X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_dir.c;h=b37c7958d72f66e00aec4f9396f760e4cbfc784f;hp=4ec29463272b3a5289fba0b92a37c0bc78f67326;hb=37cef588d339191f4ed0b8f1f11822f0231dcd78;hpb=0463179c98b81aed4f3745d7440d70b36fc86e16 diff --git a/lustre/mdd/mdd_dir.c b/lustre/mdd/mdd_dir.c index 4ec2946..b37c795 100644 --- a/lustre/mdd/mdd_dir.c +++ b/lustre/mdd/mdd_dir.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * 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, 2016, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -53,10 +49,21 @@ static const char dot[] = "."; static const char dotdot[] = ".."; static struct lu_name lname_dotdot = { - (char *) dotdot, - sizeof(dotdot) - 1 + .ln_name = (char *) dotdot, + .ln_namelen = sizeof(dotdot) - 1, }; +static inline int +mdd_name_check(struct mdd_device *m, const struct lu_name *ln) +{ + if (!lu_name_is_valid(ln)) + return -EINVAL; + else if (ln->ln_namelen > m->mdd_dt_conf.ddp_max_name_len) + return -ENAMETOOLONG; + else + return 0; +} + /* Get FID from name and parent */ static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj, @@ -74,33 +81,23 @@ __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); } - /* The common filename length check. */ - if (unlikely(lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)) - RETURN(-ENAMETOOLONG); - rc = mdd_permission_internal_locked(env, mdd_obj, pattr, mask, MOR_TGT_PARENT); if (rc) 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); @@ -114,20 +111,137 @@ 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); } +/** Read the link EA into a temp buffer. + * Uses the mdd_thread_info::mti_big_buf since it is generally large. + * A pointer to the buffer is stored in \a ldata::ld_buf. + * + * \retval 0 or error + */ +static int __mdd_links_read(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata) +{ + int rc; + + if (!mdd_object_exists(mdd_obj)) + return -ENODATA; + + /* First try a small buf */ + LASSERT(env != NULL); + ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf, + PAGE_SIZE); + if (ldata->ld_buf->lb_buf == NULL) + return -ENOMEM; + + 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); + 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); + } + if (rc < 0) { + lu_buf_free(ldata->ld_buf); + ldata->ld_buf = NULL; + return rc; + } + + return linkea_init(ldata); +} + +static int mdd_links_read(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata) +{ + int rc; + + rc = __mdd_links_read(env, mdd_obj, ldata); + if (!rc) + rc = linkea_init(ldata); + + return rc; +} + +static int mdd_links_read_with_rec(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata) +{ + int rc; + + rc = __mdd_links_read(env, mdd_obj, ldata); + if (!rc) + rc = linkea_init_with_rec(ldata); + + 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_with_rec(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); } /* @@ -169,8 +283,7 @@ 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); @@ -180,18 +293,16 @@ 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; @@ -205,7 +316,7 @@ 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; @@ -223,7 +334,7 @@ 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); @@ -253,41 +364,55 @@ int mdd_is_subdir(const struct lu_env *env, struct md_object *mo, static int mdd_dir_is_empty(const struct lu_env *env, struct mdd_object *dir) { - struct dt_it *it; - struct dt_object *obj; - const struct dt_it_ops *iops; - int result; - ENTRY; + struct dt_it *it; + struct dt_object *obj; + const struct dt_it_ops *iops; + int result; + ENTRY; - obj = mdd_object_child(dir); - if (!dt_try_as_dir(env, obj)) - RETURN(-ENOTDIR); - - iops = &obj->do_index_ops->dio_it; - it = iops->init(env, obj, LUDA_64BITHASH, BYPASS_CAPA); - if (!IS_ERR(it)) { - result = iops->get(env, it, (const void *)""); - if (result > 0) { - int i; - for (result = 0, i = 0; result == 0 && i < 3; ++i) - result = iops->next(env, it); - if (result == 0) - result = -ENOTEMPTY; - else if (result == +1) - result = 0; - } else if (result == 0) - /* - * Huh? Index contains no zero key? - */ - result = -EIO; - - iops->put(env, it); - iops->fini(env, it); - } else - result = PTR_ERR(it); - RETURN(result); + obj = mdd_object_child(dir); + if (!dt_try_as_dir(env, obj)) + RETURN(-ENOTDIR); + + iops = &obj->do_index_ops->dio_it; + it = iops->init(env, obj, LUDA_64BITHASH); + if (!IS_ERR(it)) { + result = iops->get(env, it, (const struct dt_key *)""); + if (result > 0) { + int i; + for (result = 0, i = 0; result == 0 && i < 3; ++i) + result = iops->next(env, it); + if (result == 0) + result = -ENOTEMPTY; + else if (result == 1) + result = 0; + } else if (result == 0) + /* + * Huh? Index contains no zero key? + */ + result = -EIO; + + iops->put(env, it); + iops->fini(env, it); + } else + result = PTR_ERR(it); + RETURN(result); } +/** + * Determine if the target object can be hard linked, and right now it only + * checks if the link count reach the maximum limit. Note: for ldiskfs, the + * directory nlink count might exceed the maximum link count(see + * osd_object_ref_add), so it only check nlink for non-directories. + * + * \param[in] env thread environment + * \param[in] obj object being linked to + * \param[in] la attributes of \a obj + * + * \retval 0 if \a obj can be hard linked + * \retval negative error if \a obj is a directory or has too + * many links + */ static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj, const struct lu_attr *la) { @@ -296,25 +421,31 @@ static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj, LASSERT(la != NULL); - if (!S_ISDIR(la->la_mode)) - RETURN(0); - - /* - * Subdir count limitation can be broken through. - */ - if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink) + /* Subdir