static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
{
__u32 magic;
- __u32 patten;
+ __u32 pattern;
magic = le32_to_cpu(lmm->lmm_magic);
/* If magic crashed, keep it there. Sometime later, during OST-object
* orphan handling, if some OST-object(s) back-point to it, it can be
* verified and repaired. */
- if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
- return -EINVAL;
+ if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
+ struct ost_id oi;
+ int rc;
+
+ lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
+ if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
+ rc = -EOPNOTSUPP;
+ else
+ rc = -EINVAL;
+
+ CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
+ rc == -EINVAL ? "Unknown" : "Unsupported",
+ magic, POSTID(&oi));
- patten = le32_to_cpu(lmm->lmm_pattern);
+ return rc;
+ }
+
+ pattern = le32_to_cpu(lmm->lmm_pattern);
/* XXX: currently, we only support LOV_PATTERN_RAID0. */
- if (patten != LOV_PATTERN_RAID0)
+ if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
+ struct ost_id oi;
+
+ lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
+ CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
+ pattern, POSTID(&oi));
+
return -EOPNOTSUPP;
+ }
return 0;
}
#define LFSCK_RBTREE_BITMAP_SIZE PAGE_CACHE_SIZE
#define LFSCK_RBTREE_BITMAP_WIDTH (LFSCK_RBTREE_BITMAP_SIZE << 3)
-#define LFSCK_RBTREE_BITMAP_MASK (LFSCK_RBTREE_BITMAP_SIZE - 1)
+#define LFSCK_RBTREE_BITMAP_MASK (LFSCK_RBTREE_BITMAP_WIDTH - 1)
struct lfsck_rbtree_node {
struct rb_node lrn_node;
if (oid < lrn->lrn_first_oid)
return -1;
- if (oid >= lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH)
+ if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
return 1;
return 0;
return ERR_PTR(-ENOMEM);
}
- rb_init_node(&lrn->lrn_node);
+ RB_CLEAR_NODE(&lrn->lrn_node);
lrn->lrn_seq = fid_seq(fid);
lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
atomic_set(&lrn->lrn_known_count, 0);
lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
struct lfsck_rbtree_node *lrn)
{
- struct rb_node **pos = &(llsd->llsd_rb_root.rb_node);
+ struct rb_node **pos = &llsd->llsd_rb_root.rb_node;
struct rb_node *parent = NULL;
struct lfsck_rbtree_node *tmp;
int rc;
- while (*pos) {
+ while (*pos != NULL) {
parent = *pos;
- tmp = rb_entry(*pos, struct lfsck_rbtree_node, lrn_node);
+ tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
if (rc < 0)
- pos = &((*pos)->rb_left);
+ pos = &(*pos)->rb_left;
else if (rc > 0)
- pos = &((*pos)->rb_right);
+ pos = &(*pos)->rb_right;
else
return tmp;
}
}
}
-static inline bool is_dummy_lov_ost_data(struct lov_ost_data_v1 *obj)
-{
- if (fid_is_zero(&obj->l_ost_oi.oi_fid) &&
- obj->l_ost_gen == 0 && obj->l_ost_idx == 0)
- return true;
-
- return false;
-}
-
static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
const struct lfsck_layout *src)
{
lastid = cpu_to_le64(lls->lls_lastid);
rc = dt_declare_record_write(env, lls->lls_lastid_obj,
lfsck_buf_get(env, &lastid,
- sizeof(lastid)), pos, th);
+ sizeof(lastid)),
+ pos, th);
if (rc != 0)
goto stop;
return rc;
}
+static void lfsck_layout_record_failure(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct lfsck_layout *lo)
+{
+ lo->ll_objs_failed_phase1++;
+ if (unlikely(lo->ll_pos_first_inconsistent == 0))
+ lo->ll_pos_first_inconsistent =
+ lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
+ lfsck->li_di_oit);
+}
+
static int lfsck_layout_master_async_interpret(const struct lu_env *env,
struct ptlrpc_request *req,
void *args, int rc)
break;
/* link other MDT targets locallly. */
+ ltds = &lfsck->li_mdt_descs;
spin_lock(<ds->ltd_lock);
cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
ltd = LTD_TGT(ltds, idx);
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
down_write(&com->lc_sem);
-
lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
HALF_SEC - lfsck->li_time_last_checkpoint);
lo->ll_time_last_checkpoint = cfs_time_current_sec();
lo->ll_status = LS_FAILED;
}
- if (lo->ll_status != LS_PAUSED) {
- spin_lock(&lfsck->li_lock);
- list_del_init(&com->lc_link);
- list_add_tail(&com->lc_link, &lfsck->li_list_idle);
- spin_unlock(&lfsck->li_lock);
- }
-
rc = lfsck_layout_store(env, com);
-
up_write(&com->lc_sem);
return rc;
}
/**
+ * Get the system default stripe size.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] lfsck pointer to the lfsck instance
+ * \param[out] size pointer to the default stripe size
+ *
+ * \retval 0 for success
+ * \retval negative error number on failure
+ */
+static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ __u32 *size)
+{
+ struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
+ struct dt_object *root;
+ int rc;
+
+ root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
+ if (IS_ERR(root))
+ return PTR_ERR(root);
+
+ /* Get the default stripe size via xattr_get on the backend root. */
+ rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
+ XATTR_NAME_LOV, BYPASS_CAPA);
+ if (rc > 0) {
+ /* The lum->lmm_stripe_size is LE mode. The *size also
+ * should be LE mode. So it is unnecessary to convert. */
+ *size = lum->lmm_stripe_size;
+ rc = 0;
+ } else if (unlikely(rc == 0)) {
+ rc = -EINVAL;
+ }
+
+ lfsck_object_put(env, root);
+
+ return rc;
+}
+
+/**
* \retval +1: repaired
* \retval 0: did nothing
* \retval -ve: on error
struct lov_ost_data_v1 *slot,
int fl, __u32 ost_idx)
{
- struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
- int rc;
+ struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
+ struct lov_mds_md_v1 *lmm = buf->lb_buf;
+ int rc;
fid_to_ostid(cfid, oi);
ostid_cpu_to_le(oi, &slot->l_ost_oi);
slot->l_ost_gen = cpu_to_le32(0);
slot->l_ost_idx = cpu_to_le32(ost_idx);
+
+ if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
+ struct lov_ost_data_v1 *objs;
+ int i;
+ __u16 count;
+
+ count = le16_to_cpu(lmm->lmm_stripe_count);
+ if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
+ objs = &lmm->lmm_objects[0];
+ else
+ objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
+ for (i = 0; i < count; i++, objs++) {
+ if (objs != slot && lovea_slot_is_dummy(objs))
+ break;
+ }
+
+ /* If the @slot is the last dummy slot to be refilled,
+ * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
+ if (i == count)
+ lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
+ }
+
rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV, fl, handle,
BYPASS_CAPA);
if (rc == 0)
* \retval -ve: on error
*/
static int lfsck_layout_extend_lovea(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
struct thandle *handle,
struct dt_object *parent,
struct lu_fid *cfid,
struct lu_buf *buf, int fl,
- __u32 ost_idx, __u32 ea_off)
+ __u32 ost_idx, __u32 ea_off, bool reset)
{
struct lov_mds_md_v1 *lmm = buf->lb_buf;
struct lov_ost_data_v1 *objs;
int rc;
+ __u16 count;
ENTRY;
- if (fl == LU_XATTR_CREATE) {
- LASSERT(buf->lb_len == lov_mds_md_size(ea_off + 1,
- LOV_MAGIC_V1));
