X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=de3166d0848c34c27ae743ba8eee026e8e0178fa;hb=9e29c4deb8175d47a053b5e24258a1b4b4d9fa50;hp=18bec0dc40a4980c11be102f170148b365c05e01;hpb=8c82ab5cd92ee70a4cda2fe6437861e56f3fe3d5;p=fs%2Flustre-release.git diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 18bec0d..de3166d 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -40,7 +40,6 @@ #include /* * struct OBD_{ALLOC,FREE}*() - * MDT_FAIL_CHECK */ #include /* struct ptlrpc_request */ @@ -286,10 +285,10 @@ static int mdt_getstatus(struct mdt_thread_info *info) if (rc) RETURN(err_serious(rc)); - if (MDT_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETSTATUS_PACK)) RETURN(err_serious(-ENOMEM)); - repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY); + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); rc = next->md_ops->mdo_root_get(info->mti_env, next, &repbody->fid1); if (rc != 0) RETURN(rc); @@ -304,7 +303,7 @@ static int mdt_getstatus(struct mdt_thread_info *info) if (IS_ERR(root)) RETURN(PTR_ERR(root)); - capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1); + capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA1); LASSERT(capa); capa->lc_opc = CAPA_OPC_MDS_DEFAULT; @@ -320,24 +319,28 @@ static int mdt_getstatus(struct mdt_thread_info *info) static int mdt_statfs(struct mdt_thread_info *info) { - struct md_device *next = info->mti_mdt->mdt_child; - struct obd_statfs *osfs; - int rc; + struct md_device *next = info->mti_mdt->mdt_child; + struct ptlrpc_service *svc; + struct obd_statfs *osfs; + int rc; ENTRY; + svc = info->mti_pill->rc_req->rq_rqbd->rqbd_service; + /* This will trigger a watchdog timeout */ OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP, - (MDT_SERVICE_WATCHDOG_TIMEOUT / 1000) + 1); + (MDT_SERVICE_WATCHDOG_FACTOR * + at_get(&svc->srv_at_estimate) / 1000) + 1); rc = mdt_check_ucred(info); if (rc) RETURN(err_serious(rc)); - if (MDT_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) { + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_STATFS_PACK)) { rc = err_serious(-ENOMEM); } else { - osfs = req_capsule_server_get(&info->mti_pill,&RMF_OBD_STATFS); + osfs = req_capsule_server_get(info->mti_pill, &RMF_OBD_STATFS); rc = next->md_ops->mdo_statfs(info->mti_env, next, &info->mti_u.ksfs); statfs_pack(osfs, &info->mti_u.ksfs); @@ -349,8 +352,8 @@ void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o) { struct mdt_body *b; struct lu_attr *attr = &info->mti_attr.ma_attr; - - b = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY); + + 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) && @@ -388,7 +391,13 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, if (fid) { b->fid1 = *fid; b->valid |= OBD_MD_FLID; - CDEBUG(D_INODE, ""DFID": nlink=%d, mode=%o, size="LPU64"\n", + + /* FIXME: these should be fixed when new igif ready.*/ + b->ino = fid_oid(fid); /* 1.6 compatibility */ + b->generation = fid_ver(fid); /* 1.6 compatibility */ + b->valid |= OBD_MD_FLGENER; /* 1.6 compatibility */ + + CDEBUG(D_INODE, DFID": nlink=%d, mode=%o, size="LPU64"\n", PFID(fid), b->nlink, b->mode, b->size); } @@ -412,14 +421,14 @@ static int mdt_getattr_internal(struct mdt_thread_info *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; + struct req_capsule *pill = info->mti_pill; const struct lu_env *env = info->mti_env; struct mdt_body *repbody; struct lu_buf *buffer = &info->mti_buf; int rc; ENTRY; - if (unlikely(MDT_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK))) + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_GETATTR_PACK)) RETURN(err_serious(-ENOMEM)); repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); @@ -450,6 +459,12 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, ma->ma_need = MA_LOV | MA_INODE; } + if (S_ISDIR(lu_object_attr(&next->mo_lu)) && + reqbody->valid & OBD_MD_FLDIREA && + lustre_msg_get_opc(req->rq_reqmsg) == MDS_GETATTR) { + /* get default stripe info for this dir. */ + ma->ma_need |= MA_LOV_DEF; + } rc = mo_attr_get(env, next, ma); if (unlikely(rc)) { CERROR("getattr error for "DFID": %d\n", @@ -554,7 +569,7 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, info->mti_mdt->mdt_opts.mo_mds_capa) { struct lustre_capa *capa; - capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1); + capa = req_capsule_server_get(pill, &RMF_CAPA1); LASSERT(capa); capa->lc_opc = CAPA_OPC_MDS_DEFAULT; rc = mo_capa_get(env, next, capa, 0); @@ -581,27 +596,26 @@ static int mdt_renew_capa(struct mdt_thread_info *info) if (!obj || !mdt->mdt_opts.mo_mds_capa) RETURN(0); - body = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY); + body = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(body != NULL); - c = req_capsule_client_get(&info->mti_pill, &RMF_CAPA1); + c = req_capsule_client_get(info->mti_pill, &RMF_CAPA1); LASSERT(c); - capa = req_capsule_server_get(&info->mti_pill, &RMF_CAPA1); + capa = req_capsule_server_get(info->mti_pill, &RMF_CAPA2); LASSERT(capa); *capa = *c; rc = mo_capa_get(info->mti_env, mdt_object_child(obj), capa, 1); if (rc == 0) body->valid |= OBD_MD_FLOSSCAPA; - RETURN(rc); } static int mdt_getattr(struct mdt_thread_info *info) { struct mdt_object *obj = info->mti_object; - struct req_capsule *pill = &info->mti_pill; + struct req_capsule *pill = info->mti_pill; struct mdt_body *reqbody; struct mdt_body *repbody; mode_t mode; @@ -613,14 +627,11 @@ static int mdt_getattr(struct mdt_thread_info *info) LASSERT(reqbody); if (reqbody->valid & OBD_MD_FLOSSCAPA) { - rc = req_capsule_pack(pill); + rc = req_capsule_server_pack(pill); if (unlikely(rc)) - rc = err_serious(rc); - else { - rc = mdt_renew_capa(info); - mdt_shrink_reply(info); - } - GOTO(out, rc); + RETURN(err_serious(rc)); + rc = mdt_renew_capa(info); + GOTO(out_shrink, rc); } LASSERT(obj != NULL); @@ -635,9 +646,9 @@ static int mdt_getattr(struct mdt_thread_info *info) req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER, md_size); - rc = req_capsule_pack(pill); + rc = req_capsule_server_pack(pill); if (unlikely(rc != 0)) - GOTO(out, rc = err_serious(rc)); + RETURN(err_serious(rc)); repbody = req_capsule_server_get(pill, &RMF_MDT_BODY); LASSERT(repbody != NULL); @@ -665,14 +676,13 @@ static int mdt_getattr(struct mdt_thread_info *info) EXIT; out_shrink: mdt_shrink_reply(info); -out: return rc; } static int mdt_is_subdir(struct mdt_thread_info *info) { struct mdt_object *o = info->mti_object; - struct req_capsule *pill = &info->mti_pill; + struct req_capsule *pill = info->mti_pill; const struct mdt_body *body = info->mti_body; struct mdt_body *repbody; int rc; @@ -725,11 +735,10 @@ static int mdt_raw_lookup(struct mdt_thread_info *info, } else mdt_set_disposition(info, ldlm_rep, DISP_LOOKUP_POS); - repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY); + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); #endif if (rc == 0) { - repbody = req_capsule_server_get(&info->mti_pill, - &RMF_MDT_BODY); + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); repbody->fid1 = *child_fid; repbody->valid = OBD_MD_FLID; } @@ -769,11 +778,11 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)); LASSERT(parent != NULL); - name = req_capsule_client_get(&info->mti_pill, &RMF_NAME); + name = req_capsule_client_get(info->mti_pill, &RMF_NAME); if (name == NULL) RETURN(err_serious(-EFAULT)); - namelen = req_capsule_get_size(&info->mti_pill, &RMF_NAME, + namelen = req_capsule_get_size(info->mti_pill, &RMF_NAME, RCL_CLIENT) - 1; LASSERT(namelen >= 0); @@ -781,7 +790,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, * 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, &RMF_MDT_BODY); + reqbody =req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(fid_is_sane(&reqbody->fid2)); name = NULL; @@ -855,7 +864,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, if (unlikely(rc != 0)) mdt_object_unlock(info, child, lhc, 1); } - GOTO(out, rc); + RETURN(rc); } /* step 1: lock parent */ @@ -918,6 +927,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, LDLM_LOCK_PUT(lock); rc = 0; } else { + struct md_attr *ma = &info->mti_attr; relock: mdt_lock_handle_init(lhc); mdt_lock_reg_init(lhc, LCK_PR); @@ -927,6 +937,23 @@ relock: &child->mot_obj.mo_lu, "Object doesn't exist!