nexoniarz/SchOS
nexoniarz/SchOS
Watch
1
0
Fork
You've already forked SchOS
 0
Code Issues Pull requests Projects Releases 1 Packages Wiki Activity Actions

build: fix four things a clean build found, three of them mine

Browse source 
`make` had been failing since the e1000 work and I did not see it, because I
was checking build output with `grep error` -- which matches neither "Błąd" nor
mkfs's "image full". Checking the exit status instead found all of this in a
minute.

**usbstick.img was full.** `make ldls-stage` puts a module and its BTF in
sysroot/lib/modules, and a vmlinux BTF is 5 MB on its own; the 16 MB test image
could not hold it. The size cannot move -- tools/windows/bootcheck.ps1 asserts
READ CAPACITY returns exactly 32768 sectors -- so the exclusions grew instead:
lib and usr as well as bin. That image exists to prove the USB driver can read
one block, so its contents were never the point. It also stopped a GPL module
riding into a test artifact, which RAMDISK_EXCLUDE does for the ramdisk and
nothing was doing for this one.

**toybox's httpd was shadowing ours.** /bin/httpd answered `-p` with toybox's
"Unknown option 'p'", because whichever install rule ran second won the name.
TOYBOX_SKIP existed for exactly this and held one entry, `ping`, found the same
hard way. It is now generated from cmd/ -- a list you have to remember is a
list that is wrong the next time cmd/ grows, and it already was: `chvt` had
been silently toybox's, whose VT ioctls are Linux's and not ours.

**ldlsd would not link with clang.** `*out = in;` on a 96-byte struct is a call
to memcpy as far as the C standard is concerned; gcc unrolled it and clang
emitted the call, so a freestanding program linked for months and then did not
on a machine where nothing about it had changed but the compiler. lib/glue.c
now provides memcpy, memmove, memset and memcmp -- the four an implementation
may call without being asked.

**OVMF was never found on NixOS.** `$(dir $(QEMU))share` is the *bin*
directory plus "share", so it looked in /usr/bin/share. The absolute paths
after it covered that up on distributions that keep firmware in /usr; on one
that keeps nothing there it meant the firmware was undiscoverable on a machine
that had it, and the only way to build was to copy the file in by hand. It
resolves the prefix properly now and looks in <prefix>/share/qemu too.

Verified by exit status, from a clean tree and a no-op: both 0. ldls-check
passes 27. The guest boots, serves /srv to the host and answers 403 to
%2e%2e/etc/passwd.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VQ7Qx8E1q1M1bbCq8QLHS2
This commit is contained in:
Nexoniarz 2026-09-14 19:45:22 +02:00
commit 4a032c9c62
3 changed files with 113 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

39
Makefile
View file

@ -527,9 +527,18 @@ OVMF_VARS = $(BUILD)/OVMF_VARS.fd
# rather than assume, and say something useful when nothing is found -- an
# unhelpful "cp: no such file" is a poor way to learn that a package is
# missing.
QEMU_SHARE = $(dir $(shell command -v $(QEMU) 2>/dev/null))share
OVMF_CODE_CANDIDATES = $(QEMU_SHARE)/edk2-x86_64-code.fd /usr/share/edk2/ovmf/OVMF_CODE.fd /usr/share/OVMF/OVMF_CODE.fd /usr/share/ovmf/x64/OVMF_CODE.fd /usr/share/qemu/edk2-x86_64-code.fd
OVMF_VARS_CANDIDATES = $(QEMU_SHARE)/edk2-i386-vars.fd /usr/share/edk2/ovmf/OVMF_VARS.fd /usr/share/OVMF/OVMF_VARS.fd /usr/share/ovmf/x64/OVMF_VARS.fd /usr/share/qemu/edk2-i386-vars.fd
#
# `$(dir ...)` of the binary is its *bin* directory, so the share directory is
# one level above it -- `$(dir)/share` asked for `/usr/bin/share`, which exists
# nowhere. On the distributions below that went unnoticed, because the absolute
# paths after it matched anyway; on one that keeps nothing in /usr it meant the
# firmware was never found on a machine that had it, and the only way to build
# was to copy the file in by hand.
QEMU_BIN = $(shell command -v $(QEMU) 2>/dev/null)
QEMU_PREFIX = $(patsubst %/bin/,%,$(dir $(QEMU_BIN)))
QEMU_SHARE = $(QEMU_PREFIX)/share
OVMF_CODE_CANDIDATES = $(QEMU_SHARE)/qemu/edk2-x86_64-code.fd $(QEMU_SHARE)/edk2-x86_64-code.fd /usr/share/edk2/ovmf/OVMF_CODE.fd /usr/share/OVMF/OVMF_CODE.fd /usr/share/ovmf/x64/OVMF_CODE.fd /usr/share/qemu/edk2-x86_64-code.fd
OVMF_VARS_CANDIDATES = $(QEMU_SHARE)/qemu/edk2-i386-vars.fd $(QEMU_SHARE)/edk2-i386-vars.fd /usr/share/edk2/ovmf/OVMF_VARS.fd /usr/share/OVMF/OVMF_VARS.fd /usr/share/ovmf/x64/OVMF_VARS.fd /usr/share/qemu/edk2-i386-vars.fd
# Tested through the shell rather than with $(wildcard). On Windows, make is a
# native binary using Win32 file calls while `command -v` hands back an MSYS
@ -1125,16 +1134,28 @@ $(BUILD)/sysroot.stamp: sysroot-install
# 4096. The size is load-bearing: tools/windows/bootcheck.ps1 asserts on
# READ CAPACITY returning 32768 sectors of 512 bytes, which is this.
#
# --exclude bin, because this image exists to prove the USB driver can read
# a block and for no other reason. mkfs writes each hard link as its own
# copy, so once toybox's 236 applets became links to one binary the /bin
# tree expanded to more than 60 MB on the way in -- 236 copies of something
# nothing here will ever execute.
# The exclusions matter more than they look, because the size above cannot
# move to accommodate content. This image exists to prove the USB driver
# can read a block and for no other reason, so what is in it is irrelevant
# and only its total size is not.
#
# --exclude bin: mkfs writes each hard link as its own copy, so once
# toybox's applets became links to one binary the /bin tree expanded to
# more than 60 MB on the way in -- hundreds of copies of something
# nothing here will ever execute.
# --exclude lib: `make ldls-stage` puts a kernel module and its BTF here,
# and a vmlinux BTF is 5 MB on its own. That filled this image and broke
# `make` for anyone who had staged a module -- a build failure a long way
# from its cause. It also kept somebody else's GPL module out of a test
# artifact, which RAMDISK_EXCLUDE does for the ramdisk and nothing was
# doing for this.
# --exclude usr: 7 MB of fonts, for a test that reads one block.
USBSTICK = $(BUILD)/usbstick.img
$(USBSTICK): $(MKFS) $(BUILD)/sysroot.stamp
@mkdir -p $(BUILD)
@rm -f $@
$(MKFS) $@ --size 4096 --exclude bin --sysroot $(SYSROOT)
$(MKFS) $@ --size 4096 --exclude bin --exclude lib --exclude usr \
--sysroot $(SYSROOT)
usbstick: $(USBSTICK)

