+
+int (*tgt_lfsck_in_notify)(const struct lu_env *env,
+ struct dt_device *key,
+ struct lfsck_request *lr,
+ struct thandle *th) = NULL;
+
+void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
+ struct dt_device *,
+ struct lfsck_request *,
+ struct thandle *))
+{
+ tgt_lfsck_in_notify = notify;
+}
+EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
+
+static int (*tgt_lfsck_query)(const struct lu_env *env,
+ struct dt_device *key,
+ struct lfsck_request *lr) = NULL;
+
+void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
+ struct dt_device *,
+ struct lfsck_request *))
+{
+ tgt_lfsck_query = query;
+}
+EXPORT_SYMBOL(tgt_register_lfsck_query);
+
+/* LFSCK request handlers */
+static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
+{
+ const struct lu_env *env = tsi->tsi_env;
+ struct dt_device *key = tsi->tsi_tgt->lut_bottom;
+ struct lfsck_request *lr;
+ int rc;
+ ENTRY;
+
+ lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
+ if (lr == NULL)
+ RETURN(-EPROTO);
+
+ rc = tgt_lfsck_in_notify(env, key, lr, NULL);
+
+ RETURN(rc);
+}
+
+static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
+{
+ struct lfsck_request *request;
+ struct lfsck_reply *reply;
+ int rc;
+ ENTRY;
+
+ request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
+ if (request == NULL)
+ RETURN(-EPROTO);
+
+ reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
+ if (reply == NULL)
+ RETURN(-ENOMEM);
+
+ rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
+ reply->lr_status = rc;
+
+ RETURN(rc < 0 ? rc : 0);
+}
+
+struct tgt_handler tgt_lfsck_handlers[] = {
+TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
+TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
+};
+EXPORT_SYMBOL(tgt_lfsck_handlers);
+
+/*
+ * initialize per-thread page pool (bug 5137).
+ */
+int tgt_io_thread_init(struct ptlrpc_thread *thread)
+{
+ struct tgt_thread_big_cache *tbc;
+
+ ENTRY;
+
+ LASSERT(thread != NULL);
+ LASSERT(thread->t_data == NULL);
+
+ OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
+ if (tbc == NULL)
+ RETURN(-ENOMEM);
+ thread->t_data = tbc;
+ RETURN(0);
+}
+EXPORT_SYMBOL(tgt_io_thread_init);
+
+/*
+ * free per-thread pool created by tgt_thread_init().
+ */
+void tgt_io_thread_done(struct ptlrpc_thread *thread)
+{
+ struct tgt_thread_big_cache *tbc;
+
+ ENTRY;
+
+ LASSERT(thread != NULL);
+
+ /*
+ * be prepared to handle partially-initialized pools (because this is
+ * called from ost_io_thread_init() for cleanup.
+ */
+ tbc = thread->t_data;
+ if (tbc != NULL) {
+ OBD_FREE_LARGE(tbc, sizeof(*tbc));
+ thread->t_data = NULL;
+ }
+ EXIT;
+}
+EXPORT_SYMBOL(tgt_io_thread_done);
+/**
+ * Helper function for getting server side [start, start+count] DLM lock
+ * if asked by client.
+ */
+int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
+ __u64 start, __u64 end, struct lustre_handle *lh,
+ int mode, __u64 *flags)
+{
+ ldlm_policy_data_t policy;
+ int rc;
+
+ ENTRY;
+
+ LASSERT(lh != NULL);
+ LASSERT(ns != NULL);
+ LASSERT(!lustre_handle_is_used(lh));
+
+ policy.l_extent.gid = 0;
+ policy.l_extent.start = start & CFS_PAGE_MASK;
+
+ /*
+ * If ->o_blocks is EOF it means "lock till the end of the file".
+ * Otherwise, it's size of an extent or hole being punched (in bytes).
