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>
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);
100 /* for capability renew ENOENT will be handled in
102 if (body->mbo_valid & OBD_MD_FLOSSCAPA)
107 tsi->tsi_corpus = obj;
114 tsi->tsi_fid = body->mbo_fid1;
120 * Validate oa from client.
121 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
123 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
124 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
125 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
126 * whether this is oi_fid or real ostid. So it will check
127 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
128 * c. Old FID-disable osc will send IDIF.
129 * d. new FID-enable osc/osp will send normal FID.
131 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
132 * be used for LAST_ID file, and only being accessed inside OST now.
134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
136 struct ost_id *oi = &oa->o_oi;
137 obd_seq seq = ostid_seq(oi);
138 obd_id id = ostid_id(oi);
142 if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
143 fid_seq_is_echo(seq))) {
144 /* Sigh 2.[123] client still sends echo req with oi_id = 0
145 * during create, and we will reset this to 1, since this
146 * oi_id is basically useless in the following create process,
147 * but oi_id == 0 will make it difficult to tell whether it is
148 * real FID or ost_id. */
149 oi->oi_fid.f_seq = FID_SEQ_ECHO;
150 oi->oi_fid.f_oid = id ?: 1;
151 oi->oi_fid.f_ver = 0;
153 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
155 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
156 GOTO(out, rc = -EPROTO);
158 /* Note: this check might be forced in 2.5 or 2.6, i.e.
159 * all of the requests are required to setup FLGROUP */
160 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
161 ostid_set_seq_mdt0(oi);
162 oa->o_valid |= OBD_MD_FLGROUP;
166 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
167 fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
168 GOTO(out, rc = -EPROTO);
170 rc = ostid_to_fid(&tti->tti_fid1, oi,
171 tsi->tsi_tgt->lut_lsd.lsd_osd_index);
172 if (unlikely(rc != 0))
175 oi->oi_fid = tti->tti_fid1;
181 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
182 tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
186 EXPORT_SYMBOL(tgt_validate_obdo);
188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
191 struct niobuf_remote *rnb;
192 struct obd_ioobj *ioo;
197 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
201 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
205 max_brw = ioobj_max_brw_get(ioo);
206 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
207 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
208 ": rc = %d\n", tgt_name(tsi->tsi_tgt),
209 obd_export_nid2str(tsi->tsi_exp), max_brw,
210 POSTID(oi), -EPROTO);
215 obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
216 RCL_CLIENT) / sizeof(*ioo);
217 if (obj_count == 0) {
218 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
220 } else if (obj_count > 1) {
221 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
226 if (ioo->ioo_bufcnt == 0) {
227 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
231 if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
232 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
233 "bulk has too many pages (%d)",
241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
243 struct ost_body *body;
244 struct req_capsule *pill = tsi->tsi_pill;
245 struct lustre_capa *capa;
246 struct lu_nodemap *nodemap;
251 body = req_capsule_client_get(pill, &RMF_OST_BODY);
255 rc = tgt_validate_obdo(tsi, &body->oa);
259 nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
261 body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
262 NODEMAP_CLIENT_TO_FS,
264 body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
265 NODEMAP_CLIENT_TO_FS,
268 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
269 capa = req_capsule_client_get(pill, &RMF_CAPA1);
271 CERROR("%s: OSSCAPA flag is set without capability\n",
272 tgt_name(tsi->tsi_tgt));
277 tsi->tsi_ost_body = body;
278 tsi->tsi_fid = body->oa.o_oi.oi_fid;
280 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
281 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
286 if (!(body->oa.o_valid & OBD_MD_FLID)) {
287 if (flags & HABEO_CORPUS) {
288 CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
289 "but OID/FID is mandatory with HABEO_CORPUS\n",
290 tgt_name(tsi->tsi_tgt));
297 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
300 * OST doesn't get object in advance for further use to prevent
301 * situations with nested object_find which is potential deadlock.
303 tsi->tsi_corpus = NULL;
308 * Do necessary preprocessing according to handler ->th_flags.
310 static int tgt_request_preprocess(struct tgt_session_info *tsi,
311 struct tgt_handler *h,
312 struct ptlrpc_request *req)
314 struct req_capsule *pill = tsi->tsi_pill;
315 __u32 flags = h->th_flags;
320 if (tsi->tsi_preprocessed)
323 LASSERT(h->th_act != NULL);
324 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
325 LASSERT(current->journal_info == NULL);
327 LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
329 if (h->th_fmt != NULL) {
330 req_capsule_set(pill, h->th_fmt);
331 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
332 rc = tgt_mdt_body_unpack(tsi, flags);
335 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
337 rc = tgt_ost_body_unpack(tsi, flags);
343 if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
346 if (flags & HABEO_CLAVIS) {
347 struct ldlm_request *dlm_req;
349 LASSERT(h->th_fmt != NULL);
351 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
352 if (dlm_req != NULL) {
353 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
355 dlm_req->lock_desc.l_policy_data.\
356 l_inodebits.bits == 0)) {
358 * Lock without inodebits makes no sense and
359 * will oops later in ldlm. If client miss to
360 * set such bits, do not trigger ASSERTION.
362 * For liblustre flock case, it maybe zero.
366 tsi->tsi_dlm_req = dlm_req;
372 tsi->tsi_preprocessed = 1;
377 * Invoke handler for this request opc. Also do necessary preprocessing
378 * (according to handler ->th_flags), and post-processing (setting of
379 * ->last_{xid,committed}).