+ if (fl == LU_XATTR_CREATE || reset) {
+ __u32 pattern = LOV_PATTERN_RAID0;
+
+ count = ea_off + 1;
+ LASSERT(buf->lb_len == lov_mds_md_size(count, LOV_MAGIC_V1));
+
+ if (ea_off != 0 || reset)
+ pattern |= LOV_PATTERN_F_HOLE;
memset(lmm, 0, buf->lb_len);
lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
- /* XXX: currently, we only support LOV_PATTERN_RAID0. */
- lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
+ lmm->lmm_pattern = cpu_to_le32(pattern);
fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
- /* XXX: We cannot know the stripe size,
- * then use the default value (1 MB). */
- lmm->lmm_stripe_size = cpu_to_le32(1024 * 1024);
- lmm->lmm_layout_gen = cpu_to_le16(0);
- objs = &(lmm->lmm_objects[ea_off]);
+
+ rc = lfsck_layout_get_def_stripesize(env, lfsck,
+ &lmm->lmm_stripe_size);
+ if (rc != 0)
+ RETURN(rc);
+
+ objs = &lmm->lmm_objects[ea_off];
} else {
- __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
- int gap = ea_off - count;
__u32 magic = le32_to_cpu(lmm->lmm_magic);
+ int gap;
- /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3
- * which has been verified in lfsck_layout_verify_header()
- * already. If some new magic introduced in the future,
- * then layout LFSCK needs to be updated also. */
- if (magic == LOV_MAGIC_V1) {
- objs = &(lmm->lmm_objects[count]);
- } else {
- LASSERT(magic == LOV_MAGIC_V3);
+ count = le16_to_cpu(lmm->lmm_stripe_count);
+ if (magic == LOV_MAGIC_V1)
+ objs = &lmm->lmm_objects[count];
+ else
objs = &((struct lov_mds_md_v3 *)lmm)->
lmm_objects[count];
- }
- if (gap > 0)
+ gap = ea_off - count;
+ if (gap >= 0)
+ count = ea_off + 1;
+ LASSERT(buf->lb_len == lov_mds_md_size(count, magic));
+
+ if (gap > 0) {
memset(objs, 0, gap * sizeof(*objs));
+ lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
+ }
+
lmm->lmm_layout_gen =
cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
objs += gap;
-
- LASSERT(buf->lb_len == lov_mds_md_size(ea_off + 1, magic));
}
- lmm->lmm_stripe_count = cpu_to_le16(ea_off + 1);
+ lmm->lmm_stripe_count = cpu_to_le16(count);
rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
fl, ost_idx);
pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
- /* In fact, the ff_parent::f_ver is not the real parent FID::f_ver,
- * instead, it is the OST-object index in its parent MDT-object
- * layout EA. */
- pfid->ff_parent.f_ver = cpu_to_le32(ea_off);
+ /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
+ * MDT-object's FID::f_ver, instead it is the OST-object index in its
+ * parent MDT-object's layout EA. */
+ pfid->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
}
/**
- * \retval +1: repaired
- * \retval 0: did nothing
- * \retval -ve: on error
+ * This function will create the MDT-object will the given (partial) LOV EA.
+ *
+ * Under some data corruption cases, the MDT-object of the file may be lost,
+ * but its OST-objects, or some of them are there. The layout LFSCK needs to
+ * re-create the MDT-object with the orphan OST-object(s) information.
+ *
+ * On the other hand, the LFSCK may has created some OST-object for repairing
+ * dangling LOV EA reference, but as the LFSCK processing, it may find that
+ * the old OST-object is there and should replace the former new created OST
+ * object. Unfortunately, some others have modified such newly created object.
+ * To keep the data (both new and old), the LFSCK will create MDT-object with
+ * new FID to reference the original OST-object.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] ltd pointer to target device descriptor
+ * \param[in] rec pointer to the record for the orphan OST-object
+ * \param[in] cfid pointer to FID for the orphan OST-object
+ * \param[in] infix additional information, such as the FID for original
+ * MDT-object and the stripe offset in the LOV EA
+ * \param[in] type the type for describing why the orphan MDT-object is
+ * created. The rules are as following:
+ *
+ * type "C": Multiple OST-objects claim the same MDT-object and the
+ * same slot in the layout EA. Then the LFSCK will create
+ * new MDT-object(s) to hold the conflict OST-object(s).
+ *
+ * type "N": The orphan OST-object does not know which one was the
+ * real parent MDT-object, so the LFSCK uses new FID for
+ * its parent MDT-object.
+ *
+ * type "R": The orphan OST-object knows its parent MDT-object FID,
+ * but does not know the position (the file name) in the
+ * namespace.
+ *
+ * The orphan name will be like:
+ * ${FID}-${infix}-${type}-${conflict_version}
+ *
+ * \param[in] ea_off the stripe offset in the LOV EA
+ *
+ * \retval positive on repaired something
+ * \retval if needs to repair nothing
+ * \retval negative error number on failure
*/
static int lfsck_layout_recreate_parent(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_tgt_desc *ltd,
struct lu_orphan_rec *rec,
struct lu_fid *cfid,
- const char *prefix,
- const char *postfix,
+ const char *infix,
+ const char *type,
__u32 ea_off)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct thandle *th = NULL;
struct lu_buf *pbuf = NULL;
struct lu_buf *ea_buf = &info->lti_big_buf;
+ struct lustre_handle lh = { 0 };
int buflen = ea_buf->lb_len;
+ int idx = 0;
int rc = 0;
ENTRY;
ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
- /* In fact, the ff_parent::f_ver is not the real parent FID::f_ver,
- * instead, it is the OST-object index in its parent MDT-object
- * layout EA. */
- ff->ff_parent.f_ver = cpu_to_le32(ea_off);
+ /* Currently, the filter_fid::ff_parent::f_ver is not the
+ * real parent MDT-object's FID::f_ver, instead it is the
+ * OST-object index in its parent MDT-object's layout EA. */
+ ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
pbuf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
if (IS_ERR(cobj))
RETURN(PTR_ERR(cobj));
}
- CDEBUG(D_LFSCK, "Re-create the lost MDT-object: parent "
- DFID", child "DFID", OST-index %u, stripe-index %u, "
- "prefix %s, postfix %s\n",
- PFID(pfid), PFID(cfid), ltd->ltd_index, ea_off, prefix, postfix);
+ CDEBUG(D_LFSCK, "Re-create the lost MDT-object: parent "DFID", child "
+ DFID", OST-index %u, stripe-index %u, infix %s, type %s\n",
+ PFID(pfid), PFID(cfid), ltd->ltd_index, ea_off, infix, type);
pobj = lfsck_object_find_by_dev(env, lfsck->li_bottom, pfid);
if (IS_ERR(pobj))
GOTO(put, rc = PTR_ERR(pobj));
- LASSERT(prefix != NULL);
- LASSERT(postfix != NULL);
+ LASSERT(infix != NULL);
+ LASSERT(type != NULL);
- /** name rules:
- *
- * 1. Use the MDT-object's FID as the name with prefix and postfix.
- *
- * 1.1 prefix "C-": More than one OST-objects cliam the same
- * MDT-object and the same slot in the layout EA.
- * It may be created for dangling referenced MDT
- * object or may be not.
- * 1.2 prefix "N-": The orphan OST-object does not know which one
- * is the real parent, so the LFSCK assign a new
- * FID as its parent.
