* GPL HEADER END
*/
/*
- * Copyright (c) 2013, 2016, Intel Corporation.
+ * Copyright (c) 2013, 2017, Intel Corporation.
*/
/*
* lustre/lfsck/lfsck_namespace.c
dst->ln_time_latest_reset = le64_to_cpu(src->ln_time_latest_reset);
dst->ln_linkea_overflow_cleared =
le64_to_cpu(src->ln_linkea_overflow_cleared);
+ dst->ln_agent_entries_repaired =
+ le64_to_cpu(src->ln_agent_entries_repaired);
}
static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
dst->ln_time_latest_reset = cpu_to_le64(src->ln_time_latest_reset);
dst->ln_linkea_overflow_cleared =
cpu_to_le64(src->ln_linkea_overflow_cleared);
+ dst->ln_agent_entries_repaired =
+ cpu_to_le64(src->ln_agent_entries_repaired);
}
static void lfsck_namespace_record_failure(const struct lu_env *env,
}
if (ns->ln_bitmap_size == 0) {
- lad->lad_incomplete = 0;
+ clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
CFS_RESET_BITMAP(bitmap);
RETURN(0);
RETURN(rc >= 0 ? -EINVAL : rc);
if (cfs_bitmap_check_empty(bitmap))
- lad->lad_incomplete = 0;
+ clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
else
- lad->lad_incomplete = 1;
+ set_bit(LAD_INCOMPLETE, &lad->lad_flags);
RETURN(0);
}
if (new != 0) {
rc = dt_insert(env, obj, (const struct dt_rec *)&new,
- (const struct dt_key *)key, th, 1);
+ (const struct dt_key *)key, th);
if (rc != 0)
GOTO(log, rc);
}
return rc;
}
+static inline bool linkea_reclen_is_valid(const struct linkea_data *ldata)
+{
+ if (ldata->ld_reclen <= 0)
+ return false;
+
+ if ((char *)ldata->ld_lee + ldata->ld_reclen >
+ (char *)ldata->ld_leh + ldata->ld_leh->leh_len)
+ return false;
+
+ return true;
+}
+
+static inline bool linkea_entry_is_valid(const struct linkea_data *ldata,
+ const struct lu_name *cname,
+ const struct lu_fid *pfid)
+{
+ if (!linkea_reclen_is_valid(ldata))
+ return false;
+
+ if (cname->ln_namelen <= 0 || cname->ln_namelen > NAME_MAX)
+ return false;
+
+ if (!fid_is_sane(pfid))
+ return false;
+
+ return true;
+}
+
static int lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
struct lu_name *cname,
struct lu_fid *pfid,
char *buf, const int buflen)
{
linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
- if (unlikely(ldata->ld_reclen <= 0 ||
- ldata->ld_reclen + sizeof(struct link_ea_header) >
- ldata->ld_leh->leh_len ||
- cname->ln_namelen <= 0 ||
- cname->ln_namelen > NAME_MAX ||
- cname->ln_namelen >= buflen ||
- !fid_is_sane(pfid)))
+ if (unlikely(!linkea_entry_is_valid(ldata, cname, pfid)))
return -EINVAL;
/* To guarantee the 'name' is terminated with '0'. */
/* If current record is corrupted, all the subsequent
* records will be dropped. */
- if (unlikely(ldata->ld_reclen <= 0 ||
- ldata->ld_reclen + sizeof(struct link_ea_header) >
- ldata->ld_leh->leh_len)) {
+ if (unlikely(!linkea_reclen_is_valid(ldata))) {
void *ptr = ldata->ld_lee;
ldata->ld_leh->leh_len = sizeof(struct link_ea_header);
while (ldata->ld_lee != NULL) {
ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
ldata->ld_lee->lee_reclen[1];
- if (unlikely(ldata->ld_reclen == oldlen &&
+ if (unlikely(!linkea_reclen_is_valid(ldata))) {
+ lfsck_linkea_del_buf(ldata, NULL);
+ LASSERT(ldata->ld_lee == NULL);
+ } else if (unlikely(ldata->ld_reclen == oldlen &&
memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
repeated++;
if (!remove)
rec->rec_type = S_IFDIR;
rec->rec_fid = pfid;
rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, 1);
+ (const struct dt_key *)dotdot, th);
if (rc != 0)
GOTO(unlock, rc);
}
rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
rec->rec_fid = cfid;
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)cname->ln_name, th, 1);
+ (const struct dt_key *)cname->ln_name, th);
if (rc == 0 && S_ISDIR(rec->rec_type)) {
dt_write_lock(env, parent, 0);
rc = dt_ref_add(env, parent, th);
return rc;
}
+static int lfsck_lmv_set(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct dt_object *obj,
+ struct lmv_mds_md_v1 *lmv)
+{
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ struct lu_buf buf = { lmv, sizeof(*lmv) };
+ int rc;
+
+ ENTRY;
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ RETURN(PTR_ERR(th));
+
+ rc = dt_declare_xattr_set(env, obj, &buf, XATTR_NAME_LMV".set", 0, th);
+ if (rc)
+ 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_LMV".set", 0, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ EXIT;
+stop:
+ dt_trans_stop(env, dev, th);
+
+ return rc;
+}
+
+static int lfsck_lmv_delete(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct dt_object *obj)
+{
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ int rc;
+
+ ENTRY;
+
+ th = dt_trans_create(env, dev);
+ if (IS_ERR(th))
+ RETURN(PTR_ERR(th));
+
+ rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LMV, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, th);
+ if (rc != 0)
+ GOTO(stop, rc);
+
+ rc = dt_xattr_del(env, obj, XATTR_NAME_LMV, th);
+ if (rc)
+ GOTO(stop, rc);
+
+ EXIT;
+stop:
+ dt_trans_stop(env, dev, th);
+
+ return rc;
+}
+
+static inline int lfsck_object_is_shard(const struct lu_env *env,
+ struct lfsck_instance *lfsck,
+ struct dt_object *obj,
+ const struct lu_name *lname)
+{
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
+ int rc;
+
+ rc = lfsck_shard_name_to_index(env, lname->ln_name, lname->ln_namelen,
+ lfsck_object_type(obj),
+ lfsck_dto2fid(obj));
+ if (rc < 0)
+ return 0;
+
+ rc = lfsck_read_stripe_lmv(env, lfsck, obj, lmv);
+ if (rc == -ENODATA)
+ return 0;
+
+ if (!rc && lmv->lmv_magic == LMV_MAGIC_STRIPE)
+ return 1;
+
+ return rc;
+}
+
/**
* Add the specified name entry back to namespace.
