X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_dir.c;h=9d03d171269e21d259ee97b4276a20063d3666f7;hp=3b22a66edcbd266ea73dd58b232dd19c57d1c852;hb=f625f670afbe954030ff81f0f8522137d6cdd335;hpb=de2d5808bd2987f76d2486272e1a9c192ba277d4 diff --git a/lustre/mdd/mdd_dir.c b/lustre/mdd/mdd_dir.c index 3b22a66..9d03d17 100644 --- a/lustre/mdd/mdd_dir.c +++ b/lustre/mdd/mdd_dir.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2013, Intel Corporation. + * Copyright (c) 2011, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -98,16 +98,9 @@ __mdd_lookup(const struct lu_env *env, struct md_object *pobj, RETURN(rc); if (likely(S_ISDIR(mdd_object_type(mdd_obj)) && - dt_try_as_dir(env, dir))) { - - rc = dir->do_index_ops->dio_lookup(env, dir, - (struct dt_rec *)fid, key, - mdd_object_capa(env, mdd_obj)); - if (rc > 0) - rc = 0; - else if (rc == 0) - rc = -ENOENT; - } else + dt_try_as_dir(env, dir))) + rc = dt_lookup(env, dir, (struct dt_rec *)fid, key); + else rc = -ENOTDIR; RETURN(rc); @@ -121,11 +114,12 @@ int mdd_lookup(const struct lu_env *env, int rc; ENTRY; - rc = mdd_la_get(env, md2mdd_obj(pobj), pattr, BYPASS_CAPA); + rc = mdd_la_get(env, md2mdd_obj(pobj), pattr); if (rc != 0) RETURN(rc); - rc = __mdd_lookup(env, pobj, pattr, lname, fid, MAY_EXEC); + rc = __mdd_lookup(env, pobj, pattr, lname, fid, + (spec != NULL && spec->sp_permitted) ? 0 : MAY_EXEC); RETURN(rc); } @@ -176,8 +170,7 @@ static int mdd_is_parent(const struct lu_env *env, RETURN(0); for(;;) { - /* this is done recursively, bypass capa for each obj */ - mdd_set_capainfo(env, 4, p1, BYPASS_CAPA); + /* this is done recursively */ rc = mdd_parent_fid(env, p1, attr, pfid); if (rc) GOTO(out, rc); @@ -212,7 +205,7 @@ out: * No permission check is needed. * * returns 1: if fid is ancestor of @mo; - * returns 0: if fid is not a ancestor of @mo; + * returns 0: if fid is not an ancestor of @mo; * * returns EREMOTE if remote object is found, fid of remote object is saved to * @fid; @@ -230,7 +223,7 @@ int mdd_is_subdir(const struct lu_env *env, struct md_object *mo, if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo)))) RETURN(0); - rc = mdd_la_get(env, md2mdd_obj(mo), attr, BYPASS_CAPA); + rc = mdd_la_get(env, md2mdd_obj(mo), attr); if (rc != 0) RETURN(rc); @@ -271,7 +264,7 @@ static int mdd_dir_is_empty(const struct lu_env *env, RETURN(-ENOTDIR); iops = &obj->do_index_ops->dio_it; - it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA); + it = iops->init(env, obj, LUDA_64BITHASH); if (!IS_ERR(it)) { result = iops->get(env, it, (const struct dt_key *)""); if (result > 0) { @@ -356,8 +349,7 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, /* If the parent is a sub-stripe, check whether it is dead */ xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key)); - rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV, - mdd_object_capa(env, pobj)); + rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV); if (unlikely(rc > 0)) { struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf; @@ -386,17 +378,16 @@ int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj, if (mdd_is_dead_obj(pobj)) RETURN(-ENOENT); - if ((attr->la_valid & LA_FLAGS) && - (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) + if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)) RETURN(-EPERM); rc = mdd_permission_internal_locked(env, pobj, pattr, MAY_WRITE | MAY_EXEC, MOR_TGT_PARENT); - if (rc) + if (rc != 0) RETURN(rc); - if (mdd_is_append(pobj)) + if (pattr->la_flags & LUSTRE_APPEND_FL) RETURN(-EPERM); RETURN(rc); @@ -451,7 +442,7 @@ static int mdd_may_delete_entry(const struct lu_env *env, RETURN(rc); } - if (mdd_is_append(pobj)) + if (pattr->la_flags & LUSTRE_APPEND_FL) RETURN(-EPERM); RETURN(0); @@ -488,11 +479,7 @@ int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj, if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr)) RETURN(-EPERM); - if (mdd_is_immutable(tobj) || mdd_is_append(tobj)) - RETURN(-EPERM); - - if ((tattr->la_valid & LA_FLAGS) && - (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) + if (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)) RETURN(-EPERM); /* additional check the rename case */ @@ -552,11 +539,11 @@ static int mdd_link_sanity_check(const struct lu_env *env, if (rc < 0) RETURN(rc); - if (mdd_is_immutable(src_obj) || mdd_is_append(src_obj)) - RETURN(-EPERM); + if (cattr->la_flags & (LUSTRE_IMMUTABLE_FL | LUSTRE_APPEND_FL)) + RETURN(-EPERM); - if (S_ISDIR(mdd_object_type(src_obj))) - RETURN(-EPERM); + if (S_ISDIR(mdd_object_type(src_obj))) + RETURN(-EPERM); LASSERT(src_obj != tgt_obj); rc = mdd_may_create(env, tgt_obj, tattr, NULL, true); @@ -569,42 +556,41 @@ static int mdd_link_sanity_check(const struct lu_env *env, } static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj, - const char *name, struct thandle *handle, - struct lustre_capa *capa) + const char *name, struct thandle *handle) { - struct dt_object *next = mdd_object_child(pobj); - int rc; - ENTRY; + struct dt_object *next = mdd_object_child(pobj); + int rc; + ENTRY; - if (dt_try_as_dir(env, next)) { - rc = next->do_index_ops->dio_delete(env, next, - (struct dt_key *)name, - handle, capa); - } else - rc = -ENOTDIR; + if (dt_try_as_dir(env, next)) + rc = dt_delete(env, next, (struct dt_key *)name, handle); + else + rc = -ENOTDIR; - RETURN(rc); + RETURN(rc); } static int __mdd_index_insert_only(const struct lu_env *env, struct mdd_object *pobj, - const struct lu_fid *lf, const char *name, - struct thandle *handle, - struct lustre_capa *capa) + const struct lu_fid *lf, __u32 type, + const char *name, + struct thandle *handle) { struct dt_object *next = mdd_object_child(pobj); int rc; ENTRY; if (dt_try_as_dir(env, next)) { - struct lu_ucred *uc = lu_ucred_check(env); - int ignore_quota; + struct dt_insert_rec *rec = &mdd_env_info(env)->mti_dt_rec; + struct lu_ucred *uc = lu_ucred_check(env); + int ignore_quota; + rec->rec_fid = lf; + rec->rec_type = type; ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1; - rc = next->do_index_ops->dio_insert(env, next, - (struct dt_rec*)lf, - (const struct dt_key *)name, - handle, capa, ignore_quota); + rc = dt_insert(env, next, (const struct dt_rec *)rec, + (const struct dt_key *)name, handle, + ignore_quota); } else { rc = -ENOTDIR; } @@ -613,30 +599,31 @@ static int __mdd_index_insert_only(const struct lu_env *env, /* insert named index, add reference if isdir */ static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj, - const struct lu_fid *lf, const char *name, int is_dir, - struct thandle *handle, struct lustre_capa *capa) + const struct lu_fid *lf, __u32 type, + const char *name, struct thandle *handle) { - int rc; - ENTRY; + int rc; + ENTRY; - rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa); - if (rc == 0 && is_dir) { - mdd_write_lock(env, pobj, MOR_TGT_PARENT); - mdo_ref_add(env, pobj, handle); - mdd_write_unlock(env, pobj); - } - RETURN(rc); + rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle); + if (rc == 0 && S_ISDIR(type)) { + mdd_write_lock(env, pobj, MOR_TGT_PARENT); + mdo_ref_add(env, pobj, handle); + mdd_write_unlock(env, pobj); + } + + RETURN(rc); } /* delete named index, drop reference if isdir */ static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj, - const char *name, int is_dir, struct thandle *handle, - struct lustre_capa *capa) + const char *name, int is_dir, + struct thandle *handle) { int rc; ENTRY; - rc = __mdd_index_delete_only(env, pobj, name, handle, capa); + rc = __mdd_index_delete_only(env, pobj, name, handle); if (rc == 0 && is_dir) { mdd_write_lock(env, pobj, MOR_TGT_PARENT); mdo_ref_del(env, pobj, handle); @@ -646,52 +633,37 @@ static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj, RETURN(rc); } -int mdd_declare_changelog_store(const struct lu_env *env, - struct mdd_device *mdd, - const struct lu_name *fname, - struct thandle *handle) +static int mdd_llog_record_calc_size(const struct lu_env *env, + const struct lu_name *tname, + const struct lu_name *sname) { - struct obd_device *obd = mdd2obd_dev(mdd); - struct llog_ctxt *ctxt; - struct llog_changelog_rec *rec; - struct lu_buf *buf; - int reclen; - int rc; + const struct lu_ucred *uc = lu_ucred(env); + enum changelog_rec_flags crf = 0; + size_t hdr_size = sizeof(struct llog_changelog_rec) - + sizeof(struct changelog_rec); - /* Not recording */ - if (!