X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_handler.c;h=7300d48cc23d53e5dfd38bf35fdf125914fca438;hp=90623a17c81a2f9908c989c5acb50f45601b21d1;hb=d21e502ec3a201b9b2a5b31c384e187296882d25;hpb=8266f4fc5884e62526c62ee488b7e81904211054 diff --git a/lustre/mdt/mdt_handler.c b/lustre/mdt/mdt_handler.c index 90623a1..7300d48 100644 --- a/lustre/mdt/mdt_handler.c +++ b/lustre/mdt/mdt_handler.c @@ -26,7 +26,7 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. */ /* @@ -99,7 +99,9 @@ ldlm_mode_t mdt_dlm_lock_modes[] = { /* * Initialized in mdt_mod_init(). */ -unsigned long mdt_num_threads; +static unsigned long mdt_num_threads; +static unsigned long mdt_min_threads; +static unsigned long mdt_max_threads; /* ptlrpc request handler for MDT. All handlers are * grouped into several slices - struct mdt_opc_slice, @@ -165,7 +167,7 @@ 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(struct lu_env *env, struct mdt_device *mdt, +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; @@ -368,33 +370,39 @@ static int mdt_statfs(struct mdt_thread_info *info) RETURN(rc); } -void mdt_pack_size2body(struct mdt_thread_info *info, struct mdt_object *o) +/** + * Pack SOM attributes into the reply. + * Call under a DLM UPDATE lock. + */ +static void mdt_pack_size2body(struct mdt_thread_info *info, + struct mdt_object *mo) { struct mdt_body *b; - struct lu_attr *attr = &info->mti_attr.ma_attr; + struct md_attr *ma = &info->mti_attr; + LASSERT(ma->ma_attr.la_valid & LA_MODE); b = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); - /* Check if Size-on-MDS is enabled. */ - if ((mdt_conn_flags(info) & OBD_CONNECT_SOM) && - S_ISREG(attr->la_mode) && mdt_sizeonmds_enabled(o)) { - b->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS); - b->size = attr->la_size; - b->blocks = attr->la_blocks; - } + /* Check if Size-on-MDS is supported, if this is a regular file, + * if SOM is enabled on the object and if SOM cache exists and valid. + * Otherwise do not pack Size-on-MDS attributes to the reply. */ + if (!(mdt_conn_flags(info) & OBD_CONNECT_SOM) || + !S_ISREG(ma->ma_attr.la_mode) || + !mdt_object_is_som_enabled(mo) || + !(ma->ma_valid & MA_SOM)) + return; + + b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + b->size = ma->ma_som->msd_size; + b->blocks = ma->ma_som->msd_blocks; } void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, const struct lu_attr *attr, const struct lu_fid *fid) { - /*XXX should pack the reply body according to lu_valid*/ - b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID | - OBD_MD_FLGID | OBD_MD_FLTYPE | - OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS | - OBD_MD_FLATIME | OBD_MD_FLMTIME ; + struct md_attr *ma = &info->mti_attr; - if (!S_ISREG(attr->la_mode)) - b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV; + LASSERT(ma->ma_valid & MA_INODE); b->atime = attr->la_atime; b->mtime = attr->la_mtime; @@ -408,6 +416,23 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, b->nlink = attr->la_nlink; b->rdev = attr->la_rdev; + /*XXX should pack the reply body according to lu_valid*/ + b->valid |= OBD_MD_FLCTIME | OBD_MD_FLUID | + OBD_MD_FLGID | OBD_MD_FLTYPE | + OBD_MD_FLMODE | OBD_MD_FLNLINK | OBD_MD_FLFLAGS | + OBD_MD_FLATIME | OBD_MD_FLMTIME ; + + if (!S_ISREG(attr->la_mode)) { + b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS | OBD_MD_FLRDEV; + } else if (ma->ma_need & MA_LOV && ma->ma_lmm_size == 0) { + /* means no objects are allocated on osts. */ + LASSERT(!(ma->ma_valid & MA_LOV)); + /* just ignore blocks occupied by extend attributes on MDS */ + b->blocks = 0; + /* if no object is allocated on osts, the size on mds is valid. b=22272 */ + b->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; + } + if (fid) { b->fid1 = *fid; b->valid |= OBD_MD_FLID; @@ -423,6 +448,10 @@ void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b, if (info) mdt_body_reverse_idmap(info, b); + + if (b->valid & OBD_MD_FLSIZE) + CDEBUG(D_VFSTRACE, DFID": returning size %llu\n", + PFID(fid), (unsigned long long)b->size); } static inline int mdt_body_has_lov(const struct lu_attr *la, @@ -433,7 +462,7 @@ static inline int mdt_body_has_lov(const struct lu_attr *la, } static int mdt_getattr_internal(struct mdt_thread_info *info, - struct mdt_object *o) + struct mdt_object *o, int ma_need) { struct md_object *next = mdt_object_child(o); const struct mdt_body *reqbody = info->mti_body; @@ -484,6 +513,10 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, /* get default stripe info for this dir. */ ma->ma_need |= MA_LOV_DEF; } + ma->ma_need |= ma_need; + if (ma->ma_need & MA_SOM) + ma->ma_som = &info->mti_u.som.data; + rc = mo_attr_get(env, next, ma); if (unlikely(rc)) { CERROR("getattr error for "DFID": %d\n", @@ -511,9 +544,6 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, repbody->eadatasize = ma->ma_lmv_size; repbody->valid |= (OBD_MD_FLDIREA|OBD_MD_MEA); } - if (!(ma->ma_valid & MA_LOV) && !(ma->ma_valid & MA_LMV)) { - repbody->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS; - } } else if (S_ISLNK(la->la_mode) && reqbody->valid & OBD_MD_LINKNAME) { buffer->lb_buf = ma->ma_lmm; @@ -523,6 +553,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info, CERROR("readlink failed: %d\n", rc); rc = -EFAULT; } else { + if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO)) + rc -= 2; repbody->valid |= OBD_MD_LINKNAME; repbody->eadatasize = rc; /* NULL terminate */ @@ -691,7 +723,7 @@ static int mdt_getattr(struct mdt_thread_info *info) * remote obj, and at that time no capability is available. */ mdt_set_capainfo(info, 1, &reqbody->fid1, BYPASS_CAPA); - rc = mdt_getattr_internal(info, obj); + rc = mdt_getattr_internal(info, obj, 0); if (reqbody->valid & OBD_MD_FLRMTPERM) mdt_exit_ucred(info); EXIT; @@ -790,6 +822,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, struct ldlm_lock *lock; struct ldlm_res_id *res_id; int is_resent; + int ma_need = 0; int rc; ENTRY; @@ -884,7 +917,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, /* Finally, we can get attr for child. */ mdt_set_capainfo(info, 0, mdt_object_fid(child), BYPASS_CAPA); - rc = mdt_getattr_internal(info, child); + rc = mdt_getattr_internal(info, child, 0); if (unlikely(rc != 0)) mdt_object_unlock(info, child, lhc, 1); } @@ -955,6 +988,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info, relock: ma = &info->mti_attr; + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2); mdt_lock_handle_init(lhc); mdt_lock_reg_init(lhc, LCK_PR); @@ -988,38 +1022,34 @@ relock: GOTO(out_child, rc); } + lock = ldlm_handle2lock(&lhc->mlh_reg_lh); + /* Get MA_SOM attributes if update lock is given. */ + if (lock && + lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_UPDATE && + S_ISREG(lu_object_attr(&mdt_object_child(child)->mo_lu))) + ma_need = MA_SOM; + /* finally, we can get attr for child. */ mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA); - rc = mdt_getattr_internal(info, child); + rc = mdt_getattr_internal(info, child, ma_need); if (unlikely(rc != 0)) { mdt_object_unlock(info, child, lhc, 1); - } else { - lock = ldlm_handle2lock(&lhc->mlh_reg_lh); - if (lock) { - struct mdt_body *repbody; + } else if (lock) { + /* Debugging code. */ + res_id = &lock->l_resource->lr_name; + LDLM_DEBUG(lock, "Returning lock to client"); + LASSERTF(fid_res_name_eq(mdt_object_fid(child), + &lock->l_resource->lr_name), + "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n", + (unsigned long)res_id->name[0], + (unsigned long)res_id->name[1], + (unsigned long)res_id->name[2], + PFID(mdt_object_fid(child))); + mdt_pack_size2body(info, child); + } + if (lock) + LDLM_LOCK_PUT(lock); - /* Debugging code. */ - res_id = &lock->l_resource->lr_name; - LDLM_DEBUG(lock, "Returning lock to client\n"); - LASSERTF(fid_res_name_eq(mdt_object_fid(child), - &lock->l_resource->lr_name), - "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n", - (unsigned long)res_id->name[0], - (unsigned long)res_id->name[1], - (unsigned long)res_id->name[2], - PFID(mdt_object_fid(child))); - /* - * Pack Size-on-MDS inode attributes to the body if - * update lock is given. - */ - repbody = req_capsule_server_get(info->mti_pill, - &RMF_MDT_BODY); - if (lock->l_policy_data.l_inodebits.bits & - MDS_INODELOCK_UPDATE) - mdt_pack_size2body(info, child); - LDLM_LOCK_PUT(lock); - } - } EXIT; out_child: mdt_object_put(info->mti_env, child); @@ -1103,16 +1133,17 @@ static int mdt_set_info(struct mdt_thread_info *info) vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL, RCL_CLIENT); + /* Swab any part of val you need to here */ if (KEY_IS(KEY_READ_ONLY)) { req->rq_status = 0; lustre_msg_set_status(req->rq_repmsg, 0); - spin_lock(&req->rq_export->exp_lock); + cfs_spin_lock(&req->rq_export->exp_lock); if (*(__u32 *)val) req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY; else req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY; - spin_unlock(&req->rq_export->exp_lock); + cfs_spin_unlock(&req->rq_export->exp_lock); } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) { struct changelog_setinfo *cs = @@ -1121,7 +1152,7 @@ static int mdt_set_info(struct mdt_thread_info *info) CERROR("Bad changelog_clear setinfo size %d\n", vallen); RETURN(-EINVAL); } - if (lustre_msg_swabbed(req->rq_reqmsg)) { + if (ptlrpc_req_need_swab(req)) { __swab64s(&cs->cs_recno); __swab32s(&cs->cs_id); } @@ -1172,6 +1203,7 @@ static int mdt_sendpage(struct mdt_thread_info *info, struct lu_rdpg *rdpg) { struct ptlrpc_request *req = mdt_info_req(info); + struct obd_export *exp = req->rq_export; struct ptlrpc_bulk_desc *desc; struct l_wait_info *lwi = &info->mti_u.rdpg.mti_wait_info; int tmpcount; @@ -1204,29 +1236,38 @@ static int mdt_sendpage(struct mdt_thread_info *info, if (OBD_FAIL_CHECK(OBD_FAIL_MDS_SENDPAGE)) GOTO(abort_bulk, rc = 0); - 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(cfs_time_seconds(max(timeout, 1)), NULL, NULL); - rc = l_wait_event(desc->bd_waitq, !ptlrpc_server_bulk_active(desc), lwi); - LASSERT (rc == 0 || rc == -ETIMEDOUT); + do { + 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_INTERVAL(cfs_time_seconds(max(timeout, 1)), + cfs_time_seconds(1), NULL, NULL); + rc = l_wait_event(desc->bd_waitq, + !ptlrpc_server_bulk_active(desc) || + exp->exp_failed || + exp->exp_abort_active_req, lwi); + LASSERT (rc == 0 || rc == -ETIMEDOUT); + } while ((rc == -ETIMEDOUT) && + (req->rq_deadline > cfs_time_current_sec())); if (rc == 0) { if (desc->bd_success && desc->bd_nob_transferred == rdpg->rp_count) GOTO(free_desc, rc); - rc = -ETIMEDOUT; /* XXX should this be a different errno? */ + rc = -ETIMEDOUT; + if (exp->exp_abort_active_req || exp->exp_failed) + GOTO(abort_bulk, rc); } DEBUG_REQ(D_ERROR, req, "bulk failed: %s %d(%d), evicting %s@%s", (rc == -ETIMEDOUT) ? "timeout" : "network error", desc->bd_nob_transferred, rdpg->rp_count, - req->rq_export->exp_client_uuid.uuid, - req->rq_export->exp_connection->c_remote_uuid.uuid); + exp->exp_client_uuid.uuid, + exp->exp_connection->c_remote_uuid.uuid); - class_fail_export(req->rq_export); + class_fail_export(exp); EXIT; abort_bulk: @@ -1372,7 +1413,7 @@ static int mdt_writepage(struct mdt_thread_info *info) else rc = ptlrpc_start_bulk_transfer (desc); if (rc == 0) { - *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * HZ / 4, HZ, + *lwi = LWI_TIMEOUT_INTERVAL(obd_timeout * CFS_HZ / 4, CFS_HZ, mdt_bulk_timeout, desc); rc = l_wait_event(desc->bd_waitq, !ptlrpc_bulk_active(desc) || desc->bd_export->exp_failed, lwi); @@ -1481,6 +1522,7 @@ static int mdt_reint_internal(struct mdt_thread_info *info, { struct req_capsule *pill = info->mti_pill; struct mdt_device *mdt = info->mti_mdt; + struct md_quota *mq = md_quota(info->mti_env); struct mdt_body *repbody; int rc = 0; ENTRY; @@ -1515,6 +1557,8 @@ 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)) @@ -1537,6 +1581,7 @@ static int mdt_reint_internal(struct mdt_thread_info *info, rc = lustre_msg_get_status(mdt_info_req(info)->rq_repmsg); GOTO(out_ucred, rc); } + mq->mq_exp = info->mti_exp; rc = mdt_reint_rec(info, lhc); EXIT; out_ucred: @@ -1686,13 +1731,11 @@ static int mdt_quotacheck_handle(struct mdt_thread_info *info) struct obd_quotactl *oqctl; struct req_capsule *pill = info->mti_pill; struct obd_export *exp = info->mti_exp; + struct md_quota *mq = md_quota(info->mti_env); struct md_device *next = info->mti_mdt->mdt_child; int rc; ENTRY; - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACHECK_NET)) - RETURN(0); - oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL); if (oqctl == NULL) RETURN(-EPROTO); @@ -1705,7 +1748,8 @@ static int mdt_quotacheck_handle(struct mdt_thread_info *info) if (rc) RETURN(rc); - rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next, exp, + mq->mq_exp = exp; + rc = next->md_ops->mdo_quota.mqo_check(info->mti_env, next, oqctl->qc_type); RETURN(rc); } @@ -1715,14 +1759,12 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info) struct obd_quotactl *oqctl, *repoqc; struct req_capsule *pill = info->mti_pill; struct obd_export *exp = info->mti_exp; + struct md_quota *mq = md_quota(info->mti_env); struct