\n"); } + + ma->ma_valid = 0; + ma->ma_need = MA_INODE; + rc = mo_attr_get(info->mti_env, next, ma); + if (unlikely(rc != 0)) + GOTO(out_child, rc); + + /* If the file has not been changed for some time, we return + * not only a LOOKUP lock, but also an UPDATE lock and this + * might save us RPC on later STAT. For directories, it also + * let negative dentry starts working for this dir. */ + if (ma->ma_valid & MA_INODE && + ma->ma_attr.la_valid & LA_CTIME && + info->mti_mdt->mdt_namespace->ns_ctime_age_limit + + ma->ma_attr.la_ctime < cfs_time_current_sec()) + child_bits |= MDS_INODELOCK_UPDATE; + rc = mdt_object_lock(info, child, lhc, child_bits, MDT_CROSS_LOCK); @@ -959,7 +986,7 @@ relock: * Pack Size-on-MDS inode attributes to the body if * update lock is given. */ - repbody = req_capsule_server_get(&info->mti_pill, + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); ma = &info->mti_attr.ma_attr; if (lock->l_policy_data.l_inodebits.bits & @@ -973,7 +1000,6 @@ out_child: mdt_object_put(info->mti_env, child); out_parent: mdt_object_unlock(info, parent, lhp, 1); -out: return rc; } @@ -986,9 +1012,9 @@ static int mdt_getattr_name(struct mdt_thread_info *info) int rc; ENTRY; - reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY); + reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(reqbody != NULL); - repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY); + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(repbody != NULL); info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT); @@ -998,7 +1024,7 @@ static int mdt_getattr_name(struct mdt_thread_info *info) rc = mdt_init_ucred(info, reqbody); if (unlikely(rc)) - GOTO(out, rc); + GOTO(out_shrink, rc); rc = mdt_getattr_name_lock(info, lhc, MDS_INODELOCK_UPDATE, NULL); if (lustre_handle_is_used(&lhc->mlh_reg_lh)) { @@ -1007,7 +1033,7 @@ static int mdt_getattr_name(struct mdt_thread_info *info) } mdt_exit_ucred(info); EXIT; -out: +out_shrink: mdt_shrink_reply(info); return rc; } @@ -1027,20 +1053,20 @@ static int mdt_set_info(struct mdt_thread_info *info) int keylen, rc = 0; ENTRY; - rc = lustre_pack_reply(req, 1, NULL, NULL); + rc = req_capsule_server_pack(info->mti_pill); if (rc) RETURN(rc); - key = req_capsule_client_get(&info->mti_pill, &RMF_SETINFO_KEY); + key = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_KEY); if (key == NULL) { DEBUG_REQ(D_HA, req, "no set_info key"); RETURN(-EFAULT); } - keylen = req_capsule_get_size(&info->mti_pill, &RMF_SETINFO_KEY, + keylen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_KEY, RCL_CLIENT); - val = req_capsule_client_get(&info->mti_pill, &RMF_SETINFO_VAL); + val = req_capsule_client_get(info->mti_pill, &RMF_SETINFO_VAL); if (val == NULL) { DEBUG_REQ(D_HA, req, "no set_info val"); RETURN(-EFAULT); @@ -1101,6 +1127,7 @@ static int mdt_sendpage(struct mdt_thread_info *info, struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info; int tmpcount; int tmpsize; + int timeout; int i; int rc; ENTRY; @@ -1108,7 +1135,7 @@ static int mdt_sendpage(struct mdt_thread_info *info, desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, BULK_PUT_SOURCE, MDS_BULK_PORTAL); if (desc == NULL) - GOTO(out, rc = -ENOMEM); + RETURN(-ENOMEM); for (i = 0, tmpcount = rdpg->rp_count; i < rdpg->rp_npages; i++, tmpcount -= tmpsize) { @@ -1121,10 +1148,14 @@ static int mdt_sendpage(struct mdt_thread_info *info, if (rc) GOTO(free_desc, rc); - if (MDT_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) - GOTO(abort_bulk, rc); + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) + GOTO(abort_bulk, rc = 0); - *lwi = LWI_TIMEOUT(obd_timeout * HZ / 4, NULL, NULL); + timeout = (int) req->rq_deadline - cfs_time_current_sec(); + 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); LASSERT (rc == 0 || rc == -ETIMEDOUT); @@ -1149,7 +1180,6 @@ abort_bulk: ptlrpc_abort_bulk(desc); free_desc: ptlrpc_free_bulk(desc); -out: return rc; } @@ -1187,7 +1217,7 @@ static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page, ma->ma_attr.la_valid = LA_MODE; ma->ma_valid = MA_INODE; - kmap(page); + cfs_kmap(page); dp = page_address(page); offset = (int)((__u32)lu_dirent_start(dp) - (__u32)dp); @@ -1230,7 +1260,7 @@ static int mdt_write_dir_page(struct mdt_thread_info *info, struct page *page, } EXIT; out: - kunmap(page); + cfs_kunmap(page); return rc; } @@ -1254,7 +1284,7 @@ static int mdt_writepage(struct mdt_thread_info *info) ENTRY; - reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY); + reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY); if (reqbody == NULL) RETURN(err_serious(-EFAULT)); @@ -1263,7 +1293,7 @@ static int mdt_writepage(struct mdt_thread_info *info) RETURN(err_serious(-ENOMEM)); /* allocate the page for the desc */ - page = alloc_pages(GFP_KERNEL, 0); + page = cfs_alloc_page(CFS_ALLOC_STD); if (page == NULL) GOTO(desc_cleanup, rc = -ENOMEM); @@ -1317,7 +1347,7 @@ static int mdt_writepage(struct mdt_thread_info *info) cleanup_lwi: OBD_FREE_PTR(lwi); cleanup_page: - __free_pages(page, 0); + cfs_free_page(page); desc_cleanup: ptlrpc_free_bulk(desc); RETURN(rc); @@ -1334,18 +1364,14 @@ static int mdt_readpage(struct mdt_thread_info *info) int i; ENTRY; - if (MDT_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READPAGE_PACK)) RETURN(err_serious(-ENOMEM)); - reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY); - repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY); + reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY); + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); if (reqbody == NULL || repbody == NULL) RETURN(err_serious(-EFAULT)); - rc = mdt_check_ucred(info); - if (rc) - RETURN(err_serious(rc)); - /* * prepare @rdpg before calling lower layers and transfer itself. Here * reqbody->size contains offset of where to start to read and @@ -1364,7 +1390,7 @@ static int mdt_readpage(struct mdt_thread_info *info) RETURN(-ENOMEM); for (i = 0; i < rdpg->rp_npages; ++i) { - rdpg->rp_pages[i] = alloc_pages(GFP_KERNEL, 0); + rdpg->rp_pages[i] = cfs_alloc_page(CFS_ALLOC_STD); if (rdpg->rp_pages[i] == NULL) GOTO(free_rdpg, rc = -ENOMEM); } @@ -1382,10 +1408,11 @@ free_rdpg: for (i = 0; i < rdpg->rp_npages; i++) if (rdpg->rp_pages[i] != NULL) - __free_pages(rdpg->rp_pages[i], 0); + cfs_free_page(rdpg->rp_pages[i]); OBD_FREE(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]); - MDT_FAIL_RETURN(OBD_FAIL_MDS_SENDPAGE, 0); + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) + RETURN(0); return rc; } @@ -1394,26 +1421,21 @@ static int mdt_reint_internal(struct mdt_thread_info *info, struct mdt_lock_handle *lhc, __u32 op) { - struct req_capsule *pill = &info->mti_pill; + struct req_capsule *pill = info->mti_pill; struct mdt_device *mdt = info->mti_mdt; - struct ptlrpc_request *req = mdt_info_req(info); struct mdt_body *repbody; - int need_shrink = 0; - int rc; + int rc = 0; ENTRY; /* pack reply */ - if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) { + 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); - need_shrink = 1; - } - if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) { + if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER)) req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, mdt->mdt_max_cookiesize); - need_shrink = 1; - } - rc = req_capsule_pack(pill); + + rc = req_capsule_server_pack(pill); if (rc != 0) { CERROR("Can't pack response, rc %d\n", rc); RETURN(err_serious(rc)); @@ -1426,7 +1448,7 @@ static int mdt_reint_internal(struct mdt_thread_info *info, repbody->aclsize = 0; } - if (MDT_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNPACK)) GOTO(out_shrink, rc = err_serious(-EFAULT)); rc = mdt_reint_unpack(info, op); @@ -1443,44 +1465,32 @@ static int mdt_reint_internal(struct mdt_thread_info *info, if (rc != 0) GOTO(out_ucred, rc = err_serious(rc)); - if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) { - struct mdt_client_data *mcd; - - mcd = req->rq_export->exp_mdt_data.med_mcd; - if (req_xid_is_last(req)) { - need_shrink = 0; - mdt_reconstruct(info, lhc); - rc = lustre_msg_get_status(req->rq_repmsg); - GOTO(out_ucred, rc); - } - DEBUG_REQ(D_HA, req, "no reply for RESENT (xid "LPD64")", - mcd->mcd_last_xid); + if (mdt_check_resent(info, mdt_reconstruct, lhc)) { + rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg); + GOTO(out_ucred, rc); } - - need_shrink = 0; rc = mdt_reint_rec(info, lhc); EXIT; out_ucred: mdt_exit_ucred(info); out_shrink: - if (need_shrink) - mdt_shrink_reply(info); + mdt_shrink_reply(info); return rc; } static long mdt_reint_opcode(struct mdt_thread_info *info, const struct req_format **fmt) { - __u32 *ptr; + struct mdt_rec_reint *rec; long opc; opc = err_serious(-EFAULT); - ptr = req_capsule_client_get(&info->mti_pill, &RMF_REINT_OPC); - if (ptr != NULL) { - opc = *ptr; + rec = req_capsule_client_get(info->mti_pill, &RMF_REC_REINT); + if (rec != NULL) { + opc = rec->rr_opcode; DEBUG_REQ(D_INODE, mdt_info_req(info), "reint opt = %ld", opc); if (opc < REINT_MAX && fmt[opc] != NULL) - req_capsule_extend(&info->mti_pill, fmt[opc]); + req_capsule_extend(info->mti_pill, fmt[opc]); else { CERROR("Unsupported opc: %ld\n", opc); opc = err_serious(opc); @@ -1495,12 +1505,13 @@ static int mdt_reint(struct mdt_thread_info *info) int rc; static const struct req_format *reint_fmts[REINT_MAX] = { - [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR, - [REINT_CREATE] = &RQF_MDS_REINT_CREATE, - [REINT_LINK] = &RQF_MDS_REINT_LINK, - [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK, - [REINT_RENAME] = &RQF_MDS_REINT_RENAME, - [REINT_OPEN] = &RQF_MDS_REINT_OPEN + [REINT_SETATTR] = &RQF_MDS_REINT_SETATTR, + [REINT_CREATE] = &RQF_MDS_REINT_CREATE, + [REINT_LINK] = &RQF_MDS_REINT_LINK, + [REINT_UNLINK] = &RQF_MDS_REINT_UNLINK, + [REINT_RENAME] = &RQF_MDS_REINT_RENAME, + [REINT_OPEN] = &RQF_MDS_REINT_OPEN, + [REINT_SETXATTR] = &RQF_MDS_REINT_SETXATTR }; ENTRY; @@ -1536,7 +1547,7 @@ static int mdt_object_sync(struct mdt_thread_info *info) static int mdt_sync(struct mdt_thread_info *info) { - struct req_capsule *pill = &info->mti_pill; + struct req_capsule *pill = info->mti_pill; struct mdt_body *body; int rc; ENTRY; @@ -1548,12 +1559,12 @@ static int mdt_sync(struct mdt_thread_info *info) if (body == NULL) RETURN(err_serious(-EINVAL)); - if (MDT_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK)) + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SYNC_PACK)) RETURN(err_serious(-ENOMEM)); if (fid_seq(&body->fid1) == 0) { /* sync the whole device */ - rc = req_capsule_pack(pill); + rc = req_capsule_server_pack(pill); if (rc == 0) rc = mdt_device_sync(info); else @@ -1603,6 +1614,9 @@ static int mdt_obd_ping(struct mdt_thread_info *info) { int rc; ENTRY; + + req_capsule_set(info->mti_pill, &RQF_OBD_PING); + rc = target_handle_ping(mdt_info_req(info)); if (rc < 0) rc = err_serious(rc); @@ -1641,11 +1655,6 @@ static int mdt_enqueue(struct mdt_thread_info *info) */ LASSERT(info->mti_dlm_req != NULL); - if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE)) { - info->mti_fail_id = OBD_FAIL_LDLM_ENQUEUE; - return 0; - } - req = mdt_info_req(info); /* @@ -1654,7 +1663,9 @@ static int mdt_enqueue(struct mdt_thread_info *info) * bits get corrupted somewhere in mdt_intent_policy(). */ req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits; - LASSERT(req_bits != 0); + /* This is disabled because we need to support liblustre flock. + * LASSERT(req_bits != 0); + */ rc = ldlm_handle_enqueue0(info->mti_mdt->mdt_namespace, req, info->mti_dlm_req, &cbs); @@ -1705,6 +1716,8 @@ static int mdt_sec_ctx_handle(struct mdt_thread_info *info) sptlrpc_svc_ctx_invalidate(req); } + OBD_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, obd_fail_val); + return rc; } @@ -1722,6 +1735,7 @@ struct mdt_object *mdt_object_find(const struct lu_env *env, struct mdt_object *m; ENTRY; + CDEBUG(D_INFO, "Find object for "DFID"\n", PFID(f)); o = lu_object_find(env, d->mdt_md_dev.md_lu_dev.ld_site, f); if (unlikely(IS_ERR(o))) m = (struct mdt_object *)o; @@ -1932,17 +1946,21 @@ static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags) const struct lu_env *env; struct req_capsule *pill; int rc; + ENTRY; env = info->mti_env; - pill = &info->mti_pill; + pill = info->mti_pill; body = info->mti_body = req_capsule_client_get(pill, &RMF_MDT_BODY); if (body == NULL) - return -EFAULT; + RETURN(-EFAULT); + + if (!(body->valid & OBD_MD_FLID)) + RETURN(0); if (!fid_is_sane(&body->fid1)) { CERROR("Invalid fid: "DFID"\n", PFID(&body->fid1)); - return -EINVAL; + RETURN(-EINVAL); } /* @@ -1973,16 +1991,14 @@ static int mdt_body_unpack(struct mdt_thread_info *info, __u32 flags) } else rc = PTR_ERR(obj); - return rc; + RETURN(rc); } static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags) { - struct req_capsule *pill; + struct req_capsule *pill = info->mti_pill; int rc; - ENTRY; - pill = &info->mti_pill; if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) rc = mdt_body_unpack(info, flags); @@ -2000,7 +2016,7 @@ static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags) req_capsule_set_size(pill, &RMF_LOGCOOKIES, RCL_SERVER, mdt->mdt_max_cookiesize); - rc = req_capsule_pack(pill); + rc = req_capsule_server_pack(pill); } RETURN(rc); } @@ -2034,28 +2050,25 @@ static int mdt_req_handle(struct mdt_thread_info *info, LASSERT(current->journal_info == NULL); /* - * Do not use *_FAIL_CHECK_ONCE() macros, because they will stop - * correct handling of failed req later in ldlm due to doing - * obd_fail_loc |= OBD_FAIL_ONCE | OBD_FAILED without actually - * correct actions like it is done in target_send_reply_msg(). + * 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 + * is buggy! Handlers either pack reply or return error. + * + * We return 0 here and do not send any