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) 2011, 2012, 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
38 #include <obd_class.h>
39 #include <obd_cksum.h>
40 #include <md_object.h>
42 #include "tgt_internal.h"
44 char *tgt_name(struct lu_target *tgt)
46 LASSERT(tgt->lut_obd != NULL);
47 return tgt->lut_obd->obd_name;
49 EXPORT_SYMBOL(tgt_name);
52 * Generic code handling requests that have struct mdt_body passed in:
54 * - extract mdt_body from request and save it in @tsi, if present;
56 * - create lu_object, corresponding to the fid in mdt_body, and save it in
59 * - if HABEO_CORPUS flag is set for this request type check whether object
60 * actually exists on storage (lu_object_exists()).
63 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
65 const struct mdt_body *body;
66 struct lu_object *obj;
67 struct req_capsule *pill = tsi->tsi_pill;
72 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
76 tsi->tsi_mdt_body = body;
78 if (!(body->valid & OBD_MD_FLID))
81 /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
82 * in any case in pre-2.5 clients. Fix that here if needed */
83 if (unlikely(fid_is_zero(&body->fid1)))
86 if (!fid_is_sane(&body->fid1)) {
87 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
92 obj = lu_object_find(tsi->tsi_env,
93 &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
96 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
97 lu_object_put(tsi->tsi_env, obj);
98 /* for capability renew ENOENT will be handled in
100 if (body->valid & OBD_MD_FLOSSCAPA)
105 tsi->tsi_corpus = obj;
112 tsi->tsi_fid = body->fid1;
118 * Validate oa from client.
119 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
121 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
122 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
123 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
124 * whether this is oi_fid or real ostid. So it will check
125 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
126 * c. Old FID-disable osc will send IDIF.
127 * d. new FID-enable osc/osp will send normal FID.
129 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
130 * be used for LAST_ID file, and only being accessed inside OST now.
132 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
134 struct ost_id *oi = &oa->o_oi;
135 obd_seq seq = ostid_seq(oi);
136 obd_id id = ostid_id(oi);
140 if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
141 fid_seq_is_echo(seq))) {
142 /* Sigh 2.[123] client still sends echo req with oi_id = 0
143 * during create, and we will reset this to 1, since this
144 * oi_id is basically useless in the following create process,
145 * but oi_id == 0 will make it difficult to tell whether it is
146 * real FID or ost_id. */
147 oi->oi_fid.f_seq = FID_SEQ_ECHO;
148 oi->oi_fid.f_oid = id ?: 1;
149 oi->oi_fid.f_ver = 0;
151 struct lu_fid *fid = &tsi->tsi_fid2;
153 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
154 GOTO(out, rc = -EPROTO);
156 /* Note: this check might be forced in 2.5 or 2.6, i.e.
157 * all of the requests are required to setup FLGROUP */
158 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
159 ostid_set_seq_mdt0(oi);
160 oa->o_valid |= OBD_MD_FLGROUP;
164 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
165 fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
166 GOTO(out, rc = -EPROTO);
168 rc = ostid_to_fid(fid, oi, tsi->tsi_tgt->lut_lsd.lsd_osd_index);
169 if (unlikely(rc != 0))
178 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
179 tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
183 EXPORT_SYMBOL(tgt_validate_obdo);
185 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
188 struct niobuf_remote *rnb;
189 struct obd_ioobj *ioo;
194 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
198 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
202 max_brw = ioobj_max_brw_get(ioo);
203 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
204 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
205 ": rc = %d\n", tgt_name(tsi->tsi_tgt),
206 obd_export_nid2str(tsi->tsi_exp), max_brw,
207 POSTID(oi), -EPROTO);
212 obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
213 RCL_CLIENT) / sizeof(*ioo);
214 if (obj_count == 0) {
215 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
217 } else if (obj_count > 1) {
218 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
223 if (ioo->ioo_bufcnt == 0) {
224 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
228 if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
229 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
230 "bulk has too many pages (%d)",
238 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
240 struct ost_body *body;
241 struct req_capsule *pill = tsi->tsi_pill;
242 struct lustre_capa *capa;
247 body = req_capsule_client_get(pill, &RMF_OST_BODY);
251 rc = tgt_validate_obdo(tsi, &body->oa);
255 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
256 capa = req_capsule_client_get(pill, &RMF_CAPA1);
258 CERROR("%s: OSSCAPA flag is set without capability\n",
259 tgt_name(tsi->tsi_tgt));
264 tsi->tsi_ost_body = body;
265 tsi->tsi_fid = body->oa.o_oi.oi_fid;
267 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
268 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
273 if (!(body->oa.o_valid & OBD_MD_FLID)) {
274 if (flags & HABEO_CORPUS) {
275 CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
276 "but OID/FID is mandatory with HABEO_CORPUS\n",
277 tgt_name(tsi->tsi_tgt));
284 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
287 * OST doesn't get object in advance for further use to prevent
288 * situations with nested object_find which is potential deadlock.
290 tsi->tsi_corpus = NULL;
295 * Do necessary preprocessing according to handler ->th_flags.
