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, 2017, 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 #include <linux/delay.h>
39 #include <linux/uidgid.h>
41 #include <libcfs/linux/linux-mem.h>
43 #include <obd_class.h>
44 #include <obd_cksum.h>
45 #include <lustre_lfsck.h>
46 #include <lustre_nodemap.h>
47 #include <lustre_acl.h>
49 #include "tgt_internal.h"
51 char *tgt_name(struct lu_target *tgt)
53 LASSERT(tgt->lut_obd != NULL);
54 return tgt->lut_obd->obd_name;
56 EXPORT_SYMBOL(tgt_name);
59 * Generic code handling requests that have struct mdt_body passed in:
61 * - extract mdt_body from request and save it in @tsi, if present;
63 * - create lu_object, corresponding to the fid in mdt_body, and save it in
66 * - if HAS_BODY flag is set for this request type check whether object
67 * actually exists on storage (lu_object_exists()).
70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
72 const struct mdt_body *body;
73 struct lu_object *obj;
74 struct req_capsule *pill = tsi->tsi_pill;
79 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
83 tsi->tsi_mdt_body = body;
85 if (!(body->mbo_valid & OBD_MD_FLID))
88 /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
89 * in any case in pre-2.5 clients. Fix that here if needed */
90 if (unlikely(fid_is_zero(&body->mbo_fid1)))
93 if (!fid_is_sane(&body->mbo_fid1)) {
94 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
95 PFID(&body->mbo_fid1));
99 obj = lu_object_find(tsi->tsi_env,
100 &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
101 &body->mbo_fid1, NULL);
103 if ((flags & HAS_BODY) && !lu_object_exists(obj)) {
104 lu_object_put(tsi->tsi_env, obj);
107 tsi->tsi_corpus = obj;
114 tsi->tsi_fid = body->mbo_fid1;
120 * Validate oa from client.
121 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
123 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
124 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
125 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
126 * whether this is oi_fid or real ostid. So it will check
127 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
128 * c. Old FID-disable osc will send IDIF.
129 * d. new FID-enable osc/osp will send normal FID.
131 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
132 * be used for LAST_ID file, and only being accessed inside OST now.
134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
136 struct ost_id *oi = &oa->o_oi;
137 u64 seq = ostid_seq(oi);
138 u64 id = ostid_id(oi);
142 if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
143 fid_seq_is_echo(seq))) {
144 /* Sigh 2.[123] client still sends echo req with oi_id = 0
145 * during create, and we will reset this to 1, since this
146 * oi_id is basically useless in the following create process,
147 * but oi_id == 0 will make it difficult to tell whether it is
148 * real FID or ost_id. */
149 oi->oi_fid.f_seq = FID_SEQ_ECHO;
150 oi->oi_fid.f_oid = id ?: 1;
151 oi->oi_fid.f_ver = 0;
153 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
155 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
156 GOTO(out, rc = -EPROTO);
158 /* Note: this check might be forced in 2.5 or 2.6, i.e.
159 * all of the requests are required to setup FLGROUP */
160 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
161 ostid_set_seq_mdt0(oi);
162 oa->o_valid |= OBD_MD_FLGROUP;
166 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
167 fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
168 GOTO(out, rc = -EPROTO);
170 rc = ostid_to_fid(&tti->tti_fid1, oi,
171 tsi->tsi_tgt->lut_lsd.lsd_osd_index);
172 if (unlikely(rc != 0))
175 oi->oi_fid = tti->tti_fid1;
181 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
182 tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
186 EXPORT_SYMBOL(tgt_validate_obdo);
188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
191 struct niobuf_remote *rnb;
192 struct obd_ioobj *ioo;
197 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
201 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
205 max_brw = ioobj_max_brw_get(ioo);
206 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
207 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
208 ": rc = %d\n", tgt_name(tsi->tsi_tgt),
209 obd_export_nid2str(tsi->tsi_exp), max_brw,
210 POSTID(oi), -EPROTO);
215 obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
216 RCL_CLIENT) / sizeof(*ioo);
217 if (obj_count == 0) {
218 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
220 } else if (obj_count > 1) {
221 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
226 if (ioo->ioo_bufcnt == 0) {
227 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
231 if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
232 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
233 "bulk has too many pages (%d)",
241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
243 struct ost_body *body;
244 struct req_capsule *pill = tsi->tsi_pill;
245 struct lu_nodemap *nodemap;
250 body = req_capsule_client_get(pill, &RMF_OST_BODY);
254 rc = tgt_validate_obdo(tsi, &body->oa);
258 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
260 RETURN(PTR_ERR(nodemap));
262 body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
263 NODEMAP_CLIENT_TO_FS,
265 body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
266 NODEMAP_CLIENT_TO_FS,
268 body->oa.o_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
269 NODEMAP_CLIENT_TO_FS,
271 nodemap_putref(nodemap);
273 tsi->tsi_ost_body = body;
274 tsi->tsi_fid = body->oa.o_oi.oi_fid;
276 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
277 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
282 if (!(body->oa.o_valid & OBD_MD_FLID)) {
283 if (flags & HAS_BODY) {
284 CERROR("%s: OBD_MD_FLID flag is not set in ost_body but OID/FID is mandatory with HAS_BODY\n",
285 tgt_name(tsi->tsi_tgt));
292 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
295 * OST doesn't get object in advance for further use to prevent
296 * situations with nested object_find which is potential deadlock.
298 tsi->tsi_corpus = NULL;
303 * Do necessary preprocessing according to handler ->th_flags.
305 static int tgt_request_preprocess(struct tgt_session_info *tsi,
306 struct tgt_handler *h,
307 struct ptlrpc_request *req)
309 struct req_capsule *pill = tsi->tsi_pill;
310 __u32 flags = h->th_flags;
315 if (tsi->tsi_preprocessed)
318 LASSERT(h->th_act != NULL);
319 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
320 LASSERT(current->journal_info == NULL);
322 LASSERT(ergo(flags & (HAS_BODY | HAS_REPLY),
324 if (h->th_fmt != NULL) {
325 req_capsule_set(pill, h->th_fmt);
326 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT) &&
327 req_capsule_field_present(pill, &RMF_MDT_BODY,
329 rc = tgt_mdt_body_unpack(tsi, flags);
332 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
334 req_capsule_field_present(pill, &RMF_OST_BODY,
336 rc = tgt_ost_body_unpack(tsi, flags);
342 if (flags & IS_MUTABLE && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
345 if (flags & HAS_KEY) {
346 struct ldlm_request *dlm_req;
348 LASSERT(h->th_fmt != NULL);
350 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
351 if (dlm_req != NULL) {
352 union ldlm_wire_policy_data *policy =
353 &dlm_req->lock_desc.l_policy_data;
355 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
357 (policy->l_inodebits.bits |
358 policy->l_inodebits.try_bits) == 0)) {
360 * Lock without inodebits makes no sense and
361 * will oops later in ldlm. If client miss to
362 * set such bits, do not trigger ASSERTION.
364 * For liblustre flock case, it maybe zero.
368 tsi->tsi_dlm_req = dlm_req;
374 tsi->tsi_preprocessed = 1;
379 * Invoke handler for this request opc. Also do necessary preprocessing
380 * (according to handler ->th_flags), and post-processing (setting of
381 * ->last_{xid,committed}).
383 static int tgt_handle_request0(struct tgt_session_info *tsi,
384 struct tgt_handler *h,
385 struct ptlrpc_request *req)
389 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
394 /* When dealing with sec context requests, no export is associated yet,
395 * because these requests are sent before *_CONNECT requests.
396 * A NULL req->rq_export means the normal *_common_slice handlers will
397 * not be called, because there is no reference to the target.
398 * So deal with them by hand and jump directly to target_send_reply().
402 case SEC_CTX_INIT_CONT:
404 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
409 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
410 * to put same checks into handlers like mdt_close(), mdt_reint(),
411 * etc., without talking to mdt authors first. Checking same thing
412 * there again is useless and returning 0 error without packing reply
413 * is buggy! Handlers either pack reply or return error.
415 * We return 0 here and do not send any reply in order to emulate
416 * network failure. Do not send any reply in case any of NET related
417 * fail_id has occured.
419 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
421 if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
422 OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
425 /* drop OUT_UPDATE rpc */
426 if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == OUT_UPDATE &&
427 OBD_FAIL_CHECK(OBD_FAIL_OUT_UPDATE_DROP)))
430 rc = tgt_request_preprocess(tsi, h, req);
431 /* pack reply if reply format is fixed */
432 if (rc == 0 && h->th_flags & HAS_REPLY) {
434 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
436 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
438 tsi->tsi_mdt_body->mbo_eadatasize);
439 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
441 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
443 if (req_capsule_has_field(tsi->tsi_pill, &RMF_ACL, RCL_SERVER))
444 req_capsule_set_size(tsi->tsi_pill,
445 &RMF_ACL, RCL_SERVER,
446 LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
448 if (req_capsule_has_field(tsi->tsi_pill, &RMF_SHORT_IO,
450 struct niobuf_remote *remote_nb =
451 req_capsule_client_get(tsi->tsi_pill,
453 struct ost_body *body = tsi->tsi_ost_body;
455 req_capsule_set_size(tsi->tsi_pill, &RMF_SHORT_IO,
457 (body->oa.o_valid & OBD_MD_FLFLAGS &&
458 body->oa.o_flags & OBD_FL_SHORT_IO) ?
