X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_dir.c;h=bef36b57f9bd4b4f38194082a88eef06d394c6c8;hp=773940e76e902562a73fe1821bb313c12bd6c209;hb=76f0977b7ea5d46836cb459deb7b9ad9e781d585;hpb=204c513f80e91733dc6afbfc79417111caaaa5b3 diff --git a/lustre/mdd/mdd_dir.c b/lustre/mdd/mdd_dir.c index 773940e..bef36b5 100644 --- a/lustre/mdd/mdd_dir.c +++ b/lustre/mdd/mdd_dir.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel Corporation. + * Copyright (c) 2011, 2013, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -37,7 +37,7 @@ * * Lustre Metadata Server (mdd) routines * - * Author: Wang Di + * Author: Wang Di */ #define DEBUG_SUBSYSTEM S_MDS @@ -57,63 +57,77 @@ static struct lu_name lname_dotdot = { sizeof(dotdot) - 1 }; -static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj, - const struct lu_name *lname, struct lu_fid* fid, - int mask); -static int mdd_declare_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct thandle *handle); -static inline int mdd_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *pfid, - const struct lu_name *lname, - struct thandle *handle, int first); -static inline int mdd_links_del(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *pfid, - const struct lu_name *lname, - struct thandle *handle); -static int mdd_links_rename(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *oldpfid, - const struct lu_name *oldlname, - const struct lu_fid *newpfid, - const struct lu_name *newlname, - struct thandle *handle, - int first, int check); - +/* Get FID from name and parent */ static int -__mdd_lookup_locked(const struct lu_env *env, struct md_object *pobj, - const struct lu_name *lname, struct lu_fid* fid, int mask) +__mdd_lookup(const struct lu_env *env, struct md_object *pobj, + const struct lu_attr *pattr, const struct lu_name *lname, + struct lu_fid* fid, int mask) { - const char *name = lname->ln_name; - struct mdd_object *mdd_obj = md2mdd_obj(pobj); - struct dynlock_handle *dlh; + const char *name = lname->ln_name; + const struct dt_key *key = (const struct dt_key *)name; + struct mdd_object *mdd_obj = md2mdd_obj(pobj); + struct mdd_device *m = mdo2mdd(pobj); + struct dt_object *dir = mdd_object_child(mdd_obj); int rc; + ENTRY; - dlh = mdd_pdo_read_lock(env, mdd_obj, name, MOR_TGT_PARENT); - if (unlikely(dlh == NULL)) - return -ENOMEM; - rc = __mdd_lookup(env, pobj, lname, fid, mask); - mdd_pdo_read_unlock(env, mdd_obj, dlh); + if (unlikely(mdd_is_dead_obj(mdd_obj))) + RETURN(-ESTALE); - return rc; + 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 + rc = -ENOTDIR; + + RETURN(rc); } int mdd_lookup(const struct lu_env *env, - struct md_object *pobj, const struct lu_name *lname, - struct lu_fid* fid, struct md_op_spec *spec) + struct md_object *pobj, const struct lu_name *lname, + struct lu_fid *fid, struct md_op_spec *spec) { + struct lu_attr *pattr = MDD_ENV_VAR(env, pattr); int rc; ENTRY; - rc = __mdd_lookup_locked(env, pobj, lname, fid, MAY_EXEC); + + rc = mdd_la_get(env, md2mdd_obj(pobj), pattr, BYPASS_CAPA); + if (rc != 0) + RETURN(rc); + + rc = __mdd_lookup(env, pobj, pattr, lname, fid, MAY_EXEC); RETURN(rc); } -static int mdd_parent_fid(const struct lu_env *env, struct mdd_object *obj, - struct lu_fid *fid) +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_locked(env, &obj->mod_obj, &lname_dotdot, fid, 0); + return __mdd_lookup(env, &obj->mod_obj, attr, &lname_dotdot, fid, 0); } /* @@ -136,10 +150,11 @@ int mdd_is_root(struct mdd_device *mdd, const struct lu_fid *fid) * otherwise: values < 0, errors. */ static int mdd_is_parent(const struct lu_env *env, - struct mdd_device *mdd, - struct mdd_object *p1, - const struct lu_fid *lf, - struct lu_fid *pf) + struct mdd_device *mdd, + struct mdd_object *p1, + const struct lu_attr *attr, + const struct lu_fid *lf, + struct lu_fid *pf) { struct mdd_object *parent = NULL; struct lu_fid *pfid; @@ -154,27 +169,30 @@ 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); - rc = mdd_parent_fid(env, p1, pfid); - if (rc) - GOTO(out, rc); + /* this is done recursively, bypass capa for each obj */ + mdd_set_capainfo(env, 4, p1, BYPASS_CAPA); + rc = mdd_parent_fid(env, p1, attr, pfid); + if (rc) + GOTO(out, rc); if (mdd_is_root(mdd, pfid)) GOTO(out, rc = 0); + if (lu_fid_eq(pfid, &mdd->mdd_local_root_fid)) + GOTO(out, rc = 0); if (lu_fid_eq(pfid, lf)) GOTO(out, rc = 1); if (parent) mdd_object_put(env, parent); - parent = mdd_object_find(env, mdd, pfid); - - /* cross-ref parent */ - if (parent == NULL) { - if (pf != NULL) - *pf = *pfid; - GOTO(out, rc = -EREMOTE); - } else if (IS_ERR(parent)) - GOTO(out, rc = PTR_ERR(parent)); - p1 = parent; + + parent = mdd_object_find(env, mdd, pfid); + 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); + } + p1 = parent; } EXIT; out: @@ -195,25 +213,30 @@ out: * returns < 0: if error */ int mdd_is_subdir(const struct lu_env *env, struct md_object *mo, - const struct lu_fid *fid, struct lu_fid *sfid) + const struct lu_fid *fid, struct lu_fid *sfid) { - struct mdd_device *mdd = mdo2mdd(mo); - int rc; - ENTRY; + struct mdd_device *mdd = mdo2mdd(mo); + struct lu_attr *attr = MDD_ENV_VAR(env, cattr); + int rc; + ENTRY; - if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo)))) - RETURN(0); + if (!S_ISDIR(mdd_object_type(md2mdd_obj(mo)))) + RETURN(0); - rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), fid, sfid); - if (rc == 0) { - /* found root */ - fid_zero(sfid); - } else if (rc == 1) { - /* found @fid is parent */ - *sfid = *fid; - rc = 0; - } - RETURN(rc); + rc = mdd_la_get(env, md2mdd_obj(mo), attr, BYPASS_CAPA); + if (rc != 0) + RETURN(rc); + + rc = mdd_is_parent(env, mdd, md2mdd_obj(mo), attr, fid, sfid); + if (rc == 0) { + /* found root */ + fid_zero(sfid); + } else if (rc == 1) { + /* found @fid is parent */ + *sfid = *fid; + rc = 0; + } + RETURN(rc); } /* @@ -265,79 +288,79 @@ static int mdd_dir_is_empty(const struct lu_env *env, RETURN(result); } -static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj) +static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj, + const struct lu_attr *la) { - struct mdd_device *m = mdd_obj2mdd_dev(obj); - struct lu_attr *la = &mdd_env_info(env)->mti_la; - int rc; - ENTRY; + struct mdd_device *m = mdd_obj2mdd_dev(obj); + ENTRY; - rc = mdd_la_get(env, obj, la, BYPASS_CAPA); - if (rc) - RETURN(rc); + LASSERT(la != NULL); - /* - * Subdir count limitation can be broken through. - */ - if (la->la_nlink >= m->mdd_dt_conf.ddp_max_nlink && - !S_ISDIR(la->la_mode)) - RETURN(-EMLINK); - else - RETURN(0); + 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) + RETURN(-EMLINK); + else + RETURN(0); } /* * Check whether it may create the cobj under the pobj. * cobj maybe NULL */ -int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, - 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, int check_perm, int check_nlink) { - int rc = 0; - ENTRY; + int rc = 0; + ENTRY; - if (cobj && mdd_object_exists(cobj)) - RETURN(-EEXIST); + if (cobj && mdd_object_exists(cobj)) + RETURN(-EEXIST); - if (mdd_is_dead_obj(pobj)) - RETURN(-ENOENT); + if (mdd_is_dead_obj(pobj)) + RETURN(-ENOENT); - if (check_perm) - rc = mdd_permission_internal_locked(env, pobj, NULL, - MAY_WRITE | MAY_EXEC, - MOR_TGT_PARENT); - if (!rc && check_nlink) - rc = __mdd_may_link(env, pobj); + if (check_perm) + 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); + RETURN(rc); } /* * Check whether can unlink from the pobj in the case of "cobj == NULL". */ int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj, - const struct lu_attr *attr) + const struct lu_attr *pattr, const struct lu_attr *attr) { - int rc; - ENTRY; + int rc; + ENTRY; - if (mdd_is_dead_obj(pobj)) - RETURN(-ENOENT); + if (mdd_is_dead_obj(pobj)) + RETURN(-ENOENT); if ((attr->la_valid & LA_FLAGS) && (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) - RETURN(-EPERM); + RETURN(-EPERM); - rc = mdd_permission_internal_locked(env, pobj, NULL, - MAY_WRITE | MAY_EXEC, - MOR_TGT_PARENT); - if (rc) - RETURN(rc); + rc = mdd_permission_internal_locked(env, pobj, pattr, + MAY_WRITE | MAY_EXEC, + MOR_TGT_PARENT); + if (rc) + RETURN(rc); - if (mdd_is_append(pobj)) - RETURN(-EPERM); + if (mdd_is_append(pobj)) + RETURN(-EPERM); - RETURN(rc); + RETURN(rc); } /* @@ -346,107 +369,123 @@ int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj, */ static inline int mdd_is_sticky(const struct lu_env *env, struct mdd_object *pobj, - struct mdd_object *cobj) + const struct lu_attr *pattr, + struct mdd_object *cobj, + const struct lu_attr *cattr) { - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; struct lu_ucred *uc = lu_ucred_assert(env); - int rc; - - if (pobj) { - rc = mdd_la_get(env, pobj, tmp_la, BYPASS_CAPA); - if (rc) - return rc; - if (!(tmp_la->la_mode & S_ISVTX) || - (tmp_la->la_uid == uc->uc_fsuid)) + if (pobj != NULL) { + LASSERT(pattr != NULL); + if (!