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 *process = target_queue_recovery_request(req, obd);
479 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
486 * Handle recovery. Return:
487 * +1: continue request processing;
488 * -ve: abort immediately with the given error code;
489 * 0: send reply with error code in req->rq_status;
491 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
495 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
500 case SEC_CTX_INIT_CONT:
505 if (!req->rq_export->exp_obd->obd_replayable)
508 /* sanity check: if the xid matches, the request must be marked as a
509 * resent or replayed */
510 if (req_xid_is_last(req)) {
511 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
512 (MSG_RESENT | MSG_REPLAY))) {
513 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
514 "last_xid, expected REPLAY or RESENT flag "
516 lustre_msg_get_flags(req->rq_reqmsg));
517 req->rq_status = -ENOTCONN;
521 /* else: note the opposite is not always true; a RESENT req after a
522 * failover will usually not match the last_xid, since it was likely
523 * never committed. A REPLAYed request will almost never match the
524 * last xid, however it could for a committed, but still retained,
527 /* Check for aborted recovery... */
528 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
532 DEBUG_REQ(D_INFO, req, "Got new replay");
533 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
535 if (rc != 0 || !should_process)
537 else if (should_process < 0) {
538 req->rq_status = should_process;
539 rc = ptlrpc_error(req);
546 /* Initial check for request, it is validation mostly */
547 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
549 struct tgt_handler *h;
550 struct tgt_opc_slice *s;
551 struct lu_target *tgt;
552 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
556 tgt = class_exp2tgt(req->rq_export);
557 if (unlikely(tgt == NULL)) {
558 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
559 class_exp2obd(req->rq_export)->obd_name);
560 RETURN(ERR_PTR(-EINVAL));
563 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
564 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
567 /* opcode was not found in slice */
568 if (unlikely(s->tos_hs == NULL)) {
569 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
571 RETURN(ERR_PTR(-ENOTSUPP));
574 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
575 h = s->tos_hs + (opc - s->tos_opc_start);
576 if (unlikely(h->th_opc == 0)) {
577 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
578 RETURN(ERR_PTR(-ENOTSUPP));
584 int tgt_request_handle(struct ptlrpc_request *req)
586 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
588 struct lustre_msg *msg = req->rq_reqmsg;
589 struct tgt_handler *h;
590 struct lu_target *tgt;
591 int request_fail_id = 0;
592 __u32 opc = lustre_msg_get_opc(msg);
597 /* Refill the context, to make sure all thread keys are allocated */
598 lu_env_refill(req->rq_svc_thread->t_env);
600 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
601 tsi->tsi_pill = &req->rq_pill;
602 tsi->tsi_env = req->rq_svc_thread->t_env;
604 /* if request has export then get handlers slice from corresponding
605 * target, otherwise that should be connect operation */
606 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
607 opc == MGS_CONNECT) {
608 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
609 rc = target_handle_connect(req);
611 rc = ptlrpc_error(req);
614 /* recovery-small test 18c asks to drop connect reply */
615 if (unlikely(opc == OST_CONNECT &&
616 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
620 if (unlikely(!class_connected_export(req->rq_export))) {
621 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
622 opc, libcfs_id2str(req->rq_peer));
623 req->rq_status = -ENOTCONN;
624 rc = ptlrpc_error(req);
628 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
629 tsi->tsi_exp = req->rq_export;
630 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
631 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
633 tsi->tsi_jobid = NULL;
636 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
637 class_exp2obd(req->rq_export)->obd_name);
638 req->rq_status = -EINVAL;
639 rc = ptlrpc_error(req);
643 request_fail_id = tgt->lut_request_fail_id;
644 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
646 h = tgt_handler_find_check(req);
648 req->rq_status = PTR_ERR(h);
649 rc = ptlrpc_error(req);
653 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
656 rc = lustre_msg_check_version(msg, h->th_version);
658 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
659 " %08x, expecting %08x\n", tgt_name(tgt),
660 lustre_msg_get_version(msg), h->th_version);
661 req->rq_status = -EINVAL;
662 rc = ptlrpc_error(req);
666 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
667 if (likely(rc == 1)) {
668 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
670 rc = tgt_handle_request0(tsi, h, req);
676 req_capsule_fini(tsi->tsi_pill);
677 if (tsi->tsi_corpus != NULL) {
678 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
679 tsi->tsi_corpus = NULL;
683 EXPORT_SYMBOL(tgt_request_handle);
685 /** Assign high priority operations to the request if needed. */
686 int tgt_hpreq_handler(struct ptlrpc_request *req)
688 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
689 struct tgt_handler *h;
694 if (req->rq_export == NULL)
697 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
698 tsi->tsi_pill = &req->rq_pill;
699 tsi->tsi_env = req->rq_svc_thread->t_env;
700 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
701 tsi->tsi_exp = req->rq_export;
703 h = tgt_handler_find_check(req);
709 rc = tgt_request_preprocess(tsi, h, req);
710 if (unlikely(rc != 0))
713 if (h->th_hp != NULL)
717 EXPORT_SYMBOL(tgt_hpreq_handler);
719 void tgt_counter_incr(struct obd_export *exp, int opcode)
721 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
722 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
723 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
725 EXPORT_SYMBOL(tgt_counter_incr);
728 * Unified target generic handlers.