reply in order to emulate + * network failure. Do not send any reply in case any of NET related + * fail_id has occured. */ - if (h->mh_fail_id != 0) { - /* - * Set to info->mti_fail_id to handler fail_id, it will be used - * later, and better than use default fail_id. - */ - if (OBD_FAIL_CHECK(h->mh_fail_id)) { - info->mti_fail_id = h->mh_fail_id; - RETURN(0); - } - } + if (OBD_FAIL_CHECK_ORSET(h->mh_fail_id, OBD_FAIL_ONCE)) + RETURN(0); rc = 0; flags = h->mh_flags; LASSERT(ergo(flags & (HABEO_CORPUS|HABEO_REFERO), h->mh_fmt != NULL)); if (h->mh_fmt != NULL) { - req_capsule_set(&info->mti_pill, h->mh_fmt); + req_capsule_set(info->mti_pill, h->mh_fmt); rc = mdt_unpack_req_pack_rep(info, flags); } @@ -2069,14 +2082,13 @@ static int mdt_req_handle(struct mdt_thread_info *info, LASSERT(h->mh_fmt != NULL); - dlm_req = req_capsule_client_get(&info->mti_pill, &RMF_DLM_REQ); + dlm_req = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ); if (dlm_req != NULL) { if (info->mti_mdt->mdt_opts.mo_compat_resname) rc = mdt_lock_resname_compat(info->mti_mdt, dlm_req); info->mti_dlm_req = dlm_req; } else { - CERROR("Can't unpack dlm request\n"); rc = -EFAULT; } } @@ -2096,6 +2108,13 @@ static int mdt_req_handle(struct mdt_thread_info *info, * only */ rc = h->mh_act(info); + if (rc == 0 && + !req->rq_no_reply && req->rq_reply_state == NULL) { + DEBUG_REQ(D_ERROR, req, "MDT \"handler\" %s did not " + "pack reply and returned 0 error\n", + h->mh_name); + LBUG(); + } serious = is_serious(rc); rc = clear_serious(rc); } else @@ -2116,7 +2135,7 @@ static int mdt_req_handle(struct mdt_thread_info *info, info->mti_mdt->mdt_opts.mo_compat_resname) { struct ldlm_reply *dlmrep; - dlmrep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP); + dlmrep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); if (dlmrep != NULL) rc = mdt_lock_reply_compat(info->mti_mdt, dlmrep); } @@ -2131,7 +2150,8 @@ static int mdt_req_handle(struct mdt_thread_info *info, LBUG(); } - RETURN(rc); + target_send_reply(req, rc, info->mti_fail_id); + RETURN(0); } void mdt_lock_handle_init(struct mdt_lock_handle *lh) @@ -2160,11 +2180,8 @@ static void mdt_thread_info_init(struct ptlrpc_request *req, int i; struct md_capainfo *ci; - info->mti_rep_buf_nr = ARRAY_SIZE(info->mti_rep_buf_size); - for (i = 0; i < ARRAY_SIZE(info->mti_rep_buf_size); i++) - info->mti_rep_buf_size[i] = -1; - req_capsule_init(&info->mti_pill, req, RCL_SERVER, - info->mti_rep_buf_size); + req_capsule_init(&req->rq_pill, req, RCL_SERVER); + info->mti_pill = &req->rq_pill; /* lock handle */ for (i = 0; i < ARRAY_SIZE(info->mti_lh); i++) @@ -2174,7 +2191,7 @@ static void mdt_thread_info_init(struct ptlrpc_request *req, if (req->rq_export) { info->mti_mdt = mdt_dev(req->rq_export->exp_obd->obd_lu_dev); info->mti_exp = req->rq_export; - } else + } else info->mti_mdt = NULL; info->mti_env = req->rq_svc_thread->t_env; ci = md_capainfo(info->mti_env); @@ -2200,7 +2217,7 @@ static void mdt_thread_info_fini(struct mdt_thread_info *info) { int i; - req_capsule_fini(&info->mti_pill); + req_capsule_fini(info->mti_pill); if (info->mti_object != NULL) { mdt_object_put(info->mti_env, info->mti_object); info->mti_object = NULL; @@ -2210,9 +2227,35 @@ static void mdt_thread_info_fini(struct mdt_thread_info *info) info->mti_env = NULL; } -/* mds/handler.c */ -extern int mds_filter_recovery_request(struct ptlrpc_request *req, - struct obd_device *obd, int *process); +static int mdt_filter_recovery_request(struct ptlrpc_request *req, + struct obd_device *obd, int *process) +{ + switch (lustre_msg_get_opc(req->rq_reqmsg)) { + case MDS_CONNECT: /* This will never get here, but for completeness. */ + case OST_CONNECT: /* This will never get here, but for completeness. */ + case MDS_DISCONNECT: + case OST_DISCONNECT: + *process = 1; + RETURN(0); + + case MDS_CLOSE: + case MDS_DONE_WRITING: + case MDS_SYNC: /* used in unmounting */ + case OBD_PING: + case MDS_REINT: + case SEQ_QUERY: + case FLD_QUERY: + case LDLM_ENQUEUE: + *process = target_queue_recovery_request(req, obd); + RETURN(0); + + default: + DEBUG_REQ(D_ERROR, req, "not permitted during recovery"); + *process = -EAGAIN; + RETURN(0); + } +} + /* * Handle recovery. Return: * +1: continue request processing; @@ -2290,7 +2333,7 @@ static int mdt_recovery(struct mdt_thread_info *info) int rc; int should_process; DEBUG_REQ(D_INFO, req, "Got new replay"); - rc = mds_filter_recovery_request(req, obd, &should_process); + rc = mdt_filter_recovery_request(req, obd, &should_process); if (rc != 0 || !should_process) RETURN(rc); else if (should_process < 0) { @@ -2302,24 +2345,85 @@ static int mdt_recovery(struct mdt_thread_info *info) RETURN(+1); } -static int mdt_reply(struct ptlrpc_request *req, int rc, - struct mdt_thread_info *info) +static int mdt_msg_check_version(struct lustre_msg *msg) { - ENTRY; + int rc; -#if 0 - if (req->rq_reply_state == NULL && rc == 0) { - req->rq_status = rc; - lustre_pack_reply(req, 1, NULL, NULL); + switch (lustre_msg_get_opc(msg)) { + case MDS_CONNECT: + case MDS_DISCONNECT: + case OBD_PING: + case SEC_CTX_INIT: + case SEC_CTX_INIT_CONT: + case SEC_CTX_FINI: + rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION); + if (rc) + CERROR("bad opc %u version %08x, expecting %08x\n", + lustre_msg_get_opc(msg), + lustre_msg_get_version(msg), + LUSTRE_OBD_VERSION); + break; + case MDS_GETSTATUS: + case MDS_GETATTR: + case MDS_GETATTR_NAME: + case MDS_STATFS: + case MDS_READPAGE: + case MDS_WRITEPAGE: + case MDS_IS_SUBDIR: + case MDS_REINT: + case MDS_CLOSE: + case MDS_DONE_WRITING: + case MDS_PIN: + case MDS_SYNC: + case MDS_GETXATTR: + case MDS_SETXATTR: + case MDS_SET_INFO: + case MDS_QUOTACHECK: + case MDS_QUOTACTL: + case QUOTA_DQACQ: + case QUOTA_DQREL: + case SEQ_QUERY: + case FLD_QUERY: + rc = lustre_msg_check_version(msg, LUSTRE_MDS_VERSION); + if (rc) + CERROR("bad opc %u version %08x, expecting %08x\n", + lustre_msg_get_opc(msg), + lustre_msg_get_version(msg), + LUSTRE_MDS_VERSION); + break; + case LDLM_ENQUEUE: + case LDLM_CONVERT: + case LDLM_BL_CALLBACK: + case LDLM_CP_CALLBACK: + rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION); + if (rc) + CERROR("bad opc %u version %08x, expecting %08x\n", + lustre_msg_get_opc(msg), + lustre_msg_get_version(msg), + LUSTRE_DLM_VERSION); + break; + case OBD_LOG_CANCEL: + case LLOG_ORIGIN_HANDLE_CREATE: + case LLOG_ORIGIN_HANDLE_NEXT_BLOCK: + case LLOG_ORIGIN_HANDLE_READ_HEADER: + case LLOG_ORIGIN_HANDLE_CLOSE: + case LLOG_ORIGIN_HANDLE_DESTROY: + case LLOG_ORIGIN_HANDLE_PREV_BLOCK: + case LLOG_CATINFO: + rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION); + if (rc) + CERROR("bad opc %u version %08x, expecting %08x\n", + lustre_msg_get_opc(msg), + lustre_msg_get_version(msg), + LUSTRE_LOG_VERSION); + break; + default: + CERROR("MDS unknown opcode %d\n", lustre_msg_get_opc(msg)); + rc = -ENOTSUPP; } -#endif - target_send_reply(req, rc, info->mti_fail_id); - RETURN(0); + return rc; } -/* mds/handler.c */ -extern int mds_msg_check_version(struct lustre_msg *msg); - static int mdt_handle0(struct ptlrpc_request *req, struct mdt_thread_info *info, struct mdt_opc_slice *supported) @@ -2330,12 +2434,13 @@ static int mdt_handle0(struct ptlrpc_request *req, ENTRY; - MDT_FAIL_RETURN(OBD_FAIL_MDS_ALL_REQUEST_NET | OBD_FAIL_ONCE, 0); + if (OBD_FAIL_CHECK_ORSET(OBD_FAIL_MDS_ALL_REQUEST_NET, OBD_FAIL_ONCE)) + RETURN(0); LASSERT(current->journal_info == NULL); msg = req->rq_reqmsg; - rc = mds_msg_check_version(msg); + rc = mdt_msg_check_version(msg); if (likely(rc == 0)) { rc = mdt_recovery(info); if (likely(rc == +1)) { @@ -2343,7 +2448,6 @@ static int mdt_handle0(struct ptlrpc_request *req, supported); if (likely(h != NULL)) { rc = mdt_req_handle(info, h, req); - rc = mdt_reply(req, rc, info); } else { CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) ); req->rq_status = -ENOTSUPP; @@ -2619,7 +2723,7 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT)) return; - dlmreq = req_capsule_client_get(&info->mti_pill, &RMF_DLM_REQ); + dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ); remote_hdl = dlmreq->lock_handle[0]; spin_lock(&exp->exp_ldlm_data.led_lock); @@ -2633,7 +2737,8 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, lh->mlh_reg_mode = lock->l_granted_mode; LDLM_DEBUG(lock, "restoring lock cookie"); - DEBUG_REQ(D_HA, req, "restoring lock cookie "LPX64, + DEBUG_REQ(D_DLMTRACE, req, + "restoring lock cookie "LPX64, lh->mlh_reg_lh.cookie); if (old_lock) *old_lock = LDLM_LOCK_GET(lock); @@ -2657,7 +2762,7 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, */ lustre_msg_clear_flags(req->rq_reqmsg, MSG_RESENT); - DEBUG_REQ(D_HA, req, "no existing lock with rhandle "LPX64, + DEBUG_REQ(D_DLMTRACE, req, "no existing lock with rhandle "LPX64, remote_hdl.cookie); } @@ -2676,10 +2781,10 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, int rc; ENTRY; - reqbody = req_capsule_client_get(&info->mti_pill, &RMF_MDT_BODY); + reqbody = req_capsule_client_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(reqbody); - repbody = req_capsule_server_get(&info->mti_pill, &RMF_MDT_BODY); + repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); LASSERT(repbody); info->mti_spec.sp_ck_split = !!(reqbody->valid & OBD_MD_FLCKSPLIT); @@ -2696,15 +2801,15 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, break; default: CERROR("Unhandled till now"); - GOTO(out, rc = -EINVAL); + GOTO(out_shrink, rc = -EINVAL); } rc = mdt_init_ucred(info, reqbody); if (rc) - GOTO(out, rc); + GOTO(out_shrink, rc); - req = info->mti_pill.rc_req; - ldlm_rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP); + req = info->mti_pill->rc_req; + ldlm_rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); mdt_set_disposition(info, ldlm_rep, DISP_IT_EXECD); /* Get lock from request for possible resent case. */ @@ -2725,7 +2830,7 @@ static int mdt_intent_getattr(enum mdt_it_code opcode, EXIT; out_ucred: mdt_exit_ucred(info); -out: +out_shrink: mdt_shrink_reply(info); return rc; } @@ -2749,12 +2854,12 @@ static int mdt_intent_reint(enum mdt_it_code opcode, opc = mdt_reint_opcode(info, intent_fmts); if (opc < 0) - GOTO(out, rc = opc); + RETURN(opc); if (mdt_it_flavor[opcode].it_reint != opc) { CERROR("Reint code %ld doesn't match intent: %d\n", opc, opcode); - GOTO(out, rc = err_serious(-EPROTO)); + RETURN(err_serious(-EPROTO)); } /* Get lock from request for possible resent case. */ @@ -2764,9 +2869,9 @@ static int mdt_intent_reint(enum mdt_it_code opcode, /* Check whether the reply has been packed successfully. */ if (mdt_info_req(info)->rq_repmsg != NULL) - rep = req_capsule_server_get(&info->mti_pill, &RMF_DLM_REP); + rep = req_capsule_server_get(info->mti_pill, &RMF_DLM_REP); if (rep == NULL) - GOTO(out, rc = err_serious(-EFAULT)); + RETURN(err_serious(-EFAULT)); /* MDC expects this in any case */ if (rc != 0) @@ -2777,15 +2882,13 @@ static int mdt_intent_reint(enum mdt_it_code opcode, LASSERT(lustre_handle_is_used(&lhc->mlh_reg_lh)); rep->lock_policy_res2 = 0; rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); - GOTO(out, rc); + RETURN(rc); } rep->lock_policy_res2 = clear_serious(rc); lhc->mlh_reg_lh.cookie = 0ull; rc = ELDLM_LOCK_ABORTED; - EXIT; -out: - return rc; + RETURN(rc); } static int mdt_intent_code(long itcode) @@ -2841,7 +2944,7 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, if (opc < 0) RETURN(-EINVAL); - pill = &info->mti_pill; + pill = info->mti_pill; flv = &mdt_it_flavor[opc]; if (flv->it_fmt != NULL) @@ -2852,13 +2955,14 @@ static int mdt_intent_opc(long itopc, struct mdt_thread_info *info, struct ptlrpc_request *req = mdt_info_req(info); if (flv->it_flags & MUTABOR && req->rq_export->exp_connect_flags & OBD_CONNECT_RDONLY) - rc = -EROFS; + RETURN(-EROFS); } if (rc == 0 && flv->it_act != NULL) { /* execute policy */ rc = flv->it_act(opc, info, lockp, flags); - } else + } else { rc = -EOPNOTSUPP; + } RETURN(rc); } @@ -2879,7 +2983,7 @@ static int mdt_intent_policy(struct ldlm_namespace *ns, info = lu_context_key_get(&req->rq_svc_thread->t_env->le_ctx, &mdt_thread_key); LASSERT(info != NULL); - pill = &info->mti_pill; + pill = info->mti_pill; LASSERT(pill->rc_req == req); if (req->rq_reqmsg->lm_bufcount > DLM_INTENT_IT_OFF) { @@ -2914,7 +3018,7 @@ static int mdt_intent_policy(struct ldlm_namespace *ns, } else { /* No intent was provided */ LASSERT(pill->rc_fmt == &RQF_LDLM_ENQUEUE); - rc = req_capsule_pack(pill); + rc = req_capsule_server_pack(pill); if (rc) rc = err_serious(rc); } @@ -3244,21 +3348,21 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) procfs_entry = m->mdt_md_dev.md_lu_dev.ld_obd->obd_proc_entry; conf = (typeof(conf)) { - .psc_nbufs = MDS_NBUFS, - .psc_bufsize = MDS_BUFSIZE, - .psc_max_req_size = MDS_MAXREQSIZE, - .psc_max_reply_size = MDS_MAXREPSIZE, - .psc_req_portal = MDS_REQUEST_PORTAL, - .psc_rep_portal = MDC_REPLY_PORTAL, - .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT, + .psc_nbufs = MDS_NBUFS, + .psc_bufsize = MDS_BUFSIZE, + .psc_max_req_size = MDS_MAXREQSIZE, + .psc_max_reply_size = MDS_MAXREPSIZE, + .psc_req_portal = MDS_REQUEST_PORTAL, + .psc_rep_portal = MDC_REPLY_PORTAL, + .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, /* * 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_ctx_tags = LCT_MD_THREAD + .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), + MDT_MAX_THREADS), + .psc_max_threads = MDT_MAX_THREADS, + .psc_ctx_tags = LCT_MD_THREAD }; m->mdt_ldlm_client = &m->mdt_md_dev.md_lu_dev.ld_obd->obd_ldlm_client; @@ -3267,7 +3371,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) m->mdt_regular_service = ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME, - procfs_entry, NULL, LUSTRE_MDT_NAME); + procfs_entry, target_print_req, + LUSTRE_MDT_NAME); if (m->mdt_regular_service == NULL) RETURN(-ENOMEM); @@ -3280,22 +3385,22 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) * ideally. */ conf = (typeof(conf)) { - .psc_nbufs = MDS_NBUFS, - .psc_bufsize = MDS_BUFSIZE, - .psc_max_req_size = MDS_MAXREQSIZE, - .psc_max_reply_size = MDS_MAXREPSIZE, - .psc_req_portal = MDS_READPAGE_PORTAL, - .psc_rep_portal = MDC_REPLY_PORTAL, - .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT, - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, - .psc_ctx_tags = LCT_MD_THREAD + .psc_nbufs = MDS_NBUFS, + .psc_bufsize = MDS_BUFSIZE, + .psc_max_req_size = MDS_MAXREQSIZE, + .psc_max_reply_size = MDS_MAXREPSIZE, + .