4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; if not, write to the
18 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
19 * Boston, MA 021110-1307, USA
24 * Copyright (c) 2011, 2012, Intel Corporation.
27 * lustre/target/tgt_handler.c
29 * Lustre Unified Target request handler code
31 * Author: Brian Behlendorf <behlendorf1@llnl.gov>
32 * Author: Mikhail Pershin <mike.pershin@intel.com>
35 #define DEBUG_SUBSYSTEM S_CLASS
38 #include <obd_class.h>
39 #include <obd_cksum.h>
40 #include <md_object.h>
41 #include <lustre_lfsck.h>
43 #include "tgt_internal.h"
45 char *tgt_name(struct lu_target *tgt)
47 LASSERT(tgt->lut_obd != NULL);
48 return tgt->lut_obd->obd_name;
50 EXPORT_SYMBOL(tgt_name);
53 * Generic code handling requests that have struct mdt_body passed in:
55 * - extract mdt_body from request and save it in @tsi, if present;
57 * - create lu_object, corresponding to the fid in mdt_body, and save it in
60 * - if HABEO_CORPUS flag is set for this request type check whether object
61 * actually exists on storage (lu_object_exists()).
64 static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
66 const struct mdt_body *body;
67 struct lu_object *obj;
68 struct req_capsule *pill = tsi->tsi_pill;
73 body = req_capsule_client_get(pill, &RMF_MDT_BODY);
77 tsi->tsi_mdt_body = body;
79 if (!(body->mbo_valid & OBD_MD_FLID))
82 /* mdc_pack_body() doesn't check if fid is zero and set OBD_ML_FID
83 * in any case in pre-2.5 clients. Fix that here if needed */
84 if (unlikely(fid_is_zero(&body->mbo_fid1)))
87 if (!fid_is_sane(&body->mbo_fid1)) {
88 CERROR("%s: invalid FID: "DFID"\n", tgt_name(tsi->tsi_tgt),
89 PFID(&body->mbo_fid1));
93 obj = lu_object_find(tsi->tsi_env,
94 &tsi->tsi_tgt->lut_bottom->dd_lu_dev,
95 &body->mbo_fid1, NULL);
97 if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
98 lu_object_put(tsi->tsi_env, obj);
99 /* for capability renew ENOENT will be handled in
101 if (body->mbo_valid & OBD_MD_FLOSSCAPA)
106 tsi->tsi_corpus = obj;
113 tsi->tsi_fid = body->mbo_fid1;
119 * Validate oa from client.
120 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
122 * a. objects in Single MDT FS seq = FID_SEQ_OST_MDT0, oi_id != 0
123 * b. Echo objects(seq = 2), old echo client still use oi_id/oi_seq to
124 * pack ost_id. Because non-zero oi_seq will make it diffcult to tell
125 * whether this is oi_fid or real ostid. So it will check
126 * OBD_CONNECT_FID, then convert the ostid to FID for old client.
127 * c. Old FID-disable osc will send IDIF.
128 * d. new FID-enable osc/osp will send normal FID.
130 * And also oi_id/f_oid should always start from 1. oi_id/f_oid = 0 will
131 * be used for LAST_ID file, and only being accessed inside OST now.
133 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
135 struct ost_id *oi = &oa->o_oi;
136 obd_seq seq = ostid_seq(oi);
137 obd_id id = ostid_id(oi);
141 if (unlikely(!(exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_FID) &&
142 fid_seq_is_echo(seq))) {
143 /* Sigh 2.[123] client still sends echo req with oi_id = 0
144 * during create, and we will reset this to 1, since this
145 * oi_id is basically useless in the following create process,
146 * but oi_id == 0 will make it difficult to tell whether it is
147 * real FID or ost_id. */
148 oi->oi_fid.f_seq = FID_SEQ_ECHO;
149 oi->oi_fid.f_oid = id ?: 1;
150 oi->oi_fid.f_ver = 0;
152 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
154 if (unlikely((oa->o_valid & OBD_MD_FLID) && id == 0))
155 GOTO(out, rc = -EPROTO);
157 /* Note: this check might be forced in 2.5 or 2.6, i.e.
158 * all of the requests are required to setup FLGROUP */
159 if (unlikely(!(oa->o_valid & OBD_MD_FLGROUP))) {
160 ostid_set_seq_mdt0(oi);
161 oa->o_valid |= OBD_MD_FLGROUP;
165 if (unlikely(!(fid_seq_is_idif(seq) || fid_seq_is_mdt0(seq) ||
166 fid_seq_is_norm(seq) || fid_seq_is_echo(seq))))
167 GOTO(out, rc = -EPROTO);
169 rc = ostid_to_fid(&tti->tti_fid1, oi,
170 tsi->tsi_tgt->lut_lsd.lsd_osd_index);
171 if (unlikely(rc != 0))
174 oi->oi_fid = tti->tti_fid1;
180 CERROR("%s: client %s sent bad object "DOSTID": rc = %d\n",
181 tgt_name(tsi->tsi_tgt), obd_export_nid2str(tsi->tsi_exp),
185 EXPORT_SYMBOL(tgt_validate_obdo);
187 static int tgt_io_data_unpack(struct tgt_session_info *tsi, struct ost_id *oi)
190 struct niobuf_remote *rnb;
191 struct obd_ioobj *ioo;
196 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
200 rnb = req_capsule_client_get(tsi->tsi_pill, &RMF_NIOBUF_REMOTE);
204 max_brw = ioobj_max_brw_get(ioo);
205 if (unlikely((max_brw & (max_brw - 1)) != 0)) {
206 CERROR("%s: client %s sent bad ioobj max %u for "DOSTID
207 ": rc = %d\n", tgt_name(tsi->tsi_tgt),
208 obd_export_nid2str(tsi->tsi_exp), max_brw,
209 POSTID(oi), -EPROTO);
214 obj_count = req_capsule_get_size(tsi->tsi_pill, &RMF_OBD_IOOBJ,
215 RCL_CLIENT) / sizeof(*ioo);
216 if (obj_count == 0) {
