Message-ID: <20180621024033.GK2780@desktop>
Date:   Thu, 21 Jun 2018 10:40:33 +0800
From:   Eryu Guan <guaneryu@...il.com>
To:     Ross Zwisler <ross.zwisler@...ux.intel.com>
Cc:     Eryu Guan <eguan@...hat.com>, fstests@...r.kernel.org,
        Jan Kara <jack@...e.cz>,
        Dan Williams <dan.j.williams@...el.com>,
        Dave Chinner <david@...morbit.com>,
        Christoph Hellwig <hch@....de>, linux-nvdimm@...ts.01.org,
        Jeff Moyer <jmoyer@...hat.com>, linux-ext4@...r.kernel.org
Subject: Re: [fstests PATCH 2/2] generic/999: test DAX DMA vs
 truncate/hole-punch

On Wed, Jun 20, 2018 at 04:51:47PM -0600, Ross Zwisler wrote:
> This adds a regression test for the following series:
> 
> https://lists.01.org/pipermail/linux-nvdimm/2018-June/016431.html
> 
> which added synchronization between DAX DMA in ext4 and
> truncate/hole-punch.  The intention of the test is to test those specific
> changes, but it runs fine both with XFS and without DAX, so I've put it in
> the generic tests instead of ext4 and not restricted it to only DAX
> configurations.
> 
> When run with v4.18-rc1 + DAX + ext4, this test will hit the following
> WARN_ON_ONCE() in dax_disassociate_entry():
> 
> 	WARN_ON_ONCE(trunc && page_ref_count(page) > 1);
> 
> If you change this to a WARN_ON() instead, you can see that each of the
> four paths being exercised in this test hits that condition many times in
> the one second that the subtest is being run.
> 
> Signed-off-by: Ross Zwisler <ross.zwisler@...ux.intel.com>

I haven't tested it yet, just a few quick comments below.

> ---
>  .gitignore             |   1 +
>  src/Makefile           |   2 +-
>  src/t_mmap_collision.c | 208 +++++++++++++++++++++++++++++++++++++++++++++++++
>  tests/generic/999      |  50 ++++++++++++
>  tests/generic/999.out  |   2 +
>  tests/generic/group    |   1 +
>  6 files changed, 263 insertions(+), 1 deletion(-)
>  create mode 100644 src/t_mmap_collision.c
>  create mode 100755 tests/generic/999
>  create mode 100644 tests/generic/999.out
> 
> diff --git a/.gitignore b/.gitignore
> index efc73a7c..936955e0 100644
> --- a/.gitignore
> +++ b/.gitignore
> @@ -126,6 +126,7 @@
>  /src/t_immutable
>  /src/t_locks_execve
>  /src/t_mmap_cow_race
> +/src/t_mmap_collision

Looks like this should go before t_mmap_cow_race?

>  /src/t_mmap_dio
>  /src/t_mmap_fallocate
>  /src/t_mmap_stale_pmd
> diff --git a/src/Makefile b/src/Makefile
> index b06b7e25..2dbe5e7e 100644
> --- a/src/Makefile
> +++ b/src/Makefile
> @@ -15,7 +15,7 @@ TARGETS = dirstress fill fill2 getpagesize holes lstat64 \
>  	holetest t_truncate_self t_mmap_dio af_unix t_mmap_stale_pmd \
>  	t_mmap_cow_race t_mmap_fallocate fsync-err t_mmap_write_ro \
>  	t_ext4_dax_journal_corruption t_ext4_dax_inline_corruption \
> -	t_ofd_locks t_locks_execve
> +	t_ofd_locks t_locks_execve t_mmap_collision
>  
>  LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \
>  	preallo_rw_pattern_writer ftrunc trunc fs_perms testx looptest \
> diff --git a/src/t_mmap_collision.c b/src/t_mmap_collision.c
> new file mode 100644
> index 00000000..f652e913
> --- /dev/null
> +++ b/src/t_mmap_collision.c
> @@ -0,0 +1,208 @@

Please add copyright and licence statements to c file too, use the new
SPDX tag :)

(Refer to Dave's pending patch "src/: spdx license conversion" in
patchset "fstests: SPDX license conversion round 2").