65
lib/glue.c
View file

@ -16,6 +16,71 @@ exit(void)
/* _start lives in lib/start.S — it must read incoming registers, which
* cannot be expressed portably in C. See that file. */
// The four the compiler is allowed to call without being asked.
//
// A freestanding build gets no libc, but it does not get a compiler that
// promises never to *call* one: the C standard lets an implementation emit
// calls to memcpy, memmove, memset and memcmp for things that look nothing
// like a function call. A struct assignment is the common one --
//
// *out = in; /* struct pciuser_info, 96 bytes */
//
// -- which gcc unrolled into stores and clang turned into `call memcpy`, so
// ldlsd linked for months and then did not, on a machine where nothing about
// it had changed except which compiler ran. The fix is not to avoid struct
// assignment; it is to provide the four functions, because the compiler is
// within its rights either way.
//
// Deliberately simple. These are not the hot path of anything here, and a
// clever one risks being turned back into a call to itself.
void *
memcpy(void *dst, const void *src, unsigned long n)
{
unsigned char *d = dst;
const unsigned char *s = src;
while (n--) *d++ = *s++;
return dst;
}
void *
memmove(void *dst, const void *src, unsigned long n)
{
unsigned char *d = dst;
const unsigned char *s = src;
if (d == s || n == 0)
return dst;
// Backwards when they overlap the wrong way, which is the whole reason
// this is not memcpy.
if (d < s) {
while (n--) *d++ = *s++;
} else {
d += n; s += n;
while (n--) *--d = *--s;
}
return dst;
}
void *
memset(void *dst, int c, unsigned long n)
{
unsigned char *d = dst;
while (n--) *d++ = (unsigned char)c;
return dst;
}
int
memcmp(const void *a, const void *b, unsigned long n)
{
const unsigned char *p = a, *q = b;
while (n--) {
if (*p != *q)
return *p - *q;
p++; q++;
}
return 0;
}
void
__stack_chk_fail(void)
{

22
ports/toybox/Makefile
View file

@ -18,9 +18,20 @@ include ../mk/port.mk
# applet replaced the shared inode and every applet became ping; installing
# the applet over first-party ping put it back and lost the real one.
#
# SchOS's ping talks to its own lwIP stack and knows there is no resolver yet.
# toybox's would be the wrong answer even if it worked.
TOYBOX_SKIP = ping
# SchOS's ping talks to its own lwIP stack. toybox's would be the wrong answer
# even if it worked.
#
# The list is generated rather than remembered, because remembering it failed:
# `ping` was found the hard way, and then `httpd` was found the same hard way
# months later -- /bin/httpd ran, printed toybox's "Unknown option 'p'" and
# nobody could see why the web server had no -p flag. Anything in cmd/ that
# toybox also provides belongs here, and asking the directory is the only
# version of that which stays true when cmd/ grows.
#
# `chvt` is in the list for the same reason and was never noticed: toybox's
# wants Linux's VT ioctls, ours talks to SchOS's own terminal layer, and which
# one you got depended on the order two install rules happened to run in.
TOYBOX_SKIP = $(notdir $(wildcard $(TOPDIR)/cmd/*))
TOYBOX_VERSION := 0.8.12
TOYBOX_URL := http://landley.net/toybox/downloads/toybox-$(TOYBOX_VERSION).tar.gz
@ -130,7 +141,10 @@ $(BUILDDIR)/toybox.installed: $(BUILDDIR)/toybox | $(SYSROOT)/bin
ln $(SYSROOT)/bin/toybox $(SYSROOT)/bin/"$$a" 2>/dev/null \
|| cp $< $(SYSROOT)/bin/"$$a"; \
done
@echo "toybox: installed $$($(BUILDDIR)/toybox | wc -w) applets"
@echo "toybox: installed $$($(BUILDDIR)/toybox | wc -w) applets, minus" \
"$$(for a in $$($(BUILDDIR)/toybox); do \
case " $(TOYBOX_SKIP) " in *" $$a "*) echo $$a ;; esac; \
done | tr '\n' ' ')(first-party)"
@touch $@