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_ctx_tags = LCT_MD_THREAD }; m->mdt_readpage_service = ptlrpc_init_svc_conf(&conf, mdt_readpage_handle, LUSTRE_MDT_NAME "_readpage", - procfs_entry, NULL, "mdt_rdpg"); + procfs_entry, target_print_req,"mdt_rdpg"); if (m->mdt_readpage_service == NULL) { CERROR("failed to start readpage service\n"); @@ -3308,23 +3413,23 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) * setattr service configuration. */ conf = (typeof(conf)) { - .psc_nbufs = MDS_NBUFS, - .psc_bufsize = MDS_BUFSIZE, - .psc_max_req_size = MDS_MAXREQSIZE, - .psc_max_reply_size = MDS_MAXREPSIZE, - .psc_req_portal = MDS_SETATTR_PORTAL, - .psc_rep_portal = MDC_REPLY_PORTAL, - .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT, + .psc_nbufs = MDS_NBUFS, + .psc_bufsize = MDS_BUFSIZE, + .psc_max_req_size = MDS_MAXREQSIZE, + .psc_max_reply_size = MDS_MAXREPSIZE, + .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_ctx_tags = LCT_MD_THREAD + MDT_MAX_THREADS), + .psc_max_threads = MDT_MAX_THREADS, + .psc_ctx_tags = LCT_MD_THREAD }; m->mdt_setattr_service = ptlrpc_init_svc_conf(&conf, mdt_regular_handle, LUSTRE_MDT_NAME "_setattr", - procfs_entry, NULL, "mdt_attr"); + procfs_entry, target_print_req,"mdt_attr"); if (!m->mdt_setattr_service) { CERROR("failed to start setattr service\n"); @@ -3339,22 +3444,22 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) * sequence controller service configuration */ conf = (typeof(conf)) { - .psc_nbufs = MDS_NBUFS, - .psc_bufsize = MDS_BUFSIZE, - .psc_max_req_size = SEQ_MAXREQSIZE, - .psc_max_reply_size = SEQ_MAXREPSIZE, - .psc_req_portal = SEQ_CONTROLLER_PORTAL, - .psc_rep_portal = MDC_REPLY_PORTAL, - .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, - .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD + .psc_nbufs = MDS_NBUFS, + .psc_bufsize = MDS_BUFSIZE, + .psc_max_req_size = SEQ_MAXREQSIZE, + .psc_max_reply_size = SEQ_MAXREPSIZE, + .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_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; m->mdt_mdsc_service = ptlrpc_init_svc_conf(&conf, mdt_mdsc_handle, LUSTRE_MDT_NAME"_mdsc", - procfs_entry, NULL, "mdt_mdsc"); + procfs_entry, target_print_req,"mdt_mdsc"); if (!m->mdt_mdsc_service) { CERROR("failed to start seq controller service\n"); GOTO(err_mdt_svc, rc = -ENOMEM); @@ -3368,22 +3473,22 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) * metadata sequence server service configuration */ conf = (typeof(conf)) { - .psc_nbufs = MDS_NBUFS, - .psc_bufsize = MDS_BUFSIZE, - .psc_max_req_size = SEQ_MAXREQSIZE, - .psc_max_reply_size = SEQ_MAXREPSIZE, - .psc_req_portal = SEQ_METADATA_PORTAL, - .psc_rep_portal = MDC_REPLY_PORTAL, - .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, - .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD + .psc_nbufs = MDS_NBUFS, + .psc_bufsize = MDS_BUFSIZE, + .psc_max_req_size = SEQ_MAXREQSIZE, + .psc_max_reply_size = SEQ_MAXREPSIZE, + .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_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; m->mdt_mdss_service = ptlrpc_init_svc_conf(&conf, mdt_mdss_handle, LUSTRE_MDT_NAME"_mdss", - procfs_entry, NULL, "mdt_mdss"); + procfs_entry, target_print_req,"mdt_mdss"); if (!m->mdt_mdss_service) { CERROR("failed to start metadata seq server service\n"); GOTO(err_mdt_svc, rc = -ENOMEM); @@ -3400,22 +3505,22 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) * controller which manages space. */ conf = (typeof(conf)) { - .psc_nbufs = MDS_NBUFS, - .psc_bufsize = MDS_BUFSIZE, - .psc_max_req_size = SEQ_MAXREQSIZE, - .psc_max_reply_size = SEQ_MAXREPSIZE, - .psc_req_portal = SEQ_DATA_PORTAL, - .psc_rep_portal = OSC_REPLY_PORTAL, - .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, - .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD + .psc_nbufs = MDS_NBUFS, + .psc_bufsize = MDS_BUFSIZE, + .psc_max_req_size = SEQ_MAXREQSIZE, + .psc_max_reply_size = SEQ_MAXREPSIZE, + .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_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; m->mdt_dtss_service = ptlrpc_init_svc_conf(&conf, mdt_dtss_handle, LUSTRE_MDT_NAME"_dtss", - procfs_entry, NULL, "mdt_dtss"); + procfs_entry, target_print_req,"mdt_dtss"); if (!m->mdt_dtss_service) { CERROR("failed to start data seq server service\n"); GOTO(err_mdt_svc, rc = -ENOMEM); @@ -3427,22 +3532,22 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) /* FLD service start */ conf = (typeof(conf)) { - .psc_nbufs = MDS_NBUFS, - .psc_bufsize = MDS_BUFSIZE, - .psc_max_req_size = FLD_MAXREQSIZE, - .psc_max_reply_size = FLD_MAXREPSIZE, - .psc_req_portal = FLD_REQUEST_PORTAL, - .psc_rep_portal = MDC_REPLY_PORTAL, - .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT, - .psc_min_threads = FLD_NUM_THREADS, - .psc_max_threads = FLD_NUM_THREADS, - .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD + .psc_nbufs = MDS_NBUFS, + .psc_bufsize = MDS_BUFSIZE, + .psc_max_req_size = FLD_MAXREQSIZE, + .psc_max_reply_size = FLD_MAXREPSIZE, + .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_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD }; m->mdt_fld_service = ptlrpc_init_svc_conf(&conf, mdt_fld_handle, LUSTRE_MDT_NAME"_fld", - procfs_entry, NULL, "mdt_fld"); + procfs_entry, target_print_req, "mdt_fld"); if (!m->mdt_fld_service) { CERROR("failed to start fld service\n"); GOTO(err_mdt_svc, rc = -ENOMEM); @@ -3457,21 +3562,22 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) * mds-mds requests be not blocked during recovery. */ conf = (typeof(conf)) { - .psc_nbufs = MDS_NBUFS, - .psc_bufsize = MDS_BUFSIZE, - .psc_max_req_size = MDS_MAXREQSIZE, - .psc_max_reply_size = MDS_MAXREPSIZE, - .psc_req_portal = MDS_MDS_PORTAL, - .psc_rep_portal = MDC_REPLY_PORTAL, - .psc_watchdog_timeout = MDT_SERVICE_WATCHDOG_TIMEOUT, - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, - .psc_ctx_tags = LCT_MD_THREAD + .psc_nbufs = MDS_NBUFS, + .psc_bufsize = MDS_BUFSIZE, + .psc_max_req_size = MDS_MAXREQSIZE, + .psc_max_reply_size = MDS_MAXREPSIZE, + .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_ctx_tags = LCT_MD_THREAD }; - m->mdt_xmds_service = ptlrpc_init_svc_conf(&conf, mdt_xmds_handle, - LUSTRE_MDT_NAME "_mds", - procfs_entry, NULL, "mdt_xmds"); + m->mdt_xmds_service = + ptlrpc_init_svc_conf(&conf, mdt_xmds_handle, + LUSTRE_MDT_NAME "_mds", + procfs_entry, target_print_req,"mdt_xmds"); if (m->mdt_xmds_service == NULL) { CERROR("failed to start readpage service\n"); @@ -3493,7 +3599,6 @@ err_mdt_svc: static void mdt_stack_fini(const struct lu_env *env, struct mdt_device *m, struct lu_device *top) { - struct lu_device *d = top, *n; struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd; struct lustre_cfg_bufs *bufs; struct lustre_cfg *lcfg; @@ -3522,23 +3627,7 @@ static void mdt_stack_fini(const struct lu_env *env, top->ld_ops->ldo_process_config(env, top, lcfg); lustre_cfg_free(lcfg); - lu_site_purge(env, top->ld_site, ~0); - while (d != NULL) { - struct obd_type *type; - struct lu_device_type *ldt = d->ld_type; - - /* each fini() returns next device in stack of layers - * * so we can avoid the recursion */ - n = ldt->ldt_ops->ldto_device_fini(env, d); - lu_device_put(d); - ldt->ldt_ops->ldto_device_free(env, d); - type = ldt->ldt_obd_type; - type->typ_refcnt--; - class_put_type(type); - - /* switch to the next device in the layer */ - d = n; - } + lu_stack_fini(env, top); m->mdt_child = NULL; m->mdt_bottom = NULL; } @@ -3669,20 +3758,17 