734 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
736 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
737 OBD_CONNECT_RMT_CLIENT_FORCE);
740 static int tgt_init_sec_level(struct ptlrpc_request *req)
742 struct lu_target *tgt = class_exp2tgt(req->rq_export);
744 struct obd_connect_data *data, *reply;
749 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
750 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
751 if (data == NULL || reply == NULL)
754 /* connection from MDT is always trusted */
755 if (req->rq_auth_usr_mdt) {
756 tgt_init_sec_none(reply);
760 if (unlikely(tgt == NULL)) {
761 DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
762 class_exp2obd(req->rq_export)->obd_name);
766 client = libcfs_nid2str(req->rq_peer.nid);
767 /* no GSS support case */
768 if (!req->rq_auth_gss) {
769 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
770 CWARN("client %s -> target %s does not use GSS, "
771 "can not run under security level %d.\n",
772 client, tgt_name(tgt), tgt->lut_sec_level);
775 tgt_init_sec_none(reply);
780 /* old version case */
781 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
782 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
783 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
784 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
785 CWARN("client %s -> target %s uses old version, "
786 "can not run under security level %d.\n",
787 client, tgt_name(tgt), tgt->lut_sec_level);
790 CWARN("client %s -> target %s uses old version, "
791 "run under security level %d.\n",
792 client, tgt_name(tgt), tgt->lut_sec_level);
793 tgt_init_sec_none(reply);
798 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
800 if (!req->rq_auth_remote)
801 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
802 "asked to be remote.\n", client, tgt_name(tgt));
803 } else if (req->rq_auth_remote) {
805 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
806 "as remote by default.\n", client, tgt_name(tgt));
810 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
811 CDEBUG(D_SEC, "client %s -> target %s: user is not "
812 "authenticated!\n", client, tgt_name(tgt));
818 switch (tgt->lut_sec_level) {
819 case LUSTRE_SEC_NONE:
822 "client %s -> target %s is set as remote, "
823 "can not run under security level %d.\n",
824 client, tgt_name(tgt), tgt->lut_sec_level);
827 tgt_init_sec_none(reply);
829 case LUSTRE_SEC_REMOTE:
831 tgt_init_sec_none(reply);
836 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
837 OBD_CONNECT_RMT_CLIENT_FORCE);
838 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
839 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
848 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
850 struct lu_target *tgt = class_exp2tgt(exp);
851 struct sptlrpc_flavor flvr;
855 LASSERT(tgt->lut_obd);
856 LASSERT(tgt->lut_slice);
858 /* always allow ECHO client */
859 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
860 LUSTRE_ECHO_NAME) == 0)) {
861 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
865 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
866 read_lock(&tgt->lut_sptlrpc_lock);
867 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
871 read_unlock(&tgt->lut_sptlrpc_lock);
873 spin_lock(&exp->exp_lock);
874 exp->exp_sp_peer = req->rq_sp_from;
875 exp->exp_flvr = flvr;
876 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
877 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
878 CERROR("%s: unauthorized rpc flavor %x from %s, "
879 "expect %x\n", tgt_name(tgt),
881 libcfs_nid2str(req->rq_peer.nid),
882 exp->exp_flvr.sf_rpc);
885 spin_unlock(&exp->exp_lock);
887 if (exp->exp_sp_peer != req->rq_sp_from) {
888 CERROR("%s: RPC source %s doesn't match %s\n",
890 sptlrpc_part2name(req->rq_sp_from),
891 sptlrpc_part2name(exp->exp_sp_peer));
894 rc = sptlrpc_target_export_check(exp, req);
901 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
903 struct sptlrpc_rule_set tmp_rset;
906 if (unlikely(tgt == NULL)) {
907 CERROR("No target passed");
911 sptlrpc_rule_set_init(&tmp_rset);
912 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
914 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
919 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
921 write_lock(&tgt->lut_sptlrpc_lock);
922 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
923 tgt->lut_sptlrpc_rset = tmp_rset;
924 write_unlock(&tgt->lut_sptlrpc_lock);
928 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
930 int tgt_connect(struct tgt_session_info *tsi)
932 struct ptlrpc_request *req = tgt_ses_req(tsi);
933 struct obd_connect_data *reply;
938 rc = tgt_init_sec_level(req);
942 /* XXX: better to call this check right after getting new export but
943 * before last_rcvd slot allocation to avoid server load upon insecure
944 * connects. This is to be fixed after unifiyng all targets.
