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 = tsi->tsi_exp->exp_target_data.ted_nodemap;
260 body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
261 NODEMAP_CLIENT_TO_FS,
263 body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
264 NODEMAP_CLIENT_TO_FS,
267 tsi->tsi_ost_body = body;
268 tsi->tsi_fid = body->oa.o_oi.oi_fid;
270 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
271 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
276 if (!(body->oa.o_valid & OBD_MD_FLID)) {
277 if (flags & HABEO_CORPUS) {
278 CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
279 "but OID/FID is mandatory with HABEO_CORPUS\n",
280 tgt_name(tsi->tsi_tgt));
287 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
290 * OST doesn't get object in advance for further use to prevent
291 * situations with nested object_find which is potential deadlock.
293 tsi->tsi_corpus = NULL;
298 * Do necessary preprocessing according to handler ->th_flags.
300 static int tgt_request_preprocess(struct tgt_session_info *tsi,
301 struct tgt_handler *h,
302 struct ptlrpc_request *req)
304 struct req_capsule *pill = tsi->tsi_pill;
305 __u32 flags = h->th_flags;
310 if (tsi->tsi_preprocessed)
313 LASSERT(h->th_act != NULL);
314 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
315 LASSERT(current->journal_info == NULL);
317 LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
319 if (h->th_fmt != NULL) {
320 req_capsule_set(pill, h->th_fmt);
321 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
322 rc = tgt_mdt_body_unpack(tsi, flags);
325 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
327 rc = tgt_ost_body_unpack(tsi, flags);
333 if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
336 if (flags & HABEO_CLAVIS) {
337 struct ldlm_request *dlm_req;
339 LASSERT(h->th_fmt != NULL);
341 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
342 if (dlm_req != NULL) {
343 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
345 dlm_req->lock_desc.l_policy_data.\
346 l_inodebits.bits == 0)) {
348 * Lock without inodebits makes no sense and
349 * will oops later in ldlm. If client miss to
350 * set such bits, do not trigger ASSERTION.
352 * For liblustre flock case, it maybe zero.
356 tsi->tsi_dlm_req = dlm_req;
362 tsi->tsi_preprocessed = 1;
367 * Invoke handler for this request opc. Also do necessary preprocessing
368 * (according to handler ->th_flags), and post-processing (setting of
369 * ->last_{xid,committed}).
371 static int tgt_handle_request0(struct tgt_session_info *tsi,
372 struct tgt_handler *h,
373 struct ptlrpc_request *req)
377 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
382 /* When dealing with sec context requests, no export is associated yet,
383 * because these requests are sent before *_CONNECT requests.
384 * A NULL req->rq_export means the normal *_common_slice handlers will
385 * not be called, because there is no reference to the target.
386 * So deal with them by hand and jump directly to target_send_reply().
390 case SEC_CTX_INIT_CONT:
392 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
397 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
398 * to put same checks into handlers like mdt_close(), mdt_reint(),
399 * etc., without talking to mdt authors first. Checking same thing
400 * there again is useless and returning 0 error without packing reply
401 * is buggy! Handlers either pack reply or return error.
403 * We return 0 here and do not send any reply in order to emulate
404 * network failure. Do not send any reply in case any of NET related
405 * fail_id has occured.
407 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
409 if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
410 OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
413 rc = tgt_request_preprocess(tsi, h, req);
414 /* pack reply if reply format is fixed */
415 if (rc == 0 && h->th_flags & HABEO_REFERO) {
417 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
419 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
421 tsi->tsi_mdt_body->mbo_eadatasize);
422 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
424 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
427 rc = req_capsule_server_pack(tsi->tsi_pill);
430 if (likely(rc == 0)) {
432 * Process request, there can be two types of rc:
433 * 1) errors with msg unpack/pack, other failures outside the
434 * operation itself. This is counted as serious errors;
435 * 2) errors during fs operation, should be placed in rq_status
439 if (!is_serious(rc) &&
440 !req->rq_no_reply && req->rq_reply_state == NULL) {
441 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
442 "pack reply and returned 0 error\n",
443 tgt_name(tsi->tsi_tgt), h->th_name);
446 serious = is_serious(rc);
447 rc = clear_serious(rc);
455 * ELDLM_* codes which > 0 should be in rq_status only as well as
456 * all non-serious errors.
