Skip to contentSkip to conversation

Add test 248: Check filesystem FITRIM implementation

PATCH Superseded Lukas Czerner 3 participants 7 messages Dec 10, 2010 – Dec 21, 2010

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Add test 248: Check filesystem FITRIM implementation

FITRIM ioctl is used on a mounted filesystem to discard (or "trim") blocks which are not in use by the filesystem. This is useful for solid-state drives (SSDs) and thinly-provi-sioned storage. This test helps to verify filesystem FITRIM implementation to assure that it does not corrupts data.

This test creates checksums of all files in /usr/share/doc directory and run several processes which clear its working directory on SCRATCH_MNT, then copy everything from /usr/share/doc into its working directory, create list of files in working directory and its checksums and compare it with the original list of checksums. Every process works in the loop so it repeat remove->copy->check, while fstrim tool is running simultaneously.

Fstrim is just a helper tool which uses FITRIM ioctl to actually do the filesystem discard.

I found this very useful because when the FITRIM is really buggy (thus data-destroying) the 248 test will notice, because checksums will most likely change.

---
 248          |  184 +++++++++++++++++++++++++++++++++++++++++
 248.out      |    3 +
 group        |    1 +
 src/Makefile |    2 +-
 src/fstrim.c |  257 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 446 insertions(+), 1 deletions(-)
 create mode 100755 248
 create mode 100644 248.out
 create mode 100644 src/fstrim.c
248 new file +184 -0
new file mode 100755
index 0000000..a6cfce2
@@ -0,0 +1,184 @@
+#!/bin/bash
+# FS QA Test No. 248
+#
+# This test was created in order to verify filesystem FITRIM implementation.
+# By many concurrent copy and remove operations and checking that files
+# does not change after copied into SCRATCH_MNT test if FITRIM implementation
+# corrupts the filesystem (data/metadata).
+#
+#-----------------------------------------------------------------------
+# Copyright 2010 (C) Red Hat, Inc., Lukas Czerner <[email protected]>
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation.
+#
+# This program is distributed in the hope that it would be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write the Free Software Foundation,
+# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+#-----------------------------------------------------------------------
+
[email protected]
+
+seq=`basename $0`
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=`mktemp -d`
+status=1    # failure is the default!
+trap "_cleanup; exit \$status" 0 1 3
+trap "_destroy; exit \$status" 2 15
+chpid=0
+mypid=$$
+
+# get standard environment, filters and checks
+. ./common.rc
+. ./common.filter
+
+# real QA test starts here
+_supported_fs generic
+_supported_os Linux
+_require_scratch
+_scratch_mkfs >/dev/null 2>&1
+_scratch_mount
+
+_cleanup()
+{
+	rm -rf $tmp
+}
+
+_destroy()
+{
+	kill $pids $fstrim_pid
+	wait $pids $fstrim_pid
+	rm -rf $tmp
+}
+
+_destroy_fstrim()
+{
+	kill $fpid
+	wait $fpid
+}
+
+_fail()
+{
+	echo "$1"
+	kill $mypid
+}
+
+##
+# Background FSTRIM loop. We are trimming the device in the loop and for
+# better test coverage, we are doing whole device trim followed by several
+# smaller trims.
+##
+fstrim_loop()
+{
+	trap "_destroy_fstrim; exit \$status" 2 15
+	fsize=$(df | grep $SCRATCH_MNT | grep $SCRATCH_DEV  | awk '{print $2}')
+
+	while true ; do
+		step=1048576
+		start=0
+		$here/src/fstrim $SCRATCH_MNT &
+		fpid=$!
+		wait $fpid