count limitation can be broken through + * (see osd_object_ref_add), so only check non-directory here. */ + if (!S_ISDIR(la->la_mode) && + la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink) RETURN(-EMLINK); - else - RETURN(0); + + RETURN(0); } -/* +/** * Check whether it may create the cobj under the pobj. - * cobj maybe NULL + * + * \param[in] env execution environment + * \param[in] pobj the parent directory + * \param[in] pattr the attribute of the parent directory + * \param[in] cobj the child to be created + * \param[in] check_perm if check WRITE|EXEC permission for parent + * + * \retval = 0 create the child under this dir is allowed + * \retval negative errno create the child under this dir is + * not allowed */ -int mdd_may_create(const struct lu_env *env, - struct mdd_object *pobj, const struct lu_attr *pattr, - struct mdd_object *cobj, int check_perm, int check_nlink) +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) { int rc = 0; ENTRY; @@ -329,9 +460,6 @@ int mdd_may_create(const struct lu_env *env, rc = mdd_permission_internal_locked(env, pobj, pattr, MAY_WRITE | MAY_EXEC, MOR_TGT_PARENT); - if (!rc && check_nlink) - rc = __mdd_may_link(env, pobj, pattr); - RETURN(rc); } @@ -347,17 +475,16 @@ 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); @@ -412,7 +539,7 @@ 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); @@ -449,11 +576,7 @@ 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 */ @@ -476,9 +599,20 @@ int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj, RETURN(rc); } -/* - * tgt maybe NULL - * has mdd_write_lock on src already, but not on tgt yet +/** + * Check whether it can create the link file(linked to @src_obj) under + * the target directory(@tgt_obj), and src_obj has been locked by + * mdd_write_lock. + * + * \param[in] env execution environment + * \param[in] tgt_obj the target directory + * \param[in] tattr attributes of target directory + * \param[in] lname the link name + * \param[in] src_obj source object for link + * \param[in] cattr attributes for source object + * + * \retval = 0 it is allowed to create the link file under tgt_obj + * \retval negative error not allowed to create the link file */ static int mdd_link_sanity_check(const struct lu_env *env, struct mdd_object *tgt_obj, @@ -498,21 +632,20 @@ static int mdd_link_sanity_check(const struct lu_env *env, RETURN(-ESTALE); /* Local ops, no lookup before link, check filename length here. */ - if (lname && (lname->ln_namelen > m->mdd_dt_conf.ddp_max_name_len)) - RETURN(-ENAMETOOLONG); + rc = mdd_name_check(m, lname); + 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); - if (tgt_obj) { - rc = mdd_may_create(env, tgt_obj, tattr, NULL, 1, 0); - if (rc) - RETURN(rc); - } + rc = mdd_may_create(env, tgt_obj, tattr, NULL, true); + if (rc != 0) + RETURN(rc); rc = __mdd_may_link(env, src_obj, cattr); @@ -520,42 +653,41 @@ 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; } @@ -564,30 +696,31 @@ 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); @@ -597,52 +730,37 @@ 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; - - rec = buf->lb_buf; - rec->cr_hdr.lrh_len = reclen; - rec->cr_hdr.lrh_type = CHANGELOG_REC; + if (uc != NULL && uc->uc_jobid[0] != '\0') + crf |= CLF_JOBID; - 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; @@ -650,9 +768,7 @@ 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; @@ -665,7 +781,13 @@ 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; @@ -675,7 +797,7 @@ 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, @@ -683,46 +805,12 @@ 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, 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 - */ -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(); @@ -738,72 +826,50 @@ int 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, 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. @@ -813,57 +879,65 @@ int mdd_changelog_ns_store(const struct lu_env *env, struct mdd_device *mdd, * \param spfid - source parent fid * \param tname - target name string * \param sname - source name string - * \param handle - transacion handle + * \param handle - transaction 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) +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) { - struct llog_changelog_ext_rec *rec; - struct lu_buf *buf; - int reclen; - int rc; + 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); @@ -872,10 +946,12 @@ 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)); + 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; } @@ -957,41 +1033,32 @@ static int mdd_linkea_prepare(const struct lu_env *env, struct linkea_data *ldata) { int rc = 0; - int rc2 = 0; ENTRY; if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) - return 0; + RETURN(0); LASSERT(oldpfid != NULL || newpfid != NULL); - if (mdd_obj->mod_flags & DEAD_OBJ) { - /* Prevent linkea to be updated which is NOT necessary. */ - ldata->ld_reclen = 0; - /* No more links, don't bother */ + if (mdd_obj->mod_flags & DEAD_OBJ) + /* Unnecessary to update linkEA for dead object. */ RETURN(0); - } if (oldpfid != NULL) { rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid); if (rc) { - if ((check == 1) || - (rc != -ENODATA && rc != -ENOENT)) + if ((check == 1) || (rc != -ENODATA && rc != -ENOENT)) RETURN(rc); + /* No changes done. */ rc = 0; } } /* If renaming, add the new record */ - if (newpfid != NULL) { - /* even if the add fails, we still delete the out-of-date - * old link */ - rc2 = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid, - first, check); - } - - rc = rc != 0 ? rc : rc2; + if (newpfid != NULL) + rc = __mdd_links_add(env, mdd_obj, ldata, newlname, newpfid, + first, check); RETURN(rc); } @@ -1006,7 +1073,6 @@ int mdd_links_rename(const struct lu_env *env, struct linkea_data *ldata, int first, int check) { - int rc2 = 0; int rc = 0; ENTRY; @@ -1014,42 +1080,37 @@ int mdd_links_rename(const struct lu_env *env, ldata = &mdd_env_info(env)->mti_link_data; memset(ldata, 0, sizeof(*ldata)); rc = mdd_linkea_prepare(env, mdd_obj, oldpfid, oldlname, - newpfid, newlname, first, check, - ldata); - if (rc != 0) + newpfid, newlname, first, check, ldata); + if (rc) GOTO(out, rc); } - if (ldata->ld_reclen != 0) + if (!