X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_dir.c;h=8417a75040dc60ee4d65f2e7916fced127c8c9bd;hp=4532bcddc79425a383061a436fa28120abb28dfe;hb=e1ace3751f9add26b3f01aad9c278b6bfca8f739;hpb=325505601b3a3130d38834990661b79dfd126393 diff --git a/lustre/mdd/mdd_dir.c b/lustre/mdd/mdd_dir.c index 4532bcd..8417a75 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, 2014, Intel Corporation. + * Copyright (c) 2011, 2015, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -85,11 +85,12 @@ __mdd_lookup(const struct lu_env *env, struct md_object *pobj, if (unlikely(mdd_is_dead_obj(mdd_obj))) RETURN(-ESTALE); + if (!mdd_object_exists(mdd_obj)) + RETURN(-ESTALE); + if (mdd_object_remote(mdd_obj)) { CDEBUG(D_INFO, "%s: Object "DFID" locates on remote server\n", mdd2obd_dev(m)->obd_name, PFID(mdo2fid(mdd_obj))); - } else if (!mdd_object_exists(mdd_obj)) { - RETURN(-ESTALE); } rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask, @@ -99,8 +100,7 @@ __mdd_lookup(const struct lu_env *env, struct md_object *pobj, if (likely(S_ISDIR(mdd_object_type(mdd_obj)) && dt_try_as_dir(env, dir))) - rc = dt_lookup(env, dir, (struct dt_rec *)fid, key, - mdd_object_capa(env, mdd_obj)); + rc = dt_lookup(env, dir, (struct dt_rec *)fid, key); else rc = -ENOTDIR; @@ -115,7 +115,7 @@ 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); @@ -124,12 +124,57 @@ int mdd_lookup(const struct lu_env *env, RETURN(rc); } +/** + * Get parent FID of the directory + * + * Read parent FID from linkEA, if that fails, then do lookup + * dotdot to get the parent FID. + * + * \param[in] env execution environment + * \param[in] obj object from which to find the parent FID + * \param[in] attr attribute of the object + * \param[out] fid fid to get the parent FID + * + * \retval 0 if getting the parent FID succeeds. + * \retval negative errno if getting the parent FID fails. + **/ static inline int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj, const struct lu_attr *attr, struct lu_fid *fid) { - return __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0); + struct mdd_thread_info *info = mdd_env_info(env); + struct linkea_data ldata = { NULL }; + struct lu_buf *buf = &info->mti_link_buf; + struct lu_name lname; + int rc = 0; + + ENTRY; + + LASSERT(S_ISDIR(mdd_object_type(obj))); + + buf = lu_buf_check_and_alloc(buf, PATH_MAX); + if (buf->lb_buf == NULL) + GOTO(lookup, rc = 0); + + ldata.ld_buf = buf; + rc = mdd_links_read(env, obj, &ldata); + if (rc != 0) + GOTO(lookup, rc); + + LASSERT(ldata.ld_leh != NULL); + /* Directory should only have 1 parent */ + if (ldata.ld_leh->leh_reccount > 1) + GOTO(lookup, rc); + + ldata.ld_lee = (struct link_ea_entry *)(ldata.ld_leh + 1); + + linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, fid); + if (likely(fid_is_sane(fid))) + RETURN(0); +lookup: + rc = __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0); + RETURN(rc); } /* @@ -171,8 +216,7 @@ 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); @@ -182,18 +226,16 @@ static int mdd_is_parent(const struct lu_env *env, GOTO(out, rc = 0); if (lu_fid_eq(pfid, lf)) GOTO(out, rc = 1); - if (parent) - mdd_object_put(env, parent); + if (parent != NULL) + mdd_object_put(env, parent); parent = mdd_object_find(env, mdd, pfid); - if (IS_ERR(parent)) { + if (IS_ERR(parent)) GOTO(out, rc = PTR_ERR(parent)); - } else if (mdd_object_remote(parent)) { - /*FIXME: Because of the restriction of rename in Phase I. - * If the parent is remote, we just assumed lf is not the - * parent of P1 for now */ - GOTO(out, rc = 0); - } + + if (!mdd_object_exists(parent)) + GOTO(out, rc = -EINVAL); + p1 = parent; } EXIT; @@ -207,7 +249,7 @@ 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; @@ -225,7 +267,7 @@ 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); @@ -266,7 +308,7 @@ 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) { @@ -338,8 +380,6 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, const struct lu_attr *pattr, struct mdd_object *cobj, bool check_perm) { - struct mdd_thread_info *info = mdd_env_info(env); - struct lu_buf *xbuf; int rc = 0; ENTRY; @@ -349,19 +389,6 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, if (mdd_is_dead_obj(pobj)) RETURN(-ENOENT); - /* 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)); - if (unlikely(rc > 0)) { - struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf; - - if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE && - le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD) - RETURN(-ESTALE); - } - rc = 0; - if (check_perm) rc = mdd_permission_internal_locked(env, pobj, pattr, MAY_WRITE | MAY_EXEC, @@ -559,15 +586,14 @@ 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; if (dt_try_as_dir(env, next)) - rc = dt_delete(env, next, (struct dt_key *)name, handle, capa); + rc = dt_delete(env, next, (struct dt_key *)name, handle); else rc = -ENOTDIR; @@ -578,8 +604,7 @@ static int __mdd_index_insert_only(const struct lu_env *env, struct mdd_object *pobj, const struct lu_fid *lf, __u32 type, const char *name, - struct thandle *handle, - struct lustre_capa *capa) + struct thandle *handle) { struct dt_object *next = mdd_object_child(pobj); int rc; @@ -594,7 +619,7 @@ static int __mdd_index_insert_only(const struct lu_env *env, rec->rec_type = type; ignore_quota = uc ? uc->uc_cap & CFS_CAP_SYS_RESOURCE_MASK : 1; rc = dt_insert(env, next, (const struct dt_rec *)rec, - (const struct dt_key *)name, handle, capa, + (const struct dt_key *)name, handle, ignore_quota); } else { rc = -ENOTDIR; @@ -605,13 +630,12 @@ 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, __u32 type, - const char *name, struct thandle *handle, - struct lustre_capa *capa) + const char *name, struct thandle *handle) { int rc; ENTRY; - rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle, capa); + 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); @@ -623,13 +647,13 @@ static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj, /* 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); @@ -669,6 +693,7 @@ int mdd_declare_changelog_store(const struct lu_env *env, struct llog_ctxt *ctxt; struct llog_changelog_rec *rec; struct lu_buf *buf; + struct thandle *llog_th; int reclen; int rc; @@ -689,7 +714,13 @@ int mdd_declare_changelog_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; @@ -707,6 +738,7 @@ 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) + @@ -727,12 +759,17 @@ int mdd_changelog_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; } @@ -775,7 +812,7 @@ void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid) * \param spfid - source parent fid * \param tname - target name string * \param sname - source name string - * \param handle - transacion handle + * \param handle - transaction handle */ int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd, @@ -844,8 +881,10 @@ int mdd_changelog_ns_store(const struct lu_env *env, 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)); + CERROR("%s: cannot store changelog record: type = %d, " + "name = '%s', t = "DFID", p = "DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, type, tname->ln_name, + PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid), rc); return -EFAULT; } @@ -997,7 +1036,11 @@ out: int error = 1; if (rc == -EOVERFLOW || rc == -ENOSPC) error = 0; - if (oldpfid == NULL) + if (newlname == NULL) + CDEBUG(error ? D_ERROR : D_OTHER, + "link_ea add failed %d "DFID"\n", + rc, PFID(mdd_object_fid(mdd_obj))); + else if (oldpfid == NULL) CDEBUG(error ? D_ERROR : D_OTHER, "link_ea add '%.