297 static int tgt_request_preprocess(struct tgt_session_info *tsi,
298 struct tgt_handler *h,
299 struct ptlrpc_request *req)
301 struct req_capsule *pill = tsi->tsi_pill;
302 __u32 flags = h->th_flags;
307 if (tsi->tsi_preprocessed)
310 LASSERT(h->th_act != NULL);
311 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
312 LASSERT(current->journal_info == NULL);
314 LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
316 if (h->th_fmt != NULL) {
317 req_capsule_set(pill, h->th_fmt);
318 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
319 rc = tgt_mdt_body_unpack(tsi, flags);
322 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
324 rc = tgt_ost_body_unpack(tsi, flags);
330 if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
333 if (flags & HABEO_CLAVIS) {
334 struct ldlm_request *dlm_req;
336 LASSERT(h->th_fmt != NULL);
338 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
339 if (dlm_req != NULL) {
340 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
342 dlm_req->lock_desc.l_policy_data.\
343 l_inodebits.bits == 0)) {
345 * Lock without inodebits makes no sense and
346 * will oops later in ldlm. If client miss to
347 * set such bits, do not trigger ASSERTION.
349 * For liblustre flock case, it maybe zero.
353 tsi->tsi_dlm_req = dlm_req;
359 tsi->tsi_preprocessed = 1;
364 * Invoke handler for this request opc. Also do necessary preprocessing
365 * (according to handler ->th_flags), and post-processing (setting of
366 * ->last_{xid,committed}).
368 static int tgt_handle_request0(struct tgt_session_info *tsi,
369 struct tgt_handler *h,
370 struct ptlrpc_request *req)
378 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
379 * to put same checks into handlers like mdt_close(), mdt_reint(),
380 * etc., without talking to mdt authors first. Checking same thing
381 * there again is useless and returning 0 error without packing reply
382 * is buggy! Handlers either pack reply or return error.
384 * We return 0 here and do not send any reply in order to emulate
385 * network failure. Do not send any reply in case any of NET related
386 * fail_id has occured.
388 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
391 rc = tgt_request_preprocess(tsi, h, req);
392 /* pack reply if reply format is fixed */
393 if (rc == 0 && h->th_flags & HABEO_REFERO) {
395 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
397 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
399 tsi->tsi_mdt_body->eadatasize);
400 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
402 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
405 rc = req_capsule_server_pack(tsi->tsi_pill);
408 if (likely(rc == 0)) {
410 * Process request, there can be two types of rc:
411 * 1) errors with msg unpack/pack, other failures outside the
412 * operation itself. This is counted as serious errors;
413 * 2) errors during fs operation, should be placed in rq_status
417 if (!is_serious(rc) &&
418 !req->rq_no_reply && req->rq_reply_state == NULL) {
419 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
420 "pack reply and returned 0 error\n",
421 tgt_name(tsi->tsi_tgt), h->th_name);
424 serious = is_serious(rc);
425 rc = clear_serious(rc);
433 * ELDLM_* codes which > 0 should be in rq_status only as well as
434 * all non-serious errors.
436 if (rc > 0 || !serious)
439 LASSERT(current->journal_info == NULL);
441 if (likely(rc == 0 && req->rq_export))
442 target_committed_to_req(req);
444 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
448 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
449 struct obd_device *obd, int *process)
451 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
458 case MDS_DONE_WRITING:
459 case MDS_SYNC: /* used in unmounting */
472 *process = target_queue_recovery_request(req, obd);
476 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
483 * Handle recovery. Return:
484 * +1: continue request processing;
485 * -ve: abort immediately with the given error code;
486 * 0: send reply with error code in req->rq_status;
488 int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
492 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
497 case SEC_CTX_INIT_CONT:
502 if (!req->rq_export->exp_obd->obd_replayable)
505 /* sanity check: if the xid matches, the request must be marked as a
506 * resent or replayed */
507 if (req_xid_is_last(req)) {
508 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
509 (MSG_RESENT | MSG_REPLAY))) {
510 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
511 "last_xid, expected REPLAY or RESENT flag "
513 lustre_msg_get_flags(req->rq_reqmsg));
514 req->rq_status = -ENOTCONN;
518 /* else: note the opposite is not always true; a RESENT req after a
519 * failover will usually not match the last_xid, since it was likely
520 * never committed. A REPLAYed request will almost never match the
521 * last xid, however it could for a committed, but still retained,
524 /* Check for aborted recovery... */
525 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
529 DEBUG_REQ(D_INFO, req, "Got new replay");
530 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
532 if (rc != 0 || !should_process)
534 else if (should_process < 0) {
535 req->rq_status = should_process;
536 rc = ptlrpc_error(req);
543 /* Initial check for request, it is validation mostly */
544 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
546 struct tgt_handler *h;
547 struct tgt_opc_slice *s;
548 struct lu_target *tgt;
549 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
553 tgt = class_exp2tgt(req->rq_export);
555 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
556 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
559 /* opcode was not found in slice */
560 if (unlikely(s->tos_hs == NULL)) {
561 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
563 RETURN(ERR_PTR(-ENOTSUPP));
566 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
567 h = s->tos_hs + (opc - s->tos_opc_start);
568 if (unlikely(h->th_opc == 0)) {
569 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
570 RETURN(ERR_PTR(-ENOTSUPP));
576 int tgt_request_handle(struct ptlrpc_request *req)
578 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
580 struct lustre_msg *msg = req->rq_reqmsg;
581 struct tgt_handler *h;
582 struct lu_target *tgt;
583 int request_fail_id = 0;
584 __u32 opc = lustre_msg_get_opc(msg);
589 /* Refill the context, to make sure all thread keys are allocated */
590 lu_env_refill(req->rq_svc_thread->t_env);
592 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
