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);
558 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
559 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
562 /* opcode was not found in slice */
563 if (unlikely(s->tos_hs == NULL)) {
564 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
566 RETURN(ERR_PTR(-ENOTSUPP));
569 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
570 h = s->tos_hs + (opc - s->tos_opc_start);
571 if (unlikely(h->th_opc == 0)) {
572 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
573 RETURN(ERR_PTR(-ENOTSUPP));
579 int tgt_request_handle(struct ptlrpc_request *req)
581 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
583 struct lustre_msg *msg = req->rq_reqmsg;
584 struct tgt_handler *h;
585 struct lu_target *tgt;
586 int request_fail_id = 0;
587 __u32 opc = lustre_msg_get_opc(msg);
592 /* Refill the context, to make sure all thread keys are allocated */
593 lu_env_refill(req->rq_svc_thread->t_env);
595 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
596 tsi->tsi_pill = &req->rq_pill;
597 tsi->tsi_env = req->rq_svc_thread->t_env;
599 /* if request has export then get handlers slice from corresponding
600 * target, otherwise that should be connect operation */
601 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
602 opc == MGS_CONNECT) {
603 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
604 rc = target_handle_connect(req);
606 rc = ptlrpc_error(req);
609 /* recovery-small test 18c asks to drop connect reply */
610 if (unlikely(opc == OST_CONNECT &&
611 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
615 if (unlikely(!class_connected_export(req->rq_export))) {
616 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
617 opc, libcfs_id2str(req->rq_peer));
618 req->rq_status = -ENOTCONN;
619 rc = ptlrpc_error(req);
623 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
624 tsi->tsi_exp = req->rq_export;
625 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
626 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
628 tsi->tsi_jobid = NULL;
630 request_fail_id = tgt->lut_request_fail_id;
631 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
633 h = tgt_handler_find_check(req);
635 req->rq_status = PTR_ERR(h);
636 rc = ptlrpc_error(req);
640 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
643 rc = lustre_msg_check_version(msg, h->th_version);
645 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
646 " %08x, expecting %08x\n", tgt_name(tgt),
647 lustre_msg_get_version(msg), h->th_version);
648 req->rq_status = -EINVAL;
649 rc = ptlrpc_error(req);
653 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
654 if (likely(rc == 1)) {
655 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
657 rc = tgt_handle_request0(tsi, h, req);
663 req_capsule_fini(tsi->tsi_pill);
664 if (tsi->tsi_corpus != NULL) {
665 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
666 tsi->tsi_corpus = NULL;
670 EXPORT_SYMBOL(tgt_request_handle);
672 /** Assign high priority operations to the request if needed. */
673 int tgt_hpreq_handler(struct ptlrpc_request *req)
675 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
676 struct tgt_handler *h;
681 if (req->rq_export == NULL)
684 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
685 tsi->tsi_pill = &req->rq_pill;
686 tsi->tsi_env = req->rq_svc_thread->t_env;
687 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
688 tsi->tsi_exp = req->rq_export;
690 h = tgt_handler_find_check(req);
696 rc = tgt_request_preprocess(tsi, h, req);
697 if (unlikely(rc != 0))
700 if (h->th_hp != NULL)
704 EXPORT_SYMBOL(tgt_hpreq_handler);
706 void tgt_counter_incr(struct obd_export *exp, int opcode)
708 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
709 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
710 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
712 EXPORT_SYMBOL(tgt_counter_incr);
715 * Unified target generic handlers.