+ */
+ if (end == OBD_OBJECT_EOF || end < start)
+ policy.l_extent.end = OBD_OBJECT_EOF;
+ else
+ policy.l_extent.end = end | ~CFS_PAGE_MASK;
+
+ rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
+ flags, ldlm_blocking_ast,
+ ldlm_completion_ast, ldlm_glimpse_ast,
+ NULL, 0, LVB_T_NONE, NULL, lh);
+ RETURN(rc == ELDLM_OK ? 0 : -EIO);
+}
+EXPORT_SYMBOL(tgt_extent_lock);
+
+void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
+{
+ LASSERT(lustre_handle_is_used(lh));
+ ldlm_lock_decref(lh, mode);
+}
+EXPORT_SYMBOL(tgt_extent_unlock);
+
+int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
+ struct obd_ioobj *obj, struct niobuf_remote *nb,
+ struct lustre_handle *lh, int mode)
+{
+ __u64 flags = 0;
+ int nrbufs = obj->ioo_bufcnt;
+ int i;
+
+ ENTRY;
+
+ LASSERT(mode == LCK_PR || mode == LCK_PW);
+ LASSERT(!lustre_handle_is_used(lh));
+
+ if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
+ RETURN(0);
+
+ for (i = 1; i < nrbufs; i++)
+ if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
+ RETURN(-EFAULT);
+
+ RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
+ nb[nrbufs - 1].rnb_offset +
+ nb[nrbufs - 1].rnb_len - 1,
+ lh, mode, &flags));
+}
+EXPORT_SYMBOL(tgt_brw_lock);
+
+void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
+ struct lustre_handle *lh, int mode)
+{
+ ENTRY;
+
+ LASSERT(mode == LCK_PR || mode == LCK_PW);
+ LASSERT((obj->ioo_bufcnt > 0 &&
+ (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
+ lustre_handle_is_used(lh));
+
+ if (lustre_handle_is_used(lh))
+ tgt_extent_unlock(lh, mode);
+ EXIT;
+}
+EXPORT_SYMBOL(tgt_brw_unlock);
+
+static __u32 tgt_checksum_bulk(struct lu_target *tgt,
+ struct ptlrpc_bulk_desc *desc, int opc,
+ cksum_type_t cksum_type)
+{
+ struct cfs_crypto_hash_desc *hdesc;
+ unsigned int bufsize;
+ int i, err;
+ unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
+ __u32 cksum;
+
+ hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
+ if (IS_ERR(hdesc)) {
+ CERROR("%s: unable to initialize checksum hash %s\n",
+ tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
+ return PTR_ERR(hdesc);
+ }
+
+ CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
+ for (i = 0; i < desc->bd_iov_count; i++) {
+ /* corrupt the data before we compute the checksum, to
+ * simulate a client->OST data error */
+ if (i == 0 && opc == OST_WRITE &&
+ OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
+ int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
+ int len = desc->bd_iov[i].kiov_len;
+ struct page *np = tgt_page_to_corrupt;
+ char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
+
+ if (np) {
+ char *ptr2 = kmap(np) + off;
+
+ memcpy(ptr2, ptr, len);
+ memcpy(ptr2, "bad3", min(4, len));
+ kunmap(np);
+ desc->bd_iov[i].kiov_page = np;
+ } else {
+ CERROR("%s: can't alloc page for corruption\n",
+ tgt_name(tgt));
+ }
+ }
+ cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
+ desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
+ desc->bd_iov[i].kiov_len);
+
+ /* corrupt the data after we compute the checksum, to
+ * simulate an OST->client data error */
+ if (i == 0 && opc == OST_READ &&
+ OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
+ int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
+ int len = desc->bd_iov[i].kiov_len;
+ struct page *np = tgt_page_to_corrupt;
+ char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
+
+ if (np) {
+ char *ptr2 = kmap(np) + off;
+
+ memcpy(ptr2, ptr, len);
+ memcpy(ptr2, "bad4", min(4, len));
+ kunmap(np);
+ desc->bd_iov[i].kiov_page = np;
+ } else {
+ CERROR("%s: can't alloc page for corruption\n",
+ tgt_name(tgt));
+ }
+ }
+ }
+
+ bufsize = sizeof(cksum);
+ err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
+
+ return cksum;
+}
+
+int tgt_brw_read(struct tgt_session_info *tsi)
+{
+ struct ptlrpc_request *req = tgt_ses_req(tsi);
+ struct ptlrpc_bulk_desc *desc = NULL;
+ struct obd_export *exp = tsi->tsi_exp;
+ struct niobuf_remote *remote_nb;
+ struct niobuf_local *local_nb;
+ struct obd_ioobj *ioo;
+ struct ost_body *body, *repbody;
+ struct l_wait_info lwi;
+ struct lustre_handle lockh = { 0 };
+ int npages, nob = 0, rc, i, no_reply = 0;
+ struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
+
+ ENTRY;
+
+ if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
+ CERROR("%s: deny read request from %s to portal %u\n",
+ tgt_name(tsi->tsi_tgt),
+ obd_export_nid2str(req->rq_export),
+ ptlrpc_req2svc(req)->srv_req_portal);
+ RETURN(-EPROTO);
+ }
+
+ req->rq_bulk_read = 1;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
+ RETURN(-EIO);
+
+ OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
+ cfs_fail_val : (obd_timeout + 1) / 4);
+
+ /* Check if there is eviction in progress, and if so, wait for it to
+ * finish */
+ if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
+ /* We do not care how long it takes */
+ lwi = LWI_INTR(NULL, NULL);
+ rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
+ !atomic_read(&exp->exp_obd->obd_evict_inprogress),
+ &lwi);
+ }
+