381 static int tgt_handle_request0(struct tgt_session_info *tsi,
382 struct tgt_handler *h,
383 struct ptlrpc_request *req)
391 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
392 * to put same checks into handlers like mdt_close(), mdt_reint(),
393 * etc., without talking to mdt authors first. Checking same thing
394 * there again is useless and returning 0 error without packing reply
395 * is buggy! Handlers either pack reply or return error.
397 * We return 0 here and do not send any reply in order to emulate
398 * network failure. Do not send any reply in case any of NET related
399 * fail_id has occured.
401 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
404 rc = tgt_request_preprocess(tsi, h, req);
405 /* pack reply if reply format is fixed */
406 if (rc == 0 && h->th_flags & HABEO_REFERO) {
408 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
410 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
412 tsi->tsi_mdt_body->mbo_eadatasize);
413 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
415 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
418 rc = req_capsule_server_pack(tsi->tsi_pill);
421 if (likely(rc == 0)) {
423 * Process request, there can be two types of rc:
424 * 1) errors with msg unpack/pack, other failures outside the
425 * operation itself. This is counted as serious errors;
426 * 2) errors during fs operation, should be placed in rq_status
430 if (!is_serious(rc) &&
431 !req->rq_no_reply && req->rq_reply_state == NULL) {
432 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
433 "pack reply and returned 0 error\n",
434 tgt_name(tsi->tsi_tgt), h->th_name);
437 serious = is_serious(rc);
438 rc = clear_serious(rc);
446 * ELDLM_* codes which > 0 should be in rq_status only as well as
447 * all non-serious errors.
449 if (rc > 0 || !serious)
452 LASSERT(current->journal_info == NULL);
454 if (likely(rc == 0 && req->rq_export))
455 target_committed_to_req(req);
457 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
461 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
462 struct obd_device *obd, int *process)
464 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
471 case MDS_DONE_WRITING:
472 case MDS_SYNC: /* used in unmounting */
486 *process = target_queue_recovery_request(req, obd);
490 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
497 * Handle recovery. Return:
498 * +1: continue request processing;
499 * -ve: abort immediately with the given error code;
500 * 0: send reply with error code in req->rq_status;
502 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
506 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
511 case SEC_CTX_INIT_CONT:
516 if (!req->rq_export->exp_obd->obd_replayable)
519 /* sanity check: if the xid matches, the request must be marked as a
520 * resent or replayed */
521 if (req_xid_is_last(req)) {
522 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
523 (MSG_RESENT | MSG_REPLAY))) {
524 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
525 "last_xid, expected REPLAY or RESENT flag "
527 lustre_msg_get_flags(req->rq_reqmsg));
528 req->rq_status = -ENOTCONN;
532 /* else: note the opposite is not always true; a RESENT req after a
533 * failover will usually not match the last_xid, since it was likely
534 * never committed. A REPLAYed request will almost never match the
535 * last xid, however it could for a committed, but still retained,
538 /* Check for aborted recovery... */
539 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
543 DEBUG_REQ(D_INFO, req, "Got new replay");
544 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
546 if (rc != 0 || !should_process)
548 else if (should_process < 0) {
549 req->rq_status = should_process;
550 rc = ptlrpc_error(req);
557 /* Initial check for request, it is validation mostly */
558 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
560 struct tgt_handler *h;
561 struct tgt_opc_slice *s;
562 struct lu_target *tgt;
563 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
567 tgt = class_exp2tgt(req->rq_export);
569 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
570 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
573 /* opcode was not found in slice */
574 if (unlikely(s->tos_hs == NULL)) {
575 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
577 RETURN(ERR_PTR(-ENOTSUPP));
580 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
581 h = s->tos_hs + (opc - s->tos_opc_start);
582 if (unlikely(h->th_opc == 0)) {
583 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
584 RETURN(ERR_PTR(-ENOTSUPP));
590 int tgt_request_handle(struct ptlrpc_request *req)
592 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
594 struct lustre_msg *msg = req->rq_reqmsg;
595 struct tgt_handler *h;
596 struct lu_target *tgt;
597 int request_fail_id = 0;
598 __u32 opc = lustre_msg_get_opc(msg);
603 /* Refill the context, to make sure all thread keys are allocated */
604 lu_env_refill(req->rq_svc_thread->t_env);
606 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
607 tsi->tsi_pill = &req->rq_pill;
608 tsi->tsi_env = req->rq_svc_thread->t_env;
610 /* if request has export then get handlers slice from corresponding
611 * target, otherwise that should be connect operation */
612 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
613 opc == MGS_CONNECT) {
614 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
615 rc = target_handle_connect(req);
617 rc = ptlrpc_error(req);
620 /* recovery-small test 18c asks to drop connect reply */
621 if (unlikely(opc == OST_CONNECT &&
622 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
626 if (unlikely(!class_connected_export(req->rq_export))) {
627 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
628 opc, libcfs_id2str(req->rq_peer));
629 req->rq_status = -ENOTCONN;
630 rc = ptlrpc_error(req);
634 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
635 tsi->tsi_exp = req->rq_export;
636 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
637 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
639 tsi->tsi_jobid = NULL;
641 request_fail_id = tgt->lut_request_fail_id;
642 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
644 h = tgt_handler_find_check(req);
646 req->rq_status = PTR_ERR(h);
647 rc = ptlrpc_error(req);
651 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
654 rc = lustre_msg_check_version(msg, h->th_version);
656 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
657 " %08x, expecting %08x\n", tgt_name(tgt),
658 lustre_msg_get_version(msg), h->th_version);
659 req->rq_status = -EINVAL;
660 rc = ptlrpc_error(req);
664 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
665 if (likely(rc == 1)) {
666 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
668 rc = tgt_handle_request0(tsi, h, req);
674 req_capsule_fini(tsi->tsi_pill);
675 if (tsi->tsi_corpus != NULL) {
676 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
677 tsi->tsi_corpus = NULL;
681 EXPORT_SYMBOL(tgt_request_handle);
683 /** Assign high priority operations to the request if needed. */
684 int tgt_hpreq_handler(struct ptlrpc_request *req)
686 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
687 struct tgt_handler *h;
692 if (req->rq_export == NULL)
695 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
696 tsi->tsi_pill = &req->rq_pill;
697 tsi->tsi_env = req->rq_svc_thread->t_env;
698 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
699 tsi->tsi_exp = req->rq_export;
701 h = tgt_handler_find_check(req);
707 rc = tgt_request_preprocess(tsi, h, req);
708 if (unlikely(rc != 0))
711 if (h->th_hp != NULL)
715 EXPORT_SYMBOL(tgt_hpreq_handler);
717 void tgt_counter_incr(struct obd_export *exp, int opcode)
719 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
720 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
721 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
723 EXPORT_SYMBOL(tgt_counter_incr);
726 * Unified target generic handlers.
