X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=2e3f2c15ae3fb98a0a361f262a45f9ac3a795cde;hb=721e945a96186dba1ddd5ef4b2cab8cd8c4e06c2;hp=3667afa7112ea6b09969c684d0ae8725619013f7;hpb=0ad352cfebf46499bf9d3162e6c388f47e65f25d;p=fs%2Flustre-release.git diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 3667afa..2e3f2c1 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -67,8 +67,12 @@ #include #include #include "mdt_internal.h" +#ifdef HAVE_QUOTA_SUPPORT +# include +#endif #include #include +#include mdl_mode_t mdt_mdl_lock_modes[] = { [LCK_MINMODE] = MDL_MINMODE, @@ -95,7 +99,9 @@ ldlm_mode_t mdt_dlm_lock_modes[] = { /* * Initialized in mdt_mod_init(). */ -unsigned long mdt_num_threads; +static unsigned long mdt_num_threads; +static unsigned long mdt_min_threads; +static unsigned long mdt_max_threads; /* ptlrpc request handler for MDT. All handlers are * grouped into several slices - struct mdt_opc_slice, @@ -161,8 +167,10 @@ static struct mdt_opc_slice mdt_fld_handlers[]; static struct mdt_device *mdt_dev(struct lu_device *d); static int mdt_regular_handle(struct ptlrpc_request *req); static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags); +static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt, + struct getinfo_fid2path *fp); -static struct lu_object_operations mdt_obj_ops; +static const struct lu_object_operations mdt_obj_ops; int mdt_get_disposition(struct ldlm_reply *rep, int flag) { @@ -309,7 +317,8 @@ static int mdt_getstatus(struct mdt_thread_info *info) repbody->valid |= OBD_MD_FLID; - if (mdt->mdt_opts.mo_mds_capa) { + if (mdt->mdt_opts.mo_mds_capa && + info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) { struct mdt_object *root; struct lustre_capa *capa; @@ -320,7 +329,6 @@ static int mdt_getstatus(struct mdt_thread_info *info) capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1); LASSERT(capa); capa->lc_opc = CAPA_OPC_MDS_DEFAULT; - rc = mo_capa_get(info->mti_env, mdt_object_child(root), capa, 0); mdt_object_put(info->mti_env, root); @@ -345,7 +353,7 @@ static int mdt_statfs(struct mdt_thread_info *info) /* This will trigger a watchdog timeout */ OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP, (MDT_SERVICE_WATCHDOG_FACTOR * - at_get(&svc->srv_at_estimate) / 1000) + 1); + at_get(&svc->srv_at_estimate)) + 1); rc = mdt_check_ucred(info); if (rc) @@ -362,20 +370,31 @@ static int mdt_statfs(struct mdt_thread_info *info) RETURN(rc); } -void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o) +/** + * Pack SOM attributes into the reply. + * Call under a DLM UPDATE lock. + */ +static void mdt_pack_size2body(struct mdt_thread_info *info, + struct mdt_object *mo) { struct mdt_body *b; - struct lu_attr *attr = &info->mti_attr.ma_attr; + struct md_attr *ma = &info->mti_attr; + LASSERT(ma->ma_attr.la_valid & LA_MODE); b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - /* Check if Size-on-MDS is enabled. */ - if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) && - S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) { - b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS); - b->size = attr->la_size; - b->blocks = attr->la_blocks; - } + /* Check if Size-on-MDS is supported, if this is a regular file, + * if SOM is enabled on the object and if SOM cache exists and valid. + * Otherwise do not pack Size-on-MDS attributes to the reply. */ + if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) || + !S_ISREG(ma->ma_attr.la_mode) || + !mdt_object_is_som_enabled(mo) || + !(ma->ma_valid & MA_SOM)) + return; + + b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + b->size = ma->ma_som->msd_size; + b->blocks = ma->ma_som->msd_blocks; } void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, @@ -427,12 +446,11 @@ static inline int mdt_body_has_lov(const struct lu_attr *la, } static int mdt_getattr_internal(struct mdt_thread_info *info, - struct mdt_object *o) + struct mdt_object *o, int ma_need) { struct md_object *next = mdt_object_child(o); const struct mdt_body *reqbody = info->mti_body; struct ptlrpc_request *req = mdt_info_req(info); - struct mdt_export_data *med = &req->rq_export->exp_mdt_data; struct md_attr *ma = &info->mti_attr; struct lu_attr *la = &ma->ma_attr; struct req_capsule *pill = info->mti_pill; @@ -479,6 +497,10 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, /* get default stripe info for this dir. */ ma->ma_need |= MA_LOV_DEF; } + ma->ma_need |= ma_need; + if (ma->ma_need & MA_SOM) + ma->ma_som = &info->mti_u.som.data; + rc = mo_attr_get(env, next, ma); if (unlikely(rc)) { CERROR("getattr error for "DFID": %d\n", @@ -518,6 +540,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, CERROR("readlink failed: %d\n", rc); rc = -EFAULT; } else { + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO)) + rc -= 2; repbody->valid |= OBD_MD_LINKNAME; repbody->eadatasize = rc; /* NULL terminate */ @@ -537,7 +561,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, repbody->max_cookiesize); } - if (med->med_rmtclient && (reqbody->valid & OBD_MD_FLRMTPERM)) { + if (exp_connect_rmtclient(info->mti_exp) && + reqbody->valid & OBD_MD_FLRMTPERM) { void *buf = req_capsule_server_get(pill, &RMF_ACL); /* mdt_getattr_lock only */ @@ -579,8 +604,9 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, } #endif - if ((reqbody->valid & OBD_MD_FLMDSCAPA) && - info->mti_mdt->mdt_opts.mo_mds_capa) { + if (reqbody->valid & OBD_MD_FLMDSCAPA && + info->mti_mdt->mdt_opts.mo_mds_capa && + info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) { struct lustre_capa *capa; capa = req_capsule_server_get(pill, &RMF_CAPA1); @@ -596,7 +622,6 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, static int mdt_renew_capa(struct mdt_thread_info *info) { - struct mdt_device *mdt = info->mti_mdt; struct mdt_object *obj = info->mti_object; struct mdt_body *body; struct lustre_capa *capa, *c; @@ -607,7 +632,8 @@ static int mdt_renew_capa(struct mdt_thread_info *info) * return directly, client will find body->valid OBD_MD_FLOSSCAPA * flag not set. */ - if (!obj || !mdt->mdt_opts.mo_mds_capa) + if (!obj || !info->mti_mdt->mdt_opts.mo_oss_capa || + !(info->mti_exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA)) RETURN(0); body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); @@ -684,7 +710,7 @@ static int mdt_getattr(struct mdt_thread_info *info) * remote obj, and at that time no capability is available. */ mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA); - rc = mdt_getattr_internal(info, obj); + rc = mdt_getattr_internal(info, obj, 0); if (reqbody->valid & OBD_MD_FLRMTPERM) mdt_exit_ucred(info); EXIT; @@ -783,6 +809,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, struct ldlm_lock *lock; struct ldlm_res_id *res_id; int is_resent; + int ma_need = 0; int rc; ENTRY; @@ -799,14 +826,14 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) - 1; if (!info->mti_cross_ref) { - /* - * XXX: Check for "namelen == 0" is for getattr by fid + /* + * XXX: Check for "namelen == 0" is for getattr by fid * (OBD_CONNECT_ATTRFID), otherwise do not allow empty name, * that is the name must contain at least one character and * the terminating '\0' */ if (namelen == 0) { - reqbody = req_capsule_client_get(info->mti_pill, + reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(fid_is_sane(&reqbody->fid2)); name = NULL; @@ -818,7 +845,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, } else { lname = mdt_name(info->mti_env, (char *)name, namelen); CDEBUG(D_INODE, "getattr with lock for "DFID"/%s, " - "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)), + "ldlm_rep = %p\n", PFID(mdt_object_fid(parent)), name, ldlm_rep); } } @@ -877,7 +904,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, /* Finally, we can get attr for child. */ mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA); - rc = mdt_getattr_internal(info, child); + rc = mdt_getattr_internal(info, child, 0); if (unlikely(rc != 0)) mdt_object_unlock(info, child, lhc, 1); } @@ -948,6 +975,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, relock: ma = &info->mti_attr; + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2); mdt_lock_handle_init(lhc); mdt_lock_reg_init(lhc, LCK_PR); @@ -981,38 +1009,34 @@ relock: GOTO(out_child, rc); } + lock = ldlm_handle2lock(&lhc->mlh_reg_lh); + /* Get MA_SOM attributes if update lock is given. */ + if (lock && + lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE && + S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu))) + ma_need = MA_SOM; + /* finally, we can get attr for child. */ mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA); - rc = mdt_getattr_internal(info, child); + rc = mdt_getattr_internal(info, child, ma_need); if (unlikely(rc != 0)) { mdt_object_unlock(info, child, lhc, 1); - } else { - lock = ldlm_handle2lock(&lhc->mlh_reg_lh); - if (lock) { - struct mdt_body *repbody; + } else if (lock) { + /* Debugging code. */ + res_id = &lock->l_resource->lr_name; + LDLM_DEBUG(lock, "Returning lock to client"); + LASSERTF(fid_res_name_eq(mdt_object_fid(child), + &lock->l_resource->lr_name), + "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n", + (unsigned long)res_id->name[0], + (unsigned long)res_id->name[1], + (unsigned long)res_id->name[2], + PFID(mdt_object_fid(child))); + mdt_pack_size2body(info, child); + } + if (lock) + LDLM_LOCK_PUT(lock); - /* Debugging code. */ - res_id = &lock->l_resource->lr_name; - LDLM_DEBUG(lock, "Returning lock to client\n"); - LASSERTF(fid_res_name_eq(mdt_object_fid(child), - &lock->l_resource->lr_name), - "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n", - (unsigned long)res_id->name[0], - (unsigned long)res_id->name[1], - (unsigned long)res_id->name[2], - PFID(mdt_object_fid(child))); - /* - * Pack Size-on-MDS inode attributes to the body if - * update lock is given. - */ - repbody = req_capsule_server_get(info->mti_pill, - &RMF_MDT_BODY); - if (lock->l_policy_data.l_inodebits.bits & - MDS_INODELOCK_UPDATE) - mdt_pack_size2body(info, child); - LDLM_LOCK_PUT(lock); - } - } EXIT; out_child: mdt_object_put(info->mti_env, child); @@ -1056,19 +1080,22 @@ out_shrink: return rc; } -static struct lu_device_operations mdt_lu_ops; +static const struct lu_device_operations mdt_lu_ops; static int lu_device_is_mdt(struct lu_device *d) { return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops); } +static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, + void *karg, void *uarg); + static int mdt_set_info(struct mdt_thread_info *info) { struct ptlrpc_request *req = mdt_info_req(info); char *key; - __u32 *val; - int keylen, rc = 0; + void *val; + int keylen, vallen, rc = 0; ENTRY; rc = req_capsule_server_pack(info->mti_pill); @@ -1090,19 +1117,40 @@ static int mdt_set_info(struct mdt_thread_info *info) RETURN(-EFAULT); } - if (!KEY_IS(KEY_READ_ONLY)) - RETURN(-EINVAL); + vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL, + RCL_CLIENT); - req->rq_status = 0; - lustre_msg_set_status(req->rq_repmsg, 0); + /* Swab any part of val you need to here */ + if (KEY_IS(KEY_READ_ONLY)) { + req->rq_status = 0; + lustre_msg_set_status(req->rq_repmsg, 0); - spin_lock(&req->rq_export->exp_lock); - if (*val) - req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY; - else - req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY; - spin_unlock(&req->rq_export->exp_lock); + cfs_spin_lock(&req->rq_export->exp_lock); + if (*(__u32 *)val) + req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY; + else + req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY; + cfs_spin_unlock(&req->rq_export->exp_lock); + + } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) { + struct changelog_setinfo *cs = + (struct changelog_setinfo *)val; + if (vallen != sizeof(*cs)) { + CERROR("Bad changelog_clear setinfo size %d\n", vallen); + RETURN(-EINVAL); + } + if (ptlrpc_req_need_swab(req)) { + __swab64s(&cs->cs_recno); + __swab32s(&cs->cs_id); + } + rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp, + vallen, val, NULL); + lustre_msg_set_status(req->rq_repmsg, rc); + + } else { + RETURN(-EINVAL); + } RETURN(0); } @@ -1116,12 +1164,14 @@ static int mdt_connect(struct mdt_thread_info *info) if (rc == 0) { LASSERT(req->rq_export != NULL); info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev); - rc = mdt_init_idmap(info); + rc = mdt_init_sec_level(info); + if (rc == 0) + rc = mdt_init_idmap(info); if (rc != 0) - /* if mdt_init_idmap failed, revocation for connect */ obd_disconnect(class_export_get(req->rq_export)); - } else + } else { rc = err_serious(rc); + } return rc; } @@ -1140,6 +1190,7 @@ static int mdt_sendpage(struct mdt_thread_info *info, struct lu_rdpg *rdpg) { struct ptlrpc_request *req = mdt_info_req(info); + struct obd_export *exp = req->rq_export; struct ptlrpc_bulk_desc *desc; struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info; int tmpcount; @@ -1161,6 +1212,10 @@ static int mdt_sendpage(struct mdt_thread_info *info, } LASSERT(desc->bd_nob == rdpg->rp_count); + rc = sptlrpc_svc_wrap_bulk(req, desc); + if (rc) + GOTO(free_desc, rc); + rc = ptlrpc_start_bulk_transfer(desc); if (rc) GOTO(free_desc, rc); @@ -1172,8 +1227,10 @@ static int mdt_sendpage(struct mdt_thread_info *info, if (timeout < 0) CERROR("Req deadline already passed %lu (now: %lu)\n", req->rq_deadline, cfs_time_current_sec()); - *lwi = LWI_TIMEOUT(max(timeout, 1) * HZ, NULL, NULL); - rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc), lwi); + *lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(max(timeout, 1)), + cfs_time_seconds(1), NULL, NULL); + rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc) || + exp->exp_failed || exp->exp_abort_active_req, lwi); LASSERT (rc == 0 || rc == -ETIMEDOUT); if (rc == 0) { @@ -1181,16 +1238,18 @@ static int mdt_sendpage(struct mdt_thread_info *info, desc->bd_nob_transferred == rdpg->rp_count) GOTO(free_desc, rc); - rc = -ETIMEDOUT; /* XXX should this be a different errno? */ + rc = -ETIMEDOUT; + if (exp->exp_abort_active_req || exp->exp_failed) + GOTO(abort_bulk, rc); } DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s", (rc == -ETIMEDOUT) ? "timeout" : "network error", desc->bd_nob_transferred, rdpg->rp_count, - req->rq_export->exp_client_uuid.uuid, - req->rq_export->exp_connection->c_remote_uuid.uuid); + exp->exp_client_uuid.uuid, + exp->exp_connection->c_remote_uuid.uuid); - class_fail_export(req->rq_export); + class_fail_export(exp); EXIT; abort_bulk: @@ -1258,7 +1317,7 @@ static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page, memcpy(name, ent->lde_name, le16_to_cpu(ent->lde_namelen)); lname = mdt_name(info->mti_env, name, le16_to_cpu(ent->lde_namelen)); - ma->ma_attr_flags |= MDS_PERM_BYPASS; + ma->ma_attr_flags |= (MDS_PERM_BYPASS | MDS_QUOTA_IGNORE); rc = mdo_name_insert(info->mti_env, md_object_next(&object->mot_obj), lname, lf, ma); @@ -1320,6 +1379,9 @@ static int mdt_writepage(struct mdt_thread_info *info) ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size, (int)reqbody->nlink); + rc = sptlrpc_svc_prep_bulk(req, desc); + if (rc != 0) + GOTO(cleanup_page, rc); /* * Check if client was evicted while we were doing i/o before touching * network. @@ -1333,7 +1395,7 @@ static int mdt_writepage(struct mdt_thread_info *info) else rc = ptlrpc_start_bulk_transfer (desc); if (rc == 0) { - *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ, + *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ, mdt_bulk_timeout, desc); rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) || desc->bd_export->exp_failed, lwi); @@ -1400,6 +1462,8 @@ static int mdt_readpage(struct mdt_thread_info *info) rdpg->rp_hash, reqbody->size); RETURN(-EFAULT); } + + rdpg->rp_attrs = reqbody->mode; rdpg->rp_count = reqbody->nlink; rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT; OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]); @@ -1440,6 +1504,7 @@ static int mdt_reint_internal(struct mdt_thread_info *info, { struct req_capsule *pill = info->mti_pill; struct mdt_device *mdt = info->mti_mdt; + struct md_quota *mq = md_quota(info->mti_env); struct mdt_body *repbody; int rc = 0; ENTRY; @@ -1474,6 +1539,18 @@ static int mdt_reint_internal(struct mdt_thread_info *info, GOTO(out_shrink, rc = err_serious(rc)); } + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10); + + /* for replay no cookkie / lmm need, because client have this already */ + if (info->mti_spec.no_create == 1) { + if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) + req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, 0); + + if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) + req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, + 0); + } + rc = mdt_init_ucred_reint(info); if (rc) GOTO(out_shrink, rc); @@ -1486,6 +1563,7 @@ static int mdt_reint_internal(struct mdt_thread_info *info, rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg); GOTO(out_ucred, rc); } + mq->mq_exp = info->mti_exp; rc = mdt_reint_rec(info, lhc); EXIT; out_ucred: @@ -1629,15 +1707,133 @@ static int mdt_sync(struct mdt_thread_info *info) RETURN(rc); } +#ifdef HAVE_QUOTA_SUPPORT static int mdt_quotacheck_handle(struct mdt_thread_info *info) { - return err_serious(-EOPNOTSUPP); + struct obd_quotactl *oqctl; + struct req_capsule *pill = info->mti_pill; + struct obd_export *exp = info->mti_exp; + struct md_quota *mq = md_quota(info->mti_env); + struct md_device *next = info->mti_mdt->mdt_child; + int rc; + ENTRY; + + oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL); + if (oqctl == NULL) + RETURN(-EPROTO); + + /* remote client has no permission for quotacheck */ + if (unlikely(exp_connect_rmtclient(exp))) + RETURN(-EPERM); + + rc = req_capsule_server_pack(pill); + if (rc) + RETURN(rc); + + mq->mq_exp = exp; + rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next, + oqctl->qc_type); + RETURN(rc); } static int mdt_quotactl_handle(struct mdt_thread_info *info) { - return err_serious(-EOPNOTSUPP); + struct obd_quotactl *oqctl, *repoqc; + struct req_capsule *pill = info->mti_pill; + struct obd_export *exp = info->mti_exp; + struct md_quota *mq = md_quota(info->mti_env); + struct md_device *next = info->mti_mdt->mdt_child; + const struct md_quota_operations *mqo = &next->md_ops->mdo_quota; + int id, rc; + ENTRY; + + oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL); + if (oqctl == NULL) + RETURN(-EPROTO); + + id = oqctl->qc_id; + if (exp_connect_rmtclient(exp)) { + struct ptlrpc_request *req = mdt_info_req(info); + struct mdt_export_data *med = mdt_req2med(req); + struct lustre_idmap_table *idmap = med->med_idmap; + + if (unlikely(oqctl->qc_cmd != Q_GETQUOTA && + oqctl->qc_cmd != Q_GETINFO)) + RETURN(-EPERM); + + + if (oqctl->qc_type == USRQUOTA) + id = lustre_idmap_lookup_uid(NULL, idmap, 0, + oqctl->qc_id); + else if (oqctl->qc_type == GRPQUOTA) + id = lustre_idmap_lookup_gid(NULL, idmap, 0, + oqctl->qc_id); + else + RETURN(-EINVAL); + + if (id == CFS_IDMAP_NOTFOUND) { + CDEBUG(D_QUOTA, "no mapping for id %u\n", + oqctl->qc_id); + RETURN(-EACCES); + } + } + + rc = req_capsule_server_pack(pill); + if (rc) + RETURN(rc); + + repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL); + LASSERT(repoqc != NULL); + + mq->mq_exp = exp; + switch (oqctl->qc_cmd) { + case Q_QUOTAON: + rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type); + break; + case Q_QUOTAOFF: + rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type); + break; + case Q_SETINFO: + rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id, + &oqctl->qc_dqinfo); + break; + case Q_GETINFO: + rc = mqo->mqo_getinfo(info->mti_env, next, oqctl->qc_type, id, + &oqctl->qc_dqinfo); + break; + case Q_SETQUOTA: + rc = mqo->mqo_setquota(info->mti_env, next, oqctl->qc_type, id, + &oqctl->qc_dqblk); + break; + case Q_GETQUOTA: + rc = mqo->mqo_getquota(info->mti_env, next, oqctl->qc_type, id, + &oqctl->qc_dqblk); + break; + case Q_GETOINFO: + rc = mqo->mqo_getoinfo(info->mti_env, next, oqctl->qc_type, id, + &oqctl->qc_dqinfo); + break; + case Q_GETOQUOTA: + rc = mqo->mqo_getoquota(info->mti_env, next, oqctl->qc_type, id, + &oqctl->qc_dqblk); + break; + case LUSTRE_Q_INVALIDATE: + rc = mqo->mqo_invalidate(info->mti_env, next, oqctl->qc_type); + break; + case LUSTRE_Q_FINVALIDATE: + rc = mqo->mqo_finvalidate(info->mti_env, next, oqctl->qc_type); + break; + default: + CERROR("unsupported mdt_quotactl command: %d\n", + oqctl->qc_cmd); + RETURN(-EFAULT); + } + + *repoqc = *oqctl; + RETURN(rc); } +#endif + /* * OBD PING and other handlers. @@ -1667,6 +1863,101 @@ static int mdt_obd_qc_callback(struct mdt_thread_info *info) /* + * LLOG handlers. + */ + +/** clone llog ctxt from child (mdd) + * This allows remote llog (replicator) access. + * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the + * context was originally set up, or we can handle them directly. + * I choose the latter, but that means I need any llog + * contexts set up by child to be accessable by the mdt. So we clone the + * context into our context list here. + */ +static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt, + int idx) +{ + struct md_device *next = mdt->mdt_child; + struct llog_ctxt *ctxt; + int rc; + + if (!llog_ctxt_null(mdt2obd_dev(mdt), idx)) + return 0; + + rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt); + if (rc || ctxt == NULL) { + CERROR("Can't get mdd ctxt %d\n", rc); + return rc; + } + + rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx); + if (rc) + CERROR("Can't set mdt ctxt %d\n", rc); + + return rc; +} + +static int mdt_llog_ctxt_unclone(const struct lu_env *env, + struct mdt_device *mdt, int idx) +{ + struct llog_ctxt *ctxt; + + ctxt = llog_get_context(mdt2obd_dev(mdt), idx); + if (ctxt == NULL) + return 0; + /* Put once for the get we just did, and once for the clone */ + llog_ctxt_put(ctxt); + llog_ctxt_put(ctxt); + return 0; +} + +static int mdt_llog_create(struct mdt_thread_info *info) +{ + int rc; + + req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE); + rc = llog_origin_handle_create(mdt_info_req(info)); + return (rc < 0 ? err_serious(rc) : rc); +} + +static int mdt_llog_destroy(struct mdt_thread_info *info) +{ + int rc; + + req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY); + rc = llog_origin_handle_destroy(mdt_info_req(info)); + return (rc < 0 ? err_serious(rc) : rc); +} + +static int mdt_llog_read_header(struct mdt_thread_info *info) +{ + int rc; + + req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER); + rc = llog_origin_handle_read_header(mdt_info_req(info)); + return (rc < 0 ? err_serious(rc) : rc); +} + +static int mdt_llog_next_block(struct mdt_thread_info *info) +{ + int rc; + + req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK); + rc = llog_origin_handle_next_block(mdt_info_req(info)); + return (rc < 0 ? err_serious(rc) : rc); +} + +static int mdt_llog_prev_block(struct mdt_thread_info *info) +{ + int rc; + + req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK); + rc = llog_origin_handle_prev_block(mdt_info_req(info)); + return (rc < 0 ? err_serious(rc) : rc); +} + + +/* * DLM