- * 1.3 prefix "R-": The orphan OST-object know its parent FID but
- * does not know the position in the namespace.
- *
- * 2. If there is name conflict, increase FID::f_ver for new name. */
- sprintf(name, "%s"DFID"%s", prefix, PFID(pfid), postfix);
do {
+ snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
+ type, idx++);
rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
(const struct dt_key *)name, BYPASS_CAPA);
if (rc != 0 && rc != -ENOENT)
GOTO(put, rc);
-
- if (unlikely(rc == 0)) {
- CWARN("%s: The name %s under lost+found has been used "
- "by the "DFID". Try to increase the FID version "
- "for the new file name.\n",
- lfsck_lfsck2name(lfsck), name, PFID(tfid));
- *tfid = *pfid;
- tfid->f_ver++;
- sprintf(name, "%s"DFID"%s", prefix, PFID(tfid), postfix);
- }
} while (rc == 0);
memset(la, 0, sizeof(*la));
la->la_uid = rec->lor_uid;
la->la_gid = rec->lor_gid;
- la->la_mode = S_IFREG | S_IRUSR | S_IWUSR;
+ la->la_mode = S_IFREG | S_IRUSR;
la->la_valid = LA_MODE | LA_UID | LA_GID;
memset(dof, 0, sizeof(*dof));
ea_buf->lb_len = rc;
}
+ /* Hold update lock on the .lustre/lost+found/MDTxxxx/.
+ *
+ * XXX: Currently, we do not grab the PDO lock as normal create cases,
+ * because creating MDT-object for orphan OST-object is rare, we
+ * do not much care about the performance. It can be improved in
+ * the future when needed. */
+ rc = lfsck_layout_lock(env, com, lfsck->li_lpf_obj, &lh,
+ MDS_INODELOCK_UPDATE);
+ if (rc != 0)
+ GOTO(put, rc);
+
th = dt_trans_create(env, next);
if (IS_ERR(th))
- GOTO(put, rc = PTR_ERR(th));
+ GOTO(unlock, rc = PTR_ERR(th));
/* 1a. Update OST-object's parent information remotely.
*
rc = dt_create(env, pobj, la, NULL, dof, th);
if (rc == 0)
/* 3b. Add layout EA for the MDT-object. */
- rc = lfsck_layout_extend_lovea(env, th, pobj, cfid, ea_buf,
- LU_XATTR_CREATE, ltd->ltd_index,
- ea_off);
+ rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
+ ea_buf, LU_XATTR_CREATE,
+ ltd->ltd_index, ea_off, false);
dt_write_unlock(env, pobj);
if (rc < 0)
GOTO(stop, rc);
stop:
dt_trans_stop(env, next, th);
+
+unlock:
+ lfsck_layout_unlock(&lh);
+
put:
if (cobj != NULL && !IS_ERR(cobj))
lu_object_put(env, &cobj->do_lu);
ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
if (unlikely(ltd == NULL))
- RETURN(-ENODEV);
+ RETURN(-ENXIO);
exp = ltd->ltd_exp;
if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
if (rc != ELDLM_OK)
GOTO(put, rc = -EIO);
+ dt_write_lock(env, obj, 0);
+ /* Get obj's attr within lock again. */
+ rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ if (la->la_ctime != 0)
+ GOTO(unlock, rc = -ETXTBSY);
+
th = dt_trans_create(env, dev);
if (IS_ERR(th))
- GOTO(unlock1, rc = PTR_ERR(th));
+ GOTO(unlock, rc = PTR_ERR(th));
rc = dt_declare_ref_del(env, obj, th);
if (rc != 0)
if (rc != 0)
GOTO(stop, rc);
- dt_write_lock(env, obj, 0);
- /* Get obj's attr within lock again. */
- rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
- if (rc != 0)
- GOTO(unlock2, rc);
-
- if (la->la_ctime != 0)
- GOTO(unlock2, rc = -ETXTBSY);
-
rc = dt_ref_del(env, obj, th);
if (rc != 0)
- GOTO(unlock2, rc);
+ GOTO(stop, rc);
rc = dt_destroy(env, obj, th);
if (rc == 0)
"But the original missed OST-object is found now.\n",
PFID(fid));
- GOTO(unlock2, rc);
-
-unlock2:
- dt_write_unlock(env, obj);
+ GOTO(stop, rc);
stop:
dt_trans_stop(env, dev, th);
-unlock1:
+unlock:
+ dt_write_unlock(env, obj);
ldlm_lock_decref(&lh, LCK_EX);
put:
struct lfsck_thread_info *info = lfsck_env_info(env);
struct lu_fid *cfid2 = &info->lti_fid2;
struct ost_id *oi = &info->lti_oi;
+ char *infix = info->lti_tmpbuf;
struct lov_mds_md_v1 *lmm = ea_buf->lb_buf;
struct dt_device *dev = com->lc_lfsck->li_bottom;
struct thandle *th = NULL;
struct lustre_handle lh = { 0 };
- char postfix[64];
__u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
int rc = 0;
ENTRY;
ea_buf->lb_len = ori_len;
fid_zero(&rec->lor_fid);
- snprintf(postfix, 64, "-"DFID"-%x",
+ snprintf(infix, LFSCK_TMPBUF_LEN, "-"DFID"-%x",
PFID(lu_object_fid(&parent->do_lu)), ea_off);
rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
- "C-", postfix, ea_off);
+ infix, "C", ea_off);
RETURN(rc);
}
struct lustre_handle lh = { 0 };
__u32 magic;
int fl = 0;
- int rc;
+ int rc = 0;
int rc1;
int i;
__u16 count;
+ bool locked = false;
ENTRY;
CDEBUG(D_LFSCK, "Re-create the crashed layout EA: parent "
RETURN(rc);
again:
+ if (locked) {
+ dt_write_unlock(env, parent);
+ locked = false;
+ }
+
+ if (handle != NULL) {
+ dt_trans_stop(env, dt, handle);
+ handle = NULL;
+ }
+
+ if (rc < 0)
+ GOTO(unlock_layout, rc);
+
+ if (buf->lb_len < rc) {
+ lu_buf_realloc(buf, rc);
+ buflen = buf->lb_len;
+ if (buf->lb_buf == NULL)
+ GOTO(unlock_layout, rc = -ENOMEM);
+ }
+
if (!(bk->lb_param & LPF_DRYRUN)) {
handle = dt_trans_create(env, dt);
if (IS_ERR(handle))
}
dt_write_lock(env, parent, 0);
+ locked = true;
rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
if (rc == -ERANGE) {
rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV,
BYPASS_CAPA);
LASSERT(rc != 0);
-
- dt_write_unlock(env, parent);
- if (handle != NULL) {
- dt_trans_stop(env, dt, handle);
- handle = NULL;
- }
-
- if (rc < 0)
- GOTO(unlock_layout, rc);
-
- lu_buf_realloc(buf, rc);
- buflen = buf->lb_len;
- if (buf->lb_buf == NULL)
- GOTO(unlock_layout, rc = -ENOMEM);
-
- fl = LU_XATTR_REPLACE;
goto again;
} else if (rc == -ENODATA || rc == 0) {
+ rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
+ /* If the declared is not big enough, re-try. */
+ if (buf->lb_len < rc)
+ goto again;
+
fl = LU_XATTR_CREATE;
} else if (rc < 0) {
GOTO(unlock_parent, rc);
} else if (unlikely(buf->lb_len == 0)) {
- dt_write_unlock(env, parent);
- if (handle != NULL) {
- dt_trans_stop(env, dt, handle);
- handle = NULL;
- }
-
- lu_buf_alloc(buf, rc);
- buflen = buf->lb_len;
- if (buf->lb_buf == NULL)
- GOTO(unlock_layout, rc = -ENOMEM);
-
- fl = LU_XATTR_REPLACE;
goto again;
} else {
fl = LU_XATTR_REPLACE;
if (bk->lb_param & LPF_DRYRUN)
GOTO(unlock_parent, rc = 1);
- rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
- /* If the declared is not big enough, re-try. */
- if (buf->lb_len < rc) {
- dt_write_unlock(env, parent);
- if (handle != NULL) {
- dt_trans_stop(env, dt, handle);
- handle = NULL;
- }
-
- lu_buf_realloc(buf, rc);
- buflen = buf->lb_len;
- if (buf->lb_buf == NULL)
- GOTO(unlock_layout, rc = -ENOMEM);
-
- goto again;
- }
+ LASSERT(buf->lb_len >= rc);
buf->lb_len = rc;
- rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
- fl, ost_idx, ea_off);
+ rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
+ buf, fl, ost_idx, ea_off, false);
