X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=c29948692cd6a37ec777b6c6a7f5656fe8bd55cd;hb=41ab260642f4923106da3567d7c75e1ca3c61ea4;hp=4a7dbe52765c63aa7c602b04d016455dfc8225f1;hpb=0e660eab787c3b2857e4295f1ec554e016393885;p=fs%2Flustre-release.git diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 4a7dbe5..c299486 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -165,6 +165,8 @@ 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 const struct lu_object_operations mdt_obj_ops; @@ -349,7 +351,7 @@ 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) @@ -1068,12 +1070,15 @@ 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); @@ -1095,19 +1100,39 @@ 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); + 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); + 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; + 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 (lustre_msg_swabbed(req->rq_reqmsg)) { + __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); } @@ -1168,6 +1193,10 @@ 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); @@ -1327,6 +1356,9 @@ 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. @@ -1407,6 +1439,8 @@ 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]); @@ -1481,6 +1515,8 @@ 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)) @@ -1729,6 +1765,13 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info) switch (oqctl->qc_cmd) { case Q_QUOTAON: + if (info->mti_mdt->mdt_som_conf) { + /* Quota cannot be used together with SOM while + * SOM stored blocks in i_blocks but not in SOM EA. */ + LCONSOLE_ERROR("Fail to turn Quota on: SOM is enabled " + "and temporary conflicts with quota.\n"); + RETURN(-ENOTSUPP); + } rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type); break; case Q_QUOTAOFF: @@ -1775,6 +1818,7 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info) } #endif + /* * OBD PING and other handlers. */ @@ -1803,6 +1847,101 @@ 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 = { @@ -2222,7 +2361,9 @@ 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; @@ -2326,6 +2467,7 @@ 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); @@ -2458,7 +2600,7 @@ 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) && @@ -2525,6 +2667,10 @@ static void mdt_thread_info_init(struct ptlrpc_request *req, 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; @@ -2537,6 +2683,7 @@ 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) @@ -2545,6 +2692,11 @@ 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; } @@ -2704,6 +2856,7 @@ static int mdt_msg_check_version(struct lustre_msg *msg) case MDS_GETXATTR: case MDS_SETXATTR: case MDS_SET_INFO: + case MDS_GET_INFO: case MDS_QUOTACHECK: case MDS_QUOTACTL: case QUOTA_DQACQ: @@ -2770,12 +2923,22 @@ static int mdt_handle0(struct ptlrpc_request *req, if (likely(rc == 0)) { rc = mdt_recovery(info); if (likely(rc == +1)) { + switch (lustre_msg_get_opc(msg)) { + case MDS_READPAGE: + req->rq_bulk_read = 1; + break; + case MDS_WRITEPAGE: + req->rq_bulk_write = 1; + break; + } + h = mdt_handler_find(lustre_msg_get_opc(msg), supported); 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); @@ -3216,14 +3379,15 @@ 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 { /* @@ -3234,7 +3398,14 @@ 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); + } } } @@ -4114,7 +4285,7 @@ static int mdt_obd_llog_setup(struct obd_device *obd, obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd)); if (IS_ERR(obd->obd_fsops)) - return (int) PTR_ERR(obd->obd_fsops); + return PTR_ERR(obd->obd_fsops); rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb); if (rc) { @@ -4130,7 +4301,7 @@ static int mdt_obd_llog_setup(struct obd_device *obd, 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); + CERROR("llog_setup() failed: %d\n", rc); fsfilt_put_ops(obd->obd_fsops); } @@ -4145,49 +4316,32 @@ static void mdt_obd_llog_cleanup(struct obd_device *obd) if (ctxt) llog_cleanup(ctxt); - if (obd->obd_fsops) + 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 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); - int waited = 0; ENTRY; - /* At this point, obd exports might