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) 2012, 2015, 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 <md_object.h>
46 #include <lustre_lfsck.h>
47 #include <lustre_nodemap.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 HABEO_CORPUS 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 & HABEO_CORPUS) && !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 nodemap_putref(nodemap);
270 tsi->tsi_ost_body = body;
271 tsi->tsi_fid = body->oa.o_oi.oi_fid;
273 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
274 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
279 if (!(body->oa.o_valid & OBD_MD_FLID)) {
280 if (flags & HABEO_CORPUS) {
281 CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
282 "but OID/FID is mandatory with HABEO_CORPUS\n",
283 tgt_name(tsi->tsi_tgt));
290 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
293 * OST doesn't get object in advance for further use to prevent
294 * situations with nested object_find which is potential deadlock.
296 tsi->tsi_corpus = NULL;
301 * Do necessary preprocessing according to handler ->th_flags.
303 static int tgt_request_preprocess(struct tgt_session_info *tsi,
304 struct tgt_handler *h,
305 struct ptlrpc_request *req)
307 struct req_capsule *pill = tsi->tsi_pill;
308 __u32 flags = h->th_flags;
313 if (tsi->tsi_preprocessed)
316 LASSERT(h->th_act != NULL);
317 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
318 LASSERT(current->journal_info == NULL);
320 LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
322 if (h->th_fmt != NULL) {
323 req_capsule_set(pill, h->th_fmt);
324 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
325 rc = tgt_mdt_body_unpack(tsi, flags);
328 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
330 rc = tgt_ost_body_unpack(tsi, flags);
336 if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
339 if (flags & HABEO_CLAVIS) {
340 struct ldlm_request *dlm_req;
342 LASSERT(h->th_fmt != NULL);
344 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
345 if (dlm_req != NULL) {
346 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
348 dlm_req->lock_desc.l_policy_data.\
349 l_inodebits.bits == 0)) {
351 * Lock without inodebits makes no sense and
352 * will oops later in ldlm. If client miss to
353 * set such bits, do not trigger ASSERTION.
355 * For liblustre flock case, it maybe zero.
359 tsi->tsi_dlm_req = dlm_req;
365 tsi->tsi_preprocessed = 1;
370 * Invoke handler for this request opc. Also do necessary preprocessing
371 * (according to handler ->th_flags), and post-processing (setting of
372 * ->last_{xid,committed}).
374 static int tgt_handle_request0(struct tgt_session_info *tsi,
375 struct tgt_handler *h,
376 struct ptlrpc_request *req)
380 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
385 /* When dealing with sec context requests, no export is associated yet,
386 * because these requests are sent before *_CONNECT requests.
387 * A NULL req->rq_export means the normal *_common_slice handlers will
388 * not be called, because there is no reference to the target.
389 * So deal with them by hand and jump directly to target_send_reply().
393 case SEC_CTX_INIT_CONT:
395 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
400 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
401 * to put same checks into handlers like mdt_close(), mdt_reint(),
402 * etc., without talking to mdt authors first. Checking same thing
403 * there again is useless and returning 0 error without packing reply
404 * is buggy! Handlers either pack reply or return error.
406 * We return 0 here and do not send any reply in order to emulate
407 * network failure. Do not send any reply in case any of NET related
408 * fail_id has occured.
410 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
412 if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
413 OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
416 rc = tgt_request_preprocess(tsi, h, req);
417 /* pack reply if reply format is fixed */
418 if (rc == 0 && h->th_flags & HABEO_REFERO) {
420 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
422 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
424 tsi->tsi_mdt_body->mbo_eadatasize);
425 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
427 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
430 rc = req_capsule_server_pack(tsi->tsi_pill);
433 if (likely(rc == 0)) {
435 * Process request, there can be two types of rc:
436 * 1) errors with msg unpack/pack, other failures outside the
437 * operation itself. This is counted as serious errors;
438 * 2) errors during fs operation, should be placed in rq_status
442 if (!is_serious(rc) &&
443 !req->rq_no_reply && req->rq_reply_state == NULL) {
444 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
445 "pack reply and returned 0 error\n",
446 tgt_name(tsi->tsi_tgt), h->th_name);
449 serious = is_serious(rc);
450 rc = clear_serious(rc);
458 * ELDLM_* codes which > 0 should be in rq_status only as well as
459 * all non-serious errors.
