LU-14739 quota: nodemap squashed root cannot bypass quota

When root on client is squashed via a nodemap's squash_uid/squash_gid,
its IOs must not bypass quota enforcement as it normally does without
squashing.
So on client side, do not set OBD_BRW_FROM_GRANT for every page being
used by root. And on server side, check if root is squashed via a
nodemap and remove OBD_BRW_NOQUOTA.

Signed-off-by: Sebastien Buisson <sbuisson@ddn.com>
Change-Id: I95b31277273589e363193cba8b84870f008bb079
Reviewed-on: https://review.whamcloud.com/43988
Tested-by: jenkins <devops@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Wang Shilong <wshilong@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/ofd/ofd_io.c b/lustre/ofd/ofd_io.c
index a15de42..c9811ff 100644 (file)
--- a/lustre/ofd/ofd_io.c
+++ b/lustre/ofd/ofd_io.c
@@ -1461,6 +1461,26 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
 
        if (cmd == OBD_BRW_WRITE) {
                struct lu_nodemap *nodemap;
 
        if (cmd == OBD_BRW_WRITE) {
                struct lu_nodemap *nodemap;

+               __u32 mapped_uid, mapped_gid;
+
+               nodemap = nodemap_get_from_exp(exp);
+               mapped_uid = nodemap_map_id(nodemap, NODEMAP_UID,
+                                           NODEMAP_FS_TO_CLIENT,
+                                           oa->o_uid);
+               mapped_gid = nodemap_map_id(nodemap, NODEMAP_GID,
+                                           NODEMAP_FS_TO_CLIENT,
+                                           oa->o_gid);
+
+               if (!IS_ERR_OR_NULL(nodemap)) {
+                       /* do not bypass quota enforcement if squashed uid */
+                       if (unlikely(mapped_uid == nodemap->nm_squash_uid)) {
+                               int idx;
+
+                               for (idx = 0; idx < npages; idx++)
+                                       lnb[idx].lnb_flags &= ~OBD_BRW_NOQUOTA;
+                       }
+                       nodemap_putref(nodemap);
+               }

 
                valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID |
                        OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME;
 
                valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID |
                        OBD_MD_FLATIME | OBD_MD_FLMTIME | OBD_MD_FLCTIME;


@@ -1525,16 +1545,8 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
                /* Convert back to client IDs. LU-9671.
                 * nodemap_get_from_exp() may fail due to nodemap deactivated,
                 * server ID will be returned back to client in that case. */
                /* Convert back to client IDs. LU-9671.
                 * nodemap_get_from_exp() may fail due to nodemap deactivated,
                 * server ID will be returned back to client in that case. */

-               nodemap = nodemap_get_from_exp(exp);
-               if (nodemap != NULL && !IS_ERR(nodemap)) {
-                       oa->o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
-                                                  NODEMAP_FS_TO_CLIENT,
-                                                  oa->o_uid);
-                       oa->o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
-                                                  NODEMAP_FS_TO_CLIENT,
-                                                  oa->o_gid);
-                       nodemap_putref(nodemap);
-               }
+               oa->o_uid = mapped_uid;
+               oa->o_gid = mapped_gid;

        } else if (cmd == OBD_BRW_READ) {
                rc = ofd_commitrw_read(env, ofd, fid, objcount,
                                       npages, lnb);
        } else if (cmd == OBD_BRW_READ) {
                rc = ofd_commitrw_read(env, ofd, fid, objcount,
                                       npages, lnb);

diff --git a/lustre/osc/osc_cache.c b/lustre/osc/osc_cache.c
index d193fae..658e8f4 100644 (file)
--- a/lustre/osc/osc_cache.c
+++ b/lustre/osc/osc_cache.c
@@ -2309,7 +2309,7 @@ int osc_queue_async_io(const struct lu_env *env, struct cl_io *io,
        }
 
        /* check if the file's owner/group is over quota */
        }
 
        /* check if the file's owner/group is over quota */

-       if (!(cmd & OBD_BRW_NOQUOTA)) {
+       if (!io->ci_noquota) {

                struct cl_object *obj;
                struct cl_attr   *attr;
                unsigned int qid[LL_MAXQUOTAS];
                struct cl_object *obj;
                struct cl_attr   *attr;
                unsigned int qid[LL_MAXQUOTAS];

diff --git a/lustre/tests/sanity-quota.sh b/lustre/tests/sanity-quota.sh
index 73b0132..d6b3534 100755 (executable)
--- a/lustre/tests/sanity-quota.sh
+++ b/lustre/tests/sanity-quota.sh
@@ -4921,6 +4921,93 @@ test_74()
 }
 run_test 74 "check quota pools per user"
 
