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
37 #include <linux/user_namespace.h>
38 #ifdef HAVE_UIDGID_HEADER
39 # include <linux/uidgid.h>
43 #include <obd_class.h>
44 #include <obd_cksum.h>
45 #include <md_object.h>
46 #include <lustre_lfsck.h>
47 #include <lustre_nodemap.h>
49 #include "tgt_internal.h"
51 char *tgt_name(struct lu_target *tgt)
53 LASSERT(tgt->lut_obd != NULL);
54 return tgt->lut_obd->obd_name;
56 EXPORT_SYMBOL(tgt_name);
59 * Generic code handling requests that have struct mdt_body passed in:
61 * - extract mdt_body from request and save it in @tsi, if present;
63 * - create lu_object, corresponding to the fid in mdt_body, and save it in
66 * - if HABEO_CORPUS flag is set for this request type check whether object
67 * actually exists on storage (lu_object_exists()).
70 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
72 const struct mdt_body *body;
73 struct lu_object *obj;
74 struct req_capsule *pill = tsi->tsi_pill;
79 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
83 tsi->tsi_mdt_body = body;
85 if (!(body->mbo_valid & OBD_MD_FLID))
88 /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
89 * in any case in pre-2.5 clients. Fix that here if needed */
90 if (unlikely(fid_is_zero(&body->mbo_fid1)))
93 if (!fid_is_sane(&body->mbo_fid1)) {
94 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
95 PFID(&body->mbo_fid1));
99 obj = lu_object_find(tsi->tsi_env,
100 &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
101 &body->mbo_fid1, NULL);
103 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
104 lu_object_put(tsi->tsi_env, obj);
107 tsi->tsi_corpus = obj;
114 tsi->tsi_fid = body->mbo_fid1;
120 * Validate oa from client.
121 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
123 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
124 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
125 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
126 * whether this is oi_fid or real ostid. So it will check
127 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
128 * c. Old FID-disable osc will send IDIF.
129 * d. new FID-enable osc/osp will send normal FID.
131 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
132 * be used for LAST_ID file, and only being accessed inside OST now.
134 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
136 struct ost_id *oi = &oa->o_oi;
137 u64 seq = ostid_seq(oi);
138 u64 id = ostid_id(oi);
142 if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
143 fid_seq_is_echo(seq))) {
144 /* Sigh 2.[123] client still sends echo req with oi_id = 0
145 * during create, and we will reset this to 1, since this
146 * oi_id is basically useless in the following create process,
147 * but oi_id == 0 will make it difficult to tell whether it is
148 * real FID or ost_id. */
149 oi->oi_fid.f_seq = FID_SEQ_ECHO;
150 oi->oi_fid.f_oid = id ?: 1;
151 oi->oi_fid.f_ver = 0;
153 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
155 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
156 GOTO(out, rc = -EPROTO);
158 /* Note: this check might be forced in 2.5 or 2.6, i.e.
159 * all of the requests are required to setup FLGROUP */
160 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
161 ostid_set_seq_mdt0(oi);
162 oa->o_valid |= OBD_MD_FLGROUP;
166 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
167 fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
168 GOTO(out, rc = -EPROTO);
170 rc = ostid_to_fid(&tti->tti_fid1, oi,
171 tsi->tsi_tgt->lut_lsd.lsd_osd_index);
172 if (unlikely(rc != 0))
175 oi->oi_fid = tti->tti_fid1;
181 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
182 tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
186 EXPORT_SYMBOL(tgt_validate_obdo);
188 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
191 struct niobuf_remote *rnb;
192 struct obd_ioobj *ioo;
197 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
201 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
205 max_brw = ioobj_max_brw_get(ioo);
206 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
207 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
208 ": rc = %d\n", tgt_name(tsi->tsi_tgt),
209 obd_export_nid2str(tsi->tsi_exp), max_brw,
210 POSTID(oi), -EPROTO);
215 obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
216 RCL_CLIENT) / sizeof(*ioo);
217 if (obj_count == 0) {
218 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
220 } else if (obj_count > 1) {
221 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
226 if (ioo->ioo_bufcnt == 0) {
227 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
231 if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
232 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
233 "bulk has too many pages (%d)",
241 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
243 struct ost_body *body;
244 struct req_capsule *pill = tsi->tsi_pill;
245 struct lu_nodemap *nodemap;
250 body = req_capsule_client_get(pill, &RMF_OST_BODY);
254 rc = tgt_validate_obdo(tsi, &body->oa);
258 nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
260 body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
261 NODEMAP_CLIENT_TO_FS,
263 body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
264 NODEMAP_CLIENT_TO_FS,
267 tsi->tsi_ost_body = body;
268 tsi->tsi_fid = body->oa.o_oi.oi_fid;
270 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
271 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
276 if (!(body->oa.o_valid & OBD_MD_FLID)) {
277 if (flags & HABEO_CORPUS) {
278 CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
279 "but OID/FID is mandatory with HABEO_CORPUS\n",
280 tgt_name(tsi->tsi_tgt));
287 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
290 * OST doesn't get object in advance for further use to prevent
291 * situations with nested object_find which is potential deadlock.
293 tsi->tsi_corpus = NULL;
298 * Do necessary preprocessing according to handler ->th_flags.
