#!/bin/bash
# SPDX-License-Identifier: GPL-2.0
# Copyright (c) 2021 Facebook.  All Rights Reserved.
#
# FS QA Test 250
#
# Test the new /sys/fs/btrfs/<uuid>/allocation/<block-type>/chunk_size
# setting. This setting allows the admin to change the chunk size
# setting for the next allocation.
#
# Test 1:
#   Allocate storage for all three block types (data, metadata and system)
#   with the default chunk size.
#
# Test 2:
#   Set a new chunk size to double the default size and allocate space
#   for all new block types with the new chunk size.
#
# Test 3:
#   Pick an allocation size that is used in a loop and make sure the last
#   allocation cannot be partially fullfilled.
#
# Note: Variable naming uses the following convention: if a variable name
#       ends in "_B" then its a byte value, if it ends in "_MB" then the
#       value is in megabytes.
#
. ./common/preamble
_begin_fstest auto

seq=`basename $0`
seqres="${RESULT_DIR}/${seq}"

# Parse a size string which is in the format "XX.XXMib".
#
# Parameters:
#   - (IN)    Block group type (Data, Metadata, System)
#   - (INOUT) Variable to store block group size in MB
#
parse_size_string() {
	local SIZE=$(echo "$1" | $AWK_PROG 'match($0, /([0-9.]+)/) { print substr($0, RSTART, RLENGTH) }')
	eval $2="${SIZE%.*}"
}

# Determine the size of the device in MB.
#
# Parameters:
#   - (INOUT) Variable to store device size in MB
#
device_size() {
	btrfs fi show ${SCRATCH_MNT} --mbytes | grep devid >> $seqres.full
	local SIZE=$($BTRFS_UTIL_PROG filesystem show ${SCRATCH_MNT} --mbytes | grep devid)
	parse_size_string $(echo "${SIZE}" | $AWK_PROG '{print $4}') SIZE_ALLOC
	eval $1=${SIZE_ALLOC%.*}
}

# Determine the free space of a block group in MB.
#
# Parameters:
#   - (INOUT) Variable to store free space in MB
#
free_space() {
	local SIZE=$($BTRFS_UTIL_PROG filesystem show ${SCRATCH_MNT} --mbytes | grep devid)
	parse_size_string $(echo "${SIZE}" | $AWK_PROG '{print $4}') SIZE_ALLOC
	parse_size_string $(echo "${SIZE}" | $AWK_PROG '{print $6}') SIZE_USED
	eval $1=$(expr ${SIZE_ALLOC} - ${SIZE_USED})
}

# Determine how much space in MB has been allocated to a block group.
#
# Parameters:
#   - (IN)    Block group type (Data, Metadata, System)
#   - (INOUT) Variable to store block group size in MB
#
alloc_size() {
	local SIZE_STRING=$($BTRFS_UTIL_PROG filesystem df ${SCRATCH_MNT} -m | grep  "$1" | $AWK_PROG '{print $3}')
	parse_size_string ${SIZE_STRING} BLOCK_GROUP_SIZE
	eval $2="${BLOCK_GROUP_SIZE}"
}

. ./common/filter
_supported_fs btrfs
_require_test
_require_scratch
# The chunk size on zoned mode is fixed to the zone size
_require_non_zoned_device "$SCRATCH_DEV"

# Delete log file if it exists.
rm -f "${seqres}.full"

# Make filesystem. 10GB is needed to test different chunk sizes for
# metadata and data and the default size for volumes > 5GB is different.
_scratch_mkfs_sized $((10 * 1024 * 1024 * 1024)) >> $seqres.full 2>&1
_scratch_mount >> $seqres.full 2>&1

# Check if there is sufficient sysfs support.
_require_fs_sysfs allocation/metadata/chunk_size
_require_fs_sysfs allocation/metadata/force_chunk_alloc

# Get free space.
free_space  FREE_SPACE_MB
device_size DEVICE_SIZE_MB

echo "free space = ${FREE_SPACE_MB}MB" >> ${seqres}.full

# If device is a symbolic link, get block device.
SCRATCH_BDEV=$(_real_dev $SCRATCH_DEV)

# Get chunk sizes.
echo "Capture default chunk sizes."

# The sysfs interface has only exported chunk_size (10G by default),
# but no stripe_size (1G by default) exported.
#
# Btrfs calculate the real chunk to be allocated using both limits,
# Thus the default 10G chunk size can only be fulfilled by something
# like 10 disk RAID0
#
# The proper solution should be exporting the stripe_size limit too,
# but that may take several cycles, here we just use hard coded 1G
# as the expected chunk size.
FIRST_DATA_CHUNK_SIZE_B=$((1024 * 1024 * 1024))

