4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2013, 2016, Intel Corporation.
27 * lustre/target/tgt_handler.c
29 * Lustre Unified Target request handler code
31 * Author: Brian Behlendorf <behlendorf1@llnl.gov>
32 * Author: Mikhail Pershin <mike.pershin@intel.com>
35 #define DEBUG_SUBSYSTEM S_CLASS
37 #include <linux/user_namespace.h>
38 #ifdef HAVE_UIDGID_HEADER
39 # include <linux/uidgid.h>
43 #include <obd_class.h>
44 #include <obd_cksum.h>
45 #include <lustre_lfsck.h>
46 #include <lustre_nodemap.h>
48 #include "tgt_internal.h"
50 char *tgt_name(struct lu_target *tgt)
52 LASSERT(tgt->lut_obd != NULL);
53 return tgt->lut_obd->obd_name;
55 EXPORT_SYMBOL(tgt_name);
58 * Generic code handling requests that have struct mdt_body passed in:
60 * - extract mdt_body from request and save it in @tsi, if present;
62 * - create lu_object, corresponding to the fid in mdt_body, and save it in
65 * - if HABEO_CORPUS flag is set for this request type check whether object
66 * actually exists on storage (lu_object_exists()).
69 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
71 const struct mdt_body *body;
72 struct lu_object *obj;
73 struct req_capsule *pill = tsi->tsi_pill;
78 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
82 tsi->tsi_mdt_body = body;
84 if (!(body->mbo_valid & OBD_MD_FLID))
87 /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
88 * in any case in pre-2.5 clients. Fix that here if needed */
89 if (unlikely(fid_is_zero(&body->mbo_fid1)))
92 if (!fid_is_sane(&body->mbo_fid1)) {
93 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
94 PFID(&body->mbo_fid1));
98 obj = lu_object_find(tsi->tsi_env,
99 &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
100 &body->mbo_fid1, NULL);
102 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
103 lu_object_put(tsi->tsi_env, obj);
106 tsi->tsi_corpus = obj;
113 tsi->tsi_fid = body->mbo_fid1;
119 * Validate oa from client.
120 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
122 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
123 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
124 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
125 * whether this is oi_fid or real ostid. So it will check
126 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
127 * c. Old FID-disable osc will send IDIF.
128 * d. new FID-enable osc/osp will send normal FID.
130 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
131 * be used for LAST_ID file, and only being accessed inside OST now.
133 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
135 struct ost_id *oi = &oa->o_oi;
136 u64 seq = ostid_seq(oi);
137 u64 id = ostid_id(oi);
141 if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
142 fid_seq_is_echo(seq))) {
143 /* Sigh 2.[123] client still sends echo req with oi_id = 0
144 * during create, and we will reset this to 1, since this
145 * oi_id is basically useless in the following create process,
146 * but oi_id == 0 will make it difficult to tell whether it is
147 * real FID or ost_id. */
148 oi->oi_fid.f_seq = FID_SEQ_ECHO;
149 oi->oi_fid.f_oid = id ?: 1;
150 oi->oi_fid.f_ver = 0;
152 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
154 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
155 GOTO(out, rc = -EPROTO);
157 /* Note: this check might be forced in 2.5 or 2.6, i.e.
158 * all of the requests are required to setup FLGROUP */
159 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
160 ostid_set_seq_mdt0(oi);
161 oa->o_valid |= OBD_MD_FLGROUP;
165 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
166 fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
167 GOTO(out, rc = -EPROTO);
169 rc = ostid_to_fid(&tti->tti_fid1, oi,
170 tsi->tsi_tgt->lut_lsd.lsd_osd_index);
171 if (unlikely(rc != 0))
174 oi->oi_fid = tti->tti_fid1;
180 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
181 tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
185 EXPORT_SYMBOL(tgt_validate_obdo);
187 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
190 struct niobuf_remote *rnb;
191 struct obd_ioobj *ioo;
196 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
200 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
204 max_brw = ioobj_max_brw_get(ioo);
205 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
206 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
207 ": rc = %d\n", tgt_name(tsi->tsi_tgt),
208 obd_export_nid2str(tsi->tsi_exp), max_brw,
209 POSTID(oi), -EPROTO);
214 obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
215 RCL_CLIENT) / sizeof(*ioo);
216 if (obj_count == 0) {
217 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
219 } else if (obj_count > 1) {
220 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
225 if (ioo->ioo_bufcnt == 0) {
226 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
230 if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
231 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
232 "bulk has too many pages (%d)",
240 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
242 struct ost_body *body;
243 struct req_capsule *pill = tsi->tsi_pill;
244 struct lu_nodemap *nodemap;
249 body = req_capsule_client_get(pill, &RMF_OST_BODY);
253 rc = tgt_validate_obdo(tsi, &body->oa);
257 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
259 RETURN(PTR_ERR(nodemap));
261 body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
262 NODEMAP_CLIENT_TO_FS,
264 body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
265 NODEMAP_CLIENT_TO_FS,
267 nodemap_putref(nodemap);
269 tsi->tsi_ost_body = body;
270 tsi->tsi_fid = body->oa.o_oi.oi_fid;
272 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
273 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
278 if (!(body->oa.o_valid & OBD_MD_FLID)) {
279 if (flags & HABEO_CORPUS) {
280 CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
281 "but OID/FID is mandatory with HABEO_CORPUS\n",
282 tgt_name(tsi->tsi_tgt));
289 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
292 * OST doesn't get object in advance for further use to prevent
293 * situations with nested object_find which is potential deadlock.
295 tsi->tsi_corpus = NULL;
300 * Do necessary preprocessing according to handler ->th_flags.
302 static int tgt_request_preprocess(struct tgt_session_info *tsi,
303 struct tgt_handler *h,
304 struct ptlrpc_request *req)
306 struct req_capsule *pill = tsi->tsi_pill;
307 __u32 flags = h->th_flags;
312 if (tsi->tsi_preprocessed)
315 LASSERT(h->th_act != NULL);
316 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
317 LASSERT(current->journal_info == NULL);
319 LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
321 if (h->th_fmt != NULL) {
322 req_capsule_set(pill, h->th_fmt);
323 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
324 rc = tgt_mdt_body_unpack(tsi, flags);
327 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
329 rc = tgt_ost_body_unpack(tsi, flags);
335 if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
338 if (flags & HABEO_CLAVIS) {
339 struct ldlm_request *dlm_req;
341 LASSERT(h->th_fmt != NULL);
343 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
344 if (dlm_req != NULL) {
345 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
347 dlm_req->lock_desc.l_policy_data.\
348 l_inodebits.bits == 0)) {
350 * Lock without inodebits makes no sense and
351 * will oops later in ldlm. If client miss to
352 * set such bits, do not trigger ASSERTION.