461 if (rc > 0 || !serious)
464 LASSERT(current->journal_info == NULL);
466 if (likely(rc == 0 && req->rq_export))
467 target_committed_to_req(req);
470 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
474 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
475 struct obd_device *obd, int *process)
477 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
484 case MDS_SYNC: /* used in unmounting */
498 case MDS_HSM_PROGRESS:
499 case MDS_HSM_STATE_SET:
500 case MDS_HSM_REQUEST:
501 *process = target_queue_recovery_request(req, obd);
505 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
512 * Handle recovery. Return:
513 * +1: continue request processing;
514 * -ve: abort immediately with the given error code;
515 * 0: send reply with error code in req->rq_status;
517 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
521 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
526 case SEC_CTX_INIT_CONT:
531 if (!req->rq_export->exp_obd->obd_replayable)
534 /* sanity check: if the xid matches, the request must be marked as a
535 * resent or replayed */
536 if (req_can_reconstruct(req, NULL)) {
537 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
538 (MSG_RESENT | MSG_REPLAY))) {
539 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
540 "saved xid, expected REPLAY or RESENT flag "
542 lustre_msg_get_flags(req->rq_reqmsg));
543 req->rq_status = -ENOTCONN;
547 /* else: note the opposite is not always true; a RESENT req after a
548 * failover will usually not match the last_xid, since it was likely
549 * never committed. A REPLAYed request will almost never match the
550 * last xid, however it could for a committed, but still retained,
553 /* Check for aborted recovery... */
554 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
558 DEBUG_REQ(D_INFO, req, "Got new replay");
559 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
561 if (rc != 0 || !should_process)
563 else if (should_process < 0) {
564 req->rq_status = should_process;
565 rc = ptlrpc_error(req);
572 /* Initial check for request, it is validation mostly */
573 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
575 struct tgt_handler *h;
576 struct tgt_opc_slice *s;
577 struct lu_target *tgt;
578 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
582 tgt = class_exp2tgt(req->rq_export);
583 if (unlikely(tgt == NULL)) {
584 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
585 class_exp2obd(req->rq_export)->obd_name);
586 RETURN(ERR_PTR(-EINVAL));
589 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
590 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
593 /* opcode was not found in slice */
594 if (unlikely(s->tos_hs == NULL)) {
595 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
597 RETURN(ERR_PTR(-ENOTSUPP));
600 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
601 h = s->tos_hs + (opc - s->tos_opc_start);
602 if (unlikely(h->th_opc == 0)) {
603 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
604 RETURN(ERR_PTR(-ENOTSUPP));
610 static int process_req_last_xid(struct ptlrpc_request *req)
615 /* check request's xid is consistent with export's last_xid */
616 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
617 if (last_xid > req->rq_export->exp_last_xid)
618 req->rq_export->exp_last_xid = last_xid;
620 if (req->rq_xid == 0 ||
621 (req->rq_xid <= req->rq_export->exp_last_xid)) {
622 DEBUG_REQ(D_ERROR, req, "Unexpected xid %llx vs. "
623 "last_xid %llx\n", req->rq_xid,
624 req->rq_export->exp_last_xid);
625 /* Some request is allowed to be sent during replay,
626 * such as OUT update requests, FLD requests, so it
627 * is possible that replay requests has smaller XID
628 * than the exp_last_xid.
630 * Some non-replay requests may have smaller XID as
633 * - Client send a no_resend RPC, like statfs;
634 * - The RPC timedout (or some other error) on client,
635 * then it's removed from the unreplied list;
636 * - Client send some other request to bump the
637 * exp_last_xid on server;
638 * - The former RPC got chance to be processed;
640 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
644 /* try to release in-memory reply data */
645 if (tgt_is_multimodrpcs_client(req->rq_export)) {
646 tgt_handle_received_xid(req->rq_export,
647 lustre_msg_get_last_xid(req->rq_reqmsg));
648 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
649 (MSG_RESENT | MSG_REPLAY)))
650 tgt_handle_tag(req->rq_export,
651 lustre_msg_get_tag(req->rq_reqmsg));
656 int tgt_request_handle(struct ptlrpc_request *req)
658 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
660 struct lustre_msg *msg = req->rq_reqmsg;
661 struct tgt_handler *h;
662 struct lu_target *tgt;
663 int request_fail_id = 0;
664 __u32 opc = lustre_msg_get_opc(msg);
665 struct obd_device *obd;
667 bool is_connect = false;
670 /* Refill the context, to make sure all thread keys are allocated */
671 lu_env_refill(req->rq_svc_thread->t_env);
673 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
674 tsi->tsi_pill = &req->rq_pill;
675 tsi->tsi_env = req->rq_svc_thread->t_env;
677 /* if request has export then get handlers slice from corresponding
678 * target, otherwise that should be connect operation */
679 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
680 opc == MGS_CONNECT) {
682 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
683 rc = target_handle_connect(req);
685 rc = ptlrpc_error(req);
688 /* recovery-small test 18c asks to drop connect reply */
689 if (unlikely(opc == OST_CONNECT &&
690 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
694 if (unlikely(!class_connected_export(req->rq_export))) {
695 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
696 opc == SEC_CTX_FINI) {
697 /* sec context initialization has to be handled
698 * by hand in tgt_handle_request0() */
699 tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
701 GOTO(handle_recov, rc = 0);
703 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
704 opc, libcfs_id2str(req->rq_peer));
705 req->rq_status = -ENOTCONN;
706 rc = ptlrpc_error(req);
710 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
711 tsi->tsi_exp = req->rq_export;
712 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
713 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
715 tsi->tsi_jobid = NULL;
718 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
719 class_exp2obd(req->rq_export)->obd_name);
720 req->rq_status = -EINVAL;
721 rc = ptlrpc_error(req);
725 /* Skip last_xid processing for the recovery thread, otherwise, the
726 * last_xid on same request could be processed twice: first time when
727 * processing the incoming request, second time when the request is
728 * being processed by recovery thread. */
729 obd = class_exp2obd(req->rq_export);
731 /* reset the exp_last_xid on each connection. */
732 req->rq_export->exp_last_xid = 0;
733 } else if (obd->obd_recovery_data.trd_processing_task !=
735 rc = process_req_last_xid(req);
738 rc = ptlrpc_error(req);
743 request_fail_id = tgt->lut_request_fail_id;
744 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
746 h = tgt_handler_find_check(req);
748 req->rq_status = PTR_ERR(h);
749 rc = ptlrpc_error(req);
753 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
756 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
759 rc = lustre_msg_check_version(msg, h->th_version);
761 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
762 " %08x, expecting %08x\n", tgt_name(tgt),
763 lustre_msg_get_version(msg), h->th_version);
764 req->rq_status = -EINVAL;
765 rc = ptlrpc_error(req);
770 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
771 if (likely(rc == 1)) {
772 rc = tgt_handle_request0(tsi, h, req);
778 req_capsule_fini(tsi->tsi_pill);
779 if (tsi->tsi_corpus != NULL) {
780 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
781 tsi->tsi_corpus = NULL;
785 EXPORT_SYMBOL(tgt_request_handle);
787 /** Assign high priority operations to the request if needed. */
788 int tgt_hpreq_handler(struct ptlrpc_request *req)
790 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
791 struct tgt_handler *h;
796 if (req->rq_export == NULL)
799 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
800 tsi->tsi_pill = &req->rq_pill;
801 tsi->tsi_env = req->rq_svc_thread->t_env;
802 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
803 tsi->tsi_exp = req->rq_export;
805 h = tgt_handler_find_check(req);
811 rc = tgt_request_preprocess(tsi, h, req);
812 if (unlikely(rc != 0))
815 if (h->th_hp != NULL)
819 EXPORT_SYMBOL(tgt_hpreq_handler);
821 void tgt_counter_incr(struct obd_export *exp, int opcode)
823 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
824 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
825 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
827 EXPORT_SYMBOL(tgt_counter_incr);
830 * Unified target generic handlers.