458 if (rc > 0 || !serious)
461 LASSERT(current->journal_info == NULL);
463 if (likely(rc == 0 && req->rq_export))
464 target_committed_to_req(req);
467 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
471 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
472 struct obd_device *obd, int *process)
474 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
481 case MDS_SYNC: /* used in unmounting */
495 case MDS_HSM_PROGRESS:
496 case MDS_HSM_STATE_SET:
497 case MDS_HSM_REQUEST:
498 *process = target_queue_recovery_request(req, obd);
502 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
509 * Handle recovery. Return:
510 * +1: continue request processing;
511 * -ve: abort immediately with the given error code;
512 * 0: send reply with error code in req->rq_status;
514 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
518 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
523 case SEC_CTX_INIT_CONT:
528 if (!req->rq_export->exp_obd->obd_replayable)
531 /* sanity check: if the xid matches, the request must be marked as a
532 * resent or replayed */
533 if (req_can_reconstruct(req, NULL)) {
534 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
535 (MSG_RESENT | MSG_REPLAY))) {
536 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
537 "saved xid, expected REPLAY or RESENT flag "
539 lustre_msg_get_flags(req->rq_reqmsg));
540 req->rq_status = -ENOTCONN;
544 /* else: note the opposite is not always true; a RESENT req after a
545 * failover will usually not match the last_xid, since it was likely
546 * never committed. A REPLAYed request will almost never match the
547 * last xid, however it could for a committed, but still retained,
550 /* Check for aborted recovery... */
551 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
555 DEBUG_REQ(D_INFO, req, "Got new replay");
556 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
558 if (rc != 0 || !should_process)
560 else if (should_process < 0) {
561 req->rq_status = should_process;
562 rc = ptlrpc_error(req);
569 /* Initial check for request, it is validation mostly */
570 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
572 struct tgt_handler *h;
573 struct tgt_opc_slice *s;
574 struct lu_target *tgt;
575 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
579 tgt = class_exp2tgt(req->rq_export);
580 if (unlikely(tgt == NULL)) {
581 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
582 class_exp2obd(req->rq_export)->obd_name);
583 RETURN(ERR_PTR(-EINVAL));
586 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
587 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
590 /* opcode was not found in slice */
591 if (unlikely(s->tos_hs == NULL)) {
592 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
594 RETURN(ERR_PTR(-ENOTSUPP));
597 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
598 h = s->tos_hs + (opc - s->tos_opc_start);
599 if (unlikely(h->th_opc == 0)) {
600 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
601 RETURN(ERR_PTR(-ENOTSUPP));
607 int tgt_request_handle(struct ptlrpc_request *req)
609 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
611 struct lustre_msg *msg = req->rq_reqmsg;
612 struct tgt_handler *h;
613 struct lu_target *tgt;
614 int request_fail_id = 0;
615 __u32 opc = lustre_msg_get_opc(msg);
620 /* Refill the context, to make sure all thread keys are allocated */
621 lu_env_refill(req->rq_svc_thread->t_env);
623 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
624 tsi->tsi_pill = &req->rq_pill;
625 tsi->tsi_env = req->rq_svc_thread->t_env;
627 /* if request has export then get handlers slice from corresponding
628 * target, otherwise that should be connect operation */
629 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
630 opc == MGS_CONNECT) {
631 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
632 rc = target_handle_connect(req);
634 rc = ptlrpc_error(req);
637 /* recovery-small test 18c asks to drop connect reply */
638 if (unlikely(opc == OST_CONNECT &&
639 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
643 if (unlikely(!class_connected_export(req->rq_export))) {
644 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
645 opc == SEC_CTX_FINI) {
646 /* sec context initialization has to be handled
647 * by hand in tgt_handle_request0() */
648 tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
650 GOTO(handle_recov, rc = 0);
652 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
653 opc, libcfs_id2str(req->rq_peer));
654 req->rq_status = -ENOTCONN;
655 rc = ptlrpc_error(req);
659 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
660 tsi->tsi_exp = req->rq_export;
661 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
662 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
664 tsi->tsi_jobid = NULL;
667 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
668 class_exp2obd(req->rq_export)->obd_name);
669 req->rq_status = -EINVAL;
670 rc = ptlrpc_error(req);
674 /* check request's xid is consistent with export's last_xid */
675 if (req->rq_export != NULL) {
676 __u64 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
678 req->rq_export->exp_last_xid = last_xid;
679 if (req->rq_xid == 0 ||
680 req->rq_xid <= req->rq_export->exp_last_xid) {
681 DEBUG_REQ(D_ERROR, req,
682 "Unexpected xid %llx vs. last_xid %llx\n",
683 req->rq_xid, req->rq_export->exp_last_xid);
684 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 93, 0)
687 req->rq_status = -EPROTO;
688 rc = ptlrpc_error(req);
693 request_fail_id = tgt->lut_request_fail_id;
694 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
696 /* try to release in-memory reply data */
697 if (tgt_is_multimodrpcs_client(req->rq_export)) {
698 tgt_handle_received_xid(req->rq_export,
699 lustre_msg_get_last_xid(req->rq_reqmsg));
700 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
701 (MSG_RESENT | MSG_REPLAY)))
702 tgt_handle_tag(req->rq_export,
703 lustre_msg_get_tag(req->rq_reqmsg));
706 h = tgt_handler_find_check(req);
708 req->rq_status = PTR_ERR(h);
709 rc = ptlrpc_error(req);
713 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
716 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
719 rc = lustre_msg_check_version(msg, h->th_version);
721 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
722 " %08x, expecting %08x\n", tgt_name(tgt),
723 lustre_msg_get_version(msg), h->th_version);
724 req->rq_status = -EINVAL;
725 rc = ptlrpc_error(req);
730 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
731 if (likely(rc == 1)) {
732 rc = tgt_handle_request0(tsi, h, req);
738 req_capsule_fini(tsi->tsi_pill);
739 if (tsi->tsi_corpus != NULL) {
740 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
741 tsi->tsi_corpus = NULL;
745 EXPORT_SYMBOL(tgt_request_handle);
747 /** Assign high priority operations to the request if needed. */
748 int tgt_hpreq_handler(struct ptlrpc_request *req)
750 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
751 struct tgt_handler *h;
756 if (req->rq_export == NULL)
759 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
760 tsi->tsi_pill = &req->rq_pill;
761 tsi->tsi_env = req->rq_svc_thread->t_env;
762 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
763 tsi->tsi_exp = req->rq_export;
765 h = tgt_handler_find_check(req);
771 rc = tgt_request_preprocess(tsi, h, req);
772 if (unlikely(rc != 0))
775 if (h->th_hp != NULL)
779 EXPORT_SYMBOL(tgt_hpreq_handler);
781 void tgt_counter_incr(struct obd_export *exp, int opcode)
783 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
784 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
785 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
787 EXPORT_SYMBOL(tgt_counter_incr);
790 * Unified target generic handlers.