946 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
950 /* To avoid exposing partially initialized connection flags, changes up
951 * to this point have been staged in reply->ocd_connect_flags. Now that
952 * connection handling has completed successfully, atomically update
953 * the connect flags in the shared export data structure. LU-1623 */
954 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
955 spin_lock(&tsi->tsi_exp->exp_lock);
956 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
957 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
958 spin_unlock(&tsi->tsi_exp->exp_lock);
962 obd_disconnect(class_export_get(tsi->tsi_exp));
965 EXPORT_SYMBOL(tgt_connect);
967 int tgt_disconnect(struct tgt_session_info *tsi)
973 rc = target_handle_disconnect(tgt_ses_req(tsi));
975 RETURN(err_serious(rc));
979 EXPORT_SYMBOL(tgt_disconnect);
982 * Unified target OBD handlers
984 int tgt_obd_ping(struct tgt_session_info *tsi)
990 rc = target_handle_ping(tgt_ses_req(tsi));
992 RETURN(err_serious(rc));
996 EXPORT_SYMBOL(tgt_obd_ping);
998 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
1000 return err_serious(-EOPNOTSUPP);
1003 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
1005 return err_serious(-EOPNOTSUPP);
1008 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
1010 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1011 struct ptlrpc_request *req = tgt_ses_req(tsi);
1012 struct obd_export *exp = req->rq_export;
1013 struct ptlrpc_bulk_desc *desc;
1014 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1022 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1027 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1028 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1029 * which is rdpg->rp_count */
1030 nob = rdpg->rp_count;
1032 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1033 i++, tmpcount -= tmpsize) {
1034 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1035 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1038 LASSERT(desc->bd_nob == nob);
1039 rc = target_bulk_io(exp, desc, lwi);
1040 ptlrpc_free_bulk_pin(desc);
1043 EXPORT_SYMBOL(tgt_sendpage);
1046 * OBD_IDX_READ handler
1048 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1050 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1051 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1052 struct idx_info *req_ii, *rep_ii;
1057 memset(rdpg, 0, sizeof(*rdpg));
1058 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1060 /* extract idx_info buffer from request & reply */
1061 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1062 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1063 RETURN(err_serious(-EPROTO));
1065 rc = req_capsule_server_pack(tsi->tsi_pill);
1067 RETURN(err_serious(rc));
1069 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1071 RETURN(err_serious(-EFAULT));
1072 rep_ii->ii_magic = IDX_INFO_MAGIC;
1074 /* extract hash to start with */
1075 rdpg->rp_hash = req_ii->ii_hash_start;
1077 /* extract requested attributes */
1078 rdpg->rp_attrs = req_ii->ii_attrs;
1080 /* check that fid packed in request is valid and supported */
1081 if (!fid_is_sane(&req_ii->ii_fid))
1083 rep_ii->ii_fid = req_ii->ii_fid;
1086 rep_ii->ii_flags = req_ii->ii_flags;
1088 /* compute number of pages to allocate, ii_count is the number of 4KB
1090 if (req_ii->ii_count <= 0)
1091 GOTO(out, rc = -EFAULT);
1092 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1093 exp_max_brw_size(tsi->tsi_exp));
1094 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1096 /* allocate pages to store the containers */
1097 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1098 if (rdpg->rp_pages == NULL)
1099 GOTO(out, rc = -ENOMEM);
1100 for (i = 0; i < rdpg->rp_npages; i++) {
1101 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1102 if (rdpg->rp_pages[i] == NULL)
1103 GOTO(out, rc = -ENOMEM);
1106 /* populate pages with key/record pairs */
1107 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1111 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1112 "asked %d > %d\n", rc, rdpg->rp_count);
1114 /* send pages to client */
1115 rc = tgt_sendpage(tsi, rdpg, rc);
1120 if (rdpg->rp_pages) {
1121 for (i = 0; i < rdpg->rp_npages; i++)
1122 if (rdpg->rp_pages[i])
1123 __free_page(rdpg->rp_pages[i]);
1124 OBD_FREE(rdpg->rp_pages,
1125 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1130 struct tgt_handler tgt_obd_handlers[] = {
1131 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1132 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1133 TGT_OBD_HDL_VAR(0, OBD_QC_CALLBACK, tgt_obd_qc_callback),
1134 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1136 EXPORT_SYMBOL(tgt_obd_handlers);
1138 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1139 struct dt_object *obj, __u64 start, __u64 end)
1145 /* if no objid is specified, it means "sync whole filesystem" */
1147 rc = dt_sync(env, tgt->lut_bottom);
1148 } else if (dt_version_get(env, obj) >
1149 tgt->lut_obd->obd_last_committed) {
1150 rc = dt_object_sync(env, obj, start, end);
1155 EXPORT_SYMBOL(tgt_sync);
1157 * Unified target DLM handlers.