(mdd->mdd_cl.mc_flags & CLM_ON)) - return 0; + if (sname != NULL) + crf |= CLF_RENAME; - reclen = llog_data_len(sizeof(*rec) + - (fname != NULL ? fname->ln_namelen : 0)); - buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); - if (buf->lb_buf == NULL) - return -ENOMEM; + if (uc != NULL && uc->uc_jobid[0] != '\0') + crf |= CLF_JOBID; - rec = buf->lb_buf; - rec->cr_hdr.lrh_len = reclen; - rec->cr_hdr.lrh_type = CHANGELOG_REC; - - ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); - if (ctxt == NULL) - return -ENXIO; - - rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle); - llog_ctxt_put(ctxt); - - return rc; + return llog_data_len(hdr_size + changelog_rec_offset(crf) + + (tname != NULL ? tname->ln_namelen : 0) + + (sname != NULL ? 1 + sname->ln_namelen : 0)); } -static int mdd_declare_changelog_ext_store(const struct lu_env *env, - struct mdd_device *mdd, - const struct lu_name *tname, - const struct lu_name *sname, - struct thandle *handle) +int mdd_declare_changelog_store(const struct lu_env *env, + struct mdd_device *mdd, + const struct lu_name *tname, + const struct lu_name *sname, + struct thandle *handle) { struct obd_device *obd = mdd2obd_dev(mdd); struct llog_ctxt *ctxt; - struct llog_changelog_ext_rec *rec; + struct llog_changelog_rec *rec; struct lu_buf *buf; + struct thandle *llog_th; int reclen; int rc; @@ -699,9 +671,7 @@ static int mdd_declare_changelog_ext_store(const struct lu_env *env, if (!(mdd->mdd_cl.mc_flags & CLM_ON)) return 0; - reclen = llog_data_len(sizeof(*rec) + - (tname != NULL ? tname->ln_namelen : 0) + - (sname != NULL ? 1 + sname->ln_namelen : 0)); + reclen = mdd_llog_record_calc_size(env, tname, sname); buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); if (buf->lb_buf == NULL) return -ENOMEM; @@ -714,7 +684,13 @@ static int mdd_declare_changelog_ext_store(const struct lu_env *env, if (ctxt == NULL) return -ENXIO; - rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle); + llog_th = thandle_get_sub(env, handle, ctxt->loc_handle->lgh_obj); + if (IS_ERR(llog_th)) + GOTO(out_put, rc = PTR_ERR(llog_th)); + + rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, llog_th); + +out_put: llog_ctxt_put(ctxt); return rc; @@ -724,7 +700,7 @@ static int mdd_declare_changelog_ext_store(const struct lu_env *env, * \param mdd * \param rec * \param handle - currently ignored since llogs start their own transaction; - * this will hopefully be fixed in llog rewrite + * this will hopefully be fixed in llog rewrite * \retval 0 ok */ int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd, @@ -732,46 +708,12 @@ int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd, { struct obd_device *obd = mdd2obd_dev(mdd); struct llog_ctxt *ctxt; + struct thandle *llog_th; int rc; - rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen); - rec->cr_hdr.lrh_type = CHANGELOG_REC; - rec->cr.cr_time = cl_time(); - - spin_lock(&mdd->mdd_cl.mc_lock); - /* NB: I suppose it's possible llog_add adds out of order wrt cr_index, - * but as long as the MDD transactions are ordered correctly for e.g. - * rename conflicts, I don't think this should matter. */ - rec->cr.cr_index = ++mdd->mdd_cl.mc_index; - spin_unlock(&mdd->mdd_cl.mc_lock); - - ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); - if (ctxt == NULL) - return -ENXIO; - - rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th); - llog_ctxt_put(ctxt); - if (rc > 0) - rc = 0; - return rc; -} - -/** Add a changelog_ext entry \a rec to the changelog llog - * \param mdd - * \param rec - * \param handle - currently ignored since llogs start their own transaction; - * this will hopefully be fixed in llog rewrite - * \retval 0 ok - */ -static int -mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd, - struct llog_changelog_ext_rec *rec, struct thandle *th) -{ - struct obd_device *obd = mdd2obd_dev(mdd); - struct llog_ctxt *ctxt; - int rc; + rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + + changelog_rec_varsize(&rec->cr)); - rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen); /* llog_lvfs_write_rec sets the llog tail len */ rec->cr_hdr.lrh_type = CHANGELOG_REC; rec->cr.cr_time = cl_time(); @@ -787,72 +729,50 @@ mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd, if (ctxt == NULL) return -ENXIO; + llog_th = thandle_get_sub(env, th, ctxt->loc_handle->lgh_obj); + if (IS_ERR(llog_th)) + GOTO(out_put, rc = PTR_ERR(llog_th)); + /* nested journal transaction */ - rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, th); + rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, llog_th); + +out_put: llog_ctxt_put(ctxt); if (rc > 0) rc = 0; - return rc; } -/** Store a namespace change changelog record - * If this fails, we must fail the whole transaction; we don't - * want the change to commit without the log entry. - * \param target - mdd_object of change - * \param parent - parent dir/object - * \param tname - target name string - * \param handle - transacion handle - */ -int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd, - enum changelog_rec_type type, unsigned flags, - struct mdd_object *target, struct mdd_object *parent, - const struct lu_name *tname, struct thandle *handle) -{ - struct llog_changelog_rec *rec; - struct lu_buf *buf; - int reclen; - int rc; - ENTRY; - - /* Not recording */ - if (!