721 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
723 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
724 OBD_CONNECT_RMT_CLIENT_FORCE);
727 static int tgt_init_sec_level(struct ptlrpc_request *req)
729 struct lu_target *tgt = class_exp2tgt(req->rq_export);
731 struct obd_connect_data *data, *reply;
736 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
737 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
738 if (data == NULL || reply == NULL)
741 /* connection from MDT is always trusted */
742 if (req->rq_auth_usr_mdt) {
743 tgt_init_sec_none(reply);
747 client = libcfs_nid2str(req->rq_peer.nid);
748 /* no GSS support case */
749 if (!req->rq_auth_gss) {
750 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
751 CWARN("client %s -> target %s does not use GSS, "
752 "can not run under security level %d.\n",
753 client, tgt_name(tgt), tgt->lut_sec_level);
756 tgt_init_sec_none(reply);
761 /* old version case */
762 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
763 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
764 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
765 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
766 CWARN("client %s -> target %s uses old version, "
767 "can not run under security level %d.\n",
768 client, tgt_name(tgt), tgt->lut_sec_level);
771 CWARN("client %s -> target %s uses old version, "
772 "run under security level %d.\n",
773 client, tgt_name(tgt), tgt->lut_sec_level);
774 tgt_init_sec_none(reply);
779 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
781 if (!req->rq_auth_remote)
782 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
783 "asked to be remote.\n", client, tgt_name(tgt));
784 } else if (req->rq_auth_remote) {
786 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
787 "as remote by default.\n", client, tgt_name(tgt));
791 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
792 CDEBUG(D_SEC, "client %s -> target %s: user is not "
793 "authenticated!\n", client, tgt_name(tgt));
799 switch (tgt->lut_sec_level) {
800 case LUSTRE_SEC_NONE:
803 "client %s -> target %s is set as remote, "
804 "can not run under security level %d.\n",
805 client, tgt_name(tgt), tgt->lut_sec_level);
808 tgt_init_sec_none(reply);
810 case LUSTRE_SEC_REMOTE:
812 tgt_init_sec_none(reply);
817 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
818 OBD_CONNECT_RMT_CLIENT_FORCE);
819 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
820 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
829 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
831 struct lu_target *tgt = class_exp2tgt(exp);
832 struct sptlrpc_flavor flvr;
836 LASSERT(tgt->lut_obd);
837 LASSERT(tgt->lut_slice);
839 /* always allow ECHO client */
840 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
841 LUSTRE_ECHO_NAME) == 0)) {
842 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
846 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
847 read_lock(&tgt->lut_sptlrpc_lock);
848 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
852 read_unlock(&tgt->lut_sptlrpc_lock);
854 spin_lock(&exp->exp_lock);
855 exp->exp_sp_peer = req->rq_sp_from;
856 exp->exp_flvr = flvr;
857 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
858 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
859 CERROR("%s: unauthorized rpc flavor %x from %s, "
860 "expect %x\n", tgt_name(tgt),
862 libcfs_nid2str(req->rq_peer.nid),
863 exp->exp_flvr.sf_rpc);
866 spin_unlock(&exp->exp_lock);
868 if (exp->exp_sp_peer != req->rq_sp_from) {
869 CERROR("%s: RPC source %s doesn't match %s\n",
871 sptlrpc_part2name(req->rq_sp_from),
872 sptlrpc_part2name(exp->exp_sp_peer));
875 rc = sptlrpc_target_export_check(exp, req);
882 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
884 struct sptlrpc_rule_set tmp_rset;
887 sptlrpc_rule_set_init(&tmp_rset);
888 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
890 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
895 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
897 write_lock(&tgt->lut_sptlrpc_lock);
898 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
899 tgt->lut_sptlrpc_rset = tmp_rset;
900 write_unlock(&tgt->lut_sptlrpc_lock);
904 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
906 int tgt_connect(struct tgt_session_info *tsi)
908 struct ptlrpc_request *req = tgt_ses_req(tsi);
909 struct obd_connect_data *reply;
914 rc = tgt_init_sec_level(req);
918 /* XXX: better to call this check right after getting new export but
919 * before last_rcvd slot allocation to avoid server load upon insecure
920 * connects. This is to be fixed after unifiyng all targets.
