From 80eef2f52bb92ed740ac00eeb11ee7a3e7fffff2 Mon Sep 17 00:00:00 2001
From: Raghavendra Bhat <raghavendra@redhat.com>
Date: Mon, 11 Mar 2019 12:16:50 -0400
Subject: [PATCH 459/465] features/bit-rot: Unconditionally sign the files
during oneshot crawl
Currently bit-rot feature has an issue with disabling and reenabling it
on the same volume. Consider enabling bit-rot detection which goes on to
crawl and sign all the files present in the volume. Then some files are
modified and the bit-rot daemon goes on to sign the modified files with
the correct signature. Now, disable bit-rot feature. While, signing and
scrubbing are not happening, previous checksums of the files continue to
exist as extended attributes. Now, if some files with checksum xattrs get
modified, they are not signed with new signature as the feature is off.
At this point, if the feature is enabled again, the bit rot daemon will
go and sign those files which does not have any bit-rot specific xattrs
(i.e. those files which were created after bit-rot was disabled). Whereas
the files with bit-rot xattrs wont get signed with proper new checksum.
At this point if scrubber runs, it finds the on disk checksum and the actual
checksum of the file to be different (because the file got modified) and
marks the file as corrupted.
FIX:
The fix is to unconditionally sign the files when the bit-rot daemon
comes up (instead of skipping the files with bit-rot xattrs).
upstream fix:
> patch: https://review.gluster.org/#/c/glusterfs/+/22360/
> fixes: #bz1700078
> Change-ID: Iadfb47dd39f7e2e77f22d549a4a07a385284f4f5
Change-Id: Iadfb47dd39f7e2e77f22d549a4a07a385284f4f5
BUG: 1851424
Signed-off-by: Raghavendra M <raghavendra@redhat.com>
Reviewed-on: https://code.engineering.redhat.com/gerrit/208305
Tested-by: RHGS Build Bot <nigelb@redhat.com>
Reviewed-by: Sunil Kumar Heggodu Gopala Acharya <sheggodu@redhat.com>
---
tests/bitrot/bug-1700078.t | 87 +++++++++++++++++++++++++++++
xlators/features/bit-rot/src/bitd/bit-rot.c | 15 ++++-
2 files changed, 101 insertions(+), 1 deletion(-)
create mode 100644 tests/bitrot/bug-1700078.t
diff --git a/tests/bitrot/bug-1700078.t b/tests/bitrot/bug-1700078.t
new file mode 100644
index 0000000..f273742
--- /dev/null
+++ b/tests/bitrot/bug-1700078.t
@@ -0,0 +1,87 @@
+#!/bin/bash
+
+. $(dirname $0)/../include.rc
+. $(dirname $0)/../volume.rc
+
+cleanup;
+
+## Start glusterd
+TEST glusterd;
+TEST pidof glusterd;
+
+## Lets create and start the volume
+TEST $CLI volume create $V0 $H0:$B0/${V0}1
+TEST $CLI volume start $V0
+
+## Enable bitrot for volume $V0
+TEST $CLI volume bitrot $V0 enable
+
+## Turn off quick-read so that it wont cache the contents
+# of the file in lookup. For corrupted files, it might
+# end up in reads being served from the cache instead of
+# an error.
+TEST $CLI volume set $V0 performance.quick-read off
+
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" get_bitd_count
+
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT 'Active' scrub_status $V0 'State of scrub'
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT '/var/log/glusterfs/bitd.log' scrub_status $V0 'Bitrot error log location'
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT '/var/log/glusterfs/scrub.log' scrub_status $V0 'Scrubber error log location'
+
+## Set expiry-timeout to 1 sec
+TEST $CLI volume set $V0 features.expiry-time 1
+
+##Mount $V0
+TEST glusterfs --volfile-id=$V0 --volfile-server=$H0 $M0 --attribute-timeout=0 --entry-timeout=0
+
+## Turn off quick-read xlator so that, the contents are not served from the
+# quick-read cache.
