Files
serai/orchestration/increase_default_stack_size.sh

186 lines
5.1 KiB
Bash
Raw Normal View History

#!/bin/sh
# Raises `PT_GNU_STACK`'s memory to be at least 8 MB.
#
# This causes `musl` to use a 8 MB default for new threads, resolving the primary
# compatibility issue faced when executing a program on a `musl` system.
#
# See https://wiki.musl-libc.org/functional-differences-from-glibc.html#Thread-stack-size
# for reference. This differs that instead of setting at time of link, it
# patches the binary as an already-linked ELF executable.
set -e
ELF="$1"
if [ ! -f "$ELF" ]; then
echo "\`increase_default_stack_size.sh\` [ELF binary]"
echo ""
echo "Sets the \`PT_GNU_STACK\` program header to its existing value or 8 MB,"
echo "whichever is greater."
exit 1
fi
hex() {
od -tx1 -v -A none | tr -d "[:space:]"
}
read_bytes() {
dd bs=1 skip="$1" count="$2" if="$ELF" 2> /dev/null | hex
}
Add a CI for `increase_default_stack_size.sh` This runs whenever the script is modified, or weekly to ensure the CI doesn't inadvertently decay (due to using the latest packages for a variety of shells). This runs with `sh` (presumably `dash`), `ksh`, `bash`, `dash` (explicitly), `zsh`, `ash` (Busybox), `hush` (Busybox), `mksh`, `yash`, and `brush`. While none of these guarantee this script is POSIX-compliant, as a fully and explicitly-only POSIX-compliant environment is not constructed, this does reasonably test the script itself to be POSIX-compliant. The tools called have been reviewed for being used to the POSIX standard (although not audited to that degree). The script itself is modified with the following changes for compliance with POSIX: 1) `hexdump` is replaced with `od` (`od` suggested by @PlasmaPower) 2) `printf \xFF` replaced with octal escapes, as `\x` is not part of POSIX 3) `head -c` is replaced with `cut`, as the `-c` option is not standardized under POSIX (despite it being present for `tail`). This was identified by @PlasmaPower. As we used `head -c-2` to truncate the last two characters of a string, we now use `wc -c` for a `strlen` to enable the necessary arithmetic to calculate what two bytes in from the end of the string is. This entire effort can be argued pointless, as we could simply run `monerod` on Debian. This script is useful, the journey down the rabbithole of POSIX compliance fascinating, and the methodology applicable to other potential futures though (whether running binaries on Alpine or testing other `sh` scripts for their portability). As part of this effort overall, our CI was extended with `shellcheck` for all `sh` scripts in-tree, including all of our existing `sh` scripts. That there is an actual, direct benefit past this specific effort.
2025-12-09 00:57:26 -05:00
hex_to_octal() {
printf "ibase=16; obase=8; %s\n" "$1" | bc
}
write_bytes() {
POS=$1
BYTES=$2
while [ ! "$BYTES" = "" ]; do
Add a CI for `increase_default_stack_size.sh` This runs whenever the script is modified, or weekly to ensure the CI doesn't inadvertently decay (due to using the latest packages for a variety of shells). This runs with `sh` (presumably `dash`), `ksh`, `bash`, `dash` (explicitly), `zsh`, `ash` (Busybox), `hush` (Busybox), `mksh`, `yash`, and `brush`. While none of these guarantee this script is POSIX-compliant, as a fully and explicitly-only POSIX-compliant environment is not constructed, this does reasonably test the script itself to be POSIX-compliant. The tools called have been reviewed for being used to the POSIX standard (although not audited to that degree). The script itself is modified with the following changes for compliance with POSIX: 1) `hexdump` is replaced with `od` (`od` suggested by @PlasmaPower) 2) `printf \xFF` replaced with octal escapes, as `\x` is not part of POSIX 3) `head -c` is replaced with `cut`, as the `-c` option is not standardized under POSIX (despite it being present for `tail`). This was identified by @PlasmaPower. As we used `head -c-2` to truncate the last two characters of a string, we now use `wc -c` for a `strlen` to enable the necessary arithmetic to calculate what two bytes in from the end of the string is. This entire effort can be argued pointless, as we could simply run `monerod` on Debian. This script is useful, the journey down the rabbithole of POSIX compliance fascinating, and the methodology applicable to other potential futures though (whether running binaries on Alpine or testing other `sh` scripts for their portability). As part of this effort overall, our CI was extended with `shellcheck` for all `sh` scripts in-tree, including all of our existing `sh` scripts. That there is an actual, direct benefit past this specific effort.