732 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
734 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
735 OBD_CONNECT_RMT_CLIENT_FORCE |
736 OBD_CONNECT_MDS_CAPA |
737 OBD_CONNECT_OSS_CAPA);
740 static int tgt_init_sec_level(struct ptlrpc_request *req)
742 struct lu_target *tgt = class_exp2tgt(req->rq_export);
743 char *client = libcfs_nid2str(req->rq_peer.nid);
744 struct obd_connect_data *data, *reply;
750 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
751 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
752 if (data == NULL || reply == NULL)
755 /* connection from MDT is always trusted */
756 if (req->rq_auth_usr_mdt) {
757 tgt_init_sec_none(reply);
761 /* no GSS support case */
762 if (!req->rq_auth_gss) {
763 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
764 CWARN("client %s -> target %s does not use GSS, "
765 "can not run under security level %d.\n",
766 client, tgt_name(tgt), tgt->lut_sec_level);
769 tgt_init_sec_none(reply);
774 /* old version case */
775 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
776 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
777 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
778 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
779 CWARN("client %s -> target %s uses old version, "
780 "can not run under security level %d.\n",
781 client, tgt_name(tgt), tgt->lut_sec_level);
784 CWARN("client %s -> target %s uses old version, "
785 "run under security level %d.\n",
786 client, tgt_name(tgt), tgt->lut_sec_level);
787 tgt_init_sec_none(reply);
792 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
794 if (!req->rq_auth_remote)
795 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
796 "asked to be remote.\n", client, tgt_name(tgt));
797 } else if (req->rq_auth_remote) {
799 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
800 "as remote by default.\n", client, tgt_name(tgt));
804 if (!tgt->lut_oss_capa) {
806 "client %s -> target %s is set as remote,"
807 " but OSS capabilities are not enabled: %d.\n",
808 client, tgt_name(tgt), tgt->lut_oss_capa);
812 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
813 CDEBUG(D_SEC, "client %s -> target %s: user is not "
814 "authenticated!\n", client, tgt_name(tgt));
820 switch (tgt->lut_sec_level) {
821 case LUSTRE_SEC_NONE:
824 "client %s -> target %s is set as remote, "
825 "can not run under security level %d.\n",
826 client, tgt_name(tgt), tgt->lut_sec_level);
829 tgt_init_sec_none(reply);
831 case LUSTRE_SEC_REMOTE:
833 tgt_init_sec_none(reply);
838 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
839 OBD_CONNECT_RMT_CLIENT_FORCE);
840 if (!tgt->lut_oss_capa)
841 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
842 if (!tgt->lut_mds_capa)
843 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
852 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
854 struct lu_target *tgt = class_exp2tgt(exp);
855 struct sptlrpc_flavor flvr;
859 LASSERT(tgt->lut_obd);
860 LASSERT(tgt->lut_slice);
862 /* always allow ECHO client */
863 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
864 LUSTRE_ECHO_NAME) == 0)) {
865 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
869 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
870 read_lock(&tgt->lut_sptlrpc_lock);
871 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
875 read_unlock(&tgt->lut_sptlrpc_lock);
877 spin_lock(&exp->exp_lock);
878 exp->exp_sp_peer = req->rq_sp_from;
879 exp->exp_flvr = flvr;
880 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
881 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
882 CERROR("%s: unauthorized rpc flavor %x from %s, "
883 "expect %x\n", tgt_name(tgt),
885 libcfs_nid2str(req->rq_peer.nid),
886 exp->exp_flvr.sf_rpc);
889 spin_unlock(&exp->exp_lock);
891 if (exp->exp_sp_peer != req->rq_sp_from) {
892 CERROR("%s: RPC source %s doesn't match %s\n",
894 sptlrpc_part2name(req->rq_sp_from),
895 sptlrpc_part2name(exp->exp_sp_peer));
898 rc = sptlrpc_target_export_check(exp, req);
905 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
907 struct sptlrpc_rule_set tmp_rset;
910 sptlrpc_rule_set_init(&tmp_rset);
911 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
913 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
918 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
920 write_lock(&tgt->lut_sptlrpc_lock);
921 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
922 tgt->lut_sptlrpc_rset = tmp_rset;
923 write_unlock(&tgt->lut_sptlrpc_lock);
927 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
929 int tgt_connect(struct tgt_session_info *tsi)
931 struct ptlrpc_request *req = tgt_ses_req(tsi);
932 struct obd_connect_data *reply;
937 rc = tgt_init_sec_level(req);
941 /* XXX: better to call this check right after getting new export but
942 * before last_rcvd slot allocation to avoid server load upon insecure
943 * connects. This is to be fixed after unifiyng all targets.
