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_DONE_WRITING:
462 case MDS_SYNC: /* used in unmounting */
476 *process = target_queue_recovery_request(req, obd);
480 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
487 * Handle recovery. Return:
488 * +1: continue request processing;
489 * -ve: abort immediately with the given error code;
490 * 0: send reply with error code in req->rq_status;
492 static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
496 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
501 case SEC_CTX_INIT_CONT:
506 if (!req->rq_export->exp_obd->obd_replayable)
509 /* sanity check: if the xid matches, the request must be marked as a
510 * resent or replayed */
511 if (req_xid_is_last(req)) {
512 if (!(lustre_msg_get_flags(req->rq_reqmsg) &
513 (MSG_RESENT | MSG_REPLAY))) {
514 DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
515 "last_xid, expected REPLAY or RESENT flag "
517 lustre_msg_get_flags(req->rq_reqmsg));
518 req->rq_status = -ENOTCONN;
522 /* else: note the opposite is not always true; a RESENT req after a
523 * failover will usually not match the last_xid, since it was likely
524 * never committed. A REPLAYed request will almost never match the
525 * last xid, however it could for a committed, but still retained,
528 /* Check for aborted recovery... */
529 if (unlikely(req->rq_export->exp_obd->obd_recovering)) {
533 DEBUG_REQ(D_INFO, req, "Got new replay");
534 rc = tgt_filter_recovery_request(req, req->rq_export->exp_obd,
536 if (rc != 0 || !should_process)
538 else if (should_process < 0) {
539 req->rq_status = should_process;
540 rc = ptlrpc_error(req);
547 /* Initial check for request, it is validation mostly */
548 static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
550 struct tgt_handler *h;
551 struct tgt_opc_slice *s;
552 struct lu_target *tgt;
553 __u32 opc = lustre_msg_get_opc(req->rq_reqmsg);
557 tgt = class_exp2tgt(req->rq_export);
559 for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
560 if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
563 /* opcode was not found in slice */
564 if (unlikely(s->tos_hs == NULL)) {
565 CERROR("%s: no handlers for opcode 0x%x\n", tgt_name(tgt),
567 RETURN(ERR_PTR(-ENOTSUPP));
570 LASSERT(opc >= s->tos_opc_start && opc < s->tos_opc_end);
571 h = s->tos_hs + (opc - s->tos_opc_start);
572 if (unlikely(h->th_opc == 0)) {
573 CERROR("%s: unsupported opcode 0x%x\n", tgt_name(tgt), opc);
574 RETURN(ERR_PTR(-ENOTSUPP));
580 int tgt_request_handle(struct ptlrpc_request *req)
582 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
584 struct lustre_msg *msg = req->rq_reqmsg;
585 struct tgt_handler *h;
586 struct lu_target *tgt;
587 int request_fail_id = 0;
588 __u32 opc = lustre_msg_get_opc(msg);
593 /* Refill the context, to make sure all thread keys are allocated */
594 lu_env_refill(req->rq_svc_thread->t_env);
596 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
597 tsi->tsi_pill = &req->rq_pill;
598 tsi->tsi_env = req->rq_svc_thread->t_env;
600 /* if request has export then get handlers slice from corresponding
601 * target, otherwise that should be connect operation */
602 if (opc == MDS_CONNECT || opc == OST_CONNECT ||
603 opc == MGS_CONNECT) {
604 req_capsule_set(&req->rq_pill, &RQF_CONNECT);
605 rc = target_handle_connect(req);
607 rc = ptlrpc_error(req);
610 /* recovery-small test 18c asks to drop connect reply */
611 if (unlikely(opc == OST_CONNECT &&
612 OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2)))
616 if (unlikely(!class_connected_export(req->rq_export))) {
617 CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
618 opc, libcfs_id2str(req->rq_peer));
619 req->rq_status = -ENOTCONN;
620 rc = ptlrpc_error(req);
624 tsi->tsi_tgt = tgt = class_exp2tgt(req->rq_export);
625 tsi->tsi_exp = req->rq_export;
626 if (exp_connect_flags(req->rq_export) & OBD_CONNECT_JOBSTATS)
627 tsi->tsi_jobid = lustre_msg_get_jobid(req->rq_reqmsg);
629 tsi->tsi_jobid = NULL;
631 request_fail_id = tgt->lut_request_fail_id;
632 tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
634 h = tgt_handler_find_check(req);
636 req->rq_status = PTR_ERR(h);
637 rc = ptlrpc_error(req);
641 if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
644 rc = lustre_msg_check_version(msg, h->th_version);
646 DEBUG_REQ(D_ERROR, req, "%s: drop mal-formed request, version"
647 " %08x, expecting %08x\n", tgt_name(tgt),
648 lustre_msg_get_version(msg), h->th_version);
649 req->rq_status = -EINVAL;
650 rc = ptlrpc_error(req);
654 rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
655 if (likely(rc == 1)) {
656 LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