+		while [ $start -lt $fsize ] ; do
+			$here/src/fstrim -s ${start}k -l ${step}k $SCRATCH_MNT &
+			fpid=$!
+
+			##
+			# All the waiting is done because Bash is incredibly
+			# stupid. As you know, fstrim_loop is run at background
+			# and when the test is over, or when it is killed (with
+			# ^C), because of trap, it tries to kill fstrim_loop.
+			# However, it does not kill currently running commands,
+			# so fstrim might be still running making it impossible
+			# to umount the SCRATCH_MNT and hence resulting in
+			# error.
+			##
+			wait $fpid
+			start=$(( $start + $step ))
+		done
+	done
+}
+
+function check_sums() {
+	dir=$1
+
+	(
+	cd $SCRATCH_MNT/$p
+	find -P . -xdev -type f -print0 | xargs -0 md5sum | sort -o $tmp/stress.$$.$p
+	)
+
+	diff $tmp/content.sums $tmp/stress.$$.$p
+	if [ $? -ne 0 ]; then
+		_fail "!!!Checksums has changed - Filesystem possibly corrupted!!!\n"
+	fi
+	rm -f $tmp/stress.$$.$p
+}
+
+function run_process() {
+	local p=$1
+	repeat=10
+	trap "kill $chpid; wait $chpid" 2 15
+
+	sleep $((10*$p))s &
+	export chpid=$! && wait $chpid &> /dev/null
+	chpid=0
+
+	while [ $repeat -gt 0 ]; do
+
+		# Remove old directories.
+		rm -rf $SCRATCH_MNT/$p
+		export chpid=$! && wait $chpid &> /dev/null
+
+		# Copy content -> partition.
+		mkdir $SCRATCH_MNT/$p
+		cp -ax $content/* $SCRATCH_MNT/$p
+		export chpid=$! && wait $chpid &> /dev/null
+
+		check_sums
+		repeat=$(( $repeat - 1 ))
+	done
+}
+
+nproc=20
+content=/usr/share/doc
+
+# Check for FITRIM support
+echo -n "Checking FITRIM support: "
+$here/src/fstrim -l 10M $SCRATCH_MNT
+[ $? -ne 0 ] && exit
+echo "done."
+
+mkdir -p $tmp
+
+(
+cd $content
+find -P . -xdev -type f -print0 | xargs -0 md5sum | sort -o $tmp/content.sums
+)
+
+echo -n "Running the test: "
+pids=""
+fstrim_loop &
+fstrim_pid=$!
+p=1
+while [ $p -le $nproc ]; do
+	run_process $p &
+	pids="$pids $!"
+	p=$(($p+1))
+done
+echo "done."
+
+wait $pids
+kill $fstrim_pid
+wait $fstrim_pid
+
+status=0
+
+exit
248.out new file +3 -0
new file mode 100644
index 0000000..880d9c7
@@ -0,0 +1,3 @@
+QA output created by 248
+Checking FITRIM support: done.
+Running the test: done.
group +1 -0
index 0f94dd9..19eec07 100644
@@ -361,3 +361,4 @@ deprecated
 245 auto quick dir
 246 auto quick rw
 247 auto quick rw
+248 ioctl trim
src/Makefile +1 -1
index b827bd0..885fd65 100644
@@ -17,7 +17,7 @@ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \
 	preallo_rw_pattern_writer ftrunc trunc fs_perms testx looptest \
 	locktest unwritten_mmap bulkstat_unlink_test t_stripealign \
 	bulkstat_unlink_test_modified t_dir_offset t_futimens t_immutable \
-	stale_handle
+	stale_handle fstrim
 
 SUBDIRS =
 
src/fstrim.c new file +257 -0

257 lines not on this page · raw message

Signature · 6 lines
1.7.2.3

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#1 lore raw
Quoted 13 lines · Lukas Czerner

FITRIM ioctl is used on a mounted filesystem to discard (or "trim") blocks which are not in use by the filesystem. This is useful for solid-state drives (SSDs) and thinly-provi-sioned storage. This test helps to verify filesystem FITRIM implementation to assure that it does not corrupts data.

This test creates checksums of all files in /usr/share/doc directory and run several processes which clear its working directory on SCRATCH_MNT, then copy everything from /usr/share/doc into its working directory, create list of files in working directory and its checksums and compare it with the original list of checksums. Every process works in the loop so it repeat remove->copy->check, while fstrim tool is running simultaneously.

