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)
377 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
382 /* When dealing with sec context requests, no export is associated yet,
383 * because these requests are sent before *_CONNECT requests.
384 * A NULL req->rq_export means the normal *_common_slice handlers will
385 * not be called, because there is no reference to the target.
386 * So deal with them by hand and jump directly to target_send_reply().
390 case SEC_CTX_INIT_CONT:
396 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
397 * to put same checks into handlers like mdt_close(), mdt_reint(),
398 * etc., without talking to mdt authors first. Checking same thing
399 * there again is useless and returning 0 error without packing reply
400 * is buggy! Handlers either pack reply or return error.
402 * We return 0 here and do not send any reply in order to emulate
403 * network failure. Do not send any reply in case any of NET related
404 * fail_id has occured.
406 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
409 rc = tgt_request_preprocess(tsi, h, req);
410 /* pack reply if reply format is fixed */
411 if (rc == 0 && h->th_flags & HABEO_REFERO) {
413 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
415 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
417 tsi->tsi_mdt_body->mbo_eadatasize);
418 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
420 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
423 rc = req_capsule_server_pack(tsi->tsi_pill);
426 if (likely(rc == 0)) {
428 * Process request, there can be two types of rc:
429 * 1) errors with msg unpack/pack, other failures outside the
430 * operation itself. This is counted as serious errors;
431 * 2) errors during fs operation, should be placed in rq_status
435 if (!is_serious(rc) &&
436 !req->rq_no_reply && req->rq_reply_state == NULL) {
437 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
438 "pack reply and returned 0 error\n",
439 tgt_name(tsi->tsi_tgt), h->th_name);
442 serious = is_serious(rc);
443 rc = clear_serious(rc);
451 * ELDLM_* codes which > 0 should be in rq_status only as well as
452 * all non-serious errors.
454 if (rc > 0 || !serious)
457 LASSERT(current->journal_info == NULL);
459 if (likely(rc == 0 && req->rq_export))
460 target_committed_to_req(req);
463 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
467 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
468 struct obd_device *obd, int *process)
470 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
477 case MDS_SYNC: /* used in unmounting */
491 case MDS_HSM_PROGRESS:
492 case MDS_HSM_STATE_SET:
493 case MDS_HSM_REQUEST:
494 *process = target_queue_recovery_request(req, obd);
498 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
505 * Handle recovery. Return:
506 * +1: continue request processing;
507 * -ve: abort immediately with the given error code;
508 * 0: send reply with error code in req->rq_status;
510 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
514 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
519 case SEC_CTX_INIT_CONT:
524 if (!req->rq_export->exp_obd->obd_replayable)
527 /* sanity check: if the xid matches, the request must be marked as a
528 * resent or replayed */
529 if (req_xid_is_last(req)) {
530 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
531 (MSG_RESENT | MSG_REPLAY))) {
532 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
533 "last_xid, expected REPLAY or RESENT flag "
535 lustre_msg_get_flags(req->rq_reqmsg));
536 req->rq_status = -ENOTCONN;
540 /* else: note the opposite is not always true; a RESENT req after a
541 * failover will usually not match the last_xid, since it was likely
542 * never committed. A REPLAYed request will almost never match the
543 * last xid, however it could for a committed, but still retained,
546 /* Check for aborted recovery... */
547 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
551 DEBUG_REQ(D_INFO, req, "Got new replay");
552 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
554 if (rc != 0 || !should_process)
556 else if (should_process < 0) {
557 req->rq_status = should_process;
558 rc = ptlrpc_error(req);
565 /* Initial check for request, it is validation mostly */
566 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
568 struct tgt_handler *h;
569 struct tgt_opc_slice *s;
570 struct lu_target *tgt;
571 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
575 tgt = class_exp2tgt(req->rq_export);
576 if (unlikely(tgt == NULL)) {
577 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
578 class_exp2obd(req->rq_export)->obd_name);
579 RETURN(ERR_PTR(-EINVAL));
582 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
583 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
586 /* opcode was not found in slice */
587 if (unlikely(s->tos_hs == NULL)) {
588 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
590 RETURN(ERR_PTR(-ENOTSUPP));
593 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
594 h = s->tos_hs + (opc - s->tos_opc_start);
595 if (unlikely(h->th_opc == 0)) {
596 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
597 RETURN(ERR_PTR(-ENOTSUPP));
603 int tgt_request_handle(struct ptlrpc_request *req)
605 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
607 struct lustre_msg *msg = req->rq_reqmsg;
608 struct tgt_handler *h;
609 struct lu_target *tgt;
610 int request_fail_id = 0;
611 __u32 opc = lustre_msg_get_opc(msg);
616 /* Refill the context, to make sure all thread keys are allocated */
617 lu_env_refill(req->rq_svc_thread->t_env);
619 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
620 tsi->tsi_pill = &req->rq_pill;