354 * For liblustre flock case, it maybe zero.
358 tsi->tsi_dlm_req = dlm_req;
364 tsi->tsi_preprocessed = 1;
369 * Invoke handler for this request opc. Also do necessary preprocessing
370 * (according to handler ->th_flags), and post-processing (setting of
371 * ->last_{xid,committed}).
373 static int tgt_handle_request0(struct tgt_session_info *tsi,
374 struct tgt_handler *h,
375 struct ptlrpc_request *req)
379 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
384 /* When dealing with sec context requests, no export is associated yet,
385 * because these requests are sent before *_CONNECT requests.
386 * A NULL req->rq_export means the normal *_common_slice handlers will
387 * not be called, because there is no reference to the target.
388 * So deal with them by hand and jump directly to target_send_reply().
392 case SEC_CTX_INIT_CONT:
394 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
399 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
400 * to put same checks into handlers like mdt_close(), mdt_reint(),
401 * etc., without talking to mdt authors first. Checking same thing
402 * there again is useless and returning 0 error without packing reply
403 * is buggy! Handlers either pack reply or return error.
405 * We return 0 here and do not send any reply in order to emulate
406 * network failure. Do not send any reply in case any of NET related
407 * fail_id has occured.
409 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
411 if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
412 OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
415 rc = tgt_request_preprocess(tsi, h, req);
416 /* pack reply if reply format is fixed */
417 if (rc == 0 && h->th_flags & HABEO_REFERO) {
419 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
421 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
423 tsi->tsi_mdt_body->mbo_eadatasize);
424 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
426 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
429 rc = req_capsule_server_pack(tsi->tsi_pill);
432 if (likely(rc == 0)) {
434 * Process request, there can be two types of rc:
435 * 1) errors with msg unpack/pack, other failures outside the
436 * operation itself. This is counted as serious errors;
437 * 2) errors during fs operation, should be placed in rq_status
441 if (!is_serious(rc) &&
442 !req->rq_no_reply && req->rq_reply_state == NULL) {
443 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
444 "pack reply and returned 0 error\n",
445 tgt_name(tsi->tsi_tgt), h->th_name);
448 serious = is_serious(rc);
449 rc = clear_serious(rc);
457 * ELDLM_* codes which > 0 should be in rq_status only as well as
458 * all non-serious errors.
460 if (rc > 0 || !serious)
463 LASSERT(current->journal_info == NULL);
465 if (likely(rc == 0 && req->rq_export))
466 target_committed_to_req(req);
469 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
473 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
474 struct obd_device *obd, int *process)
476 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
483 case MDS_SYNC: /* used in unmounting */
497 case MDS_HSM_PROGRESS:
498 case MDS_HSM_STATE_SET:
499 case MDS_HSM_REQUEST:
500 *process = target_queue_recovery_request(req, obd);
504 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
511 * Handle recovery. Return:
512 * +1: continue request processing;
513 * -ve: abort immediately with the given error code;
514 * 0: send reply with error code in req->rq_status;
516 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
520 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
525 case SEC_CTX_INIT_CONT:
530 if (!req->rq_export->exp_obd->obd_replayable)
533 /* sanity check: if the xid matches, the request must be marked as a
534 * resent or replayed */
535 if (req_can_reconstruct(req, NULL)) {
536 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
537 (MSG_RESENT | MSG_REPLAY))) {
538 DEBUG_REQ(D_WARNING, req, "rq_xid %llu matches "
539 "saved xid, expected REPLAY or RESENT flag "
541 lustre_msg_get_flags(req->rq_reqmsg));
542 req->rq_status = -ENOTCONN;
546 /* else: note the opposite is not always true; a RESENT req after a
547 * failover will usually not match the last_xid, since it was likely
548 * never committed. A REPLAYed request will almost never match the
549 * last xid, however it could for a committed, but still retained,
552 /* Check for aborted recovery... */
553 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
557 DEBUG_REQ(D_INFO, req, "Got new replay");
558 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
560 if (rc != 0 || !should_process)
562 else if (should_process < 0) {
563 req->rq_status = should_process;
564 rc = ptlrpc_error(req);
571 /* Initial check for request, it is validation mostly */
572 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
574 struct tgt_handler *h;
575 struct tgt_opc_slice *s;
576 struct lu_target *tgt;
577 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
581 tgt = class_exp2tgt(req->rq_export);
582 if (unlikely(tgt == NULL)) {
583 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
584 class_exp2obd(req->rq_export)->obd_name);
585 RETURN(ERR_PTR(-EINVAL));
588 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
589 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
592 /* opcode was not found in slice */
593 if (unlikely(s->tos_hs == NULL)) {
594 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
596 RETURN(ERR_PTR(-ENOTSUPP));
599 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
600 h = s->tos_hs + (opc - s->tos_opc_start);
601 if (unlikely(h->th_opc == 0)) {
602 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
603 RETURN(ERR_PTR(-ENOTSUPP));
609 static int process_req_last_xid(struct ptlrpc_request *req)
614 /* check request's xid is consistent with export's last_xid */
615 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
616 if (last_xid > req->rq_export->exp_last_xid)
617 req->rq_export->exp_last_xid = last_xid;
619 if (req->rq_xid == 0 ||
620 (req->rq_xid <= req->rq_export->exp_last_xid)) {
621 DEBUG_REQ(D_ERROR, req, "Unexpected xid %llx vs. "
622 "last_xid %llx\n", req->rq_xid,
623 req->rq_export->exp_last_xid);
624 /* Some request is allowed to be sent during replay,
625 * such as OUT update requests, FLD requests, so it
626 * is possible that replay requests has smaller XID
627 * than the exp_last_xid.