Quoted 6 lines

Fstrim is just a helper tool which uses FITRIM ioctl to actually do the filesystem discard.

I found this very useful because when the FITRIM is really buggy (thus data-destroying) the 248 test will notice, because checksums will most likely change.

Common guys, how many times I need to repost this ? :) Or do anyone have objections ?

-Lukas

Quoted 498 lines
---
 248          |  184 +++++++++++++++++++++++++++++++++++++++++
 248.out      |    3 +
 group        |    1 +
 src/Makefile |    2 +-
 src/fstrim.c |  257 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 5 files changed, 446 insertions(+), 1 deletions(-)
 create mode 100755 248
 create mode 100644 248.out
 create mode 100644 src/fstrim.c
248 new file +184 -0
new file mode 100755
index 0000000..a6cfce2
@@ -0,0 +1,184 @@
+#!/bin/bash
+# FS QA Test No. 248
+#
+# This test was created in order to verify filesystem FITRIM implementation.
+# By many concurrent copy and remove operations and checking that files
+# does not change after copied into SCRATCH_MNT test if FITRIM implementation
+# corrupts the filesystem (data/metadata).
+#
+#-----------------------------------------------------------------------
+# Copyright 2010 (C) Red Hat, Inc., Lukas Czerner <[email protected]>
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License as
+# published by the Free Software Foundation.
+#
+# This program is distributed in the hope that it would be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write the Free Software Foundation,
+# Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+#-----------------------------------------------------------------------
+
[email protected]
+
+seq=`basename $0`
+echo "QA output created by $seq"
+
+here=`pwd`
+tmp=`mktemp -d`
+status=1    # failure is the default!
+trap "_cleanup; exit \$status" 0 1 3
+trap "_destroy; exit \$status" 2 15
+chpid=0
+mypid=$$
+
+# get standard environment, filters and checks
+. ./common.rc
+. ./common.filter
+
+# real QA test starts here
+_supported_fs generic
+_supported_os Linux
+_require_scratch
+_scratch_mkfs >/dev/null 2>&1
+_scratch_mount
+
+_cleanup()
+{
+	rm -rf $tmp
+}
+
+_destroy()
+{
+	kill $pids $fstrim_pid
+	wait $pids $fstrim_pid
+	rm -rf $tmp
+}
+
+_destroy_fstrim()
+{
+	kill $fpid
+	wait $fpid
+}
+
+_fail()
+{
+	echo "$1"
+	kill $mypid
+}
+
+##
+# Background FSTRIM loop. We are trimming the device in the loop and for
+# better test coverage, we are doing whole device trim followed by several
+# smaller trims.
+##
+fstrim_loop()
+{
+	trap "_destroy_fstrim; exit \$status" 2 15
+	fsize=$(df | grep $SCRATCH_MNT | grep $SCRATCH_DEV  | awk '{print $2}')
+
+	while true ; do
+		step=1048576
+		start=0
+		$here/src/fstrim $SCRATCH_MNT &
+		fpid=$!
+		wait $fpid
+		while [ $start -lt $fsize ] ; do
+			$here/src/fstrim -s ${start}k -l ${step}k $SCRATCH_MNT &
+			fpid=$!
+
+			##
+			# All the waiting is done because Bash is incredibly
+			# stupid. As you know, fstrim_loop is run at background
+			# and when the test is over, or when it is killed (with
+			# ^C), because of trap, it tries to kill fstrim_loop.
+			# However, it does not kill currently running commands,
+			# so fstrim might be still running making it impossible
+			# to umount the SCRATCH_MNT and hence resulting in
+			# error.
+			##
+			wait $fpid
+			start=$(( $start + $step ))
+		done
+	done
+}
+
+function check_sums() {
+	dir=$1
+
+	(
+	cd $SCRATCH_MNT/$p
+	find -P . -xdev -type f -print0 | xargs -0 md5sum | sort -o $tmp/stress.$$.$p
+	)
+
+	diff $tmp/content.sums $tmp/stress.$$.$p
+	if [ $? -ne 0 ]; then
+		_fail "!!!Checksums has changed - Filesystem possibly corrupted!!!\n"
+	fi
+	rm -f $tmp/stress.$$.$p
+}
+
+function run_process() {
+	local p=$1
+	repeat=10
+	trap "kill $chpid; wait $chpid" 2 15
+
+	sleep $((10*$p))s &
+	export chpid=$! && wait $chpid &> /dev/null
+	chpid=0
+
+	while [ $repeat -gt 0 ]; do
+
+		# Remove old directories.
+		rm -rf $SCRATCH_MNT/$p
+		export chpid=$! && wait $chpid &> /dev/null
+
+		# Copy content -> partition.
+		mkdir $SCRATCH_MNT/$p
+		cp -ax $content/* $SCRATCH_MNT/$p
+		export chpid=$! && wait $chpid &> /dev/null
+
+		check_sums
+		repeat=$(( $repeat - 1 ))
+	done
+}
+
+nproc=20
+content=/usr/share/doc
+
+# Check for FITRIM support
+echo -n "Checking FITRIM support: "
+$here/src/fstrim -l 10M $SCRATCH_MNT
+[ $? -ne 0 ] && exit
+echo "done."
+
+mkdir -p $tmp
+
+(
+cd $content
+find -P . -xdev -type f -print0 | xargs -0 md5sum | sort -o $tmp/content.sums
+)
+
+echo -n "Running the test: "
+pids=""
+fstrim_loop &
+fstrim_pid=$!
+p=1
+while [ $p -le $nproc ]; do
+	run_process $p &
+	pids="$pids $!"
+	p=$(($p+1))
+done
+echo "done."
+
+wait $pids
+kill $fstrim_pid
+wait $fstrim_pid
+
+status=0
+
+exit
248.out new file +3 -0
new file mode 100644
index 0000000..880d9c7
@@ -0,0 +1,3 @@
+QA output created by 248
+Checking FITRIM support: done.
+Running the test: done.
group +1 -0
index 0f94dd9..19eec07 100644
@@ -361,3 +361,4 @@ deprecated
 245 auto quick dir
 246 auto quick rw
 247 auto quick rw
+248 ioctl trim
src/Makefile +1 -1
index b827bd0..885fd65 100644
@@ -17,7 +17,7 @@ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \
 	preallo_rw_pattern_writer ftrunc trunc fs_perms testx looptest \
 	locktest unwritten_mmap bulkstat_unlink_test t_stripealign \
 	bulkstat_unlink_test_modified t_dir_offset t_futimens t_immutable \
-	stale_handle
+	stale_handle fstrim
 