217 CERROR("%s: short ioobj\n", tgt_name(tsi->tsi_tgt));
219 } else if (obj_count > 1) {
220 CERROR("%s: too many ioobjs (%d)\n", tgt_name(tsi->tsi_tgt),
225 if (ioo->ioo_bufcnt == 0) {
226 CERROR("%s: ioo has zero bufcnt\n", tgt_name(tsi->tsi_tgt));
230 if (ioo->ioo_bufcnt > PTLRPC_MAX_BRW_PAGES) {
231 DEBUG_REQ(D_RPCTRACE, tgt_ses_req(tsi),
232 "bulk has too many pages (%d)",
240 static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
242 struct ost_body *body;
243 struct req_capsule *pill = tsi->tsi_pill;
244 struct lustre_capa *capa;
249 body = req_capsule_client_get(pill, &RMF_OST_BODY);
253 rc = tgt_validate_obdo(tsi, &body->oa);
257 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
258 capa = req_capsule_client_get(pill, &RMF_CAPA1);
260 CERROR("%s: OSSCAPA flag is set without capability\n",
261 tgt_name(tsi->tsi_tgt));
266 tsi->tsi_ost_body = body;
267 tsi->tsi_fid = body->oa.o_oi.oi_fid;
269 if (req_capsule_has_field(pill, &RMF_OBD_IOOBJ, RCL_CLIENT)) {
270 rc = tgt_io_data_unpack(tsi, &body->oa.o_oi);
275 if (!(body->oa.o_valid & OBD_MD_FLID)) {
276 if (flags & HABEO_CORPUS) {
277 CERROR("%s: OBD_MD_FLID flag is not set in ost_body "
278 "but OID/FID is mandatory with HABEO_CORPUS\n",
279 tgt_name(tsi->tsi_tgt));
286 ost_fid_build_resid(&tsi->tsi_fid, &tsi->tsi_resid);
289 * OST doesn't get object in advance for further use to prevent
290 * situations with nested object_find which is potential deadlock.
292 tsi->tsi_corpus = NULL;
297 * Do necessary preprocessing according to handler ->th_flags.
299 static int tgt_request_preprocess(struct tgt_session_info *tsi,
300 struct tgt_handler *h,
301 struct ptlrpc_request *req)
303 struct req_capsule *pill = tsi->tsi_pill;
304 __u32 flags = h->th_flags;
309 if (tsi->tsi_preprocessed)
312 LASSERT(h->th_act != NULL);
313 LASSERT(h->th_opc == lustre_msg_get_opc(req->rq_reqmsg));
314 LASSERT(current->journal_info == NULL);
316 LASSERT(ergo(flags & (HABEO_CORPUS | HABEO_REFERO),
318 if (h->th_fmt != NULL) {
319 req_capsule_set(pill, h->th_fmt);
320 if (req_capsule_has_field(pill, &RMF_MDT_BODY, RCL_CLIENT)) {
321 rc = tgt_mdt_body_unpack(tsi, flags);
324 } else if (req_capsule_has_field(pill, &RMF_OST_BODY,
326 rc = tgt_ost_body_unpack(tsi, flags);
332 if (flags & MUTABOR && tgt_conn_flags(tsi) & OBD_CONNECT_RDONLY)
335 if (flags & HABEO_CLAVIS) {
336 struct ldlm_request *dlm_req;
338 LASSERT(h->th_fmt != NULL);
340 dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
341 if (dlm_req != NULL) {
342 if (unlikely(dlm_req->lock_desc.l_resource.lr_type ==
344 dlm_req->lock_desc.l_policy_data.\
345 l_inodebits.bits == 0)) {
347 * Lock without inodebits makes no sense and
348 * will oops later in ldlm. If client miss to
349 * set such bits, do not trigger ASSERTION.
351 * For liblustre flock case, it maybe zero.
355 tsi->tsi_dlm_req = dlm_req;
361 tsi->tsi_preprocessed = 1;
366 * Invoke handler for this request opc. Also do necessary preprocessing
367 * (according to handler ->th_flags), and post-processing (setting of
368 * ->last_{xid,committed}).
370 static int tgt_handle_request0(struct tgt_session_info *tsi,
371 struct tgt_handler *h,
372 struct ptlrpc_request *req)
380 * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
381 * to put same checks into handlers like mdt_close(), mdt_reint(),
382 * etc., without talking to mdt authors first. Checking same thing
383 * there again is useless and returning 0 error without packing reply
384 * is buggy! Handlers either pack reply or return error.
386 * We return 0 here and do not send any reply in order to emulate
387 * network failure. Do not send any reply in case any of NET related
388 * fail_id has occured.
390 if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
393 rc = tgt_request_preprocess(tsi, h, req);
394 /* pack reply if reply format is fixed */
395 if (rc == 0 && h->th_flags & HABEO_REFERO) {
397 if (req_capsule_has_field(tsi->tsi_pill, &RMF_MDT_MD,
399 req_capsule_set_size(tsi->tsi_pill, &RMF_MDT_MD,
401 tsi->tsi_mdt_body->mbo_eadatasize);
402 if (req_capsule_has_field(tsi->tsi_pill, &RMF_LOGCOOKIES,
404 req_capsule_set_size(tsi->tsi_pill, &RMF_LOGCOOKIES,
407 rc = req_capsule_server_pack(tsi->tsi_pill);
410 if (likely(rc == 0)) {
412 * Process request, there can be two types of rc:
413 * 1) errors with msg unpack/pack, other failures outside the
414 * operation itself. This is counted as serious errors;
415 * 2) errors during fs operation, should be placed in rq_status
419 if (!is_serious(rc) &&
420 !req->rq_no_reply && req->rq_reply_state == NULL) {
421 DEBUG_REQ(D_ERROR, req, "%s \"handler\" %s did not "
422 "pack reply and returned 0 error\n",
423 tgt_name(tsi->tsi_tgt), h->th_name);
426 serious = is_serious(rc);
427 rc = clear_serious(rc);
435 * ELDLM_* codes which > 0 should be in rq_status only as well as
436 * all non-serious errors.