2025-12-09 00:57:26 -05:00
NEXT=$(printf "%s" "$BYTES" | cut -c-2)
# Advance to the third byte, as in, after the first two bytes
BYTES=$(printf "%s" "$BYTES" | tail -c+3)
Add a CI for `increase_default_stack_size.sh` This runs whenever the script is modified, or weekly to ensure the CI doesn't inadvertently decay (due to using the latest packages for a variety of shells). This runs with `sh` (presumably `dash`), `ksh`, `bash`, `dash` (explicitly), `zsh`, `ash` (Busybox), `hush` (Busybox), `mksh`, `yash`, and `brush`. While none of these guarantee this script is POSIX-compliant, as a fully and explicitly-only POSIX-compliant environment is not constructed, this does reasonably test the script itself to be POSIX-compliant. The tools called have been reviewed for being used to the POSIX standard (although not audited to that degree). The script itself is modified with the following changes for compliance with POSIX: 1) `hexdump` is replaced with `od` (`od` suggested by @PlasmaPower) 2) `printf \xFF` replaced with octal escapes, as `\x` is not part of POSIX 3) `head -c` is replaced with `cut`, as the `-c` option is not standardized under POSIX (despite it being present for `tail`). This was identified by @PlasmaPower. As we used `head -c-2` to truncate the last two characters of a string, we now use `wc -c` for a `strlen` to enable the necessary arithmetic to calculate what two bytes in from the end of the string is. This entire effort can be argued pointless, as we could simply run `monerod` on Debian. This script is useful, the journey down the rabbithole of POSIX compliance fascinating, and the methodology applicable to other potential futures though (whether running binaries on Alpine or testing other `sh` scripts for their portability). As part of this effort overall, our CI was extended with `shellcheck` for all `sh` scripts in-tree, including all of our existing `sh` scripts. That there is an actual, direct benefit past this specific effort.
2025-12-09 00:57:26 -05:00
NEXT=$(hex_to_octal "$NEXT")
# shellcheck disable=SC2059
Add a CI for `increase_default_stack_size.sh` This runs whenever the script is modified, or weekly to ensure the CI doesn't inadvertently decay (due to using the latest packages for a variety of shells). This runs with `sh` (presumably `dash`), `ksh`, `bash`, `dash` (explicitly), `zsh`, `ash` (Busybox), `hush` (Busybox), `mksh`, `yash`, and `brush`. While none of these guarantee this script is POSIX-compliant, as a fully and explicitly-only POSIX-compliant environment is not constructed, this does reasonably test the script itself to be POSIX-compliant. The tools called have been reviewed for being used to the POSIX standard (although not audited to that degree). The script itself is modified with the following changes for compliance with POSIX: 1) `hexdump` is replaced with `od` (`od` suggested by @PlasmaPower) 2) `printf \xFF` replaced with octal escapes, as `\x` is not part of POSIX 3) `head -c` is replaced with `cut`, as the `-c` option is not standardized under POSIX (despite it being present for `tail`). This was identified by @PlasmaPower. As we used `head -c-2` to truncate the last two characters of a string, we now use `wc -c` for a `strlen` to enable the necessary arithmetic to calculate what two bytes in from the end of the string is. This entire effort can be argued pointless, as we could simply run `monerod` on Debian. This script is useful, the journey down the rabbithole of POSIX compliance fascinating, and the methodology applicable to other potential futures though (whether running binaries on Alpine or testing other `sh` scripts for their portability). As part of this effort overall, our CI was extended with `shellcheck` for all `sh` scripts in-tree, including all of our existing `sh` scripts. That there is an actual, direct benefit past this specific effort.
2025-12-09 00:57:26 -05:00
printf \\"$NEXT" | dd conv=notrunc bs=1 seek="$POS" of="$ELF" 2> /dev/null
POS=$((POS + 1))
done
}
# Magic
MAGIC=$(read_bytes 0 4)
Add a CI for `increase_default_stack_size.sh` This runs whenever the script is modified, or weekly to ensure the CI doesn't inadvertently decay (due to using the latest packages for a variety of shells). This runs with `sh` (presumably `dash`), `ksh`, `bash`, `dash` (explicitly), `zsh`, `ash` (Busybox), `hush` (Busybox), `mksh`, `yash`, and `brush`. While none of these guarantee this script is POSIX-compliant, as a fully and explicitly-only POSIX-compliant environment is not constructed, this does reasonably test the script itself to be POSIX-compliant. The tools called have been reviewed for being used to the POSIX standard (although not audited to that degree). The script itself is modified with the following changes for compliance with POSIX: 1) `hexdump` is replaced with `od` (`od` suggested by @PlasmaPower) 2) `printf \xFF` replaced with octal escapes, as `\x` is not part of POSIX 3) `head -c` is replaced with `cut`, as the `-c` option is not standardized under POSIX (despite it being present for `tail`). This was identified by @PlasmaPower. As we used `head -c-2` to truncate the last two characters of a string, we now use `wc -c` for a `strlen` to enable the necessary arithmetic to calculate what two bytes in from the end of the string is. This entire effort can be argued pointless, as we could simply run `monerod` on Debian. This script is useful, the journey down the rabbithole of POSIX compliance fascinating, and the methodology applicable to other potential futures though (whether running binaries on Alpine or testing other `sh` scripts for their portability). As part of this effort overall, our CI was extended with `shellcheck` for all `sh` scripts in-tree, including all of our existing `sh` scripts. That there is an actual, direct benefit past this specific effort.
2025-12-09 00:57:26 -05:00
# shellcheck disable=SC2059
EXPECTED_MAGIC=$(printf \\"$(hex_to_octal 7F)"ELF | hex)
if [ ! "$MAGIC" = "$EXPECTED_MAGIC" ]; then
echo "Not ELF"
exit 2
fi
# 1 if 32-bit, 2 if 64-bit
BITS=$(read_bytes 4 1)
case $BITS in
"01") BITS=32;;
"02") BITS=64;;
*)
echo "Not 32- or 64- bit"
exit 3
;;
esac
# For `value_per_bits a b`, `a` if 32-bit and `b` if 64-bit
value_per_bits() {
RESULT=$(($1))
if [ $BITS = 64 ]; then
RESULT=$(($2))
fi
printf "%s" "$RESULT"
}
# Read an integer by its offset, differing depending on if 32- or 64-bit
read_integer_by_offset() {
OFFSET=$(value_per_bits "$1" "$2")
BYTES=$(read_bytes "$OFFSET" "$3")
BYTES=$(swap_native_endian "$BYTES")
printf "%i" $(( 0x$BYTES ))
}
# 1 if little-endian, 2 if big-endian
LITTLE_ENDIAN=$(read_bytes 5 1)
case $LITTLE_ENDIAN in
"01") LITTLE_ENDIAN=1;;
"02") LITTLE_ENDIAN=0;;
*)
echo "Not little- or big- endian"
exit 4
;;
esac
# While this script is written in big-endian, we need to work with the file in
# its declared endian. This function swaps from big to native, or vice versa,
# as necessary.
swap_native_endian() {
BYTES="$1"
if [ "$BYTES" = "" ]; then
read -r BYTES
fi
if [ "$LITTLE_ENDIAN" -eq 0 ]; then
printf "%s" "$BYTES"
return
fi
Add a CI for `increase_default_stack_size.sh` This runs whenever the script is modified, or weekly to ensure the CI doesn't inadvertently decay (due to using the latest packages for a variety of shells). This runs with `sh` (presumably `dash`), `ksh`, `bash`, `dash` (explicitly), `zsh`, `ash` (Busybox), `hush` (Busybox), `mksh`, `yash`, and `brush`. While none of these guarantee this script is POSIX-compliant, as a fully and explicitly-only POSIX-compliant environment is not constructed, this does reasonably test the script itself to be POSIX-compliant. The tools called have been reviewed for being used to the POSIX standard (although not audited to that degree). The script itself is modified with the following changes for compliance with POSIX: 1) `hexdump` is replaced with `od` (`od` suggested by @PlasmaPower) 2) `printf \xFF` replaced with octal escapes, as `\x` is not part of POSIX 3) `head -c` is replaced with `cut`, as the `-c` option is not standardized under POSIX (despite it being present for `tail`). This was identified by @PlasmaPower. As we used `head -c-2` to truncate the last two characters of a string, we now use `wc -c` for a `strlen` to enable the necessary arithmetic to calculate what two bytes in from the end of the string is. This entire effort can be argued pointless, as we could simply run `monerod` on Debian. This script is useful, the journey down the rabbithole of POSIX compliance fascinating, and the methodology applicable to other potential futures though (whether running binaries on Alpine or testing other `sh` scripts for their portability). As part of this effort overall, our CI was extended with `shellcheck` for all `sh` scripts in-tree, including all of our existing `sh` scripts. That there is an actual, direct benefit past this specific effort.
2025-12-09 00:57:26 -05:00
while :; do
printf "%s" "$BYTES" | tail -c2
Add a CI for `increase_default_stack_size.sh` This runs whenever the script is modified, or weekly to ensure the CI doesn't inadvertently decay (due to using the latest packages for a variety of shells). This runs with `sh` (presumably `dash`), `ksh`, `bash`, `dash` (explicitly), `zsh`, `ash` (Busybox), `hush` (Busybox), `mksh`, `yash`, and `brush`. While none of these guarantee this script is POSIX-compliant, as a fully and explicitly-only POSIX-compliant environment is not constructed, this does reasonably test the script itself to be POSIX-compliant. The tools called have been reviewed for being used to the POSIX standard (although not audited to that degree). The script itself is modified with the following changes for compliance with POSIX: 1) `hexdump` is replaced with `od` (`od` suggested by @PlasmaPower) 2) `printf \xFF` replaced with octal escapes, as `\x` is not part of POSIX 3) `head -c` is replaced with `cut`, as the `-c` option is not standardized under POSIX (despite it being present for `tail`). This was identified by @PlasmaPower. As we used `head -c-2` to truncate the last two characters of a string, we now use `wc -c` for a `strlen` to enable the necessary arithmetic to calculate what two bytes in from the end of the string is. This entire effort can be argued pointless, as we could simply run `monerod` on Debian. This script is useful, the journey down the rabbithole of POSIX compliance fascinating, and the methodology applicable to other potential futures though (whether running binaries on Alpine or testing other `sh` scripts for their portability). As part of this effort overall, our CI was extended with `shellcheck` for all `sh` scripts in-tree, including all of our existing `sh` scripts. That there is an actual, direct benefit past this specific effort.
2025-12-09 00:57:26 -05:00
NEW_LENGTH=$(( $(printf "%s" "$BYTES" | wc -c) - 2 ))
if [ "$NEW_LENGTH" -eq 0 ]; then
break
fi
BYTES=$(printf "%s" "$BYTES" | cut -c-$NEW_LENGTH)
done
}
ELF_VERSION=$(read_bytes 6 1)
if [ ! "$ELF_VERSION" = "01" ]; then
echo "Unknown ELF Version ($ELF_VERSION)"
exit 5
fi
ELF_VERSION_2=$(read_bytes $((0x14)) 4)
if [ ! "$ELF_VERSION_2" = "$(swap_native_endian 00000001)" ]; then
echo "Unknown secondary ELF Version ($ELF_VERSION_2)"
exit 6
fi
# Find where the program headers are
PROGRAM_HEADERS_OFFSET=$(read_integer_by_offset 0x1c 0x20 "$(value_per_bits 4 8)")
PROGRAM_HEADER_SIZE=$(value_per_bits 0x20 0x38)
DECLARED_PROGRAM_HEADER_SIZE=$(read_integer_by_offset 0x2a 0x36 2)
if [ ! "$PROGRAM_HEADER_SIZE" -eq "$DECLARED_PROGRAM_HEADER_SIZE" ]; then
echo "Unexpected size of a program header ($DECLARED_PROGRAM_HEADER_SIZE)"
exit 7
fi
program_header_start() {
printf "%i" $((PROGRAM_HEADERS_OFFSET + ($1 * PROGRAM_HEADER_SIZE)))
}
read_program_header() {
START=$(program_header_start "$1")
read_bytes "$START" "$PROGRAM_HEADER_SIZE"
}
# Iterate over each program header
PROGRAM_HEADERS=$(read_integer_by_offset 0x2c 0x38 2)
NEXT_PROGRAM_HEADER=$(( PROGRAM_HEADERS - 1 ))
FOUND=0
while [ "$NEXT_PROGRAM_HEADER" -ne -1 ]; do
THIS_PROGRAM_HEADER=$NEXT_PROGRAM_HEADER
NEXT_PROGRAM_HEADER=$(( NEXT_PROGRAM_HEADER - 1 ))
PROGRAM_HEADER=$(read_program_header "$THIS_PROGRAM_HEADER")
Add a CI for `increase_default_stack_size.sh` This runs whenever the script is modified, or weekly to ensure the CI doesn't inadvertently decay (due to using the latest packages for a variety of shells). This runs with `sh` (presumably `dash`), `ksh`, `bash`, `dash` (explicitly), `zsh`, `ash` (Busybox), `hush` (Busybox), `mksh`, `yash`, and `brush`. While none of these guarantee this script is POSIX-compliant, as a fully and explicitly-only POSIX-compliant environment is not constructed, this does reasonably test the script itself to be POSIX-compliant. The tools called have been reviewed for being used to the POSIX standard (although not audited to that degree). The script itself is modified with the following changes for compliance with POSIX: 1) `hexdump` is replaced with `od` (`od` suggested by @PlasmaPower) 2) `printf \xFF` replaced with octal escapes, as `\x` is not part of POSIX 3) `head -c` is replaced with `cut`, as the `-c` option is not standardized under POSIX (despite it being present for `tail`). This was identified by @PlasmaPower. As we used `head -c-2` to truncate the last two characters of a string, we now use `wc -c` for a `strlen` to enable the necessary arithmetic to calculate what two bytes in from the end of the string is. This entire effort can be argued pointless, as we could simply run `monerod` on Debian. This script is useful, the journey down the rabbithole of POSIX compliance fascinating, and the methodology applicable to other potential futures though (whether running binaries on Alpine or testing other `sh` scripts for their portability). As part of this effort overall, our CI was extended with `shellcheck` for all `sh` scripts in-tree, including all of our existing `sh` scripts. That there is an actual, direct benefit past this specific effort.
2025-12-09 00:57:26 -05:00
HEADER_TYPE=$(printf "%s" "$PROGRAM_HEADER" | cut -c-8)
HEADER_TYPE=$(swap_native_endian "$HEADER_TYPE")
# `PT_GNU_STACK`
# https://github.com/torvalds/linux/blob/c2f2b01b74be8b40a2173372bcd770723f87e7b2/include/uapi/linux/elf.h#L41
if [ ! "$HEADER_TYPE" = "6474e551" ]; then
continue
fi
FOUND=1
MEMSZ_OFFSET=$(( $(program_header_start "$THIS_PROGRAM_HEADER") + $(value_per_bits 0x14 0x28) ))
MEMSZ_LEN=$(value_per_bits 4 8)
# `MEMSZ_OFFSET MEMSZ_OFFSET` as we've already derived it depending on the amount of bits
MEMSZ=$(read_integer_by_offset "$MEMSZ_OFFSET" "$MEMSZ_OFFSET" "$MEMSZ_LEN")
DESIRED_STACK_SIZE=$((8 * 1024 * 1024))
# Only run if the inherent value is _smaller_
if [ "$MEMSZ" -lt "$DESIRED_STACK_SIZE" ]; then
# `2 *`, as this is its length in hexadecimal
HEX_MEMSZ=$(printf %."$((2 * MEMSZ_LEN))"x "$DESIRED_STACK_SIZE")
HEX_MEMSZ=$(swap_native_endian "$HEX_MEMSZ")
write_bytes "$MEMSZ_OFFSET" "$HEX_MEMSZ"
fi
done
if [ "$FOUND" -eq 0 ]; then
echo "\`PT_GNU_STACK\` program header not found"
exit 8
fi
echo "All instances of \`PT_GNU_STACK\` patched to be at least 8 MB"
exit 0