X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Flfsck%2Flfsck_namespace.c;h=9606a118b21d927af02e59eb0a249478f27f4b06;hp=29a901d3de5a2c2f9d05fd190d7ecc0b44174c23;hb=3cce65712d94cffe8f1626545845b95b88aef672;hpb=012834c5e7c7be50ff117cee4ac473d7fee4294d diff --git a/lustre/lfsck/lfsck_namespace.c b/lustre/lfsck/lfsck_namespace.c index 29a901d..9606a11 100644 --- a/lustre/lfsck/lfsck_namespace.c +++ b/lustre/lfsck/lfsck_namespace.c @@ -312,7 +312,7 @@ static int lfsck_namespace_load_bitmap(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); @@ -326,9 +326,9 @@ static int lfsck_namespace_load_bitmap(const struct lu_env *env, 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); } @@ -562,7 +562,7 @@ int lfsck_namespace_trace_update(const struct lu_env *env, 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); } @@ -760,19 +760,41 @@ again: 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'. */ @@ -790,9 +812,7 @@ static void lfsck_linkea_del_buf(struct linkea_data *ldata, /* 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); @@ -830,7 +850,10 @@ static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata, 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) @@ -1035,7 +1058,7 @@ again: 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); } @@ -1054,7 +1077,7 @@ again: 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); @@ -1090,6 +1113,99 @@ log: 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. * @@ -1100,13 +1216,17 @@ log: * 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 @@ -1116,19 +1236,26 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, 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, @@ -1138,9 +1265,6 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, 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)); @@ -1148,11 +1272,57 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, 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)); @@ -1160,7 +1330,7 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, 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); @@ -1170,7 +1340,13 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, 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); @@ -1186,7 +1362,7 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, 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); @@ -1198,7 +1374,13 @@ static int lfsck_namespace_insert_normal(const struct lu_env *env, 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); @@ -1211,12 +1393,15 @@ stop: 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) { @@ -1436,13 +1621,13 @@ again: 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); @@ -1460,7 +1645,7 @@ again: 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); @@ -1591,16 +1776,22 @@ again: 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); @@ -1900,7 +2091,7 @@ replace: 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)); @@ -2110,7 +2301,7 @@ int lfsck_namespace_repair_dirent(const struct lu_env *env, 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); } @@ -2238,7 +2429,7 @@ static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env, 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); @@ -2418,7 +2609,7 @@ lfsck_namespace_dsd_single(const struct lu_env *env, 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); @@ -2447,7 +2638,7 @@ lost_parent: 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 @@ -2505,7 +2696,8 @@ lost_parent: } 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; @@ -2529,7 +2721,7 @@ 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. */ @@ -3386,8 +3578,7 @@ static int lfsck_namespace_check_agent_entry(const struct lu_env *env, while (ldata.ld_lee != NULL && !remote) { linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid); - /* If parent FID is unknown, not verify agent entry. */ - if (!fid_is_sane(pfid)) + if (!linkea_entry_is_valid(&ldata, cname, pfid)) GOTO(out, rc = 0); fld_range_set_mdt(range); @@ -3639,7 +3830,7 @@ lost_parent: /* 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 @@ -3759,7 +3950,8 @@ lost_parent: 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; @@ -3785,7 +3977,7 @@ 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. */ @@ -4072,7 +4264,7 @@ static int lfsck_namespace_reset(const struct lu_env *env, 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); @@ -4112,14 +4304,19 @@ static void lfsck_namespace_close_dir(const struct lu_env *env, 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++; @@ -4128,7 +4325,7 @@ static void lfsck_namespace_close_dir(const struct lu_env *env, } 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++; @@ -4144,6 +4341,7 @@ static void lfsck_namespace_close_dir(const struct lu_env *env, 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 || @@ -4394,6 +4592,8 @@ static int lfsck_namespace_exec_oit(const struct lu_env *env, 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, @@ -5071,26 +5271,28 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, 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); @@ -5122,7 +5324,7 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, 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); @@ -5130,17 +5332,38 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, 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); @@ -5150,7 +5373,7 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, 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); @@ -5200,7 +5423,7 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, 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); } @@ -5214,6 +5437,21 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, 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)); @@ -5234,21 +5472,21 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, 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); } @@ -5257,6 +5495,23 @@ int lfsck_namespace_repair_dangling(const struct lu_env *env, /* 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); @@ -5266,6 +5521,11 @@ unlock: 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 " @@ -5926,7 +6186,7 @@ static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env, /* 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)); @@ -6519,7 +6779,7 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env, int rc = 0; ENTRY; - if (!lad->lad_incomplete) + if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags)) RETURN_EXIT; set = ptlrpc_prep_set(); @@ -6537,7 +6797,7 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env, continue; laia->laia_ltd = ltd; - rc = lfsck_async_request(env, ltd, lr, set, + rc = lfsck_async_request(env, ltd->ltd_exp, lr, set, lfsck_namespace_assistant_sync_failures_interpret, laia, LFSCK_NOTIFY); if (rc != 0) @@ -6547,7 +6807,7 @@ static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env, } up_read(<ds->ltd_rw_sem); - rc = ptlrpc_set_wait(set); + rc = ptlrpc_set_wait(env, set); ptlrpc_set_destroy(set); GOTO(out, rc); @@ -6673,6 +6933,9 @@ int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj, 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; @@ -6746,7 +7009,7 @@ int lfsck_update_name_entry(const struct lu_env *env, 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); @@ -6825,13 +7088,16 @@ int lfsck_namespace_setup(const struct lu_env *env, 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);