438 if (rc > 0 || !serious)
441 LASSERT(current->journal_info == NULL);
443 if (likely(rc == 0 && req->rq_export))
444 target_committed_to_req(req);
446 target_send_reply(req, rc, tsi->tsi_reply_fail_id);
450 static int tgt_filter_recovery_request(struct ptlrpc_request *req,
451 struct obd_device *obd, int *process)
453 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
460 case MDS_DONE_WRITING:
461 case MDS_SYNC: /* used in unmounting */
474 *process = target_queue_recovery_request(req, obd);
478 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
485 * Handle recovery. Return:
486 * +1: continue request processing;
487 * -ve: abort immediately with the given error code;
488 * 0: send reply with error code in req->rq_status;
490 int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
494 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
499 case SEC_CTX_INIT_CONT:
504 if (!req->rq_export->exp_obd->obd_replayable)
507 /* sanity check: if the xid matches, the request must be marked as a
508 * resent or replayed */
509 if (req_xid_is_last(req)) {
510 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
511 (MSG_RESENT | MSG_REPLAY))) {
512 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
513 "last_xid, expected REPLAY or RESENT flag "
515 lustre_msg_get_flags(req->rq_reqmsg));
516 req->rq_status = -ENOTCONN;
520 /* else: note the opposite is not always true; a RESENT req after a
521 * failover will usually not match the last_xid, since it was likely
522 * never committed. A REPLAYed request will almost never match the
523 * last xid, however it could for a committed, but still retained,
526 /* Check for aborted recovery... */
527 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
531 DEBUG_REQ(D_INFO, req, "Got new replay");
532 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
534 if (rc != 0 || !should_process)
536 else if (should_process < 0) {
537 req->rq_status = should_process;
538 rc = ptlrpc_error(req);
545 /* Initial check for request, it is validation mostly */
546 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
548 struct tgt_handler *h;
549 struct tgt_opc_slice *s;
550 struct lu_target *tgt;
551 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
555 tgt = class_exp2tgt(req->rq_export);
557 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
558 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
561 /* opcode was not found in slice */
562 if (unlikely(s->tos_hs == NULL)) {
563 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
565 RETURN(ERR_PTR(-ENOTSUPP));
568 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
569 h = s->tos_hs + (opc - s->tos_opc_start);
570 if (unlikely(h->th_opc == 0)) {
571 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
572 RETURN(ERR_PTR(-ENOTSUPP));
578 int tgt_request_handle(struct ptlrpc_request *req)
580 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
582 struct lustre_msg *msg = req->rq_reqmsg;
583 struct tgt_handler *h;
584 struct lu_target *tgt;
585 int request_fail_id = 0;
586 __u32 opc = lustre_msg_get_opc(msg);
591 /* Refill the context, to make sure all thread keys are allocated */
592 lu_env_refill(req->rq_svc_thread->t_env);
594 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
595 tsi->tsi_pill = &req->rq_pill;
596 tsi->tsi_env = req->rq_svc_thread->t_env;
598 /* if request has export then get handlers slice from corresponding
599 * target, otherwise that should be connect operation */
600 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
601 opc == MGS_CONNECT) {
602 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
603 rc = target_handle_connect(req);
605 rc = ptlrpc_error(req);
608 /* recovery-small test 18c asks to drop connect reply */
609 if (unlikely(opc == OST_CONNECT &&
610 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
614 if (unlikely(!class_connected_export(req->rq_export))) {
615 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
616 opc, libcfs_id2str(req->rq_peer));
617 req->rq_status = -ENOTCONN;
618 rc = ptlrpc_error(req);
622 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
623 tsi->tsi_exp = req->rq_export;
624 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
625 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
627 tsi->tsi_jobid = NULL;
629 request_fail_id = tgt->lut_request_fail_id;
630 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
632 h = tgt_handler_find_check(req);
634 req->rq_status = PTR_ERR(h);
635 rc = ptlrpc_error(req);
639 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
642 rc = lustre_msg_check_version(msg, h->th_version);
644 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
645 " %08x, expecting %08x\n", tgt_name(tgt),
646 lustre_msg_get_version(msg), h->th_version);
647 req->rq_status = -EINVAL;
648 rc = ptlrpc_error(req);
652 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
653 if (likely(rc == 1)) {
654 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
656 rc = tgt_handle_request0(tsi, h, req);
662 req_capsule_fini(tsi->tsi_pill);
663 if (tsi->tsi_corpus != NULL) {
664 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
665 tsi->tsi_corpus = NULL;
669 EXPORT_SYMBOL(tgt_request_handle);
671 /** Assign high priority operations to the request if needed. */
672 int tgt_hpreq_handler(struct ptlrpc_request *req)
674 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
675 struct tgt_handler *h;
680 if (req->rq_export == NULL)
683 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
684 tsi->tsi_pill = &req->rq_pill;
685 tsi->tsi_env = req->rq_svc_thread->t_env;
686 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
687 tsi->tsi_exp = req->rq_export;
689 h = tgt_handler_find_check(req);
695 rc = tgt_request_preprocess(tsi, h, req);
696 if (unlikely(rc != 0))
699 if (h->th_hp != NULL)
703 EXPORT_SYMBOL(tgt_hpreq_handler);
705 void tgt_counter_incr(struct obd_export *exp, int opcode)
707 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
708 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
709 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
711 EXPORT_SYMBOL(tgt_counter_incr);
714 * Unified target generic handlers.