 SUBDIRS =
 
src/fstrim.c new file +257 -0
new file mode 100644
index 0000000..45ad841
@@ -0,0 +1,257 @@
+/*
+ * fstrim.c -- discard the part (or whole) of mounted filesystem.
+ *
+ * Copyright (C) 2009 Red Hat, Inc., Lukas Czerner <[email protected]>
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ *
+ * This program uses FITRIM ioctl to discard parts or the whole filesystem
+ * online (mounted). You can specify range (start and lenght) to be
+ * discarded, or simply discard while filesystem.
+ *
+ * Usage: fstrim [options] <mount point>
+ */
+
+#include <string.h>
+#include <unistd.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <fcntl.h>
+#include <limits.h>
+#include <stdarg.h>
+#include <getopt.h>
+
+#include <sys/ioctl.h>
+#include <sys/stat.h>
+#include <linux/fs.h>
+
+#ifndef FITRIM
+struct fstrim_range {
+	uint64_t start;
+	uint64_t len;
+	uint64_t minlen;
+};
+#define FITRIM		_IOWR('X', 121, struct fstrim_range)
+#endif
+
+const char *program_name = "fstrim";
+
+struct options {
+	struct fstrim_range *range;
+	char mpoint[PATH_MAX];
+	char verbose;
+};
+
+static void usage(void)
+{
+	fprintf(stderr,
+		"Usage: %s [-s start] [-l length] [-m minimum-extent]"
+		" [-v] {mountpoint}\n\t"
+		"-s Starting Byte to discard from\n\t"
+		"-l Number of Bytes to discard from the start\n\t"
+		"-m Minimum extent length to discard\n\t"
+		"-v Verbose - number of discarded bytes\n",
+		program_name);
+}
+
+static void err_exit(const char *fmt, ...)
+{
+	va_list pvar;
+	va_start(pvar, fmt);
+	vfprintf(stderr, fmt, pvar);
+	va_end(pvar);
+	usage();
+	exit(EXIT_FAILURE);
+}
+
+/**
+ * Get the number from argument. It can be number followed by
+ * units: k|K, m|M, g|G, t|T
+ */
+static unsigned long long get_number(char **optarg)
+{
+	char *opt, *end;
+	unsigned long long number, max;
+
+	/* get the max to avoid overflow */
+	max = ULLONG_MAX / 1024;
+	number = 0;
+	opt = *optarg;
+
+	if (*opt == '-') {
+		err_exit("%s: %s (%s)\n", program_name,
+			 strerror(ERANGE), *optarg);
+	}
+
+	errno = 0;
+	number = strtoul(opt, &end , 0);
+	if (errno)
+		err_exit("%s: %s (%s)\n", program_name,
+			 strerror(errno), *optarg);
+
+	/*
+	 * Convert units to numbers. Fall-through stack is used for units
+	 * so absence of breaks is intentional.
+	 */
+	switch (*end) {
+	case 'T': /* terabytes */
+	case 't':
+		if (number > max)
+			err_exit("%s: %s (%s)\n", program_name,
+				 strerror(ERANGE), *optarg);
+		number *= 1024;
+	case 'G': /* gigabytes */
+	case 'g':
+		if (number > max)
+			err_exit("%s: %s (%s)\n", program_name,
+				 strerror(ERANGE), *optarg);
+		number *= 1024;
+	case 'M': /* megabytes */
+	case 'm':
+		if (number > max)
+			err_exit("%s: %s (%s)\n", program_name,
+				 strerror(ERANGE), *optarg);
+		number *= 1024;
+	case 'K': /* kilobytes */