300 static int tgt_request_preprocess(struct tgt_session_info *tsi,
301 struct tgt_handler *h,
302 struct ptlrpc_request *req)
304 struct req_capsule *pill = tsi->tsi_pill;
305 __u32 flags = h->th_flags;
310 if (tsi->tsi_preprocessed)
313 LASSERT(h->th_act != NULL);
314 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
315 LASSERT(current->journal_info == NULL);
317 LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
319 if (h->th_fmt != NULL) {
320 req_capsule_set(pill, h->th_fmt);
321 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
322 rc = tgt_mdt_body_unpack(tsi, flags);
325 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
327 rc = tgt_ost_body_unpack(tsi, flags);
333 if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
336 if (flags & HABEO_CLAVIS) {
337 struct ldlm_request *dlm_req;
339 LASSERT(h->th_fmt != NULL);
341 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
342 if (dlm_req != NULL) {
343 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
345 dlm_req->lock_desc.l_policy_data.\
346 l_inodebits.bits == 0)) {
348 * Lock without inodebits makes no sense and
349 * will oops later in ldlm. If client miss to
350 * set such bits, do not trigger ASSERTION.
352 * For liblustre flock case, it maybe zero.
356 tsi->tsi_dlm_req = dlm_req;
362 tsi->tsi_preprocessed = 1;
367 * Invoke handler for this request opc. Also do necessary preprocessing
368 * (according to handler ->th_flags), and post-processing (setting of
369 * ->last_{xid,committed}).
371 static int tgt_handle_request0(struct tgt_session_info *tsi,
372 struct tgt_handler *h,
373 struct ptlrpc_request *req)
381 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
382 * to put same checks into handlers like mdt_close(), mdt_reint(),
383 * etc., without talking to mdt authors first. Checking same thing
384 * there again is useless and returning 0 error without packing reply
385 * is buggy! Handlers either pack reply or return error.
387 * We return 0 here and do not send any reply in order to emulate
388 * network failure. Do not send any reply in case any of NET related
389 * fail_id has occured.
391 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
394 rc = tgt_request_preprocess(tsi, h, req);
395 /* pack reply if reply format is fixed */
396 if (rc == 0 && h->th_flags & HABEO_REFERO) {
398 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
400 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
402 tsi->tsi_mdt_body->mbo_eadatasize);
403 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
405 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
408 rc = req_capsule_server_pack(tsi->tsi_pill);
411 if (likely(rc == 0)) {
413 * Process request, there can be two types of rc:
414 * 1) errors with msg unpack/pack, other failures outside the
415 * operation itself. This is counted as serious errors;
416 * 2) errors during fs operation, should be placed in rq_status
420 if (!is_serious(rc) &&
421 !req->rq_no_reply && req->rq_reply_state == NULL) {
422 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
423 "pack reply and returned 0 error\n",
424 tgt_name(tsi->tsi_tgt), h->th_name);
427 serious = is_serious(rc);
428 rc = clear_serious(rc);
436 * ELDLM_* codes which > 0 should be in rq_status only as well as
437 * all non-serious errors.
439 if (rc > 0 || !serious)
442 LASSERT(current->journal_info == NULL);
444 if (likely(rc == 0 && req->rq_export))
445 target_committed_to_req(req);
447 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
451 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
452 struct obd_device *obd, int *process)
454 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
461 case MDS_SYNC: /* used in unmounting */
475 case MDS_HSM_PROGRESS:
476 case MDS_HSM_STATE_SET:
477 case MDS_HSM_REQUEST:
478 *process = target_queue_recovery_request(req, obd);
482 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
489 * Handle recovery. Return:
490 * +1: continue request processing;
491 * -ve: abort immediately with the given error code;
492 * 0: send reply with error code in req->rq_status;
494 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
498 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
503 case SEC_CTX_INIT_CONT:
508 if (!req->rq_export->exp_obd->obd_replayable)
511 /* sanity check: if the xid matches, the request must be marked as a
512 * resent or replayed */
513 if (req_xid_is_last(req)) {
514 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
515 (MSG_RESENT | MSG_REPLAY))) {
516 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
517 "last_xid, expected REPLAY or RESENT flag "
519 lustre_msg_get_flags(req->rq_reqmsg));
520 req->rq_status = -ENOTCONN;
524 /* else: note the opposite is not always true; a RESENT req after a
525 * failover will usually not match the last_xid, since it was likely
526 * never committed. A REPLAYed request will almost never match the
527 * last xid, however it could for a committed, but still retained,
530 /* Check for aborted recovery... */
531 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
535 DEBUG_REQ(D_INFO, req, "Got new replay");
536 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
538 if (rc != 0 || !should_process)
540 else if (should_process < 0) {
541 req->rq_status = should_process;
542 rc = ptlrpc_error(req);
549 /* Initial check for request, it is validation mostly */
550 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
552 struct tgt_handler *h;
553 struct tgt_opc_slice *s;
554 struct lu_target *tgt;
555 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
559 tgt = class_exp2tgt(req->rq_export);
560 if (unlikely(tgt == NULL)) {
561 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
562 class_exp2obd(req->rq_export)->obd_name);
563 RETURN(ERR_PTR(-EINVAL));
566 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
567 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
570 /* opcode was not found in slice */
571 if (unlikely(s->tos_hs == NULL)) {
572 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
574 RETURN(ERR_PTR(-ENOTSUPP));
577 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
578 h = s->tos_hs + (opc - s->tos_opc_start);
579 if (unlikely(h->th_opc == 0)) {
580 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
581 RETURN(ERR_PTR(-ENOTSUPP));
587 int tgt_request_handle(struct ptlrpc_request *req)
589 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
591 struct lustre_msg *msg = req->rq_reqmsg;
592 struct tgt_handler *h;
593 struct lu_target *tgt;
594 int request_fail_id = 0;
595 __u32 opc = lustre_msg_get_opc(msg);
600 /* Refill the context, to make sure all thread keys are allocated */
601 lu_env_refill(req->rq_svc_thread->t_env);
603 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
604 tsi->tsi_pill = &req->rq_pill;
605 tsi->tsi_env = req->rq_svc_thread->t_env;
607 /* if request has export then get handlers slice from corresponding
608 * target, otherwise that should be connect operation */
609 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
610 opc == MGS_CONNECT) {
611 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
612 rc = target_handle_connect(req);
614 rc = ptlrpc_error(req);
617 /* recovery-small test 18c asks to drop connect reply */
618 if (unlikely(opc == OST_CONNECT &&
619 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
