[Toybox] Fails to build with CONFIG_SYSLOGD=y (ld: syslogd.c:(.text.logmsg+0x147): undefined reference to `prioritynames')

enh enh at google.com
Fri Oct 9 16:57:52 UTC 2026


On Fri, Oct 9, 2026 at 2:05 AM Rob Landley via Toybox <
toybox at lists.landley.net> wrote:

> On 10/8/26 05:57, zyxhere💭 wrote:
> > Hi I know this is a pending command but it should build right?
>
> It does on my Devuan Excalibur laptop (which is Debian Trixie without
> systemd):
>
> $ make syslogd
> scripts/single.sh syslogd
> syslogd:
> warning: using unfinished code from toys/pending
> Compile syslogd
> ....................
> $
>
> > Using toybox-0.8.15 on Gentoo linux with musl (had to patch the build
> > because https://github.com/landley/toybox/pull/332 is still a thing).
>
> I believe the only targets that still use that are the interactive
> configs (menuconfig and legacy text "config" that asks yes/no
> questions). Everything else is scripts/genconfig.sh now.
>
> Doing a new menuconfig and getting rid of the old kconfig dir is
> currently in my top 10 todo items, and might make it into the next
> release, but my in the absence of immediate deadlines (or anybody paying
> me to do this) my development work goes off on a lot of tangents.
>
> Right after the release I wanted to go back to getting the toybox build
> and test suite to run under mkroot system images (so I can start to
> properly design root tests, and maybe deal with the fact that unshare -t
> is FUCKING USELESS ala https://mstdn.jp/@landley/117345412207364770
> which I need to design a way to deal with but am NOT playing with on my
> development system again just now). Running my scripts under mkroot has
> MOSTLY been blocked by missing toysh features, but there might be OTHER
> things missing in mkroot (kernel config for the tests or commands I
> haven't noticed I'm using) so I built a musl-static bash 5.3 to try to
> implement the rest of a mkroot/packages/testsuite build script against
> (since that bash version was the one in LFS 12.4 so I had the source
> lying around).
>
> And once I had the binary I wanted to wget it into an existing image so
> I diverged into redoing the httpd_main() plumbing so it could act as a
> standalone server without needing an inetd front end (there should be a
> function in lib/net.c to call back into an inetd handler but option
> parsing gets a little weird because -p [IP:]PORT -u USER[:GRP] -m limit
> might be common to multiple servers...?)
>
> But THEN I hit a weird bug where in mkroot "mount /dev/sda /mnt" says
> "mount /dev/sda: need -t" but it succeeds in mounting the squashfs
> _annyway_ (wha...?) and I went "I wanna strace this" but I deleted the
> old aboriginal linux binaries off my website because people built
> viruses with my toolchains so my old toolchains keep getting
> MISIDENTIFIED AS VIRUS CODE by various scanners and it caused my domain
> to wind up in spam blockers last year, and building the old aboriginal
> linux versions under my current musl toolchain is doesn't work (its
> config doesn't know what musl is and when I hacked past that the code
> was full of #ifdef staircases that kept falling through to #errors) and
> when I chopped my way through ALL THAT the link died with a bunch of ld:
> syscall.o:(.rodata+0x2e080): undefined reference to `sys_getdents64
> where I have no idea what musl thinks it's exporting (some horrible
> macro thing maybe) that this old strace code isn't matching up with...
>
> And I went "hang on, there's an strace.c in pending" so I started fixing
> THAT up and argued with dalias on #musl about "why does musl's or1k
> headers not have the mmap or lseek syscalls, I mean you've got SYS_blah
> versions of syscalls you haven't got _NR_blah versions of could it at
> least be consistent there... (answer, no: because Rich.) Plus trying to
> figure out mips architecture support is just crazy (there's O32, N32,
> N64, and whatever the hell
>
> https://github.com/torvalds/linux/blob/v7.0/arch/mips/include/asm/syscall.h#L75
> is doing let alone the function before it... v0 vs a0 map to what in the
> exported array? Anyway, I got it compiling and checked something in...)
>
> Along the way the split bug report came in which I've written half a
> reply to, and mostly think I've figured out how I want to address but it
> REALLY seems like there should be some kind of same_file() libc function
> to collate:
>
> toys/other/mountpoint.c:  toys.exitval = !(st1.st_dev != st2.st_dev ||
> st1.st_ino == st2.st_ino);
> toys/pending/sh.c:    if (stat(pwd, &st2) || st1.st_dev!=st2.st_dev ||
> st1.st_ino!=st2.st_ino)
> toys/posix/test.c:      if (i==12) return (st1.st_dev==st2.st_dev) &&
> (st1.st_ino==st2.st_ino);
>
> But there aren't QUITE enough users


three is my usual magic number for "okay, it's worth factoring this out"
... and you're talking about this because you're about to add a fouth.