720 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
722 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
723 OBD_CONNECT_RMT_CLIENT_FORCE |
724 OBD_CONNECT_MDS_CAPA |
725 OBD_CONNECT_OSS_CAPA);
728 static int tgt_init_sec_level(struct ptlrpc_request *req)
730 struct lu_target *tgt = class_exp2tgt(req->rq_export);
731 char *client = libcfs_nid2str(req->rq_peer.nid);
732 struct obd_connect_data *data, *reply;
738 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
739 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
740 if (data == NULL || reply == NULL)
743 /* connection from MDT is always trusted */
744 if (req->rq_auth_usr_mdt) {
745 tgt_init_sec_none(reply);
749 /* no GSS support case */
750 if (!req->rq_auth_gss) {
751 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
752 CWARN("client %s -> target %s does not use GSS, "
753 "can not run under security level %d.\n",
754 client, tgt_name(tgt), tgt->lut_sec_level);
757 tgt_init_sec_none(reply);
762 /* old version case */
763 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
764 !(data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) ||
765 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
766 if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
767 CWARN("client %s -> target %s uses old version, "
768 "can not run under security level %d.\n",
769 client, tgt_name(tgt), tgt->lut_sec_level);
772 CWARN("client %s -> target %s uses old version, "
773 "run under security level %d.\n",
774 client, tgt_name(tgt), tgt->lut_sec_level);
775 tgt_init_sec_none(reply);
780 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
782 if (!req->rq_auth_remote)
783 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
784 "asked to be remote.\n", client, tgt_name(tgt));
785 } else if (req->rq_auth_remote) {
787 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
788 "as remote by default.\n", client, tgt_name(tgt));
792 if (!tgt->lut_oss_capa) {
794 "client %s -> target %s is set as remote,"
795 " but OSS capabilities are not enabled: %d.\n",
796 client, tgt_name(tgt), tgt->lut_oss_capa);
800 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
801 CDEBUG(D_SEC, "client %s -> target %s: user is not "
802 "authenticated!\n", client, tgt_name(tgt));
808 switch (tgt->lut_sec_level) {
809 case LUSTRE_SEC_NONE:
812 "client %s -> target %s is set as remote, "
813 "can not run under security level %d.\n",
814 client, tgt_name(tgt), tgt->lut_sec_level);
817 tgt_init_sec_none(reply);
819 case LUSTRE_SEC_REMOTE:
821 tgt_init_sec_none(reply);
826 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
827 OBD_CONNECT_RMT_CLIENT_FORCE);
828 if (!tgt->lut_oss_capa)
829 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
830 if (!tgt->lut_mds_capa)
831 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
840 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
842 struct lu_target *tgt = class_exp2tgt(exp);
843 struct sptlrpc_flavor flvr;
847 LASSERT(tgt->lut_obd);
848 LASSERT(tgt->lut_slice);
850 /* always allow ECHO client */
851 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
852 LUSTRE_ECHO_NAME) == 0)) {
853 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
857 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
858 read_lock(&tgt->lut_sptlrpc_lock);
859 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
863 read_unlock(&tgt->lut_sptlrpc_lock);
865 spin_lock(&exp->exp_lock);
866 exp->exp_sp_peer = req->rq_sp_from;
867 exp->exp_flvr = flvr;
868 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
869 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
870 CERROR("%s: unauthorized rpc flavor %x from %s, "
871 "expect %x\n", tgt_name(tgt),
873 libcfs_nid2str(req->rq_peer.nid),
874 exp->exp_flvr.sf_rpc);
877 spin_unlock(&exp->exp_lock);
879 if (exp->exp_sp_peer != req->rq_sp_from) {
880 CERROR("%s: RPC source %s doesn't match %s\n",
882 sptlrpc_part2name(req->rq_sp_from),
883 sptlrpc_part2name(exp->exp_sp_peer));
886 rc = sptlrpc_target_export_check(exp, req);
893 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
895 struct sptlrpc_rule_set tmp_rset;
898 sptlrpc_rule_set_init(&tmp_rset);
899 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
901 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
906 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
908 write_lock(&tgt->lut_sptlrpc_lock);
909 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
910 tgt->lut_sptlrpc_rset = tmp_rset;
911 write_unlock(&tgt->lut_sptlrpc_lock);
915 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
917 int tgt_connect(struct tgt_session_info *tsi)
919 struct ptlrpc_request *req = tgt_ses_req(tsi);
920 struct obd_connect_data *reply;
925 rc = tgt_init_sec_level(req);
929 /* XXX: better to call this check right after getting new export but
930 * before last_rcvd slot allocation to avoid server load upon insecure
931 * connects. This is to be fixed after unifiyng all targets.
933 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
937 /* To avoid exposing partially initialized connection flags, changes up
938 * to this point have been staged in reply->ocd_connect_flags. Now that
939 * connection handling has completed successfully, atomically update
940 * the connect flags in the shared export data structure. LU-1623 */
941 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
942 spin_lock(&tsi->tsi_exp->exp_lock);
943 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
944 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
945 spin_unlock(&tsi->tsi_exp->exp_lock);
949 obd_disconnect(class_export_get(tsi->tsi_exp));
952 EXPORT_SYMBOL(tgt_connect);
954 int tgt_disconnect(struct tgt_session_info *tsi)
960 rc = target_handle_disconnect(tgt_ses_req(tsi));
962 RETURN(err_serious(rc));
966 EXPORT_SYMBOL(tgt_disconnect);
969 * Unified target OBD handlers
971 int tgt_obd_ping(struct tgt_session_info *tsi)
977 rc = target_handle_ping(tgt_ses_req(tsi));
979 RETURN(err_serious(rc));
983 EXPORT_SYMBOL(tgt_obd_ping);
985 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
987 return err_serious(-EOPNOTSUPP);
989 EXPORT_SYMBOL(tgt_obd_log_cancel);
991 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
993 return err_serious(-EOPNOTSUPP);
995 EXPORT_SYMBOL(tgt_obd_qc_callback);
997 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
999 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1000 struct ptlrpc_request *req = tgt_ses_req(tsi);