md_device *next = info->mti_mdt->mdt_child; const struct md_quota_operations *mqo = &next->md_ops->mdo_quota; int id, rc; ENTRY; - if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACTL_NET)) - RETURN(0); - oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL); if (oqctl == NULL) RETURN(-EPROTO); @@ -1761,6 +1803,7 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info) repoqc = req_capsule_server_get(pill, &RMF_OBD_QUOTACTL); LASSERT(repoqc != NULL); + mq->mq_exp = exp; switch (oqctl->qc_cmd) { case Q_QUOTAON: rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type); @@ -1944,7 +1987,6 @@ static struct ldlm_callback_suite cbs = { static int mdt_enqueue(struct mdt_thread_info *info) { struct ptlrpc_request *req; - __u64 req_bits; int rc; /* @@ -1954,17 +1996,6 @@ static int mdt_enqueue(struct mdt_thread_info *info) LASSERT(info->mti_dlm_req != NULL); req = mdt_info_req(info); - - /* - * Lock without inodebits makes no sense and will oops later in - * ldlm. Let's check it now to see if we have wrong lock from client or - * bits get corrupted somewhere in mdt_intent_policy(). - */ - req_bits = info->mti_dlm_req->lock_desc.l_policy_data.l_inodebits.bits; - /* 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); info->mti_fail_id = OBD_FAIL_LDLM_REPLY; @@ -2107,7 +2138,7 @@ static inline int mdt_is_lock_sync(struct ldlm_lock *lock) int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc, void *data, int flag) { - struct obd_device *obd = lock->l_resource->lr_namespace->ns_obd; + struct obd_device *obd = ldlm_lock_to_ns(lock)->ns_obd; struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev); int rc; ENTRY; @@ -2178,7 +2209,8 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o, RETURN(-ESTALE); } else { /* Non-dir object shouldn't have PDO lock */ - LASSERT(S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu))); + if (!S_ISDIR(lu_object_attr(&o->mot_obj.mo_lu))) + RETURN(-ENOTDIR); } } @@ -2224,12 +2256,12 @@ int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *o, res_id, LDLM_FL_LOCAL_ONLY | LDLM_FL_ATOMIC_CB, &info->mti_exp->exp_handle.h_cookie); if (rc) - GOTO(out, rc); - -out: - if (rc) mdt_object_unlock(info, o, lh, 1); - + else if (unlikely(OBD_FAIL_PRECHECK(OBD_FAIL_MDS_PDO_LOCK)) && + lh->mlh_pdo_hash != 0 && + (lh->mlh_reg_mode == LCK_PW || lh->mlh_reg_mode == LCK_EX)) { + OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_PDO_LOCK, 10); + } RETURN(rc); } @@ -2297,7 +2329,7 @@ void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h, * * \param info thread info object * \param o mdt object - * \param h mdt lock handle referencing regular and PDO locks + * \param lh mdt lock handle referencing regular and PDO locks * \param decref force immediate lock releasing */ void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o, @@ -2503,7 +2535,7 @@ static int mdt_req_handle(struct mdt_thread_info *info, * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try * to put same checks into handlers like mdt_close(), mdt_reint(), * etc., without talking to mdt authors first. Checking same thing - * there again is useless and returning 0 error wihtout packing reply + * there again is useless and returning 0 error without packing reply * is buggy! Handlers either pack reply or return error. * * We return 0 here and do not send any reply in order to emulate @@ -2534,10 +2566,25 @@ static int mdt_req_handle(struct mdt_thread_info *info, 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; + if (unlikely(dlm_req->lock_desc.l_resource.lr_type == + LDLM_IBITS && + dlm_req->lock_desc.l_policy_data.\ + l_inodebits.bits == 0)) { + /* + * Lock without inodebits makes no sense and + * will oops later in ldlm. If client miss to + * set such bits, do not trigger ASSERTION. + * + * For liblustre flock case, it maybe zero. + */ + rc = -EPROTO; + } else { + 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 { rc = -EFAULT; } @@ -2594,7 +2641,7 @@ static int mdt_req_handle(struct mdt_thread_info *info, 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) && + if (unlikely(req_is_replay(req) && lustre_msg_get_transno(req->rq_reqmsg) == 0)) { DEBUG_REQ(D_ERROR, req, "transno is 0 during REPLAY"); LBUG(); @@ -2658,10 +2705,7 @@ 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; + info->mti_mos = NULL; memset(&info->mti_attr, 0, sizeof(info->mti_attr)); info->mti_body = NULL; @@ -2683,6 +2727,11 @@ 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; } @@ -2729,7 +2778,6 @@ static int mdt_filter_recovery_request(struct ptlrpc_request *req, static int mdt_recovery(struct mdt_thread_info *info) { struct ptlrpc_request *req = mdt_info_req(info); - int recovering; struct obd_device *obd; ENTRY; @@ -2752,7 +2800,7 @@ static int mdt_recovery(struct mdt_thread_info *info) } } - if (unlikely(req->rq_export == NULL)) { + if (unlikely(!class_connected_export(req->rq_export))) { CERROR("operation %d on unconnected MDS from %s\n", lustre_msg_get_opc(req->rq_reqmsg), libcfs_id2str(req->rq_peer)); @@ -2790,10 +2838,7 @@ static int mdt_recovery(struct mdt_thread_info *info) obd = req->rq_export->exp_obd; /* Check for aborted recovery... */ - spin_lock_bh(&obd->obd_processing_task_lock); - recovering = obd->obd_recovering; - spin_unlock_bh(&obd->obd_processing_task_lock); - if (unlikely(recovering)) { + if (unlikely(obd->obd_recovering)) { int rc; int should_process; DEBUG_REQ(D_INFO, req, "Got new replay"); @@ -2909,15 +2954,6 @@ 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)) { @@ -3173,7 +3209,7 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info, new_lock->l_writers--; } - new_lock->l_export = class_export_get(req->rq_export); + new_lock->l_export = class_export_lock_get(req->rq_export, new_lock); new_lock->l_blocking_ast = lock->l_blocking_ast; new_lock->l_completion_ast = lock->l_completion_ast; new_lock->l_remote_handle = lock->l_remote_handle; @@ -3181,9 +3217,9 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info, unlock_res_and_lock(new_lock); - lustre_hash_add(new_lock->l_export->exp_lock_hash, - &new_lock->l_remote_handle, - &new_lock->l_exp_hash); + cfs_hash_add(new_lock->l_export->exp_lock_hash, + &new_lock->l_remote_handle, + &new_lock->l_exp_hash); LDLM_LOCK_RELEASE(new_lock); lh->mlh_reg_lh.cookie = 0; @@ -3208,7 +3244,7 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, dlmreq = req_capsule_client_get(info->mti_pill, &RMF_DLM_REQ); remote_hdl = dlmreq->lock_handle[0]; - lock = lustre_hash_lookup(exp->exp_lock_hash, &remote_hdl); + lock = cfs_hash_lookup(exp->exp_lock_hash, &remote_hdl); if (lock) { if (lock != new_lock) { lh->mlh_reg_lh.cookie = lock->l_handle.h_cookie; @@ -3220,11 +3256,11 @@ static void mdt_intent_fixup_resent(struct mdt_thread_info *info, lh->mlh_reg_lh.cookie); if (old_lock) *old_lock = LDLM_LOCK_GET(lock); - lh_put(exp->exp_lock_hash, &lock->l_exp_hash); + cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash); return; } - lh_put(exp->exp_lock_hash, &lock->l_exp_hash); + cfs_hash_put(exp->exp_lock_hash, &lock->l_exp_hash); } /* @@ -3385,6 +3421,8 @@ static int mdt_intent_reint(enum mdt_it_code opcode, * checked here. */ if (lustre_handle_is_used(&lhc->mlh_reg_lh)) { + LASSERTF(rc == 0, "Error occurred but lock handle " + "is still in use\n"); rep->lock_policy_res2 = 0; rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags); RETURN(rc); @@ -3494,23 +3532,19 @@ static int mdt_intent_policy(struct ldlm_namespace *ns, req_capsule_extend(pill, &RQF_LDLM_INTENT); it = req_capsule_client_get(pill, &RMF_LDLM_INTENT); if (it != NULL) { - const struct ldlm_request *dlmreq; - __u64 req_bits; - rc = mdt_intent_opc(it->opc, info, lockp, flags); if (rc == 0) rc = ELDLM_OK; - /* - * Lock without inodebits makes no sense and will oops + /* Lock without inodebits makes no sense and will oops * later in ldlm. Let's check it now to see if we have - * wrong lock from client or bits get corrupted - * somewhere in mdt_intent_opc(). - */ - dlmreq = info->mti_dlm_req; - req_bits = dlmreq->lock_desc.l_policy_data.l_inodebits.bits; - LASSERT(req_bits != 0); - + * ibits corrupted somewhere in mdt_intent_opc(). + * The case for client miss to set ibits has been + * processed by others. */ + LASSERT(ergo(info->mti_dlm_req->lock_desc.l_resource.\ + lr_type == LDLM_IBITS, + info->mti_dlm_req->lock_desc.\ + l_policy_data.l_inodebits.bits != 0)); } else rc = err_serious(-EFAULT); } else { @@ -3523,25 +3557,6 @@ static int mdt_intent_policy(struct ldlm_namespace *ns, RETURN(rc); } -/* - * Seq wrappers - */ -static void mdt_seq_adjust(const struct lu_env *env, - struct mdt_device *m, int lost) -{ - struct md_site *ms = mdt_md_site(m); - struct lu_seq_range out; - ENTRY; - - LASSERT(ms && ms->ms_server_seq); - LASSERT(lost >= 0); - /* get extra seq from seq_server, moving it's range up */ - while (lost-- > 0) { - seq_server_alloc_meta(ms->ms_server_seq, NULL, &out, env); - } - EXIT; -} - static int mdt_seq_fini(const struct lu_env *env, struct mdt_device *m) { @@ -3864,9 +3879,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) * We'd like to have a mechanism to set this on a per-device * basis, but alas... */ - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD }; @@ -3881,7 +3895,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) if (m->mdt_regular_service == NULL) RETURN(-ENOMEM); - rc = ptlrpc_start_threads(NULL, m->mdt_regular_service); + rc = ptlrpc_start_threads(m->mdt_regular_service); if (rc) GOTO(err_mdt_svc, rc); @@ -3897,9 +3911,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = MDS_READPAGE_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD }; m->mdt_readpage_service = @@ -3912,7 +3925,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) GOTO(err_mdt_svc, rc = -ENOMEM); } - rc = ptlrpc_start_threads(NULL, m->mdt_readpage_service); + rc = ptlrpc_start_threads(m->mdt_readpage_service); /* * setattr service configuration. @@ -3925,9 +3938,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = MDS_SETATTR_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD }; @@ -3941,7 +3953,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) GOTO(err_mdt_svc, rc = -ENOMEM); } - rc = ptlrpc_start_threads(NULL, m->mdt_setattr_service); + rc = ptlrpc_start_threads(m->mdt_setattr_service); if (rc) GOTO(err_mdt_svc, rc); @@ -3956,8 +3968,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = SEQ_CONTROLLER_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; @@ -3970,7 +3982,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) GOTO(err_mdt_svc, rc = -ENOMEM); } - rc = ptlrpc_start_threads(NULL, m->mdt_mdsc_service); + rc = ptlrpc_start_threads(m->mdt_mdsc_service); if (rc) GOTO(err_mdt_svc, rc); @@ -3985,8 +3997,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = SEQ_METADATA_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; @@ -3999,7 +4011,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) GOTO(err_mdt_svc, rc = -ENOMEM); } - rc = ptlrpc_start_threads(NULL, m->mdt_mdss_service); + rc = ptlrpc_start_threads(m->mdt_mdss_service); if (rc) GOTO(err_mdt_svc, rc); @@ -4017,8 +4029,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = SEQ_DATA_PORTAL, .psc_rep_portal = OSC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = SEQ_NUM_THREADS, - .psc_max_threads = SEQ_NUM_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD }; @@ -4031,7 +4043,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) GOTO(err_mdt_svc, rc = -ENOMEM); } - rc = ptlrpc_start_threads(NULL, m->mdt_dtss_service); + rc = ptlrpc_start_threads(m->mdt_dtss_service); if (rc) GOTO(err_mdt_svc, rc); @@ -4044,8 +4056,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = FLD_REQUEST_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = FLD_NUM_THREADS, - .psc_max_threads = FLD_NUM_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_DT_THREAD|LCT_MD_THREAD }; @@ -4058,7 +4070,7 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) GOTO(err_mdt_svc, rc = -ENOMEM); } - rc = ptlrpc_start_threads(NULL, m->mdt_fld_service); + rc = ptlrpc_start_threads(m->mdt_fld_service); if (rc) GOTO(err_mdt_svc, rc); @@ -4074,9 +4086,8 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) .psc_req_portal = MDS_MDS_PORTAL, .psc_rep_portal = MDC_REPLY_PORTAL, .psc_watchdog_factor = MDT_SERVICE_WATCHDOG_FACTOR, - .psc_min_threads = min(max(mdt_num_threads, MDT_MIN_THREADS), - MDT_MAX_THREADS), - .psc_max_threads = MDT_MAX_THREADS, + .psc_min_threads = mdt_min_threads, + .psc_max_threads = mdt_max_threads, .psc_ctx_tags = LCT_MD_THREAD }; m->mdt_xmds_service = @@ -4085,11 +4096,11 @@ static int mdt_start_ptlrpc_service(struct mdt_device *m) procfs_entry, target_print_req,"mdt_xmds"); if (m->mdt_xmds_service == NULL) { - CERROR("failed to start readpage service\n"); + CERROR("failed to start xmds service\n"); GOTO(err_mdt_svc, rc = -ENOMEM); } - rc = ptlrpc_start_threads(NULL, m->mdt_xmds_service); + rc = ptlrpc_start_threads(m->mdt_xmds_service); if (rc) GOTO(err_mdt_svc, rc); @@ -4314,36 +4325,16 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) 