GOTO(unlock_parent, rc);
}
lmm = buf->lb_buf;
rc1 = lfsck_layout_verify_header(lmm);
+
+ /* If the LOV EA crashed, the rebuild it. */
+ if (rc1 == -EINVAL) {
+ if (bk->lb_param & LPF_DRYRUN)
+ GOTO(unlock_parent, rc = 1);
+
+ LASSERT(buf->lb_len >= rc);
+
+ buf->lb_len = rc;
+ memset(lmm, 0, buf->lb_len);
+ rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
+ buf, fl, ost_idx, ea_off, true);
+
+ GOTO(unlock_parent, rc);
+ }
+
+ /* For other unknown magic/pattern, keep the current LOV EA. */
if (rc1 != 0)
GOTO(unlock_parent, rc = rc1);
* be updated also. */
magic = le32_to_cpu(lmm->lmm_magic);
if (magic == LOV_MAGIC_V1) {
- objs = &(lmm->lmm_objects[0]);
+ objs = &lmm->lmm_objects[0];
} else {
LASSERT(magic == LOV_MAGIC_V3);
objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
if (bk->lb_param & LPF_DRYRUN)
GOTO(unlock_parent, rc = 1);
- rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
+ rc = lov_mds_md_size(ea_off + 1, magic);
/* If the declared is not big enough, re-try. */
- if (buf->lb_len < rc) {
- dt_write_unlock(env, parent);
- if (handle != NULL) {
- dt_trans_stop(env, dt, handle);
- handle = NULL;
- }
-
- lu_buf_realloc(buf, rc);
- buflen = buf->lb_len;
- if (buf->lb_buf == NULL)
- GOTO(unlock_layout, rc = -ENOMEM);
-
+ if (buf->lb_len < rc)
goto again;
- }
buf->lb_len = rc;
- rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
- fl, ost_idx, ea_off);
+ rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
+ buf, fl, ost_idx, ea_off, false);
+
GOTO(unlock_parent, rc);
}
for (i = 0; i < count; i++, objs++) {
/* The MDT-object was created via lfsck_layout_recover_create()
* by others before, and we fill the dummy layout EA. */
- if (is_dummy_lov_ost_data(objs)) {
+ if (lovea_slot_is_dummy(objs)) {
if (i != ea_off)
continue;
dt_trans_stop(env, dt, handle);
lfsck_layout_unlock(&lh);
if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
- objs = &(lmm->lmm_objects[ea_off]);
+ objs = &lmm->lmm_objects[ea_off];
else
objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
RETURN(rc);
unlock_parent:
- dt_write_unlock(env, parent);
+ if (locked)
+ dt_write_unlock(env, parent);
stop:
if (handle != NULL)
struct lfsck_layout *lo = com->lc_file_ram;
struct lu_fid *pfid = &rec->lor_fid;
struct dt_object *parent = NULL;
- __u32 ea_off = pfid->f_ver;
+ __u32 ea_off = pfid->f_stripe_idx;
int rc = 0;
ENTRY;
if (fid_is_zero(pfid)) {
rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
- "N-", "", ea_off);
+ "", "N", ea_off);
GOTO(out, rc);
}
if (dt_object_exists(parent) == 0) {
lu_object_put(env, &parent->do_lu);
rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
- "R-", "", ea_off);
+ "", "R", ea_off);
GOTO(out, rc);
}
return rc > 0 ? 0 : rc;
}
-/* For the MDT-object with dangling reference, we need to re-create
- * the missed OST-object with the known FID/owner information. */
-static int lfsck_layout_recreate_ostobj(const struct lu_env *env,
+/* For the MDT-object with dangling reference, we need to repare the
+ * inconsistency according to the LFSCK sponsor's requirement:
+ *
+ * 1) Keep the inconsistency there and report the inconsistency case,
+ * then give the chance to the application to find related issues,
+ * and the users can make the decision about how to handle it with
+ * more human knownledge. (by default)
+ *
+ * 2) Re-create the missed OST-object with the FID/owner information. */
+static int lfsck_layout_repair_dangling(const struct lu_env *env,
struct lfsck_component *com,
struct lfsck_layout_req *llr,
- struct lu_attr *la)
+ const struct lu_attr *pla)
{
struct lfsck_thread_info *info = lfsck_env_info(env);
struct filter_fid *pfid = &info->lti_new_pfid;
struct dt_allocation_hint *hint = &info->lti_hint;
+ struct lu_attr *cla = &info->lti_la2;
struct dt_object *parent = llr->llr_parent->llo_obj;
struct dt_object *child = llr->llr_child;
struct dt_device *dev = lfsck_obj2dt_dev(child);
struct lu_buf *buf;
struct lustre_handle lh = { 0 };
int rc;
+ bool create;
ENTRY;
- CDEBUG(D_LFSCK, "Repair dangling reference for: parent "DFID
- ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
+ if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
+ create = true;
+ else
+ create = false;
+
+ CDEBUG(D_LFSCK, "Found dangling reference for: parent "DFID
+ ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u. %s",
PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
- llr->llr_ost_idx, llr->llr_lov_idx, la->la_uid, la->la_gid);
+ llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid,
+ create ? "Create the lost OST-object as required.\n" :
+ "Keep the MDT-object there by default.\n");
+
+ if (!create)
+ RETURN(1);
+
+ memset(cla, 0, sizeof(*cla));
+ cla->la_uid = pla->la_uid;
+ cla->la_gid = pla->la_gid;
+ cla->la_mode = S_IFREG | 0666;
+ cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
+ LA_ATIME | LA_MTIME | LA_CTIME;
rc = lfsck_layout_lock(env, com, parent, &lh,
MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
hint->dah_mode = 0;
pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
- pfid->ff_parent.f_ver = cpu_to_le32(llr->llr_lov_idx);
+ /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
+ * MDT-object's FID::f_ver, instead it is the OST-object index in its
+ * parent MDT-object's layout EA. */
+ pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
- rc = dt_declare_create(env, child, la, hint, NULL, handle);
+ rc = dt_declare_create(env, child, cla, hint, NULL, handle);
if (rc != 0)
GOTO(stop, rc);
if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
GOTO(unlock2, rc = 1);
- rc = dt_create(env, child, la, hint, NULL, handle);
+ rc = dt_create(env, child, cla, hint, NULL, handle);
if (rc != 0)
GOTO(unlock2, rc);
pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
- /* The ff_parent->f_ver is not the real parent fid->f_ver. Instead,
- * it is the OST-object index in the parent MDT-object layout. */
- pfid->ff_parent.f_ver = cpu_to_le32(llr->llr_lov_idx);
+ /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
+ * MDT-object's FID::f_ver, instead it is the OST-object index in its
+ * parent MDT-object's layout EA. */
+ pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
GOTO(unlock2, rc = 0);
lmm = buf->lb_buf;
- rc = lfsck_layout_verify_header(lmm);
- if (rc != 0)
- GOTO(unlock2, rc);
-
/* Someone change layout during the LFSCK, no need to repair then. */
if (le16_to_cpu(lmm->lmm_layout_gen) != llr->llr_parent->llo_gen)
GOTO(unlock2, rc = 0);
* be updated also. */
magic = le32_to_cpu(lmm->lmm_magic);
if (magic == LOV_MAGIC_V1) {
- objs = &(lmm->lmm_objects[0]);