621 tsi->tsi_env = req->rq_svc_thread->t_env;
623 /* if request has export then get handlers slice from corresponding
624 * target, otherwise that should be connect operation */
625 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
626 opc == MGS_CONNECT) {
627 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
628 rc = target_handle_connect(req);
630 rc = ptlrpc_error(req);
633 /* recovery-small test 18c asks to drop connect reply */
634 if (unlikely(opc == OST_CONNECT &&
635 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
639 if (unlikely(!class_connected_export(req->rq_export))) {
640 if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
641 opc == SEC_CTX_FINI) {
642 /* sec context initialization has to be handled
643 * by hand in tgt_handle_request0() */
644 tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
646 GOTO(handle_recov, rc = 0);
648 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
649 opc, libcfs_id2str(req->rq_peer));
650 req->rq_status = -ENOTCONN;
651 rc = ptlrpc_error(req);
655 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
656 tsi->tsi_exp = req->rq_export;
657 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
658 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
660 tsi->tsi_jobid = NULL;
663 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
664 class_exp2obd(req->rq_export)->obd_name);
665 req->rq_status = -EINVAL;
666 rc = ptlrpc_error(req);
670 /* check request's xid is consistent with export's last_xid */
671 if (req->rq_export != NULL) {
672 __u64 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
674 req->rq_export->exp_last_xid = last_xid;
675 if (req->rq_xid == 0 ||
676 req->rq_xid <= req->rq_export->exp_last_xid) {
677 DEBUG_REQ(D_ERROR, req,
678 "Unexpected xid %llx vs. last_xid %llx\n",
679 req->rq_xid, req->rq_export->exp_last_xid);
680 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 93, 0)
683 req->rq_status = -EPROTO;
684 rc = ptlrpc_error(req);
689 request_fail_id = tgt->lut_request_fail_id;
690 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
692 h = tgt_handler_find_check(req);
694 req->rq_status = PTR_ERR(h);
695 rc = ptlrpc_error(req);
699 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
702 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
705 rc = lustre_msg_check_version(msg, h->th_version);
707 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
708 " %08x, expecting %08x\n", tgt_name(tgt),
709 lustre_msg_get_version(msg), h->th_version);
710 req->rq_status = -EINVAL;
711 rc = ptlrpc_error(req);
716 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
717 if (likely(rc == 1)) {
718 rc = tgt_handle_request0(tsi, h, req);
724 req_capsule_fini(tsi->tsi_pill);
725 if (tsi->tsi_corpus != NULL) {
726 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
727 tsi->tsi_corpus = NULL;
731 EXPORT_SYMBOL(tgt_request_handle);
733 /** Assign high priority operations to the request if needed. */
734 int tgt_hpreq_handler(struct ptlrpc_request *req)
736 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
737 struct tgt_handler *h;
742 if (req->rq_export == NULL)
745 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
746 tsi->tsi_pill = &req->rq_pill;
747 tsi->tsi_env = req->rq_svc_thread->t_env;
748 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
749 tsi->tsi_exp = req->rq_export;
751 h = tgt_handler_find_check(req);
757 rc = tgt_request_preprocess(tsi, h, req);
758 if (unlikely(rc != 0))
761 if (h->th_hp != NULL)
765 EXPORT_SYMBOL(tgt_hpreq_handler);
767 void tgt_counter_incr(struct obd_export *exp, int opcode)
769 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
770 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
771 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
773 EXPORT_SYMBOL(tgt_counter_incr);
776 * Unified target generic handlers.
782 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
784 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
785 OBD_CONNECT_RMT_CLIENT_FORCE);
788 static int tgt_init_sec_level(struct ptlrpc_request *req)
790 struct lu_target *tgt = class_exp2tgt(req->rq_export);
792 struct obd_connect_data *data, *reply;
797 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
798 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
799 if (data == NULL || reply == NULL)
802 /* connection from MDT is always trusted */
803 if (req->rq_auth_usr_mdt) {
804 tgt_init_sec_none(reply);
808 if (unlikely(tgt == NULL)) {
809 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
810 class_exp2obd(req->rq_export)->obd_name);
814 client = libcfs_nid2str(req->rq_peer.nid);
815 /* no GSS support case */
816 if (!req->rq_auth_gss) {
817 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
818 CWARN("client %s -> target %s does not use GSS, "
819 "can not run under security level %d.\n",
820 client, tgt_name(tgt), tgt->lut_sec_level);
823 tgt_init_sec_none(reply);
828 /* old version case */
829 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
830 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
831 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
832 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
833 CWARN("client %s -> target %s uses old version, "
834 "can not run under security level %d.\n",
835 client, tgt_name(tgt), tgt->lut_sec_level);
838 CWARN("client %s -> target %s uses old version, "
839 "run under security level %d.\n",
840 client, tgt_name(tgt), tgt->lut_sec_level);
841 tgt_init_sec_none(reply);
846 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
848 if (!req->rq_auth_remote)
849 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
850 "asked to be remote.\n", client, tgt_name(tgt));