922 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
926 /* To avoid exposing partially initialized connection flags, changes up
927 * to this point have been staged in reply->ocd_connect_flags. Now that
928 * connection handling has completed successfully, atomically update
929 * the connect flags in the shared export data structure. LU-1623 */
930 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
931 spin_lock(&tsi->tsi_exp->exp_lock);
932 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
933 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
934 spin_unlock(&tsi->tsi_exp->exp_lock);
938 obd_disconnect(class_export_get(tsi->tsi_exp));
941 EXPORT_SYMBOL(tgt_connect);
943 int tgt_disconnect(struct tgt_session_info *tsi)
949 rc = target_handle_disconnect(tgt_ses_req(tsi));
951 RETURN(err_serious(rc));
955 EXPORT_SYMBOL(tgt_disconnect);
958 * Unified target OBD handlers
960 int tgt_obd_ping(struct tgt_session_info *tsi)
966 rc = target_handle_ping(tgt_ses_req(tsi));
968 RETURN(err_serious(rc));
972 EXPORT_SYMBOL(tgt_obd_ping);
974 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
976 return err_serious(-EOPNOTSUPP);
979 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
981 return err_serious(-EOPNOTSUPP);
984 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
986 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
987 struct ptlrpc_request *req = tgt_ses_req(tsi);
988 struct obd_export *exp = req->rq_export;
989 struct ptlrpc_bulk_desc *desc;
990 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
998 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1003 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1004 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1005 * which is rdpg->rp_count */
1006 nob = rdpg->rp_count;
1008 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1009 i++, tmpcount -= tmpsize) {
1010 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1011 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1014 LASSERT(desc->bd_nob == nob);
1015 rc = target_bulk_io(exp, desc, lwi);
1016 ptlrpc_free_bulk_pin(desc);
1019 EXPORT_SYMBOL(tgt_sendpage);
1022 * OBD_IDX_READ handler
1024 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1026 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1027 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1028 struct idx_info *req_ii, *rep_ii;
1033 memset(rdpg, 0, sizeof(*rdpg));
1034 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1036 /* extract idx_info buffer from request & reply */
1037 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1038 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1039 RETURN(err_serious(-EPROTO));
1041 rc = req_capsule_server_pack(tsi->tsi_pill);
1043 RETURN(err_serious(rc));
1045 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1047 RETURN(err_serious(-EFAULT));
1048 rep_ii->ii_magic = IDX_INFO_MAGIC;
1050 /* extract hash to start with */
1051 rdpg->rp_hash = req_ii->ii_hash_start;
1053 /* extract requested attributes */
1054 rdpg->rp_attrs = req_ii->ii_attrs;
1056 /* check that fid packed in request is valid and supported */
1057 if (!fid_is_sane(&req_ii->ii_fid))
1059 rep_ii->ii_fid = req_ii->ii_fid;
1062 rep_ii->ii_flags = req_ii->ii_flags;
1064 /* compute number of pages to allocate, ii_count is the number of 4KB
1066 if (req_ii->ii_count <= 0)
1067 GOTO(out, rc = -EFAULT);
1068 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1069 exp_max_brw_size(tsi->tsi_exp));
1070 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1072 /* allocate pages to store the containers */
1073 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1074 if (rdpg->rp_pages == NULL)
1075 GOTO(out, rc = -ENOMEM);
1076 for (i = 0; i < rdpg->rp_npages; i++) {
1077 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1078 if (rdpg->rp_pages[i] == NULL)
1079 GOTO(out, rc = -ENOMEM);
1082 /* populate pages with key/record pairs */
1083 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1087 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1088 "asked %d > %d\n", rc, rdpg->rp_count);
1090 /* send pages to client */
1091 rc = tgt_sendpage(tsi, rdpg, rc);
1096 if (rdpg->rp_pages) {
1097 for (i = 0; i < rdpg->rp_npages; i++)
1098 if (rdpg->rp_pages[i])
1099 __free_page(rdpg->rp_pages[i]);
1100 OBD_FREE(rdpg->rp_pages,
1101 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1106 struct tgt_handler tgt_obd_handlers[] = {
1107 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1108 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1109 TGT_OBD_HDL_VAR(0, OBD_QC_CALLBACK, tgt_obd_qc_callback),
1110 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1112 EXPORT_SYMBOL(tgt_obd_handlers);
1114 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1115 struct dt_object *obj, __u64 start, __u64 end)
1121 /* if no objid is specified, it means "sync whole filesystem" */
1123 rc = dt_sync(env, tgt->lut_bottom);
1124 } else if (dt_version_get(env, obj) >
1125 tgt->lut_obd->obd_last_committed) {
1126 rc = dt_object_sync(env, obj, start, end);
1131 EXPORT_SYMBOL(tgt_sync);
1133 * Unified target DLM handlers.