593 tsi->tsi_pill = &req->rq_pill;
594 tsi->tsi_env = req->rq_svc_thread->t_env;
596 /* if request has export then get handlers slice from corresponding
597 * target, otherwise that should be connect operation */
598 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
599 opc == MGS_CONNECT) {
600 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
601 rc = target_handle_connect(req);
603 rc = ptlrpc_error(req);
606 /* recovery-small test 18c asks to drop connect reply */
607 if (unlikely(opc == OST_CONNECT &&
608 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
612 if (unlikely(!class_connected_export(req->rq_export))) {
613 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
614 opc, libcfs_id2str(req->rq_peer));
615 req->rq_status = -ENOTCONN;
616 rc = ptlrpc_error(req);
620 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
621 tsi->tsi_exp = req->rq_export;
622 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
623 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
625 tsi->tsi_jobid = NULL;
627 request_fail_id = tgt->lut_request_fail_id;
628 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
630 h = tgt_handler_find_check(req);
632 req->rq_status = PTR_ERR(h);
633 rc = ptlrpc_error(req);
637 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
640 rc = lustre_msg_check_version(msg, h->th_version);
642 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
643 " %08x, expecting %08x\n", tgt_name(tgt),
644 lustre_msg_get_version(msg), h->th_version);
645 req->rq_status = -EINVAL;
646 rc = ptlrpc_error(req);
650 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
651 if (likely(rc == 1)) {
652 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
654 rc = tgt_handle_request0(tsi, h, req);
660 req_capsule_fini(tsi->tsi_pill);
661 if (tsi->tsi_corpus != NULL) {
662 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
663 tsi->tsi_corpus = NULL;
667 EXPORT_SYMBOL(tgt_request_handle);
669 /** Assign high priority operations to the request if needed. */
670 int tgt_hpreq_handler(struct ptlrpc_request *req)
672 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
673 struct tgt_handler *h;
678 if (req->rq_export == NULL)
681 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
682 tsi->tsi_pill = &req->rq_pill;
683 tsi->tsi_env = req->rq_svc_thread->t_env;
684 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
685 tsi->tsi_exp = req->rq_export;
687 h = tgt_handler_find_check(req);
693 rc = tgt_request_preprocess(tsi, h, req);
694 if (unlikely(rc != 0))
697 if (h->th_hp != NULL)
701 EXPORT_SYMBOL(tgt_hpreq_handler);
703 void tgt_counter_incr(struct obd_export *exp, int opcode)
705 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
706 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
707 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
709 EXPORT_SYMBOL(tgt_counter_incr);
712 * Unified target generic handlers.
718 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
720 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
721 OBD_CONNECT_RMT_CLIENT_FORCE |
722 OBD_CONNECT_MDS_CAPA |
723 OBD_CONNECT_OSS_CAPA);
726 static int tgt_init_sec_level(struct ptlrpc_request *req)
728 struct lu_target *tgt = class_exp2tgt(req->rq_export);
729 char *client = libcfs_nid2str(req->rq_peer.nid);
730 struct obd_connect_data *data, *reply;
736 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
737 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
738 if (data == NULL || reply == NULL)
741 /* connection from MDT is always trusted */
742 if (req->rq_auth_usr_mdt) {
743 tgt_init_sec_none(reply);
747 /* no GSS support case */
748 if (!req->rq_auth_gss) {
749 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
750 CWARN("client %s -> target %s does not use GSS, "
751 "can not run under security level %d.\n",
752 client, tgt_name(tgt), tgt->lut_sec_level);
755 tgt_init_sec_none(reply);
760 /* old version case */
761 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
762 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
763 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
764 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
765 CWARN("client %s -> target %s uses old version, "
766 "can not run under security level %d.\n",
767 client, tgt_name(tgt), tgt->lut_sec_level);
770 CWARN("client %s -> target %s uses old version, "
771 "run under security level %d.\n",
772 client, tgt_name(tgt), tgt->lut_sec_level);
773 tgt_init_sec_none(reply);
778 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
780 if (!req->rq_auth_remote)
781 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
782 "asked to be remote.\n", client, tgt_name(tgt));
783 } else if (req->rq_auth_remote) {
785 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
786 "as remote by default.\n", client, tgt_name(tgt));
790 if (!tgt->lut_oss_capa) {
792 "client %s -> target %s is set as remote,"
793 " but OSS capabilities are not enabled: %d.\n",
794 client, tgt_name(tgt), tgt->lut_oss_capa);
798 if (req->rq_auth_uid == INVALID_UID) {
799 CDEBUG(D_SEC, "client %s -> target %s: user is not "
800 "authenticated!\n", client, tgt_name(tgt));
806 switch (tgt->lut_sec_level) {
807 case LUSTRE_SEC_NONE:
810 "client %s -> target %s is set as remote, "
811 "can not run under security level %d.\n",
812 client, tgt_name(tgt), tgt->lut_sec_level);
815 tgt_init_sec_none(reply);
817 case LUSTRE_SEC_REMOTE:
819 tgt_init_sec_none(reply);
824 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
825 OBD_CONNECT_RMT_CLIENT_FORCE);
826 if (!tgt->lut_oss_capa)
827 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
828 if (!tgt->lut_mds_capa)
829 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
838 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
840 struct lu_target *tgt = class_exp2tgt(exp);
841 struct sptlrpc_flavor flvr;
845 LASSERT(tgt->lut_obd);
846 LASSERT(tgt->lut_slice);
848 /* always allow ECHO client */
849 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
850 LUSTRE_ECHO_NAME) == 0)) {
851 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
855 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
856 read_lock(&tgt->lut_sptlrpc_lock);
857 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
861 read_unlock(&tgt->lut_sptlrpc_lock);
863 spin_lock(&exp->exp_lock);
864 exp->exp_sp_peer = req->rq_sp_from;
865 exp->exp_flvr = flvr;
866 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
867 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
868 CERROR("%s: unauthorized rpc flavor %x from %s, "
869 "expect %x\n", tgt_name(tgt),
871 libcfs_nid2str(req->rq_peer.nid),
872 exp->exp_flvr.sf_rpc);
875 spin_unlock(&exp->exp_lock);
877 if (exp->exp_sp_peer != req->rq_sp_from) {
878 CERROR("%s: RPC source %s doesn't match %s\n",
880 sptlrpc_part2name(req->rq_sp_from),
881 sptlrpc_part2name(exp->exp_sp_peer));
884 rc = sptlrpc_target_export_check(exp, req);
891 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
893 struct sptlrpc_rule_set tmp_rset;
896 sptlrpc_rule_set_init(&tmp_rset);
897 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
899 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
904 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
906 write_lock(&tgt->lut_sptlrpc_lock);
907 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
908 tgt->lut_sptlrpc_rset = tmp_rset;
909 write_unlock(&tgt->lut_sptlrpc_lock);
913 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
915 int tgt_connect(struct tgt_session_info *tsi)
917 struct ptlrpc_request *req = tgt_ses_req(tsi);
918 struct obd_connect_data *reply;
923 rc = tgt_init_sec_level(req);
927 /* XXX: better to call this check right after getting new export but
928 * before last_rcvd slot allocation to avoid server load upon insecure
929 * connects. This is to be fixed after unifiyng all targets.
931 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
935 /* To avoid exposing partially initialized connection flags, changes up
936 * to this point have been staged in reply->ocd_connect_flags. Now that
937 * connection handling has completed successfully, atomically update
938 * the connect flags in the shared export data structure. LU-1623 */
939 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
940 spin_lock(&tsi->tsi_exp->exp_lock);
941 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
942 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
943 spin_unlock(&tsi->tsi_exp->exp_lock);
947 obd_disconnect(class_export_get(tsi->tsi_exp));
950 EXPORT_SYMBOL(tgt_connect);
952 int tgt_disconnect(struct tgt_session_info *tsi)
958 rc = target_handle_disconnect(tgt_ses_req(tsi));
960 RETURN(err_serious(rc));
964 EXPORT_SYMBOL(tgt_disconnect);
967 * Unified target OBD handlers
969 int tgt_obd_ping(struct tgt_session_info *tsi)
975 rc = target_handle_ping(tgt_ses_req(tsi));
977 RETURN(err_serious(rc));
981 EXPORT_SYMBOL(tgt_obd_ping);
983 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
985 return err_serious(-EOPNOTSUPP);
987 EXPORT_SYMBOL(tgt_obd_log_cancel);
989 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
991 return err_serious(-EOPNOTSUPP);
993 EXPORT_SYMBOL(tgt_obd_qc_callback);
995 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
997 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
998 struct ptlrpc_request *req = tgt_ses_req(tsi);
999 struct obd_export *exp = req->rq_export;
1000 struct ptlrpc_bulk_desc *desc;
1001 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1009 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1014 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1015 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1016 * which is rdpg->rp_count */
1017 nob = rdpg->rp_count;
1019 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1020 i++, tmpcount -= tmpsize) {
1021 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1022 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1025 LASSERT(desc->bd_nob == nob);
1026 rc = target_bulk_io(exp, desc, lwi);
1027 ptlrpc_free_bulk_pin(desc);
1030 EXPORT_SYMBOL(tgt_sendpage);
1033 * OBD_IDX_READ handler
1035 int tgt_obd_idx_read(struct tgt_session_info *tsi)
1037 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1038 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1039 struct idx_info *req_ii, *rep_ii;
1044 memset(rdpg, 0, sizeof(*rdpg));
1045 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1047 /* extract idx_info buffer from request & reply */
1048 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1049 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1050 RETURN(err_serious(-EPROTO));
1052 rc = req_capsule_server_pack(tsi->tsi_pill);
1054 RETURN(err_serious(rc));
1056 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1058 RETURN(err_serious(-EFAULT));
1059 rep_ii->ii_magic = IDX_INFO_MAGIC;
1061 /* extract hash to start with */
1062 rdpg->rp_hash = req_ii->ii_hash_start;
1064 /* extract requested attributes */
1065 rdpg->rp_attrs = req_ii->ii_attrs;
1067 /* check that fid packed in request is valid and supported */
1068 if (!fid_is_sane(&req_ii->ii_fid))
1070 rep_ii->ii_fid = req_ii->ii_fid;
1073 rep_ii->ii_flags = req_ii->ii_flags;
1075 /* compute number of pages to allocate, ii_count is the number of 4KB
1077 if (req_ii->ii_count <= 0)
1078 GOTO(out, rc = -EFAULT);
1079 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1080 exp_max_brw_size(tsi->tsi_exp));
1081 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1083 /* allocate pages to store the containers */
1084 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1085 if (rdpg->rp_pages == NULL)
1086 GOTO(out, rc = -ENOMEM);
1087 for (i = 0; i < rdpg->rp_npages; i++) {
1088 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1089 if (rdpg->rp_pages[i] == NULL)
1090 GOTO(out, rc = -ENOMEM);
1093 /* populate pages with key/record pairs */
1094 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1098 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1099 "asked %d > %d\n", rc, rdpg->rp_count);
1101 /* send pages to client */
1102 rc = tgt_sendpage(tsi, rdpg, rc);
1107 if (rdpg->rp_pages) {
1108 for (i = 0; i < rdpg->rp_npages; i++)
1109 if (rdpg->rp_pages[i])
1110 __free_page(rdpg->rp_pages[i]);
1111 OBD_FREE(rdpg->rp_pages,
1112 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1116 EXPORT_SYMBOL(tgt_obd_idx_read);
1118 struct tgt_handler tgt_obd_handlers[] = {
1119 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1120 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1121 TGT_OBD_HDL_VAR(0, OBD_QC_CALLBACK, tgt_obd_qc_callback),
1122 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1124 EXPORT_SYMBOL(tgt_obd_handlers);
1126 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1127 struct dt_object *obj)
1133 /* if no objid is specified, it means "sync whole filesystem" */
1135 rc = dt_sync(env, tgt->lut_bottom);
1136 } else if (dt_version_get(env, obj) >
1137 tgt->lut_obd->obd_last_committed) {
1138 rc = dt_object_sync(env, obj);
1143 EXPORT_SYMBOL(tgt_sync);
1145 * Unified target DLM handlers.
1148 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1150 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1151 void *data, int flag)
1154 struct lu_target *tgt;
1155 struct dt_object *obj;
1161 tgt = class_exp2tgt(lock->l_export);
1163 if (flag == LDLM_CB_CANCELING &&
1164 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1165 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1166 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1167 lock->l_flags & LDLM_FL_CBPENDING))) {
1168 rc = lu_env_init(&env, LCT_DT_THREAD);
1169 if (unlikely(rc != 0))
1172 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1173 tgt->lut_lsd.lsd_osd_index);
1174 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1176 GOTO(err_env, rc = PTR_ERR(obj));
1178 if (!dt_object_exists(obj))
1179 GOTO(err_put, rc = -ENOENT);
1181 rc = tgt_sync(&env, tgt, obj);
1183 CERROR("%s: sync failed on lock cancel: rc = %d\n",
1187 lu_object_put(&env, &obj->do_lu);
1192 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1196 struct ldlm_callback_suite tgt_dlm_cbs = {
1197 .lcs_completion = ldlm_server_completion_ast,
1198 .lcs_blocking = tgt_blocking_ast,
1199 .lcs_glimpse = ldlm_server_glimpse_ast
1202 int tgt_enqueue(struct tgt_session_info *tsi)
1204 struct ptlrpc_request *req = tgt_ses_req(tsi);
1209 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1210 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1212 LASSERT(tsi->tsi_dlm_req != NULL);
1213 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1214 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1216 RETURN(err_serious(rc));
1218 RETURN(req->rq_status);
1220 EXPORT_SYMBOL(tgt_enqueue);
1222 int tgt_convert(struct tgt_session_info *tsi)
1224 struct ptlrpc_request *req = tgt_ses_req(tsi);
1228 LASSERT(tsi->tsi_dlm_req);
1229 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1231 RETURN(err_serious(rc));
1233 RETURN(req->rq_status);
1235 EXPORT_SYMBOL(tgt_convert);
1237 int tgt_bl_callback(struct tgt_session_info *tsi)
1239 return err_serious(-EOPNOTSUPP);
1241 EXPORT_SYMBOL(tgt_bl_callback);
1243 int tgt_cp_callback(struct tgt_session_info *tsi)
1245 return err_serious(-EOPNOTSUPP);
1247 EXPORT_SYMBOL(tgt_cp_callback);
1249 /* generic LDLM target handler */
1250 struct tgt_handler tgt_dlm_handlers[] = {
1251 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1252 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1253 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1254 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1256 EXPORT_SYMBOL(tgt_dlm_handlers);
1259 * Unified target LLOG handlers.
1261 int tgt_llog_open(struct tgt_session_info *tsi)
1267 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1271 EXPORT_SYMBOL(tgt_llog_open);
1273 int tgt_llog_close(struct tgt_session_info *tsi)
1279 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1283 EXPORT_SYMBOL(tgt_llog_close);
1286 int tgt_llog_destroy(struct tgt_session_info *tsi)
1292 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1296 EXPORT_SYMBOL(tgt_llog_destroy);
1298 int tgt_llog_read_header(struct tgt_session_info *tsi)
1304 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1308 EXPORT_SYMBOL(tgt_llog_read_header);
1310 int tgt_llog_next_block(struct tgt_session_info *tsi)
1316 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1320 EXPORT_SYMBOL(tgt_llog_next_block);
1322 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1328 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1332 EXPORT_SYMBOL(tgt_llog_prev_block);
1334 /* generic llog target handler */
1335 struct tgt_handler tgt_llog_handlers[] = {
1336 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1337 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1338 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1339 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1340 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1341 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1343 EXPORT_SYMBOL(tgt_llog_handlers);
1346 * sec context handlers
1348 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1349 int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1354 struct tgt_handler tgt_sec_ctx_handlers[] = {
1355 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1356 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1357 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1359 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1362 * initialize per-thread page pool (bug 5137).