*s' failed %d "DFID"\n", newlname->ln_namelen, newlname->ln_name, @@ -1016,7 +1059,7 @@ out: rc, PFID(mdd_object_fid(mdd_obj))); } - if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + if (is_vmalloc_addr(ldata->ld_buf)) /* if we vmalloced a large buffer drop it */ lu_buf_free(ldata->ld_buf); @@ -1065,20 +1108,19 @@ 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); @@ -1107,18 +1149,23 @@ struct lu_buf *mdd_links_get(const struct lu_env *env, int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj, struct linkea_data *ldata, struct thandle *handle) { - const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf, - ldata->ld_leh->leh_len); + const struct lu_buf *buf; int rc; + if (ldata == NULL || ldata->ld_buf == NULL || + ldata->ld_leh == NULL) + return 0; + + buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf, + ldata->ld_leh->leh_len); if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA)) return 0; - rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle, - mdd_object_capa(env, mdd_obj)); + 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 @@ -1128,8 +1175,11 @@ int mdd_links_write(const struct lu_env *env, struct mdd_object *mdd_obj, * 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, handle); + lr, sub_th); } return rc; @@ -1160,6 +1210,7 @@ int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, 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 @@ -1169,9 +1220,12 @@ int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, * 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, handle); + lr, sub_th); } return rc; @@ -1200,9 +1254,13 @@ static int mdd_declare_link(const struct lu_env *env, struct lu_attr *la, struct linkea_data *data) { + struct lu_fid tfid = *mdo2fid(c); int rc; - rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c), + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) + tfid.f_oid = cfs_fail_val; + + rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c), name->ln_name, handle); if (rc != 0) return rc; @@ -1247,17 +1305,18 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, struct mdd_object *mdd_sobj = md2mdd_obj(src_obj); struct lu_attr *cattr = MDD_ENV_VAR(env, cattr); struct lu_attr *tattr = MDD_ENV_VAR(env, tattr); - struct mdd_device *mdd = mdo2mdd(src_obj); - struct thandle *handle; + struct mdd_device *mdd = mdo2mdd(src_obj); + struct thandle *handle; + struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2; struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; 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); @@ -1297,21 +1356,12 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, 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, - mdd_object_capa(env, mdd_tobj)); - } else { - rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), - mdd_object_type(mdd_sobj), - name, handle, - mdd_object_capa(env, mdd_tobj)); - } + *tfid = *mdo2fid(mdd_sobj); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) + tfid->f_oid = cfs_fail_val; + rc = __mdd_index_insert_only(env, mdd_tobj, tfid, + mdd_object_type(mdd_sobj), name, handle); if (rc != 0) { mdo_ref_del(env, mdd_sobj, handle); GOTO(out_unlock, rc); @@ -1335,44 +1385,39 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, * failure, reset rc here */ rc = 0; } - EXIT; + EXIT; out_unlock: - mdd_write_unlock(env, mdd_sobj); - if (rc == 0) + mdd_write_unlock(env, mdd_sobj); + if (rc == 0) rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj, mdo2fid(mdd_tobj), NULL, NULL, lname, NULL, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); - - if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + rc = mdd_trans_stop(env, mdd, rc, handle); + if (is_vmalloc_addr(ldata->ld_buf)) /* if we vmalloced a large buffer drop it */ lu_buf_free(ldata->ld_buf); out_pending: - return rc; + return rc; } -static int mdd_mark_dead_object(const struct lu_env *env, +static int mdd_mark_orphan_object(const struct lu_env *env, struct mdd_object *obj, struct thandle *handle, bool declare) { struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start); int rc; - if (!declare) - obj->mod_flags |= DEAD_OBJ; - if (!S_ISDIR(mdd_object_type(obj))) return 0; attr->la_valid = LA_FLAGS; - attr->la_flags = LUSTRE_SLAVE_DEAD_FL; + attr->la_flags = LUSTRE_ORPHAN_FL; 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; } @@ -1383,15 +1428,18 @@ static int mdd_declare_finish_unlink(const struct lu_env *env, { int rc; - rc = mdd_mark_dead_object(env, obj, handle, true); + /* Sigh, we do not know if the unlink object will become orphan in + * declare phase, but fortunately the flags here does not matter + * in current declare implementation */ + rc = mdd_mark_orphan_object(env, obj, handle, true); if (rc != 0) return rc; - rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle); + rc = mdo_declare_destroy(env, obj, handle); if (rc != 0) return rc; - rc = mdo_declare_destroy(env, obj, handle); + rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle); if (rc != 0) return rc; @@ -1406,34 +1454,37 @@ int mdd_finish_unlink(const struct lu_env *env, struct thandle *th) { int rc = 0; - int is_dir = S_ISDIR(ma->ma_attr.la_mode); - ENTRY; + int is_dir = S_ISDIR(ma->ma_attr.la_mode); + ENTRY; - LASSERT(mdd_write_locked(env, obj) != 0); + LASSERT(mdd_write_locked(env, obj) != 0); if (ma->ma_attr.la_nlink == 0 || is_dir) { - rc = mdd_mark_dead_object(env, obj, th, false); - if (rc != 0) - RETURN(rc); - - /* add new orphan and the object - * will be deleted during mdd_close() */ - if (obj->mod_count) { - rc = __mdd_orphan_add(env, obj, th); - if (rc == 0) - CDEBUG(D_HA, "Object "DFID" is inserted into " - "orphan list, open count = %d\n", - PFID(mdd_object_fid(obj)), - obj->mod_count); - else - CERROR("Object "DFID" fail to be