X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_dir.c;h=9d03d171269e21d259ee97b4276a20063d3666f7;hp=de42a5ce77775943c678d8542643f046f059b549;hb=f625f670afbe954030ff81f0f8522137d6cdd335;hpb=39a9eeb26f20bb42e9d39e361f5d0a724c199137 diff --git a/lustre/mdd/mdd_dir.c b/lustre/mdd/mdd_dir.c index de42a5c..9d03d17 100644 --- a/lustre/mdd/mdd_dir.c +++ b/lustre/mdd/mdd_dir.c @@ -27,7 +27,7 @@ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2011, 2012, Intel Corporation. + * Copyright (c) 2011, 2014, 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 +57,78 @@ 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); +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_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); + } + + 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 = dt_lookup(env, dir, (struct dt_rec *)fid, key); + 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); + if (rc != 0) + RETURN(rc); + + rc = __mdd_lookup(env, pobj, pattr, lname, fid, + (spec != NULL && spec->sp_permitted) ? 0 : 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 +151,11 @@ 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; @@ -154,13 +170,14 @@ 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 */ + 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) @@ -169,7 +186,7 @@ static int mdd_is_parent(const struct lu_env *env, parent = mdd_object_find(env, mdd, pfid); if (IS_ERR(parent)) { GOTO(out, rc = PTR_ERR(parent)); - } else if (mdd_object_exists(parent) < 0) { + } 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 */ @@ -188,7 +205,7 @@ out: * No permission check is needed. * * returns 1: if fid is ancestor of @mo; - * returns 0: if fid is not a ancestor of @mo; + * returns 0: if fid is not an ancestor of @mo; * * returns EREMOTE if remote object is found, fid of remote object is saved to * @fid; @@ -196,25 +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); + 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); } /* @@ -231,114 +253,144 @@ 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); -} - -static int __mdd_may_link(const struct lu_env *env, struct mdd_object *obj) -{ - struct mdd_device *m = mdd_obj2mdd_dev(obj); - struct lu_attr *la = &mdd_env_info(env)->mti_la; - int rc; - 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); + 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); +} - rc = mdd_la_get(env, obj, la, BYPASS_CAPA); - if (rc) - RETURN(rc); +/** + * 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) +{ + struct mdd_device *m = mdd_obj2mdd_dev(obj); + ENTRY; - /* - * 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); + LASSERT(la != NULL); + + /* 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); + + 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, - struct mdd_object *cobj, int check_perm, int check_nlink) + const struct lu_attr *pattr, struct mdd_object *cobj, + bool check_perm) { - int rc = 0; - ENTRY; + struct mdd_thread_info *info = mdd_env_info(env); + struct lu_buf *xbuf; + 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 the parent is a sub-stripe, check whether it is dead */ + xbuf = mdd_buf_get(env, info->mti_key, sizeof(info->mti_key)); + rc = mdo_xattr_get(env, pobj, xbuf, XATTR_NAME_LMV); + if (unlikely(rc > 0)) { + struct lmv_mds_md_v1 *lmv1 = xbuf->lb_buf; - RETURN(rc); + if (le32_to_cpu(lmv1->lmv_magic) == LMV_MAGIC_STRIPE && + le32_to_cpu(lmv1->lmv_hash_type) & LMV_HASH_FLAG_DEAD) + RETURN(-ESTALE); + } + rc = 0; + + if (check_perm) + rc = mdd_permission_internal_locked(env, pobj, pattr, + MAY_WRITE | MAY_EXEC, + MOR_TGT_PARENT); + 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); + if (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL)) + 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 != 0) + RETURN(rc); - if (mdd_is_append(pobj)) - RETURN(-EPERM); + if (pattr->la_flags & LUSTRE_APPEND_FL) + RETURN(-EPERM); - RETURN(rc); + RETURN(rc); } /* @@ -347,34 +399,30 @@ 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 !md_capable(uc, CFS_CAP_FOWNER); } static int mdd_may_delete_entry(const struct lu_env *env, - struct mdd_object *pobj, int check_perm) + struct mdd_object *pobj, + const struct lu_attr *pattr, + int check_perm) { ENTRY; @@ -387,14 +435,14 @@ static int mdd_may_delete_entry(const struct lu_env *env, if (check_perm) { int rc; - rc = mdd_permission_internal_locked(env, pobj, NULL, + rc = mdd_permission_internal_locked(env, pobj, pattr, MAY_WRITE | MAY_EXEC, MOR_TGT_PARENT); if (rc) RETURN(rc); } - if (mdd_is_append(pobj)) + if (pattr->la_flags & LUSTRE_APPEND_FL) RETURN(-EPERM); RETURN(0); @@ -404,67 +452,77 @@ static int mdd_may_delete_entry(const struct lu_env *env, * 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; + int rc = 0; + ENTRY; - if (pobj) { - rc = mdd_may_delete_entry(env, pobj, check_perm); + if (tpobj) { + LASSERT(tpattr != NULL); + rc = mdd_may_delete_entry(env, tpobj, tpattr, check_perm); if (rc != 0) RETURN(rc); } - if (cobj == NULL) + if (tobj == NULL) RETURN(0); - if (!mdd_object_exists(cobj)) - RETURN(-ENOENT); - - if (mdd_is_dead_obj(cobj)) - RETURN(-ESTALE); - + if (!mdd_object_exists(tobj)) + RETURN(-ENOENT); - if (mdd_is_sticky(env, pobj, cobj)) - RETURN(-EPERM); + if (mdd_is_dead_obj(tobj)) + RETURN(-ESTALE); - if (mdd_is_immutable(cobj) || mdd_is_append(cobj)) - RETURN(-EPERM); + if (mdd_is_sticky(env, tpobj, tpattr, tobj, tattr)) + RETURN(-EPERM); - if ((cattr->la_valid & LA_FLAGS) && - (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) - RETURN(-EPERM); + if (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 +/** + * 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 */ -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; @@ -477,64 +535,62 @@ 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, NULL, 1, 0); - if (rc) - RETURN(rc); - } + LASSERT(src_obj != tgt_obj); + rc = mdd_may_create(env, tgt_obj, tattr, NULL, true); + if (rc != 0) + 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, - 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; } @@ -543,30 +599,31 @@ static int __mdd_index_insert_only(const struct lu_env *env, /* insert named index, add reference if isdir */ static int __mdd_index_insert(const struct lu_env *env, struct mdd_object *pobj, - const struct lu_fid *lf, const char *name, int is_dir, - struct thandle *handle, struct lustre_capa *capa) + const struct lu_fid *lf, __u32 type, + const char *name, struct thandle *handle) { - int rc; - ENTRY; + int rc; + ENTRY; - rc = __mdd_index_insert_only(env, pobj, lf, name, handle, capa); - if (rc == 0 && is_dir) { - mdd_write_lock(env, pobj, MOR_TGT_PARENT); - mdo_ref_add(env, pobj, handle); - mdd_write_unlock(env, pobj); - } - RETURN(rc); + rc = __mdd_index_insert_only(env, pobj, lf, type, name, handle); + if (rc == 0 && S_ISDIR(type)) { + mdd_write_lock(env, pobj, MOR_TGT_PARENT); + mdo_ref_add(env, pobj, handle); + mdd_write_unlock(env, pobj); + } + + RETURN(rc); } /* delete named index, drop reference if isdir */ static int __mdd_index_delete(const struct lu_env *env, struct mdd_object *pobj, - const char *name, int is_dir, struct thandle *handle, - struct lustre_capa *capa) + const char *name, int is_dir, + struct thandle *handle) { int rc; ENTRY; - rc = __mdd_index_delete_only(env, pobj, name, handle, capa); + rc = __mdd_index_delete_only(env, pobj, name, handle); if (rc == 0 && is_dir) { mdd_write_lock(env, pobj, MOR_TGT_PARENT); mdo_ref_del(env, pobj, handle); @@ -576,69 +633,37 @@ 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) +static int mdd_llog_record_calc_size(const struct lu_env *env, + const struct lu_name *tname, + const struct lu_name *sname) { - 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; + 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); - /* new record referencing new plain llog */ - rc = dt_declare_record_write(env, mdd->mdd_capa, - DECLARE_LLOG_WRITE, 0, handle); - if (rc) - return rc; + if (sname != NULL) + crf |= CLF_RENAME; - /* catalog's header will be updated as well */ - rc = dt_declare_record_write(env, mdd->mdd_capa, - DECLARE_LLOG_REWRITE, 0, handle); + if (uc != NULL && uc->uc_jobid[0] != '\0') + crf |= CLF_JOBID; - 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)); } int mdd_declare_changelog_store(const struct lu_env *env, struct mdd_device *mdd, - const struct lu_name *fname, + 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_rec *rec; struct lu_buf *buf; + struct thandle *llog_th; int reclen; int rc; @@ -646,9 +671,8 @@ int mdd_declare_changelog_store(const struct lu_env *env, if (!