623 if (unlikely(!class_connected_export(req->rq_export))) {
624 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
625 opc, libcfs_id2str(req->rq_peer));
626 req->rq_status = -ENOTCONN;
627 rc = ptlrpc_error(req);
631 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
632 tsi->tsi_exp = req->rq_export;
633 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
634 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
636 tsi->tsi_jobid = NULL;
639 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
640 class_exp2obd(req->rq_export)->obd_name);
641 req->rq_status = -EINVAL;
642 rc = ptlrpc_error(req);
646 request_fail_id = tgt->lut_request_fail_id;
647 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
649 h = tgt_handler_find_check(req);
651 req->rq_status = PTR_ERR(h);
652 rc = ptlrpc_error(req);
656 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
659 rc = lustre_msg_check_version(msg, h->th_version);
661 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
662 " %08x, expecting %08x\n", tgt_name(tgt),
663 lustre_msg_get_version(msg), h->th_version);
664 req->rq_status = -EINVAL;
665 rc = ptlrpc_error(req);
669 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
670 if (likely(rc == 1)) {
671 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
673 rc = tgt_handle_request0(tsi, h, req);
679 req_capsule_fini(tsi->tsi_pill);
680 if (tsi->tsi_corpus != NULL) {
681 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
682 tsi->tsi_corpus = NULL;
686 EXPORT_SYMBOL(tgt_request_handle);
688 /** Assign high priority operations to the request if needed. */
689 int tgt_hpreq_handler(struct ptlrpc_request *req)
691 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
692 struct tgt_handler *h;
697 if (req->rq_export == NULL)
700 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
701 tsi->tsi_pill = &req->rq_pill;
702 tsi->tsi_env = req->rq_svc_thread->t_env;
703 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
704 tsi->tsi_exp = req->rq_export;
706 h = tgt_handler_find_check(req);
712 rc = tgt_request_preprocess(tsi, h, req);
713 if (unlikely(rc != 0))
716 if (h->th_hp != NULL)
720 EXPORT_SYMBOL(tgt_hpreq_handler);
722 void tgt_counter_incr(struct obd_export *exp, int opcode)
724 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
725 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
726 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
728 EXPORT_SYMBOL(tgt_counter_incr);
731 * Unified target generic handlers.
737 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
739 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
740 OBD_CONNECT_RMT_CLIENT_FORCE);
743 static int tgt_init_sec_level(struct ptlrpc_request *req)
745 struct lu_target *tgt = class_exp2tgt(req->rq_export);
747 struct obd_connect_data *data, *reply;
752 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
753 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
754 if (data == NULL || reply == NULL)
757 /* connection from MDT is always trusted */
758 if (req->rq_auth_usr_mdt) {
759 tgt_init_sec_none(reply);
763 if (unlikely(tgt == NULL)) {
764 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
765 class_exp2obd(req->rq_export)->obd_name);
769 client = libcfs_nid2str(req->rq_peer.nid);
770 /* no GSS support case */
771 if (!req->rq_auth_gss) {
772 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
773 CWARN("client %s -> target %s does not use GSS, "
774 "can not run under security level %d.\n",
775 client, tgt_name(tgt), tgt->lut_sec_level);
778 tgt_init_sec_none(reply);
783 /* old version case */
784 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
785 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
786 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
787 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
788 CWARN("client %s -> target %s uses old version, "
789 "can not run under security level %d.\n",
790 client, tgt_name(tgt), tgt->lut_sec_level);
793 CWARN("client %s -> target %s uses old version, "
794 "run under security level %d.\n",
795 client, tgt_name(tgt), tgt->lut_sec_level);
796 tgt_init_sec_none(reply);
801 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
803 if (!req->rq_auth_remote)
804 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
805 "asked to be remote.\n", client, tgt_name(tgt));
806 } else if (req->rq_auth_remote) {
808 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
809 "as remote by default.\n", client, tgt_name(tgt));
813 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
814 CDEBUG(D_SEC, "client %s -> target %s: user is not "
815 "authenticated!\n", client, tgt_name(tgt));
821 switch (tgt->lut_sec_level) {
822 case LUSTRE_SEC_NONE:
825 "client %s -> target %s is set as remote, "
826 "can not run under security level %d.\n",
827 client, tgt_name(tgt), tgt->lut_sec_level);
830 tgt_init_sec_none(reply);
832 case LUSTRE_SEC_REMOTE:
834 tgt_init_sec_none(reply);
839 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
840 OBD_CONNECT_RMT_CLIENT_FORCE);
841 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
842 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
851 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
853 struct lu_target *tgt = class_exp2tgt(exp);
854 struct sptlrpc_flavor flvr;
858 LASSERT(tgt->lut_obd);
859 LASSERT(tgt->lut_slice);
861 /* always allow ECHO client */
862 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
863 LUSTRE_ECHO_NAME) == 0)) {
864 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
868 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
869 read_lock(&tgt->lut_sptlrpc_lock);
870 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
874 read_unlock(&tgt->lut_sptlrpc_lock);
876 spin_lock(&exp->exp_lock);
877 exp->exp_sp_peer = req->rq_sp_from;
878 exp->exp_flvr = flvr;
879 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
880 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
881 CERROR("%s: unauthorized rpc flavor %x from %s, "
882 "expect %x\n", tgt_name(tgt),
884 libcfs_nid2str(req->rq_peer.nid),
885 exp->exp_flvr.sf_rpc);
888 spin_unlock(&exp->exp_lock);
890 if (exp->exp_sp_peer != req->rq_sp_from) {
891 CERROR("%s: RPC source %s doesn't match %s\n",
893 sptlrpc_part2name(req->rq_sp_from),
894 sptlrpc_part2name(exp->exp_sp_peer));
897 rc = sptlrpc_target_export_check(exp, req);
904 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
906 struct sptlrpc_rule_set tmp_rset;
909 if (unlikely(tgt == NULL)) {
910 CERROR("No target passed");
914 sptlrpc_rule_set_init(&tmp_rset);
915 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
917 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
922 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
924 write_lock(&tgt->lut_sptlrpc_lock);
925 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
926 tgt->lut_sptlrpc_rset = tmp_rset;
927 write_unlock(&tgt->lut_sptlrpc_lock);
931 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
933 int tgt_connect(struct tgt_session_info *tsi)
935 struct ptlrpc_request *req = tgt_ses_req(tsi);
936 struct obd_connect_data *reply;
941 rc = tgt_init_sec_level(req);
945 /* XXX: better to call this check right after getting new export but
946 * before last_rcvd slot allocation to avoid server load upon insecure
947 * connects. This is to be fixed after unifiyng all targets.
