parse.mk revision 1.4 1 1.4 rillig # $NetBSD: parse.mk,v 1.4 2023/04/28 13:09:48 rillig Exp $
2 1.1 rillig #
3 1.1 rillig # Test those parts of the parsing that do not belong in any of the other
4 1.1 rillig # categories.
5 1.1 rillig
6 1.1 rillig # expect+1: Makefile appears to contain unresolved CVS/RCS/??? merge conflicts
7 1.1 rillig <<<<<< old
8 1.2 rillig
9 1.2 rillig # No diagnostic since the following line is parsed as a variable assignment,
10 1.2 rillig # even though the variable name is empty. See also varname-empty.mk.
11 1.2 rillig ====== middle
12 1.2 rillig
13 1.2 rillig # expect+1: Makefile appears to contain unresolved CVS/RCS/??? merge conflicts
14 1.2 rillig >>>>>> new
15 1.3 rillig
16 1.3 rillig
17 1.3 rillig # Since parse.c 1.578 from 2021-12-14 and before parse.c 1.681 from
18 1.3 rillig # 2022-07-24, if a line of a makefile could only be a dependency specification
19 1.3 rillig # but didn't contain any of the dependency operators ':', '!', '::' and its
20 1.3 rillig # expansion ended with a space, make read a single byte from the memory beyond
21 1.3 rillig # the expanded line's terminating '\0'.
22 1.3 rillig #
23 1.3 rillig # https://bugs.freebsd.org/265119
24 1.3 rillig one-target ${:U }
25 1.4 rillig
26 1.4 rillig
27 1.4 rillig # Since parse.c 1.656 from 2022-01-27 and before parse.c 1.662 from
28 1.4 rillig # 2022-02-05, there was an out-of-bounds read in Parse_IsVar when looking for
29 1.4 rillig # a variable assignment in a dependency line with trailing whitespace. Lines
30 1.4 rillig # without trailing whitespace were not affected. Global variable assignments
31 1.4 rillig # were guaranteed to have no trailing whitespace and were thus not affected.
32 1.4 rillig #
33 1.4 rillig # Try to reproduce some variants that may lead to a crash, depending on the
34 1.4 rillig # memory allocator. To get a crash, the terminating '\0' of the line must be
35 1.4 rillig # the last byte of a memory page. The expression '${:U}' forces this trailing
36 1.4 rillig # whitespace.
37 1.4 rillig
38 1.4 rillig # On FreeBSD x86_64, a crash could in some cases be forced using the following
39 1.4 rillig # line, which has length 47, and if the memory for the expanded line starts at
40 1.4 rillig # 0xXXXX_XXd0, the terminating '\0' may end up at 0xXXXX_Xfff:
41 1.4 rillig Try_to_crash_FreeBSD.xxxxxxxxxxxxxxxxxx: 12345 ${:U}
42 1.4 rillig
43 1.4 rillig # The following line has length 4095 after being expanded, so line[4095] ==
44 1.4 rillig # '\0'. If the line is
45 1.4 rillig # allocated on a page boundary and the following page is not mapped, this line
46 1.4 rillig # leads to a segmentation fault.
47 1.4 rillig ${:U:range=511:@_@1234567@:ts.}: 12345 ${:U}
48 1.4 rillig
49 1.4 rillig # The following line has length 8191, so line[8191] == '\0'. If the line is
50 1.4 rillig # allocated on a page boundary and the following page is not mapped, this line
51 1.4 rillig # leads to a segmentation fault.
52 1.4 rillig ${:U:range=1023:@_@1234567@:ts.}: 12345 ${:U}
53 1.4 rillig
54 1.4 rillig 12345:
55