851 } else if (req->rq_auth_remote) {
853 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
854 "as remote by default.\n", client, tgt_name(tgt));
858 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
859 CDEBUG(D_SEC, "client %s -> target %s: user is not "
860 "authenticated!\n", client, tgt_name(tgt));
866 switch (tgt->lut_sec_level) {
867 case LUSTRE_SEC_NONE:
870 "client %s -> target %s is set as remote, "
871 "can not run under security level %d.\n",
872 client, tgt_name(tgt), tgt->lut_sec_level);
875 tgt_init_sec_none(reply);
877 case LUSTRE_SEC_REMOTE:
879 tgt_init_sec_none(reply);
884 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
885 OBD_CONNECT_RMT_CLIENT_FORCE);
886 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
887 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
896 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
898 struct lu_target *tgt = class_exp2tgt(exp);
899 struct sptlrpc_flavor flvr;
903 LASSERT(tgt->lut_obd);
904 LASSERT(tgt->lut_slice);
906 /* always allow ECHO client */
907 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
908 LUSTRE_ECHO_NAME) == 0)) {
909 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
913 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
914 read_lock(&tgt->lut_sptlrpc_lock);
915 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
919 read_unlock(&tgt->lut_sptlrpc_lock);
921 spin_lock(&exp->exp_lock);
922 exp->exp_sp_peer = req->rq_sp_from;
923 exp->exp_flvr = flvr;
924 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
925 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
926 CERROR("%s: unauthorized rpc flavor %x from %s, "
927 "expect %x\n", tgt_name(tgt),
929 libcfs_nid2str(req->rq_peer.nid),
930 exp->exp_flvr.sf_rpc);
933 spin_unlock(&exp->exp_lock);
935 if (exp->exp_sp_peer != req->rq_sp_from) {
936 CERROR("%s: RPC source %s doesn't match %s\n",
938 sptlrpc_part2name(req->rq_sp_from),
939 sptlrpc_part2name(exp->exp_sp_peer));
942 rc = sptlrpc_target_export_check(exp, req);
949 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
951 struct sptlrpc_rule_set tmp_rset;
954 if (unlikely(tgt == NULL)) {
955 CERROR("No target passed");
959 sptlrpc_rule_set_init(&tmp_rset);
960 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
962 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
967 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
969 write_lock(&tgt->lut_sptlrpc_lock);
970 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
971 tgt->lut_sptlrpc_rset = tmp_rset;
972 write_unlock(&tgt->lut_sptlrpc_lock);
976 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
978 int tgt_connect(struct tgt_session_info *tsi)
980 struct ptlrpc_request *req = tgt_ses_req(tsi);
981 struct obd_connect_data *reply;
986 rc = tgt_init_sec_level(req);
990 /* XXX: better to call this check right after getting new export but
991 * before last_rcvd slot allocation to avoid server load upon insecure
992 * connects. This is to be fixed after unifiyng all targets.
994 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
998 /* To avoid exposing partially initialized connection flags, changes up
999 * to this point have been staged in reply->ocd_connect_flags. Now that
1000 * connection handling has completed successfully, atomically update
1001 * the connect flags in the shared export data structure. LU-1623 */
1002 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
1003 spin_lock(&tsi->tsi_exp->exp_lock);
1004 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
1005 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
1006 spin_unlock(&tsi->tsi_exp->exp_lock);
1010 obd_disconnect(class_export_get(tsi->tsi_exp));
1013 EXPORT_SYMBOL(tgt_connect);
1015 int tgt_disconnect(struct tgt_session_info *tsi)
1021 rc = target_handle_disconnect(tgt_ses_req(tsi));
1023 RETURN(err_serious(rc));
1027 EXPORT_SYMBOL(tgt_disconnect);
1030 * Unified target OBD handlers
1032 int tgt_obd_ping(struct tgt_session_info *tsi)
1038 rc = target_handle_ping(tgt_ses_req(tsi));
1040 RETURN(err_serious(rc));
1044 EXPORT_SYMBOL(tgt_obd_ping);
1046 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1048 return err_serious(-EOPNOTSUPP);
1051 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1053 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1054 struct ptlrpc_request *req = tgt_ses_req(tsi);
1055 struct obd_export *exp = req->rq_export;
1056 struct ptlrpc_bulk_desc *desc;
1057 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1065 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
1066 PTLRPC_BULK_PUT_SOURCE |
1067 PTLRPC_BULK_BUF_KIOV,
1069 &ptlrpc_bulk_kiov_pin_ops);
1073 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1074 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1075 * which is rdpg->rp_count */
1076 nob = rdpg->rp_count;
1078 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1079 i++, tmpcount -= tmpsize) {
1080 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1081 desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
1085 LASSERT(desc->bd_nob == nob);
1086 rc = target_bulk_io(exp, desc, lwi);
1087 ptlrpc_free_bulk(desc);
1090 EXPORT_SYMBOL(tgt_sendpage);
1093 * OBD_IDX_READ handler
1095 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1097 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1098 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1099 struct idx_info *req_ii, *rep_ii;
1104 memset(rdpg, 0, sizeof(*rdpg));
1105 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1107 /* extract idx_info buffer from request & reply */
1108 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1109 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1110 RETURN(err_serious(-EPROTO));