836 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
840 static int tgt_init_sec_level(struct ptlrpc_request *req)
842 struct lu_target *tgt = class_exp2tgt(req->rq_export);
844 struct obd_connect_data *data, *reply;
848 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
849 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
850 if (data == NULL || reply == NULL)
853 /* connection from MDT is always trusted */
854 if (req->rq_auth_usr_mdt) {
855 tgt_init_sec_none(reply);
859 if (unlikely(tgt == NULL)) {
860 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
861 class_exp2obd(req->rq_export)->obd_name);
865 client = libcfs_nid2str(req->rq_peer.nid);
866 /* no GSS support case */
867 if (!req->rq_auth_gss) {
868 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
869 CWARN("client %s -> target %s does not use GSS, "
870 "can not run under security level %d.\n",
871 client, tgt_name(tgt), tgt->lut_sec_level);
874 tgt_init_sec_none(reply);
879 /* old version case */
880 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
881 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
882 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
883 CWARN("client %s -> target %s uses old version, "
884 "can not run under security level %d.\n",
885 client, tgt_name(tgt), tgt->lut_sec_level);
888 CWARN("client %s -> target %s uses old version, "
889 "run under security level %d.\n",
890 client, tgt_name(tgt), tgt->lut_sec_level);
891 tgt_init_sec_none(reply);
896 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
897 CDEBUG(D_SEC, "client %s -> target %s: user is not "
898 "authenticated!\n", client, tgt_name(tgt));
902 switch (tgt->lut_sec_level) {
903 case LUSTRE_SEC_NONE:
904 case LUSTRE_SEC_REMOTE:
905 tgt_init_sec_none(reply);
908 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
909 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
918 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
920 struct lu_target *tgt = class_exp2tgt(exp);
921 struct sptlrpc_flavor flvr;
925 LASSERT(tgt->lut_obd);
926 LASSERT(tgt->lut_slice);
928 /* always allow ECHO client */
929 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
930 LUSTRE_ECHO_NAME) == 0)) {
931 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
935 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
936 read_lock(&tgt->lut_sptlrpc_lock);
937 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
941 read_unlock(&tgt->lut_sptlrpc_lock);
943 spin_lock(&exp->exp_lock);
944 exp->exp_sp_peer = req->rq_sp_from;
945 exp->exp_flvr = flvr;
947 /* when on mgs, if no restriction is set, or if client
948 * is loopback, allow any flavor */
949 if ((strcmp(exp->exp_obd->obd_type->typ_name,
950 LUSTRE_MGS_NAME) == 0) &&
951 (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
952 LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
954 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
956 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
957 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
958 CERROR("%s: unauthorized rpc flavor %x from %s, "
959 "expect %x\n", tgt_name(tgt),
961 libcfs_nid2str(req->rq_peer.nid),
962 exp->exp_flvr.sf_rpc);
965 spin_unlock(&exp->exp_lock);
967 if (exp->exp_sp_peer != req->rq_sp_from) {
968 CERROR("%s: RPC source %s doesn't match %s\n",
970 sptlrpc_part2name(req->rq_sp_from),
971 sptlrpc_part2name(exp->exp_sp_peer));
974 rc = sptlrpc_target_export_check(exp, req);
981 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
983 struct sptlrpc_rule_set tmp_rset;
986 if (unlikely(tgt == NULL)) {
987 CERROR("No target passed");
991 sptlrpc_rule_set_init(&tmp_rset);
992 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
994 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
999 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
1001 write_lock(&tgt->lut_sptlrpc_lock);
1002 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
1003 tgt->lut_sptlrpc_rset = tmp_rset;
1004 write_unlock(&tgt->lut_sptlrpc_lock);
1008 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
1010 int tgt_connect(struct tgt_session_info *tsi)
1012 struct ptlrpc_request *req = tgt_ses_req(tsi);
1013 struct obd_connect_data *reply;
1018 rc = tgt_init_sec_level(req);
1022 /* XXX: better to call this check right after getting new export but
1023 * before last_rcvd slot allocation to avoid server load upon insecure
1024 * connects. This is to be fixed after unifiyng all targets.