> +#include <errno.h>
> +#include <fcntl.h>
> +#include <pthread.h>
> +#include <stdbool.h>
> +#include <stdio.h>
> +#include <stdlib.h>
> +#include <string.h>
> +#include <sys/mman.h>
> +#include <sys/stat.h>
> +#include <sys/types.h>
> +#include <unistd.h>
> +
> +#define PAGE(a) ((a)*0x1000)
> +#define FILE_SIZE PAGE(4)
> +
> +void *dax_data;
> +int nodax_fd;
> +int dax_fd;
> +bool done;
> +
> +#define err_exit(op)                                                          \
> +{                                                                             \
> +	fprintf(stderr, "%s %s: %s\n", __func__, op, strerror(errno));        \
> +	exit(1);                                                              \
> +}                                                                             \
> +
> +void punch_hole_fn(void *ptr)
> +{
> +	ssize_t read;
> +	int rc;
> +
> +	while (!done) {
> +		read = 0;
> +
> +		do {
> +			rc = pread(nodax_fd, dax_data + read, FILE_SIZE - read,
> +					read);
> +			if (rc > 0)
> +				read += rc;
> +		} while (rc > 0);
> +
> +		if (read != FILE_SIZE || rc != 0)
> +			err_exit("pread");
> +
> +		rc = fallocate(dax_fd,
> +				FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
> +				0, FILE_SIZE);
> +		if (rc < 0)
> +			err_exit("fallocate");
> +
> +		usleep(rand() % 1000);
> +	}
> +}
> +
> +void zero_range_fn(void *ptr)
> +{
> +	ssize_t read;
> +	int rc;
> +
> +	while (!done) {
> +		read = 0;
> +
> +		do {
> +			rc = pread(nodax_fd, dax_data + read, FILE_SIZE - read,
> +					read);
> +			if (rc > 0)
> +				read += rc;
> +		} while (rc > 0);
> +
> +		if (read != FILE_SIZE || rc != 0)
> +			err_exit("pread");
> +
> +		rc = fallocate(dax_fd,
> +				FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE,
> +				0, FILE_SIZE);
> +		if (rc < 0)
> +			err_exit("fallocate");
> +
> +		usleep(rand() % 1000);
> +	}
> +}
> +
> +void truncate_down_fn(void *ptr)
> +{
> +	ssize_t read;
> +	int rc;
> +
> +	while (!done) {
> +		read = 0;
> +
> +		if (ftruncate(dax_fd, 0) < 0)
> +			err_exit("ftruncate");
> +		if (fallocate(dax_fd, 0, 0, FILE_SIZE) < 0)
> +			err_exit("fallocate");
> +
> +		do {
> +			rc = pread(nodax_fd, dax_data + read, FILE_SIZE - read,
> +					read);
> +			if (rc > 0)
> +				read += rc;
> +		} while (rc > 0);
> +
> +		/*
> +		 * For this test we ignore errors from pread().  These errors
> +		 * can happen if we try and read while the other thread has
> +		 * made the file size 0.
> +		 */
> +
> +		usleep(rand() % 1000);
> +	}
> +}
> +
> +void collapse_range_fn(void *ptr)
> +{
> +	ssize_t read;
> +	int rc;
> +
> +	while (!done) {
> +		read = 0;
> +
> +		if (fallocate(dax_fd, 0, 0, FILE_SIZE) < 0)
> +			err_exit("fallocate 1");
> +		if (fallocate(dax_fd, FALLOC_FL_COLLAPSE_RANGE, 0, PAGE(1)) < 0)
> +			err_exit("fallocate 2");
> +		if (fallocate(dax_fd, 0, 0, FILE_SIZE) < 0)
> +			err_exit("fallocate 3");
> +
> +		do {
> +			rc = pread(nodax_fd, dax_data + read, FILE_SIZE - read,
> +					read);
> +			if (rc > 0)
> +				read += rc;
> +		} while (rc > 0);
> +
> +		/* For this test we ignore errors from pread. */
> +
> +		usleep(rand() % 1000);
> +	}
> +}
> +
> +void run_test(void (*test_fn)(void *))
> +{
> +	const int NUM_THREADS = 2;
> +	pthread_t worker_thread[NUM_THREADS];
> +	int i;
> +
> +	done = 0;
> +	for (i = 0; i < NUM_THREADS; i++)
> +		pthread_create(&worker_thread[i], NULL, (void*)test_fn, NULL);
> +
> +	sleep(1);
> +	done = 1;
> +
> +	for (i = 0; i < NUM_THREADS; i++)
> +		pthread_join(worker_thread[i], NULL);
> +}
> +
> +int main(int argc, char *argv[])
> +{
> +	int err;
> +
> +	if (argc != 3) {
> +		printf("Usage: %s <dax file> <non-dax file>\n",
> +				basename(argv[0]));
> +		exit(0);
> +	}
> +
> +	dax_fd = open(argv[1], O_RDWR|O_CREAT, S_IRUSR|S_IWUSR);
> +	if (dax_fd < 0)
> +		err_exit("dax_fd open");
> +
> +	nodax_fd = open(argv[2], O_RDWR|O_CREAT|O_DIRECT, S_IRUSR|S_IWUSR);
> +	if (nodax_fd < 0)
> +		err_exit("nodax_fd open");
> +
> +	if (ftruncate(dax_fd, 0) < 0)
> +		err_exit("dax_fd ftruncate");
> +	if (fallocate(dax_fd, 0, 0, FILE_SIZE) < 0)
> +		err_exit("dax_fd fallocate");
> +
> +	if (ftruncate(nodax_fd, 0) < 0)
> +		err_exit("nodax_fd ftruncate");
> +	if (fallocate(nodax_fd, 0, 0, FILE_SIZE) < 0)
> +		err_exit("nodax_fd fallocate");
> +
> +	dax_data = mmap(NULL, FILE_SIZE, PROT_READ|PROT_WRITE, MAP_SHARED,
> +			dax_fd, 0);
> +	if (dax_data == MAP_FAILED)
> +		err_exit("mmap");
> +
> +	run_test(&punch_hole_fn);
> +	run_test(&zero_range_fn);
> +	run_test(&truncate_down_fn);
> +	run_test(&collapse_range_fn);