+ objs = &lmm->lmm_objects[0];
} else {
LASSERT(magic == LOV_MAGIC_V3);
objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
GOTO(out, rc);
lmm = buf->lb_buf;
- rc = lfsck_layout_verify_header(lmm);
- if (rc != 0)
- GOTO(out, rc);
-
- /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
- * been verified in lfsck_layout_verify_header() already. If some
- * new magic introduced in the future, then layout LFSCK needs to
- * be updated also. */
magic = le32_to_cpu(lmm->lmm_magic);
if (magic == LOV_MAGIC_V1) {
- objs = &(lmm->lmm_objects[0]);
+ objs = &lmm->lmm_objects[0];
} else {
LASSERT(magic == LOV_MAGIC_V3);
objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
struct lu_fid *tfid = &info->lti_fid2;
struct ost_id *oi = &info->lti_oi;
- if (is_dummy_lov_ost_data(objs))
+ if (lovea_slot_is_dummy(objs))
continue;
ostid_le_to_cpu(&objs->l_ost_oi, oi);
fid_zero(pfid);
} else {
fid_le_to_cpu(pfid, &pea->ff_parent);
- /* OST-object does not save parent FID::f_ver, instead,
- * the OST-object index in the parent MDT-object layout
- * EA reuses the pfid->f_ver. */
- idx = pfid->f_ver;
+ /* Currently, the filter_fid::ff_parent::f_ver is not the
+ * real parent MDT-object's FID::f_ver, instead it is the
+ * OST-object index in its parent MDT-object's layout EA. */
+ idx = pfid->f_stripe_idx;
pfid->f_ver = 0;
}
switch (type) {
case LLIT_DANGLING:
- memset(cla, 0, sizeof(*cla));
- cla->la_uid = pla->la_uid;
- cla->la_gid = pla->la_gid;
- cla->la_mode = S_IFREG | 0666;
- cla->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
- LA_ATIME | LA_MTIME | LA_CTIME;
- rc = lfsck_layout_recreate_ostobj(env, com, llr, cla);
+ rc = lfsck_layout_repair_dangling(env, com, llr, pla);
break;
case LLIT_UNMATCHED_PAIR:
rc = lfsck_layout_repair_unmatched_pair(env, com, llr, pla);
out:
down_write(&com->lc_sem);
if (rc < 0) {
- /* If cannot touch the target server,
- * mark the LFSCK as INCOMPLETE. */
- if (rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -ETIMEDOUT ||
- rc == -EHOSTDOWN || rc == -EHOSTUNREACH) {
+ struct lfsck_layout_master_data *llmd = com->lc_data;
+
+ if (unlikely(llmd->llmd_exit)) {
+ rc = 0;
+ } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
+ rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
+ rc == -EHOSTUNREACH) {
+ /* If cannot touch the target server,
+ * mark the LFSCK as INCOMPLETE. */
CERROR("%s: Fail to talk with OST %x: rc = %d.\n",
lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
lo->ll_flags |= LF_INCOMPLETE;
lo->ll_objs_skipped++;
rc = 0;
} else {
- lo->ll_objs_failed_phase1++;
+ lfsck_layout_record_failure(env, lfsck, lo);
}
} else if (rc > 0) {
LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
"unknown type = %d\n", type);
lo->ll_objs_repaired[type - 1]++;
+ if (bk->lb_param & LPF_DRYRUN &&
+ unlikely(lo->ll_pos_first_inconsistent == 0))
+ lo->ll_pos_first_inconsistent =
+ lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
+ lfsck->li_di_oit);
}
up_write(&com->lc_sem);
memset(lr, 0, sizeof(*lr));
lr->lr_event = LE_START;
lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
- LSV_ASYNC_WINDOWS;
+ LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ;
lr->lr_speed = bk->lb_speed_limit;
lr->lr_version = bk->lb_version;
lr->lr_param = bk->lb_param;
while (!list_empty(&llmd->llmd_req_list)) {
bool wakeup = false;
- if (unlikely(llmd->llmd_exit))
+ if (unlikely(llmd->llmd_exit ||
+ !thread_is_running(mthread)))
GOTO(cleanup1, rc = llmd->llmd_post_result);
llr = list_entry(llmd->llmd_req_list.next,
* handled to avoid too frequent thread schedule. */
if (llmd->llmd_prefetched == 0 ||
(bk->lb_async_windows != 0 &&
- (bk->lb_async_windows >> 1) ==
+ bk->lb_async_windows / 2 ==
llmd->llmd_prefetched))
wakeup = true;
spin_unlock(&llmd->llmd_lock);
com->lc_time_last_checkpoint +
cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
+ /* flush all async updating before handling orphan. */
+ dt_sync(env, lfsck->li_next);
+
while (llmd->llmd_in_double_scan) {
struct lfsck_tgt_descs *ltds =
&lfsck->li_ost_descs;
/* Under force exit case, some requests may be just freed without
* verification, those objects should be re-handled when next run.
* So not update the on-disk tracing file under such case. */
- if (!llmd->llmd_exit)
+ if (llmd->llmd_in_double_scan && !llmd->llmd_exit)
rc1 = lfsck_layout_double_scan_result(env, com, rc);
fini:
RETURN_EXIT;
}
+/*
+ * \ret -ENODATA: unrecognized stripe
+ * \ret = 0 : recognized stripe
+ * \ret < 0 : other failures
+ */
+static int lfsck_layout_master_check_pairs(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lu_fid *cfid,
+ struct lu_fid *pfid)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_buf *buf = &info->lti_big_buf;
+ struct ost_id *oi = &info->lti_oi;
+ struct dt_object *obj;
+ struct lov_mds_md_v1 *lmm;
+ struct lov_ost_data_v1 *objs;
+ __u32 idx = pfid->f_stripe_idx;
+ __u32 magic;
+ int rc = 0;
+ int i;
+ __u16 count;
+ ENTRY;
+
+ pfid->f_ver = 0;
+ obj = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
+ if (IS_ERR(obj))
+ RETURN(PTR_ERR(obj));
+
+ dt_read_lock(env, obj, 0);
+ if (unlikely(!dt_object_exists(obj)))
+ GOTO(unlock, rc = -ENOENT);
+
+ rc = lfsck_layout_get_lovea(env, obj, buf, NULL);
+ if (rc < 0)
+ GOTO(unlock, rc);
+
+ if (rc == 0)
+ GOTO(unlock, rc = -ENODATA);
+
+ lmm = buf->lb_buf;
+ rc = lfsck_layout_verify_header(lmm);
+ if (rc != 0)
+ GOTO(unlock, rc);
+
+ /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
+ * been verified in lfsck_layout_verify_header() already. If some
+ * new magic introduced in the future, then layout LFSCK needs to
+ * be updated also. */
+ magic = le32_to_cpu(lmm->lmm_magic);
+ if (magic == LOV_MAGIC_V1) {
+ objs = &lmm->lmm_objects[0];
+ } else {
+ LASSERT(magic == LOV_MAGIC_V3);
+ objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
+ }
+
+ fid_to_ostid(cfid, oi);
+ count = le16_to_cpu(lmm->lmm_stripe_count);
+ for (i = 0; i < count; i++, objs++) {
+ struct ost_id oi2;
+
+ ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
+ if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
+ GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
+ }
+
+ GOTO(unlock, rc = -ENODATA);
+
+unlock:
+ dt_read_unlock(env, obj);
+ lu_object_put(env, &obj->do_lu);
+
+ return rc;
+}
+
+/*
+ * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
+ * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
+ * layout EA from MDT to OST. On one hand, the OST no need to understand
+ * the layout EA structure; on the other hand, it may cause trouble when
+ * transfer large layout EA from MDT to OST via normal OUT RPC.