1136 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1138 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1139 void *data, int flag)
1142 struct lu_target *tgt;
1143 struct dt_object *obj;
1149 tgt = class_exp2tgt(lock->l_export);
1151 if (flag == LDLM_CB_CANCELING &&
1152 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1153 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1154 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1155 lock->l_flags & LDLM_FL_CBPENDING))) {
1157 __u64 end = OBD_OBJECT_EOF;
1159 rc = lu_env_init(&env, LCT_DT_THREAD);
1160 if (unlikely(rc != 0))
1163 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1164 tgt->lut_lsd.lsd_osd_index);
1165 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1167 GOTO(err_env, rc = PTR_ERR(obj));
1169 if (!dt_object_exists(obj))
1170 GOTO(err_put, rc = -ENOENT);
1172 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1173 start = lock->l_policy_data.l_extent.start;
1174 end = lock->l_policy_data.l_extent.end;
1177 rc = tgt_sync(&env, tgt, obj, start, end);
1179 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1180 "cancel: rc = %d\n",
1181 tgt_name(tgt), PFID(&fid),
1182 lock->l_policy_data.l_extent.start,
1183 lock->l_policy_data.l_extent.end, rc);
1186 lu_object_put(&env, &obj->do_lu);
1191 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1195 static struct ldlm_callback_suite tgt_dlm_cbs = {
1196 .lcs_completion = ldlm_server_completion_ast,
1197 .lcs_blocking = tgt_blocking_ast,
1198 .lcs_glimpse = ldlm_server_glimpse_ast
1201 int tgt_enqueue(struct tgt_session_info *tsi)
1203 struct ptlrpc_request *req = tgt_ses_req(tsi);
1208 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1209 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1211 LASSERT(tsi->tsi_dlm_req != NULL);
1212 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1213 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1215 RETURN(err_serious(rc));
1217 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1219 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1222 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1225 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1228 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1231 RETURN(req->rq_status);
1233 EXPORT_SYMBOL(tgt_enqueue);
1235 int tgt_convert(struct tgt_session_info *tsi)
1237 struct ptlrpc_request *req = tgt_ses_req(tsi);
1241 LASSERT(tsi->tsi_dlm_req);
1242 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1244 RETURN(err_serious(rc));
1246 RETURN(req->rq_status);
1249 int tgt_bl_callback(struct tgt_session_info *tsi)
1251 return err_serious(-EOPNOTSUPP);
1254 int tgt_cp_callback(struct tgt_session_info *tsi)
1256 return err_serious(-EOPNOTSUPP);
1259 /* generic LDLM target handler */
1260 struct tgt_handler tgt_dlm_handlers[] = {
1261 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1262 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1263 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1264 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1266 EXPORT_SYMBOL(tgt_dlm_handlers);
1269 * Unified target LLOG handlers.
1271 int tgt_llog_open(struct tgt_session_info *tsi)
1277 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1281 EXPORT_SYMBOL(tgt_llog_open);
1283 int tgt_llog_close(struct tgt_session_info *tsi)
1289 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1293 EXPORT_SYMBOL(tgt_llog_close);
1296 int tgt_llog_destroy(struct tgt_session_info *tsi)
1302 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1307 int tgt_llog_read_header(struct tgt_session_info *tsi)
1313 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1317 EXPORT_SYMBOL(tgt_llog_read_header);
1319 int tgt_llog_next_block(struct tgt_session_info *tsi)
1325 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1329 EXPORT_SYMBOL(tgt_llog_next_block);
1331 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1337 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1341 EXPORT_SYMBOL(tgt_llog_prev_block);
1343 /* generic llog target handler */
1344 struct tgt_handler tgt_llog_handlers[] = {
1345 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1346 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1347 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1348 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1349 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1350 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1352 EXPORT_SYMBOL(tgt_llog_handlers);
1355 * sec context handlers
1357 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1358 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1363 struct tgt_handler tgt_sec_ctx_handlers[] = {
1364 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1365 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1366 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1368 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1370 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1371 struct dt_device *key,
1372 struct lfsck_request *lr,
1373 struct thandle *th) = NULL;
1375 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1377 struct lfsck_request *,
1380 tgt_lfsck_in_notify = notify;
1382 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1384 static int (*tgt_lfsck_query)(const struct lu_env *env,
1385 struct dt_device *key,
1386 struct lfsck_request *lr) = NULL;
1388 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1390 struct lfsck_request *))
1392 tgt_lfsck_query = query;
1394 EXPORT_SYMBOL(tgt_register_lfsck_query);
1396 /* LFSCK request handlers */
1397 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1399 const struct lu_env *env = tsi->tsi_env;
1400 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1401 struct lfsck_request *lr;
1405 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1409 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1414 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1416 struct lfsck_request *request;
1417 struct lfsck_reply *reply;
1421 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1422 if (request == NULL)
1425 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1429 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1430 reply->lr_status = rc;
1432 RETURN(rc < 0 ? rc : 0);
1435 struct tgt_handler tgt_lfsck_handlers[] = {
1436 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1437 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1439 EXPORT_SYMBOL(tgt_lfsck_handlers);
1442 * initialize per-thread page pool (bug 5137).