an orphan, " - "open count = %d, maybe cause failed " - "open replay\n", - PFID(mdd_object_fid(obj)), - obj->mod_count); - } else { + /* add new orphan and the object + * will be deleted during mdd_close() */ + obj->mod_flags |= DEAD_OBJ; + if (obj->mod_count) { + rc = __mdd_orphan_add(env, obj, th); + if (rc == 0) + CDEBUG(D_HA, "Object "DFID" is inserted into " + "orphan list, open count = %d\n", + PFID(mdd_object_fid(obj)), + obj->mod_count); + else + CERROR("Object "DFID" fail to be an orphan, " + "open count = %d, maybe cause failed " + "open replay\n", + PFID(mdd_object_fid(obj)), + obj->mod_count); + + /* mark object as an orphan here, not + * before __mdd_orphan_add() as racing + * mdd_la_get() may propagate ORPHAN_OBJ + * causing the asserition */ + rc = mdd_mark_orphan_object(env, obj, th, false); + } else { rc = mdo_destroy(env, obj, th); - } + } } else if (!is_dir) { /* old files may not have link ea; ignore errors */ mdd_links_del(env, obj, mdo2fid(pobj), lname, th); @@ -1533,14 +1584,13 @@ 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) RETURN(false); - ma->ma_valid = MA_HSM; + ma->ma_valid |= MA_HSM; } if (ma->ma_hsm.mh_flags & HS_EXISTS) RETURN(true); @@ -1568,7 +1618,7 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, struct mdd_object *mdd_cobj = NULL; struct mdd_device *mdd = mdo2mdd(pobj); struct thandle *handle; - int rc, is_dir = 0; + int rc, is_dir = 0, cl_flags = 0; ENTRY; /* cobj == NULL means only delete name entry */ @@ -1578,17 +1628,22 @@ 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); is_dir = S_ISDIR(cattr->la_mode); + /* search for an existing archive. + * we should check ahead as the object + * can be destroyed in this transaction */ + if (mdd_hsm_archive_exists(env, mdd_cobj, ma)) + cl_flags |= CLF_UNLINK_HSM_EXISTS; } rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr); @@ -1612,8 +1667,7 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD); if (likely(no_name == 0) && !OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING2)) { - 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) GOTO(cleanup, rc); } @@ -1628,8 +1682,7 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj), mdd_object_type(mdd_cobj), - name, handle, - mdd_object_capa(env, mdd_pobj)); + name, handle); GOTO(cleanup, rc); } @@ -1638,7 +1691,7 @@ 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); } @@ -1668,30 +1721,33 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, ma->ma_attr = *cattr; ma->ma_valid |= MA_INODE; rc = mdd_finish_unlink(env, mdd_cobj, ma, mdd_pobj, lname, handle); + if (rc != 0) + GOTO(cleanup, rc); /* 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 (rc == -ENOENT) { + cattr->la_nlink = 0; + rc = 0; + } - /* if object is removed then we can't get its attrs, use last get */ if (cattr->la_nlink == 0) { ma->ma_attr = *cattr; ma->ma_valid |= MA_INODE; } + EXIT; cleanup: if (likely(mdd_cobj != NULL)) mdd_write_unlock(env, mdd_cobj); if (rc == 0) { - int cl_flags = 0; - - if (cattr->la_nlink == 0) { + if (cattr->la_nlink == 0) cl_flags |= CLF_UNLINK_LAST; - /* search for an existing archive */ - if (mdd_hsm_archive_exists(env, mdd_cobj, ma)) - cl_flags |= CLF_UNLINK_HSM_EXISTS; - } + else + cl_flags &= ~CLF_UNLINK_HSM_EXISTS; rc = mdd_changelog_ns_store(env, mdd, is_dir ? CL_RMDIR : CL_UNLINK, cl_flags, @@ -1700,7 +1756,7 @@ cleanup: } stop: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); return rc; } @@ -1720,9 +1776,11 @@ static int mdd_cd_sanity_check(const struct lu_env *env, RETURN(0); } -static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, - struct md_object *cobj, const struct md_op_spec *spec, - struct md_attr *ma) +static int mdd_create_data(const struct lu_env *env, + struct md_object *pobj, + struct md_object *cobj, + const struct md_op_spec *spec, + struct md_attr *ma) { struct mdd_device *mdd = mdo2mdd(cobj); struct mdd_object *mdd_pobj = md2mdd_obj(pobj); @@ -1747,21 +1805,21 @@ 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); /* calling ->ah_make_hint() is used to transfer information from parent */ mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint); - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - GOTO(out_free, rc = PTR_ERR(handle)); + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + GOTO(out_free, rc = PTR_ERR(handle)); - /* - * XXX: Setting the lov ea is not locked but setting the attr is locked? - * Should this be fixed? - */ + /* + * XXX: Setting the lov ea is not locked but setting the attr is locked? + * Should this be fixed? + */ CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n", spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen, spec->sp_cr_flags, spec->no_create); @@ -1787,7 +1845,7 @@ 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); @@ -1795,7 +1853,8 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); + out_free: RETURN(rc); } @@ -1803,24 +1862,15 @@ out_free: static int mdd_declare_object_initialize(const struct lu_env *env, struct mdd_object *parent, struct mdd_object *child, - struct lu_attr *attr, + const struct lu_attr *attr, struct thandle *handle) { int rc; ENTRY; - /* - * inode mode has been set in creation time, and it's based on umask, - * la_mode and acl, don't set here again! (which will go wrong - * because below function doesn't consider umask). - * I'd suggest set all object attributes in creation time, see above. - */ LASSERT(attr->la_valid & (LA_MODE | LA_TYPE)); - 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)) - RETURN(rc); + if (!S_ISDIR(attr->la_mode)) + RETURN(0); rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR, dot, handle); @@ -1850,11 +1900,10 @@ static int mdd_object_initialize(const struct lu_env *env, /* Add "." and ".." for newly created dir */ mdo_ref_add(env, child, handle); rc = __mdd_index_insert_only(env, child, mdo2fid(child), - S_IFDIR, dot, handle, BYPASS_CAPA); + S_IFDIR, dot, handle); if (rc == 0) rc = __mdd_index_insert_only(env, child, pfid, S_IFDIR, - dotdot, handle, - BYPASS_CAPA); + dotdot, handle); if (rc != 0) mdo_ref_del(env, child, handle); } @@ -1894,9 +1943,9 @@ static int mdd_create_sanity_check(const struct lu_env *env, int rc; ENTRY; - /* EEXIST check */ - if (mdd_is_dead_obj(obj)) - RETURN(-ENOENT); + /* EEXIST check */ + if (mdd_is_dead_obj(obj)) + RETURN(-ENOENT); /* * In some cases this lookup is not needed - we know before if name @@ -1936,14 +1985,14 @@ static int mdd_create_sanity_check(const struct lu_env *env, RETURN(rc); switch (cattr->la_mode & S_IFMT) { - case S_IFLNK: { - unsigned int symlen = strlen(spec->u.sp_symname) + 1; + case S_IFLNK: { + unsigned int symlen = strlen(spec->u.sp_symname) + 1; - if (symlen > (1 << m->mdd_dt_conf.ddp_block_shift)) - RETURN(-ENAMETOOLONG); - else - RETURN(0); - } + if (symlen > m->mdd_dt_conf.ddp_symlink_max) + RETURN(-ENAMETOOLONG); + else + RETURN(0); + } case S_IFDIR: case S_IFREG: case S_IFCHR: @@ -1969,6 +2018,7 @@ static int mdd_declare_object_create(const struct lu_env *env, struct lu_buf *acl_buf, struct dt_allocation_hint *hint) { + const struct lu_buf *buf; int rc; rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec, @@ -2004,8 +2054,6 @@ static int mdd_declare_object_create(const struct lu_env *env, /* replay case, create LOV EA from client data */ if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA && S_ISREG(attr->la_mode))) { - const struct lu_buf *buf; - buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen); rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0, @@ -2025,6 +2073,16 @@ static int mdd_declare_object_create(const struct lu_env *env, if (rc) GOTO(out, rc); } + + if (spec->sp_cr_file_secctx_name != NULL) { + buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx, + spec->sp_cr_file_secctx_size); + rc = mdo_declare_xattr_set(env, c, buf, + spec->sp_cr_file_secctx_name, 0, + handle); + if (rc < 0) + GOTO(out, rc); + } out: return rc; } @@ -2079,13 +2137,6 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, 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 +2156,7 @@ 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 */ @@ -2142,6 +2193,7 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, struct dt_allocation_hint *hint, struct thandle *handle) { + const struct lu_buf *buf; int rc; mdd_write_lock(env, son, MOR_TGT_CHILD); @@ -2177,14 +2229,12 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, if (spec->no_create || (S_ISREG(attr->la_mode) && spec->sp_cr_flags & MDS_OPEN_HAS_EA) || S_ISDIR(attr->la_mode)) { - const struct lu_buf *buf; - buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen); 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); } @@ -2195,7 +2245,7 @@ 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); } @@ -2203,7 +2253,7 @@ 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); } @@ -2219,7 +2269,6 @@ 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); @@ -2229,6 +2278,15 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, GOTO(err_initlized, rc = -EFAULT); } + if (spec->sp_cr_file_secctx_name != NULL) { + buf = mdd_buf_get_const(env, spec->sp_cr_file_secctx, + spec->sp_cr_file_secctx_size); + rc = mdo_xattr_set(env, son, buf, spec->sp_cr_file_secctx_name, + 0, handle); + if (rc < 0) + GOTO(err_initlized, rc); + } + err_initlized: if (unlikely(rc != 0)) { int rc2; @@ -2250,6 +2308,48 @@ unlock: RETURN(rc); } +static int mdd_index_delete(const struct lu_env *env, + struct mdd_object *mdd_pobj, + struct lu_attr *cattr, + const struct lu_name *lname) +{ + struct mdd_device *mdd = mdo2mdd(&mdd_pobj->mod_obj); + struct thandle *handle; + int rc; + ENTRY; + + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + RETURN(PTR_ERR(handle)); + + rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, + handle); + if (rc != 0) + GOTO(stop, rc); + + if (S_ISDIR(cattr->la_mode)) { + rc = mdo_declare_ref_del(env, mdd_pobj, handle); + if (rc != 0) + GOTO(stop, rc); + } + + /* Since this will only be used in the error handler path, + * Let's set the thandle to be local and not mess the transno */ + handle->th_local = 1; + rc = mdd_trans_start(env, mdd, handle); + if (rc) + GOTO(stop, rc); + + rc = __mdd_index_delete(env, mdd_pobj, lname->ln_name, + S_ISDIR(cattr->la_mode), handle); + if (rc) + GOTO(stop, rc); +stop: + rc = mdd_trans_stop(env, mdd, rc, handle); + + RETURN(rc); +} + /* * Create object and insert it into namespace. */ @@ -2310,7 +2410,7 @@ 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); @@ -2370,8 +2470,7 @@ 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), - attr->la_mode, name, handle, - mdd_object_capa(env, mdd_pobj)); + attr->la_mode, name, handle); if (rc != 0) GOTO(err_created, rc); @@ -2389,14 +2488,12 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, EXIT; err_insert: if (rc != 0) { - int rc2; - if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) rc2 = __mdd_orphan_del(env, son, handle); else rc2 = __mdd_index_delete(env, mdd_pobj, name, S_ISDIR(attr->la_mode), - handle, BYPASS_CAPA); + handle); if (rc2 != 0) goto out_stop; @@ -2438,10 +2535,18 @@ out_volatile: NULL, handle); out_stop: rc2 = mdd_trans_stop(env, mdd, rc, handle); - if (rc == 0) + if (rc == 0) { + /* If creation fails, it is most likely due to the remote update + * failure, because local transaction will mostly succeed at + * this stage. There is no easy way to rollback all of previous + * updates, so let's remove the object from namespace, and + * LFSCK should handle the orphan object. */ + if (rc2 < 0 && !mdd_object_remote(mdd_pobj)) + mdd_index_delete(env, mdd_pobj, attr, lname); rc = rc2; + } out_free: - if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + if (is_vmalloc_addr(ldata->ld_buf)) /* if we vmalloced a large buffer drop it */ lu_buf_free(ldata->ld_buf); @@ -2542,8 +2647,8 @@ static int mdd_declare_rename(const struct lu_env *env, struct mdd_object *mdd_tpobj, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, - const struct lu_name *tname, const struct lu_name *sname, + const struct lu_name *tname, struct md_attr *ma, struct linkea_data *ldata, struct thandle *handle) @@ -2616,15 +2721,13 @@ static int mdd_declare_rename(const struct lu_env *env, 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 - * to balance declarations vs calls to change ... */ - rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, handle); - if (rc) - return rc; - if (mdd_tobj && mdd_object_exists(mdd_tobj)) { /* delete target child in target parent directory */ + rc = mdo_declare_index_delete(env, mdd_tpobj, tname->ln_name, + handle); + if (rc) + return rc; + rc = mdo_declare_ref_del(env, mdd_tobj, handle); if (rc) return rc; @@ -2693,22 +2796,29 @@ static int mdd_rename(const struct lu_env *env, mdd_tobj = md2mdd_obj(tobj); mdd_sobj = mdd_object_find(env, mdd, lf); + if (IS_ERR(mdd_sobj)) + RETURN(PTR_ERR(mdd_sobj)); - 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); + /* search for an existing archive. + * we should check ahead as the object + * can be destroyed in this transaction */ + if (mdd_hsm_archive_exists(env, mdd_tobj, ma)) + cl_flags |= CLF_RENAME_LAST_EXISTS; } - rc = mdd_la_get(env, mdd_tpobj, tpattr, BYPASS_CAPA); + rc = mdd_la_get(env, mdd_tpobj, tpattr); if (rc) GOTO(out_pending, rc); @@ -2721,13 +2831,18 @@ static int mdd_rename(const