*
* it is quite possible that the name entry is lost. Then the LFSCK
* should add the name entry back to the namespace.
*
+ * If \a child is shard, which means \a parent is a striped directory,
+ * if \a parent has LMV, we need to delete it before insertion because
+ * now parent's striping is broken and can't be parsed correctly.
+ *
* \param[in] env pointer to the thread context
* \param[in] com pointer to the lfsck component
* \param[in] parent pointer to the directory under which the name entry
* will be inserted into
* \param[in] child pointer to the object referenced by the name entry
* that to be inserted into the parent
- * \param[in] name the name for the child in the parent directory
+ * \param[in] lname the name for the child in the parent directory
*
* \retval positive number for repaired cases
* \retval 0 if nothing to be repaired
struct lfsck_component *com,
struct dt_object *parent,
struct dt_object *child,
- const char *name)
+ const struct lu_name *lname)
{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lu_attr *la = &info->lti_la;
- struct dt_insert_rec *rec = &info->lti_dt_rec;
- struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
/* The child and its name may be on different MDTs. */
- const struct lu_fid *pfid = lfsck_dto2fid(parent);
- const struct lu_fid *cfid = lfsck_dto2fid(child);
- struct dt_device *dev = lfsck->li_next;
- struct thandle *th = NULL;
- struct lfsck_lock_handle *llh = &info->lti_llh;
- int rc = 0;
+ const struct lu_fid *pfid = lfsck_dto2fid(parent);
+ const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ struct lfsck_lock_handle *llh = &info->lti_llh;
+ struct lmv_mds_md_v1 *lmv = &info->lti_lmv;
+ struct lu_buf buf = { lmv, sizeof(*lmv) };
+ /* whether parent's LMV is deleted before insertion */
+ bool parent_lmv_deleted = false;
+ /* whether parent's LMV is missing */
+ bool parent_lmv_lost = false;
+ int rc = 0;
+
ENTRY;
/* @parent/@child may be based on lfsck->li_bottom,
if (IS_ERR(parent))
GOTO(log, rc = PTR_ERR(parent));
- if (unlikely(!dt_try_as_dir(env, parent)))
- GOTO(log, rc = -ENOTDIR);
-
child = lfsck_object_locate(dev, child);
if (IS_ERR(child))
GOTO(log, rc = PTR_ERR(child));
if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
GOTO(log, rc = 1);
- rc = lfsck_lock(env, lfsck, parent, name, llh,
- MDS_INODELOCK_UPDATE, LCK_PW);
- if (rc != 0)
+ rc = lfsck_lock(env, lfsck, parent, lname->ln_name, llh,
+ MDS_INODELOCK_LOOKUP | MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc)
GOTO(log, rc);
+ rc = lfsck_object_is_shard(env, lfsck, child, lname);
+ if (rc < 0)
+ GOTO(unlock, rc);
+
+ if (rc == 1) {
+ rc = lfsck_read_stripe_lmv(env, lfsck, parent, lmv);
+ if (!rc) {
+ /*
+ * To add a shard, we need to convert parent to a
+ * plain directory by deleting its LMV, and after
+ * insertion set it back.