459 remote_nb[0].rnb_len : 0);
462 rc = req_capsule_server_pack(tsi->tsi_pill);
465 if (likely(rc == 0)) {
467 * Process request, there can be two types of rc:
468 * 1) errors with msg unpack/pack, other failures outside the
469 * operation itself. This is counted as serious errors;
470 * 2) errors during fs operation, should be placed in rq_status
474 if (!is_serious(rc) &&
475 !req->rq_no_reply && req->rq_reply_state == NULL) {
476 DEBUG_REQ(D_ERROR, req,
477 "%s: %s handler did not pack reply but returned no error",
478 tgt_name(tsi->tsi_tgt), h->th_name);
481 serious = is_serious(rc);
482 rc = clear_serious(rc);
490 * ELDLM_* codes which > 0 should be in rq_status only as well as
491 * all non-serious errors.
493 if (rc > 0 || !serious)
496 LASSERT(current->journal_info == NULL);
498 if (likely(rc == 0 && req->rq_export))
499 target_committed_to_req(req);
502 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
506 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
507 struct obd_device *obd, int *process)
509 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
516 case MDS_SYNC: /* used in unmounting */
530 case MDS_HSM_PROGRESS:
531 case MDS_HSM_STATE_SET:
532 case MDS_HSM_REQUEST:
534 *process = target_queue_recovery_request(req, obd);
538 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
545 * Handle recovery. Return:
546 * +1: continue request processing;
547 * -ve: abort immediately with the given error code;
548 * 0: send reply with error code in req->rq_status;
550 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
554 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
559 case SEC_CTX_INIT_CONT:
564 if (!req->rq_export->exp_obd->obd_replayable)
567 /* sanity check: if the xid matches, the request must be marked as a
568 * resent or replayed */
569 if (req_can_reconstruct(req, NULL) == 1) {
570 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
571 (MSG_RESENT | MSG_REPLAY))) {
572 DEBUG_REQ(D_WARNING, req,
573 "rq_xid=%llu matches saved XID, expected REPLAY or RESENT flag (%x)",
575 lustre_msg_get_flags(req->rq_reqmsg));
576 req->rq_status = -ENOTCONN;
580 /* else: note the opposite is not always true; a RESENT req after a
581 * failover will usually not match the last_xid, since it was likely
582 * never committed. A REPLAYed request will almost never match the
583 * last xid, however it could for a committed, but still retained,
586 /* Check for aborted recovery... */
587 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
591 DEBUG_REQ(D_INFO, req, "Got new replay");
592 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
594 if (rc != 0 || !should_process)
596 else if (should_process < 0) {
597 req->rq_status = should_process;
598 rc = ptlrpc_error(req);
605 /* Initial check for request, it is validation mostly */
606 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
608 struct tgt_handler *h;
609 struct tgt_opc_slice *s;
610 struct lu_target *tgt;
611 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
615 tgt = class_exp2tgt(req->rq_export);
616 if (unlikely(tgt == NULL)) {
617 DEBUG_REQ(D_ERROR, req, "%s: no target for connected export",
618 class_exp2obd(req->rq_export)->obd_name);
619 RETURN(ERR_PTR(-EINVAL));
622 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
623 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
626 /* opcode was not found in slice */
627 if (unlikely(s->tos_hs == NULL)) {
630 /* don't spew error messages for unhandled RPCs */
632 CERROR("%s: no handler for opcode 0x%x from %s\n",
633 tgt_name(tgt), opc, libcfs_id2str(req->rq_peer));
636 RETURN(ERR_PTR(-ENOTSUPP));
639 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
640 h = s->tos_hs + (opc - s->tos_opc_start);
641 if (unlikely(h->th_opc == 0)) {
642 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
643 RETURN(ERR_PTR(-ENOTSUPP));
649 static int process_req_last_xid(struct ptlrpc_request *req)
653 struct obd_export *exp = req->rq_export;
654 struct tg_export_data *ted = &exp->exp_target_data;
655 bool need_lock = tgt_is_multimodrpcs_client(exp);
659 mutex_lock(&ted->ted_lcd_lock);
660 /* check request's xid is consistent with export's last_xid */
661 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
662 if (last_xid > exp->exp_last_xid)
663 exp->exp_last_xid = last_xid;
665 if (req->rq_xid == 0 || req->rq_xid <= exp->exp_last_xid) {
666 /* Some request is allowed to be sent during replay,
667 * such as OUT update requests, FLD requests, so it
668 * is possible that replay requests has smaller XID
669 * than the exp_last_xid.
671 * Some non-replay requests may have smaller XID as
674 * - Client send a no_resend RPC, like statfs;
675 * - The RPC timedout (or some other error) on client,
676 * then it's removed from the unreplied list;
677 * - Client send some other request to bump the
678 * exp_last_xid on server;
679 * - The former RPC got chance to be processed;
681 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
684 DEBUG_REQ(D_WARNING, req,
685 "unexpected xid=%llx != exp_last_xid=%llx, rc = %d",
686 req->rq_xid, exp->exp_last_xid, rc);
691 /* The "last_xid" is the minimum xid among unreplied requests,
692 * if the request is from the previous connection, its xid can
693 * still be larger than "exp_last_xid", then the above check of
694 * xid is not enough to determine whether the request is delayed.
696 * For example, if some replay request was delayed and caused
697 * timeout at client and the replay is restarted, the delayed
698 * replay request will have the larger xid than "exp_last_xid"
700 if (req->rq_export->exp_conn_cnt >
701 lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
703 "Dropping request %llu from an old epoch %u/%u\n",
705 lustre_msg_get_conn_cnt(req->rq_reqmsg),
706 req->rq_export->exp_conn_cnt);
707 req->rq_no_reply = 1;
708 GOTO(out, rc = -ESTALE);
711 /* try to release in-memory reply data */
712 if (tgt_is_multimodrpcs_client(exp)) {
713 tgt_handle_received_xid(exp, last_xid);
714 rc = tgt_handle_tag(req);
719 mutex_unlock(&ted->ted_lcd_lock);
724 int tgt_request_handle(struct ptlrpc_request *req)
726 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
728 struct lustre_msg *msg = req->rq_reqmsg;
729 struct tgt_handler *h;
730 struct lu_target *tgt;
731 int request_fail_id = 0;
732 __u32 opc = lustre_msg_get_opc(msg);
733 struct obd_device *obd;
735 bool is_connect = false;
738 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_TGT_RECOVERY_REQ_RACE))) {
739 if (cfs_fail_val == 0 &&
740 lustre_msg_get_opc(msg) != OBD_PING &&
741 lustre_msg_get_flags(msg) & MSG_REQ_REPLAY_DONE) {
744 wait_event_idle(cfs_race_waitq, (cfs_race_state == 1));
748 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
749 tsi->tsi_pill = &req->rq_pill;
750 tsi->tsi_env = req->rq_svc_thread->t_env;
752 /* if request has export then get handlers slice from corresponding
753 * target, otherwise that should be connect operation */
754 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
755 opc == MGS_CONNECT) {
757 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
758 rc = target_handle_connect(req);
760 rc = ptlrpc_error(req);
763 /* recovery-small test 18c asks to drop connect reply */
764 if (unlikely(opc == OST_CONNECT &&
765 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
769 if (unlikely(!class_connected_export(req->rq_export))) {
770 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
771 opc == SEC_CTX_FINI) {
772 /* sec context initialization has to be handled
773 * by hand in tgt_handle_request0() */
774 tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
776 GOTO(handle_recov, rc = 0);
778 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
779 opc, libcfs_id2str(req->rq_peer));
780 req->rq_status = -ENOTCONN;
781 rc = ptlrpc_error(req);
785 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
786 tsi->tsi_exp = req->rq_export;
787 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
788 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
790 tsi->tsi_jobid = NULL;
793 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export",
794 class_exp2obd(req->rq_export)->obd_name);
795 req->rq_status = -EINVAL;
796 rc = ptlrpc_error(req);
800 /* Skip last_xid processing for the recovery thread, otherwise, the
801 * last_xid on same request could be processed twice: first time when
802 * processing the incoming request, second time when the request is
803 * being processed by recovery thread. */
804 obd = class_exp2obd(req->rq_export);
806 /* reset the exp_last_xid on each connection. */
807 req->rq_export->exp_last_xid = 0;
808 } else if (obd->obd_recovery_data.trd_processing_task !=
810 rc = process_req_last_xid(req);
813 rc = ptlrpc_error(req);
818 request_fail_id = tgt->lut_request_fail_id;
819 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
821 h = tgt_handler_find_check(req);
823 req->rq_status = PTR_ERR(h);
824 rc = ptlrpc_error(req);
828 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
831 if ((cfs_fail_val == 0 || cfs_fail_val == opc) &&
832 CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
835 rc = lustre_msg_check_version(msg, h->th_version);
837 DEBUG_REQ(D_ERROR, req,
838 "%s: drop malformed request version=%08x expect=%08x",
839 tgt_name(tgt), lustre_msg_get_version(msg),
841 req->rq_status = -EINVAL;
842 rc = ptlrpc_error(req);
847 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
848 if (likely(rc == 1)) {
849 rc = tgt_handle_request0(tsi, h, req);
855 req_capsule_fini(tsi->tsi_pill);
856 if (tsi->tsi_corpus != NULL) {
857 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
858 tsi->tsi_corpus = NULL;
862 EXPORT_SYMBOL(tgt_request_handle);
864 /** Assign high priority operations to the request if needed. */
865 int tgt_hpreq_handler(struct ptlrpc_request *req)
867 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
868 struct tgt_handler *h;
873 if (req->rq_export == NULL)
876 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
877 tsi->tsi_pill = &req->rq_pill;
878 tsi->tsi_env = req->rq_svc_thread->t_env;
879 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
880 tsi->tsi_exp = req->rq_export;
882 h = tgt_handler_find_check(req);
888 rc = tgt_request_preprocess(tsi, h, req);
889 if (unlikely(rc != 0))
892 if (h->th_hp != NULL)
896 EXPORT_SYMBOL(tgt_hpreq_handler);
898 void tgt_counter_incr(struct obd_export *exp, int opcode)
900 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
901 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
902 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
904 EXPORT_SYMBOL(tgt_counter_incr);
907 * Unified target generic handlers.