1026 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1030 /* To avoid exposing partially initialized connection flags, changes up
1031 * to this point have been staged in reply->ocd_connect_flags. Now that
1032 * connection handling has completed successfully, atomically update
1033 * the connect flags in the shared export data structure. LU-1623 */
1034 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1035 spin_lock(&tsi->tsi_exp->exp_lock);
1036 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1037 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1038 spin_unlock(&tsi->tsi_exp->exp_lock);
1042 obd_disconnect(class_export_get(tsi->tsi_exp));
1045 EXPORT_SYMBOL(tgt_connect);
1047 int tgt_disconnect(struct tgt_session_info *tsi)
1053 rc = target_handle_disconnect(tgt_ses_req(tsi));
1055 RETURN(err_serious(rc));
1059 EXPORT_SYMBOL(tgt_disconnect);
1062 * Unified target OBD handlers
1064 int tgt_obd_ping(struct tgt_session_info *tsi)
1070 rc = target_handle_ping(tgt_ses_req(tsi));
1072 RETURN(err_serious(rc));
1076 EXPORT_SYMBOL(tgt_obd_ping);
1078 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1080 return err_serious(-EOPNOTSUPP);
1083 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1085 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1086 struct ptlrpc_request *req = tgt_ses_req(tsi);
1087 struct obd_export *exp = req->rq_export;
1088 struct ptlrpc_bulk_desc *desc;
1089 struct l_wait_info *lwi = &tti->tti_u.update.tti_wait_info;
1095 desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1096 PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1097 MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1101 for (i = 0; i < rdbuf->rb_nbufs; i++)
1102 desc->bd_frag_ops->add_iov_frag(desc,
1103 rdbuf->rb_bufs[i].lb_buf,
1104 rdbuf->rb_bufs[i].lb_len);
1106 rc = target_bulk_io(exp, desc, lwi);
1107 ptlrpc_free_bulk(desc);
1110 EXPORT_SYMBOL(tgt_send_buffer);
1112 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1114 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1115 struct ptlrpc_request *req = tgt_ses_req(tsi);
1116 struct obd_export *exp = req->rq_export;
1117 struct ptlrpc_bulk_desc *desc;
1118 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1126 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1127 PTLRPC_BULK_PUT_SOURCE |
1128 PTLRPC_BULK_BUF_KIOV,
1130 &ptlrpc_bulk_kiov_pin_ops);
1134 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1135 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1136 * which is rdpg->rp_count */
1137 nob = rdpg->rp_count;
1139 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1140 i++, tmpcount -= tmpsize) {
1141 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1142 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1146 LASSERT(desc->bd_nob == nob);
1147 rc = target_bulk_io(exp, desc, lwi);
1148 ptlrpc_free_bulk(desc);
1151 EXPORT_SYMBOL(tgt_sendpage);
1154 * OBD_IDX_READ handler
1156 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1158 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1159 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1160 struct idx_info *req_ii, *rep_ii;
1165 memset(rdpg, 0, sizeof(*rdpg));
1166 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1168 /* extract idx_info buffer from request & reply */
1169 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1170 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1171 RETURN(err_serious(-EPROTO));
1173 rc = req_capsule_server_pack(tsi->tsi_pill);
1175 RETURN(err_serious(rc));
1177 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1179 RETURN(err_serious(-EFAULT));
1180 rep_ii->ii_magic = IDX_INFO_MAGIC;
1182 /* extract hash to start with */
1183 rdpg->rp_hash = req_ii->ii_hash_start;
1185 /* extract requested attributes */
1186 rdpg->rp_attrs = req_ii->ii_attrs;
1188 /* check that fid packed in request is valid and supported */
1189 if (!fid_is_sane(&req_ii->ii_fid))
1191 rep_ii->ii_fid = req_ii->ii_fid;
1194 rep_ii->ii_flags = req_ii->ii_flags;
1196 /* compute number of pages to allocate, ii_count is the number of 4KB
1198 if (req_ii->ii_count <= 0)
1199 GOTO(out, rc = -EFAULT);
1200 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1201 exp_max_brw_size(tsi->tsi_exp));
1202 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1204 /* allocate pages to store the containers */
1205 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1206 if (rdpg->rp_pages == NULL)
1207 GOTO(out, rc = -ENOMEM);
1208 for (i = 0; i < rdpg->rp_npages; i++) {
1209 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1210 if (rdpg->rp_pages[i] == NULL)
1211 GOTO(out, rc = -ENOMEM);
1214 /* populate pages with key/record pairs */
1215 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1219 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1220 "asked %d > %d\n", rc, rdpg->rp_count);
1222 /* send pages to client */
1223 rc = tgt_sendpage(tsi, rdpg, rc);
1228 if (rdpg->rp_pages) {
1229 for (i = 0; i < rdpg->rp_npages; i++)
1230 if (rdpg->rp_pages[i])
1231 __free_page(rdpg->rp_pages[i]);
1232 OBD_FREE(rdpg->rp_pages,
1233 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1238 struct tgt_handler tgt_obd_handlers[] = {
1239 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1240 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1241 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1243 EXPORT_SYMBOL(tgt_obd_handlers);
1245 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1246 struct dt_object *obj, __u64 start, __u64 end)
1252 /* if no objid is specified, it means "sync whole filesystem" */
1254 rc = dt_sync(env, tgt->lut_bottom);
1255 } else if (dt_version_get(env, obj) >
1256 tgt->lut_obd->obd_last_committed) {
1257 rc = dt_object_sync(env, obj, start, end);
1259 atomic_inc(&tgt->lut_sync_count);
1263 EXPORT_SYMBOL(tgt_sync);
1265 * Unified target DLM handlers.
1269 * Unified target BAST
1271 * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1272 * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1273 * but if it's cross-MDT lock, because remote object version is not set, a
1274 * filesystem sync is needed.