+ *
+ * \ret > 0: unrecognized stripe
+ * \ret = 0: recognized stripe
+ * \ret < 0: other failures
+ */
+static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lu_fid *cfid,
+ struct lu_fid *pfid)
+{
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct obd_device *obd = lfsck->li_obd;
+ struct seq_server_site *ss =
+ lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
+ struct obd_export *exp = NULL;
+ struct ptlrpc_request *req = NULL;
+ struct lfsck_request *lr;
+ struct lu_seq_range range = { 0 };
+ int rc = 0;
+ ENTRY;
+
+ if (unlikely(fid_is_idif(pfid)))
+ RETURN(1);
+
+ fld_range_set_any(&range);
+ rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), &range);
+ if (rc != 0)
+ RETURN(rc == -ENOENT ? 1 : rc);
+
+ if (unlikely(!fld_range_is_mdt(&range)))
+ RETURN(1);
+
+ exp = lustre_find_lwp_by_index(obd->obd_name, range.lsr_index);
+ if (unlikely(exp == NULL))
+ RETURN(1);
+
+ if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
+ GOTO(out, rc = -EOPNOTSUPP);
+
+ req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
+ if (req == NULL)
+ GOTO(out, rc = -ENOMEM);
+
+ rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
+ if (rc != 0) {
+ ptlrpc_request_free(req);
+
+ GOTO(out, rc);
+ }
+
+ lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
+ memset(lr, 0, sizeof(*lr));
+ lr->lr_event = LE_PAIRS_VERIFY;
+ lr->lr_active = LT_LAYOUT;
+ lr->lr_fid = *cfid; /* OST-object itself FID. */
+ lr->lr_fid2 = *pfid; /* The claimed parent FID. */
+
+ ptlrpc_request_set_replen(req);
+ rc = ptlrpc_queue_wait(req);
+ ptlrpc_req_finished(req);
+
+ if (rc == -ENOENT || rc == -ENODATA)
+ rc = 1;
+
+ GOTO(out, rc);
+
+out:
+ if (exp != NULL)
+ class_export_put(exp);
+
+ return rc;
+}
+
+static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct lfsck_request *lr)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct filter_fid *ff = &info->lti_new_pfid;
+ struct lu_buf *buf;
+ struct dt_device *dev = com->lc_lfsck->li_bottom;
+ struct dt_object *obj;
+ struct thandle *th = NULL;
+ int rc = 0;
+ ENTRY;
+
+ obj = lfsck_object_find_by_dev(env, dev, &lr->lr_fid);
+ if (IS_ERR(obj))
+ RETURN(PTR_ERR(obj));
+
+ fid_cpu_to_le(&ff->ff_parent, &lr->lr_fid2);
+ buf = lfsck_buf_get(env, ff, sizeof(*ff));
+ dt_write_lock(env, obj, 0);
+ if (unlikely(!dt_object_exists(obj)))
+ GOTO(unlock, rc = 0);
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ GOTO(unlock, rc = PTR_ERR(th));
+
+ rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_FID, 0, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_xattr_set(env, obj, buf, XATTR_NAME_FID, 0, th, BYPASS_CAPA);
+
+ GOTO(stop, rc);
+
+stop:
+ dt_trans_stop(env, dev, th);
+
+unlock:
+ dt_write_unlock(env, obj);
+ lu_object_put(env, &obj->do_lu);
+
+ return rc;
+}
+
/* layout APIs */
static int lfsck_layout_reset(const struct lu_env *env,
down_write(&com->lc_sem);
if (new_checked)
com->lc_new_checked++;
- lo->ll_objs_failed_phase1++;
- if (lo->ll_pos_first_inconsistent == 0) {
- struct lfsck_instance *lfsck = com->lc_lfsck;
-
- lo->ll_pos_first_inconsistent =
- lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
- lfsck->li_di_oit);
- }
+ lfsck_layout_record_failure(env, com->lc_lfsck, lo);
up_write(&com->lc_sem);
}
int rc;
rc = lfsck_layout_reset(env, com, false);
+ if (rc == 0)
+ rc = lfsck_set_param(env, lfsck, start, true);
+
if (rc != 0)
return rc;
}
struct lfsck_start_param *lsp)
{
struct lfsck_layout_slave_data *llsd = com->lc_data;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_layout *lo = com->lc_file_ram;
struct lfsck_start *start = lsp->lsp_start;
int rc;
rc = lfsck_layout_prep(env, com, start);
- if (rc != 0 || !lsp->lsp_index_valid)
+ if (rc != 0)
return rc;
+ if (lo->ll_flags & LF_CRASHED_LASTID &&
+ list_empty(&llsd->llsd_master_list)) {
+ LASSERT(lfsck->li_out_notify != NULL);
+
+ lfsck->li_out_notify(env, lfsck->li_out_notify_data,
+ LE_LASTID_REBUILDING);
+ }
+
+ if (!lsp->lsp_index_valid)
+ return 0;
+
rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
if (rc == 0 && start != NULL && start->ls_flags & LPF_ORPHAN) {
LASSERT(!llsd->llsd_rbtree_valid);
struct ptlrpc_thread *mthread = &lfsck->li_thread;
struct ptlrpc_thread *athread = &llmd->llmd_thread;
struct lfsck_thread_args *lta;
- long rc;
+ struct task_struct *task;
+ int rc;
ENTRY;
rc = lfsck_layout_prep(env, com, lsp->lsp_start);
if (IS_ERR(lta))
RETURN(PTR_ERR(lta));
- rc = PTR_ERR(kthread_run(lfsck_layout_assistant, lta, "lfsck_layout"));
- if (IS_ERR_VALUE(rc)) {
+ task = kthread_run(lfsck_layout_assistant, lta, "lfsck_layout");
+ if (IS_ERR(task)) {
+ rc = PTR_ERR(task);
CERROR("%s: Cannot start LFSCK layout assistant thread: "
- "rc = %ld\n", lfsck_lfsck2name(lfsck), rc);
+ "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
lfsck_thread_args_fini(lta);
} else {
struct l_wait_info lwi = { 0 };
* be updated also. */
magic = le32_to_cpu(lmm->lmm_magic);
if (magic == LOV_MAGIC_V1) {
- objs = &(lmm->lmm_objects[0]);
+ objs = &lmm->lmm_objects[0];
} else {
LASSERT(magic == LOV_MAGIC_V3);
objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
le32_to_cpu(objs->l_ost_idx);
bool wakeup = false;
- if (is_dummy_lov_ost_data(objs))
+ if (unlikely(lovea_slot_is_dummy(objs)))
continue;
l_wait_event(mthread->t_ctl_waitq,
down_write(&com->lc_sem);
com->lc_new_checked++;
if (rc < 0)
- lo->ll_objs_failed_phase1++;
+ lfsck_layout_record_failure(env, lfsck, lo);
up_write(&com->lc_sem);
if (cobj != NULL && !IS_ERR(cobj))
buf->lb_len = rc;
