From aab8a587360214432c4a2ab59134411f1d38c509 Mon Sep 17 00:00:00 2001 From: karthik-us Date: Wed, 9 Dec 2020 10:46:31 +0530 Subject: [PATCH 515/517] cluster/afr: Heal directory rename without rmdir/mkdir Problem1: When a directory is renamed while a brick is down entry-heal always did an rm -rf on that directory on the sink on old location and did mkdir and created the directory hierarchy again in the new location. This is inefficient. Problem2: Renamedir heal order may lead to a scenario where directory in the new location could be created before deleting it from old location leading to 2 directories with same gfid in posix. Fix: As part of heal, if oldlocation is healed first and is not present in source-brick always rename it into a hidden directory inside the sink-brick so that when heal is triggered in new-location shd can rename it from this hidden directory to the new-location. If new-location heal is triggered first and it detects that the directory already exists in the brick, then it should skip healing the directory until it appears in the hidden directory. Credits: Ravi for rename-data-loss.t script Upstream patch details: > Fixes: #1211 > Change-Id: I0cba2006f35cd03d314d18211ce0bd530e254843 > Signed-off-by: Pranith Kumar K Upstream patch: https://review.gluster.org/#/c/glusterfs/+/24373/ BUG: 1640148 Change-Id: I0cba2006f35cd03d314d18211ce0bd530e254843 Signed-off-by: karthik-us Reviewed-on: https://code.engineering.redhat.com/gerrit/220660 Tested-by: RHGS Build Bot Reviewed-by: Ravishankar Narayanankutty --- tests/afr.rc | 16 + tests/basic/afr/afr-anon-inode-no-quorum.t | 63 ++++ tests/basic/afr/afr-anon-inode.t | 114 ++++++ tests/basic/afr/entry-self-heal-anon-dir-off.t | 464 ++++++++++++++++++++++++ tests/basic/afr/rename-data-loss.t | 72 ++++ tests/bugs/replicate/bug-1744548-heal-timeout.t | 6 +- tests/features/trash.t | 74 ++-- xlators/cluster/afr/src/afr-common.c | 46 ++- xlators/cluster/afr/src/afr-dir-read.c | 12 +- xlators/cluster/afr/src/afr-self-heal-common.c | 182 ++++++++++ xlators/cluster/afr/src/afr-self-heal-entry.c | 206 +++++++++-- xlators/cluster/afr/src/afr-self-heal-name.c | 33 +- xlators/cluster/afr/src/afr-self-heal.h | 5 + xlators/cluster/afr/src/afr-self-heald.c | 178 ++++++++- xlators/cluster/afr/src/afr-self-heald.h | 2 +- xlators/cluster/afr/src/afr.c | 40 +- xlators/cluster/afr/src/afr.h | 11 + xlators/mgmt/glusterd/src/glusterd-volgen.c | 39 ++ xlators/mgmt/glusterd/src/glusterd-volume-set.c | 6 + 19 files changed, 1442 insertions(+), 127 deletions(-) create mode 100644 tests/basic/afr/afr-anon-inode-no-quorum.t create mode 100644 tests/basic/afr/afr-anon-inode.t create mode 100644 tests/basic/afr/entry-self-heal-anon-dir-off.t create mode 100644 tests/basic/afr/rename-data-loss.t diff --git a/tests/afr.rc b/tests/afr.rc index 35f352d..2417899 100644 --- a/tests/afr.rc +++ b/tests/afr.rc @@ -105,3 +105,19 @@ function get_quorum_type() local repl_id="$3" cat $m/.meta/graphs/active/$v-replicate-$repl_id/private|grep quorum-type|awk '{print $3}' } + +function afr_private_key_value() +{ + local v=$1 + local m=$2 + local replica_id=$3 + local key=$4 +#xargs at the end will strip leading spaces + grep -E "^${key} = " $m/.meta/graphs/active/${v}-replicate-${replica_id}/private | cut -f2 -d'=' | xargs +} + +function afr_anon_entry_count() +{ + local b=$1 + ls $b/.glusterfs-anonymous-inode* | wc -l +} diff --git a/tests/basic/afr/afr-anon-inode-no-quorum.t b/tests/basic/afr/afr-anon-inode-no-quorum.t new file mode 100644 index 0000000..896ba0c --- /dev/null +++ b/tests/basic/afr/afr-anon-inode-no-quorum.t @@ -0,0 +1,63 @@ +#!/bin/bash + +#Test that anon-inode entry is not cleaned up as long as there exists at least +#one valid entry +. $(dirname $0)/../../include.rc +. $(dirname $0)/../../volume.rc +. $(dirname $0)/../../afr.rc + +cleanup; + +TEST glusterd +TEST pidof glusterd +TEST $CLI volume create $V0 replica 2 $H0:$B0/${V0}{0,1} +TEST $CLI volume heal $V0 disable +TEST $CLI volume set $V0 performance.write-behind off +TEST $CLI volume set $V0 performance.read-ahead off +TEST $CLI volume set $V0 performance.readdir-ahead off +TEST $CLI volume set $V0 performance.open-behind off +TEST $CLI volume set $V0 performance.stat-prefetch off +TEST $CLI volume set $V0 performance.io-cache off +TEST $CLI volume set $V0 performance.quick-read off +TEST $CLI volume set $V0 cluster.entry-self-heal off +TEST $CLI volume start $V0 + +TEST $GFS --volfile-id=/$V0 --volfile-server=$H0 $M0 + +TEST touch $M0/a $M0/b + +gfid_a=$(gf_gfid_xattr_to_str $(gf_get_gfid_xattr $B0/${V0}0/a)) +gfid_b=$(gf_gfid_xattr_to_str $(gf_get_gfid_xattr $B0/${V0}0/b)) +TEST kill_brick $V0 $H0 $B0/${V0}0 +TEST mv $M0/a $M0/a-new +TEST mv $M0/b $M0/b-new + +TEST $CLI volume start $V0 force +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status $V0 0 +TEST ! ls $M0/a +TEST ! ls $M0/b +anon_inode_name=$(ls -a $B0/${V0}0 | grep glusterfs-anonymous-inode) +TEST stat $B0/${V0}0/$anon_inode_name/$gfid_a +TEST stat $B0/${V0}0/$anon_inode_name/$gfid_b +#Make sure index heal doesn't happen after enabling heal +TEST setfattr -x trusted.afr.$V0-client-0 $B0/${V0}1 +TEST rm -f $B0/${V0}1/.glusterfs/indices/xattrop/* +TEST $CLI volume heal $V0 enable +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "Y" glustershd_up_status +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status_in_shd $V0 0 +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status_in_shd $V0 1 +TEST $CLI volume heal $V0 +#Allow time for a scan +sleep 5 +TEST stat $B0/${V0}0/$anon_inode_name/$gfid_a +TEST stat $B0/${V0}0/$anon_inode_name/$gfid_b +inum_b=$(STAT_INO $B0/${V0}0/$anon_inode_name/$gfid_b) +TEST rm -f $M0/a-new +TEST stat $M0/b-new + +TEST $CLI volume heal $V0 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" afr_anon_entry_count $B0/${V0}0 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" afr_anon_entry_count $B0/${V0}1 +EXPECT "$inum_b" STAT_INO $B0/${V0}0/b-new + +cleanup diff --git a/tests/basic/afr/afr-anon-inode.t b/tests/basic/afr/afr-anon-inode.t new file mode 100644 index 0000000..f4cf37a --- /dev/null +++ b/tests/basic/afr/afr-anon-inode.t @@ -0,0 +1,114 @@ +#!/bin/bash +#Tests that afr-anon-inode test cases work fine as expected +#These are cases where in entry-heal/name-heal we dont know entry for an inode +#so these inodes are kept in a special directory + +. $(dirname $0)/../../include.rc +. $(dirname $0)/../../volume.rc +. $(dirname $0)/../../afr.rc + +cleanup; + +TEST glusterd +TEST pidof glusterd +TEST $CLI volume create $V0 replica 3 $H0:$B0/${V0}{0..2} +TEST $CLI volume set $V0 performance.quick-read off +TEST $CLI volume set $V0 performance.io-cache off +TEST $CLI volume set $V0 performance.write-behind off +TEST $CLI volume set $V0 performance.stat-prefetch off +TEST $CLI volume set $V0 performance.read-ahead off +TEST $CLI volume set $V0 performance.open-behind off +TEST $CLI volume start $V0 +TEST $GFS --volfile-id=$V0 --volfile-server=$H0 $M0; +EXPECT "^1$" afr_private_key_value $V0 $M0 0 "use-anonymous-inode" +TEST $CLI volume set $V0 cluster.use-anonymous-inode no +EXPECT_WITHIN $CONFIG_UPDATE_TIMEOUT "^0$" afr_private_key_value $V0 $M0 0 "use-anonymous-inode" +TEST $CLI volume set $V0 cluster.use-anonymous-inode yes +EXPECT_WITHIN $CONFIG_UPDATE_TIMEOUT "^1$" afr_private_key_value $V0 $M0 0 "use-anonymous-inode" +TEST mkdir -p $M0/d1/b $M0/d2/a +TEST kill_brick $V0 $H0 $B0/${V0}0 +TEST mv $M0/d2/a $M0/d1 +TEST mv $M0/d1/b $M0/d2 +TEST $CLI volume start $V0 force +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" get_pending_heal_count $V0 +anon_inode_name=$(ls -a $B0/${V0}0 | grep glusterfs-anonymous-inode) +TEST [[ -d $B0/${V0}1/$anon_inode_name ]] +TEST [[ -d $B0/${V0}2/$anon_inode_name ]] +anon_gfid=$(gf_get_gfid_xattr $B0/${V0}0/$anon_inode_name) +EXPECT "$anon_gfid" gf_get_gfid_xattr $B0/${V0}1/$anon_inode_name +EXPECT "$anon_gfid" gf_get_gfid_xattr $B0/${V0}2/$anon_inode_name + +TEST ! ls $M0/$anon_inode_name +EXPECT "^4$" echo $(ls -a $M0 | wc -l) + +#Test purging code path by shd +TEST $CLI volume heal $V0 disable +TEST mkdir $M0/l0 $M0/l1 $M0/l2 +TEST touch $M0/del-file $M0/del-file-nolink $M0/l0/file +TEST ln $M0/del-file $M0/del-file-link +TEST ln $M0/l0/file $M0/l1/file-link1 +TEST ln $M0/l0/file $M0/l2/file-link2 +TEST mkdir -p $M0/del-recursive-dir/d1 + +TEST kill_brick $V0 $H0 $B0/${V0}0 +TEST rm -f $M0/del-file $M0/del-file-nolink +TEST rm -rf $M0/del-recursive-dir +TEST mv $M0/d1/a $M0/d2 +TEST mv $M0/l0/file $M0/l0/renamed-file +TEST $CLI volume start $V0 force +EXPECT_WITHIN $CHILD_UP_TIMEOUT "^1$" afr_child_up_status $V0 0 + +nolink_gfid=$(gf_gfid_xattr_to_str $(gf_get_gfid_xattr $B0/${V0}0/del-file-nolink)) +link_gfid=$(gf_gfid_xattr_to_str $(gf_get_gfid_xattr $B0/${V0}0/del-file)) +dir_gfid=$(gf_gfid_xattr_to_str $(gf_get_gfid_xattr $B0/${V0}0/del-recursive-dir)) +rename_dir_gfid=$(gf_gfid_xattr_to_str $(gf_get_gfid_xattr $B0/${V0}0/d1/a)) +rename_file_gfid=$(gf_gfid_xattr_to_str $(gf_get_gfid_xattr $B0/${V0}0/l0/file)) +TEST ! stat $M0/del-file +TEST stat $B0/${V0}0/$anon_inode_name/$link_gfid +TEST ! stat $M0/del-file-nolink +TEST ! stat $B0/${V0}0/$anon_inode_name/$nolink_gfid +TEST ! stat $M0/del-recursive-dir +TEST stat $B0/${V0}0/$anon_inode_name/$dir_gfid +TEST ! stat $M0/d1/a +TEST stat $B0/${V0}0/$anon_inode_name/$rename_dir_gfid +TEST ! stat $M0/l0/file +TEST stat $B0/${V0}0/$anon_inode_name/$rename_file_gfid + +TEST kill_brick $V0 $H0 $B0/${V0}1 +TEST mv $M0/l1/file-link1 $M0/l1/renamed-file-link1 +TEST $CLI volume start $V0 force +EXPECT_WITHIN $CHILD_UP_TIMEOUT "^1$" afr_child_up_status $V0 1 +TEST ! stat $M0/l1/file-link1 +TEST stat $B0/${V0}1/$anon_inode_name/$rename_file_gfid + +TEST kill_brick $V0 $H0 $B0/${V0}2 +TEST mv $M0/l2/file-link2 $M0/l2/renamed-file-link2 +TEST $CLI volume start $V0 force +EXPECT_WITHIN $CHILD_UP_TIMEOUT "^1$" afr_child_up_status $V0 2 +TEST ! stat $M0/l2/file-link2 +TEST stat $B0/${V0}2/$anon_inode_name/$rename_file_gfid + +#Simulate only anon-inodes present in all bricks +TEST rm -f $M0/l0/renamed-file $M0/l1/renamed-file-link1 $M0/l2/renamed-file-link2 + +#Test that shd doesn't cleanup anon-inodes when some bricks are down +TEST kill_brick $V0 $H0 $B0/${V0}1 +TEST $CLI volume heal $V0 enable +$CLI volume heal $V0 +sleep 5 #Allow time for completion of one scan +TEST stat $B0/${V0}0/$anon_inode_name/$link_gfid +TEST stat $B0/${V0}0/$anon_inode_name/$rename_dir_gfid +TEST stat $B0/${V0}0/$anon_inode_name/$dir_gfid +rename_dir_inum=$(STAT_INO $B0/${V0}0/$anon_inode_name/$rename_dir_gfid) + +TEST $CLI volume start $V0 force +EXPECT_WITHIN $CHILD_UP_TIMEOUT "^1$" afr_child_up_status $V0 1 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" get_pending_heal_count $V0 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" afr_anon_entry_count $B0/${V0}0 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" afr_anon_entry_count $B0/${V0}1 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" afr_anon_entry_count $B0/${V0}2 + +#Test that rename indeed happened instead of rmdir/mkdir +renamed_dir_inum=$(STAT_INO $B0/${V0}0/d2/a) +EXPECT "$rename_dir_inum" echo $renamed_dir_inum +cleanup; diff --git a/tests/basic/afr/entry-self-heal-anon-dir-off.t b/tests/basic/afr/entry-self-heal-anon-dir-off.t new file mode 100644 index 0000000..0803a08 --- /dev/null +++ b/tests/basic/afr/entry-self-heal-anon-dir-off.t @@ -0,0 +1,464 @@ +#!/bin/bash + +#This file checks if missing entry self-heal and entry self-heal are working +#as expected. +. $(dirname $0)/../../include.rc +. $(dirname $0)/../../volume.rc +. $(dirname $0)/../../afr.rc + +cleanup; + +function get_file_type { + stat -c "%a:%F:%g:%t:%T:%u" $1 +} + +function diff_dirs { + diff <(ls $1 | sort) <(ls $2 | sort) +} + +function heal_status { + local f1_path="${1}/${3}" + local f2_path="${2}/${3}" + local insync="" + diff_dirs $f1_path $f2_path + if [ $? -eq 0 ]; + then + insync="Y" + else + insync="N" + fi + local xattr11=$(get_hex_xattr trusted.afr.$V0-client-0 $f1_path) + local xattr12=$(get_hex_xattr trusted.afr.$V0-client-1 $f1_path) + local xattr21=$(get_hex_xattr trusted.afr.$V0-client-0 $f2_path) + local xattr22=$(get_hex_xattr trusted.afr.$V0-client-1 $f2_path) + local dirty1=$(get_hex_xattr trusted.afr.dirty $f1_path) + local dirty2=$(get_hex_xattr trusted.afr.dirty $f2_path) + if [ -z $xattr11 ]; then xattr11="000000000000000000000000"; fi + if [ -z $xattr12 ]; then xattr12="000000000000000000000000"; fi + if [ -z $xattr21 ]; then xattr21="000000000000000000000000"; fi + if [ -z $xattr22 ]; then xattr22="000000000000000000000000"; fi + if [ -z $dirty1 ]; then dirty1="000000000000000000000000"; fi + if [ -z $dirty2 ]; then dirty2="000000000000000000000000"; fi + echo ${insync}${xattr11}${xattr12}${xattr21}${xattr22}${dirty1}${dirty2} +} + +function is_heal_done { + local zero_xattr="000000000000000000000000" + if [ "$(heal_status $@)" == "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" ]; + then + echo "Y" + else + echo "N" + fi +} + +function print_pending_heals { + local result=":" + for i in "$@"; + do + if [ "N" == $(is_heal_done $B0/${V0}0 $B0/${V0}1 $i) ]; + then + result="$result:$i" + fi + done +#To prevent any match for EXPECT_WITHIN, print a char non-existent in file-names + if [ $result == ":" ]; then result="~"; fi + echo $result +} + +zero_xattr="000000000000000000000000" +TEST glusterd +TEST pidof glusterd +TEST $CLI volume create $V0 replica 2 $H0:$B0/${V0}{0,1} +TEST $CLI volume heal $V0 disable +TEST $CLI volume set $V0 cluster.use-anonymous-inode off +TEST $CLI volume set $V0 performance.write-behind off +TEST $CLI volume set $V0 performance.read-ahead off +TEST $CLI volume set $V0 performance.readdir-ahead off +TEST $CLI volume set $V0 performance.open-behind off +TEST $CLI volume set $V0 performance.stat-prefetch off +TEST $CLI volume set $V0 performance.io-cache off +TEST $CLI volume set $V0 performance.quick-read off +TEST $CLI volume set $V0 cluster.data-self-heal on +TEST $CLI volume set $V0 cluster.metadata-self-heal on +TEST $CLI volume set $V0 cluster.entry-self-heal on +TEST $CLI volume start $V0 + +TEST $GFS --volfile-id=/$V0 --volfile-server=$H0 --use-readdirp=no $M0 +cd $M0 +#_me_ is dir on which missing entry self-heal happens, _heal is where dir self-heal happens +#spb is split-brain, fool is all fool + +#source_self_accusing means there exists source and a sink which self-accuses. +#This simulates failures where fops failed on the bricks without it going down. +#Something like EACCESS/EDQUOT etc + +TEST mkdir spb_heal spb spb_me_heal spb_me fool_heal fool_me v1_fool_heal v1_fool_me source_creations_heal source_deletions_heal source_creations_me source_deletions_me v1_dirty_me v1_dirty_heal source_self_accusing +TEST mkfifo source_deletions_heal/fifo +TEST mknod source_deletions_heal/block b 4 5 +TEST mknod source_deletions_heal/char c 1 5 +TEST touch source_deletions_heal/file +TEST ln -s source_deletions_heal/file source_deletions_heal/slink +TEST mkdir source_deletions_heal/dir1 +TEST mkdir source_deletions_heal/dir1/dir2 + +TEST mkfifo source_deletions_me/fifo +TEST mknod source_deletions_me/block b 4 5 +TEST mknod source_deletions_me/char c 1 5 +TEST touch source_deletions_me/file +TEST ln -s source_deletions_me/file source_deletions_me/slink +TEST mkdir source_deletions_me/dir1 +TEST mkdir source_deletions_me/dir1/dir2 + +TEST mkfifo source_self_accusing/fifo +TEST mknod source_self_accusing/block b 4 5 +TEST mknod source_self_accusing/char c 1 5 +TEST touch source_self_accusing/file +TEST ln -s source_self_accusing/file source_self_accusing/slink +TEST mkdir source_self_accusing/dir1 +TEST mkdir source_self_accusing/dir1/dir2 + +TEST kill_brick $V0 $H0 $B0/${V0}0 + +TEST touch spb_heal/0 spb/0 spb_me_heal/0 spb_me/0 fool_heal/0 fool_me/0 v1_fool_heal/0 v1_fool_me/0 v1_dirty_heal/0 v1_dirty_me/0 +TEST rm -rf source_deletions_heal/fifo source_deletions_heal/block source_deletions_heal/char source_deletions_heal/file source_deletions_heal/slink source_deletions_heal/dir1 +TEST rm -rf source_deletions_me/fifo source_deletions_me/block source_deletions_me/char source_deletions_me/file source_deletions_me/slink source_deletions_me/dir1 +TEST rm -rf source_self_accusing/fifo source_self_accusing/block source_self_accusing/char source_self_accusing/file source_self_accusing/slink source_self_accusing/dir1 + +#Test that the files are deleted +TEST ! stat $B0/${V0}1/source_deletions_heal/fifo +TEST ! stat $B0/${V0}1/source_deletions_heal/block +TEST ! stat $B0/${V0}1/source_deletions_heal/char +TEST ! stat $B0/${V0}1/source_deletions_heal/file +TEST ! stat $B0/${V0}1/source_deletions_heal/slink +TEST ! stat $B0/${V0}1/source_deletions_heal/dir1 +TEST ! stat $B0/${V0}1/source_deletions_me/fifo +TEST ! stat $B0/${V0}1/source_deletions_me/block +TEST ! stat $B0/${V0}1/source_deletions_me/char +TEST ! stat $B0/${V0}1/source_deletions_me/file +TEST ! stat $B0/${V0}1/source_deletions_me/slink +TEST ! stat $B0/${V0}1/source_deletions_me/dir1 +TEST ! stat $B0/${V0}1/source_self_accusing/fifo +TEST ! stat $B0/${V0}1/source_self_accusing/block +TEST ! stat $B0/${V0}1/source_self_accusing/char +TEST ! stat $B0/${V0}1/source_self_accusing/file +TEST ! stat $B0/${V0}1/source_self_accusing/slink +TEST ! stat $B0/${V0}1/source_self_accusing/dir1 + + +TEST mkfifo source_creations_heal/fifo +TEST mknod source_creations_heal/block b 4 5 +TEST mknod source_creations_heal/char c 1 5 +TEST touch source_creations_heal/file +TEST ln -s source_creations_heal/file source_creations_heal/slink +TEST mkdir source_creations_heal/dir1 +TEST mkdir source_creations_heal/dir1/dir2 + +TEST mkfifo source_creations_me/fifo +TEST mknod source_creations_me/block b 4 5 +TEST mknod source_creations_me/char c 1 5 +TEST touch source_creations_me/file +TEST ln -s source_creations_me/file source_creations_me/slink +TEST mkdir source_creations_me/dir1 +TEST mkdir source_creations_me/dir1/dir2 + +$CLI volume stop $V0 + +#simulate fool fool scenario for fool_* dirs +setfattr -x trusted.afr.