#!/bin/bash # SPDX-License-Identifier: GPL-2.0+ # # (C) Copyright 2014 Suriyan Ramasami # Invoke this test script from U-Boot base directory as ./test/fs/fs-test.sh # It currently tests the fs/sb and native commands for ext4 and fat partitions # Expected results are as follows: # EXT4 tests: # fs-test.sb.ext4.out: Summary: PASS: 24 FAIL: 0 # fs-test.ext4.out: Summary: PASS: 24 FAIL: 0 # fs-test.fs.ext4.out: Summary: PASS: 24 FAIL: 0 # FAT16 tests: # fs-test.sb.fat16.out: Summary: PASS: 24 FAIL: 0 # fs-test.fat16.out: Summary: PASS: 21 FAIL: 3 # fs-test.fs.fat16.out: Summary: PASS: 21 FAIL: 3 # FAT32 tests: # fs-test.sb.fat32.out: Summary: PASS: 24 FAIL: 0 # fs-test.fat32.out: Summary: PASS: 21 FAIL: 3 # fs-test.fs.fat32.out: Summary: PASS: 21 FAIL: 3 # Total Summary: TOTAL PASS: 204 TOTAL FAIL: 12 # pre-requisite binaries list. PREREQ_BINS="md5sum mkfs mount umount dd fallocate mkdir" # All generated output files from this test will be in $OUT_DIR # Hence everything is sandboxed. OUT_DIR="sandbox/test/fs" # Location of generated sandbox u-boot UBOOT="./sandbox/u-boot" # Our mount directory will be in the sandbox MOUNT_DIR="${OUT_DIR}/mnt" # The file system image we create will have the $IMG prefix. IMG="${OUT_DIR}/3GB" # $SMALL_FILE is the name of the 1MB file in the file system image SMALL_FILE="1MB.file" # $BIG_FILE is the name of the 2.5GB file in the file system image BIG_FILE="2.5GB.file" # $MD5_FILE will have the expected md5s when we do the test # They shall have a suffix which represents their file system (ext4/fat16/...) MD5_FILE="${OUT_DIR}/md5s.list" # $OUT shall be the prefix of the test output. Their suffix will be .out OUT="${OUT_DIR}/fs-test" # Full Path of the 1 MB file that shall be created in the fs image. MB1="${MOUNT_DIR}/${SMALL_FILE}" GB2p5="${MOUNT_DIR}/${BIG_FILE}" # ************************ # * Functions start here * # ************************ # Check if the prereq binaries exist, or exit function check_prereq() { for prereq in $PREREQ_BINS; do if [ ! -x "`which $prereq`" ]; then echo "Missing $prereq binary. Exiting!" exit fi done # We use /dev/urandom to create files. Check if it exists. if [ ! -c /dev/urandom ]; then echo "Missing character special /dev/urandom. Exiting!" exit fi } # If 1st param is "clean", then clean out the generated files and exit function check_clean() { if [ "$1" = "clean" ]; then rm -rf "$OUT_DIR" echo "Cleaned up generated files. Exiting" exit fi } # Generate sandbox U-Boot - gleaned from /test/dm/test-dm.sh function compile_sandbox() { unset CROSS_COMPILE NUM_CPUS=$(cat /proc/cpuinfo |grep -c processor) make O=sandbox sandbox_config make O=sandbox -s -j${NUM_CPUS} # Check if U-Boot exists if [ ! -x "$UBOOT" ]; then echo "$UBOOT does not exist or is not executable" echo "Build error?" echo "Please run this script as ./test/fs/`basename $0`" exit fi } # Clean out all generated files other than the file system images # We save time by not deleting and recreating the file system images function prepare_env() { rm -f ${MD5_FILE}.* ${OUT}.