1112 rc = req_capsule_server_pack(tsi->tsi_pill);
1114 RETURN(err_serious(rc));
1116 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1118 RETURN(err_serious(-EFAULT));
1119 rep_ii->ii_magic = IDX_INFO_MAGIC;
1121 /* extract hash to start with */
1122 rdpg->rp_hash = req_ii->ii_hash_start;
1124 /* extract requested attributes */
1125 rdpg->rp_attrs = req_ii->ii_attrs;
1127 /* check that fid packed in request is valid and supported */
1128 if (!fid_is_sane(&req_ii->ii_fid))
1130 rep_ii->ii_fid = req_ii->ii_fid;
1133 rep_ii->ii_flags = req_ii->ii_flags;
1135 /* compute number of pages to allocate, ii_count is the number of 4KB
1137 if (req_ii->ii_count <= 0)
1138 GOTO(out, rc = -EFAULT);
1139 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1140 exp_max_brw_size(tsi->tsi_exp));
1141 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1143 /* allocate pages to store the containers */
1144 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1145 if (rdpg->rp_pages == NULL)
1146 GOTO(out, rc = -ENOMEM);
1147 for (i = 0; i < rdpg->rp_npages; i++) {
1148 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1149 if (rdpg->rp_pages[i] == NULL)
1150 GOTO(out, rc = -ENOMEM);
1153 /* populate pages with key/record pairs */
1154 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1158 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1159 "asked %d > %d\n", rc, rdpg->rp_count);
1161 /* send pages to client */
1162 rc = tgt_sendpage(tsi, rdpg, rc);
1167 if (rdpg->rp_pages) {
1168 for (i = 0; i < rdpg->rp_npages; i++)
1169 if (rdpg->rp_pages[i])
1170 __free_page(rdpg->rp_pages[i]);
1171 OBD_FREE(rdpg->rp_pages,
1172 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1177 struct tgt_handler tgt_obd_handlers[] = {
1178 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1179 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1180 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1182 EXPORT_SYMBOL(tgt_obd_handlers);
1184 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1185 struct dt_object *obj, __u64 start, __u64 end)
1191 /* if no objid is specified, it means "sync whole filesystem" */
1193 rc = dt_sync(env, tgt->lut_bottom);
1194 } else if (dt_version_get(env, obj) >
1195 tgt->lut_obd->obd_last_committed) {
1196 rc = dt_object_sync(env, obj, start, end);
1201 EXPORT_SYMBOL(tgt_sync);
1203 * Unified target DLM handlers.
1206 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1208 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1209 void *data, int flag)
1212 struct lu_target *tgt;
1213 struct dt_object *obj;
1219 tgt = class_exp2tgt(lock->l_export);
1221 if (unlikely(tgt == NULL)) {
1222 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1223 class_exp2obd(lock->l_export)->obd_name);
1227 if (flag == LDLM_CB_CANCELING &&
1228 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1229 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1230 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1231 lock->l_flags & LDLM_FL_CBPENDING))) {
1233 __u64 end = OBD_OBJECT_EOF;
1235 rc = lu_env_init(&env, LCT_DT_THREAD);
1236 if (unlikely(rc != 0))
1239 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1240 tgt->lut_lsd.lsd_osd_index);
1241 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1243 GOTO(err_env, rc = PTR_ERR(obj));
1245 if (!dt_object_exists(obj))
1246 GOTO(err_put, rc = -ENOENT);
1248 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1249 start = lock->l_policy_data.l_extent.start;
1250 end = lock->l_policy_data.l_extent.end;
1253 rc = tgt_sync(&env, tgt, obj, start, end);
1255 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1256 "cancel: rc = %d\n",
1257 tgt_name(tgt), PFID(&fid),
1258 lock->l_policy_data.l_extent.start,
1259 lock->l_policy_data.l_extent.end, rc);
1262 lu_object_put(&env, &obj->do_lu);
1267 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1271 static struct ldlm_callback_suite tgt_dlm_cbs = {
1272 .lcs_completion = ldlm_server_completion_ast,
1273 .lcs_blocking = tgt_blocking_ast,
1274 .lcs_glimpse = ldlm_server_glimpse_ast
1277 int tgt_enqueue(struct tgt_session_info *tsi)
1279 struct ptlrpc_request *req = tgt_ses_req(tsi);
1284 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1285 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1287 LASSERT(tsi->tsi_dlm_req != NULL);
1288 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1289 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1291 RETURN(err_serious(rc));
1293 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1295 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1298 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1301 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1304 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1307 RETURN(req->rq_status);
1309 EXPORT_SYMBOL(tgt_enqueue);
1311 int tgt_convert(struct tgt_session_info *tsi)
1313 struct ptlrpc_request *req = tgt_ses_req(tsi);
1317 LASSERT(tsi->tsi_dlm_req);
1318 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1320 RETURN(err_serious(rc));
1322 RETURN(req->rq_status);
1325 int tgt_bl_callback(struct tgt_session_info *tsi)
1327 return err_serious(-EOPNOTSUPP);
1330 int tgt_cp_callback(struct tgt_session_info *tsi)
1332 return err_serious(-EOPNOTSUPP);
1335 /* generic LDLM target handler */
1336 struct tgt_handler tgt_dlm_handlers[] = {
1337 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1338 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1339 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1340 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1342 EXPORT_SYMBOL(tgt_dlm_handlers);
1345 * Unified target LLOG handlers.