949 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
953 /* To avoid exposing partially initialized connection flags, changes up
954 * to this point have been staged in reply->ocd_connect_flags. Now that
955 * connection handling has completed successfully, atomically update
956 * the connect flags in the shared export data structure. LU-1623 */
957 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
958 spin_lock(&tsi->tsi_exp->exp_lock);
959 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
960 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
961 spin_unlock(&tsi->tsi_exp->exp_lock);
965 obd_disconnect(class_export_get(tsi->tsi_exp));
968 EXPORT_SYMBOL(tgt_connect);
970 int tgt_disconnect(struct tgt_session_info *tsi)
976 rc = target_handle_disconnect(tgt_ses_req(tsi));
978 RETURN(err_serious(rc));
982 EXPORT_SYMBOL(tgt_disconnect);
985 * Unified target OBD handlers
987 int tgt_obd_ping(struct tgt_session_info *tsi)
993 rc = target_handle_ping(tgt_ses_req(tsi));
995 RETURN(err_serious(rc));
999 EXPORT_SYMBOL(tgt_obd_ping);
1001 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1003 return err_serious(-EOPNOTSUPP);
1006 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1008 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1009 struct ptlrpc_request *req = tgt_ses_req(tsi);
1010 struct obd_export *exp = req->rq_export;
1011 struct ptlrpc_bulk_desc *desc;
1012 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1020 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1021 PTLRPC_BULK_PUT_SOURCE |
1022 PTLRPC_BULK_BUF_KIOV,
1024 &ptlrpc_bulk_kiov_pin_ops);
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 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1040 LASSERT(desc->bd_nob == nob);
1041 rc = target_bulk_io(exp, desc, lwi);
1042 ptlrpc_free_bulk(desc);
1045 EXPORT_SYMBOL(tgt_sendpage);
1048 * OBD_IDX_READ handler
1050 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1052 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1053 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1054 struct idx_info *req_ii, *rep_ii;
1059 memset(rdpg, 0, sizeof(*rdpg));
1060 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1062 /* extract idx_info buffer from request & reply */
1063 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1064 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1065 RETURN(err_serious(-EPROTO));
1067 rc = req_capsule_server_pack(tsi->tsi_pill);
1069 RETURN(err_serious(rc));
1071 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1073 RETURN(err_serious(-EFAULT));
1074 rep_ii->ii_magic = IDX_INFO_MAGIC;
1076 /* extract hash to start with */
1077 rdpg->rp_hash = req_ii->ii_hash_start;
1079 /* extract requested attributes */
1080 rdpg->rp_attrs = req_ii->ii_attrs;
1082 /* check that fid packed in request is valid and supported */
1083 if (!fid_is_sane(&req_ii->ii_fid))
1085 rep_ii->ii_fid = req_ii->ii_fid;
1088 rep_ii->ii_flags = req_ii->ii_flags;
1090 /* compute number of pages to allocate, ii_count is the number of 4KB
1092 if (req_ii->ii_count <= 0)
1093 GOTO(out, rc = -EFAULT);
1094 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1095 exp_max_brw_size(tsi->tsi_exp));
1096 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1098 /* allocate pages to store the containers */
1099 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1100 if (rdpg->rp_pages == NULL)
1101 GOTO(out, rc = -ENOMEM);
1102 for (i = 0; i < rdpg->rp_npages; i++) {
1103 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1104 if (rdpg->rp_pages[i] == NULL)
1105 GOTO(out, rc = -ENOMEM);
1108 /* populate pages with key/record pairs */
1109 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1113 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1114 "asked %d > %d\n", rc, rdpg->rp_count);
1116 /* send pages to client */
1117 rc = tgt_sendpage(tsi, rdpg, rc);
1122 if (rdpg->rp_pages) {
1123 for (i = 0; i < rdpg->rp_npages; i++)
1124 if (rdpg->rp_pages[i])
1125 __free_page(rdpg->rp_pages[i]);
1126 OBD_FREE(rdpg->rp_pages,
1127 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
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 (0, OBD_IDX_READ, tgt_obd_idx_read)
1137 EXPORT_SYMBOL(tgt_obd_handlers);
1139 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1140 struct dt_object *obj, __u64 start, __u64 end)
1146 /* if no objid is specified, it means "sync whole filesystem" */
1148 rc = dt_sync(env, tgt->lut_bottom);
1149 } else if (dt_version_get(env, obj) >
1150 tgt->lut_obd->obd_last_committed) {
1151 rc = dt_object_sync(env, obj, start, end);
1156 EXPORT_SYMBOL(tgt_sync);
1158 * Unified target DLM handlers.