910 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
912 struct lu_target *tgt = class_exp2tgt(exp);
913 struct sptlrpc_flavor flvr;
917 LASSERT(tgt->lut_obd);
918 LASSERT(tgt->lut_slice);
920 /* always allow ECHO client */
921 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
922 LUSTRE_ECHO_NAME) == 0)) {
923 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
927 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
928 read_lock(&tgt->lut_sptlrpc_lock);
929 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
933 read_unlock(&tgt->lut_sptlrpc_lock);
935 spin_lock(&exp->exp_lock);
936 exp->exp_sp_peer = req->rq_sp_from;
937 exp->exp_flvr = flvr;
939 /* when on mgs, if no restriction is set, or if the client
940 * NID is on the local node, allow any flavor
942 if ((strcmp(exp->exp_obd->obd_type->typ_name,
943 LUSTRE_MGS_NAME) == 0) &&
944 (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
945 LNetIsPeerLocal(lnet_nid_to_nid4(&exp->exp_connection->c_peer.nid))))
946 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
948 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
949 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
950 CERROR("%s: unauthorized rpc flavor %x from %s, "
951 "expect %x\n", tgt_name(tgt),
953 libcfs_nid2str(req->rq_peer.nid),
954 exp->exp_flvr.sf_rpc);
957 spin_unlock(&exp->exp_lock);
959 if (exp->exp_sp_peer != req->rq_sp_from) {
960 CERROR("%s: RPC source %s doesn't match %s\n",
962 sptlrpc_part2name(req->rq_sp_from),
963 sptlrpc_part2name(exp->exp_sp_peer));
966 rc = sptlrpc_target_export_check(exp, req);
973 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt)
975 struct sptlrpc_rule_set tmp_rset;
978 if (unlikely(tgt == NULL)) {
979 CERROR("No target passed\n");
983 sptlrpc_rule_set_init(&tmp_rset);
984 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset);
986 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
991 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
993 write_lock(&tgt->lut_sptlrpc_lock);
994 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
995 tgt->lut_sptlrpc_rset = tmp_rset;
996 write_unlock(&tgt->lut_sptlrpc_lock);
1000 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
1002 int tgt_connect(struct tgt_session_info *tsi)
1004 struct ptlrpc_request *req = tgt_ses_req(tsi);
1005 struct obd_connect_data *reply;
1010 /* XXX: better to call this check right after getting new export but
1011 * before last_rcvd slot allocation to avoid server load upon insecure
1012 * connects. This is to be fixed after unifiyng all targets.
1014 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1018 /* To avoid exposing partially initialized connection flags, changes up
1019 * to this point have been staged in reply->ocd_connect_flags. Now that
1020 * connection handling has completed successfully, atomically update
1021 * the connect flags in the shared export data structure. LU-1623 */
1022 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1023 spin_lock(&tsi->tsi_exp->exp_lock);
1024 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1025 if (reply->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1026 *exp_connect_flags2_ptr(tsi->tsi_exp) =
1027 reply->ocd_connect_flags2;
1028 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1029 spin_unlock(&tsi->tsi_exp->exp_lock);
1031 if (strcmp(tsi->tsi_exp->exp_obd->obd_type->typ_name,
1032 LUSTRE_MDT_NAME) == 0) {
1033 rc = req_check_sepol(tsi->tsi_pill);
1037 if (reply->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
1038 reply->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT &&
1039 tsi->tsi_pill->rc_req->rq_export) {
1040 bool forbid_encrypt = true;
1041 struct lu_nodemap *nm =
1042 nodemap_get_from_exp(tsi->tsi_pill->rc_req->rq_export);
1045 /* nodemap_get_from_exp returns NULL in case
1046 * nodemap is not active, so we do not forbid
1048 forbid_encrypt = false;
1049 } else if (!IS_ERR(nm)) {
1050 forbid_encrypt = nm->nmf_forbid_encryption;
1055 GOTO(out, rc = -EACCES);
1061 obd_disconnect(class_export_get(tsi->tsi_exp));
1064 EXPORT_SYMBOL(tgt_connect);
1066 int tgt_disconnect(struct tgt_session_info *tsi)
1072 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DISCONNECT_DELAY, cfs_fail_val);
1074 rc = target_handle_disconnect(tgt_ses_req(tsi));
1076 RETURN(err_serious(rc));
1080 EXPORT_SYMBOL(tgt_disconnect);
1083 * Unified target OBD handlers
1085 int tgt_obd_ping(struct tgt_session_info *tsi)
1091 /* The target-specific part of OBD_PING request handling.
1092 * It controls Filter Modification Data (FMD) expiration each time
1095 * Valid only for replayable targets, e.g. MDT and OFD
1097 if (tsi->tsi_exp->exp_obd->obd_replayable)
1098 tgt_fmd_expire(tsi->tsi_exp);
1100 rc = req_capsule_server_pack(tsi->tsi_pill);
1102 RETURN(err_serious(rc));
1106 EXPORT_SYMBOL(tgt_obd_ping);
1108 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1110 return err_serious(-EOPNOTSUPP);
1113 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1115 struct ptlrpc_request *req = tgt_ses_req(tsi);
1116 struct obd_export *exp = req->rq_export;
1117 struct ptlrpc_bulk_desc *desc;
1124 for (i = 0; i < rdbuf->rb_nbufs; i++) {
1125 unsigned int offset;
1127 offset = (unsigned long)rdbuf->rb_bufs[i].lb_buf & ~PAGE_MASK;
1128 pages += DIV_ROUND_UP(rdbuf->rb_bufs[i].lb_len + offset,
1132 desc = ptlrpc_prep_bulk_exp(req, pages, 1,
1133 PTLRPC_BULK_PUT_SOURCE,
1135 &ptlrpc_bulk_kiov_nopin_ops);
1139 for (i = 0; i < rdbuf->rb_nbufs; i++)
1140 desc->bd_frag_ops->add_iov_frag(desc,
1141 rdbuf->rb_bufs[i].lb_buf,
1142 rdbuf->rb_bufs[i].lb_len);
1144 rc = target_bulk_io(exp, desc);
1145 ptlrpc_free_bulk(desc);
1148 EXPORT_SYMBOL(tgt_send_buffer);
1150 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1152 struct ptlrpc_request *req = tgt_ses_req(tsi);
1153 struct obd_export *exp = req->rq_export;
1154 struct ptlrpc_bulk_desc *desc;
1162 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1163 PTLRPC_BULK_PUT_SOURCE,
1165 &ptlrpc_bulk_kiov_pin_ops);
1169 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1170 /* old client requires reply size in it's PAGE_SIZE,
1171 * which is rdpg->rp_count */
1172 nob = rdpg->rp_count;
1174 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1175 i++, tmpcount -= tmpsize) {
1176 tmpsize = min_t(int, tmpcount, PAGE_SIZE);
1177 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1181 LASSERT(desc->bd_nob == nob);
1182 rc = target_bulk_io(exp, desc);
1183 ptlrpc_free_bulk(desc);
1186 EXPORT_SYMBOL(tgt_sendpage);
1189 * OBD_IDX_READ handler
1191 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1193 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1194 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1195 struct idx_info *req_ii, *rep_ii;
1200 memset(rdpg, 0, sizeof(*rdpg));
1201 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1203 /* extract idx_info buffer from request & reply */
1204 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1205 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1206 RETURN(err_serious(-EPROTO));
1208 rc = req_capsule_server_pack(tsi->tsi_pill);
1210 RETURN(err_serious(rc));
1212 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1214 RETURN(err_serious(-EFAULT));
1215 rep_ii->ii_magic = IDX_INFO_MAGIC;
1217 /* extract hash to start with */
1218 rdpg->rp_hash = req_ii->ii_hash_start;
1220 /* extract requested attributes */
1221 rdpg->rp_attrs = req_ii->ii_attrs;
1223 /* check that fid packed in request is valid and supported */
1224 if (!fid_is_sane(&req_ii->ii_fid))
1226 rep_ii->ii_fid = req_ii->ii_fid;
1229 rep_ii->ii_flags = req_ii->ii_flags;
1231 /* compute number of pages to allocate, ii_count is the number of 4KB
1233 if (req_ii->ii_count <= 0)
1234 GOTO(out, rc = -EFAULT);
1235 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1236 exp_max_brw_size(tsi->tsi_exp));
1237 rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >> PAGE_SHIFT;
1239 /* allocate pages to store the containers */
1240 OBD_ALLOC_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages);
1241 if (rdpg->rp_pages == NULL)
1242 GOTO(out, rc = -ENOMEM);
1243 for (i = 0; i < rdpg->rp_npages; i++) {
1244 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1245 if (rdpg->rp_pages[i] == NULL)
1246 GOTO(out, rc = -ENOMEM);
1249 /* populate pages with key/record pairs */
1250 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1254 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1255 "asked %d > %d\n", rc, rdpg->rp_count);
1257 /* send pages to client */
1258 rc = tgt_sendpage(tsi, rdpg, rc);
1263 if (rdpg->rp_pages) {
1264 for (i = 0; i < rdpg->rp_npages; i++)
1265 if (rdpg->rp_pages[i])
1266 __free_page(rdpg->rp_pages[i]);
1267 OBD_FREE_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages);
1272 struct tgt_handler tgt_obd_handlers[] = {
1273 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1274 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1276 EXPORT_SYMBOL(tgt_obd_handlers);
1278 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1279 struct dt_object *obj, __u64 start, __u64 end)
1285 /* if no objid is specified, it means "sync whole filesystem" */
1287 rc = dt_sync(env, tgt->lut_bottom);
1288 } else if (dt_version_get(env, obj) >
1289 tgt->lut_obd->obd_last_committed) {
1290 rc = dt_object_sync(env, obj, start, end);
1292 atomic_inc(&tgt->lut_sync_count);
1296 EXPORT_SYMBOL(tgt_sync);
1298 * Unified target DLM handlers.
1302 * Unified target BAST
1304 * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1305 * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1306 * but if it's cross-MDT lock, because remote object version is not set, a
1307 * filesystem sync is needed.
1309 * \param lock server side lock
1310 * \param desc lock desc
1311 * \param data ldlm_cb_set_arg
1312 * \param flag indicates whether this cancelling or blocking callback
1313 * \retval 0 on success
1314 * \retval negative number on error
1316 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1317 void *data, int flag)
1320 struct lu_target *tgt;
1321 struct dt_object *obj = NULL;
1327 tgt = class_exp2tgt(lock->l_export);
1329 if (unlikely(tgt == NULL)) {
1330 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1331 class_exp2obd(lock->l_export)->obd_name);
1335 if (flag == LDLM_CB_CANCELING &&
1336 (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1337 (tgt->lut_sync_lock_cancel == SYNC_LOCK_CANCEL_ALWAYS ||
1338 (tgt->lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING &&
1339 ldlm_is_cbpending(lock))) &&
1340 ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1341 lock->l_resource->lr_type == LDLM_EXTENT)) {
1343 __u64 end = OBD_OBJECT_EOF;
1345 rc = lu_env_init(&env, LCT_DT_THREAD);
1346 if (unlikely(rc != 0))
1349 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1350 tgt->lut_lsd.lsd_osd_index);
1352 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1353 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1355 GOTO(err_env, rc = PTR_ERR(obj));
1357 if (!dt_object_exists(obj))
1358 GOTO(err_put, rc = -ENOENT);
1360 start = lock->l_policy_data.l_extent.start;
1361 end = lock->l_policy_data.l_extent.end;
1364 rc = tgt_sync(&env, tgt, obj, start, end);
1366 CERROR("%s: syncing "DFID" (%llu-%llu) on lock "
1367 "cancel: rc = %d\n",
1368 tgt_name(tgt), PFID(&fid),
1369 lock->l_policy_data.l_extent.start,
1370 lock->l_policy_data.l_extent.end, rc);
1374 dt_object_put(&env, obj);
1379 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1383 static struct ldlm_callback_suite tgt_dlm_cbs = {
1384 .lcs_completion = ldlm_server_completion_ast,
1385 .lcs_blocking = tgt_blocking_ast,
1386 .lcs_glimpse = ldlm_server_glimpse_ast
1389 int tgt_enqueue(struct tgt_session_info *tsi)
1391 struct ptlrpc_request *req = tgt_ses_req(tsi);
1396 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1397 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1399 LASSERT(tsi->tsi_dlm_req != NULL);
1400 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1401 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1403 RETURN(err_serious(rc));
1405 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1407 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1410 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1413 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1416 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1419 RETURN(req->rq_status);
1421 EXPORT_SYMBOL(tgt_enqueue);
1423 int tgt_convert(struct tgt_session_info *tsi)
1425 struct ptlrpc_request *req = tgt_ses_req(tsi);
1429 LASSERT(tsi->tsi_dlm_req);
1430 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1432 RETURN(err_serious(rc));
1434 RETURN(req->rq_status);
1437 int tgt_bl_callback(struct tgt_session_info *tsi)
1439 return err_serious(-EOPNOTSUPP);
1442 int tgt_cp_callback(struct tgt_session_info *tsi)
1444 return err_serious(-EOPNOTSUPP);
1447 /* generic LDLM target handler */
1448 struct tgt_handler tgt_dlm_handlers[] = {
1449 TGT_DLM_HDL(HAS_KEY, LDLM_ENQUEUE, tgt_enqueue),
1450 TGT_DLM_HDL(HAS_KEY, LDLM_CONVERT, tgt_convert),
1451 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1452 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1454 EXPORT_SYMBOL(tgt_dlm_handlers);
1457 * Unified target LLOG handlers.
1459 int tgt_llog_open(struct tgt_session_info *tsi)
1465 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1469 EXPORT_SYMBOL(tgt_llog_open);
1471 int tgt_llog_read_header(struct tgt_session_info *tsi)
1477 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1481 EXPORT_SYMBOL(tgt_llog_read_header);
1483 int tgt_llog_next_block(struct tgt_session_info *tsi)
1489 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1493 EXPORT_SYMBOL(tgt_llog_next_block);
1495 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1501 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1505 EXPORT_SYMBOL(tgt_llog_prev_block);
1507 /* generic llog target handler */
1508 struct tgt_handler tgt_llog_handlers[] = {
1509 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1510 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1511 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1512 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1514 EXPORT_SYMBOL(tgt_llog_handlers);
1517 * sec context handlers
1519 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1520 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1525 struct tgt_handler tgt_sec_ctx_handlers[] = {
1526 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1527 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1528 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1530 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1532 int (*tgt_lfsck_in_notify_local)(const struct lu_env *env,
1533 struct dt_device *key,
1534 struct lfsck_req_local *lrl,
1535 struct thandle *th) = NULL;
1537 void tgt_register_lfsck_in_notify_local(int (*notify)(const struct lu_env *,
1539 struct lfsck_req_local *,
1542 tgt_lfsck_in_notify_local = notify;
1544 EXPORT_SYMBOL(tgt_register_lfsck_in_notify_local);
1546 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1547 struct dt_device *key,
1548 struct lfsck_request *lr) = NULL;
1550 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1552 struct lfsck_request *))
1554 tgt_lfsck_in_notify = notify;
1556 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1558 static int (*tgt_lfsck_query)(const struct lu_env *env,
1559 struct dt_device *key,
1560 struct lfsck_request *req,
1561 struct lfsck_reply *rep,
1562 struct lfsck_query *que) = NULL;
1564 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1566 struct lfsck_request *,
1567 struct lfsck_reply *,
1568 struct lfsck_query *))
1570 tgt_lfsck_query = query;
1572 EXPORT_SYMBOL(tgt_register_lfsck_query);
1574 /* LFSCK request handlers */
1575 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1577 const struct lu_env *env = tsi->tsi_env;
1578 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1579 struct lfsck_request *lr;
1583 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1587 rc = tgt_lfsck_in_notify(env, key, lr);
1592 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1594 struct lfsck_request *request;
1595 struct lfsck_reply *reply;
1599 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1600 if (request == NULL)
1603 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1607 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1608 request, reply, NULL);
1610 RETURN(rc < 0 ? rc : 0);
1613 struct tgt_handler tgt_lfsck_handlers[] = {
1614 TGT_LFSCK_HDL(HAS_REPLY, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1615 TGT_LFSCK_HDL(HAS_REPLY, LFSCK_QUERY, tgt_handle_lfsck_query),
1617 EXPORT_SYMBOL(tgt_lfsck_handlers);
1620 * initialize per-thread page pool (bug 5137).
1622 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1624 struct tgt_thread_big_cache *tbc;
1628 LASSERT(thread != NULL);
1629 LASSERT(thread->t_data == NULL);
1631 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1634 thread->t_data = tbc;
1637 EXPORT_SYMBOL(tgt_io_thread_init);
1640 * free per-thread pool created by tgt_thread_init().
1642 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1644 struct tgt_thread_big_cache *tbc;
1648 LASSERT(thread != NULL);
1651 * be prepared to handle partially-initialized pools (because this is
1652 * called from ost_io_thread_init() for cleanup.
1654 tbc = thread->t_data;
1656 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1657 thread->t_data = NULL;
1661 EXPORT_SYMBOL(tgt_io_thread_done);
1664 * Helper function for getting Data-on-MDT file server DLM lock
1665 * if asked by client.
1667 int tgt_mdt_data_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1668 struct lustre_handle *lh, int mode, __u64 *flags)
1670 union ldlm_policy_data policy = {
1671 .l_inodebits.bits = MDS_INODELOCK_DOM,
1677 LASSERT(lh != NULL);
1678 LASSERT(ns != NULL);
1679 LASSERT(!lustre_handle_is_used(lh));
1681 rc = ldlm_cli_enqueue_local(NULL, ns, res_id, LDLM_IBITS, &policy, mode,
1682 flags, ldlm_blocking_ast,
1683 ldlm_completion_ast, ldlm_glimpse_ast,
1684 NULL, 0, LVB_T_NONE, NULL, lh);
1686 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1688 EXPORT_SYMBOL(tgt_mdt_data_lock);
1691 * Helper function for getting server side [start, start+count] DLM lock
1692 * if asked by client.
1694 int tgt_extent_lock(const struct lu_env *env, struct ldlm_namespace *ns,
1695 struct ldlm_res_id *res_id, __u64 start, __u64 end,
1696 struct lustre_handle *lh, int mode, __u64 *flags)
1698 union ldlm_policy_data policy;
1703 LASSERT(lh != NULL);
1704 LASSERT(ns != NULL);
1705 LASSERT(!lustre_handle_is_used(lh));
1707 policy.l_extent.gid = 0;
1708 policy.l_extent.start = start & PAGE_MASK;
1711 * If ->o_blocks is EOF it means "lock till the end of the file".