+ */
+ rc = lfsck_lmv_delete(env, lfsck, parent);
+ if (rc)
+ GOTO(unlock, rc);
+ parent_lmv_deleted = true;
+ lmv->lmv_layout_version++;
+ lfsck_lmv_header_cpu_to_le(lmv, lmv);
+ } else if (rc == -ENODATA) {
+ struct lu_seq_range *range = &info->lti_range;
+ struct seq_server_site *ss = lfsck_dev_site(lfsck);
+
+ rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
+ if (rc)
+ GOTO(unlock, rc);
+
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld,
+ fid_seq(lfsck_dto2fid(parent)), range);
+ if (rc)
+ GOTO(unlock, rc);
+
+ parent_lmv_lost = true;
+ lmv->lmv_magic = LMV_MAGIC;
+ lmv->lmv_master_mdt_index = range->lsr_index;
+ lmv->lmv_layout_version++;
+ lfsck_lmv_header_cpu_to_le(lmv, lmv);
+ } else {
+ GOTO(unlock, rc);
+ }
+ }
+
+ if (unlikely(!dt_try_as_dir(env, parent)))
+ GOTO(unlock, rc = -ENOTDIR);
+
th = dt_trans_create(env, dev);
if (IS_ERR(th))
GOTO(unlock, rc = PTR_ERR(th));
rec->rec_type = lfsck_object_type(child) & S_IFMT;
rec->rec_fid = cfid;
rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th);
+ (const struct dt_key *)lname->ln_name, th);
if (rc != 0)
GOTO(stop, rc);
GOTO(stop, rc);
}
- memset(la, 0, sizeof(*la));
+ if (parent_lmv_lost) {
+ rc = dt_declare_xattr_set(env, parent, &buf,
+ XATTR_NAME_LMV".set", 0, th);
+ if (rc)
+ GOTO(stop, rc);
+ }
+
la->la_ctime = ktime_get_real_seconds();
la->la_valid = LA_CTIME;
rc = dt_declare_attr_set(env, parent, la, th);
GOTO(stop, rc);
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)lname->ln_name, th);
if (rc != 0)
GOTO(stop, rc);
GOTO(stop, rc);
}
- la->la_ctime = ktime_get_real_seconds();
+ if (parent_lmv_lost) {
+ rc = dt_xattr_set(env, parent, &buf, XATTR_NAME_LMV".set", 0,
+ th);
+ if (rc)
+ GOTO(stop, rc);
+ }
+
rc = dt_attr_set(env, parent, la, th);
if (rc != 0)
GOTO(stop, rc);
dt_trans_stop(env, dev, th);
unlock:
+ if (parent_lmv_deleted)
+ lfsck_lmv_set(env, lfsck, parent, lmv);
+
lfsck_unlock(llh);
log:
CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
"the name %s and type %o to the parent "DFID": rc = %d\n",
- lfsck_lfsck2name(lfsck), PFID(cfid), name,
+ lfsck_lfsck2name(lfsck), PFID(cfid), lname->ln_name,
lfsck_object_type(child) & S_IFMT, PFID(pfid), rc);
if (rc != 0) {
rec->rec_fid = cfid;
rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
- (const struct dt_key *)dot, th, 1);
+ (const struct dt_key *)dot, th);
if (rc != 0)
GOTO(unlock2, rc);
rec->rec_fid = lfsck_dto2fid(parent);
rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, 1);
+ (const struct dt_key *)dotdot, th);
if (rc != 0)
GOTO(unlock2, rc);
rec->rec_fid = cfid;
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)name, th);
if (rc == 0) {
dt_write_lock(env, parent, 0);
rc = dt_ref_add(env, parent, th);
lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
true);
- if (buflen < ldata_new.ld_leh->leh_len) {
+ /*
+ * linkea may change because it doesn't take lock in the first read, if
+ * it becomes larger, restart from beginning.
+ */
+ if ((ldata_new.ld_leh->leh_reccount > 0 ||
+ unlikely(ldata_new.ld_leh->leh_overflow_time)) &&
+ buflen < ldata_new.ld_leh->leh_len) {
dt_write_unlock(env, obj);
dt_trans_stop(env, dev, th);
lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
ldata_new.ld_leh->leh_len);
+ buflen = linkea_buf.lb_len;
goto again;
}
- if (ldata_new.ld_leh->leh_reccount > 0 ||
- unlikely(ldata->ld_leh->leh_overflow_time))
+ if (buflen)
rc = lfsck_links_write(env, obj, &ldata_new, th);
else
rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th);
GOTO(stop, rc);
rc = dt_insert(env, parent, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)name, th);
GOTO(stop, rc = (rc == 0 ? 1 : rc));
struct lfsck_thread_info *info = lfsck_env_info(env);
struct dt_insert_rec *rec = &info->lti_dt_rec;
const struct lu_fid *pfid = lfsck_dto2fid(parent);
- const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct lu_fid cfid = {0};
struct lu_fid tfid;
struct lfsck_instance *lfsck = com->lc_lfsck;
struct dt_device *dev = lfsck->li_next;
int rc = 0;
ENTRY;
+ if (child)
+ cfid = *lfsck_dto2fid(child);
parent = lfsck_object_locate(dev, parent);
if (IS_ERR(parent))
GOTO(log, rc = PTR_ERR(parent));
if (update) {
rec->rec_type = lfsck_object_type(child) & S_IFMT;
- rec->rec_fid = cfid;
+ LASSERT(!fid_is_zero(&cfid));
+ rec->rec_fid = &cfid;
rc = dt_declare_insert(env, parent,
(const struct dt_rec *)rec,
(const struct dt_key *)name2, th);
/* Someone has removed the bad name entry and reused it for other
* object by race. */
- if (!lu_fid_eq(&tfid, cfid))
+ if (!lu_fid_eq(&tfid, &cfid))
GOTO(unlock2, rc = 0);
if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
if (update) {
rc = dt_insert(env, parent,
(const struct dt_rec *)rec,
- (const struct dt_key *)name2, th, 1);
+ (const struct dt_key *)name2, th);
if (rc != 0)
GOTO(unlock2, rc);
}
/* We are not sure whether the child will become orphan or not.