629 * Some non-replay requests may have smaller XID as
632 * - Client send a no_resend RPC, like statfs;
633 * - The RPC timedout (or some other error) on client,
634 * then it's removed from the unreplied list;
635 * - Client send some other request to bump the
636 * exp_last_xid on server;
637 * - The former RPC got chance to be processed;
639 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
643 /* try to release in-memory reply data */
644 if (tgt_is_multimodrpcs_client(req->rq_export)) {
645 tgt_handle_received_xid(req->rq_export,
646 lustre_msg_get_last_xid(req->rq_reqmsg));
647 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
648 (MSG_RESENT | MSG_REPLAY)))
649 tgt_handle_tag(req->rq_export,
650 lustre_msg_get_tag(req->rq_reqmsg));
655 int tgt_request_handle(struct ptlrpc_request *req)
657 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
659 struct lustre_msg *msg = req->rq_reqmsg;
660 struct tgt_handler *h;
661 struct lu_target *tgt;
662 int request_fail_id = 0;
663 __u32 opc = lustre_msg_get_opc(msg);
664 struct obd_device *obd;
666 bool is_connect = false;
669 /* Refill the context, to make sure all thread keys are allocated */
670 lu_env_refill(req->rq_svc_thread->t_env);
672 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
673 tsi->tsi_pill = &req->rq_pill;
674 tsi->tsi_env = req->rq_svc_thread->t_env;
676 /* if request has export then get handlers slice from corresponding
677 * target, otherwise that should be connect operation */
678 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
679 opc == MGS_CONNECT) {
681 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
682 rc = target_handle_connect(req);
684 rc = ptlrpc_error(req);
687 /* recovery-small test 18c asks to drop connect reply */
688 if (unlikely(opc == OST_CONNECT &&
689 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
693 if (unlikely(!class_connected_export(req->rq_export))) {
694 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
695 opc == SEC_CTX_FINI) {
696 /* sec context initialization has to be handled
697 * by hand in tgt_handle_request0() */
698 tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
700 GOTO(handle_recov, rc = 0);
702 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
703 opc, libcfs_id2str(req->rq_peer));
704 req->rq_status = -ENOTCONN;
705 rc = ptlrpc_error(req);
709 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
710 tsi->tsi_exp = req->rq_export;
711 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
712 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
714 tsi->tsi_jobid = NULL;
717 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
718 class_exp2obd(req->rq_export)->obd_name);
719 req->rq_status = -EINVAL;
720 rc = ptlrpc_error(req);
724 /* Skip last_xid processing for the recovery thread, otherwise, the
725 * last_xid on same request could be processed twice: first time when
726 * processing the incoming request, second time when the request is
727 * being processed by recovery thread. */
728 obd = class_exp2obd(req->rq_export);
730 /* reset the exp_last_xid on each connection. */
731 req->rq_export->exp_last_xid = 0;
732 } else if (obd->obd_recovery_data.trd_processing_task !=
734 rc = process_req_last_xid(req);
737 rc = ptlrpc_error(req);
742 request_fail_id = tgt->lut_request_fail_id;
743 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
745 h = tgt_handler_find_check(req);
747 req->rq_status = PTR_ERR(h);
748 rc = ptlrpc_error(req);
752 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
755 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
758 rc = lustre_msg_check_version(msg, h->th_version);
760 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
761 " %08x, expecting %08x\n", tgt_name(tgt),
762 lustre_msg_get_version(msg), h->th_version);
763 req->rq_status = -EINVAL;
764 rc = ptlrpc_error(req);
769 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
770 if (likely(rc == 1)) {
771 rc = tgt_handle_request0(tsi, h, req);
777 req_capsule_fini(tsi->tsi_pill);
778 if (tsi->tsi_corpus != NULL) {
779 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
780 tsi->tsi_corpus = NULL;
784 EXPORT_SYMBOL(tgt_request_handle);
786 /** Assign high priority operations to the request if needed. */
787 int tgt_hpreq_handler(struct ptlrpc_request *req)
789 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
790 struct tgt_handler *h;
795 if (req->rq_export == NULL)
798 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
799 tsi->tsi_pill = &req->rq_pill;
800 tsi->tsi_env = req->rq_svc_thread->t_env;
801 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
802 tsi->tsi_exp = req->rq_export;
804 h = tgt_handler_find_check(req);
810 rc = tgt_request_preprocess(tsi, h, req);
811 if (unlikely(rc != 0))
814 if (h->th_hp != NULL)
818 EXPORT_SYMBOL(tgt_hpreq_handler);
820 void tgt_counter_incr(struct obd_export *exp, int opcode)
822 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
823 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
824 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
826 EXPORT_SYMBOL(tgt_counter_incr);
829 * Unified target generic handlers.
832 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
834 struct lu_target *tgt = class_exp2tgt(exp);
835 struct sptlrpc_flavor flvr;
839 LASSERT(tgt->lut_obd);
840 LASSERT(tgt->lut_slice);
842 /* always allow ECHO client */
843 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
844 LUSTRE_ECHO_NAME) == 0)) {
845 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
849 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
850 read_lock(&tgt->lut_sptlrpc_lock);
851 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
855 read_unlock(&tgt->lut_sptlrpc_lock);
857 spin_lock(&exp->exp_lock);
858 exp->exp_sp_peer = req->rq_sp_from;
859 exp->exp_flvr = flvr;
861 /* when on mgs, if no restriction is set, or if client
862 * is loopback, allow any flavor */
863 if ((strcmp(exp->exp_obd->obd_type->typ_name,
864 LUSTRE_MGS_NAME) == 0) &&
865 (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
866 LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
868 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
870 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
871 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
872 CERROR("%s: unauthorized rpc flavor %x from %s, "
873 "expect %x\n", tgt_name(tgt),
875 libcfs_nid2str(req->rq_peer.nid),
876 exp->exp_flvr.sf_rpc);
879 spin_unlock(&exp->exp_lock);
881 if (exp->exp_sp_peer != req->rq_sp_from) {
882 CERROR("%s: RPC source %s doesn't match %s\n",
884 sptlrpc_part2name(req->rq_sp_from),
885 sptlrpc_part2name(exp->exp_sp_peer));
888 rc = sptlrpc_target_export_check(exp, req);
895 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt)
897 struct sptlrpc_rule_set tmp_rset;
900 if (unlikely(tgt == NULL)) {
901 CERROR("No target passed");
905 sptlrpc_rule_set_init(&tmp_rset);
906 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset);
908 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
913 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
915 write_lock(&tgt->lut_sptlrpc_lock);
916 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
917 tgt->lut_sptlrpc_rset = tmp_rset;
918 write_unlock(&tgt->lut_sptlrpc_lock);
922 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
924 int tgt_connect(struct tgt_session_info *tsi)
926 struct ptlrpc_request *req = tgt_ses_req(tsi);
927 struct obd_connect_data *reply;
932 /* XXX: better to call this check right after getting new export but
933 * before last_rcvd slot allocation to avoid server load upon insecure
934 * connects. This is to be fixed after unifiyng all targets.