$V0-client-0 $B0/${V0}1/{fool_heal,fool_me} +setfattr -n trusted.afr.dirty -v 0x000000000000000000000001 $B0/${V0}1/{fool_heal,fool_me} +setfattr -n trusted.afr.$V0-client-1 -v 0x000000000000000000000001 $B0/${V0}1/{v1_fool_heal,v1_fool_me} + +#Simulate v1-dirty(self-accusing but no pending ops on others) scenario for v1-dirty +setfattr -x trusted.afr.$V0-client-0 $B0/${V0}1/v1_dirty_{heal,me} +setfattr -n trusted.afr.$V0-client-1 -v 0x000000000000000000000001 $B0/${V0}1/v1_dirty_{heal,me} + +$CLI volume start $V0 force +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 0 +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 1 +TEST kill_brick $V0 $H0 $B0/${V0}1 + +TEST touch spb_heal/1 spb/0 spb_me_heal/1 spb_me/0 fool_heal/1 fool_me/1 v1_fool_heal/1 v1_fool_me/1 + +$CLI volume stop $V0 + +#simulate fool fool scenario for fool_* dirs +setfattr -x trusted.afr.$V0-client-1 $B0/${V0}0/{fool_heal,fool_me} +setfattr -n trusted.afr.dirty -v 0x000000000000000000000001 $B0/${V0}1/{fool_heal,fool_me} +setfattr -n trusted.afr.$V0-client-0 -v 0x000000000000000000000001 $B0/${V0}1/{v1_fool_heal,v1_fool_me} + +#simulate self-accusing for source_self_accusing +TEST setfattr -n trusted.afr.$V0-client-0 -v 0x000000000000000000000006 $B0/${V0}0/source_self_accusing + +$CLI volume start $V0 force +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 1 +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 0 + +# Check if conservative merges happened correctly on _me_ dirs +TEST stat spb_me_heal/1 +TEST stat $B0/${V0}0/spb_me_heal/1 +TEST stat $B0/${V0}1/spb_me_heal/1 + +TEST stat spb_me_heal/0 +TEST stat $B0/${V0}0/spb_me_heal/0 +TEST stat $B0/${V0}1/spb_me_heal/0 + +TEST stat fool_me/1 +TEST stat $B0/${V0}0/fool_me/1 +TEST stat $B0/${V0}1/fool_me/1 + +TEST stat fool_me/0 +TEST stat $B0/${V0}0/fool_me/0 +TEST stat $B0/${V0}1/fool_me/0 + +TEST stat v1_fool_me/0 +TEST stat $B0/${V0}0/v1_fool_me/0 +TEST stat $B0/${V0}1/v1_fool_me/0 + +TEST stat v1_fool_me/1 +TEST stat $B0/${V0}0/v1_fool_me/1 +TEST stat $B0/${V0}1/v1_fool_me/1 + +TEST stat v1_dirty_me/0 +TEST stat $B0/${V0}0/v1_dirty_me/0 +TEST stat $B0/${V0}1/v1_dirty_me/0 + +#Check if files that have gfid-mismatches in _me_ are giving EIO +TEST ! stat spb_me/0 + +#Check if stale files are deleted on access +TEST ! stat source_deletions_me/fifo +TEST ! stat $B0/${V0}0/source_deletions_me/fifo +TEST ! stat $B0/${V0}1/source_deletions_me/fifo +TEST ! stat source_deletions_me/block +TEST ! stat $B0/${V0}0/source_deletions_me/block +TEST ! stat $B0/${V0}1/source_deletions_me/block +TEST ! stat source_deletions_me/char +TEST ! stat $B0/${V0}0/source_deletions_me/char +TEST ! stat $B0/${V0}1/source_deletions_me/char +TEST ! stat source_deletions_me/file +TEST ! stat $B0/${V0}0/source_deletions_me/file +TEST ! stat $B0/${V0}1/source_deletions_me/file +TEST ! stat source_deletions_me/file +TEST ! stat $B0/${V0}0/source_deletions_me/file +TEST ! stat $B0/${V0}1/source_deletions_me/file +TEST ! stat source_deletions_me/dir1/dir2 +TEST ! stat $B0/${V0}0/source_deletions_me/dir1/dir2 +TEST ! stat $B0/${V0}1/source_deletions_me/dir1/dir2 +TEST ! stat source_deletions_me/dir1 +TEST ! stat $B0/${V0}0/source_deletions_me/dir1 +TEST ! stat $B0/${V0}1/source_deletions_me/dir1 + +#Test if the files created as part of access are healed correctly +r=$(get_file_type source_creations_me/fifo) +EXPECT "$r" get_file_type $B0/${V0}0/source_creations_me/fifo +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_me/fifo +TEST [ -p source_creations_me/fifo ] + +r=$(get_file_type source_creations_me/block) +EXPECT "$r" get_file_type $B0/${V0}0/source_creations_me/block +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_me/block +EXPECT "^4 5$" stat -c "%t %T" $B0/${V0}1/source_creations_me/block +EXPECT "^4 5$" stat -c "%t %T" $B0/${V0}0/source_creations_me/block +TEST [ -b source_creations_me/block ] + +r=$(get_file_type source_creations_me/char) +EXPECT "$r" get_file_type $B0/${V0}0/source_creations_me/char +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_me/char +EXPECT "^1 5$" stat -c "%t %T" $B0/${V0}1/source_creations_me/char +EXPECT "^1 5$" stat -c "%t %T" $B0/${V0}0/source_creations_me/char +TEST [ -c source_creations_me/char ] + +r=$(get_file_type source_creations_me/file) +EXPECT "$r" get_file_type $B0/${V0}0/source_creations_me/file +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_me/file +TEST [ -f source_creations_me/file ] + +r=$(get_file_type source_creations_me/slink) +EXPECT "$r" get_file_type $B0/${V0}0/source_creations_me/slink +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_me/slink +TEST [ -h source_creations_me/slink ] + +r=$(get_file_type source_creations_me/dir1/dir2) +EXPECT "$r" get_file_type $B0/${V0}0/source_creations_me/dir1/dir2 +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_me/dir1/dir2 +TEST [ -d source_creations_me/dir1/dir2 ] + +r=$(get_file_type source_creations_me/dir1) +EXPECT "$r" get_file_type $B0/${V0}0/source_creations_me/dir1 +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_me/dir1 +TEST [ -d source_creations_me/dir1 ] + +#Trigger heal and check _heal dirs are healed properly +#Trigger change in event generation number. That way inodes would get refreshed during lookup +TEST kill_brick $V0 $H0 $B0/${V0}1 +$CLI volume start $V0 force +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 1 +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 0 + +TEST stat spb_heal +TEST stat spb_me_heal +TEST stat fool_heal +TEST stat fool_me +TEST stat v1_fool_heal +TEST stat v1_fool_me +TEST stat source_deletions_heal +TEST stat source_deletions_me +TEST stat source_self_accusing +TEST stat source_creations_heal +TEST stat source_creations_me +TEST stat v1_dirty_heal +TEST stat v1_dirty_me +TEST $CLI volume stop $V0 +TEST rm -rf $B0/${V0}{0,1}/.glusterfs/indices/xattrop/* + +$CLI volume start $V0 +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 1 +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 0 + +#Create base entry in indices/xattrop +echo "Data" > $M0/FILE +rm -f $M0/FILE +EXPECT "1" count_index_entries $B0/${V0}0 +EXPECT "1" count_index_entries $B0/${V0}1 + +TEST $CLI volume stop $V0; + +#Create entries for fool_heal and fool_me to ensure they are fully healed and dirty xattrs erased, before triggering index heal +create_brick_xattrop_entry $B0/${V0}0 fool_heal fool_me source_creations_heal/dir1 + +$CLI volume start $V0 +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 1 +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "1" afr_child_up_status $V0 0 + +$CLI volume heal $V0 enable +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "Y" glustershd_up_status +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status_in_shd $V0 0 +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status_in_shd $V0 1 + +TEST $CLI volume heal $V0; +EXPECT_WITHIN $HEAL_TIMEOUT "~" print_pending_heals spb_heal spb_me_heal fool_heal fool_me v1_fool_heal v1_fool_me source_deletions_heal source_deletions_me source_creations_heal source_creations_me v1_dirty_heal v1_dirty_me source_self_accusing + +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 spb_heal +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 spb_me_heal +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 fool_heal +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 fool_me +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 v1_fool_heal +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 v1_fool_me +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 source_deletions_heal +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 source_deletions_me +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 source_self_accusing +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 source_creations_heal +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 source_creations_me +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 v1_dirty_heal +EXPECT "Y${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}${zero_xattr}" heal_status $B0/${V0}0 $B0/${V0}1 v1_dirty_me + +#Don't access the files/dirs from mount point as that may cause self-heals +# Check if conservative merges happened correctly on heal dirs +TEST stat $B0/${V0}0/spb_heal/1 +TEST stat $B0/${V0}1/spb_heal/1 + +TEST stat $B0/${V0}0/spb_heal/0 +TEST stat $B0/${V0}1/spb_heal/0 + +TEST stat $B0/${V0}0/fool_heal/1 +TEST stat $B0/${V0}1/fool_heal/1 + +TEST stat $B0/${V0}0/fool_heal/0 +TEST stat $B0/${V0}1/fool_heal/0 + +TEST stat $B0/${V0}0/v1_fool_heal/0 +TEST stat $B0/${V0}1/v1_fool_heal/0 + +TEST stat $B0/${V0}0/v1_fool_heal/1 +TEST stat $B0/${V0}1/v1_fool_heal/1 + +TEST stat $B0/${V0}0/v1_dirty_heal/0 +TEST stat $B0/${V0}1/v1_dirty_heal/0 + +#Check if files that have gfid-mismatches in spb are giving EIO +TEST ! stat spb/0 + +#Check if stale files are deleted on access +TEST ! stat $B0/${V0}0/source_deletions_heal/fifo +TEST ! stat $B0/${V0}1/source_deletions_heal/fifo +TEST ! stat $B0/${V0}0/source_deletions_heal/block +TEST ! stat $B0/${V0}1/source_deletions_heal/block +TEST ! stat $B0/${V0}0/source_deletions_heal/char +TEST ! stat $B0/${V0}1/source_deletions_heal/char +TEST ! stat $B0/${V0}0/source_deletions_heal/file +TEST ! stat $B0/${V0}1/source_deletions_heal/file +TEST ! stat $B0/${V0}0/source_deletions_heal/file +TEST ! stat $B0/${V0}1/source_deletions_heal/file +TEST ! stat $B0/${V0}0/source_deletions_heal/dir1/dir2 +TEST ! stat $B0/${V0}1/source_deletions_heal/dir1/dir2 +TEST ! stat $B0/${V0}0/source_deletions_heal/dir1 +TEST ! stat $B0/${V0}1/source_deletions_heal/dir1 + +#Check if stale files are deleted on access +TEST ! stat $B0/${V0}0/source_self_accusing/fifo +TEST ! stat $B0/${V0}1/source_self_accusing/fifo +TEST ! stat $B0/${V0}0/source_self_accusing/block +TEST ! stat $B0/${V0}1/source_self_accusing/block +TEST ! stat $B0/${V0}0/source_self_accusing/char +TEST ! stat $B0/${V0}1/source_self_accusing/char +TEST ! stat $B0/${V0}0/source_self_accusing/file +TEST ! stat $B0/${V0}1/source_self_accusing/file +TEST ! stat $B0/${V0}0/source_self_accusing/file +TEST ! stat $B0/${V0}1/source_self_accusing/file +TEST ! stat $B0/${V0}0/source_self_accusing/dir1/dir2 +TEST ! stat $B0/${V0}1/source_self_accusing/dir1/dir2 +TEST ! stat $B0/${V0}0/source_self_accusing/dir1 +TEST ! stat $B0/${V0}1/source_self_accusing/dir1 + +#Test if the files created as part of full self-heal correctly +r=$(get_file_type $B0/${V0}0/source_creations_heal/fifo) +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_heal/fifo +TEST [ -p $B0/${V0}0/source_creations_heal/fifo ] +EXPECT "^4 5$" stat -c "%t %T" $B0/${V0}1/source_creations_heal/block +EXPECT "^4 5$" stat -c "%t %T" $B0/${V0}0/source_creations_heal/block + +r=$(get_file_type $B0/${V0}0/source_creations_heal/block) +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_heal/block + +r=$(get_file_type $B0/${V0}0/source_creations_heal/char) +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_heal/char +EXPECT "^1 5$" stat -c "%t %T" $B0/${V0}1/source_creations_heal/char +EXPECT "^1 5$" stat -c "%t %T" $B0/${V0}0/source_creations_heal/char + +r=$(get_file_type $B0/${V0}0/source_creations_heal/file) +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_heal/file +TEST [ -f $B0/${V0}0/source_creations_heal/file ] + +r=$(get_file_type source_creations_heal/file $B0/${V0}0/slink) +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_heal/file slink +TEST [ -h $B0/${V0}0/source_creations_heal/slink ] + +r=$(get_file_type $B0/${V0}0/source_creations_heal/dir1/dir2) +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_heal/dir1/dir2 +TEST [ -d $B0/${V0}0/source_creations_heal/dir1/dir2 ] + +r=$(get_file_type $B0/${V0}0/source_creations_heal/dir1) +EXPECT "$r" get_file_type $B0/${V0}1/source_creations_heal/dir1 +TEST [ -d $B0/${V0}0/source_creations_heal/dir1 ] + +cd - + +#Anonymous directory shouldn't be created +TEST mkdir $M0/rename-dir +before_rename=$(STAT_INO $B0/${V0}1/rename-dir) +TEST kill_brick $V0 $H0 $B0/${V0}1 +TEST mv $M0/rename-dir $M0/new-name +TEST $CLI volume start $V0 force +#Since features.ctime is not enabled by default in downstream, the below test +#will fail. If ctime feature is enabled, there will be trusted.glusterfs.mdata +#xattr set which will differ for the parent in the gfid split-brain scenario +#and when lookup is triggered, the gfid gets added to indices/xattrop leading +#the below test to pass in upstream. Hence commenting it here. +#'spb' is in split-brain so pending-heal-count will be 2 +#EXPECT_WITHIN $HEAL_TIMEOUT "^2$" get_pending_heal_count $V0 +after_rename=$(STAT_INO $B0/${V0}1/new-name) +EXPECT "0" echo $(ls -a $B0/${V0}0/ | grep anonymous-inode | wc -l) +EXPECT "0" echo $(ls -a $B0/${V0}1/ | grep anonymous-inode | wc -l) +EXPECT_NOT "$before_rename" echo $after_rename +cleanup diff --git a/tests/basic/afr/rename-data-loss.t b/tests/basic/afr/rename-data-loss.t new file mode 100644 index 0000000..256ee2a --- /dev/null +++ b/tests/basic/afr/rename-data-loss.t @@ -0,0 +1,72 @@ +#!/bin/bash +#Self-heal tests +. $(dirname $0)/../../include.rc +. $(dirname $0)/../../volume.rc +. $(dirname $0)/../../afr.rc + +cleanup; + +TEST glusterd +TEST pidof glusterd +TEST $CLI volume create $V0 replica 2 $H0:$B0/brick{0,1} +TEST $CLI volume set $V0 write-behind off +TEST $CLI volume set $V0 self-heal-daemon off +TEST $CLI volume set $V0 data-self-heal off +TEST $CLI volume set $V0 metadata-self-heal off +TEST $CLI volume set $V0 entry-self-heal off +TEST $CLI volume start $V0 +EXPECT 'Started' volinfo_field $V0 'Status' +TEST glusterfs --volfile-id=$V0 --volfile-server=$H0 $M0; + +cd $M0 +TEST `echo "line1" >> file1` +TEST mkdir dir1 +TEST mkdir dir2 +TEST mkdir -p dir1/dira/dirb +TEST `echo "line1">>dir1/dira/dirb/file1` +TEST mkdir delete_me +TEST `echo "line1" >> delete_me/file1` + +#brick0 has witnessed the second write while brick1 is down. +TEST kill_brick $V0 $H0 $B0/brick1 +TEST `echo "line2" >> file1` +TEST `echo "line2" >> dir1/dira/dirb/file1` +TEST `echo "line2" >> delete_me/file1` + +#Toggle the bricks that are up/down. +TEST $CLI volume start $V0 force +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status $V0 1 +TEST kill_brick $V0 $H0 $B0/brick0 + +#Rename when the 'source' brick0 for data-selfheals is down. +mv file1 file2 +mv dir1/dira dir2 + +#Delete a dir when brick0 is down. +rm -rf delete_me +cd - + +#Bring everything up and trigger heal +TEST $CLI volume set $V0 self-heal-daemon on +TEST $CLI volume start $V0 force +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status $V0 0 +EXPECT_WITHIN $PROCESS_UP_TIMEOUT "Y" glustershd_up_status +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status_in_shd $V0 0 +EXPECT_WITHIN $CHILD_UP_TIMEOUT "1" afr_child_up_status_in_shd $V0 1 +TEST $CLI volume heal $V0 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" get_pending_heal_count $V0 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" afr_anon_entry_count $B0/brick0 +EXPECT_WITHIN $HEAL_TIMEOUT "^0$" afr_anon_entry_count $B0/brick1 + +#Remount to avoid reading from caches +EXPECT_WITHIN $UMOUNT_TIMEOUT "Y" force_umount $M0 +TEST glusterfs --volfile-id=$V0 --volfile-server=$H0 $M0; +EXPECT "line2" tail -1 $M0/file2 +EXPECT "line2" tail -1 $M0/dir2/dira/dirb/file1 +TEST ! stat $M0/delete_me/file1 +TEST ! stat $M0/delete_me + +anon_inode_name=$(ls -a $B0/brick0 | grep glusterfs-anonymous-inode) +TEST [[ -d $B0/brick0/$anon_inode_name ]] +TEST [[ -d $B0/brick1/$anon_inode_name ]] +cleanup diff --git a/tests/bugs/replicate/bug-1744548-heal-timeout.t b/tests/bugs/replicate/bug-1744548-heal-timeout.t index c208112..0115350 100644 --- a/tests/bugs/replicate/bug-1744548-heal-timeout.t +++ b/tests/bugs/replicate/bug-1744548-heal-timeout.t @@ -25,14 +25,14 @@ TEST ! $CLI volume heal $V0 TEST $CLI volume profile $V0 start TEST $CLI volume profile $V0 info clear TEST $CLI volume heal $V0 enable -# Each brick does 3 opendirs, corresponding to dirty, xattrop and entry-changes -EXPECT_WITHIN $HEAL_TIMEOUT "^333$" get_cumulative_opendir_count +# Each brick does 4 opendirs, corresponding to dirty, xattrop and entry-changes, anonymous-inode +EXPECT_WITHIN 4 "^444$" get_cumulative_opendir_count # Check that a change in heal-timeout is honoured immediately. TEST $CLI volume set $V0 cluster.heal-timeout 5 sleep 10 # Two crawls must have happened. -EXPECT_WITHIN $HEAL_TIMEOUT "^999$" get_cumulative_opendir_count +EXPECT_WITHIN $CHILD_UP_TIMEOUT "^121212$" get_cumulative_opendir_count # shd must not heal if it is disabled and heal-timeout is changed. TEST $CLI volume heal $V0 disable diff --git a/tests/features/trash.t b/tests/features/trash.t index 472e909..da5b50b 100755 --- a/tests/features/trash.t +++ b/tests/features/trash.t @@ -94,105 +94,105 @@ wildcard_not_exists() { if [ $? -eq 0 ]; then echo "Y"; else echo "N"; fi } -# testing glusterd [1-3] +# testing glusterd TEST glusterd TEST pidof glusterd TEST $CLI volume info -# creating distributed volume [4] +# creating distributed volume TEST $CLI volume create $V0 $H0:$B0/${V0}{1,2} -# checking volume status [5-7] +# checking volume status EXPECT "$V0" volinfo_field $V0 'Volume Name' EXPECT 'Created' volinfo_field $V0 'Status' EXPECT '2' brick_count $V0 -# test without enabling trash translator [8] +# test without enabling trash translator TEST start_vol $V0 $M0 -# test on enabling trash translator [9-10] +# test on enabling trash translator TEST $CLI volume set $V0 features.trash on EXPECT 'on' volinfo_field $V0 'features.trash' -# files directly under mount point [11] +# files directly under mount point create_files $M0/file1 $M0/file2 TEST file_exists $V0 file1 file2 -# perform unlink [12] +# perform unlink TEST unlink_op file1 -# perform truncate [13] +# perform truncate TEST truncate_op file2 4 -# create files directory hierarchy and check [14] +# create files directory hierarchy and check mkdir -p $M0/1/2/3 create_files $M0/1/2/3/foo1 $M0/1/2/3/foo2 TEST file_exists $V0 1/2/3/foo1 1/2/3/foo2 -# perform unlink [15] +# perform unlink TEST unlink_op 1/2/3/foo1 -# perform truncate [16] +# perform truncate TEST truncate_op 1/2/3/foo2 4 # create a directory for eliminate pattern mkdir $M0/a -# set the eliminate pattern [17-18] +# set the eliminate pattern TEST $CLI volume set $V0 features.trash-eliminate-path /a EXPECT '/a' volinfo_field $V0 'features.trash-eliminate-path' -# create two files and check [19] +# create two files and check create_files $M0/a/test1 $M0/a/test2 TEST file_exists $V0 a/test1 a/test2 -# remove from eliminate pattern [20] +# remove from eliminate pattern rm -f $M0/a/test1 EXPECT "Y" wildcard_not_exists $M0/.trashcan/a/test1* -# truncate from eliminate path [21-23] +# truncate from eliminate path truncate -s 2 $M0/a/test2 TEST [ -e $M0/a/test2 ] TEST [ `ls -l $M0/a/test2 | awk '{print $5}'` -eq 2 ] EXPECT "Y" wildcard_not_exists $M0/.trashcan/a/test2* -# set internal op on [24-25] +# set internal op on TEST $CLI volume set $V0 features.trash-internal-op on EXPECT 'on' volinfo_field $V0 'features.trash-internal-op' -# again create two files and check [26] +# again create two files and check create_files $M0/inop1 $M0/inop2 TEST file_exists $V0 inop1 inop2 -# perform unlink [27] +# perform unlink TEST unlink_op inop1 -# perform truncate [28] +# perform truncate TEST truncate_op inop2 4 -# remove one brick and restart the volume [28-31] +# remove one brick and restart the volume TEST $CLI volume remove-brick $V0 $H0:$B0/${V0}2 force EXPECT_WITHIN $UMOUNT_TIMEOUT "Y" force_umount $M0 TEST $CLI volume stop $V0 TEST start_vol $V0 $M0 $M0/.trashcan -# again create two files and check [33] +# again create two files and check create_files $M0/rebal1 $M0/rebal2 TEST file_exists $V0 rebal1 rebal2 -# add one brick [34-35] +# add one brick TEST $CLI volume add-brick $V0 $H0:$B0/${V0}3 TEST [ -d $B0/${V0}3 ] -# perform rebalance [36] +# perform rebalance TEST $CLI volume rebalance $V0 start force EXPECT_WITHIN $REBALANCE_TIMEOUT "0" rebalance_completed #Find out which file was migrated to the new brick file_name=$(ls $B0/${V0}3/rebal*| xargs basename) -# check whether rebalance was succesful [37-40] +# check whether rebalance was succesful EXPECT "Y" wildcard_exists $B0/${V0}3/$file_name* EXPECT "Y" wildcard_exists $B0/${V0}1/.trashcan/internal_op/$file_name* @@ -201,52 +201,42 @@ EXPECT_WITHIN $UMOUNT_TIMEOUT "Y" force_umount $M0 # force required in case rebalance is not over TEST $CLI volume stop $V0 force -# create a replicated volume [41] +# create a replicated volume TEST $CLI volume create $V1 replica 2 $H0:$B0/${V1}{1,2} -# checking volume status [42-45] +# checking volume status EXPECT "$V1" volinfo_field $V1 'Volume Name' EXPECT 'Replicate' volinfo_field $V1 'Type' EXPECT 'Created' volinfo_field $V1 'Status' EXPECT '2' brick_count $V1 -# enable trash with options and start the replicate volume by disabling automatic self-heal [46-50] +# enable trash with options and start the replicate volume by disabling automatic self-heal TEST $CLI volume set $V1 features.trash on TEST $CLI volume set $V1 features.trash-internal-op on EXPECT 'on' volinfo_field $V1 'features.trash' EXPECT 'on' volinfo_field $V1 'features.trash-internal-op' TEST start_vol $V1 $M1 $M1/.trashcan -# mount and check for trash directory [51] +# mount and check for trash directory TEST [ -d $M1/.trashcan/internal_op ] -# create a file and check [52] +# create a file and check touch $M1/self TEST [ -e $B0/${V1}1/self -a -e $B0/${V1}2/self ] -# kill one brick and delete the file from mount point [53-54] +# kill one brick and delete the file from mount point kill_brick $V1 $H0 $B0/${V1}1 EXPECT_WITHIN ${PROCESS_UP_TIMEOUT} "1" online_brick_count rm -f $M1/self EXPECT "Y" wildcard_exists $B0/${V1}2/.trashcan/self* -# force start the volume and trigger the self-heal manually [55-57] -TEST $CLI volume start $V1 force -EXPECT_WITHIN $PROCESS_UP_TIMEOUT "2" online_brick_count -EXPECT_WITHIN $PROCESS_UP_TIMEOUT "Y" glustershd_up_status -# Since we created the file under root of the volume, it will be -# healed automatically - -# check for the removed file in trashcan [58] -EXPECT_WITHIN $HEAL_TIMEOUT "Y" wildcard_exists $B0/${V1}1/.trashcan/internal_op/self* - -# check renaming of trash directory through cli [59-62] +# check renaming of trash directory through cli TEST $CLI volume set $V0 trash-dir abc TEST start_vol $V0 $M0 $M0/abc TEST [ -e $M0/abc -a ! -e $M0/.trashcan ] EXPECT "Y" wildcard_exists $B0/${V0}1/abc/internal_op/rebal* -# ensure that rename and delete operation on trash directory fails [63-65] +# ensure that rename and delete operation on trash directory fails rm -rf $M0/abc/internal_op TEST [ -e $M0/abc/internal_op ] rm -rf $M0/abc/ diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c index 90b4f14..6f2da11 100644 --- a/xlators/cluster/afr/src/afr-common.c +++ b/xlators/cluster/afr/src/afr-common.c @@ -47,6 +47,41 @@ afr_quorum_errno(afr_private_t *priv) return ENOTCONN; } +gf_boolean_t +afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name, + pid_t pid) +{ + if (!__is_root_gfid(pargfid)) { + return _gf_false; + } + + if (strcmp(name, GF_REPLICATE_TRASH_DIR) == 0) { + /*For backward compatibility /.landfill is private*/ + return _gf_true; + } + + if (pid == GF_CLIENT_PID_GSYNCD) { + /*geo-rep needs to create/sync private directory on slave because + * it appears in changelog*/ + return _gf_false; + } + + if (pid == GF_CLIENT_PID_GLFS_HEAL || pid == GF_CLIENT_PID_SELF_HEALD) { + if (strcmp(name, priv->anon_inode_name) == 0) { + /* anonymous-inode dir is private*/ + return _gf_true; + } + } else { + if (strncmp(name, AFR_ANON_DIR_PREFIX, strlen(AFR_ANON_DIR_PREFIX)) == + 0) { + /* anonymous-inode dir prefix is private for geo-rep to work*/ + return _gf_true; + } + } + + return _gf_false; +} + int afr_fav_child_reset_sink_xattrs(void *opaque); @@ -3301,11 +3336,10 @@ afr_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req) return 0; } - if (__is_root_gfid(loc->parent->gfid)) { - if (!strcmp(loc->name, GF_REPLICATE_TRASH_DIR)) { - op_errno = EPERM; - goto out; - } + if (afr_is_private_directory(this->private, loc->parent->gfid, loc->name, + frame->root->pid)) { + op_errno = EPERM; + goto out; } local = AFR_FRAME_INIT(frame, op_errno); @@ -4832,6 +4866,7 @@ afr_priv_dump(xlator_t *this) priv->background_self_heal_count); gf_proc_dump_write("healers", "%d", priv->healers); gf_proc_dump_write("read-hash-mode", "%d", priv->hash_mode); + gf_proc_dump_write("use-anonymous-inode", "%d", priv->use_anon_inode); if (priv->quorum_count == AFR_QUORUM_AUTO) { gf_proc_dump_write("quorum-type", "auto"); } else if (priv->quorum_count == 0) { @@ -5792,6 +5827,7 @@ afr_priv_destroy(afr_private_t *priv) GF_FREE(priv->local); GF_FREE(priv->pending_key); GF_FREE(priv->children); + GF_FREE(priv->anon_inode); GF_FREE(priv->child_up); GF_FREE(priv->child_latency); LOCK_DESTROY(&priv->lock); diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c index 6307b63..d64b6a9 100644 --- a/xlators/cluster/afr/src/afr-dir-read.c +++ b/xlators/cluster/afr/src/afr-dir-read.c @@ -158,8 +158,8 @@ afr_validate_read_subvol(inode_t *inode, xlator_t *this, int par_read_subvol) } static void -afr_readdir_transform_entries(gf_dirent_t *subvol_entries, int subvol, - gf_dirent_t *entries, fd_t *fd) +afr_readdir_transform_entries(call_frame_t *frame, gf_dirent_t *subvol_entries, + int subvol, gf_dirent_t *entries, fd_t *fd) { int ret = -1; gf_dirent_t *entry = NULL; @@ -177,8 +177,8 @@ afr_readdir_transform_entries(gf_dirent_t *subvol_entries, int subvol, list_for_each_entry_safe(entry, tmp, &subvol_entries->list, list) { - if (__is_root_gfid(fd->inode->gfid) && - !strcmp(entry->d_name, GF_REPLICATE_TRASH_DIR)) { + if (afr_is_private_directory(priv, fd->inode->gfid, entry->d_name, + frame->root->pid)) { continue; } @@ -222,8 +222,8 @@ afr_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, } if (op_ret >= 0) - afr_readdir_transform_entries(subvol_entries, (long)cookie, &entries, - local->fd); + afr_readdir_transform_entries(frame, subvol_entries, (long)cookie, + &entries, local->fd); AFR_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, xdata); diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c index 9b6575f..0a8a7fd 100644 --- a/xlators/cluster/afr/src/afr-self-heal-common.c +++ b/xlators/cluster/afr/src/afr-self-heal-common.c @@ -2753,3 +2753,185 @@ afr_choose_source_by_policy(afr_private_t *priv, unsigned char *sources, out: return source; } + +static int +afr_anon_inode_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, + int32_t op_ret, int32_t op_errno, inode_t *inode, + struct iatt *buf, struct iatt *preparent, + struct iatt *postparent, dict_t *xdata) +{ + afr_local_t *local = frame->local; + int i = (long)cookie; + + local->replies[i].valid = 1; + local->replies[i].op_ret = op_ret; + local->replies[i].op_errno = op_errno; + if (op_ret == 0) { + local->op_ret = 0; + local->replies[i].poststat = *buf; + local->replies[i].preparent = *preparent; + local->replies[i].postparent = *postparent; + } + if (xdata) { + local->replies[i].xdata = dict_ref(xdata); + } + + syncbarrier_wake(&local->barrier); + return 0; +} + +int +afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode) +{ + call_frame_t *frame = NULL; + afr_local_t *local = NULL; + afr_private_t *priv = this->private; + unsigned char *mkdir_on = alloca0(priv->child_count); + unsigned char *lookup_on = alloca0(priv->child_count); + loc_t loc = {0}; + int32_t op_errno = 0; + int32_t child_op_errno = 0; + struct iatt iatt = {0}; + dict_t *xdata = NULL; + uuid_t anon_inode_gfid = {0}; + int mkdir_count = 0; + int i = 0; + + /*Try to mkdir everywhere and return success if the dir exists on 'child' + */ + + if (!priv->use_anon_inode) { + op_errno = EINVAL; + goto out; + } + + frame = afr_frame_create(this, &op_errno); + if (op_errno) { + goto out; + } + local = frame->local; + if (!local->child_up[child]) { + /*Other bricks may need mkdir so don't error out yet*/ + child_op_errno = ENOTCONN; + } + gf_uuid_parse(priv->anon_gfid_str, anon_inode_gfid); + for (i = 0; i < priv->child_count; i++) { + if (!local->child_up[i]) + continue; + + if (priv->anon_inode[i]) { + mkdir_on[i] = 0; + } else { + mkdir_on[i] = 1; + mkdir_count++; + } + } + + if (mkdir_count == 0) { + *linked_inode = inode_find(this->itable, anon_inode_gfid); + if (*linked_inode) { + op_errno = 0; + goto out; + } + } + + loc.parent = inode_ref(this->itable->root); + loc.name = priv->anon_inode_name; + loc.inode = inode_new(this->itable); + if (!loc.inode) { + op_errno = ENOMEM; + goto out; + } + + xdata = dict_new(); + if (!xdata) { + op_errno = ENOMEM; + goto out; + } + + op_errno = -dict_set_gfuuid(xdata, "gfid-req", anon_inode_gfid, _gf_true); + if (op_errno) { + goto out; + } + + if (mkdir_count == 0) { + memcpy(lookup_on, local->child_up, priv->child_count); + goto lookup; + } + + AFR_ONLIST(mkdir_on, frame, afr_anon_inode_mkdir_cbk, mkdir, &loc, 0755, 0, + xdata); + + for (i = 0; i < priv->child_count; i++) { + if (!mkdir_on[i]) { + continue; + } + + if (local->replies[i].op_ret == 0) { + priv->anon_inode[i] = 1; + iatt = local->replies[i].poststat; + } else if (local->replies[i].op_ret < 0 && + local->replies[i].op_errno == EEXIST) { + lookup_on[i] = 1; + } else if (i == child) { + child_op_errno = local->replies[i].op_errno; + } + } + + if (AFR_COUNT(lookup_on, priv->child_count) == 0) { + goto link; + } + +lookup: + AFR_ONLIST(lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc, + xdata); + for (i = 0; i < priv->child_count; i++) { + if (!lookup_on[i]) { + continue; + } + + if (local->replies[i].op_ret == 0) { + if (gf_uuid_compare(anon_inode_gfid, + local->replies[i].poststat.ia_gfid) == 0) { + priv->anon_inode[i] = 1; + iatt = local->replies[i].poststat; + } else { + if (i == child) + child_op_errno = EINVAL; + gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_DATA, + "%s has gfid: %s", priv->anon_inode_name, + uuid_utoa(local->replies[i].poststat.ia_gfid)); + } + } else if (i == child) { + child_op_errno = local->replies[i].op_errno; + } + } +link: + if (!gf_uuid_is_null(iatt.ia_gfid)) { + *linked_inode = inode_link(loc.inode, loc.parent, loc.name, &iatt); + if (*linked_inode) { + op_errno = 0; + inode_lookup(*linked_inode); + } else { + op_errno = ENOMEM; + } + goto out; + } + +out: + if (xdata) + dict_unref(xdata); + loc_wipe(&loc); + /*child_op_errno takes precedence*/ + if (child_op_errno == 0) { + child_op_errno = op_errno; + } + + if (child_op_errno && *linked_inode) { + inode_unref(*linked_inode); + *linked_inode = NULL; + } + if (frame) + AFR_STACK_DESTROY(frame); + return -child_op_errno; +} diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c index 00b5b2d..20b07dd 100644 --- a/xlators/cluster/afr/src/afr-self-heal-entry.c +++ b/xlators/cluster/afr/src/afr-self-heal-entry.c @@ -16,54 +16,170 @@ #include #include -static int -afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name, - inode_t *inode, int child, struct afr_reply *replies) +int +afr_selfheal_entry_anon_inode(xlator_t *this, inode_t *dir, const char *name, + inode_t *inode, int child, + struct afr_reply *replies, + gf_boolean_t *anon_inode) { afr_private_t *priv = NULL; + afr_local_t *local = NULL; xlator_t *subvol = NULL; int ret = 0; + int i = 0; + char g[64] = {0}; + unsigned char *lookup_success = NULL; + call_frame_t *frame = NULL; + loc_t loc2 = { + 0, + }; loc_t loc = { 0, }; - char g[64]; priv = this->private; - subvol = priv->children[child]; + lookup_success = alloca0(priv->child_count); + uuid_utoa_r(replies[child].poststat.ia_gfid, g); + loc.inode = inode_new(inode->table); + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + + if (replies[child].poststat.ia_type == IA_IFDIR) { + /* This directory may have sub-directory hierarchy which may need to + * be preserved for subsequent heals. So unconditionally move the + * directory to anonymous-inode directory*/ + *anon_inode = _gf_true; + goto anon_inode; + } + + frame = afr_frame_create(this, &ret); + if (!frame) { + ret = -ret; + goto out; + } + local = frame->local; + gf_uuid_copy(loc.gfid, replies[child].poststat.ia_gfid); + AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc, + NULL); + for (i = 0; i < priv->child_count; i++) { + if (local->replies[i].op_ret == 0) { + lookup_success[i] = 1; + } else if (local->replies[i].op_errno != ENOENT && + local->replies[i].op_errno != ESTALE) { + ret = -local->replies[i].op_errno; + } + } + + if (priv->quorum_count) { + if (afr_has_quorum(lookup_success, this, NULL)) { + *anon_inode = _gf_true; + } + } else if (AFR_COUNT(lookup_success, priv->child_count) > 1) { + *anon_inode = _gf_true; + } else if (ret) { + goto out; + } + +anon_inode: + if (!*anon_inode) { + ret = 0; + goto out; + } loc.parent = inode_ref(dir); gf_uuid_copy(loc.pargfid, dir->gfid); loc.name = name; - loc.inode = inode_ref(inode); - if (replies[child].valid && replies[child].op_ret == 0) { - switch (replies[child].poststat.ia_type) { - case IA_IFDIR: - gf_msg(this->name, GF_LOG_WARNING, 0, - AFR_MSG_EXPUNGING_FILE_OR_DIR, - "expunging dir %s/%s (%s) on %s", uuid_utoa(dir->gfid), - name, uuid_utoa_r(replies[child].poststat.ia_gfid, g), - subvol->name); - ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL); - break; - default: - gf_msg(this->name, GF_LOG_WARNING, 0, - AFR_MSG_EXPUNGING_FILE_OR_DIR, - "expunging file %s/%s (%s) on %s", uuid_utoa(dir->gfid), - name, uuid_utoa_r(replies[child].poststat.ia_gfid, g), - subvol->name); - ret = syncop_unlink(subvol, &loc, NULL, NULL); - break; - } + ret = afr_anon_inode_create(this, child, &loc2.parent); + if (ret < 0) + goto out; + + loc2.name = g; + ret = syncop_rename(subvol, &loc, &loc2, NULL, NULL); + if (ret < 0) { + gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_EXPUNGING_FILE_OR_DIR, + "Rename to %s dir %s/%s (%s) on %s failed", + priv->anon_inode_name, uuid_utoa(dir->gfid), name, g, + subvol->name); + } else { + gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR, + "Rename to %s dir %s/%s (%s) on %s successful", + priv->anon_inode_name, uuid_utoa(dir->gfid), name, g, + subvol->name); } +out: loc_wipe(&loc); + loc_wipe(&loc2); + if (frame) { + AFR_STACK_DESTROY(frame); + } return ret; } int +afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name, + inode_t *inode, int child, struct afr_reply *replies) +{ + char g[64] = {0}; + afr_private_t *priv = NULL; + xlator_t *subvol = NULL; + int ret = 0; + loc_t loc = { + 0, + }; + gf_boolean_t anon_inode = _gf_false; + + priv = this->private; + subvol = priv->children[child]; + + if ((!replies[child].valid) || (replies[child].op_ret < 0)) { + /*Nothing to do*/ + ret = 0; + goto out; + } + + if (priv->use_anon_inode) { + ret = afr_selfheal_entry_anon_inode(this, dir, name, inode, child, + replies, &anon_inode); + if (ret < 0 || anon_inode) + goto out; + } + + loc.parent = inode_ref(dir); + loc.inode = inode_new(inode->table); + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + loc.name = name; + switch (replies[child].poststat.ia_type) { + case IA_IFDIR: + gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR, + "expunging dir %s/%s (%s) on %s", uuid_utoa(dir->gfid), name, + uuid_utoa_r(replies[child].poststat.ia_gfid, g), + subvol->name); + ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL); + break; + default: + gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR, + "expunging file %s/%s (%s) on %s", uuid_utoa(dir->gfid), + name, uuid_utoa_r(replies[child].poststat.ia_gfid, g), + subvol->name); + ret = syncop_unlink(subvol, &loc, NULL, NULL); + break; + } + +out: + loc_wipe(&loc); + return ret; +} + +int afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source, unsigned char *sources, inode_t *dir, const char *name, inode_t *inode, @@ -76,6 +192,9 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source, loc_t srcloc = { 0, }; + loc_t anonloc = { + 0, + }; xlator_t *this = frame->this; afr_private_t *priv = NULL; dict_t *xdata = NULL; @@ -86,15 +205,18 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source, 0, }; unsigned char *newentry = NULL; + char iatt_uuid_str[64] = {0}; + char dir_uuid_str[64] = {0}; priv = this->private; iatt = &replies[source].poststat; + uuid_utoa_r(iatt->ia_gfid, iatt_uuid_str); if (iatt->ia_type == IA_INVAL || gf_uuid_is_null(iatt->ia_gfid)) { gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SELF_HEAL_FAILED, "Invalid ia_type (%d) or gfid(%s). source brick=%d, " "pargfid=%s, name=%s", - iatt->ia_type, uuid_utoa(iatt->ia_gfid), source, - uuid_utoa(dir->gfid), name); + iatt->ia_type, iatt_uuid_str, source, + uuid_utoa_r(dir->gfid, dir_uuid_str), name); ret = -EINVAL; goto out; } @@ -119,14 +241,24 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source, srcloc.inode = inode_ref(inode); gf_uuid_copy(srcloc.gfid, iatt->ia_gfid); - if (iatt->ia_type != IA_IFDIR) - ret = syncop_lookup(priv->children[dst], &srcloc, 0, 0, 0, 0); - if (iatt->ia_type == IA_IFDIR || ret == -ENOENT || ret == -ESTALE) { + ret = syncop_lookup(priv->children[dst], &srcloc, 0, 0, 0, 0); + if (ret == -ENOENT || ret == -ESTALE) { newentry[dst] = 1; ret = afr_selfheal_newentry_mark(frame, this, inode, source, replies, sources, newentry); if (ret) goto out; + } else if (ret == 0 && iatt->ia_type == IA_IFDIR && priv->use_anon_inode) { + // Try rename from hidden directory + ret = afr_anon_inode_create(this, dst, &anonloc.parent); + if (ret < 0) + goto out; + anonloc.inode = inode_ref(inode); + anonloc.name = iatt_uuid_str; + ret = syncop_rename(priv->children[dst], &anonloc, &loc, NULL, NULL); + if (ret == -ENOENT || ret == -ESTALE) + ret = -1; /*This sets 'mismatch' to true*/ + goto out; } mode = st_mode_from_ia(iatt->ia_prot, iatt->ia_type); @@ -165,6 +297,7 @@ out: GF_FREE(linkname); loc_wipe(&loc); loc_wipe(&srcloc); + loc_wipe(&anonloc); return ret; } @@ -580,6 +713,11 @@ afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd, priv = this->private; + if (afr_is_private_directory(priv, fd->inode->gfid, name, + GF_CLIENT_PID_SELF_HEALD)) { + return 0; + } + xattr = dict_new(); if (!xattr) return -ENOMEM; @@ -628,7 +766,7 @@ afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd, replies); if ((ret == 0) && (priv->esh_granular) && parent_idx_inode) { - ret = afr_shd_index_purge(subvol, parent_idx_inode, name, + ret = afr_shd_entry_purge(subvol, parent_idx_inode, name, inode->ia_type); /* Why is ret force-set to 0? We do not care about * index purge failing for full heal as it is quite @@ -758,10 +896,6 @@ afr_selfheal_entry_do_subvol(call_frame_t *frame, xlator_t *this, fd_t *fd, if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, "..")) continue; - if (__is_root_gfid(fd->inode->gfid) && - !strcmp(entry->d_name, GF_REPLICATE_TRASH_DIR)) - continue; - ret = afr_selfheal_entry_dirent(iter_frame, this, fd, entry->d_name, loc.inode, subvol, local->need_full_crawl); @@ -824,7 +958,7 @@ afr_selfheal_entry_granular_dirent(xlator_t *subvol, gf_dirent_t *entry, /* The name indices under the pgfid index dir are guaranteed * to be regular files. Hence the hardcoding. */ - afr_shd_index_purge(subvol, parent->inode, entry->d_name, IA_IFREG); + afr_shd_entry_purge(subvol, parent->inode, entry->d_name, IA_IFREG); ret = 0; goto out; } diff --git a/xlators/cluster/afr/src/afr-self-heal-name.c b/xlators/cluster/afr/src/afr-self-heal-name.c index dace071..51e3d8c 100644 --- a/xlators/cluster/afr/src/afr-self-heal-name.c +++ b/xlators/cluster/afr/src/afr-self-heal-name.c @@ -98,21 +98,12 @@ __afr_selfheal_name_expunge(xlator_t *this, inode_t *parent, uuid_t pargfid, const char *bname, inode_t *inode, struct afr_reply *replies) { - loc_t loc = { - 0, - }; int i = 0; afr_private_t *priv = NULL; - char g[64]; int ret = 0; priv = this->private; - loc.parent = inode_ref(parent); - gf_uuid_copy(loc.pargfid, pargfid); - loc.name = bname; - loc.inode = inode_ref(inode); - for (i = 0; i < priv->child_count; i++) { if (!replies[i].valid) continue; @@ -120,30 +111,10 @@ __afr_selfheal_name_expunge(xlator_t *this, inode_t *parent, uuid_t pargfid, if (replies[i].op_ret) continue; - switch (replies[i].poststat.ia_type) { - case IA_IFDIR: - gf_msg(this->name, GF_LOG_WARNING, 0, - AFR_MSG_EXPUNGING_FILE_OR_DIR, - "expunging dir %s/%s (%s) on %s", uuid_utoa(pargfid), - bname, uuid_utoa_r(replies[i].poststat.ia_gfid, g), - priv->children[i]->name); - - ret |= syncop_rmdir(priv->children[i], &loc, 1, NULL, NULL); - break; - default: - gf_msg(this->name, GF_LOG_WARNING, 0, - AFR_MSG_EXPUNGING_FILE_OR_DIR, - "expunging file %s/%s (%s) on %s", uuid_utoa(pargfid), - bname, uuid_utoa_r(replies[i].poststat.ia_gfid, g), - priv->children[i]->name); - - ret |= syncop_unlink(priv->children[i], &loc, NULL, NULL); - break; - } + ret |= afr_selfheal_entry_delete(this, parent, bname, inode, i, + replies); } - loc_wipe(&loc); - return ret; } diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h index 8f6fb00..c8dc384 100644 --- a/xlators/cluster/afr/src/afr-self-heal.h +++ b/xlators/cluster/afr/src/afr-self-heal.h @@ -370,4 +370,9 @@ gf_boolean_t afr_is_file_empty_on_all_children(afr_private_t *priv, struct afr_reply *replies); +int +afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name, + inode_t *inode, int child, struct afr_reply *replies); +int +afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode); #endif /* !_AFR_SELFHEAL_H */ diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c index 95ac5f2..939a135 100644 --- a/xlators/cluster/afr/src/afr-self-heald.c +++ b/xlators/cluster/afr/src/afr-self-heald.c @@ -222,7 +222,7 @@ out: } int -afr_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name, +afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name, ia_type_t type) { int ret = 0; @@ -422,7 +422,7 @@ afr_shd_index_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, ret = afr_shd_selfheal(healer, healer->subvol, gfid); if (ret == -ENOENT || ret == -ESTALE) - afr_shd_index_purge(subvol, parent->inode, entry->d_name, val); + afr_shd_entry_purge(subvol, parent->inode, entry->d_name, val); if (ret == 2) /* If bricks crashed in pre-op after creating indices/xattrop @@ -798,6 +798,176 @@ afr_bricks_available_for_heal(afr_private_t *priv) return _gf_true; } +static int +afr_shd_anon_inode_cleaner(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent, + void *data) +{ + struct subvol_healer *healer = data; + afr_private_t *priv = healer->this->private; + call_frame_t *frame = NULL; + afr_local_t *local = NULL; + int ret = 0; + loc_t loc = {0}; + int count = 0; + int i = 0; + int op_errno = 0; + struct iatt *iatt = NULL; + gf_boolean_t multiple_links = _gf_false; + unsigned char *gfid_present = alloca0(priv->child_count); + unsigned char *entry_present = alloca0(priv->child_count); + char *type = "file"; + + frame = afr_frame_create(healer->this, &ret); + if (!frame) { + ret = -ret; + goto out; + } + local = frame->local; + if (AFR_COUNT(local->child_up, priv->child_count) != priv->child_count) { + gf_msg_debug(healer->this->name, 0, + "Not all bricks are up. Skipping " + "cleanup of %s on %s", + entry->d_name, subvol->name); + ret = 0; + goto out; + } + + loc.inode = inode_new(parent->inode->table); + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + ret = gf_uuid_parse(entry->d_name, loc.gfid); + if (ret) { + ret = 0; + goto out; + } + AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc, + NULL); + for (i = 0; i < priv->child_count; i++) { + if (local->replies[i].op_ret == 0) { + count++; + gfid_present[i] = 1; + iatt = &local->replies[i].poststat; + if (iatt->ia_type == IA_IFDIR) { + type = "dir"; + } + + if (i == healer->subvol) { + if (local->replies[i].poststat.ia_nlink > 1) { + multiple_links = _gf_true; + } + } + } else if (local->replies[i].op_errno != ENOENT && + local->replies[i].op_errno != ESTALE) { + /*We don't have complete view. Skip the entry*/ + gf_msg_debug(healer->this->name, local->replies[i].op_errno, + "Skipping cleanup of %s on %s", entry->d_name, + subvol->name); + ret = 0; + goto out; + } + } + + /*Inode is deleted from subvol*/ + if (count == 1 || (iatt->ia_type != IA_IFDIR && multiple_links)) { + gf_msg(healer->this->name, GF_LOG_WARNING, 0, + AFR_MSG_EXPUNGING_FILE_OR_DIR, "expunging %s %s/%s on %s", type, + priv->anon_inode_name, entry->d_name, subvol->name); + ret = afr_shd_entry_purge(subvol, parent->inode, entry->d_name, + iatt->ia_type); + if (ret == -ENOENT || ret == -ESTALE) + ret = 0; + } else if (count > 1) { + loc_wipe(&loc); + loc.parent = inode_ref(parent->inode); + loc.name = entry->d_name; + loc.inode = inode_new(parent->inode->table); + if (!loc.inode) { + ret = -ENOMEM; + goto out; + } + AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, + &loc, NULL); + count = 0; + for (i = 0; i < priv->child_count; i++) { + if (local->replies[i].op_ret == 0) { + count++; + entry_present[i] = 1; + iatt = &local->replies[i].poststat; + } else if (local->replies[i].op_errno != ENOENT && + local->replies[i].op_errno != ESTALE) { + /*We don't have complete view. Skip the entry*/ + gf_msg_debug(healer->this->name, local->replies[i].op_errno, + "Skipping cleanup of %s on %s", entry->d_name, + subvol->name); + ret = 0; + goto out; + } + } + for (i = 0; i < priv->child_count; i++) { + if (gfid_present[i] && !entry_present[i]) { + /*Entry is not anonymous on at least one subvol*/ + gf_msg_debug(healer->this->name, 0, + "Valid entry present on %s " + "Skipping cleanup of %s on %s", + priv->children[i]->name, entry->d_name, + subvol->name); + ret = 0; + goto out; + } + } + + gf_msg(healer->this->name, GF_LOG_WARNING, 0, + AFR_MSG_EXPUNGING_FILE_OR_DIR, + "expunging %s %s/%s on all subvols", type, priv->anon_inode_name, + entry->d_name); + ret = 0; + for (i = 0; i < priv->child_count; i++) { + op_errno = -afr_shd_entry_purge(priv->children[i], loc.parent, + entry->d_name, iatt->ia_type); + if (op_errno != ENOENT && op_errno != ESTALE) { + ret |= -op_errno; + } + } + } + +out: + if (frame) + AFR_STACK_DESTROY(frame); + loc_wipe(&loc); + return ret; +} + +static void +afr_cleanup_anon_inode_dir(struct subvol_healer *healer) +{ + int ret = 0; + call_frame_t *frame = NULL; + afr_private_t *priv = healer->this->private; + loc_t loc = {0}; + + ret = afr_anon_inode_create(healer->this, healer->subvol, &loc.inode); + if (ret) + goto out; + + frame = afr_frame_create(healer->this, &ret); + if (!frame) { + ret = -ret; + goto out; + } + + ret = syncop_mt_dir_scan(frame, priv->children[healer->subvol], &loc, + GF_CLIENT_PID_SELF_HEALD, healer, + afr_shd_anon_inode_cleaner, NULL, + priv->shd.max_threads, priv->shd.wait_qlength); +out: + if (frame) + AFR_STACK_DESTROY(frame); + loc_wipe(&loc); + return; +} + void * afr_shd_index_healer(void *data) { @@ -854,6 +1024,10 @@ afr_shd_index_healer(void *data) sleep(1); } while (ret > 0); + if (ret == 0) { + afr_cleanup_anon_inode_dir(healer); + } + if (pre_crawl_xdata && !healer->crawl_event.heal_failed_count) { afr_shd_ta_check_and_unset_xattrs(this, &loc, healer, pre_crawl_xdata); diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h index 1990539..acd567e 100644 --- a/xlators/cluster/afr/src/afr-self-heald.h +++ b/xlators/cluster/afr/src/afr-self-heald.h @@ -70,6 +70,6 @@ afr_shd_gfid_to_path(xlator_t *this, xlator_t *subvol, uuid_t gfid, char **path_p); int -afr_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name, +afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name, ia_type_t type); #endif /* !_AFR_SELF_HEALD_H */ diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c index bfa464f..33fe4d8 100644 --- a/xlators/cluster/afr/src/afr.c +++ b/xlators/cluster/afr/src/afr.c @@ -135,6 +135,27 @@ set_data_self_heal_algorithm(afr_private_t *priv, char *algo) } } +void +afr_handle_anon_inode_options(afr_private_t *priv, dict_t *options) +{ + char *volfile_id_str = NULL; + uuid_t anon_inode_gfid = {0}; + + /*If volume id is not present don't enable anything*/ + if (dict_get_str(options, "volume-id", &volfile_id_str)) + return; + GF_ASSERT(strlen(AFR_ANON_DIR_PREFIX) + strlen(volfile_id_str) <= NAME_MAX); + /*anon_inode_name is not supposed to change once assigned*/ + if (!priv->anon_inode_name[0]) { + snprintf(priv->anon_inode_name, sizeof(priv->anon_inode_name), "%s-%s", + AFR_ANON_DIR_PREFIX, volfile_id_str); + gf_uuid_parse(volfile_id_str, anon_inode_gfid); + /*Flip a bit to make sure volfile-id and anon-gfid are not same*/ + anon_inode_gfid[0] ^= 1; + uuid_utoa_r(anon_inode_gfid, priv->anon_gfid_str); + } +} + int reconfigure(xlator_t *this, dict_t *options) { @@ -287,6 +308,10 @@ reconfigure(xlator_t *this, dict_t *options) consistent_io = _gf_false; priv->consistent_io = consistent_io; + afr_handle_anon_inode_options(priv, options); + + GF_OPTION_RECONF("use-anonymous-inode", priv->use_anon_inode, options, bool, + out); if (priv->shd.enabled) { if ((priv->shd.enabled != enabled_old) || (timeout_old != priv->shd.timeout)) @@ -535,7 +560,9 @@ init(xlator_t *this) GF_OPTION_INIT("consistent-metadata", priv->consistent_metadata, bool, out); GF_OPTION_INIT("consistent-io", priv->consistent_io, bool, out); + afr_handle_anon_inode_options(priv, this->options); + GF_OPTION_INIT("use-anonymous-inode", priv->use_anon_inode, bool, out); if (priv->quorum_count != 0) priv->consistent_io = _gf_false; @@ -547,13 +574,16 @@ init(xlator_t *this) goto out; } + priv->anon_inode = GF_CALLOC(sizeof(unsigned char), child_count, + gf_afr_mt_char); + priv->child_up = GF_CALLOC(sizeof(unsigned char), child_count, gf_afr_mt_char); priv->child_latency = GF_MALLOC(sizeof(*priv->child_latency) * child_count, gf_afr_mt_child_latency_t); - if (!priv->child_up || !priv->child_latency) { + if (!priv->child_up || !priv->child_latency || !priv->anon_inode) { ret = -ENOMEM; goto out; } @@ -1218,6 +1248,14 @@ struct volume_options options[] = { .tags = {"replicate"}, .description = "This option exists only for backward compatibility " "and configuring it doesn't have any effect"}, + {.key = {"use-anonymous-inode"}, + .type = GF_OPTION_TYPE_BOOL, + .default_value = "no", + .op_version = {GD_OP_VERSION_7_0}, + .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE, + .tags = {"replicate"}, + .description = "Setting this option heals directory renames efficiently"}, + {.key = {NULL}}, }; diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h index 3a2b26d..6a9a763 100644 --- a/xlators/cluster/afr/src/afr.h +++ b/xlators/cluster/afr/src/afr.h @@ -40,6 +40,8 @@ #define AFR_TA_DOM_MODIFY "afr.ta.dom-modify" #define AFR_HALO_MAX_LATENCY 99999 +#define AFR_ANON_DIR_PREFIX ".glusterfs-anonymous-inode" + #define PFLAG_PENDING (1 << 0) #define PFLAG_SBRAIN (1 << 1) @@ -155,6 +157,7 @@ typedef struct _afr_private { struct list_head ta_waitq; struct list_head ta_onwireq; + unsigned char *anon_inode; unsigned char *child_up; int64_t *child_latency; unsigned char *local; @@ -240,6 +243,11 @@ typedef struct _afr_private { gf_boolean_t esh_granular; gf_boolean_t consistent_io; gf_boolean_t data_self_heal; /* on/off */ + gf_boolean_t use_anon_inode; + + /*For anon-inode handling */ + char anon_inode_name[NAME_MAX + 1]; + char anon_gfid_str[UUID_SIZE + 1]; } afr_private_t; typedef enum { @@ -1341,4 +1349,7 @@ afr_selfheal_childup(xlator_t *this, afr_private_t *priv); void afr_fill_success_replies(afr_local_t *local, afr_private_t *priv, unsigned char *replies); +gf_boolean_t +afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name, + pid_t pid); #endif /* __AFR_H__ */ diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c index 094a71f..1920284 100644 --- a/xlators/mgmt/glusterd/src/glusterd-volgen.c +++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c @@ -3867,6 +3867,38 @@ out: } static int +set_volfile_id_option(volgen_graph_t *graph, glusterd_volinfo_t *volinfo, + int clusters) +{ + xlator_t *xlator = NULL; + int i = 0; + int ret = -1; + glusterd_conf_t *conf = NULL; + xlator_t *this = NULL; + + this = THIS; + GF_VALIDATE_OR_GOTO("glusterd", this, out); + conf = this->private; + GF_VALIDATE_OR_GOTO(this->name, conf, out); + + if (conf->op_version < GD_OP_VERSION_7_1) + return 0; + xlator = first_of(graph); + + for (i = 0; i < clusters; i++) { + ret = xlator_set_fixed_option(xlator, "volume-id", + uuid_utoa(volinfo->volume_id)); + if (ret) + goto out; + + xlator = xlator->next; + } + +out: + return ret; +} + +static int volgen_graph_build_afr_clusters(volgen_graph_t *graph, glusterd_volinfo_t *volinfo) { @@ -3906,6 +3938,13 @@ volgen_graph_build_afr_clusters(volgen_graph_t *graph, clusters = -1; goto out; } + + ret = set_volfile_id_option(graph, volinfo, clusters); + if (ret) { + clusters = -1; + goto out; + } + if (!volinfo->arbiter_count) goto out; diff --git a/xlators/mgmt/glusterd/src/glusterd-volume-set.c b/xlators/mgmt/glusterd/src/glusterd-volume-set.c index 62acadf..c1ca190 100644 --- a/xlators/mgmt/glusterd/src/glusterd-volume-set.c +++ b/xlators/mgmt/glusterd/src/glusterd-volume-set.c @@ -3789,4 +3789,10 @@ struct volopt_map_entry glusterd_volopt_map[] = { .voltype = "features/cloudsync", .op_version = GD_OP_VERSION_7_0, .flags = VOLOPT_FLAG_CLIENT_OPT}, + + {.key = "cluster.use-anonymous-inode", + .voltype = "cluster/replicate", + .op_version = GD_OP_VERSION_7_1, + .value = "yes", + .flags = VOLOPT_FLAG_CLIENT_OPT}, {.key = NULL}}; -- 1.8.3.1