* Record it in the LFSCK trace file for further checking in
* the second-stage scanning. */
- if (!update && !dec && rc == 0)
- lfsck_namespace_trace_update(env, com, cfid,
+ if (!update && !dec && child && rc == 0)
+ lfsck_namespace_trace_update(env, com, &cfid,
LNTF_CHECK_LINKEA, true);
unlock1:
"entry for: parent "DFID", child "DFID", name %s, type "
"in name entry %o, type claimed by child %o. repair it "
"by %s with new name2 %s: rc = %d\n",
- lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
+ lfsck_lfsck2name(lfsck), PFID(pfid), PFID(&cfid),
name, type, update ? lfsck_object_type(child) : 0,
update ? "updating" : "removing", name2, rc);
dt_delete(env, obj, (const struct dt_key *)dotdot, th);
rc = dt_insert(env, obj, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, 1);
+ (const struct dt_key *)dotdot, th);
if (rc != 0)
GOTO(unlock, rc);
lost_parent:
lmv = &info->lti_lmv;
- rc = lfsck_read_stripe_lmv(env, child, lmv);
+ rc = lfsck_read_stripe_lmv(env, lfsck, child, lmv);
if (rc != 0 && rc != -ENODATA)
GOTO(out, rc);
if (rc >= 0) {
/* Add the missing name entry to the parent. */
rc = lfsck_namespace_insert_normal(env, com, parent,
- child, cname->ln_name);
+ child, cname);
if (unlikely(rc == -EEXIST)) {
/* Unfortunately, someone reused the name
* under the parent by race. So we have
}
lfsck_ibits_unlock(lh, LCK_EX);
- rc = lfsck_namespace_check_name(env, parent, child, cname);
+ rc = lfsck_namespace_check_name(env, lfsck, parent, child,
+ cname);
if (rc == -ENOENT)
goto lost_parent;
/* Add the missing name entry back to the namespace. */
rc = lfsck_namespace_insert_normal(env, com, parent, child,
- cname->ln_name);
+ cname);
if (unlikely(rc == -ESTALE))
/* It may happen when the remote object has been
* removed, but the local MDT is not aware of that. */
}
/**
+ * Verify the object's agent entry.
+ *
+ * If the object claims to have agent entry but the linkEA does not contain
+ * remote parent, then remove the agent entry. Otherwise, if the object has
+ * no agent entry but its linkEA contains remote parent, then will generate
+ * agent entry for it.
+ *
+ * \param[in] env pointer to the thread context
+ * \param[in] com pointer to the lfsck component
+ * \param[in] obj pointer to the dt_object to be handled
+ *
+ * \retval positive number for repaired cases
+ * \retval 0 if nothing to be repaired
+ * \retval negative error number on failure
+ */
+static int lfsck_namespace_check_agent_entry(const struct lu_env *env,
+ struct lfsck_component *com,
+ struct dt_object *obj)
+{
+ struct linkea_data ldata = { NULL };
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lfsck_namespace *ns = com->lc_file_ram;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lu_fid *pfid = &info->lti_fid2;
+ struct lu_name *cname = &info->lti_name;
+ struct lu_seq_range *range = &info->lti_range;
+ struct seq_server_site *ss = lfsck_dev_site(lfsck);
+ __u32 idx = lfsck_dev_idx(lfsck);
+ int rc;
+ bool remote = false;
+ ENTRY;
+
+ if (!(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT))
+ RETURN(0);
+
+ rc = lfsck_links_read_with_rec(env, obj, &ldata);
+ if (rc == -ENOENT || rc == -ENODATA)
+ RETURN(0);
+
+ if (rc && rc != -EINVAL)
+ GOTO(out, rc);
+
+ /* We check the agent entry again after verifying the linkEA
+ * successfully. So invalid linkEA should be dryrun mode. */
+ if (rc == -EINVAL || unlikely(!ldata.ld_leh->leh_reccount))
+ RETURN(0);
+
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL && !remote) {
+ linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
+ cname, pfid);
+ if (!linkea_entry_is_valid(&ldata, cname, pfid))
+ GOTO(out, rc = 0);
+
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld,
+ fid_seq(pfid), range);
+ if (rc)
+ GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
+
+ if (range->lsr_index != idx)
+ remote = true;
+ else
+ linkea_next_entry(&ldata);
+ }
+
+ if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
+ (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
+ struct dt_device *dev = lfsck_obj2dev(obj);
