dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
dst->ln_lost_dirent_repaired =
le64_to_cpu(src->ln_lost_dirent_repaired);
+ dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
}
static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
dst->ln_lost_dirent_repaired =
cpu_to_le64(src->ln_lost_dirent_repaired);
+ dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
}
static void lfsck_namespace_record_failure(const struct lu_env *env,
}
/**
+ * Load the MDT bitmap from the lfsck_namespace tracing file.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ *
+ * \retval positive number for data corruption
+ * \retval 0 for success
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_load_bitmap(const struct lu_env *env,
+ struct lfsck_component *com)
+{
+ struct dt_object *obj = com->lc_obj;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ cfs_bitmap_t *bitmap = lad->lad_bitmap;
+ ssize_t size;
+ __u32 nbits;
+ int rc;
+ ENTRY;
+
+ if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
+ ns->ln_bitmap_size)
+ nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
+ else
+ nbits = ns->ln_bitmap_size;
+
+ if (unlikely(nbits < BITS_PER_LONG))
+ nbits = BITS_PER_LONG;
+
+ if (nbits > bitmap->size) {
+ __u32 new_bits = bitmap->size;
+ cfs_bitmap_t *new_bitmap;
+
+ while (new_bits < nbits)
+ new_bits <<= 1;
+
+ new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
+ if (new_bitmap == NULL)
+ RETURN(-ENOMEM);
+
+ lad->lad_bitmap = new_bitmap;
+ CFS_FREE_BITMAP(bitmap);
+ bitmap = new_bitmap;
+ }
+
+ if (ns->ln_bitmap_size == 0) {
+ lad->lad_incomplete = 0;
+ CFS_RESET_BITMAP(bitmap);
+
+ RETURN(0);
+ }
+
+ size = (ns->ln_bitmap_size + 7) >> 3;
+ rc = dt_xattr_get(env, obj,
+ lfsck_buf_get(env, bitmap->data, size),
+ XATTR_NAME_LFSCK_BITMAP, BYPASS_CAPA);
+ if (rc == -ERANGE || rc == -ENODATA || rc == 0)
+ RETURN(1);
+
+ if (rc < 0)
+ RETURN(rc);
+
+ if (rc != size)
+ RETURN(rc);
+
+ if (cfs_bitmap_check_empty(bitmap))
+ lad->lad_incomplete = 0;
+ else
+ lad->lad_incomplete = 1;
+
+ RETURN(0);
+}
+
+/**
* \retval +ve: the lfsck_namespace is broken, the caller should reset it.
* \retval 0: succeed.
* \retval -ve: failed cases.
}
static int lfsck_namespace_store(const struct lu_env *env,
- struct lfsck_component *com, bool init)
+ struct lfsck_component *com)
{
- struct dt_object *obj = com->lc_obj;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct thandle *handle;
- int len = com->lc_file_size;
- int rc;
+ struct dt_object *obj = com->lc_obj;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ cfs_bitmap_t *bitmap = NULL;
+ struct thandle *handle;
+ __u32 nbits = 0;
+ int len = com->lc_file_size;
+ int rc;
ENTRY;
+ if (lad != NULL) {
+ bitmap = lad->lad_bitmap;
+ nbits = bitmap->size;
+
+ LASSERT(nbits > 0);
+ LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
+ }
+
+ ns->ln_bitmap_size = nbits;
lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
- (struct lfsck_namespace *)com->lc_file_ram);
+ ns);
handle = dt_trans_create(env, lfsck->li_bottom);
if (IS_ERR(handle))
GOTO(log, rc = PTR_ERR(handle));
if (rc != 0)
GOTO(out, rc);
+ if (bitmap != NULL) {
+ rc = dt_declare_xattr_set(env, obj,
+ lfsck_buf_get(env, bitmap->data, nbits >> 3),
+ XATTR_NAME_LFSCK_BITMAP, 0, handle);
+ if (rc != 0)
+ GOTO(out, rc);
+ }
+
rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
if (rc != 0)
GOTO(out, rc);
rc = dt_xattr_set(env, obj,
lfsck_buf_get(env, com->lc_file_disk, len),
- XATTR_NAME_LFSCK_NAMESPACE,
- init ? LU_XATTR_CREATE : LU_XATTR_REPLACE,
- handle, BYPASS_CAPA);
+ XATTR_NAME_LFSCK_NAMESPACE, 0, handle, BYPASS_CAPA);
+ if (rc == 0 && bitmap != NULL)
+ rc = dt_xattr_set(env, obj,
+ lfsck_buf_get(env, bitmap->data, nbits >> 3),
+ XATTR_NAME_LFSCK_BITMAP, 0, handle,
+ BYPASS_CAPA);
GOTO(out, rc);
ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
ns->ln_status = LS_INIT;
down_write(&com->lc_sem);
- rc = lfsck_namespace_store(env, com, true);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
return rc;
}
enum lfsck_namespace_inconsistency_type *type)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lfsck_namespace *ns = com->lc_file_ram;
int rc;
ENTRY;
RETURN(rc);
*type = LNIT_MUL_REF;
+
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