still be on the "obd_zombie_exports" - * list, and obd_zombie_impexp_thread() is trying to destroy them. - * We wait a little bit until all exports (except the self-export) - * have been destroyed, because the whole mdt stack might be accessed - * in mdt_destroy_export(). This will not be a long time, maybe one or - * two seconds are enough. This is not a problem while umounting. - * - * The three references that should be remaining are the - * obd_self_export and the attach and setup references. - */ - while (atomic_read(&obd->obd_refcount) > 3) { - cfs_schedule_timeout(CFS_TASK_UNINT, cfs_time_seconds(1)); - ++waited; - if (waited > 5 && IS_PO2(waited)) - LCONSOLE_WARN("Waiting for obd_zombie_impexp_thread " - "more than %d seconds to destroy all " - "the exports. The current obd refcount =" - " %d. Is it stuck there?\n", - waited, atomic_read(&obd->obd_refcount)); - } + target_recovery_fini(obd); ping_evictor_stop(); - target_recovery_fini(obd); mdt_stop_ptlrpc_service(m); + mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT); mdt_obd_llog_cleanup(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; @@ -4207,22 +4361,27 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) mdt_seq_fini(env, m); mdt_seq_fini_cli(m); mdt_fld_fini(env, m); - mdt_procfs_fini(m); - lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry); - lprocfs_free_per_client_stats(obd); - lprocfs_free_obd_stats(obd); - ptlrpc_lprocfs_unregister_obd(d->ld_obd); - lprocfs_obd_cleanup(d->ld_obd); - 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)); + mdt_procfs_fini(m); + if (obd->obd_proc_exports_entry) { + lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry); + obd->obd_proc_exports_entry = NULL; + } + lprocfs_free_per_client_stats(obd); + lprocfs_free_obd_stats(obd); + ptlrpc_lprocfs_unregister_obd(obd); + lprocfs_obd_cleanup(obd); + if (ls) { struct md_site *mite; @@ -4349,6 +4508,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, 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; @@ -4451,10 +4611,14 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, 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); @@ -4498,6 +4662,10 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, 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 @@ -4510,7 +4678,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, server_put_mount_2(dev, lmi->lmi_mnt); lmi = NULL; - target_recovery_init(obd, mdt_recovery_handle); + target_recovery_init(&m->mdt_lut, mdt_recovery_handle); rc = mdt_start_ptlrpc_service(m); if (rc) @@ -4542,8 +4710,9 @@ err_recovery: target_recovery_fini(obd); #ifdef HAVE_QUOTA_SUPPORT next->md_ops->mdo_quota.mqo_cleanup(env, next); -err_llog_cleanup: #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); @@ -4559,12 +4728,16 @@ 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); - if (obd->obd_proc_exports_entry) + if (obd->obd_proc_exports_entry) { lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry); + obd->obd_proc_exports_entry = NULL; + } ptlrpc_lprocfs_unregister_obd(obd); lprocfs_obd_cleanup(obd); err_fini_site: @@ -4572,7 +4745,7 @@ err_fini_site: err_free_site: OBD_FREE_PTR(mite); err_lmi: - if (lmi) + if (lmi) server_put_mount_2(dev, lmi->lmi_mnt); return (rc); } @@ -4694,6 +4867,16 @@ static void mdt_object_free(const struct lu_env *env, struct lu_object *o) EXIT; } +static int mdt_object_print(const struct lu_env *env, void *cookie, + lu_printer_t p, const struct lu_object *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_epochcount, mdto->mot_writecount); +} + static const struct lu_device_operations mdt_lu_ops = { .ldo_object_alloc = mdt_object_alloc, .ldo_process_config = mdt_process_config, @@ -4701,7 +4884,8 @@ static const struct lu_device_operations mdt_lu_ops = { static const struct lu_object_operations mdt_obj_ops = { .loo_object_init = mdt_object_init, - .loo_object_free = mdt_object_free + .loo_object_free = mdt_object_free, + .loo_object_print = mdt_object_print }; static int mdt_obd_set_info_async(struct obd_export *exp, @@ -4745,6 +4929,9 @@ 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_som_conf) + data->ocd_connect_flags &= ~OBD_CONNECT_SOM; + spin_lock(&exp->exp_lock); exp->exp_connect_flags = data->ocd_connect_flags; spin_unlock(&exp->exp_lock); @@ -4767,6 +4954,14 @@ static int mdt_connect_internal(struct obd_export *exp, return -EBADE; } + if (mdt->mdt_som_conf && + !