936 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
940 /* To avoid exposing partially initialized connection flags, changes up
941 * to this point have been staged in reply->ocd_connect_flags. Now that
942 * connection handling has completed successfully, atomically update
943 * the connect flags in the shared export data structure. LU-1623 */
944 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
945 spin_lock(&tsi->tsi_exp->exp_lock);
946 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
947 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
948 spin_unlock(&tsi->tsi_exp->exp_lock);
952 obd_disconnect(class_export_get(tsi->tsi_exp));
955 EXPORT_SYMBOL(tgt_connect);
957 int tgt_disconnect(struct tgt_session_info *tsi)
963 rc = target_handle_disconnect(tgt_ses_req(tsi));
965 RETURN(err_serious(rc));
969 EXPORT_SYMBOL(tgt_disconnect);
972 * Unified target OBD handlers
974 int tgt_obd_ping(struct tgt_session_info *tsi)
980 rc = target_handle_ping(tgt_ses_req(tsi));
982 RETURN(err_serious(rc));
986 EXPORT_SYMBOL(tgt_obd_ping);
988 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
990 return err_serious(-EOPNOTSUPP);
993 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
995 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
996 struct ptlrpc_request *req = tgt_ses_req(tsi);
997 struct obd_export *exp = req->rq_export;
998 struct ptlrpc_bulk_desc *desc;
999 struct l_wait_info *lwi = &tti->tti_u.update.tti_wait_info;
1005 desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1006 PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1007 MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1011 for (i = 0; i < rdbuf->rb_nbufs; i++)
1012 desc->bd_frag_ops->add_iov_frag(desc,
1013 rdbuf->rb_bufs[i].lb_buf,
1014 rdbuf->rb_bufs[i].lb_len);
1016 rc = target_bulk_io(exp, desc, lwi);
1017 ptlrpc_free_bulk(desc);
1020 EXPORT_SYMBOL(tgt_send_buffer);
1022 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1024 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1025 struct ptlrpc_request *req = tgt_ses_req(tsi);
1026 struct obd_export *exp = req->rq_export;
1027 struct ptlrpc_bulk_desc *desc;
1028 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1036 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1037 PTLRPC_BULK_PUT_SOURCE |
1038 PTLRPC_BULK_BUF_KIOV,
1040 &ptlrpc_bulk_kiov_pin_ops);
1044 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1045 /* old client requires reply size in it's PAGE_SIZE,
1046 * which is rdpg->rp_count */
1047 nob = rdpg->rp_count;
1049 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1050 i++, tmpcount -= tmpsize) {
1051 tmpsize = min_t(int, tmpcount, PAGE_SIZE);
1052 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1056 LASSERT(desc->bd_nob == nob);
1057 rc = target_bulk_io(exp, desc, lwi);
1058 ptlrpc_free_bulk(desc);
1061 EXPORT_SYMBOL(tgt_sendpage);
1064 * OBD_IDX_READ handler
1066 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1068 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1069 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1070 struct idx_info *req_ii, *rep_ii;
1075 memset(rdpg, 0, sizeof(*rdpg));
1076 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1078 /* extract idx_info buffer from request & reply */
1079 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1080 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1081 RETURN(err_serious(-EPROTO));
1083 rc = req_capsule_server_pack(tsi->tsi_pill);
1085 RETURN(err_serious(rc));
1087 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1089 RETURN(err_serious(-EFAULT));
1090 rep_ii->ii_magic = IDX_INFO_MAGIC;
1092 /* extract hash to start with */
1093 rdpg->rp_hash = req_ii->ii_hash_start;
1095 /* extract requested attributes */
1096 rdpg->rp_attrs = req_ii->ii_attrs;
1098 /* check that fid packed in request is valid and supported */
1099 if (!fid_is_sane(&req_ii->ii_fid))
1101 rep_ii->ii_fid = req_ii->ii_fid;
1104 rep_ii->ii_flags = req_ii->ii_flags;
1106 /* compute number of pages to allocate, ii_count is the number of 4KB
1108 if (req_ii->ii_count <= 0)
1109 GOTO(out, rc = -EFAULT);
1110 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1111 exp_max_brw_size(tsi->tsi_exp));
1112 rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >> PAGE_SHIFT;
1114 /* allocate pages to store the containers */
1115 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1116 if (rdpg->rp_pages == NULL)
1117 GOTO(out, rc = -ENOMEM);
1118 for (i = 0; i < rdpg->rp_npages; i++) {
1119 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1120 if (rdpg->rp_pages[i] == NULL)
1121 GOTO(out, rc = -ENOMEM);
1124 /* populate pages with key/record pairs */
1125 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1129 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1130 "asked %d > %d\n", rc, rdpg->rp_count);
1132 /* send pages to client */
1133 rc = tgt_sendpage(tsi, rdpg, rc);
1138 if (rdpg->rp_pages) {
1139 for (i = 0; i < rdpg->rp_npages; i++)
1140 if (rdpg->rp_pages[i])
1141 __free_page(rdpg->rp_pages[i]);
1142 OBD_FREE(rdpg->rp_pages,
1143 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1148 struct tgt_handler tgt_obd_handlers[] = {
1149 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1150 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1151 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1153 EXPORT_SYMBOL(tgt_obd_handlers);
1155 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1156 struct dt_object *obj, __u64 start, __u64 end)
1162 /* if no objid is specified, it means "sync whole filesystem" */
1164 rc = dt_sync(env, tgt->lut_bottom);
1165 } else if (dt_version_get(env, obj) >
1166 tgt->lut_obd->obd_last_committed) {
1167 rc = dt_object_sync(env, obj, start, end);
1169 atomic_inc(&tgt->lut_sync_count);
1173 EXPORT_SYMBOL(tgt_sync);
1175 * Unified target DLM handlers.