+ * ever tried to verify some remote MDT-object that resides on this
+ * MDT, but this MDT failed to respond such request. So means there
+ * may be some remote name entry on other MDT that references this
+ * object with another name, so we cannot know whether this linkEA
+ * is valid or not. So keep it there and maybe resolved when next
+ * LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ RETURN(0);
+
/* The unique linkEA is invalid, even if the ".." name entry may be
* valid, we still cannot know via which name entry this directory
* will be referenced. Then handle it as pure orphan. */
struct lu_name *cname = &info->lti_name;
const struct lu_fid *cfid = lfsck_dto2fid(child);
struct lu_fid *tfid = &info->lti_fid3;
+ struct lfsck_namespace *ns = com->lc_file_ram;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct dt_object *parent = NULL;
int rc = 0;
* name entry the child will be referenced, since all known entries
* have been verified during the first-stage scanning. */
if (!dt_object_exists(parent)) {
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
+ * has ever tried to verify some remote MDT-object that resides
+ * on this MDT, but this MDT failed to respond such request. So
+ * means there may be some remote name entry on other MDT that
+ * references this object with another name, so we cannot know
+ * whether this linkEA is valid or not. So keep it there and
+ * maybe resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ GOTO(out, rc = 0);
+
if (!lustre_handle_is_used(lh) && retry != NULL) {
*retry = true;
rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
(const struct dt_key *)cname->ln_name, BYPASS_CAPA);
if (rc == -ENOENT) {
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
+ * has ever tried to verify some remote MDT-object that resides
+ * on this MDT, but this MDT failed to respond such request. So
+ * means there may be some remote name entry on other MDT that
+ * references this object with another name, so we cannot know
+ * whether this linkEA is valid or not. So keep it there and
+ * maybe resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ GOTO(out, rc = 0);
+
if (!lustre_handle_is_used(lh) && retry != NULL) {
*retry = true;
LASSERT(!dt_object_remote(child));
- if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
+ if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
+ !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
"the namespace LFSCK, then the LFSCK cannot guarantee"
"all the name entries have been verified in first-stage"
* directory, then handle it as orphan. */
lfsck_ibits_unlock(&lh, LCK_EX);
type = LNIT_MUL_REF;
+
+ /* If the LFSCK is marked as LF_INCOMPLETE,
+ * then means some MDT has ever tried to
+ * verify some remote MDT-object that resides
+ * on this MDT, but this MDT failed to respond
+ * such request. So means there may be some
+ * remote name entry on other MDT that
+ * references this object with another name,
+ * so we cannot know whether this linkEA is
+ * valid or not. So keep it there and maybe
+ * resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE)
+ GOTO(out, rc = 0);
+
snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
"-"DFID, PFID(pfid));
rc = lfsck_namespace_insert_orphan(env, com, child,
rc = lfsck_namespace_shrink_linkea(env, com,
child, &ldata, cname, pfid, true);
} else {
+ /* If the LFSCK is marked as LF_INCOMPLETE,
+ * then means some MDT has ever tried to
+ * verify some remote MDT-object that resides
+ * on this MDT, but this MDT failed to respond
+ * such request. So means there may be some
+ * remote name entry on other MDT that
+ * references this object with another name,
+ * so we cannot know whether this linkEA is
+ * valid or not. So keep it there and maybe
+ * resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc = 0);
+ }
+
/* Create the lost parent as an orphan. */
rc = lfsck_namespace_create_orphan(env, com,
parent);
continue;
}
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some
+ * MDT has ever tried to verify some remote MDT-object that
+ * resides on this MDT, but this MDT failed to respond such
+ * request. So means there may be some remote name entry on
+ * other MDT that references this object with another name,
+ * so we cannot know whether this linkEA is valid or not.