+ struct linkea_data ldata2 = { NULL };
+ struct lustre_handle lh = { 0 };
+ struct lu_buf linkea_buf;
+ struct thandle *handle;
+
+ if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
+ GOTO(out, rc = 1);
+
+ rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
+ MDS_INODELOCK_UPDATE |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc)
+ GOTO(out, rc);
+
+ handle = dt_trans_create(env, dev);
+ if (IS_ERR(handle))
+ GOTO(unlock, rc = PTR_ERR(handle));
+
+ lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len);
+ rc = dt_declare_xattr_set(env, obj, &linkea_buf,
+ XATTR_NAME_LINK, LU_XATTR_REPLACE, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_trans_start_local(env, dev, handle);
+ if (rc)
+ GOTO(stop, rc);
+
+ dt_write_lock(env, obj, 0);
+ rc = lfsck_links_read2_with_rec(env, obj, &ldata2);
+ if (rc) {
+ if (rc == -ENOENT || rc == -ENODATA)
+ rc = 0;
+ GOTO(unlock2, rc);
+ }
+
+ /* If someone changed linkEA by race, then the agent
+ * entry will be updated by lower layer automatically. */
+ if (ldata.ld_leh->leh_len != ldata2.ld_leh->leh_len ||
+ memcmp(ldata.ld_buf->lb_buf, ldata2.ld_buf->lb_buf,
+ ldata.ld_leh->leh_len) != 0)
+ GOTO(unlock2, rc = 0);
+
+ rc = dt_xattr_set(env, obj, &linkea_buf, XATTR_NAME_LINK,
+ LU_XATTR_REPLACE, handle);
+ if (!rc)
+ rc = 1;
+
+ GOTO(unlock2, rc);
+
+unlock2:
+ dt_write_unlock(env, obj);
+stop:
+ dt_trans_stop(env, dev, handle);
+unlock:
+ lfsck_ibits_unlock(&lh, LCK_EX);
+ }
+
+ GOTO(out, rc);
+
+out:
+ if (rc > 0)
+ ns->ln_agent_entries_repaired++;
+ if (rc)
+ CDEBUG(D_LFSCK, "%s: repair agent entry for "DFID": rc = %d\n",
+ lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
+ return rc;
+}
+
+/**
* Double scan the MDT-object for namespace LFSCK.
*
* If the MDT-object contains invalid or repeated linkEA entries, then drop
if (S_ISDIR(lfsck_object_type(child))) {
dt_read_unlock(env, child);
rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
+ if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
+ rc = lfsck_namespace_check_agent_entry(env, com, child);
RETURN(rc);
}
/* Add the missing name entry to the parent. */
rc = lfsck_namespace_insert_normal(env, com,
- parent, child, cname->ln_name);
+ parent, child, cname);
if (unlikely(rc == -EEXIST))
/* Unfortunately, someone reused the
* name under the parent by race. So we
GOTO(out, rc = 0);
}
- rc = lfsck_namespace_check_name(env, parent, child, cname);
+ rc = lfsck_namespace_check_name(env, lfsck, parent, child,
+ cname);
if (rc == -ENOENT)
goto lost_parent;
/* Add the missing name entry back to the namespace. */
rc = lfsck_namespace_insert_normal(env, com, parent, child,
- cname->ln_name);
+ cname);
if (unlikely(rc == -ESTALE))
/* It may happen when the remote object has been
* removed, but the local MDT is not aware of that. */
rc = 1;
}
+ if (!rc && flags & LNTF_CHECK_AGENT_ENTRY)
+ rc = lfsck_namespace_check_agent_entry(env, com, child);
+
return rc;
}
"striped_shards_skipped: %llu\n"
"name_hash_%s: %llu\n"
"linkea_overflow_%s: %llu\n"
+ "agent_entries_%s: %llu\n"
"success_count: %u\n"
"run_time_phase1: %lld seconds\n"
"run_time_phase2: %lld seconds\n",
postfix, ns->ln_name_hash_repaired,
dryrun ? "inconsistent" : "cleared",
ns->ln_linkea_overflow_cleared,
+ postfix, ns->ln_agent_entries_repaired,
ns->ln_success_count,
time_phase1,
time_phase2);
if (rc != 0)
GOTO(out, rc);
- lad->lad_incomplete = 0;
+ clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
CFS_RESET_BITMAP(lad->lad_bitmap);
rc = lfsck_namespace_store(env, com);
struct lfsck_instance *lfsck = com->lc_lfsck;
struct lfsck_lmv *llmv = lfsck->li_lmv;
struct lfsck_namespace_req *lnr;
- __u32 size =
- sizeof(*lnr) + LFSCK_TMPBUF_LEN;
- bool wakeup = false;
+ struct lu_attr *la = &lfsck_env_info(env)->lti_la2;
+ __u32 size = sizeof(*lnr) + LFSCK_TMPBUF_LEN;
+ int rc;
+ bool wakeup = false;
ENTRY;
if (llmv == NULL)
RETURN_EXIT;
+ rc = dt_attr_get(env, lfsck->li_obj_dir, la);
+ if (rc)
+ RETURN_EXIT;
+
OBD_ALLOC(lnr, size);
if (lnr == NULL) {
ns->ln_striped_dirs_skipped++;
}