1179 * Unified target BAST
1181 * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1182 * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1183 * but if it's cross-MDT lock, because remote object version is not set, a
1184 * filesystem sync is needed.
1186 * \param lock server side lock
1187 * \param desc lock desc
1188 * \param data ldlm_cb_set_arg
1189 * \param flag indicates whether this cancelling or blocking callback
1190 * \retval 0 on success
1191 * \retval negative number on error
1193 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1194 void *data, int flag)
1197 struct lu_target *tgt;
1198 struct dt_object *obj = NULL;
1204 tgt = class_exp2tgt(lock->l_export);
1206 if (unlikely(tgt == NULL)) {
1207 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1208 class_exp2obd(lock->l_export)->obd_name);
1212 if (flag == LDLM_CB_CANCELING &&
1213 (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1214 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1215 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1216 ldlm_is_cbpending(lock))) &&
1217 ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1218 lock->l_resource->lr_type == LDLM_EXTENT)) {
1220 __u64 end = OBD_OBJECT_EOF;
1222 rc = lu_env_init(&env, LCT_DT_THREAD);
1223 if (unlikely(rc != 0))
1226 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1227 tgt->lut_lsd.lsd_osd_index);
1229 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1230 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1232 GOTO(err_env, rc = PTR_ERR(obj));
1234 if (!dt_object_exists(obj))
1235 GOTO(err_put, rc = -ENOENT);
1237 start = lock->l_policy_data.l_extent.start;
1238 end = lock->l_policy_data.l_extent.end;
1241 rc = tgt_sync(&env, tgt, obj, start, end);
1243 CERROR("%s: syncing "DFID" (%llu-%llu) on lock "
1244 "cancel: rc = %d\n",
1245 tgt_name(tgt), PFID(&fid),
1246 lock->l_policy_data.l_extent.start,
1247 lock->l_policy_data.l_extent.end, rc);
1251 dt_object_put(&env, obj);
1256 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1260 static struct ldlm_callback_suite tgt_dlm_cbs = {
1261 .lcs_completion = ldlm_server_completion_ast,
1262 .lcs_blocking = tgt_blocking_ast,
1263 .lcs_glimpse = ldlm_server_glimpse_ast
1266 int tgt_enqueue(struct tgt_session_info *tsi)
1268 struct ptlrpc_request *req = tgt_ses_req(tsi);
1273 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1274 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1276 LASSERT(tsi->tsi_dlm_req != NULL);
1277 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1278 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1280 RETURN(err_serious(rc));
1282 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1284 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1287 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1290 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1293 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1296 RETURN(req->rq_status);
1298 EXPORT_SYMBOL(tgt_enqueue);
1300 int tgt_convert(struct tgt_session_info *tsi)
1302 struct ptlrpc_request *req = tgt_ses_req(tsi);
1306 LASSERT(tsi->tsi_dlm_req);
1307 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1309 RETURN(err_serious(rc));
1311 RETURN(req->rq_status);
1314 int tgt_bl_callback(struct tgt_session_info *tsi)
1316 return err_serious(-EOPNOTSUPP);
1319 int tgt_cp_callback(struct tgt_session_info *tsi)
1321 return err_serious(-EOPNOTSUPP);
1324 /* generic LDLM target handler */
1325 struct tgt_handler tgt_dlm_handlers[] = {
1326 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1327 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1328 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1329 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1331 EXPORT_SYMBOL(tgt_dlm_handlers);
1334 * Unified target LLOG handlers.
1336 int tgt_llog_open(struct tgt_session_info *tsi)
1342 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1346 EXPORT_SYMBOL(tgt_llog_open);
1348 int tgt_llog_close(struct tgt_session_info *tsi)
1354 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1358 EXPORT_SYMBOL(tgt_llog_close);
1361 int tgt_llog_destroy(struct tgt_session_info *tsi)
1367 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1372 int tgt_llog_read_header(struct tgt_session_info *tsi)
1378 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1382 EXPORT_SYMBOL(tgt_llog_read_header);
1384 int tgt_llog_next_block(struct tgt_session_info *tsi)
1390 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1394 EXPORT_SYMBOL(tgt_llog_next_block);
1396 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1402 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1406 EXPORT_SYMBOL(tgt_llog_prev_block);
1408 /* generic llog target handler */
1409 struct tgt_handler tgt_llog_handlers[] = {
1410 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1411 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1412 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1413 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1414 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1415 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1417 EXPORT_SYMBOL(tgt_llog_handlers);
1420 * sec context handlers
1422 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1423 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1428 struct tgt_handler tgt_sec_ctx_handlers[] = {
1429 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1430 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1431 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1433 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1435 int (*tgt_lfsck_in_notify_local)(const struct lu_env *env,
1436 struct dt_device *key,
1437 struct lfsck_req_local *lrl,
1438 struct thandle *th) = NULL;
1440 void tgt_register_lfsck_in_notify_local(int (*notify)(const struct lu_env *,
1442 struct lfsck_req_local *,
1445 tgt_lfsck_in_notify_local = notify;
1447 EXPORT_SYMBOL(tgt_register_lfsck_in_notify_local);
1449 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1450 struct dt_device *key,
1451 struct lfsck_request *lr) = NULL;
1453 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1455 struct lfsck_request *))
1457 tgt_lfsck_in_notify = notify;
1459 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1461 static int (*tgt_lfsck_query)(const struct lu_env *env,
1462 struct dt_device *key,
1463 struct lfsck_request *req,
1464 struct lfsck_reply *rep,
1465 struct lfsck_query *que) = NULL;
1467 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1469 struct lfsck_request *,
1470 struct lfsck_reply *,
1471 struct lfsck_query *))
1473 tgt_lfsck_query = query;
1475 EXPORT_SYMBOL(tgt_register_lfsck_query);
1477 /* LFSCK request handlers */
1478 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1480 const struct lu_env *env = tsi->tsi_env;
1481 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1482 struct lfsck_request *lr;
1486 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1490 rc = tgt_lfsck_in_notify(env, key, lr);
1495 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1497 struct lfsck_request *request;
1498 struct lfsck_reply *reply;
1502 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1503 if (request == NULL)
1506 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1510 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1511 request, reply, NULL);
1513 RETURN(rc < 0 ? rc : 0);
1516 struct tgt_handler tgt_lfsck_handlers[] = {
1517 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1518 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1520 EXPORT_SYMBOL(tgt_lfsck_handlers);
1523 * initialize per-thread page pool (bug 5137).