1160 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1162 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1163 void *data, int flag)
1166 struct lu_target *tgt;
1167 struct dt_object *obj;
1173 tgt = class_exp2tgt(lock->l_export);
1175 if (unlikely(tgt == NULL)) {
1176 CDEBUG(D_ERROR, "%s: No target for connected export\n",
1177 class_exp2obd(lock->l_export)->obd_name);
1181 if (flag == LDLM_CB_CANCELING &&
1182 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1183 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1184 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1185 lock->l_flags & LDLM_FL_CBPENDING))) {
1187 __u64 end = OBD_OBJECT_EOF;
1189 rc = lu_env_init(&env, LCT_DT_THREAD);
1190 if (unlikely(rc != 0))
1193 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1194 tgt->lut_lsd.lsd_osd_index);
1195 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1197 GOTO(err_env, rc = PTR_ERR(obj));
1199 if (!dt_object_exists(obj))
1200 GOTO(err_put, rc = -ENOENT);
1202 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1203 start = lock->l_policy_data.l_extent.start;
1204 end = lock->l_policy_data.l_extent.end;
1207 rc = tgt_sync(&env, tgt, obj, start, end);
1209 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1210 "cancel: rc = %d\n",
1211 tgt_name(tgt), PFID(&fid),
1212 lock->l_policy_data.l_extent.start,
1213 lock->l_policy_data.l_extent.end, rc);
1216 lu_object_put(&env, &obj->do_lu);
1221 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1225 static struct ldlm_callback_suite tgt_dlm_cbs = {
1226 .lcs_completion = ldlm_server_completion_ast,
1227 .lcs_blocking = tgt_blocking_ast,
1228 .lcs_glimpse = ldlm_server_glimpse_ast
1231 int tgt_enqueue(struct tgt_session_info *tsi)
1233 struct ptlrpc_request *req = tgt_ses_req(tsi);
1238 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1239 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1241 LASSERT(tsi->tsi_dlm_req != NULL);
1242 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1243 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1245 RETURN(err_serious(rc));
1247 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1249 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1252 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1255 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1258 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1261 RETURN(req->rq_status);
1263 EXPORT_SYMBOL(tgt_enqueue);
1265 int tgt_convert(struct tgt_session_info *tsi)
1267 struct ptlrpc_request *req = tgt_ses_req(tsi);
1271 LASSERT(tsi->tsi_dlm_req);
1272 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1274 RETURN(err_serious(rc));
1276 RETURN(req->rq_status);
1279 int tgt_bl_callback(struct tgt_session_info *tsi)
1281 return err_serious(-EOPNOTSUPP);
1284 int tgt_cp_callback(struct tgt_session_info *tsi)
1286 return err_serious(-EOPNOTSUPP);
1289 /* generic LDLM target handler */
1290 struct tgt_handler tgt_dlm_handlers[] = {
1291 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1292 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1293 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1294 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1296 EXPORT_SYMBOL(tgt_dlm_handlers);
1299 * Unified target LLOG handlers.