1001 struct obd_export *exp = req->rq_export;
1002 struct ptlrpc_bulk_desc *desc;
1003 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
1011 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1016 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1017 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1018 * which is rdpg->rp_count */
1019 nob = rdpg->rp_count;
1021 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1022 i++, tmpcount -= tmpsize) {
1023 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1024 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1027 LASSERT(desc->bd_nob == nob);
1028 rc = target_bulk_io(exp, desc, lwi);
1029 ptlrpc_free_bulk_pin(desc);
1032 EXPORT_SYMBOL(tgt_sendpage);
1035 * OBD_IDX_READ handler
1037 int tgt_obd_idx_read(struct tgt_session_info *tsi)
1039 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1040 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1041 struct idx_info *req_ii, *rep_ii;
1046 memset(rdpg, 0, sizeof(*rdpg));
1047 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1049 /* extract idx_info buffer from request & reply */
1050 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1051 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1052 RETURN(err_serious(-EPROTO));
1054 rc = req_capsule_server_pack(tsi->tsi_pill);
1056 RETURN(err_serious(rc));
1058 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1060 RETURN(err_serious(-EFAULT));
1061 rep_ii->ii_magic = IDX_INFO_MAGIC;
1063 /* extract hash to start with */
1064 rdpg->rp_hash = req_ii->ii_hash_start;
1066 /* extract requested attributes */
1067 rdpg->rp_attrs = req_ii->ii_attrs;
1069 /* check that fid packed in request is valid and supported */
1070 if (!fid_is_sane(&req_ii->ii_fid))
1072 rep_ii->ii_fid = req_ii->ii_fid;
1075 rep_ii->ii_flags = req_ii->ii_flags;
1077 /* compute number of pages to allocate, ii_count is the number of 4KB
1079 if (req_ii->ii_count <= 0)
1080 GOTO(out, rc = -EFAULT);
1081 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1082 exp_max_brw_size(tsi->tsi_exp));
1083 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1085 /* allocate pages to store the containers */
1086 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1087 if (rdpg->rp_pages == NULL)
1088 GOTO(out, rc = -ENOMEM);
1089 for (i = 0; i < rdpg->rp_npages; i++) {
1090 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1091 if (rdpg->rp_pages[i] == NULL)
1092 GOTO(out, rc = -ENOMEM);
1095 /* populate pages with key/record pairs */
1096 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1100 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1101 "asked %d > %d\n", rc, rdpg->rp_count);
1103 /* send pages to client */
1104 rc = tgt_sendpage(tsi, rdpg, rc);
1109 if (rdpg->rp_pages) {
1110 for (i = 0; i < rdpg->rp_npages; i++)
1111 if (rdpg->rp_pages[i])
1112 __free_page(rdpg->rp_pages[i]);
1113 OBD_FREE(rdpg->rp_pages,
1114 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1118 EXPORT_SYMBOL(tgt_obd_idx_read);
1120 struct tgt_handler tgt_obd_handlers[] = {
1121 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1122 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1123 TGT_OBD_HDL_VAR(0, OBD_QC_CALLBACK, tgt_obd_qc_callback),
1124 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1126 EXPORT_SYMBOL(tgt_obd_handlers);
1128 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1129 struct dt_object *obj, __u64 start, __u64 end)
1135 /* if no objid is specified, it means "sync whole filesystem" */
1137 rc = dt_sync(env, tgt->lut_bottom);
1138 } else if (dt_version_get(env, obj) >
1139 tgt->lut_obd->obd_last_committed) {
1140 rc = dt_object_sync(env, obj, start, end);
1145 EXPORT_SYMBOL(tgt_sync);
1147 * Unified target DLM handlers.
1150 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1152 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1153 void *data, int flag)
1156 struct lu_target *tgt;
1157 struct dt_object *obj;
1163 tgt = class_exp2tgt(lock->l_export);
1165 if (flag == LDLM_CB_CANCELING &&
1166 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1167 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1168 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1169 lock->l_flags & LDLM_FL_CBPENDING))) {
1171 __u64 end = OBD_OBJECT_EOF;
1173 rc = lu_env_init(&env, LCT_DT_THREAD);
1174 if (unlikely(rc != 0))
1177 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1178 tgt->lut_lsd.lsd_osd_index);
1179 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1181 GOTO(err_env, rc = PTR_ERR(obj));
1183 if (!dt_object_exists(obj))
1184 GOTO(err_put, rc = -ENOENT);
1186 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1187 start = lock->l_policy_data.l_extent.start;
1188 end = lock->l_policy_data.l_extent.end;
1191 rc = tgt_sync(&env, tgt, obj, start, end);
1193 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1194 "cancel: rc = %d\n",
1195 tgt_name(tgt), PFID(&fid),
1196 lock->l_policy_data.l_extent.start,
1197 lock->l_policy_data.l_extent.end, rc);
1200 lu_object_put(&env, &obj->do_lu);
1205 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1209 struct ldlm_callback_suite tgt_dlm_cbs = {
1210 .lcs_completion = ldlm_server_completion_ast,
1211 .lcs_blocking = tgt_blocking_ast,
1212 .lcs_glimpse = ldlm_server_glimpse_ast
1215 int tgt_enqueue(struct tgt_session_info *tsi)
1217 struct ptlrpc_request *req = tgt_ses_req(tsi);
1222 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1223 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1225 LASSERT(tsi->tsi_dlm_req != NULL);
1226 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1227 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1229 RETURN(err_serious(rc));
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);
1248 EXPORT_SYMBOL(tgt_convert);
1250 int tgt_bl_callback(struct tgt_session_info *tsi)
1252 return err_serious(-EOPNOTSUPP);
1254 EXPORT_SYMBOL(tgt_bl_callback);
1256 int tgt_cp_callback(struct tgt_session_info *tsi)
1258 return err_serious(-EOPNOTSUPP);
1260 EXPORT_SYMBOL(tgt_cp_callback);
1262 /* generic LDLM target handler */
1263 struct tgt_handler tgt_dlm_handlers[] = {
1264 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1265 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1266 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1267 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1269 EXPORT_SYMBOL(tgt_dlm_handlers);
1272 * Unified target LLOG handlers.
