4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2013, 2017, Intel Corporation.
27 * lustre/target/tgt_handler.c
29 * Lustre Unified Target request handler code
31 * Author: Brian Behlendorf <behlendorf1@llnl.gov>
32 * Author: Mikhail Pershin <mike.pershin@intel.com>
35 #define DEBUG_SUBSYSTEM S_CLASS
37 #include <linux/user_namespace.h>
38 #include <linux/delay.h>
39 #include <linux/uidgid.h>
41 #include <libcfs/linux/linux-mem.h>
43 #include <obd_class.h>
44 #include <obd_cksum.h>
45 #include <lustre_lfsck.h>
46 #include <lustre_nodemap.h>
47 #include <lustre_acl.h>
49 #include "tgt_internal.h"
51 char *tgt_name(struct lu_target *tgt)
53 LASSERT(tgt->lut_obd != NULL);
54 return tgt->lut_obd->obd_name;
56 EXPORT_SYMBOL(tgt_name);
59 * Generic code handling requests that have struct mdt_body passed in:
61 * - extract mdt_body from request and save it in @tsi, if present;
63 * - create lu_object, corresponding to the fid in mdt_body, and save it in
66 * - if HAS_BODY flag is set for this request type check whether object
67 * actually exists on storage (lu_object_exists()).
70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
72 const struct mdt_body *body;
73 struct lu_object *obj;
74 struct req_capsule *pill = tsi->tsi_pill;
79 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
83 tsi->tsi_mdt_body = body;
85 if (!(body->mbo_valid & OBD_MD_FLID))
88 /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
89 * in any case in pre-2.5 clients. Fix that here if needed */
90 if (unlikely(fid_is_zero(&body->mbo_fid1)))
93 if (!fid_is_sane(&body->mbo_fid1)) {
94 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
95 PFID(&body->mbo_fid1));
99 obj = lu_object_find(tsi->tsi_env,
100 &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
101 &body->mbo_fid1, NULL);
103 if ((flags & HAS_BODY) && !lu_object_exists(obj)) {
104 lu_object_put(tsi->tsi_env, obj);
107 tsi->tsi_corpus = obj;
114 tsi->tsi_fid = body->mbo_fid1;
120 * Validate oa from client.
121 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
123 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
124 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
125 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
126 * whether this is oi_fid or real ostid. So it will check
127 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
128 * c. Old FID-disable osc will send IDIF.
129 * d. new FID-enable osc/osp will send normal FID.
131 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
132 * be used for LAST_ID file, and only being accessed inside OST now.
134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
136 struct ost_id *oi = &oa->o_oi;
137 u64 seq = ostid_seq(oi);
138 u64 id = ostid_id(oi);
142 if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
143 fid_seq_is_echo(seq))) {
144 /* Sigh 2.[123] client still sends echo req with oi_id = 0
145 * during create, and we will reset this to 1, since this
146 * oi_id is basically useless in the following create process,
147 * but oi_id == 0 will make it difficult to tell whether it is
148 * real FID or ost_id. */
149 oi->oi_fid.f_seq = FID_SEQ_ECHO;
150 oi->oi_fid.f_oid = id ?: 1;
151 oi->oi_fid.f_ver = 0;
153 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
155 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
156 GOTO(out, rc = -EPROTO);
158 /* Note: this check might be forced in 2.5 or 2.6, i.e.
159 * all of the requests are required to setup FLGROUP */
160 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
161 ostid_set_seq_mdt0(oi);
162 oa->o_valid |= OBD_MD_FLGROUP;
166 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
167 fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
168 GOTO(out, rc = -EPROTO);
170 rc = ostid_to_fid(&tti->tti_fid1, oi,
171 tsi->tsi_tgt->lut_lsd.lsd_osd_index);
172 if (unlikely(rc != 0))
175 oi->oi_fid = tti->tti_fid1;
181 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
182 tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
186 EXPORT_SYMBOL(tgt_validate_obdo);
188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
191 struct niobuf_remote *rnb;
192 struct obd_ioobj *ioo;
197 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
201 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
205 max_brw = ioobj_max_brw_get(ioo);
206 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
207 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
208 ": rc = %d\n", tgt_name(tsi->tsi_tgt),
209 obd_export_nid2str(tsi->tsi_exp), max_brw,
210 POSTID(oi), -EPROTO);
215 obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
216 RCL_CLIENT) / sizeof(*ioo);
217 if (obj_count == 0) {
218 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
220 } else if (obj_count > 1) {
221 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
226 if (ioo->ioo_bufcnt == 0) {
227 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
231 if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
232 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
233 "bulk has too many pages (%d)",
241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
243 struct ost_body *body;
244 struct req_capsule *pill = tsi->tsi_pill;
245 struct lu_nodemap *nodemap;
250 body = req_capsule_client_get(pill, &RMF_OST_BODY);
254 rc = tgt_validate_obdo(tsi, &body->oa);
258 nodemap = nodemap_get_from_exp(tsi->tsi_exp);
260 RETURN(PTR_ERR(nodemap));
262 body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
263 NODEMAP_CLIENT_TO_FS,
265 body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
266 NODEMAP_CLIENT_TO_FS,
268 body->oa.o_projid = nodemap_map_id(nodemap, NODEMAP_PROJID,
269 NODEMAP_CLIENT_TO_FS,
271 nodemap_putref(nodemap);
273 tsi->tsi_ost_body = body;
274 tsi->tsi_fid = body->oa.o_oi.oi_fid;
276 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
277 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
282 if (!(body->oa.o_valid & OBD_MD_FLID)) {
283 if (flags & HAS_BODY) {
284 CERROR("%s: OBD_MD_FLID flag is not set in ost_body but OID/FID is mandatory with HAS_BODY\n",
285 tgt_name(tsi->tsi_tgt));
292 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
295 * OST doesn't get object in advance for further use to prevent
296 * situations with nested object_find which is potential deadlock.
298 tsi->tsi_corpus = NULL;
303 * Do necessary preprocessing according to handler ->th_flags.
305 static int tgt_request_preprocess(struct tgt_session_info *tsi,
306 struct tgt_handler *h,
307 struct ptlrpc_request *req)
309 struct req_capsule *pill = tsi->tsi_pill;
310 __u32 flags = h->th_flags;
315 if (tsi->tsi_preprocessed)
318 LASSERT(h->th_act != NULL);
319 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
320 LASSERT(current->journal_info == NULL);
322 LASSERT(ergo(flags & (HAS_BODY | HAS_REPLY),
324 if (h->th_fmt != NULL) {
325 req_capsule_set(pill, h->th_fmt);
326 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT) &&
327 req_capsule_field_present(pill, &RMF_MDT_BODY,
329 rc = tgt_mdt_body_unpack(tsi, flags);
332 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
334 req_capsule_field_present(pill, &RMF_OST_BODY,
336 rc = tgt_ost_body_unpack(tsi, flags);
342 if (flags & IS_MUTABLE && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
345 if (flags & HAS_KEY) {
346 struct ldlm_request *dlm_req;
348 LASSERT(h->th_fmt != NULL);
350 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
351 if (dlm_req != NULL) {
352 union ldlm_wire_policy_data *policy =
353 &dlm_req->lock_desc.l_policy_data;
355 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
357 (policy->l_inodebits.bits |
358 policy->l_inodebits.try_bits) == 0)) {
360 * Lock without inodebits makes no sense and
361 * will oops later in ldlm. If client miss to
362 * set such bits, do not trigger ASSERTION.
364 * For liblustre flock case, it maybe zero.
368 tsi->tsi_dlm_req = dlm_req;
374 tsi->tsi_preprocessed = 1;
379 * Invoke handler for this request opc. Also do necessary preprocessing
380 * (according to handler ->th_flags), and post-processing (setting of
381 * ->last_{xid,committed}).
383 static int tgt_handle_request0(struct tgt_session_info *tsi,
384 struct tgt_handler *h,
385 struct ptlrpc_request *req)
389 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
394 /* When dealing with sec context requests, no export is associated yet,
395 * because these requests are sent before *_CONNECT requests.
396 * A NULL req->rq_export means the normal *_common_slice handlers will
397 * not be called, because there is no reference to the target.
398 * So deal with them by hand and jump directly to target_send_reply().
402 case SEC_CTX_INIT_CONT:
404 CFS_FAIL_TIMEOUT(OBD_FAIL_SEC_CTX_HDL_PAUSE, cfs_fail_val);
409 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
410 * to put same checks into handlers like mdt_close(), mdt_reint(),
411 * etc., without talking to mdt authors first. Checking same thing
412 * there again is useless and returning 0 error without packing reply
413 * is buggy! Handlers either pack reply or return error.
415 * We return 0 here and do not send any reply in order to emulate
416 * network failure. Do not send any reply in case any of NET related
417 * fail_id has occured.
419 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
421 if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
422 OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
425 /* drop OUT_UPDATE rpc */
426 if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == OUT_UPDATE &&
427 OBD_FAIL_CHECK(OBD_FAIL_OUT_UPDATE_DROP)))
430 rc = tgt_request_preprocess(tsi, h, req);
431 /* pack reply if reply format is fixed */
432 if (rc == 0 && h->th_flags & HAS_REPLY) {
434 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
436 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
438 tsi->tsi_mdt_body->mbo_eadatasize);
439 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
441 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
443 if (req_capsule_has_field(tsi->tsi_pill, &RMF_ACL, RCL_SERVER))
444 req_capsule_set_size(tsi->tsi_pill,
445 &RMF_ACL, RCL_SERVER,
446 LUSTRE_POSIX_ACL_MAX_SIZE_OLD);
448 if (req_capsule_has_field(tsi->tsi_pill, &RMF_SHORT_IO,
450 struct niobuf_remote *remote_nb =
451 req_capsule_client_get(tsi->tsi_pill,
453 struct ost_body *body = tsi->tsi_ost_body;
455 req_capsule_set_size(tsi->tsi_pill, &RMF_SHORT_IO,
457 (body->oa.o_valid & OBD_MD_FLFLAGS &&
458 body->oa.o_flags & OBD_FL_SHORT_IO) ?