796 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
798 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
799 OBD_CONNECT_RMT_CLIENT_FORCE);
802 static int tgt_init_sec_level(struct ptlrpc_request *req)
804 struct lu_target *tgt = class_exp2tgt(req->rq_export);
806 struct obd_connect_data *data, *reply;
811 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
812 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
813 if (data == NULL || reply == NULL)
816 /* connection from MDT is always trusted */
817 if (req->rq_auth_usr_mdt) {
818 tgt_init_sec_none(reply);
822 if (unlikely(tgt == NULL)) {
823 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
824 class_exp2obd(req->rq_export)->obd_name);
828 client = libcfs_nid2str(req->rq_peer.nid);
829 /* no GSS support case */
830 if (!req->rq_auth_gss) {
831 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
832 CWARN("client %s -> target %s does not use GSS, "
833 "can not run under security level %d.\n",
834 client, tgt_name(tgt), tgt->lut_sec_level);
837 tgt_init_sec_none(reply);
842 /* old version case */
843 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
844 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
845 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
846 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
847 CWARN("client %s -> target %s uses old version, "
848 "can not run under security level %d.\n",
849 client, tgt_name(tgt), tgt->lut_sec_level);
852 CWARN("client %s -> target %s uses old version, "
853 "run under security level %d.\n",
854 client, tgt_name(tgt), tgt->lut_sec_level);
855 tgt_init_sec_none(reply);
860 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
862 if (!req->rq_auth_remote)
863 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
864 "asked to be remote.\n", client, tgt_name(tgt));
865 } else if (req->rq_auth_remote) {
867 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
868 "as remote by default.\n", client, tgt_name(tgt));
872 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
873 CDEBUG(D_SEC, "client %s -> target %s: user is not "
874 "authenticated!\n", client, tgt_name(tgt));
880 switch (tgt->lut_sec_level) {
881 case LUSTRE_SEC_NONE:
884 "client %s -> target %s is set as remote, "
885 "can not run under security level %d.\n",
886 client, tgt_name(tgt), tgt->lut_sec_level);
889 tgt_init_sec_none(reply);
891 case LUSTRE_SEC_REMOTE:
893 tgt_init_sec_none(reply);
898 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
899 OBD_CONNECT_RMT_CLIENT_FORCE);
900 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
901 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
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 client
940 * is loopback, allow any flavor */
941 if ((strcmp(exp->exp_obd->obd_type->typ_name,
942 LUSTRE_MGS_NAME) == 0) &&
943 (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
944 LNET_NETTYP(LNET_NIDNET(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, int initial)
975 struct sptlrpc_rule_set tmp_rset;
978 if (unlikely(tgt == NULL)) {
979 CERROR("No target passed");
983 sptlrpc_rule_set_init(&tmp_rset);
984 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
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 rc = tgt_init_sec_level(req);
1014 /* XXX: better to call this check right after getting new export but
1015 * before last_rcvd slot allocation to avoid server load upon insecure
1016 * connects. This is to be fixed after unifiyng all targets.
1018 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1022 /* To avoid exposing partially initialized connection flags, changes up
1023 * to this point have been staged in reply->ocd_connect_flags. Now that
1024 * connection handling has completed successfully, atomically update
1025 * the connect flags in the shared export data structure. LU-1623 */
1026 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1027 spin_lock(&tsi->tsi_exp->exp_lock);
1028 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1029 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1030 spin_unlock(&tsi->tsi_exp->exp_lock);
1034 obd_disconnect(class_export_get(tsi->tsi_exp));
1037 EXPORT_SYMBOL(tgt_connect);
1039 int tgt_disconnect(struct tgt_session_info *tsi)
1045 rc = target_handle_disconnect(tgt_ses_req(tsi));
1047 RETURN(err_serious(rc));
1051 EXPORT_SYMBOL(tgt_disconnect);
1054 * Unified target OBD handlers
1056 int tgt_obd_ping(struct tgt_session_info *tsi)
1062 rc = target_handle_ping(tgt_ses_req(tsi));
1064 RETURN(err_serious(rc));
1068 EXPORT_SYMBOL(tgt_obd_ping);
1070 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1072 return err_serious(-EOPNOTSUPP);
1075 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1077 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1078 struct ptlrpc_request *req = tgt_ses_req(tsi);
1079 struct obd_export *exp = req->rq_export;
1080 struct ptlrpc_bulk_desc *desc;
1081 struct l_wait_info *lwi = &tti->tti_u.rdbuf.tti_wait_info;
1087 desc = ptlrpc_prep_bulk_exp(req, rdbuf->rb_nbufs, 1,
1088 PTLRPC_BULK_PUT_SOURCE | PTLRPC_BULK_BUF_KVEC,
1089 MDS_BULK_PORTAL, &ptlrpc_bulk_kvec_ops);
1093 for (i = 0; i < rdbuf->rb_nbufs; i++)
1094 desc->bd_frag_ops->add_iov_frag(desc,
1095 rdbuf->rb_bufs[i]->lb_buf,
1096 rdbuf->rb_bufs[i]->lb_len);
1098 rc = target_bulk_io(exp, desc, lwi);
1099 ptlrpc_free_bulk(desc);
1102 EXPORT_SYMBOL(tgt_send_buffer);
1104 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1106 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1107 struct ptlrpc_request *req = tgt_ses_req(tsi);
1108 struct obd_export *exp = req->rq_export;
1109 struct ptlrpc_bulk_desc *desc;
1110 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1118 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1119 PTLRPC_BULK_PUT_SOURCE |
1120 PTLRPC_BULK_BUF_KIOV,
1122 &ptlrpc_bulk_kiov_pin_ops);
1126 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1127 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1128 * which is rdpg->rp_count */
1129 nob = rdpg->rp_count;
1131 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1132 i++, tmpcount -= tmpsize) {
1133 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1134 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1138 LASSERT(desc->bd_nob == nob);