658 rc = tgt_handle_request0(tsi, h, req);
664 req_capsule_fini(tsi->tsi_pill);
665 if (tsi->tsi_corpus != NULL) {
666 lu_object_put(tsi->tsi_env, tsi->tsi_corpus);
667 tsi->tsi_corpus = NULL;
671 EXPORT_SYMBOL(tgt_request_handle);
673 /** Assign high priority operations to the request if needed. */
674 int tgt_hpreq_handler(struct ptlrpc_request *req)
676 struct tgt_session_info *tsi = tgt_ses_info(req->rq_svc_thread->t_env);
677 struct tgt_handler *h;
682 if (req->rq_export == NULL)
685 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
686 tsi->tsi_pill = &req->rq_pill;
687 tsi->tsi_env = req->rq_svc_thread->t_env;
688 tsi->tsi_tgt = class_exp2tgt(req->rq_export);
689 tsi->tsi_exp = req->rq_export;
691 h = tgt_handler_find_check(req);
697 rc = tgt_request_preprocess(tsi, h, req);
698 if (unlikely(rc != 0))
701 if (h->th_hp != NULL)
705 EXPORT_SYMBOL(tgt_hpreq_handler);
707 void tgt_counter_incr(struct obd_export *exp, int opcode)
709 lprocfs_counter_incr(exp->exp_obd->obd_stats, opcode);
710 if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
711 lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
713 EXPORT_SYMBOL(tgt_counter_incr);
716 * Unified target generic handlers.
722 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
724 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
725 OBD_CONNECT_RMT_CLIENT_FORCE);
728 static int tgt_init_sec_level(struct ptlrpc_request *req)
730 struct lu_target *tgt = class_exp2tgt(req->rq_export);
732 struct obd_connect_data *data, *reply;
737 data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
738 reply = req_capsule_server_get(&req->rq_pill, &RMF_CONNECT_DATA);
739 if (data == NULL || reply == NULL)
742 /* connection from MDT is always trusted */
743 if (req->rq_auth_usr_mdt) {
744 tgt_init_sec_none(reply);
748 client = libcfs_nid2str(req->rq_peer.nid);
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 (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
793 CDEBUG(D_SEC, "client %s -> target %s: user is not "
794 "authenticated!\n", client, tgt_name(tgt));
800 switch (tgt->lut_sec_level) {
801 case LUSTRE_SEC_NONE:
804 "client %s -> target %s is set as remote, "
805 "can not run under security level %d.\n",
806 client, tgt_name(tgt), tgt->lut_sec_level);
809 tgt_init_sec_none(reply);
811 case LUSTRE_SEC_REMOTE:
813 tgt_init_sec_none(reply);
818 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
819 OBD_CONNECT_RMT_CLIENT_FORCE);
820 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
821 reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
830 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp)
832 struct lu_target *tgt = class_exp2tgt(exp);
833 struct sptlrpc_flavor flvr;
837 LASSERT(tgt->lut_obd);
838 LASSERT(tgt->lut_slice);
840 /* always allow ECHO client */
841 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
842 LUSTRE_ECHO_NAME) == 0)) {
843 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
847 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
848 read_lock(&tgt->lut_sptlrpc_lock);
849 sptlrpc_target_choose_flavor(&tgt->lut_sptlrpc_rset,
853 read_unlock(&tgt->lut_sptlrpc_lock);
855 spin_lock(&exp->exp_lock);
856 exp->exp_sp_peer = req->rq_sp_from;
857 exp->exp_flvr = flvr;
858 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
859 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
860 CERROR("%s: unauthorized rpc flavor %x from %s, "
861 "expect %x\n", tgt_name(tgt),
863 libcfs_nid2str(req->rq_peer.nid),
864 exp->exp_flvr.sf_rpc);
867 spin_unlock(&exp->exp_lock);
869 if (exp->exp_sp_peer != req->rq_sp_from) {
870 CERROR("%s: RPC source %s doesn't match %s\n",
872 sptlrpc_part2name(req->rq_sp_from),
873 sptlrpc_part2name(exp->exp_sp_peer));
876 rc = sptlrpc_target_export_check(exp, req);
883 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
885 struct sptlrpc_rule_set tmp_rset;
888 sptlrpc_rule_set_init(&tmp_rset);
889 rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
891 CERROR("%s: failed get sptlrpc rules: rc = %d\n",
896 sptlrpc_target_update_exp_flavor(tgt->lut_obd, &tmp_rset);
898 write_lock(&tgt->lut_sptlrpc_lock);
899 sptlrpc_rule_set_free(&tgt->lut_sptlrpc_rset);
900 tgt->lut_sptlrpc_rset = tmp_rset;
901 write_unlock(&tgt->lut_sptlrpc_lock);
905 EXPORT_SYMBOL(tgt_adapt_sptlrpc_conf);
907 int tgt_connect(struct tgt_session_info *tsi)
909 struct ptlrpc_request *req = tgt_ses_req(tsi);
910 struct obd_connect_data *reply;
915 rc = tgt_init_sec_level(req);
919 /* XXX: better to call this check right after getting new export but
920 * before last_rcvd slot allocation to avoid server load upon insecure
921 * connects. This is to be fixed after unifiyng all targets.