1712 * Otherwise, it's size of an extent or hole being punched (in bytes).
1714 if (end == OBD_OBJECT_EOF || end < start)
1715 policy.l_extent.end = OBD_OBJECT_EOF;
1717 policy.l_extent.end = end | ~PAGE_MASK;
1719 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_EXTENT, &policy,
1720 mode, flags, ldlm_blocking_ast,
1721 ldlm_completion_ast, ldlm_glimpse_ast,
1722 NULL, 0, LVB_T_NONE, NULL, lh);
1723 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1725 EXPORT_SYMBOL(tgt_extent_lock);
1727 static int tgt_data_lock(const struct lu_env *env, struct obd_export *exp,
1728 struct ldlm_res_id *res_id, __u64 start, __u64 end,
1729 struct lustre_handle *lh, enum ldlm_mode mode)
1731 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
1734 /* MDT IO for data-on-mdt */
1735 if (exp->exp_connect_data.ocd_connect_flags & OBD_CONNECT_IBITS)
1736 return tgt_mdt_data_lock(ns, res_id, lh, mode, &flags);
1738 return tgt_extent_lock(env, ns, res_id, start, end, lh, mode, &flags);
1741 void tgt_data_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1743 LASSERT(lustre_handle_is_used(lh));
1744 ldlm_lock_decref(lh, mode);
1746 EXPORT_SYMBOL(tgt_data_unlock);
1748 static int tgt_brw_lock(const struct lu_env *env, struct obd_export *exp,
1749 struct ldlm_res_id *res_id, struct obd_ioobj *obj,
1750 struct niobuf_remote *nb, struct lustre_handle *lh,
1751 enum ldlm_mode mode)
1753 int nrbufs = obj->ioo_bufcnt;
1758 LASSERT(mode == LCK_PR || mode == LCK_PW);
1759 LASSERT(!lustre_handle_is_used(lh));
1761 if (exp->exp_obd->obd_recovering)
1764 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1767 for (i = 1; i < nrbufs; i++)
1768 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1771 return tgt_data_lock(env, exp, res_id, nb[0].rnb_offset,
1772 nb[nrbufs - 1].rnb_offset +
1773 nb[nrbufs - 1].rnb_len - 1, lh, mode);
1776 static void tgt_brw_unlock(struct obd_export *exp, struct obd_ioobj *obj,
1777 struct niobuf_remote *niob,
1778 struct lustre_handle *lh, enum ldlm_mode mode)
1782 LASSERT(mode == LCK_PR || mode == LCK_PW);
1783 LASSERT((!exp->exp_obd->obd_recovering && obj->ioo_bufcnt &&
1784 niob[0].rnb_flags & OBD_BRW_SRVLOCK) ==
1785 lustre_handle_is_used(lh));
1787 if (lustre_handle_is_used(lh))
1788 tgt_data_unlock(lh, mode);
1792 static int tgt_checksum_niobuf(struct lu_target *tgt,
1793 struct niobuf_local *local_nb, int npages,
1794 int opc, enum cksum_types cksum_type,
1797 struct ahash_request *req;
1798 unsigned int bufsize;
1800 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1802 req = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1804 CERROR("%s: unable to initialize checksum hash %s\n",
1805 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1806 return PTR_ERR(req);
1809 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1810 for (i = 0; i < npages; i++) {
1811 /* corrupt the data before we compute the checksum, to
1812 * simulate a client->OST data error */
1813 if (i == 0 && opc == OST_WRITE &&
1814 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1815 int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
1816 int len = local_nb[i].lnb_len;
1817 struct page *np = tgt_page_to_corrupt;
1820 char *ptr = kmap_atomic(local_nb[i].lnb_page);
1821 char *ptr2 = page_address(np);
1823 memcpy(ptr2 + off, ptr + off, len);
1824 memcpy(ptr2 + off, "bad3", min(4, len));
1827 /* LU-8376 to preserve original index for
1828 * display in dump_all_bulk_pages() */
1831 cfs_crypto_hash_update_page(req, np, off,
1835 CERROR("%s: can't alloc page for corruption\n",
1839 cfs_crypto_hash_update_page(req, local_nb[i].lnb_page,
1840 local_nb[i].lnb_page_offset & ~PAGE_MASK,
1841 local_nb[i].lnb_len);
1843 /* corrupt the data after we compute the checksum, to
1844 * simulate an OST->client data error */
1845 if (i == 0 && opc == OST_READ &&
1846 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1847 int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
1848 int len = local_nb[i].lnb_len;
1849 struct page *np = tgt_page_to_corrupt;
1852 char *ptr = kmap_atomic(local_nb[i].lnb_page);
1853 char *ptr2 = page_address(np);
1855 memcpy(ptr2 + off, ptr + off, len);
1856 memcpy(ptr2 + off, "bad4", min(4, len));
1859 /* LU-8376 to preserve original index for
1860 * display in dump_all_bulk_pages() */
1863 cfs_crypto_hash_update_page(req, np, off,
1867 CERROR("%s: can't alloc page for corruption\n",
1873 bufsize = sizeof(*cksum);
1874 err = cfs_crypto_hash_final(req, (unsigned char *)cksum, &bufsize);
1879 char dbgcksum_file_name[PATH_MAX];
1881 static void dump_all_bulk_pages(struct obdo *oa, int count,
1882 struct niobuf_local *local_nb,
1883 __u32 server_cksum, __u32 client_cksum)
1890 /* will only keep dump of pages on first error for the same range in
1891 * file/fid, not during the resends/retries. */
1892 snprintf(dbgcksum_file_name, sizeof(dbgcksum_file_name),
1893 "%s-checksum_dump-ost-"DFID":[%llu-%llu]-%x-%x",
1894 (strncmp(libcfs_debug_file_path, "NONE", 4) != 0 ?
1895 libcfs_debug_file_path : LIBCFS_DEBUG_FILE_PATH_DEFAULT),
1896 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1897 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1898 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1899 local_nb[0].lnb_file_offset,
1900 local_nb[count-1].lnb_file_offset +
1901 local_nb[count-1].lnb_len - 1, client_cksum, server_cksum);
1902 CWARN("dumping checksum data to %s\n", dbgcksum_file_name);
1903 filp = filp_open(dbgcksum_file_name,
1904 O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE, 0600);
1908 CDEBUG(D_INFO, "%s: can't open to dump pages with "
1909 "checksum error: rc = %d\n", dbgcksum_file_name,
1912 CERROR("%s: can't open to dump pages with checksum "
1913 "error: rc = %d\n", dbgcksum_file_name, rc);
1917 for (i = 0; i < count; i++) {
1918 len = local_nb[i].lnb_len;
1919 buf = kmap(local_nb[i].lnb_page);
1921 rc = cfs_kernel_write(filp, buf, len, &filp->f_pos);
1923 CERROR("%s: wanted to write %u but got %d "
1924 "error\n", dbgcksum_file_name, len, rc);
1930 kunmap(local_nb[i].lnb_page);
1933 rc = vfs_fsync_range(filp, 0, LLONG_MAX, 1);
1935 CERROR("%s: sync returns %d\n", dbgcksum_file_name, rc);
1936 filp_close(filp, NULL);
1938 libcfs_debug_dumplog();
1941 static int check_read_checksum(struct niobuf_local *local_nb, int npages,
1942 struct obd_export *exp, struct obdo *oa,
1943 const struct lnet_process_id *peer,
1944 __u32 client_cksum, __u32 server_cksum,
1945 enum cksum_types server_cksum_type)
1948 enum cksum_types cksum_type;
1951 if (unlikely(npages <= 0))
1954 /* unlikely to happen and only if resend does not occur due to cksum
1955 * control failure on Client */
1956 if (unlikely(server_cksum == client_cksum)) {
1957 CDEBUG(D_PAGE, "checksum %x confirmed upon retry\n",
1962 if (exp->exp_obd->obd_checksum_dump)
1963 dump_all_bulk_pages(oa, npages, local_nb, server_cksum,
1966 cksum_type = obd_cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
1969 if (cksum_type != server_cksum_type)
1970 msg = "the server may have not used the checksum type specified"
1971 " in the original request - likely a protocol problem";
1973 msg = "should have changed on the client or in transit";
1975 start = local_nb[0].lnb_file_offset;
1976 end = local_nb[npages-1].lnb_file_offset +
1977 local_nb[npages-1].lnb_len - 1;
1979 LCONSOLE_ERROR_MSG(0x132, "%s: BAD READ CHECKSUM: %s: from %s inode "
1980 DFID " object "DOSTID" extent [%llu-%llu], client returned csum"
1981 " %x (type %x), server csum %x (type %x)\n",