> and would the two stat bufs be the
> argument or should it fstat() a supplied file descriptor or...?
>

add the one that takes two stat bufs now, because you need that, and add
the "i'll stat for you variant if _that_ hits the requisite number of
duplicates to warrant it"?

Oh, and as long as I've got a static bash I once again tried to build a
> static busybox to throw in the $PATH in a mkroot image on the theory
> that versions of tools known to have built LFS 12.4 means I can
> implement the LFS 12.4 buidl and then swap the toybox versions over one
> by one like I did with busybox and gnu back in the day, but busybox
> STILL refuses to build without bzip2 compressor in the $PATH bit it
> turns out that gratuitous build dependency only goes back a couple
> releases: 1.35 complained about its absence but built anyway, and then
> died because "diff" isn't in root/build/airlock and _that_ dependency
> goes back quite a ways so I wanted to rebuild adding that but the
> PENDING= argument to mkroot only adds commands to the target
> filesystem's toybox config so I added an AIRPEND= argument to add
> pending commands to the airlock build (which involved moving be2csv and
> csv2cfg up with announce() and die() and redoing the comments), and when
> I noticed this email, I was trying to figure out why:
>
>    PREFIX="$AIRLOCK" KCONFIG_CONFIG=.singleconfig_airlock CROSS_COMPILE=\
>      make ${AIRPEND:-KCONFIG_ALLCONFIG=<(csv2cfg $(be2csv $AIRPEND) y)} \
>      clean defconfig toybox install_airlock && # see scripts/install.sh&&
>
> Is turning into:
>
> PREFIX=/home/landley/toybox/kleen/root/build/airlock
> KCONFIG_CONFIG=.singleconfig_airlock CROSS_COMPILE= make diff clean
> defconfig toybox install_airlock
>
> And thus trying to do a single build of "diff", which is wrong. (That
> "diff" SHOULD be "KCONFIG_ALLCONFIG=/dev/fd/63" or similar.) In the
> filesystem build of toybox, it's doing:
>
>    PREFIX="$ROOT" KCONFIG_CONFIG="$TOYCFG" make clean \
>      ${CONF:-defconfig KCONFIG_ALLCONFIG=<(csv2cfg $(be2csv $PENDING SH
> ROUTE) y)} \
>      toybox install || exit 1
>
> Which is working as expected. (Can I not use $($AIRPEND) inside
> ${AIRPEND:-} ? Do I need a temporary variable? No, don't bother
> answering that, rhetorical question, I'll presumably have figured it out
> or bypassed it by then...)
>
> Anyway, let's throw your interrupt on top of the pile:
>
> > .config and build.log attached
>
> What does that say...
>
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld: cannot find -liconv:
> Invalid argument
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld: cannot find -lselinux:
> Invalid argument
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld: cannot find -ltls:
> Invalid argument
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld: cannot find -lsmack:
> Invalid argument
> collect2: error: ld returned 1 exit status
> collect2: error: ld returned 1 exit status
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld: cannot find -llog:
> Invalid argument
> collect2: error: ld returned 1 exit status
> collect2: error: ld returned 1 exit status
> collect2: error: ld returned 1 exit status
>
> That's probe debris from scripts/make.sh. It's not actually an error, it
> just means probes to see which libraries are available did not find some
> of the libraries. It's this code:
>
> > # Respond to V= by echoing command lines as well as running them
> > do_loudly()
> > {
> >   { [ -n "$V" ] && echo "$@" || echo -n "$DOTPROG" ; } >&2
> >   "$@"
> > }
> >
> >
> > # Respond to V= by echoing command lines as well as running them
> > do_loudly()
> > {
> >   { [ -n "$V" ] && echo "$@" || echo -n "$DOTPROG" ; } >&2
> >   "$@"
> > }
> >
> > # --as-needed removes libraries we don't use any symbols out of, but the
> > # compiler has no way to ignore a library that doesn't exist, so detect
> > # and skip nonexistent libraries for it (probing in parallel).
> > LIBRARIES=$(
> >   [ -z "$V" ] && X=/dev/null || X=/dev/stderr
> >   for i in util crypt m resolv selinux smack attr crypto z log iconv tls
> ssl
> >   do
> >     do_loudly ${CROSS_COMPILE}${CC} $CFLAGS $LDFLAGS -xc - -l$i >>$X
> 2>&1 \
> >       -o /dev/null <<<"int main(int argc,char*argv[]){return 0;}" &&
> >       echo -l$i &
> >   done | sort | xargs
> > )
>
> It really looks like you have V=1 set, because >>$X 2>&1 should sent the
> error output to $X which is either /dev/null or /dev/stderr depending on
> what V= is set to.
>
> Your linux kernel builds would similarly be insanely chatty if you have
> V=1 exported in your environment. I dunno how much of them would look