1274 int tgt_llog_open(struct tgt_session_info *tsi)
1280 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1284 EXPORT_SYMBOL(tgt_llog_open);
1286 int tgt_llog_close(struct tgt_session_info *tsi)
1292 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1296 EXPORT_SYMBOL(tgt_llog_close);
1299 int tgt_llog_destroy(struct tgt_session_info *tsi)
1305 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1309 EXPORT_SYMBOL(tgt_llog_destroy);
1311 int tgt_llog_read_header(struct tgt_session_info *tsi)
1317 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1321 EXPORT_SYMBOL(tgt_llog_read_header);
1323 int tgt_llog_next_block(struct tgt_session_info *tsi)
1329 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1333 EXPORT_SYMBOL(tgt_llog_next_block);
1335 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1341 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1345 EXPORT_SYMBOL(tgt_llog_prev_block);
1347 /* generic llog target handler */
1348 struct tgt_handler tgt_llog_handlers[] = {
1349 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1350 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1351 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1352 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1353 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1354 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1356 EXPORT_SYMBOL(tgt_llog_handlers);
1359 * sec context handlers
1361 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1362 int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1367 struct tgt_handler tgt_sec_ctx_handlers[] = {
1368 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1369 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1370 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1372 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1374 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1375 struct dt_device *key,
1376 struct lfsck_request *lr) = NULL;
1378 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1380 struct lfsck_request *))
1382 tgt_lfsck_in_notify = notify;
1384 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1386 static int (*tgt_lfsck_query)(const struct lu_env *env,
1387 struct dt_device *key,
1388 struct lfsck_request *lr) = NULL;
1390 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1392 struct lfsck_request *))
1394 tgt_lfsck_query = query;
1396 EXPORT_SYMBOL(tgt_register_lfsck_query);
1398 /* LFSCK request handlers */
1399 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1401 const struct lu_env *env = tsi->tsi_env;
1402 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1403 struct lfsck_request *lr;
1407 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1411 rc = tgt_lfsck_in_notify(env, key, lr);
1416 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1418 struct lfsck_request *request;
1419 struct lfsck_reply *reply;
1423 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1424 if (request == NULL)
1427 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1431 reply->lr_status = tgt_lfsck_query(tsi->tsi_env,
1432 tsi->tsi_tgt->lut_bottom, request);
1433 if (reply->lr_status < 0)
1434 rc = reply->lr_status;
1439 struct tgt_handler tgt_lfsck_handlers[] = {
1440 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1441 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1443 EXPORT_SYMBOL(tgt_lfsck_handlers);
1446 * initialize per-thread page pool (bug 5137).
1448 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1450 struct tgt_thread_big_cache *tbc;
1454 LASSERT(thread != NULL);
1455 LASSERT(thread->t_data == NULL);
1457 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1460 thread->t_data = tbc;
1463 EXPORT_SYMBOL(tgt_io_thread_init);
1466 * free per-thread pool created by tgt_thread_init().
1468 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1470 struct tgt_thread_big_cache *tbc;
1474 LASSERT(thread != NULL);
1477 * be prepared to handle partially-initialized pools (because this is
1478 * called from ost_io_thread_init() for cleanup.
1480 tbc = thread->t_data;
1482 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1483 thread->t_data = NULL;
1487 EXPORT_SYMBOL(tgt_io_thread_done);
1489 * Helper function for getting server side [start, start+count] DLM lock
1490 * if asked by client.
1492 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1493 __u64 start, __u64 end, struct lustre_handle *lh,
1494 int mode, __u64 *flags)
1496 ldlm_policy_data_t policy;
1501 LASSERT(lh != NULL);
1502 LASSERT(ns != NULL);
1503 LASSERT(!lustre_handle_is_used(lh));
1505 policy.l_extent.gid = 0;
1506 policy.l_extent.start = start & CFS_PAGE_MASK;
1509 * If ->o_blocks is EOF it means "lock till the end of the file".
1510 * Otherwise, it's size of an extent or hole being punched (in bytes).
1512 if (end == OBD_OBJECT_EOF || end < start)
1513 policy.l_extent.end = OBD_OBJECT_EOF;
1515 policy.l_extent.end = end | ~CFS_PAGE_MASK;
1517 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1518 flags, ldlm_blocking_ast,
1519 ldlm_completion_ast, ldlm_glimpse_ast,
1520 NULL, 0, LVB_T_NONE, NULL, lh);
1521 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1523 EXPORT_SYMBOL(tgt_extent_lock);
1525 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1527 LASSERT(lustre_handle_is_used(lh));
1528 ldlm_lock_decref(lh, mode);
1530 EXPORT_SYMBOL(tgt_extent_unlock);
1532 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1533 struct obd_ioobj *obj, struct niobuf_remote *nb,
1534 struct lustre_handle *lh, int mode)
1537 int nrbufs = obj->ioo_bufcnt;
1542 LASSERT(mode == LCK_PR || mode == LCK_PW);
1543 LASSERT(!lustre_handle_is_used(lh));
1545 if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
1548 for (i = 1; i < nrbufs; i++)
1549 if (!(nb[i].flags & OBD_BRW_SRVLOCK))
1552 RETURN(tgt_extent_lock(ns, res_id, nb[0].offset,
1553 nb[nrbufs - 1].offset + nb[nrbufs - 1].len - 1,
1556 EXPORT_SYMBOL(tgt_brw_lock);
1558 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1559 struct lustre_handle *lh, int mode)
1563 LASSERT(mode == LCK_PR || mode == LCK_PW);
1564 LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
1565 lustre_handle_is_used(lh));
1566 if (lustre_handle_is_used(lh))
1567 tgt_extent_unlock(lh, mode);
1570 EXPORT_SYMBOL(tgt_brw_unlock);
1572 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1573 struct ptlrpc_bulk_desc *desc, int opc,
1574 cksum_type_t cksum_type)
1576 struct cfs_crypto_hash_desc *hdesc;
1577 unsigned int bufsize;
1579 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1582 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1583 if (IS_ERR(hdesc)) {
1584 CERROR("%s: unable to initialize checksum hash %s\n",
1585 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1586 return PTR_ERR(hdesc);
1589 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1590 for (i = 0; i < desc->bd_iov_count; i++) {
1591 /* corrupt the data before we compute the checksum, to
1592 * simulate a client->OST data error */
1593 if (i == 0 && opc == OST_WRITE &&
1594 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1595 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1596 int len = desc->bd_iov[i].kiov_len;
1597 struct page *np = tgt_page_to_corrupt;