(pattr->la_mode & S_ISVTX) || + (pattr->la_uid == uc->uc_fsuid)) return 0; } - rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA); - if (rc) - return rc; - - if (tmp_la->la_uid == uc->uc_fsuid) + LASSERT(cattr != NULL); + if (cattr->la_uid == uc->uc_fsuid) return 0; - return !mdd_capable(uc, CFS_CAP_FOWNER); + return !md_capable(uc, CFS_CAP_FOWNER); +} + +static int mdd_may_delete_entry(const struct lu_env *env, + struct mdd_object *pobj, + const struct lu_attr *pattr, + int check_perm) +{ + ENTRY; + + LASSERT(pobj != NULL); + if (!mdd_object_exists(pobj)) + RETURN(-ENOENT); + + if (mdd_is_dead_obj(pobj)) + RETURN(-ENOENT); + + if (check_perm) { + int rc; + rc = mdd_permission_internal_locked(env, pobj, pattr, + MAY_WRITE | MAY_EXEC, + MOR_TGT_PARENT); + if (rc) + RETURN(rc); + } + + if (mdd_is_append(pobj)) + RETURN(-EPERM); + + RETURN(0); } /* * Check whether it may delete the cobj from the pobj. * pobj maybe NULL */ -int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj, - struct mdd_object *cobj, struct lu_attr *cattr, - struct lu_attr *src_attr, int check_perm, int check_empty) +int mdd_may_delete(const struct lu_env *env, struct mdd_object *tpobj, + const struct lu_attr *tpattr, struct mdd_object *tobj, + const struct lu_attr *tattr, const struct lu_attr *cattr, + int check_perm, int check_empty) { - int rc = 0; - ENTRY; - - LASSERT(cobj); - if (!mdd_object_exists(cobj)) - RETURN(-ENOENT); - - if (mdd_is_dead_obj(cobj)) - RETURN(-ESTALE); + int rc = 0; + ENTRY; - if (pobj) { - if (!mdd_object_exists(pobj)) - RETURN(-ENOENT); + if (tpobj) { + LASSERT(tpattr != NULL); + rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm); + if (rc != 0) + RETURN(rc); + } - if (mdd_is_dead_obj(pobj)) - RETURN(-ENOENT); + if (tobj == NULL) + RETURN(0); - if (check_perm) { - rc = mdd_permission_internal_locked(env, pobj, NULL, - MAY_WRITE | MAY_EXEC, - MOR_TGT_PARENT); - if (rc) - RETURN(rc); - } + if (!mdd_object_exists(tobj)) + RETURN(-ENOENT); - if (mdd_is_append(pobj)) - RETURN(-EPERM); - } + if (mdd_is_dead_obj(tobj)) + RETURN(-ESTALE); - if (mdd_is_sticky(env, pobj, cobj)) - RETURN(-EPERM); + if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr)) + RETURN(-EPERM); - if (mdd_is_immutable(cobj) || mdd_is_append(cobj)) - RETURN(-EPERM); + if (mdd_is_immutable(tobj) || mdd_is_append(tobj)) + RETURN(-EPERM); - if ((cattr->la_valid & LA_FLAGS) && - (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) - RETURN(-EPERM); + if ((tattr->la_valid & LA_FLAGS) && + (tattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) + RETURN(-EPERM); /* additional check the rename case */ - if (src_attr) { - if (S_ISDIR(src_attr->la_mode)) { - struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj); + if (cattr) { + if (S_ISDIR(cattr->la_mode)) { + struct mdd_device *mdd = mdo2mdd(&tobj->mod_obj); - if (!S_ISDIR(cattr->la_mode)) + if (!S_ISDIR(tattr->la_mode)) RETURN(-ENOTDIR); - if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid)) + if (lu_fid_eq(mdo2fid(tobj), &mdd->mdd_root_fid)) RETURN(-EBUSY); - } else if (S_ISDIR(cattr->la_mode)) + } else if (S_ISDIR(tattr->la_mode)) RETURN(-EISDIR); } - if (S_ISDIR(cattr->la_mode) && check_empty) - rc = mdd_dir_is_empty(env, cobj); + if (S_ISDIR(tattr->la_mode) && check_empty) + rc = mdd_dir_is_empty(env, tobj); - RETURN(rc); + RETURN(rc); } /* * tgt maybe NULL * has mdd_write_lock on src already, but not on tgt yet */ -int mdd_link_sanity_check(const struct lu_env *env, - struct mdd_object *tgt_obj, - const struct lu_name *lname, - struct mdd_object *src_obj) +static int mdd_link_sanity_check(const struct lu_env *env, + struct mdd_object *tgt_obj, + const struct lu_attr *tattr, + const struct lu_name *lname, + struct mdd_object *src_obj, + const struct lu_attr *cattr) { struct mdd_device *m = mdd_obj2mdd_dev(src_obj); int rc = 0; @@ -468,16 +507,16 @@ int mdd_link_sanity_check(const struct lu_env *env, 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, NULL, 1, 0); - if (rc) - RETURN(rc); - } + 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_link(env, src_obj); + rc = __mdd_may_link(env, src_obj, cattr); - RETURN(rc); + RETURN(rc); } static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object *pobj, @@ -558,60 +597,6 @@ static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj, RETURN(rc); } -int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd, - int reclen, struct thandle *handle) -{ - int rc; - - /* XXX: this is a temporary solution to declare llog changes - * will be fixed in 2.3 with new llog implementation */ - - LASSERT(mdd->mdd_capa); - - /* XXX: Since we use the 'mdd_capa' as fake llog object here, we - * have to set the parameter 'size' as INT_MAX or 0 to inform - * OSD that this record write is for a llog write or catalog - * header update, and osd declare function will reserve less - * credits for optimization purpose. - * - * Reserve 6 blocks for a llog write, since the llog file is - * usually small, reserve 2 blocks for catalog header update, - * because we know for sure that catalog header is already - * allocated. - * - * This hack should be removed in 2.3. - */ - - /* record itself */ - rc = dt_declare_record_write(env, mdd->mdd_capa, - DECLARE_LLOG_WRITE, 0, handle); - if (rc) - return rc; - - /* header will be updated as well */ - rc = dt_declare_record_write(env, mdd->mdd_capa, - DECLARE_LLOG_WRITE, 0, handle); - if (rc) - return rc; - - /* also we should be able to create new plain log */ - rc = dt_declare_create(env, mdd->mdd_capa, NULL, NULL, NULL, handle); - if (rc) - return rc; - - /* new record referencing new plain llog */ - rc = dt_declare_record_write(env, mdd->mdd_capa, - DECLARE_LLOG_WRITE, 0, handle); - if (rc) - return rc; - - /* catalog's header will be updated as well */ - rc = dt_declare_record_write(env, mdd->mdd_capa, - DECLARE_LLOG_REWRITE, 0, handle); - - return rc; -} - int mdd_declare_changelog_store(const struct lu_env *env, struct mdd_device *mdd, const struct lu_name *fname, @@ -630,7 +615,7 @@ int mdd_declare_changelog_store(const struct lu_env *env, reclen = llog_data_len(sizeof(*rec) + (fname != NULL ? fname->ln_namelen : 0)); - buf = mdd_buf_alloc(env, reclen); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); if (buf->lb_buf == NULL) return -ENOMEM; @@ -668,7 +653,7 @@ static int mdd_declare_changelog_ext_store(const struct lu_env *env, reclen = llog_data_len(sizeof(*rec) + (tname != NULL ? tname->ln_namelen : 0) + (sname != NULL ? 1 + sname->ln_namelen : 0)); - buf = mdd_buf_alloc(env, reclen); + buf = lu_buf_check_and_alloc(&mdd_env_info(env)->mti_big_buf, reclen); if (buf->lb_buf == NULL) return -ENOMEM; @@ -770,14 +755,10 @@ int mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd, * \param tname - target name string * \param handle - transacion handle */ -static 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) +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; @@ -797,7 +778,7 @@ static int mdd_changelog_ns_store(const struct lu_env *env, LASSERT(handle != NULL); reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen); - buf = mdd_buf_alloc(env, reclen); + 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; @@ -865,7 +846,7 @@ static int mdd_changelog_ext_ns_store(const struct lu_env *env, reclen = llog_data_len(sizeof(*rec) + sname != NULL ? 1 + sname->ln_namelen : 0); - buf = mdd_buf_alloc(env, reclen); + 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; @@ -901,113 +882,432 @@ static int mdd_changelog_ext_ns_store(const struct lu_env *env, return 0; } -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 thandle *handle) +static int __mdd_links_add(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata, + const struct lu_name *lname, + const struct lu_fid *pfid, + int first, int check) { - int rc; - - rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle); - if (rc) - return rc; + int rc; - rc = mdo_declare_ref_add(env, c, handle); - if (rc) - return rc; + if (ldata->ld_leh == NULL) { + rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata); + if (rc) { + if (rc != -ENODATA) + return rc; + rc = linkea_data_new(ldata, + &mdd_env_info(env)->mti_link_buf); + if (rc) + return rc; + } + } - rc = mdo_declare_attr_set(env, p, NULL, handle); - if (rc) - return rc; + if (check) { + rc = linkea_links_find(ldata, lname, pfid); + if (rc && rc != -ENOENT) + return rc; + if (rc == 0) + return -EEXIST; + } - rc = mdo_declare_attr_set(env, c, NULL, handle); - if (rc) - return rc; + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) { + struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2; - rc = mdd_declare_links_add(env, c, handle); - if (rc) - return rc; + *tfid = *pfid; + tfid->f_ver = ~0; + linkea_add_buf(ldata, lname, tfid); + } - rc = mdd_declare_changelog_store(env, mdd, name, handle); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2)) + linkea_add_buf(ldata, lname, pfid); - return rc; + return linkea_add_buf(ldata, lname, pfid); } -static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, - struct md_object *src_obj, const struct lu_name *lname, - struct md_attr *ma) +static int __mdd_links_del(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct linkea_data *ldata, + const struct lu_name *lname, + const struct lu_fid *pfid) { - const char *name = lname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_tobj = md2mdd_obj(tgt_obj); - struct mdd_object *mdd_sobj = md2mdd_obj(src_obj); - struct mdd_device *mdd = mdo2mdd(src_obj); - struct dynlock_handle *dlh; - struct thandle *handle; - int rc; - ENTRY; - - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - GOTO(out_pending, rc = PTR_ERR(handle)); + int rc; - rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle); - if (rc) - GOTO(stop, rc); + if (ldata->ld_leh == NULL) { + rc = mdd_links_read(env, mdd_obj, ldata); + if (rc) + return rc; + } - rc = mdd_trans_start(env, mdd, handle); - if (rc) - GOTO(stop, rc); + rc = linkea_links_find(ldata, lname, pfid); + if (rc) + return rc; - dlh = mdd_pdo_write_lock(env, mdd_tobj, name, MOR_TGT_CHILD); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD); + linkea_del_buf(ldata, lname); + return 0; +} - rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj); - if (rc) - GOTO(out_unlock, rc); +static int mdd_linkea_prepare(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *oldpfid, + const struct lu_name *oldlname, + const struct lu_fid *newpfid, + const struct lu_name *newlname, + int first, int check, + struct linkea_data *ldata) +{ + int rc = 0; + int rc2 = 0; + ENTRY; - rc = mdo_ref_add(env, mdd_sobj, handle); - if (rc) - GOTO(out_unlock, rc); + if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) + return 0; + LASSERT(oldpfid != NULL || newpfid != NULL); - rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), - name, handle, - mdd_object_capa(env, mdd_tobj)); - if (rc != 0) { - mdo_ref_del(env, mdd_sobj, handle); - GOTO(out_unlock, rc); + 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 */ + RETURN(0); } - LASSERT(ma->ma_attr.la_valid & LA_CTIME); - la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + if (oldpfid != NULL) { + rc = __mdd_links_del(env, mdd_obj, ldata, oldlname, oldpfid); + if (rc) { + 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; + + RETURN(rc); +} + +int mdd_links_rename(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *oldpfid, + const struct lu_name *oldlname, + const struct lu_fid *newpfid, + const struct lu_name *newlname, + struct thandle *handle, + struct linkea_data *ldata, + int first, int check) +{ + int rc2 = 0; + int rc = 0; + ENTRY; + + if (ldata == NULL) { + 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) + GOTO(out, rc); + } + + if (ldata->ld_reclen != 0) + rc = mdd_links_write(env, mdd_obj, ldata, handle); + EXIT; +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, + 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))); + 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))); + } + + if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + lu_buf_free(ldata->ld_buf); + + return rc; +} + +static inline int mdd_links_add(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *pfid, + const struct lu_name *lname, + struct thandle *handle, + struct linkea_data *data, int first) +{ + return mdd_links_rename(env, mdd_obj, NULL, NULL, + pfid, lname, handle, data, first, 0); +} + +static inline int mdd_links_del(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *pfid, + const struct lu_name *lname, + struct thandle *handle) +{ + return mdd_links_rename(env, mdd_obj, pfid, lname, + NULL, NULL, handle, NULL, 0, 0); +} + +/** 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; + + /* 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; + + if (!mdd_object_exists(mdd_obj)) + return -ENODATA; + + 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) + return rc; + + return linkea_init(ldata); +} - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); +/** 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) + */ +struct lu_buf *mdd_links_get(const struct lu_env *env, + struct mdd_object *mdd_obj) +{ + struct linkea_data ldata = { 0 }; + int rc; + + rc = mdd_links_read(env, mdd_obj, &ldata); + return rc ? ERR_PTR(rc) : ldata.ld_buf; +} + +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)); +} + +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; + void *linkea; + + if (ldata != NULL && ldata->ld_lee != NULL) { + ea_len = ldata->ld_leh->leh_len; + linkea = ldata->ld_buf->lb_buf; + } else { + ea_len = DEFAULT_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; +} + +static inline int mdd_declare_links_del(const struct lu_env *env, + struct mdd_object *c, + struct thandle *handle) +{ + int rc = 0; + + /* For directory, the linkEA will be removed together + * with the object. */ + if (!S_ISDIR(mdd_object_type(c))) + rc = mdd_declare_links_add(env, c, handle, NULL); + + return rc; +} + +static int mdd_declare_link(const struct lu_env *env, + struct mdd_device *mdd, + struct mdd_object *p, + struct mdd_object *c, + const struct lu_name *name, + struct thandle *handle, + struct lu_attr *la, + struct linkea_data *data) +{ + int rc; + + rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle); if (rc) - GOTO(out_unlock, rc); + return rc; - la->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0); - if (rc == 0) { - mdd_links_add(env, mdd_sobj, - mdo2fid(mdd_tobj), lname, handle, 0); - } + rc = mdo_declare_ref_add(env, c, handle); + if (rc) + return rc; + + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdo_declare_attr_set(env, p, la, handle); + if (rc != 0) + return rc; + + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, c, la, handle); + if (rc) + return rc; + + rc = mdd_declare_links_add(env, c, handle, data); + if (rc) + return rc; + + rc = mdd_declare_changelog_store(env, mdd, name, handle); + + return rc; +} + +static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, + struct md_object *src_obj, const struct lu_name *lname, + struct md_attr *ma) +{ + const char *name = lname->ln_name; + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + struct mdd_object *mdd_tobj = md2mdd_obj(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 linkea_data *ldata = &mdd_env_info(env)->mti_link_data; + int rc; + ENTRY; + rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA); + if (rc != 0) + RETURN(rc); + + rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); + if (rc != 0) + RETURN(rc); + + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + GOTO(out_pending, rc = PTR_ERR(handle)); + + memset(ldata, 0, sizeof(*ldata)); + + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + + rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, handle, + la, ldata); + if (rc) + GOTO(stop, rc); + + rc = mdd_trans_start(env, mdd, handle); + if (rc) + GOTO(stop, rc); + + mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD); + rc = mdd_link_sanity_check(env, mdd_tobj, tattr, lname, mdd_sobj, + cattr); + if (rc) + GOTO(out_unlock, rc); + + rc = mdo_ref_add(env, mdd_sobj, handle); + if (rc) + GOTO(out_unlock, rc); + + + rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), + name, handle, + mdd_object_capa(env, mdd_tobj)); + if (rc != 0) { + mdo_ref_del(env, mdd_sobj, handle); + GOTO(out_unlock, rc); + } + + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdd_update_time(env, mdd_tobj, tattr, la, handle); + if (rc) + GOTO(out_unlock, rc); + + 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; out_unlock: mdd_write_unlock(env, mdd_sobj); - mdd_pdo_write_unlock(env, mdd_tobj, dlh); -out_trans: if (rc == 0) rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj, mdd_tobj, lname, handle); stop: - mdd_trans_stop(env, mdd, rc, handle); + mdd_trans_stop(env, mdd, rc, handle); + + if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + lu_buf_free(ldata->ld_buf); out_pending: return rc; } @@ -1037,8 +1337,9 @@ int mdd_finish_unlink(const struct lu_env *env, LASSERT(mdd_write_locked(env, obj) != 0); - if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) { + 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) { @@ -1067,170 +1368,227 @@ int mdd_finish_unlink(const struct lu_env *env, * has mdd_write_lock on cobj already, but not on pobj yet */ int mdd_unlink_sanity_check(const struct lu_env *env, struct mdd_object *pobj, - struct mdd_object *cobj, struct lu_attr *cattr) + const struct lu_attr *pattr, + struct mdd_object *cobj, + const struct lu_attr *cattr) { - int rc; - ENTRY; - - rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1); - - RETURN(rc); -} - -static inline int mdd_declare_links_del(const struct lu_env *env, - struct mdd_object *c, - struct thandle *handle) -{ - int rc = 0; + int rc; + ENTRY; - /* For directory, the linkEA will be removed together with the object. */ - if (!S_ISDIR(mdd_object_type(c))) - rc = mdd_declare_links_add(env, c, handle); + rc = mdd_may_delete(env, pobj, pattr, cobj, cattr, NULL, 1, 1); - return rc; + RETURN(rc); } 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) + struct mdd_object *p, struct mdd_object *c, + const struct lu_name *name, struct md_attr *ma, + struct thandle *handle, int no_name) { + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; int rc; - rc = mdo_declare_index_delete(env, p, name->ln_name, handle); - if (rc) - return rc; + if (likely(no_name == 0)) { + rc = mdo_declare_index_delete(env, p, name->ln_name, handle); + if (rc) + return rc; + } rc = mdo_declare_ref_del(env, p, handle); if (rc) return rc; - rc = mdo_declare_ref_del(env, c, handle); - if (rc) - return rc; + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdo_declare_attr_set(env, p, la, handle); + if (rc) + return rc; - rc = mdo_declare_ref_del(env, c, handle); - if (rc) - return rc; + if (c != NULL) { + rc = mdo_declare_ref_del(env, c, handle); + if (rc) + return rc; - rc = mdo_declare_attr_set(env, p, NULL, handle); - if (rc) - return rc; + rc = mdo_declare_ref_del(env, c, handle); + if (rc) + return rc; - rc = mdo_declare_attr_set(env, c, NULL, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, c, la, handle); + if (rc) + return rc; - rc = mdd_declare_finish_unlink(env, c, ma, handle); - if (rc) - return rc; + rc = mdd_declare_finish_unlink(env, c, ma, handle); + if (rc) + return rc; - rc = mdd_declare_links_del(env, c, handle); - if (rc != 0) - return rc; + rc = mdd_declare_links_del(env, c, handle); + if (rc != 0) + return rc; - rc = mdd_declare_changelog_store(env, mdd, name, handle); + /* FIXME: need changelog for remove entry */ + rc = mdd_declare_changelog_store(env, mdd, name, handle); + } return rc; } -static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, - struct md_object *cobj, const struct lu_name *lname, - struct md_attr *ma) +/* + * test if a file has an HSM archive + * if HSM attributes are not found in ma update them from + * HSM xattr + */ +static bool mdd_hsm_archive_exists(const struct lu_env *env, + struct mdd_object *obj, + struct md_attr *ma) { - const char *name = lname->ln_name; - struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_pobj = md2mdd_obj(pobj); - struct mdd_object *mdd_cobj = md2mdd_obj(cobj); - struct mdd_device *mdd = mdo2mdd(pobj); - struct dynlock_handle *dlh; - struct thandle *handle; - int rc, is_dir; - ENTRY; + ENTRY; - if (mdd_object_exists(mdd_cobj) <= 0) - RETURN(-ENOENT); + if (!(ma->ma_valid & MA_HSM)) { + /* no HSM MD provided, read xattr */ + struct lu_buf *hsm_buf; + const size_t buflen = sizeof(struct hsm_attrs); + int rc; + + 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 = lustre_buf2hsm(hsm_buf->lb_buf, rc, &ma->ma_hsm); + lu_buf_free(hsm_buf); + if (rc < 0) + RETURN(false); - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - RETURN(PTR_ERR(handle)); + ma->ma_valid = MA_HSM; + } + if (ma->ma_hsm.mh_flags & HS_EXISTS) + RETURN(true); + RETURN(false); +} - rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj, - lname, ma, handle); - if (rc) - GOTO(stop, rc); +/** + * Delete name entry and the object. + * Note: no_name == 1 means it only destory the object, i.e. name_entry + * does not exist for this object, and it could only happen during resending + * of remote unlink. see the comments in mdt_reint_unlink. Unfortunately, lname + * is also needed in this case(needed by changelog), so we have to add another + * parameter(no_name)here. XXX: this is only needed in DNE phase I, on Phase II, + * the ENOENT failure should be able to be fixed by redo mechanism. + */ +static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, + struct md_object *cobj, const struct lu_name *lname, + struct md_attr *ma, int no_name) +{ + const char *name = lname->ln_name; + struct lu_attr *pattr = MDD_ENV_VAR(env, pattr); + struct lu_attr *cattr = MDD_ENV_VAR(env, cattr); + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + struct mdd_object *mdd_pobj = md2mdd_obj(pobj); + struct mdd_object *mdd_cobj = NULL; + struct mdd_device *mdd = mdo2mdd(pobj); + struct thandle *handle; + int rc, is_dir = 0; + ENTRY; + + /* cobj == NULL means only delete name entry */ + if (likely(cobj != NULL)) { + mdd_cobj = md2mdd_obj(cobj); + if (mdd_object_exists(mdd_cobj) == 0) + RETURN(-ENOENT); + } + + rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA); + if (rc) + RETURN(rc); - rc = mdd_trans_start(env, mdd, handle); - if (rc) - GOTO(stop, rc); + if (likely(mdd_cobj != NULL)) { + /* fetch cattr */ + rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA); + if (rc) + RETURN(rc); - dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(stop, rc = -ENOMEM); - mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD); + is_dir = S_ISDIR(cattr->la_mode); + } - /* fetch cattr */ - rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj)); - if (rc) - GOTO(cleanup, rc); + rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr); + if (rc) + RETURN(rc); - is_dir = S_ISDIR(cattr->la_mode); + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + RETURN(PTR_ERR(handle)); - rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr); - if (rc) - GOTO(cleanup, rc); + rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj, + lname, ma, handle, no_name); + if (rc) + GOTO(stop, rc); - rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle, - mdd_object_capa(env, mdd_pobj)); + rc = mdd_trans_start(env, mdd, handle); if (rc) - GOTO(cleanup, rc); + GOTO(stop, rc); - rc = mdo_ref_del(env, mdd_cobj, handle); - if (rc != 0) { - __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj), - name, handle, + 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)); - GOTO(cleanup, rc); + if (rc) + GOTO(cleanup, rc); } - if (is_dir) - /* unlink dot */ - mdo_ref_del(env, mdd_cobj, handle); + 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)); + GOTO(cleanup, rc); + } + + if (is_dir) + /* unlink dot */ + mdo_ref_del(env, mdd_cobj, handle); - /* fetch updated nlink */ - rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj)); + /* fetch updated nlink */ + rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA); + if (rc) + GOTO(cleanup, rc); + } + + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; + + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdd_update_time(env, mdd_pobj, pattr, la, handle); if (rc) GOTO(cleanup, rc); - LASSERT(ma->ma_attr.la_valid & LA_CTIME); - la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; - - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0); - if (rc) - GOTO(cleanup, rc); + /* Enough for only unlink the entry */ + if (unlikely(mdd_cobj == NULL)) + GOTO(stop, rc); if (cattr->la_nlink > 0 || mdd_cobj->mod_count > 0) { - /* update ctime of an unlinked file only if it is still - * opened or a link still exists */ - la->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal(env, mdd_cobj, la, handle, 0); - if (rc) - GOTO(cleanup, rc); - } + /* update ctime of an unlinked file only if it is still + * opened or a link still exists */ + la->la_valid = LA_CTIME; + rc = mdd_update_time(env, mdd_cobj, cattr, la, handle); + if (rc) + GOTO(cleanup, rc); + } /* 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, handle); /* fetch updated nlink */ if (rc == 0) - rc = mdd_la_get(env, mdd_cobj, cattr, - mdd_object_capa(env, mdd_cobj)); + rc = mdd_la_get(env, mdd_cobj, cattr, BYPASS_CAPA); - if (!is_dir) + if (!is_dir) /* old files may not have link ea; ignore errors */ mdd_links_del(env, mdd_cobj, mdo2fid(mdd_pobj), lname, handle); @@ -1239,27 +1597,28 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, ma->ma_attr = *cattr; ma->ma_valid |= MA_INODE; } - EXIT; + EXIT; cleanup: - mdd_write_unlock(env, mdd_cobj); - mdd_pdo_write_unlock(env, mdd_pobj, dlh); - if (rc == 0) { - int cl_flags; - - cl_flags = (cattr->la_nlink == 0) ? CLF_UNLINK_LAST : 0; - if ((ma->ma_valid & MA_HSM) && - (ma->ma_hsm.mh_flags & HS_EXISTS)) - cl_flags |= CLF_UNLINK_HSM_EXISTS; + mdd_write_unlock(env, mdd_cobj); + if (rc == 0) { + int cl_flags = 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; + } rc = mdd_changelog_ns_store(env, mdd, is_dir ? CL_RMDIR : CL_UNLINK, cl_flags, mdd_cobj, mdd_pobj, lname, handle); - } + } stop: - mdd_trans_stop(env, mdd, rc, handle); + mdd_trans_stop(env, mdd, rc, handle); - return rc; + return rc; } /* @@ -1275,33 +1634,27 @@ static int mdd_cd_sanity_check(const struct lu_env *env, RETURN(-ENOENT); 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) { - struct mdd_device *mdd = mdo2mdd(cobj); - struct mdd_object *mdd_pobj = md2mdd_obj(pobj); - struct mdd_object *son = md2mdd_obj(cobj); - struct thandle *handle; + struct mdd_device *mdd = mdo2mdd(cobj); + struct mdd_object *mdd_pobj = md2mdd_obj(pobj); + struct mdd_object *son = md2mdd_obj(cobj); + struct thandle *handle; const struct lu_buf *buf; - struct lu_attr *attr = &mdd_env_info(env)->mti_cattr; - int rc; - ENTRY; - - /* do not let users to create stripes via .lustre/ - * mdd_obf_setup() sets IMMUTE_OBJ on this directory */ - if (pobj && mdd_pobj->mod_flags & IMMUTE_OBJ) - RETURN(-ENOENT); + struct lu_attr *attr = MDD_ENV_VAR(env, cattr); + int rc; + ENTRY; - rc = mdd_cd_sanity_check(env, son); - if (rc) - RETURN(rc); + rc = mdd_cd_sanity_check(env, son); + if (rc) + RETURN(rc); - if (!md_should_create(spec->sp_cr_flags)) - RETURN(0); + if (!md_should_create(spec->sp_cr_flags)) + RETURN(0); /* * there are following use cases for this function: @@ -1309,12 +1662,12 @@ 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, mdd_object_capa(env, son)); + rc = mdd_la_get(env, son, attr, BYPASS_CAPA); if (rc) RETURN(rc); /* calling ->ah_make_hint() is used to transfer information from parent */ - mdd_object_make_hint(env, mdd_pobj, son, attr); + mdd_object_make_hint(env, mdd_pobj, son, attr, spec); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) @@ -1324,12 +1677,11 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, * 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 %Lo, no_create %u\n", + CDEBUG(D_OTHER, "ea %p/%u, cr_flags "LPO64", no_create %u\n", spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen, spec->sp_cr_flags, spec->no_create); - if (spec->no_create || spec->sp_cr_flags & MDS_OPEN_HAS_EA) { - /* replay case or lfs setstripe */ + if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) { buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, spec->u.sp_ea.eadatalen); } else { @@ -1341,74 +1693,37 @@ 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); + if (rc) + GOTO(stop, rc); + rc = mdd_trans_start(env, mdd, handle); if (rc) GOTO(stop, rc); rc = dt_xattr_set(env, mdd_object_child(son), buf, XATTR_NAME_LOV, 0, handle, mdd_object_capa(env, son)); + + if (rc) + GOTO(stop, rc); + + rc = mdd_changelog_data_store(env, mdd, CL_LAYOUT, 0, son, handle); + stop: mdd_trans_stop(env, mdd, rc, handle); out_free: RETURN(rc); } -/* Get fid from name and parent */ -static int -__mdd_lookup(const struct lu_env *env, struct md_object *pobj, - const struct lu_name *lname, struct lu_fid* fid, int mask) -{ - const char *name = lname->ln_name; - const struct dt_key *key = (const struct dt_key *)name; - struct mdd_object *mdd_obj = md2mdd_obj(pobj); - struct mdd_device *m = mdo2mdd(pobj); - struct dt_object *dir = mdd_object_child(mdd_obj); - int rc; - ENTRY; - - if (unlikely(mdd_is_dead_obj(mdd_obj))) - RETURN(-ESTALE); - - rc = mdd_object_exists(mdd_obj); - if (unlikely(rc == 0)) - RETURN(-ESTALE); - else if (unlikely(rc < 0)) { - CERROR("Object "DFID" locates on remote server\n", - PFID(mdo2fid(mdd_obj))); - RETURN(-EINVAL); - } - - /* 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, NULL, 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 - rc = -ENOTDIR; - - RETURN(rc); -} - -int mdd_declare_object_initialize(const struct lu_env *env, - struct mdd_object *child, - struct lu_attr *attr, - struct thandle *handle) +static int mdd_declare_object_initialize(const struct lu_env *env, + struct mdd_object *parent, + struct mdd_object *child, + struct lu_attr *attr, + struct thandle *handle, + struct linkea_data *ldata) { int rc; + ENTRY; /* * inode mode has been set in creation time, and it's based on umask, @@ -1425,18 +1740,24 @@ int mdd_declare_object_initialize(const struct lu_env *env, 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 (rc == 0) - mdd_declare_links_add(env, child, handle); + if (rc == 0) + mdd_declare_links_add(env, child, handle, ldata); - return rc; + RETURN(rc); } -int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, - const struct lu_name *lname, struct mdd_object *child, - struct lu_attr *attr, struct thandle *handle, - const struct md_op_spec *spec) +static int mdd_object_initialize(const struct lu_env *env, + const struct lu_fid *pfid, + const struct lu_name *lname, + struct mdd_object *child, + struct lu_attr *attr, struct thandle *handle, + const struct md_op_spec *spec, + struct linkea_data *ldata) { int rc; ENTRY; @@ -1449,20 +1770,8 @@ int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, * (2) maybe, the child attributes should be set in OSD when creation. */ - /* - * 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 = mdd_attr_set_internal(env, child, attr, handle, 0); /* arguments are supposed to stay the same */ - attr->la_valid |= LA_MODE | LA_TYPE; - if (rc != 0) - RETURN(rc); - if (S_ISDIR(attr->la_mode)) { /* Add "." and ".." for newly created dir */ mdo_ref_add(env, child, handle); @@ -1475,16 +1784,17 @@ int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, if (rc != 0) mdo_ref_del(env, child, handle); } - if (rc == 0) - mdd_links_add(env, child, pfid, lname, handle, 1); - RETURN(rc); + if (rc == 0) + mdd_links_add(env, child, pfid, lname, handle, ldata, 1); + + RETURN(rc); } /* has not lock on pobj yet */ static int mdd_create_sanity_check(const struct lu_env *env, struct md_object *pobj, - struct lu_attr *pattr, + const struct lu_attr *pattr, const struct lu_name *lname, struct lu_attr *cattr, struct md_op_spec *spec) @@ -1511,21 +1821,21 @@ static int mdd_create_sanity_check(const struct lu_env *env, * _index_insert also, for avoiding rolling back if exists * _index_insert. */ - rc = __mdd_lookup_locked(env, pobj, lname, fid, - MAY_WRITE | MAY_EXEC); + 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. - */ + /* + * 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) + RETURN(rc); + } /* sgid check */ if (pattr->la_mode & S_ISGID) { @@ -1564,9 +1874,11 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, struct mdd_object *p, struct mdd_object *c, const struct lu_name *name, struct lu_attr *attr, - int got_def_acl, struct thandle *handle, - const struct md_op_spec *spec) + const struct md_op_spec *spec, + struct linkea_data *ldata, + struct lu_buf *def_acl_buf, + struct lu_buf *acl_buf) { int rc; @@ -1575,19 +1887,16 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, GOTO(out, rc); #ifdef CONFIG_FS_POSIX_ACL - if (got_def_acl > 0) { - struct lu_buf *acl_buf; - - acl_buf = mdd_buf_get(env, NULL, got_def_acl); + if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) { /* if dir, then can inherit default ACl */ - if (S_ISDIR(attr->la_mode)) { - rc = mdo_declare_xattr_set(env, c, acl_buf, - XATTR_NAME_ACL_DEFAULT, - 0, handle); - if (rc) - GOTO(out, rc); - } + rc = mdo_declare_xattr_set(env, c, def_acl_buf, + XATTR_NAME_ACL_DEFAULT, + 0, handle); + if (rc) + GOTO(out, rc); + } + if (acl_buf->lb_len > 0) { rc = mdo_declare_attr_set(env, c, attr, handle); if (rc) GOTO(out, rc); @@ -1605,7 +1914,7 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, GOTO(out, rc); } - rc = mdd_declare_object_initialize(env, c, attr, handle); + rc = mdd_declare_object_initialize(env, p, c, attr, handle, ldata); if (rc) GOTO(out, rc); @@ -1618,13 +1927,14 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, GOTO(out, rc); /* replay case, create LOV EA from client data */ - if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) { + 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, handle); + rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, 0, + handle); if (rc) GOTO(out, rc); } @@ -1638,7 +1948,11 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, } if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) { - rc = mdo_declare_attr_set(env, p, attr, handle); + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + + *la = *attr; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdo_declare_attr_set(env, p, la, handle); if (rc) return rc; } @@ -1647,10 +1961,57 @@ static int mdd_declare_create(const struct lu_env *env, struct mdd_device *mdd, if (rc) return rc; + /* XXX: For remote create, it should indicate the remote RPC + * will be sent after local transaction is finished, which + * is not very nice, but it will be removed once we fully support + * async update */ + if (mdd_object_remote(p) && handle->th_update != NULL) + handle->th_update->tu_sent_after_local_trans = 1; out: - return rc; + return rc; } +static int mdd_acl_init(const struct lu_env *env, struct mdd_object *pobj, + struct lu_attr *la, struct lu_buf *def_acl_buf, + struct lu_buf *acl_buf) +{ + int rc; + ENTRY; + + if (S_ISLNK(la->la_mode)) { + acl_buf->lb_len = 0; + def_acl_buf->lb_len = 0; + RETURN(0); + } + + mdd_read_lock(env, pobj, MOR_TGT_PARENT); + rc = mdo_xattr_get(env, pobj, def_acl_buf, + XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA); + mdd_read_unlock(env, pobj); + if (rc > 0) { + /* If there are default ACL, fix mode/ACL by default ACL */ + def_acl_buf->lb_len = rc; + LASSERT(def_acl_buf->lb_len <= acl_buf->lb_len); + memcpy(acl_buf->lb_buf, def_acl_buf->lb_buf, rc); + acl_buf->lb_len = rc; + rc = __mdd_fix_mode_acl(env, acl_buf, &la->la_mode); + if (rc < 0) + RETURN(rc); + } else if (rc == -ENODATA || rc == -EOPNOTSUPP) { + /* If there are no default ACL, fix mode by mask */ + struct lu_ucred *uc = lu_ucred(env); + + /* The create triggered by MDT internal events, such as + * LFSCK reset, will not contain valid "uc". */ + if (unlikely(uc != NULL)) + la->la_mode &= ~uc->uc_umask; + rc = 0; + acl_buf->lb_len = 0; + def_acl_buf->lb_len = 0; + } + + RETURN(rc); +} /* * Create object and insert it into namespace. @@ -1667,10 +2028,11 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, struct lu_attr *attr = &ma->ma_attr; struct thandle *handle; struct lu_attr *pattr = &info->mti_pattr; - struct dynlock_handle *dlh; + struct lu_buf acl_buf; + struct lu_buf def_acl_buf; + struct linkea_data *ldata = &info->mti_link_data; const char *name = lname->ln_name; int rc, created = 0, initialized = 0, inserted = 0; - int got_def_acl = 0; ENTRY; /* @@ -1713,37 +2075,34 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, if (rc != 0) RETURN(rc); - /* Sanity checks before big job. */ + /* Sanity checks before big job. */ rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec); - if (rc) - RETURN(rc); + if (rc) + RETURN(rc); if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET)) GOTO(out_free, rc = -EINPROGRESS); - if (!S_ISLNK(attr->la_mode)) { - struct lu_buf *acl_buf; - - acl_buf = mdd_buf_get(env, info->mti_xattr_buf, - sizeof(info->mti_xattr_buf)); - mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT); - rc = mdo_xattr_get(env, mdd_pobj, acl_buf, - XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA); - mdd_read_unlock(env, mdd_pobj); - if (rc > 0) - got_def_acl = rc; - else if (rc < 0 && rc != -EOPNOTSUPP && rc != -ENODATA) - GOTO(out_free, rc); - } + acl_buf.lb_buf = info->mti_xattr_buf; + acl_buf.lb_len = sizeof(info->mti_xattr_buf); + def_acl_buf.lb_buf = info->mti_key; + def_acl_buf.lb_len = sizeof(info->mti_key); + rc = mdd_acl_init(env, mdd_pobj, attr, &def_acl_buf, &acl_buf); + if (rc < 0) + GOTO(out_free, rc); - mdd_object_make_hint(env, mdd_pobj, son, attr); + mdd_object_make_hint(env, mdd_pobj, son, attr, spec); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_free, rc = PTR_ERR(handle)); + memset(ldata, 0, sizeof(*ldata)); + 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, - got_def_acl, handle, spec); + handle, spec, ldata, &def_acl_buf, &acl_buf); if (rc) GOTO(out_stop, rc); @@ -1751,10 +2110,6 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, if (rc) GOTO(out_stop, rc); - dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - mdd_write_lock(env, son, MOR_TGT_CHILD); rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec); if (rc) { @@ -1765,11 +2120,21 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, created = 1; #ifdef CONFIG_FS_POSIX_ACL - if (got_def_acl) { - struct lu_buf *acl_buf; - - acl_buf = mdd_buf_get(env, info->mti_xattr_buf, got_def_acl); - rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle); + if (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); + if (rc) { + mdd_write_unlock(env, son); + GOTO(cleanup, rc); + } + } + /* set its own acl */ + if (acl_buf.lb_len > 0) { + rc = mdo_xattr_set(env, son, &acl_buf, + XATTR_NAME_ACL_ACCESS, + 0, handle, BYPASS_CAPA); if (rc) { mdd_write_unlock(env, son); GOTO(cleanup, rc); @@ -1778,7 +2143,7 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, #endif rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname, - son, attr, handle, spec); + son, attr, handle, spec, ldata); /* * in case of replay we just set LOVEA provided by the client @@ -1787,13 +2152,14 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, * probably this way we code can be made better. */ if (rc == 0 && (spec->no_create || - (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) { + (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); + BYPASS_CAPA); } if (rc == 0 && spec->sp_cr_flags & MDS_OPEN_VOLATILE) @@ -1847,7 +2213,7 @@ static int mdd_create(const struct lu_env *env, struct md_object *pobj, /* update parent directory mtime/ctime */ *la = *attr; la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0); + rc = mdd_update_time(env, mdd_pobj, pattr, la, handle); if (rc) GOTO(cleanup, rc); @@ -1888,9 +2254,7 @@ cleanup: mdd_write_unlock(env, son); } - mdd_pdo_write_unlock(env, mdd_pobj, dlh); -out_trans: - if (rc == 0 && fid_is_client_mdt_visible(mdo2fid(son))) + if (rc == 0 && fid_is_namespace_visible(mdo2fid(son))) rc = mdd_changelog_ns_store(env, mdd, S_ISDIR(attr->la_mode) ? CL_MKDIR : S_ISREG(attr->la_mode) ? CL_CREATE : @@ -1899,6 +2263,10 @@ out_trans: out_stop: mdd_trans_stop(env, mdd, rc, handle); out_free: + if (ldata->ld_buf && ldata->ld_buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + lu_buf_free(ldata->ld_buf); + /* The child object shouldn't be cached anymore */ if (rc) set_bit(LU_OBJECT_HEARD_BANSHEE, @@ -1919,6 +2287,7 @@ enum rename_order { static int mdd_rename_order(const struct lu_env *env, struct mdd_device *mdd, struct mdd_object *src_pobj, + const struct lu_attr *pattr, struct mdd_object *tgt_pobj) { /* order of locking, 1 - tgt-src, 0 - src-tgt*/ @@ -1934,7 +2303,8 @@ static int mdd_rename_order(const struct lu_env *env, } else if (lu_fid_eq(&mdd->mdd_root_fid, mdo2fid(tgt_pobj))) { rc = MDD_RN_TGTSRC; } else { - rc = mdd_is_parent(env, mdd, src_pobj, mdo2fid(tgt_pobj), NULL); + rc = mdd_is_parent(env, mdd, src_pobj, pattr, mdo2fid(tgt_pobj), + NULL); if (rc == -EREMOTE) rc = 0; @@ -1950,55 +2320,61 @@ static int mdd_rename_order(const struct lu_env *env, /* has not mdd_write{read}_lock on any obj yet. */ static int mdd_rename_sanity_check(const struct lu_env *env, struct mdd_object *src_pobj, + const struct lu_attr *pattr, struct mdd_object *tgt_pobj, + const struct lu_attr *tpattr, struct mdd_object *sobj, + const struct lu_attr *cattr, struct mdd_object *tobj, - struct lu_attr *so_attr, - struct lu_attr *tg_attr) + const struct lu_attr *tattr) { - int rc = 0; - ENTRY; + int rc = 0; + ENTRY; - /* XXX: when get here, sobj must NOT be NULL, - * the other case has been processed in cml_rename - * before mdd_rename and enable MDS_PERM_BYPASS. */ - LASSERT(sobj); + /* XXX: when get here, sobj must NOT be NULL, + * the other case has been processed in cld_rename + * before mdd_rename and enable MDS_PERM_BYPASS. */ + LASSERT(sobj); - rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0); - if (rc) - RETURN(rc); + rc = mdd_may_delete(env, src_pobj, pattr, sobj, cattr, NULL, 1, 0); + if (rc) + RETURN(rc); - /* XXX: when get here, "tobj == NULL" means tobj must - * NOT exist (neither on remote MDS, such case has been - * processed in cml_rename before mdd_rename and enable - * MDS_PERM_BYPASS). - * So check may_create, but not check may_unlink. */ - if (!tobj) - rc = mdd_may_create(env, tgt_pobj, NULL, - (src_pobj != tgt_pobj), 0); - else - rc = mdd_may_delete(env, tgt_pobj, tobj, tg_attr, so_attr, - (src_pobj != tgt_pobj), 1); - - if (!rc && !tobj && (src_pobj != tgt_pobj) && - S_ISDIR(so_attr->la_mode)) - rc = __mdd_may_link(env, tgt_pobj); + /* XXX: when get here, "tobj == NULL" means tobj must + * NOT exist (neither on remote MDS, such case has been + * processed in cld_rename before mdd_rename and enable + * MDS_PERM_BYPASS). + * So check may_create, but not check may_unlink. */ + if (!tobj) + rc = mdd_may_create(env, tgt_pobj, tpattr, NULL, + (src_pobj != tgt_pobj), 0); + else + rc = mdd_may_delete(env, tgt_pobj, tpattr, tobj, tattr, cattr, + (src_pobj != tgt_pobj), 1); - RETURN(rc); + if (!rc && !tobj && (src_pobj != tgt_pobj) && S_ISDIR(cattr->la_mode)) + rc = __mdd_may_link(env, tgt_pobj, tpattr); + + RETURN(rc); } static int mdd_declare_rename(const struct lu_env *env, - struct mdd_device *mdd, - struct mdd_object *mdd_spobj, - struct mdd_object *mdd_tpobj, - struct mdd_object *mdd_sobj, - struct mdd_object *mdd_tobj, - const struct lu_name *tname, + struct mdd_device *mdd, + struct mdd_object *mdd_spobj, + 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, - struct md_attr *ma, - struct thandle *handle) + struct md_attr *ma, + struct linkea_data *ldata, + struct thandle *handle) { - int rc; + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + int rc; + + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; LASSERT(mdd_spobj); LASSERT(mdd_tpobj); @@ -2016,38 +2392,42 @@ static int mdd_declare_rename(const struct lu_env *env, rc = mdo_declare_ref_del(env, mdd_spobj, handle); if (rc) return rc; + if (mdd_spobj != mdd_tpobj) { + rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, + handle); + if (rc) + return rc; - rc = mdo_declare_index_delete(env, mdd_sobj, dotdot, handle); - if (rc) - return rc; - - rc = mdo_declare_index_insert(env, mdd_sobj, mdo2fid(mdd_tpobj), - dotdot, handle); - if (rc) - return rc; - + rc = mdo_declare_index_insert(env, mdd_sobj, + mdo2fid(mdd_tpobj), + dotdot, 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) return rc; - } - rc = mdo_declare_attr_set(env, mdd_spobj, NULL, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdo_declare_attr_set(env, mdd_spobj, la, handle); + if (rc != 0) + return rc; - rc = mdo_declare_attr_set(env, mdd_sobj, NULL, handle); - if (rc) - return rc; - mdd_declare_links_add(env, mdd_sobj, handle); - if (rc) - return rc; + rc = mdo_declare_attr_set(env, mdd_tpobj, la, handle); + if (rc != 0) + return rc; - rc = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, mdd_sobj, la, handle); + if (rc) + return rc; + + rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata); + if (rc) + return rc; /* new name */ rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj), @@ -2081,17 +2461,18 @@ static int mdd_declare_rename(const struct lu_env *env, return rc; } - rc = mdo_declare_attr_set(env, mdd_tobj, NULL, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, mdd_tobj, la, handle); + if (rc) + return rc; - mdd_declare_links_add(env, mdd_tobj, handle); - 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, ma, handle); - if (rc) - return rc; + rc = mdd_declare_finish_unlink(env, mdd_tobj, ma, handle); + if (rc) + return rc; } rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle); @@ -2108,88 +2489,79 @@ static int mdd_rename(const struct lu_env *env, struct md_object *tobj, const struct lu_name *ltname, struct md_attr *ma) { - const char *sname = lsname->ln_name; - const char *tname = ltname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct lu_attr *so_attr = &mdd_env_info(env)->mti_cattr; - struct lu_attr *tg_attr = &mdd_env_info(env)->mti_pattr; - struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */ - struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj); - struct mdd_device *mdd = mdo2mdd(src_pobj); - struct mdd_object *mdd_sobj = NULL; /* source object */ - struct mdd_object *mdd_tobj = NULL; - struct dynlock_handle *sdlh, *tdlh; - struct thandle *handle; - const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj); - const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj); - bool is_dir; + const char *sname = lsname->ln_name; + const char *tname = ltname->ln_name; + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + struct mdd_object *mdd_spobj = md2mdd_obj(src_pobj); /* source parent */ + struct mdd_object *mdd_tpobj = md2mdd_obj(tgt_pobj); + struct mdd_device *mdd = mdo2mdd(src_pobj); + struct mdd_object *mdd_sobj = NULL; /* source object */ + struct mdd_object *mdd_tobj = NULL; + struct lu_attr *cattr = MDD_ENV_VAR(env, cattr); + struct lu_attr *pattr = MDD_ENV_VAR(env, pattr); + struct lu_attr *tattr = MDD_ENV_VAR(env, tattr); + struct lu_attr *tpattr = MDD_ENV_VAR(env, tpattr); + struct thandle *handle; + struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; + const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj); + const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj); + bool is_dir; bool tobj_ref = 0; bool tobj_locked = 0; unsigned cl_flags = 0; - int rc, rc2; - ENTRY; + int rc, rc2; + ENTRY; + + if (tobj) + mdd_tobj = md2mdd_obj(tobj); + + mdd_sobj = mdd_object_find(env, mdd, lf); + + rc = mdd_la_get(env, mdd_sobj, cattr, BYPASS_CAPA); + if (rc) + GOTO(out_pending, rc); + + rc = mdd_la_get(env, mdd_spobj, pattr, BYPASS_CAPA); + if (rc) + GOTO(out_pending, rc); - if (tobj) - mdd_tobj = md2mdd_obj(tobj); + if (mdd_tobj) { + rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); + if (rc) + GOTO(out_pending, rc); + } + + rc = mdd_la_get(env, mdd_tpobj, tpattr, BYPASS_CAPA); + if (rc) + GOTO(out_pending, rc); - mdd_sobj = mdd_object_find(env, mdd, lf); + rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr, + mdd_sobj, cattr, mdd_tobj, tattr); + if (rc) + GOTO(out_pending, rc); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_pending, rc = PTR_ERR(handle)); - rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj, - mdd_tobj, lsname, ltname, ma, handle); - if (rc) - GOTO(stop, rc); + memset(ldata, 0, sizeof(*ldata)); + mdd_linkea_prepare(env, mdd_sobj, NULL, NULL, mdd_object_fid(mdd_tpobj), + ltname, 1, 0, ldata); + rc = mdd_declare_rename(env, mdd, mdd_spobj, mdd_tpobj, mdd_sobj, + mdd_tobj, lsname, ltname, ma, ldata, handle); + if (rc) + GOTO(stop, rc); rc = mdd_trans_start(env, mdd, handle); if (rc) GOTO(stop, rc); /* FIXME: Should consider tobj and sobj too in rename_lock. */ - rc = mdd_rename_order(env, mdd, mdd_spobj, mdd_tpobj); - if (rc < 0) - GOTO(cleanup_unlocked, rc); - - /* Get locks in determined order */ - if (rc == MDD_RN_SAME) { - sdlh = mdd_pdo_write_lock(env, mdd_spobj, - sname, MOR_SRC_PARENT); - /* check hashes to determine do we need one lock or two */ - if (mdd_name2hash(sname) != mdd_name2hash(tname)) - tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname, - MOR_TGT_PARENT); - else - tdlh = sdlh; - } else if (rc == MDD_RN_SRCTGT) { - sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_SRC_PARENT); - tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_TGT_PARENT); - } else { - tdlh = mdd_pdo_write_lock(env, mdd_tpobj, tname,MOR_SRC_PARENT); - sdlh = mdd_pdo_write_lock(env, mdd_spobj, sname,MOR_TGT_PARENT); - } - if (sdlh == NULL || tdlh == NULL) - GOTO(cleanup, rc = -ENOMEM); - - rc = mdd_la_get(env, mdd_sobj, so_attr, - mdd_object_capa(env, mdd_sobj)); - if (rc) - GOTO(cleanup, rc); - - if (mdd_tobj) { - rc = mdd_la_get(env, mdd_tobj, tg_attr, - mdd_object_capa(env, mdd_tobj)); - if (rc) - GOTO(cleanup, rc); - } - - rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, mdd_sobj, - mdd_tobj, so_attr, tg_attr); - if (rc) - GOTO(cleanup, rc); + rc = mdd_rename_order(env, mdd, mdd_spobj, pattr, mdd_tpobj); + if (rc < 0) + GOTO(cleanup_unlocked, rc); - is_dir = S_ISDIR(so_attr->la_mode); + is_dir = S_ISDIR(cattr->la_mode); /* Remove source name from source directory */ rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle, @@ -2235,12 +2607,12 @@ static int mdd_rename(const struct lu_env *env, 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_attr_check_set_internal(env, mdd_sobj, la, handle, 0); - if (rc) - GOTO(fixup_tpobj, rc); - } + if (mdd_sobj) { + la->la_valid = LA_CTIME; + rc = mdd_update_time(env, mdd_sobj, cattr, la, handle); + if (rc) + GOTO(fixup_tpobj, rc); + } /* Remove old target object * For tobj is remote case cmm layer has processed @@ -2259,13 +2631,12 @@ static int mdd_rename(const struct lu_env *env, mdo_ref_del(env, mdd_tobj, handle); /* Remove dot reference. */ - if (S_ISDIR(tg_attr->la_mode)) + if (S_ISDIR(tattr->la_mode)) mdo_ref_del(env, mdd_tobj, handle); tobj_ref = 1; /* fetch updated nlink */ - rc = mdd_la_get(env, mdd_tobj, tg_attr, - mdd_object_capa(env, mdd_tobj)); + rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); if (rc != 0) { CERROR("%s: Failed to get nlink for tobj " DFID": rc = %d\n", @@ -2275,7 +2646,7 @@ static int mdd_rename(const struct lu_env *env, } la->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); + rc = mdd_update_time(env, mdd_tobj, tattr, la, handle); if (rc != 0) { CERROR("%s: Failed to set ctime for tobj " DFID": rc = %d\n", @@ -2285,7 +2656,7 @@ static int mdd_rename(const struct lu_env *env, } /* XXX: this transfer to ma will be removed with LOD/OSP */ - ma->ma_attr = *tg_attr; + ma->ma_attr = *tattr; ma->ma_valid |= MA_INODE; rc = mdd_finish_unlink(env, mdd_tobj, ma, handle); if (rc != 0) { @@ -2297,8 +2668,7 @@ static int mdd_rename(const struct lu_env *env, } /* fetch updated nlink */ - rc = mdd_la_get(env, mdd_tobj, tg_attr, - mdd_object_capa(env, mdd_tobj)); + rc = mdd_la_get(env, mdd_tobj, tattr, BYPASS_CAPA); if (rc != 0) { CERROR("%s: Failed to get nlink for tobj " DFID": rc = %d\n", @@ -2307,32 +2677,35 @@ static int mdd_rename(const struct lu_env *env, GOTO(fixup_tpobj, rc); } /* XXX: this transfer to ma will be removed with LOD/OSP */ - ma->ma_attr = *tg_attr; + ma->ma_attr = *tattr; ma->ma_valid |= MA_INODE; - if (so_attr->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; + } } - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0); - if (rc) - GOTO(fixup_tpobj, rc); + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdd_update_time(env, mdd_spobj, pattr, la, handle); + if (rc) + GOTO(fixup_tpobj, rc); - if (mdd_spobj != mdd_tpobj) { - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal(env, mdd_tpobj, la, - handle, 0); - } + 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 && mdd_sobj) { - mdd_write_lock(env, mdd_sobj, MOR_SRC_CHILD); + 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, 0, 0); + 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, 0); + 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. */ @@ -2367,19 +2740,21 @@ fixup_tpobj: } fixup_spobj: - if (rc && is_dir && mdd_sobj) { - rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle, - BYPASS_CAPA); + if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) { + rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle, + BYPASS_CAPA); - if (rc2) - CWARN("sp obj dotdot delete error %d\n",rc2); + 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) - CWARN("sp obj dotdot insert error %d\n",rc2); - } + rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, + dotdot, handle, BYPASS_CAPA); + if (rc2) + CWARN("%s: sp obj dotdot insert error: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc2); + } fixup_spobj2: if (rc) { @@ -2391,10 +2766,6 @@ fixup_spobj2: cleanup: if (tobj_locked) mdd_write_unlock(env, mdd_tobj); - if (likely(tdlh) && sdlh != tdlh) - mdd_pdo_write_unlock(env, mdd_tpobj, tdlh); - if (likely(sdlh)) - mdd_pdo_write_unlock(env, mdd_spobj, sdlh); cleanup_unlocked: if (rc == 0) rc = mdd_changelog_ext_ns_store(env, mdd, CL_RENAME, cl_flags, @@ -2404,429 +2775,11 @@ cleanup_unlocked: stop: mdd_trans_stop(env, mdd, rc, handle); - if (mdd_sobj) - mdd_object_put(env, mdd_sobj); out_pending: - return rc; -} - -/** - * The data that link search is done on. - */ -struct mdd_link_data { - /** - * Buffer to keep link EA body. - */ - struct lu_buf *ml_buf; - /** - * The matched header, entry and its lenght in the EA - */ - struct link_ea_header *ml_leh; - struct link_ea_entry *ml_lee; - int ml_reclen; -}; - -static int mdd_links_new(const struct lu_env *env, - struct mdd_link_data *ldata) -{ - ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE); - if (ldata->ml_buf->lb_buf == NULL) - return -ENOMEM; - ldata->ml_leh = ldata->ml_buf->lb_buf; - ldata->ml_leh->leh_magic = LINK_EA_MAGIC; - ldata->ml_leh->leh_len = sizeof(struct link_ea_header); - ldata->ml_leh->leh_reccount = 0; - return 0; -} - -/** 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::ml_buf. - * - * \retval 0 or error - */ -int mdd_links_read(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct mdd_link_data *ldata) -{ - struct lustre_capa *capa; - struct link_ea_header *leh; - int rc; - - /* First try a small buf */ - LASSERT(env != NULL); - ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE); - if (ldata->ml_buf->lb_buf == NULL) - return -ENOMEM; - - if (!mdd_object_exists(mdd_obj)) - return -ENODATA; - - capa = mdd_object_capa(env, mdd_obj); - rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf, - XATTR_NAME_LINK, capa); - if (rc == -ERANGE) { - /* Buf was too small, figure out what we need. */ - mdd_buf_put(ldata->ml_buf); - rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf, - XATTR_NAME_LINK, capa); - if (rc < 0) - return rc; - ldata->ml_buf = mdd_buf_alloc(env, rc); - if (ldata->ml_buf->lb_buf == NULL) - return -ENOMEM; - rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf, - XATTR_NAME_LINK, capa); - } - if (rc < 0) - return rc; - - leh = ldata->ml_buf->lb_buf; - if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) { - leh->leh_magic = LINK_EA_MAGIC; - leh->leh_reccount = __swab32(leh->leh_reccount); - leh->leh_len = __swab64(leh->leh_len); - /* entries are swabbed by mdd_lee_unpack */ - } - if (leh->leh_magic != LINK_EA_MAGIC) - return -EINVAL; - if (leh->leh_reccount == 0) - return -ENODATA; - - ldata->ml_leh = leh; - return 0; -} - -/** 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) - */ -struct lu_buf *mdd_links_get(const struct lu_env *env, - struct mdd_object *mdd_obj) -{ - struct mdd_link_data ldata = { 0 }; - int rc; - - rc = mdd_links_read(env, mdd_obj, &ldata); - return rc ? ERR_PTR(rc) : ldata.ml_buf; -} - -static int mdd_links_write(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct mdd_link_data *ldata, - struct thandle *handle) -{ - const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ml_buf->lb_buf, - ldata->ml_leh->leh_len); - return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle, - mdd_object_capa(env, mdd_obj)); -} - -/** Pack a link_ea_entry. - * All elements are stored as chars to avoid alignment issues. - * Numbers are always big-endian - * \retval record length - */ -static int mdd_lee_pack(struct link_ea_entry *lee, const struct lu_name *lname, - const struct lu_fid *pfid) -{ - struct lu_fid tmpfid; - int reclen; - - fid_cpu_to_be(&tmpfid, pfid); - memcpy(&lee->lee_parent_fid, &tmpfid, sizeof(tmpfid)); - memcpy(lee->lee_name, lname->ln_name, lname->ln_namelen); - reclen = sizeof(struct link_ea_entry) + lname->ln_namelen; - - lee->lee_reclen[0] = (reclen >> 8) & 0xff; - lee->lee_reclen[1] = reclen & 0xff; - return reclen; -} - -void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen, - struct lu_name *lname, struct lu_fid *pfid) -{ - *reclen = (lee->lee_reclen[0] << 8) | lee->lee_reclen[1]; - memcpy(pfid, &lee->lee_parent_fid, sizeof(*pfid)); - fid_be_to_cpu(pfid, pfid); - lname->ln_name = lee->lee_name; - lname->ln_namelen = *reclen - sizeof(struct link_ea_entry); -} - -static int mdd_declare_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct thandle *handle) -{ - int rc; - - /* XXX: max size? */ - rc = mdo_declare_xattr_set(env, mdd_obj, - mdd_buf_get_const(env, NULL, 4096), - XATTR_NAME_LINK, 0, handle); - - return rc; -} - -/* For pathologic linkers, we don't want to spend lots of time scanning the - * link ea. Limit ourseleves to something reasonable; links not in the EA - * can be looked up via (slower) parent lookup. - */ -#define LINKEA_MAX_COUNT 128 - -/** Add a record to the end of link ea buf */ -static int mdd_links_add_buf(const struct lu_env *env, - struct mdd_link_data *ldata, - const struct lu_name *lname, - const struct lu_fid *pfid) -{ - LASSERT(ldata->ml_leh != NULL); - - if (lname == NULL || pfid == NULL) - return -EINVAL; - - /* Make sure our buf is big enough for the new one */ - if (ldata->ml_leh->leh_reccount > LINKEA_MAX_COUNT) - return -EOVERFLOW; - - ldata->ml_reclen = lname->ln_namelen + sizeof(struct link_ea_entry); - if (ldata->ml_leh->leh_len + ldata->ml_reclen > - ldata->ml_buf->lb_len) { - if (mdd_buf_grow(env, ldata->ml_leh->leh_len + - ldata->ml_reclen) < 0) - return -ENOMEM; - } - - ldata->ml_leh = ldata->ml_buf->lb_buf; - ldata->ml_lee = ldata->ml_buf->lb_buf + ldata->ml_leh->leh_len; - ldata->ml_reclen = mdd_lee_pack(ldata->ml_lee, lname, pfid); - ldata->ml_leh->leh_len += ldata->ml_reclen; - ldata->ml_leh->leh_reccount++; - CDEBUG(D_INODE, "New link_ea name '%.*s' is added\n", - lname->ln_namelen, lname->ln_name); - return 0; -} - -/** Del the current record from the link ea buf */ -static void mdd_links_del_buf(const struct lu_env *env, - struct mdd_link_data *ldata, - const struct lu_name *lname) -{ - LASSERT(ldata->ml_leh != NULL); - - ldata->ml_leh->leh_reccount--; - ldata->ml_leh->leh_len -= ldata->ml_reclen; - memmove(ldata->ml_lee, (char *)ldata->ml_lee + ldata->ml_reclen, - (char *)ldata->ml_leh + ldata->ml_leh->leh_len - - (char *)ldata->ml_lee); - CDEBUG(D_INODE, "Old link_ea name '%.*s' is removed\n", - lname->ln_namelen, lname->ln_name); - -} - -/** - * Check if such a link exists in linkEA. - * - * \param mdd_obj object being handled - * \param pfid parent fid the link to be found for - * \param lname name in the parent's directory entry pointing to this object - * \param ldata link data the search to be done on - * - * \retval 0 success - * \retval -ENOENT link does not exist - * \retval -ve on error - */ -static int mdd_links_find(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct mdd_link_data *ldata, - const struct lu_name *lname, - const struct lu_fid *pfid) -{ - struct lu_name *tmpname = &mdd_env_info(env)->mti_name2; - struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid; - int count; - - LASSERT(ldata->ml_leh != NULL); - - /* link #0 */ - ldata->ml_lee = (struct link_ea_entry *)(ldata->ml_leh + 1); - - for (count = 0; count < ldata->ml_leh->leh_reccount; count++) { - mdd_lee_unpack(ldata->ml_lee, &ldata->ml_reclen, - tmpname, tmpfid); - if (tmpname->ln_namelen == lname->ln_namelen && - lu_fid_eq(tmpfid, pfid) && - (strncmp(tmpname->ln_name, lname->ln_name, - tmpname->ln_namelen) == 0)) - break; - ldata->ml_lee = (struct link_ea_entry *)((char *)ldata->ml_lee + - ldata->ml_reclen); - } - - if (count == ldata->ml_leh->leh_reccount) { - CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n", - lname->ln_namelen, lname->ln_name); - return -ENOENT; - } - return 0; -} - -static int __mdd_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct mdd_link_data *ldata, - const struct lu_name *lname, - const struct lu_fid *pfid, - int first, int check) -{ - int rc; - - if (ldata->ml_leh == NULL) { - rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata); - if (rc) { - if (rc != -ENODATA) - return rc; - rc = mdd_links_new(env, ldata); - if (rc) - return rc; - } - } - - if (check) { - rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid); - if (rc && rc != -ENOENT) - return rc; - if (rc == 0) - return -EEXIST; - } - - return mdd_links_add_buf(env, ldata, lname, pfid); -} - -static int __mdd_links_del(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct mdd_link_data *ldata, - const struct lu_name *lname, - const struct lu_fid *pfid) -{ - int rc; - - if (ldata->ml_leh == NULL) { - rc = mdd_links_read(env, mdd_obj, ldata); - if (rc) - return rc; - } - - rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid); - if (rc) - return rc; - - mdd_links_del_buf(env, ldata, lname); - return 0; -} - -static int mdd_links_rename(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *oldpfid, - const struct lu_name *oldlname, - const struct lu_fid *newpfid, - const struct lu_name *newlname, - struct thandle *handle, - int first, int check) -{ - struct mdd_link_data ldata = { 0 }; - int updated = 0; - int rc2 = 0; - int rc = 0; - ENTRY; - - LASSERT(oldpfid != NULL || newpfid != NULL); - - if (mdd_obj->mod_flags & DEAD_OBJ) - /* No more links, don't bother */ - RETURN(0); - - if (oldpfid != NULL) { - rc = __mdd_links_del(env, mdd_obj, &ldata, - oldlname, oldpfid); - if (rc) { - if ((check == 0) || - (rc != -ENODATA && rc != -ENOENT)) - GOTO(out, rc); - /* No changes done. */ - rc = 0; - } else { - updated = 1; - } - } - - /* 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); - if (rc2 == -EEXIST) - rc2 = 0; - else if (rc2 == 0) - updated = 1; - } - - if (updated) - rc = mdd_links_write(env, mdd_obj, &ldata, handle); - EXIT; -out: - if (rc == 0) - rc = rc2; - if (rc) { - int error = 1; - if (rc == -EOVERFLOW || rc == - ENOENT) - 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, - 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))); - 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))); - } - - if (ldata.ml_buf && ldata.ml_buf->lb_len > OBD_ALLOC_BIG) - /* if we vmalloced a large buffer drop it */ - mdd_buf_put(ldata.ml_buf); - + mdd_object_put(env, mdd_sobj); return rc; } -static inline int mdd_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *pfid, - const struct lu_name *lname, - struct thandle *handle, int first) -{ - return mdd_links_rename(env, mdd_obj, NULL, NULL, - pfid, lname, handle, first, 0); -} - -static inline int mdd_links_del(const struct lu_env *env, - struct mdd_object *mdd_obj, - const struct lu_fid *pfid, - const struct lu_name *lname, - struct thandle *handle) -{ - return mdd_links_rename(env, mdd_obj, pfid, lname, - NULL, NULL, handle, 0, 0); -} - const struct md_dir_operations mdd_dir_ops = { .mdo_is_subdir = mdd_is_subdir, .mdo_lookup = mdd_lookup,