1347 int tgt_llog_open(struct tgt_session_info *tsi)
1353 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1357 EXPORT_SYMBOL(tgt_llog_open);
1359 int tgt_llog_close(struct tgt_session_info *tsi)
1365 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1369 EXPORT_SYMBOL(tgt_llog_close);
1372 int tgt_llog_destroy(struct tgt_session_info *tsi)
1378 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1383 int tgt_llog_read_header(struct tgt_session_info *tsi)
1389 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1393 EXPORT_SYMBOL(tgt_llog_read_header);
1395 int tgt_llog_next_block(struct tgt_session_info *tsi)
1401 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1405 EXPORT_SYMBOL(tgt_llog_next_block);
1407 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1413 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1417 EXPORT_SYMBOL(tgt_llog_prev_block);
1419 /* generic llog target handler */
1420 struct tgt_handler tgt_llog_handlers[] = {
1421 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1422 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1423 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1424 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1425 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1426 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1428 EXPORT_SYMBOL(tgt_llog_handlers);
1430 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1431 struct dt_device *key,
1432 struct lfsck_request *lr,
1433 struct thandle *th) = NULL;
1435 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1437 struct lfsck_request *,
1440 tgt_lfsck_in_notify = notify;
1442 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1444 static int (*tgt_lfsck_query)(const struct lu_env *env,
1445 struct dt_device *key,
1446 struct lfsck_request *lr) = NULL;
1448 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1450 struct lfsck_request *))
1452 tgt_lfsck_query = query;
1454 EXPORT_SYMBOL(tgt_register_lfsck_query);
1456 /* LFSCK request handlers */
1457 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1459 const struct lu_env *env = tsi->tsi_env;
1460 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1461 struct lfsck_request *lr;
1465 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1469 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1474 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1476 struct lfsck_request *request;
1477 struct lfsck_reply *reply;
1481 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1482 if (request == NULL)
1485 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1489 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1490 reply->lr_status = rc;
1492 RETURN(rc < 0 ? rc : 0);
1495 struct tgt_handler tgt_lfsck_handlers[] = {
1496 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1497 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1499 EXPORT_SYMBOL(tgt_lfsck_handlers);
1502 * initialize per-thread page pool (bug 5137).
1504 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1506 struct tgt_thread_big_cache *tbc;
1510 LASSERT(thread != NULL);
1511 LASSERT(thread->t_data == NULL);
1513 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1516 thread->t_data = tbc;
1519 EXPORT_SYMBOL(tgt_io_thread_init);
1522 * free per-thread pool created by tgt_thread_init().
1524 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1526 struct tgt_thread_big_cache *tbc;
1530 LASSERT(thread != NULL);
1533 * be prepared to handle partially-initialized pools (because this is
1534 * called from ost_io_thread_init() for cleanup.
1536 tbc = thread->t_data;
1538 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1539 thread->t_data = NULL;
1543 EXPORT_SYMBOL(tgt_io_thread_done);
1545 * Helper function for getting server side [start, start+count] DLM lock
1546 * if asked by client.
1548 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1549 __u64 start, __u64 end, struct lustre_handle *lh,
1550 int mode, __u64 *flags)
1552 ldlm_policy_data_t policy;
1557 LASSERT(lh != NULL);
1558 LASSERT(ns != NULL);
1559 LASSERT(!lustre_handle_is_used(lh));
1561 policy.l_extent.gid = 0;
1562 policy.l_extent.start = start & PAGE_MASK;
1565 * If ->o_blocks is EOF it means "lock till the end of the file".
1566 * Otherwise, it's size of an extent or hole being punched (in bytes).