1444 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1446 struct tgt_thread_big_cache *tbc;
1450 LASSERT(thread != NULL);
1451 LASSERT(thread->t_data == NULL);
1453 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1456 thread->t_data = tbc;
1459 EXPORT_SYMBOL(tgt_io_thread_init);
1462 * free per-thread pool created by tgt_thread_init().
1464 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1466 struct tgt_thread_big_cache *tbc;
1470 LASSERT(thread != NULL);
1473 * be prepared to handle partially-initialized pools (because this is
1474 * called from ost_io_thread_init() for cleanup.
1476 tbc = thread->t_data;
1478 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1479 thread->t_data = NULL;
1483 EXPORT_SYMBOL(tgt_io_thread_done);
1485 * Helper function for getting server side [start, start+count] DLM lock
1486 * if asked by client.
1488 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1489 __u64 start, __u64 end, struct lustre_handle *lh,
1490 int mode, __u64 *flags)
1492 ldlm_policy_data_t policy;
1497 LASSERT(lh != NULL);
1498 LASSERT(ns != NULL);
1499 LASSERT(!lustre_handle_is_used(lh));
1501 policy.l_extent.gid = 0;
1502 policy.l_extent.start = start & PAGE_MASK;
1505 * If ->o_blocks is EOF it means "lock till the end of the file".
1506 * Otherwise, it's size of an extent or hole being punched (in bytes).
1508 if (end == OBD_OBJECT_EOF || end < start)
1509 policy.l_extent.end = OBD_OBJECT_EOF;
1511 policy.l_extent.end = end | ~PAGE_MASK;
1513 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1514 flags, ldlm_blocking_ast,
1515 ldlm_completion_ast, ldlm_glimpse_ast,
1516 NULL, 0, LVB_T_NONE, NULL, lh);
1517 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1519 EXPORT_SYMBOL(tgt_extent_lock);
1521 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1523 LASSERT(lustre_handle_is_used(lh));
1524 ldlm_lock_decref(lh, mode);
1526 EXPORT_SYMBOL(tgt_extent_unlock);
1528 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1529 struct obd_ioobj *obj, struct niobuf_remote *nb,
1530 struct lustre_handle *lh, int mode)
1533 int nrbufs = obj->ioo_bufcnt;
1538 LASSERT(mode == LCK_PR || mode == LCK_PW);
1539 LASSERT(!lustre_handle_is_used(lh));
1541 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1544 for (i = 1; i < nrbufs; i++)
1545 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1548 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1549 nb[nrbufs - 1].rnb_offset +
1550 nb[nrbufs - 1].rnb_len - 1,
1554 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1555 struct lustre_handle *lh, int mode)
1559 LASSERT(mode == LCK_PR || mode == LCK_PW);
1560 LASSERT((obj->ioo_bufcnt > 0 &&
1561 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1562 lustre_handle_is_used(lh));
1564 if (lustre_handle_is_used(lh))
1565 tgt_extent_unlock(lh, mode);
1569 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1570 struct ptlrpc_bulk_desc *desc, int opc,
1571 cksum_type_t cksum_type)
1573 struct cfs_crypto_hash_desc *hdesc;
1574 unsigned int bufsize;
1576 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1579 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1580 if (IS_ERR(hdesc)) {
1581 CERROR("%s: unable to initialize checksum hash %s\n",
1582 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1583 return PTR_ERR(hdesc);
1586 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1587 for (i = 0; i < desc->bd_iov_count; i++) {
1588 /* corrupt the data before we compute the checksum, to
1589 * simulate a client->OST data error */
1590 if (i == 0 && opc == OST_WRITE &&
1591 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1592 int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1593 int len = desc->bd_iov[i].kiov_len;
1594 struct page *np = tgt_page_to_corrupt;
1595 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1598 char *ptr2 = kmap(np) + off;
1600 memcpy(ptr2, ptr, len);
1601 memcpy(ptr2, "bad3", min(4, len));
1603 desc->bd_iov[i].kiov_page = np;
1605 CERROR("%s: can't alloc page for corruption\n",
1609 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1610 desc->bd_iov[i].kiov_offset & ~PAGE_MASK,
1611 desc->bd_iov[i].kiov_len);
1613 /* corrupt the data after we compute the checksum, to
1614 * simulate an OST->client data error */
1615 if (i == 0 && opc == OST_READ &&
1616 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1617 int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1618 int len = desc->bd_iov[i].kiov_len;