* mkdir -p ${OUT_DIR} } # 1st parameter is the name of the image file to be created # 2nd parameter is the filesystem - fat16 ext4 etc # -F cant be used with fat as it means something else. function create_image() { # Create image if not already present - saves time, while debugging case "$2" in fat16) MKFS_OPTION="-F 16" FS_TYPE="fat" ;; fat32) MKFS_OPTION="-F 32" FS_TYPE="fat" ;; ext4) MKFS_OPTION="-F" FS_TYPE="ext4" ;; esac if [ ! -f "$1" ]; then fallocate -l 3G "$1" &> /dev/null if [ $? -ne 0 ]; then echo fallocate failed - using dd instead dd if=/dev/zero of=$1 bs=1024 count=$((3 * 1024 * 1024)) if [ $? -ne 0 ]; then echo Could not create empty disk image exit $? fi fi mkfs -t "$FS_TYPE" $MKFS_OPTION "$1" &> /dev/null if [ $? -ne 0 -a "$FS_TYPE" = "fat" ]; then # If we fail and we did fat, try vfat. mkfs -t vfat $MKFS_OPTION "$1" &> /dev/null fi if [ $? -ne 0 ]; then echo Could not create filesystem exit $? fi fi } # 1st parameter is image file # 2nd parameter is file system type - fat16/ext4/... # 3rd parameter is name of small file # 4th parameter is name of big file # 5th parameter is fs/nonfs/sb - to dictate generic fs commands or # otherwise or sb hostfs # 6th parameter is the directory path for the files. Its "" for generic # fs and ext4/fat and full patch for sb hostfs # UBOOT is set in env function test_image() { addr="0x01000008" length="0x00100000" case "$2" in fat*) FPATH="" PREFIX="fat" WRITE="write" ;; ext4) # ext4 needs absolute path FPATH="/" PREFIX="ext4" WRITE="write" ;; *) echo "Unhandled filesystem $2. Exiting!" exit ;; esac case "$5" in fs) PREFIX="" WRITE="save" SUFFIX=" 0:0" ;; nonfs) SUFFIX=" 0:0" ;; sb) PREFIX="sb " WRITE="save" SUFFIX="fs -" ;; *) echo "Unhandled mode $5. Exiting!" exit ;; esac # sb always uses full path to mointpoint, irrespective of filesystem if [ "$5" = "sb" ]; then FPATH=${6}/ fi FILE_WRITE=${3}.w FILE_SMALL=$3 FILE_BIG=$4 # In u-boot commands, <interface> stands for host or hostfs # hostfs maps to the host fs. # host maps to the "sb bind" that we do $UBOOT << EOF sb=$5 setenv bind 'if test "\$sb" != sb; then sb bind 0 "$1"; fi' run bind # Test Case 1 - ls ${PREFIX}ls host${SUFFIX} $6 # In addition, test with a nonexistent directory to see if we crash. ${PREFIX}ls host${SUFFIX} invalid_d # # We want ${PREFIX}size host 0:0 $3 for host commands and # sb size hostfs - $3 for hostfs commands. # 1MB is 0x0010 0000 # Test Case 2a - size of small file ${PREFIX}size host${SUFFIX} ${FPATH}$FILE_SMALL printenv filesize setenv filesize # Test Case 2b - size of small file via a path using '..' ${PREFIX}size host${SUFFIX} ${FPATH}SUBDIR/../$FILE_SMALL printenv filesize setenv filesize # 2.5GB (1024*1024*2500) is 0x9C40 0000 # Test Case 3 - size of big file ${PREFIX}size host${SUFFIX} ${FPATH}$FILE_BIG printenv filesize setenv filesize # Notes about load operation # If I use 0x01000000 I get DMA misaligned error message # Last two parameters are size and offset. # Test Case 4a - Read full 1MB of small file ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_SMALL printenv filesize # Test Case 4b - Read full 1MB of small file md5sum $addr \$filesize setenv filesize # Test Case 5a - First 1MB of big file ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_BIG $length 0x0 printenv filesize # Test Case 5b - First 1MB of big file md5sum $addr \$filesize setenv filesize # fails for ext as no offset support # Test Case 6a - Last 1MB of big file ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_BIG $length 0x9C300000 printenv filesize # Test Case 6b - Last 1MB of big file md5sum $addr \$filesize setenv filesize # fails for ext as no offset support # Test Case 7a - One from the last 1MB chunk of 2GB ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_BIG $length 0x7FF00000 printenv filesize # Test Case 7b - One from the last 1MB chunk of 2GB md5sum $addr \$filesize setenv filesize # fails for ext as no offset support # Test Case 8a - One from the start 1MB chunk from 2GB ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_BIG $length 0x80000000 printenv filesize # Test Case 8b - One from the start 1MB chunk from 2GB md5sum $addr \$filesize setenv filesize # fails for ext as no offset support # Test Case 9a - One 1MB chunk crossing the 2GB boundary ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_BIG $length 0x7FF80000 printenv filesize # Test Case 9b - One 1MB chunk crossing the 2GB boundary md5sum $addr \$filesize setenv filesize # Generic failure case # Test Case 10 - 2MB chunk from the last 1MB of big file ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_BIG 0x00200000 0x9C300000 printenv filesize # # Read 1MB from small file ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_SMALL # Write it back to test the writes # Test Case 11a - Check that the write succeeded ${PREFIX}${WRITE} host${SUFFIX} $addr ${FPATH}$FILE_WRITE \$filesize mw.b $addr 00 100 ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_WRITE # Test Case 11b - Check md5 of written to is same as the one read from md5sum $addr \$filesize setenv filesize # # Next test case checks writing a file whose dirent # is the first in the block, which is always true for "." # The write should fail, but the lookup should work # Test Case 12 - Check directory traversal ${PREFIX}${WRITE} host${SUFFIX} $addr ${FPATH}. 0x10 # Read 1MB from small file ${PREFIX}load host${SUFFIX} $addr ${FPATH}$FILE_SMALL # Write it via "same directory", i.e. "." dirent # Test Case 13a - Check directory traversal ${PREFIX}${WRITE} host${SUFFIX} $addr ${FPATH}./${FILE_WRITE}2 \$filesize mw.b $addr 00 100 ${PREFIX}load host${SUFFIX} $addr ${FPATH}./${FILE_WRITE}2 # Test Case 13b - Check md5 of written to is same as the one read from md5sum $addr \$filesize setenv filesize mw.b $addr 00 100 ${PREFIX}load host${SUFFIX} $addr ${FPATH}${FILE_WRITE}2 # Test Case 13c - Check md5 of written to is same as the one read from md5sum $addr \$filesize setenv filesize # reset EOF } # 1st argument is the name of the image file. # 2nd argument is the file where we generate the md5s of the files # generated with the appropriate start and length that we use to test. # It creates the necessary files in the image to test. # $GB2p5 is the path of the big file (2.5 GB) # $MB1 is the path of the small file (1 MB) # $MOUNT_DIR is the path we can use to mount the image file. function create_files() { # Mount the image so we can populate it. mkdir -p "$MOUNT_DIR" sudo mount -o loop,rw "$1" "$MOUNT_DIR" # Create a subdirectory. sudo mkdir -p "$MOUNT_DIR/SUBDIR" # Create big file in this image. # Note that we work only on the start 1MB, couple MBs in the 2GB range # and the last 1 MB of the huge 2.5GB file. # So, just put random values only in those areas. if [ ! -f "${GB2p5}" ]; then sudo dd if=/dev/urandom of="${GB2p5}" bs=1M count=1 \ &> /dev/null sudo dd if=/dev/urandom of="${GB2p5}" bs=1M count=2 seek=2047 \ &> /dev/null sudo dd if=/dev/urandom of="${GB2p5}" bs=1M count=1 seek=2499 \ &> /dev/null fi # Create a small file in this image. if [ ! -f "${MB1}" ]; then sudo dd if=/dev/urandom of="${MB1}" bs=1M count=1 \ &> /dev/null fi # Delete the small file copies which possibly are written as part of a # previous test. sudo rm -f "${MB1}.w" sudo rm -f "${MB1}.w2" # Generate the md5sums of reads that we will test against small file dd if="${MB1}" bs=1M skip=0 count=1 2> /dev/null | md5sum > "$2" # Generate the md5sums of reads that we will test against big file # One from beginning of file. dd if="${GB2p5}" bs=1M skip=0 count=1 \ 2> /dev/null | md5sum >> "$2" # One from end of file. dd if="${GB2p5}" bs=1M skip=2499 count=1 \ 2> /dev/null | md5sum >> "$2" # One