945 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
949 /* To avoid exposing partially initialized connection flags, changes up
950 * to this point have been staged in reply->ocd_connect_flags. Now that
951 * connection handling has completed successfully, atomically update
952 * the connect flags in the shared export data structure. LU-1623 */
953 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
954 spin_lock(&tsi->tsi_exp->exp_lock);
955 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
956 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
957 spin_unlock(&tsi->tsi_exp->exp_lock);
961 obd_disconnect(class_export_get(tsi->tsi_exp));
964 EXPORT_SYMBOL(tgt_connect);
966 int tgt_disconnect(struct tgt_session_info *tsi)
972 rc = target_handle_disconnect(tgt_ses_req(tsi));
974 RETURN(err_serious(rc));
978 EXPORT_SYMBOL(tgt_disconnect);
981 * Unified target OBD handlers
983 int tgt_obd_ping(struct tgt_session_info *tsi)
989 rc = target_handle_ping(tgt_ses_req(tsi));
991 RETURN(err_serious(rc));
995 EXPORT_SYMBOL(tgt_obd_ping);
997 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
999 return err_serious(-EOPNOTSUPP);
1001 EXPORT_SYMBOL(tgt_obd_log_cancel);
1003 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
1005 return err_serious(-EOPNOTSUPP);
1007 EXPORT_SYMBOL(tgt_obd_qc_callback);
1009 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1011 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1012 struct ptlrpc_request *req = tgt_ses_req(tsi);
1013 struct obd_export *exp = req->rq_export;
1014 struct ptlrpc_bulk_desc *desc;
1015 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1023 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1028 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1029 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1030 * which is rdpg->rp_count */
1031 nob = rdpg->rp_count;
1033 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1034 i++, tmpcount -= tmpsize) {
1035 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1036 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1039 LASSERT(desc->bd_nob == nob);
1040 rc = target_bulk_io(exp, desc, lwi);
1041 ptlrpc_free_bulk_pin(desc);
1044 EXPORT_SYMBOL(tgt_sendpage);
1047 * OBD_IDX_READ handler
1049 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1051 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1052 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1053 struct idx_info *req_ii, *rep_ii;
1058 memset(rdpg, 0, sizeof(*rdpg));
1059 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1061 /* extract idx_info buffer from request & reply */
1062 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1063 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1064 RETURN(err_serious(-EPROTO));
1066 rc = req_capsule_server_pack(tsi->tsi_pill);
1068 RETURN(err_serious(rc));
1070 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1072 RETURN(err_serious(-EFAULT));
1073 rep_ii->ii_magic = IDX_INFO_MAGIC;
1075 /* extract hash to start with */
1076 rdpg->rp_hash = req_ii->ii_hash_start;
1078 /* extract requested attributes */
1079 rdpg->rp_attrs = req_ii->ii_attrs;
1081 /* check that fid packed in request is valid and supported */
1082 if (!fid_is_sane(&req_ii->ii_fid))
1084 rep_ii->ii_fid = req_ii->ii_fid;
1087 rep_ii->ii_flags = req_ii->ii_flags;
1089 /* compute number of pages to allocate, ii_count is the number of 4KB
1091 if (req_ii->ii_count <= 0)
1092 GOTO(out, rc = -EFAULT);
1093 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1094 exp_max_brw_size(tsi->tsi_exp));
1095 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1097 /* allocate pages to store the containers */
1098 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1099 if (rdpg->rp_pages == NULL)
1100 GOTO(out, rc = -ENOMEM);
1101 for (i = 0; i < rdpg->rp_npages; i++) {
1102 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1103 if (rdpg->rp_pages[i] == NULL)
1104 GOTO(out, rc = -ENOMEM);
1107 /* populate pages with key/record pairs */
1108 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1112 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1113 "asked %d > %d\n", rc, rdpg->rp_count);
1115 /* send pages to client */
1116 rc = tgt_sendpage(tsi, rdpg, rc);
1121 if (rdpg->rp_pages) {
1122 for (i = 0; i < rdpg->rp_npages; i++)
1123 if (rdpg->rp_pages[i])
1124 __free_page(rdpg->rp_pages[i]);
1125 OBD_FREE(rdpg->rp_pages,
1126 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1130 EXPORT_SYMBOL(tgt_obd_idx_read);
1132 struct tgt_handler tgt_obd_handlers[] = {
1133 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1134 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1135 TGT_OBD_HDL_VAR(0, OBD_QC_CALLBACK, tgt_obd_qc_callback),
1136 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1138 EXPORT_SYMBOL(tgt_obd_handlers);
1140 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1141 struct dt_object *obj, __u64 start, __u64 end)
1147 /* if no objid is specified, it means "sync whole filesystem" */
1149 rc = dt_sync(env, tgt->lut_bottom);
1150 } else if (dt_version_get(env, obj) >
1151 tgt->lut_obd->obd_last_committed) {
1152 rc = dt_object_sync(env, obj, start, end);
1157 EXPORT_SYMBOL(tgt_sync);
1159 * Unified target DLM handlers.