1525 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1527 struct tgt_thread_big_cache *tbc;
1531 LASSERT(thread != NULL);
1532 LASSERT(thread->t_data == NULL);
1534 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1537 thread->t_data = tbc;
1540 EXPORT_SYMBOL(tgt_io_thread_init);
1543 * free per-thread pool created by tgt_thread_init().
1545 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1547 struct tgt_thread_big_cache *tbc;
1551 LASSERT(thread != NULL);
1554 * be prepared to handle partially-initialized pools (because this is
1555 * called from ost_io_thread_init() for cleanup.
1557 tbc = thread->t_data;
1559 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1560 thread->t_data = NULL;
1564 EXPORT_SYMBOL(tgt_io_thread_done);
1566 * Helper function for getting server side [start, start+count] DLM lock
1567 * if asked by client.
1569 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1570 __u64 start, __u64 end, struct lustre_handle *lh,
1571 int mode, __u64 *flags)
1573 union ldlm_policy_data policy;
1578 LASSERT(lh != NULL);
1579 LASSERT(ns != NULL);
1580 LASSERT(!lustre_handle_is_used(lh));
1582 policy.l_extent.gid = 0;
1583 policy.l_extent.start = start & PAGE_MASK;
1586 * If ->o_blocks is EOF it means "lock till the end of the file".
1587 * Otherwise, it's size of an extent or hole being punched (in bytes).
1589 if (end == OBD_OBJECT_EOF || end < start)
1590 policy.l_extent.end = OBD_OBJECT_EOF;
1592 policy.l_extent.end = end | ~PAGE_MASK;
1594 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1595 flags, ldlm_blocking_ast,
1596 ldlm_completion_ast, ldlm_glimpse_ast,
1597 NULL, 0, LVB_T_NONE, NULL, lh);
1598 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1600 EXPORT_SYMBOL(tgt_extent_lock);
1602 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1604 LASSERT(lustre_handle_is_used(lh));
1605 ldlm_lock_decref(lh, mode);
1607 EXPORT_SYMBOL(tgt_extent_unlock);
1609 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1610 struct obd_ioobj *obj, struct niobuf_remote *nb,
1611 struct lustre_handle *lh, enum ldlm_mode mode)
1614 int nrbufs = obj->ioo_bufcnt;
1619 LASSERT(mode == LCK_PR || mode == LCK_PW);
1620 LASSERT(!lustre_handle_is_used(lh));
1622 if (ns->ns_obd->obd_recovering)
1625 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1628 for (i = 1; i < nrbufs; i++)
1629 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1632 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1633 nb[nrbufs - 1].rnb_offset +
1634 nb[nrbufs - 1].rnb_len - 1,
1638 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1639 struct lustre_handle *lh, enum ldlm_mode mode)
1643 LASSERT(mode == LCK_PR || mode == LCK_PW);
1644 LASSERT((obj->ioo_bufcnt > 0 &&
1645 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1646 lustre_handle_is_used(lh));
1648 if (lustre_handle_is_used(lh))
1649 tgt_extent_unlock(lh, mode);
1653 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1654 struct ptlrpc_bulk_desc *desc, int opc,
1655 cksum_type_t cksum_type)
1657 struct cfs_crypto_hash_desc *hdesc;
1658 unsigned int bufsize;
1660 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1663 LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1665 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1666 if (IS_ERR(hdesc)) {
1667 CERROR("%s: unable to initialize checksum hash %s\n",
1668 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1669 return PTR_ERR(hdesc);
1672 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1673 for (i = 0; i < desc->bd_iov_count; i++) {
1674 /* corrupt the data before we compute the checksum, to
1675 * simulate a client->OST data error */
1676 if (i == 0 && opc == OST_WRITE &&
1677 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1678 int off = BD_GET_KIOV(desc, i).kiov_offset &
1680 int len = BD_GET_KIOV(desc, i).kiov_len;
1681 struct page *np = tgt_page_to_corrupt;
1682 char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1685 char *ptr2 = kmap(np) + off;
1687 memcpy(ptr2, ptr, len);
1688 memcpy(ptr2, "bad3", min(4, len));
1690 BD_GET_KIOV(desc, i).kiov_page = np;
1692 CERROR("%s: can't alloc page for corruption\n",
1696 cfs_crypto_hash_update_page(hdesc,
1697 BD_GET_KIOV(desc, i).kiov_page,
1698 BD_GET_KIOV(desc, i).kiov_offset &
1700 BD_GET_KIOV(desc, i).kiov_len);
1702 /* corrupt the data after we compute the checksum, to
1703 * simulate an OST->client data error */
1704 if (i == 0 && opc == OST_READ &&
1705 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1706 int off = BD_GET_KIOV(desc, i).kiov_offset
1708 int len = BD_GET_KIOV(desc, i).kiov_len;
1709 struct page *np = tgt_page_to_corrupt;
1711 kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1714 char *ptr2 = kmap(np) + off;
1716 memcpy(ptr2, ptr, len);
1717 memcpy(ptr2, "bad4", min(4, len));
1719 BD_GET_KIOV(desc, i).kiov_page = np;
1721 CERROR("%s: can't alloc page for corruption\n",
1727 bufsize = sizeof(cksum);
1728 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1733 int tgt_brw_read(struct tgt_session_info *tsi)
1735 struct ptlrpc_request *req = tgt_ses_req(tsi);
1736 struct ptlrpc_bulk_desc *desc = NULL;
1737 struct obd_export *exp = tsi->tsi_exp;
1738 struct niobuf_remote *remote_nb;
1739 struct niobuf_local *local_nb;
1740 struct obd_ioobj *ioo;
1741 struct ost_body *body, *repbody;
1742 struct l_wait_info lwi;
1743 struct lustre_handle lockh = { 0 };
1744 int npages, nob = 0, rc, i, no_reply = 0;
1745 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1749 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1750 CERROR("%s: deny read request from %s to portal %u\n",