1139 rc = target_bulk_io(exp, desc, lwi);
1140 ptlrpc_free_bulk(desc);
1143 EXPORT_SYMBOL(tgt_sendpage);
1146 * OBD_IDX_READ handler
1148 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1150 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1151 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1152 struct idx_info *req_ii, *rep_ii;
1157 memset(rdpg, 0, sizeof(*rdpg));
1158 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1160 /* extract idx_info buffer from request & reply */
1161 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1162 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1163 RETURN(err_serious(-EPROTO));
1165 rc = req_capsule_server_pack(tsi->tsi_pill);
1167 RETURN(err_serious(rc));
1169 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1171 RETURN(err_serious(-EFAULT));
1172 rep_ii->ii_magic = IDX_INFO_MAGIC;
1174 /* extract hash to start with */
1175 rdpg->rp_hash = req_ii->ii_hash_start;
1177 /* extract requested attributes */
1178 rdpg->rp_attrs = req_ii->ii_attrs;
1180 /* check that fid packed in request is valid and supported */
1181 if (!fid_is_sane(&req_ii->ii_fid))
1183 rep_ii->ii_fid = req_ii->ii_fid;
1186 rep_ii->ii_flags = req_ii->ii_flags;
1188 /* compute number of pages to allocate, ii_count is the number of 4KB
1190 if (req_ii->ii_count <= 0)
1191 GOTO(out, rc = -EFAULT);
1192 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1193 exp_max_brw_size(tsi->tsi_exp));
1194 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1196 /* allocate pages to store the containers */
1197 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1198 if (rdpg->rp_pages == NULL)
1199 GOTO(out, rc = -ENOMEM);
1200 for (i = 0; i < rdpg->rp_npages; i++) {
1201 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1202 if (rdpg->rp_pages[i] == NULL)
1203 GOTO(out, rc = -ENOMEM);
1206 /* populate pages with key/record pairs */
1207 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1211 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1212 "asked %d > %d\n", rc, rdpg->rp_count);
1214 /* send pages to client */
1215 rc = tgt_sendpage(tsi, rdpg, rc);
1220 if (rdpg->rp_pages) {
1221 for (i = 0; i < rdpg->rp_npages; i++)
1222 if (rdpg->rp_pages[i])
1223 __free_page(rdpg->rp_pages[i]);
1224 OBD_FREE(rdpg->rp_pages,
1225 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1230 struct tgt_handler tgt_obd_handlers[] = {
1231 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1232 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1233 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1235 EXPORT_SYMBOL(tgt_obd_handlers);
1237 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1238 struct dt_object *obj, __u64 start, __u64 end)
1244 /* if no objid is specified, it means "sync whole filesystem" */
1246 rc = dt_sync(env, tgt->lut_bottom);
1247 } else if (dt_version_get(env, obj) >
1248 tgt->lut_obd->obd_last_committed) {
1249 rc = dt_object_sync(env, obj, start, end);
1254 EXPORT_SYMBOL(tgt_sync);
1256 * Unified target DLM handlers.
1259 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1261 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1262 void *data, int flag)
1265 struct lu_target *tgt;
1266 struct dt_object *obj;
1272 tgt = class_exp2tgt(lock->l_export);
1274 if (unlikely(tgt == NULL)) {
1275 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1276 class_exp2obd(lock->l_export)->obd_name);
1280 if (flag == LDLM_CB_CANCELING &&
1281 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1282 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1283 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1284 lock->l_flags & LDLM_FL_CBPENDING))) {
1286 __u64 end = OBD_OBJECT_EOF;
1288 rc = lu_env_init(&env, LCT_DT_THREAD);
1289 if (unlikely(rc != 0))
1292 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1293 tgt->lut_lsd.lsd_osd_index);
1294 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1296 GOTO(err_env, rc = PTR_ERR(obj));
1298 if (!dt_object_exists(obj))
1299 GOTO(err_put, rc = -ENOENT);
1301 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1302 start = lock->l_policy_data.l_extent.start;
1303 end = lock->l_policy_data.l_extent.end;
1306 rc = tgt_sync(&env, tgt, obj, start, end);
1308 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1309 "cancel: rc = %d\n",
1310 tgt_name(tgt), PFID(&fid),
1311 lock->l_policy_data.l_extent.start,
1312 lock->l_policy_data.l_extent.end, rc);
1315 lu_object_put(&env, &obj->do_lu);
1320 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1324 static struct ldlm_callback_suite tgt_dlm_cbs = {
1325 .lcs_completion = ldlm_server_completion_ast,
1326 .lcs_blocking = tgt_blocking_ast,
1327 .lcs_glimpse = ldlm_server_glimpse_ast
1330 int tgt_enqueue(struct tgt_session_info *tsi)
1332 struct ptlrpc_request *req = tgt_ses_req(tsi);
1337 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1338 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1340 LASSERT(tsi->tsi_dlm_req != NULL);
1341 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1342 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1344 RETURN(err_serious(rc));
1346 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1348 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1351 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1354 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1357 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1360 RETURN(req->rq_status);
1362 EXPORT_SYMBOL(tgt_enqueue);
1364 int tgt_convert(struct tgt_session_info *tsi)
1366 struct ptlrpc_request *req = tgt_ses_req(tsi);
1370 LASSERT(tsi->tsi_dlm_req);
1371 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1373 RETURN(err_serious(rc));
1375 RETURN(req->rq_status);
1378 int tgt_bl_callback(struct tgt_session_info *tsi)
1380 return err_serious(-EOPNOTSUPP);
1383 int tgt_cp_callback(struct tgt_session_info *tsi)
1385 return err_serious(-EOPNOTSUPP);
1388 /* generic LDLM target handler */
1389 struct tgt_handler tgt_dlm_handlers[] = {
1390 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1391 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1392 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1393 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1395 EXPORT_SYMBOL(tgt_dlm_handlers);
1398 * Unified target LLOG handlers.