1161 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1163 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1164 void *data, int flag)
1167 struct lu_target *tgt;
1168 struct dt_object *obj;
1174 tgt = class_exp2tgt(lock->l_export);
1176 if (unlikely(tgt == NULL)) {
1177 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1178 class_exp2obd(lock->l_export)->obd_name);
1182 if (flag == LDLM_CB_CANCELING &&
1183 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1184 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1185 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1186 lock->l_flags & LDLM_FL_CBPENDING))) {
1188 __u64 end = OBD_OBJECT_EOF;
1190 rc = lu_env_init(&env, LCT_DT_THREAD);
1191 if (unlikely(rc != 0))
1194 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1195 tgt->lut_lsd.lsd_osd_index);
1196 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1198 GOTO(err_env, rc = PTR_ERR(obj));
1200 if (!dt_object_exists(obj))
1201 GOTO(err_put, rc = -ENOENT);
1203 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1204 start = lock->l_policy_data.l_extent.start;
1205 end = lock->l_policy_data.l_extent.end;
1208 rc = tgt_sync(&env, tgt, obj, start, end);
1210 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1211 "cancel: rc = %d\n",
1212 tgt_name(tgt), PFID(&fid),
1213 lock->l_policy_data.l_extent.start,
1214 lock->l_policy_data.l_extent.end, rc);
1217 lu_object_put(&env, &obj->do_lu);
1222 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1226 static struct ldlm_callback_suite tgt_dlm_cbs = {
1227 .lcs_completion = ldlm_server_completion_ast,
1228 .lcs_blocking = tgt_blocking_ast,
1229 .lcs_glimpse = ldlm_server_glimpse_ast
1232 int tgt_enqueue(struct tgt_session_info *tsi)
1234 struct ptlrpc_request *req = tgt_ses_req(tsi);
1239 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1240 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1242 LASSERT(tsi->tsi_dlm_req != NULL);
1243 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1244 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1246 RETURN(err_serious(rc));
1248 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1250 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1253 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1256 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1259 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1262 RETURN(req->rq_status);
1264 EXPORT_SYMBOL(tgt_enqueue);
1266 int tgt_convert(struct tgt_session_info *tsi)
1268 struct ptlrpc_request *req = tgt_ses_req(tsi);
1272 LASSERT(tsi->tsi_dlm_req);
1273 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1275 RETURN(err_serious(rc));
1277 RETURN(req->rq_status);
1280 int tgt_bl_callback(struct tgt_session_info *tsi)
1282 return err_serious(-EOPNOTSUPP);
1285 int tgt_cp_callback(struct tgt_session_info *tsi)
1287 return err_serious(-EOPNOTSUPP);
1290 /* generic LDLM target handler */
1291 struct tgt_handler tgt_dlm_handlers[] = {
1292 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1293 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1294 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1295 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1297 EXPORT_SYMBOL(tgt_dlm_handlers);
1300 * Unified target LLOG handlers.
1302 int tgt_llog_open(struct tgt_session_info *tsi)
1308 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1312 EXPORT_SYMBOL(tgt_llog_open);
1314 int tgt_llog_close(struct tgt_session_info *tsi)
1320 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1324 EXPORT_SYMBOL(tgt_llog_close);
1327 int tgt_llog_destroy(struct tgt_session_info *tsi)
1333 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1338 int tgt_llog_read_header(struct tgt_session_info *tsi)
1344 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1348 EXPORT_SYMBOL(tgt_llog_read_header);
1350 int tgt_llog_next_block(struct tgt_session_info *tsi)
1356 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1360 EXPORT_SYMBOL(tgt_llog_next_block);
1362 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1368 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1372 EXPORT_SYMBOL(tgt_llog_prev_block);
1374 /* generic llog target handler */
1375 struct tgt_handler tgt_llog_handlers[] = {
1376 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1377 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1378 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1379 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1380 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1381 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1383 EXPORT_SYMBOL(tgt_llog_handlers);
1386 * sec context handlers
1388 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1389 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1394 struct tgt_handler tgt_sec_ctx_handlers[] = {
1395 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1396 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1397 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1399 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1401 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1402 struct dt_device *key,
1403 struct lfsck_request *lr,
1404 struct thandle *th) = NULL;
1406 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1408 struct lfsck_request *,
1411 tgt_lfsck_in_notify = notify;
1413 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1415 static int (*tgt_lfsck_query)(const struct lu_env *env,
1416 struct dt_device *key,
1417 struct lfsck_request *lr) = NULL;
1419 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1421 struct lfsck_request *))
1423 tgt_lfsck_query = query;
1425 EXPORT_SYMBOL(tgt_register_lfsck_query);
1427 /* LFSCK request handlers */
1428 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1430 const struct lu_env *env = tsi->tsi_env;
1431 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1432 struct lfsck_request *lr;
1436 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1440 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1445 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1447 struct lfsck_request *request;
1448 struct lfsck_reply *reply;
1452 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1453 if (request == NULL)
1456 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1460 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1461 reply->lr_status = rc;
1463 RETURN(rc < 0 ? rc : 0);
1466 struct tgt_handler tgt_lfsck_handlers[] = {
1467 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1468 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1470 EXPORT_SYMBOL(tgt_lfsck_handlers);
1473 * initialize per-thread page pool (bug 5137).