(mdd_obj->mod_flags & DEAD_OBJ)) rc = mdd_links_write(env, mdd_obj, ldata, handle); - EXIT; + + GOTO(out, rc); + out: - if (rc == 0) - rc = rc2; - if (rc) { - int error = 1; - if (rc == -EOVERFLOW || rc == -ENOSPC) - error = 0; - if (oldpfid == NULL) - CDEBUG(error ? D_ERROR : D_OTHER, - "link_ea add '%.*s' failed %d "DFID"\n", - newlname->ln_namelen, newlname->ln_name, + if (rc != 0) { + if (newlname == NULL) + CERROR("link_ea add failed %d "DFID"\n", rc, PFID(mdd_object_fid(mdd_obj))); + else if (oldpfid == NULL) + CERROR("link_ea add '%.*s' failed %d "DFID"\n", + newlname->ln_namelen, newlname->ln_name, rc, + PFID(mdd_object_fid(mdd_obj))); else if (newpfid == NULL) - CDEBUG(error ? D_ERROR : D_OTHER, - "link_ea del '%.*s' failed %d "DFID"\n", - oldlname->ln_namelen, oldlname->ln_name, - rc, PFID(mdd_object_fid(mdd_obj))); + CERROR("link_ea del '%.*s' failed %d "DFID"\n", + oldlname->ln_namelen, oldlname->ln_name, rc, + PFID(mdd_object_fid(mdd_obj))); else - CDEBUG(error ? D_ERROR : D_OTHER, - "link_ea rename '%.*s'->'%.*s' failed %d " - DFID"\n", - oldlname->ln_namelen, oldlname->ln_name, - newlname->ln_namelen, newlname->ln_name, - rc, PFID(mdd_object_fid(mdd_obj))); + CERROR("link_ea rename '%.*s'->'%.*s' failed %d "DFID + "\n", oldlname->ln_namelen, oldlname->ln_name, + newlname->ln_namelen, newlname->ln_name, 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); @@ -1061,10 +1122,10 @@ static inline int mdd_links_add(const struct lu_env *env, const struct lu_fid *pfid, const struct lu_name *lname, struct thandle *handle, - struct linkea_data *data, int first) + struct linkea_data *ldata, int first) { return mdd_links_rename(env, mdd_obj, NULL, NULL, - pfid, lname, handle, data, first, 0); + pfid, lname, handle, ldata, first, 0); } static inline int mdd_links_del(const struct lu_env *env, @@ -1078,51 +1139,6 @@ static inline int mdd_links_del(const struct lu_env *env, } /** Read the link EA into a temp buffer. - * Uses the mdd_thread_info::mti_big_buf since it is generally large. - * A pointer to the buffer is stored in \a ldata::ld_buf. - * - * \retval 0 or error - */ -int mdd_links_read(const struct lu_env *env, struct mdd_object *mdd_obj, - struct linkea_data *ldata) -{ - int rc; - - if (!mdd_object_exists(mdd_obj)) - return -ENODATA; - - /* First try a small buf */ - LASSERT(env != NULL); - ldata->ld_buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_link_buf, - PAGE_CACHE_SIZE); - if (ldata->ld_buf->lb_buf == NULL) - return -ENOMEM; - - rc = mdo_xattr_get(env, mdd_obj, ldata->ld_buf, XATTR_NAME_LINK, - BYPASS_CAPA); - 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); - 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); - } - if (rc < 0) { - lu_buf_free(ldata->ld_buf); - ldata->ld_buf = NULL; - return rc; - } - - return linkea_init(ldata); -} - -/** Read the link EA into a temp buffer. * Uses the name_buf since it is generally large. * \retval IS_ERR err * \retval ptr to \a lu_buf (always \a mti_big_buf) @@ -1130,7 +1146,7 @@ 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); @@ -1140,14 +1156,33 @@ 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); - return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle, - mdd_object_capa(env, mdd_obj)); + const struct lu_buf *buf; + int rc; + + if (ldata == NULL || ldata->ld_buf == NULL || + ldata->ld_leh == NULL) + return 0; + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA)) + return 0; + +again: + buf = mdd_buf_get_const(env, ldata->ld_buf->lb_buf, + ldata->ld_leh->leh_len); + rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle); + if (unlikely(rc == -ENOSPC)) { + rc = linkea_overflow_shrink(ldata); + if (likely(rc > 0)) + goto again; + } + + return rc; } -int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, - struct thandle *handle, struct linkea_data *ldata) +static int mdd_declare_links_add(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct thandle *handle, + struct linkea_data *ldata) { int rc; int ea_len; @@ -1157,14 +1192,14 @@ int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, ea_len = ldata->ld_leh->leh_len; linkea = ldata->ld_buf->lb_buf; } else { - ea_len = DEFAULT_LINKEA_SIZE; + ea_len = MAX_LINKEA_SIZE; linkea = NULL; } - /* XXX: max size? */ rc = mdo_declare_xattr_set(env, mdd_obj, mdd_buf_get_const(env, linkea, ea_len), XATTR_NAME_LINK, 0, handle); + return rc; } @@ -1183,23 +1218,28 @@ static inline int mdd_declare_links_del(const struct lu_env *env, } 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; + struct lu_fid tfid = *mdo2fid(c); + int rc; - rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle); - if (rc) - return rc; + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) + tfid.f_oid = cfs_fail_val; - rc = mdo_declare_ref_add(env, c, handle); - if (rc) - return rc; + rc = mdo_declare_index_insert(env, p, &tfid, mdd_object_type(c), + name->ln_name, handle); + if (rc != 0) + return rc; + + 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); @@ -1208,16 +1248,16 @@ 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; + 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, @@ -1230,17 +1270,18 @@ 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); @@ -1253,6 +1294,14 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, LASSERT(ma->ma_attr.la_valid & LA_CTIME); la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + /* Note: even this function will change ldata, but it comes from + * thread_info, which is completely temporary and only seen in + * this function, so we do not need reset ldata once it fails.