1619 struct page *np = tgt_page_to_corrupt;
1620 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1623 char *ptr2 = kmap(np) + off;
1625 memcpy(ptr2, ptr, len);
1626 memcpy(ptr2, "bad4", min(4, len));
1628 desc->bd_iov[i].kiov_page = np;
1630 CERROR("%s: can't alloc page for corruption\n",
1636 bufsize = sizeof(cksum);
1637 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1642 int tgt_brw_read(struct tgt_session_info *tsi)
1644 struct ptlrpc_request *req = tgt_ses_req(tsi);
1645 struct ptlrpc_bulk_desc *desc = NULL;
1646 struct obd_export *exp = tsi->tsi_exp;
1647 struct niobuf_remote *remote_nb;
1648 struct niobuf_local *local_nb;
1649 struct obd_ioobj *ioo;
1650 struct ost_body *body, *repbody;
1651 struct l_wait_info lwi;
1652 struct lustre_handle lockh = { 0 };
1653 int npages, nob = 0, rc, i, no_reply = 0;
1654 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1658 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1659 CERROR("%s: deny read request from %s to portal %u\n",
1660 tgt_name(tsi->tsi_tgt),
1661 obd_export_nid2str(req->rq_export),
1662 ptlrpc_req2svc(req)->srv_req_portal);
1666 req->rq_bulk_read = 1;
1668 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1671 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1672 cfs_fail_val : (obd_timeout + 1) / 4);
1674 /* Check if there is eviction in progress, and if so, wait for it to
1676 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1677 /* We do not care how long it takes */
1678 lwi = LWI_INTR(NULL, NULL);
1679 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1680 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1684 /* There must be big cache in current thread to process this request
1685 * if it is NULL then something went wrong and it wasn't allocated,
1686 * report -ENOMEM in that case */
1690 body = tsi->tsi_ost_body;
1691 LASSERT(body != NULL);
1693 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1694 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1696 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1697 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1699 local_nb = tbc->local;
1701 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1702 remote_nb, &lockh, LCK_PR);
1707 * If getting the lock took more time than
1708 * client was willing to wait, drop it. b=11330
1710 if (cfs_time_current_sec() > req->rq_deadline ||
1711 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1713 CERROR("Dropping timed-out read from %s because locking"
1714 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1715 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1716 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1717 req->rq_deadline - req->rq_arrival_time.tv_sec);
1718 GOTO(out_lock, rc = -ETIMEDOUT);
1721 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1722 repbody->oa = body->oa;
1724 npages = PTLRPC_MAX_BRW_PAGES;
1725 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1726 ioo, remote_nb, &npages, local_nb, NULL);
1730 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1731 BULK_PUT_SOURCE, OST_BULK_PORTAL);
1733 GOTO(out_commitrw, rc = -ENOMEM);
1736 for (i = 0; i < npages; i++) {
1737 int page_rc = local_nb[i].lnb_rc;
1745 if (page_rc != 0) { /* some data! */
1746 LASSERT(local_nb[i].lnb_page != NULL);
1747 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1748 local_nb[i].lnb_page_offset,
1752 if (page_rc != local_nb[i].lnb_len) { /* short read */
1753 /* All subsequent pages should be 0 */
1754 while (++i < npages)
1755 LASSERT(local_nb[i].lnb_rc == 0);
1760 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1761 cksum_type_t cksum_type =
1762 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1763 body->oa.o_flags : 0);
1764 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1765 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1766 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1767 OST_READ, cksum_type);
1768 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1769 repbody->oa.o_cksum);
1771 repbody->oa.o_valid = 0;
1773 /* We're finishing using body->oa as an input variable */
1775 /* Check if client was evicted while we were doing i/o before touching
1777 if (likely(rc == 0 &&
1778 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1779 rc = target_bulk_io(exp, desc, &lwi);
1784 /* Must commit after prep above in all cases */
1785 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1786 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1789 tgt_drop_id(exp, &repbody->oa);