(mdd->mdd_cl.mc_flags & CLM_ON)) - RETURN(0); - if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) - RETURN(0); +static void mdd_changelog_rec_ext_rename(struct changelog_rec *rec, + const struct lu_fid *sfid, + const struct lu_fid *spfid, + const struct lu_name *sname) +{ + struct changelog_ext_rename *rnm = changelog_rec_rename(rec); + size_t extsize = sname->ln_namelen + 1; - LASSERT(target != NULL); - LASSERT(parent != NULL); - LASSERT(tname != NULL); - LASSERT(handle != NULL); + LASSERT(sfid != NULL); + LASSERT(spfid != NULL); + LASSERT(sname != NULL); - reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen); - buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); - if (buf->lb_buf == NULL) - RETURN(-ENOMEM); - rec = buf->lb_buf; + rnm->cr_sfid = *sfid; + rnm->cr_spfid = *spfid; - rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags); - rec->cr.cr_type = (__u32)type; - rec->cr.cr_tfid = *mdo2fid(target); - rec->cr.cr_pfid = *mdo2fid(parent); - rec->cr.cr_namelen = tname->ln_namelen; - memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen); + changelog_rec_name(rec)[rec->cr_namelen] = '\0'; + strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize); + rec->cr_namelen += extsize; +} - target->mod_cltime = cfs_time_current_64(); +void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid) +{ + struct changelog_ext_jobid *jid = changelog_rec_jobid(rec); - rc = mdd_changelog_store(env, mdd, rec, handle); - if (rc < 0) { - CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n", - rc, type, tname->ln_name, PFID(&rec->cr.cr_tfid), - PFID(&rec->cr.cr_pfid)); - RETURN(-EFAULT); - } + if (jobid == NULL || jobid[0] == '\0') + return; - RETURN(0); + strlcpy(jid->cr_jobid, jobid, sizeof(jid->cr_jobid)); } - /** Store a namespace change changelog record * If this fails, we must fail the whole transaction; we don't * want the change to commit without the log entry. @@ -864,55 +784,63 @@ int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd, * \param sname - source name string * \param handle - transacion handle */ -static int mdd_changelog_ext_ns_store(const struct lu_env *env, - struct mdd_device *mdd, - enum changelog_rec_type type, - unsigned flags, - struct mdd_object *target, - const struct lu_fid *tpfid, - const struct lu_fid *sfid, - const struct lu_fid *spfid, - const struct lu_name *tname, - const struct lu_name *sname, - struct thandle *handle) -{ - struct llog_changelog_ext_rec *rec; - struct lu_buf *buf; - int reclen; - int rc; +int mdd_changelog_ns_store(const struct lu_env *env, + struct mdd_device *mdd, + enum changelog_rec_type type, + enum changelog_rec_flags crf, + struct mdd_object *target, + const struct lu_fid *tpfid, + const struct lu_fid *sfid, + const struct lu_fid *spfid, + const struct lu_name *tname, + const struct lu_name *sname, + struct thandle *handle) +{ + const struct lu_ucred *uc = lu_ucred(env); + struct llog_changelog_rec *rec; + struct lu_buf *buf; + int reclen; + int rc; ENTRY; /* Not recording */ if (!(mdd->mdd_cl.mc_flags & CLM_ON)) RETURN(0); + if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) RETURN(0); - LASSERT(sfid != NULL); LASSERT(tpfid != NULL); LASSERT(tname != NULL); LASSERT(handle != NULL); - reclen = llog_data_len(sizeof(*rec) + - sname != NULL ? 1 + sname->ln_namelen : 0); + reclen = mdd_llog_record_calc_size(env, tname, sname); buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); if (buf->lb_buf == NULL) RETURN(-ENOMEM); rec = buf->lb_buf; - rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags); + crf &= CLF_FLAGMASK; + + if (uc != NULL && uc->uc_jobid[0] != '\0') + crf |= CLF_JOBID; + + if (sname != NULL) + crf |= CLF_RENAME; + else + crf |= CLF_VERSION; + + rec->cr.cr_flags = crf; rec->cr.cr_type = (__u32)type; rec->cr.cr_pfid = *tpfid; - rec->cr.cr_sfid = *sfid; - rec->cr.cr_spfid = *spfid; rec->cr.cr_namelen = tname->ln_namelen; - memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen); - if (sname) { - rec->cr.cr_name[tname->ln_namelen] = '\0'; - memcpy(rec->cr.cr_name + tname->ln_namelen + 1, sname->ln_name, - sname->ln_namelen); - rec->cr.cr_namelen += 1 + sname->ln_namelen; - } + memcpy(changelog_rec_name(&rec->cr), tname->ln_name, tname->ln_namelen); + + if (crf & CLF_RENAME) + mdd_changelog_rec_ext_rename(&rec->cr, sfid, spfid, sname); + + if (crf & CLF_JOBID) + mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid); if (likely(target != NULL)) { rec->cr.cr_tfid = *mdo2fid(target); @@ -921,7 +849,7 @@ static int mdd_changelog_ext_ns_store(const struct lu_env *env, fid_zero(&rec->cr.cr_tfid); } - rc = mdd_changelog_ext_store(env, mdd, rec, handle); + rc = mdd_changelog_store(env, mdd, rec, handle); if (rc < 0) { CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n", rc, type, tname->ln_name, PFID(sfid), PFID(tpfid)); @@ -1144,20 +1072,19 @@ int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj, if (ldata->ld_buf->lb_buf == NULL) return -ENOMEM; - rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK, - BYPASS_CAPA); + rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK); if (rc == -ERANGE) { /* Buf was too small, figure out what we need. */ lu_buf_free(ldata->ld_buf); rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, - XATTR_NAME_LINK, BYPASS_CAPA); + XATTR_NAME_LINK); if (rc < 0) return rc; ldata->ld_buf = lu_buf_check_and_alloc(ldata->ld_buf, rc); if (ldata->ld_buf->lb_buf == NULL) return -ENOMEM; rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, - XATTR_NAME_LINK, BYPASS_CAPA); + XATTR_NAME_LINK); } if (rc < 0) { lu_buf_free(ldata->ld_buf); @@ -1176,7 +1103,7 @@ int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj, struct lu_buf *mdd_links_get(const struct lu_env *env, struct mdd_object *mdd_obj) { - struct linkea_data ldata = { 0 }; + struct linkea_data ldata = { NULL }; int rc; rc = mdd_links_read(env, mdd_obj, &ldata); @@ -1188,12 +1115,38 @@ int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj, { const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf, ldata->ld_leh->leh_len); - return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle, - mdd_object_capa(env, mdd_obj)); + int rc; + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA)) + return 0; + + rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle); + if (unlikely(rc == -ENOSPC) && S_ISREG(mdd_object_type(mdd_obj)) && + mdd_object_remote(mdd_obj) == 0) { + struct lfsck_request *lr = &mdd_env_info(env)->mti_lr; + struct thandle *sub_th; + + /* XXX: If the linkEA is overflow, then we need to notify the + * namespace LFSCK to skip "nlink" attribute verification + * on this object to avoid the "nlink" to be shrinked by + * wrong. It may be not good an interaction with LFSCK + * like this. We will consider to replace it with other + * mechanism in future. LU-5802. */ + lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK, + LFSCK_TYPE_NAMESPACE); + + sub_th = thandle_get_sub_by_dt(env, handle, + mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom); + lfsck_in_notify(env, mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom, + lr, sub_th); + } + + return rc; } int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, - struct thandle *handle, struct linkea_data *ldata) + struct thandle *handle, struct linkea_data *ldata, + enum mdd_links_add_overflow overflow) { int rc; int ea_len; @@ -1211,6 +1164,29 @@ int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, rc = mdo_declare_xattr_set(env, mdd_obj, mdd_buf_get_const(env, linkea, ea_len), XATTR_NAME_LINK, 0, handle); + if (rc != 0) + return rc; + + if (mdd_object_remote(mdd_obj) == 0 && overflow == MLAO_CHECK) { + struct lfsck_request *lr = &mdd_env_info(env)->mti_lr; + struct thandle *sub_th; + + /* XXX: If the linkEA is overflow, then we need to notify the + * namespace LFSCK to skip "nlink" attribute verification + * on this object to avoid the "nlink" to be shrinked by + * wrong. It may be not good an interaction with LFSCK + * like this. We will consider to replace it with other + * mechanism in future. LU-5802. */ + lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK_DECLARE, + LFSCK_TYPE_NAMESPACE); + + sub_th = thandle_get_sub_by_dt(env, handle, + mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom); + rc = lfsck_in_notify(env, + mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom, + lr, sub_th); + } + return rc; } @@ -1223,29 +1199,36 @@ static inline int mdd_declare_links_del(const