lmm = buf->lb_buf;
rc = lfsck_layout_verify_header(lmm);
+ /* If the LOV EA crashed, then it is possible to be rebuilt later
+ * when handle orphan OST-objects. */
if (rc != 0)
GOTO(out, rc);
down_write(&com->lc_sem);
com->lc_new_checked++;
if (rc < 0)
- lo->ll_objs_failed_phase1++;
+ lfsck_layout_record_failure(env, lfsck, lo);
up_write(&com->lc_sem);
}
buf->lb_len = buflen;
LASSERT(llsd != NULL);
+ if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
+ cfs_fail_val == lfsck_dev_idx(lfsck->li_bottom)) {
+ struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(1),
+ NULL, NULL);
+ struct ptlrpc_thread *thread = &lfsck->li_thread;
+
+ l_wait_event(thread->t_ctl_waitq,
+ !thread_is_running(thread),
+ &lwi);
+ }
+
lfsck_rbtree_update_bitmap(env, com, fid, false);
down_write(&com->lc_sem);
}
static int lfsck_layout_dump(const struct lu_env *env,
- struct lfsck_component *com, char *buf, int len)
+ struct lfsck_component *com, struct seq_file *m)
{
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
struct lfsck_layout *lo = com->lc_file_ram;
- int save = len;
- int ret = -ENOSPC;
int rc;
down_read(&com->lc_sem);
- rc = snprintf(buf, len,
- "name: lfsck_layout\n"
+ seq_printf(m, "name: lfsck_layout\n"
"magic: %#x\n"
"version: %d\n"
"status: %s\n",
lo->ll_magic,
bk->lb_version,
lfsck_status2names(lo->ll_status));
- if (rc <= 0)
- goto out;
- buf += rc;
- len -= rc;
- rc = lfsck_bits_dump(&buf, &len, lo->ll_flags, lfsck_flags_names,
- "flags");
+ rc = lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
if (rc < 0)
goto out;
- rc = lfsck_bits_dump(&buf, &len, bk->lb_param, lfsck_param_names,
- "param");
+ rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
if (rc < 0)
goto out;
- rc = lfsck_time_dump(&buf, &len, lo->ll_time_last_complete,
+ rc = lfsck_time_dump(m, lo->ll_time_last_complete,
"time_since_last_completed");
if (rc < 0)
goto out;
- rc = lfsck_time_dump(&buf, &len, lo->ll_time_latest_start,
+ rc = lfsck_time_dump(m, lo->ll_time_latest_start,
"time_since_latest_start");
if (rc < 0)
goto out;
- rc = lfsck_time_dump(&buf, &len, lo->ll_time_last_checkpoint,
+ rc = lfsck_time_dump(m, lo->ll_time_last_checkpoint,
"time_since_last_checkpoint");
if (rc < 0)
goto out;
- rc = snprintf(buf, len,
- "latest_start_position: "LPU64"\n"
+ seq_printf(m, "latest_start_position: "LPU64"\n"
"last_checkpoint_position: "LPU64"\n"
"first_failure_position: "LPU64"\n",
lo->ll_pos_latest_start,
lo->ll_pos_last_checkpoint,
lo->ll_pos_first_inconsistent);
- if (rc <= 0)
- goto out;
- buf += rc;
- len -= rc;
-
- rc = snprintf(buf, len,
- "success_count: %u\n"
+ seq_printf(m, "success_count: %u\n"
"repaired_dangling: "LPU64"\n"
"repaired_unmatched_pair: "LPU64"\n"
"repaired_multiple_referenced: "LPU64"\n"
lo->ll_objs_skipped,
lo->ll_objs_failed_phase1,
lo->ll_objs_failed_phase2);
- if (rc <= 0)
- goto out;
-
- buf += rc;
- len -= rc;
if (lo->ll_status == LS_SCANNING_PHASE1) {
__u64 pos;
const struct dt_it_ops *iops;
cfs_duration_t duration = cfs_time_current() -
lfsck->li_time_last_checkpoint;
- __u64 checked = lo->ll_objs_checked_phase1 + com->lc_new_checked;
+ __u64 checked = lo->ll_objs_checked_phase1 +
+ com->lc_new_checked;
__u64 speed = checked;
__u64 new_checked = com->lc_new_checked * HZ;
__u32 rtime = lo->ll_run_time_phase1 +
do_div(new_checked, duration);
if (rtime != 0)
do_div(speed, rtime);
- rc = snprintf(buf, len,
- "checked_phase1: "LPU64"\n"
+ seq_printf(m, "checked_phase1: "LPU64"\n"
"checked_phase2: "LPU64"\n"
"run_time_phase1: %u seconds\n"
"run_time_phase2: %u seconds\n"
lo->ll_run_time_phase2,
speed,
new_checked);
- if (rc <= 0)
- goto out;
-
- buf += rc;
- len -= rc;
LASSERT(lfsck->li_di_oit != NULL);
pos = iops->store(env, lfsck->li_di_oit);
if (!lfsck->li_current_oit_processed)
pos--;
- rc = snprintf(buf, len, "current_position: "LPU64"\n", pos);
- if (rc <= 0)
- goto out;
+ seq_printf(m, "current_position: "LPU64"\n", pos);
- buf += rc;
- len -= rc;
} else if (lo->ll_status == LS_SCANNING_PHASE2) {
cfs_duration_t duration = cfs_time_current() -
lfsck->li_time_last_checkpoint;
- __u64 checked = lo->ll_objs_checked_phase1 + com->lc_new_checked;
- __u64 speed = checked;
+ __u64 checked = lo->ll_objs_checked_phase2 +
+ com->lc_new_checked;
+ __u64 speed1 = lo->ll_objs_checked_phase1;
+ __u64 speed2 = checked;
__u64 new_checked = com->lc_new_checked * HZ;
- __u32 rtime = lo->ll_run_time_phase1 +
+ __u32 rtime = lo->ll_run_time_phase2 +
cfs_duration_sec(duration + HALF_SEC);
if (duration != 0)
do_div(new_checked, duration);
+ if (lo->ll_run_time_phase1 != 0)
+ do_div(speed1, lo->ll_run_time_phase1);
if (rtime != 0)
- do_div(speed, rtime);
- rc = snprintf(buf, len,
- "checked_phase1: "LPU64"\n"
- "checked_phase2: "LPU64"\n"
- "run_time_phase1: %u seconds\n"
- "run_time_phase2: %u seconds\n"
- "average_speed_phase1: "LPU64" items/sec\n"
- "average_speed_phase2: N/A\n"
- "real-time_speed_phase1: "LPU64" items/sec\n"
- "real-time_speed_phase2: N/A\n"
- "current_position: "DFID"\n",
- checked,
- lo->ll_objs_checked_phase2,
- rtime,
- lo->ll_run_time_phase2,
- speed,
- new_checked,
- PFID(&com->lc_fid_latest_scanned_phase2));
+ do_div(speed2, rtime);
+ rc = seq_printf(m, "checked_phase1: "LPU64"\n"
+ "checked_phase2: "LPU64"\n"
+ "run_time_phase1: %u seconds\n"
+ "run_time_phase2: %u seconds\n"
+ "average_speed_phase1: "LPU64" items/sec\n"
+ "average_speed_phase2: "LPU64" items/sec\n"
+ "real-time_speed_phase1: N/A\n"
+ "real-time_speed_phase2: "LPU64" items/sec\n"