1162 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1164 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1165 void *data, int flag)
1168 struct lu_target *tgt;
1169 struct dt_object *obj;
1175 tgt = class_exp2tgt(lock->l_export);
1177 if (flag == LDLM_CB_CANCELING &&
1178 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1179 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1180 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1181 lock->l_flags & LDLM_FL_CBPENDING))) {
1183 __u64 end = OBD_OBJECT_EOF;
1185 rc = lu_env_init(&env, LCT_DT_THREAD);
1186 if (unlikely(rc != 0))
1189 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1190 tgt->lut_lsd.lsd_osd_index);
1191 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1193 GOTO(err_env, rc = PTR_ERR(obj));
1195 if (!dt_object_exists(obj))
1196 GOTO(err_put, rc = -ENOENT);
1198 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1199 start = lock->l_policy_data.l_extent.start;
1200 end = lock->l_policy_data.l_extent.end;
1203 rc = tgt_sync(&env, tgt, obj, start, end);
1205 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1206 "cancel: rc = %d\n",
1207 tgt_name(tgt), PFID(&fid),
1208 lock->l_policy_data.l_extent.start,
1209 lock->l_policy_data.l_extent.end, rc);
1212 lu_object_put(&env, &obj->do_lu);
1217 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1221 static struct ldlm_callback_suite tgt_dlm_cbs = {
1222 .lcs_completion = ldlm_server_completion_ast,
1223 .lcs_blocking = tgt_blocking_ast,
1224 .lcs_glimpse = ldlm_server_glimpse_ast
1227 int tgt_enqueue(struct tgt_session_info *tsi)
1229 struct ptlrpc_request *req = tgt_ses_req(tsi);
1234 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1235 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1237 LASSERT(tsi->tsi_dlm_req != NULL);
1238 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1239 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1241 RETURN(err_serious(rc));
1243 RETURN(req->rq_status);
1245 EXPORT_SYMBOL(tgt_enqueue);
1247 int tgt_convert(struct tgt_session_info *tsi)
1249 struct ptlrpc_request *req = tgt_ses_req(tsi);
1253 LASSERT(tsi->tsi_dlm_req);
1254 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1256 RETURN(err_serious(rc));
1258 RETURN(req->rq_status);
1260 EXPORT_SYMBOL(tgt_convert);
1262 int tgt_bl_callback(struct tgt_session_info *tsi)
1264 return err_serious(-EOPNOTSUPP);
1266 EXPORT_SYMBOL(tgt_bl_callback);
1268 int tgt_cp_callback(struct tgt_session_info *tsi)
1270 return err_serious(-EOPNOTSUPP);
1272 EXPORT_SYMBOL(tgt_cp_callback);
1274 /* generic LDLM target handler */
1275 struct tgt_handler tgt_dlm_handlers[] = {
1276 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1277 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1278 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1279 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1281 EXPORT_SYMBOL(tgt_dlm_handlers);
1284 * Unified target LLOG handlers.
1286 int tgt_llog_open(struct tgt_session_info *tsi)
1292 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1296 EXPORT_SYMBOL(tgt_llog_open);
1298 int tgt_llog_close(struct tgt_session_info *tsi)
1304 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1308 EXPORT_SYMBOL(tgt_llog_close);
1311 int tgt_llog_destroy(struct tgt_session_info *tsi)
1317 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1321 EXPORT_SYMBOL(tgt_llog_destroy);
1323 int tgt_llog_read_header(struct tgt_session_info *tsi)
1329 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1333 EXPORT_SYMBOL(tgt_llog_read_header);
1335 int tgt_llog_next_block(struct tgt_session_info *tsi)
1341 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1345 EXPORT_SYMBOL(tgt_llog_next_block);
1347 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1353 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1357 EXPORT_SYMBOL(tgt_llog_prev_block);
1359 /* generic llog target handler */
1360 struct tgt_handler tgt_llog_handlers[] = {
1361 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1362 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1363 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1364 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1365 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1366 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1368 EXPORT_SYMBOL(tgt_llog_handlers);
1371 * sec context handlers
1373 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1374 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1379 struct tgt_handler tgt_sec_ctx_handlers[] = {
1380 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1381 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1382 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1384 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1386 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1387 struct dt_device *key,
1388 struct lfsck_request *lr,
1389 struct thandle *th) = NULL;
1391 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1393 struct lfsck_request *,
1396 tgt_lfsck_in_notify = notify;
1398 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1400 static int (*tgt_lfsck_query)(const struct lu_env *env,
1401 struct dt_device *key,
1402 struct lfsck_request *lr) = NULL;
1404 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1406 struct lfsck_request *))
1408 tgt_lfsck_query = query;
1410 EXPORT_SYMBOL(tgt_register_lfsck_query);
1412 /* LFSCK request handlers */
1413 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1415 const struct lu_env *env = tsi->tsi_env;
1416 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1417 struct lfsck_request *lr;
1421 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1425 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1430 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1432 struct lfsck_request *request;
1433 struct lfsck_reply *reply;
1437 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1438 if (request == NULL)
1441 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1445 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1446 reply->lr_status = rc;
1448 RETURN(rc < 0 ? rc : 0);
1451 struct tgt_handler tgt_lfsck_handlers[] = {
1452 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1453 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1455 EXPORT_SYMBOL(tgt_lfsck_handlers);
1458 * initialize per-thread page pool (bug 5137).