1301 int tgt_llog_open(struct tgt_session_info *tsi)
1307 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1311 EXPORT_SYMBOL(tgt_llog_open);
1313 int tgt_llog_close(struct tgt_session_info *tsi)
1319 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1323 EXPORT_SYMBOL(tgt_llog_close);
1326 int tgt_llog_destroy(struct tgt_session_info *tsi)
1332 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1337 int tgt_llog_read_header(struct tgt_session_info *tsi)
1343 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1347 EXPORT_SYMBOL(tgt_llog_read_header);
1349 int tgt_llog_next_block(struct tgt_session_info *tsi)
1355 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1359 EXPORT_SYMBOL(tgt_llog_next_block);
1361 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1367 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1371 EXPORT_SYMBOL(tgt_llog_prev_block);
1373 /* generic llog target handler */
1374 struct tgt_handler tgt_llog_handlers[] = {
1375 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1376 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1377 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1378 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1379 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1380 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1382 EXPORT_SYMBOL(tgt_llog_handlers);
1385 * sec context handlers
1387 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1388 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1393 struct tgt_handler tgt_sec_ctx_handlers[] = {
1394 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1395 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1396 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1398 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1400 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1401 struct dt_device *key,
1402 struct lfsck_request *lr,
1403 struct thandle *th) = NULL;
1405 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1407 struct lfsck_request *,
1410 tgt_lfsck_in_notify = notify;
1412 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1414 static int (*tgt_lfsck_query)(const struct lu_env *env,
1415 struct dt_device *key,
1416 struct lfsck_request *lr) = NULL;
1418 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1420 struct lfsck_request *))
1422 tgt_lfsck_query = query;
1424 EXPORT_SYMBOL(tgt_register_lfsck_query);
1426 /* LFSCK request handlers */
1427 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1429 const struct lu_env *env = tsi->tsi_env;
1430 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1431 struct lfsck_request *lr;
1435 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1439 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1444 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1446 struct lfsck_request *request;
1447 struct lfsck_reply *reply;
1451 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1452 if (request == NULL)
1455 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1459 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1460 reply->lr_status = rc;
1462 RETURN(rc < 0 ? rc : 0);
1465 struct tgt_handler tgt_lfsck_handlers[] = {
1466 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1467 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1469 EXPORT_SYMBOL(tgt_lfsck_handlers);
1472 * initialize per-thread page pool (bug 5137).
1474 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1476 struct tgt_thread_big_cache *tbc;
1480 LASSERT(thread != NULL);
1481 LASSERT(thread->t_data == NULL);
1483 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1486 thread->t_data = tbc;
1489 EXPORT_SYMBOL(tgt_io_thread_init);
1492 * free per-thread pool created by tgt_thread_init().
1494 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1496 struct tgt_thread_big_cache *tbc;
1500 LASSERT(thread != NULL);
1503 * be prepared to handle partially-initialized pools (because this is
1504 * called from ost_io_thread_init() for cleanup.
1506 tbc = thread->t_data;
1508 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1509 thread->t_data = NULL;
1513 EXPORT_SYMBOL(tgt_io_thread_done);
1515 * Helper function for getting server side [start, start+count] DLM lock
1516 * if asked by client.
1518 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1519 __u64 start, __u64 end, struct lustre_handle *lh,
1520 int mode, __u64 *flags)
1522 ldlm_policy_data_t policy;
1527 LASSERT(lh != NULL);
1528 LASSERT(ns != NULL);
1529 LASSERT(!lustre_handle_is_used(lh));
1531 policy.l_extent.gid = 0;
1532 policy.l_extent.start = start & PAGE_MASK;
1535 * If ->o_blocks is EOF it means "lock till the end of the file".
1536 * Otherwise, it's size of an extent or hole being punched (in bytes).
1538 if (end == OBD_OBJECT_EOF || end < start)
1539 policy.l_extent.end = OBD_OBJECT_EOF;
1541 policy.l_extent.end = end | ~PAGE_MASK;
1543 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1544 flags, ldlm_blocking_ast,
1545 ldlm_completion_ast, ldlm_glimpse_ast,
1546 NULL, 0, LVB_T_NONE, NULL, lh);
1547 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1549 EXPORT_SYMBOL(tgt_extent_lock);
1551 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1553 LASSERT(lustre_handle_is_used(lh));
1554 ldlm_lock_decref(lh, mode);
1556 EXPORT_SYMBOL(tgt_extent_unlock);