+ /* There must be big cache in current thread to process this request
+ * if it is NULL then something went wrong and it wasn't allocated,
+ * report -ENOMEM in that case */
+ if (tbc == NULL)
+ RETURN(-ENOMEM);
+
+ body = tsi->tsi_ost_body;
+ LASSERT(body != NULL);
+
+ ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
+ LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
+
+ remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
+ LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
+
+ local_nb = tbc->local;
+
+ rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
+ remote_nb, &lockh, LCK_PR);
+ if (rc != 0)
+ RETURN(rc);
+
+ /*
+ * If getting the lock took more time than
+ * client was willing to wait, drop it. b=11330
+ */
+ if (cfs_time_current_sec() > req->rq_deadline ||
+ OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
+ no_reply = 1;
+ CERROR("Dropping timed-out read from %s because locking"
+ "object "DOSTID" took %ld seconds (limit was %ld).\n",
+ libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
+ cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
+ req->rq_deadline - req->rq_arrival_time.tv_sec);
+ GOTO(out_lock, rc = -ETIMEDOUT);
+ }
+
+ repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+ repbody->oa = body->oa;
+
+ npages = PTLRPC_MAX_BRW_PAGES;
+ rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
+ ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
+ if (rc != 0)
+ GOTO(out_lock, rc);
+
+ desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
+ BULK_PUT_SOURCE, OST_BULK_PORTAL);
+ if (desc == NULL)
+ GOTO(out_commitrw, rc = -ENOMEM);
+
+ nob = 0;
+ for (i = 0; i < npages; i++) {
+ int page_rc = local_nb[i].lnb_rc;
+
+ if (page_rc < 0) {
+ rc = page_rc;
+ break;
+ }
+
+ nob += page_rc;
+ if (page_rc != 0) { /* some data! */
+ LASSERT(local_nb[i].lnb_page != NULL);
+ ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
+ local_nb[i].lnb_page_offset,
+ page_rc);
+ }
+
+ if (page_rc != local_nb[i].lnb_len) { /* short read */
+ /* All subsequent pages should be 0 */
+ while (++i < npages)
+ LASSERT(local_nb[i].lnb_rc == 0);
+ break;
+ }
+ }
+
+ if (body->oa.o_valid & OBD_MD_FLCKSUM) {
+ cksum_type_t cksum_type =
+ cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
+ body->oa.o_flags : 0);
+ repbody->oa.o_flags = cksum_type_pack(cksum_type);
+ repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
+ repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
+ OST_READ, cksum_type);
+ CDEBUG(D_PAGE, "checksum at read origin: %x\n",
+ repbody->oa.o_cksum);
+ } else {
+ repbody->oa.o_valid = 0;
+ }
+ /* We're finishing using body->oa as an input variable */
+
+ /* Check if client was evicted while we were doing i/o before touching
+ * network */
+ if (likely(rc == 0 &&
+ !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
+ rc = target_bulk_io(exp, desc, &lwi);
+ no_reply = rc != 0;
+ }
+
+out_commitrw:
+ /* Must commit after prep above in all cases */
+ rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
+ &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
+ NULL, rc);
+ if (rc == 0)
+ tgt_drop_id(exp, &repbody->oa);
+out_lock:
+ tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
+
+ if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
+ ptlrpc_free_bulk_nopin(desc);
+
+ LASSERT(rc <= 0);
+ if (rc == 0) {
+ rc = nob;
+ ptlrpc_lprocfs_brw(req, nob);
+ } else if (no_reply) {
+ req->rq_no_reply = 1;
+ /* reply out callback would free */
+ ptlrpc_req_drop_rs(req);
+ LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
+ "client will retry: rc %d\n",
+ exp->exp_obd->obd_name,
+ obd_uuid2str(&exp->exp_client_uuid),
+ obd_export_nid2str(exp), rc);
+ }
+ /* send a bulk after reply to simulate a network delay or reordering
+ * by a router */
+ if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
+ wait_queue_head_t waitq;
+ struct l_wait_info lwi1;
+
+ CDEBUG(D_INFO, "reorder BULK\n");
+ init_waitqueue_head(&waitq);
+
+ lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
+ l_wait_event(waitq, 0, &lwi1);
+ target_bulk_io(exp, desc, &lwi);
+ ptlrpc_free_bulk_nopin(desc);
+ }
+
+ RETURN(rc);
+}
+EXPORT_SYMBOL(tgt_brw_read);
+
+static void tgt_warn_on_cksum(struct ptlrpc_request *req,
+ struct ptlrpc_bulk_desc *desc,
+ struct niobuf_local *local_nb, int npages,
+ u32 client_cksum, u32 server_cksum,
+ bool mmap)
+{
+ struct obd_export *exp = req->rq_export;
+ struct ost_body *body;
+ char *router;
+ char *via;
+
+ body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
+ LASSERT(body != NULL);
+
+ if (req->rq_peer.nid == desc->bd_sender) {
+ via = router = "";
+ } else {
+ via = " via ";
+ router = libcfs_nid2str(desc->bd_sender);
+ }
+
+ if (mmap) {
+ CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
+ client_cksum, server_cksum);
+ return;
+ }
+
+ LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
+ DFID" object "DOSTID" extent ["LPU64"-"LPU64
+ "]: client csum %x, server csum %x\n",
+ exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
+ via, router,
+ body->oa.o_valid & OBD_MD_FLFID ?