handlers. */ static struct ldlm_callback_suite cbs = { @@ -1776,6 +2067,108 @@ struct mdt_object *mdt_object_find(const struct lu_env *env, RETURN(m); } +/** + * Asyncronous commit for mdt device. + * + * Pass asynchonous commit call down the MDS stack. + * + * \param env environment + * \param mdt the mdt device + */ +static void mdt_device_commit_async(const struct lu_env *env, + struct mdt_device *mdt) +{ + struct dt_device *dt = mdt->mdt_bottom; + int rc; + + rc = dt->dd_ops->dt_commit_async(env, dt); + if (unlikely(rc != 0)) + CWARN("async commit start failed with rc = %d", rc); +} + +/** + * Mark the lock as "synchonous". + * + * Mark the lock to deffer transaction commit to the unlock time. + * + * \param lock the lock to mark as "synchonous" + * + * \see mdt_is_lock_sync + * \see mdt_save_lock + */ +static inline void mdt_set_lock_sync(struct ldlm_lock *lock) +{ + lock->l_ast_data = (void*)1; +} + +/** + * Check whehter the lock "synchonous" or not. + * + * \param lock the lock to check + * \retval 1 the lock is "synchonous" + * \retval 0 the lock isn't "synchronous" + * + * \see mdt_set_lock_sync + * \see mdt_save_lock + */ +static inline int mdt_is_lock_sync(struct ldlm_lock *lock) +{ + return lock->l_ast_data != NULL; +} + +/** + * Blocking AST for mdt locks. + * + * Starts transaction commit if in case of COS lock conflict or + * deffers such a commit to the mdt_save_lock. + * + * \param lock the lock which blocks a request or cancelling lock + * \param desc unused + * \param data unused + * \param flag indicates whether this cancelling or blocking callback + * \retval 0 + * \see ldlm_blocking_ast_nocheck + */ +int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, + void *data, int flag) +{ + struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd; + struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev); + int rc; + ENTRY; + + if (flag == LDLM_CB_CANCELING) + RETURN(0); + lock_res_and_lock(lock); + if (lock->l_blocking_ast != mdt_blocking_ast) { + unlock_res_and_lock(lock); + RETURN(0); + } + if (mdt_cos_is_enabled(mdt) && + lock->l_req_mode & (LCK_PW | LCK_EX) && + lock->l_blocking_lock != NULL && + lock->l_client_cookie != lock->l_blocking_lock->l_client_cookie) { + mdt_set_lock_sync(lock); + } + rc = ldlm_blocking_ast_nocheck(lock); + + /* There is no lock conflict if l_blocking_lock == NULL, + * it indicates a blocking ast sent from ldlm_lock_decref_internal + * when the last reference to a local lock was released */ + if (lock->l_req_mode == LCK_COS && lock->l_blocking_lock != NULL) { + struct lu_env env; + + rc = lu_env_init(&env, LCT_MD_THREAD); + if (unlikely(rc != 0)) + CWARN("lu_env initialization failed with rc = %d," + "cannot start asynchronous commit\n", rc); + else + mdt_device_commit_async(&env, mdt); + lu_env_fini(&env); + } + RETURN(rc); +} + int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o, struct mdt_lock_handle *lh, __u64 ibits, int locality) { @@ -1832,7 +2225,8 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o, */ policy->l_inodebits.bits = MDS_INODELOCK_UPDATE; rc = mdt_fid_lock(ns, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, - policy, res_id, LDLM_FL_ATOMIC_CB); + policy, res_id, LDLM_FL_ATOMIC_CB, + &info->mti_exp->exp_handle.h_cookie); if (unlikely(rc)) RETURN(rc); } @@ -1852,8 +2246,8 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o, * fix it up and turn FL_LOCAL flag off. */ rc = mdt_fid_lock(ns, &lh->mlh_reg_lh, lh->mlh_reg_mode, policy, - res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB); - + res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB, + &info->mti_exp->exp_handle.h_cookie); if (rc) GOTO(out, rc); @@ -1865,36 +2259,79 @@ out: RETURN(rc); } -static inline -void mdt_save_lock(struct ptlrpc_request *req, struct lustre_handle *h, +/** + * Save a lock within request object. + * + * Keep the lock referenced until whether client ACK or transaction + * commit happens or release the lock immediately depending on input + * parameters. If COS is ON, a write lock is converted to COS lock + * before saving. + * + * \param info thead info object + * \param h lock handle + * \param mode lock mode + * \param decref force immediate lock releasing + */ +static +void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, ldlm_mode_t mode, int decref) { ENTRY; if (lustre_handle_is_used(h)) { - if (decref) + if (decref || !info->mti_has_trans || + !(mode & (LCK_PW | LCK_EX))){ mdt_fid_unlock(h, mode); - else - ptlrpc_save_lock(req, h, mode); + } else { + struct mdt_device *mdt = info->mti_mdt; + struct ldlm_lock *lock = ldlm_handle2lock(h); + struct ptlrpc_request *req = mdt_info_req(info); + int no_ack = 0; + + LASSERTF(lock != NULL, "no lock for cookie "LPX64"\n", + h->cookie); + CDEBUG(D_HA, "request = %p reply state = %p" + " transno = "LPD64"\n", + req, req->rq_reply_state, req->rq_transno); + if (mdt_cos_is_enabled(mdt)) { + no_ack = 1; + ldlm_lock_downgrade(lock, LCK_COS); + mode = LCK_COS; + } + ptlrpc_save_lock(req, h, mode, no_ack); + if (mdt_is_lock_sync(lock)) { + CDEBUG(D_HA, "found sync-lock," + " async commit started\n"); + mdt_device_commit_async(info->mti_env, + mdt); + } + LDLM_LOCK_PUT(lock); + } h->cookie = 0ull; } EXIT; } -/* - * Just call ldlm_lock_decref() if decref, else we only call ptlrpc_save_lock() - * to save this lock in req. when transaction committed, req will be released, - * and lock will, too. +/** + * Unlock mdt object. + * + * Immeditely release the regular lock and the PDO lock or save the + * lock in reqeuest and keep them referenced until client ACK or + * transaction commit. + * + * \param info thread info object + * \param o mdt object + * \param lh mdt lock handle referencing regular and PDO locks + * \param decref force immediate lock releasing */ void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o, struct mdt_lock_handle *lh, int decref) { - struct ptlrpc_request *req = mdt_info_req(info); ENTRY; - mdt_save_lock(req, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref); - mdt_save_lock(req, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref); + mdt_save_lock(info, &lh->mlh_pdo_lh, lh->mlh_pdo_mode, decref); + mdt_save_lock(info, &lh->mlh_reg_lh, lh->mlh_reg_mode, decref); EXIT; } @@ -1940,7 +2377,9 @@ static struct mdt_handler *mdt_handler_find(__u32 opc, if (s->mos_opc_start <= opc && opc < s->mos_opc_end) { h = s->mos_hs + (opc - s->mos_opc_start); if (likely(h->mh_opc != 0)) - LASSERT(h->mh_opc == opc); + LASSERTF(h->mh_opc == opc, + "opcode mismatch %d != %d\n", + h->mh_opc, opc); else h = NULL; /* unsupported opc */ break; @@ -2044,6 +2483,7 @@ static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags) struct mdt_device *mdt = info->mti_mdt; /* Pack reply. */ + if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, mdt->mdt_max_mdsize); @@ -2088,7 +2528,7 @@ static int mdt_req_handle(struct mdt_thread_info *info, * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try * to put same checks into handlers like mdt_close(), mdt_reint(), * etc., without talking to mdt authors first. Checking same thing - * there again is useless and returning 0 error wihtout packing reply + * there again is useless and returning 0 error without packing reply * is buggy! Handlers either pack reply or return error. * * We return 0 here and do not send any reply in order to emulate @@ -2176,7 +2616,7 @@ static int mdt_req_handle(struct mdt_thread_info *info, } /* If we're DISCONNECTing, the mdt_export_data is already freed */ - if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT)) + if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT)) target_committed_to_req(req); if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) && @@ -2231,9 +2671,22 @@ static void mdt_thread_info_init(struct ptlrpc_request *req, info->mti_env = req->rq_svc_thread->t_env; ci = md_capainfo(info->mti_env); memset(ci, 0, sizeof *ci); + if (req->rq_export) { + if (exp_connect_rmtclient(req->rq_export)) + ci->mc_auth = LC_ID_CONVERT; + else if (req->rq_export->exp_connect_flags & + OBD_CONNECT_MDS_CAPA) + ci->mc_auth = LC_ID_PLAIN; + else + ci->mc_auth = LC_ID_NONE; + } info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET; info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg); + info->mti_mos[0] = NULL; + info->mti_mos[1] = NULL; + info->mti_mos[2] = NULL; + info->mti_mos[3] = NULL; memset(&info->mti_attr, 0, sizeof(info->mti_attr)); info->mti_body = NULL; @@ -2246,6 +2699,7 @@ static void mdt_thread_info_init(struct ptlrpc_request *req, /* To not check for split by default. */ info->mti_spec.sp_ck_split = 0; + info->mti_spec.no_create = 0; } static void mdt_thread_info_fini(struct mdt_thread_info *info) @@ -2254,6 +2708,11 @@ static void mdt_thread_info_fini(struct mdt_thread_info *info) req_capsule_fini(info->mti_pill); if (info->mti_object != NULL) { + /* + * freeing an object may lead to OSD level transaction, do not + * let it mess with MDT. bz19385. + */ + info->mti_no_need_trans = 1; mdt_object_put(info->mti_env, info->mti_object); info->mti_object = NULL; } @@ -2300,7 +2759,6 @@ static int mdt_filter_recovery_request(struct ptlrpc_request *req, static int mdt_recovery(struct mdt_thread_info *info) { struct ptlrpc_request *req = mdt_info_req(info); - int recovering; struct obd_device *obd; ENTRY; @@ -2323,7 +2781,7 @@ static int mdt_recovery(struct mdt_thread_info *info) } } - if (unlikely(req->rq_export == NULL)) { + if (unlikely(!class_connected_export(req->rq_export))) { CERROR("operation %d on unconnected MDS from %s\n", lustre_msg_get_opc(req->rq_reqmsg), libcfs_id2str(req->rq_peer)); @@ -2361,10 +2819,7 @@ static int mdt_recovery(struct mdt_thread_info *info) obd = req->rq_export->exp_obd; /* Check for aborted recovery... */ - spin_lock_bh(&obd->obd_processing_task_lock); - recovering = obd->obd_recovering; - spin_unlock_bh(&obd->obd_processing_task_lock); - if (unlikely(recovering)) { + if (unlikely(obd->obd_recovering)) { int rc; int should_process; DEBUG_REQ(D_INFO, req, "Got new replay"); @@ -2387,6 +2842,7 @@ static int mdt_msg_check_version(struct lustre_msg *msg) switch (lustre_msg_get_opc(msg)) { case MDS_CONNECT: case MDS_DISCONNECT: + case MDS_SET_INFO: case OBD_PING: case SEC_CTX_INIT: case SEC_CTX_INIT_CONT: @@ -2412,7 +2868,7 @@ static int mdt_msg_check_version(struct lustre_msg *msg) case MDS_SYNC: case MDS_GETXATTR: case MDS_SETXATTR: - case MDS_SET_INFO: + case MDS_GET_INFO: case MDS_QUOTACHECK: case MDS_QUOTACTL: case QUOTA_DQACQ: @@ -2484,7 +2940,8 @@ static int mdt_handle0(struct ptlrpc_request *req, if (likely(h != NULL)) { rc = mdt_req_handle(info, h, req); } else { - CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) ); + CERROR("The unsupported opc: 0x%x\n", + lustre_msg_get_opc(msg) ); req->rq_status = -ENOTSUPP; rc = ptlrpc_error(req); RETURN(rc); @@ -2733,17 +3190,18 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info, new_lock->l_writers--; } - new_lock->l_export = class_export_get(req->rq_export); + new_lock->l_export = class_export_lock_get(req->rq_export, new_lock); new_lock->l_blocking_ast = lock->l_blocking_ast; new_lock->l_completion_ast = lock->l_completion_ast; new_lock->l_remote_handle = lock->l_remote_handle; new_lock->l_flags &= ~LDLM_FL_LOCAL; - lustre_hash_add(new_lock->l_export->exp_lock_hash, - &new_lock->l_remote_handle, - &new_lock->l_exp_hash); - unlock_res_and_lock(new_lock); + + cfs_hash_add(new_lock->l_export->exp_lock_hash, + &new_lock->l_remote_handle, + &new_lock->l_exp_hash); + LDLM_LOCK_RELEASE(new_lock); lh->mlh_reg_lh.cookie = 0; @@ -2767,7 +3225,7 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ); remote_hdl = dlmreq->lock_handle[0]; - lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl); + lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl); if (lock) { if (lock != new_lock) { lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie; @@ -2779,11 +3237,11 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, lh->mlh_reg_lh.cookie); if (old_lock) *old_lock = LDLM_LOCK_GET(lock); - lh_put(exp->exp_lock_hash, &lock->l_exp_hash); + cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash); return; } - lh_put(exp->exp_lock_hash, &lock->l_exp_hash); + cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash); } /* @@ -2924,14 +3382,15 @@ static int mdt_intent_reint(enum mdt_it_code opcode, } rep->lock_policy_res2 = clear_serious(rc); - lhc->mlh_reg_lh.cookie = 0ull; - if (rc == -ENOTCONN || rc == -ENODEV) { + if (rc == -ENOTCONN || rc == -ENODEV || + rc == -EOVERFLOW) { /**< if VBR failure then return error */ /* * If it is the disconnect error (ENODEV & ENOCONN), the error * will be returned by rq_status, and client at ptlrpc layer * will detect this, then disconnect, reconnect the import * immediately, instead of impacting the following the rpc. */ + lhc->mlh_reg_lh.cookie = 0ull; RETURN(rc); } else { /* @@ -2942,7 +3401,14 @@ static int mdt_intent_reint(enum mdt_it_code opcode, * FIXME: when open lock is finished, that should be * checked here. */ - RETURN(ELDLM_LOCK_ABORTED); + if (lustre_handle_is_used(&lhc->mlh_reg_lh)) { + rep->lock_policy_res2 = 0; + rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); + RETURN(rc); + } else { + lhc->mlh_reg_lh.cookie = 0ull; + RETURN(ELDLM_LOCK_ABORTED); + } } } @@ -3080,15 +3546,15 @@ static int mdt_intent_policy(struct ldlm_namespace *ns, static void mdt_seq_adjust(const struct lu_env *env, struct mdt_device *m, int lost) { - struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site; - struct lu_range out; + struct md_site *ms = mdt_md_site(m); + struct lu_seq_range out; ENTRY; - LASSERT(ls && ls->ls_server_seq); + LASSERT(ms && ms->ms_server_seq); LASSERT(lost >= 0); /* get extra seq from seq_server, moving it's range up */ while (lost-- > 0) { - seq_server_alloc_meta(ls->ls_server_seq, NULL, &out, env); + seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env); } EXIT; } @@ -3096,25 +3562,27 @@ static void mdt_seq_adjust(const struct lu_env *env, static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *m) { - struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site; + struct md_site *ms = mdt_md_site(m); ENTRY; - if (ls && ls->ls_server_seq) { - seq_server_fini(ls->ls_server_seq, env); - OBD_FREE_PTR(ls->ls_server_seq); - ls->ls_server_seq = NULL; + if (ms != NULL) { + if (ms->ms_server_seq) { + seq_server_fini(ms->ms_server_seq, env); + OBD_FREE_PTR(ms->ms_server_seq); + ms->ms_server_seq = NULL; } - if (ls && ls->ls_control_seq) { - seq_server_fini(ls->ls_control_seq, env); - OBD_FREE_PTR(ls->ls_control_seq); - ls->ls_control_seq = NULL; + if (ms->ms_control_seq) { + seq_server_fini(ms->ms_control_seq, env); + OBD_FREE_PTR(ms->ms_control_seq); + ms->ms_control_seq = NULL; } - if (ls && ls->ls_client_seq) { - seq_client_fini(ls->ls_client_seq); - OBD_FREE_PTR(ls->ls_client_seq); - ls->ls_client_seq = NULL; + if (ms->ms_client_seq) { + seq_client_fini(ms->ms_client_seq); + OBD_FREE_PTR(ms->ms_client_seq); + ms->ms_client_seq = NULL; + } } RETURN(0); @@ -3124,39 +3592,40 @@ static int mdt_seq_init(const struct lu_env *env, const char *uuid, struct mdt_device *m) { - struct lu_site *ls; + struct md_site *ms; char *prefix; int rc; ENTRY; - ls = m->mdt_md_dev.md_lu_dev.ld_site; + ms = mdt_md_site(m); /* * This is sequence-controller node. Init seq-controller server on local * MDT. */ - if (ls->ls_node_id == 0) { - LASSERT(ls->ls_control_seq == NULL); + if (ms->ms_node_id == 0) { + LASSERT(ms->ms_control_seq == NULL); - OBD_ALLOC_PTR(ls->ls_control_seq); - if (ls->ls_control_seq == NULL) + OBD_ALLOC_PTR(ms->ms_control_seq); + if (ms->ms_control_seq == NULL) RETURN(-ENOMEM); - rc = seq_server_init(ls->ls_control_seq, + rc = seq_server_init(ms->ms_control_seq, m->mdt_bottom, uuid, LUSTRE_SEQ_CONTROLLER, + ms, env); if (rc) GOTO(out_seq_fini, rc); - OBD_ALLOC_PTR(ls->ls_client_seq); - if (ls->ls_client_seq == NULL) + OBD_ALLOC_PTR(ms->ms_client_seq); + if (ms->ms_client_seq == NULL) GOTO(out_seq_fini, rc = -ENOMEM); OBD_ALLOC(prefix, MAX_OBD_NAME + 5); if (prefix == NULL) { - OBD_FREE_PTR(ls->ls_client_seq); + OBD_FREE_PTR(ms->ms_client_seq); GOTO(out_seq_fini, rc = -ENOMEM); } @@ -3165,11 +3634,11 @@ static int mdt_seq_init(const struct lu_env *env, /* * Init seq-controller client after seq-controller server is - * ready. Pass ls->ls_control_seq to it for direct talking. + * ready. Pass ms->ms_control_seq to it for direct talking. */ - rc = seq_client_init(ls->ls_client_seq, NULL, + rc = seq_client_init(ms->ms_client_seq, NULL, LUSTRE_SEQ_METADATA, prefix, - ls->ls_control_seq); + ms->ms_control_seq); OBD_FREE(prefix, MAX_OBD_NAME + 5); if (rc) @@ -3177,25 +3646,26 @@ static int mdt_seq_init(const struct lu_env *env, } /* Init seq-server on local MDT */ - LASSERT(ls->ls_server_seq == NULL); + LASSERT(ms->ms_server_seq == NULL); - OBD_ALLOC_PTR(ls->ls_server_seq); - if (ls->ls_server_seq == NULL) + OBD_ALLOC_PTR(ms->ms_server_seq); + if (ms->ms_server_seq == NULL) GOTO(out_seq_fini, rc = -ENOMEM); - rc = seq_server_init(ls->ls_server_seq, + rc = seq_server_init(ms->ms_server_seq, m->mdt_bottom, uuid, LUSTRE_SEQ_SERVER, + ms, env); if (rc) GOTO(out_seq_fini, rc = -ENOMEM); /* Assign seq-controller client to local seq-server. */ - if (ls->ls_node_id == 0) { - LASSERT(ls->ls_client_seq != NULL); + if (ms->ms_node_id == 0) { + LASSERT(ms->ms_client_seq != NULL); - rc = seq_server_set_cli(ls->ls_server_seq, - ls->ls_client_seq, + rc = seq_server_set_cli(ms->ms_server_seq, + ms->ms_client_seq, env); } @@ -3214,7 +3684,7 @@ static int mdt_seq_init_cli(const struct lu_env *env, struct mdt_device *m, struct lustre_cfg *cfg) { - struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site; + struct md_site *ms = mdt_md_site(m); struct obd_device *mdc; struct obd_uuid *uuidp, *mdcuuidp; char *uuid_str, *mdc_uuid_str; @@ -3238,7 +3708,7 @@ static int mdt_seq_init_cli(const struct lu_env *env, /* check if this is adding the first MDC and controller is not yet * initialized. */ - if (index != 0 || ls->ls_client_seq) + if (index != 0 || ms->ms_client_seq) RETURN(0); uuid_str = lustre_cfg_string(cfg, 1); @@ -3255,9 +3725,9 @@ static int mdt_seq_init_cli(const struct lu_env *env, CERROR("target %s not set up\n", mdc->obd_name); rc = -EINVAL; } else { - LASSERT(ls->ls_control_exp); - OBD_ALLOC_PTR(ls->ls_client_seq); - if (ls->ls_client_seq != NULL) { + LASSERT(ms->ms_control_exp); + OBD_ALLOC_PTR(ms->ms_client_seq); + if (ms->ms_client_seq != NULL) { char *prefix; OBD_ALLOC(prefix, MAX_OBD_NAME + 5); @@ -3267,8 +3737,8 @@ static int mdt_seq_init_cli(const struct lu_env *env, snprintf(prefix, MAX_OBD_NAME + 5, "ctl-%s", mdc->obd_name); - rc = seq_client_init(ls->ls_client_seq, - ls->ls_control_exp, + rc = seq_client_init(ms->ms_client_seq, + ms->ms_control_exp, LUSTRE_SEQ_METADATA, prefix, NULL); OBD_FREE(prefix, MAX_OBD_NAME + 5); @@ -3278,8 +3748,8 @@ static int mdt_seq_init_cli(const struct lu_env *env, if (rc) RETURN(rc); - LASSERT(ls->ls_server_seq != NULL); - rc = seq_server_set_cli(ls->ls_server_seq, ls->ls_client_seq, + LASSERT(ms->ms_server_seq != NULL); + rc = seq_server_set_cli(ms->ms_server_seq, ms->ms_client_seq, env); } @@ -3288,19 +3758,21 @@ static int mdt_seq_init_cli(const struct lu_env *env, static void mdt_seq_fini_cli(struct mdt_device *m) { - struct lu_site *ls; + struct md_site *ms; ENTRY; - ls = m->mdt_md_dev.md_lu_dev.ld_site; + ms = mdt_md_site(m); - if (ls && ls->ls_server_seq) - seq_server_set_cli(ls->ls_server_seq, + if (ms != NULL) { + if (ms->ms_server_seq) + seq_server_set_cli(ms->ms_server_seq, NULL, NULL); - if (ls && ls->ls_control_exp) { - class_export_put(ls->ls_control_exp); - ls->ls_control_exp = NULL; + if (ms->ms_control_exp) { + class_export_put(ms->ms_control_exp); + ms->ms_control_exp = NULL; + } } EXIT; } @@ -3311,13 +3783,13 @@ static void mdt_seq_fini_cli(struct mdt_device *m) static int mdt_fld_fini(const struct lu_env *env, struct mdt_device *m) { - struct lu_site *ls = m->mdt_md_dev.md_lu_dev.ld_site; + struct md_site *ms = mdt_md_site(m); ENTRY; - if (ls && ls->ls_server_fld) { - fld_server_fini(ls->ls_server_fld, env); - OBD_FREE_PTR(ls->ls_server_fld); - ls->ls_server_fld = NULL; + if (ms && ms->ms_server_fld) { + fld_server_fini(ms->ms_server_fld, env); + OBD_FREE_PTR(ms->ms_server_fld); + ms->ms_server_fld = NULL; } RETURN(0); @@ -3327,21 +3799,22 @@ static int mdt_fld_init(const struct lu_env *env, const char *uuid, struct mdt_device *m) { - struct lu_site *ls; + struct md_site *ms; int rc; ENTRY; - ls = m->mdt_md_dev.md_lu_dev.ld_site; + ms = mdt_md_site(m); - OBD_ALLOC_PTR(ls->ls_server_fld); - if (ls->ls_server_fld == NULL) + OBD_ALLOC_PTR(ms->ms_server_fld); + if (ms->ms_server_fld == NULL) RETURN(rc = -ENOMEM); - rc = fld_server_init(ls->ls_server_fld, - m->mdt_bottom, uuid, env); + rc = fld_server_init(ms->ms_server_fld, + m->mdt_bottom, uuid, + env, ms->ms_node_id); if (rc) { - OBD_FREE_PTR(ls->ls_server_fld); - ls->ls_server_fld = NULL; + OBD_FREE_PTR(ms->ms_server_fld); + ms->ms_server_fld = NULL; RETURN(rc); } @@ -3384,7 +3857,7 @@ static void mdt_stop_ptlrpc_service(struct mdt_device *m) ptlrpc_unregister_service(m->mdt_fld_service); m->mdt_fld_service = NULL; } - ENTRY; + EXIT; } static int mdt_start_ptlrpc_service(struct mdt_device *m) @@ -3408,9 +3881,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) * We'd like to have a mechanism to set this on a per-device * basis, but alas... */ - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD }; @@ -3441,9 +3913,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = MDS_READPAGE_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD }; m->mdt_readpage_service = @@ -3469,9 +3940,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = MDS_SETATTR_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD }; @@ -3500,8 +3970,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = SEQ_CONTROLLER_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; @@ -3529,8 +3999,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = SEQ_METADATA_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; @@ -3561,8 +4031,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = SEQ_DATA_PORTAL, .psc_rep_portal = OSC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; @@ -3588,8 +4058,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = FLD_REQUEST_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = FLD_NUM_THREADS, - .psc_max_threads = FLD_NUM_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD }; @@ -3618,9 +4088,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = MDS_MDS_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD }; m->mdt_xmds_service = @@ -3629,7 +4098,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) procfs_entry, target_print_req,"mdt_xmds"); if (m->mdt_xmds_service == NULL) { - CERROR("failed to start readpage service\n"); + CERROR("failed to start xmds service\n"); GOTO(err_mdt_svc, rc = -ENOMEM); } @@ -3648,7 +4117,7 @@ err_mdt_svc: static void mdt_stack_fini(const