lso = lfsck_assistant_object_init(env, lfsck_dto2fid(lfsck->li_obj_dir),
- NULL, lfsck->li_pos_current.lp_oit_cookie, true);
+ la, lfsck->li_pos_current.lp_oit_cookie, true);
if (IS_ERR(lso)) {
OBD_FREE(lnr, size);
ns->ln_striped_dirs_skipped++;
lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
lnr->lnr_size = size;
+ lnr->lnr_type = lso->lso_attr.la_mode;
spin_lock(&lad->lad_lock);
- if (lad->lad_assistant_status < 0) {
+ if (lad->lad_assistant_status < 0 ||
+ unlikely(!thread_is_running(&lfsck->li_thread) ||
+ !thread_is_running(&lad->lad_thread))) {
spin_unlock(&lad->lad_lock);
lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
ns->ln_striped_dirs_skipped++;
struct linkea_data ldata = { NULL };
__u32 idx = lfsck_dev_idx(lfsck);
int rc;
+ bool remote = false;
ENTRY;
rc = lfsck_links_read(env, obj, &ldata);
GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
}
+ if (rc && rc != -ENODATA)
+ GOTO(out, rc);
+
if (rc == -ENODATA || unlikely(!ldata.ld_leh->leh_reccount)) {
rc = lfsck_namespace_check_for_double_scan(env, com, obj);
GOTO(out, rc);
}
- if (rc != 0)
- GOTO(out, rc);
+ linkea_first_entry(&ldata);
+ while (ldata.ld_lee != NULL) {
+ linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
+ cname, pfid);
+ if (!fid_is_sane(pfid)) {
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_PARENT, true);
+ } else if (!linkea_entry_is_valid(&ldata, cname, pfid)) {
+ GOTO(out, rc);
+ } else {
+ fld_range_set_mdt(range);
+ rc = fld_server_lookup(env, ss->ss_server_fld,
+ fid_seq(pfid), range);
+ if ((rc == -ENOENT) ||
+ (!rc && range->lsr_index != idx)) {
+ remote = true;
+ break;
+ }
+ }
+ if (rc)
+ GOTO(out, rc);
+
+ linkea_next_entry(&ldata);
+ }
+
+ if ((lu_object_has_agent_entry(&obj->do_lu) && !remote) ||
+ (!lu_object_has_agent_entry(&obj->do_lu) && remote)) {
+ rc = lfsck_namespace_trace_update(env, com, fid,
+ LNTF_CHECK_AGENT_ENTRY, true);
+ if (rc)
+ GOTO(out, rc);
+ }
/* Record multiple-linked object. */
if (ldata.ld_leh->leh_reccount > 1) {
GOTO(out, rc);
}
- linkea_first_entry(&ldata);
- linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
- if (!fid_is_sane(pfid)) {
+ if (remote)
rc = lfsck_namespace_trace_update(env, com, fid,
- LNTF_CHECK_PARENT, true);
- } else {
- fld_range_set_mdt(range);
- rc = fld_local_lookup(env, ss->ss_server_fld,
- fid_seq(pfid), range);
- if ((rc == -ENOENT) ||
- (rc == 0 && range->lsr_index != idx))
- rc = lfsck_namespace_trace_update(env, com, fid,
- LNTF_CHECK_LINKEA, true);
- else
- rc = lfsck_namespace_check_for_double_scan(env, com,
- obj);
- }
+ LNTF_CHECK_LINKEA, true);
+ else
+ rc = lfsck_namespace_check_for_double_scan(env, com, obj);
GOTO(out, rc);
l_wait_event(mthread->t_ctl_waitq,
lad->lad_prefetched < bk->lb_async_windows ||
!thread_is_running(mthread) ||
- thread_is_stopped(athread),
+ !thread_is_running(athread),
&lwi);
- if (unlikely(!thread_is_running(mthread)) ||
- thread_is_stopped(athread))
+ if (unlikely(!thread_is_running(mthread) ||
+ !thread_is_running(athread)))
return 0;
if (unlikely(lfsck_is_dead_obj(lfsck->li_obj_dir)))
}
spin_lock(&lad->lad_lock);
- if (lad->lad_assistant_status < 0) {
+ if (lad->lad_assistant_status < 0 ||
+ unlikely(!thread_is_running(mthread) ||
+ !thread_is_running(athread))) {
spin_unlock(&lad->lad_lock);
lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
return lad->lad_assistant_status;
time64_t rtime = ns->ln_run_time_phase1 + duration;
if (duration != 0)
- div_u64(new_checked, duration);
+ new_checked = div64_s64(new_checked, duration);
if (rtime != 0)
- div_u64(speed, rtime);
+ speed = div64_s64(speed, rtime);
lfsck_namespace_dump_statistics(m, ns, checked, 0, rtime, 0,
bk->lb_param & LPF_DRYRUN);
time64_t time0 = ns->ln_run_time_phase1 + rtime;
if (duration != 0)
- div_u64(new_checked, duration);
+ new_checked = div64_s64(new_checked, duration);
if (ns->ln_run_time_phase1 != 0)
- div_u64(speed1, ns->ln_run_time_phase1);
+ speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
else if (ns->ln_items_checked != 0)
time0++;
if (rtime != 0)
- div_u64(speed2, rtime);