+ body->oa.o_parent_seq : (__u64)0,
+ body->oa.o_valid & OBD_MD_FLFID ?
+ body->oa.o_parent_oid : 0,
+ body->oa.o_valid & OBD_MD_FLFID ?
+ body->oa.o_parent_ver : 0,
+ POSTID(&body->oa.o_oi),
+ local_nb[0].lnb_file_offset,
+ local_nb[npages-1].lnb_file_offset +
+ local_nb[npages - 1].lnb_len - 1,
+ client_cksum, server_cksum);
+}
+
+int tgt_brw_write(struct tgt_session_info *tsi)
+{
+ struct ptlrpc_request *req = tgt_ses_req(tsi);
+ struct ptlrpc_bulk_desc *desc = NULL;
+ struct obd_export *exp = req->rq_export;
+ struct niobuf_remote *remote_nb;
+ struct niobuf_local *local_nb;
+ struct obd_ioobj *ioo;
+ struct ost_body *body, *repbody;
+ struct l_wait_info lwi;
+ struct lustre_handle lockh = {0};
+ __u32 *rcs;
+ int objcount, niocount, npages;
+ int rc, i, j;
+ cksum_type_t cksum_type = OBD_CKSUM_CRC32;
+ bool no_reply = false, mmap;
+ struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
+
+ ENTRY;
+
+ if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
+ CERROR("%s: deny write request from %s to portal %u\n",
+ tgt_name(tsi->tsi_tgt),
+ obd_export_nid2str(req->rq_export),
+ ptlrpc_req2svc(req)->srv_req_portal);
+ RETURN(err_serious(-EPROTO));
+ }
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
+ RETURN(err_serious(-ENOSPC));
+ if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
+ RETURN(err_serious(-EROFS));
+
+ req->rq_bulk_write = 1;
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
+ RETURN(err_serious(-EIO));
+ if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
+ RETURN(err_serious(-EFAULT));
+
+ /* pause before transaction has been started */
+ CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
+ cfs_fail_val : (obd_timeout + 1) / 4);
+
+ /* There must be big cache in current thread to process this request
+ * if it is NULL then something went wrong and it wasn't allocated,
+ * report -ENOMEM in that case */
+ if (tbc == NULL)
+ RETURN(-ENOMEM);
+
+ body = tsi->tsi_ost_body;
+ LASSERT(body != NULL);
+
+ ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
+ LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
+
+ objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
+ RCL_CLIENT) / sizeof(*ioo);
+
+ for (niocount = i = 0; i < objcount; i++)
+ niocount += ioo[i].ioo_bufcnt;
+
+ remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
+ LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
+ if (niocount != req_capsule_get_size(&req->rq_pill,
+ &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
+ sizeof(*remote_nb))
+ RETURN(err_serious(-EPROTO));
+
+ if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
+ (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
+ memory_pressure_set();
+
+ req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
+ niocount * sizeof(*rcs));
+ rc = req_capsule_server_pack(&req->rq_pill);
+ if (rc != 0)
+ GOTO(out, rc = err_serious(rc));
+
+ CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
+ rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
+
+ local_nb = tbc->local;
+
+ rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
+ remote_nb, &lockh, LCK_PW);
+ if (rc != 0)
+ GOTO(out, rc);
+
+ /*
+ * If getting the lock took more time than
+ * client was willing to wait, drop it. b=11330
+ */
+ if (cfs_time_current_sec() > req->rq_deadline ||
+ OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
+ no_reply = true;
+ CERROR("%s: Dropping timed-out write from %s because locking "
+ "object "DOSTID" took %ld seconds (limit was %ld).\n",
+ tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
+ POSTID(&ioo->ioo_oid),
+ cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
+ req->rq_deadline - req->rq_arrival_time.tv_sec);
+ GOTO(out_lock, rc = -ETIMEDOUT);