 }
 run_test 74 "check quota pools per user"
 

+function cleanup_quota_test_75()
+{
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property admin --value 1
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property trusted --value 1
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property squash_uid --value 99
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property squash_gid --value 99
+
+       wait_nm_sync default admin_nodemap
+       wait_nm_sync default trusted_nodemap
+
+       do_facet mgs $LCTL nodemap_activate 0
+       wait_nm_sync active
+
+       resetquota -u $TSTUSR
+
+       cleanup_quota_test
+}
+
+test_75()
+{
+       local limit=10 # MB
+       local testfile="$DIR/$tdir/$tfile-0"
+
+       setup_quota_test || error "setup quota failed with $?"
+       stack_trap cleanup_quota_test_75 EXIT
+
+       # enable ost quota
+       set_ost_qtype $QTYPE || error "enable ost quota failed"
+
+       # test for user
+       log "User $TSTUSR quota block hardlimit:$limit MB"
+       $LFS setquota -u $TSTID -b 0 -B ${limit}M -i 0 -I 0 $DIR ||
+               error "set user quota failed"
+
+       # make sure the system is clean
+       local used=$(getquota -u $TSTID global curspace)
+       [ $used -ne 0 ] && error "Used space ($used) for user $TSTUSR not 0."
+
+       chmod 777 $DIR/$tdir || error "chmod 777 $DIR/$tdir failed"
+
+       do_facet mgs $LCTL nodemap_activate 1
+       wait_nm_sync active
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property admin --value 0
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property trusted --value 0
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property deny_unknown --value 0
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property squash_uid --value $TSTID
+       do_facet mgs $LCTL nodemap_modify --name default \
+               --property squash_gid --value $TSTID
+       cancel_lru_locks mdc
+       wait_nm_sync default admin_nodemap
+       wait_nm_sync default trusted_nodemap
+       wait_nm_sync default squash_uid
+
+       log "Write..."
+       $DD of=$testfile bs=1M count=$((limit/2)) ||
+               quota_error u $TSTID \
+                       "root write failure, but expect success"
+
+       log "Write out of block quota ..."
+       # possibly a cache write, ignore failure
+       $DD of=$testfile bs=1M count=$((limit/2)) seek=$((limit/2)) || true
+       # flush cache, ensure noquota flag is set on client
+       cancel_lru_locks osc
+       sync; sync_all_data || true
+       # sync forced cache flush, but did not guarantee that slave
+       # got new edquot through glimpse, so wait to make sure
+       sleep 5
+       $DD of=$testfile bs=1M count=1 seek=$limit conv=fsync &&
+               quota_error u $TSTID \
+                       "user write success, but expect EDQUOT"
+       rm -f $testfile
+       wait_delete_completed || error "wait_delete_completed failed"
+       sync_all_data || true
+       used=$(getquota -u $TSTUSR global curspace)
+       [ $used -eq 0 ] || quota_error u $TSTID \
+               "user quota not released after deletion"
+}
+run_test 75 "nodemap squashed root respects quota enforcement"
+

 quota_fini()
 {
        do_nodes $(comma_list $(nodes_list)) \
 quota_fini()
 {
        do_nodes $(comma_list $(nodes_list)) \

diff --git a/lustre/tests/sanity-sec.sh b/lustre/tests/sanity-sec.sh
index b8964c0..9744b9a 100755 (executable)
--- a/lustre/tests/sanity-sec.sh
+++ b/lustre/tests/sanity-sec.sh
@@ -475,71 +475,6 @@ squash_id() {
        fi
 }
 