1475 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1477 struct tgt_thread_big_cache *tbc;
1481 LASSERT(thread != NULL);
1482 LASSERT(thread->t_data == NULL);
1484 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1487 thread->t_data = tbc;
1490 EXPORT_SYMBOL(tgt_io_thread_init);
1493 * free per-thread pool created by tgt_thread_init().
1495 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1497 struct tgt_thread_big_cache *tbc;
1501 LASSERT(thread != NULL);
1504 * be prepared to handle partially-initialized pools (because this is
1505 * called from ost_io_thread_init() for cleanup.
1507 tbc = thread->t_data;
1509 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1510 thread->t_data = NULL;
1514 EXPORT_SYMBOL(tgt_io_thread_done);
1516 * Helper function for getting server side [start, start+count] DLM lock
1517 * if asked by client.
1519 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1520 __u64 start, __u64 end, struct lustre_handle *lh,
1521 int mode, __u64 *flags)
1523 ldlm_policy_data_t policy;
1528 LASSERT(lh != NULL);
1529 LASSERT(ns != NULL);
1530 LASSERT(!lustre_handle_is_used(lh));
1532 policy.l_extent.gid = 0;
1533 policy.l_extent.start = start & PAGE_MASK;
1536 * If ->o_blocks is EOF it means "lock till the end of the file".
1537 * Otherwise, it's size of an extent or hole being punched (in bytes).
1539 if (end == OBD_OBJECT_EOF || end < start)
1540 policy.l_extent.end = OBD_OBJECT_EOF;
1542 policy.l_extent.end = end | ~PAGE_MASK;
1544 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1545 flags, ldlm_blocking_ast,
1546 ldlm_completion_ast, ldlm_glimpse_ast,
1547 NULL, 0, LVB_T_NONE, NULL, lh);
1548 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1550 EXPORT_SYMBOL(tgt_extent_lock);
1552 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1554 LASSERT(lustre_handle_is_used(lh));
1555 ldlm_lock_decref(lh, mode);
1557 EXPORT_SYMBOL(tgt_extent_unlock);
1559 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1560 struct obd_ioobj *obj, struct niobuf_remote *nb,
1561 struct lustre_handle *lh, int mode)
1564 int nrbufs = obj->ioo_bufcnt;
1569 LASSERT(mode == LCK_PR || mode == LCK_PW);
1570 LASSERT(!lustre_handle_is_used(lh));
1572 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1575 for (i = 1; i < nrbufs; i++)
1576 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1579 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1580 nb[nrbufs - 1].rnb_offset +
1581 nb[nrbufs - 1].rnb_len - 1,
1585 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1586 struct lustre_handle *lh, int mode)
1590 LASSERT(mode == LCK_PR || mode == LCK_PW);
1591 LASSERT((obj->ioo_bufcnt > 0 &&
1592 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1593 lustre_handle_is_used(lh));
1595 if (lustre_handle_is_used(lh))
1596 tgt_extent_unlock(lh, mode);
1600 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1601 struct ptlrpc_bulk_desc *desc, int opc,
1602 cksum_type_t cksum_type)
1604 struct cfs_crypto_hash_desc *hdesc;
1605 unsigned int bufsize;
1607 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1610 LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1612 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1613 if (IS_ERR(hdesc)) {
1614 CERROR("%s: unable to initialize checksum hash %s\n",
1615 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1616 return PTR_ERR(hdesc);
1619 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1620 for (i = 0; i < desc->bd_iov_count; i++) {
1621 /* corrupt the data before we compute the checksum, to
1622 * simulate a client->OST data error */
1623 if (i == 0 && opc == OST_WRITE &&
1624 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1625 int off = BD_GET_KIOV(desc, i).kiov_offset &
1627 int len = BD_GET_KIOV(desc, i).kiov_len;
1628 struct page *np = tgt_page_to_corrupt;
1629 char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1632 char *ptr2 = kmap(np) + off;
1634 memcpy(ptr2, ptr, len);
1635 memcpy(ptr2, "bad3", min(4, len));
1637 BD_GET_KIOV(desc, i).kiov_page = np;
1639 CERROR("%s: can't alloc page for corruption\n",
1643 cfs_crypto_hash_update_page(hdesc,
1644 BD_GET_KIOV(desc, i).kiov_page,
1645 BD_GET_KIOV(desc, i).kiov_offset &
1647 BD_GET_KIOV(desc, i).kiov_len);
1649 /* corrupt the data after we compute the checksum, to
1650 * simulate an OST->client data error */
1651 if (i == 0 && opc == OST_READ &&
1652 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1653 int off = BD_GET_KIOV(desc, i).kiov_offset
1655 int len = BD_GET_KIOV(desc, i).kiov_len;
1656 struct page *np = tgt_page_to_corrupt;
1658 kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1661 char *ptr2 = kmap(np) + off;
1663 memcpy(ptr2, ptr, len);
1664 memcpy(ptr2, "bad4", min(4, len));
1666 BD_GET_KIOV(desc, i).kiov_page = np;
1668 CERROR("%s: can't alloc page for corruption\n",
1674 bufsize = sizeof(cksum);
1675 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1680 int tgt_brw_read(struct tgt_session_info *tsi)
1682 struct ptlrpc_request *req = tgt_ses_req(tsi);
1683 struct ptlrpc_bulk_desc *desc = NULL;
1684 struct obd_export *exp = tsi->tsi_exp;
1685 struct niobuf_remote *remote_nb;
1686 struct niobuf_local *local_nb;
1687 struct obd_ioobj *ioo;
1688 struct ost_body *body, *repbody;
1689 struct l_wait_info lwi;
1690 struct lustre_handle lockh = { 0 };
1691 int npages, nob = 0, rc, i, no_reply = 0;
1692 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1696 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1697 CERROR("%s: deny read request from %s to portal %u\n",