1276 * \param lock server side lock
1277 * \param desc lock desc
1278 * \param data ldlm_cb_set_arg
1279 * \param flag indicates whether this cancelling or blocking callback
1280 * \retval 0 on success
1281 * \retval negative number on error
1283 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1284 void *data, int flag)
1287 struct lu_target *tgt;
1288 struct dt_object *obj = NULL;
1294 tgt = class_exp2tgt(lock->l_export);
1296 if (unlikely(tgt == NULL)) {
1297 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1298 class_exp2obd(lock->l_export)->obd_name);
1302 if (flag == LDLM_CB_CANCELING &&
1303 (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1304 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1305 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1306 ldlm_is_cbpending(lock))) &&
1307 ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1308 lock->l_resource->lr_type == LDLM_EXTENT)) {
1310 __u64 end = OBD_OBJECT_EOF;
1312 rc = lu_env_init(&env, LCT_DT_THREAD);
1313 if (unlikely(rc != 0))
1316 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1317 tgt->lut_lsd.lsd_osd_index);
1319 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1320 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1322 GOTO(err_env, rc = PTR_ERR(obj));
1324 if (!dt_object_exists(obj))
1325 GOTO(err_put, rc = -ENOENT);
1327 start = lock->l_policy_data.l_extent.start;
1328 end = lock->l_policy_data.l_extent.end;
1331 rc = tgt_sync(&env, tgt, obj, start, end);
1333 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1334 "cancel: rc = %d\n",
1335 tgt_name(tgt), PFID(&fid),
1336 lock->l_policy_data.l_extent.start,
1337 lock->l_policy_data.l_extent.end, rc);
1341 lu_object_put(&env, &obj->do_lu);
1346 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1350 static struct ldlm_callback_suite tgt_dlm_cbs = {
1351 .lcs_completion = ldlm_server_completion_ast,
1352 .lcs_blocking = tgt_blocking_ast,
1353 .lcs_glimpse = ldlm_server_glimpse_ast
1356 int tgt_enqueue(struct tgt_session_info *tsi)
1358 struct ptlrpc_request *req = tgt_ses_req(tsi);
1363 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1364 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1366 LASSERT(tsi->tsi_dlm_req != NULL);
1367 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1368 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1370 RETURN(err_serious(rc));
1372 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1374 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1377 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1380 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1383 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1386 RETURN(req->rq_status);
1388 EXPORT_SYMBOL(tgt_enqueue);
1390 int tgt_convert(struct tgt_session_info *tsi)
1392 struct ptlrpc_request *req = tgt_ses_req(tsi);
1396 LASSERT(tsi->tsi_dlm_req);
1397 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1399 RETURN(err_serious(rc));
1401 RETURN(req->rq_status);
1404 int tgt_bl_callback(struct tgt_session_info *tsi)
1406 return err_serious(-EOPNOTSUPP);
1409 int tgt_cp_callback(struct tgt_session_info *tsi)
1411 return err_serious(-EOPNOTSUPP);
1414 /* generic LDLM target handler */
1415 struct tgt_handler tgt_dlm_handlers[] = {
1416 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1417 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1418 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1419 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1421 EXPORT_SYMBOL(tgt_dlm_handlers);
1424 * Unified target LLOG handlers.
1426 int tgt_llog_open(struct tgt_session_info *tsi)
1432 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1436 EXPORT_SYMBOL(tgt_llog_open);
1438 int tgt_llog_close(struct tgt_session_info *tsi)
1444 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1448 EXPORT_SYMBOL(tgt_llog_close);
1451 int tgt_llog_destroy(struct tgt_session_info *tsi)
1457 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1462 int tgt_llog_read_header(struct tgt_session_info *tsi)
1468 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1472 EXPORT_SYMBOL(tgt_llog_read_header);
1474 int tgt_llog_next_block(struct tgt_session_info *tsi)
1480 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1484 EXPORT_SYMBOL(tgt_llog_next_block);
1486 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1492 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1496 EXPORT_SYMBOL(tgt_llog_prev_block);
1498 /* generic llog target handler */
1499 struct tgt_handler tgt_llog_handlers[] = {
1500 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1501 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1502 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1503 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1504 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1505 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1507 EXPORT_SYMBOL(tgt_llog_handlers);
1510 * sec context handlers
1512 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1513 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1518 struct tgt_handler tgt_sec_ctx_handlers[] = {
1519 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1520 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1521 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1523 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1525 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1526 struct dt_device *key,
1527 struct lfsck_request *lr,
1528 struct thandle *th) = NULL;
1530 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1532 struct lfsck_request *,
1535 tgt_lfsck_in_notify = notify;
1537 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1539 static int (*tgt_lfsck_query)(const struct lu_env *env,
1540 struct dt_device *key,
1541 struct lfsck_request *req,
1542 struct lfsck_reply *rep,
1543 struct lfsck_query *que) = NULL;
1545 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1547 struct lfsck_request *,
1548 struct lfsck_reply *,
1549 struct lfsck_query *))
1551 tgt_lfsck_query = query;
1553 EXPORT_SYMBOL(tgt_register_lfsck_query);
1555 /* LFSCK request handlers */
1556 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1558 const struct lu_env *env = tsi->tsi_env;
1559 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1560 struct lfsck_request *lr;
1564 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1568 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1573 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1575 struct lfsck_request *request;
1576 struct lfsck_reply *reply;
1580 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1581 if (request == NULL)
1584 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1588 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1589 request, reply, NULL);
1591 RETURN(rc < 0 ? rc : 0);
1594 struct tgt_handler tgt_lfsck_handlers[] = {
1595 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1596 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1598 EXPORT_SYMBOL(tgt_lfsck_handlers);
1601 * initialize per-thread page pool (bug 5137).
1603 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1605 struct tgt_thread_big_cache *tbc;
1609 LASSERT(thread != NULL);
1610 LASSERT(thread->t_data == NULL);
1612 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1615 thread->t_data = tbc;
1618 EXPORT_SYMBOL(tgt_io_thread_init);
1621 * free per-thread pool created by tgt_thread_init().
1623 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1625 struct tgt_thread_big_cache *tbc;
1629 LASSERT(thread != NULL);
1632 * be prepared to handle partially-initialized pools (because this is
1633 * called from ost_io_thread_init() for cleanup.
1635 tbc = thread->t_data;
1637 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1638 thread->t_data = NULL;
1642 EXPORT_SYMBOL(tgt_io_thread_done);
1644 * Helper function for getting server side [start, start+count] DLM lock
1645 * if asked by client.
1647 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1648 __u64 start, __u64 end, struct lustre_handle *lh,
1649 int mode, __u64 *flags)
1651 union ldlm_policy_data policy;
1656 LASSERT(lh != NULL);
1657 LASSERT(ns != NULL);
1658 LASSERT(!lustre_handle_is_used(lh));
1660 policy.l_extent.gid = 0;
1661 policy.l_extent.start = start & PAGE_MASK;
1664 * If ->o_blocks is EOF it means "lock till the end of the file".