struct lu_env *env, struct mdt_device *m, struct lu_device *top) { - struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd; + struct obd_device *obd = mdt2obd_dev(m); struct lustre_cfg_bufs *bufs; struct lustre_cfg *lcfg; struct mdt_thread_info *info; @@ -3700,20 +4169,12 @@ static struct lu_device *mdt_layer_setup(struct lu_env *env, GOTO(out, rc = -ENODEV); } - rc = lu_context_refill(&env->le_ctx); + rc = lu_env_refill((struct lu_env *)env); if (rc != 0) { - CERROR("Failure to refill context: '%d'\n", rc); + CERROR("Failure to refill session: '%d'\n", rc); GOTO(out_type, rc); } - if (env->le_ses != NULL) { - rc = lu_context_refill(env->le_ses); - if (rc != 0) { - CERROR("Failure to refill session: '%d'\n", rc); - GOTO(out_type, rc); - } - } - ldt = type->typ_lu; if (ldt == NULL) { CERROR("type: '%s'\n", typename); @@ -3737,6 +4198,7 @@ static struct lu_device *mdt_layer_setup(struct lu_env *env, GOTO(out_alloc, rc); } lu_device_get(d); + lu_ref_add(&d->ld_reference, "lu-stack", &lu_site_init); RETURN(d); @@ -3750,11 +4212,14 @@ out: } static int mdt_stack_init(struct lu_env *env, - struct mdt_device *m, struct lustre_cfg *cfg) + struct mdt_device *m, + struct lustre_cfg *cfg, + struct lustre_mount_info *lmi) { struct lu_device *d = &m->mdt_md_dev.md_lu_dev; struct lu_device *tmp; struct md_device *md; + struct lu_device *child_lu_dev; int rc; ENTRY; @@ -3789,7 +4254,15 @@ static int mdt_stack_init(struct lu_env *env, /* process setup config */ tmp = &m->mdt_md_dev.md_lu_dev; rc = tmp->ld_ops->ldo_process_config(env, tmp, cfg); - GOTO(out, rc); + if (rc) + GOTO(out, rc); + + /* initialize local objects */ + child_lu_dev = &m->mdt_child->md_lu_dev; + + rc = child_lu_dev->ld_ops->ldo_prepare(env, + &m->mdt_md_dev.md_lu_dev, + child_lu_dev); out: /* fini from last known good lu_device */ if (rc) @@ -3798,60 +4271,155 @@ out: return rc; } -static void mdt_fini(const struct lu_env *env, struct mdt_device *m) +/** + * setup CONFIG_ORIG context, used to access local config log. + * this may need to be rewrite as part of llog rewrite for lu-api. + */ +static int mdt_obd_llog_setup(struct obd_device *obd, + struct lustre_sb_info *lsi) { - struct md_device *next = m->mdt_child; - struct lu_device *d = &m->mdt_md_dev.md_lu_dev; - struct lu_site *ls = d->ld_site; - struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd; - ENTRY; + int rc; - ping_evictor_stop(); + LASSERT(obd->obd_fsops == NULL); - target_recovery_fini(obd); - mdt_stop_ptlrpc_service(m); + obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd)); + if (IS_ERR(obd->obd_fsops)) + return PTR_ERR(obd->obd_fsops); - mdt_fs_cleanup(env, m); + rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb); + if (rc) { + fsfilt_put_ops(obd->obd_fsops); + return rc; + } - upcall_cache_cleanup(m->mdt_identity_cache); - m->mdt_identity_cache = NULL; + OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt); + obd->obd_lvfs_ctxt.pwdmnt = lsi->lsi_srv_mnt; + obd->obd_lvfs_ctxt.pwd = lsi->lsi_srv_mnt->mnt_root; + obd->obd_lvfs_ctxt.fs = get_ds(); - if (m->mdt_namespace != NULL) { - ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force); - d->ld_obd->obd_namespace = m->mdt_namespace = NULL; + rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd, + 0, NULL, &llog_lvfs_ops); + if (rc) { + CERROR("llog_setup() failed: %d\n", rc); + fsfilt_put_ops(obd->obd_fsops); } - mdt_seq_fini(env, m); - mdt_seq_fini_cli(m); - mdt_fld_fini(env, m); - mdt_procfs_fini(m); - ptlrpc_lprocfs_unregister_obd(d->ld_obd); - lprocfs_obd_cleanup(d->ld_obd); + return rc; +} + +static void mdt_obd_llog_cleanup(struct obd_device *obd) +{ + struct llog_ctxt *ctxt; + + ctxt = llog_get_context(obd, LLOG_CONFIG_ORIG_CTXT); + if (ctxt) + llog_cleanup(ctxt); + + if (obd->obd_fsops) { + fsfilt_put_ops(obd->obd_fsops); + obd->obd_fsops = NULL; + } +} + +static void mdt_fini(const struct lu_env *env, struct mdt_device *m) +{ + struct md_device *next = m->mdt_child; + struct lu_device *d = &m->mdt_md_dev.md_lu_dev; + struct lu_site *ls = d->ld_site; + struct obd_device *obd = mdt2obd_dev(m); + ENTRY; + + target_recovery_fini(obd); + + ping_evictor_stop(); + + mdt_stop_ptlrpc_service(m); + mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT); + mdt_obd_llog_cleanup(obd); + obd_exports_barrier(obd); + obd_zombie_barrier(); +#ifdef HAVE_QUOTA_SUPPORT + next->md_ops->mdo_quota.mqo_cleanup(env, next); +#endif + lut_fini(env, &m->mdt_lut); + mdt_fs_cleanup(env, m); + upcall_cache_cleanup(m->mdt_identity_cache); + m->mdt_identity_cache = NULL; + + if (m->mdt_namespace != NULL) { + ldlm_namespace_free(m->mdt_namespace, NULL, + d->ld_obd->obd_force); + d->ld_obd->obd_namespace = m->mdt_namespace = NULL; + } + cfs_free_nidlist(&m->mdt_nosquash_nids); + if (m->mdt_nosquash_str) { + OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen); + m->mdt_nosquash_str = NULL; + m->mdt_nosquash_strlen = 0; + } + + mdt_seq_fini(env, m); + mdt_seq_fini_cli(m); + mdt_fld_fini(env, m); sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset); next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL); cfs_timer_disarm(&m->mdt_ck_timer); mdt_ck_thread_stop(m); - /* finish the stack */ + /* + * Finish the stack + */ mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); + lprocfs_free_per_client_stats(obd); + lprocfs_free_obd_stats(obd); + mdt_procfs_fini(m); + if (ls) { + struct md_site *mite; + lu_site_fini(ls); - OBD_FREE_PTR(ls); + mite = lu_site2md(ls); + OBD_FREE_PTR(mite); d->ld_site = NULL; } - LASSERT(atomic_read(&d->ld_ref) == 0); - md_device_fini(&m->mdt_md_dev); + LASSERT(cfs_atomic_read(&d->ld_ref) == 0); EXIT; } +static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial) +{ + struct mdt_device *m = mdt_dev(obd->obd_lu_dev); + struct sptlrpc_rule_set tmp_rset; + int rc; + + sptlrpc_rule_set_init(&tmp_rset); + rc = sptlrpc_conf_target_get_rules(obd, &tmp_rset, initial); + if (rc) { + CERROR("mdt %s: failed get sptlrpc rules: %d\n", + obd->obd_name, rc); + return rc; + } + + sptlrpc_target_update_exp_flavor(obd, &tmp_rset); + + cfs_write_lock(&m->mdt_sptlrpc_lock); + sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset); + m->mdt_sptlrpc_rset = tmp_rset; + cfs_write_unlock(&m->mdt_sptlrpc_lock); + + return 0; +} + static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options) { char *p = options; + m->mdt_opts.mo_mds_capa = 1; + m->mdt_opts.mo_oss_capa = 1; #ifdef CONFIG_FS_POSIX_ACL /* ACLs should be enabled by default (b=13829) */ m->mdt_opts.mo_acl = 1; @@ -3894,31 +4462,49 @@ int mdt_postrecov(const struct lu_env *, struct mdt_device *); static int mdt_init0(const struct lu_env *env, struct mdt_device *m, struct lu_device_type *ldt, struct lustre_cfg *cfg) { - struct lprocfs_static_vars lvars; struct mdt_thread_info *info; struct obd_device *obd; const char *dev = lustre_cfg_string(cfg, 0); const char *num = lustre_cfg_string(cfg, 2); - struct lustre_mount_info *lmi; + struct lustre_mount_info *lmi = NULL; struct lustre_sb_info *lsi; + struct lustre_disk_data *ldd; struct lu_site *s; + struct md_site *mite; const char *identity_upcall = "NONE"; +#ifdef HAVE_QUOTA_SUPPORT + struct md_device *next; +#endif int rc; + int node_id; ENTRY; + md_device_init(&m->mdt_md_dev, ldt); + /* + * Environment (env) might be missing mdt_thread_key values at that + * point, if device is allocated when mdt_thread_key is in QUIESCENT + * mode. + * + * Usually device allocation path doesn't use module key values, but + * mdt has to do a lot of work here, so allocate key value. + */ + rc = lu_env_refill((struct lu_env *)env); + if (rc != 0) + RETURN(rc); + info = lu_context_key_get(&env->le_ctx, &mdt_thread_key); LASSERT(info != NULL); obd = class_name2obd(dev); LASSERT(obd != NULL); - spin_lock_init(&m->mdt_transno_lock); - m->mdt_max_mdsize = MAX_MD_SIZE; m->mdt_max_cookiesize = sizeof(struct llog_cookie); + m->mdt_som_conf = 0; m->mdt_opts.mo_user_xattr = 0; m->mdt_opts.mo_acl = 0; + m->mdt_opts.mo_cos = MDT_COS_DEFAULT; lmi = server_get_mount_2(dev); if (lmi == NULL) { CERROR("Cannot get mount info for %s!\n", dev); @@ -3926,25 +4512,37 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, } else { lsi = s2lsi(lmi->lmi_sb); fsoptions_to_mdt_flags(m, lsi->lsi_lmd->lmd_opts); - server_put_mount_2(dev, lmi->lmi_mnt); + /* CMD is supported only in IAM mode */ + ldd = lsi->lsi_ldd; + LASSERT(num); + node_id = simple_strtol(num, NULL, 10); + if (!(ldd->ldd_flags & LDD_F_IAM_DIR) && node_id) { + CERROR("CMD Operation not allowed in IOP mode\n"); + GOTO(err_lmi, rc = -EINVAL); + } } - m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED; + cfs_rwlock_init(&m->mdt_sptlrpc_lock); sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset); - spin_lock_init(&m->mdt_ioepoch_lock); + cfs_spin_lock_init(&m->mdt_ioepoch_lock); m->mdt_opts.mo_compat_resname = 0; m->mdt_capa_timeout = CAPA_TIMEOUT; m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1; m->mdt_ck_timeout = CAPA_KEY_TIMEOUT; + m->mdt_squash_uid = 0; + m->mdt_squash_gid = 0; + CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids); + m->mdt_nosquash_str = NULL; + m->mdt_nosquash_strlen = 0; + cfs_init_rwsem(&m->mdt_squash_sem); - spin_lock_init(&m->mdt_client_bitmap_lock); + OBD_ALLOC_PTR(mite); + if (mite == NULL) + GOTO(err_lmi, rc = -ENOMEM); - OBD_ALLOC_PTR(s); - if (s == NULL) - RETURN(-ENOMEM); + s = &mite->ms_lu; - md_device_init(&m->mdt_md_dev, ldt); m->mdt_md_dev.md_lu_dev.ld_ops = &mdt_lu_ops; m->mdt_md_dev.md_lu_dev.ld_obd = obd; /* set this lu_device to obd, because error handling need it */ @@ -3956,14 +4554,6 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, GOTO(err_free_site, rc); } - lprocfs_mdt_init_vars(&lvars); - rc = lprocfs_obd_setup(obd, lvars.obd_vars); - if (rc) { - CERROR("Can't init lprocfs, rc %d\n", rc); - GOTO(err_fini_site, rc); - } - ptlrpc_lprocfs_register_obd(obd); - rc = mdt_procfs_init(m, dev); if (rc) { CERROR("Can't init MDT lprocfs, rc %d\n", rc); @@ -3971,12 +4561,11 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, } /* set server index */ - LASSERT(num); - s->ls_node_id = simple_strtol(num, NULL, 10); + lu_site2md(s)->ms_node_id = node_id; /* failover is the default * FIXME: we do not failout mds0/mgs, which may cause some problems. - * assumed whose ls_node_id == 0 XXX + * assumed whose ms_node_id == 0 XXX * */ obd->obd_replayable = 1; /* No connection accepted until configurations will finish */ @@ -3991,16 +4580,20 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, } /* init the stack */ - rc = mdt_stack_init((struct lu_env *)env, m, cfg); + rc = mdt_stack_init((struct lu_env *)env, m, cfg, lmi); if (rc) { CERROR("Can't init device stack, rc %d\n", rc); GOTO(err_fini_proc, rc); } - rc = mdt_fld_init(env, obd->obd_name, m); + rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom); if (rc) GOTO(err_fini_stack, rc); + rc = mdt_fld_init(env, obd->obd_name, m); + if (rc) + GOTO(err_lut, rc); + rc = mdt_seq_init(env, obd->obd_name, m); if (rc) GOTO(err_fini_fld, rc); @@ -4036,15 +4629,35 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, if (rc) GOTO(err_free_ns, rc); - rc = mdt_fs_setup(env, m, obd); + rc = mdt_fs_setup(env, m, obd, lsi); if (rc) GOTO(err_capa, rc); - target_recovery_init(obd, mdt_recovery_handle); + rc = mdt_obd_llog_setup(obd, lsi); + if (rc) + GOTO(err_fs_cleanup, rc); + + rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT); + if (rc) + GOTO(err_llog_cleanup, rc); + + mdt_adapt_sptlrpc_conf(obd, 1); + +#ifdef HAVE_QUOTA_SUPPORT + next = m->mdt_child; + rc = next->md_ops->mdo_quota.mqo_setup(env, next, lmi->lmi_mnt); + if (rc) + GOTO(err_llog_cleanup, rc); +#endif + + server_put_mount_2(dev, lmi->lmi_mnt); + lmi = NULL; + + target_recovery_init(&m->mdt_lut, mdt_recovery_handle); rc = mdt_start_ptlrpc_service(m); if (rc) - GOTO(err_fs_cleanup, rc); + GOTO(err_recovery, rc); ping_evictor_start(); @@ -4068,8 +4681,15 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, err_stop_service: ping_evictor_stop(); mdt_stop_ptlrpc_service(m); -err_fs_cleanup: +err_recovery: target_recovery_fini(obd); +#ifdef HAVE_QUOTA_SUPPORT + next->md_ops->mdo_quota.mqo_cleanup(env, next); +#endif +err_llog_cleanup: + mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT); + mdt_obd_llog_cleanup(obd); +err_fs_cleanup: mdt_fs_cleanup(env, m); err_capa: cfs_timer_disarm(&m->mdt_ck_timer); @@ -4083,18 +4703,18 @@ err_fini_seq: mdt_seq_fini(env, m); err_fini_fld: mdt_fld_fini(env, m); +err_lut: + lut_fini(env, &m->mdt_lut); err_fini_stack: mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); err_fini_proc: mdt_procfs_fini(m); - ptlrpc_lprocfs_unregister_obd(obd); - lprocfs_obd_cleanup(obd); -err_fini_site: lu_site_fini(s); err_free_site: - OBD_FREE_PTR(s); - - md_device_fini(&m->mdt_md_dev); + OBD_FREE_PTR(mite); +err_lmi: + if (lmi) + server_put_mount_2(dev, lmi->lmi_mnt); return (rc); } @@ -4109,42 +4729,27 @@ static int mdt_process_config(const struct lu_env *env, ENTRY; switch (cfg->lcfg_command) { - case LCFG_SPTLRPC_CONF: { - struct sptlrpc_conf_log *log; - struct sptlrpc_rule_set tmp_rset; - - log = sptlrpc_conf_log_extract(cfg); - if (IS_ERR(log)) { - rc = PTR_ERR(log); - break; - } - - sptlrpc_rule_set_init(&tmp_rset); - - rc = sptlrpc_rule_set_from_log(&tmp_rset, log); - if (rc) { - CERROR("mdt %p: failed get sptlrpc rules: %d\n", m, rc); - break; - } - - write_lock(&m->mdt_sptlrpc_lock); - sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset); - m->mdt_sptlrpc_rset = tmp_rset; - write_unlock(&m->mdt_sptlrpc_lock); - - sptlrpc_target_update_exp_flavor( - md2lu_dev(&m->mdt_md_dev)->ld_obd, &tmp_rset); - - break; - } case LCFG_PARAM: { struct lprocfs_static_vars lvars; struct obd_device *obd = d->ld_obd; + /* + * For interoperability between 1.8 and 2.0, + * skip old "mdt.group_upcall" param. + */ + { + char *param = lustre_cfg_string(cfg, 1); + if (param && !strncmp("mdt.group_upcall", param, 16)) { + CWARN("For 1.8 interoperability, skip this" + " mdt.group_upcall. It is obsolete\n"); + break; + } + } + lprocfs_mdt_init_vars(&lvars); rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd); - if (rc == -ENOSYS) + if (rc > 0 || rc == -ENOSYS) /* we don't understand; pass it on */ rc = next->ld_ops->ldo_process_config(env, next, cfg); break; @@ -4187,13 +4792,14 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env, lu_object_init(o, h, d); lu_object_add_top(h, o); o->lo_ops = &mdt_obj_ops; + cfs_sema_init(&mo->mot_ioepoch_sem, 1); RETURN(o); } else RETURN(NULL); } static int mdt_object_init(const struct lu_env *env, struct lu_object *o, - const struct lu_object_conf *_) + const struct lu_object_conf *unused) { struct mdt_device *d = mdt_dev(o->lo_dev); struct lu_device *under; @@ -4233,27 +4839,48 @@ static void mdt_object_free(const struct lu_env *env, struct lu_object *o) static int mdt_object_print(const struct lu_env *env, void *cookie, lu_printer_t p, const struct lu_object *o) { - return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p", o); + struct mdt_object *mdto = mdt_obj((struct lu_object *)o); + return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch=%llu " + "flags=%llx, epochcount=%d, writecount=%d)", + mdto, mdto->mot_ioepoch, mdto->mot_flags, + mdto->mot_ioepoch_count, mdto->mot_writecount); } -static struct lu_device_operations mdt_lu_ops = { +static const struct lu_device_operations mdt_lu_ops = { .ldo_object_alloc = mdt_object_alloc, - .ldo_process_config = mdt_process_config + .ldo_process_config = mdt_process_config, }; -static struct lu_object_operations mdt_obj_ops = { +static const struct lu_object_operations mdt_obj_ops = { .loo_object_init = mdt_object_init, .loo_object_free = mdt_object_free, .loo_object_print = mdt_object_print }; +static int mdt_obd_set_info_async(struct obd_export *exp, + __u32 keylen, void *key, + __u32 vallen, void *val, + struct ptlrpc_request_set *set) +{ + struct obd_device *obd = exp->exp_obd; + int rc; + ENTRY; + + LASSERT(obd); + + if (KEY_IS(KEY_SPTLRPC_CONF)) { + rc = mdt_adapt_sptlrpc_conf(obd, 0); + RETURN(rc); + } + + RETURN(0); +} + /* mds_connect_internal */ static int mdt_connect_internal(struct obd_export *exp, struct mdt_device *mdt, struct obd_connect_data *data) { - __u64 flags; - if (data != NULL) { data->ocd_connect_flags &= MDT_CONNECT_SUPPORTED; data->ocd_ibits_known &= MDS_INODELOCK_FULL; @@ -4271,15 +4898,12 @@ static int mdt_connect_internal(struct obd_export *exp, if (!mdt->mdt_opts.mo_user_xattr) data->ocd_connect_flags &= ~OBD_CONNECT_XATTR; - if (!mdt->mdt_opts.mo_mds_capa) - data->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA; - - if (!mdt->mdt_opts.mo_oss_capa) - data->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA; + if (!mdt->mdt_som_conf) + data->ocd_connect_flags &= ~OBD_CONNECT_SOM; - spin_lock(&exp->exp_lock); + cfs_spin_lock(&exp->exp_lock); exp->exp_connect_flags = data->ocd_connect_flags; - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&exp->exp_lock); data->ocd_version = LUSTRE_VERSION_CODE; exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known; } @@ -4293,88 +4917,102 @@ static int mdt_connect_internal(struct obd_export *exp, } #endif - flags = OBD_CONNECT_LCL_CLIENT | OBD_CONNECT_RMT_CLIENT; - if ((exp->exp_connect_flags & flags) == flags) { - CWARN("%s: both local and remote client flags are set\n", + if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) { + CWARN("%s: MDS requires FID support, but client not\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name); return -EBADE; } - if (mdt->mdt_opts.mo_mds_capa && - ((exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA) == 0)) { - CWARN("%s: MDS requires capability support, but client not\n", - mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name); + if (mdt->mdt_som_conf && !exp_connect_som(exp) && + !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) { + CWARN("%s: MDS has SOM enabled, but client does not support " + "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name); return -EBADE; } - if (mdt->mdt_opts.mo_oss_capa && - ((exp->exp_connect_flags & OBD_CONNECT_OSS_CAPA) == 0)) { - CWARN("%s: MDS requires OSS capability support, " - "but client not\n", - mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name); - return -EBADE; - } + return 0; +} - if ((exp->exp_connect_flags & OBD_CONNECT_FID) == 0) { - CWARN("%s: MDS requires FID support, but client not\n", - mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name); - return -EBADE; +static int mdt_connect_check_sptlrpc(struct mdt_device *mdt, + struct obd_export *exp, + struct ptlrpc_request *req) +{ + struct sptlrpc_flavor flvr; + int rc = 0; + + if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) { + cfs_read_lock(&mdt->mdt_sptlrpc_lock); + sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset, + req->rq_sp_from, + req->rq_peer.nid, + &flvr); + cfs_read_unlock(&mdt->mdt_sptlrpc_lock); + + cfs_spin_lock(&exp->exp_lock); + + exp->exp_sp_peer = req->rq_sp_from; + exp->exp_flvr = flvr; + + if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY && + exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) { + CERROR("unauthorized rpc flavor %x from %s, " + "expect %x\n", req->rq_flvr.sf_rpc, + libcfs_nid2str(req->rq_peer.nid), + exp->exp_flvr.sf_rpc); + rc = -EACCES; + } + + cfs_spin_unlock(&exp->exp_lock); + } else { + if (exp->exp_sp_peer != req->rq_sp_from) { + CERROR("RPC source %s doesn't match %s\n", + sptlrpc_part2name(req->rq_sp_from), + sptlrpc_part2name(exp->exp_sp_peer)); + rc = -EACCES; + } else { + rc = sptlrpc_target_export_check(exp, req); + } } - return 0; + return rc; } /* mds_connect copy */ static int mdt_obd_connect(const struct lu_env *env, - struct lustre_handle *conn, struct obd_device *obd, + struct obd_export **exp, struct obd_device *obd, struct obd_uuid *cluuid, struct obd_connect_data *data, void *localdata) { struct mdt_thread_info *info; struct lsd_client_data *lcd; - struct obd_export *exp; + struct obd_export *lexp; + struct lustre_handle conn = { 0 }; struct mdt_device *mdt; struct ptlrpc_request *req; int rc; ENTRY; LASSERT(env != NULL); - if (!conn || !obd || !cluuid) + if (!exp || !obd || !cluuid) RETURN(-EINVAL); info = lu_context_key_get(&env->le_ctx, &mdt_thread_key); req = info->mti_pill->rc_req; mdt = mdt_dev(obd->obd_lu_dev); - rc = class_connect(conn, obd, cluuid); + rc = class_connect(&conn, obd, cluuid); if (rc) RETURN(rc); - exp = class_conn2export(conn); - LASSERT(exp != NULL); - - CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from)); + lexp = class_conn2export(&conn); + LASSERT(lexp != NULL); - spin_lock(&exp->exp_lock); - exp->exp_sp_peer = req->rq_sp_from; - - read_lock(&mdt->mdt_sptlrpc_lock); - sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, exp->exp_sp_peer, - req->rq_peer.nid, &exp->exp_flvr); - read_unlock(&mdt->mdt_sptlrpc_lock); - - if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) { - CERROR("invalid rpc flavor %x, expect %x, from %s\n", - req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc, - libcfs_nid2str(req->rq_peer.nid)); - exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID; - spin_unlock(&exp->exp_lock); - RETURN(-EACCES); - } - spin_unlock(&exp->exp_lock); + rc = mdt_connect_check_sptlrpc(mdt, lexp, req); + if (rc) + GOTO(out, rc); - rc = mdt_connect_internal(exp, mdt, data); + rc = mdt_connect_internal(lexp, mdt, data); if (rc == 0) { OBD_ALLOC_PTR(lcd); if (lcd != NULL) { @@ -4382,22 +5020,22 @@ static int mdt_obd_connect(const struct lu_env *env, mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); LASSERT(mti != NULL); - mti->mti_exp = exp; + mti->mti_exp = lexp; memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid); - exp->exp_mdt_data.med_lcd = lcd; + lexp->exp_mdt_data.med_lcd = lcd; rc = mdt_client_new(env, mdt); - if (rc != 0) { - OBD_FREE_PTR(lcd); - exp->exp_mdt_data.med_lcd = NULL; - } - } else + if (rc == 0) + mdt_export_stats_init(obd, lexp, localdata); + } else { rc = -ENOMEM; + } } +out: if (rc != 0) - class_disconnect(exp); + class_disconnect(lexp); else - class_export_put(exp); + *exp = lexp; RETURN(rc); } @@ -4405,7 +5043,8 @@ static int mdt_obd_connect(const struct lu_env *env, static int mdt_obd_reconnect(const struct lu_env *env, struct obd_export *exp, struct obd_device *obd, struct obd_uuid *cluuid, - struct obd_connect_data *data) + struct obd_connect_data *data, + void *localdata) { struct mdt_thread_info *info; struct mdt_device *mdt; @@ -4420,66 +5059,109 @@ static int mdt_obd_reconnect(const struct lu_env *env, req = info->mti_pill->rc_req; mdt = mdt_dev(obd->obd_lu_dev); - CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from)); + rc = mdt_connect_check_sptlrpc(mdt, exp, req); + if (rc) + RETURN(rc); - spin_lock(&exp->exp_lock); - if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) { - exp->exp_sp_peer = req->rq_sp_from; + rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data); + if (rc == 