+ * So keep it there and maybe resolved when next LFSCK run. */
+ if (ns->ln_flags & LF_INCOMPLETE) {
+ lfsck_object_put(env, parent);
+
+ GOTO(out, rc = 0);
+ }
+
/* Add the missing name entry back to the namespace. */
rc = lfsck_namespace_insert_normal(env, com, parent, child,
cname->ln_name);
count = ldata.ld_leh->leh_reccount;
}
- if (count == 0) {
+ /* If the LFSCK is marked as LF_INCOMPLETE, then means some
+ * MDT has ever tried to verify some remote MDT-object that
+ * resides on this MDT, but this MDT failed to respond such
+ * request. So means there may be some remote name entry on
+ * other MDT that references this object with another name,
+ * so we cannot know whether this linkEA is valid or not.
+ * So keep it there and maybe resolved when next LFSCK run. */
+ if (count == 0 && !(ns->ln_flags & LF_INCOMPLETE)) {
/* If the child becomes orphan, then insert it into
* the global .lustre/lost+found/MDTxxxx directory. */
rc = lfsck_namespace_insert_orphan(env, com, child, "", "O",
static int lfsck_namespace_reset(const struct lu_env *env,
struct lfsck_component *com, bool init)
{
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_namespace *ns = com->lc_file_ram;
- struct dt_object *root;
- struct dt_object *dto;
- int rc;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct dt_object *root;
+ struct dt_object *dto;
+ int rc;
ENTRY;
root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
if (rc != 0)
GOTO(out, rc);
- rc = lfsck_namespace_store(env, com, true);
+ lad->lad_incomplete = 0;
+ CFS_RESET_BITMAP(lad->lad_bitmap);
+
+ rc = lfsck_namespace_store(env, com);
GOTO(out, rc);
com->lc_new_checked = 0;
}
- rc = lfsck_namespace_store(env, com, false);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
log:
struct lfsck_position *pos = &com->lc_pos_start;
int rc;
- if (ns->ln_status == LS_COMPLETED) {
+ rc = lfsck_namespace_load_bitmap(env, com);
+ if (rc > 0 || (rc == 0 && ns->ln_status == LS_COMPLETED)) {
rc = lfsck_namespace_reset(env, com, false);
if (rc == 0)
rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
+ }
- if (rc != 0) {
- CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
- "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
+ if (rc != 0) {
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: rc = %d\n",
+ lfsck_lfsck2name(lfsck), rc);
- return rc;
- }
+ return rc;
}
down_write(&com->lc_sem);
com->lc_new_checked = 0;
}
- rc = lfsck_namespace_store(env, com, false);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
}
CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
- "status %d\n", lfsck_lfsck2name(lfsck), lr->lr_event,
- lr->lr_index, lr->lr_status);
+ "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
+ lr->lr_index, lr->lr_status, lr->lr_flags2);
spin_lock(<ds->ltd_lock);
ltd = LTD_TGT(ltds, lr->lr_index);
break;
}
+ if (lr->lr_flags2 & LF_INCOMPLETE)
+ ns->ln_flags |= LF_INCOMPLETE;
+
if (list_empty(<d->ltd_namespace_list))
list_add_tail(<d->ltd_namespace_list,
&lad->lad_mdt_list);
struct dt_object *dir = lnr->lnr_obj;
struct dt_object *obj = NULL;
const struct lu_fid *pfid = lfsck_dto2fid(dir);
- struct dt_device *dev;
+ struct dt_device *dev = NULL;
struct lustre_handle lh = { 0 };
bool repaired = false;
bool dtlocked = false;
bool remove;
bool newdata;
bool log = false;
- int idx;
+ int idx = 0;
int count = 0;
int rc;
enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
"did not join the namespace LFSCK\n",
lfsck_lfsck2name(lfsck), idx);
- ns->ln_flags |= LF_INCOMPLETE;
+ lfsck_lad_set_bitmap(env, com, idx);
GOTO(out, rc = -ENODEV);
}
lnr->lnr_namelen, lnr->lnr_name, rc);
lfsck_namespace_record_failure(env, lfsck, ns);
+ if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
+ rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
+ rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
+ dev != NULL && dev != lfsck->li_next)
+ lfsck_lad_set_bitmap(env, com, idx);
+
if (!(bk->lb_param & LPF_FAILOUT))
rc = 0;
} else {
if (obj != NULL && !IS_ERR(obj))
lfsck_object_put(env, obj);
+
return rc;
}
ns->ln_time_last_checkpoint = cfs_time_current_sec();
ns->ln_objs_checked_phase2 += com->lc_new_checked;
com->lc_new_checked = 0;
- rc = lfsck_namespace_store(env, com, false);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
if (rc != 0)
GOTO(put, rc);
ns->ln_status = LS_FAILED;
}
- rc = lfsck_namespace_store(env, com, false);
+ rc = lfsck_namespace_store(env, com);
up_write(&com->lc_sem);
return rc;
}
+static int
+lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
+ struct ptlrpc_request *req,
+ void *args, int rc)
+{
+ return 0;
+}
+
+/**
+ * Notify remote LFSCK instances about former failures.
