Re: [PATCH] e2fsck: Correct ext4 dates generated by old kernels

[David Turner <novalis@novalis.org>]

[-- Attachment #1: Type: text/plain, Size: 1188 bytes --]

On Mon, 2016-03-07 at 10:27 -0500, Theodore Ts'o wrote:
> On Mon, Mar 07, 2016 at 04:34:04AM -0700, Andreas Dilger wrote:
> > On maint or master? I'd applied and tested it on maint before sending it.
> 
> At least on my tree David Turner's 64-bit date changes only landed on
> the master/next branch.  It was never on the maint branch....
> 
>     			    	      	      - Ted


Sorry for the delay in responding; I hadn't had time to look at this
until today.  I've figured out the problem.
In commit 188960ea4, the following change was made:

@@ -224,15 +232,19 @@ time_t string_to_time(const char *arg)
                /* interpret it as an integer */
                arg++;
        fallback:
-               ret = strtoul(arg, &tmp, 0);
+               ret = strtoll(arg+1, &tmp, 0);
                if (*tmp)


This is wrong at least for the non-goto case, because arg was just
incremented just above.  So it removes the first character *after* the @
as well.  I think it's wrong in the general case too; there's no reason
to skip the first character.

I've attached new version of the patch which corrects this error (it's
against master -- let me know if that's not right).

Thanks.

[-- Attachment #2: 0001-e2fsck-Correct-ext4-dates-generated-by-old-kernels.patch --]
[-- Type: text/x-patch, Size: 10125 bytes --]

>From 9447baaf5a111dbecf03f2e2be470be02acad2f1 Mon Sep 17 00:00:00 2001
From: David Turner <novalis@novalis.org>
Date: Thu, 18 Feb 2016 12:57:56 -0700
Subject: [PATCH] e2fsck: Correct ext4 dates generated by old kernels

Older kernels on 64-bit machines would incorrectly encode pre-1970
ext4 dates as post-2311 dates.  Detect and correct this (assuming the
current date is before 2242).

Include tests for this, as well as changes to debugfs to correctly
set crtimes.

Fix an off-by-one bug in debugfs's literal unix time parsing.

Signed-off-by: David Turner <novalis@novalis.org>

- ext2_fs.h: declare EXT4_EPOCH_BITS/EXT4_EPOCH_MASK like the kernel
	instead of in a separate header in an unusual location
- problem.h: move PR_1_EA_TIME_OUT_OF_RANGE to avoid master conflict
- problem.c: fix PR_1_EA_TIME_OUT_OF_RANGE PR_*_OK flag usage
- f_pre_1970_date_encoding/script: run debugfs less often,
	use $MKE2FS instead of mkfs.ext4, fit within 80 columns

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
---
 debugfs/util.c                        |  2 +-
 e2fsck/pass1.c                        | 41 +++++++++++++++
 e2fsck/problem.c                      |  5 ++
 e2fsck/problem.h                      |  3 ++
 lib/ext2fs/ext2_fs.h                  |  3 ++
 tests/f_pre_1970_date_encoding/expect | 45 +++++++++++++++++
 tests/f_pre_1970_date_encoding/name   |  1 +
 tests/f_pre_1970_date_encoding/script | 94 +++++++++++++++++++++++++++++++++++
 8 files changed, 193 insertions(+), 1 deletion(-)
 create mode 100644 tests/f_pre_1970_date_encoding/expect
 create mode 100644 tests/f_pre_1970_date_encoding/name
 create mode 100644 tests/f_pre_1970_date_encoding/script

diff --git a/debugfs/util.c b/debugfs/util.c
index 770e7e1..bd5de79 100644
--- a/debugfs/util.c
+++ b/debugfs/util.c
@@ -232,7 +232,7 @@ extern __s64 string_to_time(const char *arg)
 		/* interpret it as an integer */
 		arg++;
 	fallback:
-		ret = strtoll(arg+1, &tmp, 0);
+		ret = strtoll(arg, &tmp, 0);
 		if (*tmp)
 			return -1;
 		return ret;
diff --git a/e2fsck/pass1.c b/e2fsck/pass1.c
index 631d735..60ed41f 100644
--- a/e2fsck/pass1.c
+++ b/e2fsck/pass1.c
@@ -448,6 +448,21 @@ fix:
 				EXT2_INODE_SIZE(sb), "pass1");
 }
 