# For metadata, we are safe to use the exported value, as the default
# metadata chunk size limit is already smaller than its stripe size.
FIRST_METADATA_CHUNK_SIZE_B=$(_get_fs_sysfs_attr ${SCRATCH_BDEV} allocation/metadata/chunk_size)

echo "Orig Data chunk size    = ${FIRST_DATA_CHUNK_SIZE_B}"     >> ${seqres}.full
echo "Orig Metaata chunk size = ${FIRST_METADATA_CHUNK_SIZE_B}" >> ${seqres}.full

INIT_ALLOC_SIZE_MB=$(expr \( ${FIRST_DATA_CHUNK_SIZE_B} + ${FIRST_METADATA_CHUNK_SIZE_B} \) / 1024 / 1024)
echo "Allocation size for initial allocation = ${INIT_ALLOC_SIZE_MB}MB" >> $seqres.full

#
# Do first allocation with the default chunk sizes for the different block
# types.
#
echo "First allocation."
alloc_size "Data"     DATA_SIZE_START_MB
alloc_size "Metadata" METADATA_SIZE_START_MB

echo "Block group Data alloc size     = ${DATA_SIZE_START_MB}MB"     >> $seqres.full
echo "Block group Metadata alloc size = ${METADATA_SIZE_START_MB}MB" >> $seqres.full

_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/data/force_chunk_alloc 1
_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/metadata/force_chunk_alloc 1

alloc_size "Data"     FIRST_DATA_SIZE_MB
alloc_size "Metadata" FIRST_METADATA_SIZE_MB

echo "First block group Data alloc size     = ${FIRST_DATA_SIZE_MB}MB"     >> ${seqres}.full
echo "First block group Metadata alloc size = ${FIRST_METADATA_SIZE_MB}MB" >> ${seqres}.full

free_space FREE_SPACE_AFTER
echo "Free space after first allocation = ${FREE_SPACE_AFTER}MB" >> ${seqres}.full

#
# Do allocation with the doubled chunk sizes for the different block types.
#
echo "Second allocation."
SECOND_DATA_CHUNK_SIZE_B=$(expr ${FIRST_DATA_CHUNK_SIZE_B} \* 2)
SECOND_METADATA_CHUNK_SIZE_B=$(expr ${FIRST_METADATA_CHUNK_SIZE_B} \* 2)

echo "Second block group Data calc alloc size     = ${SECOND_DATA_CHUNK_SIZE_B}"     >> $seqres.full
echo "Second block group Metadata calc alloc size = ${SECOND_METADATA_CHUNK_SIZE_B}" >> $seqres.full

# Stripe size is limited to 10% of device size.
if [[ ${SECOND_DATA_CHUNK_SIZE_B} -gt $(expr ${DEVICE_SIZE_MB} \* 10 \* 1024 \* 1024 / 100) ]]; then
	SECOND_DATA_CHUNK_SIZE_B=$(expr ${DEVICE_SIZE_MB} \* 10 / 100 \* 1024 \* 1024)
fi
if [[ ${SECOND_METADATA_CHUNK_SIZE_B} -gt $(expr ${DEVICE_SIZE_MB} \* 10 \* 1024 \* 1024 / 100) ]]; then
	SECOND_METADATA_CHUNK_SIZE_B=$(expr ${DEVICE_SIZE_MB} \* 10 / 100 \* 1024 \* 1024)
fi

echo "Second block group Data alloc size     = ${SECOND_DATA_CHUNK_SIZE_B}"     >> $seqres.full
echo "Second block group Metadata alloc size = ${SECOND_METADATA_CHUNK_SIZE_B}" >> $seqres.full

_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/data/chunk_size ${SECOND_DATA_CHUNK_SIZE_B}
_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/metadata/chunk_size ${SECOND_METADATA_CHUNK_SIZE_B}

SECOND_DATA_CHUNK_SIZE_READ_B=$(_get_fs_sysfs_attr ${SCRATCH_BDEV} allocation/data/chunk_size)
SECOND_METADATA_CHUNK_SIZE_READ_B=$(_get_fs_sysfs_attr ${SCRATCH_BDEV} allocation/metadata/chunk_size)

_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/data/force_chunk_alloc 1
echo "Allocated data chunk" >> $seqres.full
_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/metadata/force_chunk_alloc 1
echo "Allocated metadata chunk" >> $seqres.full

alloc_size "Data"     SECOND_DATA_SIZE_MB
alloc_size "Metadata" SECOND_METADATA_SIZE_MB
alloc_size "System"   SECOND_SYSTEM_SIZE_MB

echo "Calculate request size so last memory allocation cannot be completely fullfilled."
free_space FREE_SPACE_MB