1751 tgt_name(tsi->tsi_tgt),
1752 obd_export_nid2str(req->rq_export),
1753 ptlrpc_req2svc(req)->srv_req_portal);
1757 req->rq_bulk_read = 1;
1759 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1762 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1763 cfs_fail_val : (obd_timeout + 1) / 4);
1765 /* Check if there is eviction in progress, and if so, wait for it to
1767 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1768 /* We do not care how long it takes */
1769 lwi = LWI_INTR(NULL, NULL);
1770 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1771 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1775 /* There must be big cache in current thread to process this request
1776 * if it is NULL then something went wrong and it wasn't allocated,
1777 * report -ENOMEM in that case */
1781 body = tsi->tsi_ost_body;
1782 LASSERT(body != NULL);
1784 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1785 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1787 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1788 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1790 local_nb = tbc->local;
1792 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1793 remote_nb, &lockh, LCK_PR);
1798 * If getting the lock took more time than
1799 * client was willing to wait, drop it. b=11330
1801 if (cfs_time_current_sec() > req->rq_deadline ||
1802 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1804 CERROR("Dropping timed-out read from %s because locking"
1805 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1806 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1807 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1808 req->rq_deadline - req->rq_arrival_time.tv_sec);
1809 GOTO(out_lock, rc = -ETIMEDOUT);
1812 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1813 repbody->oa = body->oa;
1815 npages = PTLRPC_MAX_BRW_PAGES;
1816 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1817 ioo, remote_nb, &npages, local_nb);
1821 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1822 PTLRPC_BULK_PUT_SOURCE |
1823 PTLRPC_BULK_BUF_KIOV,
1825 &ptlrpc_bulk_kiov_nopin_ops);
1827 GOTO(out_commitrw, rc = -ENOMEM);
1830 for (i = 0; i < npages; i++) {
1831 int page_rc = local_nb[i].lnb_rc;
1839 if (page_rc != 0) { /* some data! */
1840 LASSERT(local_nb[i].lnb_page != NULL);
1841 desc->bd_frag_ops->add_kiov_frag
1842 (desc, local_nb[i].lnb_page,
1843 local_nb[i].lnb_page_offset,
1847 if (page_rc != local_nb[i].lnb_len) { /* short read */
1848 /* All subsequent pages should be 0 */
1849 while (++i < npages)
1850 LASSERT(local_nb[i].lnb_rc == 0);
1854 if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) &&
1855 nob != cfs_fail_val)
1858 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1859 cksum_type_t cksum_type =
1860 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1861 body->oa.o_flags : 0);
1862 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1863 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1864 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1865 OST_READ, cksum_type);
1866 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1867 repbody->oa.o_cksum);
1869 repbody->oa.o_valid = 0;
1871 /* We're finishing using body->oa as an input variable */
1873 /* Check if client was evicted while we were doing i/o before touching
1875 if (likely(rc == 0 &&
1876 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2) &&
1877 !CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_BULK))) {
1878 rc = target_bulk_io(exp, desc, &lwi);
1883 /* Must commit after prep above in all cases */
1884 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1885 remote_nb, npages, local_nb, rc);
1887 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1889 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1890 ptlrpc_free_bulk(desc);
1895 ptlrpc_lprocfs_brw(req, nob);
1896 } else if (no_reply) {
1897 req->rq_no_reply = 1;
1898 /* reply out callback would free */
1899 ptlrpc_req_drop_rs(req);
1900 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1901 "client will retry: rc %d\n",
1902 exp->exp_obd->obd_name,
1903 obd_uuid2str(&exp->exp_client_uuid),
1904 obd_export_nid2str(exp), rc);
1906 /* send a bulk after reply to simulate a network delay or reordering
1908 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1909 wait_queue_head_t waitq;
1910 struct l_wait_info lwi1;
1912 CDEBUG(D_INFO, "reorder BULK\n");
1913 init_waitqueue_head(&waitq);
1915 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1916 l_wait_event(waitq, 0, &lwi1);
1917 target_bulk_io(exp, desc, &lwi);
1918 ptlrpc_free_bulk(desc);
1923 EXPORT_SYMBOL(tgt_brw_read);
1925 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1926 struct ptlrpc_bulk_desc *desc,
1927 struct niobuf_local *local_nb, int npages,
1928 u32 client_cksum, u32 server_cksum,
1931 struct obd_export *exp = req->rq_export;
1932 struct ost_body *body;
1936 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1937 LASSERT(body != NULL);
1939 if (req->rq_peer.nid != desc->bd_sender) {
1941 router = libcfs_nid2str(desc->bd_sender);
1945 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1946 client_cksum, server_cksum);
1950 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1951 DFID" object "DOSTID" extent [%llu-%llu"
1952 "]: client csum %x, server csum %x\n",
1953 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1955 body->oa.o_valid & OBD_MD_FLFID ?
1956 body->oa.o_parent_seq : (__u64)0,
1957 body->oa.o_valid & OBD_MD_FLFID ?
1958 body->oa.o_parent_oid : 0,
1959 body->oa.o_valid & OBD_MD_FLFID ?