1982 exp->exp_obd->obd_name,
1983 msg, libcfs_nid2str(peer->nid),
1984 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : 0ULL,
1985 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1986 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1988 start, end, client_cksum, cksum_type, server_cksum,
1994 static int tgt_pages2shortio(struct niobuf_local *local, int npages,
1995 unsigned char *buf, int size)
1997 int i, off, len, copied = size;
2000 for (i = 0; i < npages; i++) {
2001 off = local[i].lnb_page_offset & ~PAGE_MASK;
2002 len = local[i].lnb_len;
2004 CDEBUG(D_PAGE, "index %d offset = %d len = %d left = %d\n",
2009 ptr = kmap_atomic(local[i].lnb_page);
2010 memcpy(buf, ptr + off, len);
2015 return copied - size;
2018 static int tgt_checksum_niobuf_t10pi(struct lu_target *tgt,
2019 struct niobuf_local *local_nb, int npages,
2020 int opc, obd_dif_csum_fn *fn,
2021 int sector_size, u32 *check_sum,
2024 enum cksum_types t10_cksum_type = tgt->lut_dt_conf.ddp_t10_cksum_type;
2025 unsigned char cfs_alg = cksum_obd2cfs(OBD_CKSUM_T10_TOP);
2026 const char *obd_name = tgt->lut_obd->obd_name;
2027 struct ahash_request *req;
2028 unsigned int bufsize;
2029 unsigned char *buffer;
2030 struct page *__page;
2033 int used_number = 0;
2039 __page = alloc_page(GFP_KERNEL);
2043 req = cfs_crypto_hash_init(cfs_alg, NULL, 0);
2045 CERROR("%s: unable to initialize checksum hash %s\n",
2046 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
2047 return PTR_ERR(req);
2050 buffer = kmap(__page);
2051 guard_start = (__u16 *)buffer;
2052 guard_number = PAGE_SIZE / sizeof(*guard_start);
2053 if (unlikely(resend))
2054 CDEBUG(D_PAGE | D_HA, "GRD tags per page = %u\n", guard_number);
2055 for (i = 0; i < npages; i++) {
2058 /* corrupt the data before we compute the checksum, to
2059 * simulate a client->OST data error */
2060 if (i == 0 && opc == OST_WRITE &&
2061 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
2062 int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
2063 int len = local_nb[i].lnb_len;
2064 struct page *np = tgt_page_to_corrupt;
2067 char *ptr = kmap_atomic(local_nb[i].lnb_page);
2068 char *ptr2 = page_address(np);
2070 memcpy(ptr2 + off, ptr + off, len);
2071 memcpy(ptr2 + off, "bad3", min(4, len));
2074 /* LU-8376 to preserve original index for
2075 * display in dump_all_bulk_pages() */
2078 cfs_crypto_hash_update_page(req, np, off,
2082 CERROR("%s: can't alloc page for corruption\n",
2088 * The left guard number should be able to hold checksums of a
2091 use_t10_grd = t10_cksum_type && opc == OST_READ &&
2092 local_nb[i].lnb_len == PAGE_SIZE &&
2093 local_nb[i].lnb_guard_disk;
2095 used = DIV_ROUND_UP(local_nb[i].lnb_len, sector_size);
2096 if (used > (guard_number - used_number)) {
2100 memcpy(guard_start + used_number,
2101 local_nb[i].lnb_guards,
2102 used * sizeof(*guard_start));
2103 if (unlikely(resend))
2104 CDEBUG(D_PAGE | D_HA,
2105 "lnb[%u]: used %u off %u+%u lnb checksum: %*phN\n",
2107 local_nb[i].lnb_page_offset,
2108 local_nb[i].lnb_len,
2109 (int)(used * sizeof(*guard_start)),
2110 guard_start + used_number);
2112 if (!use_t10_grd || unlikely(resend)) {
2113 __u16 guard_tmp[MAX_GUARD_NUMBER];
2114 __u16 *guards = guard_start + used_number;
2115 int used_tmp = -1, *usedp = &used;
2117 if (unlikely(use_t10_grd)) {
2121 rc = obd_page_dif_generate_buffer(obd_name,
2122 local_nb[i].lnb_page,
2123 local_nb[i].lnb_page_offset & ~PAGE_MASK,
2124 local_nb[i].lnb_len, guards,
2125 guard_number - used_number, usedp, sector_size,
2127 if (unlikely(resend)) {
2128 bool bad = use_t10_grd &&
2130 local_nb[i].lnb_guards,
2131 used_tmp * sizeof(*guard_tmp));
2134 CERROR("lnb[%u]: used %u/%u off %u+%u tmp checksum: %*phN\n",
2136 local_nb[i].lnb_page_offset,
2137 local_nb[i].lnb_len,
2138 (int)(used_tmp * sizeof(*guard_start)),
2140 CDEBUG_LIMIT(D_PAGE | D_HA | (bad ? D_ERROR : 0),
2141 "lnb[%u]: used %u/%u off %u+%u gen checksum: %*phN\n",
2143 local_nb[i].lnb_page_offset,
2144 local_nb[i].lnb_len,
2145 (int)(used * sizeof(*guard_start)),
2146 guard_start + used_number);
2152 LASSERT(used <= MAX_GUARD_NUMBER);
2154 * If disk support T10PI checksum, copy guards to local_nb.
2155 * If the write is partial page, do not use the guards for bio
2156 * submission since the data might not be full-sector. The bio
2157 * guards will be generated later based on the full sectors. If
2158 * the sector size is 512B rather than 4 KB, or the page size
2159 * is larger than 4KB, this might drop some useful guards for
2160 * partial page write, but it will only add minimal extra time
2161 * of checksum calculation.
2163 if (t10_cksum_type && opc == OST_WRITE &&
2164 local_nb[i].lnb_len == PAGE_SIZE) {
2165 local_nb[i].lnb_guard_rpc = 1;
2166 memcpy(local_nb[i].lnb_guards,
2167 guard_start + used_number,
2168 used * sizeof(*local_nb[i].lnb_guards));
2171 used_number += used;
2172 if (used_number == guard_number) {
2173 cfs_crypto_hash_update_page(req, __page, 0,
2174 used_number * sizeof(*guard_start));
2178 /* corrupt the data after we compute the checksum, to
2179 * simulate an OST->client data error */
2180 if (unlikely(i == 0 && opc == OST_READ &&
2181 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND))) {
2182 int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
2183 int len = local_nb[i].lnb_len;
2184 struct page *np = tgt_page_to_corrupt;
2187 char *ptr = kmap_atomic(local_nb[i].lnb_page);
2188 char *ptr2 = page_address(np);
2190 memcpy(ptr2 + off, ptr + off, len);
2191 memcpy(ptr2 + off, "bad4", min(4, len));
2194 /* LU-8376 to preserve original index for
2195 * display in dump_all_bulk_pages() */
2198 cfs_crypto_hash_update_page(req, np, off,
2202 CERROR("%s: can't alloc page for corruption\n",
2211 if (used_number != 0)
2212 cfs_crypto_hash_update_page(req, __page, 0,
2213 used_number * sizeof(*guard_start));
2215 bufsize = sizeof(cksum);
2216 rc = cfs_crypto_hash_final(req, (unsigned char *)&cksum, &bufsize);
2221 __free_page(__page);
2225 static int tgt_checksum_niobuf_rw(struct lu_target *tgt,
2226 enum cksum_types cksum_type,
2227 struct niobuf_local *local_nb,
2228 int npages, int opc, u32 *check_sum,
2231 obd_dif_csum_fn *fn = NULL;
2232 int sector_size = 0;
2236 obd_t10_cksum2dif(cksum_type, &fn, §or_size);
2239 rc = tgt_checksum_niobuf_t10pi(tgt, local_nb, npages,
2240 opc, fn, sector_size,
2243 rc = tgt_checksum_niobuf(tgt, local_nb, npages, opc,
2244 cksum_type, check_sum);
2249 int tgt_brw_read(struct tgt_session_info *tsi)
2251 struct ptlrpc_request *req = tgt_ses_req(tsi);
2252 struct ptlrpc_bulk_desc *desc = NULL;
2253 struct obd_export *exp = tsi->tsi_exp;
2254 struct niobuf_remote *remote_nb;
2255 struct niobuf_local *local_nb;
2256 struct obd_ioobj *ioo;
2257 struct ost_body *body, *repbody;
2258 struct lustre_handle lockh = { 0 };
2259 int npages, nob = 0, rc, i, no_reply = 0,
2261 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2262 const char *obd_name = exp->exp_obd->obd_name;
2266 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL &&
2267 ptlrpc_req2svc(req)->srv_req_portal != MDS_IO_PORTAL) {
2268 CERROR("%s: deny read request from %s to portal %u\n",
2269 tgt_name(tsi->tsi_tgt),
2270 obd_export_nid2str(req->rq_export),
2271 ptlrpc_req2svc(req)->srv_req_portal);
2275 req->rq_bulk_read = 1;
2277 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK)) {
2278 /* optionally use cfs_fail_val - 1 to select a specific OST on
2279 * this server to fail requests.