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS) && + !(exp->exp_connect_flags & OBD_CONNECT_SOM)) { + 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; + } + return 0; } @@ -4816,39 +5011,40 @@ static int mdt_connect_check_sptlrpc(struct mdt_device *mdt, /* 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); + lexp = class_conn2export(&conn); + LASSERT(lexp != NULL); - rc = mdt_connect_check_sptlrpc(mdt, exp, req); + 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) { @@ -4856,15 +5052,15 @@ 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; + lexp->exp_mdt_data.med_lcd = NULL; } else { - mdt_export_stats_init(obd, exp, localdata); + mdt_export_stats_init(obd, lexp, localdata); } } else rc = -ENOMEM; @@ -4872,9 +5068,9 @@ static int mdt_obd_connect(const struct lu_env *env, out: if (rc != 0) - class_disconnect(exp); + class_disconnect(lexp); else - class_export_put(exp); + *exp = lexp; RETURN(rc); } @@ -4908,6 +5104,81 @@ static int mdt_obd_reconnect(const struct lu_env *env, RETURN(rc); } +static int mdt_mfd_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; + + spin_lock(&med->med_open_lock); + while (!list_empty(&med->med_open_head)) { + struct list_head *tmp = med->med_open_head.next; + 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_move_tail(&mfd->mfd_list, &closing_list); + } + spin_unlock(&med->med_open_lock); + mdt = mdt_dev(obd->obd_lu_dev); + LASSERT(mdt != NULL); + + 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 (!list_empty(&closing_list)) { + struct md_attr *ma = &info->mti_attr; + int lmm_size; + int cookie_size; + + lmm_size = mdt->mdt_max_mdsize; + cookie_size = mdt->mdt_max_cookiesize; + OBD_ALLOC(ma->ma_lmm, lmm_size); + if (ma->ma_lmm == NULL) + GOTO(out_lmm, rc = -ENOMEM); + 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). */ + list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) { + list_del_init(&mfd->mfd_list); + /* TODO: if we close the unlinked file, + * we need to remove its objects from OST */ + memset(&ma->ma_attr, 0, sizeof(ma->ma_attr)); + ma->ma_lmm_size = lmm_size; + ma->ma_cookie_size = cookie_size; + ma->ma_need = MA_LOV | MA_COOKIE; + ma->ma_valid = 0; + mdt_mfd_close(info, mfd); + } + info->mti_mdt = NULL; + 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: + lu_env_fini(&env); + + RETURN(rc); +} static int mdt_obd_disconnect(struct obd_export *exp) { @@ -4942,7 +5213,7 @@ static int mdt_obd_disconnect(struct obd_export *exp) spin_unlock(&svc->srv_lock); } spin_unlock(&exp->exp_lock); - + rc = mdt_mfd_cleanup(exp); class_export_put(exp); RETURN(rc); } @@ -4974,11 +5245,6 @@ static int mdt_destroy_export(struct obd_export *export) struct mdt_device *mdt; struct mdt_thread_info *info; struct lu_env env; - struct md_attr *ma; - int lmm_size; - int cookie_size; - CFS_LIST_HEAD(closing_list); - struct mdt_file_data *mfd, *n; int rc = 0; ENTRY; @@ -4989,6 +5255,8 @@ static int mdt_destroy_export(struct obd_export *export) target_destroy_export(export); ldlm_destroy_export(export); + LASSERT(list_empty(&export->exp_outstanding_replies)); + LASSERT(list_empty(&med->med_open_head)); if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid)) RETURN(0); @@ -5003,62 +5271,12 @@ static int mdt_destroy_export(struct obd_export *export) 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; - 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_move_tail(&mfd->mfd_list, &closing_list); - } - spin_unlock(&med->med_open_lock); - - list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) { - list_del_init(&mfd->mfd_list); - mdt_mfd_close(info, mfd); - /* TODO: if we close the unlinked file, - * we need to remove its 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; + RETURN(rc); } static void mdt_allow_cli(struct mdt_device *m, unsigned int flag) @@ -5076,7 +5294,7 @@ static void mdt_allow_cli(struct mdt_device *m, unsigned int flag) } 