from the last 1MB chunk of 2GB dd if="${GB2p5}" bs=1M skip=2047 count=1 \ 2> /dev/null | md5sum >> "$2" # One from the start 1MB chunk from 2GB dd if="${GB2p5}" bs=1M skip=2048 count=1 \ 2> /dev/null | md5sum >> "$2" # One 1MB chunk crossing the 2GB boundary dd if="${GB2p5}" bs=512K skip=4095 count=2 \ 2> /dev/null | md5sum >> "$2" sync sudo umount "$MOUNT_DIR" rmdir "$MOUNT_DIR" } # 1st parameter is the text to print # if $? is 0 its a pass, else a fail # As a side effect it shall update env variable PASS and FAIL function pass_fail() { if [ $? -eq 0 ]; then echo pass - "$1" PASS=$((PASS + 1)) else echo FAIL - "$1" FAIL=$((FAIL + 1)) fi } # 1st parameter is the string which leads to an md5 generation # 2nd parameter is the file we grep, for that string # 3rd parameter is the name of the file which has md5s in it # 4th parameter is the line # in the md5 file that we match it against # This function checks if the md5 of the file in the sandbox matches # that calculated while generating the file # 5th parameter is the string to print with the result check_md5() { # md5sum in u-boot has output of form: # md5 for 01000008 ... 01100007 ==> <md5> # the 7th field is the actual md5 md5_src=`grep -A2 "$1" "$2" | grep "md5 for" | tr -d '\r'` md5_src=($md5_src) md5_src=${md5_src[6]} # The md5 list, each line is of the form: # - <md5> # the 2nd field is the actual md5 md5_dst=`sed -n $4p $3` md5_dst=($md5_dst) md5_dst=${md5_dst[0]} # For a pass they should match. [ "$md5_src" = "$md5_dst" ] pass_fail "$5" } # 1st parameter is the name of the output file to check # 2nd parameter is the name of the file containing the md5 expected # 3rd parameter is the name of the small file # 4th parameter is the name of the big file # 5th paramter is the name of the written file # This function checks the output file for correct results. function check_results() { echo "** Start $1" PASS=0 FAIL=0 # Check if the ls is showing correct results for 2.5 gb file grep -A7 "Test Case 1 " "$1" | egrep -iq "2621440000 *$4" pass_fail "TC1: ls of $4" # Check if the ls is showing correct results for 1 mb file grep -A7 "Test Case 1 " "$1" | egrep -iq "1048576 *$3" pass_fail "TC1: ls of $3" # Check size command on 1MB.file egrep -A3 "Test Case 2a " "$1" | grep -q "filesize=100000" pass_fail "TC2: size of $3" # Check size command on 1MB.file via a path using '..' egrep -A3 "Test Case 2b " "$1" | grep -q "filesize=100000" pass_fail "TC2: size of $3 via a path using '..'" # Check size command on 2.5GB.file egrep -A3 "Test Case 3 " "$1" | grep -q "filesize=9c400000" pass_fail "TC3: size of $4" # Check read full mb of 1MB.file grep -A4 "Test Case 4a " "$1" | grep -q "filesize=100000" pass_fail "TC4: load of $3 size" check_md5 "Test Case 4b " "$1" "$2" 1 "TC4: load from $3" # Check first mb of 2.5GB.file grep -A4 "Test Case 5a " "$1" | grep -q "filesize=100000" pass_fail "TC5: load of 1st MB from $4 size" check_md5 "Test Case 5b " "$1" "$2" 2 "TC5: load of 1st MB from $4" # Check last mb of 2.5GB.file grep -A4 "Test Case 6a " "$1" | grep -q "filesize=100000" pass_fail "TC6: load of last MB from $4 size" check_md5 "Test Case 6b " "$1" "$2" 3 "TC6: load of last MB from $4" # Check last 1mb chunk of 2gb from 2.5GB file grep -A4 "Test Case 7a " "$1" | grep -q "filesize=100000" pass_fail "TC7: load of last 1mb chunk of 2GB from $4 size" check_md5 "Test Case 7b " "$1" "$2" 4 \ "TC7: load of last 1mb chunk of 2GB from $4" # Check first 1mb chunk after 2gb from 2.5GB file grep -A4 "Test Case 8a " "$1" | grep -q "filesize=100000" pass_fail "TC8: load 1st MB chunk