1568 if (end == OBD_OBJECT_EOF || end < start)
1569 policy.l_extent.end = OBD_OBJECT_EOF;
1571 policy.l_extent.end = end | ~PAGE_MASK;
1573 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1574 flags, ldlm_blocking_ast,
1575 ldlm_completion_ast, ldlm_glimpse_ast,
1576 NULL, 0, LVB_T_NONE, NULL, lh);
1577 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1579 EXPORT_SYMBOL(tgt_extent_lock);
1581 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1583 LASSERT(lustre_handle_is_used(lh));
1584 ldlm_lock_decref(lh, mode);
1586 EXPORT_SYMBOL(tgt_extent_unlock);
1588 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1589 struct obd_ioobj *obj, struct niobuf_remote *nb,
1590 struct lustre_handle *lh, int mode)
1593 int nrbufs = obj->ioo_bufcnt;
1598 LASSERT(mode == LCK_PR || mode == LCK_PW);
1599 LASSERT(!lustre_handle_is_used(lh));
1601 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1604 for (i = 1; i < nrbufs; i++)
1605 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1608 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1609 nb[nrbufs - 1].rnb_offset +
1610 nb[nrbufs - 1].rnb_len - 1,
1614 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1615 struct lustre_handle *lh, int mode)
1619 LASSERT(mode == LCK_PR || mode == LCK_PW);
1620 LASSERT((obj->ioo_bufcnt > 0 &&
1621 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1622 lustre_handle_is_used(lh));
1624 if (lustre_handle_is_used(lh))
1625 tgt_extent_unlock(lh, mode);
1629 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1630 struct ptlrpc_bulk_desc *desc, int opc,
1631 cksum_type_t cksum_type)
1633 struct cfs_crypto_hash_desc *hdesc;
1634 unsigned int bufsize;
1636 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1639 LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
1641 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1642 if (IS_ERR(hdesc)) {
1643 CERROR("%s: unable to initialize checksum hash %s\n",
1644 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1645 return PTR_ERR(hdesc);
1648 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1649 for (i = 0; i < desc->bd_iov_count; i++) {
1650 /* corrupt the data before we compute the checksum, to
1651 * simulate a client->OST data error */
1652 if (i == 0 && opc == OST_WRITE &&
1653 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1654 int off = BD_GET_KIOV(desc, i).kiov_offset &
1656 int len = BD_GET_KIOV(desc, i).kiov_len;
1657 struct page *np = tgt_page_to_corrupt;
1658 char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1661 char *ptr2 = kmap(np) + off;
1663 memcpy(ptr2, ptr, len);
1664 memcpy(ptr2, "bad3", min(4, len));
1666 BD_GET_KIOV(desc, i).kiov_page = np;
1668 CERROR("%s: can't alloc page for corruption\n",
1672 cfs_crypto_hash_update_page(hdesc,
1673 BD_GET_KIOV(desc, i).kiov_page,
1674 BD_GET_KIOV(desc, i).kiov_offset &
1676 BD_GET_KIOV(desc, i).kiov_len);
1678 /* corrupt the data after we compute the checksum, to
1679 * simulate an OST->client data error */
1680 if (i == 0 && opc == OST_READ &&
1681 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1682 int off = BD_GET_KIOV(desc, i).kiov_offset
1684 int len = BD_GET_KIOV(desc, i).kiov_len;
1685 struct page *np = tgt_page_to_corrupt;
1687 kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
1690 char *ptr2 = kmap(np) + off;
1692 memcpy(ptr2, ptr, len);
1693 memcpy(ptr2, "bad4", min(4, len));
1695 BD_GET_KIOV(desc, i).kiov_page = np;
1697 CERROR("%s: can't alloc page for corruption\n",
1703 bufsize = sizeof(cksum);
1704 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1709 int tgt_brw_read(struct tgt_session_info *tsi)
1711 struct ptlrpc_request *req = tgt_ses_req(tsi);
1712 struct ptlrpc_bulk_desc *desc = NULL;
1713 struct obd_export *exp = tsi->tsi_exp;
1714 struct niobuf_remote *remote_nb;
1715 struct niobuf_local *local_nb;
1716 struct obd_ioobj *ioo;
1717 struct ost_body *body, *repbody;
1718 struct l_wait_info lwi;
1719 struct lustre_handle lockh = { 0 };
1720 int npages, nob = 0, rc, i, no_reply = 0;
1721 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1725 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1726 CERROR("%s: deny read request from %s to portal %u\n",
1727 tgt_name(tsi->tsi_tgt),
1728 obd_export_nid2str(req->rq_export),
1729 ptlrpc_req2svc(req)->srv_req_portal);
1733 req->rq_bulk_read = 1;
1735 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1738 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1739 cfs_fail_val : (obd_timeout + 1) / 4);
1741 /* Check if there is eviction in progress, and if so, wait for it to
1743 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1744 /* We do not care how long it takes */
1745 lwi = LWI_INTR(NULL, NULL);
1746 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1747 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1751 /* There must be big cache in current thread to process this request