struct lu_env *env, if (rc < 0) GOTO(out_pending, rc); + /* FIXME: Should consider tobj and sobj too in rename_lock. */ + rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj); + if (rc < 0) + GOTO(out_pending, rc); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_pending, rc = PTR_ERR(handle)); memset(ldata, 0, sizeof(*ldata)); - mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj), - ltname, 1, 0, ldata); + mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), lsname, + mdd_object_fid(mdd_tpobj), ltname, 1, 0, ldata); rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj, mdd_tobj, lsname, ltname, ma, ldata, handle); if (rc) @@ -2737,64 +2852,60 @@ static int mdd_rename(const struct lu_env *env, if (rc) GOTO(stop, rc); - /* FIXME: Should consider tobj and sobj too in rename_lock. */ - rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj); - if (rc < 0) - GOTO(cleanup_unlocked, rc); - 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)); - if (rc) - GOTO(cleanup, rc); + /* Remove source name from source directory */ + rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle); + if (rc != 0) + GOTO(stop, 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)); + /* "mv dir1 dir2" needs "dir1/.." link update */ + if (is_dir && !lu_fid_eq(spobj_fid, tpobj_fid)) { + 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, S_IFDIR, - dotdot, handle, - mdd_object_capa(env, mdd_sobj)); + dotdot, handle); if (rc != 0) GOTO(fixup_spobj, rc); } - /* Remove target name from target directory - * 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)); - if (rc != 0) { - if (mdd_tobj) { - /* tname might been renamed to something else */ - GOTO(fixup_spobj, rc); - } - if (rc != -ENOENT) - GOTO(fixup_spobj, rc); - } + if (mdd_tobj != NULL && mdd_object_exists(mdd_tobj)) { + rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle); + if (rc != 0) + /* tname might been renamed to something else */ + GOTO(fixup_spobj, rc); + } /* Insert new fid with target name into target dir */ rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode, - tname, handle, mdd_object_capa(env, mdd_tpobj)); + tname, handle); if (rc != 0) GOTO(fixup_tpobj, rc); LASSERT(ma->ma_attr.la_valid & LA_CTIME); la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; - /* XXX: mdd_sobj must be local one if it is NOT NULL. */ - if (mdd_sobj) { - la->la_valid = LA_CTIME; - rc = mdd_update_time(env, mdd_sobj, cattr, la, handle); - if (rc) - GOTO(fixup_tpobj, rc); - } + /* XXX: mdd_sobj must be local one if it is NOT NULL. */ + la->la_valid = LA_CTIME; + rc = mdd_update_time(env, mdd_sobj, cattr, la, handle); + if (rc) + GOTO(fixup_tpobj, rc); + + /* Update the linkEA for the source object */ + mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD); + rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname, + mdo2fid(mdd_tpobj), ltname, handle, ldata, + 0, 0); + if (rc == -ENOENT) + /* Old files might not have EA entry */ + mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj), + lsname, handle, NULL, 0); + mdd_write_unlock(env, mdd_sobj); + /* We don't fail the transaction if the link ea can't be + updated -- fid2path will use alternate lookup method. */ + rc = 0; /* Remove old target object * For tobj is remote case cmm layer has processed @@ -2818,7 +2929,7 @@ 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", @@ -2851,8 +2962,13 @@ static int mdd_rename(const struct lu_env *env, } /* fetch updated nlink */ - rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); - if (rc != 0) { + rc = mdd_la_get(env, mdd_tobj, tattr); + if (rc == -ENOENT) { + /* the object got removed, let's + * return the latest known attributes */ + tattr->la_nlink = 0; + rc = 0; + } else if (rc != 0) { CERROR("%s: Failed to get nlink for tobj " DFID": rc = %d\n", mdd2obd_dev(mdd)->obd_name, @@ -2863,11 +2979,10 @@ static int mdd_rename(const struct lu_env *env, ma->ma_attr = *tattr; ma->ma_valid |= MA_INODE; - if (tattr->la_nlink == 0) { + if (tattr->la_nlink == 0) cl_flags |= CLF_RENAME_LAST; - if (mdd_hsm_archive_exists(env, mdd_tobj, ma)) - cl_flags |= CLF_RENAME_LAST_EXISTS; - } + else + cl_flags &= ~CLF_RENAME_LAST_EXISTS; } la->la_valid = LA_CTIME | LA_MTIME; @@ -2878,29 +2993,15 @@ static int mdd_rename(const struct lu_env *env, if (mdd_spobj != mdd_tpobj) { la->la_valid = LA_CTIME | LA_MTIME; rc = mdd_update_time(env, mdd_tpobj, tpattr, la, handle); + if (rc != 0) + GOTO(fixup_tpobj, rc); } - if (rc == 0 && mdd_sobj) { - mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD); - rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname, - mdo2fid(mdd_tpobj), ltname, handle, NULL, - 0, 0); - if (rc == -ENOENT) - /* Old files might not have EA entry */ - mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj), - lsname, handle, NULL, 0); - mdd_write_unlock(env, mdd_sobj); - /* We don't fail the transaction if the link ea can't be - updated -- fid2path will use alternate lookup method. */ - rc = 0; - } - EXIT; 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); @@ -2915,7 +3016,7 @@ fixup_tpobj: rc2 = __mdd_index_insert(env, mdd_tpobj, mdo2fid(mdd_tobj), mdd_object_type(mdd_tobj), - tname, handle, BYPASS_CAPA); + tname, handle); if (rc2 != 0) CWARN("tp obj fix error: rc = %d\n", rc2); } @@ -2923,16 +3024,14 @@ fixup_tpobj: 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, S_IFDIR, - dotdot, handle, BYPASS_CAPA); + dotdot, handle); if (rc2 != 0) CWARN("%s: sp obj dotdot insert error: rc = %d\n", mdd2obd_dev(mdd)->obd_name, rc2); @@ -2942,7 +3041,7 @@ fixup_spobj2: if (rc != 0) { rc2 = __mdd_index_insert(env, mdd_spobj, lf, mdd_object_type(mdd_sobj), sname, - handle, BYPASS_CAPA); + handle); if (rc2 != 0) CWARN("sp obj fix error: rc = %d\n", rc2); } @@ -2951,14 +3050,13 @@ cleanup: if (tobj_locked) mdd_write_unlock(env, mdd_tobj); -cleanup_unlocked: 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); + rc = mdd_trans_stop(env, mdd, rc, handle); out_pending: mdd_object_put(env, mdd_sobj); @@ -2971,6 +3069,7 @@ out_pending: **/ static int mdd_linkea_update_child_internal(const struct lu_env *env, struct mdd_object *parent, + struct mdd_object *newparent, struct mdd_object *child, const char *name, int namelen, struct thandle *handle, @@ -3007,7 +3106,7 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, &lname, &fid); if (strncmp(lname.ln_name, name, namelen) != 0 || - lu_fid_eq(&fid, mdd_object_fid(parent))) { + !lu_fid_eq(&fid, mdd_object_fid(parent))) { ldata.ld_lee = (struct link_ea_entry *) ((char *)ldata.ld_lee + ldata.ld_reclen); @@ -3017,10 +3116,10 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, CDEBUG(D_INFO, "%s: update "DFID" with %.