1558 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1559 struct obd_ioobj *obj, struct niobuf_remote *nb,
1560 struct lustre_handle *lh, int mode)
1563 int nrbufs = obj->ioo_bufcnt;
1568 LASSERT(mode == LCK_PR || mode == LCK_PW);
1569 LASSERT(!lustre_handle_is_used(lh));
1571 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1574 for (i = 1; i < nrbufs; i++)
1575 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1578 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1579 nb[nrbufs - 1].rnb_offset +
1580 nb[nrbufs - 1].rnb_len - 1,
1584 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1585 struct lustre_handle *lh, int mode)
1589 LASSERT(mode == LCK_PR || mode == LCK_PW);
1590 LASSERT((obj->ioo_bufcnt > 0 &&
1591 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1592 lustre_handle_is_used(lh));
1594 if (lustre_handle_is_used(lh))
1595 tgt_extent_unlock(lh, mode);
1599 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1600 struct ptlrpc_bulk_desc *desc, int opc,
1601 cksum_type_t cksum_type)
1603 struct cfs_crypto_hash_desc *hdesc;
1604 unsigned int bufsize;
1606 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1609 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1610 if (IS_ERR(hdesc)) {
1611 CERROR("%s: unable to initialize checksum hash %s\n",
1612 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1613 return PTR_ERR(hdesc);
1616 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1617 for (i = 0; i < desc->bd_iov_count; i++) {
1618 /* corrupt the data before we compute the checksum, to
1619 * simulate a client->OST data error */
1620 if (i == 0 && opc == OST_WRITE &&
1621 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1622 int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1623 int len = desc->bd_iov[i].kiov_len;
1624 struct page *np = tgt_page_to_corrupt;
1625 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1628 char *ptr2 = kmap(np) + off;
1630 memcpy(ptr2, ptr, len);
1631 memcpy(ptr2, "bad3", min(4, len));
1633 desc->bd_iov[i].kiov_page = np;
1635 CERROR("%s: can't alloc page for corruption\n",
1639 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1640 desc->bd_iov[i].kiov_offset & ~PAGE_MASK,
1641 desc->bd_iov[i].kiov_len);
1643 /* corrupt the data after we compute the checksum, to
1644 * simulate an OST->client data error */
1645 if (i == 0 && opc == OST_READ &&
1646 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1647 int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1648 int len = desc->bd_iov[i].kiov_len;
1649 struct page *np = tgt_page_to_corrupt;
1650 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1653 char *ptr2 = kmap(np) + off;
1655 memcpy(ptr2, ptr, len);
1656 memcpy(ptr2, "bad4", min(4, len));
1658 desc->bd_iov[i].kiov_page = np;
1660 CERROR("%s: can't alloc page for corruption\n",
1666 bufsize = sizeof(cksum);
1667 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1672 int tgt_brw_read(struct tgt_session_info *tsi)
1674 struct ptlrpc_request *req = tgt_ses_req(tsi);
1675 struct ptlrpc_bulk_desc *desc = NULL;
1676 struct obd_export *exp = tsi->tsi_exp;
1677 struct niobuf_remote *remote_nb;
1678 struct niobuf_local *local_nb;
1679 struct obd_ioobj *ioo;
1680 struct ost_body *body, *repbody;
1681 struct l_wait_info lwi;
1682 struct lustre_handle lockh = { 0 };
1683 int npages, nob = 0, rc, i, no_reply = 0;
1684 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1688 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1689 CERROR("%s: deny read request from %s to portal %u\n",
1690 tgt_name(tsi->tsi_tgt),
1691 obd_export_nid2str(req->rq_export),
1692 ptlrpc_req2svc(req)->srv_req_portal);
1696 req->rq_bulk_read = 1;
1698 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1701 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1702 cfs_fail_val : (obd_timeout + 1) / 4);
1704 /* Check if there is eviction in progress, and if so, wait for it to
1706 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1707 /* We do not care how long it takes */
1708 lwi = LWI_INTR(NULL, NULL);
1709 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1710 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1714 /* There must be big cache in current thread to process this request
1715 * if it is NULL then something went wrong and it wasn't allocated,
1716 * report -ENOMEM in that case */
1720 body = tsi->tsi_ost_body;
1721 LASSERT(body != NULL);
1723 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1724 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1726 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1727 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1729 local_nb = tbc->local;
1731 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1732 remote_nb, &lockh, LCK_PR);
1737 * If getting the lock took more time than
1738 * client was willing to wait, drop it. b=11330
1740 if (cfs_time_current_sec() > req->rq_deadline ||
1741 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1743 CERROR("Dropping timed-out read from %s because locking"
1744 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1745 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1746 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1747 req->rq_deadline - req->rq_arrival_time.tv_sec);
1748 GOTO(out_lock, rc = -ETIMEDOUT);