+static int check_inode_extra_negative_epoch(__u32 xtime, __u32 extra) {
+	return (xtime & (1 << 31)) != 0 &&
+		(extra & EXT4_EPOCH_MASK) == EXT4_EPOCH_MASK;
+}
+
+#define CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, xtime) \
+	check_inode_extra_negative_epoch(inode->i_##xtime, \
+					 inode->i_##xtime##_extra)
+
+/* When today's date is earlier than 2242, we assume that atimes,
+ * ctimes, crtimes, and mtimes with years in the range 2310..2378 are
+ * actually pre-1970 dates mis-encoded.
+ */
+#define EXT4_EXTRA_NEGATIVE_DATE_CUTOFF 2 * (1LL << 32)
+
 static void check_inode_extra_space(e2fsck_t ctx, struct problem_context *pctx)
 {
 	struct ext2_super_block *sb = ctx->fs->super;
@@ -492,6 +507,32 @@ static void check_inode_extra_space(e2fsck_t ctx, struct problem_context *pctx)
 		/* it seems inode has an extended attribute(s) in body */
 		check_ea_in_inode(ctx, pctx);
 	}
+
+	/*
+	 * If the inode's extended atime (ctime, crtime, mtime) is stored in
+	 * the old, invalid format, repair it.
+	 */
+	if (sizeof(time_t) > 4 && ctx->now < EXT4_EXTRA_NEGATIVE_DATE_CUTOFF &&
+	    (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, atime) ||
+	     CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, ctime) ||
+	     CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, crtime) ||
+	     CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, mtime))) {
+
+		if (!fix_problem(ctx, PR_1_EA_TIME_OUT_OF_RANGE, pctx))
+			return;
+
+		if (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, atime))
+			inode->i_atime_extra &= ~EXT4_EPOCH_MASK;
+		if (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, ctime))
+			inode->i_ctime_extra &= ~EXT4_EPOCH_MASK;
+		if (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, crtime))
+			inode->i_crtime_extra &= ~EXT4_EPOCH_MASK;
+		if (CHECK_INODE_EXTRA_NEGATIVE_EPOCH(inode, mtime))
+			inode->i_mtime_extra &= ~EXT4_EPOCH_MASK;
+		e2fsck_write_inode_full(ctx, pctx->ino, pctx->inode,
+					EXT2_INODE_SIZE(sb), "pass1");
+	}
+
 }
 
 /*
diff --git a/e2fsck/problem.c b/e2fsck/problem.c
index b39eab2..1e645e4 100644
--- a/e2fsck/problem.c
+++ b/e2fsck/problem.c
@@ -1124,6 +1124,11 @@ static struct e2fsck_problem problem_table[] = {
 	  N_("@i %i has corrupt @x header.  "),
 	  PROMPT_CLEAR_INODE, 0 },
 
+	/* Timestamp(s) on inode beyond 2310-04-04 are likely pre-1970. */
+	{ PR_1_EA_TIME_OUT_OF_RANGE,
+	  N_("Timestamp(s) on @i %i beyond 2310-04-04 are likely pre-1970.\n"),
+	  PROMPT_FIX, PR_PREEN_OK | PR_NO_OK },
+
 	/* Pass 1b errors */
 