> like failures because there's just so much output...
>
> Did the build fail, or did you just see error messages in your log output?
>
> Ah, right: mousepad just showed me the start of the file without an
> obvious scrollbar, cursoring down there's SO MUCH MORE debris. Let's see...
>
> lib/hash.c: In function 'hash_by_name':
> lib/hash.c:33:5: warning: 'MD5_Init' is deprecated: Since OpenSSL 3.0
> [-Wdeprecated-declarations]
>     33 |     USE_MD5SUM(HASH_INIT("md5sum", MD5),)
>        |     ^~~~~~~~~~
>
> I haven't actually tried building syslogd with openssl support. No idea
> what might go boing...
>
> Ok, here's the relevant part. (Like the last 5% of the file.) It did
> basically this:
>
> cc -Wall -Wundef -Werror=implicit-function-declaration
> -Wno-char-subscripts -Wno-pointer-sign -funsigned-char -Wno-restrict
> -Wno-format-overflow -Wno-discarded-qualifiers -I . -Os
> -ffunction-sections -fdata-sections -fno-asynchronous-unwind-tables
> -fno-strict-aliasing generated/unstripped/obj/*.o -Wl,--gc-sections
> -Wl,--as-needed -lattr -lcrypt -lcrypto -lm -lresolv -lssl -lutil -lz -o
> generated/unstripped/toybox
>
> And got:
>
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld:
> generated/unstripped/obj/syslogd.o: warning: relocation against
> `prioritynames' in read-only section `.text.logger_lookup'
>
> $ grep -r prioritynames
> x86_64-linux-gnu/sys/syslog.h:CODE prioritynames[] =
>
> That's out of debian's libc. I'm using a symbol in a way that seems
> normal to me, and your libc/headers get mad about it. (String constants
> are also in a read only section, I can stick 'em in a normal char
> pointer and the segfault's on me if I write to it. I told it
> -Wno-discarded-qualifiers to get it to shut up, and it didn't obviously
> complain in other commands where I took out the (void *) typecasts after
> doing that, so...)
>
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld:
> generated/unstripped/obj/syslogd.o: in function `logmsg':
> syslogd.c:(.text.logmsg+0x104): undefined reference to `facilitynames'
>
> Let's see...
>
> $ grep -r facilitynames /usr/include
> /usr/include/x86_64-linux-gnu/sys/syslog.h:CODE facilitynames[] =
>
> $ cd musl/musl/
> ~/musl/musl$ grep -r facilitynames
> include/syslog.h:#define facilitynames ((CODE *)(const CODE []){ \
>
> $ grep -r prioritynames android-ndk-r29-beta2
> android-ndk-r29-beta2/toolchains/llvm/prebuilt/linux-x86_64/sysroot/usr/include/syslog.h:static
>
> const CODE prioritynames[] = {
>
> I've built against debian's glibc, musl, and bionic. Your libc remains
> different than any of those.
>
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld:
> syslogd.c:(.text.logmsg+0x147): undefined reference to `prioritynames'
>
> Ok, hang on. Above your log was complaining that prioritynames was in a
> read only segement, but NOW it's saying it doesn't exist?
>
> PICK ONE.
>
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld:
> generated/unstripped/obj/syslogd.o: in function `logger_lookup':
> syslogd.c:(.text.logger_lookup+0x4): undefined reference to `facilitynames'
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld:
> syslogd.c:(.text.logger_lookup+0x29): undefined reference to
> `prioritynames'
>
> Same bug.
>
> /usr/x86_64-pc-linux-musl/binutils-bin/2.47/ld: read-only segment has
> dynamic relocations
>
> And again, it simultaneously doesn't exist AND is in a read only segment.
>
> Your link environment is contradicting itself.
>
> Sigh, let's search the web for that first weird link message:
>
> https://github.com/vmagnin/gtk-fortran/issues/183
>
> Hmmm, it looks like this sort of thing happens when mixing pic and
> non-pic code? Or possibly when putting a nonrelocatable symbol into a
> shared library, it later barfs if you try to _use_ that symbol out of
> the library because it can't relocate it...
>
> And this one is implying that your libraries may depend on each other
> and get confused when loaded out of order:
>
>
> https://stackoverflow.com/questions/65407610/relocation-against-xxx-in-read-only-section-text-wrong-compiler-or-linux-se
>
> I have no idea what the case is here, but toys/pending/syslogd.c seems
> (to me) to be making fairly straightforward use of a library symbol in a
> way that's previously worked on glibc, musl, and bionic.
>
> Rob
> _______________________________________________
> Toybox mailing list
> Toybox at lists.landley.net
> https://lists.landley.net/mailman/listinfo/toybox
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.landley.net/pipermail/toybox/attachments/20261009/74a2fac7/attachment-0001.htm>


More information about the Toybox mailing list