struct lu_env *env, /* For directory, the linkEA will be removed together * with the object. */ if (!S_ISDIR(mdd_object_type(c))) - rc = mdd_declare_links_add(env, c, handle, NULL); + rc = mdd_declare_links_add(env, c, handle, NULL, MLAO_IGNORE); return rc; } static int mdd_declare_link(const struct lu_env *env, - struct mdd_device *mdd, - struct mdd_object *p, - struct mdd_object *c, - const struct lu_name *name, + struct mdd_device *mdd, + struct mdd_object *p, + struct mdd_object *c, + const struct lu_name *name, struct thandle *handle, struct lu_attr *la, struct linkea_data *data) { - int rc; + int rc; - rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle); - if (rc) - return rc; + rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c), + name->ln_name, handle); + if (rc != 0) + return rc; - rc = mdo_declare_ref_add(env, c, handle); - if (rc) - return rc; + rc = mdo_declare_ref_add(env, c, handle); + if (rc != 0) + return rc; + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) { + rc = mdo_declare_ref_add(env, c, handle); + if (rc != 0) + return rc; + } la->la_valid = LA_CTIME | LA_MTIME; rc = mdo_declare_attr_set(env, p, la, handle); @@ -1254,16 +1237,17 @@ static int mdd_declare_link(const struct lu_env *env, la->la_valid = LA_CTIME; rc = mdo_declare_attr_set(env, c, la, handle); - if (rc) - return rc; + if (rc != 0) + return rc; - rc = mdd_declare_links_add(env, c, handle, data); - if (rc) - return rc; + rc = mdd_declare_links_add(env, c, handle, data, + S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE); + if (rc != 0) + return rc; - rc = mdd_declare_changelog_store(env, mdd, name, handle); + rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle); - return rc; + return rc; } static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, @@ -1282,11 +1266,11 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, int rc; ENTRY; - rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_sobj, cattr); if (rc != 0) RETURN(rc); - rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_tobj, tattr); if (rc != 0) RETURN(rc); @@ -1314,14 +1298,31 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, if (rc) GOTO(out_unlock, rc); - rc = mdo_ref_add(env, mdd_sobj, handle); - if (rc) - GOTO(out_unlock, rc); + if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) { + rc = mdo_ref_add(env, mdd_sobj, handle); + if (rc != 0) + GOTO(out_unlock, rc); + } + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) { + rc = mdo_ref_add(env, mdd_sobj, handle); + if (rc != 0) + GOTO(out_unlock, rc); + } + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) { + struct lu_fid tfid = *mdo2fid(mdd_sobj); + + tfid.f_oid++; + rc = __mdd_index_insert_only(env, mdd_tobj, &tfid, + mdd_object_type(mdd_sobj), + name, handle); + } else { + rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), + mdd_object_type(mdd_sobj), + name, handle); + } - rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), - name, handle, - mdd_object_capa(env, mdd_tobj)); if (rc != 0) { mdo_ref_del(env, mdd_sobj, handle); GOTO(out_unlock, rc); @@ -1350,7 +1351,8 @@ out_unlock: mdd_write_unlock(env, mdd_sobj); if (rc == 0) rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj, - mdd_tobj, lname, handle); + mdo2fid(mdd_tobj), NULL, NULL, + lname, NULL, handle); stop: mdd_trans_stop(env, mdd, rc, handle); @@ -1380,8 +1382,7 @@ static int mdd_mark_dead_object(const struct lu_env *env, if (declare) rc = mdo_declare_attr_set(env, obj, attr, handle); else - rc = mdo_attr_set(env, obj, attr, handle, - mdd_object_capa(env, obj)); + rc = mdo_attr_set(env, obj, attr, handle); return rc; } @@ -1471,20 +1472,25 @@ int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj, static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd, struct mdd_object *p, struct mdd_object *c, const struct lu_name *name, struct md_attr *ma, - struct thandle *handle, int no_name) + struct thandle *handle, int no_name, int is_dir) { - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - int rc; + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + int rc; - if (likely(no_name == 0)) { - rc = mdo_declare_index_delete(env, p, name->ln_name, handle); - if (rc) - return rc; - } + if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) { + if (likely(no_name == 0)) { + rc = mdo_declare_index_delete(env, p, name->ln_name, + handle); + if (rc != 0) + return rc; + } - rc = mdo_declare_ref_del(env, p, handle); - if (rc) - return rc; + if (is_dir != 0) { + rc = mdo_declare_ref_del(env, p, handle); + if (rc != 0) + return rc; + } + } LASSERT(ma->ma_attr.la_valid & LA_CTIME); la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; @@ -1512,7 +1518,7 @@ static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd, return rc; /* FIXME: need changelog for remove entry */ - rc = mdd_declare_changelog_store(env, mdd, name, handle); + rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle); } return rc; @@ -1537,8 +1543,7 @@ static bool mdd_hsm_archive_exists(const struct lu_env *env, hsm_buf = mdd_buf_get(env, NULL, 0); lu_buf_alloc(hsm_buf, buflen); - rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM, - mdd_object_capa(env, obj)); + rc = mdo_xattr_get(env, obj, hsm_buf, XATTR_NAME_HSM); rc = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm); lu_buf_free(hsm_buf); if (rc < 0) @@ -1582,13 +1587,13 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, RETURN(-ENOENT); } - rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_pobj, pattr); if (rc) RETURN(rc); if (likely(mdd_cobj != NULL)) { /* fetch cattr */ - rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_cobj, cattr); if (rc) RETURN(rc); @@ -1604,7 +1609,7 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, RETURN(PTR_ERR(handle)); rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj, - lname, ma, handle, no_name); + lname, ma, handle, no_name, is_dir); if (rc) GOTO(stop, rc); @@ -1615,20 +1620,23 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, if (likely(mdd_cobj != NULL)) mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD); - if (likely(no_name == 0)) { - rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle, - mdd_object_capa(env, mdd_pobj)); + if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) { + rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle); if (rc) GOTO(cleanup, rc); } + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MUL_REF) || + OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_NAMEENTRY)) + GOTO(cleanup, rc = 0); + if (likely(mdd_cobj != NULL)) { rc = mdo_ref_del(env, mdd_cobj, handle); if (rc != 0) { __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj), - name, handle, - mdd_object_capa(env, mdd_pobj)); + mdd_object_type(mdd_cobj), + name, handle); GOTO(cleanup, rc); } @@ -1637,7 +1645,7 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, mdo_ref_del(env, mdd_cobj, handle); /* fetch updated nlink */ - rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_cobj, cattr); if (rc) GOTO(cleanup, rc); } @@ -1670,7 +1678,7 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, /* fetch updated nlink */ if (rc == 0) - rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_cobj, cattr); /* if object is removed then we can't get its attrs, use last get */ if (cattr->la_nlink == 0) { @@ -1679,7 +1687,9 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, } EXIT; cleanup: - mdd_write_unlock(env, mdd_cobj); + if (likely(mdd_cobj != NULL)) + mdd_write_unlock(env, mdd_cobj); + if (rc == 0) { int cl_flags = 0; @@ -1692,7 +1702,8 @@ cleanup: rc = mdd_changelog_ns_store(env, mdd, is_dir ? CL_RMDIR : CL_UNLINK, cl_flags, - mdd_cobj, mdd_pobj, lname, handle); + mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL, + handle); } stop: @@ -1743,7 +1754,7 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, * striping can be specified or not * 2) CMD? */ - rc = mdd_la_get(env, son, attr, BYPASS_CAPA); + rc = mdd_la_get(env, son, attr); if (rc) RETURN(rc); @@ -1774,7 +1785,7 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, if (rc) GOTO(stop, rc); - rc = mdd_declare_changelog_store(env, mdd, NULL, handle); + rc = mdd_declare_changelog_store(env, mdd, NULL, NULL, handle); if (rc) GOTO(stop, rc); @@ -1783,7 +1794,7 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, GOTO(stop, rc); rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV, - 0, handle, mdd_object_capa(env, son)); + 0, handle); if (rc) GOTO(stop, rc); @@ -1802,7 +1813,7 @@ static int mdd_declare_object_initialize(const struct lu_env *env, struct lu_attr *attr, struct thandle *handle) { - int rc; + int rc; ENTRY; /* @@ -1815,15 +1826,20 @@ static int mdd_declare_object_initialize(const struct lu_env *env, attr->la_valid &= ~(LA_MODE | LA_TYPE); rc = mdo_declare_attr_set(env, child, attr, handle); attr->la_valid |= LA_MODE | LA_TYPE; - if (rc == 0 && S_ISDIR(attr->la_mode)) { - rc = mdo_declare_index_insert(env, child, mdo2fid(child), - dot, handle); - if (rc == 0) - rc = mdo_declare_ref_add(env, child, handle); + if (rc != 0 || !S_ISDIR(attr->la_mode)) + RETURN(rc); - rc = mdo_declare_index_insert(env, child, mdo2fid(parent), - dotdot, handle); - } + rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR, + dot, handle); + if (rc != 0) + RETURN(rc); + + rc = mdo_declare_ref_add(env, child, handle); + if (rc != 0) + RETURN(rc); + + rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR, + dotdot, handle); RETURN(rc); } @@ -1834,31 +1850,20 @@ static int mdd_object_initialize(const struct lu_env *env, struct lu_attr *attr, struct thandle *handle, const struct md_op_spec *spec) { - int rc; - ENTRY; - - /* - * Update attributes for child. - * - * FIXME: - * (1) the valid bits should be converted between Lustre and Linux; - * (2) maybe, the child attributes should be set in OSD when creation. - */ + int rc = 0; + ENTRY; - rc = mdd_attr_set_internal(env, child, attr, handle, 0); - /* arguments are supposed to stay the same */ if (S_ISDIR(attr->la_mode)) { /* Add "." and ".." for newly created dir */ mdo_ref_add(env, child, handle); rc = __mdd_index_insert_only(env, child, mdo2fid(child), - dot, handle, BYPASS_CAPA); - if (rc == 0) - rc = __mdd_index_insert_only(env, child, pfid, - dotdot, handle, - BYPASS_CAPA); - if (rc != 0) - mdo_ref_del(env, child, handle); - } + S_IFDIR, dot, handle); + if (rc == 0) + rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR, + dotdot, handle); + if (rc != 0) + mdo_ref_del(env, child, handle); + } RETURN(rc); } @@ -2061,11 +2066,12 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, } else { struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name, - handle); - if (rc) + rc = mdo_declare_index_insert(env, p, mdo2fid(c), attr->la_mode, + name->ln_name, handle); + if (rc != 0) return rc; - rc = mdd_declare_links_add(env, c, handle, ldata); + + rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE); if (rc) return rc; @@ -2075,17 +2081,10 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, if (rc) return rc; - rc = mdd_declare_changelog_store(env, mdd, name, handle); + rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle); if (rc) return rc; } - - /* XXX: For remote create, it should indicate the remote RPC - * will be sent after local transaction is finished, which - * is not very nice, but it will be removed once we fully support - * async update */ - if (mdd_object_remote(p) && handle->th_update != NULL) - handle->th_update->tu_sent_after_local_trans = 1; out: return rc; } @@ -2105,7 +2104,7 @@ static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj, mdd_read_lock(env, pobj, MOR_TGT_PARENT); rc = mdo_xattr_get(env, pobj, def_acl_buf, - XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA); + XATTR_NAME_ACL_DEFAULT); mdd_read_unlock(env, pobj); if (rc > 0) { /* If there are default ACL, fix mode/ACL by default ACL */ @@ -2169,9 +2168,11 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, /* During creation, there are only a few cases we need do xattr_set to * create stripes. * 1. regular file: see comments above. - * 2. create striped directory with provided stripeEA. - * 3. create striped directory because inherit default layout from the - * parent. */ + * 2. dir: inherit default striping or pool settings from parent. + * 3. create striped directory with provided stripeEA. + * 4. create striped directory because inherit default layout from the + * parent. + */ if (spec->no_create || (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) || S_ISDIR(attr->la_mode)) { @@ -2182,7 +2183,7 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, rc = mdo_xattr_set(env, son, buf, S_ISDIR(attr->la_mode) ? XATTR_NAME_LMV : XATTR_NAME_LOV, 0, - handle, BYPASS_CAPA); + handle); if (rc != 0) GOTO(err_destroy, rc); } @@ -2193,7 +2194,7 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, /* set default acl */ rc = mdo_xattr_set(env, son, def_acl_buf, XATTR_NAME_ACL_DEFAULT, 0, - handle, BYPASS_CAPA); + handle); if (rc) GOTO(err_destroy, rc); } @@ -2201,7 +2202,7 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, if (acl_buf != NULL && acl_buf->lb_len > 0) { rc = mdo_xattr_set(env, son, acl_buf, XATTR_NAME_ACL_ACCESS, - 0, handle, BYPASS_CAPA); + 0, handle); if (rc) GOTO(err_destroy, rc); } @@ -2217,7 +2218,6 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, buf = mdd_buf_get_const(env, target_name, sym_len); rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle, - mdd_object_capa(env, son), uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK); @@ -2308,7 +2308,7 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, * 2. insert (__mdd_index_insert(), lookup again) */ - rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_pobj, pattr); if (rc != 0) RETURN(rc); @@ -2335,9 +2335,17 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint); memset(ldata, 0, sizeof(*ldata)); - rc = mdd_linkea_prepare(env, son, NULL, NULL, - mdd_object_fid(mdd_pobj), - lname, 1, 0, ldata); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) { + struct lu_fid tfid = *mdd_object_fid(mdd_pobj); + + tfid.f_oid--; + rc = mdd_linkea_prepare(env, son, NULL, NULL, + &tfid, lname, 1, 0, ldata); + } else { + rc = mdd_linkea_prepare(env, son, NULL, NULL, + mdd_object_fid(mdd_pobj), + lname, 1, 0, ldata); + } rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr, handle, spec, ldata, &def_acl_buf, &acl_buf, @@ -2360,8 +2368,7 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, GOTO(out_volatile, rc); } else { rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son), - name, S_ISDIR(attr->la_mode), handle, - mdd_object_capa(env, mdd_pobj)); + attr->la_mode, name, handle); if (rc != 0) GOTO(err_created, rc); @@ -2386,7 +2393,7 @@ err_insert: else rc2 = __mdd_index_delete(env, mdd_pobj, name, S_ISDIR(attr->la_mode), - handle, BYPASS_CAPA); + handle); if (rc2 != 0) goto out_stop; @@ -2424,7 +2431,8 @@ out_volatile: S_ISDIR(attr->la_mode) ? CL_MKDIR : S_ISREG(attr->la_mode) ? CL_CREATE : S_ISLNK(attr->la_mode) ? CL_SOFTLINK : CL_MKNOD, - 0, son, mdd_pobj, lname, handle); + 0, son, mdo2fid(mdd_pobj), NULL, NULL, lname, + NULL, handle); out_stop: rc2 = mdd_trans_stop(env, mdd, rc, handle); if (rc == 0) @@ -2562,21 +2570,22 @@ static int mdd_declare_rename(const struct lu_env *env, if (mdd_spobj != mdd_tpobj) { rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle); - if (rc) + if (rc != 0) return rc; rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj), - dotdot, handle); - if (rc) + S_IFDIR, dotdot, handle); + if (rc != 0) return rc; } - /* new target child can be directory, - * counted by target dir's nlink */ - rc = mdo_declare_ref_add(env, mdd_tpobj, handle); - if (rc) - return rc; - } + + /* new target child can be directory, + * counted by target dir's nlink */ + rc = mdo_declare_ref_add(env, mdd_tpobj, handle); + if (rc != 0) + return rc; + } la->la_valid = LA_CTIME | LA_MTIME; rc = mdo_declare_attr_set(env, mdd_spobj, la, handle); @@ -2592,15 +2601,17 @@ static int mdd_declare_rename(const struct lu_env *env, if (rc) return rc; - rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata); + rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata, + S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE); if (rc) return rc; - /* new name */ - rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj), - tname->ln_name, handle); - if (rc) - return rc; + /* new name */ + rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj), + mdd_object_type(mdd_sobj), + tname->ln_name, handle); + if (rc != 0) + return rc; /* name from target dir (old name), we declare it unconditionally * as mdd_rename() calls delete unconditionally as well. so just @@ -2638,7 +2649,7 @@ static int mdd_declare_rename(const struct lu_env *env, return rc; } - rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle); + rc = mdd_declare_changelog_store(env, mdd, tname, sname, handle); if (rc) return rc; @@ -2680,21 +2691,21 @@ static int mdd_rename(const struct lu_env *env, mdd_sobj = mdd_object_find(env, mdd, lf); - rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_sobj, cattr); if (rc) GOTO(out_pending, rc); - rc = mdd_la_get(env, mdd_spobj, pattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_spobj, pattr); if (rc) GOTO(out_pending, rc); if (mdd_tobj) { - rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_tobj, tattr); if (rc) GOTO(out_pending, rc); } - rc = mdd_la_get(env, mdd_tpobj, tpattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_tpobj, tpattr); if (rc) GOTO(out_pending, rc); @@ -2731,21 +2742,19 @@ static int mdd_rename(const struct lu_env *env, is_dir = S_ISDIR(cattr->la_mode); /* Remove source name from source directory */ - rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle, - mdd_object_capa(env, mdd_spobj)); + rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle); if (rc) GOTO(cleanup, rc); /* "mv dir1 dir2" needs "dir1/.." link update */ if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) { - rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle, - mdd_object_capa(env, mdd_sobj)); - if (rc) - GOTO(fixup_spobj2, rc); + rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle); + if (rc != 0) + GOTO(fixup_spobj2, rc); - rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot, - handle, mdd_object_capa(env, mdd_sobj)); - if (rc) + rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR, + dotdot, handle); + if (rc != 0) GOTO(fixup_spobj, rc); } @@ -2753,8 +2762,7 @@ static int mdd_rename(const struct lu_env *env, * Here tobj can be remote one, so we do index_delete unconditionally * and -ENOENT is allowed. */ - rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle, - mdd_object_capa(env, mdd_tpobj)); + rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle); if (rc != 0) { if (mdd_tobj) { /* tname might been renamed to something else */ @@ -2765,9 +2773,9 @@ static int mdd_rename(const struct lu_env *env, } /* Insert new fid with target name into target dir */ - rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle, - mdd_object_capa(env, mdd_tpobj)); - if (rc) + rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode, + tname, handle); + if (rc != 0) GOTO(fixup_tpobj, rc); LASSERT(ma->ma_attr.la_valid & LA_CTIME); @@ -2803,7 +2811,7 @@ static int mdd_rename(const struct lu_env *env, tobj_ref = 1; /* fetch updated nlink */ - rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_tobj, tattr); if (rc != 0) { CERROR("%s: Failed to get nlink for tobj " DFID": rc = %d\n", @@ -2836,7 +2844,7 @@ static int mdd_rename(const struct lu_env *env, } /* fetch updated nlink */ - rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_tobj, tattr); if (rc != 0) { CERROR("%s: Failed to get nlink for tobj " DFID": rc = %d\n", @@ -2884,8 +2892,7 @@ static int mdd_rename(const struct lu_env *env, fixup_tpobj: if (rc) { - rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle, - BYPASS_CAPA); + rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle); if (rc2) CWARN("tp obj fix error %d\n",rc2); @@ -2897,52 +2904,52 @@ fixup_tpobj: mdo_ref_add(env, mdd_tobj, handle); } - rc2 = __mdd_index_insert(env, mdd_tpobj, - mdo2fid(mdd_tobj), tname, - is_dir, handle, - BYPASS_CAPA); - - if (rc2) - CWARN("tp obj fix error %d\n",rc2); - } - } + rc2 = __mdd_index_insert(env, mdd_tpobj, + mdo2fid(mdd_tobj), + mdd_object_type(mdd_tobj), + tname, handle); + if (rc2 != 0) + CWARN("tp obj fix error: rc = %d\n", rc2); + } + } fixup_spobj: if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) { - rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle, - BYPASS_CAPA); - + rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle); if (rc2) CWARN("%s: sp obj dotdot delete error: rc = %d\n", mdd2obd_dev(mdd)->obd_name, rc2); - rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, - dotdot, handle, BYPASS_CAPA); - if (rc2) + rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR, + dotdot, handle); + if (rc2 != 0) CWARN("%s: sp obj dotdot insert error: rc = %d\n", mdd2obd_dev(mdd)->obd_name, rc2); } fixup_spobj2: - if (rc) { - rc2 = __mdd_index_insert(env, mdd_spobj, - lf, sname, is_dir, handle, BYPASS_CAPA); - if (rc2) - CWARN("sp obj fix error %d\n",rc2); - } + if (rc != 0) { + rc2 = __mdd_index_insert(env, mdd_spobj, lf, + mdd_object_type(mdd_sobj), sname, + handle); + if (rc2 != 0) + CWARN("sp obj fix error: rc = %d\n", rc2); + } + cleanup: if (tobj_locked) mdd_write_unlock(env, mdd_tobj); + cleanup_unlocked: - if (rc == 0) - rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags, - mdd_tobj, tpobj_fid, lf, - spobj_fid, ltname, lsname, - handle); + if (rc == 0) + rc = mdd_changelog_ns_store(env, mdd, CL_RENAME, cl_flags, + mdd_tobj, tpobj_fid, lf, spobj_fid, + ltname, lsname, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); + mdd_trans_stop(env, mdd, rc, handle); + out_pending: mdd_object_put(env, mdd_sobj); return rc; @@ -2960,7 +2967,7 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, bool declare) { struct mdd_thread_info *info = mdd_env_info(env); - struct linkea_data ldata = {0}; + struct linkea_data ldata = { NULL }; struct lu_buf *buf = &info->mti_link_buf; int