1751 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1752 repbody->oa = body->oa;
1754 npages = PTLRPC_MAX_BRW_PAGES;
1755 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1756 ioo, remote_nb, &npages, local_nb, NULL);
1760 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1761 BULK_PUT_SOURCE, OST_BULK_PORTAL);
1763 GOTO(out_commitrw, rc = -ENOMEM);
1766 for (i = 0; i < npages; i++) {
1767 int page_rc = local_nb[i].lnb_rc;
1775 if (page_rc != 0) { /* some data! */
1776 LASSERT(local_nb[i].lnb_page != NULL);
1777 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1778 local_nb[i].lnb_page_offset,
1782 if (page_rc != local_nb[i].lnb_len) { /* short read */
1783 /* All subsequent pages should be 0 */
1784 while (++i < npages)
1785 LASSERT(local_nb[i].lnb_rc == 0);
1790 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1791 cksum_type_t cksum_type =
1792 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1793 body->oa.o_flags : 0);
1794 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1795 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1796 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1797 OST_READ, cksum_type);
1798 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1799 repbody->oa.o_cksum);
1801 repbody->oa.o_valid = 0;
1803 /* We're finishing using body->oa as an input variable */
1805 /* Check if client was evicted while we were doing i/o before touching
1807 if (likely(rc == 0 &&
1808 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1809 rc = target_bulk_io(exp, desc, &lwi);
1814 /* Must commit after prep above in all cases */
1815 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1816 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1819 tgt_drop_id(exp, &repbody->oa);
1821 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1823 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1824 ptlrpc_free_bulk_nopin(desc);
1829 ptlrpc_lprocfs_brw(req, nob);
1830 } else if (no_reply) {
1831 req->rq_no_reply = 1;
1832 /* reply out callback would free */
1833 ptlrpc_req_drop_rs(req);
1834 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1835 "client will retry: rc %d\n",
1836 exp->exp_obd->obd_name,
1837 obd_uuid2str(&exp->exp_client_uuid),
1838 obd_export_nid2str(exp), rc);
1840 /* send a bulk after reply to simulate a network delay or reordering
1842 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1843 wait_queue_head_t waitq;
1844 struct l_wait_info lwi1;
1846 CDEBUG(D_INFO, "reorder BULK\n");
1847 init_waitqueue_head(&waitq);
1849 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1850 l_wait_event(waitq, 0, &lwi1);
1851 target_bulk_io(exp, desc, &lwi);
1852 ptlrpc_free_bulk_nopin(desc);
1857 EXPORT_SYMBOL(tgt_brw_read);
1859 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1860 struct ptlrpc_bulk_desc *desc,
1861 struct niobuf_local *local_nb, int npages,
1862 u32 client_cksum, u32 server_cksum,
1865 struct obd_export *exp = req->rq_export;
1866 struct ost_body *body;
1870 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1871 LASSERT(body != NULL);
1873 if (req->rq_peer.nid != desc->bd_sender) {
1875 router = libcfs_nid2str(desc->bd_sender);
1879 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1880 client_cksum, server_cksum);
1884 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1885 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1886 "]: client csum %x, server csum %x\n",
1887 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1889 body->oa.o_valid & OBD_MD_FLFID ?
1890 body->oa.o_parent_seq : (__u64)0,
1891 body->oa.o_valid & OBD_MD_FLFID ?
1892 body->oa.o_parent_oid : 0,
1893 body->oa.o_valid & OBD_MD_FLFID ?
1894 body->oa.o_parent_ver : 0,
1895 POSTID(&body->oa.o_oi),
1896 local_nb[0].lnb_file_offset,
1897 local_nb[npages-1].lnb_file_offset +
1898 local_nb[npages - 1].lnb_len - 1,
1899 client_cksum, server_cksum);
1902 int tgt_brw_write(struct tgt_session_info *tsi)
1904 struct ptlrpc_request *req = tgt_ses_req(tsi);
1905 struct ptlrpc_bulk_desc *desc = NULL;
1906 struct obd_export *exp = req->rq_export;
1907 struct niobuf_remote *remote_nb;
1908 struct niobuf_local *local_nb;
1909 struct obd_ioobj *ioo;
1910 struct ost_body *body, *repbody;
1911 struct l_wait_info lwi;
1912 struct lustre_handle lockh = {0};
1914 int objcount, niocount, npages;
1916 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1917 bool no_reply = false, mmap;
1918 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1922 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1923 CERROR("%s: deny write request from %s to portal %u\n",
1924 tgt_name(tsi->tsi_tgt),
1925 obd_export_nid2str(req->rq_export),
1926 ptlrpc_req2svc(req)->srv_req_portal);
1927 RETURN(err_serious(-EPROTO));
1930 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1931 RETURN(err_serious(-ENOSPC));
1932 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1933 RETURN(err_serious(-EROFS));
1935 req->rq_bulk_write = 1;
1937 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1938 RETURN(err_serious(-EIO));
1939 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1940 RETURN(err_serious(-EFAULT));
1942 /* pause before transaction has been started */
1943 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1944 cfs_fail_val : (obd_timeout + 1) / 4);
1946 /* There must be big cache in current thread to process this request
