6 The FRR library provides an introspection facility called "xrefs." The intent
7 is to provide structured access to annotated entities in the compiled binary,
8 such as log messages and thread scheduling calls.
13 Support for emitting an xref is included in the macros for the specific
14 entities, e.g. :c:func:`zlog_info` contains the relevant statements. The only
15 requirement for the system to work is a GNU compatible linker that supports
16 section start/end symbols. (The only known linker on any system FRR supports
17 that does not do this is the Solaris linker.)
19 To verify xrefs have been included in a binary or dynamic library, run
20 ``readelf -n binary``. For individual object files, it's
21 ``readelf -S object.o | grep xref_array`` instead.
23 Structure and contents
24 ----------------------
26 As a slight improvement to security and fault detection, xrefs are divided into
27 a ``const struct xref *`` and an optional ``struct xrefdata *``. The required
30 .. c:member:: enum xref_type xref.type
32 Identifies what kind of object the xref points to.
34 .. c:member:: int line
35 .. c:member:: const char *xref.file
36 .. c:member:: const char *xref.func
38 Source code location of the xref. ``func`` will be ``<global>`` for
39 xrefs outside of a function.
41 .. c:member:: struct xrefdata *xref.xrefdata
43 The optional writable part of the xref. NULL if no non-const part exists.
45 The optional non-const part has:
47 .. c:member:: const struct xref *xrefdata.xref
49 Pointer back to the constant part. Since circular pointers are close to
50 impossible to emit from inside a function body's static variables, this
51 is initialized at startup.
53 .. c:member:: char xrefdata.uid[16]
55 Unique identifier, see below.
57 .. c:member:: const char *xrefdata.hashstr
58 .. c:member:: uint32_t xrefdata.hashu32[2]
60 Input to unique identifier calculation. These should encompass all
61 details needed to make an xref unique. If more than one string should
62 be considered, use string concatenation for the initializer.
64 Both structures can be extended by embedding them in a larger type-specific
65 struct, e.g. ``struct xref_logmsg *``.
70 All xrefs that have a writable ``struct xrefdata *`` part are assigned an
71 unique identifier, which is formed as base32 (crockford) SHA256 on:
74 - the ``hashstr`` field
75 - the ``hashu32`` fields
79 Function names and line numbers are intentionally not included to allow
80 moving items within a file without affecting the identifier.
82 For running executables, this hash is calculated once at startup. When
83 directly reading from an ELF file with external tooling, the value must be
84 calculated when necessary.
86 The identifiers have the form ``AXXXX-XXXXX`` where ``X`` is
87 ``0-9, A-Z except I,L,O,U`` and ``A`` is ``G-Z except I,L,O,U`` (i.e. the
88 identifiers always start with a letter.) When reading identifiers from user
89 input, ``I`` and ``L`` should be replaced with ``1`` and ``O`` should be
90 replaced with ``0``. There are 49 bits of entropy in this identifier.
95 Xrefs are nothing other than global variables with some extra glue to make
96 them possible to find from the outside by looking at the binary. The first
97 non-obvious part is that they can occur inside of functions, since they're
98 defined as ``static``. They don't have a visible name -- they don't need one.
100 To make finding these variables possible, another global variable, a pointer
101 to the first one, is created in the same way. However, it is put in a special
102 ELF section through ``__attribute__((section("xref_array")))``. This is the
103 section you can see with readelf.
105 Finally, on the level of a whole executable or library, the linker will stuff
106 the individual pointers consecutive to each other since they're in the same
107 section — hence the array. Start and end of this array is given by the
108 linker-autogenerated ``__start_xref_array`` and ``__stop_xref_array`` symbols.
109 Using these, both a constructor to run at startup as well as an ELF note are
112 The ELF note is the entrypoint for externally retrieving xrefs from a binary
113 without having to run it. It can be found by walking through the ELF data
114 structures even if the binary has been fully stripped of debug and section
115 information. SystemTap's SDT probes & LTTng's trace points work in the same
116 way (though they emit 1 note for each probe, while xrefs only emit one note
117 in total which refers to the array.) Using xrefs does not impact SystemTap
118 or LTTng, the notes have identifiers they can be distinguished by.