+ *
+ * The local LFSCK instance has recorded which MDTs have ever failed to respond
+ * some LFSCK verification requests (maybe because of network issues or the MDT
+ * itself trouble). During the respond gap the MDT may missed some name entries
+ * verification, then the MDT cannot know whether related MDT-objects have been
+ * referenced by related name entries or not, then in the second-stage scanning,
+ * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
+ * linkEA for back reference, then it will misguide the LFSCK to generate wrong
+ * name entry for repairing the orphan.
+ *
+ * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
+ * it will scan the bitmap for the ever failed MDTs, and notify them that they
+ * have ever missed some name entries verification and should skip the handling
+ * for orphan MDT-objects.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] lr pointer to the lfsck request
+ */
static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_request *lr)
{
- /* XXX: TBD */
+ struct lfsck_async_interpret_args *laia =
+ &lfsck_env_info(env)->lti_laia2;
+ struct lfsck_assistant_data *lad = com->lc_data;
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
+ struct lfsck_tgt_desc *ltd;
+ struct ptlrpc_request_set *set;
+ int rc = 0;
+ ENTRY;
+
+ set = ptlrpc_prep_set();
+ if (set == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
+ memset(laia, 0, sizeof(*laia));
+ lad->lad_touch_gen++;
+
+ spin_lock(<ds->ltd_lock);
+ while (!list_empty(&lad->lad_mdt_list)) {
+ ltd = list_entry(lad->lad_mdt_list.next,
+ struct lfsck_tgt_desc,
+ ltd_namespace_list);
+ if (ltd->ltd_namespace_gen == lad->lad_touch_gen)
+ break;
+
+ ltd->ltd_namespace_gen = lad->lad_touch_gen;
+ list_move_tail(<d->ltd_namespace_list,
+ &lad->lad_mdt_list);
+ if (!lad->lad_incomplete ||
+ !cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
+ ltd->ltd_namespace_failed = 0;
+ continue;
+ }
+
+ ltd->ltd_namespace_failed = 1;
+ spin_unlock(<ds->ltd_lock);
+ rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
+ lfsck_namespace_assistant_sync_failures_interpret,
+ laia, LFSCK_NOTIFY);
+ if (rc != 0)
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
+ "to sync failure with MDT %x: rc = %d\n",
+ lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
+
+ spin_lock(<ds->ltd_lock);
+ }
+ spin_unlock(<ds->ltd_lock);
+
+ rc = ptlrpc_set_wait(set);
+ ptlrpc_set_destroy(set);
+
+ GOTO(out, rc);
+
+out:
+ if (rc != 0)
+ CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
+ "to sync failure with MDTs, and related MDTs "
+ "may handle orphan un-properly: rc = %d\n",
+ lfsck_lfsck2name(lfsck), rc);
+
+ EXIT;
}
struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
ALWAYS_EXCEPT="$ALWAYS_EXCEPT 11 12 13 14 15 16 17 18 19 20 21"
[[ $(lustre_version_code $SINGLEMDS) -lt $(version_code 2.6.50) ]] &&
- ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2d 2e 3 22 23 24 25 26 27"
+ ALWAYS_EXCEPT="$ALWAYS_EXCEPT 2d 2e 3 22 23 24 25 26 27 28"
build_test_filter
}
run_test 27b "LFSCK can recreate the lost remote parent directory as orphan"
+test_28() {
+ [ $MDSCOUNT -lt 2 ] &&
+ skip "The test needs at least 2 MDTs" && return
+
+ echo "#####"
+ echo "The target name entry is lost. The LFSCK should insert the"
+ echo "orphan MDT-object under .lustre/lost+found/MDTxxxx. But if"
+ echo "the MDT (on which the orphan MDT-object resides) has ever"
+ echo "failed to respond some name entry verification durin the"
+ echo "first stage-scanning, then the LFSCK should skip to handle"
+ echo "orphan MDT-object on this MDT. But other MDTs should not"
+ echo "be affected."
+ echo "#####"
+
+ check_mount_and_prep
+ $LFS mkdir -i 0 $DIR/$tdir/d1
+ $LFS mkdir -i 1 $DIR/$tdir/d1/a1
+ $LFS mkdir -i 1 $DIR/$tdir/d1/a2
+
+ $LFS mkdir -i 1 $DIR/$tdir/d2
+ $LFS mkdir -i 0 $DIR/$tdir/d2/a1
+ $LFS mkdir -i 0 $DIR/$tdir/d2/a2
+
+ echo "Inject failure stub on MDT0 to simulate the case that"
+ echo "d1/a1's name entry will be removed, but the d1/a1's object"
+ echo "and its linkEA are kept in the system. And the case that"
+ echo "d2/a2's name entry will be removed, but the d2/a2's object"
+ echo "and its linkEA are kept in the system."
+
+ #define OBD_FAIL_LFSCK_NO_NAMEENTRY 0x1624
+ do_facet mds1 $LCTL set_param fail_loc=0x1624
+ do_facet mds2 $LCTL set_param fail_loc=0x1624
+ rmdir $DIR/$tdir/d1/a1 || error "(1) Fail to rmdir $DIR/$tdir/d1/a1"
+ rmdir $DIR/$tdir/d2/a2 || error "(2) Fail to rmdir $DIR/$tdir/d2/a2"
+ do_facet mds1 $LCTL set_param fail_loc=0
+ do_facet mds2 $LCTL set_param fail_loc=0
+
+ cancel_lru_locks mdc
+ cancel_lru_locks osc
+
+ echo "Inject failure, to simulate the MDT0 fail to handle"
+ echo "MDT1 LFSCK request during the first-stage scanning."
+ #define OBD_FAIL_LFSCK_BAD_NETWORK 0x161c
+ do_facet mds2 $LCTL set_param fail_loc=0x161c fail_val=0
+
+ echo "Trigger namespace LFSCK on all devices to find out orphan object"
+ $START_NAMESPACE -r -A ||
+ error "(3) Fail to start LFSCK for namespace"
+
+ wait_update_facet mds1 "$LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "partial" 32 || {
+ error "(4) mds1 is not the expected 'partial'"
+ }
+
+ wait_update_facet mds2 "$LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 || {
+ error "(5) mds2 is not the expected 'completed'"
+ }
+
+ do_facet mds2 $LCTL set_param fail_loc=0 fail_val=0
+
+ local repaired=$(do_facet mds1 $LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_namespace |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(6) Expect 0 fixed on mds1, but got: $repaired"
+
+ repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(7) Expect 1 fixed on mds2, but got: $repaired"
+
+ echo "Trigger namespace LFSCK on all devices again to cleanup"
+ $START_NAMESPACE -r -A ||
+ error "(8) Fail to start LFSCK for namespace"
+
+ for k in $(seq $MDSCOUNT); do
+ # The LFSCK status query internal is 30 seconds. For the case
+ # of some LFSCK_NOTIFY RPCs failure/lost, we will wait enough
+ # time to guarantee the status sync up.
+ wait_update_facet mds${k} "$LCTL get_param -n \
+ mdd.$(facet_svc mds${k}).lfsck_namespace |
+ awk '/^status/ { print \\\$2 }'" "completed" 32 ||
+ error "(9) MDS${k} is not the expected 'completed'"
+ done
+
+ local repaired=$(do_facet mds1 $LCTL get_param -n \
+ mdd.$(facet_svc mds1).lfsck_namespace |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 1 ] ||
+ error "(10) Expect 1 fixed on mds1, but got: $repaired"
+
+ repaired=$(do_facet mds2 $LCTL get_param -n \
+ mdd.$(facet_svc mds2).lfsck_namespace |
+ awk '/^lost_dirent_repaired/ { print $2 }')
+ [ $repaired -eq 0 ] ||
+ error "(11) Expect 0 fixed on mds2, but got: $repaired"
+}
+run_test 28 "Skip the failed MDT(s) when handle orphan MDT-objects"
+
$LCTL set_param debug=-lfsck > /dev/null || true
# restore MDS/OST size
git://git.whamcloud.com / fs / lustre-release.git / commitdiff