Skip to content

fileinfo: built-in magic database php_magic_database is not aligned, breaks on strict-alignment CPUs (ARMv5, 32-bit) #25029

Closed

Description

@EmrsnFrbo

Description

The following code:

<?php
var_dump(mime_content_type('/tmp/file.zip'));

Resulted in this output:

ValueError: Path must not be empty.

But I expected this output instead:

application/zip

Description:
On ARMv5 (32-bit cpu), built with GCC and -Os, finfo_open() and mime_content_type() fail in PHP 8.5.x with:
ValueError: Path must not be empty.

Cause:
php_magic_database (ext/fileinfo/data_file.c) is declared as const unsigned char[][], so the compiler only guarantees 1-byte alignment.
libmagic apprentice_map() casts the buffer to uint32_t * to check MAGICNO/VERSIONNO, then uses it as an array of struct magic (8-byte aligned).
On CPUs that don't support unaligned access, the header reads back wrong and the built-in database is rejected.
libmagic then falls back to apprentice_load(NULL), which calls the stream layer with an empty path and throws the ValueError.
Whether the bug appears depends on where the linker places the symbol (nm shows an address that isn't a multiple of 4 or 8). GCC on ARM normally word-aligns large arrays, but not with -Os.

Fix: declare the array with ZEND_SET_ALIGNED(8, ...) in create_data_file.php and regenerate data_file.c

Since ext/fileinfo/create_data_file.php is generated I will not provide a patch for it, but this is the AI generated proposal:
-echo 'const unsigned char php_magic_database[' . count($chunks) . '][' . CHUNK_SIZE . "] ZEND_NONSTRING = {\n";
+/* libmagic accesses the database as uint32_t and struct magic, which need 8-byte alignment */
+echo 'ZEND_SET_ALIGNED(8, const unsigned char php_magic_database[' . count($chunks) . '][' . CHUNK_SIZE . "]) ZEND_NONSTRING = {\n";

My own patch to solve my problem: directly in data_file.c
--- php-8.5.12/ext/fileinfo/data_file.c 2026-10-28 16:30:57.011450103 +0100
+++ php-8.5.12.modified/ext/fileinfo/data_file.c 2026-10-28 16:22:31.851759910 +0100
@@ -3 +3 @@
-const unsigned char php_magic_database[10113][1024] ZEND_NONSTRING = {
+const unsigned char php_magic_database[10113][1024] ZEND_NONSTRING attribute((aligned(8))) = {

PHP Version

PHP 8.5.12 (cli) (built: Oct 28 2026 16:52:49) (NTS)
Copyright (c) The PHP Group
Zend Engine v4.5.12, Copyright (c) Zend Technologies
    with Zend OPcache v8.5.12, Copyright (c), by Zend Technologies

Operating System

No response

Activity

  1. added 2 commits that reference this issue on Oct 30, 2026
    a077e8a
    ad687b6
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions