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, 2014, 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>
41 #include <lustre_lfsck.h>
42 #include <lustre_nodemap.h>
44 #include "tgt_internal.h"
46 char *tgt_name(struct lu_target *tgt)
48 LASSERT(tgt->lut_obd != NULL);
49 return tgt->lut_obd->obd_name;
51 EXPORT_SYMBOL(tgt_name);
54 * Generic code handling requests that have struct mdt_body passed in:
56 * - extract mdt_body from request and save it in @tsi, if present;
58 * - create lu_object, corresponding to the fid in mdt_body, and save it in
61 * - if HABEO_CORPUS flag is set for this request type check whether object
62 * actually exists on storage (lu_object_exists()).
65 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
67 const struct mdt_body *body;
68 struct lu_object *obj;
69 struct req_capsule *pill = tsi->tsi_pill;
74 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
78 tsi->tsi_mdt_body = body;
80 if (!(body->mbo_valid & OBD_MD_FLID))
83 /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
84 * in any case in pre-2.5 clients. Fix that here if needed */
85 if (unlikely(fid_is_zero(&body->mbo_fid1)))
88 if (!fid_is_sane(&body->mbo_fid1)) {
89 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
90 PFID(&body->mbo_fid1));
94 obj = lu_object_find(tsi->tsi_env,
95 &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
96 &body->mbo_fid1, NULL);
98 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
99 lu_object_put(tsi->tsi_env, obj);
102 tsi->tsi_corpus = obj;
109 tsi->tsi_fid = body->mbo_fid1;
115 * Validate oa from client.
116 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
118 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
119 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
120 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
121 * whether this is oi_fid or real ostid. So it will check
122 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
123 * c. Old FID-disable osc will send IDIF.
124 * d. new FID-enable osc/osp will send normal FID.
126 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
127 * be used for LAST_ID file, and only being accessed inside OST now.
129 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
131 struct ost_id *oi = &oa->o_oi;
132 u64 seq = ostid_seq(oi);
133 u64 id = ostid_id(oi);
137 if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
138 fid_seq_is_echo(seq))) {
139 /* Sigh 2.[123] client still sends echo req with oi_id = 0
140 * during create, and we will reset this to 1, since this
141 * oi_id is basically useless in the following create process,
142 * but oi_id == 0 will make it difficult to tell whether it is
143 * real FID or ost_id. */
144 oi->oi_fid.f_seq = FID_SEQ_ECHO;
145 oi->oi_fid.f_oid = id ?: 1;
146 oi->oi_fid.f_ver = 0;
148 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
150 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
151 GOTO(out, rc = -EPROTO);
153 /* Note: this check might be forced in 2.5 or 2.6, i.e.
154 * all of the requests are required to setup FLGROUP */
155 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
156 ostid_set_seq_mdt0(oi);
157 oa->o_valid |= OBD_MD_FLGROUP;
161 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
162 fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
163 GOTO(out, rc = -EPROTO);
165 rc = ostid_to_fid(&tti->tti_fid1, oi,
166 tsi->tsi_tgt->lut_lsd.lsd_osd_index);
167 if (unlikely(rc != 0))
170 oi->oi_fid = tti->tti_fid1;
176 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
177 tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
181 EXPORT_SYMBOL(tgt_validate_obdo);
183 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
186 struct niobuf_remote *rnb;
187 struct obd_ioobj *ioo;
192 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
196 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
200 max_brw = ioobj_max_brw_get(ioo);
201 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
202 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
203 ": rc = %d\n", tgt_name(tsi->tsi_tgt),
204 obd_export_nid2str(tsi->tsi_exp), max_brw,
205 POSTID(oi), -EPROTO);
210 obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
211 RCL_CLIENT) / sizeof(*ioo);
212 if (obj_count == 0) {
213 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
215 } else if (obj_count > 1) {
216 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
221 if (ioo->ioo_bufcnt == 0) {
222 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
226 if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
227 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
228 "bulk has too many pages (%d)",
236 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
238 struct ost_body *body;
239 struct req_capsule *pill = tsi->tsi_pill;
240 struct lu_nodemap *nodemap;
245 body = req_capsule_client_get(pill, &RMF_OST_BODY);
249 rc = tgt_validate_obdo(tsi, &body->oa);
253 nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
255 body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
256 NODEMAP_CLIENT_TO_FS,
258 body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
259 NODEMAP_CLIENT_TO_FS,
262 tsi->tsi_ost_body = body;
263 tsi->tsi_fid = body->oa.o_oi.oi_fid;
265 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
266 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
271 if (!(body->oa.o_valid & OBD_MD_FLID)) {
272 if (flags & HABEO_CORPUS) {
273 CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
274 "but OID/FID is mandatory with HABEO_CORPUS\n",
275 tgt_name(tsi->tsi_tgt));
282 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
285 * OST doesn't get object in advance for further use to prevent
286 * situations with nested object_find which is potential deadlock.
288 tsi->tsi_corpus = NULL;
293 * Do necessary preprocessing according to handler ->th_flags.
295 static int tgt_request_preprocess(struct tgt_session_info *tsi,
296 struct tgt_handler *h,
297 struct ptlrpc_request *req)
299 struct req_capsule *pill = tsi->tsi_pill;
300 __u32 flags = h->th_flags;
305 if (tsi->tsi_preprocessed)
308 LASSERT(h->th_act != NULL);
309 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
310 LASSERT(current->journal_info == NULL);
312 LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
314 if (h->th_fmt != NULL) {
315 req_capsule_set(pill, h->th_fmt);
316 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
317 rc = tgt_mdt_body_unpack(tsi, flags);
320 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
322 rc = tgt_ost_body_unpack(tsi, flags);
328 if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
331 if (flags & HABEO_CLAVIS) {
332 struct ldlm_request *dlm_req;
334 LASSERT(h->th_fmt != NULL);
336 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
337 if (dlm_req != NULL) {
338 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
340 dlm_req->lock_desc.l_policy_data.\
341 l_inodebits.bits == 0)) {
343 * Lock without inodebits makes no sense and
344 * will oops later in ldlm. If client miss to
345 * set such bits, do not trigger ASSERTION.
347 * For liblustre flock case, it maybe zero.
351 tsi->tsi_dlm_req = dlm_req;
357 tsi->tsi_preprocessed = 1;
362 * Invoke handler for this request opc. Also do necessary preprocessing
363 * (according to handler ->th_flags), and post-processing (setting of
364 * ->last_{xid,committed}).
366 static int tgt_handle_request0(struct tgt_session_info *tsi,
367 struct tgt_handler *h,
368 struct ptlrpc_request *req)
376 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
377 * to put same checks into handlers like mdt_close(), mdt_reint(),
378 * etc., without talking to mdt authors first. Checking same thing
379 * there again is useless and returning 0 error without packing reply
380 * is buggy! Handlers either pack reply or return error.
382 * We return 0 here and do not send any reply in order to emulate
383 * network failure. Do not send any reply in case any of NET related
384 * fail_id has occured.
386 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
389 rc = tgt_request_preprocess(tsi, h, req);
390 /* pack reply if reply format is fixed */
391 if (rc == 0 && h->th_flags & HABEO_REFERO) {
393 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
395 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
397 tsi->tsi_mdt_body->mbo_eadatasize);
398 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
400 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
403 rc = req_capsule_server_pack(tsi->tsi_pill);
406 if (likely(rc == 0)) {
408 * Process request, there can be two types of rc:
409 * 1) errors with msg unpack/pack, other failures outside the
410 * operation itself. This is counted as serious errors;
411 * 2) errors during fs operation, should be placed in rq_status
415 if (!is_serious(rc) &&
416 !req->rq_no_reply && req->rq_reply_state == NULL) {
417 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
418 "pack reply and returned 0 error\n",
419 tgt_name(tsi->tsi_tgt), h->th_name);
422 serious = is_serious(rc);
423 rc = clear_serious(rc);
431 * ELDLM_* codes which > 0 should be in rq_status only as well as
432 * all non-serious errors.
434 if (rc > 0 || !serious)
437 LASSERT(current->journal_info == NULL);
439 if (likely(rc == 0 && req->rq_export))
440 target_committed_to_req(req);
442 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
446 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
447 struct obd_device *obd, int *process)
449 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
456 case MDS_DONE_WRITING:
457 case MDS_SYNC: /* used in unmounting */
471 *process = target_queue_recovery_request(req, obd);
475 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
482 * Handle recovery. Return:
483 * +1: continue request processing;
484 * -ve: abort immediately with the given error code;
485 * 0: send reply with error code in req->rq_status;
487 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
491 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
496 case SEC_CTX_INIT_CONT:
501 if (!req->rq_export->exp_obd->obd_replayable)
504 /* sanity check: if the xid matches, the request must be marked as a
505 * resent or replayed */
506 if (req_xid_is_last(req)) {
507 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
508 (MSG_RESENT | MSG_REPLAY))) {
509 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
510 "last_xid, expected REPLAY or RESENT flag "
512 lustre_msg_get_flags(req->rq_reqmsg));
513 req->rq_status = -ENOTCONN;
517 /* else: note the opposite is not always true; a RESENT req after a
518 * failover will usually not match the last_xid, since it was likely
519 * never committed. A REPLAYed request will almost never match the
520 * last xid, however it could for a committed, but still retained,
523 /* Check for aborted recovery... */
524 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
528 DEBUG_REQ(D_INFO, req, "Got new replay");
529 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
531 if (rc != 0 || !should_process)
533 else if (should_process < 0) {
534 req->rq_status = should_process;
535 rc = ptlrpc_error(req);
542 /* Initial check for request, it is validation mostly */
543 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
545 struct tgt_handler *h;
546 struct tgt_opc_slice *s;
547 struct lu_target *tgt;
548 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
552 tgt = class_exp2tgt(req->rq_export);
554 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
555 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
558 /* opcode was not found in slice */
559 if (unlikely(s->tos_hs == NULL)) {
560 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
562 RETURN(ERR_PTR(-ENOTSUPP));
565 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
566 h = s->tos_hs + (opc - s->tos_opc_start);
567 if (unlikely(h->th_opc == 0)) {
568 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
569 RETURN(ERR_PTR(-ENOTSUPP));
575 int tgt_request_handle(struct ptlrpc_request *req)
577 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
579 struct lustre_msg *msg = req->rq_reqmsg;
580 struct tgt_handler *h;
581 struct lu_target *tgt;
582 int request_fail_id = 0;
583 __u32 opc = lustre_msg_get_opc(msg);
588 /* Refill the context, to make sure all thread keys are allocated */
589 lu_env_refill(req->rq_svc_thread->t_env);
591 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
592 tsi->tsi_pill = &req->rq_pill;
593 tsi->tsi_env = req->rq_svc_thread->t_env;
595 /* if request has export then get handlers slice from corresponding
596 * target, otherwise that should be connect operation */
597 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
598 opc == MGS_CONNECT) {
599 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
600 rc = target_handle_connect(req);
602 rc = ptlrpc_error(req);
605 /* recovery-small test 18c asks to drop connect reply */
606 if (unlikely(opc == OST_CONNECT &&
607 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
611 if (unlikely(!class_connected_export(req->rq_export))) {
612 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
613 opc, libcfs_id2str(req->rq_peer));
614 req->rq_status = -ENOTCONN;
615 rc = ptlrpc_error(req);
619 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
620 tsi->tsi_exp = req->rq_export;
621 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
622 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
624 tsi->tsi_jobid = NULL;
626 request_fail_id = tgt->lut_request_fail_id;
627 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
629 h = tgt_handler_find_check(req);
631 req->rq_status = PTR_ERR(h);
632 rc = ptlrpc_error(req);
636 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
639 rc = lustre_msg_check_version(msg, h->th_version);
641 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
642 " %08x, expecting %08x\n", tgt_name(tgt),
643 lustre_msg_get_version(msg), h->th_version);
644 req->rq_status = -EINVAL;
645 rc = ptlrpc_error(req);
649 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
650 if (likely(rc == 1)) {
651 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
653 rc = tgt_handle_request0(tsi, h, req);
659 req_capsule_fini(tsi->tsi_pill);
660 if (tsi->tsi_corpus != NULL) {
661 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
662 tsi->tsi_corpus = NULL;
666 EXPORT_SYMBOL(tgt_request_handle);
668 /** Assign high priority operations to the request if needed. */
669 int tgt_hpreq_handler(struct ptlrpc_request *req)
671 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
672 struct tgt_handler *h;
677 if (req->rq_export == NULL)
680 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
681 tsi->tsi_pill = &req->rq_pill;
682 tsi->tsi_env = req->rq_svc_thread->t_env;
683 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
684 tsi->tsi_exp = req->rq_export;
686 h = tgt_handler_find_check(req);
692 rc = tgt_request_preprocess(tsi, h, req);
693 if (unlikely(rc != 0))
696 if (h->th_hp != NULL)
700 EXPORT_SYMBOL(tgt_hpreq_handler);
702 void tgt_counter_incr(struct obd_export *exp, int opcode)
704 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
705 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
706 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
708 EXPORT_SYMBOL(tgt_counter_incr);
711 * Unified target generic handlers.
717 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
719 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
720 OBD_CONNECT_RMT_CLIENT_FORCE);
723 static int tgt_init_sec_level(struct ptlrpc_request *req)
725 struct lu_target *tgt = class_exp2tgt(req->rq_export);
727 struct obd_connect_data *data, *reply;
732 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
733 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
734 if (data == NULL || reply == NULL)
737 /* connection from MDT is always trusted */
738 if (req->rq_auth_usr_mdt) {
739 tgt_init_sec_none(reply);
743 client = libcfs_nid2str(req->rq_peer.nid);
744 /* no GSS support case */
745 if (!req->rq_auth_gss) {
746 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
747 CWARN("client %s -> target %s does not use GSS, "
748 "can not run under security level %d.\n",
749 client, tgt_name(tgt), tgt->lut_sec_level);
752 tgt_init_sec_none(reply);
757 /* old version case */
758 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
759 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
760 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
761 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
762 CWARN("client %s -> target %s uses old version, "
763 "can not run under security level %d.\n",
764 client, tgt_name(tgt), tgt->lut_sec_level);
767 CWARN("client %s -> target %s uses old version, "
768 "run under security level %d.\n",
769 client, tgt_name(tgt), tgt->lut_sec_level);
770 tgt_init_sec_none(reply);
775 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
777 if (!req->rq_auth_remote)
778 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
779 "asked to be remote.\n", client, tgt_name(tgt));
780 } else if (req->rq_auth_remote) {
782 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
783 "as remote by default.\n", client, tgt_name(tgt));
787 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
788 CDEBUG(D_SEC, "client %s -> target %s: user is not "
789 "authenticated!\n", client, tgt_name(tgt));
795 switch (tgt->lut_sec_level) {
796 case LUSTRE_SEC_NONE:
799 "client %s -> target %s is set as remote, "
800 "can not run under security level %d.\n",
801 client, tgt_name(tgt), tgt->lut_sec_level);
804 tgt_init_sec_none(reply);
806 case LUSTRE_SEC_REMOTE:
808 tgt_init_sec_none(reply);
813 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
814 OBD_CONNECT_RMT_CLIENT_FORCE);
815 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
816 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
825 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
827 struct lu_target *tgt = class_exp2tgt(exp);
828 struct sptlrpc_flavor flvr;
832 LASSERT(tgt->lut_obd);
833 LASSERT(tgt->lut_slice);
835 /* always allow ECHO client */
836 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
837 LUSTRE_ECHO_NAME) == 0)) {
838 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
842 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
843 read_lock(&tgt->lut_sptlrpc_lock);
844 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
848 read_unlock(&tgt->lut_sptlrpc_lock);
850 spin_lock(&exp->exp_lock);
851 exp->exp_sp_peer = req->rq_sp_from;
852 exp->exp_flvr = flvr;
853 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
854 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
855 CERROR("%s: unauthorized rpc flavor %x from %s, "
856 "expect %x\n", tgt_name(tgt),
858 libcfs_nid2str(req->rq_peer.nid),
859 exp->exp_flvr.sf_rpc);
862 spin_unlock(&exp->exp_lock);
864 if (exp->exp_sp_peer != req->rq_sp_from) {
865 CERROR("%s: RPC source %s doesn't match %s\n",
867 sptlrpc_part2name(req->rq_sp_from),
868 sptlrpc_part2name(exp->exp_sp_peer));
871 rc = sptlrpc_target_export_check(exp, req);
878 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
880 struct sptlrpc_rule_set tmp_rset;
883 sptlrpc_rule_set_init(&tmp_rset);
884 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
886 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
891 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
893 write_lock(&tgt->lut_sptlrpc_lock);
894 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
895 tgt->lut_sptlrpc_rset = tmp_rset;
896 write_unlock(&tgt->lut_sptlrpc_lock);
900 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
902 int tgt_connect(struct tgt_session_info *tsi)
904 struct ptlrpc_request *req = tgt_ses_req(tsi);
905 struct obd_connect_data *reply;
910 rc = tgt_init_sec_level(req);
914 /* XXX: better to call this check right after getting new export but
915 * before last_rcvd slot allocation to avoid server load upon insecure
916 * connects. This is to be fixed after unifiyng all targets.
918 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
922 /* To avoid exposing partially initialized connection flags, changes up
923 * to this point have been staged in reply->ocd_connect_flags. Now that
924 * connection handling has completed successfully, atomically update
925 * the connect flags in the shared export data structure. LU-1623 */
926 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
927 spin_lock(&tsi->tsi_exp->exp_lock);
928 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
929 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
930 spin_unlock(&tsi->tsi_exp->exp_lock);
934 obd_disconnect(class_export_get(tsi->tsi_exp));
937 EXPORT_SYMBOL(tgt_connect);
939 int tgt_disconnect(struct tgt_session_info *tsi)
945 rc = target_handle_disconnect(tgt_ses_req(tsi));
947 RETURN(err_serious(rc));
951 EXPORT_SYMBOL(tgt_disconnect);
954 * Unified target OBD handlers
956 int tgt_obd_ping(struct tgt_session_info *tsi)
962 rc = target_handle_ping(tgt_ses_req(tsi));
964 RETURN(err_serious(rc));
968 EXPORT_SYMBOL(tgt_obd_ping);
970 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
972 return err_serious(-EOPNOTSUPP);
975 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
977 return err_serious(-EOPNOTSUPP);
980 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
982 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
983 struct ptlrpc_request *req = tgt_ses_req(tsi);
984 struct obd_export *exp = req->rq_export;
985 struct ptlrpc_bulk_desc *desc;
986 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
994 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
999 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1000 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1001 * which is rdpg->rp_count */
1002 nob = rdpg->rp_count;
1004 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1005 i++, tmpcount -= tmpsize) {
1006 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1007 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1010 LASSERT(desc->bd_nob == nob);
1011 rc = target_bulk_io(exp, desc, lwi);
1012 ptlrpc_free_bulk_pin(desc);
1015 EXPORT_SYMBOL(tgt_sendpage);
1018 * OBD_IDX_READ handler
1020 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1022 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1023 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1024 struct idx_info *req_ii, *rep_ii;
1029 memset(rdpg, 0, sizeof(*rdpg));
1030 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1032 /* extract idx_info buffer from request & reply */
1033 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1034 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1035 RETURN(err_serious(-EPROTO));
1037 rc = req_capsule_server_pack(tsi->tsi_pill);
1039 RETURN(err_serious(rc));
1041 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1043 RETURN(err_serious(-EFAULT));
1044 rep_ii->ii_magic = IDX_INFO_MAGIC;
1046 /* extract hash to start with */
1047 rdpg->rp_hash = req_ii->ii_hash_start;
1049 /* extract requested attributes */
1050 rdpg->rp_attrs = req_ii->ii_attrs;
1052 /* check that fid packed in request is valid and supported */
1053 if (!fid_is_sane(&req_ii->ii_fid))
1055 rep_ii->ii_fid = req_ii->ii_fid;
1058 rep_ii->ii_flags = req_ii->ii_flags;
1060 /* compute number of pages to allocate, ii_count is the number of 4KB
1062 if (req_ii->ii_count <= 0)
1063 GOTO(out, rc = -EFAULT);
1064 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1065 exp_max_brw_size(tsi->tsi_exp));
1066 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1068 /* allocate pages to store the containers */
1069 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1070 if (rdpg->rp_pages == NULL)
1071 GOTO(out, rc = -ENOMEM);
1072 for (i = 0; i < rdpg->rp_npages; i++) {
1073 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1074 if (rdpg->rp_pages[i] == NULL)
1075 GOTO(out, rc = -ENOMEM);
1078 /* populate pages with key/record pairs */
1079 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1083 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1084 "asked %d > %d\n", rc, rdpg->rp_count);
1086 /* send pages to client */
1087 rc = tgt_sendpage(tsi, rdpg, rc);
1092 if (rdpg->rp_pages) {
1093 for (i = 0; i < rdpg->rp_npages; i++)
1094 if (rdpg->rp_pages[i])
1095 __free_page(rdpg->rp_pages[i]);
1096 OBD_FREE(rdpg->rp_pages,
1097 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1102 struct tgt_handler tgt_obd_handlers[] = {
1103 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1104 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1105 TGT_OBD_HDL_VAR(0, OBD_QC_CALLBACK, tgt_obd_qc_callback),
1106 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1108 EXPORT_SYMBOL(tgt_obd_handlers);
1110 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1111 struct dt_object *obj, __u64 start, __u64 end)
1117 /* if no objid is specified, it means "sync whole filesystem" */
1119 rc = dt_sync(env, tgt->lut_bottom);
1120 } else if (dt_version_get(env, obj) >
1121 tgt->lut_obd->obd_last_committed) {
1122 rc = dt_object_sync(env, obj, start, end);
1127 EXPORT_SYMBOL(tgt_sync);
1129 * Unified target DLM handlers.
1132 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1134 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1135 void *data, int flag)
1138 struct lu_target *tgt;
1139 struct dt_object *obj;
1145 tgt = class_exp2tgt(lock->l_export);
1147 if (flag == LDLM_CB_CANCELING &&
1148 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1149 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1150 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1151 lock->l_flags & LDLM_FL_CBPENDING))) {
1153 __u64 end = OBD_OBJECT_EOF;
1155 rc = lu_env_init(&env, LCT_DT_THREAD);
1156 if (unlikely(rc != 0))
1159 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1160 tgt->lut_lsd.lsd_osd_index);
1161 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1163 GOTO(err_env, rc = PTR_ERR(obj));
1165 if (!dt_object_exists(obj))
1166 GOTO(err_put, rc = -ENOENT);
1168 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1169 start = lock->l_policy_data.l_extent.start;
1170 end = lock->l_policy_data.l_extent.end;
1173 rc = tgt_sync(&env, tgt, obj, start, end);
1175 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1176 "cancel: rc = %d\n",
1177 tgt_name(tgt), PFID(&fid),
1178 lock->l_policy_data.l_extent.start,
1179 lock->l_policy_data.l_extent.end, rc);
1182 lu_object_put(&env, &obj->do_lu);
1187 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1191 static struct ldlm_callback_suite tgt_dlm_cbs = {
1192 .lcs_completion = ldlm_server_completion_ast,
1193 .lcs_blocking = tgt_blocking_ast,
1194 .lcs_glimpse = ldlm_server_glimpse_ast
1197 int tgt_enqueue(struct tgt_session_info *tsi)
1199 struct ptlrpc_request *req = tgt_ses_req(tsi);
1204 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1205 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1207 LASSERT(tsi->tsi_dlm_req != NULL);
1208 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1209 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1211 RETURN(err_serious(rc));
1213 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1215 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1218 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1221 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1224 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1227 RETURN(req->rq_status);
1229 EXPORT_SYMBOL(tgt_enqueue);
1231 int tgt_convert(struct tgt_session_info *tsi)
1233 struct ptlrpc_request *req = tgt_ses_req(tsi);
1237 LASSERT(tsi->tsi_dlm_req);
1238 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1240 RETURN(err_serious(rc));
1242 RETURN(req->rq_status);
1245 int tgt_bl_callback(struct tgt_session_info *tsi)
1247 return err_serious(-EOPNOTSUPP);
1250 int tgt_cp_callback(struct tgt_session_info *tsi)
1252 return err_serious(-EOPNOTSUPP);
1255 /* generic LDLM target handler */
1256 struct tgt_handler tgt_dlm_handlers[] = {
1257 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1258 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1259 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1260 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1262 EXPORT_SYMBOL(tgt_dlm_handlers);
1265 * Unified target LLOG handlers.
1267 int tgt_llog_open(struct tgt_session_info *tsi)
1273 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1277 EXPORT_SYMBOL(tgt_llog_open);
1279 int tgt_llog_close(struct tgt_session_info *tsi)
1285 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1289 EXPORT_SYMBOL(tgt_llog_close);
1292 int tgt_llog_destroy(struct tgt_session_info *tsi)
1298 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1303 int tgt_llog_read_header(struct tgt_session_info *tsi)
1309 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1313 EXPORT_SYMBOL(tgt_llog_read_header);
1315 int tgt_llog_next_block(struct tgt_session_info *tsi)
1321 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1325 EXPORT_SYMBOL(tgt_llog_next_block);
1327 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1333 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1337 EXPORT_SYMBOL(tgt_llog_prev_block);
1339 /* generic llog target handler */
1340 struct tgt_handler tgt_llog_handlers[] = {
1341 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1342 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1343 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1344 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1345 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1346 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1348 EXPORT_SYMBOL(tgt_llog_handlers);
1351 * sec context handlers
1353 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1354 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1359 struct tgt_handler tgt_sec_ctx_handlers[] = {
1360 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1361 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1362 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1364 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1366 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1367 struct dt_device *key,
1368 struct lfsck_request *lr,
1369 struct thandle *th) = NULL;
1371 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1373 struct lfsck_request *,
1376 tgt_lfsck_in_notify = notify;
1378 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1380 static int (*tgt_lfsck_query)(const struct lu_env *env,
1381 struct dt_device *key,
1382 struct lfsck_request *lr) = NULL;
1384 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1386 struct lfsck_request *))
1388 tgt_lfsck_query = query;
1390 EXPORT_SYMBOL(tgt_register_lfsck_query);
1392 /* LFSCK request handlers */
1393 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1395 const struct lu_env *env = tsi->tsi_env;
1396 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1397 struct lfsck_request *lr;
1401 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1405 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1410 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1412 struct lfsck_request *request;
1413 struct lfsck_reply *reply;
1417 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1418 if (request == NULL)
1421 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1425 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1426 reply->lr_status = rc;
1428 RETURN(rc < 0 ? rc : 0);
1431 struct tgt_handler tgt_lfsck_handlers[] = {
1432 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1433 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1435 EXPORT_SYMBOL(tgt_lfsck_handlers);
1438 * initialize per-thread page pool (bug 5137).
1440 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1442 struct tgt_thread_big_cache *tbc;
1446 LASSERT(thread != NULL);
1447 LASSERT(thread->t_data == NULL);
1449 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1452 thread->t_data = tbc;
1455 EXPORT_SYMBOL(tgt_io_thread_init);
1458 * free per-thread pool created by tgt_thread_init().
1460 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1462 struct tgt_thread_big_cache *tbc;
1466 LASSERT(thread != NULL);
1469 * be prepared to handle partially-initialized pools (because this is
1470 * called from ost_io_thread_init() for cleanup.
1472 tbc = thread->t_data;
1474 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1475 thread->t_data = NULL;
1479 EXPORT_SYMBOL(tgt_io_thread_done);
1481 * Helper function for getting server side [start, start+count] DLM lock
1482 * if asked by client.
1484 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1485 __u64 start, __u64 end, struct lustre_handle *lh,
1486 int mode, __u64 *flags)
1488 ldlm_policy_data_t policy;
1493 LASSERT(lh != NULL);
1494 LASSERT(ns != NULL);
1495 LASSERT(!lustre_handle_is_used(lh));
1497 policy.l_extent.gid = 0;
1498 policy.l_extent.start = start & PAGE_MASK;
1501 * If ->o_blocks is EOF it means "lock till the end of the file".
1502 * Otherwise, it's size of an extent or hole being punched (in bytes).
1504 if (end == OBD_OBJECT_EOF || end < start)
1505 policy.l_extent.end = OBD_OBJECT_EOF;
1507 policy.l_extent.end = end | ~PAGE_MASK;
1509 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1510 flags, ldlm_blocking_ast,
1511 ldlm_completion_ast, ldlm_glimpse_ast,
1512 NULL, 0, LVB_T_NONE, NULL, lh);
1513 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1515 EXPORT_SYMBOL(tgt_extent_lock);
1517 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1519 LASSERT(lustre_handle_is_used(lh));
1520 ldlm_lock_decref(lh, mode);
1522 EXPORT_SYMBOL(tgt_extent_unlock);
1524 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1525 struct obd_ioobj *obj, struct niobuf_remote *nb,
1526 struct lustre_handle *lh, int mode)
1529 int nrbufs = obj->ioo_bufcnt;
1534 LASSERT(mode == LCK_PR || mode == LCK_PW);
1535 LASSERT(!lustre_handle_is_used(lh));
1537 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1540 for (i = 1; i < nrbufs; i++)
1541 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1544 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1545 nb[nrbufs - 1].rnb_offset +
1546 nb[nrbufs - 1].rnb_len - 1,
1550 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1551 struct lustre_handle *lh, int mode)
1555 LASSERT(mode == LCK_PR || mode == LCK_PW);
1556 LASSERT((obj->ioo_bufcnt > 0 &&
1557 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1558 lustre_handle_is_used(lh));
1560 if (lustre_handle_is_used(lh))
1561 tgt_extent_unlock(lh, mode);
1565 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1566 struct ptlrpc_bulk_desc *desc, int opc,
1567 cksum_type_t cksum_type)
1569 struct cfs_crypto_hash_desc *hdesc;
1570 unsigned int bufsize;
1572 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1575 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1576 if (IS_ERR(hdesc)) {
1577 CERROR("%s: unable to initialize checksum hash %s\n",
1578 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1579 return PTR_ERR(hdesc);
1582 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1583 for (i = 0; i < desc->bd_iov_count; i++) {
1584 /* corrupt the data before we compute the checksum, to
1585 * simulate a client->OST data error */
1586 if (i == 0 && opc == OST_WRITE &&
1587 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1588 int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1589 int len = desc->bd_iov[i].kiov_len;
1590 struct page *np = tgt_page_to_corrupt;
1591 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1594 char *ptr2 = kmap(np) + off;
1596 memcpy(ptr2, ptr, len);
1597 memcpy(ptr2, "bad3", min(4, len));
1599 desc->bd_iov[i].kiov_page = np;
1601 CERROR("%s: can't alloc page for corruption\n",
1605 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1606 desc->bd_iov[i].kiov_offset & ~PAGE_MASK,
1607 desc->bd_iov[i].kiov_len);
1609 /* corrupt the data after we compute the checksum, to
1610 * simulate an OST->client data error */
1611 if (i == 0 && opc == OST_READ &&
1612 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1613 int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1614 int len = desc->bd_iov[i].kiov_len;
1615 struct page *np = tgt_page_to_corrupt;
1616 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1619 char *ptr2 = kmap(np) + off;
1621 memcpy(ptr2, ptr, len);
1622 memcpy(ptr2, "bad4", min(4, len));
1624 desc->bd_iov[i].kiov_page = np;
1626 CERROR("%s: can't alloc page for corruption\n",
1632 bufsize = sizeof(cksum);
1633 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1638 int tgt_brw_read(struct tgt_session_info *tsi)
1640 struct ptlrpc_request *req = tgt_ses_req(tsi);
1641 struct ptlrpc_bulk_desc *desc = NULL;
1642 struct obd_export *exp = tsi->tsi_exp;
1643 struct niobuf_remote *remote_nb;
1644 struct niobuf_local *local_nb;
1645 struct obd_ioobj *ioo;
1646 struct ost_body *body, *repbody;
1647 struct l_wait_info lwi;
1648 struct lustre_handle lockh = { 0 };
1649 int npages, nob = 0, rc, i, no_reply = 0;
1650 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1654 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1655 CERROR("%s: deny read request from %s to portal %u\n",
1656 tgt_name(tsi->tsi_tgt),
1657 obd_export_nid2str(req->rq_export),
1658 ptlrpc_req2svc(req)->srv_req_portal);
1662 req->rq_bulk_read = 1;
1664 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1667 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1668 cfs_fail_val : (obd_timeout + 1) / 4);
1670 /* Check if there is eviction in progress, and if so, wait for it to
1672 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1673 /* We do not care how long it takes */
1674 lwi = LWI_INTR(NULL, NULL);
1675 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1676 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1680 /* There must be big cache in current thread to process this request
1681 * if it is NULL then something went wrong and it wasn't allocated,
1682 * report -ENOMEM in that case */
1686 body = tsi->tsi_ost_body;
1687 LASSERT(body != NULL);
1689 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1690 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1692 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1693 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1695 local_nb = tbc->local;
1697 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1698 remote_nb, &lockh, LCK_PR);
1703 * If getting the lock took more time than
1704 * client was willing to wait, drop it. b=11330
1706 if (cfs_time_current_sec() > req->rq_deadline ||
1707 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1709 CERROR("Dropping timed-out read from %s because locking"
1710 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1711 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1712 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1713 req->rq_deadline - req->rq_arrival_time.tv_sec);
1714 GOTO(out_lock, rc = -ETIMEDOUT);
1717 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1718 repbody->oa = body->oa;
1720 npages = PTLRPC_MAX_BRW_PAGES;
1721 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1722 ioo, remote_nb, &npages, local_nb, NULL);
1726 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1727 BULK_PUT_SOURCE, OST_BULK_PORTAL);
1729 GOTO(out_commitrw, rc = -ENOMEM);
1732 for (i = 0; i < npages; i++) {
1733 int page_rc = local_nb[i].lnb_rc;
1741 if (page_rc != 0) { /* some data! */
1742 LASSERT(local_nb[i].lnb_page != NULL);
1743 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1744 local_nb[i].lnb_page_offset,
1748 if (page_rc != local_nb[i].lnb_len) { /* short read */
1749 /* All subsequent pages should be 0 */
1750 while (++i < npages)
1751 LASSERT(local_nb[i].lnb_rc == 0);
1756 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1757 cksum_type_t cksum_type =
1758 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1759 body->oa.o_flags : 0);
1760 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1761 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1762 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1763 OST_READ, cksum_type);
1764 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1765 repbody->oa.o_cksum);
1767 repbody->oa.o_valid = 0;
1769 /* We're finishing using body->oa as an input variable */
1771 /* Check if client was evicted while we were doing i/o before touching
1773 if (likely(rc == 0 &&
1774 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1775 rc = target_bulk_io(exp, desc, &lwi);
1780 /* Must commit after prep above in all cases */
1781 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1782 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1785 tgt_drop_id(exp, &repbody->oa);
1787 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1789 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1790 ptlrpc_free_bulk_nopin(desc);
1795 ptlrpc_lprocfs_brw(req, nob);
1796 } else if (no_reply) {
1797 req->rq_no_reply = 1;
1798 /* reply out callback would free */
1799 ptlrpc_req_drop_rs(req);
1800 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1801 "client will retry: rc %d\n",
1802 exp->exp_obd->obd_name,
1803 obd_uuid2str(&exp->exp_client_uuid),
1804 obd_export_nid2str(exp), rc);
1806 /* send a bulk after reply to simulate a network delay or reordering
1808 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1809 wait_queue_head_t waitq;
1810 struct l_wait_info lwi1;
1812 CDEBUG(D_INFO, "reorder BULK\n");
1813 init_waitqueue_head(&waitq);
1815 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1816 l_wait_event(waitq, 0, &lwi1);
1817 target_bulk_io(exp, desc, &lwi);
1818 ptlrpc_free_bulk_nopin(desc);
1823 EXPORT_SYMBOL(tgt_brw_read);
1825 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1826 struct ptlrpc_bulk_desc *desc,
1827 struct niobuf_local *local_nb, int npages,
1828 u32 client_cksum, u32 server_cksum,
1831 struct obd_export *exp = req->rq_export;
1832 struct ost_body *body;
1836 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1837 LASSERT(body != NULL);
1839 if (req->rq_peer.nid != desc->bd_sender) {
1841 router = libcfs_nid2str(desc->bd_sender);
1845 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1846 client_cksum, server_cksum);
1850 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1851 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1852 "]: client csum %x, server csum %x\n",
1853 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1855 body->oa.o_valid & OBD_MD_FLFID ?
1856 body->oa.o_parent_seq : (__u64)0,
1857 body->oa.o_valid & OBD_MD_FLFID ?
1858 body->oa.o_parent_oid : 0,
1859 body->oa.o_valid & OBD_MD_FLFID ?
1860 body->oa.o_parent_ver : 0,
1861 POSTID(&body->oa.o_oi),
1862 local_nb[0].lnb_file_offset,
1863 local_nb[npages-1].lnb_file_offset +
1864 local_nb[npages - 1].lnb_len - 1,
1865 client_cksum, server_cksum);
1868 int tgt_brw_write(struct tgt_session_info *tsi)
1870 struct ptlrpc_request *req = tgt_ses_req(tsi);
1871 struct ptlrpc_bulk_desc *desc = NULL;
1872 struct obd_export *exp = req->rq_export;
1873 struct niobuf_remote *remote_nb;
1874 struct niobuf_local *local_nb;
1875 struct obd_ioobj *ioo;
1876 struct ost_body *body, *repbody;
1877 struct l_wait_info lwi;
1878 struct lustre_handle lockh = {0};
1880 int objcount, niocount, npages;
1882 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1883 bool no_reply = false, mmap;
1884 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1888 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1889 CERROR("%s: deny write request from %s to portal %u\n",
1890 tgt_name(tsi->tsi_tgt),
1891 obd_export_nid2str(req->rq_export),
1892 ptlrpc_req2svc(req)->srv_req_portal);
1893 RETURN(err_serious(-EPROTO));
1896 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1897 RETURN(err_serious(-ENOSPC));
1898 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1899 RETURN(err_serious(-EROFS));
1901 req->rq_bulk_write = 1;
1903 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1904 RETURN(err_serious(-EIO));
1905 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1906 RETURN(err_serious(-EFAULT));
1908 /* pause before transaction has been started */
1909 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1910 cfs_fail_val : (obd_timeout + 1) / 4);
1912 /* There must be big cache in current thread to process this request
1913 * if it is NULL then something went wrong and it wasn't allocated,
1914 * report -ENOMEM in that case */
1918 body = tsi->tsi_ost_body;
1919 LASSERT(body != NULL);
1921 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1922 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1924 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1925 RCL_CLIENT) / sizeof(*ioo);
1927 for (niocount = i = 0; i < objcount; i++)
1928 niocount += ioo[i].ioo_bufcnt;
1930 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1931 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1932 if (niocount != req_capsule_get_size(&req->rq_pill,
1933 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1935 RETURN(err_serious(-EPROTO));
1937 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1938 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1939 memory_pressure_set();
1941 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1942 niocount * sizeof(*rcs));
1943 rc = req_capsule_server_pack(&req->rq_pill);
1945 GOTO(out, rc = err_serious(rc));
1947 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1948 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1950 local_nb = tbc->local;
1952 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1953 remote_nb, &lockh, LCK_PW);
1958 * If getting the lock took more time than
1959 * client was willing to wait, drop it. b=11330
1961 if (cfs_time_current_sec() > req->rq_deadline ||
1962 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1964 CERROR("%s: Dropping timed-out write from %s because locking "
1965 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1966 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1967 POSTID(&ioo->ioo_oid),
1968 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1969 req->rq_deadline - req->rq_arrival_time.tv_sec);
1970 GOTO(out_lock, rc = -ETIMEDOUT);
1973 /* Because we already sync grant info with client when reconnect,
1974 * grant info will be cleared for resent req, then fed_grant and
1975 * total_grant will not be modified in following preprw_write */
1976 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1977 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1978 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1981 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1982 if (repbody == NULL)
1983 GOTO(out_lock, rc = -ENOMEM);
1984 repbody->oa = body->oa;
1986 npages = PTLRPC_MAX_BRW_PAGES;
1987 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1988 objcount, ioo, remote_nb, &npages, local_nb, NULL);
1992 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1993 BULK_GET_SINK, OST_BULK_PORTAL);
1995 GOTO(skip_transfer, rc = -ENOMEM);
1997 /* NB Having prepped, we must commit... */
1998 for (i = 0; i < npages; i++)
1999 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2000 local_nb[i].lnb_page_offset,
2001 local_nb[i].lnb_len);
2003 rc = sptlrpc_svc_prep_bulk(req, desc);
2005 GOTO(skip_transfer, rc);
2007 rc = target_bulk_io(exp, desc, &lwi);
2011 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2012 static int cksum_counter;
2014 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2015 cksum_type = cksum_type_unpack(body->oa.o_flags);
2017 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2018 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2019 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2020 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2021 OST_WRITE, cksum_type);
2024 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2025 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2026 body->oa.o_flags & OBD_FL_MMAP);
2028 tgt_warn_on_cksum(req, desc, local_nb, npages,
2030 repbody->oa.o_cksum, mmap);
2032 } else if ((cksum_counter & (-cksum_counter)) ==
2034 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2035 cksum_counter, libcfs_id2str(req->rq_peer),
2036 repbody->oa.o_cksum);
2040 /* Must commit after prep above in all cases */
2041 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2042 objcount, ioo, remote_nb, npages, local_nb, NULL,
2044 if (rc == -ENOTCONN)
2045 /* quota acquire process has been given up because
2046 * either the client has been evicted or the client
2047 * has timed out the request already */
2051 * Disable sending mtime back to the client. If the client locked the
2052 * whole object, then it has already updated the mtime on its side,
2053 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2055 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2060 /* set per-requested niobuf return codes */
2061 for (i = j = 0; i < niocount; i++) {
2062 int len = remote_nb[i].rnb_len;
2067 LASSERT(j < npages);
2068 if (local_nb[j].lnb_rc < 0)
2069 rcs[i] = local_nb[j].lnb_rc;
2070 len -= local_nb[j].lnb_len;
2075 LASSERT(j == npages);
2076 ptlrpc_lprocfs_brw(req, nob);
2078 tgt_drop_id(exp, &repbody->oa);
2081 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2083 ptlrpc_free_bulk_nopin(desc);
2086 req->rq_no_reply = 1;
2087 /* reply out callback would free */
2088 ptlrpc_req_drop_rs(req);
2089 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2090 "client will retry: rc %d\n",
2091 exp->exp_obd->obd_name,
2092 obd_uuid2str(&exp->exp_client_uuid),
2093 obd_export_nid2str(exp), rc);
2095 memory_pressure_clr();
2098 EXPORT_SYMBOL(tgt_brw_write);