*/ + rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdo2fid(mdd_tobj), + lname, 0, 0, ldata); + if (rc != 0) + GOTO(stop, rc); + rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle, la, ldata); if (rc) @@ -1268,14 +1317,18 @@ 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); + } + *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, mdo2fid(mdd_sobj), - name, handle, - mdd_object_capa(env, mdd_tobj)); + 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); @@ -1288,31 +1341,47 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, la->la_valid = LA_CTIME; rc = mdd_update_time(env, mdd_sobj, cattr, la, handle); - if (rc == 0) { - rc = mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, - mdo2fid(mdd_tobj), lname, 0, 0, - ldata); - if (rc == 0) - mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), - lname, handle, ldata, 0); - /* The failure of links_add should not cause the link - * failure, reset rc here */ - rc = 0; - } - EXIT; + if (rc == 0) + /* Note: The failure of links_add should not cause the + * link failure, so do not check return value. */ + mdd_links_add(env, mdd_sobj, mdo2fid(mdd_tobj), + lname, handle, ldata, 0); + + 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, - mdd_tobj, lname, handle); + 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_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 (!S_ISDIR(mdd_object_type(obj))) + return 0; + + attr->la_valid = LA_FLAGS; + 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); + + return rc; } static int mdd_declare_finish_unlink(const struct lu_env *env, @@ -1321,48 +1390,69 @@ static int mdd_declare_finish_unlink(const struct lu_env *env, { int rc; + /* 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 = mdo_declare_destroy(env, obj, handle); + if (rc != 0) + return rc; + rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle); - if (rc) + if (rc != 0) return rc; - return mdo_declare_destroy(env, obj, handle); + return mdd_declare_links_del(env, obj, handle); } /* caller should take a lock before calling */ int mdd_finish_unlink(const struct lu_env *env, - struct mdd_object *obj, struct md_attr *ma, - struct thandle *th) + struct mdd_object *obj, struct md_attr *ma, + const struct mdd_object *pobj, + const struct lu_name *lname, + 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) { - obj->mod_flags |= DEAD_OBJ; - - /* 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); + } - RETURN(rc); + RETURN(rc); } /* @@ -1385,20 +1475,25 @@ 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; @@ -1425,12 +1520,8 @@ static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd, if (rc) return rc; - rc = mdd_declare_links_del(env, c, handle); - if (rc != 0) - 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; @@ -1455,14 +1546,13 @@ 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); @@ -1490,7 +1580,7 @@ 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 */ @@ -1500,17 +1590,22 @@ 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); @@ -1522,7 +1617,7 @@ 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); @@ -1533,20 +1628,23 @@ 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); } @@ -1555,7 +1653,7 @@ 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); } @@ -1584,41 +1682,43 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, /* XXX: this transfer to ma will be removed with LOD/OSP */ ma->ma_attr = *cattr; ma->ma_valid |= MA_INODE; - rc = mdd_finish_unlink(env, mdd_cobj, ma, handle); + 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); - - if (!is_dir) - /* old files may not have link ea; ignore errors */ - mdd_links_del(env, mdd_cobj, mdo2fid(mdd_pobj), lname, handle); + 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: - mdd_write_unlock(env, mdd_cobj); - if (rc == 0) { - int cl_flags = 0; + if (likely(mdd_cobj != NULL)) + mdd_write_unlock(env, mdd_cobj); - if (cattr->la_nlink == 0) { + if (rc == 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, - mdd_cobj, mdd_pobj, lname, handle); + mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL, + handle); } stop: - mdd_trans_stop(env, mdd, rc, handle); + rc = mdd_trans_stop(env, mdd, rc, handle); return rc; } @@ -1638,9 +1738,11 @@ 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); @@ -1665,22 +1767,22 @@ 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? - */ - CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n", + /* + * 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 %#llo, no_create %u\n", spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen, spec->sp_cr_flags, spec->no_create); @@ -1696,7 +1798,7 @@ 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); @@ -1705,7 +1807,7 @@ 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); @@ -1713,7 +1815,8 @@ 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); } @@ -1721,31 +1824,27 @@ 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; + 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)) { - rc = mdo_declare_index_insert(env, child, mdo2fid(child), - dot, handle); - if (rc == 0) - rc = mdo_declare_ref_add(env, child, handle); - - rc = mdo_declare_index_insert(env, child, mdo2fid(parent), - dotdot, handle); - } + if (!S_ISDIR(attr->la_mode)) + RETURN(0); + + 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); } @@ -1756,36 +1855,41 @@ 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); } -/* has not lock on pobj yet */ +/** + * This