1400 int tgt_llog_open(struct tgt_session_info *tsi)
1406 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1410 EXPORT_SYMBOL(tgt_llog_open);
1412 int tgt_llog_close(struct tgt_session_info *tsi)
1418 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1422 EXPORT_SYMBOL(tgt_llog_close);
1425 int tgt_llog_destroy(struct tgt_session_info *tsi)
1431 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1436 int tgt_llog_read_header(struct tgt_session_info *tsi)
1442 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1446 EXPORT_SYMBOL(tgt_llog_read_header);
1448 int tgt_llog_next_block(struct tgt_session_info *tsi)
1454 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1458 EXPORT_SYMBOL(tgt_llog_next_block);
1460 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1466 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1470 EXPORT_SYMBOL(tgt_llog_prev_block);
1472 /* generic llog target handler */
1473 struct tgt_handler tgt_llog_handlers[] = {
1474 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1475 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1476 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1477 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1478 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1479 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1481 EXPORT_SYMBOL(tgt_llog_handlers);
1484 * sec context handlers
1486 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1487 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1492 struct tgt_handler tgt_sec_ctx_handlers[] = {
1493 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1494 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1495 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1497 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1499 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1500 struct dt_device *key,
1501 struct lfsck_request *lr,
1502 struct thandle *th) = NULL;
1504 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1506 struct lfsck_request *,
1509 tgt_lfsck_in_notify = notify;
1511 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1513 static int (*tgt_lfsck_query)(const struct lu_env *env,
1514 struct dt_device *key,
1515 struct lfsck_request *lr) = NULL;
1517 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1519 struct lfsck_request *))
1521 tgt_lfsck_query = query;
1523 EXPORT_SYMBOL(tgt_register_lfsck_query);
1525 /* LFSCK request handlers */
1526 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1528 const struct lu_env *env = tsi->tsi_env;
1529 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1530 struct lfsck_request *lr;
1534 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1538 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1543 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1545 struct lfsck_request *request;
1546 struct lfsck_reply *reply;
1550 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1551 if (request == NULL)
1554 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1558 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1559 reply->lr_status = rc;
1561 RETURN(rc < 0 ? rc : 0);
1564 struct tgt_handler tgt_lfsck_handlers[] = {
1565 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1566 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1568 EXPORT_SYMBOL(tgt_lfsck_handlers);
1571 * initialize per-thread page pool (bug 5137).
1573 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1575 struct tgt_thread_big_cache *tbc;
1579 LASSERT(thread != NULL);
1580 LASSERT(thread->t_data == NULL);
1582 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1585 thread->t_data = tbc;
1588 EXPORT_SYMBOL(tgt_io_thread_init);
1591 * free per-thread pool created by tgt_thread_init().
1593 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1595 struct tgt_thread_big_cache *tbc;
1599 LASSERT(thread != NULL);
1602 * be prepared to handle partially-initialized pools (because this is
1603 * called from ost_io_thread_init() for cleanup.
1605 tbc = thread->t_data;
1607 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1608 thread->t_data = NULL;
1612 EXPORT_SYMBOL(tgt_io_thread_done);
1614 * Helper function for getting server side [start, start+count] DLM lock
1615 * if asked by client.
1617 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1618 __u64 start, __u64 end, struct lustre_handle *lh,
1619 int mode, __u64 *flags)
1621 union ldlm_policy_data policy;
1626 LASSERT(lh != NULL);
1627 LASSERT(ns != NULL);
1628 LASSERT(!lustre_handle_is_used(lh));
1630 policy.l_extent.gid = 0;
1631 policy.l_extent.start = start & PAGE_MASK;
1634 * If ->o_blocks is EOF it means "lock till the end of the file".
1635 * Otherwise, it's size of an extent or hole being punched (in bytes).