+	case 'k':
+		if (number > max)
+			err_exit("%s: %s (%s)\n", program_name,
+				 strerror(ERANGE), *optarg);
+		number *= 1024;
+		end++;
+	case '\0': /* end of the string */
+		break;
+	default:
+		err_exit("%s: %s (%s)\n", program_name,
+			 strerror(EINVAL), *optarg);
+		return 0;
+	}
+
+	if (*end != '\0') {
+		err_exit("%s: %s (%s)\n", program_name,
+			 strerror(EINVAL), *optarg);
+	}
+
+	return number;
+}
+
+static int parse_opts(int argc, char **argv, struct options *opts)
+{
+	int c;
+
+	while ((c = getopt(argc, argv, "s:l:m:v")) != EOF) {
+		switch (c) {
+		case 's': /* starting point */
+			opts->range->start = get_number(&optarg);
+			break;
+		case 'l': /* length */
+			opts->range->len = get_number(&optarg);
+			break;
+		case 'm': /* minlen */
+			opts->range->minlen = get_number(&optarg);
+			break;
+		case 'v': /* verbose */
+			opts->verbose = 1;
+			break;
+		default:
+			return EXIT_FAILURE;
+		}
+	}
+
+	return 0;
+}
+
+int main(int argc, char **argv)
+{
+	struct options *opts;
+	struct stat sb;
+	int fd, err = 0, ret = EXIT_FAILURE;
+
+	opts = malloc(sizeof(struct options));
+	if (!opts)
+		err_exit("%s: malloc(): %s\n", program_name, strerror(errno));
+
+	opts->range = NULL;
+	opts->verbose = 0;
+
+	if (argc > 1)
+		strncpy(opts->mpoint, argv[argc - 1], sizeof(opts->mpoint));
+
+	opts->range = calloc(1, sizeof(struct fstrim_range));
+	if (!opts->range) {
+		fprintf(stderr, "%s: calloc(): %s\n", program_name,
+			strerror(errno));
+		goto free_opts;
+	}
+
+	opts->range->len = ULLONG_MAX;
+
+	if (argc > 2)
+		err = parse_opts(argc, argv, opts);
+
+	if (err) {
+		usage();
+		goto free_opts;
+	}
+
+	if (strnlen(opts->mpoint, 1) < 1) {
+		fprintf(stderr, "%s: You have to specify mount point.\n",
+			program_name);
+		usage();
+		goto free_opts;
+	}
+
+	if (stat(opts->mpoint, &sb) == -1) {
+		fprintf(stderr, "%s: %s: %s\n", program_name,
+			opts->mpoint, strerror(errno));
+		usage();
+		goto free_opts;
+	}
+
+	if (!S_ISDIR(sb.st_mode)) {
+		fprintf(stderr, "%s: %s: (%s)\n", program_name,
+			opts->mpoint, strerror(ENOTDIR));
+		usage();
+		goto free_opts;
+	}
+
+	fd = open(opts->mpoint, O_RDONLY);
+	if (fd < 0) {
+		fprintf(stderr, "%s: open(%s): %s\n", program_name,
+			opts->mpoint, strerror(errno));
+		goto free_opts;
+	}
+
+	if (ioctl(fd, FITRIM, opts->range)) {
+		fprintf(stderr, "%s: FSTRIM: %s\n", program_name,
+			strerror(errno));
+		goto free_opts;
+	}
+
+	if ((opts->verbose) && (opts->range))
+		fprintf(stdout, "%lu Bytes were trimmed\n", opts->range->len);
+
+	ret = EXIT_SUCCESS;
+
+free_opts:
+	if (opts) {
+		if (opts->range)
+			free(opts->range);
+		free(opts);
+	}
+
+	return ret;
+}
Signature · 4 lines
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#2 lore raw
In reply to Lukas Czerner