+ }
+
+ /* Because we already sync grant info with client when reconnect,
+ * grant info will be cleared for resent req, then fed_grant and
+ * total_grant will not be modified in following preprw_write */
+ if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
+ DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
+ body->oa.o_valid &= ~OBD_MD_FLGRANT;
+ }
+
+ repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+ if (repbody == NULL)
+ GOTO(out_lock, rc = -ENOMEM);
+ repbody->oa = body->oa;
+
+ npages = PTLRPC_MAX_BRW_PAGES;
+ rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
+ objcount, ioo, remote_nb, &npages, local_nb, NULL,
+ BYPASS_CAPA);
+ if (rc < 0)
+ GOTO(out_lock, rc);
+
+ desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
+ BULK_GET_SINK, OST_BULK_PORTAL);
+ if (desc == NULL)
+ GOTO(skip_transfer, rc = -ENOMEM);
+
+ /* NB Having prepped, we must commit... */
+ for (i = 0; i < npages; i++)
+ ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
+ local_nb[i].lnb_page_offset,
+ local_nb[i].lnb_len);
+
+ rc = sptlrpc_svc_prep_bulk(req, desc);
+ if (rc != 0)
+ GOTO(skip_transfer, rc);
+
+ rc = target_bulk_io(exp, desc, &lwi);
+ no_reply = rc != 0;
+
+skip_transfer:
+ if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
+ static int cksum_counter;
+
+ if (body->oa.o_valid & OBD_MD_FLFLAGS)
+ cksum_type = cksum_type_unpack(body->oa.o_flags);
+
+ repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
+ repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
+ repbody->oa.o_flags |= cksum_type_pack(cksum_type);
+ repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
+ OST_WRITE, cksum_type);
+ cksum_counter++;
+
+ if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
+ mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
+ body->oa.o_flags & OBD_FL_MMAP);
+
+ tgt_warn_on_cksum(req, desc, local_nb, npages,
+ body->oa.o_cksum,
+ repbody->oa.o_cksum, mmap);
+ cksum_counter = 0;
+ } else if ((cksum_counter & (-cksum_counter)) ==
+ cksum_counter) {
+ CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
+ cksum_counter, libcfs_id2str(req->rq_peer),
+ repbody->oa.o_cksum);
+ }
+ }
+
+ /* Must commit after prep above in all cases */
+ rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
+ objcount, ioo, remote_nb, npages, local_nb, NULL,
+ rc);
+ if (rc == -ENOTCONN)
+ /* quota acquire process has been given up because
+ * either the client has been evicted or the client
+ * has timed out the request already */
+ no_reply = true;
+
+ /*
+ * Disable sending mtime back to the client. If the client locked the
+ * whole object, then it has already updated the mtime on its side,
+ * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
+ */
+ repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
+
+ if (rc == 0) {
+ int nob = 0;
+
+ /* set per-requested niobuf return codes */
+ for (i = j = 0; i < niocount; i++) {
+ int len = remote_nb[i].rnb_len;
+
+ nob += len;
+ rcs[i] = 0;
+ do {
+ LASSERT(j < npages);
+ if (local_nb[j].lnb_rc < 0)
+ rcs[i] = local_nb[j].lnb_rc;
+ len -= local_nb[j].lnb_len;
+ j++;
+ } while (len > 0);
+ LASSERT(len == 0);
+ }
+ LASSERT(j == npages);
+ ptlrpc_lprocfs_brw(req, nob);
+
+ tgt_drop_id(exp, &repbody->oa);
+ }
+out_lock:
+ tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
+ if (desc)
+ ptlrpc_free_bulk_nopin(desc);
+out:
+ if (no_reply) {
+ req->rq_no_reply = 1;
+ /* reply out callback would free */
+ ptlrpc_req_drop_rs(req);
+ LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
+ "client will retry: rc %d\n",
+ exp->exp_obd->obd_name,
+ obd_uuid2str(&exp->exp_client_uuid),
+ obd_export_nid2str(exp), rc);
+ }
+ memory_pressure_clr();
+ RETURN(rc);
+}
+EXPORT_SYMBOL(tgt_brw_write);