*s:"DFID"\n", mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(child)), lname.ln_namelen, lname.ln_name, - PFID(mdd_object_fid(parent))); + PFID(mdd_object_fid(newparent))); /* update to the new parent fid */ linkea_entry_pack(ldata.ld_lee, &lname, - mdd_object_fid(parent)); + mdd_object_fid(newparent)); if (declare) rc = mdd_declare_links_add(env, child, handle, &ldata, MLAO_IGNORE); @@ -3033,21 +3132,25 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, static int mdd_linkea_declare_update_child(const struct lu_env *env, struct mdd_object *parent, + struct mdd_object *newparent, struct mdd_object *child, const char *name, int namelen, struct thandle *handle) { - return mdd_linkea_update_child_internal(env, parent, child, name, + return mdd_linkea_update_child_internal(env, parent, newparent, + child, name, namelen, handle, true); } static int mdd_linkea_update_child(const struct lu_env *env, struct mdd_object *parent, + struct mdd_object *newparent, struct mdd_object *child, const char *name, int namelen, struct thandle *handle) { - return mdd_linkea_update_child_internal(env, parent, child, name, + return mdd_linkea_update_child_internal(env, parent, newparent, + child, name, namelen, handle, false); } @@ -3056,22 +3159,18 @@ static int mdd_update_linkea_internal(const struct lu_env *env, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, const struct lu_name *child_name, + struct linkea_data *ldata, struct thandle *handle, int declare) { struct mdd_thread_info *info = mdd_env_info(env); - struct linkea_data *ldata = &info->mti_link_data; int count; int rc = 0; ENTRY; - rc = mdd_links_read(env, mdd_sobj, ldata); - if (rc != 0) { - if (rc == -ENOENT || rc == -ENODATA) - rc = 0; - RETURN(rc); - } + LASSERT(ldata->ld_buf != NULL); +again: /* If it is mulitple links file, we need update the name entry for * all parent */ LASSERT(ldata->ld_leh != NULL); @@ -3084,20 +3183,21 @@ static int mdd_update_linkea_internal(const struct lu_env *env, linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname, &fid); - ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee + - ldata->ld_reclen); pobj = mdd_object_find(env, mdd, &fid); if (IS_ERR(pobj)) { CWARN("%s: cannot find obj "DFID": rc = %ld\n", mdd2obd_dev(mdd)->obd_name, PFID(&fid), PTR_ERR(pobj)); - continue; + linkea_del_buf(ldata, &lname); + goto again; } if (!mdd_object_exists(pobj)) { CDEBUG(D_INFO, "%s: obj "DFID" does not exist\n", mdd2obd_dev(mdd)->obd_name, PFID(&fid)); - GOTO(next_put, rc); + linkea_del_buf(ldata, &lname); + mdd_object_put(env, pobj); + goto again; } if (pobj == mdd_pobj && @@ -3107,7 +3207,9 @@ static int mdd_update_linkea_internal(const struct lu_env *env, CDEBUG(D_INFO, "%s: skip its own %s: "DFID"\n", mdd2obd_dev(mdd)->obd_name, child_name->ln_name, PFID(&fid)); - GOTO(next_put, rc); + linkea_del_buf(ldata, &lname); + mdd_object_put(env, pobj); + goto again; } CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n", @@ -3137,17 +3239,45 @@ static int mdd_update_linkea_internal(const struct lu_env *env, if (rc) GOTO(next_put, rc); } else { - rc = __mdd_index_delete(env, pobj, lname.ln_name, - 0, handle, - mdd_object_capa(env, pobj)); - if (rc) + char *tmp_name = info->mti_key; + + if (lname.ln_namelen >= sizeof(info->mti_key)) { + /* lnamelen is too big(> NAME_MAX + 16), + * something wrong about this linkea, let's + * skip it */ + linkea_del_buf(ldata, &lname); + mdd_object_put(env, pobj); + goto again; + } + + /* Note: lname might be without \0 at the end, see + * linkea_entry_unpack(), let's add extra \0 by + * snprintf */ + snprintf(tmp_name, sizeof(info->mti_key), "%.*s", + lname.ln_namelen, lname.ln_name); + lname.ln_name = tmp_name; + + /* Let's check if this linkEA still valid, before + * it might be packed into the RPC buffer. */ + rc = mdd_lookup(env, &pobj->mod_obj, &lname, + &info->mti_fid, NULL); + if (rc < 0 || + !lu_fid_eq(&info->mti_fid, + mdd_object_fid(mdd_sobj))) { + /* skip invalid linkea entry */ + linkea_del_buf(ldata, &lname); + mdd_object_put(env, pobj); + goto again; + } + + rc = __mdd_index_delete(env, pobj, tmp_name, 0, handle); + if (rc != 0) GOTO(next_put, rc); rc = __mdd_index_insert(env, pobj, mdd_object_fid(mdd_tobj), mdd_object_type(mdd_tobj), - lname.ln_name, handle, - mdd_object_capa(env, pobj)); + tmp_name, handle); if (rc != 0) GOTO(next_put, rc); @@ -3165,6 +3295,9 @@ next_put: mdd_object_put(env, pobj); if (rc != 0) break; + + ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee + + ldata->ld_reclen); } RETURN(rc); @@ -3187,8 +3320,7 @@ 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; @@ -3201,8 +3333,7 @@ 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; @@ -3220,9 +3351,7 @@ 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) @@ -3234,8 +3363,7 @@ 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) @@ -3257,15 +3385,14 @@ 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); if (rc != 0) GOTO(stop_trans, rc); stop_trans: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); if (rc != 0) GOTO(out, rc); next: @@ -3311,7 +3438,7 @@ 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)) { + } else if (S_ISDIR(la->la_mode) && ldata != NULL) { rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata, MLAO_IGNORE); if (rc != 0) @@ -3383,7 +3510,7 @@ static int mdd_migrate_create(const struct lu_env *env, RETURN(rc); } spec->u.sp_symname = link_buf.lb_buf; - } else if S_ISREG(la->la_mode) { + } else if (S_ISREG(la->la_mode)) { /* retrieve lov of the old object */ rc = mdd_get_lov_ea(env, mdd_sobj, &lmm_buf); if (rc != 0 && rc != -ENODATA) @@ -3395,7 +3522,12 @@ static int mdd_migrate_create(const struct lu_env *env, } } else if (S_ISDIR(la->la_mode)) { rc = mdd_links_read(env, mdd_sobj, ldata); - if (rc < 0 && rc != -ENODATA) + if (rc == -ENODATA) { + /* ignore the non-linkEA error */ + ldata = NULL; + rc = 0; + } + if (rc < 0) RETURN(rc); } @@ -3429,13 +3561,16 @@ static int mdd_migrate_create(const struct lu_env *env, if (rc != 0) GOTO(stop_trans, rc); + /* don't set nlink from the original object */ + la->la_valid &= ~LA_NLINK; + /* create the target object */ rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL, hint, handle); if (rc != 0) GOTO(stop_trans, rc); - if (S_ISDIR(la->la_mode)) { + if (S_ISDIR(la->la_mode) && ldata != NULL) { rc = mdd_links_write(env, mdd_tobj, ldata, handle); if (rc != 0) GOTO(stop_trans, rc); @@ -3446,8 +3581,7 @@ static int mdd_migrate_create(const struct lu_env *env, * 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); @@ -3458,11 +3592,10 @@ 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; - 