1665 * Otherwise, it's size of an extent or hole being punched (in bytes).
1667 if (end == OBD_OBJECT_EOF || end < start)
1668 policy.l_extent.end = OBD_OBJECT_EOF;
1670 policy.l_extent.end = end | ~PAGE_MASK;
1672 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1673 flags, ldlm_blocking_ast,
1674 ldlm_completion_ast, ldlm_glimpse_ast,
1675 NULL, 0, LVB_T_NONE, NULL, lh);
1676 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1678 EXPORT_SYMBOL(tgt_extent_lock);
1680 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1682 LASSERT(lustre_handle_is_used(lh));
1683 ldlm_lock_decref(lh, mode);
1685 EXPORT_SYMBOL(tgt_extent_unlock);
1687 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1688 struct obd_ioobj *obj, struct niobuf_remote *nb,
1689 struct lustre_handle *lh, enum ldlm_mode mode)
1692 int nrbufs = obj->ioo_bufcnt;
1697 LASSERT(mode == LCK_PR || mode == LCK_PW);
1698 LASSERT(!lustre_handle_is_used(lh));
1700 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1703 for (i = 1; i < nrbufs; i++)
1704 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1707 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1708 nb[nrbufs - 1].rnb_offset +
1709 nb[nrbufs - 1].rnb_len - 1,
1713 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1714 struct lustre_handle *lh, enum ldlm_mode mode)
1718 LASSERT(mode == LCK_PR || mode == LCK_PW);
1719 LASSERT((obj->ioo_bufcnt > 0 &&
1720 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1721 lustre_handle_is_used(lh));
1723 if (lustre_handle_is_used(lh))
1724 tgt_extent_unlock(lh, mode);
1728 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1729 struct ptlrpc_bulk_desc *desc, int opc,
1730 cksum_type_t cksum_type)
1732 struct cfs_crypto_hash_desc *hdesc;
1733 unsigned int bufsize;
1735 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1738 LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1740 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1741 if (IS_ERR(hdesc)) {
1742 CERROR("%s: unable to initialize checksum hash %s\n",
1743 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1744 return PTR_ERR(hdesc);
1747 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1748 for (i = 0; i < desc->bd_iov_count; i++) {
1749 /* corrupt the data before we compute the checksum, to
1750 * simulate a client->OST data error */
1751 if (i == 0 && opc == OST_WRITE &&
1752 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1753 int off = BD_GET_KIOV(desc, i).kiov_offset &
1755 int len = BD_GET_KIOV(desc, i).kiov_len;
1756 struct page *np = tgt_page_to_corrupt;
1757 char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1760 char *ptr2 = kmap(np) + off;
1762 memcpy(ptr2, ptr, len);
1763 memcpy(ptr2, "bad3", min(4, len));
1765 BD_GET_KIOV(desc, i).kiov_page = np;
1767 CERROR("%s: can't alloc page for corruption\n",
1771 cfs_crypto_hash_update_page(hdesc,
1772 BD_GET_KIOV(desc, i).kiov_page,
1773 BD_GET_KIOV(desc, i).kiov_offset &
1775 BD_GET_KIOV(desc, i).kiov_len);
1777 /* corrupt the data after we compute the checksum, to
1778 * simulate an OST->client data error */
1779 if (i == 0 && opc == OST_READ &&
1780 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1781 int off = BD_GET_KIOV(desc, i).kiov_offset
1783 int len = BD_GET_KIOV(desc, i).kiov_len;
1784 struct page *np = tgt_page_to_corrupt;
1786 kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1789 char *ptr2 = kmap(np) + off;
1791 memcpy(ptr2, ptr, len);
1792 memcpy(ptr2, "bad4", min(4, len));
1794 BD_GET_KIOV(desc, i).kiov_page = np;
1796 CERROR("%s: can't alloc page for corruption\n",
1802 bufsize = sizeof(cksum);
1803 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1808 int tgt_brw_read(struct tgt_session_info *tsi)
1810 struct ptlrpc_request *req = tgt_ses_req(tsi);
1811 struct ptlrpc_bulk_desc *desc = NULL;
1812 struct obd_export *exp = tsi->tsi_exp;
1813 struct niobuf_remote *remote_nb;
1814 struct niobuf_local *local_nb;
1815 struct obd_ioobj *ioo;
1816 struct ost_body *body, *repbody;
1817 struct l_wait_info lwi;
1818 struct lustre_handle lockh = { 0 };
1819 int npages, nob = 0, rc, i, no_reply = 0;
1820 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1824 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1825 CERROR("%s: deny read request from %s to portal %u\n",
1826 tgt_name(tsi->tsi_tgt),
1827 obd_export_nid2str(req->rq_export),
1828 ptlrpc_req2svc(req)->srv_req_portal);
1832 req->rq_bulk_read = 1;
1834 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1837 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1838 cfs_fail_val : (obd_timeout + 1) / 4);
1840 /* Check if there is eviction in progress, and if so, wait for it to
1842 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1843 /* We do not care how long it takes */
1844 lwi = LWI_INTR(NULL, NULL);
1845 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1846 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1850 /* There must be big cache in current thread to process this request
1851 * if it is NULL then something went wrong and it wasn't allocated,
1852 * report -ENOMEM in that case */
1856 body = tsi->tsi_ost_body;
1857 LASSERT(body != NULL);
1859 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1860 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1862 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1863 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1865 local_nb = tbc->local;
1867 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1868 remote_nb, &lockh, LCK_PR);
1873 * If getting the lock took more time than
1874 * client was willing to wait, drop it. b=11330
1876 if (cfs_time_current_sec() > req->rq_deadline ||
1877 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1879 CERROR("Dropping timed-out read from %s because locking"
1880 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1881 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1882 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1883 req->rq_deadline - req->rq_arrival_time.tv_sec);
1884 GOTO(out_lock, rc = -ETIMEDOUT);