1752 * if it is NULL then something went wrong and it wasn't allocated,
1753 * report -ENOMEM in that case */
1757 body = tsi->tsi_ost_body;
1758 LASSERT(body != NULL);
1760 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1761 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1763 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1764 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1766 local_nb = tbc->local;
1768 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1769 remote_nb, &lockh, LCK_PR);
1774 * If getting the lock took more time than
1775 * client was willing to wait, drop it. b=11330
1777 if (cfs_time_current_sec() > req->rq_deadline ||
1778 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1780 CERROR("Dropping timed-out read from %s because locking"
1781 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1782 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1783 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1784 req->rq_deadline - req->rq_arrival_time.tv_sec);
1785 GOTO(out_lock, rc = -ETIMEDOUT);
1788 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1789 repbody->oa = body->oa;
1791 npages = PTLRPC_MAX_BRW_PAGES;
1792 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1793 ioo, remote_nb, &npages, local_nb);
1797 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1798 PTLRPC_BULK_PUT_SOURCE |
1799 PTLRPC_BULK_BUF_KIOV,
1801 &ptlrpc_bulk_kiov_nopin_ops);
1803 GOTO(out_commitrw, rc = -ENOMEM);
1806 for (i = 0; i < npages; i++) {
1807 int page_rc = local_nb[i].lnb_rc;
1815 if (page_rc != 0) { /* some data! */
1816 LASSERT(local_nb[i].lnb_page != NULL);
1817 desc->bd_frag_ops->add_kiov_frag
1818 (desc, local_nb[i].lnb_page,
1819 local_nb[i].lnb_page_offset,
1823 if (page_rc != local_nb[i].lnb_len) { /* short read */
1824 /* All subsequent pages should be 0 */
1825 while (++i < npages)
1826 LASSERT(local_nb[i].lnb_rc == 0);
1831 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1832 cksum_type_t cksum_type =
1833 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1834 body->oa.o_flags : 0);
1835 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1836 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1837 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1838 OST_READ, cksum_type);
1839 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1840 repbody->oa.o_cksum);
1842 repbody->oa.o_valid = 0;
1844 /* We're finishing using body->oa as an input variable */
1846 /* Check if client was evicted while we were doing i/o before touching
1848 if (likely(rc == 0 &&
1849 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1850 rc = target_bulk_io(exp, desc, &lwi);
1855 /* Must commit after prep above in all cases */
1856 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
1857 remote_nb, npages, local_nb, rc);
1859 tgt_drop_id(exp, &repbody->oa);
1861 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1863 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1864 ptlrpc_free_bulk(desc);
1869 ptlrpc_lprocfs_brw(req, nob);
1870 } else if (no_reply) {
1871 req->rq_no_reply = 1;
1872 /* reply out callback would free */
1873 ptlrpc_req_drop_rs(req);
1874 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1875 "client will retry: rc %d\n",
1876 exp->exp_obd->obd_name,
1877 obd_uuid2str(&exp->exp_client_uuid),
1878 obd_export_nid2str(exp), rc);
1880 /* send a bulk after reply to simulate a network delay or reordering
1882 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1883 wait_queue_head_t waitq;
1884 struct l_wait_info lwi1;
1886 CDEBUG(D_INFO, "reorder BULK\n");
1887 init_waitqueue_head(&waitq);
1889 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1890 l_wait_event(waitq, 0, &lwi1);
1891 target_bulk_io(exp, desc, &lwi);
1892 ptlrpc_free_bulk(desc);
1897 EXPORT_SYMBOL(tgt_brw_read);
1899 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1900 struct ptlrpc_bulk_desc *desc,
1901 struct niobuf_local *local_nb, int npages,
1902 u32 client_cksum, u32 server_cksum,
1905 struct obd_export *exp = req->rq_export;
1906 struct ost_body *body;
1910 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1911 LASSERT(body != NULL);
1913 if (req->rq_peer.nid != desc->bd_sender) {
1915 router = libcfs_nid2str(desc->bd_sender);
1919 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1920 client_cksum, server_cksum);
1924 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1925 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1926 "]: client csum %x, server csum %x\n",
1927 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1929 body->oa.o_valid & OBD_MD_FLFID ?
1930 body->oa.o_parent_seq : (__u64)0,
1931 body->oa.o_valid & OBD_MD_FLFID ?
1932 body->oa.o_parent_oid : 0,
1933 body->oa.o_valid & OBD_MD_FLFID ?