923 rc = tgt_connect_check_sptlrpc(req, tsi->tsi_exp);
927 /* To avoid exposing partially initialized connection flags, changes up
928 * to this point have been staged in reply->ocd_connect_flags. Now that
929 * connection handling has completed successfully, atomically update
930 * the connect flags in the shared export data structure. LU-1623 */
931 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_CONNECT_DATA);
932 spin_lock(&tsi->tsi_exp->exp_lock);
933 *exp_connect_flags_ptr(tsi->tsi_exp) = reply->ocd_connect_flags;
934 tsi->tsi_exp->exp_connect_data.ocd_brw_size = reply->ocd_brw_size;
935 spin_unlock(&tsi->tsi_exp->exp_lock);
939 obd_disconnect(class_export_get(tsi->tsi_exp));
942 EXPORT_SYMBOL(tgt_connect);
944 int tgt_disconnect(struct tgt_session_info *tsi)
950 rc = target_handle_disconnect(tgt_ses_req(tsi));
952 RETURN(err_serious(rc));
956 EXPORT_SYMBOL(tgt_disconnect);
959 * Unified target OBD handlers
961 int tgt_obd_ping(struct tgt_session_info *tsi)
967 rc = target_handle_ping(tgt_ses_req(tsi));
969 RETURN(err_serious(rc));
973 EXPORT_SYMBOL(tgt_obd_ping);
975 int tgt_obd_log_cancel(struct tgt_session_info *tsi)
977 return err_serious(-EOPNOTSUPP);
980 int tgt_obd_qc_callback(struct tgt_session_info *tsi)
982 return err_serious(-EOPNOTSUPP);
985 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
987 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
988 struct ptlrpc_request *req = tgt_ses_req(tsi);
989 struct obd_export *exp = req->rq_export;
990 struct ptlrpc_bulk_desc *desc;
991 struct l_wait_info *lwi = &tti->tti_u.rdpg.tti_wait_info;
999 desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
1004 if (!(exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE))
1005 /* old client requires reply size in it's PAGE_CACHE_SIZE,
1006 * which is rdpg->rp_count */
1007 nob = rdpg->rp_count;
1009 for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
1010 i++, tmpcount -= tmpsize) {
1011 tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
1012 ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
1015 LASSERT(desc->bd_nob == nob);
1016 rc = target_bulk_io(exp, desc, lwi);
1017 ptlrpc_free_bulk_pin(desc);
1020 EXPORT_SYMBOL(tgt_sendpage);
1023 * OBD_IDX_READ handler
1025 static int tgt_obd_idx_read(struct tgt_session_info *tsi)
1027 struct tgt_thread_info *tti = tgt_th_info(tsi->tsi_env);
1028 struct lu_rdpg *rdpg = &tti->tti_u.rdpg.tti_rdpg;
1029 struct idx_info *req_ii, *rep_ii;
1034 memset(rdpg, 0, sizeof(*rdpg));
1035 req_capsule_set(tsi->tsi_pill, &RQF_OBD_IDX_READ);
1037 /* extract idx_info buffer from request & reply */
1038 req_ii = req_capsule_client_get(tsi->tsi_pill, &RMF_IDX_INFO);
1039 if (req_ii == NULL || req_ii->ii_magic != IDX_INFO_MAGIC)
1040 RETURN(err_serious(-EPROTO));
1042 rc = req_capsule_server_pack(tsi->tsi_pill);
1044 RETURN(err_serious(rc));
1046 rep_ii = req_capsule_server_get(tsi->tsi_pill, &RMF_IDX_INFO);
1048 RETURN(err_serious(-EFAULT));
1049 rep_ii->ii_magic = IDX_INFO_MAGIC;
1051 /* extract hash to start with */
1052 rdpg->rp_hash = req_ii->ii_hash_start;
1054 /* extract requested attributes */
1055 rdpg->rp_attrs = req_ii->ii_attrs;
1057 /* check that fid packed in request is valid and supported */
1058 if (!fid_is_sane(&req_ii->ii_fid))
1060 rep_ii->ii_fid = req_ii->ii_fid;
1063 rep_ii->ii_flags = req_ii->ii_flags;
1065 /* compute number of pages to allocate, ii_count is the number of 4KB
1067 if (req_ii->ii_count <= 0)
1068 GOTO(out, rc = -EFAULT);
1069 rdpg->rp_count = min_t(unsigned int, req_ii->ii_count << LU_PAGE_SHIFT,