2281 char fail_ost_name[MAX_OBD_NAME];
2283 if (cfs_fail_val > 0) {
2284 snprintf(fail_ost_name, MAX_OBD_NAME, "OST%04X",
2287 if (strstr(obd_name, fail_ost_name))
2288 RETURN(err_serious(-EIO));
2290 RETURN(err_serious(-EIO));
2294 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2295 cfs_fail_val : (obd_timeout + 1) / 4);
2297 /* Check if there is eviction in progress, and if so, wait for it to
2299 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
2300 /* We do not care how long it takes */
2302 exp->exp_obd->obd_evict_inprogress_waitq,
2303 !atomic_read(&exp->exp_obd->obd_evict_inprogress));
2306 /* There must be big cache in current thread to process this request
2307 * if it is NULL then something went wrong and it wasn't allocated,
2308 * report -ENOMEM in that case */
2312 body = tsi->tsi_ost_body;
2313 LASSERT(body != NULL);
2315 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2316 body->oa.o_flags & OBD_FL_NORPC)
2319 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2320 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2322 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2323 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2325 local_nb = tbc->local;
2327 rc = tgt_brw_lock(tsi->tsi_env, exp, &tsi->tsi_resid, ioo, remote_nb,
2333 * If getting the lock took more time than
2334 * client was willing to wait, drop it. b=11330
2336 if (ktime_get_real_seconds() > req->rq_deadline ||
2337 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2339 CERROR("Dropping timed-out read from %s because locking object " DOSTID " took %lld seconds (limit was %lld).\n",
2340 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
2341 ktime_get_real_seconds() - req->rq_arrival_time.tv_sec,
2342 req->rq_deadline - req->rq_arrival_time.tv_sec);
2343 GOTO(out_lock, rc = -ETIMEDOUT);
2347 * Because we already sync grant info with client when
2348 * reconnect, grant info will be cleared for resent req,
2349 * otherwise, outdated grant count in the rpc would de-sync
2350 * grant counters in case of shrink
2352 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2353 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2354 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2357 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2358 repbody->oa = body->oa;
2360 npages = PTLRPC_MAX_BRW_PAGES;
2361 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
2362 ioo, remote_nb, &npages, local_nb);
2366 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2367 body->oa.o_flags & OBD_FL_SHORT_IO) {
2370 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2371 PTLRPC_BULK_PUT_SOURCE,
2373 &ptlrpc_bulk_kiov_nopin_ops);
2375 GOTO(out_commitrw, rc = -ENOMEM);
2379 npages_read = npages;
2380 for (i = 0; i < npages; i++) {
2381 int page_rc = local_nb[i].lnb_rc;
2390 if (page_rc != 0 && desc != NULL) { /* some data! */
2391 LASSERT(local_nb[i].lnb_page != NULL);
2392 desc->bd_frag_ops->add_kiov_frag
2393 (desc, local_nb[i].lnb_page,
2394 local_nb[i].lnb_page_offset & ~PAGE_MASK,
2398 if (page_rc != local_nb[i].lnb_len) { /* short read */
2399 local_nb[i].lnb_len = page_rc;
2400 npages_read = i + (page_rc != 0 ? 1 : 0);
2401 /* All subsequent pages should be 0 */
2402 while (++i < npages)
2403 LASSERT(local_nb[i].lnb_rc == 0);
2408 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
2409 u32 flag = body->oa.o_valid & OBD_MD_FLFLAGS ?
2410 body->oa.o_flags : 0;
2411 enum cksum_types cksum_type = obd_cksum_type_unpack(flag);
2412 bool resend = (body->oa.o_valid & OBD_MD_FLFLAGS) &&
2413 (body->oa.o_flags & OBD_FL_RECOV_RESEND);
2415 repbody->oa.o_flags = obd_cksum_type_pack(obd_name,
2417 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2419 rc = tgt_checksum_niobuf_rw(tsi->tsi_tgt, cksum_type,
2420 local_nb, npages_read, OST_READ,
2421 &repbody->oa.o_cksum, resend);
2423 GOTO(out_commitrw, rc);
2424 CDEBUG(D_PAGE | (resend ? D_HA : 0),
2425 "checksum at read origin: %x (%x)\n",
2426 repbody->oa.o_cksum, cksum_type);
2428 /* if a resend it could be for a cksum error, so check Server
2429 * cksum with returned Client cksum (this should even cover
2430 * zero-cksum case) */
2432 check_read_checksum(local_nb, npages_read, exp,
2433 &body->oa, &req->rq_peer,
2435 repbody->oa.o_cksum, cksum_type);
2437 repbody->oa.o_valid = 0;
2439 if (body->oa.o_valid & OBD_MD_FLGRANT)
2440 repbody->oa.o_valid |= OBD_MD_FLGRANT;
2441 /* We're finishing using body->oa as an input variable */
2443 /* Check if client was evicted while we were doing i/o before touching
2446 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2447 body->oa.o_flags & OBD_FL_SHORT_IO) {
2448 unsigned char *short_io_buf;
2451 short_io_buf = req_capsule_server_get(&req->rq_pill,
2453 short_io_size = req_capsule_get_size(&req->rq_pill,
2456 rc = tgt_pages2shortio(local_nb, npages_read,
2457 short_io_buf, short_io_size);
2459 req_capsule_shrink(&req->rq_pill,
2462 rc = rc > 0 ? 0 : rc;
2463 } else if (!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) {
2464 rc = target_bulk_io(exp, desc);
2468 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2469 body->oa.o_flags & OBD_FL_SHORT_IO)
2470 req_capsule_shrink(&req->rq_pill, &RMF_SHORT_IO, 0,
2475 /* Must commit after prep above in all cases */
2476 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
2477 remote_nb, npages, local_nb, rc);
2479 tgt_brw_unlock(exp, ioo, remote_nb, &lockh, LCK_PR);
2481 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
2482 ptlrpc_free_bulk(desc);
2487 ptlrpc_lprocfs_brw(req, nob);
2488 } else if (no_reply) {
2489 req->rq_no_reply = 1;
2490 /* reply out callback would free */
2491 ptlrpc_req_drop_rs(req);
2492 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
2493 "client will retry: rc %d\n",
2495 obd_uuid2str(&exp->exp_client_uuid),
2496 obd_export_nid2str(exp), rc);
2498 /* send a bulk after reply to simulate a network delay or reordering
2499 * by a router - Note that !desc implies short io, so there is no bulk
2501 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) &&
2503 /* Calculate checksum before request transfer, original
2504 * it is done by target_bulk_io() */
2505 rc = sptlrpc_svc_wrap_bulk(req, desc);
2506 if (OCD_HAS_FLAG(&exp->exp_connect_data, BULK_MBITS))
2507 req->rq_mbits = lustre_msg_get_mbits(req->rq_reqmsg);
2508 else /* old version, bulk matchbits is rq_xid */
2509 req->rq_mbits = req->rq_xid;
2511 req->rq_status = rc;
2512 target_committed_to_req(req);
2513 target_send_reply(req, 0, 0);
2515 CDEBUG(D_INFO, "reorder BULK\n");
2516 OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2,
2517 cfs_fail_val ? : 3);
2519 target_bulk_io(exp, desc);
2520 ptlrpc_free_bulk(desc);
2525 EXPORT_SYMBOL(tgt_brw_read);
2527 static int tgt_shortio2pages(struct niobuf_local *local, int npages,
2528 unsigned char *buf, unsigned int size)
2533 for (i = 0; i < npages; i++) {
2534 off = local[i].lnb_page_offset & ~PAGE_MASK;
2535 len = local[i].lnb_len;
2540 CDEBUG(D_PAGE, "index %d offset = %d len = %d left = %d\n",
2542 ptr = kmap_atomic(local[i].lnb_page);
2545 memcpy(ptr + off, buf, len < size ? len : size);
2553 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
2554 struct ptlrpc_bulk_desc *desc,
2555 struct niobuf_local *local_nb, int npages,
2556 u32 client_cksum, u32 server_cksum,
2559 struct obd_export *exp = req->rq_export;
2560 struct ost_body *body;
2564 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2565 LASSERT(body != NULL);
2567 if (desc && req->rq_peer.nid != desc->bd_sender) {
2569 router = libcfs_nid2str(desc->bd_sender);
2572 if (exp->exp_obd->obd_checksum_dump)
2573 dump_all_bulk_pages(&body->oa, npages, local_nb, server_cksum,
2577 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
2578 client_cksum, server_cksum);
2582 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: from %s%s%s inode "
2583 DFID" object "DOSTID" extent [%llu-%llu"
2584 "]: client csum %x, server csum %x\n",
2585 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
2587 body->oa.o_valid & OBD_MD_FLFID ?
2588 body->oa.o_parent_seq : (__u64)0,
2589 body->oa.o_valid & OBD_MD_FLFID ?
2590 body->oa.o_parent_oid : 0,
2591 body->oa.o_valid & OBD_MD_FLFID ?
2592 body->oa.o_parent_ver : 0,
2593 POSTID(&body->oa.o_oi),
2594 local_nb[0].lnb_file_offset,
2595 local_nb[npages-1].lnb_file_offset +
2596 local_nb[npages - 1].lnb_len - 1,
2597 client_cksum, server_cksum);
2600 int tgt_brw_write(struct tgt_session_info *tsi)
2602 struct ptlrpc_request *req = tgt_ses_req(tsi);
2603 struct ptlrpc_bulk_desc *desc = NULL;
2604 struct obd_export *exp = req->rq_export;
2605 struct niobuf_remote *remote_nb;
2606 struct niobuf_local *local_nb;
2607 struct obd_ioobj *ioo;
2608 struct ost_body *body, *repbody;
2609 struct lustre_handle lockh = {0};
2611 int objcount, niocount, npages;
2614 enum cksum_types cksum_type = OBD_CKSUM_CRC32;
2615 bool no_reply = false, mmap;
2616 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2617 bool wait_sync = false;
2618 const char *obd_name = exp->exp_obd->obd_name;
2619 /* '1' for consistency with code that checks !mpflag to restore */
2620 unsigned int mpflags = 1;
2624 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL &&
2625 ptlrpc_req2svc(req)->srv_req_portal != MDS_IO_PORTAL) {
2626 CERROR("%s: deny write request from %s to portal %u\n",
2627 tgt_name(tsi->tsi_tgt),
2628 obd_export_nid2str(req->rq_export),
2629 ptlrpc_req2svc(req)->srv_req_portal);
2630 RETURN(err_serious(-EPROTO));
2633 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2634 RETURN(err_serious(-ENOSPC));
2635 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2636 RETURN(err_serious(-EROFS));
2638 req->rq_bulk_write = 1;
2640 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2642 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2645 /* optionally use cfs_fail_val - 1 to select a specific OST on
2646 * this server to fail requests.
