULES all select\nit.\n\nNote that this is not limited to CONFIG_MODULES=n. Unlike arm64, riscv\nselects CONFIG_ARCH_OPTIONAL_KERNEL_RWX, so CONFIG_STRICT_MODULE_RWX can\nbe disabled with CONFIG_MODULES=y as well, and the bug is reachable in\nthat configuration too.\n\nThe !CONFIG_MMU build is unaffected: it has its own __patch_insn_set()\nand __patch_insn_write() implementations and never calls patch_map().","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/562a35b9e0b3d9483d9472995219cadb3592db3d","tags":[]},{"url":"https://git.kernel.org/stable/c/9e8fc7972b398e9dc2415ece283d8f64edac3858","tags":[]},{"url":"https://git.kernel.org/stable/c/a510e1ae8c9030750c2edf7b0bc9b186cd8c7a06","tags":[]},{"url":"https://git.kernel.org/stable/c/f71d6ea3843c8038c431c2699482f246aea37ca0","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}