Check for FALLOC_FL_* definition and compile
(punch_hole|zero_range|collapse_range)_fn out if the needed flag is not
defined?

> +
> +	if (munmap(dax_data, FILE_SIZE) != 0)
> +		err_exit("munmap");
> +
> +	err = close(dax_fd);
> +	if (err < 0)
> +		err_exit("dax_fd close");
> +
> +	err = close(nodax_fd);
> +	if (err < 0)
> +		err_exit("nodax_fd close");
> +
> +	return 0;
> +}
> diff --git a/tests/generic/999 b/tests/generic/999
> new file mode 100755
> index 00000000..8f488cb5
> --- /dev/null
> +++ b/tests/generic/999
> @@ -0,0 +1,50 @@
> +#! /bin/bash
> +# SPDX-License-Identifier: GPL-2.0
> +# Copyright (c) 2018 Intel Corporation.  All Rights Reserved.
> +#
> +# FS QA Test generic/999
> +#
> +# This is a regression test for kernel patch:
> +#   ext4: handle layout changes to pinned DAX mapping
> +# created by Ross Zwisler <ross.zwisler@...ux.intel.com>
> +#
> +seq=`basename $0`
> +seqres=$RESULT_DIR/$seq
> +echo "QA output created by $seq"
> +
> +here=`pwd`
> +tmp=/tmp/$$
> +status=1	# failure is the default!
> +trap "_cleanup; exit \$status" 0 1 2 3 15
> +
> +_cleanup()
> +{
> +	cd /
> +	rm -f $tmp.*
> +}
> +
> +# get standard environment, filters and checks
> +. ./common/rc
> +. ./common/filter
> +
> +# remove previous $seqres.full before test
> +rm -f $seqres.full
> +
> +# Modify as appropriate.
> +_supported_fs generic
> +_supported_os Linux
> +_require_test
> +_require_test_program "t_mmap_collision"

_require_xfs_io_command "fpunch"
_require_xfs_io_command "fcollapse"
_require_xfs_io_command "fzero"

> +
> +# turn off DAX on our scratch device so we can get normal O_DIRECT behavior
> +export MOUNT_OPTIONS=""
> +_scratch_unmount >> $seqres.full 2>&1
> +_scratch_mount >> $seqres.full 2>&1
> +
> +# real QA test starts here
> +$here/src/t_mmap_collision $TEST_DIR/testfile $SCRATCH_MNT/testfile
> +
> +# success, all done
> +echo "Silence is golden"
> +status=0
> +exit
> diff --git a/tests/generic/999.out b/tests/generic/999.out
> new file mode 100644
> index 00000000..3b276ca8
> --- /dev/null
> +++ b/tests/generic/999.out
> @@ -0,0 +1,2 @@
> +QA output created by 999
> +Silence is golden
> diff --git a/tests/generic/group b/tests/generic/group
> index 83a6fdab..793f71ed 100644
> --- a/tests/generic/group
> +++ b/tests/generic/group
> @@ -501,3 +501,4 @@
>  496 auto quick swap
>  497 auto quick swap collapse
>  498 auto quick log
> +999 auto quick dax

Also need "punch collapse zero" groups.

Thanks,
Eryu

> -- 
> 2.14.4
> 

Powered by blists - more mailing lists