2648 char fail_ost_name[MAX_OBD_NAME];
2650 if (cfs_fail_val > 0) {
2651 snprintf(fail_ost_name, MAX_OBD_NAME, "OST%04X",
2654 if (strstr(obd_name, fail_ost_name))
2655 RETURN(err_serious(rc));
2657 RETURN(err_serious(rc));
2661 /* pause before transaction has been started */
2662 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2663 cfs_fail_val : (obd_timeout + 1) / 4);
2665 /* Delay write commit to show stale size information */
2666 CFS_FAIL_TIMEOUT(OBD_FAIL_OSC_NO_SIZE_DATA, cfs_fail_val);
2668 /* There must be big cache in current thread to process this request
2669 * if it is NULL then something went wrong and it wasn't allocated,
2670 * report -ENOMEM in that case */
2674 body = tsi->tsi_ost_body;
2675 LASSERT(body != NULL);
2677 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2678 body->oa.o_flags & OBD_FL_NORPC)
2682 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2683 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2685 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2686 RCL_CLIENT) / sizeof(*ioo);
2688 for (niocount = i = 0; i < objcount; i++)
2689 niocount += ioo[i].ioo_bufcnt;
2691 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2692 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2693 if (niocount != req_capsule_get_size(&req->rq_pill,
2694 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2696 RETURN(err_serious(-EPROTO));
2698 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2699 ptlrpc_connection_is_local(exp->exp_connection))
2700 mpflags = memalloc_noreclaim_save();
2702 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2703 niocount * sizeof(*rcs));
2704 rc = req_capsule_server_pack(&req->rq_pill);
2706 GOTO(out, rc = err_serious(rc));
2708 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2709 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2711 local_nb = tbc->local;
2713 rc = tgt_brw_lock(tsi->tsi_env, exp, &tsi->tsi_resid, ioo, remote_nb,
2719 * If getting the lock took more time than
2720 * client was willing to wait, drop it. b=11330
2722 if (ktime_get_real_seconds() > req->rq_deadline ||
2723 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2725 CERROR("%s: Dropping timed-out write from %s because locking object " DOSTID " took %lld seconds (limit was %lld).\n",
2726 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2727 POSTID(&ioo->ioo_oid),
2728 ktime_get_real_seconds() - req->rq_arrival_time.tv_sec,
2729 req->rq_deadline - req->rq_arrival_time.tv_sec);
2730 GOTO(out_lock, rc = -ETIMEDOUT);
2733 /* Because we already sync grant info with client when reconnect,
2734 * grant info will be cleared for resent req, then fed_grant and
2735 * total_grant will not be modified in following preprw_write */
2736 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2737 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2738 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2741 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2742 if (repbody == NULL)
2743 GOTO(out_lock, rc = -ENOMEM);
2744 repbody->oa = body->oa;
2746 npages = PTLRPC_MAX_BRW_PAGES;
2747 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2748 objcount, ioo, remote_nb, &npages, local_nb);
2751 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2752 body->oa.o_flags & OBD_FL_SHORT_IO) {
2753 unsigned int short_io_size;
2754 unsigned char *short_io_buf;
2756 short_io_size = req_capsule_get_size(&req->rq_pill,
2759 short_io_buf = req_capsule_client_get(&req->rq_pill,
2761 CDEBUG(D_INFO, "Client use short io for data transfer,"
2762 " size = %d\n", short_io_size);
2764 /* Copy short io buf to pages */
2765 rc = tgt_shortio2pages(local_nb, npages, short_io_buf,
2769 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2770 PTLRPC_BULK_GET_SINK,
2772 &ptlrpc_bulk_kiov_nopin_ops);
2774 GOTO(skip_transfer, rc = -ENOMEM);
2776 /* NB Having prepped, we must commit... */
2777 for (i = 0; i < npages; i++)
2778 desc->bd_frag_ops->add_kiov_frag(desc,
2779 local_nb[i].lnb_page,
2780 local_nb[i].lnb_page_offset & ~PAGE_MASK,
2781 local_nb[i].lnb_len);
2783 rc = sptlrpc_svc_prep_bulk(req, desc);
2785 GOTO(skip_transfer, rc);
2787 rc = target_bulk_io(exp, desc);
2793 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2794 static int cksum_counter;
2796 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2797 cksum_type = obd_cksum_type_unpack(body->oa.o_flags);
2799 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2800 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2801 repbody->oa.o_flags |= obd_cksum_type_pack(obd_name,
2804 rc = tgt_checksum_niobuf_rw(tsi->tsi_tgt, cksum_type,
2805 local_nb, npages, OST_WRITE,
2806 &repbody->oa.o_cksum, false);
2808 GOTO(out_commitrw, rc);
2812 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2813 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2814 body->oa.o_flags & OBD_FL_MMAP);
2816 tgt_warn_on_cksum(req, desc, local_nb, npages,
2818 repbody->oa.o_cksum, mmap);
2820 } else if ((cksum_counter & (-cksum_counter)) ==
2822 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2823 cksum_counter, libcfs_id2str(req->rq_peer),
2824 repbody->oa.o_cksum);
2828 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK2, cfs_fail_val);
2831 /* Must commit after prep above in all cases */
2832 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2833 objcount, ioo, remote_nb, npages, local_nb, rc);
2834 if (rc == -ENOTCONN)
2835 /* quota acquire process has been given up because
2836 * either the client has been evicted or the client
2837 * has timed out the request already */
2840 for (i = 0; i < niocount; i++) {
2841 if (!(local_nb[i].lnb_flags & OBD_BRW_ASYNC)) {
2847 * Disable sending mtime back to the client. If the client locked the
2848 * whole object, then it has already updated the mtime on its side,
2849 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2851 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2856 /* set per-requested niobuf return codes */
2857 for (i = j = 0; i < niocount; i++) {
2858 int len = remote_nb[i].rnb_len;
2863 LASSERT(j < npages);
2864 if (local_nb[j].lnb_rc < 0)
2865 rcs[i] = local_nb[j].lnb_rc;
2866 len -= local_nb[j].lnb_len;
2871 LASSERT(j == npages);
2872 ptlrpc_lprocfs_brw(req, nob);
2875 tgt_brw_unlock(exp, ioo, remote_nb, &lockh, LCK_PW);
2877 ptlrpc_free_bulk(desc);
2879 if (unlikely(no_reply || (exp->exp_obd->obd_no_transno && wait_sync))) {
2880 req->rq_no_reply = 1;
2881 /* reply out callback would free */
2882 ptlrpc_req_drop_rs(req);
2883 if (!exp->exp_obd->obd_no_transno)
2884 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s),"
2885 " client will retry: rc = %d\n",
2887 obd_uuid2str(&exp->exp_client_uuid),
2888 obd_export_nid2str(exp), rc);
2892 memalloc_noreclaim_restore(mpflags);
2896 EXPORT_SYMBOL(tgt_brw_write);
2899 * Common request handler for OST_SEEK RPC.
2901 * Unified request handling for OST_SEEK RPC.
2902 * It takes object by its FID, does needed lseek and packs result
2903 * into reply. Only SEEK_HOLE and SEEK_DATA are supported.
2905 * \param[in] tsi target session environment for this request
2907 * \retval 0 if successful
2908 * \retval negative value on error
2910 int tgt_lseek(struct tgt_session_info *tsi)
2912 struct lustre_handle lh = { 0 };
2913 struct dt_object *dob;
2914 struct ost_body *repbody;
2915 loff_t offset = tsi->tsi_ost_body->oa.o_size;
2916 int whence = tsi->tsi_ost_body->oa.o_mode;
2922 if (whence != SEEK_HOLE && whence != SEEK_DATA)
2925 /* Negative offset is prohibited on wire and must be handled on client
2926 * prior sending RPC.
2931 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2932 if (repbody == NULL)
2934 repbody->oa = tsi->tsi_ost_body->oa;
2936 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2937 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
2939 rc = tgt_data_lock(tsi->tsi_env, tsi->tsi_exp, &tsi->tsi_resid,
2940 offset, OBD_OBJECT_EOF, &lh, LCK_PR);
2945 dob = dt_locate(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, &tsi->tsi_fid);
2947 GOTO(out, rc = PTR_ERR(dob));
2949 if (!dt_object_exists(dob))
2950 GOTO(obj_put, rc = -ENOENT);
2952 repbody->oa.o_size = dt_lseek(tsi->tsi_env, dob, offset, whence);
2955 dt_object_put(tsi->tsi_env, dob);
2958 tgt_data_unlock(&lh, LCK_PR);
2962 EXPORT_SYMBOL(tgt_lseek);
2964 /* Check if request can be reconstructed from saved reply data
2965 * A copy of the reply data is returned in @trd if the pointer is not NULL
2967 int req_can_reconstruct(struct ptlrpc_request *req,
2968 struct tg_reply_data *trd)
2970 struct tg_export_data *ted = &req->rq_export->exp_target_data;
2971 struct lsd_client_data *lcd = ted->ted_lcd;
2974 if (tgt_is_multimodrpcs_client(req->rq_export))
2975 return tgt_lookup_reply(req, trd);
2977 mutex_lock(&ted->ted_lcd_lock);
2978 found = req->rq_xid == lcd->lcd_last_xid ||
2979 req->rq_xid == lcd->lcd_last_close_xid;
2981 if (found && trd != NULL) {
2982 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2983 trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2984 trd->trd_reply.lrd_transno =
2985 lcd->lcd_last_close_transno;
2986 trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2988 trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2989 trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2990 trd->trd_reply.lrd_result = lcd->lcd_last_result;
2991 trd->trd_reply.lrd_data = lcd->lcd_last_data;
2992 trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2993 trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2994 trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2995 trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2998 mutex_unlock(&ted->ted_lcd_lock);
3002 EXPORT_SYMBOL(req_can_reconstruct);