1637 if (end == OBD_OBJECT_EOF || end < start)
1638 policy.l_extent.end = OBD_OBJECT_EOF;
1640 policy.l_extent.end = end | ~PAGE_MASK;
1642 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1643 flags, ldlm_blocking_ast,
1644 ldlm_completion_ast, ldlm_glimpse_ast,
1645 NULL, 0, LVB_T_NONE, NULL, lh);
1646 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1648 EXPORT_SYMBOL(tgt_extent_lock);
1650 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1652 LASSERT(lustre_handle_is_used(lh));
1653 ldlm_lock_decref(lh, mode);
1655 EXPORT_SYMBOL(tgt_extent_unlock);
1657 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1658 struct obd_ioobj *obj, struct niobuf_remote *nb,
1659 struct lustre_handle *lh, enum ldlm_mode mode)
1662 int nrbufs = obj->ioo_bufcnt;
1667 LASSERT(mode == LCK_PR || mode == LCK_PW);
1668 LASSERT(!lustre_handle_is_used(lh));
1670 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1673 for (i = 1; i < nrbufs; i++)
1674 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1677 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1678 nb[nrbufs - 1].rnb_offset +
1679 nb[nrbufs - 1].rnb_len - 1,
1683 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1684 struct lustre_handle *lh, enum ldlm_mode mode)
1688 LASSERT(mode == LCK_PR || mode == LCK_PW);
1689 LASSERT((obj->ioo_bufcnt > 0 &&
1690 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1691 lustre_handle_is_used(lh));
1693 if (lustre_handle_is_used(lh))
1694 tgt_extent_unlock(lh, mode);
1698 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1699 struct ptlrpc_bulk_desc *desc, int opc,
1700 cksum_type_t cksum_type)
1702 struct cfs_crypto_hash_desc *hdesc;
1703 unsigned int bufsize;
1705 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1708 LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1710 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1711 if (IS_ERR(hdesc)) {
1712 CERROR("%s: unable to initialize checksum hash %s\n",
1713 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1714 return PTR_ERR(hdesc);
1717 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1718 for (i = 0; i < desc->bd_iov_count; i++) {
1719 /* corrupt the data before we compute the checksum, to
1720 * simulate a client->OST data error */
1721 if (i == 0 && opc == OST_WRITE &&
1722 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1723 int off = BD_GET_KIOV(desc, i).kiov_offset &
1725 int len = BD_GET_KIOV(desc, i).kiov_len;
1726 struct page *np = tgt_page_to_corrupt;
1727 char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1730 char *ptr2 = kmap(np) + off;
1732 memcpy(ptr2, ptr, len);
1733 memcpy(ptr2, "bad3", min(4, len));
1735 BD_GET_KIOV(desc, i).kiov_page = np;
1737 CERROR("%s: can't alloc page for corruption\n",
1741 cfs_crypto_hash_update_page(hdesc,
1742 BD_GET_KIOV(desc, i).kiov_page,
1743 BD_GET_KIOV(desc, i).kiov_offset &
1745 BD_GET_KIOV(desc, i).kiov_len);
1747 /* corrupt the data after we compute the checksum, to
1748 * simulate an OST->client data error */
1749 if (i == 0 && opc == OST_READ &&
1750 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1751 int off = BD_GET_KIOV(desc, i).kiov_offset
1753 int len = BD_GET_KIOV(desc, i).kiov_len;
1754 struct page *np = tgt_page_to_corrupt;
1756 kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1759 char *ptr2 = kmap(np) + off;
1761 memcpy(ptr2, ptr, len);
1762 memcpy(ptr2, "bad4", min(4, len));
1764 BD_GET_KIOV(desc, i).kiov_page = np;
1766 CERROR("%s: can't alloc page for corruption\n",
1772 bufsize = sizeof(cksum);
1773 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1778 int tgt_brw_read(struct tgt_session_info *tsi)
1780 struct ptlrpc_request *req = tgt_ses_req(tsi);
1781 struct ptlrpc_bulk_desc *desc = NULL;
1782 struct obd_export *exp = tsi->tsi_exp;
1783 struct niobuf_remote *remote_nb;
1784 struct niobuf_local *local_nb;
1785 struct obd_ioobj *ioo;
1786 struct ost_body *body, *repbody;
1787 struct l_wait_info lwi;
1788 struct lustre_handle lockh = { 0 };
1789 int npages, nob = 0, rc, i, no_reply = 0;
1790 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1794 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1795 CERROR("%s: deny read request from %s to portal %u\n",
1796 tgt_name(tsi->tsi_tgt),
1797 obd_export_nid2str(req->rq_export),
1798 ptlrpc_req2svc(req)->srv_req_portal);
1802 req->rq_bulk_read = 1;
1804 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1807 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1808 cfs_fail_val : (obd_timeout + 1) / 4);
1810 /* Check if there is eviction in progress, and if so, wait for it to
1812 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1813 /* We do not care how long it takes */
1814 lwi = LWI_INTR(NULL, NULL);
1815 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1816 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1820 /* There must be big cache in current thread to process this request
1821 * if it is NULL then something went wrong and it wasn't allocated,
1822 * report -ENOMEM in that case */
1826 body = tsi->tsi_ost_body;
1827 LASSERT(body != NULL);
1829 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1830 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1832 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1833 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1835 local_nb = tbc->local;
1837 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1838 remote_nb, &lockh, LCK_PR);
1843 * If getting the lock took more time than
1844 * client was willing to wait, drop it. b=11330
1846 if (cfs_time_current_sec() > req->rq_deadline ||
1847 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1849 CERROR("Dropping timed-out read from %s because locking"
1850 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1851 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1852 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1853 req->rq_deadline - req->rq_arrival_time.tv_sec);
1854 GOTO(out_lock, rc = -ETIMEDOUT);
1857 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1858 repbody->oa = body->oa;