1460 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1462 struct tgt_thread_big_cache *tbc;
1466 LASSERT(thread != NULL);
1467 LASSERT(thread->t_data == NULL);
1469 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1472 thread->t_data = tbc;
1475 EXPORT_SYMBOL(tgt_io_thread_init);
1478 * free per-thread pool created by tgt_thread_init().
1480 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1482 struct tgt_thread_big_cache *tbc;
1486 LASSERT(thread != NULL);
1489 * be prepared to handle partially-initialized pools (because this is
1490 * called from ost_io_thread_init() for cleanup.
1492 tbc = thread->t_data;
1494 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1495 thread->t_data = NULL;
1499 EXPORT_SYMBOL(tgt_io_thread_done);
1501 * Helper function for getting server side [start, start+count] DLM lock
1502 * if asked by client.
1504 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1505 __u64 start, __u64 end, struct lustre_handle *lh,
1506 int mode, __u64 *flags)
1508 ldlm_policy_data_t policy;
1513 LASSERT(lh != NULL);
1514 LASSERT(ns != NULL);
1515 LASSERT(!lustre_handle_is_used(lh));
1517 policy.l_extent.gid = 0;
1518 policy.l_extent.start = start & CFS_PAGE_MASK;
1521 * If ->o_blocks is EOF it means "lock till the end of the file".
1522 * Otherwise, it's size of an extent or hole being punched (in bytes).
1524 if (end == OBD_OBJECT_EOF || end < start)
1525 policy.l_extent.end = OBD_OBJECT_EOF;
1527 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1529 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1530 flags, ldlm_blocking_ast,
1531 ldlm_completion_ast, ldlm_glimpse_ast,
1532 NULL, 0, LVB_T_NONE, NULL, lh);
1533 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1535 EXPORT_SYMBOL(tgt_extent_lock);
1537 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1539 LASSERT(lustre_handle_is_used(lh));
1540 ldlm_lock_decref(lh, mode);
1542 EXPORT_SYMBOL(tgt_extent_unlock);
1544 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1545 struct obd_ioobj *obj, struct niobuf_remote *nb,
1546 struct lustre_handle *lh, int mode)
1549 int nrbufs = obj->ioo_bufcnt;
1554 LASSERT(mode == LCK_PR || mode == LCK_PW);
1555 LASSERT(!lustre_handle_is_used(lh));
1557 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1560 for (i = 1; i < nrbufs; i++)
1561 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1564 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1565 nb[nrbufs - 1].rnb_offset +
1566 nb[nrbufs - 1].rnb_len - 1,
1569 EXPORT_SYMBOL(tgt_brw_lock);
1571 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1572 struct lustre_handle *lh, int mode)
1576 LASSERT(mode == LCK_PR || mode == LCK_PW);
1577 LASSERT((obj->ioo_bufcnt > 0 &&
1578 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1579 lustre_handle_is_used(lh));
1581 if (lustre_handle_is_used(lh))
1582 tgt_extent_unlock(lh, mode);
1585 EXPORT_SYMBOL(tgt_brw_unlock);
1587 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1588 struct ptlrpc_bulk_desc *desc, int opc,
1589 cksum_type_t cksum_type)
1591 struct cfs_crypto_hash_desc *hdesc;
1592 unsigned int bufsize;
1594 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1597 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1598 if (IS_ERR(hdesc)) {
1599 CERROR("%s: unable to initialize checksum hash %s\n",
1600 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1601 return PTR_ERR(hdesc);
1604 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1605 for (i = 0; i < desc->bd_iov_count; i++) {
1606 /* corrupt the data before we compute the checksum, to
1607 * simulate a client->OST data error */
1608 if (i == 0 && opc == OST_WRITE &&
1609 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1610 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1611 int len = desc->bd_iov[i].kiov_len;
1612 struct page *np = tgt_page_to_corrupt;
1613 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1616 char *ptr2 = kmap(np) + off;
1618 memcpy(ptr2, ptr, len);
1619 memcpy(ptr2, "bad3", min(4, len));
1621 desc->bd_iov[i].kiov_page = np;
1623 CERROR("%s: can't alloc page for corruption\n",
1627 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1628 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1629 desc->bd_iov[i].kiov_len);
1631 /* corrupt the data after we compute the checksum, to
1632 * simulate an OST->client data error */
1633 if (i == 0 && opc == OST_READ &&
1634 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1635 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1636 int len = desc->bd_iov[i].kiov_len;
1637 struct page *np = tgt_page_to_corrupt;
1638 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1641 char *ptr2 = kmap(np) + off;
1643 memcpy(ptr2, ptr, len);
1644 memcpy(ptr2, "bad4", min(4, len));
1646 desc->bd_iov[i].kiov_page = np;
1648 CERROR("%s: can't alloc page for corruption\n",
1654 bufsize = sizeof(cksum);
1655 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1660 int tgt_brw_read(struct tgt_session_info *tsi)
1662 struct ptlrpc_request *req = tgt_ses_req(tsi);
1663 struct ptlrpc_bulk_desc *desc = NULL;
1664 struct obd_export *exp = tsi->tsi_exp;
1665 struct niobuf_remote *remote_nb;
1666 struct niobuf_local *local_nb;
1667 struct obd_ioobj *ioo;
1668 struct ost_body *body, *repbody;
1669 struct l_wait_info lwi;
1670 struct lustre_handle lockh = { 0 };
1671 int npages, nob = 0, rc, i, no_reply = 0;
1672 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1676 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1677 CERROR("%s: deny read request from %s to portal %u\n",
1678 tgt_name(tsi->tsi_tgt),
1679 obd_export_nid2str(req->rq_export),
1680 ptlrpc_req2svc(req)->srv_req_portal);
1684 req->rq_bulk_read = 1;
1686 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1689 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1690 cfs_fail_val : (obd_timeout + 1) / 4);
1692 /* Check if there is eviction in progress, and if so, wait for it to
1694 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1695 /* We do not care how long it takes */
1696 lwi = LWI_INTR(NULL, NULL);
1697 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1698 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1702 /* There must be big cache in current thread to process this request
1703 * if it is NULL then something went wrong and it wasn't allocated,
1704 * report -ENOMEM in that case */
1708 body = tsi->tsi_ost_body;
1709 LASSERT(body != NULL);
1711 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1712 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1714 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1715 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1717 local_nb = tbc->local;
1719 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1720 remote_nb, &lockh, LCK_PR);
1725 * If getting the lock took more time than
1726 * client was willing to wait, drop it. b=11330
1728 if (cfs_time_current_sec() > req->rq_deadline ||
1729 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1731 CERROR("Dropping timed-out read from %s because locking"
1732 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1733 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1734 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1735 req->rq_deadline - req->rq_arrival_time.tv_sec);
1736 GOTO(out_lock, rc = -ETIMEDOUT);
1739 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1740 repbody->oa = body->oa;
1742 npages = PTLRPC_MAX_BRW_PAGES;
1743 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1744 ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1748 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1749 BULK_PUT_SOURCE, OST_BULK_PORTAL);
1751 GOTO(out_commitrw, rc = -ENOMEM);
1754 for (i = 0; i < npages; i++) {
1755 int page_rc = local_nb[i].lnb_rc;
1763 if (page_rc != 0) { /* some data! */
1764 LASSERT(local_nb[i].lnb_page != NULL);
1765 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1766 local_nb[i].lnb_page_offset,
1770 if (page_rc != local_nb[i].lnb_len) { /* short read */
1771 /* All subsequent pages should be 0 */
1772 while (++i < npages)
1773 LASSERT(local_nb[i].lnb_rc == 0);
1778 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1779 cksum_type_t cksum_type =
1780 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1781 body->oa.o_flags : 0);
1782 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1783 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1784 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1785 OST_READ, cksum_type);
1786 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1787 repbody->oa.o_cksum);
1789 repbody->oa.o_valid = 0;
1791 /* We're finishing using body->oa as an input variable */
1793 /* Check if client was evicted while we were doing i/o before touching
1795 if (likely(rc == 0 &&
1796 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1797 rc = target_bulk_io(exp, desc, &lwi);
1802 /* Must commit after prep above in all cases */
1803 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1804 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1807 tgt_drop_id(exp, &repbody->oa);
1809 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1811 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1812 ptlrpc_free_bulk_nopin(desc);
1817 ptlrpc_lprocfs_brw(req, nob);
1818 } else if (no_reply) {
1819 req->rq_no_reply = 1;
1820 /* reply out callback would free */
1821 ptlrpc_req_drop_rs(req);
1822 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1823 "client will retry: rc %d\n",
1824 exp->exp_obd->obd_name,
1825 obd_uuid2str(&exp->exp_client_uuid),
1826 obd_export_nid2str(exp), rc);
1828 /* send a bulk after reply to simulate a network delay or reordering
1830 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1831 wait_queue_head_t waitq;
1832 struct l_wait_info lwi1;
1834 CDEBUG(D_INFO, "reorder BULK\n");
1835 init_waitqueue_head(&waitq);
1837 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1838 l_wait_event(waitq, 0, &lwi1);
1839 target_bulk_io(exp, desc, &lwi);
1840 ptlrpc_free_bulk_nopin(desc);
1845 EXPORT_SYMBOL(tgt_brw_read);
1847 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1848 struct ptlrpc_bulk_desc *desc,
1849 struct niobuf_local *local_nb, int npages,
1850 obd_count client_cksum, obd_count server_cksum,
1853 struct obd_export *exp = req->rq_export;
1854 struct ost_body *body;
1858 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1859 LASSERT(body != NULL);
1861 if (req->rq_peer.nid == desc->bd_sender) {
1865 router = libcfs_nid2str(desc->bd_sender);
1869 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1870 client_cksum, server_cksum);
1874 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1875 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1876 "]: client csum %x, server csum %x\n",
1877 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1879 body->oa.o_valid & OBD_MD_FLFID ?
1880 body->oa.o_parent_seq : (__u64)0,
1881 body->oa.o_valid & OBD_MD_FLFID ?
1882 body->oa.o_parent_oid : 0,
1883 body->oa.o_valid & OBD_MD_FLFID ?
