Hi Darrick, all, > On Tue, Oct 04, 2022 at 11:08:10AM +0200, Petr Vorel wrote: > > df01.sh started to fail on XFS on certain configuration since mkfs.xfs > > and kernel 5.19. Implement fsfreeze instead of introducing external > ...since the introduction of background garbage collection on XFS in > kernel 5.19. ;) Thank you, I'll update the commit message before merge. > > dependency. NOTE: implementation could fail on other filesystems > > (EOPNOTSUPP on exfat, ntfs, vfat). > > Suggested-by: Darrick J. Wong <djwong@xxxxxxxxxx> > > Suggested-by: Eric Sandeen <sandeen@xxxxxxxxxx> > > Signed-off-by: Petr Vorel <pvorel@xxxxxxx> > > --- > > Hi, > > FYI the background of this issue: > > https://lore.kernel.org/ltp/Yv5oaxsX6z2qxxF3@magnolia/ > > https://lore.kernel.org/ltp/974cc110-d47e-5fae-af5f-e2e610720e2d@xxxxxxxxxx/ > > @LTP developers: not sure if the consensus is to avoid LTP API > > completely (even use it just with TST_NO_DEFAULT_MAIN), if required I > > can rewrite to use it just to get SAFE_*() macros (like > > testcases/lib/tst_checkpoint.c) or even with tst_test workarounds > > (testcases/lib/tst_get_free_pids.c). > > Kind regards, > > Petr > > testcases/commands/df/Makefile | 4 +- > > testcases/commands/df/df01.sh | 3 ++ > > testcases/commands/df/df01_fsfreeze.c | 55 +++++++++++++++++++++++++++ > > 3 files changed, 61 insertions(+), 1 deletion(-) > > create mode 100644 testcases/commands/df/df01_fsfreeze.c > > diff --git a/testcases/commands/df/Makefile b/testcases/commands/df/Makefile > > index 2787bb43a..1e0b4283a 100644 > > --- a/testcases/commands/df/Makefile > > +++ b/testcases/commands/df/Makefile > > @@ -1,11 +1,13 @@ > > # SPDX-License-Identifier: GPL-2.0-or-later > > +# Copyright (c) Linux Test Project, 2021-2022 > > # Copyright (c) 2015 Fujitsu Ltd. > > -# Author:Zhang Jin <jy_zhangjin@xxxxxxxxxxxxxx> > > +# Author: Zhang Jin <jy_zhangjin@xxxxxxxxxxxxxx> > > top_srcdir ?= ../../.. > > include $(top_srcdir)/include/mk/env_pre.mk > > INSTALL_TARGETS := df01.sh > > +MAKE_TARGETS := df01_fsfreeze > > include $(top_srcdir)/include/mk/generic_leaf_target.mk > > diff --git a/testcases/commands/df/df01.sh b/testcases/commands/df/df01.sh > > index ae0449c3c..c59d2a01d 100755 > > --- a/testcases/commands/df/df01.sh > > +++ b/testcases/commands/df/df01.sh > > @@ -46,6 +46,9 @@ df_test() > > ROD_SILENT rm -rf $TST_MNTPOINT/testimg > > + # ensure free space change can be seen by statfs > > + [ "$fs" = "xfs" ] && ROD_SILENT df01_fsfreeze $TST_MNTPOINT > > + > > # flush file system buffers, then we can get the actual sizes. > > sync > > } > > diff --git a/testcases/commands/df/df01_fsfreeze.c b/testcases/commands/df/df01_fsfreeze.c > > new file mode 100644 > > index 000000000..d47e1b01a > > --- /dev/null > > +++ b/testcases/commands/df/df01_fsfreeze.c > > @@ -0,0 +1,55 @@ > > +// SPDX-License-Identifier: GPL-2.0-or-later > > +/* > > + * Copyright (c) 2010 Hajime Taira <htaira@xxxxxxxxxx> > > + * Copyright (c) 2010 Masatake Yamato <yamato@xxxxxxxxxx> > > + * Copyright (c) 2022 Petr Vorel <pvorel@xxxxxxx> > > + */ > > + > > +#include <errno.h> > > +#include <fcntl.h> > > +#include <linux/fs.h> > > +#include <stdio.h> > > +#include <stdlib.h> > > +#include <string.h> > > +#include <sys/ioctl.h> > > +#include <sys/stat.h> > > +#include <unistd.h> > > + > > +#define err_exit(...) ({ \ > > + fprintf(stderr, __VA_ARGS__); \ > > + if (errno) \ > > + fprintf(stderr, ": %s (%d)", strerror(errno), errno); \ > > + fprintf(stderr, "\n"); \ > > + exit(EXIT_FAILURE); \ > > +}) > > + > > +int main(int argc, char *argv[]) > > +{ > > + int fd; > > + struct stat sb; > > + > > + if (argc < 2) > > + err_exit("USAGE: df01_fsfreeze <mountpoint>"); > > + > > + fd = open(argv[1], O_RDONLY); > > + if (fd < 0) > > + err_exit("open '%s' failed", argv[1]); > > + > > + if (fstat(fd, &sb) == -1) > > + err_exit("stat of '%s' failed", argv[1]); > > + > > + if (!S_ISDIR(sb.st_mode)) > Note that XFS is perfectly happy to let you call FIFREEZE on a > nondirectory. Thanks for info. I added this because the tool is intended to be used on $TST_MNTPOINT, which is directory. But as not needed, I might remove it. > > + err_exit("%s: is not a directory", argv[1]); > > + > > + if (ioctl(fd, FIFREEZE, 0) < 0) > > + err_exit("ioctl FIFREEZE on '%s' failed", argv[1]); > > + > > + usleep(100); > The usleep shouldn't be necessary here. All the work necessary > (background gc flushing, log quiescing, etc.) to stabilize the free > space counters are performed synchronously before the FIFREEZE ioctl > returns. > If that's not been your experience, please let us know. Indeed usleep() is not needed. I added it when I was debugging failures on other filesystems. Kind regards, Petr > --D > > + if (ioctl(fd, FITHAW, 0) < 0) > > + err_exit("ioctl FITHAW on '%s' failed", argv[1]); > > + > > + close(fd); > > + > > + return EXIT_SUCCESS; > > +} > > -- > > 2.37.3