1598 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1601 char *ptr2 = kmap(np) + off;
1603 memcpy(ptr2, ptr, len);
1604 memcpy(ptr2, "bad3", min(4, len));
1606 desc->bd_iov[i].kiov_page = np;
1608 CERROR("%s: can't alloc page for corruption\n",
1612 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1613 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
1614 desc->bd_iov[i].kiov_len);
1616 /* corrupt the data after we compute the checksum, to
1617 * simulate an OST->client data error */
1618 if (i == 0 && opc == OST_READ &&
1619 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1620 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
1621 int len = desc->bd_iov[i].kiov_len;
1622 struct page *np = tgt_page_to_corrupt;
1623 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1626 char *ptr2 = kmap(np) + off;
1628 memcpy(ptr2, ptr, len);
1629 memcpy(ptr2, "bad4", min(4, len));
1631 desc->bd_iov[i].kiov_page = np;
1633 CERROR("%s: can't alloc page for corruption\n",
1639 bufsize = sizeof(cksum);
1640 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1645 int tgt_brw_read(struct tgt_session_info *tsi)
1647 struct ptlrpc_request *req = tgt_ses_req(tsi);
1648 struct ptlrpc_bulk_desc *desc = NULL;
1649 struct obd_export *exp = tsi->tsi_exp;
1650 struct niobuf_remote *remote_nb;
1651 struct niobuf_local *local_nb;
1652 struct obd_ioobj *ioo;
1653 struct ost_body *body, *repbody;
1654 struct l_wait_info lwi;
1655 struct lustre_handle lockh = { 0 };
1656 int niocount, npages, nob = 0, rc, i;
1658 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1662 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1663 CERROR("%s: deny read request from %s to portal %u\n",
1664 tgt_name(tsi->tsi_tgt),
1665 obd_export_nid2str(req->rq_export),
1666 ptlrpc_req2svc(req)->srv_req_portal);
1670 req->rq_bulk_read = 1;
1672 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1675 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1676 cfs_fail_val : (obd_timeout + 1) / 4);
1678 /* Check if there is eviction in progress, and if so, wait for it to
1680 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1681 /* We do not care how long it takes */
1682 lwi = LWI_INTR(NULL, NULL);
1683 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1684 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1688 /* There must be big cache in current thread to process this request
1689 * if it is NULL then something went wrong and it wasn't allocated,
1690 * report -ENOMEM in that case */
1694 body = tsi->tsi_ost_body;
1695 LASSERT(body != NULL);
1697 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1698 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1700 niocount = ioo->ioo_bufcnt;
1701 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1702 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1704 local_nb = tbc->local;
1706 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1707 remote_nb, &lockh, LCK_PR);
1712 * If getting the lock took more time than
1713 * client was willing to wait, drop it. b=11330
1715 if (cfs_time_current_sec() > req->rq_deadline ||
1716 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1718 CERROR("Dropping timed-out read from %s because locking"
1719 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1720 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1721 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1722 req->rq_deadline - req->rq_arrival_time.tv_sec);
1723 GOTO(out_lock, rc = -ETIMEDOUT);
1726 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1727 repbody->oa = body->oa;
1729 npages = PTLRPC_MAX_BRW_PAGES;
1730 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1731 ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
1735 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1736 BULK_PUT_SOURCE, OST_BULK_PORTAL);
1738 GOTO(out_commitrw, rc = -ENOMEM);
1741 for (i = 0; i < npages; i++) {
1742 int page_rc = local_nb[i].rc;
1750 if (page_rc != 0) { /* some data! */
1751 LASSERT(local_nb[i].page != NULL);
1752 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
1753 local_nb[i].lnb_page_offset,
1757 if (page_rc != local_nb[i].len) { /* short read */
1758 /* All subsequent pages should be 0 */
1759 while (++i < npages)
1760 LASSERT(local_nb[i].rc == 0);
1765 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1766 cksum_type_t cksum_type =
1767 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1768 body->oa.o_flags : 0);
1769 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1770 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1771 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1772 OST_READ, cksum_type);
1773 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1774 repbody->oa.o_cksum);
1776 repbody->oa.o_valid = 0;
1778 /* We're finishing using body->oa as an input variable */
1780 /* Check if client was evicted while we were doing i/o before touching
1782 if (likely(rc == 0 &&
1783 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1784 rc = target_bulk_io(exp, desc, &lwi);
1789 /* Must commit after prep above in all cases */
1790 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1791 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1794 tgt_drop_id(exp, &repbody->oa);
1796 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1798 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1799 ptlrpc_free_bulk_nopin(desc);
1804 ptlrpc_lprocfs_brw(req, nob);
1805 } else if (no_reply) {
1806 req->rq_no_reply = 1;
1807 /* reply out callback would free */
1808 ptlrpc_req_drop_rs(req);
1809 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1810 "client will retry: rc %d\n",
1811 exp->exp_obd->obd_name,
1812 obd_uuid2str(&exp->exp_client_uuid),
1813 obd_export_nid2str(exp), rc);
1815 /* send a bulk after reply to simulate a network delay or reordering
1817 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1818 wait_queue_head_t waitq;
1819 struct l_wait_info lwi1;
1821 CDEBUG(D_INFO, "reorder BULK\n");
1822 init_waitqueue_head(&waitq);
1824 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1825 l_wait_event(waitq, 0, &lwi1);
1826 target_bulk_io(exp, desc, &lwi);
1827 ptlrpc_free_bulk_nopin(desc);
1832 EXPORT_SYMBOL(tgt_brw_read);
1834 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1835 struct ptlrpc_bulk_desc *desc,
1836 struct niobuf_local *local_nb, int npages,
1837 obd_count client_cksum, obd_count server_cksum,
1840 struct obd_export *exp = req->rq_export;
1841 struct ost_body *body;
1845 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1846 LASSERT(body != NULL);
1848 if (req->rq_peer.nid == desc->bd_sender) {
1852 router = libcfs_nid2str(desc->bd_sender);
1856 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1857 client_cksum, server_cksum);
1861 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1862 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1863 "]: client csum %x, server csum %x\n",
1864 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1866 body->oa.o_valid & OBD_MD_FLFID ?
1867 body->oa.o_parent_seq : (__u64)0,
1868 body->oa.o_valid & OBD_MD_FLFID ?
1869 body->oa.o_parent_oid : 0,
1870 body->oa.o_valid & OBD_MD_FLFID ?