1884 body->oa.o_parent_ver : 0,
1885 POSTID(&body->oa.o_oi),
1886 local_nb[0].lnb_file_offset,
1887 local_nb[npages-1].lnb_file_offset +
1888 local_nb[npages - 1].lnb_len - 1,
1889 client_cksum, server_cksum);
1892 int tgt_brw_write(struct tgt_session_info *tsi)
1894 struct ptlrpc_request *req = tgt_ses_req(tsi);
1895 struct ptlrpc_bulk_desc *desc = NULL;
1896 struct obd_export *exp = req->rq_export;
1897 struct niobuf_remote *remote_nb;
1898 struct niobuf_local *local_nb;
1899 struct obd_ioobj *ioo;
1900 struct ost_body *body, *repbody;
1901 struct l_wait_info lwi;
1902 struct lustre_handle lockh = {0};
1904 int objcount, niocount, npages;
1906 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1907 bool no_reply = false, mmap;
1908 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1912 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1913 CERROR("%s: deny write request from %s to portal %u\n",
1914 tgt_name(tsi->tsi_tgt),
1915 obd_export_nid2str(req->rq_export),
1916 ptlrpc_req2svc(req)->srv_req_portal);
1917 RETURN(err_serious(-EPROTO));
1920 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1921 RETURN(err_serious(-ENOSPC));
1922 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1923 RETURN(err_serious(-EROFS));
1925 req->rq_bulk_write = 1;
1927 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1928 RETURN(err_serious(-EIO));
1929 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1930 RETURN(err_serious(-EFAULT));
1932 /* pause before transaction has been started */
1933 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1934 cfs_fail_val : (obd_timeout + 1) / 4);
1936 /* There must be big cache in current thread to process this request
1937 * if it is NULL then something went wrong and it wasn't allocated,
1938 * report -ENOMEM in that case */
1942 body = tsi->tsi_ost_body;
1943 LASSERT(body != NULL);
1945 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1946 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1948 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1949 RCL_CLIENT) / sizeof(*ioo);
1951 for (niocount = i = 0; i < objcount; i++)
1952 niocount += ioo[i].ioo_bufcnt;
1954 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1955 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1956 if (niocount != req_capsule_get_size(&req->rq_pill,
1957 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1959 RETURN(err_serious(-EPROTO));
1961 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1962 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1963 memory_pressure_set();
1965 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1966 niocount * sizeof(*rcs));
1967 rc = req_capsule_server_pack(&req->rq_pill);
1969 GOTO(out, rc = err_serious(rc));
1971 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1972 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1974 local_nb = tbc->local;
1976 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1977 remote_nb, &lockh, LCK_PW);
1982 * If getting the lock took more time than
1983 * client was willing to wait, drop it. b=11330
1985 if (cfs_time_current_sec() > req->rq_deadline ||
1986 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1988 CERROR("%s: Dropping timed-out write from %s because locking "
1989 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1990 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1991 POSTID(&ioo->ioo_oid),
1992 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1993 req->rq_deadline - req->rq_arrival_time.tv_sec);
1994 GOTO(out_lock, rc = -ETIMEDOUT);
1997 /* Because we already sync grant info with client when reconnect,
1998 * grant info will be cleared for resent req, then fed_grant and
1999 * total_grant will not be modified in following preprw_write */
2000 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2001 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2002 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2005 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2006 if (repbody == NULL)
2007 GOTO(out_lock, rc = -ENOMEM);
2008 repbody->oa = body->oa;
2010 npages = PTLRPC_MAX_BRW_PAGES;
2011 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2012 objcount, ioo, remote_nb, &npages, local_nb, NULL,
2017 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2018 BULK_GET_SINK, OST_BULK_PORTAL);
2020 GOTO(skip_transfer, rc = -ENOMEM);
2022 /* NB Having prepped, we must commit... */
2023 for (i = 0; i < npages; i++)
2024 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2025 local_nb[i].lnb_page_offset,
2026 local_nb[i].lnb_len);
2028 rc = sptlrpc_svc_prep_bulk(req, desc);
2030 GOTO(skip_transfer, rc);
2032 rc = target_bulk_io(exp, desc, &lwi);
2036 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2037 static int cksum_counter;
2039 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2040 cksum_type = cksum_type_unpack(body->oa.o_flags);
2042 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2043 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2044 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2045 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2046 OST_WRITE, cksum_type);
2049 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2050 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2051 body->oa.o_flags & OBD_FL_MMAP);
2053 tgt_warn_on_cksum(req, desc, local_nb, npages,
2055 repbody->oa.o_cksum, mmap);
2057 } else if ((cksum_counter & (-cksum_counter)) ==
2059 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2060 cksum_counter, libcfs_id2str(req->rq_peer),
2061 repbody->oa.o_cksum);
2065 /* Must commit after prep above in all cases */
2066 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2067 objcount, ioo, remote_nb, npages, local_nb, NULL,
2069 if (rc == -ENOTCONN)
2070 /* quota acquire process has been given up because
2071 * either the client has been evicted or the client
2072 * has timed out the request already */
2076 * Disable sending mtime back to the client. If the client locked the
2077 * whole object, then it has already updated the mtime on its side,
2078 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2080 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2085 /* set per-requested niobuf return codes */
2086 for (i = j = 0; i < niocount; i++) {
2087 int len = remote_nb[i].rnb_len;
2092 LASSERT(j < npages);
2093 if (local_nb[j].lnb_rc < 0)
2094 rcs[i] = local_nb[j].lnb_rc;
2095 len -= local_nb[j].lnb_len;
2100 LASSERT(j == npages);
2101 ptlrpc_lprocfs_brw(req, nob);
2103 tgt_drop_id(exp, &repbody->oa);
2106 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2108 ptlrpc_free_bulk_nopin(desc);
2111 req->rq_no_reply = 1;
2112 /* reply out callback would free */
2113 ptlrpc_req_drop_rs(req);
2114 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2115 "client will retry: rc %d\n",
2116 exp->exp_obd->obd_name,
2117 obd_uuid2str(&exp->exp_client_uuid),
2118 obd_export_nid2str(exp), rc);
2120 memory_pressure_clr();
2123 EXPORT_SYMBOL(tgt_brw_write);