1960 body->oa.o_parent_ver : 0,
1961 POSTID(&body->oa.o_oi),
1962 local_nb[0].lnb_file_offset,
1963 local_nb[npages-1].lnb_file_offset +
1964 local_nb[npages - 1].lnb_len - 1,
1965 client_cksum, server_cksum);
1968 int tgt_brw_write(struct tgt_session_info *tsi)
1970 struct ptlrpc_request *req = tgt_ses_req(tsi);
1971 struct ptlrpc_bulk_desc *desc = NULL;
1972 struct obd_export *exp = req->rq_export;
1973 struct niobuf_remote *remote_nb;
1974 struct niobuf_local *local_nb;
1975 struct obd_ioobj *ioo;
1976 struct ost_body *body, *repbody;
1977 struct l_wait_info lwi;
1978 struct lustre_handle lockh = {0};
1980 int objcount, niocount, npages;
1982 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1983 bool no_reply = false, mmap;
1984 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1985 bool wait_sync = false;
1989 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1990 CERROR("%s: deny write request from %s to portal %u\n",
1991 tgt_name(tsi->tsi_tgt),
1992 obd_export_nid2str(req->rq_export),
1993 ptlrpc_req2svc(req)->srv_req_portal);
1994 RETURN(err_serious(-EPROTO));
1997 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1998 RETURN(err_serious(-ENOSPC));
1999 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2000 RETURN(err_serious(-EROFS));
2002 req->rq_bulk_write = 1;
2004 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2005 RETURN(err_serious(-EIO));
2006 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2007 RETURN(err_serious(-EFAULT));
2009 /* pause before transaction has been started */
2010 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2011 cfs_fail_val : (obd_timeout + 1) / 4);
2013 /* There must be big cache in current thread to process this request
2014 * if it is NULL then something went wrong and it wasn't allocated,
2015 * report -ENOMEM in that case */
2019 body = tsi->tsi_ost_body;
2020 LASSERT(body != NULL);
2022 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2023 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2025 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2026 RCL_CLIENT) / sizeof(*ioo);
2028 for (niocount = i = 0; i < objcount; i++)
2029 niocount += ioo[i].ioo_bufcnt;
2031 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2032 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2033 if (niocount != req_capsule_get_size(&req->rq_pill,
2034 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2036 RETURN(err_serious(-EPROTO));
2038 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2039 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2040 memory_pressure_set();
2042 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2043 niocount * sizeof(*rcs));
2044 rc = req_capsule_server_pack(&req->rq_pill);
2046 GOTO(out, rc = err_serious(rc));
2048 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2049 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2051 local_nb = tbc->local;
2053 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2054 remote_nb, &lockh, LCK_PW);
2059 * If getting the lock took more time than
2060 * client was willing to wait, drop it. b=11330
2062 if (cfs_time_current_sec() > req->rq_deadline ||
2063 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2065 CERROR("%s: Dropping timed-out write from %s because locking "
2066 "object "DOSTID" took %ld seconds (limit was %ld).\n",
2067 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2068 POSTID(&ioo->ioo_oid),
2069 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2070 req->rq_deadline - req->rq_arrival_time.tv_sec);
2071 GOTO(out_lock, rc = -ETIMEDOUT);
2074 /* Because we already sync grant info with client when reconnect,
2075 * grant info will be cleared for resent req, then fed_grant and
2076 * total_grant will not be modified in following preprw_write */
2077 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2078 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2079 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2082 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2083 if (repbody == NULL)
2084 GOTO(out_lock, rc = -ENOMEM);
2085 repbody->oa = body->oa;
2087 npages = PTLRPC_MAX_BRW_PAGES;
2088 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2089 objcount, ioo, remote_nb, &npages, local_nb);
2093 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2094 PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2096 &ptlrpc_bulk_kiov_nopin_ops);
2098 GOTO(skip_transfer, rc = -ENOMEM);
2100 /* NB Having prepped, we must commit... */
2101 for (i = 0; i < npages; i++)
2102 desc->bd_frag_ops->add_kiov_frag(desc,
2103 local_nb[i].lnb_page,
2104 local_nb[i].lnb_page_offset,
2105 local_nb[i].lnb_len);
2107 rc = sptlrpc_svc_prep_bulk(req, desc);
2109 GOTO(skip_transfer, rc);
2111 rc = target_bulk_io(exp, desc, &lwi);
2115 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2116 static int cksum_counter;
2118 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2119 cksum_type = cksum_type_unpack(body->oa.o_flags);
2121 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2122 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2123 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2124 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2125 OST_WRITE, cksum_type);
2128 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2129 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2130 body->oa.o_flags & OBD_FL_MMAP);
2132 tgt_warn_on_cksum(req, desc, local_nb, npages,
2134 repbody->oa.o_cksum, mmap);
2136 } else if ((cksum_counter & (-cksum_counter)) ==
2138 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2139 cksum_counter, libcfs_id2str(req->rq_peer),
2140 repbody->oa.o_cksum);
2144 /* Must commit after prep above in all cases */
2145 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2146 objcount, ioo, remote_nb, npages, local_nb, rc);
2147 if (rc == -ENOTCONN)
2148 /* quota acquire process has been given up because
2149 * either the client has been evicted or the client
2150 * has timed out the request already */
2153 for (i = 0; i < niocount; i++) {
2154 if (!(local_nb[i].lnb_flags & OBD_BRW_ASYNC)) {
2160 * Disable sending mtime back to the client. If the client locked the
2161 * whole object, then it has already updated the mtime on its side,
2162 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2164 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2169 /* set per-requested niobuf return codes */
2170 for (i = j = 0; i < niocount; i++) {
2171 int len = remote_nb[i].rnb_len;
2176 LASSERT(j < npages);
2177 if (local_nb[j].lnb_rc < 0)
2178 rcs[i] = local_nb[j].lnb_rc;
2179 len -= local_nb[j].lnb_len;
2184 LASSERT(j == npages);
2185 ptlrpc_lprocfs_brw(req, nob);
2188 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2190 ptlrpc_free_bulk(desc);
2192 if (unlikely(no_reply || (exp->exp_obd->obd_no_transno && wait_sync))) {
2193 req->rq_no_reply = 1;
2194 /* reply out callback would free */
2195 ptlrpc_req_drop_rs(req);
2196 if (!exp->exp_obd->obd_no_transno)
2197 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s),"
2198 " client will retry: rc = %d\n",
2199 exp->exp_obd->obd_name,
2200 obd_uuid2str(&exp->exp_client_uuid),
2201 obd_export_nid2str(exp), rc);
2203 memory_pressure_clr();
2206 EXPORT_SYMBOL(tgt_brw_write);
2208 /* Check if request can be reconstructed from saved reply data
2209 * A copy of the reply data is returned in @trd if the pointer is not NULL
2211 bool req_can_reconstruct(struct ptlrpc_request *req,
2212 struct tg_reply_data *trd)
2214 struct tg_export_data *ted = &req->rq_export->exp_target_data;
2215 struct lsd_client_data *lcd = ted->ted_lcd;
2218 if (tgt_is_multimodrpcs_client(req->rq_export))
2219 return tgt_lookup_reply(req, trd);
2221 mutex_lock(&ted->ted_lcd_lock);
2222 found = req->rq_xid == lcd->lcd_last_xid ||
2223 req->rq_xid == lcd->lcd_last_close_xid;
2225 if (found && trd != NULL) {
2226 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2227 trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2228 trd->trd_reply.lrd_transno =
2229 lcd->lcd_last_close_transno;
2230 trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2232 trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2233 trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2234 trd->trd_reply.lrd_result = lcd->lcd_last_result;
2235 trd->trd_reply.lrd_data = lcd->lcd_last_data;
2236 trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2237 trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2238 trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2239 trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2242 mutex_unlock(&ted->ted_lcd_lock);
2246 EXPORT_SYMBOL(req_can_reconstruct);