1887 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1888 repbody->oa = body->oa;
1890 npages = PTLRPC_MAX_BRW_PAGES;
1891 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1892 ioo, remote_nb, &npages, local_nb);
1896 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1897 PTLRPC_BULK_PUT_SOURCE |
1898 PTLRPC_BULK_BUF_KIOV,
1900 &ptlrpc_bulk_kiov_nopin_ops);
1902 GOTO(out_commitrw, rc = -ENOMEM);
1905 for (i = 0; i < npages; i++) {
1906 int page_rc = local_nb[i].lnb_rc;
1914 if (page_rc != 0) { /* some data! */
1915 LASSERT(local_nb[i].lnb_page != NULL);
1916 desc->bd_frag_ops->add_kiov_frag
1917 (desc, local_nb[i].lnb_page,
1918 local_nb[i].lnb_page_offset,
1922 if (page_rc != local_nb[i].lnb_len) { /* short read */
1923 /* All subsequent pages should be 0 */
1924 while (++i < npages)
1925 LASSERT(local_nb[i].lnb_rc == 0);
1929 if (OBD_FAIL_CHECK(OBD_FAIL_OST_READ_SIZE) &&
1930 nob != cfs_fail_val)
1933 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1934 cksum_type_t cksum_type =
1935 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1936 body->oa.o_flags : 0);
1937 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1938 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1939 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1940 OST_READ, cksum_type);
1941 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1942 repbody->oa.o_cksum);
1944 repbody->oa.o_valid = 0;
1946 /* We're finishing using body->oa as an input variable */
1948 /* Check if client was evicted while we were doing i/o before touching
1950 if (likely(rc == 0 &&
1951 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2) &&
1952 !CFS_FAIL_CHECK(OBD_FAIL_PTLRPC_DROP_BULK))) {
1953 rc = target_bulk_io(exp, desc, &lwi);
1958 /* Must commit after prep above in all cases */
1959 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1960 remote_nb, npages, local_nb, rc);
1962 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1964 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1965 ptlrpc_free_bulk(desc);
1970 ptlrpc_lprocfs_brw(req, nob);
1971 } else if (no_reply) {
1972 req->rq_no_reply = 1;
1973 /* reply out callback would free */
1974 ptlrpc_req_drop_rs(req);
1975 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1976 "client will retry: rc %d\n",
1977 exp->exp_obd->obd_name,
1978 obd_uuid2str(&exp->exp_client_uuid),
1979 obd_export_nid2str(exp), rc);
1981 /* send a bulk after reply to simulate a network delay or reordering
1983 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1984 wait_queue_head_t waitq;
1985 struct l_wait_info lwi1;
1987 CDEBUG(D_INFO, "reorder BULK\n");
1988 init_waitqueue_head(&waitq);
1990 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1991 l_wait_event(waitq, 0, &lwi1);
1992 target_bulk_io(exp, desc, &lwi);
1993 ptlrpc_free_bulk(desc);
1998 EXPORT_SYMBOL(tgt_brw_read);
2000 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
2001 struct ptlrpc_bulk_desc *desc,
2002 struct niobuf_local *local_nb, int npages,
2003 u32 client_cksum, u32 server_cksum,
2006 struct obd_export *exp = req->rq_export;
2007 struct ost_body *body;
2011 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2012 LASSERT(body != NULL);
2014 if (req->rq_peer.nid != desc->bd_sender) {
2016 router = libcfs_nid2str(desc->bd_sender);
2020 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
2021 client_cksum, server_cksum);
2025 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
2026 DFID" object "DOSTID" extent ["LPU64"-"LPU64
2027 "]: client csum %x, server csum %x\n",
2028 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
2030 body->oa.o_valid & OBD_MD_FLFID ?
2031 body->oa.o_parent_seq : (__u64)0,
2032 body->oa.o_valid & OBD_MD_FLFID ?
2033 body->oa.o_parent_oid : 0,
2034 body->oa.o_valid & OBD_MD_FLFID ?
2035 body->oa.o_parent_ver : 0,
2036 POSTID(&body->oa.o_oi),
2037 local_nb[0].lnb_file_offset,
2038 local_nb[npages-1].lnb_file_offset +
2039 local_nb[npages - 1].lnb_len - 1,
2040 client_cksum, server_cksum);
2043 int tgt_brw_write(struct tgt_session_info *tsi)
2045 struct ptlrpc_request *req = tgt_ses_req(tsi);
2046 struct ptlrpc_bulk_desc *desc = NULL;
2047 struct obd_export *exp = req->rq_export;
2048 struct niobuf_remote *remote_nb;
2049 struct niobuf_local *local_nb;
2050 struct obd_ioobj *ioo;
2051 struct ost_body *body, *repbody;
2052 struct l_wait_info lwi;
2053 struct lustre_handle lockh = {0};
2055 int objcount, niocount, npages;
2057 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
2058 bool no_reply = false, mmap;
2059 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2063 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2064 CERROR("%s: deny write request from %s to portal %u\n",
2065 tgt_name(tsi->tsi_tgt),
2066 obd_export_nid2str(req->rq_export),
2067 ptlrpc_req2svc(req)->srv_req_portal);
2068 RETURN(err_serious(-EPROTO));
2071 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2072 RETURN(err_serious(-ENOSPC));
2073 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2074 RETURN(err_serious(-EROFS));
2076 req->rq_bulk_write = 1;
2078 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2079 RETURN(err_serious(-EIO));
2080 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2081 RETURN(err_serious(-EFAULT));
2083 /* pause before transaction has been started */
2084 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2085 cfs_fail_val : (obd_timeout + 1) / 4);
2087 /* There must be big cache in current thread to process this request
2088 * if it is NULL then something went wrong and it wasn't allocated,
2089 * report -ENOMEM in that case */
2093 body = tsi->tsi_ost_body;
2094 LASSERT(body != NULL);
2096 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2097 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2099 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2100 RCL_CLIENT) / sizeof(*ioo);
2102 for (niocount = i = 0; i < objcount; i++)
2103 niocount += ioo[i].ioo_bufcnt;
2105 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2106 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2107 if (niocount != req_capsule_get_size(&req->rq_pill,