+ speed2 = div64_s64(speed2, rtime);
else if (checked != 0)
time0++;
if (time0 != 0)
- div_u64(speed0, time0);
+ speed0 = div64_s64(speed0, time0);
lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
checked,
time64_t time0 = ns->ln_run_time_phase1 + ns->ln_run_time_phase2;
if (ns->ln_run_time_phase1 != 0)
- div_u64(speed1, ns->ln_run_time_phase1);
+ speed1 = div64_s64(speed1, ns->ln_run_time_phase1);
else if (ns->ln_items_checked != 0)
time0++;
if (ns->ln_run_time_phase2 != 0)
- div_u64(speed2, ns->ln_run_time_phase2);
+ speed2 = div64_s64(speed2, ns->ln_run_time_phase2);
else if (ns->ln_objs_checked_phase2 != 0)
time0++;
if (time0 != 0)
- div_u64(speed0, time0);
+ speed0 = div64_s64(speed0, time0);
lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
ns->ln_objs_checked_phase2,
struct dt_object *child,
struct lfsck_namespace_req *lnr)
{
- struct lfsck_thread_info *info = lfsck_env_info(env);
- struct lu_attr *la = &info->lti_la;
- struct dt_allocation_hint *hint = &info->lti_hint;
- struct dt_object_format *dof = &info->lti_dof;
- struct dt_insert_rec *rec = &info->lti_dt_rec;
- struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
- const struct lu_name *cname;
- const struct lu_fid *pfid = lfsck_dto2fid(parent);
- const struct lu_fid *cfid = lfsck_dto2fid(child);
- struct linkea_data ldata = { NULL };
- struct lfsck_lock_handle *llh = &info->lti_llh;
- struct lu_buf linkea_buf;
- struct lu_buf lmv_buf;
- struct lfsck_instance *lfsck = com->lc_lfsck;
- struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
- struct dt_device *dev = lfsck->li_next;
- struct thandle *th = NULL;
- int rc = 0;
- __u16 type = lnr->lnr_type;
- bool create;
+ struct lfsck_thread_info *info = lfsck_env_info(env);
+ struct lu_attr *la = &info->lti_la;
+ struct dt_allocation_hint *hint = &info->lti_hint;
+ struct dt_object_format *dof = &info->lti_dof;
+ struct dt_insert_rec *rec = &info->lti_dt_rec;
+ struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
+ const struct lu_name *cname;
+ const struct lu_fid *pfid = lfsck_dto2fid(parent);
+ const struct lu_fid *cfid = lfsck_dto2fid(child);
+ struct linkea_data ldata = { NULL };
+ struct lfsck_lock_handle *llh = &info->lti_llh;
+ struct lustre_handle rlh = { 0 };
+ struct lustre_handle clh = { 0 };
+ struct lu_buf linkea_buf;
+ struct lu_buf lmv_buf;
+ struct lfsck_instance *lfsck = com->lc_lfsck;
+ struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
+ struct dt_device *dev = lfsck->li_next;
+ struct thandle *th = NULL;
+ int rc = 0;
+ __u16 type = lnr->lnr_type;
+ bool create;
ENTRY;
cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
GOTO(log, rc);
rc = lfsck_lock(env, lfsck, parent, lnr->lnr_name, llh,
- MDS_INODELOCK_UPDATE, LCK_PR);
+ MDS_INODELOCK_UPDATE, LCK_PW);
if (rc != 0)
GOTO(log, rc);
if (rc != 0)
GOTO(log, rc);
+ if (dt_object_remote(child)) {
+ rc = lfsck_remote_lookup_lock(env, lfsck, parent, child, &rlh,
+ LCK_EX);
+ if (rc != 0)
+ GOTO(log, rc);
+ }
+
+ rc = lfsck_ibits_lock(env, lfsck, child, &clh,
+ MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP |
+ MDS_INODELOCK_XATTR, LCK_EX);
+ if (rc != 0)
+ GOTO(unlock_remote_lookup, rc);
+
/* Set the ctime as zero, then others can know it is created for
* repairing dangling name entry by LFSCK. And if the LFSCK made
* wrong decision and the real MDT-object has been found later,
* then the LFSCK has chance to fix the incosistency properly. */
memset(la, 0, sizeof(*la));
- la->la_mode = (type & S_IFMT) | 0600;
- la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
- LA_ATIME | LA_MTIME | LA_CTIME;
-
- child->do_ops->do_ah_init(env, hint, parent, child,
- la->la_mode & S_IFMT);
+ if (S_ISDIR(type))
+ la->la_mode = (type & S_IFMT) | 0700;
+ else
+ la->la_mode = (type & S_IFMT) | 0600;
+ la->la_valid = LA_TYPE | LA_MODE | LA_CTIME;
+
+ /*
+ * if it's directory, skip do_ah_init() to create a plain directory
+ * because it may have shards already, which will be inserted back
+ * later, besides, it may be remote, and creating stripe directory
+ * remotely is not supported.