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) ENTRY; ping_evictor_stop(); - + target_recovery_fini(obd); mdt_stop_ptlrpc_service(m); mdt_fs_cleanup(env, m); - upcall_cache_cleanup(m->mdt_rmtacl_cache); - m->mdt_rmtacl_cache = NULL; - upcall_cache_cleanup(m->mdt_identity_cache); m->mdt_identity_cache = NULL; if (m->mdt_namespace != NULL) { - ldlm_namespace_free(m->mdt_namespace, d->ld_obd->obd_force); + ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force); d->ld_obd->obd_namespace = m->mdt_namespace = NULL; } @@ -3693,28 +3779,16 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) ptlrpc_lprocfs_unregister_obd(d->ld_obd); lprocfs_obd_cleanup(d->ld_obd); - if (m->mdt_rootsquash_info) { - OBD_FREE_PTR(m->mdt_rootsquash_info); - m->mdt_rootsquash_info = NULL; - } + sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset); next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL); - del_timer(&m->mdt_ck_timer); + cfs_timer_disarm(&m->mdt_ck_timer); mdt_ck_thread_stop(m); /* finish the stack */ mdt_stack_fini(env, m, md2lu_dev(m->mdt_child)); if (ls) { - if (!list_empty(&ls->ls_lru) || ls->ls_total != 0) { - /* - * Uh-oh, objects still exist. - */ - static DECLARE_LU_CDEBUG_PRINT_INFO(cookie, D_ERROR); - - lu_site_print(env, ls, &cookie, lu_cdebug_printer); - } - lu_site_fini(ls); OBD_FREE_PTR(ls); d->ld_site = NULL; @@ -3729,6 +3803,15 @@ static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options) { char *p = options; +#ifdef CONFIG_FS_POSIX_ACL + /* ACLs should be enabled by default (b=13829) */ + m->mdt_opts.mo_acl = 1; + LCONSOLE_INFO("Enabling ACL\n"); +#else + m->mdt_opts.mo_acl = 0; + LCONSOLE_INFO("Disabling ACL\n"); +#endif + if (!options) return; @@ -3747,16 +3830,6 @@ static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options) (memcmp(options, "nouser_xattr", len) == 0)) { m->mdt_opts.mo_user_xattr = 0; LCONSOLE_INFO("Disabling user_xattr\n"); - } else if ((len == sizeof("acl") - 1) && - (memcmp(options, "acl", len) == 0)) { -#ifdef CONFIG_FS_POSIX_ACL - m->mdt_opts.mo_acl = 1; - LCONSOLE_INFO("Enabling ACL\n"); -#else - m->mdt_opts.mo_acl = 0; - CWARN("ignoring unsupported acl mount option\n"); - LCONSOLE_INFO("Disabling ACL\n"); -#endif } else if ((len == sizeof("noacl") - 1) && (memcmp(options, "noacl", len) == 0)) { m->mdt_opts.mo_acl = 0; @@ -3780,6 +3853,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, struct lustre_mount_info *lmi; struct lustre_sb_info *lsi; struct lu_site *s; + const char *identity_upcall = "NONE"; int rc; ENTRY; @@ -3806,6 +3880,9 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, server_put_mount_2(dev, lmi->lmi_mnt); } + m->mdt_sptlrpc_lock = RW_LOCK_UNLOCKED; + sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset); + spin_lock_init(&m->mdt_ioepoch_lock); m->mdt_opts.mo_compat_resname = 0; m->mdt_capa_timeout = CAPA_TIMEOUT; @@ -3830,7 +3907,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, GOTO(err_free_site, rc); } - lprocfs_init_vars(mdt, &lvars); + lprocfs_mdt_init_vars(&lvars); rc = lprocfs_obd_setup(obd, lvars.obd_vars); if (rc) { CERROR("Can't init lprocfs, rc %d\n", rc); @@ -3881,8 +3958,9 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, snprintf(info->mti_u.ns_name, sizeof info->mti_u.ns_name, LUSTRE_MDT_NAME"-%p", m); - m->mdt_namespace = ldlm_namespace_new(info->mti_u.ns_name, - LDLM_NAMESPACE_SERVER); + m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name, + LDLM_NAMESPACE_SERVER, + LDLM_NAMESPACE_GREEDY); if (m->mdt_namespace == NULL) GOTO(err_fini_seq, rc = -ENOMEM); @@ -3890,8 +3968,12 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, /* set obd_namespace for compatibility with old code */ obd->obd_namespace = m->mdt_namespace; - m->mdt_identity_cache = upcall_cache_init(obd->obd_name, - "NONE", + /* XXX: to support suppgid for ACL, we enable identity_upcall + * by default, otherwise, maybe got unexpected -EACCESS. */ + if (m->mdt_opts.mo_acl) + identity_upcall = MDT_IDENTITY_UPCALL_PATH; + + m->mdt_identity_cache = upcall_cache_init(obd->obd_name, identity_upcall, &mdt_identity_upcall_cache_ops); if (IS_ERR(m->mdt_identity_cache)) { rc = PTR_ERR(m->mdt_identity_cache); @@ -3899,18 +3981,8 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, GOTO(err_free_ns, rc); } - m->mdt_rmtacl_cache = upcall_cache_init(obd->obd_name, - MDT_RMTACL_UPCALL_PATH, - &mdt_rmtacl_upcall_cache_ops); - if (IS_ERR(m->mdt_rmtacl_cache)) { - rc = PTR_ERR(m->mdt_rmtacl_cache); - m->mdt_rmtacl_cache = NULL; - GOTO(err_free_ns, rc); - } + cfs_timer_init(&m->mdt_ck_timer, mdt_ck_timer_callback, m); - m->mdt_ck_timer.function = mdt_ck_timer_callback; - m->mdt_ck_timer.data = (unsigned long)m; - init_timer(&m->mdt_ck_timer); rc = mdt_ck_thread_start(m); if (rc) GOTO(err_free_ns, rc); @@ -3936,8 +4008,10 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, mdt_init_capa_ctxt(env, m); + /* we use a shorter ldlm_timeout on MDS for keep bumping on + * might-be slow processing OST */ if (ldlm_timeout == LDLM_TIMEOUT_DEFAULT) - ldlm_timeout = 6; + ldlm_timeout = MDS_LDLM_TIMEOUT_DEFAULT; RETURN(0); @@ -3948,14 +4022,12 @@ err_fs_cleanup: target_recovery_fini(obd); mdt_fs_cleanup(env, m); err_capa: - del_timer(&m->mdt_ck_timer); + cfs_timer_disarm(&m->mdt_ck_timer); mdt_ck_thread_stop(m); err_free_ns: - upcall_cache_cleanup(m->mdt_rmtacl_cache); - m->mdt_rmtacl_cache = NULL; upcall_cache_cleanup(m->mdt_identity_cache); m->mdt_identity_cache = NULL; - ldlm_namespace_free(m->mdt_namespace, 0); + ldlm_namespace_free(m->mdt_namespace, NULL, 0); obd->obd_namespace = m->mdt_namespace = NULL; err_fini_seq: mdt_seq_fini(env, m); @@ -3986,11 +4058,39 @@ 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; - lprocfs_init_vars(mdt, &lvars); + lprocfs_mdt_init_vars(&lvars); rc = class_process_proc_param(PARAM_MDT, lvars.obd_vars, cfg, obd); if (rc) /* others are passed further */ @@ -4162,6 +4262,12 @@ static int mdt_connect_internal(struct obd_export *exp, return -EBADE; } + 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; + } + return 0; } @@ -4169,11 +4275,14 @@ static int mdt_connect_internal(struct obd_export *exp, static int mdt_obd_connect(const struct lu_env *env, struct lustre_handle *conn, struct obd_device *obd, struct obd_uuid *cluuid, - struct obd_connect_data *data) + struct obd_connect_data *data, + void *localdata) { - struct mdt_client_data *mcd; + struct mdt_thread_info *info; + struct lsd_client_data *lcd; struct obd_export *exp; struct mdt_device *mdt; + struct ptlrpc_request *req; int rc; ENTRY; @@ -4181,6 +4290,8 @@ static int mdt_obd_connect(const struct lu_env *env, if (!conn || !