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); + rc = mdd_trans_stop(env, mdd, rc, handle); out_free: if (lmm_buf.lb_buf != NULL) OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len); @@ -3479,9 +3612,9 @@ static int mdd_migrate_entries(const struct lu_env *env, struct thandle *handle; struct dt_it *it; const struct dt_it_ops *iops; - int rc; int result; struct lu_dirent *ent; + int rc; ENTRY; OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1); @@ -3494,8 +3627,7 @@ 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)); @@ -3545,6 +3677,7 @@ static int mdd_migrate_entries(const struct lu_env *env, if (IS_ERR(child)) GOTO(out, rc = PTR_ERR(child)); + mdd_write_lock(env, child, MOR_SRC_CHILD); is_dir = S_ISDIR(mdd_object_type(child)); snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name); @@ -3554,15 +3687,14 @@ static int mdd_migrate_entries(const struct lu_env *env, struct lu_fid *fid = &mdd_env_info(env)->mti_fid2; rc = dt_lookup(env, dt_tobj, (struct dt_rec *)fid, - (struct dt_key *)name, - mdd_object_capa(env, mdd_tobj)); + (struct dt_key *)name); if (unlikely(rc == 0)) target_exist = true; } handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) - GOTO(out, rc = PTR_ERR(handle)); + GOTO(out_put, rc = PTR_ERR(handle)); /* Note: this transaction is part of migration, and it is not * the last step of migration, so we set th_local = 1 to avoid @@ -3605,7 +3737,7 @@ static int mdd_migrate_entries(const struct lu_env *env, GOTO(out_put, rc); } - rc = mdd_linkea_declare_update_child(env, mdd_tobj, + rc = mdd_linkea_declare_update_child(env, mdd_sobj,mdd_tobj, child, name, strlen(name), handle); @@ -3622,44 +3754,36 @@ static int mdd_migrate_entries(const struct lu_env *env, if (likely(!target_exist)) { rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid, mdd_object_type(child), - name, handle, - mdd_object_capa(env, mdd_tobj)); + 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), S_IFDIR, - dotdot, handle, - mdd_object_capa(env, child)); + dotdot, handle); if (rc != 0) GOTO(out_put, rc); } - rc = mdd_linkea_update_child(env, mdd_tobj, child, name, + rc = mdd_linkea_update_child(env, mdd_sobj, mdd_tobj, + child, name, strlen(name), handle); out_put: + mdd_write_unlock(env, child); mdd_object_put(env, child); - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); if (rc != 0) GOTO(out, rc); next: @@ -3682,22 +3806,24 @@ static int mdd_declare_update_linkea(const struct lu_env *env, struct mdd_object *mdd_pobj, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, - struct thandle *handle, - const struct lu_name *child_name) + const struct lu_name *child_name, + struct linkea_data *ldata, + struct thandle *handle) { return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj, - child_name, handle, 1); + child_name, ldata, handle, 1); } static int mdd_update_linkea(const struct lu_env *env, struct mdd_object *mdd_pobj, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, - struct thandle *handle, - const struct lu_name *child_name) + const struct lu_name *child_name, + struct linkea_data *ldata, + struct thandle *handle) { return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj, - child_name, handle, 0); + child_name, ldata, handle, 0); } static int mdd_declare_migrate_update_name(const struct lu_env *env, @@ -3707,10 +3833,12 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, const struct lu_name *lname, struct lu_attr *la, struct lu_attr *parent_la, + struct linkea_data *ldata, struct thandle *handle) { - struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr); - int rc; + struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj); + struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr); + int rc; /* Revert IMMUTABLE flag */ la_flag->la_valid = LA_FLAGS; @@ -3724,16 +3852,25 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, if (rc != 0) return rc; - rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj, - mdd_tobj, handle, lname); - if (rc != 0) - return rc; + if (ldata->ld_buf != NULL) { + rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj, + mdd_tobj, lname, ldata, handle); + if (rc != 0) + return rc; + } if (S_ISREG(mdd_object_type(mdd_sobj))) { rc = mdo_declare_xattr_del(env, mdd_sobj, XATTR_NAME_LOV, handle); if (rc != 0) return rc; + + handle->th_complex = 1; + rc = mdo_declare_xattr_set(env, mdd_tobj, NULL, + XATTR_NAME_FID, + LU_XATTR_REPLACE, handle); + if (rc < 0) + return rc; } if (S_ISDIR(mdd_object_type(mdd_sobj))) { @@ -3749,6 +3886,10 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, 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) @@ -3776,6 +3917,10 @@ static int mdd_declare_migrate_update_name(const struct lu_env *env, return rc; rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle); + if (rc != 0) + return rc; + + rc = mdd_declare_changelog_store(env, mdd, lname, NULL, handle); return rc; } @@ -3791,6 +3936,7 @@ static int mdd_migrate_update_name(const struct lu_env *env, struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr); struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr); struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj); + struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; struct thandle *handle; int is_dir = S_ISDIR(mdd_object_type(mdd_sobj)); const char *name = lname->ln_name; @@ -3802,16 +3948,22 @@ 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); + ldata->ld_buf = NULL; + rc = mdd_links_read(env, mdd_sobj, ldata); + if (rc != 0 && rc != -ENOENT && rc != -ENODATA) + RETURN(rc); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) RETURN(PTR_ERR(handle)); rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, - lname, so_attr, p_la, handle); + lname, so_attr, p_la, ldata, + handle); if (rc != 0) { /* If the migration can not be fit in one transaction, just * leave it in the original MDT */ @@ -3833,74 +3985,122 @@ 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; - 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); - rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj, - handle, lname); - if (rc != 0) - GOTO(stop_trans, rc); + if (ldata->ld_buf != NULL) { + rc = mdd_update_linkea(env, mdd_pobj, mdd_sobj, mdd_tobj, + lname, ldata, handle); + if (rc != 0) + GOTO(stop_trans, rc); + + /* linkea update might decrease the source object + * nlink, let's get the attr again after ref_del */ + rc = mdd_la_get(env, mdd_sobj, so_attr); + if (rc != 0) + GOTO(stop_trans, rc); + } 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); + + rc = mdo_xattr_set(env, mdd_tobj, NULL, + XATTR_NAME_FID, + LU_XATTR_REPLACE, handle); + if (rc < 0) + 