Hi Lukas,

I've looked over it a bit again, and I can't really comment on the code too much, as my bash fu is nowhere near a level to actually make sense
of most of the testcase. I have however ran it in a few setups and
can comment based on that.

First your current first discard to check if we actually support it is not handled correctly. Running the test on a filesystem that doesn't support it currently fails the test for me with:

@@ -1,3 +1,2 @@
QA output created by 249
-Checking FITRIM support: done.
-Running the test: done.
+Checking FITRIM support: fstrim: FSTRIM: Inappropriate ioctl for device

This should be easily fixable by redirecting the output of the test command to /dev/null and do a _notrun if it exited with an error value, just like other tests that require non-standard behaviour.

Second using data from /usr/share/doc seems a bit non-deterministic to me, as the content will be different on every system. Any chance you could just use the xfstests source code instead?

Third the testcase runs forever, e.g. 93 minutes in my KVM setup, even using a no-op discard implementation. While this is useful for burn in testing having a shorter run time version that can be run automatically would be really useful. I tried to figure out what paramters control the runtime, but as mentioned above my bash skill really aren't enough to do that.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#3 lore raw
Quoted 25 lines · Christoph Hellwig

Hi Lukas,

I've looked over it a bit again, and I can't really comment on the code too much, as my bash fu is nowhere near a level to actually make sense
of most of the testcase. I have however ran it in a few setups and
can comment based on that.

First your current first discard to check if we actually support it is not handled correctly. Running the test on a filesystem that doesn't support it currently fails the test for me with:

(quoted diff) +0 -0
@@ -1,3 +1,2 @@

QA output created by 249

(quoted diff) +1 -2
-Checking FITRIM support: done.
-Running the test: done.
+Checking FITRIM support: fstrim: FSTRIM: Inappropriate ioctl for device

This should be easily fixable by redirecting the output of the test command to /dev/null and do a _notrun if it exited with an error value, just like other tests that require non-standard behaviour.