1698 tgt_name(tsi->tsi_tgt),
1699 obd_export_nid2str(req->rq_export),
1700 ptlrpc_req2svc(req)->srv_req_portal);
1704 req->rq_bulk_read = 1;
1706 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1709 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1710 cfs_fail_val : (obd_timeout + 1) / 4);
1712 /* Check if there is eviction in progress, and if so, wait for it to
1714 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1715 /* We do not care how long it takes */
1716 lwi = LWI_INTR(NULL, NULL);
1717 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1718 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1722 /* There must be big cache in current thread to process this request
1723 * if it is NULL then something went wrong and it wasn't allocated,
1724 * report -ENOMEM in that case */
1728 body = tsi->tsi_ost_body;
1729 LASSERT(body != NULL);
1731 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1732 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1734 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1735 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1737 local_nb = tbc->local;
1739 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1740 remote_nb, &lockh, LCK_PR);
1745 * If getting the lock took more time than
1746 * client was willing to wait, drop it. b=11330
1748 if (cfs_time_current_sec() > req->rq_deadline ||
1749 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1751 CERROR("Dropping timed-out read from %s because locking"
1752 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1753 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1754 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1755 req->rq_deadline - req->rq_arrival_time.tv_sec);
1756 GOTO(out_lock, rc = -ETIMEDOUT);
1759 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1760 repbody->oa = body->oa;
1762 npages = PTLRPC_MAX_BRW_PAGES;
1763 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1764 ioo, remote_nb, &npages, local_nb);
1768 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1769 PTLRPC_BULK_PUT_SOURCE |
1770 PTLRPC_BULK_BUF_KIOV,
1772 &ptlrpc_bulk_kiov_nopin_ops);
1774 GOTO(out_commitrw, rc = -ENOMEM);
1777 for (i = 0; i < npages; i++) {
1778 int page_rc = local_nb[i].lnb_rc;
1786 if (page_rc != 0) { /* some data! */
1787 LASSERT(local_nb[i].lnb_page != NULL);
1788 desc->bd_frag_ops->add_kiov_frag
1789 (desc, local_nb[i].lnb_page,
1790 local_nb[i].lnb_page_offset,
1794 if (page_rc != local_nb[i].lnb_len) { /* short read */
1795 /* All subsequent pages should be 0 */
1796 while (++i < npages)
1797 LASSERT(local_nb[i].lnb_rc == 0);
1802 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1803 cksum_type_t cksum_type =
1804 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1805 body->oa.o_flags : 0);
1806 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1807 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1808 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1809 OST_READ, cksum_type);
1810 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1811 repbody->oa.o_cksum);
1813 repbody->oa.o_valid = 0;
1815 /* We're finishing using body->oa as an input variable */
1817 /* Check if client was evicted while we were doing i/o before touching
1819 if (likely(rc == 0 &&
1820 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1821 rc = target_bulk_io(exp, desc, &lwi);
1826 /* Must commit after prep above in all cases */
1827 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1828 remote_nb, npages, local_nb, rc);
1830 tgt_drop_id(exp, &repbody->oa);
1832 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1834 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1835 ptlrpc_free_bulk(desc);
1840 ptlrpc_lprocfs_brw(req, nob);
1841 } else if (no_reply) {
1842 req->rq_no_reply = 1;
1843 /* reply out callback would free */
1844 ptlrpc_req_drop_rs(req);
1845 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1846 "client will retry: rc %d\n",
1847 exp->exp_obd->obd_name,
1848 obd_uuid2str(&exp->exp_client_uuid),
1849 obd_export_nid2str(exp), rc);
1851 /* send a bulk after reply to simulate a network delay or reordering
1853 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1854 wait_queue_head_t waitq;
1855 struct l_wait_info lwi1;
1857 CDEBUG(D_INFO, "reorder BULK\n");
1858 init_waitqueue_head(&waitq);
1860 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1861 l_wait_event(waitq, 0, &lwi1);
1862 target_bulk_io(exp, desc, &lwi);
1863 ptlrpc_free_bulk(desc);
1868 EXPORT_SYMBOL(tgt_brw_read);
1870 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1871 struct ptlrpc_bulk_desc *desc,
1872 struct niobuf_local *local_nb, int npages,
1873 u32 client_cksum, u32 server_cksum,
1876 struct obd_export *exp = req->rq_export;
1877 struct ost_body *body;
1881 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1882 LASSERT(body != NULL);
1884 if (req->rq_peer.nid != desc->bd_sender) {
1886 router = libcfs_nid2str(desc->bd_sender);
1890 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1891 client_cksum, server_cksum);
1895 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1896 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1897 "]: client csum %x, server csum %x\n",
1898 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1900 body->oa.o_valid & OBD_MD_FLFID ?
1901 body->oa.o_parent_seq : (__u64)0,
1902 body->oa.o_valid & OBD_MD_FLFID ?
1903 body->oa.o_parent_oid : 0,
1904 body->oa.o_valid & OBD_MD_FLFID ?
1905 body->oa.o_parent_ver : 0,
1906 POSTID(&body->oa.o_oi),
1907 local_nb[0].lnb_file_offset,
1908 local_nb[npages-1].lnb_file_offset +
1909 local_nb[npages - 1].lnb_len - 1,
1910 client_cksum, server_cksum);
1913 int tgt_brw_write(struct tgt_session_info *tsi)
1915 struct ptlrpc_request *req = tgt_ses_req(tsi);
1916 struct ptlrpc_bulk_desc *desc = NULL;
1917 struct obd_export *exp = req->rq_export;
1918 struct niobuf_remote *remote_nb;
1919 struct niobuf_local *local_nb;
1920 struct obd_ioobj *ioo;
1921 struct ost_body *body, *repbody;
1922 struct l_wait_info lwi;
1923 struct lustre_handle lockh = {0};
1925 int objcount, niocount, npages;
1927 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1928 bool no_reply = false, mmap;
1929 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1933 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1934 CERROR("%s: deny write request from %s to portal %u\n",
1935 tgt_name(tsi->tsi_tgt),
1936 obd_export_nid2str(req->rq_export),
1937 ptlrpc_req2svc(req)->srv_req_portal);
1938 RETURN(err_serious(-EPROTO));
1941 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1942 RETURN(err_serious(-ENOSPC));
1943 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1944 RETURN(err_serious(-EROFS));
1946 req->rq_bulk_write = 1;
1948 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1949 RETURN(err_serious(-EIO));
1950 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1951 RETURN(err_serious(-EFAULT));
1953 /* pause before transaction has been started */
1954 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1955 cfs_fail_val : (obd_timeout + 1) / 4);
1957 /* There must be big cache in current thread to process this request
1958 * if it is NULL then something went wrong and it wasn't allocated,
1959 * report -ENOMEM in that case */
1963 body = tsi->tsi_ost_body;
1964 LASSERT(body != NULL);
1966 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1967 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1969 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1970 RCL_CLIENT) / sizeof(*ioo);
1972 for (niocount = i = 0; i < objcount; i++)
1973 niocount += ioo[i].ioo_bufcnt;
1975 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1976 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1977 if (niocount != req_capsule_get_size(&req->rq_pill,
1978 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1980 RETURN(err_serious(-EPROTO));
1982 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1983 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1984 memory_pressure_set();
1986 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1987 niocount * sizeof(*rcs));
1988 rc = req_capsule_server_pack(&req->rq_pill);
1990 GOTO(out, rc = err_serious(rc));
1992 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1993 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1995 local_nb = tbc->local;
1997 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1998 remote_nb, &lockh, LCK_PW);
2003 * If getting the lock took more time than
2004 * client was willing to wait, drop it. b=11330
2006 if (cfs_time_current_sec() > req->rq_deadline ||
2007 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2009 CERROR("%s: Dropping timed-out write from %s because locking "
2010 "object "DOSTID" took %ld seconds (limit was %ld).\n",
2011 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2012 POSTID(&ioo->ioo_oid),
2013 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2014 req->rq_deadline - req->rq_arrival_time.tv_sec);
2015 GOTO(out_lock, rc = -ETIMEDOUT);
2018 /* Because we already sync grant info with client when reconnect,
2019 * grant info will be cleared for resent req, then fed_grant and
2020 * total_grant will not be modified in following preprw_write */
2021 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2022 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2023 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2026 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2027 if (repbody == NULL)
2028 GOTO(out_lock, rc = -ENOMEM);
2029 repbody->oa = body->oa;
2031 npages = PTLRPC_MAX_BRW_PAGES;
2032 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2033 objcount, ioo, remote_nb, &npages, local_nb);
2037 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2038 PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2040 &ptlrpc_bulk_kiov_nopin_ops);
2042 GOTO(skip_transfer, rc = -ENOMEM);
2044 /* NB Having prepped, we must commit... */
2045 for (i = 0; i < npages; i++)
2046 desc->bd_frag_ops->add_kiov_frag(desc,
2047 local_nb[i].lnb_page,
2048 local_nb[i].lnb_page_offset,
2049 local_nb[i].lnb_len);
2051 rc = sptlrpc_svc_prep_bulk(req, desc);
2053 GOTO(skip_transfer, rc);
2055 rc = target_bulk_io(exp, desc, &lwi);
2059 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2060 static int cksum_counter;
2062 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2063 cksum_type = cksum_type_unpack(body->oa.o_flags);
2065 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2066 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2067 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2068 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2069 OST_WRITE, cksum_type);
2072 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2073 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2074 body->oa.o_flags & OBD_FL_MMAP);
2076 tgt_warn_on_cksum(req, desc, local_nb, npages,
2078 repbody->oa.o_cksum, mmap);
2080 } else if ((cksum_counter & (-cksum_counter)) ==
2082 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2083 cksum_counter, libcfs_id2str(req->rq_peer),
2084 repbody->oa.o_cksum);
2088 /* Must commit after prep above in all cases */
2089 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2090 objcount, ioo, remote_nb, npages, local_nb, rc);
2091 if (rc == -ENOTCONN)
2092 /* quota acquire process has been given up because
2093 * either the client has been evicted or the client
2094 * has timed out the request already */
2098 * Disable sending mtime back to the client. If the client locked the
2099 * whole object, then it has already updated the mtime on its side,
2100 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2102 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2107 /* set per-requested niobuf return codes */
2108 for (i = j = 0; i < niocount; i++) {
2109 int len = remote_nb[i].rnb_len;
2114 LASSERT(j < npages);
2115 if (local_nb[j].lnb_rc < 0)
2116 rcs[i] = local_nb[j].lnb_rc;
2117 len -= local_nb[j].lnb_len;
2122 LASSERT(j == npages);
2123 ptlrpc_lprocfs_brw(req, nob);
2125 tgt_drop_id(exp, &repbody->oa);
2128 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2130 ptlrpc_free_bulk(desc);
2133 req->rq_no_reply = 1;
2134 /* reply out callback would free */
2135 ptlrpc_req_drop_rs(req);
2136 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2137 "client will retry: rc %d\n",
2138 exp->exp_obd->obd_name,
2139 obd_uuid2str(&exp->exp_client_uuid),
2140 obd_export_nid2str(exp), rc);
2142 memory_pressure_clr();
2145 EXPORT_SYMBOL(tgt_brw_write);