+ "current_position: "DFID"\n",
+ lo->ll_objs_checked_phase1,
+ checked,
+ lo->ll_run_time_phase1,
+ rtime,
+ speed1,
+ speed2,
+ new_checked,
+ PFID(&com->lc_fid_latest_scanned_phase2));
if (rc <= 0)
goto out;
- buf += rc;
- len -= rc;
} else {
__u64 speed1 = lo->ll_objs_checked_phase1;
__u64 speed2 = lo->ll_objs_checked_phase2;
do_div(speed1, lo->ll_run_time_phase1);
if (lo->ll_run_time_phase2 != 0)
do_div(speed2, lo->ll_run_time_phase2);
- rc = snprintf(buf, len,
- "checked_phase1: "LPU64"\n"
- "checked_phase2: "LPU64"\n"
- "run_time_phase1: %u seconds\n"
- "run_time_phase2: %u seconds\n"
- "average_speed_phase1: "LPU64" items/sec\n"
- "average_speed_phase2: "LPU64" objs/sec\n"
- "real-time_speed_phase1: N/A\n"
- "real-time_speed_phase2: N/A\n"
- "current_position: N/A\n",
- lo->ll_objs_checked_phase1,
- lo->ll_objs_checked_phase2,
- lo->ll_run_time_phase1,
- lo->ll_run_time_phase2,
- speed1,
- speed2);
- if (rc <= 0)
- goto out;
-
- buf += rc;
- len -= rc;
+ seq_printf(m, "checked_phase1: "LPU64"\n"
+ "checked_phase2: "LPU64"\n"
+ "run_time_phase1: %u seconds\n"
+ "run_time_phase2: %u seconds\n"
+ "average_speed_phase1: "LPU64" items/sec\n"
+ "average_speed_phase2: "LPU64" objs/sec\n"
+ "real-time_speed_phase1: N/A\n"
+ "real-time_speed_phase2: N/A\n"
+ "current_position: N/A\n",
+ lo->ll_objs_checked_phase1,
+ lo->ll_objs_checked_phase2,
+ lo->ll_run_time_phase1,
+ lo->ll_run_time_phase2,
+ speed1,
+ speed2);
}
- ret = save - len;
-
out:
up_read(&com->lc_sem);
- return ret;
+ return rc;
}
static int lfsck_layout_master_double_scan(const struct lu_env *env,
ltd_layout_list) {
list_del_init(<d->ltd_layout_list);
}
+ spin_unlock(<ds->ltd_lock);
+
+ ltds = &lfsck->li_mdt_descs;
+ spin_lock(<ds->ltd_lock);
list_for_each_entry_safe(ltd, next, &llmd->llmd_mdt_phase1_list,
ltd_layout_phase_list) {
list_del_init(<d->ltd_layout_phase_list);
bool fail = false;
ENTRY;
+ if (lr->lr_event == LE_PAIRS_VERIFY) {
+ int rc;
+
+ rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
+ &lr->lr_fid2);
+
+ RETURN(rc);
+ }
+
if (lr->lr_event != LE_PHASE1_DONE &&
lr->lr_event != LE_PHASE2_DONE &&
lr->lr_event != LE_PEER_EXIT)
if (ltd == NULL) {
spin_unlock(<ds->ltd_lock);
- RETURN(-ENODEV);
+ RETURN(-ENXIO);
}
list_del_init(<d->ltd_layout_phase_list);
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_layout_slave_data *llsd = com->lc_data;
struct lfsck_layout_slave_target *llst;
+ int rc;
ENTRY;
- if (lr->lr_event == LE_FID_ACCESSED) {
+ switch (lr->lr_event) {
+ case LE_FID_ACCESSED:
lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
-
RETURN(0);
- }
+ case LE_CONDITIONAL_DESTROY:
+ rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
+ RETURN(rc);
+ case LE_PAIRS_VERIFY: {
+ lr->lr_status = LPVS_INIT;
+ /* Firstly, if the MDT-object which is claimed via OST-object
+ * local stored PFID xattr recognizes the OST-object, then it
+ * must be that the client given PFID is wrong. */
+ rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
+ &lr->lr_fid3);
+ if (rc <= 0)
+ RETURN(0);
- if (lr->lr_event == LE_CONDITIONAL_DESTROY) {
- int rc;
+ lr->lr_status = LPVS_INCONSISTENT;
+ /* The OST-object local stored PFID xattr is stale. We need to
+ * check whether the MDT-object that is claimed via the client
+ * given PFID information recognizes the OST-object or not. If
+ * matches, then need to update the OST-object's PFID xattr. */
+ rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
+ &lr->lr_fid2);
+ /* For rc < 0 case:
+ * We are not sure whether the client given PFID information
+ * is correct or not, do nothing to avoid improper fixing.
+ *
+ * For rc > 0 case:
+ * The client given PFID information is also invalid, we can
+ * NOT fix the OST-object inconsistency.
+ */
+ if (rc != 0)
+ RETURN(rc);
- rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
+ lr->lr_status = LPVS_INCONSISTENT_TOFIX;
+ rc = lfsck_layout_slave_repair_pfid(env, com, lr);
RETURN(rc);
}
-
- if (lr->lr_event != LE_PHASE2_DONE && lr->lr_event != LE_PEER_EXIT)
+ case LE_PHASE2_DONE:
+ case LE_PEER_EXIT:
+ break;
+ default:
RETURN(-EINVAL);
+ }
llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
if (llst == NULL)
- RETURN(-ENODEV);
+ RETURN(-ENXIO);
lfsck_layout_llst_put(llst);
if (list_empty(&llsd->llsd_master_list))
lfsck = lfsck_instance_find(dev, true, false);
if (unlikely(lfsck == NULL))
- RETURN(ERR_PTR(-ENODEV));
+ RETURN(ERR_PTR(-ENXIO));
com = lfsck_component_find(lfsck, LT_LAYOUT);
if (unlikely(com == NULL))
it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
if (it->loi_llst == NULL)
- GOTO(out, rc = -ENODEV);
+ GOTO(out, rc = -ENXIO);
if (dev->dd_record_fid_accessed) {
/* The first iteration against the rbtree, scan the whole rbtree
GOTO(out, rc = -EINVAL);
fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
- /* In fact, the ff_parent::f_ver is not the real parent FID::f_ver,
- * instead, it is the OST-object index in its parent MDT-object
- * layout EA. */
- save = rec->lor_fid.f_ver;
+ /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
+ * MDT-object's FID::f_ver, instead it is the OST-object index in its
+ * parent MDT-object's layout EA. */
+ save = rec->lor_fid.f_stripe_idx;
rec->lor_fid.f_ver = 0;
rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
/* If the orphan OST-object does not claim the MDT, then next.
goto again1;
}
- rec->lor_fid.f_ver = save;
+ rec->lor_fid.f_stripe_idx = save;
rec->lor_uid = la->la_uid;
rec->lor_gid = la->la_gid;