function checks whether it can create a file/dir under the + * directory(@pobj). The directory(@pobj) is not being locked by + * mdd lock. + * + * \param[in] env execution environment + * \param[in] pobj the directory to create files + * \param[in] pattr the attributes of the directory + * \param[in] lname the name of the created file/dir + * \param[in] cattr the attributes of the file/dir + * \param[in] spec create specification + * + * \retval = 0 it is allowed to create file/dir under + * the directory + * \retval negative error not allowed to create file/dir + * under the directory + */ static int mdd_create_sanity_check(const struct lu_env *env, struct md_object *pobj, const struct lu_attr *pattr, @@ -1793,16 +1897,17 @@ static int mdd_create_sanity_check(const struct lu_env *env, struct lu_attr *cattr, struct md_op_spec *spec) { - struct mdd_thread_info *info = mdd_env_info(env); - struct lu_fid *fid = &info->mti_fid; - struct mdd_object *obj = md2mdd_obj(pobj); - struct mdd_device *m = mdo2mdd(pobj); - int rc; - ENTRY; + struct mdd_thread_info *info = mdd_env_info(env); + struct lu_fid *fid = &info->mti_fid; + struct mdd_object *obj = md2mdd_obj(pobj); + struct mdd_device *m = mdo2mdd(pobj); + bool check_perm = true; + 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 @@ -1817,20 +1922,17 @@ static int mdd_create_sanity_check(const struct lu_env *env, */ rc = __mdd_lookup(env, pobj, pattr, lname, fid, MAY_WRITE | MAY_EXEC); - if (rc != -ENOENT) - RETURN(rc ? : -EEXIST); - } else { - /* - * Check WRITE permission for the parent. - * EXEC permission have been checked - * when lookup before create already. - */ - rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE, - MOR_TGT_PARENT); - if (rc) - RETURN(rc); + if (rc != -ENOENT) + RETURN(rc ? : -EEXIST); + + /* Permission is already being checked in mdd_lookup */ + check_perm = false; } + rc = mdd_may_create(env, obj, pattr, NULL, check_perm); + if (rc != 0) + RETURN(rc); + /* sgid check */ if (pattr->la_mode & S_ISGID) { cattr->la_gid = pattr->la_gid; @@ -1840,15 +1942,19 @@ static int mdd_create_sanity_check(const struct lu_env *env, } } + rc = mdd_name_check(m, lname); + if (rc < 0) + 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: @@ -1874,6 +1980,7 @@ 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, @@ -1909,8 +2016,6 @@ 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, @@ -1930,6 +2035,16 @@ 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; } @@ -1958,17 +2073,18 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, GOTO(out, rc); } - if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) { + if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) { rc = orph_declare_index_insert(env, c, attr->la_mode, handle); if (rc) GOTO(out, rc); } 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); if (rc) return rc; @@ -1978,18 +2094,11 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, rc = mdo_declare_attr_set(env, p, la, handle); if (rc) return rc; - } - - rc = mdd_declare_changelog_store(env, mdd, name, 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; + rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle); + if (rc) + return rc; + } out: return rc; } @@ -2009,7 +2118,7 @@ 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 */ @@ -2046,6 +2155,7 @@ 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); @@ -2054,13 +2164,50 @@ static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, if (rc) GOTO(unlock, rc); + /* Note: In DNE phase I, for striped dir, though sub-stripes will be + * created in declare phase, they also needs to be added to master + * object as sub-directory entry. So it has to initialize the master + * object, then set dir striped EA.(in mdo_xattr_set) */ + rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle, + spec); + if (rc != 0) + GOTO(err_destroy, rc); + + /* + * in case of replay we just set LOVEA provided by the client + * XXX: I think it would be interesting to try "old" way where + * MDT calls this xattr_set(LOV) in a different transaction. + * probably this way we code can be made better. + */ + + /* During creation, there are only a few cases we need do xattr_set to + * create stripes. + * 1. regular file: see comments above. + * 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)) { + 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); + if (rc != 0) + GOTO(err_destroy, rc); + } + #ifdef CONFIG_FS_POSIX_ACL if (def_acl_buf != NULL && def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) { /* 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); } @@ -2068,35 +2215,12 @@ 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); } #endif - rc = mdd_object_initialize(env, mdo2fid(pobj), son, attr, handle, - spec); - if (rc != 0) - GOTO(err_destroy, rc); - - /* - * in case of replay we just set LOVEA provided by the client - * XXX: I think it would be interesting to try "old" way where - * MDT calls this xattr_set(LOV) in a different transaction. - * probably this way we code can be made better. - */ - 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_xattr_set(env, son, buf, XATTR_NAME_LOV, 0, handle, - BYPASS_CAPA); - if (rc != 0) - GOTO(err_destroy, rc); - } - if (S_ISLNK(attr->la_mode)) { struct lu_ucred *uc = lu_ucred_assert(env); struct dt_object *dt = mdd_object_child(son); @@ -2107,7 +2231,6 @@ 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); @@ -2117,6 +2240,15 @@ 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; @@ -2138,6 +2270,48 @@ 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. */ @@ -2159,6 +2333,7 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, const char *name = lname->ln_name; struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint; int rc; + int rc2; ENTRY; /* @@ -2197,7 +2372,7 @@ 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); @@ -2224,9 +2399,17 @@ 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, @@ -2246,13 +2429,10 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) { mdd_write_lock(env, son, MOR_TGT_CHILD); rc = __mdd_orphan_add(env, son, handle); - mdd_write_unlock(env, son); - if (rc != 0) - GOTO(err_created, rc); + 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); @@ -2270,14 +2450,12 @@ 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; @@ -2285,41 +2463,60 @@ err_created: mdd_write_lock(env, son, MOR_TGT_CHILD); if (S_ISDIR(attr->la_mode)) { /* Drop the reference, no need to delete "."/"..", - * because the object to be destroied directly. */ + * because the object is to be destroyed directly. */ rc2 = mdo_ref_del(env, son, handle); if (rc2 != 0) { mdd_write_unlock(env, son); goto out_stop; } } +out_volatile: + /* For volatile files drop one link immediately, since there is + * no filename in the namespace, and save any error returned. */ rc2 = mdo_ref_del(env, son, handle); if (rc2 != 0) { mdd_write_unlock(env, son); + if (unlikely(rc == 0)) + rc = rc2; goto out_stop; } - mdo_destroy(env, son, handle); + /* Don't destroy the volatile object on success */ + if (likely(rc != 0)) + mdo_destroy(env, son, handle); mdd_write_unlock(env, son); - } + } - if (rc == 0 && fid_is_namespace_visible(mdo2fid(son))) + if (rc == 0 && fid_is_namespace_visible(mdo2fid(son)) && + likely((spec->sp_cr_flags & MDS_OPEN_VOLATILE) == 0)) rc = mdd_changelog_ns_store(env, mdd, - 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); + 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, mdo2fid(mdd_pobj), NULL, NULL, lname, + NULL, handle); out_stop: - mdd_trans_stop(env, mdd, rc, handle); + rc2 = mdd_trans_stop(env, mdd, rc, handle); + 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); - /* The child object shouldn't be cached anymore */ - if (rc) + /* The child object shouldn't be cached anymore */ + if (rc) set_bit(LU_OBJECT_HEARD_BANSHEE, - &child->mo_lu.lo_header->loh_flags); - return rc; + &child->mo_lu.lo_header->loh_flags); + return rc; } /* @@ -2393,9 +2590,9 @@ static int mdd_rename_sanity_check(const struct lu_env *env, * processed in cld_rename before mdd_rename and enable * MDS_PERM_BYPASS). * So check may_create, but not check may_unlink. */ - if (!tobj) + if (tobj == NULL) rc = mdd_may_create(env, tgt_pobj, tpattr, NULL, - (src_pobj != tgt_pobj), 0); + (src_pobj != tgt_pobj)); else rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr, (src_pobj != tgt_pobj), 1); @@ -2412,8 +2609,8 @@ 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) @@ -2443,21 +2640,22 @@ 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); @@ -2477,21 +2675,20 @@ static int mdd_declare_rename(const struct lu_env *env, 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; - - /* 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; + /* 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; 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; @@ -2514,16 +2711,12 @@ static int mdd_declare_rename(const struct lu_env *env, if (rc) return rc; - rc = mdd_declare_links_del(env, mdd_tobj, handle); - if (rc) - return rc; - rc = mdd_declare_finish_unlink(env, mdd_tobj, handle); if (rc) 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; @@ -2564,22 +2757,29 @@ 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); @@ -2588,13 +2788,26 @@ static int mdd_rename(const struct lu_env *env, if (rc) GOTO(out_pending, rc); + rc = mdd_name_check(mdd, ltname); + 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); + rc = mdd_linkea_prepare(env, mdd_sobj, mdd_object_fid(mdd_spobj), + lsname, mdd_object_fid(mdd_tpobj), ltname, + 1, 0, ldata); + if (rc) + GOTO(stop, rc); + rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj, mdd_tobj, lsname, ltname, ma, ldata, handle); if (rc) @@ -2604,63 +2817,60 @@ 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)); - if (rc) - GOTO(fixup_spobj2, rc); + /* "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, 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); } - /* 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, 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); 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 @@ -2684,7 +2894,7 @@ 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", @@ -2706,7 +2916,8 @@ static int mdd_rename(const struct lu_env *env, /* XXX: this transfer to ma will be removed with LOD/OSP */ ma->ma_attr = *tattr; ma->ma_valid |= MA_INODE; - rc = mdd_finish_unlink(env, mdd_tobj, ma, handle); + rc = mdd_finish_unlink(env, mdd_tobj, ma, mdd_tpobj, ltname, + handle); if (rc != 0) { CERROR("%s: Failed to unlink tobj " DFID": rc = %d\n", @@ -2716,8 +2927,13 @@ 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, @@ -2728,11 +2944,10 @@ 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; @@ -2743,29 +2958,15 @@ 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); @@ -2777,52 +2978,51 @@ 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); + rc = mdd_trans_stop(env, mdd, rc, handle); + out_pending: mdd_object_put(env, mdd_sobj); return rc; @@ -2834,13 +3034,14 @@ 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, 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; @@ -2869,8 +3070,8 @@ static int mdd_linkea_update_child_internal(const struct lu_env *env, linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, &lname, &fid); - if (strncmp(lname.ln_name, name, namelen) || - lu_fid_eq(&fid, mdd_object_fid(parent))) { + if (strncmp(lname.ln_name, name, namelen) != 0 || + !lu_fid_eq(&fid, mdd_object_fid(parent))) { ldata.ld_lee = (struct link_ea_entry *) ((char *)ldata.ld_lee + ldata.ld_reclen); @@ -2880,10 +3081,10 @@ 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); else @@ -2895,21 +3096,25 @@ 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); } @@ -2918,33 +3123,20 @@ 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); - } - - 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); + LASSERT(ldata->ld_buf != NULL); + LASSERT(ldata->ld_leh != NULL); /* If it is mulitple links file, we need update the name entry for * all parent */ - 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(&mdd_sobj->mod_obj); @@ -2954,8 +3146,6 @@ 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", @@ -2967,7 +3157,7 @@ static int mdd_update_linkea_internal(const struct lu_env *env, 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); + goto next_put; } if (pobj == mdd_pobj && @@ -2977,7 +3167,7 @@ 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); + goto next_put; } CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n", @@ -2989,13 +3179,14 @@ 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); @@ -3006,17 +3197,43 @@ 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 */ + CWARN("%s: the name %.*s is too long under " + DFID"\n", mdd2obd_dev(mdd)->obd_name, + lname.ln_namelen, lname.ln_name, + PFID(&fid)); + goto next_put; + } + + /* 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))) + GOTO(next_put, rc == -ENOENT ? 0 : rc); + + 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), - lname.ln_name, 0, handle, - mdd_object_capa(env, pobj)); - if (rc) + mdd_object_fid(mdd_tobj), + mdd_object_type(mdd_tobj), + tmp_name, handle); + if (rc != 0) GOTO(next_put, rc); mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD); @@ -3033,6 +3250,9 @@ 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); @@ -3055,8 +3275,7 @@ 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; @@ -3069,8 +3288,7 @@ 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; @@ -3078,8 +3296,7 @@ static int mdd_migrate_xattrs(const struct lu_env *env, xname = list_xbuf.lb_buf; while (rem > 0) { xlen = strnlen(xname, rem - 1) + 1; - if (strcmp(XATTR_NAME_LINK, xname) == 0 || - strcmp(XATTR_NAME_LMA, xname) == 0 || + if (strcmp(XATTR_NAME_LMA, xname) == 0 || strcmp(XATTR_NAME_LMV, xname) == 0) goto next; @@ -3088,9 +3305,7 @@ 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) @@ -3102,8 +3317,7 @@ 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) @@ -3125,15 +3339,25 @@ 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)); +again: + rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle); if (rc == -EEXIST) GOTO(stop_trans, rc = 0); + if (unlikely(rc == -ENOSPC && + strcmp(xname, XATTR_NAME_LINK) == 0)) { + rc = linkea_overflow_shrink( + (struct linkea_data *)(xbuf.lb_buf)); + if (likely(rc > 0)) { + xbuf.lb_len = rc; + goto again; + } + } + 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: @@ -3151,6 +3375,7 @@ 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); @@ -3178,6 +3403,10 @@ 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) && ldata != NULL) { + rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata); + if (rc != 0) + return rc; } if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) { @@ -3189,7 +3418,7 @@ static int mdd_declare_migrate_create(const struct lu_env *env, return rc; } - mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_MIGRATE); + mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1); buf = mdd_buf_get_const(env, mgr_ea, mgr_easize); rc = mdo_declare_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle); @@ -3197,8 +3426,7 @@ static int mdd_declare_migrate_create(const struct lu_env *env, return rc; la_flag->la_valid = LA_FLAGS; - la_flag->la_flags = LUSTRE_IMMUTABLE_FL; - mdd_flags_xlate(mdd_sobj, la_flag->la_flags); + la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL; rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle); return rc; @@ -3208,18 +3436,21 @@ 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; @@ -3233,6 +3464,7 @@ 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; @@ -3242,7 +3474,7 @@ static int mdd_migrate_create(const struct lu_env *env, RETURN(rc); } spec->u.sp_symname = link_buf.lb_buf; - } else{ + } 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) @@ -3252,16 +3484,28 @@ 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_with_rec(env, mdd_sobj, ldata); + if (rc == -ENODATA) { + /* ignore the non-linkEA error */ + ldata = NULL; + rc = 0; + } + if (rc < 0) + RETURN(rc); } mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf; - mgr_ea->lmv_magic = cpu_to_le32(LMV_MAGIC_MIGRATE); + 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; - mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_TYPE_MIGRATION); + mgr_ea->lmv_hash_type = cpu_to_le32(LMV_HASH_FLAG_MIGRATION); fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[0], mdd_object_fid(mdd_sobj)); fid_cpu_to_le(&mgr_ea->lmv_stripe_fids[1], mdd_object_fid(mdd_tobj)); + mdd_object_make_hint(env, mdd_pobj, mdd_tobj, la, spec, hint); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_free, rc = PTR_ERR(handle)); @@ -3273,7 +3517,7 @@ 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); @@ -3281,16 +3525,17 @@ 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, - NULL, handle); + hint, handle); if (rc != 0) GOTO(stop_trans, rc); - if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) { - buf = mdd_buf_get_const(env, lmm_buf.lb_buf, lmm_buf.lb_len); - rc = mdo_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV, - 0, handle, mdd_object_capa(env, mdd_sobj)); + if (S_ISDIR(la->la_mode) && ldata != NULL) { + rc = mdd_links_write(env, mdd_tobj, ldata, handle); if (rc != 0) GOTO(stop_trans, rc); } @@ -3298,10 +3543,9 @@ static int mdd_migrate_create(const struct lu_env *env, /* 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_MIGRATE); + 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); @@ -3311,13 +3555,11 @@ static int mdd_migrate_create(const struct lu_env *env, * IMMUTALBE flag and MIGRATE EA, it need to clear IMMUTABLE * flag and approve the migration */ la_flag->la_valid = LA_FLAGS; - la_flag->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)); + la_flag->la_flags = la->la_flags | LUSTRE_IMMUTABLE_FL; + 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); @@ -3334,9 +3576,9 @@ 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); @@ -3349,8 +3591,7 @@ 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)); @@ -3370,7 +3611,6 @@ static int mdd_migrate_entries(const struct lu_env *env, struct mdd_object *child; char *name = mdd_env_info(env)->mti_key; int len; - int recsize; int is_dir; bool target_exist = false; @@ -3388,7 +3628,6 @@ static int mdd_migrate_entries(const struct lu_env *env, } fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid); - recsize = le16_to_cpu(ent->lde_reclen); /* Insert new fid with target name into target dir */ if ((ent->lde_namelen == 1 && ent->lde_name[0] == '.') || @@ -3400,7 +3639,8 @@ 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)); + 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); @@ -3408,17 +3648,15 @@ 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; } 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 @@ -3427,6 +3665,7 @@ 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); @@ -3455,12 +3694,12 @@ 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); } - 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); @@ -3476,44 +3715,37 @@ 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); } - 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: @@ -3536,22 +3768,24 @@ 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, @@ -3561,15 +3795,16 @@ 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; 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; @@ -3579,16 +3814,25 @@ 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))) { @@ -3599,10 +3843,15 @@ 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); + 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) @@ -3630,6 +3879,10 @@ 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; } @@ -3645,6 +3898,7 @@ 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; @@ -3656,16 +3910,22 @@ 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 */ @@ -3687,67 +3947,116 @@ 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); - 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), 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); - rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle, - NULL, 1); + 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(stop_trans, rc); + GOTO(out_unlock, rc); + + rc = __mdd_orphan_add(env, mdd_sobj, handle); + if (rc != 0) + GOTO(out_unlock, rc); - mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD); 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); + 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, handle); - if (rc != 0) - GOTO(stop_trans, rc); rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0); if (rc != 0) - GOTO(stop_trans, rc); + 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 @@ -3764,20 +4073,20 @@ 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_MIGRATE); + mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1); mgr_buf = lu_buf_check_and_alloc(&info->mti_big_buf, mgr_easize); 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; @@ -3785,12 +4094,11 @@ static int mdd_migrate_sanity_check(const struct lu_env *env, * is being set by previous migration process, so it * needs to override the IMMUTE flag, otherwise the * following sanity check will fail */ - if (le32_to_cpu(lmm->lmv_md_v1.lmv_magic) == - LMV_MAGIC_MIGRATE) { + if (le32_to_cpu(lmm->lmv_md_v1.lmv_hash_type) & + LMV_HASH_FLAG_MIGRATION) { 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))); @@ -3812,46 +4120,46 @@ 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); + + /* If the linkEA is overflow, then means there are some unknown name + * entries under unknown parents, that will prevent the migration. */ + if (unlikely(ldata->ld_leh->leh_overflow_time)) + RETURN(1); + 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; } @@ -3860,27 +4168,25 @@ static int mdd_migrate_sanity_check(const struct lu_env *env, } static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, - const struct lu_fid *lf, const struct lu_name *lname, + struct md_object *sobj, const struct lu_name *lname, struct md_object *tobj, struct md_attr *ma) { struct mdd_object *mdd_pobj = md2mdd_obj(pobj); struct mdd_device *mdd = mdo2mdd(pobj); - struct mdd_object *mdd_sobj = NULL; - struct mdd_object *mdd_tobj = NULL; + struct mdd_object *mdd_sobj = md2mdd_obj(sobj); + 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; - /* object has to be locked by mdt, so it must exist */ - mdd_sobj = mdd_object_find(env, mdd, lf); - LASSERT(mdd_sobj != NULL); - /* 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); @@ -3889,11 +4195,11 @@ 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); @@ -3926,14 +4232,15 @@ 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, - so_attr); + 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) @@ -3947,17 +4254,21 @@ 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); } + 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); -put: - if (mdd_sobj) - mdd_object_put(env, mdd_sobj); + /* newly created target was not locked, don't cache its attributes */ + if (created) + mdd_invalidate(env, tobj); +put: RETURN(rc); }