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), - mdd_object_type(mdd_tobj), name, - handle, mdd_object_capa(env, mdd_pobj)); + mdd_object_type(mdd_tobj), name, handle); if (rc != 0) GOTO(stop_trans, rc); + linkea_add_buf(ldata, lname, mdd_object_fid(mdd_pobj)); rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle, - NULL, 1); + ldata, 1); if (rc != 0) GOTO(stop_trans, rc); - mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD); + mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD); + + mdd_sobj->mod_flags |= DEAD_OBJ; + rc = mdd_mark_orphan_object(env, mdd_sobj, handle, false); + if (rc != 0) + GOTO(out_unlock, rc); + + rc = __mdd_orphan_add(env, mdd_sobj, handle); + if (rc != 0) + GOTO(out_unlock, rc); + mdo_ref_del(env, mdd_sobj, handle); if (is_dir) 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(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(out_unlock, rc); rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0); if (rc != 0) GOTO(out_unlock, rc); + rc = mdd_changelog_ns_store(env, mdd, CL_MIGRATE, 0, mdd_tobj, + mdo2fid(mdd_pobj), mdo2fid(mdd_sobj), + mdo2fid(mdd_pobj), lname, lname, handle); + if (rc != 0) { + CWARN("%s: changelog for migrate %s "DFID + "under "DFID" failed: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, lname->ln_name, + PFID(mdd_object_fid(mdd_sobj)), + PFID(mdd_object_fid(mdd_pobj)), rc); + /* Sigh, there are no easy way to migrate back the object, so + * let's reset the result to 0 for now XXX */ + rc = 0; + } out_unlock: mdd_write_unlock(env, mdd_sobj); stop_trans: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); RETURN(rc); } +static int mdd_fld_lookup(const struct lu_env *env, struct mdd_device *mdd, + const struct lu_fid *fid, __u32 *mdt_index) +{ + struct lu_seq_range *range = &mdd_env_info(env)->mti_range; + struct seq_server_site *ss; + int rc; + + ss = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site; + + range->lsr_flags = LU_SEQ_RANGE_MDT; + rc = fld_server_lookup(env, ss->ss_server_fld, fid->f_seq, range); + if (rc != 0) + return rc; + + *mdt_index = range->lsr_index; + + return 0; +} /** * Check whether we should migrate the file/dir * return val @@ -3917,11 +4117,12 @@ static int mdd_migrate_sanity_check(const struct lu_env *env, { struct mdd_thread_info *info = mdd_env_info(env); struct linkea_data *ldata = &info->mti_link_data; + struct mdd_device *mdd = mdo2mdd(&pobj->mod_obj); int mgr_easize; struct lu_buf *mgr_buf; int count; int rc; - + __u64 mdt_index; ENTRY; mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1); @@ -3929,8 +4130,7 @@ 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; @@ -3964,46 +4164,40 @@ static int mdd_migrate_sanity_check(const struct lu_env *env, if (rc != 0) { /* For multiple links files, if there are no linkEA data at all, * means the file might be created before linkEA is enabled, and - * all all of its links should not be migrated yet, otherwise - * it should have some linkEA there */ + * all of its links should not be migrated yet, otherwise it + * should have some linkEA there */ if (rc == -ENOENT || rc == -ENODATA) RETURN(1); RETURN(rc); } - /* If it is mulitple links file, we need update the name entry for - * all parent */ + mdt_index = mdd->mdd_md_dev.md_lu_dev.ld_site->ld_seq_site->ss_node_id; + /* If there are still links locally, then the file will not be + * migrated. */ LASSERT(ldata->ld_leh != NULL); ldata->ld_lee = (struct link_ea_entry *)(ldata->ld_leh + 1); for (count = 0; count < ldata->ld_leh->leh_reccount; count++) { - struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj); - struct mdd_object *lpobj; struct lu_name lname; struct lu_fid fid; + __u32 parent_mdt_index; linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, &lname, &fid); ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee + ldata->ld_reclen); - lpobj = mdd_object_find(env, mdd, &fid); - if (IS_ERR(lpobj)) { - CWARN("%s: cannot find obj "DFID": rc = %ld\n", - mdd2obd_dev(mdd)->obd_name, PFID(&fid), - PTR_ERR(lpobj)); - continue; - } - if (!mdd_object_exists(lpobj) || mdd_object_remote(lpobj)) { - CDEBUG(D_INFO, DFID"%.*s: is on remote MDT.\n", - PFID(&fid), lname.ln_namelen, lname.ln_name); - mdd_object_put(env, lpobj); + rc = mdd_fld_lookup(env, mdd, &fid, &parent_mdt_index); + if (rc != 0) + RETURN(rc); + + /* Migrate the object only if none of its parents are on the + * current MDT. */ + if (parent_mdt_index != mdt_index) continue; - } CDEBUG(D_INFO, DFID"still has local entry %.*s "DFID"\n", PFID(mdd_object_fid(sobj)), lname.ln_namelen, lname.ln_name, PFID(&fid)); - mdd_object_put(env, lpobj); rc = 1; break; } @@ -4018,17 +4212,19 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, struct mdd_object *mdd_pobj = md2mdd_obj(pobj); struct mdd_device *mdd = mdo2mdd(pobj); struct mdd_object *mdd_sobj = md2mdd_obj(sobj); - struct mdd_object *mdd_tobj = NULL; + struct mdd_object *mdd_tobj = md2mdd_obj(tobj); struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr); struct lu_attr *pattr = MDD_ENV_VAR(env, pattr); + bool created = false; int rc; ENTRY; /* If the file will being migrated, it will check whether * the file is being opened by someone else right now */ mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD); - if (mdd_sobj->mod_count >= 1) { - CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n", + if (mdd_sobj->mod_count > 0) { + CDEBUG(D_OTHER, + "%s: "DFID"%s is already opened count %d: rc = %d\n", mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_sobj)), lname->ln_name, mdd_sobj->mod_count, -EBUSY); @@ -4037,11 +4233,11 @@ 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); @@ -4074,14 +4270,15 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, /* step 1: Check whether the orphan object has been created, and create * orphan object on the remote MDT if needed */ - mdd_tobj = md2mdd_obj(tobj); if (!mdd_object_exists(mdd_tobj)) { rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, lname, so_attr); if (rc != 0) GOTO(put, rc); + created = true; } + LASSERT(mdd_object_exists(mdd_tobj)); /* step 2: migrate xattr */ rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj); if (rc != 0) @@ -4099,11 +4296,16 @@ static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, OBD_FAIL_TIMEOUT(OBD_FAIL_MIGRATE_DELAY, cfs_fail_val); } + LASSERT(mdd_object_exists(mdd_tobj)); /* step 4: update name entry to the new object */ rc = mdd_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, lname, ma); if (rc != 0) GOTO(put, rc); + + /* newly created target was not locked, don't cache its attributes */ + if (created) + mdd_invalidate(env, tobj); put: RETURN(rc); }