(mdd->mdd_cl.mc_flags & CLM_ON)) return 0; - reclen = llog_data_len(sizeof(*rec) + - (fname != NULL ? fname->ln_namelen : 0)); - buf = mdd_buf_alloc(env, reclen); + 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; @@ -660,45 +684,13 @@ int mdd_declare_changelog_store(const struct lu_env *env, if (ctxt == NULL) return -ENXIO; - rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle); - llog_ctxt_put(ctxt); - - return rc; -} - -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) -{ - struct obd_device *obd = mdd2obd_dev(mdd); - struct llog_ctxt *ctxt; - struct llog_changelog_ext_rec *rec; - struct lu_buf *buf; - int reclen; - int rc; - - /* Not recording */ - 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)); - buf = mdd_buf_alloc(env, 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; + 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)); - 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, llog_th); - rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle); +out_put: llog_ctxt_put(ctxt); return rc; @@ -708,7 +700,7 @@ static int mdd_declare_changelog_ext_store(const struct lu_env *env, * \param mdd * \param rec * \param handle - currently ignored since llogs start their own transaction; - * this will hopefully be fixed in llog rewrite + * this will hopefully be fixed in llog rewrite * \retval 0 ok */ int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd, @@ -716,9 +708,13 @@ 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_len = llog_data_len(sizeof(*rec) + + changelog_rec_varsize(&rec->cr)); + + /* llog_lvfs_write_rec sets the llog tail len */ rec->cr_hdr.lrh_type = CHANGELOG_REC; rec->cr.cr_time = cl_time(); @@ -733,224 +729,525 @@ int mdd_changelog_store(const struct lu_env *env, struct mdd_device *mdd, if (ctxt == NULL) return -ENXIO; - rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th); + 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, llog_th); + +out_put: 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) +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 obd_device *obd = mdd2obd_dev(mdd); - struct llog_ctxt *ctxt; - int rc; + struct changelog_ext_rename *rnm = changelog_rec_rename(rec); + size_t extsize = sname->ln_namelen + 1; - 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(); + LASSERT(sfid != NULL); + LASSERT(spfid != NULL); + LASSERT(sname != NULL); - 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); + rnm->cr_sfid = *sfid; + rnm->cr_spfid = *spfid; - ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); - if (ctxt == NULL) - return -ENXIO; + changelog_rec_name(rec)[rec->cr_namelen] = '\0'; + strlcpy(changelog_rec_sname(rec), sname->ln_name, extsize); + rec->cr_namelen += extsize; +} - /* nested journal transaction */ - rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th); - llog_ctxt_put(ctxt); - if (rc > 0) - rc = 0; +void mdd_changelog_rec_ext_jobid(struct changelog_rec *rec, const char *jobid) +{ + struct changelog_ext_jobid *jid = changelog_rec_jobid(rec); - return rc; + if (jobid == NULL || jobid[0] == '\0') + return; + + 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. * \param target - mdd_object of change - * \param parent - parent dir/object + * \param tpfid - target parent dir/object fid + * \param sfid - source object fid + * \param spfid - source parent fid * \param tname - target name string + * \param sname - source 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) -{ - struct llog_changelog_rec *rec; - struct lu_buf *buf; - int reclen; - int rc; +int mdd_changelog_ns_store(const struct lu_env *env, + struct mdd_device *mdd, + enum changelog_rec_type type, + enum changelog_rec_flags crf, + struct mdd_object *target, + const struct lu_fid *tpfid, + const struct lu_fid *sfid, + const struct lu_fid *spfid, + const struct lu_name *tname, + const struct lu_name *sname, + struct thandle *handle) +{ + const struct lu_ucred *uc = lu_ucred(env); + struct llog_changelog_rec *rec; + struct lu_buf *buf; + int reclen; + int rc; ENTRY; /* Not recording */ if (!(mdd->mdd_cl.mc_flags & CLM_ON)) RETURN(0); + if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) RETURN(0); - LASSERT(target != NULL); - LASSERT(parent != NULL); + LASSERT(tpfid != NULL); LASSERT(tname != NULL); LASSERT(handle != NULL); - reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen); - buf = mdd_buf_alloc(env, reclen); + 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_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_tfid = *mdo2fid(target); - rec->cr.cr_pfid = *mdo2fid(parent); + rec->cr.cr_pfid = *tpfid; rec->cr.cr_namelen = tname->ln_namelen; - memcpy(rec->cr.cr_name, tname->ln_name, tname->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); - target->mod_cltime = cfs_time_current_64(); + if (crf & CLF_JOBID) + mdd_changelog_rec_ext_jobid(&rec->cr, uc->uc_jobid); + + if (likely(target != NULL)) { + rec->cr.cr_tfid = *mdo2fid(target); + target->mod_cltime = cfs_time_current_64(); + } else { + fid_zero(&rec->cr.cr_tfid); + } 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); + rc, type, tname->ln_name, PFID(sfid), PFID(tpfid)); + return -EFAULT; } - RETURN(0); + return 0; } +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; -/** 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 tpfid - target parent dir/object fid - * \param sfid - source object fid - * \param spfid - source parent fid - * \param tname - target name string - * \param sname - source name string - * \param handle - transacion handle - */ -static int mdd_changelog_ext_ns_store(const struct lu_env *env, - struct mdd_device *mdd, - enum changelog_rec_type type, - unsigned flags, - struct mdd_object *target, - const struct lu_fid *tpfid, - const struct lu_fid *sfid, - const struct lu_fid *spfid, - const struct lu_name *tname, - const struct lu_name *sname, - struct thandle *handle) + 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; + } + } + + if (check) { + rc = linkea_links_find(ldata, lname, pfid); + if (rc && rc != -ENOENT) + return rc; + if (rc == 0) + return -EEXIST; + } + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE)) { + struct lu_fid *tfid = &mdd_env_info(env)->mti_fid2; + + *tfid = *pfid; + tfid->f_ver = ~0; + linkea_add_buf(ldata, lname, tfid); + } + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_MORE2)) + linkea_add_buf(ldata, lname, pfid); + + return linkea_add_buf(ldata, lname, pfid); +} + +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) { - struct llog_changelog_ext_rec *rec; - struct lu_buf *buf; - int reclen; int rc; + + if (ldata->ld_leh == NULL) { + rc = mdd_links_read(env, mdd_obj, ldata); + if (rc) + return rc; + } + + rc = linkea_links_find(ldata, lname, pfid); + if (rc) + return rc; + + linkea_del_buf(ldata, lname); + return 0; +} + +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; - /* Not recording */ - if (!(mdd->mdd_cl.mc_flags & CLM_ON)) - RETURN(0); - if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) + if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) + 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 */ RETURN(0); + } - LASSERT(sfid != NULL); - LASSERT(tpfid != NULL); - LASSERT(tname != NULL); - LASSERT(handle != NULL); + 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; + } + } - reclen = llog_data_len(sizeof(*rec) + - sname != NULL ? 1 + sname->ln_namelen : 0); - buf = mdd_buf_alloc(env, reclen); - if (buf->lb_buf == NULL) - RETURN(-ENOMEM); - rec = buf->lb_buf; + /* 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); + } - rec->cr.cr_flags = CLF_EXT_VERSION | (CLF_FLAGMASK & flags); - 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; + 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 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 (likely(target != NULL)) { - rec->cr.cr_tfid = *mdo2fid(target); - target->mod_cltime = cfs_time_current_64(); - } else { - fid_zero(&rec->cr.cr_tfid); + if (ldata->ld_reclen != 0) + rc = mdd_links_write(env, mdd_obj, ldata, handle); + EXIT; +out: + if (rc != 0) { + 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))); } - rc = mdd_changelog_ext_store(env, mdd, rec, 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); + + 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 *ldata, int first) +{ + return mdd_links_rename(env, mdd_obj, NULL, NULL, + pfid, lname, handle, ldata, 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; + + 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); + 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) { - CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n", - rc, type, tname->ln_name, PFID(sfid), PFID(tpfid)); - return -EFAULT; + lu_buf_free(ldata->ld_buf); + ldata->ld_buf = NULL; + return rc; } - return 0; + 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) + */ +struct lu_buf *mdd_links_get(const struct lu_env *env, + struct mdd_object *mdd_obj) +{ + struct linkea_data ldata = { NULL }; + 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); + int rc; + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_LINKEA)) + return 0; + + rc = mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle); + if (unlikely(rc == -ENOSPC) && S_ISREG(mdd_object_type(mdd_obj)) && + mdd_object_remote(mdd_obj) == 0) { + struct lfsck_request *lr = &mdd_env_info(env)->mti_lr; + struct thandle *sub_th; + + /* XXX: If the linkEA is overflow, then we need to notify the + * namespace LFSCK to skip "nlink" attribute verification + * on this object to avoid the "nlink" to be shrinked by + * wrong. It may be not good an interaction with LFSCK + * like this. We will consider to replace it with other + * mechanism in future. LU-5802. */ + lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK, + LFSCK_TYPE_NAMESPACE); + + sub_th = thandle_get_sub_by_dt(env, handle, + mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom); + lfsck_in_notify(env, mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom, + lr, sub_th); + } + + return rc; +} + +int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, + struct thandle *handle, struct linkea_data *ldata, + enum mdd_links_add_overflow overflow) +{ + int rc; + int ea_len; + void *linkea; + + if (ldata != NULL && ldata->ld_leh != 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); + if (rc != 0) + return rc; + + if (mdd_object_remote(mdd_obj) == 0 && overflow == MLAO_CHECK) { + struct lfsck_request *lr = &mdd_env_info(env)->mti_lr; + struct thandle *sub_th; + + /* XXX: If the linkEA is overflow, then we need to notify the + * namespace LFSCK to skip "nlink" attribute verification + * on this object to avoid the "nlink" to be shrinked by + * wrong. It may be not good an interaction with LFSCK + * like this. We will consider to replace it with other + * mechanism in future. LU-5802. */ + lfsck_pack_rfa(lr, mdo2fid(mdd_obj), LE_SKIP_NLINK_DECLARE, + LFSCK_TYPE_NAMESPACE); + + sub_th = thandle_get_sub_by_dt(env, handle, + mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom); + rc = lfsck_in_notify(env, + mdo2mdd(&mdd_obj->mod_obj)->mdd_bottom, + lr, sub_th); + } + + return rc; +} + +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, MLAO_IGNORE); + + return rc; } static int mdd_declare_link(const struct lu_env *env, - struct mdd_device *mdd, - struct mdd_object *p, - struct mdd_object *c, - const struct lu_name *name, - struct thandle *handle) + struct mdd_device *mdd, + struct mdd_object *p, + struct mdd_object *c, + const struct lu_name *name, + struct thandle *handle, + struct lu_attr *la, + struct linkea_data *data) { - int rc; + int rc; - rc = mdo_declare_index_insert(env, p, mdo2fid(c), name->ln_name,handle); - if (rc) - return rc; + rc = mdo_declare_index_insert(env, p, mdo2fid(c), mdd_object_type(c), + name->ln_name, handle); + if (rc != 0) + return rc; - rc = mdo_declare_ref_add(env, c, handle); - if (rc) - return rc; + rc = mdo_declare_ref_add(env, c, handle); + if (rc != 0) + return rc; - rc = mdo_declare_attr_set(env, p, NULL, handle); - if (rc) - return rc; + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) { + rc = mdo_declare_ref_add(env, c, handle); + if (rc != 0) + return rc; + } - rc = mdo_declare_attr_set(env, c, NULL, 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; - rc = mdd_declare_links_add(env, c, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, c, la, handle); + if (rc != 0) + return rc; - rc = mdd_declare_changelog_store(env, mdd, name, handle); + rc = mdd_declare_links_add(env, c, handle, data, + S_ISREG(mdd_object_type(c)) ? MLAO_CHECK : MLAO_IGNORE); + if (rc != 0) + return rc; - return rc; + rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle); + + return rc; } static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, @@ -961,17 +1258,33 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, 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 dynlock_handle *dlh; struct thandle *handle; - int rc; - ENTRY; + struct linkea_data *ldata = &mdd_env_info(env)->mti_link_data; + int rc; + ENTRY; + + rc = mdd_la_get(env, mdd_sobj, cattr); + if (rc != 0) + RETURN(rc); + + rc = mdd_la_get(env, mdd_tobj, tattr); + if (rc != 0) + RETURN(rc); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_pending, rc = PTR_ERR(handle)); - rc = mdd_declare_link(env, mdd, mdd_tobj, mdd_sobj, lname, 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); @@ -979,75 +1292,128 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, if (rc) GOTO(stop, 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); - - rc = mdd_link_sanity_check(env, mdd_tobj, lname, mdd_sobj); - if (rc) - GOTO(out_unlock, rc); - - rc = mdo_ref_add(env, mdd_sobj, handle); + 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); + if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LESS_NLINK)) { + rc = mdo_ref_add(env, mdd_sobj, handle); + if (rc != 0) + GOTO(out_unlock, rc); + } + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_MORE_NLINK)) { + rc = mdo_ref_add(env, mdd_sobj, handle); + if (rc != 0) + GOTO(out_unlock, rc); + } + + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DANGLING3)) { + struct lu_fid tfid = *mdo2fid(mdd_sobj); + + tfid.f_oid++; + rc = __mdd_index_insert_only(env, mdd_tobj, &tfid, + mdd_object_type(mdd_sobj), + name, handle); + } else { + rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), + mdd_object_type(mdd_sobj), + name, handle); + } - rc = __mdd_index_insert_only(env, mdd_tobj, mdo2fid(mdd_sobj), - name, handle, - mdd_object_capa(env, mdd_tobj)); if (rc != 0) { mdo_ref_del(env, mdd_sobj, handle); GOTO(out_unlock, rc); } - 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_tobj, la, handle, 0); - if (rc) - GOTO(out_unlock, 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); - } + 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); + mdo2fid(mdd_tobj), NULL, NULL, + lname, NULL, 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; } -int mdd_declare_finish_unlink(const struct lu_env *env, - struct mdd_object *obj, - struct md_attr *ma, - struct thandle *handle) +static int mdd_mark_dead_object(const struct lu_env *env, + struct mdd_object *obj, struct thandle *handle, + bool declare) +{ + struct lu_attr *attr = MDD_ENV_VAR(env, la_for_start); + int rc; + + if (!declare) + obj->mod_flags |= DEAD_OBJ; + + if (!S_ISDIR(mdd_object_type(obj))) + return 0; + + attr->la_valid = LA_FLAGS; + attr->la_flags = LUSTRE_SLAVE_DEAD_FL; + + 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, + struct mdd_object *obj, + struct thandle *handle) { int rc; + rc = mdd_mark_dead_object(env, obj, handle, true); + if (rc != 0) + return rc; + rc = orph_declare_index_insert(env, obj, mdd_object_type(obj), handle); - if (rc) + if (rc != 0) + return rc; + + rc = mdo_declare_destroy(env, obj, handle); + 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); @@ -1055,8 +1421,11 @@ 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)) { - obj->mod_flags |= DEAD_OBJ; + if (ma->ma_attr.la_nlink == 0 || is_dir) { + rc = mdd_mark_dead_object(env, obj, th, false); + if (rc != 0) + RETURN(rc); + /* add new orphan and the object * will be deleted during mdd_close() */ if (obj->mod_count) { @@ -1075,9 +1444,12 @@ int mdd_finish_unlink(const struct lu_env *env, } 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); } /* @@ -1085,44 +1457,40 @@ 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) -{ - 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) + const struct lu_attr *pattr, + struct mdd_object *cobj, + const struct lu_attr *cattr) { - 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, int is_dir) { - 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; + struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; + int 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; @@ -1140,56 +1508,108 @@ static int mdd_declare_unlink(const struct lu_env *env, struct mdd_device *mdd, if (rc) return rc; - rc = mdo_declare_attr_set(env, c, NULL, handle); + 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); + rc = mdd_declare_finish_unlink(env, c, handle); 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; } +/* + * 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) +{ + ENTRY; + + 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); + 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; + } + if (ma->ma_hsm.mh_flags & HS_EXISTS) + RETURN(true); + RETURN(false); +} + +/** + * 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) + 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 *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); + 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 dynlock_handle *dlh; - struct thandle *handle; + struct mdd_device *mdd = mdo2mdd(pobj); + struct thandle *handle; int rc, is_dir = 0; - ENTRY; + 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); - /* currently it is assume, it could only delete - * name entry of remote directory */ - is_dir = 1; } + 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); + if (rc) + RETURN(rc); + + is_dir = S_ISDIR(cattr->la_mode); + } + + rc = mdd_unlink_sanity_check(env, mdd_pobj, pattr, mdd_cobj, cattr); + if (rc) + RETURN(rc); + handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - RETURN(PTR_ERR(handle)); + if (IS_ERR(handle)) + RETURN(PTR_ERR(handle)); rc = mdd_declare_unlink(env, mdd, mdd_pobj, mdd_cobj, - lname, ma, handle); + lname, ma, handle, no_name, is_dir); if (rc) GOTO(stop, rc); @@ -1197,39 +1617,26 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, if (rc) GOTO(stop, rc); - dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(stop, rc = -ENOMEM); - - if (likely(mdd_cobj != NULL)) { + if (likely(mdd_cobj != NULL)) mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD); - /* fetch cattr */ - rc = mdd_la_get(env, mdd_cobj, cattr, - mdd_object_capa(env, mdd_cobj)); + 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); - - is_dir = S_ISDIR(cattr->la_mode); - } - rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr); - if (rc) - GOTO(cleanup, rc); - - rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle, - mdd_object_capa(env, mdd_pobj)); - 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); } @@ -1238,8 +1645,7 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, mdo_ref_del(env, mdd_cobj, handle); /* fetch updated nlink */ - rc = mdd_la_get(env, mdd_cobj, cattr, - mdd_object_capa(env, mdd_cobj)); + rc = mdd_la_get(env, mdd_cobj, cattr); if (rc) GOTO(cleanup, rc); } @@ -1248,65 +1654,62 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, 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); + rc = mdd_update_time(env, mdd_pobj, pattr, la, handle); if (rc) GOTO(cleanup, rc); /* Enough for only unlink the entry */ - if (unlikely(mdd_cobj == NULL)) { - mdd_pdo_write_unlock(env, mdd_pobj, dlh); + 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, mdd_pobj, lname, handle); /* fetch updated nlink */ if (rc == 0) - rc = mdd_la_get(env, mdd_cobj, cattr, - mdd_object_capa(env, mdd_cobj)); - - 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 (cattr->la_nlink == 0) { 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; + if (likely(mdd_cobj != NULL)) + mdd_write_unlock(env, mdd_cobj); + + if (rc == 0) { + int cl_flags = 0; - 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; + 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); - } + mdd_cobj, mdo2fid(mdd_pobj), NULL, NULL, lname, NULL, + handle); + } stop: - mdd_trans_stop(env, mdd, rc, handle); + mdd_trans_stop(env, mdd, rc, handle); - return rc; + return rc; } /* @@ -1322,33 +1725,28 @@ 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); + struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint; + 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: @@ -1356,12 +1754,12 @@ 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); 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, hint); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) @@ -1371,12 +1769,11 @@ 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 { @@ -1388,74 +1785,36 @@ 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, 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)); + 0, handle); + + 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) { - int rc; + int rc; + ENTRY; /* * inode mode has been set in creation time, and it's based on umask, @@ -1467,81 +1826,79 @@ int mdd_declare_object_initialize(const struct lu_env *env, attr->la_valid &= ~(LA_MODE | LA_TYPE); rc = mdo_declare_attr_set(env, child, attr, handle); attr->la_valid |= LA_MODE | LA_TYPE; - if (rc == 0 && S_ISDIR(attr->la_mode)) { - rc = mdo_declare_index_insert(env, child, mdo2fid(child), - dot, handle); - if (rc == 0) - rc = mdo_declare_ref_add(env, child, handle); - } + if (rc != 0 || !S_ISDIR(attr->la_mode)) + RETURN(rc); - if (rc == 0) - mdd_declare_links_add(env, child, handle); + rc = mdo_declare_index_insert(env, child, mdo2fid(child), S_IFDIR, + dot, handle); + if (rc != 0) + RETURN(rc); - return rc; -} + rc = mdo_declare_ref_add(env, child, handle); + if (rc != 0) + 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) -{ - int rc; - ENTRY; + rc = mdo_declare_index_insert(env, child, mdo2fid(parent), S_IFDIR, + dotdot, handle); - /* - * 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. - */ + RETURN(rc); +} - /* - * 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); +static int mdd_object_initialize(const struct lu_env *env, + const struct lu_fid *pfid, + struct mdd_object *child, + struct lu_attr *attr, struct thandle *handle, + const struct md_op_spec *spec) +{ + int rc = 0; + ENTRY; 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); - } - if (rc == 0) - mdd_links_add(env, child, pfid, lname, handle, 1); + 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); + 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, - struct lu_attr *pattr, + const struct lu_attr *pattr, const struct lu_name *lname, 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)) @@ -1558,21 +1915,18 @@ 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); - 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); - } + rc = __mdd_lookup(env, pobj, pattr, lname, fid, + MAY_WRITE | MAY_EXEC); + 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) { @@ -1583,6 +1937,10 @@ 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; @@ -1607,95 +1965,287 @@ static int mdd_create_sanity_check(const struct lu_env *env, RETURN(rc); } +static int mdd_declare_object_create(const struct lu_env *env, + struct mdd_device *mdd, + struct mdd_object *p, struct mdd_object *c, + struct lu_attr *attr, + struct thandle *handle, + const struct md_op_spec *spec, + struct lu_buf *def_acl_buf, + struct lu_buf *acl_buf, + struct dt_allocation_hint *hint) +{ + int rc; + + rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec, + hint); + if (rc) + GOTO(out, rc); + +#ifdef CONFIG_FS_POSIX_ACL + if (def_acl_buf->lb_len > 0 && S_ISDIR(attr->la_mode)) { + /* if dir, then can inherit default ACl */ + 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); + + rc = mdo_declare_xattr_set(env, c, acl_buf, + XATTR_NAME_ACL_ACCESS, 0, handle); + if (rc) + GOTO(out, rc); + } +#endif + rc = mdd_declare_object_initialize(env, p, c, attr, handle); + if (rc) + GOTO(out, rc); + + /* 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, + handle); + if (rc) + GOTO(out, rc); + } + + if (S_ISLNK(attr->la_mode)) { + const char *target_name = spec->u.sp_symname; + int sym_len = strlen(target_name); + const struct lu_buf *buf; + + buf = mdd_buf_get_const(env, target_name, sym_len); + rc = dt_declare_record_write(env, mdd_object_child(c), + buf, 0, handle); + if (rc) + GOTO(out, rc); + } +out: + return rc; +} + 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, + struct dt_allocation_hint *hint) { int rc; - rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec); - if (rc) - GOTO(out, rc); + rc = mdd_declare_object_create(env, mdd, p, c, attr, handle, spec, + def_acl_buf, acl_buf, hint); + if (rc) + GOTO(out, rc); -#ifdef CONFIG_FS_POSIX_ACL - if (got_def_acl > 0) { - struct lu_buf *acl_buf; + if (S_ISDIR(attr->la_mode)) { + rc = mdo_declare_ref_add(env, p, handle); + if (rc) + GOTO(out, rc); + } - acl_buf = mdd_buf_get(env, NULL, got_def_acl); - /* 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); - } + 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), attr->la_mode, + name->ln_name, handle); + if (rc != 0) + return rc; + + rc = mdd_declare_links_add(env, c, handle, ldata, MLAO_IGNORE); + if (rc) + return rc; + + *la = *attr; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdo_declare_attr_set(env, p, la, handle); + if (rc) + return rc; + + rc = mdd_declare_changelog_store(env, mdd, name, NULL, handle); + if (rc) + return rc; + } +out: + 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; - rc = mdo_declare_attr_set(env, c, attr, handle); - if (rc) - GOTO(out, rc); + if (S_ISLNK(la->la_mode)) { + acl_buf->lb_len = 0; + def_acl_buf->lb_len = 0; + RETURN(0); + } - rc = mdo_declare_xattr_set(env, c, acl_buf, - XATTR_NAME_ACL_ACCESS, 0, handle); - if (rc) - GOTO(out, rc); + mdd_read_lock(env, pobj, MOR_TGT_PARENT); + rc = mdo_xattr_get(env, pobj, def_acl_buf, + XATTR_NAME_ACL_DEFAULT); + 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; } -#endif - if (S_ISDIR(attr->la_mode)) { - rc = mdo_declare_ref_add(env, p, handle); - if (rc) - GOTO(out, rc); - } + RETURN(rc); +} - rc = mdd_declare_object_initialize(env, c, attr, handle); - if (rc) - GOTO(out, rc); +/** + * Create a metadata object and initialize it, set acl, xattr. + **/ +static int mdd_object_create(const struct lu_env *env, struct mdd_object *pobj, + struct mdd_object *son, struct lu_attr *attr, + struct md_op_spec *spec, struct lu_buf *acl_buf, + struct lu_buf *def_acl_buf, + struct dt_allocation_hint *hint, + struct thandle *handle) +{ + int rc; - if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) - rc = orph_declare_index_insert(env, c, attr->la_mode, handle); - else - rc = mdo_declare_index_insert(env, p, mdo2fid(c), - name->ln_name, handle); + mdd_write_lock(env, son, MOR_TGT_CHILD); + rc = mdd_object_create_internal(env, NULL, son, attr, handle, spec, + hint); if (rc) - GOTO(out, 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); - /* replay case, create LOV EA from client data */ - if (spec->no_create || (spec->sp_cr_flags & MDS_OPEN_HAS_EA)) { + /* + * 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)) { 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_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); if (rc) - GOTO(out, rc); + GOTO(err_destroy, rc); + } + /* set its own acl */ + if (acl_buf != NULL && acl_buf->lb_len > 0) { + rc = mdo_xattr_set(env, son, acl_buf, + XATTR_NAME_ACL_ACCESS, + 0, handle); + if (rc) + GOTO(err_destroy, rc); } +#endif if (S_ISLNK(attr->la_mode)) { - rc = dt_declare_record_write(env, mdd_object_child(c), - strlen(spec->u.sp_symname), 0, - handle); - if (rc) - GOTO(out, rc); - } + struct lu_ucred *uc = lu_ucred_assert(env); + struct dt_object *dt = mdd_object_child(son); + const char *target_name = spec->u.sp_symname; + int sym_len = strlen(target_name); + const struct lu_buf *buf; + loff_t pos = 0; - if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) { - rc = mdo_declare_attr_set(env, p, attr, handle); - if (rc) - return rc; - } + buf = mdd_buf_get_const(env, target_name, sym_len); + rc = dt->do_body_ops->dbo_write(env, dt, buf, &pos, handle, + uc->uc_cap & + CFS_CAP_SYS_RESOURCE_MASK); - rc = mdd_declare_changelog_store(env, mdd, name, handle); - if (rc) - return rc; + if (rc == sym_len) + rc = 0; + else + GOTO(err_initlized, rc = -EFAULT); + } -out: - return rc; +err_initlized: + if (unlikely(rc != 0)) { + int rc2; + if (S_ISDIR(attr->la_mode)) { + /* Drop the reference, no need to delete "."/"..", + * because the object to be destroied directly. */ + rc2 = mdo_ref_del(env, son, handle); + if (rc2 != 0) + GOTO(unlock, rc); + } + rc2 = mdo_ref_del(env, son, handle); + if (rc2 != 0) + GOTO(unlock, rc); +err_destroy: + mdo_destroy(env, son, handle); + } +unlock: + mdd_write_unlock(env, son); + RETURN(rc); } /* @@ -1713,10 +2263,13 @@ 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; + struct dt_allocation_hint *hint = &mdd_env_info(env)->mti_hint; + int rc; + int rc2; ENTRY; /* @@ -1755,41 +2308,48 @@ 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); - /* 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; + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + GOTO(out_free, rc = PTR_ERR(handle)); + + 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_stop, rc); - 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); - } + mdd_object_make_hint(env, mdd_pobj, son, attr, spec, hint); - mdd_object_make_hint(env, mdd_pobj, son, attr); + memset(ldata, 0, sizeof(*ldata)); + if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) { + struct lu_fid tfid = *mdd_object_fid(mdd_pobj); - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - GOTO(out_free, rc = PTR_ERR(handle)); + 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, - got_def_acl, handle, spec); + handle, spec, ldata, &def_acl_buf, &acl_buf, + hint); if (rc) GOTO(out_stop, rc); @@ -1797,159 +2357,96 @@ 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) { - mdd_write_unlock(env, son); - GOTO(cleanup, rc); - } - - 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 (rc) { - mdd_write_unlock(env, son); - GOTO(cleanup, rc); - } - } -#endif - - rc = mdd_object_initialize(env, mdo2fid(mdd_pobj), lname, - son, attr, handle, spec); - - /* - * 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 (rc == 0 && (spec->no_create || - (spec->sp_cr_flags & MDS_OPEN_HAS_EA))) { - 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 && spec->sp_cr_flags & MDS_OPEN_VOLATILE) - rc = __mdd_orphan_add(env, son, handle); - - mdd_write_unlock(env, son); - + rc = mdd_object_create(env, mdd_pobj, son, attr, spec, &acl_buf, + &def_acl_buf, hint, handle); if (rc != 0) - /* - * Object has no links, so it will be destroyed when last - * reference is released. (XXX not now.) - */ - GOTO(cleanup, rc); + GOTO(out_stop, rc); - initialized = 1; - - if (!(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) + if (unlikely(spec->sp_cr_flags & MDS_OPEN_VOLATILE)) { + mdd_write_lock(env, son, MOR_TGT_CHILD); + rc = __mdd_orphan_add(env, son, handle); + 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)); - - if (rc != 0) - GOTO(cleanup, rc); - - inserted = 1; - - if (S_ISLNK(attr->la_mode)) { - struct lu_ucred *uc = lu_ucred_assert(env); - struct dt_object *dt = mdd_object_child(son); - const char *target_name = spec->u.sp_symname; - int sym_len = strlen(target_name); - const struct lu_buf *buf; - loff_t pos = 0; - - 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); - - if (rc == sym_len) - rc = 0; - else - GOTO(cleanup, rc = -EFAULT); - } + attr->la_mode, name, handle); + if (rc != 0) + GOTO(err_created, rc); - /* volatile file creation does not update parent directory times */ - if (spec->sp_cr_flags & MDS_OPEN_VOLATILE) - GOTO(cleanup, rc = 0); + mdd_links_add(env, son, mdo2fid(mdd_pobj), lname, handle, + ldata, 1); - /* 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); - if (rc) - GOTO(cleanup, rc); + /* update parent directory mtime/ctime */ + *la = *attr; + la->la_valid = LA_CTIME | LA_MTIME; + rc = mdd_update_time(env, mdd_pobj, pattr, la, handle); + if (rc) + GOTO(err_insert, rc); + } - EXIT; -cleanup: - if (rc != 0 && created != 0) { + EXIT; +err_insert: + if (rc != 0) { int rc2; - if (inserted != 0) { - 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); - if (rc2 != 0) - goto out_stop; - } + 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); + if (rc2 != 0) + goto out_stop; +err_created: mdd_write_lock(env, son, MOR_TGT_CHILD); - if (initialized != 0 && S_ISDIR(attr->la_mode)) { + 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); - } + } - 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)) && + 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) + rc = rc2; out_free: - /* The child object shouldn't be cached anymore */ - if (rc) + 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, - &child->mo_lu.lo_header->loh_flags); - return rc; + &child->mo_lu.lo_header->loh_flags); + return rc; } /* @@ -1965,6 +2462,7 @@ 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*/ @@ -1980,7 +2478,8 @@ 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; @@ -1996,55 +2495,61 @@ 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 == NULL) + rc = mdd_may_create(env, tgt_pobj, tpattr, NULL, + (src_pobj != tgt_pobj)); + 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); @@ -2062,44 +2567,51 @@ 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 != 0) + 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), + 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); + if (rc != 0) + return rc; - rc = mdo_declare_attr_set(env, mdd_spobj, NULL, 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_sobj, NULL, handle); - if (rc) - return rc; - mdd_declare_links_add(env, mdd_sobj, 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 = mdo_declare_attr_set(env, mdd_tpobj, NULL, handle); - if (rc) - return rc; + rc = mdd_declare_links_add(env, mdd_sobj, handle, ldata, + S_ISREG(mdd_object_type(mdd_sobj)) ? MLAO_CHECK : MLAO_IGNORE); + if (rc) + return rc; - /* new name */ - rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj), - tname->ln_name, handle); - if (rc) - return rc; + /* new name */ + rc = mdo_declare_index_insert(env, mdd_tpobj, mdo2fid(mdd_sobj), + mdd_object_type(mdd_sobj), + tname->ln_name, handle); + if (rc != 0) + return rc; /* name from target dir (old name), we declare it unconditionally * as mdd_rename() calls delete unconditionally as well. so just @@ -2127,20 +2639,17 @@ 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; - - mdd_declare_links_add(env, mdd_tobj, handle); - if (rc) - return rc; + la->la_valid = LA_CTIME; + rc = mdo_declare_attr_set(env, mdd_tobj, la, 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, 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; @@ -2154,105 +2663,98 @@ 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); + if (rc) + GOTO(out_pending, rc); + + 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); + if (rc) + GOTO(out_pending, rc); + } + + rc = mdd_la_get(env, mdd_tpobj, tpattr); + if (rc) + GOTO(out_pending, rc); - if (tobj) - mdd_tobj = md2mdd_obj(tobj); + rc = mdd_rename_sanity_check(env, mdd_spobj, pattr, mdd_tpobj, tpattr, + mdd_sobj, cattr, mdd_tobj, tattr); + if (rc) + GOTO(out_pending, rc); - mdd_sobj = mdd_object_find(env, mdd, lf); + rc = mdd_name_check(mdd, ltname); + if (rc < 0) + 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, - mdd_object_capa(env, mdd_spobj)); + rc = __mdd_index_delete(env, mdd_spobj, sname, is_dir, handle); if (rc) GOTO(cleanup, rc); /* "mv dir1 dir2" needs "dir1/.." link update */ if (is_dir && mdd_sobj && !lu_fid_eq(spobj_fid, tpobj_fid)) { - rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle, - mdd_object_capa(env, mdd_sobj)); - if (rc) - GOTO(fixup_spobj2, rc); + rc = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle); + if (rc != 0) + GOTO(fixup_spobj2, rc); - rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, dotdot, - handle, mdd_object_capa(env, mdd_sobj)); - if (rc) + rc = __mdd_index_insert_only(env, mdd_sobj, tpobj_fid, S_IFDIR, + dotdot, handle); + if (rc != 0) GOTO(fixup_spobj, rc); } @@ -2260,8 +2762,7 @@ static int mdd_rename(const struct lu_env *env, * Here tobj can be remote one, so we do index_delete unconditionally * and -ENOENT is allowed. */ - rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle, - mdd_object_capa(env, mdd_tpobj)); + rc = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle); if (rc != 0) { if (mdd_tobj) { /* tname might been renamed to something else */ @@ -2272,21 +2773,21 @@ static int mdd_rename(const struct lu_env *env, } /* Insert new fid with target name into target dir */ - rc = __mdd_index_insert(env, mdd_tpobj, lf, tname, is_dir, handle, - mdd_object_capa(env, mdd_tpobj)); - if (rc) + rc = __mdd_index_insert(env, mdd_tpobj, lf, cattr->la_mode, + tname, handle); + if (rc != 0) GOTO(fixup_tpobj, rc); LASSERT(ma->ma_attr.la_valid & LA_CTIME); 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 @@ -2305,13 +2806,12 @@ 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); if (rc != 0) { CERROR("%s: Failed to get nlink for tobj " DFID": rc = %d\n", @@ -2321,7 +2821,7 @@ 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", @@ -2331,9 +2831,10 @@ 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); + 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", @@ -2343,8 +2844,7 @@ 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); if (rc != 0) { CERROR("%s: Failed to get nlink for tobj " DFID": rc = %d\n", @@ -2353,32 +2853,35 @@ 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. */ @@ -2389,8 +2892,7 @@ static int mdd_rename(const struct lu_env *env, fixup_tpobj: if (rc) { - rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle, - BYPASS_CAPA); + rc2 = __mdd_index_delete(env, mdd_tpobj, tname, is_dir, handle); if (rc2) CWARN("tp obj fix error %d\n",rc2); @@ -2402,483 +2904,1183 @@ 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) { - rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle, - BYPASS_CAPA); - - if (rc2) - CWARN("sp obj dotdot delete error %d\n",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); - } + if (rc && is_dir && mdd_sobj && mdd_spobj != mdd_tpobj) { + rc2 = __mdd_index_delete_only(env, mdd_sobj, dotdot, handle); + if (rc2) + CWARN("%s: sp obj dotdot delete error: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc2); + + + rc2 = __mdd_index_insert_only(env, mdd_sobj, spobj_fid, S_IFDIR, + dotdot, handle); + 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); - 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, - 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); - if (mdd_sobj) - mdd_object_put(env, mdd_sobj); + mdd_trans_stop(env, mdd, rc, handle); + out_pending: - return rc; + mdd_object_put(env, mdd_sobj); + return rc; +} + +/** + * During migration once the parent FID has been changed, + * we need update the parent FID in linkea. + **/ +static int mdd_linkea_update_child_internal(const struct lu_env *env, + struct mdd_object *parent, + 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 = { NULL }; + struct lu_buf *buf = &info->mti_link_buf; + int count; + int rc = 0; + + ENTRY; + + buf = lu_buf_check_and_alloc(buf, PATH_MAX); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + + ldata.ld_buf = buf; + rc = mdd_links_read(env, child, &ldata); + if (rc != 0) { + if (rc == -ENOENT || rc == -ENODATA) + rc = 0; + RETURN(rc); + } + + 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(&child->mod_obj); + struct lu_name lname; + struct lu_fid fid; + + linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, + &lname, &fid); + + 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); + continue; + } + + 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))); + /* update to the new parent fid */ + linkea_entry_pack(ldata.ld_lee, &lname, + mdd_object_fid(parent)); + if (declare) + rc = mdd_declare_links_add(env, child, handle, &ldata, + MLAO_IGNORE); + else + rc = mdd_links_write(env, child, &ldata, handle); + break; + } + RETURN(rc); +} + +static int mdd_linkea_declare_update_child(const struct lu_env *env, + struct mdd_object *parent, + struct mdd_object *child, + const char *name, int namelen, + struct thandle *handle) +{ + return mdd_linkea_update_child_internal(env, parent, child, name, + namelen, handle, true); +} + +static int mdd_linkea_update_child(const struct lu_env *env, + struct mdd_object *parent, + struct mdd_object *child, + const char *name, int namelen, + struct thandle *handle) +{ + return mdd_linkea_update_child_internal(env, parent, child, name, + namelen, handle, false); +} + +static int mdd_update_linkea_internal(const struct lu_env *env, + struct mdd_object *mdd_pobj, + struct mdd_object *mdd_sobj, + struct mdd_object *mdd_tobj, + const struct lu_name *child_name, + 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 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); + struct mdd_object *pobj; + struct lu_name lname; + struct lu_fid fid; + + linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, + &lname, &fid); + ldata->ld_lee = (struct link_ea_entry *)((char *)ldata->ld_lee + + ldata->ld_reclen); + pobj = mdd_object_find(env, mdd, &fid); + if (IS_ERR(pobj)) { + CWARN("%s: cannot find obj "DFID": rc = %ld\n", + mdd2obd_dev(mdd)->obd_name, PFID(&fid), + PTR_ERR(pobj)); + continue; + } + + 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); + } + + if (pobj == mdd_pobj && + lname.ln_namelen == child_name->ln_namelen && + strncmp(lname.ln_name, child_name->ln_name, + lname.ln_namelen) == 0) { + 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); + } + + CDEBUG(D_INFO, "%s: update "DFID" with "DNAME":"DFID"\n", + mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(pobj)), + PNAME(&lname), PFID(mdd_object_fid(mdd_tobj))); + + if (declare) { + /* Remove source name from source directory */ + /* Insert new fid with target name into target dir */ + rc = mdo_declare_index_delete(env, pobj, lname.ln_name, + handle); + if (rc != 0) + GOTO(next_put, rc); + + rc = mdo_declare_index_insert(env, pobj, + 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); + if (rc) + GOTO(next_put, rc); + + rc = mdo_declare_ref_del(env, mdd_sobj, handle); + if (rc) + GOTO(next_put, rc); + } else { + rc = __mdd_index_delete(env, pobj, lname.ln_name, + 0, handle); + if (rc) + GOTO(next_put, rc); + + rc = __mdd_index_insert(env, pobj, + mdd_object_fid(mdd_tobj), + mdd_object_type(mdd_tobj), + lname.ln_name, handle); + if (rc != 0) + GOTO(next_put, rc); + + mdd_write_lock(env, mdd_tobj, MOR_SRC_CHILD); + rc = mdo_ref_add(env, mdd_tobj, handle); + mdd_write_unlock(env, mdd_tobj); + if (rc) + GOTO(next_put, rc); + + mdd_write_lock(env, mdd_sobj, MOR_TGT_CHILD); + mdo_ref_del(env, mdd_sobj, handle); + mdd_write_unlock(env, mdd_sobj); + } +next_put: + mdd_object_put(env, pobj); + if (rc != 0) + break; + } + + RETURN(rc); +} + +static int mdd_migrate_xattrs(const struct lu_env *env, + struct mdd_object *mdd_sobj, + struct mdd_object *mdd_tobj) +{ + struct mdd_thread_info *info = mdd_env_info(env); + struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj); + char *xname; + struct thandle *handle; + struct lu_buf xbuf; + int xlen; + int rem; + int xsize; + int list_xsize; + struct lu_buf list_xbuf; + int rc; + + /* retrieve xattr list from the old object */ + list_xsize = mdo_xattr_list(env, mdd_sobj, &LU_BUF_NULL); + if (list_xsize == -ENODATA) + return 0; + + if (list_xsize < 0) + return list_xsize; + + lu_buf_check_and_alloc(&info->mti_big_buf, list_xsize); + if (info->mti_big_buf.lb_buf == NULL) + return -ENOMEM; + + 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); + if (rc < 0) + return rc; + rc = 0; + rem = list_xsize; + 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 || + strcmp(XATTR_NAME_LMV, xname) == 0) + goto next; + + /* For directory, if there are default layout, migrate here */ + if (strcmp(XATTR_NAME_LOV, xname) == 0 && + !S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) + goto next; + + xsize = mdo_xattr_get(env, mdd_sobj, &LU_BUF_NULL, xname); + if (xsize == -ENODATA) + goto next; + if (xsize < 0) + GOTO(out, rc); + + lu_buf_check_and_alloc(&info->mti_link_buf, xsize); + if (info->mti_link_buf.lb_buf == NULL) + GOTO(out, rc = -ENOMEM); + + xbuf.lb_len = xsize; + xbuf.lb_buf = info->mti_link_buf.lb_buf; + rc = mdo_xattr_get(env, mdd_sobj, &xbuf, xname); + if (rc == -ENODATA) + goto next; + if (rc < 0) + GOTO(out, rc); + + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + GOTO(out, rc = PTR_ERR(handle)); + + rc = mdo_declare_xattr_set(env, mdd_tobj, &xbuf, xname, 0, + handle); + if (rc != 0) + GOTO(stop_trans, rc); + /* Note: this transaction is part of migration, and it is not + * the last step of migration, so we set th_local = 1 to avoid + * update last rcvd for this transaction */ + handle->th_local = 1; + rc = mdd_trans_start(env, mdd, handle); + if (rc != 0) + GOTO(stop_trans, rc); + + rc = mdo_xattr_set(env, mdd_tobj, &xbuf, xname, 0, handle); + if (rc == -EEXIST) + GOTO(stop_trans, rc = 0); + + if (rc != 0) + GOTO(stop_trans, rc); +stop_trans: + mdd_trans_stop(env, mdd, rc, handle); + if (rc != 0) + GOTO(out, rc); +next: + rem -= xlen; + memmove(xname, xname + xlen, rem); + } +out: + return rc; +} + +static int mdd_declare_migrate_create(const struct lu_env *env, + struct mdd_object *mdd_pobj, + struct mdd_object *mdd_sobj, + struct mdd_object *mdd_tobj, + 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); + const struct lu_buf *buf; + int rc; + int mgr_easize; + + rc = mdd_declare_object_create_internal(env, mdd_pobj, mdd_tobj, la, + handle, spec, NULL); + if (rc != 0) + return rc; + + rc = mdd_declare_object_initialize(env, mdd_pobj, mdd_tobj, la, + handle); + if (rc != 0) + return rc; + + if (S_ISLNK(la->la_mode)) { + const char *target_name = spec->u.sp_symname; + int sym_len = strlen(target_name); + const struct lu_buf *buf; + + buf = mdd_buf_get_const(env, target_name, sym_len); + rc = dt_declare_record_write(env, mdd_object_child(mdd_tobj), + buf, 0, handle); + if (rc != 0) + return rc; + } else if (S_ISDIR(la->la_mode)) { + rc = mdd_declare_links_add(env, mdd_tobj, handle, ldata, + MLAO_IGNORE); + if (rc != 0) + return rc; + } + + if (spec->u.sp_ea.eadata != NULL && spec->u.sp_ea.eadatalen != 0) { + buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, + spec->u.sp_ea.eadatalen); + rc = mdo_declare_xattr_set(env, mdd_tobj, buf, XATTR_NAME_LOV, + 0, handle); + if (rc) + return rc; + } + + 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); + if (rc) + return rc; + + la_flag->la_valid = 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; +} + +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 = { 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; + + /* prepare spec for create */ + memset(spec, 0, sizeof(*spec)); + spec->sp_cr_lookup = 0; + spec->sp_feat = &dt_directory_features; + if (S_ISLNK(la->la_mode)) { + buf = lu_buf_check_and_alloc( + &mdd_env_info(env)->mti_big_buf, + 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; + CERROR("%s: "DFID" readlink failed: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(mdd_object_fid(mdd_sobj)), rc); + RETURN(rc); + } + spec->u.sp_symname = link_buf.lb_buf; + } 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) + RETURN(rc); + if (lmm_buf.lb_buf != NULL && lmm_buf.lb_len != 0) { + spec->u.sp_ea.eadata = lmm_buf.lb_buf; + spec->u.sp_ea.eadatalen = lmm_buf.lb_len; + spec->sp_cr_flags |= MDS_OPEN_HAS_EA; + } + } else if (S_ISDIR(la->la_mode)) { + rc = mdd_links_read(env, mdd_sobj, ldata); + if (rc < 0 && rc != -ENODATA) + RETURN(rc); + } + + mgr_ea = (struct lmv_mds_md_v1 *)info->mti_xattr_buf; + memset(mgr_ea, 0, sizeof(*mgr_ea)); + 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_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)); + + /* Note: this transaction is part of migration, and it is not + * the last step of migration, so we set th_local = 1 to avoid + * update last rcvd for this transaction */ + handle->th_local = 1; + rc = mdd_declare_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, + spec, la, + (union lmv_mds_md *)info->mti_xattr_buf, + ldata, handle); + if (rc != 0) + GOTO(stop_trans, rc); + + rc = mdd_trans_start(env, mdd, handle); + if (rc != 0) + GOTO(stop_trans, rc); + + /* create the target object */ + rc = mdd_object_create(env, mdd_pobj, mdd_tobj, la, spec, NULL, NULL, + hint, handle); + if (rc != 0) + GOTO(stop_trans, rc); + + if (S_ISDIR(la->la_mode)) { + rc = mdd_links_write(env, mdd_tobj, ldata, handle); + if (rc != 0) + GOTO(stop_trans, rc); + } + + /* Set MIGRATE EA on the source inode, so once the migration needs + * to be re-done during failover, the re-do process can locate the + * target object which is already being created. */ + mgr_easize = lmv_mds_md_size(2, LMV_MAGIC_V1); + buf = mdd_buf_get_const(env, mgr_ea, mgr_easize); + rc = mdo_xattr_set(env, mdd_sobj, buf, XATTR_NAME_LMV, 0, handle); + if (rc != 0) + GOTO(stop_trans, rc); + + /* Set immutable flag, so any modification is disabled until + * the migration is done. Once the migration is interrupted, + * if the resume process find the migrating object has both + * 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 = 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); +out_free: + if (lmm_buf.lb_buf != NULL) + OBD_FREE(lmm_buf.lb_buf, lmm_buf.lb_len); + RETURN(rc); +} + +static int mdd_migrate_entries(const struct lu_env *env, + struct mdd_object *mdd_sobj, + struct mdd_object *mdd_tobj) +{ + struct dt_object *next = mdd_object_child(mdd_sobj); + struct mdd_device *mdd = mdo2mdd(&mdd_sobj->mod_obj); + struct dt_object *dt_tobj = mdd_object_child(mdd_tobj); + struct thandle *handle; + struct dt_it *it; + const struct dt_it_ops *iops; + int rc; + int result; + struct lu_dirent *ent; + ENTRY; + + OBD_ALLOC(ent, NAME_MAX + sizeof(*ent) + 1); + if (ent == NULL) + RETURN(-ENOMEM); + + if (!dt_try_as_dir(env, next)) + GOTO(out_ent, rc = -ENOTDIR); + /* + * iterate directories + */ + iops = &next->do_index_ops->dio_it; + it = iops->init(env, next, LUDA_FID | LUDA_TYPE); + if (IS_ERR(it)) + GOTO(out_ent, rc = PTR_ERR(it)); + + rc = iops->load(env, it, 0); + if (rc == 0) + rc = iops->next(env, it); + else if (rc > 0) + rc = 0; + /* + * At this point and across for-loop: + * + * rc == 0 -> ok, proceed. + * rc > 0 -> end of directory. + * rc < 0 -> error. + */ + do { + struct mdd_object *child; + char *name = mdd_env_info(env)->mti_key; + int len; + int recsize; + int is_dir; + bool target_exist = false; + + len = iops->key_size(env, it); + if (len == 0) + goto next; + + result = iops->rec(env, it, (struct dt_rec *)ent, + LUDA_FID | LUDA_TYPE); + if (result == -ESTALE) + goto next; + if (result != 0) { + rc = result; + goto out; + } + + 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] == '.') || + (ent->lde_namelen == 2 && ent->lde_name[0] == '.' && + ent->lde_name[1] == '.')) + goto next; + + child = mdd_object_find(env, mdd, &ent->lde_fid); + if (IS_ERR(child)) + GOTO(out, rc = PTR_ERR(child)); + + is_dir = S_ISDIR(mdd_object_type(child)); + + snprintf(name, ent->lde_namelen + 1, "%s", ent->lde_name); + + /* Check whether the name has been inserted to the target */ + if (dt_try_as_dir(env, dt_tobj)) { + struct lu_fid *fid = &mdd_env_info(env)->mti_fid2; + + 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)); + + /* Note: this transaction is part of migration, and it is not + * the last step of migration, so we set th_local = 1 to avoid + * updating last rcvd for this transaction */ + handle->th_local = 1; + 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); + + if (is_dir) { + rc = mdo_declare_ref_add(env, mdd_tobj, handle); + if (rc != 0) + GOTO(out_put, rc); + } + } + + rc = mdo_declare_index_delete(env, mdd_sobj, name, handle); + if (rc != 0) + GOTO(out_put, rc); + + if (is_dir) { + rc = mdo_declare_ref_del(env, mdd_sobj, handle); + if (rc != 0) + GOTO(out_put, rc); + + /* Update .. for child */ + rc = mdo_declare_index_delete(env, child, dotdot, + handle); + if (rc != 0) + GOTO(out_put, rc); + + rc = mdo_declare_index_insert(env, child, + mdd_object_fid(mdd_tobj), + S_IFDIR, dotdot, handle); + if (rc != 0) + GOTO(out_put, rc); + } + + rc = mdd_linkea_declare_update_child(env, mdd_tobj, + child, name, + strlen(name), + handle); + if (rc != 0) + GOTO(out_put, rc); + + rc = mdd_trans_start(env, mdd, handle); + if (rc != 0) { + CERROR("%s: transaction start failed: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, rc); + GOTO(out_put, rc); + } + + if (likely(!target_exist)) { + rc = __mdd_index_insert(env, mdd_tobj, &ent->lde_fid, + mdd_object_type(child), + name, handle); + if (rc != 0) + GOTO(out_put, rc); + } + + 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); + if (rc != 0) + GOTO(out_put, rc); + + rc = __mdd_index_insert_only(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, + strlen(name), handle); + +out_put: + mdd_object_put(env, child); + mdd_trans_stop(env, mdd, rc, handle); + if (rc != 0) + GOTO(out, rc); +next: + result = iops->next(env, it); + if (OBD_FAIL_CHECK(OBD_FAIL_MIGRATE_ENTRIES)) + GOTO(out, rc = -EINTR); + + if (result == -ESTALE) + goto next; + } while (result == 0); +out: + iops->put(env, it); + iops->fini(env, it); +out_ent: + OBD_FREE(ent, NAME_MAX + sizeof(*ent) + 1); + 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_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) +{ + return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj, + child_name, handle, 1); +} -static int mdd_links_new(const struct lu_env *env, - struct mdd_link_data *ldata) +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) { - 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; + return mdd_update_linkea_internal(env, mdd_pobj, mdd_sobj, mdd_tobj, + child_name, handle, 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) +static int mdd_declare_migrate_update_name(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 lu_attr *parent_la, + struct thandle *handle) { - struct lustre_capa *capa; - struct link_ea_header *leh; - int rc; + struct lu_attr *la_flag = MDD_ENV_VAR(env, tattr); + 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; + /* Revert IMMUTABLE flag */ + la_flag->la_valid = LA_FLAGS; + la_flag->la_flags = la->la_flags & ~LUSTRE_IMMUTABLE_FL; + rc = mdo_declare_attr_set(env, mdd_sobj, la_flag, handle); + if (rc != 0) + return rc; - if (!mdd_object_exists(mdd_obj)) - return -ENODATA; + /* delete entry from source dir */ + rc = mdo_declare_index_delete(env, mdd_pobj, lname->ln_name, handle); + if (rc != 0) + return rc; - 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) + rc = mdd_declare_update_linkea(env, mdd_pobj, mdd_sobj, + mdd_tobj, handle, lname); + 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; - 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 (S_ISDIR(mdd_object_type(mdd_sobj))) { + rc = mdo_declare_ref_del(env, mdd_pobj, handle); + if (rc != 0) + return rc; } - if (leh->leh_magic != LINK_EA_MAGIC) - return -EINVAL; - if (leh->leh_reccount == 0) - return -ENODATA; - ldata->ml_leh = leh; - return 0; -} + /* 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; -/** 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_declare_links_add(env, mdd_tobj, handle, NULL, MLAO_IGNORE); + if (rc != 0) + return rc; - rc = mdd_links_read(env, mdd_obj, &ldata); - return rc ? ERR_PTR(rc) : ldata.ml_buf; -} + if (S_ISDIR(mdd_object_type(mdd_sobj))) { + rc = mdo_declare_ref_add(env, mdd_pobj, handle); + if (rc != 0) + return rc; + } -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)); -} + /* delete old object */ + rc = mdo_declare_ref_del(env, mdd_sobj, handle); + if (rc != 0) + return rc; -/** 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; + if (S_ISDIR(mdd_object_type(mdd_sobj))) { + /* delete old object */ + rc = mdo_declare_ref_del(env, mdd_sobj, handle); + if (rc != 0) + return rc; + /* set nlink to 0 */ + rc = mdo_declare_attr_set(env, mdd_sobj, la, handle); + if (rc != 0) + return rc; + } - 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; + rc = mdd_declare_finish_unlink(env, mdd_sobj, handle); + if (rc) + return rc; - lee->lee_reclen[0] = (reclen >> 8) & 0xff; - lee->lee_reclen[1] = reclen & 0xff; - return reclen; -} + rc = mdo_declare_attr_set(env, mdd_pobj, parent_la, handle); -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); + return rc; } -static int mdd_declare_links_add(const struct lu_env *env, - struct mdd_object *mdd_obj, - struct thandle *handle) +static int mdd_migrate_update_name(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 md_attr *ma) { - int rc; + struct lu_attr *p_la = MDD_ENV_VAR(env, la_for_fix); + 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 thandle *handle; + int is_dir = S_ISDIR(mdd_object_type(mdd_sobj)); + const char *name = lname->ln_name; + int rc; + ENTRY; - /* XXX: max size? */ - rc = mdo_declare_xattr_set(env, mdd_obj, - mdd_buf_get_const(env, NULL, 4096), - XATTR_NAME_LINK, 0, handle); + /* update time for parent */ + LASSERT(ma->ma_attr.la_valid & LA_CTIME); + p_la->la_ctime = p_la->la_mtime = ma->ma_attr.la_ctime; + p_la->la_valid = LA_CTIME; - return rc; -} + rc = mdd_la_get(env, mdd_sobj, so_attr); + if (rc != 0) + 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 + handle = mdd_trans_create(env, mdd); + if (IS_ERR(handle)) + RETURN(PTR_ERR(handle)); -/** 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); + rc = mdd_declare_migrate_update_name(env, mdd_pobj, mdd_sobj, mdd_tobj, + lname, so_attr, p_la, handle); + if (rc != 0) { + /* If the migration can not be fit in one transaction, just + * leave it in the original MDT */ + if (rc == -E2BIG) + GOTO(stop_trans, rc = 0); + else + GOTO(stop_trans, rc); + } - if (lname == NULL || pfid == NULL) - return -EINVAL; + CDEBUG(D_INFO, "%s: update "DFID"/"DFID" with %s:"DFID"\n", + mdd2obd_dev(mdd)->obd_name, PFID(mdd_object_fid(mdd_pobj)), + PFID(mdd_object_fid(mdd_sobj)), lname->ln_name, + PFID(mdd_object_fid(mdd_tobj))); + + rc = mdd_trans_start(env, mdd, handle); + if (rc != 0) + GOTO(stop_trans, rc); - /* Make sure our buf is big enough for the new one */ - if (ldata->ml_leh->leh_reccount > LINKEA_MAX_COUNT) - return -EOVERFLOW; + /* Revert IMMUTABLE flag */ + la_flag->la_valid = LA_FLAGS; + la_flag->la_flags = so_attr->la_flags & ~LUSTRE_IMMUTABLE_FL; + rc = mdo_attr_set(env, mdd_sobj, la_flag, handle); + if (rc != 0) + GOTO(stop_trans, rc); - 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; + /* Remove source name from source directory */ + 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 (S_ISREG(so_attr->la_mode)) { + if (so_attr->la_nlink == 1) { + rc = mdo_xattr_del(env, mdd_sobj, XATTR_NAME_LOV, + handle); + if (rc != 0 && rc != -ENODATA) + GOTO(stop_trans, rc); + } } - 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; -} + /* Insert new fid with target name into target dir */ + rc = __mdd_index_insert(env, mdd_pobj, mdd_object_fid(mdd_tobj), + mdd_object_type(mdd_tobj), name, handle); + if (rc != 0) + GOTO(stop_trans, rc); -/** 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); + rc = mdd_links_add(env, mdd_tobj, mdo2fid(mdd_pobj), lname, handle, + NULL, 1); + if (rc != 0) + GOTO(stop_trans, rc); - 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); + 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); -/** - * 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); - } + ma->ma_attr = *so_attr; + ma->ma_valid |= MA_INODE; + rc = mdd_finish_unlink(env, mdd_sobj, ma, mdd_pobj, lname, handle); + if (rc != 0) + GOTO(out_unlock, rc); - 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; + rc = mdd_attr_set_internal(env, mdd_pobj, p_la, handle, 0); + if (rc != 0) + GOTO(out_unlock, rc); + +out_unlock: + mdd_write_unlock(env, mdd_sobj); + +stop_trans: + mdd_trans_stop(env, mdd, rc, handle); + + RETURN(rc); } -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) +/** + * Check whether we should migrate the file/dir + * return val + * < 0 permission check failed or other error. + * = 0 the file can be migrated. + * > 0 the file does not need to be migrated, mostly + * for multiple link file + **/ +static int mdd_migrate_sanity_check(const struct lu_env *env, + struct mdd_object *pobj, + const struct lu_attr *pattr, + struct mdd_object *sobj, + struct lu_attr *sattr) { - int rc; + struct mdd_thread_info *info = mdd_env_info(env); + struct linkea_data *ldata = &info->mti_link_data; + int mgr_easize; + struct lu_buf *mgr_buf; + int count; + 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; + ENTRY; + + 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); + if (rc > 0) { + union lmv_mds_md *lmm = mgr_buf->lb_buf; + + /* If the object has migrateEA, it means IMMUTE flag + * 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_hash_type) & + LMV_HASH_FLAG_MIGRATION) { + struct mdd_device *mdd = mdo2mdd(&sobj->mod_obj); + + sattr->la_flags &= ~LUSTRE_IMMUTABLE_FL; + CDEBUG(D_HA, "%s: "DFID" override IMMUTE FLAG\n", + mdd2obd_dev(mdd)->obd_name, + PFID(mdd_object_fid(sobj))); } } - if (check) { - rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid); - if (rc && rc != -ENOENT) - return rc; - if (rc == 0) - return -EEXIST; + rc = mdd_rename_sanity_check(env, pobj, pattr, pobj, pattr, + sobj, sattr, NULL, NULL); + if (rc != 0) + RETURN(rc); + + /* Then it will check if the file should be migrated. If the file + * has mulitple links, we only need migrate the file if all of its + * entries has been migrated to the remote MDT */ + if (!S_ISREG(sattr->la_mode) || sattr->la_nlink < 2) + RETURN(0); + + rc = mdd_links_read(env, sobj, ldata); + 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 */ + if (rc == -ENOENT || rc == -ENODATA) + RETURN(1); + RETURN(rc); } - return mdd_links_add_buf(env, ldata, lname, pfid); -} + /* 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(&sobj->mod_obj); + struct mdd_object *lpobj; + struct lu_name lname; + struct lu_fid fid; + + 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; + } -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 (!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); + continue; + } - if (ldata->ml_leh == NULL) { - rc = mdd_links_read(env, mdd_obj, ldata); - if (rc) - return rc; + 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; } - rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid); - if (rc) - return rc; - - mdd_links_del_buf(env, ldata, lname); - return 0; + RETURN(rc); } -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) +static int mdd_migrate(const struct lu_env *env, struct md_object *pobj, + struct md_object *sobj, const struct lu_name *lname, + struct md_object *tobj, struct md_attr *ma) { - struct mdd_link_data ldata = { 0 }; - int updated = 0; - int rc2 = 0; - int rc = 0; + struct mdd_object *mdd_pobj = md2mdd_obj(pobj); + struct mdd_device *mdd = mdo2mdd(pobj); + struct mdd_object *mdd_sobj = md2mdd_obj(sobj); + struct mdd_object *mdd_tobj = NULL; + struct lu_attr *so_attr = MDD_ENV_VAR(env, cattr); + struct lu_attr *pattr = MDD_ENV_VAR(env, pattr); + int rc; + ENTRY; + /* If the file will being migrated, it will check whether + * the file is being opened by someone else right now */ + mdd_read_lock(env, mdd_sobj, MOR_SRC_CHILD); + if (mdd_sobj->mod_count >= 1) { + CERROR("%s: "DFID"%s is already opened count %d: rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(mdd_object_fid(mdd_sobj)), lname->ln_name, + mdd_sobj->mod_count, -EBUSY); + mdd_read_unlock(env, mdd_sobj); + GOTO(put, rc = -EBUSY); + } + mdd_read_unlock(env, mdd_sobj); - LASSERT(oldpfid != NULL || newpfid != NULL); + rc = mdd_la_get(env, mdd_sobj, so_attr); + if (rc != 0) + GOTO(put, rc); - if (mdd_obj->mod_flags & DEAD_OBJ) - /* No more links, don't bother */ - RETURN(0); + rc = mdd_la_get(env, mdd_pobj, pattr); + if (rc != 0) + GOTO(put, rc); - 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 = mdd_migrate_sanity_check(env, mdd_pobj, pattr, mdd_sobj, so_attr); + if (rc != 0) { + if (rc > 0) rc = 0; - } else { - updated = 1; - } + GOTO(put, rc); } - /* 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; - } + /* Sigh, it is impossible to finish all of migration in a single + * transaction, for example migrating big directory entries to the + * new MDT, it needs insert all of name entries of children in the + * new directory. + * + * So migration will be done in multiple steps and transactions. + * + * 1. create an orphan object on the remote MDT in one transaction. + * 2. migrate extend attributes to the new target file/directory. + * 3. For directory, migrate the entries to the new MDT and update + * linkEA of each children. Because we can not migrate all entries + * in a single transaction, so the migrating directory will become + * a striped directory during migration, so once the process is + * interrupted, the directory is still accessible. (During lookup, + * client will locate the name by searching both original and target + * object). + * 4. Finally, update the name/FID to point to the new file/directory + * in a separate transaction. + */ - 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))); + /* step 1: Check whether the orphan object has been created, and create + * orphan object on the remote MDT if needed */ + mdd_tobj = md2mdd_obj(tobj); + if (!mdd_object_exists(mdd_tobj)) { + rc = mdd_migrate_create(env, mdd_pobj, mdd_sobj, mdd_tobj, + lname, so_attr); + if (rc != 0) + GOTO(put, rc); } - 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); - - return rc; -} + /* step 2: migrate xattr */ + rc = mdd_migrate_xattrs(env, mdd_sobj, mdd_tobj); + if (rc != 0) + GOTO(put, 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); -} + /* step 3: migrate name entries to the orphan object */ + if (S_ISDIR(lu_object_attr(&mdd_sobj->mod_obj.mo_lu))) { + rc = mdd_migrate_entries(env, mdd_sobj, mdd_tobj); + if (rc != 0) + GOTO(put, rc); + 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); + } -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); + /* 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: + RETURN(rc); } const struct md_dir_operations mdd_dir_ops = { - .mdo_is_subdir = mdd_is_subdir, - .mdo_lookup = mdd_lookup, - .mdo_create = mdd_create, - .mdo_rename = mdd_rename, - .mdo_link = mdd_link, - .mdo_unlink = mdd_unlink, - .mdo_create_data = mdd_create_data, + .mdo_is_subdir = mdd_is_subdir, + .mdo_lookup = mdd_lookup, + .mdo_create = mdd_create, + .mdo_rename = mdd_rename, + .mdo_link = mdd_link, + .mdo_unlink = mdd_unlink, + .mdo_create_data = mdd_create_data, + .mdo_migrate = mdd_migrate, };