1934 body->oa.o_parent_ver : 0,
1935 POSTID(&body->oa.o_oi),
1936 local_nb[0].lnb_file_offset,
1937 local_nb[npages-1].lnb_file_offset +
1938 local_nb[npages - 1].lnb_len - 1,
1939 client_cksum, server_cksum);
1942 int tgt_brw_write(struct tgt_session_info *tsi)
1944 struct ptlrpc_request *req = tgt_ses_req(tsi);
1945 struct ptlrpc_bulk_desc *desc = NULL;
1946 struct obd_export *exp = req->rq_export;
1947 struct niobuf_remote *remote_nb;
1948 struct niobuf_local *local_nb;
1949 struct obd_ioobj *ioo;
1950 struct ost_body *body, *repbody;
1951 struct l_wait_info lwi;
1952 struct lustre_handle lockh = {0};
1954 int objcount, niocount, npages;
1956 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1957 bool no_reply = false, mmap;
1958 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1962 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1963 CERROR("%s: deny write request from %s to portal %u\n",
1964 tgt_name(tsi->tsi_tgt),
1965 obd_export_nid2str(req->rq_export),
1966 ptlrpc_req2svc(req)->srv_req_portal);
1967 RETURN(err_serious(-EPROTO));
1970 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1971 RETURN(err_serious(-ENOSPC));
1972 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1973 RETURN(err_serious(-EROFS));
1975 req->rq_bulk_write = 1;
1977 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1978 RETURN(err_serious(-EIO));
1979 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1980 RETURN(err_serious(-EFAULT));
1982 /* pause before transaction has been started */
1983 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1984 cfs_fail_val : (obd_timeout + 1) / 4);
1986 /* There must be big cache in current thread to process this request
1987 * if it is NULL then something went wrong and it wasn't allocated,
1988 * report -ENOMEM in that case */
1992 body = tsi->tsi_ost_body;
1993 LASSERT(body != NULL);
1995 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1996 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1998 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1999 RCL_CLIENT) / sizeof(*ioo);
2001 for (niocount = i = 0; i < objcount; i++)
2002 niocount += ioo[i].ioo_bufcnt;
2004 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
2005 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
2006 if (niocount != req_capsule_get_size(&req->rq_pill,
2007 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
2009 RETURN(err_serious(-EPROTO));
2011 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
2012 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
2013 memory_pressure_set();
2015 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
2016 niocount * sizeof(*rcs));
2017 rc = req_capsule_server_pack(&req->rq_pill);
2019 GOTO(out, rc = err_serious(rc));
2021 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
2022 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
2024 local_nb = tbc->local;
2026 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
2027 remote_nb, &lockh, LCK_PW);
2032 * If getting the lock took more time than
2033 * client was willing to wait, drop it. b=11330
2035 if (cfs_time_current_sec() > req->rq_deadline ||
2036 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
2038 CERROR("%s: Dropping timed-out write from %s because locking "
2039 "object "DOSTID" took %ld seconds (limit was %ld).\n",
2040 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2041 POSTID(&ioo->ioo_oid),
2042 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2043 req->rq_deadline - req->rq_arrival_time.tv_sec);
2044 GOTO(out_lock, rc = -ETIMEDOUT);
2047 /* Because we already sync grant info with client when reconnect,
2048 * grant info will be cleared for resent req, then fed_grant and
2049 * total_grant will not be modified in following preprw_write */
2050 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2051 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2052 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2055 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2056 if (repbody == NULL)
2057 GOTO(out_lock, rc = -ENOMEM);
2058 repbody->oa = body->oa;
2060 npages = PTLRPC_MAX_BRW_PAGES;
2061 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2062 objcount, ioo, remote_nb, &npages, local_nb);
2066 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2067 PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
2069 &ptlrpc_bulk_kiov_nopin_ops);
2071 GOTO(skip_transfer, rc = -ENOMEM);
2073 /* NB Having prepped, we must commit... */
2074 for (i = 0; i < npages; i++)
2075 desc->bd_frag_ops->add_kiov_frag(desc,
2076 local_nb[i].lnb_page,
2077 local_nb[i].lnb_page_offset,
2078 local_nb[i].lnb_len);
2080 rc = sptlrpc_svc_prep_bulk(req, desc);
2082 GOTO(skip_transfer, rc);
2084 rc = target_bulk_io(exp, desc, &lwi);
2088 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2089 static int cksum_counter;
2091 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2092 cksum_type = cksum_type_unpack(body->oa.o_flags);
2094 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2095 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2096 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2097 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2098 OST_WRITE, cksum_type);
2101 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2102 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2103 body->oa.o_flags & OBD_FL_MMAP);
2105 tgt_warn_on_cksum(req, desc, local_nb, npages,
2107 repbody->oa.o_cksum, mmap);
2109 } else if ((cksum_counter & (-cksum_counter)) ==
2111 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2112 cksum_counter, libcfs_id2str(req->rq_peer),
2113 repbody->oa.o_cksum);
2117 /* Must commit after prep above in all cases */
2118 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2119 objcount, ioo, remote_nb, npages, local_nb, rc);
2120 if (rc == -ENOTCONN)
2121 /* quota acquire process has been given up because
2122 * either the client has been evicted or the client
2123 * has timed out the request already */
2127 * Disable sending mtime back to the client. If the client locked the
2128 * whole object, then it has already updated the mtime on its side,
2129 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2131 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2136 /* set per-requested niobuf return codes */
2137 for (i = j = 0; i < niocount; i++) {
2138 int len = remote_nb[i].rnb_len;
2143 LASSERT(j < npages);
2144 if (local_nb[j].lnb_rc < 0)
2145 rcs[i] = local_nb[j].lnb_rc;
2146 len -= local_nb[j].lnb_len;
2151 LASSERT(j == npages);
2152 ptlrpc_lprocfs_brw(req, nob);
2154 tgt_drop_id(exp, &repbody->oa);
2157 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2159 ptlrpc_free_bulk(desc);
2162 req->rq_no_reply = 1;
2163 /* reply out callback would free */
2164 ptlrpc_req_drop_rs(req);
2165 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2166 "client will retry: rc %d\n",
2167 exp->exp_obd->obd_name,
2168 obd_uuid2str(&exp->exp_client_uuid),
2169 obd_export_nid2str(exp), rc);
2171 memory_pressure_clr();
2174 EXPORT_SYMBOL(tgt_brw_write);