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; @@ -5092,10 +5310,8 @@ 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); -#ifdef HAVE_QUOTA_SUPPORT - if (md->md_lu_dev.ld_obd->obd_recovering == 0) - next->md_ops->mdo_quota.mqo_recovery(env, next); -#endif + if (data) + (*(__u64 *)data) = m->mdt_mount_count; break; case MD_NO_TRANS: mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); @@ -5106,6 +5322,13 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md, /* Check that MDT is not yet configured */ LASSERT(!m->mdt_fl_cfglog); 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; @@ -5140,48 +5363,45 @@ static int mdt_obd_notify(struct obd_device *host, RETURN(0); } -static int mdt_ioc_fid2path(struct lu_env *env, struct mdt_device *mdt, - struct obd_ioctl_data *data) +static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, + void *val, int vallen) { - struct lu_context ioctl_session; - struct mdt_object *obj; - struct lu_fid *fid; - char *path = NULL; - __u64 recno; - int pathlen = data->ioc_plen1; - int linkno; - int rc; - ENTRY; + struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev); + struct getinfo_fid2path *fpout, *fpin; + int rc = 0; + fpin = key + size_round(sizeof(KEY_FID2PATH)); + fpout = val; - fid = (struct lu_fid *)data->ioc_inlbuf1; - memcpy(&recno, data->ioc_inlbuf2, sizeof(recno)); - memcpy(&linkno, data->ioc_inlbuf3, sizeof(linkno)); - CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n", - PFID(fid), recno, linkno); + if (lustre_msg_swabbed(mdt_info_req(info)->rq_reqmsg)) + lustre_swab_fid2path(fpin); - if (!fid_is_sane(fid)) + memcpy(fpout, fpin, sizeof(*fpin)); + if (fpout->gf_pathlen != vallen - sizeof(*fpin)) RETURN(-EINVAL); - if (pathlen < 3) - RETURN(-EOVERFLOW); + rc = mdt_fid2path(info->mti_env, mdt, fpout); + RETURN(rc); +} - 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; +static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt, + struct getinfo_fid2path *fp) +{ + struct mdt_object *obj; + int rc; + ENTRY; - OBD_ALLOC(path, pathlen); - if (path == NULL) - GOTO(out_context, rc = -ENOMEM); + 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, fid); + obj = mdt_object_find(env, mdt, &fp->gf_fid); if (obj == NULL || IS_ERR(obj)) { - CDEBUG(D_IOCTL, "no object "DFID": %ld\n", PFID(fid), + CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid), PTR_ERR(obj)); - GOTO(out_free, rc = -EINVAL); + RETURN(-EINVAL); } rc = lu_object_exists(&obj->mot_obj.mo_lu); @@ -5191,31 +5411,86 @@ static int mdt_ioc_fid2path(struct lu_env *env, struct mdt_device *mdt, else rc = -ENOENT; mdt_object_put(env, obj); - CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n", PFID(fid), - rc); - GOTO(out_free, rc); + 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), path, pathlen, recno, - &linkno); + 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); - if (rc) - GOTO(out_free, rc); - if (copy_to_user(data->ioc_pbuf1, path, pathlen)) - rc = -EFAULT; + RETURN(rc); +} - memcpy(data->ioc_inlbuf3, &linkno, sizeof(linkno)); +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); - EXIT; -out_free: - OBD_FREE(path, pathlen); -out_context: lu_context_exit(&ioctl_session); lu_context_fini(&ioctl_session); - return rc; + RETURN(rc); } +/* ioctls on obd dev */ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void *uarg) { @@ -5243,8 +5518,10 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, target_stop_recovery_thread(obd); rc = 0; break; - case OBD_IOC_FID2PATH: - rc = mdt_ioc_fid2path(&env, mdt, karg); + 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; default: CERROR("Not supported cmd = %d for device %s\n", @@ -5272,7 +5549,8 @@ int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) rc = ld->ld_ops->ldo_recovery_complete(env, ld); #ifdef HAVE_QUOTA_SUPPORT - next->md_ops->mdo_quota.mqo_recovery(env, next); + if (likely(obd->obd_stopping == 0)) + next->md_ops->mdo_quota.mqo_recovery(env, next); #endif RETURN(rc); } @@ -5494,6 +5772,7 @@ 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_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), @@ -5533,7 +5812,19 @@ 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) \