Second using data from /usr/share/doc seems a bit non-deterministic to me, as the content will be different on every system. Any chance you could just use the xfstests source code instead?

Quoted 6 lines

Third the testcase runs forever, e.g. 93 minutes in my KVM setup, even using a no-op discard implementation. While this is useful for burn in testing having a shorter run time version that can be run automatically would be really useful. I tried to figure out what paramters control the runtime, but as mentioned above my bash skill really aren't enough to do that.

You can lower the "nproc" variable, optionally you can also lower "repeat" variable in run_process().

Hi Christoph,

thanks a lot for review. I certainly can use xfstests source code, so I'll change that. Regarding the duration of the test, it is kind of hard to determine the right amount of load, because the devices and machines differs a lot. It also involves a LOT of copying files so I guess it might be a little slow. However I need to be careful with reducing number of concurrent processes, because with less stress it might be harder to hit potential problem. (you can always

But I'll try to test it an find just the right balance. Thanks again.

-Lukas

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#4 lore raw

What is the plan for TRIM support in XFS? I don't see any mention in the 2.6.36.2 sources, so I guess the code you guys have been discussing must not be considered stable yet.

How will it eventually work? Will the filesystem automatically issue the commands to the drive whenever a file is deleted, or will it always require some sort of prodding from a userspace program?

And what kind of experiences are people having with XFS on current SSDs? I just put an Intel drive in my newly upgraded box and I'm wondering if this blazing performance will screech to a halt when my cumulative writes exceed its 120 GB capacity. Does anyone have a standalone "drive optimizer" that understands XFS, i.e, that can examine an unmounted file system and issue the appropriate TRIM commands for every block on the free list?

Or is the Intel drive somehow really good at avoiding write amplification even without TRIM help from the file system? I've heard that may be the case, but I don't understand how that can be.

Phil

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

#5 lore raw
Quoted 7 lines · Phil Karn

What is the plan for TRIM support in XFS? I don't see any mention in the 2.6.36.2 sources, so I guess the code you guys have been discussing must not be considered stable yet.

How will it eventually work? Will the filesystem automatically issue the commands to the drive whenever a file is deleted, or will it always require some sort of prodding from a userspace program?

I submitted a patch to add FITRIM support, which should go into 2.6.38 if I get a positive review. That is the variant that is triggered by the fstim userspace program and what the test we're discussing here exercises. I've also prototyped various forms of online discard, but none of them is quite ready yet. I've started porting some allocator patches from Dave forward, which are a requirement for reliable online discard. After that I need to decide which of the variants we'll actually use. So far the "dumb" discard at transaction commit time shows the best results.

Quoted 6 lines

And what kind of experiences are people having with XFS on current SSDs? I just put an Intel drive in my newly upgraded box and I'm wondering if this blazing performance will screech to a halt when my cumulative writes exceed its 120 GB capacity. Does anyone have a standalone "drive optimizer" that understands XFS, i.e, that can examine an unmounted file system and issue the appropriate TRIM commands for every block on the free list?

The intel SSDs do quite fine without constant trimming, as do most current SSDs. I only know of one sample that areally needs periodic trims, but AFAIK it's not generally available yet (sorry, can't name the vendor due to NDAs). But even with the current ones a TRIM once in a while is useful, but about once every couple of weeks is enough. For a system not constantly loaded you can use Mark Lord's wiper.sh script, but be aware it only works on disks / partitions and not on LVM volumes. Once the FITRIM support is in you can use that for the same purpose, with the benefit of also working over LVM, and also working nicely on a system under heavy use, as it doesn't have to allocate all free space as the userspace hack done by wiper.sh does.

Quoted 3 lines

Or is the Intel drive somehow really good at avoiding write amplification even without TRIM help from the file system? I've heard that may be the case, but I don't understand how that can be.

No, the Intel disks actually degrade a lot after heavy use. Unfortunately even regular trims don't always seem to solve this.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs