X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdd%2Fmdd_dir.c;h=6170636cdde738ce16f7ae2d3b8d08b0687fd1c3;hp=9e440ba2b4b92117c0dd037f44b903cb5b06210e;hb=a82057b43d5f6680d0be386126465ac88be19564;hpb=f30581aa19447ea58f0120488acaa57bcd55144d diff --git a/lustre/mdd/mdd_dir.c b/lustre/mdd/mdd_dir.c index 9e440ba..6170636 100644 --- a/lustre/mdd/mdd_dir.c +++ b/lustre/mdd/mdd_dir.c @@ -42,16 +42,9 @@ #define DEBUG_SUBSYSTEM S_MDS -#include -#include -#include #include -#include #include -#include -#include #include -#include #include #include "mdd_internal.h" @@ -70,18 +63,24 @@ static int __mdd_lookup(const struct lu_env *env, struct md_object *pobj, static int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, struct thandle *handle); -static 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_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); + 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_lookup_locked(const struct lu_env *env, struct md_object *pobj, @@ -307,7 +306,6 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, 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); @@ -318,7 +316,7 @@ int mdd_may_create(const struct lu_env *env, struct mdd_object *pobj, * 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 md_attr *ma) + const struct lu_attr *attr) { int rc; ENTRY; @@ -326,8 +324,8 @@ int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj, if (mdd_is_dead_obj(pobj)) RETURN(-ENOENT); - if ((ma->ma_attr.la_valid & LA_FLAGS) && - (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) + if ((attr->la_valid & LA_FLAGS) && + (attr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) RETURN(-EPERM); rc = mdd_permission_internal_locked(env, pobj, NULL, @@ -347,31 +345,31 @@ int mdd_may_unlink(const struct lu_env *env, struct mdd_object *pobj, * VTX feature has been checked already, no need check again. */ static inline int mdd_is_sticky(const struct lu_env *env, - struct mdd_object *pobj, - struct mdd_object *cobj) + struct mdd_object *pobj, + struct mdd_object *cobj) { - struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la; - struct md_ucred *uc = md_ucred(env); - int rc; + 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 (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->mu_fsuid)) - return 0; - } + if (!(tmp_la->la_mode & S_ISVTX) || + (tmp_la->la_uid == uc->uc_fsuid)) + return 0; + } - rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA); - if (rc) - return rc; + rc = mdd_la_get(env, cobj, tmp_la, BYPASS_CAPA); + if (rc) + return rc; - if (tmp_la->la_uid == uc->mu_fsuid) - return 0; + if (tmp_la->la_uid == uc->uc_fsuid) + return 0; - return !mdd_capable(uc, CFS_CAP_FOWNER); + return !mdd_capable(uc, CFS_CAP_FOWNER); } /* @@ -379,8 +377,8 @@ static inline int mdd_is_sticky(const struct lu_env *env, * pobj maybe NULL */ int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj, - struct mdd_object *cobj, struct md_attr *ma, - int check_perm, int check_empty) + struct mdd_object *cobj, struct lu_attr *cattr, + struct lu_attr *src_attr, int check_perm, int check_empty) { int rc = 0; ENTRY; @@ -411,29 +409,31 @@ int mdd_may_delete(const struct lu_env *env, struct mdd_object *pobj, RETURN(-EPERM); } - if (!(ma->ma_attr_flags & MDS_VTX_BYPASS) && - mdd_is_sticky(env, pobj, cobj)) + if (mdd_is_sticky(env, pobj, cobj)) RETURN(-EPERM); if (mdd_is_immutable(cobj) || mdd_is_append(cobj)) RETURN(-EPERM); - if ((ma->ma_attr.la_valid & LA_FLAGS) && - (ma->ma_attr.la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) + if ((cattr->la_valid & LA_FLAGS) && + (cattr->la_flags & (LUSTRE_APPEND_FL | LUSTRE_IMMUTABLE_FL))) RETURN(-EPERM); - if (S_ISDIR(ma->ma_attr.la_mode)) { - struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj); + /* additional check the rename case */ + if (src_attr) { + if (S_ISDIR(src_attr->la_mode)) { + struct mdd_device *mdd = mdo2mdd(&cobj->mod_obj); - if (!S_ISDIR(mdd_object_type(cobj))) - RETURN(-ENOTDIR); + if (!S_ISDIR(cattr->la_mode)) + RETURN(-ENOTDIR); - if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid)) - RETURN(-EBUSY); - } else if (S_ISDIR(mdd_object_type(cobj))) - RETURN(-EISDIR); + if (lu_fid_eq(mdo2fid(cobj), &mdd->mdd_root_fid)) + RETURN(-EBUSY); + } else if (S_ISDIR(cattr->la_mode)) + RETURN(-EISDIR); + } - if (S_ISDIR(ma->ma_attr.la_mode) && check_empty) + if (S_ISDIR(cattr->la_mode) && check_empty) rc = mdd_dir_is_empty(env, cobj); RETURN(rc); @@ -499,27 +499,28 @@ static int __mdd_index_delete_only(const struct lu_env *env, struct mdd_object * } 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) + struct mdd_object *pobj, + const struct lu_fid *lf, const char *name, + struct thandle *handle, + struct lustre_capa *capa) { - struct dt_object *next = mdd_object_child(pobj); - int rc; - ENTRY; - - if (dt_try_as_dir(env, next)) { - struct md_ucred *uc = md_ucred(env); - - rc = next->do_index_ops->dio_insert(env, next, - (struct dt_rec*)lf, - (const struct dt_key *)name, - handle, capa, uc->mu_cap & - CFS_CAP_SYS_RESOURCE_MASK); - } else { - rc = -ENOTDIR; - } - RETURN(rc); + 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; + + 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); + } else { + rc = -ENOTDIR; + } + RETURN(rc); } /* insert named index, add reference if isdir */ @@ -612,22 +613,153 @@ int mdd_declare_llog_record(const struct lu_env *env, struct mdd_device *mdd, } int mdd_declare_changelog_store(const struct lu_env *env, - struct mdd_device *mdd, - const struct lu_name *fname, - struct thandle *handle) + struct mdd_device *mdd, + const struct lu_name *fname, + struct thandle *handle) { - int reclen; + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; + struct llog_changelog_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) + + (fname != NULL ? fname->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; + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle); + llog_ctxt_put(ctxt); + + return rc; +} - /* Not recording */ - if (!(mdd->mdd_cl.mc_flags & CLM_ON)) - return 0; +static int mdd_declare_changelog_ext_store(const struct lu_env *env, + struct mdd_device *mdd, + const struct lu_name *tname, + const struct lu_name *sname, + struct thandle *handle) +{ + 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; + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + rc = llog_declare_add(env, ctxt->loc_handle, &rec->cr_hdr, handle); + llog_ctxt_put(ctxt); + + return rc; +} - /* we'll be writing payload + llog header */ - reclen = sizeof(struct llog_changelog_rec); - if (fname) - reclen += llog_data_len(fname->ln_namelen); +/** Add a changelog 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_store(const struct lu_env *env, struct mdd_device *mdd, + struct llog_changelog_rec *rec, struct thandle *th) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; + int rc; + + rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + rec->cr.cr_namelen); + rec->cr_hdr.lrh_type = CHANGELOG_REC; + rec->cr.cr_time = cl_time(); + + spin_lock(&mdd->mdd_cl.mc_lock); + /* NB: I suppose it's possible llog_add adds out of order wrt cr_index, + * but as long as the MDD transactions are ordered correctly for e.g. + * rename conflicts, I don't think this should matter. */ + rec->cr.cr_index = ++mdd->mdd_cl.mc_index; + spin_unlock(&mdd->mdd_cl.mc_lock); + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + rc = llog_add(env, ctxt->loc_handle, &rec->cr_hdr, NULL, NULL, th); + llog_ctxt_put(ctxt); + if (rc > 0) + rc = 0; + return rc; +} - return mdd_declare_llog_record(env, mdd, reclen, handle); +/** Add a changelog_ext entry \a rec to the changelog llog + * \param mdd + * \param rec + * \param handle - currently ignored since llogs start their own transaction; + * this will hopefully be fixed in llog rewrite + * \retval 0 ok + */ +int mdd_changelog_ext_store(const struct lu_env *env, struct mdd_device *mdd, + struct llog_changelog_ext_rec *rec, + struct thandle *th) +{ + struct obd_device *obd = mdd2obd_dev(mdd); + struct llog_ctxt *ctxt; + int rc; + + rec->cr_hdr.lrh_len = llog_data_len(sizeof(*rec) + 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(); + + spin_lock(&mdd->mdd_cl.mc_lock); + /* NB: I suppose it's possible llog_add adds out of order wrt cr_index, + * but as long as the MDD transactions are ordered correctly for e.g. + * rename conflicts, I don't think this should matter. */ + rec->cr.cr_index = ++mdd->mdd_cl.mc_index; + spin_unlock(&mdd->mdd_cl.mc_lock); + + ctxt = llog_get_context(obd, LLOG_CHANGELOG_ORIG_CTXT); + if (ctxt == NULL) + return -ENXIO; + + /* 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; + + return rc; } /** Store a namespace change changelog record @@ -635,63 +767,139 @@ int mdd_declare_changelog_store(const struct lu_env *env, * want the change to commit without the log entry. * \param target - mdd_object of change * \param parent - parent dir/object - * \param tf - target lu_fid, overrides fid of \a target if this is non-null * \param tname - target name string * \param handle - transacion handle */ static int mdd_changelog_ns_store(const struct lu_env *env, - struct mdd_device *mdd, - enum changelog_rec_type type, - int flags, - struct mdd_object *target, - struct mdd_object *parent, - const struct lu_fid *tf, - const struct lu_name *tname, - struct thandle *handle) + 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) { - const struct lu_fid *tfid; - const struct lu_fid *tpfid = mdo2fid(parent); - 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); + 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(tname != NULL); + LASSERT(handle != NULL); + + reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen); + buf = mdd_buf_alloc(env, reclen); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + rec = buf->lb_buf; + + rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags); + rec->cr.cr_type = (__u32)type; + rec->cr.cr_tfid = *mdo2fid(target); + rec->cr.cr_pfid = *mdo2fid(parent); + rec->cr.cr_namelen = tname->ln_namelen; + memcpy(rec->cr.cr_name, tname->ln_name, tname->ln_namelen); + + target->mod_cltime = cfs_time_current_64(); + + 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); + } + + RETURN(0); +} - LASSERT(parent != NULL); - LASSERT(tname != NULL); - LASSERT(handle != NULL); - - /* target */ - reclen = llog_data_len(sizeof(*rec) + tname->ln_namelen); - buf = mdd_buf_alloc(env, reclen); - if (buf->lb_buf == NULL) - RETURN(-ENOMEM); - rec = (struct llog_changelog_rec *)buf->lb_buf; - - rec->cr.cr_flags = CLF_VERSION | (CLF_FLAGMASK & flags); - rec->cr.cr_type = (__u32)type; - tfid = tf ? tf : mdo2fid(target); - rec->cr.cr_tfid = *tfid; - rec->cr.cr_pfid = *tpfid; - rec->cr.cr_namelen = tname->ln_namelen; - memcpy(rec->cr.cr_name, tname->ln_name, rec->cr.cr_namelen); - if (likely(target)) - target->mod_cltime = cfs_time_current_64(); - - rc = mdd_changelog_llog_write(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(tfid), PFID(tpfid)); - return -EFAULT; - } - return 0; +/** 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) +{ + struct llog_changelog_ext_rec *rec; + struct lu_buf *buf; + int reclen; + int rc; + ENTRY; + + /* Not recording */ + if (!(mdd->mdd_cl.mc_flags & CLM_ON)) + RETURN(0); + if ((mdd->mdd_cl.mc_mask & (1 << type)) == 0) + RETURN(0); + + LASSERT(sfid != NULL); + LASSERT(tpfid != NULL); + LASSERT(tname != NULL); + LASSERT(handle != NULL); + + reclen = llog_data_len(sizeof(*rec) + + sname != NULL ? 1 + sname->ln_namelen : 0); + buf = mdd_buf_alloc(env, reclen); + if (buf->lb_buf == NULL) + RETURN(-ENOMEM); + rec = buf->lb_buf; + + 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) { + LASSERT(sfid != NULL); + 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; + } + + 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_ext_store(env, mdd, rec, handle); + if (rc < 0) { + CERROR("changelog failed: rc=%d, op%d %s c"DFID" p"DFID"\n", + rc, type, tname->ln_name, PFID(sfid), PFID(tpfid)); + return -EFAULT; + } + + return 0; } static int mdd_declare_link(const struct lu_env *env, @@ -739,34 +947,9 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, struct mdd_device *mdd = mdo2mdd(src_obj); struct dynlock_handle *dlh; struct thandle *handle; -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 }; -#endif int rc; ENTRY; -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_tobj, la_tmp, BYPASS_CAPA); - if (!rc) { - void *data = NULL; - mdd_data_get(env, mdd_tobj, &data); - quota_opc = FSFILT_OP_LINK; - mdd_quota_wrapper(la_tmp, qids); - /* get block quota for parent */ - lquota_chkquota(mds_quota_interface_ref, obd, exp, - qids, rec_pending, 1, NULL, - LQUOTA_FLAGS_BLK, data, 1); - } - } -#endif - handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_pending, rc = PTR_ERR(handle)); @@ -794,13 +977,18 @@ static int mdd_link(const struct lu_env *env, struct md_object *tgt_obj, if (rc) GOTO(out_unlock, rc); - mdo_ref_add(env, mdd_sobj, handle); + rc = mdo_ref_add(env, mdd_sobj, handle); + if (rc != 0) { + __mdd_index_delete_only(env, mdd_tobj, name, handle, + mdd_object_capa(env, mdd_tobj)); + 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_locked(env, mdd_tobj, la, handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); if (rc) GOTO(out_unlock, rc); @@ -817,21 +1005,11 @@ out_unlock: 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, NULL, lname, handle); + rc = mdd_changelog_ns_store(env, mdd, CL_HARDLINK, 0, mdd_sobj, + mdd_tobj, lname, handle); stop: mdd_trans_stop(env, mdd, rc, handle); out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (quota_opc) { - lquota_pending_commit(mds_quota_interface_ref, obd, - qids, rec_pending, 1); - /* Trigger dqacq for the parent owner. If failed, - * the next call for lquota_chkquota will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc, - quota_opc); - } -#endif return rc; } @@ -846,7 +1024,7 @@ int mdd_declare_finish_unlink(const struct lu_env *env, if (rc) return rc; - return mdd_declare_object_kill(env, obj, ma, handle); + return mdo_declare_destroy(env, obj, handle); } /* caller should take a lock before calling */ @@ -854,17 +1032,13 @@ int mdd_finish_unlink(const struct lu_env *env, struct mdd_object *obj, struct md_attr *ma, struct thandle *th) { - int rc; - int reset = 1; + int rc = 0; int is_dir = S_ISDIR(ma->ma_attr.la_mode); ENTRY; LASSERT(mdd_write_locked(env, obj) != 0); - /* read HSM flags, needed to set changelogs flags */ - ma->ma_need = MA_HSM | MA_INODE; - rc = mdd_attr_get_internal(env, obj, ma); - if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) { + if (rc == 0 && (ma->ma_attr.la_nlink == 0 || is_dir)) { obj->mod_flags |= DEAD_OBJ; /* add new orphan and the object * will be deleted during mdd_close() */ @@ -882,17 +1056,9 @@ int mdd_finish_unlink(const struct lu_env *env, PFID(mdd_object_fid(obj)), obj->mod_count); } else { - rc = mdd_object_kill(env, obj, ma, th); - if (rc == 0) - reset = 0; + rc = mdo_destroy(env, obj, th); } - - /* get the i_nlink */ - ma->ma_need = MA_INODE; - rc = mdd_attr_get_internal(env, obj, ma); } - if (reset) - ma->ma_valid &= ~(MA_LOV | MA_COOKIE); RETURN(rc); } @@ -902,12 +1068,12 @@ 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 md_attr *ma) + struct mdd_object *cobj, struct lu_attr *cattr) { int rc; ENTRY; - rc = mdd_may_delete(env, pobj, cobj, ma, 1, 1); + rc = mdd_may_delete(env, pobj, cobj, cattr, NULL, 1, 1); RETURN(rc); } @@ -961,21 +1127,14 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, struct md_attr *ma) { const char *name = lname->ln_name; + struct lu_attr *cattr = &mdd_env_info(env)->mti_cattr; struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; struct mdd_object *mdd_pobj = md2mdd_obj(pobj); struct mdd_object *mdd_cobj = md2mdd_obj(cobj); struct mdd_device *mdd = mdo2mdd(pobj); struct dynlock_handle *dlh; struct thandle *handle; -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct mds_obd *mds = &obd->u.mds; - unsigned int qcids[MAXQUOTAS] = { 0, 0 }; - unsigned int qpids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0; -#endif - int is_dir = S_ISDIR(ma->ma_attr.la_mode); - int rc; + int rc, is_dir; ENTRY; if (mdd_object_exists(mdd_cobj) <= 0) @@ -996,32 +1155,51 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, dlh = mdd_pdo_write_lock(env, mdd_pobj, name, MOR_TGT_PARENT); if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); + GOTO(stop, rc = -ENOMEM); mdd_write_lock(env, mdd_cobj, MOR_TGT_CHILD); - rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, ma); + /* fetch cattr */ + rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj)); if (rc) GOTO(cleanup, rc); - rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle, - mdd_object_capa(env, mdd_pobj)); + is_dir = S_ISDIR(cattr->la_mode); + + rc = mdd_unlink_sanity_check(env, mdd_pobj, mdd_cobj, cattr); if (rc) GOTO(cleanup, rc); - mdo_ref_del(env, mdd_cobj, handle); + rc = mdo_ref_del(env, mdd_cobj, handle); + if (rc != 0) { + __mdd_index_insert_only(env, mdd_pobj, mdo2fid(mdd_cobj), + name, handle, + mdd_object_capa(env, mdd_pobj)); + GOTO(cleanup, rc); + } + + rc = __mdd_index_delete(env, mdd_pobj, name, is_dir, handle, + mdd_object_capa(env, mdd_pobj)); + if (rc) + GOTO(cleanup, rc); + if (is_dir) /* unlink dot */ mdo_ref_del(env, mdd_cobj, handle); + /* fetch updated nlink */ + rc = mdd_la_get(env, mdd_cobj, cattr, mdd_object_capa(env, mdd_cobj)); + if (rc) + GOTO(cleanup, rc); + LASSERT(ma->ma_attr.la_valid & LA_CTIME); la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0); if (rc) GOTO(cleanup, rc); - if (ma->ma_attr.la_nlink > 0 || mdd_cobj->mod_count > 0) { + 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; @@ -1030,466 +1208,45 @@ static int mdd_unlink(const struct lu_env *env, struct md_object *pobj, 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); -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota && ma->ma_valid & MA_INODE && - ma->ma_attr.la_nlink == 0) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA); - if (!rc) { - mdd_quota_wrapper(la_tmp, qpids); - if (mdd_cobj->mod_count == 0) { - quota_opc = FSFILT_OP_UNLINK; - mdd_quota_wrapper(&ma->ma_attr, qcids); - } else { - quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT; - } - } - } -#endif - if (!is_dir) - /* old files may not have link ea; ignore errors */ - mdd_links_rename(env, mdd_cobj, mdo2fid(mdd_pobj), - lname, NULL, NULL, 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); + + /* if object is removed then we can't get its attrs, use last get */ + if (cattr->la_nlink == 0) { + ma->ma_attr = *cattr; + ma->ma_valid |= MA_INODE; + } EXIT; cleanup: mdd_write_unlock(env, mdd_cobj); mdd_pdo_write_unlock(env, mdd_pobj, dlh); -out_trans: if (rc == 0) { int cl_flags; - cl_flags = (ma->ma_attr.la_nlink == 0) ? CLF_UNLINK_LAST : 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; - rc = mdd_changelog_ns_store(env, mdd, - is_dir ? CL_RMDIR : CL_UNLINK, cl_flags, - mdd_cobj, mdd_pobj, NULL, lname, handle); + rc = mdd_changelog_ns_store(env, mdd, + is_dir ? CL_RMDIR : CL_UNLINK, cl_flags, + mdd_cobj, mdd_pobj, lname, handle); } stop: mdd_trans_stop(env, mdd, rc, handle); -#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,3,50,0) - if (rc == 0 && ma->ma_valid & MA_COOKIE && ma->ma_valid & MA_LOV && - ma->ma_valid & MA_FLAGS && ma->ma_attr_flags & MDS_UNLINK_DESTROY) - /* Since echo client is incapable of destorying ost object, - * it will destory the object here. */ - rc = mdd_lovobj_unlink(env, mdd, mdd_cobj, la, - ma->ma_lmm, ma->ma_lmm_size, - ma->ma_cookie, 1); -#else -#warning "please remove this after 2.4 (LOD/OSP)." -#endif - -#ifdef HAVE_QUOTA_SUPPORT - if (quota_opc) - /* Trigger dqrel on the owner of child and parent. If failed, - * the next call for lquota_chkquota will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc, - quota_opc); -#endif - return rc; -} - -/* has not lock on pobj yet */ -static int mdd_ni_sanity_check(const struct lu_env *env, - struct md_object *pobj, - const struct md_attr *ma) -{ - struct mdd_object *obj = md2mdd_obj(pobj); - int rc; - ENTRY; - - if (ma->ma_attr_flags & MDS_PERM_BYPASS) - RETURN(0); - - rc = mdd_may_create(env, obj, NULL, 1, S_ISDIR(ma->ma_attr.la_mode)); - - RETURN(rc); -} - -/* - * Partial operation. - */ -static int mdd_name_insert(const struct lu_env *env, - struct md_object *pobj, - const struct lu_name *lname, - const struct lu_fid *fid, - const struct md_attr *ma) -{ - const char *name = lname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_obj = md2mdd_obj(pobj); - struct dynlock_handle *dlh; - struct thandle *handle; - int is_dir = S_ISDIR(ma->ma_attr.la_mode); -#ifdef HAVE_QUOTA_SUPPORT - struct mdd_device *mdd = mdo2mdd(pobj); - struct md_ucred *uc = md_ucred(env); - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0, rec_pending[MAXQUOTAS] = { 0, 0 }; - cfs_cap_t save = uc->mu_cap; -#endif - int rc; - ENTRY; - - /* XXX: this code won't be used ever: - * DNE uses slightly different approach */ - LBUG(); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - if (!(ma->ma_attr_flags & MDS_QUOTA_IGNORE)) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA); - if (!rc) { - void *data = NULL; - mdd_data_get(env, mdd_obj, &data); - quota_opc = FSFILT_OP_LINK; - mdd_quota_wrapper(la_tmp, qids); - /* get block quota for parent */ - lquota_chkquota(mds_quota_interface_ref, obd, - exp, qids, rec_pending, 1, NULL, - LQUOTA_FLAGS_BLK, data, 1); - } - } else { - uc->mu_cap |= CFS_CAP_SYS_RESOURCE_MASK; - } - } -#endif - handle = mdd_trans_create(env, mdo2mdd(pobj)); - if (IS_ERR(handle)) - GOTO(out_pending, rc = PTR_ERR(handle)); - - rc = mdd_trans_start(env, mdo2mdd(pobj), handle); - - dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - - rc = mdd_ni_sanity_check(env, pobj, ma); - if (rc) - GOTO(out_unlock, rc); - - rc = __mdd_index_insert(env, mdd_obj, fid, name, is_dir, - handle, BYPASS_CAPA); - if (rc) - GOTO(out_unlock, rc); - - /* - * For some case, no need update obj's ctime (LA_CTIME is not set), - * e.g. split_dir. - * For other cases, update obj's ctime (LA_CTIME is set), - * e.g. cmr_link. - */ - if (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_locked(env, mdd_obj, la, - handle, 0); - } - EXIT; -out_unlock: - mdd_pdo_write_unlock(env, mdd_obj, dlh); -out_trans: - mdd_trans_stop(env, mdo2mdd(pobj), rc, handle); -out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - if (quota_opc) { - lquota_pending_commit(mds_quota_interface_ref, - obd, qids, rec_pending, 1); - /* Trigger dqacq for the parent owner. If failed, - * the next call for lquota_chkquota will process it*/ - lquota_adjust(mds_quota_interface_ref, obd, 0, qids, - rc, quota_opc); - } else { - uc->mu_cap = save; - } - } -#endif - return rc; -} - -/* has not lock on pobj yet */ -static int mdd_nr_sanity_check(const struct lu_env *env, - struct md_object *pobj, - const struct md_attr *ma) -{ - struct mdd_object *obj = md2mdd_obj(pobj); - int rc; - ENTRY; - - if (ma->ma_attr_flags & MDS_PERM_BYPASS) - RETURN(0); - - rc = mdd_may_unlink(env, obj, ma); - - RETURN(rc); -} - -/* - * Partial operation. - */ -static int mdd_name_remove(const struct lu_env *env, - struct md_object *pobj, - const struct lu_name *lname, - const struct md_attr *ma) -{ - const char *name = lname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_obj = md2mdd_obj(pobj); - struct mdd_device *mdd = mdo2mdd(pobj); - struct dynlock_handle *dlh; - struct thandle *handle; - int is_dir = S_ISDIR(ma->ma_attr.la_mode); -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct mds_obd *mds = &obd->u.mds; - unsigned int qids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0; -#endif - int rc; - ENTRY; - - /* XXX: this code won't be used ever: - * DNE uses slightly different approach */ - LBUG(); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_obj, la_tmp, BYPASS_CAPA); - if (!rc) { - quota_opc = FSFILT_OP_UNLINK_PARTIAL_PARENT; - mdd_quota_wrapper(la_tmp, qids); - } - } -#endif - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - GOTO(out_pending, rc = PTR_ERR(handle)); - - rc = mdd_trans_start(env, mdd, handle); - - dlh = mdd_pdo_write_lock(env, mdd_obj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - - rc = mdd_nr_sanity_check(env, pobj, ma); - if (rc) - GOTO(out_unlock, rc); - - rc = __mdd_index_delete(env, mdd_obj, name, is_dir, - handle, BYPASS_CAPA); - if (rc) - GOTO(out_unlock, rc); - - /* - * For some case, no need update obj's ctime (LA_CTIME is not set), - * e.g. split_dir. - * For other cases, update obj's ctime (LA_CTIME is set), - * e.g. cmr_unlink. - */ - if (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_locked(env, mdd_obj, la, - handle, 0); - } - EXIT; -out_unlock: - mdd_pdo_write_unlock(env, mdd_obj, dlh); -out_trans: - mdd_trans_stop(env, mdd, rc, handle); -out_pending: -#ifdef HAVE_QUOTA_SUPPORT - /* Trigger dqrel for the parent owner. - * If failed, the next call for lquota_chkquota will process it. */ - if (quota_opc) - lquota_adjust(mds_quota_interface_ref, obd, 0, qids, rc, - quota_opc); -#endif - return rc; -} - -/* - * tobj maybe NULL - * has mdd_write_lock on tobj alreay, but not on tgt_pobj yet - */ -static int mdd_rt_sanity_check(const struct lu_env *env, - struct mdd_object *tgt_pobj, - struct mdd_object *tobj, - struct md_attr *ma) -{ - int rc; - ENTRY; - - if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS)) - RETURN(0); - - /* XXX: for mdd_rename_tgt, "tobj == NULL" does not mean tobj not - * exist. In fact, tobj must exist, otherwise the call trace will be: - * mdt_reint_rename_tgt -> mdo_name_insert -> ... -> mdd_name_insert. - * When get here, tobj must be NOT NULL, the other case has been - * processed in cmr_rename_tgt before mdd_rename_tgt and enable - * MDS_PERM_BYPASS. - * So check may_delete, but not check nlink of tgt_pobj. */ - - rc = mdd_may_delete(env, tgt_pobj, tobj, ma, 1, 1); - - RETURN(rc); -} - -/* Partial rename op on slave MDD */ -static int mdd_rename_tgt(const struct lu_env *env, - struct md_object *pobj, struct md_object *tobj, - const struct lu_fid *lf, const struct lu_name *lname, - struct md_attr *ma) -{ - const char *name = lname->ln_name; - struct lu_attr *la = &mdd_env_info(env)->mti_la_for_fix; - struct mdd_object *mdd_tpobj = md2mdd_obj(pobj); - struct mdd_object *mdd_tobj = md2mdd_obj(tobj); - struct mdd_device *mdd = mdo2mdd(pobj); - struct dynlock_handle *dlh; - struct thandle *handle; -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qcids[MAXQUOTAS] = { 0, 0 }; - unsigned int qpids[MAXQUOTAS] = { 0, 0 }; - int quota_copc = 0, quota_popc = 0; - int rec_pending[MAXQUOTAS] = { 0, 0 }; -#endif - int rc; - ENTRY; - - /* XXX: this code won't be used ever: - * DNE uses slightly different approach */ - LBUG(); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota && !tobj) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_tpobj, la_tmp, BYPASS_CAPA); - if (!rc) { - void *data = NULL; - mdd_data_get(env, mdd_tpobj, &data); - quota_popc = FSFILT_OP_LINK; - mdd_quota_wrapper(la_tmp, qpids); - /* get block quota for target parent */ - lquota_chkquota(mds_quota_interface_ref, obd, exp, - qpids, rec_pending, 1, NULL, - LQUOTA_FLAGS_BLK, data, 1); - } - } -#endif - handle = mdd_trans_create(env, mdd); - if (IS_ERR(handle)) - GOTO(out_pending, rc = PTR_ERR(handle)); - - rc = mdd_trans_start(env, mdd, handle); - - dlh = mdd_pdo_write_lock(env, mdd_tpobj, name, MOR_TGT_PARENT); - if (dlh == NULL) - GOTO(out_trans, rc = -ENOMEM); - if (tobj) - mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD); - - rc = mdd_rt_sanity_check(env, mdd_tpobj, mdd_tobj, ma); - if (rc) - GOTO(cleanup, rc); - - /* - * If rename_tgt is called then we should just re-insert name with - * correct fid, no need to dec/inc parent nlink if obj is dir. - */ - rc = __mdd_index_delete(env, mdd_tpobj, name, 0, handle, BYPASS_CAPA); - if (rc) - GOTO(cleanup, rc); - - rc = __mdd_index_insert_only(env, mdd_tpobj, lf, name, handle, - BYPASS_CAPA); - if (rc) - GOTO(cleanup, rc); - - LASSERT(ma->ma_attr.la_valid & LA_CTIME); - la->la_ctime = la->la_mtime = ma->ma_attr.la_ctime; - - la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, handle, 0); - if (rc) - GOTO(cleanup, rc); - - /* - * For tobj is remote case cmm layer has processed - * and pass NULL tobj to here. So when tobj is NOT NULL, - * it must be local one. - */ - if (tobj && mdd_object_exists(mdd_tobj)) { - mdo_ref_del(env, mdd_tobj, handle); - - /* Remove dot reference. */ - if (S_ISDIR(ma->ma_attr.la_mode)) - mdo_ref_del(env, mdd_tobj, handle); - - la->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); - if (rc) - GOTO(cleanup, rc); - - rc = mdd_finish_unlink(env, mdd_tobj, ma, handle); - if (rc) - GOTO(cleanup, rc); -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota && ma->ma_valid & MA_INODE && - ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) { - quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD; - mdd_quota_wrapper(&ma->ma_attr, qcids); - } -#endif - } - EXIT; -cleanup: - if (tobj) - mdd_write_unlock(env, mdd_tobj); - mdd_pdo_write_unlock(env, mdd_tpobj, dlh); -out_trans: - if (rc == 0) - /* Bare EXT record with no RENAME in front of it signifies - a partial slave op */ - rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj, - mdd_tpobj, NULL, lname, handle); - - mdd_trans_stop(env, mdd, rc, handle); -out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - if (quota_popc) - lquota_pending_commit(mds_quota_interface_ref, obd, - qpids, rec_pending, 1); - - if (quota_copc) - /* Trigger dqrel on the target owner of child. - * If failed, the next call for lquota_chkquota - * will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, - rc, quota_copc); - } -#endif return rc; } @@ -1509,27 +1266,6 @@ static int mdd_cd_sanity_check(const struct lu_env *env, } -static int mdd_declare_create_data(const struct lu_env *env, - struct mdd_device *mdd, - struct mdd_object *obj, - int lmm_size, - struct thandle *handle) -{ - struct lu_buf *buf = &mdd_env_info(env)->mti_buf; - int rc; - - buf->lb_buf = NULL; - buf->lb_len = lmm_size; - rc = mdo_declare_xattr_set(env, obj, buf, XATTR_NAME_LOV, - 0, handle); - if (rc) - return rc; - - rc = mdd_declare_lov_objid_update(env, mdd, handle); - - return rc; -} - 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) @@ -1537,69 +1273,81 @@ static int mdd_create_data(const struct lu_env *env, struct md_object *pobj, struct mdd_device *mdd = mdo2mdd(cobj); struct mdd_object *mdd_pobj = md2mdd_obj(pobj); struct mdd_object *son = md2mdd_obj(cobj); - struct lov_mds_md *lmm = NULL; - int lmm_size = 0; 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); + rc = mdd_cd_sanity_check(env, son); if (rc) RETURN(rc); if (!md_should_create(spec->sp_cr_flags)) RETURN(0); - lmm_size = ma->ma_lmm_size; - rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, spec, ma); - if (rc) - RETURN(rc); + /* + * there are following use cases for this function: + * 1) late striping - file was created with MDS_OPEN_DELAY_CREATE + * striping can be specified or not + * 2) CMD? + */ + rc = mdd_la_get(env, son, attr, mdd_object_capa(env, son)); + if (rc) + RETURN(rc); + + /* calling ->ah_make_hint() is used to transfer information from parent */ + mdd_object_make_hint(env, mdd_pobj, son, attr); handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_free, rc = PTR_ERR(handle)); - rc = mdd_declare_create_data(env, mdd, son, lmm_size, handle); - if (rc) - GOTO(stop, rc); - - rc = mdd_trans_start(env, mdd, handle); - if (rc) - GOTO(stop, rc); - /* * XXX: Setting the lov ea is not locked but setting the attr is locked? * Should this be fixed? */ - - /* Replay creates has objects already */ -#if 0 - if (spec->no_create) { - CDEBUG(D_INFO, "we already have lov ea\n"); - rc = mdd_lov_set_md(env, mdd_pobj, son, - (struct lov_mds_md *)spec->u.sp_ea.eadata, - spec->u.sp_ea.eadatalen, handle, 0); - } else -#endif - /* No need mdd_lsm_sanity_check here */ - rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, - lmm_size, handle, 0); - - if (rc == 0) - rc = mdd_attr_get_internal_locked(env, son, ma); - - /* update lov_objid data, must be before transaction stop! */ - if (rc == 0) - mdd_lov_objid_update(mdd, lmm); - + CDEBUG(D_OTHER, "ea %p/%u, cr_flags %Lo, 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) { + /* replay case */ + buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, + spec->u.sp_ea.eadatalen); + } else if (!(spec->sp_cr_flags & MDS_OPEN_HAS_OBJS)) { + if (spec->sp_cr_flags & MDS_OPEN_HAS_EA) { + /* lfs setstripe */ + buf = mdd_buf_get_const(env, spec->u.sp_ea.eadata, + spec->u.sp_ea.eadatalen); + } else { + buf = &LU_BUF_NULL; + } + } else { + /* MDS_OPEN_HAS_OBJS is not used anymore ? */ + LBUG(); + } + + rc = dt_declare_xattr_set(env, mdd_object_child(son), buf, + XATTR_NAME_LOV, 0, 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)); stop: - mdd_trans_stop(env, mdd, rc, handle); + mdd_trans_stop(env, mdd, rc, handle); out_free: - /* Finish mdd_lov_create() stuff. */ - /* if no_create == 0 (not replay), we free lmm allocated by - * mdd_lov_create() */ - mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec); - RETURN(rc); + RETURN(rc); } /* Get fid from name and parent */ @@ -1653,19 +1401,20 @@ __mdd_lookup(const struct lu_env *env, struct md_object *pobj, } int mdd_declare_object_initialize(const struct lu_env *env, - struct mdd_object *child, - struct md_attr *ma, - struct thandle *handle) + struct mdd_object *child, + struct lu_attr *attr, + struct thandle *handle) { int rc; - rc = mdo_declare_attr_set(env, child, &ma->ma_attr, handle); - if (rc == 0 && S_ISDIR(ma->ma_attr.la_mode)) { - rc = mdo_declare_index_insert(env, child, mdo2fid(child), - dot, handle); + rc = mdo_declare_attr_set(env, child, attr, handle); + 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) mdd_declare_links_add(env, child, handle); @@ -1673,9 +1422,9 @@ int mdd_declare_object_initialize(const struct lu_env *env, } int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, - const struct lu_name *lname, struct mdd_object *child, - struct md_attr *ma, struct thandle *handle, - const struct md_op_spec *spec) + 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; @@ -1688,11 +1437,18 @@ int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, * (2) maybe, the child attributes should be set in OSD when creation. */ - rc = mdd_attr_set_internal(env, child, &ma->ma_attr, handle, 0); + /* + * 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. + */ + attr->la_valid &= ~LA_MODE; + rc = mdd_attr_set_internal(env, child, attr, handle, 0); if (rc != 0) RETURN(rc); - if (S_ISDIR(ma->ma_attr.la_mode)) { + 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), @@ -1713,12 +1469,12 @@ int mdd_object_initialize(const struct lu_env *env, const struct lu_fid *pfid, /* has not lock on pobj yet */ static int mdd_create_sanity_check(const struct lu_env *env, struct md_object *pobj, + struct lu_attr *pattr, const struct lu_name *lname, - struct md_attr *ma, + struct lu_attr *cattr, struct md_op_spec *spec) { struct mdd_thread_info *info = mdd_env_info(env); - struct lu_attr *la = &info->mti_la; struct lu_fid *fid = &info->mti_fid; struct mdd_object *obj = md2mdd_obj(pobj); struct mdd_device *m = mdo2mdd(pobj); @@ -1751,26 +1507,22 @@ static int mdd_create_sanity_check(const struct lu_env *env, * EXEC permission have been checked * when lookup before create already. */ - rc = mdd_permission_internal_locked(env, obj, NULL, MAY_WRITE, - MOR_TGT_PARENT); + rc = mdd_permission_internal_locked(env, obj, pattr, MAY_WRITE, + MOR_TGT_PARENT); if (rc) RETURN(rc); } /* sgid check */ - rc = mdd_la_get(env, obj, la, BYPASS_CAPA); - if (rc != 0) - RETURN(rc); - - if (la->la_mode & S_ISGID) { - ma->ma_attr.la_gid = la->la_gid; - if (S_ISDIR(ma->ma_attr.la_mode)) { - ma->ma_attr.la_mode |= S_ISGID; - ma->ma_attr.la_valid |= LA_MODE; - } - } - - switch (ma->ma_attr.la_mode & S_IFMT) { + if (pattr->la_mode & S_ISGID) { + cattr->la_gid = pattr->la_gid; + if (S_ISDIR(cattr->la_mode)) { + cattr->la_mode |= S_ISGID; + cattr->la_valid |= LA_MODE; + } + } + + switch (cattr->la_mode & S_IFMT) { case S_IFLNK: { unsigned int symlen = strlen(spec->u.sp_symname) + 1; @@ -1794,41 +1546,53 @@ static int mdd_create_sanity_check(const struct lu_env *env, 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 md_attr *ma, - int lmm_size, - struct thandle *handle, - const struct md_op_spec *spec) +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) { - struct lu_buf *buf = &mdd_env_info(env)->mti_buf; + struct mdd_thread_info *info = mdd_env_info(env); int rc = 0; - rc = mdd_declare_object_create_internal(env, p, c, ma, handle, spec); + rc = mdd_declare_object_create_internal(env, p, c, attr, handle, spec); if (rc) GOTO(out, rc); - /* if dir, then can inherit default ACl */ - buf->lb_buf = NULL; - buf->lb_len = lmm_size; - if (S_ISDIR(ma->ma_attr.la_mode)) { - rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_DEFAULT, - 0, handle); - if (rc == 0) - rc = mdo_declare_ref_add(env, p, handle); - } - if (rc) - GOTO(out, rc); +#ifdef CONFIG_FS_POSIX_ACL + if (got_def_acl > 0) { + struct lu_buf *acl_buf; + + acl_buf = mdd_buf_get(env, NULL, got_def_acl); + /* if dir, then can inherit default ACl */ + if (S_ISDIR(attr->la_mode)) { + rc = mdo_declare_xattr_set(env, c, acl_buf, + XATTR_NAME_ACL_DEFAULT, + 0, handle); + if (rc) + GOTO(out, rc); + } + + rc = mdo_declare_attr_set(env, c, &info->mti_pattr, 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 = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_ACL_ACCESS, - 0, handle); - if (rc) - GOTO(out, rc); + if (S_ISDIR(attr->la_mode)) { + rc = mdo_declare_ref_add(env, p, handle); + if (rc) + GOTO(out, rc); + } - rc = mdd_declare_object_initialize(env, c, ma, handle); + rc = mdd_declare_object_initialize(env, c, attr, handle); if (rc) GOTO(out, rc); @@ -1837,19 +1601,27 @@ static int mdd_declare_create(const struct lu_env *env, if (rc) GOTO(out, rc); - rc = mdo_declare_xattr_set(env, c, buf, XATTR_NAME_LOV, - 0, 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)) { + 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(ma->ma_attr.la_mode)) { + 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); } - rc = mdo_declare_attr_set(env, p, &ma->ma_attr, handle); + + rc = mdo_declare_attr_set(env, p, attr, handle); if (rc) return rc; @@ -1857,8 +1629,6 @@ static int mdd_declare_create(const struct lu_env *env, if (rc) return rc; - rc = mdd_declare_lov_objid_update(env, mdd, handle); - out: return rc; } @@ -1867,37 +1637,22 @@ out: /* * Create object and insert it into namespace. */ -static int mdd_create(const struct lu_env *env, - struct md_object *pobj, - const struct lu_name *lname, - struct md_object *child, - struct md_op_spec *spec, - struct md_attr* ma) +static int mdd_create(const struct lu_env *env, struct md_object *pobj, + const struct lu_name *lname, struct md_object *child, + struct md_op_spec *spec, struct md_attr* ma) { struct mdd_thread_info *info = mdd_env_info(env); struct lu_attr *la = &info->mti_la_for_fix; - struct md_attr *ma_acl = &info->mti_ma; struct mdd_object *mdd_pobj = md2mdd_obj(pobj); struct mdd_object *son = md2mdd_obj(child); struct mdd_device *mdd = mdo2mdd(pobj); struct lu_attr *attr = &ma->ma_attr; - struct lov_mds_md *lmm = NULL; struct thandle *handle; + struct lu_attr *pattr = &info->mti_pattr; struct dynlock_handle *dlh; const char *name = lname->ln_name; - int rc, created = 0, initialized = 0, inserted = 0, lmm_size = 0; + int rc, created = 0, initialized = 0, inserted = 0; int got_def_acl = 0; -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qcids[MAXQUOTAS] = { 0, 0 }; - unsigned int qpids[MAXQUOTAS] = { 0, 0 }; - int quota_opc = 0, block_count = 0; - int inode_pending[MAXQUOTAS] = { 0, 0 }; - int block_pending[MAXQUOTAS] = { 0, 0 }; - int parent_pending[MAXQUOTAS] = { 0, 0 }; -#endif ENTRY; /* @@ -1936,91 +1691,41 @@ static int mdd_create(const struct lu_env *env, * 2. insert (__mdd_index_insert(), lookup again) */ + rc = mdd_la_get(env, mdd_pobj, pattr, BYPASS_CAPA); + if (rc != 0) + RETURN(rc); + /* Sanity checks before big job. */ - rc = mdd_create_sanity_check(env, pobj, lname, ma, spec); + rc = mdd_create_sanity_check(env, pobj, pattr, lname, attr, spec); if (rc) RETURN(rc); -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_pobj, la_tmp, BYPASS_CAPA); - if (!rc) { - int same = 0; - - quota_opc = FSFILT_OP_CREATE; - mdd_quota_wrapper(&ma->ma_attr, qcids); - mdd_quota_wrapper(la_tmp, qpids); - /* get file quota for child */ - lquota_chkquota(mds_quota_interface_ref, obd, exp, - qcids, inode_pending, 1, NULL, 0, NULL, - 0); - switch (ma->ma_attr.la_mode & S_IFMT) { - case S_IFLNK: - case S_IFDIR: - block_count = 2; - break; - case S_IFREG: - block_count = 1; - break; - } - if (qcids[USRQUOTA] == qpids[USRQUOTA] && - qcids[GRPQUOTA] == qpids[GRPQUOTA]) { - block_count += 1; - same = 1; - } - /* get block quota for child and parent */ - if (block_count) - lquota_chkquota(mds_quota_interface_ref, obd, - exp, qcids, block_pending, - block_count, NULL, - LQUOTA_FLAGS_BLK, NULL, 0); - if (!same) - lquota_chkquota(mds_quota_interface_ref, obd, - exp, qpids, parent_pending, 1, - NULL, LQUOTA_FLAGS_BLK, NULL, - 0); - } - } -#endif - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_DQACQ_NET)) - GOTO(out_pending, rc = -EINPROGRESS); - - /* - * No RPC inside the transaction, so OST objects should be created at - * first. - */ - if (S_ISREG(attr->la_mode)) { - lmm_size = ma->ma_lmm_size; - rc = mdd_lov_create(env, mdd, mdd_pobj, son, &lmm, &lmm_size, - spec, ma); - if (rc) - GOTO(out_pending, rc); - } + GOTO(out_free, rc = -EINPROGRESS); if (!S_ISLNK(attr->la_mode)) { - ma_acl->ma_acl_size = sizeof info->mti_xattr_buf; - ma_acl->ma_acl = info->mti_xattr_buf; - ma_acl->ma_need = MA_ACL_DEF; - ma_acl->ma_valid = 0; + struct lu_buf *acl_buf; + acl_buf = mdd_buf_get(env, info->mti_xattr_buf, + sizeof(info->mti_xattr_buf)); mdd_read_lock(env, mdd_pobj, MOR_TGT_PARENT); - rc = mdd_def_acl_get(env, mdd_pobj, ma_acl); + rc = mdo_xattr_get(env, mdd_pobj, acl_buf, + XATTR_NAME_ACL_DEFAULT, BYPASS_CAPA); mdd_read_unlock(env, mdd_pobj); - if (rc) - GOTO(out_free, rc); - else if (ma_acl->ma_valid & MA_ACL_DEF) - got_def_acl = 1; + 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); + handle = mdd_trans_create(env, mdd); if (IS_ERR(handle)) GOTO(out_free, rc = PTR_ERR(handle)); - rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, ma, - lmm_size, handle, spec); + rc = mdd_declare_create(env, mdd, mdd_pobj, son, lname, attr, + got_def_acl, handle, spec); if (rc) GOTO(out_stop, rc); @@ -2033,7 +1738,7 @@ static int mdd_create(const struct lu_env *env, GOTO(out_trans, rc = -ENOMEM); mdd_write_lock(env, son, MOR_TGT_CHILD); - rc = mdd_object_create_internal(env, mdd_pobj, son, ma, handle, spec); + rc = mdd_object_create_internal(env, mdd_pobj, son, attr, handle, spec); if (rc) { mdd_write_unlock(env, son); GOTO(cleanup, rc); @@ -2043,22 +1748,35 @@ static int mdd_create(const struct lu_env *env, #ifdef CONFIG_FS_POSIX_ACL if (got_def_acl) { - struct lu_buf *acl_buf = &info->mti_buf; - acl_buf->lb_buf = ma_acl->ma_acl; - acl_buf->lb_len = ma_acl->ma_acl_size; + struct lu_buf *acl_buf; - rc = __mdd_acl_init(env, son, acl_buf, &attr->la_mode, handle); - if (rc) { - mdd_write_unlock(env, son); - GOTO(cleanup, rc); - } else { - ma->ma_attr.la_valid |= LA_MODE; - } + 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, ma, handle, spec); + 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); + } mdd_write_unlock(env, son); if (rc) /* @@ -2072,35 +1790,13 @@ static int mdd_create(const struct lu_env *env, rc = __mdd_index_insert(env, mdd_pobj, mdo2fid(son), name, S_ISDIR(attr->la_mode), handle, mdd_object_capa(env, mdd_pobj)); - if (rc) GOTO(cleanup, rc); inserted = 1; - /* No need mdd_lsm_sanity_check here */ - rc = mdd_lov_set_md(env, mdd_pobj, son, lmm, lmm_size, handle, 0); - if (rc) { - CERROR("error on stripe info copy %d \n", rc); - GOTO(cleanup, rc); - } - if (lmm && lmm_size > 0) { - /* Set Lov here, do not get lmm again later */ - if (lmm_size > ma->ma_lmm_size) { - /* Reply buffer is smaller, need bigger one */ - mdd_max_lmm_buffer(env, lmm_size); - if (unlikely(info->mti_max_lmm == NULL)) - GOTO(cleanup, rc = -ENOMEM); - ma->ma_lmm = info->mti_max_lmm; - ma->ma_big_lmm_used = 1; - } - memcpy(ma->ma_lmm, lmm, lmm_size); - ma->ma_lmm_size = lmm_size; - ma->ma_valid |= MA_LOV; - } - if (S_ISLNK(attr->la_mode)) { - struct md_ucred *uc = md_ucred(env); + 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); @@ -2108,10 +1804,10 @@ static int mdd_create(const struct lu_env *env, 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->mu_cap & - CFS_CAP_SYS_RESOURCE_MASK); + 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; @@ -2119,72 +1815,60 @@ static int mdd_create(const struct lu_env *env, GOTO(cleanup, rc = -EFAULT); } - *la = ma->ma_attr; + *la = *attr; la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_pobj, la, handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_pobj, la, handle, 0); if (rc) GOTO(cleanup, rc); - /* Return attr back. */ - rc = mdd_attr_get_internal_locked(env, son, ma); EXIT; cleanup: - if (rc && created) { - int rc2 = 0; - - if (inserted) { - rc2 = __mdd_index_delete(env, mdd_pobj, name, - S_ISDIR(attr->la_mode), - handle, BYPASS_CAPA); - if (rc2) - CERROR("error can not cleanup destroy %d\n", - rc2); - } - - if (rc2 == 0) { - mdd_write_lock(env, son, MOR_TGT_CHILD); - mdo_ref_del(env, son, handle); - if (initialized && S_ISDIR(attr->la_mode)) - mdo_ref_del(env, son, handle); - mdd_write_unlock(env, son); - } + if (rc != 0 && created != 0) { + int rc2; + + if (inserted != 0) { + rc2 = __mdd_index_delete(env, mdd_pobj, name, + S_ISDIR(attr->la_mode), + handle, BYPASS_CAPA); + if (rc2 != 0) + goto out_stop; + } + + mdd_write_lock(env, son, MOR_TGT_CHILD); + if (initialized != 0 && 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) { + mdd_write_unlock(env, son); + goto out_stop; + } + } + + rc2 = mdo_ref_del(env, son, handle); + if (rc2 != 0) { + mdd_write_unlock(env, son); + goto out_stop; + } + + mdo_destroy(env, son, handle); + mdd_write_unlock(env, son); } - /* update lov_objid data, must be before transaction stop! */ - if (rc == 0) - mdd_lov_objid_update(mdd, lmm); - mdd_pdo_write_unlock(env, mdd_pobj, dlh); out_trans: if (rc == 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, NULL, lname, handle); + 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); out_stop: mdd_trans_stop(env, mdd, rc, handle); out_free: - /* finish lov_create stuff, free all temporary data */ - mdd_lov_create_finish(env, mdd, lmm, lmm_size, spec); -out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (quota_opc) { - lquota_pending_commit(mds_quota_interface_ref, obd, qcids, - inode_pending, 0); - lquota_pending_commit(mds_quota_interface_ref, obd, qcids, - block_pending, 1); - lquota_pending_commit(mds_quota_interface_ref, obd, qpids, - parent_pending, 1); - /* Trigger dqacq on the owner of child and parent. If failed, - * the next call for lquota_chkquota will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, qcids, qpids, rc, - quota_opc); - } -#endif /* The child object shouldn't be cached anymore */ if (rc) - cfs_set_bit(LU_OBJECT_HEARD_BANSHEE, + set_bit(LU_OBJECT_HEARD_BANSHEE, &child->mo_lu.lo_header->loh_flags); return rc; } @@ -2236,20 +1920,18 @@ static int mdd_rename_sanity_check(const struct lu_env *env, struct mdd_object *tgt_pobj, struct mdd_object *sobj, struct mdd_object *tobj, - struct md_attr *ma) + struct lu_attr *so_attr, + struct lu_attr *tg_attr) { int rc = 0; ENTRY; - if (unlikely(ma->ma_attr_flags & MDS_PERM_BYPASS)) - RETURN(0); - /* 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); - rc = mdd_may_delete(env, src_pobj, sobj, ma, 1, 0); + rc = mdd_may_delete(env, src_pobj, sobj, so_attr, NULL, 1, 0); if (rc) RETURN(rc); @@ -2262,11 +1944,11 @@ static int mdd_rename_sanity_check(const struct lu_env *env, rc = mdd_may_create(env, tgt_pobj, NULL, (src_pobj != tgt_pobj), 0); else - rc = mdd_may_delete(env, tgt_pobj, tobj, ma, + 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(ma->ma_attr.la_mode)) + S_ISDIR(so_attr->la_mode)) rc = __mdd_may_link(env, tgt_pobj); RETURN(rc); @@ -2278,8 +1960,8 @@ static int mdd_declare_rename(const struct lu_env *env, struct mdd_object *mdd_tpobj, struct mdd_object *mdd_sobj, struct mdd_object *mdd_tobj, - const struct lu_name *sname, const struct lu_name *tname, + const struct lu_name *sname, struct md_attr *ma, struct thandle *handle) { @@ -2379,11 +2061,7 @@ static int mdd_declare_rename(const struct lu_env *env, return rc; } - rc = mdd_declare_changelog_store(env, mdd, tname, handle); - if (rc) - return rc; - - rc = mdd_declare_changelog_store(env, mdd, sname, handle); + rc = mdd_declare_changelog_ext_store(env, mdd, tname, sname, handle); if (rc) return rc; @@ -2400,6 +2078,8 @@ static int mdd_rename(const struct lu_env *env, 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); @@ -2409,53 +2089,16 @@ static int mdd_rename(const struct lu_env *env, struct thandle *handle; const struct lu_fid *tpobj_fid = mdo2fid(mdd_tpobj); const struct lu_fid *spobj_fid = mdo2fid(mdd_spobj); - int is_dir; + bool is_dir; + bool tobj_ref = 0; + bool tobj_locked = 0; + unsigned cl_flags = 0; int rc, rc2; - -#ifdef HAVE_QUOTA_SUPPORT - struct obd_device *obd = mdd->mdd_obd_dev; - struct obd_export *exp = md_quota(env)->mq_exp; - struct mds_obd *mds = &obd->u.mds; - unsigned int qspids[MAXQUOTAS] = { 0, 0 }; - unsigned int qtcids[MAXQUOTAS] = { 0, 0 }; - unsigned int qtpids[MAXQUOTAS] = { 0, 0 }; - int quota_copc = 0, quota_popc = 0; - int rec_pending[MAXQUOTAS] = { 0, 0 }; -#endif ENTRY; - LASSERT(ma->ma_attr.la_mode & S_IFMT); - is_dir = S_ISDIR(ma->ma_attr.la_mode); - if (tobj) mdd_tobj = md2mdd_obj(tobj); -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - struct lu_attr *la_tmp = &mdd_env_info(env)->mti_la; - - rc = mdd_la_get(env, mdd_spobj, la_tmp, BYPASS_CAPA); - if (!rc) { - mdd_quota_wrapper(la_tmp, qspids); - if (!tobj) { - rc = mdd_la_get(env, mdd_tpobj, la_tmp, - BYPASS_CAPA); - if (!rc) { - void *data = NULL; - mdd_data_get(env, mdd_tpobj, &data); - quota_popc = FSFILT_OP_LINK; - mdd_quota_wrapper(la_tmp, qtpids); - /* get block quota for target parent */ - lquota_chkquota(mds_quota_interface_ref, - obd, exp, qtpids, - rec_pending, 1, NULL, - LQUOTA_FLAGS_BLK, - data, 1); - } - } - } - } -#endif mdd_sobj = mdd_object_find(env, mdd, lf); handle = mdd_trans_create(env, mdd); @@ -2496,11 +2139,25 @@ static int mdd_rename(const struct lu_env *env, if (sdlh == NULL || tdlh == NULL) GOTO(cleanup, rc = -ENOMEM); - rc = mdd_rename_sanity_check(env, mdd_spobj, mdd_tpobj, - mdd_sobj, mdd_tobj, ma); + 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); + is_dir = S_ISDIR(so_attr->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)); @@ -2547,8 +2204,7 @@ static int mdd_rename(const struct lu_env *env, /* 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_locked(env, mdd_sobj, la, - handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_sobj, la, handle, 0); if (rc) GOTO(fixup_tpobj, rc); } @@ -2560,8 +2216,8 @@ static int mdd_rename(const struct lu_env *env, */ if (tobj && mdd_object_exists(mdd_tobj)) { mdd_write_lock(env, mdd_tobj, MOR_TGT_CHILD); + tobj_locked = 1; if (mdd_is_dead_obj(mdd_tobj)) { - mdd_write_unlock(env, mdd_tobj); /* shld not be dead, something is wrong */ CERROR("tobj is dead, something is wrong\n"); rc = -EINVAL; @@ -2570,43 +2226,76 @@ static int mdd_rename(const struct lu_env *env, mdo_ref_del(env, mdd_tobj, handle); /* Remove dot reference. */ - if (is_dir) + if (S_ISDIR(tg_attr->la_mode)) mdo_ref_del(env, mdd_tobj, handle); - - la->la_valid = LA_CTIME; - rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); - if (rc) - GOTO(fixup_tpobj, rc); - - rc = mdd_finish_unlink(env, mdd_tobj, ma, handle); - mdd_write_unlock(env, mdd_tobj); - if (rc) - GOTO(fixup_tpobj, rc); - -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota && ma->ma_valid & MA_INODE && - ma->ma_attr.la_nlink == 0 && mdd_tobj->mod_count == 0) { - quota_copc = FSFILT_OP_UNLINK_PARTIAL_CHILD; - mdd_quota_wrapper(&ma->ma_attr, qtcids); - } -#endif + tobj_ref = 1; + + /* fetch updated nlink */ + rc = mdd_la_get(env, mdd_tobj, tg_attr, + mdd_object_capa(env, mdd_tobj)); + if (rc != 0) { + CERROR("%s: Failed to get nlink for tobj " + DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(tpobj_fid), rc); + GOTO(fixup_tpobj, rc); + } + + la->la_valid = LA_CTIME; + rc = mdd_attr_check_set_internal(env, mdd_tobj, la, handle, 0); + if (rc != 0) { + CERROR("%s: Failed to set ctime for tobj " + DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(tpobj_fid), rc); + GOTO(fixup_tpobj, rc); + } + + /* XXX: this transfer to ma will be removed with LOD/OSP */ + ma->ma_attr = *tg_attr; + ma->ma_valid |= MA_INODE; + rc = mdd_finish_unlink(env, mdd_tobj, ma, handle); + if (rc != 0) { + CERROR("%s: Failed to unlink tobj " + DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(tpobj_fid), rc); + GOTO(fixup_tpobj, rc); + } + + /* fetch updated nlink */ + rc = mdd_la_get(env, mdd_tobj, tg_attr, + mdd_object_capa(env, mdd_tobj)); + if (rc != 0) { + CERROR("%s: Failed to get nlink for tobj " + DFID": rc = %d\n", + mdd2obd_dev(mdd)->obd_name, + PFID(tpobj_fid), rc); + GOTO(fixup_tpobj, rc); + } + /* XXX: this transfer to ma will be removed with LOD/OSP */ + ma->ma_attr = *tg_attr; + ma->ma_valid |= MA_INODE; + + if (so_attr->la_nlink == 0) + cl_flags |= CLF_RENAME_LAST; } la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_spobj, la, handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_spobj, la, handle, 0); if (rc) GOTO(fixup_tpobj, rc); if (mdd_spobj != mdd_tpobj) { la->la_valid = LA_CTIME | LA_MTIME; - rc = mdd_attr_check_set_internal_locked(env, mdd_tpobj, la, - handle, 0); + rc = mdd_attr_check_set_internal(env, mdd_tpobj, la, + handle, 0); } 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); + rc = mdd_links_rename(env, mdd_sobj, mdo2fid(mdd_spobj), lsname, + mdo2fid(mdd_tpobj), ltname, handle, 0, 0); if (rc == -ENOENT) /* Old files might not have EA entry */ mdd_links_add(env, mdd_sobj, mdo2fid(mdd_spobj), @@ -2628,6 +2317,12 @@ fixup_tpobj: if (mdd_tobj && mdd_object_exists(mdd_tobj) && !mdd_is_dead_obj(mdd_tobj)) { + if (tobj_ref) { + mdo_ref_add(env, mdd_tobj, handle); + if (is_dir) + mdo_ref_add(env, mdd_tobj, handle); + } + rc2 = __mdd_index_insert(env, mdd_tpobj, mdo2fid(mdd_tobj), tname, is_dir, handle, @@ -2661,58 +2356,113 @@ fixup_spobj2: CWARN("sp obj fix error %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_ns_store(env, mdd, CL_RENAME, 0, mdd_tobj, - mdd_spobj, lf, lsname, handle); - if (rc == 0) { - struct lu_fid zero_fid; - fid_zero(&zero_fid); - /* If the rename target exist, The CL_EXT record should save - * the target fid as tfid, otherwise, use zero fid. LU-543 */ - rc = mdd_changelog_ns_store(env, mdd, CL_EXT, 0, mdd_tobj, - mdd_tpobj, - mdd_tobj ? NULL : &zero_fid, - ltname, handle); - } + rc = mdd_changelog_ext_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); out_pending: -#ifdef HAVE_QUOTA_SUPPORT - if (mds->mds_quota) { - if (quota_popc) - lquota_pending_commit(mds_quota_interface_ref, obd, - qtpids, rec_pending, 1); - - if (quota_copc) { - /* Trigger dqrel on the source owner of parent. - * If failed, the next call for lquota_chkquota will - * process it. */ - lquota_adjust(mds_quota_interface_ref, obd, 0, qspids, rc, - FSFILT_OP_UNLINK_PARTIAL_PARENT); - - /* Trigger dqrel on the target owner of child. - * If failed, the next call for lquota_chkquota - * will process it. */ - lquota_adjust(mds_quota_interface_ref, obd, qtcids, - qtpids, rc, quota_copc); - } - } -#endif return rc; } -/** enable/disable storing of hardlink info */ -int mdd_linkea_enable = 1; -CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644, - "record hardlink info in EAs"); +/** + * The data that link search is done on. + */ +struct mdd_link_data { + /** + * Buffer to keep link EA body. + */ + struct lu_buf *ml_buf; + /** + * The matched header, entry and its lenght in the EA + */ + struct link_ea_header *ml_leh; + struct link_ea_entry *ml_lee; + int ml_reclen; +}; + +static int mdd_links_new(const struct lu_env *env, + struct mdd_link_data *ldata) +{ + ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE); + if (ldata->ml_buf->lb_buf == NULL) + return -ENOMEM; + ldata->ml_leh = ldata->ml_buf->lb_buf; + ldata->ml_leh->leh_magic = LINK_EA_MAGIC; + ldata->ml_leh->leh_len = sizeof(struct link_ea_header); + ldata->ml_leh->leh_reccount = 0; + return 0; +} + +/** Read the link EA into a temp buffer. + * Uses the mdd_thread_info::mti_big_buf since it is generally large. + * A pointer to the buffer is stored in \a ldata::ml_buf. + * + * \retval 0 or error + */ +int mdd_links_read(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct mdd_link_data *ldata) +{ + struct lustre_capa *capa; + struct link_ea_header *leh; + int rc; + + /* First try a small buf */ + LASSERT(env != NULL); + ldata->ml_buf = mdd_buf_alloc(env, CFS_PAGE_SIZE); + if (ldata->ml_buf->lb_buf == NULL) + return -ENOMEM; + + if (!mdd_object_exists(mdd_obj)) + return -ENODATA; + + capa = mdd_object_capa(env, mdd_obj); + rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf, + XATTR_NAME_LINK, capa); + if (rc == -ERANGE) { + /* Buf was too small, figure out what we need. */ + mdd_buf_put(ldata->ml_buf); + rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf, + XATTR_NAME_LINK, capa); + if (rc < 0) + return rc; + ldata->ml_buf = mdd_buf_alloc(env, rc); + if (ldata->ml_buf->lb_buf == NULL) + return -ENOMEM; + rc = mdo_xattr_get(env, mdd_obj, ldata->ml_buf, + XATTR_NAME_LINK, capa); + } + if (rc < 0) + return rc; + + leh = ldata->ml_buf->lb_buf; + if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) { + leh->leh_magic = LINK_EA_MAGIC; + leh->leh_reccount = __swab32(leh->leh_reccount); + leh->leh_len = __swab64(leh->leh_len); + /* entries are swabbed by mdd_lee_unpack */ + } + if (leh->leh_magic != LINK_EA_MAGIC) + return -EINVAL; + if (leh->leh_reccount == 0) + return -ENODATA; + + ldata->ml_leh = leh; + return 0; +} /** Read the link EA into a temp buffer. * Uses the name_buf since it is generally large. @@ -2720,47 +2470,24 @@ CFS_MODULE_PARM(mdd_linkea_enable, "d", int, 0644, * \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_object *mdd_obj) { - struct lu_buf *buf; - struct lustre_capa *capa; - struct link_ea_header *leh; - int rc; + struct mdd_link_data ldata = { 0 }; + int rc; - /* First try a small buf */ - buf = mdd_buf_alloc(env, CFS_PAGE_SIZE); - if (buf->lb_buf == NULL) - return ERR_PTR(-ENOMEM); - - capa = mdd_object_capa(env, mdd_obj); - rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa); - if (rc == -ERANGE) { - /* Buf was too small, figure out what we need. */ - mdd_buf_put(buf); - rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa); - if (rc < 0) - return ERR_PTR(rc); - buf = mdd_buf_alloc(env, rc); - if (buf->lb_buf == NULL) - return ERR_PTR(-ENOMEM); - rc = mdo_xattr_get(env, mdd_obj, buf, XATTR_NAME_LINK, capa); - } - if (rc < 0) - return ERR_PTR(rc); - - leh = buf->lb_buf; - if (leh->leh_magic == __swab32(LINK_EA_MAGIC)) { - leh->leh_magic = LINK_EA_MAGIC; - leh->leh_reccount = __swab32(leh->leh_reccount); - leh->leh_len = __swab64(leh->leh_len); - /* entries are swabbed by mdd_lee_unpack */ - } - if (leh->leh_magic != LINK_EA_MAGIC) - return ERR_PTR(-EINVAL); - if (leh->leh_reccount == 0) - return ERR_PTR(-ENODATA); + rc = mdd_links_read(env, mdd_obj, &ldata); + return rc ? ERR_PTR(rc) : ldata.ml_buf; +} - return buf; +static int mdd_links_write(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct mdd_link_data *ldata, + struct thandle *handle) +{ + const struct lu_buf *buf = mdd_buf_get_const(env, ldata->ml_buf->lb_buf, + ldata->ml_leh->leh_len); + return mdo_xattr_set(env, mdd_obj, buf, XATTR_NAME_LINK, 0, handle, + mdd_object_capa(env, mdd_obj)); } /** Pack a link_ea_entry. @@ -2794,34 +2521,6 @@ void mdd_lee_unpack(const struct link_ea_entry *lee, int *reclen, lname->ln_namelen = *reclen - sizeof(struct link_ea_entry); } -/** Add a record to the end of link ea buf */ -static int __mdd_links_add(const struct lu_env *env, struct lu_buf *buf, - const struct lu_fid *pfid, - const struct lu_name *lname) -{ - struct link_ea_header *leh; - struct link_ea_entry *lee; - int reclen; - - if (lname == NULL || pfid == NULL) - return -EINVAL; - - /* Make sure our buf is big enough for the new one */ - leh = buf->lb_buf; - reclen = lname->ln_namelen + sizeof(struct link_ea_entry); - if (leh->leh_len + reclen > buf->lb_len) { - if (mdd_buf_grow(env, leh->leh_len + reclen) < 0) - return -ENOMEM; - } - - leh = buf->lb_buf; - lee = buf->lb_buf + leh->leh_len; - reclen = mdd_lee_pack(lee, lname, pfid); - leh->leh_len += reclen; - leh->leh_reccount++; - return 0; -} - static int mdd_declare_links_add(const struct lu_env *env, struct mdd_object *mdd_obj, struct thandle *handle) @@ -2842,151 +2541,257 @@ static int mdd_declare_links_add(const struct lu_env *env, */ #define LINKEA_MAX_COUNT 128 -static 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) +/** 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) { - struct lu_buf *buf; - struct link_ea_header *leh; - int rc; - ENTRY; - - if (!mdd_linkea_enable) - RETURN(0); - - buf = first ? ERR_PTR(-ENODATA) : mdd_links_get(env, mdd_obj); - if (IS_ERR(buf)) { - rc = PTR_ERR(buf); - if (rc != -ENODATA) { - CERROR("link_ea read failed %d "DFID"\n", rc, - PFID(mdd_object_fid(mdd_obj))); - RETURN (rc); - } - /* empty EA; start one */ - buf = mdd_buf_alloc(env, CFS_PAGE_SIZE); - if (buf->lb_buf == NULL) - RETURN(-ENOMEM); - leh = buf->lb_buf; - leh->leh_magic = LINK_EA_MAGIC; - leh->leh_len = sizeof(struct link_ea_header); - leh->leh_reccount = 0; - } - - leh = buf->lb_buf; - if (leh->leh_reccount > LINKEA_MAX_COUNT) - RETURN(-EOVERFLOW); - - rc = __mdd_links_add(env, buf, pfid, lname); - if (rc) - RETURN(rc); + LASSERT(ldata->ml_leh != NULL); + + if (lname == NULL || pfid == NULL) + return -EINVAL; + + /* Make sure our buf is big enough for the new one */ + if (ldata->ml_leh->leh_reccount > LINKEA_MAX_COUNT) + return -EOVERFLOW; + + ldata->ml_reclen = lname->ln_namelen + sizeof(struct link_ea_entry); + if (ldata->ml_leh->leh_len + ldata->ml_reclen > + ldata->ml_buf->lb_len) { + if (mdd_buf_grow(env, ldata->ml_leh->leh_len + + ldata->ml_reclen) < 0) + return -ENOMEM; + } + + ldata->ml_leh = ldata->ml_buf->lb_buf; + ldata->ml_lee = ldata->ml_buf->lb_buf + ldata->ml_leh->leh_len; + ldata->ml_reclen = mdd_lee_pack(ldata->ml_lee, lname, pfid); + ldata->ml_leh->leh_len += ldata->ml_reclen; + ldata->ml_leh->leh_reccount++; + CDEBUG(D_INODE, "New link_ea name '%.*s' is added\n", + lname->ln_namelen, lname->ln_name); + return 0; +} - leh = buf->lb_buf; - rc = __mdd_xattr_set(env, mdd_obj, - mdd_buf_get_const(env, buf->lb_buf, leh->leh_len), - XATTR_NAME_LINK, 0, handle); - if (rc) { - if (rc == -ENOSPC) - CDEBUG(D_INODE, "link_ea add failed %d "DFID"\n", rc, - PFID(mdd_object_fid(mdd_obj))); - else - CERROR("link_ea add failed %d "DFID"\n", rc, - PFID(mdd_object_fid(mdd_obj))); - } +/** 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); - if (buf->lb_len > OBD_ALLOC_BIG) - /* if we vmalloced a large buffer drop it */ - mdd_buf_put(buf); + 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); - 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) +/** + * 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_buf *buf; - struct link_ea_header *leh; - struct link_ea_entry *lee; - struct lu_name *tmpname = &mdd_env_info(env)->mti_name; - struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid; - int reclen = 0; - int count; - int rc, rc2 = 0; - ENTRY; + struct lu_name *tmpname = &mdd_env_info(env)->mti_name2; + struct lu_fid *tmpfid = &mdd_env_info(env)->mti_fid; + int count; + + LASSERT(ldata->ml_leh != NULL); + + /* link #0 */ + ldata->ml_lee = (struct link_ea_entry *)(ldata->ml_leh + 1); + + for (count = 0; count < ldata->ml_leh->leh_reccount; count++) { + mdd_lee_unpack(ldata->ml_lee, &ldata->ml_reclen, + tmpname, tmpfid); + if (tmpname->ln_namelen == lname->ln_namelen && + lu_fid_eq(tmpfid, pfid) && + (strncmp(tmpname->ln_name, lname->ln_name, + tmpname->ln_namelen) == 0)) + break; + ldata->ml_lee = (struct link_ea_entry *)((char *)ldata->ml_lee + + ldata->ml_reclen); + } + + if (count == ldata->ml_leh->leh_reccount) { + CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n", + lname->ln_namelen, lname->ln_name); + return -ENOENT; + } + return 0; +} - if (!mdd_linkea_enable) - RETURN(0); +static int __mdd_links_add(const struct lu_env *env, + struct mdd_object *mdd_obj, + struct mdd_link_data *ldata, + const struct lu_name *lname, + const struct lu_fid *pfid, + int first, int check) +{ + int rc; + + if (ldata->ml_leh == NULL) { + rc = first ? -ENODATA : mdd_links_read(env, mdd_obj, ldata); + if (rc) { + if (rc != -ENODATA) + return rc; + rc = mdd_links_new(env, ldata); + if (rc) + return rc; + } + } + + if (check) { + rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid); + if (rc && rc != -ENOENT) + return rc; + if (rc == 0) + return -EEXIST; + } + + return mdd_links_add_buf(env, ldata, lname, pfid); +} - if (mdd_obj->mod_flags & DEAD_OBJ) - /* No more links, don't bother */ - RETURN(0); +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; - buf = mdd_links_get(env, mdd_obj); - if (IS_ERR(buf)) { - rc = PTR_ERR(buf); - if (rc == -ENODATA) - CDEBUG(D_INODE, "link_ea read failed %d "DFID"\n", - rc, PFID(mdd_object_fid(mdd_obj))); - else - CERROR("link_ea read failed %d "DFID"\n", - rc, PFID(mdd_object_fid(mdd_obj))); - RETURN(rc); - } - leh = buf->lb_buf; - lee = (struct link_ea_entry *)(leh + 1); /* link #0 */ - - /* Find the old record */ - for(count = 0; count < leh->leh_reccount; count++) { - mdd_lee_unpack(lee, &reclen, tmpname, tmpfid); - if (tmpname->ln_namelen == oldlname->ln_namelen && - lu_fid_eq(tmpfid, oldpfid) && - (strncmp(tmpname->ln_name, oldlname->ln_name, - tmpname->ln_namelen) == 0)) - break; - lee = (struct link_ea_entry *)((char *)lee + reclen); - } - if ((count + 1) > leh->leh_reccount) { - CDEBUG(D_INODE, "Old link_ea name '%.*s' not found\n", - oldlname->ln_namelen, oldlname->ln_name); - GOTO(out, rc = -ENOENT); - } + if (ldata->ml_leh == NULL) { + rc = mdd_links_read(env, mdd_obj, ldata); + if (rc) + return rc; + } - /* Remove the old record */ - leh->leh_reccount--; - leh->leh_len -= reclen; - memmove(lee, (char *)lee + reclen, (char *)leh + leh->leh_len - - (char *)lee); - - /* If renaming, add the new record */ - if (newpfid != NULL) { - /* if the add fails, we still delete the out-of-date old link */ - rc2 = __mdd_links_add(env, buf, newpfid, newlname); - leh = buf->lb_buf; - } + rc = mdd_links_find(env, mdd_obj, ldata, lname, pfid); + if (rc) + return rc; - rc = __mdd_xattr_set(env, mdd_obj, - mdd_buf_get_const(env, buf->lb_buf, leh->leh_len), - XATTR_NAME_LINK, 0, handle); + mdd_links_del_buf(env, ldata, lname); + return 0; +} +static int mdd_links_rename(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *oldpfid, + const struct lu_name *oldlname, + const struct lu_fid *newpfid, + const struct lu_name *newlname, + struct thandle *handle, + int first, int check) +{ + struct mdd_link_data ldata = { 0 }; + int updated = 0; + int rc2 = 0; + int rc = 0; + ENTRY; + + LASSERT(oldpfid != NULL || newpfid != NULL); + + if (mdd_obj->mod_flags & DEAD_OBJ) + /* No more links, don't bother */ + RETURN(0); + + if (oldpfid != NULL) { + rc = __mdd_links_del(env, mdd_obj, &ldata, + oldlname, oldpfid); + if (rc) { + if ((check == 0) || + (rc != -ENODATA && rc != -ENOENT)) + GOTO(out, rc); + /* No changes done. */ + rc = 0; + } else { + updated = 1; + } + } + + /* If renaming, add the new record */ + if (newpfid != NULL) { + /* even if the add fails, we still delete the out-of-date + * old link */ + rc2 = __mdd_links_add(env, mdd_obj, &ldata, + newlname, newpfid, first, check); + if (rc2 == -EEXIST) + rc2 = 0; + else if (rc2 == 0) + updated = 1; + } + + if (updated) + rc = mdd_links_write(env, mdd_obj, &ldata, handle); + EXIT; out: - if (rc == 0) - rc = rc2; - if (rc) - CDEBUG(D_INODE, "link_ea mv/unlink '%.*s' failed %d "DFID"\n", - oldlname->ln_namelen, oldlname->ln_name, rc, - PFID(mdd_object_fid(mdd_obj))); + if (rc == 0) + rc = rc2; + if (rc) { + int error = 1; + if (rc == -EOVERFLOW || rc == - ENOENT) + error = 0; + if (oldpfid == NULL) + CDEBUG(error ? D_ERROR : D_OTHER, + "link_ea add '%.*s' failed %d "DFID"\n", + newlname->ln_namelen, newlname->ln_name, + rc, PFID(mdd_object_fid(mdd_obj))); + else if (newpfid == NULL) + CDEBUG(error ? D_ERROR : D_OTHER, + "link_ea del '%.*s' failed %d "DFID"\n", + oldlname->ln_namelen, oldlname->ln_name, + rc, PFID(mdd_object_fid(mdd_obj))); + else + CDEBUG(error ? D_ERROR : D_OTHER, + "link_ea rename '%.*s'->'%.*s' failed %d " + DFID"\n", + oldlname->ln_namelen, oldlname->ln_name, + newlname->ln_namelen, newlname->ln_name, + rc, PFID(mdd_object_fid(mdd_obj))); + } + + if (ldata.ml_buf && ldata.ml_buf->lb_len > OBD_ALLOC_BIG) + /* if we vmalloced a large buffer drop it */ + mdd_buf_put(ldata.ml_buf); + + return rc; +} - if (buf->lb_len > OBD_ALLOC_BIG) - /* if we vmalloced a large buffer drop it */ - mdd_buf_put(buf); +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); +} - RETURN (rc); +static inline int mdd_links_del(const struct lu_env *env, + struct mdd_object *mdd_obj, + const struct lu_fid *pfid, + const struct lu_name *lname, + struct thandle *handle) +{ + return mdd_links_rename(env, mdd_obj, pfid, lname, + NULL, NULL, handle, 0, 0); } const struct md_dir_operations mdd_dir_ops = { @@ -2996,9 +2801,5 @@ const struct md_dir_operations mdd_dir_ops = { .mdo_rename = mdd_rename, .mdo_link = mdd_link, .mdo_unlink = mdd_unlink, - .mdo_lum_lmm_cmp = mdd_lum_lmm_cmp, - .mdo_name_insert = mdd_name_insert, - .mdo_name_remove = mdd_name_remove, - .mdo_rename_tgt = mdd_rename_tgt, .mdo_create_data = mdd_create_data, };