459 remote_nb[0].rnb_len : 0);
461 if (req_capsule_has_field(tsi->tsi_pill, &RMF_FILE_ENCCTX,
463 req_capsule_set_size(tsi->tsi_pill, &RMF_FILE_ENCCTX,
466 rc = req_capsule_server_pack(tsi->tsi_pill);
469 if (likely(rc == 0)) {
471 * Process request, there can be two types of rc:
472 * 1) errors with msg unpack/pack, other failures outside the
473 * operation itself. This is counted as serious errors;
474 * 2) errors during fs operation, should be placed in rq_status
478 if (!is_serious(rc) &&
479 !req->rq_no_reply && req->rq_reply_state == NULL) {
480 DEBUG_REQ(D_ERROR, req,
481 "%s: %s handler did not pack reply but returned no error",
482 tgt_name(tsi->tsi_tgt), h->th_name);
485 serious = is_serious(rc);
486 rc = clear_serious(rc);
494 * ELDLM_* codes which > 0 should be in rq_status only as well as
495 * all non-serious errors.
497 if (rc > 0 || !serious)
500 LASSERT(current->journal_info == NULL);
502 if (likely(rc == 0 && req->rq_export))
503 target_committed_to_req(req);
506 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
510 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
511 struct obd_device *obd, int *process)
513 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
520 case MDS_SYNC: /* used in unmounting */
534 case MDS_HSM_PROGRESS:
535 case MDS_HSM_STATE_SET:
536 case MDS_HSM_REQUEST:
538 *process = target_queue_recovery_request(req, obd);
542 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
549 * Handle recovery. Return:
550 * +1: continue request processing;
551 * -ve: abort immediately with the given error code;
552 * 0: send reply with error code in req->rq_status;
554 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
558 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
563 case SEC_CTX_INIT_CONT:
568 if (!req->rq_export->exp_obd->obd_replayable)
571 /* sanity check: if the xid matches, the request must be marked as a
572 * resent or replayed */
573 if (req_can_reconstruct(req, NULL) == 1) {
574 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
575 (MSG_RESENT | MSG_REPLAY))) {
576 DEBUG_REQ(D_WARNING, req,
577 "rq_xid=%llu matches saved XID, expected REPLAY or RESENT flag (%x)",
579 lustre_msg_get_flags(req->rq_reqmsg));
580 req->rq_status = -ENOTCONN;
584 /* else: note the opposite is not always true; a RESENT req after a
585 * failover will usually not match the last_xid, since it was likely
586 * never committed. A REPLAYed request will almost never match the
587 * last xid, however it could for a committed, but still retained,
590 /* Check for aborted recovery... */
591 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
595 DEBUG_REQ(D_INFO, req, "Got new replay");
596 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
598 if (rc != 0 || !should_process)
600 else if (should_process < 0) {
601 req->rq_status = should_process;
602 rc = ptlrpc_error(req);
609 /* Initial check for request, it is validation mostly */
610 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
612 struct tgt_handler *h;
613 struct tgt_opc_slice *s;
614 struct lu_target *tgt;
615 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
616 /* don't spew error messages for unhandled RPCs */
621 tgt = class_exp2tgt(req->rq_export);
622 if (unlikely(tgt == NULL)) {
623 DEBUG_REQ(D_ERROR, req, "%s: no target for connected export",
624 class_exp2obd(req->rq_export)->obd_name);
625 RETURN(ERR_PTR(-EINVAL));
628 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
629 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
632 /* opcode was not found in slice */
633 if (unlikely(s->tos_hs == NULL)) {
635 CERROR("%s: no handler for opcode 0x%x from %s\n",
636 tgt_name(tgt), opc, libcfs_id2str(req->rq_peer));
639 goto err_unsupported;
642 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
643 h = s->tos_hs + (opc - s->tos_opc_start);
644 if (unlikely(h->th_opc == 0)) {
646 CERROR("%s: unsupported opcode 0x%x\n",
650 goto err_unsupported;
653 if (OBD_FAIL_CHECK(OBD_FAIL_OST_OPCODE) && opc == cfs_fail_val)
654 goto err_unsupported;
659 * Unknown opcode does not necessarily means insane client. A new
660 * client might send RPCs with new opcodes to an old server. The
661 * client might desperately stuck there waiting for a reply. So,
662 * send an error back here.
664 * An old client might also send RPCs with deprecated opcodes (e.g.
667 * Error in ptlrpc_send_error() is ignored.
669 req->rq_status = -EOPNOTSUPP;
670 ptlrpc_send_error(req, PTLRPC_REPLY_MAYBE_DIFFICULT);
671 RETURN(ERR_PTR(-EOPNOTSUPP));
674 static int process_req_last_xid(struct ptlrpc_request *req)
678 struct obd_export *exp = req->rq_export;
679 struct tg_export_data *ted = &exp->exp_target_data;
680 bool need_lock = tgt_is_multimodrpcs_client(exp);
684 mutex_lock(&ted->ted_lcd_lock);
685 /* check request's xid is consistent with export's last_xid */
686 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
687 if (last_xid > exp->exp_last_xid)
688 exp->exp_last_xid = last_xid;
690 if (req->rq_xid == 0 || req->rq_xid <= exp->exp_last_xid) {
691 /* Some request is allowed to be sent during replay,
692 * such as OUT update requests, FLD requests, so it
693 * is possible that replay requests has smaller XID
694 * than the exp_last_xid.
696 * Some non-replay requests may have smaller XID as
699 * - Client send a no_resend RPC, like statfs;
700 * - The RPC timedout (or some other error) on client,
701 * then it's removed from the unreplied list;
702 * - Client send some other request to bump the
703 * exp_last_xid on server;
704 * - The former RPC got chance to be processed;
706 if (!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY))
709 DEBUG_REQ(D_WARNING, req,
710 "unexpected xid=%llx != exp_last_xid=%llx, rc = %d",
711 req->rq_xid, exp->exp_last_xid, rc);
716 /* The "last_xid" is the minimum xid among unreplied requests,
717 * if the request is from the previous connection, its xid can
718 * still be larger than "exp_last_xid", then the above check of
719 * xid is not enough to determine whether the request is delayed.
721 * For example, if some replay request was delayed and caused
722 * timeout at client and the replay is restarted, the delayed
723 * replay request will have the larger xid than "exp_last_xid"
725 if (req->rq_export->exp_conn_cnt >
726 lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
728 "Dropping request %llu from an old epoch %u/%u\n",
730 lustre_msg_get_conn_cnt(req->rq_reqmsg),
731 req->rq_export->exp_conn_cnt);
732 req->rq_no_reply = 1;
733 GOTO(out, rc = -ESTALE);
736 /* try to release in-memory reply data */
737 if (tgt_is_multimodrpcs_client(exp)) {
738 tgt_handle_received_xid(exp, last_xid);
739 rc = tgt_handle_tag(req);
744 mutex_unlock(&ted->ted_lcd_lock);
749 int tgt_request_handle(struct ptlrpc_request *req)
751 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
753 struct lustre_msg *msg = req->rq_reqmsg;
754 struct tgt_handler *h;
755 struct lu_target *tgt;
756 int request_fail_id = 0;
757 __u32 opc = lustre_msg_get_opc(msg);
758 struct obd_device *obd;
760 bool is_connect = false;
763 if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_TGT_RECOVERY_REQ_RACE))) {
764 if (cfs_fail_val == 0 &&
765 lustre_msg_get_opc(msg) != OBD_PING &&
766 lustre_msg_get_flags(msg) & MSG_REQ_REPLAY_DONE) {
769 wait_event_idle(cfs_race_waitq, (cfs_race_state == 1));
773 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
774 tsi->tsi_pill = &req->rq_pill;
775 tsi->tsi_env = req->rq_svc_thread->t_env;
777 /* if request has export then get handlers slice from corresponding
778 * target, otherwise that should be connect operation */
779 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
780 opc == MGS_CONNECT) {
782 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
783 rc = target_handle_connect(req);
785 rc = ptlrpc_error(req);
788 /* recovery-small test 18c asks to drop connect reply */
789 if (unlikely(opc == OST_CONNECT &&
790 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
794 if (unlikely(!class_connected_export(req->rq_export))) {
795 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
796 opc == SEC_CTX_FINI) {
797 /* sec context initialization has to be handled
798 * by hand in tgt_handle_request0() */
799 tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
801 GOTO(handle_recov, rc = 0);
803 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
804 opc, libcfs_id2str(req->rq_peer));
805 req->rq_status = -ENOTCONN;
806 rc = ptlrpc_error(req);
810 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
811 tsi->tsi_exp = req->rq_export;
812 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
813 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
815 tsi->tsi_jobid = NULL;
818 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export",
819 class_exp2obd(req->rq_export)->obd_name);
820 req->rq_status = -EINVAL;
821 rc = ptlrpc_error(req);
825 /* Skip last_xid processing for the recovery thread, otherwise, the
826 * last_xid on same request could be processed twice: first time when
827 * processing the incoming request, second time when the request is
828 * being processed by recovery thread. */
829 obd = class_exp2obd(req->rq_export);
831 /* reset the exp_last_xid on each connection. */
832 req->rq_export->exp_last_xid = 0;
833 } else if (obd->obd_recovery_data.trd_processing_task !=
835 rc = process_req_last_xid(req);
838 rc = ptlrpc_error(req);
843 request_fail_id = tgt->lut_request_fail_id;
844 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
846 h = tgt_handler_find_check(req);
848 req->rq_status = PTR_ERR(h);
849 rc = ptlrpc_error(req);
853 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
856 if ((cfs_fail_val == 0 || cfs_fail_val == opc) &&
857 CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
860 rc = lustre_msg_check_version(msg, h->th_version);
862 DEBUG_REQ(D_ERROR, req,
863 "%s: drop malformed request version=%08x expect=%08x",
864 tgt_name(tgt), lustre_msg_get_version(msg),
866 req->rq_status = -EINVAL;
867 rc = ptlrpc_error(req);
872 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
873 if (likely(rc == 1)) {
874 rc = tgt_handle_request0(tsi, h, req);
880 req_capsule_fini(tsi->tsi_pill);
881 if (tsi->tsi_corpus != NULL) {
882 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
883 tsi->tsi_corpus = NULL;
887 EXPORT_SYMBOL(tgt_request_handle);
889 /** Assign high priority operations to the request if needed. */
890 int tgt_hpreq_handler(struct ptlrpc_request *req)
892 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
893 struct tgt_handler *h;
898 if (req->rq_export == NULL)
901 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
902 tsi->tsi_pill = &req->rq_pill;
903 tsi->tsi_env = req->rq_svc_thread->t_env;
904 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
905 tsi->tsi_exp = req->rq_export;
907 h = tgt_handler_find_check(req);
913 rc = tgt_request_preprocess(tsi, h, req);
914 if (unlikely(rc != 0))
917 if (h->th_hp != NULL)
921 EXPORT_SYMBOL(tgt_hpreq_handler);
923 void tgt_counter_incr(struct obd_export *exp, int opcode)
925 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
926 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
927 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
929 EXPORT_SYMBOL(tgt_counter_incr);
932 * Unified target generic handlers.