 	/* Pass 1B: Rescan for duplicate/bad blocks */
diff --git a/e2fsck/problem.h b/e2fsck/problem.h
index b3f5b8f..edc381d 100644
--- a/e2fsck/problem.h
+++ b/e2fsck/problem.h
@@ -657,6 +657,9 @@ struct problem_context {
 /* Missing extent header */
 #define PR_1_MISSING_EXTENT_HEADER		0x010081
 
+/* Timestamp(s) on inode beyond 2310-04-04 are likely pre-1970. */
+#define PR_1_EA_TIME_OUT_OF_RANGE		0x010082
+
 /*
  * Pass 1b errors
  */
diff --git a/lib/ext2fs/ext2_fs.h b/lib/ext2fs/ext2_fs.h
index cdb68e8..9918356 100644
--- a/lib/ext2fs/ext2_fs.h
+++ b/lib/ext2fs/ext2_fs.h
@@ -481,6 +481,9 @@ struct ext2_inode_large {
 	(offsetof(struct ext2_inode_large, i_checksum_hi) + sizeof(__u16) - \
 	 EXT2_GOOD_OLD_INODE_SIZE)
 
+#define EXT4_EPOCH_BITS 2
+#define EXT4_EPOCH_MASK ((1 << EXT4_EPOCH_BITS) - 1)
+
 #define i_dir_acl	i_size_high
 