1860 npages = PTLRPC_MAX_BRW_PAGES;
1861 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1862 ioo, remote_nb, &npages, local_nb);
1866 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1867 PTLRPC_BULK_PUT_SOURCE |
1868 PTLRPC_BULK_BUF_KIOV,
1870 &ptlrpc_bulk_kiov_nopin_ops);
1872 GOTO(out_commitrw, rc = -ENOMEM);
1875 for (i = 0; i < npages; i++) {
1876 int page_rc = local_nb[i].lnb_rc;
1884 if (page_rc != 0) { /* some data! */
1885 LASSERT(local_nb[i].lnb_page != NULL);
1886 desc->bd_frag_ops->add_kiov_frag
1887 (desc, local_nb[i].lnb_page,
1888 local_nb[i].lnb_page_offset,
1892 if (page_rc != local_nb[i].lnb_len) { /* short read */
1893 /* All subsequent pages should be 0 */
1894 while (++i < npages)
1895 LASSERT(local_nb[i].lnb_rc == 0);
1900 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1901 cksum_type_t cksum_type =
1902 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1903 body->oa.o_flags : 0);
1904 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1905 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1906 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1907 OST_READ, cksum_type);
1908 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1909 repbody->oa.o_cksum);
1911 repbody->oa.o_valid = 0;
1913 /* We're finishing using body->oa as an input variable */
1915 /* Check if client was evicted while we were doing i/o before touching
1917 if (likely(rc == 0 &&
1918 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1919 rc = target_bulk_io(exp, desc, &lwi);
1924 /* Must commit after prep above in all cases */
1925 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1926 remote_nb, npages, local_nb, rc);
1928 tgt_drop_id(exp, &repbody->oa);
1930 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1932 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1933 ptlrpc_free_bulk(desc);
1938 ptlrpc_lprocfs_brw(req, nob);
1939 } else if (no_reply) {
1940 req->rq_no_reply = 1;
1941 /* reply out callback would free */
1942 ptlrpc_req_drop_rs(req);
1943 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1944 "client will retry: rc %d\n",
1945 exp->exp_obd->obd_name,
1946 obd_uuid2str(&exp->exp_client_uuid),
1947 obd_export_nid2str(exp), rc);
1949 /* send a bulk after reply to simulate a network delay or reordering
1951 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1952 wait_queue_head_t waitq;
1953 struct l_wait_info lwi1;
1955 CDEBUG(D_INFO, "reorder BULK\n");
1956 init_waitqueue_head(&waitq);
1958 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1959 l_wait_event(waitq, 0, &lwi1);
1960 target_bulk_io(exp, desc, &lwi);
1961 ptlrpc_free_bulk(desc);
1966 EXPORT_SYMBOL(tgt_brw_read);
1968 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1969 struct ptlrpc_bulk_desc *desc,
1970 struct niobuf_local *local_nb, int npages,
1971 u32 client_cksum, u32 server_cksum,
1974 struct obd_export *exp = req->rq_export;
1975 struct ost_body *body;
1979 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1980 LASSERT(body != NULL);
1982 if (req->rq_peer.nid != desc->bd_sender) {
1984 router = libcfs_nid2str(desc->bd_sender);
1988 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1989 client_cksum, server_cksum);
1993 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1994 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1995 "]: client csum %x, server csum %x\n",
1996 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1998 body->oa.o_valid & OBD_MD_FLFID ?
1999 body->oa.o_parent_seq : (__u64)0,
2000 body->oa.o_valid & OBD_MD_FLFID ?
2001 body->oa.o_parent_oid : 0,
2002 body->oa.o_valid & OBD_MD_FLFID ?
2003 body->oa.o_parent_ver : 0,
2004 POSTID(&body->oa.o_oi),
2005 local_nb[0].lnb_file_offset,
2006 local_nb[npages-1].lnb_file_offset +
2007 local_nb[npages - 1].lnb_len - 1,
2008 client_cksum, server_cksum);
2011 int tgt_brw_write(struct tgt_session_info *tsi)
2013 struct ptlrpc_request *req = tgt_ses_req(tsi);
2014 struct ptlrpc_bulk_desc *desc = NULL;
2015 struct obd_export *exp = req->rq_export;
2016 struct niobuf_remote *remote_nb;
2017 struct niobuf_local *local_nb;
2018 struct obd_ioobj *ioo;
2019 struct ost_body *body, *repbody;
2020 struct l_wait_info lwi;
2021 struct lustre_handle lockh = {0};
2023 int objcount, niocount, npages;
2025 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
2026 bool no_reply = false, mmap;
2027 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2031 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
2032 CERROR("%s: deny write request from %s to portal %u\n",
2033 tgt_name(tsi->tsi_tgt),
2034 obd_export_nid2str(req->rq_export),
2035 ptlrpc_req2svc(req)->srv_req_portal);
2036 RETURN(err_serious(-EPROTO));
2039 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2040 RETURN(err_serious(-ENOSPC));
2041 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2042 RETURN(err_serious(-EROFS));
2044 req->rq_bulk_write = 1;
2046 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2047 RETURN(err_serious(-EIO));
2048 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2049 RETURN(err_serious(-EFAULT));
2051 /* pause before transaction has been started */
2052 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2053 cfs_fail_val : (obd_timeout + 1) / 4);
2055 /* There must be big cache in current thread to process this request
2056 * if it is NULL then something went wrong and it wasn't allocated,
2057 * report -ENOMEM in that case */
2061 body = tsi->tsi_ost_body;
2062 LASSERT(body != NULL);
2064 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2065 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2067 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2068 RCL_CLIENT) / sizeof(*ioo);
2070 for (niocount = i = 0; i < objcount; i++)
2071 niocount += ioo[i].ioo_bufcnt;
2073 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2074 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2075 if (niocount != req_capsule_get_size(&req->rq_pill,
2076 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2078 RETURN(err_serious(-EPROTO));