935 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
937 struct lu_target *tgt = class_exp2tgt(exp);
938 struct sptlrpc_flavor flvr;
942 LASSERT(tgt->lut_obd);
943 LASSERT(tgt->lut_slice);
945 /* always allow ECHO client */
946 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
947 LUSTRE_ECHO_NAME) == 0)) {
948 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
952 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
953 read_lock(&tgt->lut_sptlrpc_lock);
954 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
958 read_unlock(&tgt->lut_sptlrpc_lock);
960 spin_lock(&exp->exp_lock);
961 exp->exp_sp_peer = req->rq_sp_from;
962 exp->exp_flvr = flvr;
964 /* when on mgs, if no restriction is set, or if the client
965 * NID is on the local node, allow any flavor
967 if ((strcmp(exp->exp_obd->obd_type->typ_name,
968 LUSTRE_MGS_NAME) == 0) &&
969 (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
970 LNetIsPeerLocal(&exp->exp_connection->c_peer.nid)))
971 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
973 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
974 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
975 CERROR("%s: unauthorized rpc flavor %x from %s, "
976 "expect %x\n", tgt_name(tgt),
978 libcfs_nid2str(req->rq_peer.nid),
979 exp->exp_flvr.sf_rpc);
982 spin_unlock(&exp->exp_lock);
984 if (exp->exp_sp_peer != req->rq_sp_from) {
985 CERROR("%s: RPC source %s doesn't match %s\n",
987 sptlrpc_part2name(req->rq_sp_from),
988 sptlrpc_part2name(exp->exp_sp_peer));
991 rc = sptlrpc_target_export_check(exp, req);
998 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt)
1000 struct sptlrpc_rule_set tmp_rset;
1003 if (unlikely(tgt == NULL)) {
1004 CERROR("No target passed\n");
1008 sptlrpc_rule_set_init(&tmp_rset);
1009 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset);
1011 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
1016 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
1018 write_lock(&tgt->lut_sptlrpc_lock);
1019 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
1020 tgt->lut_sptlrpc_rset = tmp_rset;
1021 write_unlock(&tgt->lut_sptlrpc_lock);
1025 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
1027 int tgt_connect(struct tgt_session_info *tsi)
1029 struct ptlrpc_request *req = tgt_ses_req(tsi);
1030 struct obd_connect_data *reply;
1035 /* XXX: better to call this check right after getting new export but
1036 * before last_rcvd slot allocation to avoid server load upon insecure
1037 * connects. This is to be fixed after unifiyng all targets.
1039 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
1043 /* To avoid exposing partially initialized connection flags, changes up
1044 * to this point have been staged in reply->ocd_connect_flags. Now that
1045 * connection handling has completed successfully, atomically update
1046 * the connect flags in the shared export data structure. LU-1623 */
1047 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1048 spin_lock(&tsi->tsi_exp->exp_lock);
1049 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1050 if (reply->ocd_connect_flags & OBD_CONNECT_FLAGS2)
1051 *exp_connect_flags2_ptr(tsi->tsi_exp) =
1052 reply->ocd_connect_flags2;
1053 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1054 spin_unlock(&tsi->tsi_exp->exp_lock);
1056 if (strcmp(tsi->tsi_exp->exp_obd->obd_type->typ_name,
1057 LUSTRE_MDT_NAME) == 0) {
1058 struct lu_nodemap *nm = NULL;
1060 rc = req_check_sepol(tsi->tsi_pill);
1064 if (tsi->tsi_pill->rc_req->rq_export)
1066 nodemap_get_from_exp(tsi->tsi_pill->rc_req->rq_export);
1068 if (reply->ocd_connect_flags & OBD_CONNECT_FLAGS2 &&
1069 reply->ocd_connect_flags2 & OBD_CONNECT2_ENCRYPT) {
1070 bool forbid_encrypt = true;
1073 /* nodemap_get_from_exp returns NULL in case
1074 * nodemap is not active, so we do not forbid
1076 forbid_encrypt = false;
1077 else if (!IS_ERR(nm))
1078 forbid_encrypt = nm->nmf_forbid_encryption;
1080 GOTO(put_nm, rc = -EACCES);
1083 if (!(reply->ocd_connect_flags & OBD_CONNECT_RDONLY)) {
1084 bool readonly = false;
1086 if (!IS_ERR_OR_NULL(nm))
1087 readonly = nm->nmf_readonly_mount;
1088 if (unlikely(readonly))
1089 GOTO(put_nm, rc = -EROFS);
1093 if (!IS_ERR_OR_NULL(nm))
1101 obd_disconnect(class_export_get(tsi->tsi_exp));
1104 EXPORT_SYMBOL(tgt_connect);
1106 int tgt_disconnect(struct tgt_session_info *tsi)
1112 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DISCONNECT_DELAY, cfs_fail_val);
1114 rc = target_handle_disconnect(tgt_ses_req(tsi));
1116 RETURN(err_serious(rc));
1120 EXPORT_SYMBOL(tgt_disconnect);
1123 * Unified target OBD handlers
1125 int tgt_obd_ping(struct tgt_session_info *tsi)
1131 /* The target-specific part of OBD_PING request handling.
1132 * It controls Filter Modification Data (FMD) expiration each time
1135 * Valid only for replayable targets, e.g. MDT and OFD
1137 if (tsi->tsi_exp->exp_obd->obd_replayable)
1138 tgt_fmd_expire(tsi->tsi_exp);
1140 rc = req_capsule_server_pack(tsi->tsi_pill);
1142 RETURN(err_serious(rc));
1146 EXPORT_SYMBOL(tgt_obd_ping);
1148 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1150 return err_serious(-EOPNOTSUPP);
1153 int tgt_send_buffer(struct tgt_session_info *tsi, struct lu_rdbuf *rdbuf)
1155 struct ptlrpc_request *req = tgt_ses_req(tsi);
1156 struct obd_export *exp = req->rq_export;
1157 struct ptlrpc_bulk_desc *desc;
1164 for (i = 0; i < rdbuf->rb_nbufs; i++) {
1165 unsigned int offset;
1167 offset = (unsigned long)rdbuf->rb_bufs[i].lb_buf & ~PAGE_MASK;
1168 pages += DIV_ROUND_UP(rdbuf->rb_bufs[i].lb_len + offset,
1172 desc = ptlrpc_prep_bulk_exp(req, pages, 1,
1173 PTLRPC_BULK_PUT_SOURCE,
1175 &ptlrpc_bulk_kiov_nopin_ops);
1179 for (i = 0; i < rdbuf->rb_nbufs; i++)
1180 desc->bd_frag_ops->add_iov_frag(desc,
1181 rdbuf->rb_bufs[i].lb_buf,
1182 rdbuf->rb_bufs[i].lb_len);
1184 rc = target_bulk_io(exp, desc);
1185 ptlrpc_free_bulk(desc);
1188 EXPORT_SYMBOL(tgt_send_buffer);
1190 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1192 struct ptlrpc_request *req = tgt_ses_req(tsi);
1193 struct obd_export *exp = req->rq_export;
1194 struct ptlrpc_bulk_desc *desc;
1202 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1203 PTLRPC_BULK_PUT_SOURCE,
1205 &ptlrpc_bulk_kiov_pin_ops);
1209 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1210 /* old client requires reply size in it's PAGE_SIZE,
1211 * which is rdpg->rp_count */
1212 nob = rdpg->rp_count;
1214 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1215 i++, tmpcount -= tmpsize) {
1216 tmpsize = min_t(int, tmpcount, PAGE_SIZE);
1217 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1221 LASSERT(desc->bd_nob == nob);
1222 rc = target_bulk_io(exp, desc);
1223 ptlrpc_free_bulk(desc);
1226 EXPORT_SYMBOL(tgt_sendpage);
1229 * OBD_IDX_READ handler
1231 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1233 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1234 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1235 struct idx_info *req_ii, *rep_ii;
1240 memset(rdpg, 0, sizeof(*rdpg));
1241 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1243 /* extract idx_info buffer from request & reply */
1244 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1245 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1246 RETURN(err_serious(-EPROTO));
1248 rc = req_capsule_server_pack(tsi->tsi_pill);
1250 RETURN(err_serious(rc));
1252 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1254 RETURN(err_serious(-EFAULT));
1255 rep_ii->ii_magic = IDX_INFO_MAGIC;
1257 /* extract hash to start with */
1258 rdpg->rp_hash = req_ii->ii_hash_start;
1260 /* extract requested attributes */
1261 rdpg->rp_attrs = req_ii->ii_attrs;
1263 /* check that fid packed in request is valid and supported */
1264 if (!fid_is_sane(&req_ii->ii_fid))
1266 rep_ii->ii_fid = req_ii->ii_fid;
1269 rep_ii->ii_flags = req_ii->ii_flags;
1271 /* compute number of pages to allocate, ii_count is the number of 4KB
1273 if (req_ii->ii_count <= 0)
1274 GOTO(out, rc = -EFAULT);
1275 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1276 exp_max_brw_size(tsi->tsi_exp));
1277 rdpg->rp_npages = (rdpg->rp_count + PAGE_SIZE - 1) >> PAGE_SHIFT;
1279 /* allocate pages to store the containers */
1280 OBD_ALLOC_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages);
1281 if (rdpg->rp_pages == NULL)
1282 GOTO(out, rc = -ENOMEM);
1283 for (i = 0; i < rdpg->rp_npages; i++) {
1284 rdpg->rp_pages[i] = alloc_page(GFP_NOFS);
1285 if (rdpg->rp_pages[i] == NULL)
1286 GOTO(out, rc = -ENOMEM);
1289 /* populate pages with key/record pairs */
1290 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1294 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1295 "asked %d > %d\n", rc, rdpg->rp_count);
1297 /* send pages to client */
1298 rc = tgt_sendpage(tsi, rdpg, rc);
1303 if (rdpg->rp_pages) {
1304 for (i = 0; i < rdpg->rp_npages; i++)
1305 if (rdpg->rp_pages[i])
1306 __free_page(rdpg->rp_pages[i]);
1307 OBD_FREE_PTR_ARRAY(rdpg->rp_pages, rdpg->rp_npages);
1312 struct tgt_handler tgt_obd_handlers[] = {
1313 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1314 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1316 EXPORT_SYMBOL(tgt_obd_handlers);
1318 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1319 struct dt_object *obj, __u64 start, __u64 end)
1325 /* if no objid is specified, it means "sync whole filesystem" */
1327 rc = dt_sync(env, tgt->lut_bottom);
1328 } else if (dt_version_get(env, obj) >
1329 tgt->lut_obd->obd_last_committed) {
1330 rc = dt_object_sync(env, obj, start, end);
1332 atomic_inc(&tgt->lut_sync_count);
1336 EXPORT_SYMBOL(tgt_sync);
1338 * Unified target DLM handlers.