 #define i_checksum_lo	osd2.linux2.l_i_checksum_lo
diff --git a/tests/f_pre_1970_date_encoding/expect b/tests/f_pre_1970_date_encoding/expect
new file mode 100644
index 0000000..1a71571
--- /dev/null
+++ b/tests/f_pre_1970_date_encoding/expect
@@ -0,0 +1,45 @@
+times for year-1909 =
+ ctime: 0x8e475440:00000003
+ atime: 0x8e475440:00000003
+ mtime: 0x8e475440:00000003
+crtime: 0x8e475440:00000003
+times for year-1979 =
+ ctime: 0x11db6940:00000000
+ atime: 0x11db6940:00000000
+ mtime: 0x11db6940:00000000
+crtime: 0x11db6940:00000000
+times for year-2039 =
+ ctime: 0x82a37b40:00000001
+ atime: 0x82a37b40:00000001
+ mtime: 0x82a37b40:00000001
+crtime: 0x82a37b40:00000001
+times for year-2139 =
+ ctime: 0x3e9b9940:00000001
+ atime: 0x3e9b9940:00000001
+ mtime: 0x3e9b9940:00000001
+crtime: 0x3e9b9940:00000001
+times for year-1909 =
+ ctime: 0x8e475440:00000000
+ atime: 0x8e475440:00000000
+ mtime: 0x8e475440:00000000
+crtime: 0x8e475440:00000000
+times for year-1979 =
+ ctime: 0x11db6940:00000000
+ atime: 0x11db6940:00000000
+ mtime: 0x11db6940:00000000
+crtime: 0x11db6940:00000000
+times for year-2039 =
+ ctime: 0x82a37b40:00000001
+ atime: 0x82a37b40:00000001
+ mtime: 0x82a37b40:00000001
+crtime: 0x82a37b40:00000001
+times for year-2139 =
+ ctime: 0x3e9b9940:00000001
+ atime: 0x3e9b9940:00000001
+ mtime: 0x3e9b9940:00000001
+crtime: 0x3e9b9940:00000001
+times for year-1909 =
+ ctime: 0x8e475440:00000003
+ atime: 0x8e475440:00000003
+ mtime: 0x8e475440:00000003
+crtime: 0x8e475440:00000003
diff --git a/tests/f_pre_1970_date_encoding/name b/tests/f_pre_1970_date_encoding/name
new file mode 100644
index 0000000..9805324
--- /dev/null
+++ b/tests/f_pre_1970_date_encoding/name
@@ -0,0 +1 @@
+correct mis-encoded pre-1970 dates
diff --git a/tests/f_pre_1970_date_encoding/script b/tests/f_pre_1970_date_encoding/script
new file mode 100644
index 0000000..e6d7bbd
--- /dev/null
+++ b/tests/f_pre_1970_date_encoding/script
@@ -0,0 +1,94 @@
+if ! test -x $DEBUGFS_EXE; then
+	echo "$test_name: $test_description: skipped (no debugfs)"
+	return 0
+fi
+
+OUT=$test_name.log
+TIMESTAMPS=$test_name.timestamps.log
+EXP=$test_dir/expect
+FSCK_OPT=-yf
+
+create_file_with_xtime_and_extra() {
+	name=$1
+	time=$2
+	extra=$3
+	{
+		echo "write /dev/null $name"
+		for xtime in atime ctime mtime crtime; do
+			echo "set_inode_field $name $xtime @$time"
+			echo "set_inode_field $name ${xtime}_extra $extra"
+		done
+	} | $DEBUGFS -w -f /dev/stdin $TMPFILE >> $OUT 2>&1
+}
+
+get_file_xtime_and_extra() {
+	name=$1
+	echo "times for $name =" >> $TIMESTAMPS
+	$DEBUGFS -R "stat $name" $TMPFILE 2>&1 | egrep '^( a| c| m|cr)time:' |
+		sed 's/ --.*//' >> $TIMESTAMPS
+}
+
+rm -f $OUT $TIMESTAMPS
+
+# create an empty ext4 filesystem with 256-byte inodes for testing
+> $TMPFILE
+echo mkfs.ext4 -b 1024 -q -I 256 $TMPFILE 5000 >> $OUT
+$MKE2FS -t ext4 -b 1024 -q -I 256 -F $TMPFILE 5000 >> $OUT 2>&1
+
+# this is a pre-1970 file encoded with the old encoding.
+# fsck should repair this
+create_file_with_xtime_and_extra year-1909 -1907928000 3
+
+# these are all already encoded correctly
+create_file_with_xtime_and_extra year-1979   299592000 0
+create_file_with_xtime_and_extra year-2039  2191752000 1
+create_file_with_xtime_and_extra year-2139  5345352000 1
+
+# confirm that the xtime is wrong on the pre-1970 file
+get_file_xtime_and_extra year-1909
+
+# and confirm that it is right on the remaining files
+get_file_xtime_and_extra year-1979
+get_file_xtime_and_extra year-2039
+get_file_xtime_and_extra year-2139
+
+# before we repair the filesystem, save off a copy so that
+# we can use it later
+
+cp -a $TMPFILE $TMPFILE.sav
+
+# repair the filesystem
+E2FSCK_TIME=1386393539 $FSCK $FSCK_OPT $TMPFILE >> $OUT 2>&1
+
+# check that the dates and xtime_extra on the file is now correct
+get_file_xtime_and_extra year-1909
+
+# check that the remaining dates have not been altered
+get_file_xtime_and_extra year-1979
+get_file_xtime_and_extra year-2039
+get_file_xtime_and_extra year-2139
+
+# now we need to check that after the year 2242, e2fsck does not
+# modify dates with extra_xtime=3
+
+# restore the unrepaired filesystem
+mv $TMPFILE.sav $TMPFILE
+
+#retry the repair
+E2FSCK_TIME=9270393539 $FSCK $FSCK_OPT $TMPFILE >> $OUT 2>&1
+
+# check that the 1909 file is unaltered (i.e. it has a post-2378 date)
+get_file_xtime_and_extra year-1909
+
+cmp -s $TIMESTAMPS $EXP
+status=$?
+
+if [ "$status" = 0 ]; then
+	echo "$test_name: $test_description: ok"
+	touch $test_name.ok
+else
+	echo "$test_name: $test_description: failed"
+	diff $DIFF_OPTS $EXP $TIMESTAMPS > $test_name.failed
+fi
+
+unset OUT TIMESTAMPS EXP FSCK_OPT
-- 
1.9.1