1791 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1793 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1794 ptlrpc_free_bulk_nopin(desc);
1799 ptlrpc_lprocfs_brw(req, nob);
1800 } else if (no_reply) {
1801 req->rq_no_reply = 1;
1802 /* reply out callback would free */
1803 ptlrpc_req_drop_rs(req);
1804 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1805 "client will retry: rc %d\n",
1806 exp->exp_obd->obd_name,
1807 obd_uuid2str(&exp->exp_client_uuid),
1808 obd_export_nid2str(exp), rc);
1810 /* send a bulk after reply to simulate a network delay or reordering
1812 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1813 wait_queue_head_t waitq;
1814 struct l_wait_info lwi1;
1816 CDEBUG(D_INFO, "reorder BULK\n");
1817 init_waitqueue_head(&waitq);
1819 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1820 l_wait_event(waitq, 0, &lwi1);
1821 target_bulk_io(exp, desc, &lwi);
1822 ptlrpc_free_bulk_nopin(desc);
1827 EXPORT_SYMBOL(tgt_brw_read);
1829 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1830 struct ptlrpc_bulk_desc *desc,
1831 struct niobuf_local *local_nb, int npages,
1832 u32 client_cksum, u32 server_cksum,
1835 struct obd_export *exp = req->rq_export;
1836 struct ost_body *body;
1840 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1841 LASSERT(body != NULL);
1843 if (req->rq_peer.nid != desc->bd_sender) {
1845 router = libcfs_nid2str(desc->bd_sender);
1849 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1850 client_cksum, server_cksum);
1854 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1855 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1856 "]: client csum %x, server csum %x\n",
1857 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1859 body->oa.o_valid & OBD_MD_FLFID ?
1860 body->oa.o_parent_seq : (__u64)0,
1861 body->oa.o_valid & OBD_MD_FLFID ?
1862 body->oa.o_parent_oid : 0,
1863 body->oa.o_valid & OBD_MD_FLFID ?
1864 body->oa.o_parent_ver : 0,
1865 POSTID(&body->oa.o_oi),
1866 local_nb[0].lnb_file_offset,
1867 local_nb[npages-1].lnb_file_offset +
1868 local_nb[npages - 1].lnb_len - 1,
1869 client_cksum, server_cksum);
1872 int tgt_brw_write(struct tgt_session_info *tsi)
1874 struct ptlrpc_request *req = tgt_ses_req(tsi);
1875 struct ptlrpc_bulk_desc *desc = NULL;
1876 struct obd_export *exp = req->rq_export;
1877 struct niobuf_remote *remote_nb;
1878 struct niobuf_local *local_nb;
1879 struct obd_ioobj *ioo;
1880 struct ost_body *body, *repbody;
1881 struct l_wait_info lwi;
1882 struct lustre_handle lockh = {0};
1884 int objcount, niocount, npages;
1886 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1887 bool no_reply = false, mmap;
1888 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1892 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1893 CERROR("%s: deny write request from %s to portal %u\n",
1894 tgt_name(tsi->tsi_tgt),
1895 obd_export_nid2str(req->rq_export),
1896 ptlrpc_req2svc(req)->srv_req_portal);
1897 RETURN(err_serious(-EPROTO));
1900 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1901 RETURN(err_serious(-ENOSPC));
1902 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1903 RETURN(err_serious(-EROFS));
1905 req->rq_bulk_write = 1;
1907 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1908 RETURN(err_serious(-EIO));
1909 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1910 RETURN(err_serious(-EFAULT));
1912 /* pause before transaction has been started */
1913 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1914 cfs_fail_val : (obd_timeout + 1) / 4);
1916 /* There must be big cache in current thread to process this request
1917 * if it is NULL then something went wrong and it wasn't allocated,
1918 * report -ENOMEM in that case */
1922 body = tsi->tsi_ost_body;
1923 LASSERT(body != NULL);
1925 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1926 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1928 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1929 RCL_CLIENT) / sizeof(*ioo);
1931 for (niocount = i = 0; i < objcount; i++)
1932 niocount += ioo[i].ioo_bufcnt;
1934 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1935 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1936 if (niocount != req_capsule_get_size(&req->rq_pill,
1937 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1939 RETURN(err_serious(-EPROTO));
1941 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1942 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1943 memory_pressure_set();