1342 * Unified target BAST
1344 * Ensure data and metadata are synced to disk when lock is canceled if Sync on
1345 * Cancel (SOC) is enabled. If it's extent lock, normally sync obj is enough,
1346 * but if it's cross-MDT lock, because remote object version is not set, a
1347 * filesystem sync is needed.
1349 * \param lock server side lock
1350 * \param desc lock desc
1351 * \param data ldlm_cb_set_arg
1352 * \param flag indicates whether this cancelling or blocking callback
1353 * \retval 0 on success
1354 * \retval negative number on error
1356 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1357 void *data, int flag)
1360 struct lu_target *tgt;
1361 struct dt_object *obj = NULL;
1367 tgt = class_exp2tgt(lock->l_export);
1369 if (unlikely(tgt == NULL)) {
1370 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1371 class_exp2obd(lock->l_export)->obd_name);
1375 if (flag == LDLM_CB_CANCELING &&
1376 (lock->l_granted_mode & (LCK_EX | LCK_PW | LCK_GROUP)) &&
1377 (tgt->lut_sync_lock_cancel == SYNC_LOCK_CANCEL_ALWAYS ||
1378 (tgt->lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING &&
1379 ldlm_is_cbpending(lock))) &&
1380 ((exp_connect_flags(lock->l_export) & OBD_CONNECT_MDS_MDS) ||
1381 lock->l_resource->lr_type == LDLM_EXTENT)) {
1383 __u64 end = OBD_OBJECT_EOF;
1385 rc = lu_env_init(&env, LCT_DT_THREAD);
1386 if (unlikely(rc != 0))
1389 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1390 tgt->lut_lsd.lsd_osd_index);
1392 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1393 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1395 GOTO(err_env, rc = PTR_ERR(obj));
1397 if (!dt_object_exists(obj))
1398 GOTO(err_put, rc = -ENOENT);
1400 start = lock->l_policy_data.l_extent.start;
1401 end = lock->l_policy_data.l_extent.end;
1404 rc = tgt_sync(&env, tgt, obj, start, end);
1406 CERROR("%s: syncing "DFID" (%llu-%llu) on lock "
1407 "cancel: rc = %d\n",
1408 tgt_name(tgt), PFID(&fid),
1409 lock->l_policy_data.l_extent.start,
1410 lock->l_policy_data.l_extent.end, rc);
1414 dt_object_put(&env, obj);
1419 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1423 static struct ldlm_callback_suite tgt_dlm_cbs = {
1424 .lcs_completion = ldlm_server_completion_ast,
1425 .lcs_blocking = tgt_blocking_ast,
1426 .lcs_glimpse = ldlm_server_glimpse_ast
1429 int tgt_enqueue(struct tgt_session_info *tsi)
1431 struct ptlrpc_request *req = tgt_ses_req(tsi);
1436 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1437 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1439 LASSERT(tsi->tsi_dlm_req != NULL);
1440 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1441 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1443 RETURN(err_serious(rc));
1445 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1447 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1450 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1453 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1456 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1459 RETURN(req->rq_status);
1461 EXPORT_SYMBOL(tgt_enqueue);
1463 int tgt_convert(struct tgt_session_info *tsi)
1465 struct ptlrpc_request *req = tgt_ses_req(tsi);
1469 LASSERT(tsi->tsi_dlm_req);
1470 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1472 RETURN(err_serious(rc));
1474 RETURN(req->rq_status);
1477 int tgt_bl_callback(struct tgt_session_info *tsi)
1479 return err_serious(-EOPNOTSUPP);
1482 int tgt_cp_callback(struct tgt_session_info *tsi)
1484 return err_serious(-EOPNOTSUPP);
1487 /* generic LDLM target handler */
1488 struct tgt_handler tgt_dlm_handlers[] = {
1489 TGT_DLM_HDL(HAS_KEY, LDLM_ENQUEUE, tgt_enqueue),
1490 TGT_DLM_HDL(HAS_KEY, LDLM_CONVERT, tgt_convert),
1491 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1492 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1494 EXPORT_SYMBOL(tgt_dlm_handlers);
1497 * Unified target LLOG handlers.
1499 int tgt_llog_open(struct tgt_session_info *tsi)
1505 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1509 EXPORT_SYMBOL(tgt_llog_open);
1511 int tgt_llog_read_header(struct tgt_session_info *tsi)
1517 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1521 EXPORT_SYMBOL(tgt_llog_read_header);
1523 int tgt_llog_next_block(struct tgt_session_info *tsi)
1529 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1533 EXPORT_SYMBOL(tgt_llog_next_block);
1535 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1541 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1545 EXPORT_SYMBOL(tgt_llog_prev_block);
1547 /* generic llog target handler */
1548 struct tgt_handler tgt_llog_handlers[] = {
1549 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1550 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1551 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1552 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1554 EXPORT_SYMBOL(tgt_llog_handlers);
1557 * sec context handlers
1559 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1560 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1565 struct tgt_handler tgt_sec_ctx_handlers[] = {
1566 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1567 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1568 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1570 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1572 int (*tgt_lfsck_in_notify_local)(const struct lu_env *env,
1573 struct dt_device *key,
1574 struct lfsck_req_local *lrl,
1575 struct thandle *th) = NULL;
1577 void tgt_register_lfsck_in_notify_local(int (*notify)(const struct lu_env *,
1579 struct lfsck_req_local *,
1582 tgt_lfsck_in_notify_local = notify;
1584 EXPORT_SYMBOL(tgt_register_lfsck_in_notify_local);
1586 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1587 struct dt_device *key,
1588 struct lfsck_request *lr) = NULL;
1590 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1592 struct lfsck_request *))
1594 tgt_lfsck_in_notify = notify;
1596 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1598 static int (*tgt_lfsck_query)(const struct lu_env *env,
1599 struct dt_device *key,
1600 struct lfsck_request *req,
1601 struct lfsck_reply *rep,
1602 struct lfsck_query *que) = NULL;
1604 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1606 struct lfsck_request *,
1607 struct lfsck_reply *,
1608 struct lfsck_query *))
1610 tgt_lfsck_query = query;
1612 EXPORT_SYMBOL(tgt_register_lfsck_query);
1614 /* LFSCK request handlers */
1615 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1617 const struct lu_env *env = tsi->tsi_env;
1618 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1619 struct lfsck_request *lr;
1623 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1627 rc = tgt_lfsck_in_notify(env, key, lr);
1632 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1634 struct lfsck_request *request;
1635 struct lfsck_reply *reply;
1639 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1640 if (request == NULL)
1643 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1647 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom,
1648 request, reply, NULL);
1650 RETURN(rc < 0 ? rc : 0);
1653 struct tgt_handler tgt_lfsck_handlers[] = {
1654 TGT_LFSCK_HDL(HAS_REPLY, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1655 TGT_LFSCK_HDL(HAS_REPLY, LFSCK_QUERY, tgt_handle_lfsck_query),
1657 EXPORT_SYMBOL(tgt_lfsck_handlers);
1660 * initialize per-thread page pool (bug 5137).
1662 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1664 struct tgt_thread_big_cache *tbc;
1668 LASSERT(thread != NULL);
1669 LASSERT(thread->t_data == NULL);
1671 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1674 thread->t_data = tbc;
1677 EXPORT_SYMBOL(tgt_io_thread_init);
1680 * free per-thread pool created by tgt_thread_init().
1682 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1684 struct tgt_thread_big_cache *tbc;
1688 LASSERT(thread != NULL);
1691 * be prepared to handle partially-initialized pools (because this is
1692 * called from ost_io_thread_init() for cleanup.
1694 tbc = thread->t_data;
1696 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1697 thread->t_data = NULL;
1701 EXPORT_SYMBOL(tgt_io_thread_done);
1704 * Helper function for getting Data-on-MDT file server DLM lock
1705 * if asked by client.
1707 int tgt_mdt_data_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1708 struct lustre_handle *lh, int mode, __u64 *flags)
1710 union ldlm_policy_data policy = {
1711 .l_inodebits.bits = MDS_INODELOCK_DOM,
1717 LASSERT(lh != NULL);
1718 LASSERT(ns != NULL);
1719 LASSERT(!lustre_handle_is_used(lh));
1721 rc = ldlm_cli_enqueue_local(NULL, ns, res_id, LDLM_IBITS, &policy, mode,
1722 flags, ldlm_blocking_ast,
1723 ldlm_completion_ast, ldlm_glimpse_ast,
1724 NULL, 0, LVB_T_NONE, NULL, lh);
1726 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1728 EXPORT_SYMBOL(tgt_mdt_data_lock);
1731 * Helper function for getting server side [start, start+count] DLM lock
1732 * if asked by client.