2108 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2110 RETURN(err_serious(-EPROTO));
2112 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2113 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2114 memory_pressure_set();
2116 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2117 niocount * sizeof(*rcs));
2118 rc = req_capsule_server_pack(&req->rq_pill);
2120 GOTO(out, rc = err_serious(rc));
2122 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2123 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2125 local_nb = tbc->local;
2127 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2128 remote_nb, &lockh, LCK_PW);
2133 * If getting the lock took more time than
2134 * client was willing to wait, drop it. b=11330
2136 if (cfs_time_current_sec() > req->rq_deadline ||
2137 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2139 CERROR("%s: Dropping timed-out write from %s because locking "
2140 "object "DOSTID" took %ld seconds (limit was %ld).\n",
2141 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2142 POSTID(&ioo->ioo_oid),
2143 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2144 req->rq_deadline - req->rq_arrival_time.tv_sec);
2145 GOTO(out_lock, rc = -ETIMEDOUT);
2148 /* Because we already sync grant info with client when reconnect,
2149 * grant info will be cleared for resent req, then fed_grant and
2150 * total_grant will not be modified in following preprw_write */
2151 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2152 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2153 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2156 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2157 if (repbody == NULL)
2158 GOTO(out_lock, rc = -ENOMEM);
2159 repbody->oa = body->oa;
2161 npages = PTLRPC_MAX_BRW_PAGES;
2162 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2163 objcount, ioo, remote_nb, &npages, local_nb);
2167 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2168 PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2170 &ptlrpc_bulk_kiov_nopin_ops);
2172 GOTO(skip_transfer, rc = -ENOMEM);
2174 /* NB Having prepped, we must commit... */
2175 for (i = 0; i < npages; i++)
2176 desc->bd_frag_ops->add_kiov_frag(desc,
2177 local_nb[i].lnb_page,
2178 local_nb[i].lnb_page_offset,
2179 local_nb[i].lnb_len);
2181 rc = sptlrpc_svc_prep_bulk(req, desc);
2183 GOTO(skip_transfer, rc);
2185 rc = target_bulk_io(exp, desc, &lwi);
2189 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2190 static int cksum_counter;
2192 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2193 cksum_type = cksum_type_unpack(body->oa.o_flags);
2195 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2196 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2197 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2198 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2199 OST_WRITE, cksum_type);
2202 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2203 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2204 body->oa.o_flags & OBD_FL_MMAP);
2206 tgt_warn_on_cksum(req, desc, local_nb, npages,
2208 repbody->oa.o_cksum, mmap);
2210 } else if ((cksum_counter & (-cksum_counter)) ==
2212 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2213 cksum_counter, libcfs_id2str(req->rq_peer),
2214 repbody->oa.o_cksum);
2218 /* Must commit after prep above in all cases */
2219 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2220 objcount, ioo, remote_nb, npages, local_nb, rc);
2221 if (rc == -ENOTCONN)
2222 /* quota acquire process has been given up because
2223 * either the client has been evicted or the client
2224 * has timed out the request already */
2228 * Disable sending mtime back to the client. If the client locked the
2229 * whole object, then it has already updated the mtime on its side,
2230 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2232 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2237 /* set per-requested niobuf return codes */
2238 for (i = j = 0; i < niocount; i++) {
2239 int len = remote_nb[i].rnb_len;
2244 LASSERT(j < npages);
2245 if (local_nb[j].lnb_rc < 0)
2246 rcs[i] = local_nb[j].lnb_rc;
2247 len -= local_nb[j].lnb_len;
2252 LASSERT(j == npages);
2253 ptlrpc_lprocfs_brw(req, nob);
2256 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2258 ptlrpc_free_bulk(desc);
2261 req->rq_no_reply = 1;
2262 /* reply out callback would free */
2263 ptlrpc_req_drop_rs(req);
2264 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2265 "client will retry: rc %d\n",
2266 exp->exp_obd->obd_name,
2267 obd_uuid2str(&exp->exp_client_uuid),
2268 obd_export_nid2str(exp), rc);
2270 memory_pressure_clr();
2273 EXPORT_SYMBOL(tgt_brw_write);
2275 /* Check if request can be reconstructed from saved reply data
2276 * A copy of the reply data is returned in @trd if the pointer is not NULL
2278 bool req_can_reconstruct(struct ptlrpc_request *req,
2279 struct tg_reply_data *trd)
2281 struct tg_export_data *ted = &req->rq_export->exp_target_data;
2282 struct lsd_client_data *lcd = ted->ted_lcd;
2285 if (tgt_is_multimodrpcs_client(req->rq_export))
2286 return tgt_lookup_reply(req, trd);
2288 mutex_lock(&ted->ted_lcd_lock);
2289 found = req->rq_xid == lcd->lcd_last_xid ||
2290 req->rq_xid == lcd->lcd_last_close_xid;
2292 if (found && trd != NULL) {
2293 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2294 trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2295 trd->trd_reply.lrd_transno =
2296 lcd->lcd_last_close_transno;
2297 trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2299 trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2300 trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2301 trd->trd_reply.lrd_result = lcd->lcd_last_result;
2302 trd->trd_reply.lrd_data = lcd->lcd_last_data;
2303 trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2304 trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2305 trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2306 trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2309 mutex_unlock(&ted->ted_lcd_lock);
2313 EXPORT_SYMBOL(req_can_reconstruct);