1945 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1946 niocount * sizeof(*rcs));
1947 rc = req_capsule_server_pack(&req->rq_pill);
1949 GOTO(out, rc = err_serious(rc));
1951 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1952 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1954 local_nb = tbc->local;
1956 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1957 remote_nb, &lockh, LCK_PW);
1962 * If getting the lock took more time than
1963 * client was willing to wait, drop it. b=11330
1965 if (cfs_time_current_sec() > req->rq_deadline ||
1966 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1968 CERROR("%s: Dropping timed-out write from %s because locking "
1969 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1970 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1971 POSTID(&ioo->ioo_oid),
1972 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1973 req->rq_deadline - req->rq_arrival_time.tv_sec);
1974 GOTO(out_lock, rc = -ETIMEDOUT);
1977 /* Because we already sync grant info with client when reconnect,
1978 * grant info will be cleared for resent req, then fed_grant and
1979 * total_grant will not be modified in following preprw_write */
1980 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1981 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1982 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1985 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1986 if (repbody == NULL)
1987 GOTO(out_lock, rc = -ENOMEM);
1988 repbody->oa = body->oa;
1990 npages = PTLRPC_MAX_BRW_PAGES;
1991 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1992 objcount, ioo, remote_nb, &npages, local_nb, NULL);
1996 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1997 BULK_GET_SINK, OST_BULK_PORTAL);
1999 GOTO(skip_transfer, rc = -ENOMEM);
2001 /* NB Having prepped, we must commit... */
2002 for (i = 0; i < npages; i++)
2003 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2004 local_nb[i].lnb_page_offset,
2005 local_nb[i].lnb_len);
2007 rc = sptlrpc_svc_prep_bulk(req, desc);
2009 GOTO(skip_transfer, rc);
2011 rc = target_bulk_io(exp, desc, &lwi);
2015 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2016 static int cksum_counter;
2018 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2019 cksum_type = cksum_type_unpack(body->oa.o_flags);
2021 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2022 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2023 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2024 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2025 OST_WRITE, cksum_type);
2028 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2029 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2030 body->oa.o_flags & OBD_FL_MMAP);
2032 tgt_warn_on_cksum(req, desc, local_nb, npages,
2034 repbody->oa.o_cksum, mmap);
2036 } else if ((cksum_counter & (-cksum_counter)) ==
2038 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2039 cksum_counter, libcfs_id2str(req->rq_peer),
2040 repbody->oa.o_cksum);
2044 /* Must commit after prep above in all cases */
2045 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2046 objcount, ioo, remote_nb, npages, local_nb, NULL,
2048 if (rc == -ENOTCONN)
2049 /* quota acquire process has been given up because
2050 * either the client has been evicted or the client
2051 * has timed out the request already */
2055 * Disable sending mtime back to the client. If the client locked the
2056 * whole object, then it has already updated the mtime on its side,
2057 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2059 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2064 /* set per-requested niobuf return codes */
2065 for (i = j = 0; i < niocount; i++) {
2066 int len = remote_nb[i].rnb_len;
2071 LASSERT(j < npages);
2072 if (local_nb[j].lnb_rc < 0)
2073 rcs[i] = local_nb[j].lnb_rc;
2074 len -= local_nb[j].lnb_len;
2079 LASSERT(j == npages);
2080 ptlrpc_lprocfs_brw(req, nob);
2082 tgt_drop_id(exp, &repbody->oa);
2085 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2087 ptlrpc_free_bulk_nopin(desc);
2090 req->rq_no_reply = 1;
2091 /* reply out callback would free */
2092 ptlrpc_req_drop_rs(req);
2093 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2094 "client will retry: rc %d\n",
2095 exp->exp_obd->obd_name,
2096 obd_uuid2str(&exp->exp_client_uuid),
2097 obd_export_nid2str(exp), rc);
2099 memory_pressure_clr();
2102 EXPORT_SYMBOL(tgt_brw_write);