+TEST $CLI volume set $V0 performance.quick-read off
+
+#Create sample file
+TEST `echo "1234" > $M0/FILE1`
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT 'trusted.bit-rot.signature' check_for_xattr 'trusted.bit-rot.signature' "/$B0/${V0}1/FILE1"
+
+##disable bitrot
+TEST $CLI volume bitrot $V0 disable
+
+## modify the file
+TEST `echo "write" >> $M0/FILE1`
+
+# unmount and remount when the file has to be accessed.
+# This is to ensure that, when the remount happens,
+# and the file is read, its contents are served from the
+# brick instead of cache.
+EXPECT_WITHIN $UMOUNT_TIMEOUT "Y" force_umount $M0
+
+##enable bitrot
+TEST $CLI volume bitrot $V0 enable
+
+# expiry time is set to 1 second. Hence sleep for 2 seconds for the
+# oneshot crawler to finish its crawling and sign the file properly.
+sleep 2
+
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" get_bitd_count
+
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT 'Active' scrub_status $V0 'State of scrub'
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT '/var/log/glusterfs/bitd.log' scrub_status $V0 'Bitrot error log location'
+EXPECT_WITHIN $PROCESS_UP_TIMEOUT '/var/log/glusterfs/scrub.log' scrub_status $V0 'Scrubber error log location'
+
+## Ondemand scrub
+TEST $CLI volume bitrot $V0 scrub ondemand
+
+# the scrub ondemand CLI command, just ensures that
+# the scrubber has received the ondemand scrub directive
+# and started. sleep for 2 seconds for scrubber to finish
+# crawling and marking file(s) as bad (if if finds that
+# corruption has happened) which are filesystem operations.
+sleep 2
+
+TEST ! getfattr -n 'trusted.bit-rot.bad-file' $B0/${V0}1/FILE1
+
+##Mount $V0
+TEST glusterfs --volfile-id=$V0 --volfile-server=$H0 $M0 --attribute-timeout=0 --entry-timeout=0
+
+TEST cat $M0/FILE1
+
+cleanup;
diff --git a/xlators/features/bit-rot/src/bitd/bit-rot.c b/xlators/features/bit-rot/src/bitd/bit-rot.c
index b8feef7..424c0d5 100644
--- a/xlators/features/bit-rot/src/bitd/bit-rot.c
+++ b/xlators/features/bit-rot/src/bitd/bit-rot.c
@@ -973,6 +973,7 @@ bitd_oneshot_crawl(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
int32_t ret = -1;
inode_t *linked_inode = NULL;
gf_boolean_t need_signing = _gf_false;
+ gf_boolean_t need_reopen = _gf_true;
GF_VALIDATE_OR_GOTO("bit-rot", subvol, out);
GF_VALIDATE_OR_GOTO("bit-rot", data, out);
@@ -1046,6 +1047,18 @@ bitd_oneshot_crawl(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
uuid_utoa(linked_inode->gfid));
} else {
need_signing = br_check_object_need_sign(this, xattr, child);
+
+ /*
+ * If we are here means, bitrot daemon has started. Is it just
+ * a simple restart of the daemon or is it started because the
+ * feature is enabled is something hard to determine. Hence,
+ * if need_signing is false (because bit-rot version and signature
+ * are present), then still go ahead and sign it.
+ */
+ if (!need_signing) {
+ need_signing = _gf_true;
+ need_reopen = _gf_true;
+ }
}
if (!need_signing)
@@ -1054,7 +1067,7 @@ bitd_oneshot_crawl(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
gf_msg(this->name, GF_LOG_INFO, 0, BRB_MSG_TRIGGER_SIGN,
"Triggering signing for %s [GFID: %s | Brick: %s]", loc.path,
uuid_utoa(linked_inode->gfid), child->brick_path);
- br_trigger_sign(this, child, linked_inode, &loc, _gf_true);
+ br_trigger_sign(this, child, linked_inode, &loc, need_reopen);
ret = 0;
--
1.8.3.1