1734 int tgt_extent_lock(const struct lu_env *env, struct ldlm_namespace *ns,
1735 struct ldlm_res_id *res_id, __u64 start, __u64 end,
1736 struct lustre_handle *lh, int mode, __u64 *flags)
1738 union ldlm_policy_data policy;
1743 LASSERT(lh != NULL);
1744 LASSERT(ns != NULL);
1745 LASSERT(!lustre_handle_is_used(lh));
1747 policy.l_extent.gid = 0;
1748 policy.l_extent.start = start & PAGE_MASK;
1751 * If ->o_blocks is EOF it means "lock till the end of the file".
1752 * Otherwise, it's size of an extent or hole being punched (in bytes).
1754 if (end == OBD_OBJECT_EOF || end < start)
1755 policy.l_extent.end = OBD_OBJECT_EOF;
1757 policy.l_extent.end = end | ~PAGE_MASK;
1759 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_EXTENT, &policy,
1760 mode, flags, ldlm_blocking_ast,
1761 ldlm_completion_ast, ldlm_glimpse_ast,
1762 NULL, 0, LVB_T_NONE, NULL, lh);
1763 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1765 EXPORT_SYMBOL(tgt_extent_lock);
1767 static int tgt_data_lock(const struct lu_env *env, struct obd_export *exp,
1768 struct ldlm_res_id *res_id, __u64 start, __u64 end,
1769 struct lustre_handle *lh, enum ldlm_mode mode)
1771 struct ldlm_namespace *ns = exp->exp_obd->obd_namespace;
1774 /* MDT IO for data-on-mdt */
1775 if (exp->exp_connect_data.ocd_connect_flags & OBD_CONNECT_IBITS)
1776 return tgt_mdt_data_lock(ns, res_id, lh, mode, &flags);
1778 return tgt_extent_lock(env, ns, res_id, start, end, lh, mode, &flags);
1781 void tgt_data_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1783 LASSERT(lustre_handle_is_used(lh));
1784 ldlm_lock_decref(lh, mode);
1786 EXPORT_SYMBOL(tgt_data_unlock);
1788 static int tgt_brw_lock(const struct lu_env *env, struct obd_export *exp,
1789 struct ldlm_res_id *res_id, struct obd_ioobj *obj,
1790 struct niobuf_remote *nb, struct lustre_handle *lh,
1791 enum ldlm_mode mode)
1793 int nrbufs = obj->ioo_bufcnt;
1798 LASSERT(mode == LCK_PR || mode == LCK_PW);
1799 LASSERT(!lustre_handle_is_used(lh));
1801 if (exp->exp_obd->obd_recovering)
1804 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1807 for (i = 1; i < nrbufs; i++)
1808 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1811 return tgt_data_lock(env, exp, res_id, nb[0].rnb_offset,
1812 nb[nrbufs - 1].rnb_offset +
1813 nb[nrbufs - 1].rnb_len - 1, lh, mode);
1816 static void tgt_brw_unlock(struct obd_export *exp, struct obd_ioobj *obj,
1817 struct niobuf_remote *niob,
1818 struct lustre_handle *lh, enum ldlm_mode mode)
1822 LASSERT(mode == LCK_PR || mode == LCK_PW);
1823 LASSERT((!exp->exp_obd->obd_recovering && obj->ioo_bufcnt &&
1824 niob[0].rnb_flags & OBD_BRW_SRVLOCK) ==
1825 lustre_handle_is_used(lh));
1827 if (lustre_handle_is_used(lh))
1828 tgt_data_unlock(lh, mode);
1832 static int tgt_checksum_niobuf(struct lu_target *tgt,
1833 struct niobuf_local *local_nb, int npages,
1834 int opc, enum cksum_types cksum_type,
1837 struct ahash_request *req;
1838 unsigned int bufsize;
1840 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1842 req = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1844 CERROR("%s: unable to initialize checksum hash %s\n",
1845 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1846 return PTR_ERR(req);
1849 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1850 for (i = 0; i < npages; i++) {
1851 /* corrupt the data before we compute the checksum, to
1852 * simulate a client->OST data error */
1853 if (i == 0 && opc == OST_WRITE &&
1854 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1855 int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
1856 int len = local_nb[i].lnb_len;
1857 struct page *np = tgt_page_to_corrupt;
1860 char *ptr = kmap_atomic(local_nb[i].lnb_page);
1861 char *ptr2 = page_address(np);
1863 memcpy(ptr2 + off, ptr + off, len);
1864 memcpy(ptr2 + off, "bad3", min(4, len));
1867 /* LU-8376 to preserve original index for
1868 * display in dump_all_bulk_pages() */
1871 cfs_crypto_hash_update_page(req, np, off,
1875 CERROR("%s: can't alloc page for corruption\n",
1879 cfs_crypto_hash_update_page(req, local_nb[i].lnb_page,
1880 local_nb[i].lnb_page_offset & ~PAGE_MASK,
1881 local_nb[i].lnb_len);
1883 /* corrupt the data after we compute the checksum, to
1884 * simulate an OST->client data error */
1885 if (i == 0 && opc == OST_READ &&
1886 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1887 int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
1888 int len = local_nb[i].lnb_len;
1889 struct page *np = tgt_page_to_corrupt;
1892 char *ptr = kmap_atomic(local_nb[i].lnb_page);
1893 char *ptr2 = page_address(np);
1895 memcpy(ptr2 + off, ptr + off, len);
1896 memcpy(ptr2 + off, "bad4", min(4, len));
1899 /* LU-8376 to preserve original index for
1900 * display in dump_all_bulk_pages() */
1903 cfs_crypto_hash_update_page(req, np, off,
1907 CERROR("%s: can't alloc page for corruption\n",
1913 bufsize = sizeof(*cksum);
1914 err = cfs_crypto_hash_final(req, (unsigned char *)cksum, &bufsize);
1919 char dbgcksum_file_name[PATH_MAX];
1921 static void dump_all_bulk_pages(struct obdo *oa, int count,
1922 struct niobuf_local *local_nb,
1923 __u32 server_cksum, __u32 client_cksum)
1930 /* will only keep dump of pages on first error for the same range in
1931 * file/fid, not during the resends/retries. */
1932 snprintf(dbgcksum_file_name, sizeof(dbgcksum_file_name),
1933 "%s-checksum_dump-ost-"DFID":[%llu-%llu]-%x-%x",
1934 (strncmp(libcfs_debug_file_path, "NONE", 4) != 0 ?
1935 libcfs_debug_file_path : LIBCFS_DEBUG_FILE_PATH_DEFAULT),
1936 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : (__u64)0,
1937 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
1938 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
1939 local_nb[0].lnb_file_offset,
1940 local_nb[count-1].lnb_file_offset +
1941 local_nb[count-1].lnb_len - 1, client_cksum, server_cksum);
1942 CWARN("dumping checksum data to %s\n", dbgcksum_file_name);
1943 filp = filp_open(dbgcksum_file_name,
1944 O_CREAT | O_EXCL | O_WRONLY | O_LARGEFILE, 0600);
1948 CDEBUG(D_INFO, "%s: can't open to dump pages with "
1949 "checksum error: rc = %d\n", dbgcksum_file_name,
1952 CERROR("%s: can't open to dump pages with checksum "
1953 "error: rc = %d\n", dbgcksum_file_name, rc);
1957 for (i = 0; i < count; i++) {
1958 len = local_nb[i].lnb_len;
1959 buf = kmap(local_nb[i].lnb_page);
1961 rc = cfs_kernel_write(filp, buf, len, &filp->f_pos);
1963 CERROR("%s: wanted to write %u but got %d "
1964 "error\n", dbgcksum_file_name, len, rc);
1970 kunmap(local_nb[i].lnb_page);
1973 rc = vfs_fsync_range(filp, 0, LLONG_MAX, 1);
1975 CERROR("%s: sync returns %d\n", dbgcksum_file_name, rc);
1976 filp_close(filp, NULL);
1978 libcfs_debug_dumplog();
1981 static int check_read_checksum(struct niobuf_local *local_nb, int npages,
1982 struct obd_export *exp, struct obdo *oa,
1983 const struct lnet_process_id *peer,
1984 __u32 client_cksum, __u32 server_cksum,
1985 enum cksum_types server_cksum_type)
1988 enum cksum_types cksum_type;
1991 if (unlikely(npages <= 0))
1994 /* unlikely to happen and only if resend does not occur due to cksum
1995 * control failure on Client */
1996 if (unlikely(server_cksum == client_cksum)) {
1997 CDEBUG(D_PAGE, "checksum %x confirmed upon retry\n",
2002 if (exp->exp_obd->obd_checksum_dump)
2003 dump_all_bulk_pages(oa, npages, local_nb, server_cksum,
2006 cksum_type = obd_cksum_type_unpack(oa->o_valid & OBD_MD_FLFLAGS ?
