ext4/056: add a check to make sure ext4 uuid ioctls get/set during fsstress.
PATCH v5 1/1 ext4/056: add a check to make sure ext4 uuid ioctls get/set during fsstress.
Adds a utility to set/get uuid through ext4 ioctl. Executes the ioctls while running fsstress. These ioctls are used by tune2fs to safely change the uuid without racing other filesystem modifications.
---
Changes in v5:
Added reviewed-by tag.
Added wait after killing fsstress pids.
Removed _scratch_mount output redirection.
.gitignore | 1 + src/Makefile | 4 +- src/uuid_ioctl.c | 105 +++++++++++++++++++++++++++++++++++++++++++++ tests/ext4/056 | 64 +++++++++++++++++++++++++++ tests/ext4/056.out | 2 + 5 files changed, 174 insertions(+), 2 deletions(-) create mode 100644 src/uuid_ioctl.c create mode 100755 tests/ext4/056 create mode 100644 tests/ext4/056.out
.gitignore +1 -0
@@ -169,6 +169,7 @@ tags /src/unwritten_mmap /src/unwritten_sync /src/usemem +/src/uuid_ioctl /src/writemod /src/writev_on_pagefault /src/xfsctl
src/Makefile +2 -2
@@ -31,14 +31,14 @@ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \ dio-invalidate-cache stat_test t_encrypted_d_revalidate \ attr_replace_test swapon mkswap t_attr_corruption t_open_tmpfiles \ fscrypt-crypt-util bulkstat_null_ocount splice-test chprojid_fail \ - detached_mounts_propagation ext4_resize + detached_mounts_propagation ext4_resize uuid_ioctl EXTRA_EXECS = dmerror fill2attr fill2fs fill2fs_check scaleread.sh \ btrfs_crc32c_forged_name.py SUBDIRS = log-writes perf -LLDLIBS = $(LIBHANDLE) $(LIBACL) -lpthread -lrt +LLDLIBS = $(LIBHANDLE) $(LIBACL) -lpthread -lrt -luuid ifeq ($(HAVE_XLOG_ASSIGN_LSN), true) LINUX_TARGETS += loggen
src/uuid_ioctl.c new file +105 -0
105 lines not on this page · raw message
tests/ext4/056 new file +64 -0
64 lines not on this page · raw message
tests/ext4/056.out new file +2 -0
@@ -0,0 +1,2 @@ +QA output created by 056 +Silence is golden
Signature · 1 line
2.37.0.170.g444d1eabd0-goog
Quoted 183 lines · Jeremy Bongio
Adds a utility to set/get uuid through ext4 ioctl. Executes the ioctls while running fsstress. These ioctls are used by tune2fs to safely change the uuid without racing other filesystem modifications.
Reviewed-by Darrick J. Wong <[email protected]>Signed-off-by Jeremy Bongio <[email protected]>---
Changes in v5:
Added reviewed-by tag.
Added wait after killing fsstress pids.
Removed
_scratch_mountoutput redirection..gitignore | 1 + src/Makefile | 4 +- src/uuid_ioctl.c | 105 +++++++++++++++++++++++++++++++++++++++++++++ tests/ext4/056 | 64 +++++++++++++++++++++++++++ tests/ext4/056.out | 2 + 5 files changed, 174 insertions(+), 2 deletions(-) create mode 100644 src/uuid_ioctl.c create mode 100755 tests/ext4/056 create mode 100644 tests/ext4/056.out.gitignore +1
-0@@ -169,6 +169,7 @@ tags /src/unwritten_mmap /src/unwritten_sync /src/usemem +/src/uuid_ioctl /src/writemod /src/writev_on_pagefault /src/xfsctlsrc/Makefile +2
-2@@ -31,14 +31,14 @@ LINUX_TARGETS = xfsctl bstat t_mtab getdevicesize preallo_rw_pattern_reader \ dio-invalidate-cache stat_test t_encrypted_d_revalidate \ attr_replace_test swapon mkswap t_attr_corruption t_open_tmpfiles \ fscrypt-crypt-util bulkstat_null_ocount splice-test chprojid_fail \ - detached_mounts_propagation ext4_resize + detached_mounts_propagation ext4_resize uuid_ioctl EXTRA_EXECS = dmerror fill2attr fill2fs fill2fs_check scaleread.sh \ btrfs_crc32c_forged_name.py SUBDIRS = log-writes perf -LLDLIBS = $(LIBHANDLE) $(LIBACL) -lpthread -lrt +LLDLIBS = $(LIBHANDLE) $(LIBACL) -lpthread -lrt -luuid ifeq ($(HAVE_XLOG_ASSIGN_LSN), true) LINUX_TARGETS += loggensrc/uuid_ioctl.c new file +105
-0@@ -0,0 +1,105 @@ +// SPDX-License-Identifier: GPL-2.0 +/* +* Copyright (c) 2022 Google, Inc. All Rights Reserved. +* +* Test program which uses the raw ext4 set_fsuuid ioctl directly. +* SYNOPSIS: +* $0 COMMAND MOUNT_POINT [UUID] +* +* COMMAND must be either "get" or "set". +* The UUID must be a 16 octet sequence represented as 32 hexadecimal digits in +* canonical textual representation, e.g. output from `uuidgen`. +* +*/ + +#include <stdio.h> +#include <fcntl.h> +#include <errno.h> +#include <unistd.h> +#include <stdint.h> +#include <stdlib.h> +#include <string.h> +#include <sys/ioctl.h> +#include <uuid/uuid.h> +#include <linux/fs.h> + +struct fsuuid { +__u32 fsu_len; +__u32 fsu_flags; +__u8 fsu_uuid[]; +}; + +#ifndef EXT4_IOC_GETFSUUID +#define EXT4_IOC_GETFSUUID _IOR('f', 44, struct fsuuid) +#endif + +#ifndef EXT4_IOC_SETFSUUID +#define EXT4_IOC_SETFSUUID _IOW('f', 44, struct fsuuid) +#endif + +int main (int argc,char **argv) +{ +int error, fd; +struct fsuuid *fsuuid =NULL ; + +if (argc <3 ) { +fprintf (stderr,"Invalid arguments\n" ); +return 1 ; + } + + fd =open (argv[2 ],O_RDONLY ); +if (!fd) { +perror (argv[2 ]); +return 1 ; + } + + fsuuid =malloc (sizeof (*fsuuid) +sizeof (uuid_t)); +if (!fsuuid) { +perror ("malloc" ); +return 1 ; + } + fsuuid->fsu_len =sizeof (uuid_t); + fsuuid->fsu_flags =0 ; + +if (strcmp (argv[1 ],"get" ) ==0 ) { +uuid_t uuid; +char uuid_str[37 ]; + +if (ioctl (fd,EXT4_IOC_GETFSUUID , fsuuid)) { +fprintf (stderr,"%s while trying to get fs uuid\n" , +strerror (errno)); +return 1 ; + } + +memcpy (&uuid, fsuuid->fsu_uuid,sizeof (uuid)); +uuid_unparse (uuid, uuid_str); +printf ("%s\n" , uuid_str); + }else if (strcmp (argv[1 ],"set" ) ==0 ) { +uuid_t uuid; + +if (argc !=4 ) { +fprintf (stderr,"UUID argument missing.\n" ); +return 1 ; + } + + error =uuid_parse (argv[3 ], uuid); +if (error <0 ) { +fprintf (stderr,"Invalid UUID. The UUID should be in " +"canonical format. Example: " +"8c628557-6987-42b2-ba16-b7cc79ddfb43\n" ); +return 1 ; + } + +memcpy (&fsuuid->fsu_uuid, uuid,sizeof (uuid)); +if (ioctl (fd,EXT4_IOC_SETFSUUID , fsuuid)) { +fprintf (stderr,"%s while trying to set fs uuid\n" , +strerror (errno)); +return 1 ; + } + }else { +fprintf (stderr,"Invalid command\n" ); +return 1 ; + } + +return 0 ; +}tests/ext4/056 new file +7
-0@@ -0,0 +1,64 @@ +#! /bin/bash +# SPDX-License-Identifier: GPL-2.0 +# Copyright (c) 2022 Google, Inc. All Rights Reserved. +# +# Test the set/get UUID ioctl. +# +
tests/ext4/056 line 17
Quoted 10 lines
tests/ext4/056 new file +10
-0+. ./common/preamble +_begin_fstest auto ioctl + +# Override the default cleanup function. +_cleanup() +{ + cd / + rm -r -f $tmp.* + kill -9 $fsstress_pid 2>/dev/null; + wait $fsstress_pid > /dev/null 2>&1
I think "wait" is enough. With this change, it's good to me.
Quoted 58 lines
tests/ext4/056 new file +47
-0+} + +# Import common functions. +. ./common/filter + +# real QA test starts here +_supported_fs ext4 +_require_scratch +_require_test_program uuid_ioctl +_require_command $UUIDGEN_PROG uuidgen + +UUID_IOCTL=$here/src/uuid_ioctl + +# If the ioctl is not supported by the kernel, then skip test. +current_uuid=$($UUID_IOCTL get $SCRATCH_MNT 2>&1) +if [[ "$current_uuid" =~ ^Inappropriate[[:space:]]ioctl ]]; then + _notrun "UUID ioctls are not supported by kernel." +fi + +# metadata_csum_seed must be set to decouple checksums from the uuid. +# Otherwise, checksums need to be recomputed when the uuid changes, which +# is not supported by the ioctl. +_scratch_mkfs_ext4 -O metadata_csum_seed >> $seqres.full 2>&1 +_scratch_mount + +# Begin fsstress while modifying UUID +fsstress_args=$(_scale_fsstress_args -d $SCRATCH_MNT -p 15 -n 999999) +$FSSTRESS_PROG $fsstress_args > /dev/null 2>&1 & +fsstress_pid=$! + +for n in $(seq 1 20); do + new_uuid=$($UUIDGEN_PROG) + + echo "Setting UUID to ${new_uuid}" >> $seqres.full 2>&1 + $UUID_IOCTL set $SCRATCH_MNT $new_uuid + + current_uuid=$($UUID_IOCTL get $SCRATCH_MNT) + echo "$UUID_IOCTL get $SCARTCH_MNT: $current_uuid" >> $seqres.full 2>&1 + if [[ "$current_uuid" != "$new_uuid" ]]; then + echo "Current UUID ($current_uuid) does not equal what was sent with the ioctl ($new_uuid)" + fi +done + +# success, all done +echo "Silence is golden" +status=0 +exittests/ext4/056.out new file +2
-0@@ -0,0 +1,2 @@ +QA output created by 056 +Silence is goldenSignature · 1 line
2.37.0.170.g444d1eabd0-goog
Theodore Ts'o
Reviewed-byQuoted 10 lines · Zorro Lang
tests/ext4/056 line 17
Quoted 7 lines · Jeremy Bongio
tests/ext4/056 new file +7
-0+# Override the default cleanup function. +_cleanup() +{ + cd / + rm -r -f $tmp.* + kill -9 $fsstress_pid 2>/dev/null; + wait $fsstress_pid > /dev/null 2>&1I think "wait" is enough. With this change, it's good to me.
The kill -9 is needed, because otherwise the test will run for a **very** long time. The reason for it is because of the -n 999999 in fstress_args:
Quoted 4 lines
tests/ext4/056 line 46
Quoted 4 lines
tests/ext4/056 new file +4
-0+# Begin fsstress while modifying UUID +fsstress_args=$(_scale_fsstress_args -d $SCRATCH_MNT -p 15 -n 999999) +$FSSTRESS_PROG $fsstress_args > /dev/null 2>&1 & +fsstress_pid=$!
We could adjust the number of loops to a more reasonable number, but then test becomes less reliable, since depending on the storage device (e.g., cheap USB thumb drive found in the checkout counter at a convenience store, vs. a high-end NVMe SSD) and the overall speed of the system, a different number of loops will be needed.
Given that we're *only* using the fsstress as an antogonist while we are changing the UUID of the file system 20 times, killing the fsstress once we're done with the UUID runs is sufficient, I would argue.
Also, Jeremy, it looks like you haven't updated your xfstests-dev repository in a few weeks. Since you started this project, ext4/056 has been assigned, and there has been some new helper programs added which caused patch conflicts in src/Makefile and in .gitignore. They were pretty trivial to fix up the patch conflicts (which I've done in my xfstests-dev tree), but it's best practice to rebase on top of origin/for-next and re-test just to make sure there haven't been some major change in the fstests common scripts that might catch your test out.
Also, feel free to add my:
Cheers,
- Ted
Quoted 12 lines · Theodore Ts'o
Quoted 10 lines · Zorro Lang
tests/ext4/056 line 17
Quoted 7 lines · Jeremy Bongio
tests/ext4/056 new file +7
-0+# Override the default cleanup function. +_cleanup() +{ + cd / + rm -r -f $tmp.* + kill -9 $fsstress_pid 2>/dev/null; + wait $fsstress_pid > /dev/null 2>&1I think "wait" is enough. With this change, it's good to me.
Quoted 2 lines
The kill -9 is needed, because otherwise the test will run for a **very** long time. The reason for it is because of the -n 999999 in
Sure, I mean:
kill -9 $fsstress_pid 2>/dev/null
wait
Not remove the "kill" line :)
Quoted 18 lines
fstress_args:Quoted 4 lines
tests/ext4/056 line 46
Quoted 4 lines
tests/ext4/056 new file +4
-0+# Begin fsstress while modifying UUID +fsstress_args=$(_scale_fsstress_args -d $SCRATCH_MNT -p 15 -n 999999) +$FSSTRESS_PROG $fsstress_args > /dev/null 2>&1 & +fsstress_pid=$!We could adjust the number of loops to a more reasonable number, but then test becomes less reliable, since depending on the storage device (e.g., cheap USB thumb drive found in the checkout counter at a convenience store, vs. a high-end NVMe SSD) and the overall speed of the system, a different number of loops will be needed.
Given that we're *only* using the fsstress as an antogonist while we are changing the UUID of the file system 20 times, killing the fsstress once we're done with the UUID runs is sufficient, I would argue.
Quoted 9 lines
Also, Jeremy, it looks like you haven't updated your xfstests-dev repository in a few weeks. Since you started this project, ext4/056 has been assigned, and there has been some new helper programs added which caused patch conflicts in src/Makefile and in .gitignore. They were pretty trivial to fix up the patch conflicts (which I've done in my xfstests-dev tree), but it's best practice to rebase on top of origin/for-next and re-test just to make sure there haven't been some major change in the fstests common scripts that might catch your test out.
Thanks for pointing out that, yes, better to rebase to latest fstests for-next branch.
Quoted 4 lines
Also, feel free to add my:
Reviewed-by Theodore Ts'o <[email protected]>
Sure,
Thanks,
Zorro
Quoted 5 lines
Cheers,
- Ted
Quoted 9 lines · Zorro Lang
Quoted 2 lines
The kill -9 is needed, because otherwise the test will run for a **very** long time. The reason for it is because of the -n 999999 in
Sure, I mean:
kill -9 $fsstress_pid 2>/dev/null
waitNot remove the "kill" line :)
Ah yes, sorry, I misunderstood what you meant.
Quoted 10 lines
Quoted 9 lines
Also, Jeremy, it looks like you haven't updated your xfstests-dev repository in a few weeks. Since you started this project, ext4/056 has been assigned, and there has been some new helper programs added which caused patch conflicts in src/Makefile and in .gitignore. They were pretty trivial to fix up the patch conflicts (which I've done in my xfstests-dev tree), but it's best practice to rebase on top of origin/for-next and re-test just to make sure there haven't been some major change in the fstests common scripts that might catch your test out.
Quoted 2 lines
Thanks for pointing out that, yes, better to rebase to latest fstests for-next branch.
Jeremy, for your convenience, my version of the change which is rebased on for-next, fixes the merge conflicts and uses ext4/057 instead of ext4/056 can be found here:
https://github.com/tytso/xfstests/commit/330bf72dc67dd39e0fd413ecea78ab18b5405fb9
- Ted
Quoted 24 lines · Theodore Ts'o
Quoted 9 lines · Zorro Lang
Quoted 2 lines
The kill -9 is needed, because otherwise the test will run for a **very** long time. The reason for it is because of the -n 999999 in
Sure, I mean:
kill -9 $fsstress_pid 2>/dev/null
waitNot remove the "kill" line :)
Ah yes, sorry, I misunderstood what you meant.
Quoted 10 lines
Quoted 9 lines
Also, Jeremy, it looks like you haven't updated your xfstests-dev repository in a few weeks. Since you started this project, ext4/056 has been assigned, and there has been some new helper programs added which caused patch conflicts in src/Makefile and in .gitignore. They were pretty trivial to fix up the patch conflicts (which I've done in my xfstests-dev tree), but it's best practice to rebase on top of origin/for-next and re-test just to make sure there haven't been some major change in the fstests common scripts that might catch your test out.
Quoted 8 lines
Quoted 2 lines
Thanks for pointing out that, yes, better to rebase to latest fstests for-next branch.
Jeremy, for your convenience, my version of the change which is rebased on for-next, fixes the merge conflicts and uses ext4/057 instead of ext4/056 can be found here:
https://github.com/tytso/xfstests/commit/330bf72dc67dd39e0fd413ecea78ab18b5405fb9
Great. Thank you everyone! I'll use your merged version, fix the wait line, and add Reviewed-by: Zorro Lang <[email protected]>.
Quoted 2 lines
- Ted