after 2GB from $4 size" check_md5 "Test Case 8b " "$1" "$2" 5 \ "TC8: load 1st MB chunk after 2GB from $4" # Check 1mb chunk crossing the 2gb boundary from 2.5GB file grep -A4 "Test Case 9a " "$1" | grep -q "filesize=100000" pass_fail "TC9: load 1MB chunk crossing 2GB boundary from $4 size" check_md5 "Test Case 9b " "$1" "$2" 6 \ "TC9: load 1MB chunk crossing 2GB boundary from $4" # Check 2mb chunk from the last 1MB of 2.5GB file loads 1MB grep -A5 "Test Case 10 " "$1" | grep -q "filesize=100000" pass_fail "TC10: load 2MB from the last 1MB of $4 loads 1MB" # Check 1mb chunk write grep -A2 "Test Case 11a " "$1" | grep -q '1048576 bytes written' pass_fail "TC11: 1MB write to $3.w - write succeeded" check_md5 "Test Case 11b " "$1" "$2" 1 \ "TC11: 1MB write to $3.w - content verified" # Check lookup of 'dot' directory grep -A4 "Test Case 12 " "$1" | grep -q 'Unable to write file' pass_fail "TC12: 1MB write to . - write denied" # Check directory traversal grep -A2 "Test Case 13a " "$1" | grep -q '1048576 bytes written' pass_fail "TC13: 1MB write to ./$3.w2 - write succeeded" check_md5 "Test Case 13b " "$1" "$2" 1 \ "TC13: 1MB read from ./$3.w2 - content verified" check_md5 "Test Case 13c " "$1" "$2" 1 \ "TC13: 1MB read from $3.w2 - content verified" echo "** End $1" } # Takes in one parameter which is "fs" or "nonfs", which then dictates # if a fs test (size/load/save) or a nonfs test (fatread/extread) needs to # be performed. function test_fs_nonfs() { echo "Creating files in $fs image if not already present." create_files $IMAGE $MD5_FILE_FS OUT_FILE="${OUT}.$1.${fs}.out" test_image $IMAGE $fs $SMALL_FILE $BIG_FILE $1 "" \ > ${OUT_FILE} 2>&1 # strip out noise from fs code grep -v -e "File System is consistent\|update journal finished" \ -e "reading .*\.file\|writing .*\.file.w" \ < ${OUT_FILE} > ${OUT_FILE}_clean check_results ${OUT_FILE}_clean $MD5_FILE_FS $SMALL_FILE \ $BIG_FILE TOTAL_FAIL=$((TOTAL_FAIL + FAIL)) TOTAL_PASS=$((TOTAL_PASS + PASS)) echo "Summary: PASS: $PASS FAIL: $FAIL" echo "--------------------------------------------" } # ******************** # * End of functions * # ******************** check_clean "$1" check_prereq compile_sandbox prepare_env # Track TOTAL_FAIL and TOTAL_PASS TOTAL_FAIL=0 TOTAL_PASS=0 # In each loop, for a given file system image, we test both the # fs command, like load/size/write, the file system specific command # like: ext4load/ext4size/ext4write and the sb load/ls/save commands. for fs in ext4 fat16 fat32; do echo "Creating $fs image if not already present." IMAGE=${IMG}.${fs}.img MD5_FILE_FS="${MD5_FILE}.${fs}" create_image $IMAGE $fs # sb commands test echo "Creating files in $fs image if not already present." create_files $IMAGE $MD5_FILE_FS # Lets mount the image and test sb hostfs commands mkdir -p "$MOUNT_DIR" case "$fs" in fat*) uid="uid=`id -u`" ;; *) uid="" ;; esac sudo mount -o loop,rw,$uid "$IMAGE" "$MOUNT_DIR" sudo chmod 777 "$MOUNT_DIR" OUT_FILE="${OUT}.sb.${fs}.out" test_image $IMAGE $fs $SMALL_FILE $BIG_FILE sb `pwd`/$MOUNT_DIR \ > ${OUT_FILE} 2>&1 sudo umount "$MOUNT_DIR" rmdir "$MOUNT_DIR" check_results $OUT_FILE $MD5_FILE_FS $SMALL_FILE $BIG_FILE TOTAL_FAIL=$((TOTAL_FAIL + FAIL)) TOTAL_PASS=$((TOTAL_PASS + PASS)) echo "Summary: PASS: $PASS FAIL: $FAIL" echo "--------------------------------------------" test_fs_nonfs nonfs test_fs_nonfs fs done echo "Total Summary: TOTAL PASS: $TOTAL_PASS TOTAL FAIL: $TOTAL_FAIL" echo "--------------------------------------------" if [ $TOTAL_FAIL -eq 0 ]; then echo "PASSED" exit 0 else echo "FAILED" exit 1 fi