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); @@ -4190,21 +4301,41 @@ static int mdt_obd_connect(const struct lu_env *env, exp = class_conn2export(conn); LASSERT(exp != NULL); + CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from)); + + 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_internal(exp, mdt, data); if (rc == 0) { - OBD_ALLOC_PTR(mcd); - if (mcd != NULL) { + OBD_ALLOC_PTR(lcd); + if (lcd != NULL) { struct mdt_thread_info *mti; mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); LASSERT(mti != NULL); mti->mti_exp = exp; - memcpy(mcd->mcd_uuid, cluuid, sizeof mcd->mcd_uuid); - exp->exp_mdt_data.med_mcd = mcd; + memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid); + exp->exp_mdt_data.med_lcd = lcd; rc = mdt_client_new(env, mdt); if (rc != 0) { - OBD_FREE_PTR(mcd); - exp->exp_mdt_data.med_mcd = NULL; + OBD_FREE_PTR(lcd); + exp->exp_mdt_data.med_lcd = NULL; } } else rc = -ENOMEM; @@ -4218,16 +4349,47 @@ static int mdt_obd_connect(const struct lu_env *env, RETURN(rc); } -static int mdt_obd_reconnect(struct obd_export *exp, struct obd_device *obd, +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) { - int rc; + struct mdt_thread_info *info; + struct mdt_device *mdt; + struct ptlrpc_request *req; + int rc; ENTRY; if (exp == NULL || obd == NULL || cluuid == NULL) 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); + + CDEBUG(D_SEC, "from %s\n", sptlrpc_part2name(req->rq_sp_from)); + + spin_lock(&exp->exp_lock); + if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) { + 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_internal(exp, mdt_dev(obd->obd_lu_dev), data); RETURN(rc); @@ -4272,8 +4434,10 @@ static int mdt_init_export(struct obd_export *exp) struct mdt_export_data *med = &exp->exp_mdt_data; ENTRY; - INIT_LIST_HEAD(&med->med_open_head); + CFS_INIT_LIST_HEAD(&med->med_open_head); spin_lock_init(&med->med_open_lock); + sema_init(&med->med_idmap_sem, 1); + med->med_idmap = NULL; spin_lock(&exp->exp_lock); exp->exp_connecting = 1; spin_unlock(&exp->exp_lock); @@ -4367,6 +4531,18 @@ out: return rc; } +static void mdt_allow_cli(struct mdt_device *m, unsigned int flag) +{ + if (flag & CONFIG_LOG) + m->mdt_fl_cfglog = 1; + if (flag & CONFIG_SYNC) + m->mdt_fl_synced = 1; + + if (m->mdt_fl_cfglog /* bz11778: && m->mdt_fl_synced */) + /* Open for clients */ + m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0; +} + static int mdt_upcall(const struct lu_env *env, struct md_device *md, enum md_upcall_event ev) { @@ -4383,12 +4559,17 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md, &m->mdt_max_cookiesize); 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); break; case MD_NO_TRANS: mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); mti->mti_no_need_trans = 1; CDEBUG(D_INFO, "disable mdt trans for this thread\n"); break; + case MD_LOV_CONFIG: + /* Check that MDT is not yet configured */ + LASSERT(!m->mdt_fl_cfglog); + break; default: CERROR("invalid event\n"); rc = -EINVAL; @@ -4399,15 +4580,16 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md, static int mdt_obd_notify(struct obd_device *host, struct obd_device *watched, - enum obd_notify_event ev, void *owner) + enum obd_notify_event ev, void *data) { ENTRY; switch (ev) { case OBD_NOTIFY_CONFIG: - host->obd_no_conn = 0; + mdt_allow_cli(mdt_dev(host->obd_lu_dev), (unsigned long)data); + break; default: - CDEBUG(D_INFO, "Notification 0x%x\n", ev); + CDEBUG(D_INFO, "Unhandled notification %#x\n", ev); } RETURN(0); } @@ -4456,11 +4638,11 @@ int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) struct obd_device *obd = mdt->mdt_md_dev.md_lu_dev.ld_obd; int rc, lost; ENTRY; - /* if some clients didn't participate in recovery then we can possibly + /* if some clients didn't participate in recovery then we can possibly * lost sequence. Now we should increase sequence for safe value */ lost = obd->obd_max_recoverable_clients - obd->obd_connected_clients; mdt_seq_adjust(env, mdt, lost); - + rc = ld->ld_ops->ldo_recovery_complete(env, ld); RETURN(rc); } @@ -4500,11 +4682,14 @@ static struct lu_device* mdt_device_fini(const struct lu_env *env, RETURN(NULL); } -static void mdt_device_free(const struct lu_env *env, struct lu_device *d) +static struct lu_device *mdt_device_free(const struct lu_env *env, + struct lu_device *d) { struct mdt_device *m = mdt_dev(d); + ENTRY; OBD_FREE_PTR(m); + RETURN(NULL); } static struct lu_device *mdt_device_alloc(const struct lu_env *env, @@ -4531,59 +4716,14 @@ static struct lu_device *mdt_device_alloc(const struct lu_env *env, return l; } -/* - * context key constructor/destructor - */ -static void *mdt_key_init(const struct lu_context *ctx, - struct lu_context_key *key) -{ - struct mdt_thread_info *info; - - /* - * check that no high order allocations are incurred. - */ - CLASSERT(CFS_PAGE_SIZE >= sizeof *info); - OBD_ALLOC_PTR(info); - if (info == NULL) - info = ERR_PTR(-ENOMEM); - return info; -} - -static void mdt_key_fini(const struct lu_context *ctx, - struct lu_context_key *key, void *data) -{ - struct mdt_thread_info *info = data; - OBD_FREE_PTR(info); -} +/* context key constructor/destructor: mdt_key_init, mdt_key_fini */ +LU_KEY_INIT_FINI(mdt, struct mdt_thread_info); -struct lu_context_key mdt_thread_key = { - .lct_tags = LCT_MD_THREAD, - .lct_init = mdt_key_init, - .lct_fini = mdt_key_fini -}; +/* context key: mdt_thread_key */ +LU_CONTEXT_KEY_DEFINE(mdt, LCT_MD_THREAD); -static void *mdt_txn_key_init(const struct lu_context *ctx, - struct lu_context_key *key) -{ - struct mdt_txn_info *txi; - - /* - * check that no high order allocations are incurred. - */ - CLASSERT(CFS_PAGE_SIZE >= sizeof *txi); - OBD_ALLOC_PTR(txi); - if (txi == NULL) - txi = ERR_PTR(-ENOMEM); - memset(txi, 0, sizeof(*txi)); - return txi; -} - -static void mdt_txn_key_fini(const struct lu_context *ctx, - struct lu_context_key *key, void *data) -{ - struct mdt_txn_info *txi = data; - OBD_FREE_PTR(txi); -} +/* context key constructor/destructor: mdt_txn_key_init, mdt_txn_key_fini */ +LU_KEY_INIT_FINI(mdt_txn, struct mdt_txn_info); struct lu_context_key mdt_txn_key = { .lct_tags = LCT_TX_HANDLE, @@ -4596,24 +4736,8 @@ struct md_ucred *mdt_ucred(const struct mdt_thread_info *info) return md_ucred(info->mti_env); } -static int mdt_type_init(struct lu_device_type *t) -{ - int rc; - - LU_CONTEXT_KEY_INIT(&mdt_thread_key); - rc = lu_context_key_register(&mdt_thread_key); - if (rc == 0) { - LU_CONTEXT_KEY_INIT(&mdt_txn_key); - rc = lu_context_key_register(&mdt_txn_key); - } - return rc; -} - -static void mdt_type_fini(struct lu_device_type *t) -{ - lu_context_key_degister(&mdt_thread_key); - lu_context_key_degister(&mdt_txn_key); -} +/* type constructor/destructor: mdt_type_init, mdt_type_fini */ +LU_TYPE_INIT_FINI(mdt, &mdt_thread_key, &mdt_txn_key); static struct lu_device_type_operations mdt_device_type_ops = { .ldto_init = mdt_type_init, @@ -4637,7 +4761,7 @@ static int __init mdt_mod_init(void) int rc; mdt_num_threads = MDT_NUM_THREADS; - lprocfs_init_vars(mdt, &lvars); + lprocfs_mdt_init_vars(&lvars); rc = class_register_type(&mdt_obd_device_ops, NULL, lvars.module_vars, LUSTRE_MDT_NAME, &mdt_device_type); @@ -4693,7 +4817,6 @@ DEF_MDT_HNDL_F(0, SET_INFO, mdt_set_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), -DEF_MDT_HNDL_F(HABEO_CORPUS|MUTABOR, SETXATTR, mdt_setxattr), DEF_MDT_HNDL_F(HABEO_CORPUS, GETXATTR, mdt_getxattr), DEF_MDT_HNDL_F(0 |HABEO_REFERO, STATFS, mdt_statfs), DEF_MDT_HNDL_F(0 |MUTABOR, REINT, mdt_reint), @@ -4702,8 +4825,8 @@ 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), -DEF_MDT_HNDL_0(0, QUOTACHECK, mdt_quotacheck_handle), -DEF_MDT_HNDL_0(0, QUOTACTL, mdt_quotactl_handle) +DEF_MDT_HNDL_F(0, QUOTACHECK, mdt_quotacheck_handle), +DEF_MDT_HNDL_F(0, QUOTACTL, mdt_quotactl_handle) }; #define DEF_OBD_HNDL(flags, name, fn) \ @@ -4816,6 +4939,11 @@ static struct mdt_opc_slice mdt_xmds_handlers[] = { .mos_hs = mdt_obd_ops }, { + .mos_opc_start = SEC_CTX_INIT, + .mos_opc_end = SEC_LAST_OPC, + .mos_hs = mdt_sec_ctx_ops + }, + { .mos_hs = NULL } };