1364 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1366 struct tgt_thread_big_cache *tbc;
1370 LASSERT(thread != NULL);
1371 LASSERT(thread->t_data == NULL);
1373 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1376 thread->t_data = tbc;
1379 EXPORT_SYMBOL(tgt_io_thread_init);
1382 * free per-thread pool created by tgt_thread_init().
1384 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1386 struct tgt_thread_big_cache *tbc;
1390 LASSERT(thread != NULL);
1393 * be prepared to handle partially-initialized pools (because this is
1394 * called from ost_io_thread_init() for cleanup.
1396 tbc = thread->t_data;
1398 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1399 thread->t_data = NULL;
1403 EXPORT_SYMBOL(tgt_io_thread_done);
1405 * Helper function for getting server side [start, start+count] DLM lock
1406 * if asked by client.
1408 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1409 __u64 start, __u64 end, struct lustre_handle *lh,
1410 int mode, __u64 *flags)
1412 ldlm_policy_data_t policy;
1417 LASSERT(lh != NULL);
1418 LASSERT(ns != NULL);
1419 LASSERT(!lustre_handle_is_used(lh));
1421 policy.l_extent.gid = 0;
1422 policy.l_extent.start = start & CFS_PAGE_MASK;
1425 * If ->o_blocks is EOF it means "lock till the end of the file".
1426 * Otherwise, it's size of an extent or hole being punched (in bytes).
1428 if (end == OBD_OBJECT_EOF || end < start)
1429 policy.l_extent.end = OBD_OBJECT_EOF;
1431 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1433 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1434 flags, ldlm_blocking_ast,
1435 ldlm_completion_ast, ldlm_glimpse_ast,
1436 NULL, 0, LVB_T_NONE, NULL, lh);
1437 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1439 EXPORT_SYMBOL(tgt_extent_lock);
1441 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1443 LASSERT(lustre_handle_is_used(lh));
1444 ldlm_lock_decref(lh, mode);
1446 EXPORT_SYMBOL(tgt_extent_unlock);
1448 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1449 struct obd_ioobj *obj, struct niobuf_remote *nb,
1450 struct lustre_handle *lh, int mode)
1453 int nrbufs = obj->ioo_bufcnt;
1458 LASSERT(mode == LCK_PR || mode == LCK_PW);
1459 LASSERT(!lustre_handle_is_used(lh));
1461 if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1464 for (i = 1; i < nrbufs; i++)
1465 if (!(nb[i].flags & OBD_BRW_SRVLOCK))
1468 RETURN(tgt_extent_lock(ns, res_id, nb[0].offset,
1469 nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1,
1472 EXPORT_SYMBOL(tgt_brw_lock);
1474 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1475 struct lustre_handle *lh, int mode)
1479 LASSERT(mode == LCK_PR || mode == LCK_PW);
1480 LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
1481 lustre_handle_is_used(lh));
1482 if (lustre_handle_is_used(lh))
1483 tgt_extent_unlock(lh, mode);
1486 EXPORT_SYMBOL(tgt_brw_unlock);
1488 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1489 struct ptlrpc_bulk_desc *desc, int opc,
1490 cksum_type_t cksum_type)
1492 struct cfs_crypto_hash_desc *hdesc;
1493 unsigned int bufsize;
1495 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1498 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1499 if (IS_ERR(hdesc)) {
1500 CERROR("%s: unable to initialize checksum hash %s\n",
1501 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1502 return PTR_ERR(hdesc);
1505 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1506 for (i = 0; i < desc->bd_iov_count; i++) {
1507 /* corrupt the data before we compute the checksum, to
1508 * simulate a client->OST data error */
1509 if (i == 0 && opc == OST_WRITE &&
1510 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1511 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1512 int len = desc->bd_iov[i].kiov_len;
1513 struct page *np = tgt_page_to_corrupt;
1514 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1517 char *ptr2 = kmap(np) + off;
1519 memcpy(ptr2, ptr, len);
1520 memcpy(ptr2, "bad3", min(4, len));
1522 desc->bd_iov[i].kiov_page = np;
1524 CERROR("%s: can't alloc page for corruption\n",
1528 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1529 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1530 desc->bd_iov[i].kiov_len);
1532 /* corrupt the data after we compute the checksum, to
1533 * simulate an OST->client data error */
1534 if (i == 0 && opc == OST_READ &&
1535 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1536 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1537 int len = desc->bd_iov[i].kiov_len;
1538 struct page *np = tgt_page_to_corrupt;
1539 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1542 char *ptr2 = kmap(np) + off;
1544 memcpy(ptr2, ptr, len);
1545 memcpy(ptr2, "bad4", min(4, len));
1547 desc->bd_iov[i].kiov_page = np;
1549 CERROR("%s: can't alloc page for corruption\n",
1556 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1558 cfs_crypto_hash_final(hdesc, NULL, NULL);
1563 int tgt_brw_read(struct tgt_session_info *tsi)
1565 struct ptlrpc_request *req = tgt_ses_req(tsi);
1566 struct ptlrpc_bulk_desc *desc = NULL;
1567 struct obd_export *exp = tsi->tsi_exp;
1568 struct niobuf_remote *remote_nb;
1569 struct niobuf_local *local_nb;
1570 struct obd_ioobj *ioo;
1571 struct ost_body *body, *repbody;
1572 struct l_wait_info lwi;
1573 struct lustre_handle lockh = { 0 };
1574 int niocount, npages, nob = 0, rc, i;
1576 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1580 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1581 CERROR("%s: deny read request from %s to portal %u\n",
1582 tgt_name(tsi->tsi_tgt),
1583 obd_export_nid2str(req->rq_export),
1584 ptlrpc_req2svc(req)->srv_req_portal);
1588 req->rq_bulk_read = 1;
1590 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1593 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1594 cfs_fail_val : (obd_timeout + 1) / 4);
1596 /* Check if there is eviction in progress, and if so, wait for it to
1598 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1599 /* We do not care how long it takes */
1600 lwi = LWI_INTR(NULL, NULL);
1601 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1602 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1606 /* There must be big cache in current thread to process this request
1607 * if it is NULL then something went wrong and it wasn't allocated,
1608 * report -ENOMEM in that case */
1612 body = tsi->tsi_ost_body;
1613 LASSERT(body != NULL);
1615 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1616 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1618 niocount = ioo->ioo_bufcnt;
1619 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1620 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1622 local_nb = tbc->local;
1624 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1625 remote_nb, &lockh, LCK_PR);
1630 * If getting the lock took more time than
1631 * client was willing to wait, drop it. b=11330
1633 if (cfs_time_current_sec() > req->rq_deadline ||
1634 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1636 CERROR("Dropping timed-out read from %s because locking"
1637 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1638 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1639 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1640 req->rq_deadline - req->rq_arrival_time.tv_sec);
1641 GOTO(out_lock, rc = -ETIMEDOUT);
1644 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1645 repbody->oa = body->oa;
1647 npages = PTLRPC_MAX_BRW_PAGES;
1648 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1649 ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1653 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1654 BULK_PUT_SOURCE, OST_BULK_PORTAL);
1656 GOTO(out_commitrw, rc = -ENOMEM);
1659 for (i = 0; i < npages; i++) {
1660 int page_rc = local_nb[i].rc;
1668 if (page_rc != 0) { /* some data! */
1669 LASSERT(local_nb[i].page != NULL);
1670 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1671 local_nb[i].lnb_page_offset,
1675 if (page_rc != local_nb[i].len) { /* short read */
1676 /* All subsequent pages should be 0 */
1677 while (++i < npages)
1678 LASSERT(local_nb[i].rc == 0);
1683 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1684 cksum_type_t cksum_type =
1685 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1686 body->oa.o_flags : 0);
1687 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1688 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1689 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1690 OST_READ, cksum_type);
1691 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1692 repbody->oa.o_cksum);
1694 repbody->oa.o_valid = 0;
1696 /* We're finishing using body->oa as an input variable */
1698 /* Check if client was evicted while we were doing i/o before touching
1700 if (likely(rc == 0 &&
1701 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1702 rc = target_bulk_io(exp, desc, &lwi);
1707 /* Must commit after prep above in all cases */
1708 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1709 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1712 tgt_drop_id(exp, &repbody->oa);
1714 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1716 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1717 ptlrpc_free_bulk_nopin(desc);
1722 ptlrpc_lprocfs_brw(req, nob);
1723 } else if (no_reply) {
1724 req->rq_no_reply = 1;
1725 /* reply out callback would free */
1726 ptlrpc_req_drop_rs(req);
1727 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1728 "client will retry: rc %d\n",
1729 exp->exp_obd->obd_name,
1730 obd_uuid2str(&exp->exp_client_uuid),
1731 obd_export_nid2str(exp), rc);
1733 /* send a bulk after reply to simulate a network delay or reordering
1735 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1736 wait_queue_head_t waitq;
1737 struct l_wait_info lwi1;
1739 CDEBUG(D_INFO, "reorder BULK\n");
1740 init_waitqueue_head(&waitq);
1742 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1743 l_wait_event(waitq, 0, &lwi1);
1744 target_bulk_io(exp, desc, &lwi);
1745 ptlrpc_free_bulk_nopin(desc);
1750 EXPORT_SYMBOL(tgt_brw_read);
1752 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1753 struct ptlrpc_bulk_desc *desc,
1754 struct niobuf_local *local_nb, int npages,
1755 obd_count client_cksum, obd_count server_cksum,
1758 struct obd_export *exp = req->rq_export;
1759 struct ost_body *body;
1763 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1764 LASSERT(body != NULL);
1766 if (req->rq_peer.nid == desc->bd_sender) {
1770 router = libcfs_nid2str(desc->bd_sender);
1774 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1775 client_cksum, server_cksum);
1779 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1780 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1781 "]: client csum %x, server csum %x\n",
1782 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1784 body->oa.o_valid & OBD_MD_FLFID ?
1785 body->oa.o_parent_seq : (__u64)0,
1786 body->oa.o_valid & OBD_MD_FLFID ?
1787 body->oa.o_parent_oid : 0,
1788 body->oa.o_valid & OBD_MD_FLFID ?
1789 body->oa.o_parent_ver : 0,
1790 POSTID(&body->oa.o_oi),
1791 local_nb[0].lnb_file_offset,
1792 local_nb[npages-1].lnb_file_offset +
1793 local_nb[npages-1].len - 1,
1794 client_cksum, server_cksum);
1797 int tgt_brw_write(struct tgt_session_info *tsi)
1799 struct ptlrpc_request *req = tgt_ses_req(tsi);
1800 struct ptlrpc_bulk_desc *desc = NULL;
1801 struct obd_export *exp = req->rq_export;
1802 struct niobuf_remote *remote_nb;
1803 struct niobuf_local *local_nb;
1804 struct obd_ioobj *ioo;
1805 struct ost_body *body, *repbody;
1806 struct l_wait_info lwi;
1807 struct lustre_handle lockh = {0};
1809 int objcount, niocount, npages;
1811 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1812 bool no_reply = false, mmap;
1813 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1817 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1818 CERROR("%s: deny write request from %s to portal %u\n",
1819 tgt_name(tsi->tsi_tgt),
1820 obd_export_nid2str(req->rq_export),
1821 ptlrpc_req2svc(req)->srv_req_portal);
1822 RETURN(err_serious(-EPROTO));
1825 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1826 RETURN(err_serious(-ENOSPC));
1827 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1828 RETURN(err_serious(-EROFS));
1830 req->rq_bulk_write = 1;
1832 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1833 RETURN(err_serious(-EIO));
1834 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1835 RETURN(err_serious(-EFAULT));
1837 /* pause before transaction has been started */
1838 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1839 cfs_fail_val : (obd_timeout + 1) / 4);
1841 /* There must be big cache in current thread to process this request
1842 * if it is NULL then something went wrong and it wasn't allocated,
1843 * report -ENOMEM in that case */
1847 body = tsi->tsi_ost_body;
1848 LASSERT(body != NULL);
1850 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1851 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1853 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1854 RCL_CLIENT) / sizeof(*ioo);
1856 for (niocount = i = 0; i < objcount; i++)
1857 niocount += ioo[i].ioo_bufcnt;
1859 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1860 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1861 if (niocount != req_capsule_get_size(&req->rq_pill,
1862 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1864 RETURN(err_serious(-EPROTO));
1866 if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1867 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1868 memory_pressure_set();
1870 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1871 niocount * sizeof(*rcs));
1872 rc = req_capsule_server_pack(&req->rq_pill);
1874 GOTO(out, rc = err_serious(rc));
1876 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1877 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1879 local_nb = tbc->local;
1881 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1882 remote_nb, &lockh, LCK_PW);
1887 * If getting the lock took more time than
1888 * client was willing to wait, drop it. b=11330
1890 if (cfs_time_current_sec() > req->rq_deadline ||
1891 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1893 CERROR("%s: Dropping timed-out write from %s because locking "
1894 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1895 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1896 POSTID(&ioo->ioo_oid),
1897 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1898 req->rq_deadline - req->rq_arrival_time.tv_sec);
1899 GOTO(out_lock, rc = -ETIMEDOUT);
1902 /* Because we already sync grant info with client when reconnect,
1903 * grant info will be cleared for resent req, then fed_grant and
1904 * total_grant will not be modified in following preprw_write */
1905 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1906 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1907 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1910 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1911 if (repbody == NULL)
1912 GOTO(out_lock, rc = -ENOMEM);
1913 repbody->oa = body->oa;
1915 npages = PTLRPC_MAX_BRW_PAGES;
1916 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1917 objcount, ioo, remote_nb, &npages, local_nb, NULL,
1922 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1923 BULK_GET_SINK, OST_BULK_PORTAL);
1925 GOTO(skip_transfer, rc = -ENOMEM);
1927 /* NB Having prepped, we must commit... */
1928 for (i = 0; i < npages; i++)
1929 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1930 local_nb[i].lnb_page_offset,
1933 rc = sptlrpc_svc_prep_bulk(req, desc);
1935 GOTO(skip_transfer, rc);
1937 rc = target_bulk_io(exp, desc, &lwi);
1941 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
1942 static int cksum_counter;
1944 if (body->oa.o_valid & OBD_MD_FLFLAGS)
1945 cksum_type = cksum_type_unpack(body->oa.o_flags);
1947 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1948 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
1949 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
1950 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1951 OST_WRITE, cksum_type);
1954 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
1955 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
1956 body->oa.o_flags & OBD_FL_MMAP);
1958 tgt_warn_on_cksum(req, desc, local_nb, npages,
1960 repbody->oa.o_cksum, mmap);
1962 } else if ((cksum_counter & (-cksum_counter)) ==
1964 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1965 cksum_counter, libcfs_id2str(req->rq_peer),
1966 repbody->oa.o_cksum);
1970 /* Must commit after prep above in all cases */
1971 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1972 objcount, ioo, remote_nb, npages, local_nb, NULL,
1974 if (rc == -ENOTCONN)
1975 /* quota acquire process has been given up because
1976 * either the client has been evicted or the client
1977 * has timed out the request already */
1981 * Disable sending mtime back to the client. If the client locked the
1982 * whole object, then it has already updated the mtime on its side,
1983 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1985 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1990 /* set per-requested niobuf return codes */
1991 for (i = j = 0; i < niocount; i++) {
1992 int len = remote_nb[i].len;
1997 LASSERT(j < npages);
1998 if (local_nb[j].rc < 0)
1999 rcs[i] = local_nb[j].rc;
2000 len -= local_nb[j].len;
2005 LASSERT(j == npages);
2006 ptlrpc_lprocfs_brw(req, nob);
2008 tgt_drop_id(exp, &repbody->oa);
2011 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2013 ptlrpc_free_bulk_nopin(desc);
2016 req->rq_no_reply = 1;
2017 /* reply out callback would free */
2018 ptlrpc_req_drop_rs(req);
2019 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2020 "client will retry: rc %d\n",
2021 exp->exp_obd->obd_name,
2022 obd_uuid2str(&exp->exp_client_uuid),
2023 obd_export_nid2str(exp), rc);
2025 memory_pressure_clr();
2028 EXPORT_SYMBOL(tgt_brw_write);