1947 * if it is NULL then something went wrong and it wasn't allocated,
1948 * report -ENOMEM in that case */
1952 body = tsi->tsi_ost_body;
1953 LASSERT(body != NULL);
1955 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1956 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1958 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1959 RCL_CLIENT) / sizeof(*ioo);
1961 for (niocount = i = 0; i < objcount; i++)
1962 niocount += ioo[i].ioo_bufcnt;
1964 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1965 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1966 if (niocount != req_capsule_get_size(&req->rq_pill,
1967 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1969 RETURN(err_serious(-EPROTO));
1971 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1972 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1973 memory_pressure_set();
1975 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1976 niocount * sizeof(*rcs));
1977 rc = req_capsule_server_pack(&req->rq_pill);
1979 GOTO(out, rc = err_serious(rc));
1981 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1982 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1984 local_nb = tbc->local;
1986 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1987 remote_nb, &lockh, LCK_PW);
1992 * If getting the lock took more time than
1993 * client was willing to wait, drop it. b=11330
1995 if (cfs_time_current_sec() > req->rq_deadline ||
1996 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1998 CERROR("%s: Dropping timed-out write from %s because locking "
1999 "object "DOSTID" took %ld seconds (limit was %ld).\n",
2000 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
2001 POSTID(&ioo->ioo_oid),
2002 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
2003 req->rq_deadline - req->rq_arrival_time.tv_sec);
2004 GOTO(out_lock, rc = -ETIMEDOUT);
2007 /* Because we already sync grant info with client when reconnect,
2008 * grant info will be cleared for resent req, then fed_grant and
2009 * total_grant will not be modified in following preprw_write */
2010 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
2011 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
2012 body->oa.o_valid &= ~OBD_MD_FLGRANT;
2015 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
2016 if (repbody == NULL)
2017 GOTO(out_lock, rc = -ENOMEM);
2018 repbody->oa = body->oa;
2020 npages = PTLRPC_MAX_BRW_PAGES;
2021 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2022 objcount, ioo, remote_nb, &npages, local_nb, NULL);
2026 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2027 BULK_GET_SINK, OST_BULK_PORTAL);
2029 GOTO(skip_transfer, rc = -ENOMEM);
2031 /* NB Having prepped, we must commit... */
2032 for (i = 0; i < npages; i++)
2033 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2034 local_nb[i].lnb_page_offset,
2035 local_nb[i].lnb_len);
2037 rc = sptlrpc_svc_prep_bulk(req, desc);
2039 GOTO(skip_transfer, rc);
2041 rc = target_bulk_io(exp, desc, &lwi);
2045 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2046 static int cksum_counter;
2048 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2049 cksum_type = cksum_type_unpack(body->oa.o_flags);
2051 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2052 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2053 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2054 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2055 OST_WRITE, cksum_type);
2058 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2059 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2060 body->oa.o_flags & OBD_FL_MMAP);
2062 tgt_warn_on_cksum(req, desc, local_nb, npages,
2064 repbody->oa.o_cksum, mmap);
2066 } else if ((cksum_counter & (-cksum_counter)) ==
2068 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2069 cksum_counter, libcfs_id2str(req->rq_peer),
2070 repbody->oa.o_cksum);
2074 /* Must commit after prep above in all cases */
2075 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2076 objcount, ioo, remote_nb, npages, local_nb, NULL,
2078 if (rc == -ENOTCONN)
2079 /* quota acquire process has been given up because
2080 * either the client has been evicted or the client
2081 * has timed out the request already */
2085 * Disable sending mtime back to the client. If the client locked the
2086 * whole object, then it has already updated the mtime on its side,
2087 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2089 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2094 /* set per-requested niobuf return codes */
2095 for (i = j = 0; i < niocount; i++) {
2096 int len = remote_nb[i].rnb_len;
2101 LASSERT(j < npages);
2102 if (local_nb[j].lnb_rc < 0)
2103 rcs[i] = local_nb[j].lnb_rc;
2104 len -= local_nb[j].lnb_len;
2109 LASSERT(j == npages);
2110 ptlrpc_lprocfs_brw(req, nob);
2112 tgt_drop_id(exp, &repbody->oa);
2115 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2117 ptlrpc_free_bulk_nopin(desc);
2120 req->rq_no_reply = 1;
2121 /* reply out callback would free */
2122 ptlrpc_req_drop_rs(req);
2123 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2124 "client will retry: rc %d\n",
2125 exp->exp_obd->obd_name,
2126 obd_uuid2str(&exp->exp_client_uuid),
2127 obd_export_nid2str(exp), rc);
2129 memory_pressure_clr();
2132 EXPORT_SYMBOL(tgt_brw_write);