2080 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2081 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2082 memory_pressure_set();
2084 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2085 niocount * sizeof(*rcs));
2086 rc = req_capsule_server_pack(&req->rq_pill);
2088 GOTO(out, rc = err_serious(rc));
2090 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2091 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2093 local_nb = tbc->local;
2095 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2096 remote_nb, &lockh, LCK_PW);
2101 * If getting the lock took more time than
2102 * client was willing to wait, drop it. b=11330
2104 if (cfs_time_current_sec() > req->rq_deadline ||
2105 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2107 CERROR("%s: Dropping timed-out write from %s because locking "
2108 "object "DOSTID" took %ld seconds (limit was %ld).\n",
2109 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2110 POSTID(&ioo->ioo_oid),
2111 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2112 req->rq_deadline - req->rq_arrival_time.tv_sec);
2113 GOTO(out_lock, rc = -ETIMEDOUT);
2116 /* Because we already sync grant info with client when reconnect,
2117 * grant info will be cleared for resent req, then fed_grant and
2118 * total_grant will not be modified in following preprw_write */
2119 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2120 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2121 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2124 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2125 if (repbody == NULL)
2126 GOTO(out_lock, rc = -ENOMEM);
2127 repbody->oa = body->oa;
2129 npages = PTLRPC_MAX_BRW_PAGES;
2130 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2131 objcount, ioo, remote_nb, &npages, local_nb);
2135 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2136 PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2138 &ptlrpc_bulk_kiov_nopin_ops);
2140 GOTO(skip_transfer, rc = -ENOMEM);
2142 /* NB Having prepped, we must commit... */
2143 for (i = 0; i < npages; i++)
2144 desc->bd_frag_ops->add_kiov_frag(desc,
2145 local_nb[i].lnb_page,
2146 local_nb[i].lnb_page_offset,
2147 local_nb[i].lnb_len);
2149 rc = sptlrpc_svc_prep_bulk(req, desc);
2151 GOTO(skip_transfer, rc);
2153 rc = target_bulk_io(exp, desc, &lwi);
2157 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2158 static int cksum_counter;
2160 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2161 cksum_type = cksum_type_unpack(body->oa.o_flags);
2163 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2164 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2165 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2166 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2167 OST_WRITE, cksum_type);
2170 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2171 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2172 body->oa.o_flags & OBD_FL_MMAP);
2174 tgt_warn_on_cksum(req, desc, local_nb, npages,
2176 repbody->oa.o_cksum, mmap);
2178 } else if ((cksum_counter & (-cksum_counter)) ==
2180 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2181 cksum_counter, libcfs_id2str(req->rq_peer),
2182 repbody->oa.o_cksum);
2186 /* Must commit after prep above in all cases */
2187 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2188 objcount, ioo, remote_nb, npages, local_nb, rc);
2189 if (rc == -ENOTCONN)
2190 /* quota acquire process has been given up because
2191 * either the client has been evicted or the client
2192 * has timed out the request already */
2196 * Disable sending mtime back to the client. If the client locked the
2197 * whole object, then it has already updated the mtime on its side,
2198 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2200 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2205 /* set per-requested niobuf return codes */
2206 for (i = j = 0; i < niocount; i++) {
2207 int len = remote_nb[i].rnb_len;
2212 LASSERT(j < npages);
2213 if (local_nb[j].lnb_rc < 0)
2214 rcs[i] = local_nb[j].lnb_rc;
2215 len -= local_nb[j].lnb_len;
2220 LASSERT(j == npages);
2221 ptlrpc_lprocfs_brw(req, nob);
2223 tgt_drop_id(exp, &repbody->oa);
2226 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2228 ptlrpc_free_bulk(desc);
2231 req->rq_no_reply = 1;
2232 /* reply out callback would free */
2233 ptlrpc_req_drop_rs(req);
2234 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2235 "client will retry: rc %d\n",
2236 exp->exp_obd->obd_name,
2237 obd_uuid2str(&exp->exp_client_uuid),
2238 obd_export_nid2str(exp), rc);
2240 memory_pressure_clr();
2243 EXPORT_SYMBOL(tgt_brw_write);
2245 /* Check if request can be reconstructed from saved reply data
2246 * A copy of the reply data is returned in @trd if the pointer is not NULL
2248 bool req_can_reconstruct(struct ptlrpc_request *req,
2249 struct tg_reply_data *trd)
2251 struct tg_export_data *ted = &req->rq_export->exp_target_data;
2252 struct lsd_client_data *lcd = ted->ted_lcd;
2255 if (tgt_is_multimodrpcs_client(req->rq_export))
2256 return tgt_lookup_reply(req, trd);
2258 mutex_lock(&ted->ted_lcd_lock);
2259 found = req->rq_xid == lcd->lcd_last_xid ||
2260 req->rq_xid == lcd->lcd_last_close_xid;
2262 if (found && trd != NULL) {
2263 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
2264 trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
2265 trd->trd_reply.lrd_transno =
2266 lcd->lcd_last_close_transno;
2267 trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
2269 trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
2270 trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
2271 trd->trd_reply.lrd_result = lcd->lcd_last_result;
2272 trd->trd_reply.lrd_data = lcd->lcd_last_data;
2273 trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
2274 trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
2275 trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
2276 trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
2279 mutex_unlock(&ted->ted_lcd_lock);
2283 EXPORT_SYMBOL(req_can_reconstruct);