120 The ELF structure of a linked binary (library or executable) will look like
123 $ readelf --wide -l -n lib/.libs/libfrr.so
125 Elf file type is DYN (Shared object file)
127 There are 12 program headers, starting at offset 64
130 Type Offset VirtAddr PhysAddr FileSiz MemSiz Flg Align
131 PHDR 0x000040 0x0000000000000040 0x0000000000000040 0x0002a0 0x0002a0 R 0x8
132 INTERP 0x125560 0x0000000000125560 0x0000000000125560 0x00001c 0x00001c R 0x10
133 [Requesting program interpreter: /lib64/ld-linux-x86-64.so.2]
134 LOAD 0x000000 0x0000000000000000 0x0000000000000000 0x02aff0 0x02aff0 R 0x1000
135 LOAD 0x02b000 0x000000000002b000 0x000000000002b000 0x0b2889 0x0b2889 R E 0x1000
136 LOAD 0x0de000 0x00000000000de000 0x00000000000de000 0x070048 0x070048 R 0x1000
137 LOAD 0x14e428 0x000000000014f428 0x000000000014f428 0x00fb70 0x01a2b8 RW 0x1000
138 DYNAMIC 0x157a40 0x0000000000158a40 0x0000000000158a40 0x000270 0x000270 RW 0x8
139 NOTE 0x0002e0 0x00000000000002e0 0x00000000000002e0 0x00004c 0x00004c R 0x4
140 TLS 0x14e428 0x000000000014f428 0x000000000014f428 0x000000 0x000008 R 0x8
141 GNU_EH_FRAME 0x12557c 0x000000000012557c 0x000000000012557c 0x00819c 0x00819c R 0x4
142 GNU_STACK 0x000000 0x0000000000000000 0x0000000000000000 0x000000 0x000000 RW 0x10
143 GNU_RELRO 0x14e428 0x000000000014f428 0x000000000014f428 0x009bd8 0x009bd8 R 0x1
147 Displaying notes found in: .note.gnu.build-id
148 Owner Data size Description
149 GNU 0x00000014 NT_GNU_BUILD_ID (unique build ID bitstring) Build ID: 6a1f66be38b523095ebd6ec13cc15820cede903d
151 Displaying notes found in: .note.FRR
152 Owner Data size Description
153 FRRouting 0x00000010 Unknown note type: (0x46455258) description data: 6c eb 15 00 00 00 00 00 74 ec 15 00 00 00 00 00
155 Where 0x15eb6c…0x15ec74 are the offsets (relative to the note itself) where
156 the xref array is in the file. Also note the owner is clearly marked as
157 "FRRouting" and the type is "XREF" in hex.
159 For SystemTap's use of ELF notes, refer to
160 https://libstapsdt.readthedocs.io/en/latest/how-it-works/internals.html as an
165 Due to GCC bug 41091, the "xref_array" section is not correctly generated
166 for C++ code when compiled by GCC. A workaround is present for runtime
167 functionality, but to extract the xrefs from a C++ source file, it needs
168 to be built with clang (or a future fixed version of GCC) instead.
173 The FRR source contains a matching tool to extract xref data from compiled ELF
174 binaries in ``python/xrelfo.py``. This tool uses CPython extensions
175 implemented in ``clippy`` and must therefore be executed with that.
177 ``xrelfo.py`` processes input from one or more ELF file (.o, .so, executable),
178 libtool object (.lo, .la, executable wrapper script) or JSON (output from
179 ``xrelfo.py``) and generates an output JSON file. During standard FRR build,
180 it is invoked on all binaries and libraries and the result is combined into
183 ELF files from any operating system, CPU architecture and endianness can be
184 processed on any host. Any issues with this are bugs in ``xrelfo.py``
185 (or clippy's ELF code.)
187 ``xrelfo.py`` also performs some sanity checking, particularly on log
188 messages. The following options are available:
190 .. option:: -o OUTPUT
192 Filename to write JSON output to. As a convention, a ``.xref`` filename
195 .. option:: -Wlog-format
197 Performs extra checks on log message format strings, particularly checks
198 for ``\t`` and ``\n`` characters (which should not be used in log messages).
200 .. option:: -Wlog-args
202 Generates cleanup hints for format string arguments where
203 :c:func:`printfrr()` extensions could be used, e.g. replacing ``inet_ntoa``
206 .. option:: --profile
208 Runs the Python profiler to identify hotspots in the ``xrelfo.py`` code.
210 ``xrelfo.py`` uses information about C structure definitions saved in
211 ``python/xrefstructs.json``. This file is included with the FRR sources and
212 only needs to be regenerated when some of the ``struct xref_*`` definitions
213 are changed (which should be almost never). The file is written by
214 ``python/tiabwarfo.py``, which uses ``pahole`` to extract the necessary data
215 from DWARF information.