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; target_recovery_fini(obd); - /* 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)); - } ping_evictor_stop(); mdt_stop_ptlrpc_service(m); mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT); mdt_obd_llog_cleanup(obd); + obd_exports_barrier(obd); obd_zombie_barrier(); #ifdef HAVE_QUOTA_SUPPORT next->md_ops->mdo_quota.mqo_cleanup(env, next); @@ -4354,7 +4345,8 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) m->mdt_identity_cache = NULL; if (m->mdt_namespace != NULL) { - ldlm_namespace_free(m->mdt_namespace, NULL, 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; } @@ -4379,15 +4371,9 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) */ 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); + mdt_procfs_fini(m); if (ls) { struct md_site *mite; @@ -4397,7 +4383,7 @@ static void mdt_fini(const struct lu_env *env, struct mdt_device *m) OBD_FREE_PTR(mite); d->ld_site = NULL; } - LASSERT(atomic_read(&d->ld_ref) == 0); + LASSERT(cfs_atomic_read(&d->ld_ref) == 0); EXIT; } @@ -4418,10 +4404,10 @@ static int mdt_adapt_sptlrpc_conf(struct obd_device *obd, int initial) sptlrpc_target_update_exp_flavor(obd, &tmp_rset); - write_lock(&m->mdt_sptlrpc_lock); + cfs_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); + cfs_write_unlock(&m->mdt_sptlrpc_lock); return 0; } @@ -4465,6 +4451,9 @@ static void fsoptions_to_mdt_flags(struct mdt_device *m, char *options) LCONSOLE_INFO("Disabling ACL\n"); } + if (!*p) + break; + options = ++p; } } @@ -4474,7 +4463,6 @@ int mdt_postrecov(const struct lu_env *, struct mdt_device *); static int mdt_init0(const struct lu_env *env, struct mdt_device *m, struct lu_device_type *ldt, struct lustre_cfg *cfg) { - struct lprocfs_static_vars lvars; struct mdt_thread_info *info; struct obd_device *obd; const char *dev = lustre_cfg_string(cfg, 0); @@ -4511,10 +4499,9 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, obd = class_name2obd(dev); LASSERT(obd != NULL); - spin_lock_init(&m->mdt_transno_lock); - m->mdt_max_mdsize = MAX_MD_SIZE; m->mdt_max_cookiesize = sizeof(struct llog_cookie); + m->mdt_som_conf = 0; m->mdt_opts.mo_user_xattr = 0; m->mdt_opts.mo_acl = 0; @@ -4534,12 +4521,20 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, CERROR("CMD Operation not allowed in IOP mode\n"); GOTO(err_lmi, rc = -EINVAL); } + /* Read recovery timeouts */ + if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_soft) + obd->obd_recovery_timeout = + lsi->lsi_lmd->lmd_recovery_time_soft; + + if (lsi->lsi_lmd && lsi->lsi_lmd->lmd_recovery_time_hard) + obd->obd_recovery_time_hard = + lsi->lsi_lmd->lmd_recovery_time_hard; } - rwlock_init(&m->mdt_sptlrpc_lock); + cfs_rwlock_init(&m->mdt_sptlrpc_lock); sptlrpc_rule_set_init(&m->mdt_sptlrpc_rset); - spin_lock_init(&m->mdt_ioepoch_lock); + cfs_spin_lock_init(&m->mdt_ioepoch_lock); m->mdt_opts.mo_compat_resname = 0; m->mdt_capa_timeout = CAPA_TIMEOUT; m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1; @@ -4549,9 +4544,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids); m->mdt_nosquash_str = NULL; m->mdt_nosquash_strlen = 0; - init_rwsem(&m->mdt_squash_sem); - - spin_lock_init(&m->mdt_client_bitmap_lock); + cfs_init_rwsem(&m->mdt_squash_sem); OBD_ALLOC_PTR(mite); if (mite == NULL) @@ -4570,27 +4563,12 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, GOTO(err_free_site, rc); } - lprocfs_mdt_init_vars(&lvars); - rc = lprocfs_obd_setup(obd, lvars.obd_vars); - if (rc) { - CERROR("Can't init lprocfs, rc %d\n", rc); - GOTO(err_fini_site, rc); - } - ptlrpc_lprocfs_register_obd(obd); - rc = mdt_procfs_init(m, dev); if (rc) { CERROR("Can't init MDT lprocfs, rc %d\n", rc); GOTO(err_fini_proc, rc); } - obd->obd_proc_exports_entry = proc_mkdir("exports", - obd->obd_proc_entry); - if (obd->obd_proc_exports_entry) - lprocfs_add_simple(obd->obd_proc_exports_entry, - "clear", lprocfs_nid_stats_clear_read, - lprocfs_nid_stats_clear_write, obd, NULL); - /* set server index */ lu_site2md(s)->ms_node_id = node_id; @@ -4633,7 +4611,8 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m, LUSTRE_MDT_NAME"-%p", m); m->mdt_namespace = ldlm_namespace_new(obd, info->mti_u.ns_name, LDLM_NAMESPACE_SERVER, - LDLM_NAMESPACE_GREEDY); + LDLM_NAMESPACE_GREEDY, + LDLM_NS_TYPE_MDT); if (m->mdt_namespace == NULL) GOTO(err_fini_seq, rc = -ENOMEM); @@ -4740,13 +4719,6 @@ 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) { - 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: lu_site_fini(s); err_free_site: OBD_FREE_PTR(mite); @@ -4830,13 +4802,14 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env, lu_object_init(o, h, d); lu_object_add_top(h, o); o->lo_ops = &mdt_obj_ops; + cfs_sema_init(&mo->mot_ioepoch_sem, 1); RETURN(o); } else RETURN(NULL); } static int mdt_object_init(const struct lu_env *env, struct lu_object *o, - const struct lu_object_conf *_) + const struct lu_object_conf *unused) { struct mdt_device *d = mdt_dev(o->lo_dev); struct lu_device *under; @@ -4877,10 +4850,10 @@ 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)", + return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch="LPU64" " + "flags="LPX64", epochcount=%d, writecount=%d)", mdto, mdto->mot_ioepoch, mdto->mot_flags, - mdto->mot_epochcount, mdto->mot_writecount); + mdto->mot_ioepoch_count, mdto->mot_writecount); } static const struct lu_device_operations mdt_lu_ops = { @@ -4935,9 +4908,12 @@ static int mdt_connect_internal(struct obd_export *exp, if (!mdt->mdt_opts.mo_user_xattr) data->ocd_connect_flags &= ~OBD_CONNECT_XATTR; - spin_lock(&exp->exp_lock); + if (!mdt->mdt_som_conf) + data->ocd_connect_flags &= ~OBD_CONNECT_SOM; + + cfs_spin_lock(&exp->exp_lock); exp->exp_connect_flags = data->ocd_connect_flags; - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&exp->exp_lock); data->ocd_version = LUSTRE_VERSION_CODE; exp->exp_mdt_data.med_ibits_known = data->ocd_ibits_known; } @@ -4957,6 +4933,13 @@ static int mdt_connect_internal(struct obd_export *exp, return -EBADE; } + if (mdt->mdt_som_conf && !exp_connect_som(exp) && + !