count; int rc = 0; @@ -3005,7 +3012,8 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, linkea_entry_pack(ldata.ld_lee, &lname, mdd_object_fid(parent)); if (declare) - rc = mdd_declare_links_add(env, child, handle, &ldata); + rc = mdd_declare_links_add(env, child, handle, &ldata, + MLAO_IGNORE); else rc = mdd_links_write(env, child, &ldata, handle); break; @@ -3054,14 +3062,6 @@ static int mdd_update_linkea_internal(const struct lu_env *env, RETURN(rc); } - if (declare) - rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata); - else - rc = mdd_links_write(env, mdd_tobj, ldata, handle); - - if (rc != 0) - RETURN(rc); - /* If it is mulitple links file, we need update the name entry for * all parent */ LASSERT(ldata->ld_leh != NULL); @@ -3109,13 +3109,14 @@ static int mdd_update_linkea_internal(const struct lu_env *env, /* Insert new fid with target name into target dir */ rc = mdo_declare_index_delete(env, pobj, lname.ln_name, handle); - if (rc) + if (rc != 0) GOTO(next_put, rc); rc = mdo_declare_index_insert(env, pobj, - mdd_object_fid(mdd_tobj), - lname.ln_name, handle); - if (rc) + mdd_object_fid(mdd_tobj), + mdd_object_type(mdd_tobj), + lname.ln_name, handle); + if (rc != 0) GOTO(next_put, rc); rc = mdo_declare_ref_add(env, mdd_tobj, handle); @@ -3127,16 +3128,15 @@ static int mdd_update_linkea_internal(const struct lu_env *env, GOTO(next_put, rc); } else { rc = __mdd_index_delete(env, pobj, lname.ln_name, - 0, handle, - mdd_object_capa(env, pobj)); + 0, handle); if (rc) GOTO(next_put, rc); rc = __mdd_index_insert(env, pobj, - mdd_object_fid(mdd_tobj), - lname.ln_name, 0, handle, - mdd_object_capa(env, pobj)); - if (rc) + mdd_object_fid(mdd_tobj), + mdd_object_type(mdd_tobj), + lname.ln_name, handle); + if (rc != 0) GOTO(next_put, rc); mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD); @@ -3175,8 +3175,7 @@ static int mdd_migrate_xattrs(const struct lu_env *env, int rc; /* retrieve xattr list from the old object */ - list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL, - mdd_object_capa(env, mdd_sobj)); + list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL); if (list_xsize == -ENODATA) return 0; @@ -3189,8 +3188,7 @@ static int mdd_migrate_xattrs(const struct lu_env *env, list_xbuf.lb_buf = info->mti_big_buf.lb_buf; list_xbuf.lb_len = list_xsize; - rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf, - mdd_object_capa(env, mdd_sobj)); + rc = mdo_xattr_list(env, mdd_sobj, &list_xbuf); if (rc < 0) return rc; rc = 0; @@ -3208,9 +3206,7 @@ static int mdd_migrate_xattrs(const struct lu_env *env, !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) goto next; - xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, - xname, - mdd_object_capa(env, mdd_sobj)); + xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname); if (xsize == -ENODATA) goto next; if (xsize < 0) @@ -3222,8 +3218,7 @@ static int mdd_migrate_xattrs(const struct lu_env *env, xbuf.lb_len = xsize; xbuf.lb_buf = info->mti_link_buf.lb_buf; - rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname, - mdd_object_capa(env, mdd_sobj)); + rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname); if (rc == -ENODATA) goto next; if (rc < 0) @@ -3245,8 +3240,7 @@ static int mdd_migrate_xattrs(const struct lu_env *env, if (rc != 0) GOTO(stop_trans, rc); - rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle, - mdd_object_capa(env, mdd_sobj)); + rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle); if (rc == -EEXIST) GOTO(stop_trans, rc = 0); @@ -3271,6 +3265,7 @@ static int mdd_declare_migrate_create(const struct lu_env *env, struct md_op_spec *spec, struct lu_attr *la, union lmv_mds_md *mgr_ea, + struct linkea_data *ldata, struct thandle *handle) { struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix); @@ -3298,6 +3293,11 @@ static int mdd_declare_migrate_create(const struct lu_env *env, buf, 0, handle); if (rc != 0) return rc; + } else if (S_ISDIR(la->la_mode)) { + rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata, + MLAO_IGNORE); + if (rc != 0) + return rc; } if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) { @@ -3318,7 +3318,6 @@ static int mdd_declare_migrate_create(const struct lu_env *env, la_flag->la_valid = LA_FLAGS; la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL; - mdd_flags_xlate(mdd_sobj, la_flag->la_flags); rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle); return rc; @@ -3328,19 +3327,21 @@ static int mdd_migrate_create(const struct lu_env *env, struct mdd_object *mdd_pobj, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, + const struct lu_name *lname, struct lu_attr *la) { struct mdd_thread_info *info = mdd_env_info(env); struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj); struct md_op_spec *spec = &info->mti_spec; - struct lu_buf lmm_buf = { 0 }; - struct lu_buf link_buf = { 0 }; + struct lu_buf lmm_buf = { NULL }; + struct lu_buf link_buf = { NULL }; const struct lu_buf *buf; struct thandle *handle; struct lmv_mds_md_v1 *mgr_ea; struct lu_attr *la_flag = MDD_ENV_VAR(env, la_for_fix); struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint; int mgr_easize; + struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; int rc; ENTRY; @@ -3354,6 +3355,7 @@ static int mdd_migrate_create(const struct lu_env *env, la->la_size + 1); link_buf = *buf; link_buf.lb_len = la->la_size + 1; + memset(link_buf.lb_buf, 0, link_buf.lb_len); rc = mdd_readlink(env, &mdd_sobj->mod_obj, &link_buf); if (rc <= 0) { rc = rc != 0 ? rc : -EFAULT; @@ -3373,10 +3375,14 @@ static int mdd_migrate_create(const struct lu_env *env, spec->u.sp_ea.eadatalen = lmm_buf.lb_len; spec->sp_cr_flags |= MDS_OPEN_HAS_EA; } + } else if (S_ISDIR(la->la_mode)) { + rc = mdd_links_read(env, mdd_sobj, ldata); + if (rc < 0 && rc != -ENODATA) + RETURN(rc); } mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf; - memset(mgr_ea, 0, sizeof(mgr_ea)); + memset(mgr_ea, 0, sizeof(*mgr_ea)); mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_V1); mgr_ea->lmv_stripe_count = cpu_to_le32(2); mgr_ea->lmv_master_mdt_index = mdd_seq_site(mdd)->ss_node_id; @@ -3397,7 +3403,7 @@ static int mdd_migrate_create(const struct lu_env *env, rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, spec, la, (union lmv_mds_md *)info->mti_xattr_buf, - handle); + ldata, handle); if (rc != 0) GOTO(stop_trans, rc); @@ -3411,13 +3417,18 @@ static int mdd_migrate_create(const struct lu_env *env, if (rc != 0) GOTO(stop_trans, rc); + if (S_ISDIR(la->la_mode)) { + rc = mdd_links_write(env, mdd_tobj, ldata, handle); + if (rc != 0) + GOTO(stop_trans, rc); + } + /* Set MIGRATE EA on the source inode, so once the migration needs * to be re-done during failover, the re-do process can locate the * target object which is already being created. */ mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1); buf = mdd_buf_get_const(env, mgr_ea, mgr_easize); - rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, - handle, mdd_object_capa(env, mdd_sobj)); + rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle); if (rc != 0) GOTO(stop_trans, rc); @@ -3428,9 +3439,7 @@ static int mdd_migrate_create(const struct lu_env *env, * flag and approve the migration */ la_flag->la_valid = LA_FLAGS; la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL; - mdd_flags_xlate(mdd_sobj, la_flag->la_flags); - rc = mdo_attr_set(env, mdd_sobj, la_flag, handle, - mdd_object_capa(env, mdd_sobj)); + rc = mdo_attr_set(env, mdd_sobj, la_flag, handle); stop_trans: if (handle != NULL) mdd_trans_stop(env, mdd, rc, handle); @@ -3465,8 +3474,7 @@ static int mdd_migrate_entries(const struct lu_env *env, * iterate directories */ iops = &next->do_index_ops->dio_it; - it = iops->init(env, next, LUDA_FID | LUDA_TYPE, - mdd_object_capa(env, mdd_sobj)); + it = iops->init(env, next, LUDA_FID | LUDA_TYPE); if (IS_ERR(it)) GOTO(out_ent, rc = PTR_ERR(it)); @@ -3516,7 +3524,7 @@ static int mdd_migrate_entries(const struct lu_env *env, if (IS_ERR(child)) GOTO(out, rc = PTR_ERR(child)); - is_dir = S_ISDIR(lu_object_attr(&child->mod_obj.mo_lu)); + is_dir = S_ISDIR(mdd_object_type(child)); snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name); @@ -3524,10 +3532,8 @@ static int mdd_migrate_entries(const struct lu_env *env, if (dt_try_as_dir(env, dt_tobj)) { struct lu_fid *fid = &mdd_env_info(env)->mti_fid2; - rc = dt_tobj->do_index_ops->dio_lookup(env, dt_tobj, - (struct dt_rec *)fid, - (struct dt_key *)name, - mdd_object_capa(env, mdd_tobj)); + rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid, + (struct dt_key *)name); if (unlikely(rc == 0)) target_exist = true; } @@ -3543,6 +3549,7 @@ static int mdd_migrate_entries(const struct lu_env *env, if (likely(!target_exist)) { rc = mdo_declare_index_insert(env, mdd_tobj, &ent->lde_fid, + mdd_object_type(child), name, handle); if (rc != 0) GOTO(out_put, rc); @@ -3571,7 +3578,7 @@ static int mdd_migrate_entries(const struct lu_env *env, rc = mdo_declare_index_insert(env, child, mdd_object_fid(mdd_tobj), - dotdot, handle); + S_IFDIR, dotdot, handle); if (rc != 0) GOTO(out_put, rc); } @@ -3592,34 +3599,25 @@ static int mdd_migrate_entries(const struct lu_env *env, if (likely(!target_exist)) { rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid, - name, is_dir, handle, - mdd_object_capa(env, mdd_tobj)); + mdd_object_type(child), + name, handle); if (rc != 0) GOTO(out_put, rc); - - if (is_dir) { - rc = mdo_ref_add(env, mdd_tobj, handle); - if (rc != 0) - GOTO(out_put, rc); - - } } - rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle, - mdd_object_capa(env, mdd_sobj)); + rc = __mdd_index_delete(env, mdd_sobj, name, is_dir, handle); if (rc != 0) GOTO(out_put, rc); if (is_dir) { - rc = __mdd_index_delete_only(env, child, dotdot, handle, - mdd_object_capa(env, child)); + rc = __mdd_index_delete_only(env, child, dotdot, + handle); if (rc != 0) GOTO(out_put, rc); rc = __mdd_index_insert_only(env, child, - mdd_object_fid(mdd_tobj), - dotdot, handle, - mdd_object_capa(env, child)); + mdd_object_fid(mdd_tobj), S_IFDIR, + dotdot, handle); if (rc != 0) GOTO(out_put, rc); } @@ -3685,7 +3683,6 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, /* Revert IMMUTABLE flag */ la_flag->la_valid = LA_FLAGS; la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL; - mdd_flags_xlate(mdd_sobj, la_flag->la_flags); rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle); if (rc != 0) return rc; @@ -3715,10 +3712,15 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, /* new name */ rc = mdo_declare_index_insert(env, mdd_pobj, mdo2fid(mdd_tobj), + mdd_object_type(mdd_tobj), lname->ln_name, handle); if (rc != 0) return rc; + rc = mdd_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE); + if (rc != 0) + return rc; + if (S_ISDIR(mdd_object_type(mdd_sobj))) { rc = mdo_declare_ref_add(env, mdd_pobj, handle); if (rc != 0) @@ -3772,7 +3774,7 @@ static int mdd_migrate_update_name(const struct lu_env *env, p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime; p_la->la_valid = LA_CTIME; - rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj)); + rc = mdd_la_get(env, mdd_sobj, so_attr); if (rc != 0) RETURN(rc); @@ -3803,15 +3805,12 @@ static int mdd_migrate_update_name(const struct lu_env *env, /* Revert IMMUTABLE flag */ la_flag->la_valid = LA_FLAGS; la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL; - mdd_flags_xlate(mdd_sobj, la_flag->la_flags); - rc = mdo_attr_set(env, mdd_sobj, la_flag, handle, - mdd_object_capa(env, mdd_pobj)); + rc = mdo_attr_set(env, mdd_sobj, la_flag, handle); if (rc != 0) GOTO(stop_trans, rc); /* Remove source name from source directory */ - rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle, - mdd_object_capa(env, mdd_pobj)); + rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle); if (rc != 0) GOTO(stop_trans, rc); @@ -3823,16 +3822,15 @@ static int mdd_migrate_update_name(const struct lu_env *env, if (S_ISREG(so_attr->la_mode)) { if (so_attr->la_nlink == 1) { rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV, - handle, - mdd_object_capa(env, mdd_sobj)); + handle); if (rc != 0 && rc != -ENODATA) GOTO(stop_trans, rc); } } /* Insert new fid with target name into target dir */ - rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj), name, - is_dir, handle, mdd_object_capa(env, mdd_pobj)); + rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj), + mdd_object_type(mdd_tobj), name, handle); if (rc != 0) GOTO(stop_trans, rc); @@ -3847,20 +3845,21 @@ static int mdd_migrate_update_name(const struct lu_env *env, mdo_ref_del(env, mdd_sobj, handle); /* Get the attr again after ref_del */ - rc = mdd_la_get(env, mdd_sobj, so_attr, - mdd_object_capa(env, mdd_sobj)); + rc = mdd_la_get(env, mdd_sobj, so_attr); if (rc != 0) - GOTO(stop_trans, rc); + GOTO(out_unlock, rc); + ma->ma_attr = *so_attr; ma->ma_valid |= MA_INODE; rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle); if (rc != 0) - GOTO(stop_trans, rc); + GOTO(out_unlock, rc); rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0); if (rc != 0) - GOTO(stop_trans, rc); + GOTO(out_unlock, rc); +out_unlock: mdd_write_unlock(env, mdd_sobj); stop_trans: @@ -3897,8 +3896,7 @@ static int mdd_migrate_sanity_check(const struct lu_env *env, if (mgr_buf->lb_buf == NULL) RETURN(-ENOMEM); - rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV, - mdd_object_capa(env, sobj)); + rc = mdo_xattr_get(env, sobj, mgr_buf, XATTR_NAME_LMV); if (rc > 0) { union lmv_mds_md *lmm = mgr_buf->lb_buf; @@ -3911,7 +3909,6 @@ static int mdd_migrate_sanity_check(const struct lu_env *env, struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj); sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL; - sobj->mod_flags &= ~IMMUTE_OBJ; CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n", mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(sobj))); @@ -4006,11 +4003,11 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, } mdd_read_unlock(env, mdd_sobj); - rc = mdd_la_get(env, mdd_sobj, so_attr, mdd_object_capa(env, mdd_sobj)); + rc = mdd_la_get(env, mdd_sobj, so_attr); if (rc != 0) GOTO(put, rc); - rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_pobj, pattr); if (rc != 0) GOTO(put, rc); @@ -4046,7 +4043,7 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, mdd_tobj = md2mdd_obj(tobj); if (!mdd_object_exists(mdd_tobj)) { rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, - so_attr); + lname, so_attr); if (rc != 0) GOTO(put, rc); } @@ -4064,6 +4061,8 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, if (unlikely(OBD_FAIL_CHECK_RESET(OBD_FAIL_MIGRATE_NET_REP, OBD_FAIL_MDS_REINT_NET_REP))) GOTO(put, rc = 0); + } else { + OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val); } /* step 4: update name entry to the new object */