# Find request size whose space allocation cannot be completely fullfilled.
THIRD_DATA_CHUNK_SIZE_MB=$(expr 256)
until [ ${THIRD_DATA_CHUNK_SIZE_MB} -gt $(expr 7 \* 1024) ]; do
	if [ $((FREE_SPACE_MB%THIRD_DATA_CHUNK_SIZE_MB)) -ne 0 ]; then
		break
	fi
	THIRD_DATA_CHUNK_SIZE_MB=$((THIRD_DATA_CHUNK_SIZE_MB+256))
done

if [[ ${THIRD_DATA_CHUNK_SIZE_MB} -eq $(expr 7 \* 1024) ]]; then
	_fail "Cannot find allocation size for partial block allocation."
fi

THIRD_DATA_CHUNK_SIZE_B=$(expr ${THIRD_DATA_CHUNK_SIZE_MB} \* 1024 \* 1024)
echo "Allocation size in mb    = ${THIRD_DATA_CHUNK_SIZE_MB}" >> ${seqres}.full
echo "Allocation size in bytes = ${THIRD_DATA_CHUNK_SIZE_B}"  >> ${seqres}.full

#
# Do allocation until free space is exhausted.
#
echo "Third allocation."
_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/data/chunk_size ${THIRD_DATA_CHUNK_SIZE_B}

free_space FREE_SPACE_MB
while [ $FREE_SPACE_MB -gt $THIRD_DATA_CHUNK_SIZE_MB ]; do
	alloc_size "Data" THIRD_DATA_SIZE_MB
	_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/data/force_chunk_alloc 1

	free_space FREE_SPACE_MB
done

alloc_size "Data" FOURTH_DATA_SIZE_MB

#
# Force chunk allocation of system block type must fail.
#
echo "Force allocation of system block type must fail."
_set_fs_sysfs_attr ${SCRATCH_BDEV} allocation/system/force_chunk_alloc 1 2>/dev/null

#
# Verification of initial allocation.
#
echo "Verify first allocation."
FIRST_DATA_CHUNK_SIZE_MB=$(expr ${FIRST_DATA_CHUNK_SIZE_B} / 1024 / 1024)
FIRST_METADATA_CHUNK_SIZE_MB=$(expr ${FIRST_METADATA_CHUNK_SIZE_B} / 1024 / 1024)

# if [[ $(expr ${FIRST_DATA_CHUNK_SIZE_MB} + ${DATA_SIZE_START_MB}) -ne $(expr ${FIRST_DATA_SIZE_MB}) ]]; then
if [[ $(expr ${FIRST_DATA_CHUNK_SIZE_MB} + ${DATA_SIZE_START_MB}) -ne ${FIRST_DATA_SIZE_MB} ]]; then
	echo "Initial data allocation not correct."
fi

if [[ $(expr ${FIRST_METADATA_CHUNK_SIZE_MB} + ${METADATA_SIZE_START_MB}) -ne ${FIRST_METADATA_SIZE_MB} ]]; then
	echo "Initial metadata allocation not correct."
fi

#
# Verification of second allocation.
#
echo "Verify second allocation."
SECOND_DATA_CHUNK_SIZE_MB=$(expr ${SECOND_DATA_CHUNK_SIZE_B} / 1024 / 1024)
SECOND_METADATA_CHUNK_SIZE_MB=$(expr ${SECOND_METADATA_CHUNK_SIZE_B} / 1024 / 1024)

if [[ ${SECOND_DATA_CHUNK_SIZE_B} -ne ${SECOND_DATA_CHUNK_SIZE_READ_B} ]]; then
	echo "Value written to allocation/data/chunk_size and read value are different."
fi

if [[ ${SECOND_METADATA_CHUNK_SIZE_B} -ne ${SECOND_METADATA_CHUNK_SIZE_READ_B} ]]; then
	echo "Value written to allocation/metadata/chunk_size and read value are different."
fi

if [[ $(expr ${SECOND_DATA_CHUNK_SIZE_MB} + ${FIRST_DATA_SIZE_MB}) -ne ${SECOND_DATA_SIZE_MB} ]]; then
	echo "Data allocation after chunk size change not correct."
fi

if [[ $(expr ${SECOND_METADATA_CHUNK_SIZE_MB} + ${FIRST_METADATA_SIZE_MB}) -ne ${SECOND_METADATA_SIZE_MB} ]]; then
	echo "Metadata allocation after chunk size change not correct."
fi

#
# Verification of third allocation.
#
echo "Verify third allocation."
if [[ ${FREE_SPACE_MB} -gt ${THIRD_DATA_CHUNK_SIZE_MB} ]]; then
	echo "Free space after allocating over memlimit is too high."
fi

# The + 1 is required as 1MB is always kept as free space.
if [[ $(expr ${THIRD_DATA_CHUNK_SIZE_MB} + ${THIRD_DATA_SIZE_MB} + 1) -le $(expr ${FOURTH_DATA_SIZE_MB}) ]]; then
	echo "Allocation until out of memory: last memory allocation size is not correct."
fi

status=0
exit