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)) { + CWARN("%s: MDS has SOM enabled, but client does not support " + "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name); + return -EBADE; + } + return 0; } @@ -4968,14 +4951,14 @@ static int mdt_connect_check_sptlrpc(struct mdt_device *mdt, int rc = 0; if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) { - read_lock(&mdt->mdt_sptlrpc_lock); + cfs_read_lock(&mdt->mdt_sptlrpc_lock); sptlrpc_target_choose_flavor(&mdt->mdt_sptlrpc_rset, req->rq_sp_from, req->rq_peer.nid, &flvr); - read_unlock(&mdt->mdt_sptlrpc_lock); + cfs_read_unlock(&mdt->mdt_sptlrpc_lock); - spin_lock(&exp->exp_lock); + cfs_spin_lock(&exp->exp_lock); exp->exp_sp_peer = req->rq_sp_from; exp->exp_flvr = flvr; @@ -4989,7 +4972,7 @@ static int mdt_connect_check_sptlrpc(struct mdt_device *mdt, rc = -EACCES; } - spin_unlock(&exp->exp_lock); + cfs_spin_unlock(&exp->exp_lock); } else { if (exp->exp_sp_peer != req->rq_sp_from) { CERROR("RPC source %s doesn't match %s\n", @@ -5012,7 +4995,6 @@ static int mdt_obd_connect(const struct lu_env *env, void *localdata) { struct mdt_thread_info *info; - struct lsd_client_data *lcd; struct obd_export *lexp; struct lustre_handle conn = { 0 }; struct mdt_device *mdt; @@ -5041,31 +5023,25 @@ static int mdt_obd_connect(const struct lu_env *env, rc = mdt_connect_internal(lexp, mdt, data); if (rc == 0) { - 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 = lexp; - memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid); - lexp->exp_mdt_data.med_lcd = lcd; - rc = mdt_client_new(env, mdt); - if (rc != 0) { - OBD_FREE_PTR(lcd); - lexp->exp_mdt_data.med_lcd = NULL; - } else { - mdt_export_stats_init(obd, lexp, localdata); - } - } else - rc = -ENOMEM; + struct mdt_thread_info *mti; + struct lsd_client_data *lcd = lexp->exp_target_data.ted_lcd; + LASSERT(lcd); + mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); + LASSERT(mti != NULL); + mti->mti_exp = lexp; + memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid); + rc = mdt_client_new(env, mdt); + if (rc == 0) + mdt_export_stats_init(obd, lexp, 0, localdata); } out: - if (rc != 0) + if (rc != 0) { class_disconnect(lexp); - else + *exp = NULL; + } else { *exp = lexp; + } RETURN(rc); } @@ -5095,45 +5071,105 @@ static int mdt_obd_reconnect(const struct lu_env *env, rc = mdt_connect_internal(exp, mdt_dev(obd->obd_lu_dev), data); if (rc == 0) - mdt_export_stats_init(obd, exp, localdata); + mdt_export_stats_init(obd, exp, 1, localdata); + + RETURN(rc); +} +static int mdt_export_cleanup(struct obd_export *exp) +{ + struct mdt_export_data *med = &exp->exp_mdt_data; + struct obd_device *obd = exp->exp_obd; + struct mdt_device *mdt; + struct mdt_thread_info *info; + struct lu_env env; + CFS_LIST_HEAD(closing_list); + struct mdt_file_data *mfd, *n; + int rc = 0; + ENTRY; + + cfs_spin_lock(&med->med_open_lock); + while (!cfs_list_empty(&med->med_open_head)) { + cfs_list_t *tmp = med->med_open_head.next; + mfd = cfs_list_entry(tmp, struct mdt_file_data, mfd_list); + + /* Remove mfd handle so it can't be found again. + * We are consuming the mfd_list reference here. */ + class_handle_unhash(&mfd->mfd_handle); + cfs_list_move_tail(&mfd->mfd_list, &closing_list); + } + cfs_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 (!cfs_list_empty(&closing_list)) { + struct md_attr *ma = &info->mti_attr; + int lmm_size; + int cookie_size; + + lmm_size = mdt->mdt_max_mdsize; + OBD_ALLOC(ma->ma_lmm, lmm_size); + if (ma->ma_lmm == NULL) + GOTO(out_lmm, rc = -ENOMEM); + + cookie_size = mdt->mdt_max_cookiesize; + OBD_ALLOC(ma->ma_cookie, cookie_size); + if (ma->ma_cookie == NULL) + GOTO(out_cookie, rc = -ENOMEM); + + /* Close any open files (which may also cause orphan unlinking). */ + cfs_list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) { + cfs_list_del_init(&mfd->mfd_list); + memset(&ma->ma_attr, 0, sizeof(ma->ma_attr)); + ma->ma_lmm_size = lmm_size; + ma->ma_cookie_size = cookie_size; + ma->ma_need = 0; + /* It is not for setattr, just tell MDD to send + * DESTROY RPC to OSS if needed */ + ma->ma_attr_flags = MDS_CLOSE_CLEANUP; + ma->ma_valid = MA_FLAGS; + mdt_mfd_close(info, mfd); + } + OBD_FREE(ma->ma_cookie, cookie_size); + ma->ma_cookie = NULL; +out_cookie: + OBD_FREE(ma->ma_lmm, lmm_size); + ma->ma_lmm = NULL; + } +out_lmm: + info->mti_mdt = NULL; + /* cleanup client slot early */ + /* Do not erase record for recoverable client. */ + if (!obd->obd_fail || exp->exp_failed) + mdt_client_del(&env, mdt); + lu_env_fini(&env); RETURN(rc); } static int mdt_obd_disconnect(struct obd_export *exp) { - struct mdt_device *mdt = mdt_dev(exp->exp_obd->obd_lu_dev); int rc; ENTRY; LASSERT(exp); class_export_get(exp); - /* Disconnect early so that clients can't keep using export */ - rc = class_disconnect(exp); - if (mdt->mdt_namespace != NULL || exp->exp_obd->obd_namespace != NULL) - ldlm_cancel_locks_for_export(exp); - - /* release nid stat refererence */ - lprocfs_exp_cleanup(exp); - - /* complete all outstanding replies */ - spin_lock(&exp->exp_lock); - while (!list_empty(&exp->exp_outstanding_replies)) { - struct ptlrpc_reply_state *rs = - list_entry(exp->exp_outstanding_replies.next, - struct ptlrpc_reply_state, rs_exp_list); - struct ptlrpc_service *svc = rs->rs_service; - - spin_lock(&svc->srv_lock); - list_del_init(&rs->rs_exp_list); - spin_lock(&rs->rs_lock); - ptlrpc_schedule_difficult_reply(rs); - spin_unlock(&rs->rs_lock); - spin_unlock(&svc->srv_lock); - } - spin_unlock(&exp->exp_lock); + rc = server_disconnect_export(exp); + if (rc != 0) + CDEBUG(D_IOCTL, "server disconnect error: %d\n", rc); + rc = mdt_export_cleanup(exp); class_export_put(exp); RETURN(rc); } @@ -5146,128 +5182,67 @@ static int mdt_init_export(struct obd_export *exp) ENTRY; CFS_INIT_LIST_HEAD(&med->med_open_head); - spin_lock_init(&med->med_open_lock); - sema_init(&med->med_idmap_sem, 1); + cfs_spin_lock_init(&med->med_open_lock); + cfs_sema_init(&med->med_idmap_sem, 1); med->med_idmap = NULL; - spin_lock(&exp->exp_lock); + cfs_spin_lock(&exp->exp_lock); exp->exp_connecting = 1; - spin_unlock(&exp->exp_lock); - rc = ldlm_init_export(exp); + cfs_spin_unlock(&exp->exp_lock); + rc = lut_client_alloc(exp); + if (rc == 0) + rc = ldlm_init_export(exp); + if (rc) CERROR("Error %d while initializing export\n", rc); RETURN(rc); } -static int mdt_destroy_export(struct obd_export *export) +static int mdt_destroy_export(struct obd_export *exp) { struct mdt_export_data *med; - struct obd_device *obd = export->exp_obd; - struct mdt_device *mdt; - struct mdt_thread_info *info; - struct lu_env env; - struct md_attr *ma; - int lmm_size; - int cookie_size; - CFS_LIST_HEAD(closing_list); - struct mdt_file_data *mfd, *n; int rc = 0; ENTRY; - med = &export->exp_mdt_data; - if (exp_connect_rmtclient(export)) - mdt_cleanup_idmap(med); + med = &exp->exp_mdt_data; + if (exp_connect_rmtclient(exp)) + mdt_cleanup_idmap(&exp->exp_mdt_data); - target_destroy_export(export); - ldlm_destroy_export(export); + target_destroy_export(exp); + ldlm_destroy_export(exp); + lut_client_free(exp); - if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid)) + LASSERT(cfs_list_empty(&exp->exp_outstanding_replies)); + LASSERT(cfs_list_empty(&exp->exp_mdt_data.med_open_head)); + if (obd_uuid_equals(&exp->exp_client_uuid, &exp->exp_obd->obd_uuid)) RETURN(0); - mdt = mdt_dev(obd->obd_lu_dev); - LASSERT(mdt != NULL); - - rc = lu_env_init(&env, LCT_MD_THREAD); - if (rc) - RETURN(rc); - - info = lu_context_key_get(&env.le_ctx, &mdt_thread_key); - LASSERT(info != NULL); - memset(info, 0, sizeof *info); - info->mti_env = &env; - info->mti_mdt = mdt; - info->mti_exp = export; - - ma = &info->mti_attr; - lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize; - cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize; - OBD_ALLOC(ma->ma_lmm, lmm_size); - OBD_ALLOC(ma->ma_cookie, cookie_size); - - if (ma->ma_lmm == NULL || ma->ma_cookie == NULL) - GOTO(out, rc = -ENOMEM); - ma->ma_need = MA_LOV | MA_COOKIE; - ma->ma_valid = 0; - /* Close any open files (which may also cause orphan unlinking). */ - spin_lock(&med->med_open_lock); - while (!list_empty(&med->med_open_head)) { - struct list_head *tmp = med->med_open_head.next; - 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) { if (flag & CONFIG_LOG) - m->mdt_fl_cfglog = 1; + cfs_set_bit(MDT_FL_CFGLOG, &m->mdt_state); /* also notify active event */ if (flag & CONFIG_SYNC) - m->mdt_fl_synced = 1; + cfs_set_bit(MDT_FL_SYNCED, &m->mdt_state); - if (m->mdt_fl_cfglog && m->mdt_fl_synced) + if (cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state) && + cfs_test_bit(MDT_FL_SYNCED, &m->mdt_state)) { + struct obd_device *obd = m->mdt_md_dev.md_lu_dev.ld_obd; + /* Open for clients */ - m->mdt_md_dev.md_lu_dev.ld_obd->obd_no_conn = 0; + if (obd->obd_no_conn) { + cfs_spin_lock(&obd->obd_dev_lock); + obd->obd_no_conn = 0; + cfs_spin_unlock(&obd->obd_dev_lock); + } + } } static int mdt_upcall(const struct lu_env *env, struct md_device *md, - enum md_upcall_event ev) + enum md_upcall_event ev, void *data) { struct mdt_device *m = mdt_dev(&md->md_lu_dev); struct md_device *next = m->mdt_child; @@ -5283,6 +5258,9 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md, CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n", m->mdt_max_mdsize, m->mdt_max_cookiesize); mdt_allow_cli(m, CONFIG_SYNC); + if (data) + (*(__u64 *)data) = + m->mdt_lut.lut_obd->u.obt.obt_mount_count; break; case MD_NO_TRANS: mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key); @@ -5291,7 +5269,7 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md, break; case MD_LOV_CONFIG: /* Check that MDT is not yet configured */ - LASSERT(!m->mdt_fl_cfglog); + LASSERT(!cfs_test_bit(MDT_FL_CFGLOG, &m->mdt_state)); break; #ifdef HAVE_QUOTA_SUPPORT case MD_LOV_QUOTA: @@ -5334,81 +5312,30 @@ static int mdt_obd_notify(struct obd_device *host, RETURN(0); } -/* This routine is quite similar to mdc_ioc_fid2path() */ -static int mdt_ioc_fid2path(struct lu_env *env, struct mdt_device *mdt, - struct obd_ioctl_data *data) -{ - struct getinfo_fid2path *fp = NULL; - int pathlen; - int alloclen; - int rc; - - ENTRY; - - pathlen = data->ioc_plen1; - if (pathlen > PATH_MAX) - GOTO(out, rc = -EINVAL); - if (pathlen < 2) - GOTO(out, rc = -EINVAL); - - alloclen = sizeof(*fp) + pathlen; - OBD_ALLOC(fp, alloclen); - if (fp == NULL) - GOTO(out, rc = -ENOMEM); - - memcpy(&fp->gf_fid, data->ioc_inlbuf1, sizeof(struct lu_fid)); - memcpy(&fp->gf_recno, data->ioc_inlbuf2, sizeof(fp->gf_recno)); - memcpy(&fp->gf_linkno, data->ioc_inlbuf3, - sizeof(fp->gf_linkno)); - fp->gf_pathlen = pathlen; - - if (!fid_is_sane(&fp->gf_fid)) - GOTO(out, rc = -EINVAL); - - rc = mdt_fid2path(env, mdt, fp); - - if (!rc && copy_to_user(data->ioc_pbuf1, fp->gf_path, pathlen)) - GOTO(out, rc = -EFAULT); - memcpy(data->ioc_inlbuf2, &fp->gf_recno, sizeof(fp->gf_recno)); - memcpy(data->ioc_inlbuf3, &fp->gf_linkno, sizeof(fp->gf_linkno)); -out: - if (fp) - OBD_FREE(fp, alloclen); - RETURN(rc); -} - static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key, - int keylen, void *val, int vallen) + void *val, int vallen) { - struct lu_env env; struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev); - struct ptlrpc_request *req = mdt_info_req(info); struct getinfo_fid2path *fpout, *fpin; int rc = 0; - fpin = key + size_round(sizeof(KEY_FID2PATH)); + fpin = key + cfs_size_round(sizeof(KEY_FID2PATH)); fpout = val; - if (lustre_msg_swabbed(req->rq_reqmsg)) + if (ptlrpc_req_need_swab(info->mti_pill->rc_req)) lustre_swab_fid2path(fpin); memcpy(fpout, fpin, sizeof(*fpin)); if (fpout->gf_pathlen != vallen - sizeof(*fpin)) RETURN(-EINVAL); - rc = lu_env_init(&env, LCT_MD_THREAD); - if (rc) - RETURN(rc); - rc = mdt_fid2path(&env, mdt, fpout); - lu_env_fini(&env); - + rc = mdt_fid2path(info->mti_env, mdt, fpout); RETURN(rc); } -static int mdt_fid2path(struct lu_env *env, struct mdt_device *mdt, +static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt, struct getinfo_fid2path *fp) { - struct lu_context ioctl_session; struct mdt_object *obj; int rc; ENTRY; @@ -5419,18 +5346,11 @@ static int mdt_fid2path(struct lu_env *env, struct mdt_device *mdt, if (!fid_is_sane(&fp->gf_fid)) RETURN(-EINVAL); - 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; - obj = mdt_object_find(env, mdt, &fp->gf_fid); if (obj == NULL || IS_ERR(obj)) { CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid), PTR_ERR(obj)); - GOTO(out, rc = -EINVAL); + RETURN(-EINVAL); } rc = lu_object_exists(&obj->mot_obj.mo_lu); @@ -5442,33 +5362,24 @@ static int mdt_fid2path(struct lu_env *env, struct mdt_device *mdt, mdt_object_put(env, obj); CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n", PFID(&fp->gf_fid), rc); - GOTO(out, rc); + RETURN(rc); } rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path, fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno); mdt_object_put(env, obj); - EXIT; - -out: - lu_context_exit(&ioctl_session); - lu_context_fini(&ioctl_session); - return rc; + RETURN(rc); } static int mdt_get_info(struct mdt_thread_info *info) { + struct ptlrpc_request *req = mdt_info_req(info); char *key; - int keysize; int keylen; - __u32 *valin; - __u32 vallen; - int valsize; + __u32 *vallen; void *valout; int rc; - struct ptlrpc_request *req = mdt_info_req(info); - ENTRY; key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY); @@ -5476,25 +5387,17 @@ static int mdt_get_info(struct mdt_thread_info *info) CDEBUG(D_IOCTL, "No GETINFO key"); RETURN(-EFAULT); } - keysize = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY, + keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY, RCL_CLIENT); - keylen = strnlen(key, keysize); - valin = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN); - if (valin == NULL) { + 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); } - valsize = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_VALLEN, - RCL_CLIENT); - if (valsize != sizeof(vallen)) { - CDEBUG(D_IOCTL, "RMF_GETINFO_VALLEN has invalid size"); - RETURN(-EINVAL); - } - vallen = *valin; req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER, - vallen); + *vallen); rc = req_capsule_server_pack(info->mti_pill); valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL); if (valout == NULL) { @@ -5503,7 +5406,7 @@ static int mdt_get_info(struct mdt_thread_info *info) } if (KEY_IS(KEY_FID2PATH)) - rc = mdt_rpc_fid2path(info, key, keysize, valout, vallen); + rc = mdt_rpc_fid2path(info, key, valout, *vallen); else rc = -EINVAL; @@ -5536,6 +5439,43 @@ static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt, RETURN(rc); } +static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg) +{ + struct obd_ioctl_data *data = karg; + struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1; + __u64 version; + struct mdt_object *obj; + struct mdt_lock_handle *lh; + int rc; + ENTRY; + CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid)); + if (!fid_is_sane(fid)) + RETURN(-EINVAL); + + lh = &mti->mti_lh[MDT_LH_PARENT]; + mdt_lock_reg_init(lh, LCK_CR); + + obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE); + if (IS_ERR(obj)) + RETURN(PTR_ERR(obj)); + + rc = mdt_object_exists(obj); + if (rc < 0) { + rc = -EREMOTE; + /** + * before calling version get the correct MDS should be + * fid, this is error to find remote object here + */ + CERROR("nonlocal object "DFID"\n", PFID(fid)); + } else { + version = mo_version_get(mti->mti_env, mdt_object_child(obj)); + *(__u64 *)data->ioc_inlbuf2 = version; + rc = 0; + } + mdt_object_unlock_put(mti, obj, lh, 1); + RETURN(rc); +} + /* ioctls on obd dev */ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, void *karg, void *uarg) @@ -5564,14 +5504,22 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, target_stop_recovery_thread(obd); rc = 0; break; - case OBD_IOC_FID2PATH: - rc = mdt_ioc_fid2path(&env, mdt, karg); - break; case OBD_IOC_CHANGELOG_REG: case OBD_IOC_CHANGELOG_DEREG: case OBD_IOC_CHANGELOG_CLEAR: rc = mdt_ioc_child(&env, mdt, cmd, len, karg); break; + case OBD_IOC_GET_OBJ_VERSION: { + struct mdt_thread_info *mti; + mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key); + memset(mti, 0, sizeof *mti); + mti->mti_env = &env; + mti->mti_mdt = mdt; + mti->mti_exp = exp; + + rc = mdt_ioc_version_get(mti, karg); + break; + } default: CERROR("Not supported cmd = %d for device %s\n", cmd, obd->obd_name); @@ -5585,16 +5533,12 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len, int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt) { struct lu_device *ld = md2lu_dev(mdt->mdt_child); - struct obd_device *obd = mdt2obd_dev(mdt); #ifdef HAVE_QUOTA_SUPPORT + struct obd_device *obd = mdt2obd_dev(mdt); struct md_device *next = mdt->mdt_child; #endif - int rc, lost; + int rc; ENTRY; - /* 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); #ifdef HAVE_QUOTA_SUPPORT @@ -5617,6 +5561,63 @@ int mdt_obd_postrecov(struct obd_device *obd) return rc; } +/** + * Send a copytool req to a client + * Note this sends a request RPC from a server (MDT) to a client (MDC), + * backwards of normal comms. + */ +int mdt_hsm_copytool_send(struct obd_export *exp) +{ + struct kuc_hdr *lh; + struct hsm_action_list *hal; + struct hsm_action_item *hai; + int rc, len; + ENTRY; + + CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name, + libcfs_nid2str(exp->exp_connection->c_peer.nid)); + + len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN + + /* for mockup below */ 2 * cfs_size_round(sizeof(*hai)); + OBD_ALLOC(lh, len); + if (lh == NULL) + RETURN(-ENOMEM); + + lh->kuc_magic = KUC_MAGIC; + lh->kuc_transport = KUC_TRANSPORT_HSM; + lh->kuc_msgtype = HMT_ACTION_LIST; + lh->kuc_msglen = len; + + hal = (struct hsm_action_list *)(lh + 1); + hal->hal_version = HAL_VERSION; + hal->hal_archive_num = 1; + obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name, + MTI_NAME_MAXLEN); + + /* mock up an action list */ + hal->hal_count = 2; + hai = hai_zero(hal); + hai->hai_action = HSMA_ARCHIVE; + hai->hai_fid.f_oid = 0xA00A; + hai->hai_len = sizeof(*hai); + hai = hai_next(hai); + hai->hai_action = HSMA_RESTORE; + hai->hai_fid.f_oid = 0xB00B; + hai->hai_len = sizeof(*hai); + + /* Uses the ldlm reverse import; this rpc will be seen by + the ldlm_callback_handler */ + rc = do_set_info_async(exp->exp_imp_reverse, + LDLM_SET_INFO, LUSTRE_OBD_VERSION, + sizeof(KEY_HSM_COPYTOOL_SEND), + KEY_HSM_COPYTOOL_SEND, + len, lh, NULL); + + OBD_FREE(lh, len); + + RETURN(rc); +} + static struct obd_ops mdt_obd_device_ops = { .o_owner = THIS_MODULE, .o_set_info_async = mdt_obd_set_info_async, @@ -5696,7 +5697,7 @@ struct md_ucred *mdt_ucred(const struct mdt_thread_info *info) } /** - * Enable/disable COS. + * Enable/disable COS (Commit On Sharing). * * Set/Clear the COS flag in mdt options. * @@ -5720,9 +5721,9 @@ void mdt_enable_cos(struct mdt_device *mdt, int val) } /** - * Check COS status. + * Check COS (Commit On Sharing) status. * - * Return COS flag status/ + * Return COS flag status. * * \param mdt mdt device */ @@ -5766,7 +5767,19 @@ static int __init mdt_mod_init(void) llo_local_obj_register(&mdt_last_recv); - mdt_num_threads = MDT_NUM_THREADS; + if (mdt_num_threads > 0) { + if (mdt_num_threads > MDT_MAX_THREADS) + mdt_num_threads = MDT_MAX_THREADS; + if (mdt_num_threads < MDT_MIN_THREADS) + mdt_num_threads = MDT_MIN_THREADS; + mdt_max_threads = mdt_min_threads = mdt_num_threads; + } else { + mdt_max_threads = MDT_MAX_THREADS; + mdt_min_threads = MDT_MIN_THREADS; + if (mdt_min_threads < MDT_NUM_THREADS) + mdt_min_threads = MDT_NUM_THREADS; + } + lprocfs_mdt_init_vars(&lvars); rc = class_register_type(&mdt_obd_device_ops, NULL, lvars.module_vars, LUSTRE_MDT_NAME, @@ -5820,7 +5833,8 @@ static void __exit mdt_mod_exit(void) static struct mdt_handler mdt_mds_ops[] = { DEF_MDT_HNDL_F(0, CONNECT, mdt_connect), DEF_MDT_HNDL_F(0, DISCONNECT, mdt_disconnect), -DEF_MDT_HNDL_F(0, SET_INFO, mdt_set_info), +DEF_MDT_HNDL (0, SET_INFO, mdt_set_info, + &RQF_OBD_SET_INFO), DEF_MDT_HNDL_F(0, GET_INFO, mdt_get_info), DEF_MDT_HNDL_F(0 |HABEO_REFERO, GETSTATUS, mdt_getstatus), DEF_MDT_HNDL_F(HABEO_CORPUS, GETATTR, mdt_getattr),