        fi
 }
 

-wait_nm_sync() {
-       local nodemap_name=$1
-       local key=$2
-       local value=$3
-       local opt=$4
-       local proc_param
-       local is_active=$(do_facet mgs $LCTL get_param -n nodemap.active)
-       local max_retries=20
-       local is_sync
-       local out1=""
-       local out2
-       local mgs_ip=$(host_nids_address $mgs_HOST $NETTYPE | cut -d' ' -f1)
-       local i
-
-       if [ "$nodemap_name" == "active" ]; then
-               proc_param="active"
-       elif [ -z "$key" ]; then
-               proc_param=${nodemap_name}
-       else
-               proc_param="${nodemap_name}.${key}"
-       fi
-       if [ "$opt" == "inactive" ]; then
-               # check nm sync even if nodemap is not activated
-               is_active=1
-               opt=""
-       fi
-       (( is_active == 0 )) && [ "$proc_param" != "active" ] && return
-
-       if [ -z "$value" ]; then
-               out1=$(do_facet mgs $LCTL get_param $opt \
-                       nodemap.${proc_param} 2>/dev/null)
-               echo "On MGS ${mgs_ip}, ${proc_param} = $out1"
-       else
-               out1=$value;
-       fi
-
-       # wait up to 10 seconds for other servers to sync with mgs
-       for i in $(seq 1 10); do
-               for node in $(all_server_nodes); do
-                   local node_ip=$(host_nids_address $node $NETTYPE |
-                                   cut -d' ' -f1)
-
-                   is_sync=true
-                   if [ -z "$value" ]; then
-                       [ $node_ip == $mgs_ip ] && continue
-                   fi
-
-                   out2=$(do_node $node_ip $LCTL get_param $opt \
-                                  nodemap.$proc_param 2>/dev/null)
-                   echo "On $node ${node_ip}, ${proc_param} = $out2"
-                   [ "$out1" != "$out2" ] && is_sync=false && break
-               done
-               $is_sync && break
-               sleep 1
-       done
-       if ! $is_sync; then
-               echo MGS
-               echo $out1
-               echo OTHER - IP: $node_ip
-               echo $out2
-               error "mgs and $nodemap_name ${key} mismatch, $i attempts"
-       fi
-       echo "waited $((i - 1)) seconds for sync"
-}
-

 # ensure that the squash defaults are the expected defaults
 squash_id default 99 0
 wait_nm_sync default squash_uid '' inactive
 # ensure that the squash defaults are the expected defaults
 squash_id default 99 0
 wait_nm_sync default squash_uid '' inactive

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index f75aaea..bc18328 100755 (executable)
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -10466,3 +10466,78 @@ function restore_opencache()
 mkdir_on_mdt0() {
        $LFS mkdir -i 0 -c 1 $*
 }
 mkdir_on_mdt0() {
        $LFS mkdir -i 0 -c 1 $*
 }

+
+# Wait for nodemap synchronization
+wait_nm_sync() {
+       local nodemap_name=$1
+       local key=$2
+       local value=$3
+       local opt=$4
+       local proc_param
+       local is_active=$(do_facet mgs $LCTL get_param -n nodemap.active)
+       local max_retries=20
+       local is_sync
+       local out1=""
+       local out2
+       local mgs_ip=$(host_nids_address $mgs_HOST $NETTYPE | cut -d' ' -f1)
+       local i
+
+       if [ "$nodemap_name" == "active" ]; then
+               proc_param="active"
+       elif [ -z "$key" ]; then
+               proc_param=${nodemap_name}
+       else
+               proc_param="${nodemap_name}.${key}"
+       fi
+       if [ "$opt" == "inactive" ]; then
+               # check nm sync even if nodemap is not activated
+               is_active=1
+               opt=""
+       fi
+       (( is_active == 0 )) && [ "$proc_param" != "active" ] && return
+
+       if [ -z "$value" ]; then
+               out1=$(do_facet mgs $LCTL get_param $opt \
+                       nodemap.${proc_param} 2>/dev/null)
+               echo "On MGS ${mgs_ip}, ${proc_param} = $out1"
+       else
+               out1=$value;
+       fi
+
+       # if servers run on the same node, it is impossible to tell if they get
+       # synced with the mgs, so just wait an arbitrary 10 seconds
+       if [ $(facet_active_host mgs) == $(facet_active_host mds) ] &&
+          [ $(facet_active_host mgs) == $(facet_active_host ost1) ]; then
+               echo "waiting 10 secs for sync"
+               sleep 10
+               return
+       fi
+
+       # wait up to 10 seconds for other servers to sync with mgs
+       for i in $(seq 1 10); do
+               for node in $(all_server_nodes); do
+                       local node_ip=$(host_nids_address $node $NETTYPE |
+                                       cut -d' ' -f1)
+
+                       is_sync=true
+                       if [ -z "$value" ]; then
+                               [ $node_ip == $mgs_ip ] && continue
+                       fi
+
+                       out2=$(do_node $node_ip $LCTL get_param $opt \
+                              nodemap.$proc_param 2>/dev/null)
+                       echo "On $node ${node_ip}, ${proc_param} = $out2"
+                       [ "$out1" != "$out2" ] && is_sync=false && break
+               done
+               $is_sync && break
+               sleep 1
+       done
+       if ! $is_sync; then
+               echo MGS
+               echo $out1
+               echo OTHER - IP: $node_ip
+               echo $out2
+               error "mgs and $nodemap_name ${key} mismatch, $i attempts"
+       fi
+       echo "waited $((i - 1)) seconds for sync"
+}