0) + mdt_export_stats_init(obd, exp, localdata); - read_lock(&mdt->mdt_sptlrpc_lock); - sptlrpc_rule_set_choose(&mdt->mdt_sptlrpc_rset, - exp->exp_sp_peer, - req->rq_peer.nid, &exp->exp_flvr); - read_unlock(&mdt->mdt_sptlrpc_lock); - - if (exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) { - CERROR("invalid rpc flavor %x, expect %x, from %s\n", - req->rq_flvr.sf_rpc, exp->exp_flvr.sf_rpc, - libcfs_nid2str(req->rq_peer.nid)); - exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_INVALID; - spin_unlock(&exp->exp_lock); - RETURN(-EACCES); - } + RETURN(rc); +} +static int mdt_export_cleanup(struct obd_export *exp) +{ + struct mdt_export_data *med = &exp->exp_mdt_data; + struct obd_device *obd = exp->exp_obd; + struct mdt_device *mdt; + struct mdt_thread_info *info; + struct lu_env env; + CFS_LIST_HEAD(closing_list); + struct mdt_file_data *mfd, *n; + int rc = 0; + ENTRY; + + cfs_spin_lock(&med->med_open_lock); + while (!cfs_list_empty(&med->med_open_head)) { + cfs_list_t *tmp = med->med_open_head.next; + mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list); + + /* Remove mfd handle so it can't be found again. + * We are consuming the mfd_list reference here. */ + class_handle_unhash(&mfd->mfd_handle); + cfs_list_move_tail(&mfd->mfd_list, &closing_list); } - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&med->med_open_lock); + mdt = mdt_dev(obd->obd_lu_dev); + LASSERT(mdt != NULL); - rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data); + rc = lu_env_init(&env, LCT_MD_THREAD); + if (rc) + RETURN(rc); + + info = lu_context_key_get(&env.le_ctx, &mdt_thread_key); + LASSERT(info != NULL); + memset(info, 0, sizeof *info); + info->mti_env = &env; + info->mti_mdt = mdt; + info->mti_exp = exp; + + if (!cfs_list_empty(&closing_list)) { + struct md_attr *ma = &info->mti_attr; + int lmm_size; + int cookie_size; + + lmm_size = mdt->mdt_max_mdsize; + OBD_ALLOC(ma->ma_lmm, lmm_size); + if (ma->ma_lmm == NULL) + GOTO(out_lmm, rc = -ENOMEM); + + cookie_size = mdt->mdt_max_cookiesize; + OBD_ALLOC(ma->ma_cookie, cookie_size); + if (ma->ma_cookie == NULL) + GOTO(out_cookie, rc = -ENOMEM); + + /* Close any open files (which may also cause orphan unlinking). */ + cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) { + cfs_list_del_init(&mfd->mfd_list); + memset(&ma->ma_attr, 0, sizeof(ma->ma_attr)); + ma->ma_lmm_size = lmm_size; + ma->ma_cookie_size = cookie_size; + ma->ma_need = 0; + /* It is not for setattr, just tell MDD to send + * DESTROY RPC to OSS if needed */ + ma->ma_attr_flags = MDS_CLOSE_CLEANUP; + ma->ma_valid = MA_FLAGS; + mdt_mfd_close(info, mfd); + } + OBD_FREE(ma->ma_cookie, cookie_size); + ma->ma_cookie = NULL; +out_cookie: + OBD_FREE(ma->ma_lmm, lmm_size); + ma->ma_lmm = NULL; + } +out_lmm: + info->mti_mdt = NULL; + /* cleanup client slot early */ + mdt_client_del(&env, mdt); + lu_env_fini(&env); RETURN(rc); } static int mdt_obd_disconnect(struct obd_export *exp) { - struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev); int rc; ENTRY; LASSERT(exp); class_export_get(exp); - /* Disconnect early so that clients can't keep using export */ - rc = class_disconnect(exp); - if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL) - ldlm_cancel_locks_for_export(exp); - - /* release nid stat refererence */ - lprocfs_exp_cleanup(exp); - - /* complete all outstanding replies */ - spin_lock(&exp->exp_lock); - while (!list_empty(&exp->exp_outstanding_replies)) { - struct ptlrpc_reply_state *rs = - list_entry(exp->exp_outstanding_replies.next, - struct ptlrpc_reply_state, rs_exp_list); - struct ptlrpc_service *svc = rs->rs_service; - - spin_lock(&svc->srv_lock); - list_del_init(&rs->rs_exp_list); - ptlrpc_schedule_difficult_reply(rs); - spin_unlock(&svc->srv_lock); - } - spin_unlock(&exp->exp_lock); + rc = server_disconnect_export(exp); + if (rc != 0) + CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc); + rc = mdt_export_cleanup(exp); class_export_put(exp); RETURN(rc); } @@ -4492,120 +5174,64 @@ static int mdt_init_export(struct obd_export *exp) ENTRY; CFS_INIT_LIST_HEAD(&med->med_open_head); - spin_lock_init(&med->med_open_lock); - sema_init(&med->med_idmap_sem, 1); + cfs_spin_lock_init(&med->med_open_lock); + cfs_sema_init(&med->med_idmap_sem, 1); med->med_idmap = NULL; - spin_lock(&exp->exp_lock); + cfs_spin_lock(&exp->exp_lock); exp->exp_connecting = 1; - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&exp->exp_lock); rc = ldlm_init_export(exp); if (rc) CERROR("Error %d while initializing export\n", rc); RETURN(rc); } -static int mdt_destroy_export(struct obd_export *export) +static int mdt_destroy_export(struct obd_export *exp) { struct mdt_export_data *med; - struct obd_device *obd = export->exp_obd; - struct mdt_device *mdt; - struct mdt_thread_info *info; - struct lu_env env; - struct md_attr *ma; - int lmm_size; - int cookie_size; int rc = 0; ENTRY; - med = &export->exp_mdt_data; - if (med->med_rmtclient) - mdt_cleanup_idmap(med); + med = &exp->exp_mdt_data; + if (exp_connect_rmtclient(exp)) + mdt_cleanup_idmap(&exp->exp_mdt_data); - target_destroy_export(export); - ldlm_destroy_export(export); + target_destroy_export(exp); + ldlm_destroy_export(exp); - if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid)) + LASSERT(cfs_list_empty(&exp->exp_outstanding_replies)); + LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head)); + if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid)) RETURN(0); - mdt = mdt_dev(obd->obd_lu_dev); - LASSERT(mdt != NULL); - - rc = lu_env_init(&env, NULL, LCT_MD_THREAD); - if (rc) - RETURN(rc); - - info = lu_context_key_get(&env.le_ctx, &mdt_thread_key); - LASSERT(info != NULL); - memset(info, 0, sizeof *info); - info->mti_env = &env; - info->mti_mdt = mdt; - info->mti_exp = export; - - ma = &info->mti_attr; - lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize; - cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize; - OBD_ALLOC(ma->ma_lmm, lmm_size); - OBD_ALLOC(ma->ma_cookie, cookie_size); - - if (ma->ma_lmm == NULL || ma->ma_cookie == NULL) - GOTO(out, rc = -ENOMEM); - ma->ma_need = MA_LOV | MA_COOKIE; - ma->ma_valid = 0; - /* Close any open files (which may also cause orphan unlinking). */ - spin_lock(&med->med_open_lock); - while (!list_empty(&med->med_open_head)) { - struct list_head *tmp = med->med_open_head.next; - struct mdt_file_data *mfd = - list_entry(tmp, struct mdt_file_data, mfd_list); - - /* Remove mfd handle so it can't be found again. - * We are consuming the mfd_list reference here. */ - class_handle_unhash(&mfd->mfd_handle); - list_del_init(&mfd->mfd_list); - spin_unlock(&med->med_open_lock); - mdt_mfd_close(info, mfd); - /* TODO: if we close the unlinked file, - * we need to remove it's objects from OST */ - memset(&ma->ma_attr, 0, sizeof(ma->ma_attr)); - spin_lock(&med->med_open_lock); - ma->ma_lmm_size = lmm_size; - ma->ma_cookie_size = cookie_size; - ma->ma_need = MA_LOV | MA_COOKIE; - ma->ma_valid = 0; - } - spin_unlock(&med->med_open_lock); - info->mti_mdt = NULL; - mdt_client_del(&env, mdt); - - EXIT; -out: - if (lmm_size) { - OBD_FREE(ma->ma_lmm, lmm_size); - ma->ma_lmm = NULL; - } - if (cookie_size) { - OBD_FREE(ma->ma_cookie, cookie_size); - ma->ma_cookie = NULL; - } - lu_env_fini(&env); - - return rc; + lut_client_free(exp); + RETURN(rc); } static void mdt_allow_cli(struct mdt_device *m, unsigned int flag) { if (flag & CONFIG_LOG) - m->mdt_fl_cfglog = 1; + cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state); + + /* also notify active event */ if (flag & CONFIG_SYNC) - m->mdt_fl_synced = 1; + cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state); - if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */) + if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) && + cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) { + struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd; + /* Open for clients */ - m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0; + if (obd->obd_no_conn) { + cfs_spin_lock_bh(&obd->obd_processing_task_lock); + obd->obd_no_conn = 0; + cfs_spin_unlock_bh(&obd->obd_processing_task_lock); + } + } } static int mdt_upcall(const struct lu_env *env, struct md_device *md, - enum md_upcall_event ev) + enum md_upcall_event ev, void *data) { struct mdt_device *m = mdt_dev(&md->md_lu_dev); struct md_device *next = m->mdt_child; @@ -4621,6 +5247,8 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md, CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n", m->mdt_max_mdsize, m->mdt_max_cookiesize); mdt_allow_cli(m, CONFIG_SYNC); + if (data) + (*(__u64 *)data) = m->mdt_mount_count; break; case MD_NO_TRANS: mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); @@ -4629,8 +5257,15 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md, break; case MD_LOV_CONFIG: /* Check that MDT is not yet configured */ - LASSERT(!m->mdt_fl_cfglog); + LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state)); + break; +#ifdef HAVE_QUOTA_SUPPORT + case MD_LOV_QUOTA: + if (md->md_lu_dev.ld_obd->obd_recovering == 0 && + likely(md->md_lu_dev.ld_obd->obd_stopping == 0)) + next->md_ops->mdo_quota.mqo_recovery(env, next); break; +#endif default: CERROR("invalid event\n"); rc = -EINVAL; @@ -4643,11 +5278,21 @@ static int mdt_obd_notify(struct obd_device *host, struct obd_device *watched, enum obd_notify_event ev, void *data) { + struct mdt_device *mdt = mdt_dev(host->obd_lu_dev); +#ifdef HAVE_QUOTA_SUPPORT + struct md_device *next = mdt->mdt_child; +#endif ENTRY; switch (ev) { case OBD_NOTIFY_CONFIG: - mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned long)data); + mdt_allow_cli(mdt, (unsigned long)data); + +#ifdef HAVE_QUOTA_SUPPORT + /* quota_type has been processed, we can now handle + * incoming quota requests */ + next->md_ops->mdo_quota.mqo_notify(NULL, next); +#endif break; default: CDEBUG(D_INFO, "Unhandled notification %#x\n", ev); @@ -4655,18 +5300,190 @@ static int mdt_obd_notify(struct obd_device *host, RETURN(0); } +static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, + void *val, int vallen) +{ + struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev); + struct getinfo_fid2path *fpout, *fpin; + int rc = 0; + + fpin = key + cfs_size_round(sizeof(KEY_FID2PATH)); + fpout = val; + + if (ptlrpc_req_need_swab(info->mti_pill->rc_req)) + lustre_swab_fid2path(fpin); + + memcpy(fpout, fpin, sizeof(*fpin)); + if (fpout->gf_pathlen != vallen - sizeof(*fpin)) + RETURN(-EINVAL); + + rc = mdt_fid2path(info->mti_env, mdt, fpout); + RETURN(rc); +} + +static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt, + struct getinfo_fid2path *fp) +{ + struct mdt_object *obj; + int rc; + ENTRY; + + CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n", + PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno); + + if (!fid_is_sane(&fp->gf_fid)) + RETURN(-EINVAL); + + obj = mdt_object_find(env, mdt, &fp->gf_fid); + if (obj == NULL || IS_ERR(obj)) { + CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid), + PTR_ERR(obj)); + RETURN(-EINVAL); + } + + rc = lu_object_exists(&obj->mot_obj.mo_lu); + if (rc <= 0) { + if (rc == -1) + rc = -EREMOTE; + else + rc = -ENOENT; + mdt_object_put(env, obj); + CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n", + PFID(&fp->gf_fid), rc); + RETURN(rc); + } + + rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path, + fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno); + mdt_object_put(env, obj); + + RETURN(rc); +} + +static int mdt_get_info(struct mdt_thread_info *info) +{ + struct ptlrpc_request *req = mdt_info_req(info); + char *key; + int keylen; + __u32 *vallen; + void *valout; + int rc; + ENTRY; + + key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY); + if (key == NULL) { + CDEBUG(D_IOCTL, "No GETINFO key"); + RETURN(-EFAULT); + } + keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY, + RCL_CLIENT); + + vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN); + if (vallen == NULL) { + CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer"); + RETURN(-EFAULT); + } + + req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER, + *vallen); + rc = req_capsule_server_pack(info->mti_pill); + valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL); + if (valout == NULL) { + CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer"); + RETURN(-EFAULT); + } + + if (KEY_IS(KEY_FID2PATH)) + rc = mdt_rpc_fid2path(info, key, valout, *vallen); + else + rc = -EINVAL; + + lustre_msg_set_status(req->rq_repmsg, rc); + + RETURN(rc); +} + +/* Pass the ioc down */ +static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt, + unsigned int cmd, int len, void *data) +{ + struct lu_context ioctl_session; + struct md_device *next = mdt->mdt_child; + int rc; + ENTRY; + + rc = lu_context_init(&ioctl_session, LCT_SESSION); + if (rc) + RETURN(rc); + ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current(); + lu_context_enter(&ioctl_session); + env->le_ses = &ioctl_session; + + LASSERT(next->md_ops->mdo_iocontrol); + rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data); + + lu_context_exit(&ioctl_session); + lu_context_fini(&ioctl_session); + RETURN(rc); +} + +static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg) +{ + struct obd_ioctl_data *data = karg; + struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1; + __u64 version; + struct mdt_object *obj; + struct mdt_lock_handle *lh; + int rc; + ENTRY; + CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid)); + if (!fid_is_sane(fid)) + RETURN(-EINVAL); + + lh = &mti->mti_lh[MDT_LH_PARENT]; + mdt_lock_reg_init(lh, LCK_CR); + + obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE); + if (IS_ERR(obj)) + RETURN(PTR_ERR(obj)); + + rc = mdt_object_exists(obj); + if (rc < 0) { + rc = -EREMOTE; + /** + * before calling version get the correct MDS should be + * fid, this is error to find remote object here + */ + CERROR("nonlocal object "DFID"\n", PFID(fid)); + } else if (rc == 0) { + rc = -ENOENT; + CDEBUG(D_IOCTL, "no such object: "DFID"\n", PFID(fid)); + } else { + version = mo_version_get(mti->mti_env, mdt_object_child(obj)); + if (version < 0) { + rc = (int)version; + } else { + *(__u64 *)data->ioc_inlbuf2 = version; + rc = 0; + } + } + mdt_object_unlock_put(mti, obj, lh, 1); + RETURN(rc); +} + +/* ioctls on obd dev */ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void *uarg) { struct lu_env env; - struct obd_device *obd= exp->exp_obd; + struct obd_device *obd = exp->exp_obd; struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev); struct dt_device *dt = mdt->mdt_bottom; int rc; ENTRY; CDEBUG(D_IOCTL, "handling ioctl cmd %#x\n", cmd); - rc = lu_env_init(&env, NULL, LCT_MD_THREAD); + rc = lu_env_init(&env, LCT_MD_THREAD); if (rc) RETURN(rc); @@ -4675,7 +5492,6 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, rc = mdt_device_sync(&env, mdt); break; case OBD_IOC_SET_READONLY: - rc = dt->dd_ops->dt_sync(&env, dt); dt->dd_ops->dt_ro(&env, dt); break; case OBD_IOC_ABORT_RECOVERY: @@ -4683,6 +5499,22 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, target_stop_recovery_thread(obd); rc = 0; break; + case OBD_IOC_CHANGELOG_REG: + case OBD_IOC_CHANGELOG_DEREG: + case OBD_IOC_CHANGELOG_CLEAR: + rc = mdt_ioc_child(&env, mdt, cmd, len, karg); + break; + case OBD_IOC_GET_OBJ_VERSION: { + struct mdt_thread_info *mti; + mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key); + memset(mti, 0, sizeof *mti); + mti->mti_env = &env; + mti->mti_mdt = mdt; + mti->mti_exp = exp; + + rc = mdt_ioc_version_get(mti, karg); + break; + } default: CERROR("Not supported cmd = %d for device %s\n", cmd, obd->obd_name); @@ -4696,7 +5528,10 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) { struct lu_device *ld = md2lu_dev(mdt->mdt_child); - struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd; + struct obd_device *obd = mdt2obd_dev(mdt); +#ifdef HAVE_QUOTA_SUPPORT + struct md_device *next = mdt->mdt_child; +#endif int rc, lost; ENTRY; /* if some clients didn't participate in recovery then we can possibly @@ -4705,6 +5540,10 @@ int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) mdt_seq_adjust(env, mdt, lost); rc = ld->ld_ops->ldo_recovery_complete(env, ld); +#ifdef HAVE_QUOTA_SUPPORT + if (likely(obd->obd_stopping == 0)) + next->md_ops->mdo_quota.mqo_recovery(env, next); +#endif RETURN(rc); } @@ -4713,7 +5552,7 @@ int mdt_obd_postrecov(struct obd_device *obd) struct lu_env env; int rc; - rc = lu_env_init(&env, NULL, LCT_MD_THREAD); + rc = lu_env_init(&env, LCT_MD_THREAD); if (rc) RETURN(rc); rc = mdt_postrecov(&env, mdt_dev(obd->obd_lu_dev)); @@ -4721,8 +5560,65 @@ int mdt_obd_postrecov(struct obd_device *obd) return rc; } +/** + * Send a copytool req to a client + * Note this sends a request RPC from a server (MDT) to a client (MDC), + * backwards of normal comms. + */ +int mdt_hsm_copytool_send(struct obd_export *exp) +{ + struct kuc_hdr *lh; + struct hsm_action_list *hal; + struct hsm_action_item *hai; + int rc, len; + ENTRY; + + CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name, + libcfs_nid2str(exp->exp_connection->c_peer.nid)); + + len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN + + /* for mockup below */ 2 * cfs_size_round(sizeof(*hai)); + OBD_ALLOC(lh, len); + if (lh == NULL) + RETURN(-ENOMEM); + + lh->kuc_magic = KUC_MAGIC; + lh->kuc_transport = KUC_TRANSPORT_HSM; + lh->kuc_msgtype = HMT_ACTION_LIST; + lh->kuc_msglen = len; + + hal = (struct hsm_action_list *)(lh + 1); + hal->hal_version = HAL_VERSION; + hal->hal_archive_num = 1; + obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name, + MTI_NAME_MAXLEN); + + /* mock up an action list */ + hal->hal_count = 2; + hai = hai_zero(hal); + hai->hai_action = HSMA_ARCHIVE; + hai->hai_fid.f_oid = 0xA00A; + hai->hai_len = sizeof(*hai); + hai = hai_next(hai); + hai->hai_action = HSMA_RESTORE; + hai->hai_fid.f_oid = 0xB00B; + hai->hai_len = sizeof(*hai); + + /* Uses the ldlm reverse import; this rpc will be seen by + the ldlm_callback_handler */ + rc = target_set_info_rpc(exp->exp_imp_reverse, LDLM_SET_INFO, + sizeof(KEY_HSM_COPYTOOL_SEND), + KEY_HSM_COPYTOOL_SEND, + len, lh, NULL); + + OBD_FREE(lh, len); + + RETURN(rc); +} + static struct obd_ops mdt_obd_device_ops = { .o_owner = THIS_MODULE, + .o_set_info_async = mdt_obd_set_info_async, .o_connect = mdt_obd_connect, .o_reconnect = mdt_obd_reconnect, .o_disconnect = mdt_obd_disconnect, @@ -4749,6 +5645,7 @@ static struct lu_device *mdt_device_free(const struct lu_env *env, struct mdt_device *m = mdt_dev(d); ENTRY; + md_device_fini(&m->mdt_md_dev); OBD_FREE_PTR(m); RETURN(NULL); } @@ -4767,7 +5664,7 @@ static struct lu_device *mdt_device_alloc(const struct lu_env *env, l = &m->mdt_md_dev.md_lu_dev; rc = mdt_init0(env, m, t, cfg); if (rc != 0) { - OBD_FREE_PTR(m); + mdt_device_free(env, l); l = ERR_PTR(rc); return l; } @@ -4797,6 +5694,42 @@ struct md_ucred *mdt_ucred(const struct mdt_thread_info *info) return md_ucred(info->mti_env); } +/** + * Enable/disable COS (Commit On Sharing). + * + * Set/Clear the COS flag in mdt options. + * + * \param mdt mdt device + * \param val 0 disables COS, other values enable COS + */ +void mdt_enable_cos(struct mdt_device *mdt, int val) +{ + struct lu_env env; + int rc; + + mdt->mdt_opts.mo_cos = !!val; + rc = lu_env_init(&env, LCT_MD_THREAD); + if (unlikely(rc != 0)) { + CWARN("lu_env initialization failed with rc = %d," + "cannot sync\n", rc); + return; + } + mdt_device_sync(&env, mdt); + lu_env_fini(&env); +} + +/** + * Check COS (Commit On Sharing) status. + * + * Return COS flag status. + * + * \param mdt mdt device + */ +int mdt_cos_is_enabled(struct mdt_device *mdt) +{ + return mdt->mdt_opts.mo_cos != 0; +} + /* type constructor/destructor: mdt_type_init, mdt_type_fini */ LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key); @@ -4804,6 +5737,9 @@ static struct lu_device_type_operations mdt_device_type_ops = { .ldto_init = mdt_type_init, .ldto_fini = mdt_type_fini, + .ldto_start = mdt_type_start, + .ldto_stop = mdt_type_stop, + .ldto_device_alloc = mdt_device_alloc, .ldto_device_free = mdt_device_free, .ldto_device_fini = mdt_device_fini @@ -4816,12 +5752,32 @@ static struct lu_device_type mdt_device_type = { .ldt_ctx_tags = LCT_MD_THREAD }; +static struct lu_local_obj_desc mdt_last_recv = { + .llod_name = LAST_RCVD, + .llod_oid = MDT_LAST_RECV_OID, + .llod_is_index = 0, +}; + static int __init mdt_mod_init(void) { struct lprocfs_static_vars lvars; int rc; - mdt_num_threads = MDT_NUM_THREADS; + llo_local_obj_register(&mdt_last_recv); + + if (mdt_num_threads > 0) { + if (mdt_num_threads > MDT_MAX_THREADS) + mdt_num_threads = MDT_MAX_THREADS; + if (mdt_num_threads < MDT_MIN_THREADS) + mdt_num_threads = MDT_MIN_THREADS; + mdt_max_threads = mdt_min_threads = mdt_num_threads; + } else { + mdt_max_threads = MDT_MAX_THREADS; + mdt_min_threads = MDT_MIN_THREADS; + if (mdt_min_threads < MDT_NUM_THREADS) + mdt_min_threads = MDT_NUM_THREADS; + } + lprocfs_mdt_init_vars(&lvars); rc = class_register_type(&mdt_obd_device_ops, NULL, lvars.module_vars, LUSTRE_MDT_NAME, @@ -4832,6 +5788,7 @@ static int __init mdt_mod_init(void) static void __exit mdt_mod_exit(void) { + llo_local_obj_unregister(&mdt_last_recv); class_unregister_type(LUSTRE_MDT_NAME); } @@ -4874,7 +5831,9 @@ static void __exit mdt_mod_exit(void) static struct mdt_handler mdt_mds_ops[] = { DEF_MDT_HNDL_F(0, CONNECT, mdt_connect), DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect), -DEF_MDT_HNDL_F(0, SET_INFO, mdt_set_info), +DEF_MDT_HNDL (0, SET_INFO, mdt_set_info, + &RQF_OBD_SET_INFO), +DEF_MDT_HNDL_F(0, GET_INFO, mdt_get_info), DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus), DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr), DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name), @@ -4886,8 +5845,10 @@ DEF_MDT_HNDL_F(HABEO_CORPUS, DONE_WRITING, mdt_done_writing), DEF_MDT_HNDL_F(0 |HABEO_REFERO, PIN, mdt_pin), DEF_MDT_HNDL_0(0, SYNC, mdt_sync), DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, IS_SUBDIR, mdt_is_subdir), +#ifdef HAVE_QUOTA_SUPPORT DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle), DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle) +#endif }; #define DEF_OBD_HNDL(flags, name, fn) \ @@ -4912,7 +5873,19 @@ static struct mdt_handler mdt_dlm_ops[] = { DEF_DLM_HNDL_0(0, CP_CALLBACK, mdt_cp_callback) }; +#define DEF_LLOG_HNDL(flags, name, fn) \ + DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL) + static struct mdt_handler mdt_llog_ops[] = { + DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE, mdt_llog_create), + DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK, mdt_llog_next_block), + DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header), + DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC, NULL), + DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE, NULL), + DEF_LLOG_HNDL(0, ORIGIN_CONNECT, NULL), + DEF_LLOG_HNDL(0, CATINFO, NULL), + DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK, mdt_llog_prev_block), + DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY, mdt_llog_destroy), }; #define DEF_SEC_CTX_HNDL(name, fn) \