+ */
+ if (S_ISREG(type))
+ child->do_ops->do_ah_init(env, hint, parent, child, type);
memset(dof, 0, sizeof(*dof));
dof->dof_type = dt_mode_to_dft(type);
th = dt_trans_create(env, dev);
if (IS_ERR(th))
- GOTO(log, rc = PTR_ERR(th));
+ GOTO(unlock_child, rc = PTR_ERR(th));
/* 1a. create child. */
rc = dt_declare_create(env, child, la, hint, dof, th);
lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
rc = dt_declare_xattr_set(env, child, &lmv_buf,
- XATTR_NAME_LMV, 0, th);
+ XATTR_NAME_LMV".set", 0, th);
if (rc != 0)
GOTO(stop, rc);
}
if (rc != 0)
GOTO(stop, rc);
+ /* 7a. if child is remote, delete and insert to generate local agent */
+ if (dt_object_remote(child)) {
+ rc = dt_declare_delete(env, parent,
+ (const struct dt_key *)lnr->lnr_name,
+ th);
+ if (rc)
+ GOTO(stop, rc);
+
+ rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)lnr->lnr_name,
+ th);
+ if (rc)
+ GOTO(stop, rc);
+ }
+
rc = dt_trans_start_local(env, dev, th);
if (rc != 0)
GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
rec->rec_type = S_IFDIR;
rec->rec_fid = cfid;
rc = dt_insert(env, child, (const struct dt_rec *)rec,
- (const struct dt_key *)dot, th, 1);
+ (const struct dt_key *)dot, th);
if (rc != 0)
GOTO(unlock, rc);
/* 4b. insert dotdot into child dir */
rec->rec_fid = pfid;
rc = dt_insert(env, child, (const struct dt_rec *)rec,
- (const struct dt_key *)dotdot, th, 1);
+ (const struct dt_key *)dotdot, th);
if (rc != 0)
GOTO(unlock, rc);
/* 5b. generate slave LMV EA. */
if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
- rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
- 0, th);
+ rc = dt_xattr_set(env, child, &lmv_buf,
+ XATTR_NAME_LMV".set", 0, th);
if (rc != 0)
GOTO(unlock, rc);
}
/* 6b. insert linkEA for child. */
rc = dt_xattr_set(env, child, &linkea_buf,
XATTR_NAME_LINK, 0, th);
+ if (rc)
+ GOTO(unlock, rc);
+
+ /* 7b. if child is remote, delete and insert to generate local agent */
+ if (dt_object_remote(child)) {
+ rc = dt_delete(env, parent,
+ (const struct dt_key *)lnr->lnr_name, th);
+ if (rc)
+ GOTO(unlock, rc);
+
+ rec->rec_type = type;
+ rec->rec_fid = cfid;
+ rc = dt_insert(env, parent, (const struct dt_rec *)rec,
+ (const struct dt_key *)lnr->lnr_name, th);
+ if (rc)
+ GOTO(unlock, rc);
+ }
GOTO(unlock, rc);
stop:
dt_trans_stop(env, dev, th);
+unlock_child:
+ lfsck_ibits_unlock(&clh, LCK_EX);
+unlock_remote_lookup:
+ if (dt_object_remote(child))
+ lfsck_ibits_unlock(&rlh, LCK_EX);
log:
lfsck_unlock(llh);
CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
/* b5. insert child's FID into the LFSCK trace file. */
rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
- (const struct dt_key *)key, th, 1);
+ (const struct dt_key *)key, th);
GOTO(stop, rc = (rc == 0 ? 1 : rc));
int rc = 0;
ENTRY;
- if (!lad->lad_incomplete)
+ if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
RETURN_EXIT;
set = ptlrpc_prep_set();
}
up_read(<ds->ltd_rw_sem);
- rc = ptlrpc_set_wait(set);
+ rc = ptlrpc_set_wait(env, set);
ptlrpc_set_destroy(set);
GOTO(out, rc);
linkea_first_entry(&ldata);
linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
+ if (!linkea_entry_is_valid(&ldata, cname, pfid))
+ return -EINVAL;
+
/* To guarantee the 'name' is terminated with '0'. */
memcpy(name, cname->ln_name, cname->ln_namelen);
name[cname->ln_namelen] = 0;
GOTO(stop, rc);
rc = dt_insert(env, dir, (const struct dt_rec *)rec,
- (const struct dt_key *)name, th, 1);
+ (const struct dt_key *)name, th);
if (rc == 0 && S_ISDIR(type) && !exists) {
dt_write_lock(env, dir, 0);
rc = dt_ref_add(env, dir, th);
com->lc_obj = obj;
rc = lfsck_namespace_load(env, com);
- if (rc == -ENODATA)
+ if (rc == -ENODATA) {
rc = lfsck_namespace_init(env, com);
- else if (rc < 0)
+ } else if (rc < 0) {
rc = lfsck_namespace_reset(env, com, true);
- else
+ } else {
rc = lfsck_load_sub_trace_files(env, com,
&dt_lfsck_namespace_features, LFSCK_NAMESPACE, false);
+ if (rc)
+ rc = lfsck_namespace_reset(env, com, true);
+ }
if (rc != 0)
GOTO(out, rc);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff