Skip to contentSkip to conversation

generic: add test for garbled stat output from negative nanoseconds

PATCH In review koraynilay 2 participants 3 messages Nov 21

lore
Share & tools
lore

koraynilay

author
#0 lore raw

PATCH v2 1/1 generic: add test for garbled stat output from negative nanoseconds

Some filesystems don't use set_normalized_timespec64() to handle their timestamps format conversion, leading to a garbled stat command output for timestamps.

For example:

     from cache:   1960-01-01 01:01:01.123456789 +0000

after remount: 1960-01-01 01:01:02.()*+,--00 +0000

This bug also violates a struct timespec64 invariant, as documented in the set_normalized_timespec64() kernel-doc:
[...]
* Note: The tv_nsec part is always in the range of 0 <= tv_nsec <
* NSEC_PER_SEC.
* For negative values only the tv_sec field is negative ! */

Example patches that fix the problem[1][2].

---
v2 changes:
- use _fixed_by_fs_commit for ksmbd and cifs
- add _fixed_by_fs_commit for ntfs3
- fix cleanup function
- remove useless comment
- wrap more the other comments so they're easier to read

Thanks.

 tests/generic/812     | 48 +++++++++++++++++++++++++++++++++++++++++++
 tests/generic/812.out |  3 +++
 2 files changed, 51 insertions(+)
 create mode 100755 tests/generic/812
 create mode 100644 tests/generic/812.out
tests/generic/812 new file +48 -0
new file mode 100755
index 00000000..eb76f2e5
@@ -0,0 +1,48 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2026 koraynilay.  All Rights Reserved.
+#
+# FS QA Test No. 812
+#
+# Based on generic/258, but tests timestamps prior to epoch with sub-second
+# precision.
+#
+# ksmbd, cifs, ntfs3 and other filesystem drivers violate timespec64
+# non-negative nanoseconds constraint, garbling output of stat.
+#
+. ./common/preamble
+_begin_fstest auto quick bigtime
+
+_fixed_by_fs_commit cifs 6a75ce0866bb \
+	"smb: client: fix negative timestamps being read from smb"
+_fixed_by_fs_commit cifs f2f961e0983c \
+	"ksmbd: fix negative timestamps being saved to disk"
+_fixed_by_fs_commit ntfs3 xxxxxxxxxxxx \
+	"ntfs3: fix negative timestamps corrupting stat output"
+
+_require_test
+_require_negative_timestamps
+
+TESTFILE=$TEST_DIR/timestamp-test.txt
+# we use 123456700 instead of 123456789 because some
+# filesystems support only 100ns precision
+DATE="1960-01-01 01:01:01.123456700 UTC"
+
+_cleanup()
+{
+	cd /
+	rm -f "$TESTFILE"
+}
+
+# create a file with a timestamp prior to the
+# epoch with nanosecond precision
+touch -d "$DATE" "$TESTFILE"
+
+TZ=UTC stat -c "%y" "$TESTFILE"
+
+# unmount, remount, and check the timestamp
+_test_cycle_mount
+
+TZ=UTC stat -c "%y" "$TESTFILE"
+
+_exit 0
tests/generic/812.out new file +3 -0
new file mode 100644
index 00000000..f0267fbb
@@ -0,0 +1,3 @@
+QA output created by 812
+1960-01-01 01:01:01.123456700 +0000
+1960-01-01 01:01:01.123456700 +0000

base-commit: 71641219a506ac62bfa231592394265ca812670e

Signature · 1 line
2.56.0
Reviewed-by on v2 1/1
#1 lore raw
Quoted 69 lines · koraynilay

Some filesystems don't use set_normalized_timespec64() to handle their timestamps format conversion, leading to a garbled stat command output for timestamps.

For example:

     from cache:   1960-01-01 01:01:01.123456789 +0000

after remount: 1960-01-01 01:01:02.()*+,--00 +0000

This bug also violates a struct timespec64 invariant, as documented in the set_normalized_timespec64() kernel-doc:
[...]
* Note: The tv_nsec part is always in the range of 0 <= tv_nsec <
* NSEC_PER_SEC.
* For negative values only the tv_sec field is negative ! */

Example patches that fix the problem[1][2].

---
v2 changes:
- use _fixed_by_fs_commit for ksmbd and cifs
- add _fixed_by_fs_commit for ntfs3
- fix cleanup function
- remove useless comment
- wrap more the other comments so they're easier to read

Thanks.

 tests/generic/812     | 48 +++++++++++++++++++++++++++++++++++++++++++
 tests/generic/812.out |  3 +++
 2 files changed, 51 insertions(+)
 create mode 100755 tests/generic/812
 create mode 100644 tests/generic/812.out
tests/generic/812 new file +26 -0
new file mode 100755
index 00000000..eb76f2e5
@@ -0,0 +1,48 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2026 koraynilay.  All Rights Reserved.
+#
+# FS QA Test No. 812
+#
+# Based on generic/258, but tests timestamps prior to epoch with sub-second
+# precision.
+#
+# ksmbd, cifs, ntfs3 and other filesystem drivers violate timespec64
+# non-negative nanoseconds constraint, garbling output of stat.
+#
+. ./common/preamble
+_begin_fstest auto quick bigtime
+
+_fixed_by_fs_commit cifs 6a75ce0866bb \
+	"smb: client: fix negative timestamps being read from smb"
+_fixed_by_fs_commit cifs f2f961e0983c \
+	"ksmbd: fix negative timestamps being saved to disk"
+_fixed_by_fs_commit ntfs3 xxxxxxxxxxxx \
+	"ntfs3: fix negative timestamps corrupting stat output"
+
+_require_test
+_require_negative_timestamps
+
+TESTFILE=$TEST_DIR/timestamp-test.txt

tests/generic/812 line 36

Quoted 10 lines
tests/generic/812 new file +10 -0
+# we use 123456700 instead of 123456789 because some
+# filesystems support only 100ns precision
+DATE="1960-01-01 01:01:01.123456700 UTC"
+
+_cleanup()
+{
+	cd /
+	rm -f "$TESTFILE"
+}
+

Haha, I probably didn't explain that clearly. I didn't mean changing the rm line in _cleanup. What I meant forcefully running rm -f $TESTFILE right before touch -d below, so that any existing file doesn't interfere with the test, since we don't strictly require TEST_DIR to be completely cleaned up after every test case run.

Don't worry about sending a new version, I'll fix it when I merge. Then...

Quoted 26 lines
tests/generic/812 new file +12 -0
+# create a file with a timestamp prior to the
+# epoch with nanosecond precision
+touch -d "$DATE" "$TESTFILE"
+
+TZ=UTC stat -c "%y" "$TESTFILE"
+
+# unmount, remount, and check the timestamp
+_test_cycle_mount
+
+TZ=UTC stat -c "%y" "$TESTFILE"
+
+_exit 0
tests/generic/812.out new file +3 -0
new file mode 100644
index 00000000..f0267fbb
@@ -0,0 +1,3 @@
+QA output created by 812
+1960-01-01 01:01:01.123456700 +0000
+1960-01-01 01:01:01.123456700 +0000

base-commit: 71641219a506ac62bfa231592394265ca812670e

Signature · 1 line
2.56.0
author
#2 lore raw
Quoted 78 lines · Zorro Lang
Quoted 77 lines · koraynilay

Some filesystems don't use set_normalized_timespec64() to handle their timestamps format conversion, leading to a garbled stat command output for timestamps.

For example:

     from cache:   1960-01-01 01:01:01.123456789 +0000

after remount: 1960-01-01 01:01:02.()*+,--00 +0000

This bug also violates a struct timespec64 invariant, as documented in the set_normalized_timespec64() kernel-doc:
[...]
* Note: The tv_nsec part is always in the range of 0 <= tv_nsec <
* NSEC_PER_SEC.
* For negative values only the tv_sec field is negative ! */

Example patches that fix the problem[1][2].

---
v2 changes:
- use _fixed_by_fs_commit for ksmbd and cifs
- add _fixed_by_fs_commit for ntfs3
- fix cleanup function
- remove useless comment
- wrap more the other comments so they're easier to read

Thanks.

 tests/generic/812     | 48 +++++++++++++++++++++++++++++++++++++++++++
 tests/generic/812.out |  3 +++
 2 files changed, 51 insertions(+)
 create mode 100755 tests/generic/812
 create mode 100644 tests/generic/812.out
tests/generic/812 new file +34 -0
new file mode 100755
index 00000000..eb76f2e5
@@ -0,0 +1,48 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2026 koraynilay.  All Rights Reserved.
+#
+# FS QA Test No. 812
+#
+# Based on generic/258, but tests timestamps prior to epoch with sub-second
+# precision.
+#
+# ksmbd, cifs, ntfs3 and other filesystem drivers violate timespec64
+# non-negative nanoseconds constraint, garbling output of stat.
+#
+. ./common/preamble
+_begin_fstest auto quick bigtime
+
+_fixed_by_fs_commit cifs 6a75ce0866bb \
+	"smb: client: fix negative timestamps being read from smb"
+_fixed_by_fs_commit cifs f2f961e0983c \
+	"ksmbd: fix negative timestamps being saved to disk"
+_fixed_by_fs_commit ntfs3 xxxxxxxxxxxx \
+	"ntfs3: fix negative timestamps corrupting stat output"
+
+_require_test
+_require_negative_timestamps
+
+TESTFILE=$TEST_DIR/timestamp-test.txt
+# we use 123456700 instead of 123456789 because some
+# filesystems support only 100ns precision
+DATE="1960-01-01 01:01:01.123456700 UTC"
+
+_cleanup()
+{
+	cd /
+	rm -f "$TESTFILE"
Quoted 8 lines

tests/generic/812 line 36

Quoted 2 lines
tests/generic/812 new file +2 -0
+}
+

Haha, I probably didn't explain that clearly. I didn't mean changing the rm line in _cleanup. What I meant forcefully running rm -f $TESTFILE right before touch -d below, so that any existing file doesn't interfere with the test, since we don't strictly require TEST_DIR to be completely cleaned up after every test case run.

Oh, yeah that makes more sense considering the template and other tests all keep the rm -f $tmp.* ahah.

Quoted 4 lines

Don't worry about sending a new version, I'll fix it when I merge. Then...

Thank you.

Best,
koraynilay

Quoted 26 lines
Quoted 26 lines
tests/generic/812 new file +12 -0
+# create a file with a timestamp prior to the
+# epoch with nanosecond precision
+touch -d "$DATE" "$TESTFILE"
+
+TZ=UTC stat -c "%y" "$TESTFILE"
+
+# unmount, remount, and check the timestamp
+_test_cycle_mount
+
+TZ=UTC stat -c "%y" "$TESTFILE"
+
+_exit 0
tests/generic/812.out new file +3 -0
new file mode 100644
index 00000000..f0267fbb
@@ -0,0 +1,3 @@
+QA output created by 812
+1960-01-01 01:01:01.123456700 +0000
+1960-01-01 01:01:01.123456700 +0000

base-commit: 71641219a506ac62bfa231592394265ca812670e

Signature · 1 line
2.56.0