2009 if (cksum_type != server_cksum_type)
2010 msg = "the server may have not used the checksum type specified"
2011 " in the original request - likely a protocol problem";
2013 msg = "should have changed on the client or in transit";
2015 start = local_nb[0].lnb_file_offset;
2016 end = local_nb[npages-1].lnb_file_offset +
2017 local_nb[npages-1].lnb_len - 1;
2019 LCONSOLE_ERROR_MSG(0x132, "%s: BAD READ CHECKSUM: %s: from %s inode "
2020 DFID " object "DOSTID" extent [%llu-%llu], client returned csum"
2021 " %x (type %x), server csum %x (type %x)\n",
2022 exp->exp_obd->obd_name,
2023 msg, libcfs_nid2str(peer->nid),
2024 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_seq : 0ULL,
2025 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_oid : 0,
2026 oa->o_valid & OBD_MD_FLFID ? oa->o_parent_ver : 0,
2028 start, end, client_cksum, cksum_type, server_cksum,
2034 static int tgt_pages2shortio(struct niobuf_local *local, int npages,
2035 unsigned char *buf, int size)
2037 int i, off, len, copied = size;
2040 for (i = 0; i < npages; i++) {
2041 off = local[i].lnb_page_offset & ~PAGE_MASK;
2042 len = local[i].lnb_len;
2044 CDEBUG(D_PAGE, "index %d offset = %d len = %d left = %d\n",
2049 ptr = kmap_atomic(local[i].lnb_page);
2050 memcpy(buf, ptr + off, len);
2055 return copied - size;
2058 static int tgt_checksum_niobuf_t10pi(struct lu_target *tgt,
2059 struct niobuf_local *local_nb, int npages,
2060 int opc, obd_dif_csum_fn *fn,
2061 int sector_size, u32 *check_sum,
2064 enum cksum_types t10_cksum_type = tgt->lut_dt_conf.ddp_t10_cksum_type;
2065 unsigned char cfs_alg = cksum_obd2cfs(OBD_CKSUM_T10_TOP);
2066 const char *obd_name = tgt->lut_obd->obd_name;
2067 struct ahash_request *req;
2068 unsigned int bufsize;
2069 unsigned char *buffer;
2070 struct page *__page;
2073 int used_number = 0;
2079 __page = alloc_page(GFP_KERNEL);
2083 req = cfs_crypto_hash_init(cfs_alg, NULL, 0);
2086 CERROR("%s: unable to initialize checksum hash %s: rc = %d\n",
2087 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg), rc);
2091 buffer = kmap(__page);
2092 guard_start = (__u16 *)buffer;
2093 guard_number = PAGE_SIZE / sizeof(*guard_start);
2094 if (unlikely(resend))
2095 CDEBUG(D_PAGE | D_HA, "GRD tags per page = %u\n", guard_number);
2096 for (i = 0; i < npages; i++) {
2099 /* corrupt the data before we compute the checksum, to
2100 * simulate a client->OST data error */
2101 if (i == 0 && opc == OST_WRITE &&
2102 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
2103 int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
2104 int len = local_nb[i].lnb_len;
2105 struct page *np = tgt_page_to_corrupt;
2108 char *ptr = kmap_atomic(local_nb[i].lnb_page);
2109 char *ptr2 = page_address(np);
2111 memcpy(ptr2 + off, ptr + off, len);
2112 memcpy(ptr2 + off, "bad3", min(4, len));
2115 /* LU-8376 to preserve original index for
2116 * display in dump_all_bulk_pages() */
2119 cfs_crypto_hash_update_page(req, np, off,
2123 CERROR("%s: can't alloc page for corruption\n",
2129 * The left guard number should be able to hold checksums of a
2132 use_t10_grd = t10_cksum_type && opc == OST_READ &&
2133 local_nb[i].lnb_len == PAGE_SIZE &&
2134 local_nb[i].lnb_guard_disk;
2136 used = DIV_ROUND_UP(local_nb[i].lnb_len, sector_size);
2137 if (used > (guard_number - used_number)) {
2141 memcpy(guard_start + used_number,
2142 local_nb[i].lnb_guards,
2143 used * sizeof(*guard_start));
2144 if (unlikely(resend))
2145 CDEBUG(D_PAGE | D_HA,
2146 "lnb[%u]: used %u off %u+%u lnb checksum: %*phN\n",
2148 local_nb[i].lnb_page_offset,
2149 local_nb[i].lnb_len,
2150 (int)(used * sizeof(*guard_start)),
2151 guard_start + used_number);
2153 if (!use_t10_grd || unlikely(resend)) {
2154 __u16 guard_tmp[MAX_GUARD_NUMBER];
2155 __u16 *guards = guard_start + used_number;
2156 int used_tmp = -1, *usedp = &used;
2158 if (unlikely(use_t10_grd)) {
2162 rc = obd_page_dif_generate_buffer(obd_name,
2163 local_nb[i].lnb_page,
2164 local_nb[i].lnb_page_offset & ~PAGE_MASK,
2165 local_nb[i].lnb_len, guards,
2166 guard_number - used_number, usedp, sector_size,
2168 if (unlikely(resend)) {
2169 bool bad = use_t10_grd &&
2171 local_nb[i].lnb_guards,
2172 used_tmp * sizeof(*guard_tmp));
2175 CERROR("lnb[%u]: used %u/%u off %u+%u tmp checksum: %*phN\n",
2177 local_nb[i].lnb_page_offset,
2178 local_nb[i].lnb_len,
2179 (int)(used_tmp * sizeof(*guard_start)),
2181 CDEBUG_LIMIT(D_PAGE | D_HA | (bad ? D_ERROR : 0),
2182 "lnb[%u]: used %u/%u off %u+%u gen checksum: %*phN\n",
2184 local_nb[i].lnb_page_offset,
2185 local_nb[i].lnb_len,
2186 (int)(used * sizeof(*guard_start)),
2187 guard_start + used_number);
2193 LASSERT(used <= MAX_GUARD_NUMBER);
2195 * If disk support T10PI checksum, copy guards to local_nb.
2196 * If the write is partial page, do not use the guards for bio
2197 * submission since the data might not be full-sector. The bio
2198 * guards will be generated later based on the full sectors. If
2199 * the sector size is 512B rather than 4 KB, or the page size
2200 * is larger than 4KB, this might drop some useful guards for
2201 * partial page write, but it will only add minimal extra time
2202 * of checksum calculation.
2204 if (t10_cksum_type && opc == OST_WRITE &&
2205 local_nb[i].lnb_len == PAGE_SIZE) {
2206 local_nb[i].lnb_guard_rpc = 1;
2207 memcpy(local_nb[i].lnb_guards,
2208 guard_start + used_number,
2209 used * sizeof(*local_nb[i].lnb_guards));
2212 used_number += used;
2213 if (used_number == guard_number) {
2214 cfs_crypto_hash_update_page(req, __page, 0,
2215 used_number * sizeof(*guard_start));
2219 /* corrupt the data after we compute the checksum, to
2220 * simulate an OST->client data error */
2221 if (unlikely(i == 0 && opc == OST_READ &&
2222 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND))) {
2223 int off = local_nb[i].lnb_page_offset & ~PAGE_MASK;
2224 int len = local_nb[i].lnb_len;
2225 struct page *np = tgt_page_to_corrupt;
2228 char *ptr = kmap_atomic(local_nb[i].lnb_page);
2229 char *ptr2 = page_address(np);
2231 memcpy(ptr2 + off, ptr + off, len);
2232 memcpy(ptr2 + off, "bad4", min(4, len));
2235 /* LU-8376 to preserve original index for
2236 * display in dump_all_bulk_pages() */
2239 cfs_crypto_hash_update_page(req, np, off,
2243 CERROR("%s: can't alloc page for corruption\n",
2252 if (used_number != 0)
2253 cfs_crypto_hash_update_page(req, __page, 0,
2254 used_number * sizeof(*guard_start));
2256 bufsize = sizeof(cksum);
2257 rc = cfs_crypto_hash_final(req, (unsigned char *)&cksum, &bufsize);
2262 __free_page(__page);
2266 static int tgt_checksum_niobuf_rw(struct lu_target *tgt,
2267 enum cksum_types cksum_type,
2268 struct niobuf_local *local_nb,
2269 int npages, int opc, u32 *check_sum,
2272 obd_dif_csum_fn *fn = NULL;
2273 int sector_size = 0;
2277 obd_t10_cksum2dif(cksum_type, &fn, §or_size);
2280 rc = tgt_checksum_niobuf_t10pi(tgt, local_nb, npages,
2281 opc, fn, sector_size,
2284 rc = tgt_checksum_niobuf(tgt, local_nb, npages, opc,
2285 cksum_type, check_sum);
2290 int tgt_brw_read(struct tgt_session_info *tsi)
2292 struct ptlrpc_request *req = tgt_ses_req(tsi);
2293 struct ptlrpc_bulk_desc *desc = NULL;
2294 struct obd_export *exp = tsi->tsi_exp;
2295 struct niobuf_remote *remote_nb;
2296 struct niobuf_local *local_nb;
2297 struct obd_ioobj *ioo;
2298 struct ost_body *body, *repbody;
2299 struct lustre_handle lockh = { 0 };
2300 int npages, nob = 0, rc, i, no_reply = 0,
2302 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2303 const char *obd_name = exp->exp_obd->obd_name;
2308 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL &&
2309 ptlrpc_req2svc(req)->srv_req_portal != MDS_IO_PORTAL) {
2310 CERROR("%s: deny read request from %s to portal %u\n",
2311 tgt_name(tsi->tsi_tgt),
2312 obd_export_nid2str(req->rq_export),
2313 ptlrpc_req2svc(req)->srv_req_portal);
2317 req->rq_bulk_read = 1;
2319 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK)) {
2320 /* optionally use cfs_fail_val - 1 to select a specific OST on
2321 * this server to fail requests.
2323 char fail_ost_name[MAX_OBD_NAME];
2325 if (cfs_fail_val > 0) {
2326 snprintf(fail_ost_name, MAX_OBD_NAME, "OST%04X",
2329 if (strstr(obd_name, fail_ost_name))
2330 RETURN(err_serious(-EIO));
2332 RETURN(err_serious(-EIO));
2336 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2337 cfs_fail_val : (obd_timeout + 1) / 4);
2339 /* Check if there is eviction in progress, and if so, wait for it to
2341 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
2342 /* We do not care how long it takes */
2344 exp->exp_obd->obd_evict_inprogress_waitq,
2345 !atomic_read(&exp->exp_obd->obd_evict_inprogress));
2348 /* There must be big cache in current thread to process this request
2349 * if it is NULL then something went wrong and it wasn't allocated,
2350 * report -ENOMEM in that case */
2354 body = tsi->tsi_ost_body;
2355 LASSERT(body != NULL);
2357 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2358 body->oa.o_flags & OBD_FL_NORPC)
2361 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
2362 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2364 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2365 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2367 local_nb = tbc->local;
2369 rc = tgt_brw_lock(tsi->tsi_env, exp, &tsi->tsi_resid, ioo, remote_nb,
2375 * If getting the lock took more time than
2376 * client was willing to wait, drop it. b=11330
2378 if (ktime_get_real_seconds() > req->rq_deadline ||
2379 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2381 CERROR("Dropping timed-out read from %s because locking object " DOSTID " took %lld seconds (limit was %lld).\n",
2382 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
2383 ktime_get_real_seconds() - req->rq_arrival_time.tv_sec,
2384 req->rq_deadline - req->rq_arrival_time.tv_sec);
2385 GOTO(out_lock, rc = -ETIMEDOUT);
2389 * Because we already sync grant info with client when
2390 * reconnect, grant info will be cleared for resent req,
2391 * otherwise, outdated grant count in the rpc would de-sync
2392 * grant counters in case of shrink
2394 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2395 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2396 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2399 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2400 repbody->oa = body->oa;
2402 npages = PTLRPC_MAX_BRW_PAGES;
2403 kstart = ktime_get();
2404 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
2405 ioo, remote_nb, &npages, local_nb);
2409 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2410 body->oa.o_flags & OBD_FL_SHORT_IO) {
2413 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2414 PTLRPC_BULK_PUT_SOURCE,
2416 &ptlrpc_bulk_kiov_nopin_ops);
2418 GOTO(out_commitrw, rc = -ENOMEM);
2421 npages_read = npages;
2422 for (i = 0; i < npages; i++) {
2423 int page_rc = local_nb[i].lnb_rc;
2432 if (page_rc != 0 && desc != NULL) { /* some data! */
2433 LASSERT(local_nb[i].lnb_page != NULL);
2434 desc->bd_frag_ops->add_kiov_frag
2435 (desc, local_nb[i].lnb_page,
2436 local_nb[i].lnb_page_offset & ~PAGE_MASK,
2440 if (page_rc != local_nb[i].lnb_len) { /* short read */
2441 local_nb[i].lnb_len = page_rc;
2442 npages_read = i + (page_rc != 0 ? 1 : 0);
2443 /* All subsequent pages should be 0 */
2444 while (++i < npages)
2445 LASSERT(local_nb[i].lnb_rc == 0);
2450 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
2451 u32 flag = body->oa.o_valid & OBD_MD_FLFLAGS ?
2452 body->oa.o_flags : 0;
2453 enum cksum_types cksum_type = obd_cksum_type_unpack(flag);
2454 bool resend = (body->oa.o_valid & OBD_MD_FLFLAGS) &&
2455 (body->oa.o_flags & OBD_FL_RECOV_RESEND);
2457 repbody->oa.o_flags = obd_cksum_type_pack(obd_name,
2459 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2461 rc = tgt_checksum_niobuf_rw(tsi->tsi_tgt, cksum_type,
2462 local_nb, npages_read, OST_READ,
2463 &repbody->oa.o_cksum, resend);
2465 GOTO(out_commitrw, rc);
2466 CDEBUG(D_PAGE | (resend ? D_HA : 0),
2467 "checksum at read origin: %x (%x)\n",
2468 repbody->oa.o_cksum, cksum_type);
2470 /* if a resend it could be for a cksum error, so check Server
2471 * cksum with returned Client cksum (this should even cover
2472 * zero-cksum case) */
2474 check_read_checksum(local_nb, npages_read, exp,
2475 &body->oa, &req->rq_peer,
2477 repbody->oa.o_cksum, cksum_type);
2479 repbody->oa.o_valid = 0;
2481 if (body->oa.o_valid & OBD_MD_FLGRANT)
2482 repbody->oa.o_valid |= OBD_MD_FLGRANT;
2483 /* We're finishing using body->oa as an input variable */
2485 /* Check if client was evicted while we were doing i/o before touching
2488 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2489 body->oa.o_flags & OBD_FL_SHORT_IO) {
2490 unsigned char *short_io_buf;
2493 short_io_buf = req_capsule_server_get(&req->rq_pill,
2495 short_io_size = req_capsule_get_size(&req->rq_pill,
2498 rc = tgt_pages2shortio(local_nb, npages_read,
2499 short_io_buf, short_io_size);
2501 req_capsule_shrink(&req->rq_pill,
2504 rc = rc > 0 ? 0 : rc;
2505 } else if (!CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) {
2506 rc = target_bulk_io(exp, desc);
2510 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2511 body->oa.o_flags & OBD_FL_SHORT_IO)
2512 req_capsule_shrink(&req->rq_pill, &RMF_SHORT_IO, 0,
2517 /* Must commit after prep above in all cases */
2518 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
2519 remote_nb, npages, local_nb, rc, nob, kstart);
2521 tgt_brw_unlock(exp, ioo, remote_nb, &lockh, LCK_PR);
2523 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
2524 ptlrpc_free_bulk(desc);
2529 ptlrpc_lprocfs_brw(req, nob);
2530 } else if (no_reply) {
2531 req->rq_no_reply = 1;
2532 /* reply out callback would free */
2533 ptlrpc_req_drop_rs(req);
2534 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
2535 "client will retry: rc %d\n",
2537 obd_uuid2str(&exp->exp_client_uuid),
2538 obd_export_nid2str(exp), rc);
2540 /* send a bulk after reply to simulate a network delay or reordering
2541 * by a router - Note that !desc implies short io, so there is no bulk
2543 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2)) &&
2545 /* Calculate checksum before request transfer, original
2546 * it is done by target_bulk_io() */
2547 rc = sptlrpc_svc_wrap_bulk(req, desc);
2548 if (OCD_HAS_FLAG(&exp->exp_connect_data, BULK_MBITS))
2549 req->rq_mbits = lustre_msg_get_mbits(req->rq_reqmsg);
2550 else /* old version, bulk matchbits is rq_xid */
2551 req->rq_mbits = req->rq_xid;
2553 req->rq_status = rc;
2554 target_committed_to_req(req);
2555 target_send_reply(req, 0, 0);
2557 CDEBUG(D_INFO, "reorder BULK\n");
2558 OBD_FAIL_TIMEOUT(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2,
2559 cfs_fail_val ? : 3);
2561 target_bulk_io(exp, desc);
2562 ptlrpc_free_bulk(desc);
2567 EXPORT_SYMBOL(tgt_brw_read);
2569 static int tgt_shortio2pages(struct niobuf_local *local, int npages,
2570 unsigned char *buf, unsigned int size)
2575 for (i = 0; i < npages; i++) {
2576 off = local[i].lnb_page_offset & ~PAGE_MASK;
2577 len = local[i].lnb_len;
2582 CDEBUG(D_PAGE, "index %d offset = %d len = %d left = %d\n",
2584 ptr = kmap_atomic(local[i].lnb_page);
2587 memcpy(ptr + off, buf, len < size ? len : size);
2595 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
2596 struct ptlrpc_bulk_desc *desc,
2597 struct niobuf_local *local_nb, int npages,
2598 u32 client_cksum, u32 server_cksum,
2601 struct obd_export *exp = req->rq_export;
2602 struct ost_body *body;
2606 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
2607 LASSERT(body != NULL);
2609 if (desc && req->rq_peer.nid != desc->bd_sender) {
2611 router = libcfs_nid2str(desc->bd_sender);
2614 if (exp->exp_obd->obd_checksum_dump)
2615 dump_all_bulk_pages(&body->oa, npages, local_nb, server_cksum,
2619 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
2620 client_cksum, server_cksum);
2624 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: from %s%s%s inode "
2625 DFID" object "DOSTID" extent [%llu-%llu"
2626 "]: client csum %x, server csum %x\n",
2627 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
2629 body->oa.o_valid & OBD_MD_FLFID ?
2630 body->oa.o_parent_seq : (__u64)0,
2631 body->oa.o_valid & OBD_MD_FLFID ?
2632 body->oa.o_parent_oid : 0,
2633 body->oa.o_valid & OBD_MD_FLFID ?
2634 body->oa.o_parent_ver : 0,
2635 POSTID(&body->oa.o_oi),
2636 local_nb[0].lnb_file_offset,
2637 local_nb[npages-1].lnb_file_offset +
2638 local_nb[npages - 1].lnb_len - 1,
2639 client_cksum, server_cksum);
2642 int tgt_brw_write(struct tgt_session_info *tsi)
2644 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
2645 struct ptlrpc_request *req = tgt_ses_req(tsi);
2646 struct ptlrpc_bulk_desc *desc = NULL;
2647 struct obd_export *exp = req->rq_export;
2648 struct niobuf_remote *remote_nb;
2649 struct niobuf_local *local_nb;
2650 struct obd_ioobj *ioo;
2651 struct ost_body *body, *repbody;
2652 struct lustre_handle lockh = {0};
2654 int objcount, niocount, npages;
2657 enum cksum_types cksum_type = OBD_CKSUM_CRC32;
2658 bool no_reply = false, mmap;
2659 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
2660 bool wait_sync = false;
2661 const char *obd_name = exp->exp_obd->obd_name;
2662 /* '1' for consistency with code that checks !mpflag to restore */
2663 unsigned int mpflags = 1;
2669 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL &&
2670 ptlrpc_req2svc(req)->srv_req_portal != MDS_IO_PORTAL) {
2671 CERROR("%s: deny write request from %s to portal %u\n",
2672 tgt_name(tsi->tsi_tgt),
2673 obd_export_nid2str(req->rq_export),
2674 ptlrpc_req2svc(req)->srv_req_portal);
2675 RETURN(err_serious(-EPROTO));
2678 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2679 RETURN(err_serious(-ENOSPC));
2680 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
2681 RETURN(err_serious(-EROFS));
2683 req->rq_bulk_write = 1;
2685 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
2687 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
2690 /* optionally use cfs_fail_val - 1 to select a specific OST on
2691 * this server to fail requests.
2693 char fail_ost_name[MAX_OBD_NAME];
2695 if (cfs_fail_val > 0) {
2696 snprintf(fail_ost_name, MAX_OBD_NAME, "OST%04X",
2699 if (strstr(obd_name, fail_ost_name))
2700 RETURN(err_serious(rc));
2702 RETURN(err_serious(rc));
2706 /* pause before transaction has been started */
2707 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
2708 cfs_fail_val : (obd_timeout + 1) / 4);
2710 /* Delay write commit to show stale size information */
2711 CFS_FAIL_TIMEOUT(OBD_FAIL_OSC_NO_SIZE_DATA, cfs_fail_val);
2713 /* There must be big cache in current thread to process this request
2714 * if it is NULL then something went wrong and it wasn't allocated,
2715 * report -ENOMEM in that case */
2719 body = tsi->tsi_ost_body;
2720 LASSERT(body != NULL);
2722 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2723 body->oa.o_flags & OBD_FL_NORPC)
2727 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
2728 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
2730 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
2731 RCL_CLIENT) / sizeof(*ioo);
2733 for (niocount = i = 0; i < objcount; i++)
2734 niocount += ioo[i].ioo_bufcnt;
2736 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2737 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2738 if (niocount != req_capsule_get_size(&req->rq_pill,
2739 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2741 RETURN(err_serious(-EPROTO));
2743 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2744 ptlrpc_connection_is_local(exp->exp_connection))
2745 mpflags = memalloc_noreclaim_save();
2747 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2748 niocount * sizeof(*rcs));
2749 rc = req_capsule_server_pack(&req->rq_pill);
2751 GOTO(out, rc = err_serious(rc));
2753 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2754 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2756 local_nb = tbc->local;
2758 rc = tgt_brw_lock(tsi->tsi_env, exp, &tsi->tsi_resid, ioo, remote_nb,
2764 * If getting the lock took more time than
2765 * client was willing to wait, drop it. b=11330
2767 if (ktime_get_real_seconds() > req->rq_deadline ||
2768 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2770 CERROR("%s: Dropping timed-out write from %s because locking object " DOSTID " took %lld seconds (limit was %lld).\n",
2771 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2772 POSTID(&ioo->ioo_oid),
2773 ktime_get_real_seconds() - req->rq_arrival_time.tv_sec,
2774 req->rq_deadline - req->rq_arrival_time.tv_sec);
2775 GOTO(out_lock, rc = -ETIMEDOUT);
2778 /* Because we already sync grant info with client when reconnect,
2779 * grant info will be cleared for resent req, then fed_grant and
2780 * total_grant will not be modified in following preprw_write */
2781 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2782 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2783 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2786 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2787 if (repbody == NULL)
2788 GOTO(out_lock, rc = -ENOMEM);
2789 repbody->oa = body->oa;
2791 npages = PTLRPC_MAX_BRW_PAGES;
2792 kstart = ktime_get();
2793 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2794 objcount, ioo, remote_nb, &npages, local_nb);
2797 if (body->oa.o_valid & OBD_MD_FLFLAGS &&
2798 body->oa.o_flags & OBD_FL_SHORT_IO) {
2799 unsigned int short_io_size;
2800 unsigned char *short_io_buf;
2802 short_io_size = req_capsule_get_size(&req->rq_pill,
2805 short_io_buf = req_capsule_client_get(&req->rq_pill,
2807 CDEBUG(D_INFO, "Client use short io for data transfer,"
2808 " size = %d\n", short_io_size);
2810 /* Copy short io buf to pages */
2811 rc = tgt_shortio2pages(local_nb, npages, short_io_buf,
2815 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2816 PTLRPC_BULK_GET_SINK,
2818 &ptlrpc_bulk_kiov_nopin_ops);
2820 GOTO(skip_transfer, rc = -ENOMEM);
2822 /* NB Having prepped, we must commit... */
2823 for (i = 0; i < npages; i++)
2824 desc->bd_frag_ops->add_kiov_frag(desc,
2825 local_nb[i].lnb_page,
2826 local_nb[i].lnb_page_offset & ~PAGE_MASK,
2827 local_nb[i].lnb_len);
2829 rc = sptlrpc_svc_prep_bulk(req, desc);
2831 GOTO(skip_transfer, rc);
2833 rc = target_bulk_io(exp, desc);
2839 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2840 static int cksum_counter;
2842 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2843 cksum_type = obd_cksum_type_unpack(body->oa.o_flags);
2845 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2846 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2847 repbody->oa.o_flags |= obd_cksum_type_pack(obd_name,
2850 rc = tgt_checksum_niobuf_rw(tsi->tsi_tgt, cksum_type,
2851 local_nb, npages, OST_WRITE,
2852 &repbody->oa.o_cksum, false);
2854 GOTO(out_commitrw, rc);
2858 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2859 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2860 body->oa.o_flags & OBD_FL_MMAP);
2862 tgt_warn_on_cksum(req, desc, local_nb, npages,
2864 repbody->oa.o_cksum, mmap);
2866 } else if ((cksum_counter & (-cksum_counter)) ==
2868 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2869 cksum_counter, libcfs_id2str(req->rq_peer),
2870 repbody->oa.o_cksum);
2874 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK2, cfs_fail_val);
2877 /* calculate the expected actual write bytes (nob) for OFD stats.
2878 * Technically, if commit fails this would be wrong, but that should be
2881 for (i = 0; i < niocount; i++) {
2882 int len = remote_nb[i].rnb_len;
2887 /* multiple transactions can be assigned during write commit */
2888 tti->tti_mult_trans = 1;
2890 /* Must commit after prep above in all cases */
2891 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2892 objcount, ioo, remote_nb, npages, local_nb, rc, nob,
2894 if (rc == -ENOTCONN)
2895 /* quota acquire process has been given up because
2896 * either the client has been evicted or the client
2897 * has timed out the request already
2901 for (i = 0; i < niocount; i++) {
2902 if (!(local_nb[i].lnb_flags & OBD_BRW_ASYNC)) {
2908 * Disable sending mtime back to the client. If the client locked the
2909 * whole object, then it has already updated the mtime on its side,
2910 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2912 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2915 /* set per-requested niobuf return codes */
2916 for (i = j = 0; i < niocount; i++) {
2917 int len = remote_nb[i].rnb_len;
2921 LASSERT(j < npages);
2922 if (local_nb[j].lnb_rc < 0)
2923 rcs[i] = local_nb[j].lnb_rc;
2924 len -= local_nb[j].lnb_len;
2929 LASSERT(j == npages);
2930 ptlrpc_lprocfs_brw(req, nob);
2933 tgt_brw_unlock(exp, ioo, remote_nb, &lockh, LCK_PW);
2935 ptlrpc_free_bulk(desc);
2937 if (unlikely(no_reply || (exp->exp_obd->obd_no_transno && wait_sync))) {
2938 req->rq_no_reply = 1;
2939 /* reply out callback would free */
2940 ptlrpc_req_drop_rs(req);
2941 if (!exp->exp_obd->obd_no_transno)
2942 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s),"
2943 " client will retry: rc = %d\n",
2945 obd_uuid2str(&exp->exp_client_uuid),
2946 obd_export_nid2str(exp), rc);
2950 memalloc_noreclaim_restore(mpflags);
2954 EXPORT_SYMBOL(tgt_brw_write);
2957 * Common request handler for OST_SEEK RPC.
2959 * Unified request handling for OST_SEEK RPC.
2960 * It takes object by its FID, does needed lseek and packs result
2961 * into reply. Only SEEK_HOLE and SEEK_DATA are supported.
2963 * \param[in] tsi target session environment for this request
2965 * \retval 0 if successful
2966 * \retval negative value on error
2968 int tgt_lseek(struct tgt_session_info *tsi)
2970 struct lustre_handle lh = { 0 };
2971 struct dt_object *dob;
2972 struct ost_body *repbody;
2973 loff_t offset = tsi->tsi_ost_body->oa.o_size;
2974 int whence = tsi->tsi_ost_body->oa.o_mode;
2980 if (whence != SEEK_HOLE && whence != SEEK_DATA)
2983 /* Negative offset is prohibited on wire and must be handled on client
2984 * prior sending RPC.
2989 repbody = req_capsule_server_get(tsi->tsi_pill, &RMF_OST_BODY);
2990 if (repbody == NULL)
2992 repbody->oa = tsi->tsi_ost_body->oa;
2994 srvlock = tsi->tsi_ost_body->oa.o_valid & OBD_MD_FLFLAGS &&
2995 tsi->tsi_ost_body->oa.o_flags & OBD_FL_SRVLOCK;
2997 rc = tgt_data_lock(tsi->tsi_env, tsi->tsi_exp, &tsi->tsi_resid,
2998 offset, OBD_OBJECT_EOF, &lh, LCK_PR);
3003 dob = dt_locate(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, &tsi->tsi_fid);
3005 GOTO(out, rc = PTR_ERR(dob));
3007 if (!dt_object_exists(dob))
3008 GOTO(obj_put, rc = -ENOENT);
3010 repbody->oa.o_size = dt_lseek(tsi->tsi_env, dob, offset, whence);
3013 dt_object_put(tsi->tsi_env, dob);
3016 tgt_data_unlock(&lh, LCK_PR);
3020 EXPORT_SYMBOL(tgt_lseek);
3022 /* Check if request can be reconstructed from saved reply data
3023 * A copy of the reply data is returned in @trd if the pointer is not NULL
3025 int req_can_reconstruct(struct ptlrpc_request *req,
3026 struct tg_reply_data *trd)
3028 struct tg_export_data *ted = &req->rq_export->exp_target_data;
3029 struct lsd_client_data *lcd = ted->ted_lcd;
3032 if (tgt_is_multimodrpcs_client(req->rq_export))
3033 return tgt_lookup_reply(req, trd);
3035 mutex_lock(&ted->ted_lcd_lock);
3036 found = req->rq_xid == lcd->lcd_last_xid ||
3037 req->rq_xid == lcd->lcd_last_close_xid;
3039 if (found && trd != NULL) {
3040 if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
3041 trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
3042 trd->trd_reply.lrd_transno =
3043 lcd->lcd_last_close_transno;
3044 trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
3046 trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
3047 trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
3048 trd->trd_reply.lrd_result = lcd->lcd_last_result;
3049 trd->trd_reply.lrd_data = lcd->lcd_last_data;
3050 trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
3051 trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
3052 trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
3053 trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
3056 mutex_unlock(&ted->ted_lcd_lock);
3060 EXPORT_SYMBOL(req_can_reconstruct);