1871 body->oa.o_parent_ver : 0,
1872 POSTID(&body->oa.o_oi),
1873 local_nb[0].lnb_file_offset,
1874 local_nb[npages-1].lnb_file_offset +
1875 local_nb[npages-1].len - 1,
1876 client_cksum, server_cksum);
1879 int tgt_brw_write(struct tgt_session_info *tsi)
1881 struct ptlrpc_request *req = tgt_ses_req(tsi);
1882 struct ptlrpc_bulk_desc *desc = NULL;
1883 struct obd_export *exp = req->rq_export;
1884 struct niobuf_remote *remote_nb;
1885 struct niobuf_local *local_nb;
1886 struct obd_ioobj *ioo;
1887 struct ost_body *body, *repbody;
1888 struct l_wait_info lwi;
1889 struct lustre_handle lockh = {0};
1891 int objcount, niocount, npages;
1893 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1894 bool no_reply = false, mmap;
1895 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1899 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1900 CERROR("%s: deny write request from %s to portal %u\n",
1901 tgt_name(tsi->tsi_tgt),
1902 obd_export_nid2str(req->rq_export),
1903 ptlrpc_req2svc(req)->srv_req_portal);
1904 RETURN(err_serious(-EPROTO));
1907 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1908 RETURN(err_serious(-ENOSPC));
1909 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1910 RETURN(err_serious(-EROFS));
1912 req->rq_bulk_write = 1;
1914 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1915 RETURN(err_serious(-EIO));
1916 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1917 RETURN(err_serious(-EFAULT));
1919 /* pause before transaction has been started */
1920 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1921 cfs_fail_val : (obd_timeout + 1) / 4);
1923 /* There must be big cache in current thread to process this request
1924 * if it is NULL then something went wrong and it wasn't allocated,
1925 * report -ENOMEM in that case */
1929 body = tsi->tsi_ost_body;
1930 LASSERT(body != NULL);
1932 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1933 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1935 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1936 RCL_CLIENT) / sizeof(*ioo);
1938 for (niocount = i = 0; i < objcount; i++)
1939 niocount += ioo[i].ioo_bufcnt;
1941 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1942 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1943 if (niocount != req_capsule_get_size(&req->rq_pill,
1944 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1946 RETURN(err_serious(-EPROTO));
1948 if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
1949 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1950 memory_pressure_set();
1952 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1953 niocount * sizeof(*rcs));
1954 rc = req_capsule_server_pack(&req->rq_pill);
1956 GOTO(out, rc = err_serious(rc));
1958 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1959 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1961 local_nb = tbc->local;
1963 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1964 remote_nb, &lockh, LCK_PW);
1969 * If getting the lock took more time than
1970 * client was willing to wait, drop it. b=11330
1972 if (cfs_time_current_sec() > req->rq_deadline ||
1973 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1975 CERROR("%s: Dropping timed-out write from %s because locking "
1976 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1977 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1978 POSTID(&ioo->ioo_oid),
1979 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1980 req->rq_deadline - req->rq_arrival_time.tv_sec);
1981 GOTO(out_lock, rc = -ETIMEDOUT);
1984 /* Because we already sync grant info with client when reconnect,
1985 * grant info will be cleared for resent req, then fed_grant and
1986 * total_grant will not be modified in following preprw_write */
1987 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1988 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1989 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1992 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1993 if (repbody == NULL)
1994 GOTO(out_lock, rc = -ENOMEM);
1995 repbody->oa = body->oa;
1997 npages = PTLRPC_MAX_BRW_PAGES;
1998 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1999 objcount, ioo, remote_nb, &npages, local_nb, NULL,
2004 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
2005 BULK_GET_SINK, OST_BULK_PORTAL);
2007 GOTO(skip_transfer, rc = -ENOMEM);
2009 /* NB Having prepped, we must commit... */
2010 for (i = 0; i < npages; i++)
2011 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].page,
2012 local_nb[i].lnb_page_offset,
2015 rc = sptlrpc_svc_prep_bulk(req, desc);
2017 GOTO(skip_transfer, rc);
2019 rc = target_bulk_io(exp, desc, &lwi);
2023 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2024 static int cksum_counter;
2026 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2027 cksum_type = cksum_type_unpack(body->oa.o_flags);
2029 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2030 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2031 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2032 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2033 OST_WRITE, cksum_type);
2036 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2037 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2038 body->oa.o_flags & OBD_FL_MMAP);
2040 tgt_warn_on_cksum(req, desc, local_nb, npages,
2042 repbody->oa.o_cksum, mmap);
2044 } else if ((cksum_counter & (-cksum_counter)) ==
2046 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2047 cksum_counter, libcfs_id2str(req->rq_peer),
2048 repbody->oa.o_cksum);
2052 /* Must commit after prep above in all cases */
2053 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2054 objcount, ioo, remote_nb, npages, local_nb, NULL,
2056 if (rc == -ENOTCONN)
2057 /* quota acquire process has been given up because
2058 * either the client has been evicted or the client
2059 * has timed out the request already */
2063 * Disable sending mtime back to the client. If the client locked the
2064 * whole object, then it has already updated the mtime on its side,
2065 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2067 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2072 /* set per-requested niobuf return codes */
2073 for (i = j = 0; i < niocount; i++) {
2074 int len = remote_nb[i].len;
2079 LASSERT(j < npages);
2080 if (local_nb[j].rc < 0)
2081 rcs[i] = local_nb[j].rc;
2082 len -= local_nb[j].len;
2087 LASSERT(j == npages);
2088 ptlrpc_lprocfs_brw(req, nob);
2090 tgt_drop_id(exp, &repbody->oa);
2093 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2095 ptlrpc_free_bulk_nopin(desc);
2098 req->rq_no_reply = 1;
2099 /* reply out callback would free */
2100 ptlrpc_req_drop_rs(req);
2101 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2102 "client will retry: rc %d\n",
2103 exp->exp_obd->obd_name,
2104 obd_uuid2str(&exp->exp_client_uuid),
2105 obd_export_nid2str(exp), rc);
2107 memory_pressure_clr();
2110 EXPORT_SYMBOL(tgt_brw_write);