1070 exp_max_brw_size(tsi->tsi_exp));
1071 rdpg->rp_npages = (rdpg->rp_count + PAGE_CACHE_SIZE -1) >> PAGE_CACHE_SHIFT;
1073 /* allocate pages to store the containers */
1074 OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1075 if (rdpg->rp_pages == NULL)
1076 GOTO(out, rc = -ENOMEM);
1077 for (i = 0; i < rdpg->rp_npages; i++) {
1078 rdpg->rp_pages[i] = alloc_page(GFP_IOFS);
1079 if (rdpg->rp_pages[i] == NULL)
1080 GOTO(out, rc = -ENOMEM);
1083 /* populate pages with key/record pairs */
1084 rc = dt_index_read(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, rep_ii, rdpg);
1088 LASSERTF(rc <= rdpg->rp_count, "dt_index_read() returned more than "
1089 "asked %d > %d\n", rc, rdpg->rp_count);
1091 /* send pages to client */
1092 rc = tgt_sendpage(tsi, rdpg, rc);
1097 if (rdpg->rp_pages) {
1098 for (i = 0; i < rdpg->rp_npages; i++)
1099 if (rdpg->rp_pages[i])
1100 __free_page(rdpg->rp_pages[i]);
1101 OBD_FREE(rdpg->rp_pages,
1102 rdpg->rp_npages * sizeof(rdpg->rp_pages[0]));
1107 struct tgt_handler tgt_obd_handlers[] = {
1108 TGT_OBD_HDL (0, OBD_PING, tgt_obd_ping),
1109 TGT_OBD_HDL_VAR(0, OBD_LOG_CANCEL, tgt_obd_log_cancel),
1110 TGT_OBD_HDL_VAR(0, OBD_QC_CALLBACK, tgt_obd_qc_callback),
1111 TGT_OBD_HDL (0, OBD_IDX_READ, tgt_obd_idx_read)
1113 EXPORT_SYMBOL(tgt_obd_handlers);
1115 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
1116 struct dt_object *obj, __u64 start, __u64 end)
1122 /* if no objid is specified, it means "sync whole filesystem" */
1124 rc = dt_sync(env, tgt->lut_bottom);
1125 } else if (dt_version_get(env, obj) >
1126 tgt->lut_obd->obd_last_committed) {
1127 rc = dt_object_sync(env, obj, start, end);
1132 EXPORT_SYMBOL(tgt_sync);
1134 * Unified target DLM handlers.
1137 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1139 static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1140 void *data, int flag)
1143 struct lu_target *tgt;
1144 struct dt_object *obj;
1150 tgt = class_exp2tgt(lock->l_export);
1152 if (flag == LDLM_CB_CANCELING &&
1153 (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
1154 (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1155 (tgt->lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1156 lock->l_flags & LDLM_FL_CBPENDING))) {
1158 __u64 end = OBD_OBJECT_EOF;
1160 rc = lu_env_init(&env, LCT_DT_THREAD);
1161 if (unlikely(rc != 0))
1164 ost_fid_from_resid(&fid, &lock->l_resource->lr_name,
1165 tgt->lut_lsd.lsd_osd_index);
1166 obj = dt_locate(&env, tgt->lut_bottom, &fid);
1168 GOTO(err_env, rc = PTR_ERR(obj));
1170 if (!dt_object_exists(obj))
1171 GOTO(err_put, rc = -ENOENT);
1173 if (lock->l_resource->lr_type == LDLM_EXTENT) {
1174 start = lock->l_policy_data.l_extent.start;
1175 end = lock->l_policy_data.l_extent.end;
1178 rc = tgt_sync(&env, tgt, obj, start, end);
1180 CERROR("%s: syncing "DFID" ("LPU64"-"LPU64") on lock "
1181 "cancel: rc = %d\n",
1182 tgt_name(tgt), PFID(&fid),
1183 lock->l_policy_data.l_extent.start,
1184 lock->l_policy_data.l_extent.end, rc);
1187 lu_object_put(&env, &obj->do_lu);
1192 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1196 static struct ldlm_callback_suite tgt_dlm_cbs = {
1197 .lcs_completion = ldlm_server_completion_ast,
1198 .lcs_blocking = tgt_blocking_ast,
1199 .lcs_glimpse = ldlm_server_glimpse_ast
1202 int tgt_enqueue(struct tgt_session_info *tsi)
1204 struct ptlrpc_request *req = tgt_ses_req(tsi);
1209 * tsi->tsi_dlm_req was already swapped and (if necessary) converted,
1210 * tsi->tsi_dlm_cbs was set by the *_req_handle() function.
1212 LASSERT(tsi->tsi_dlm_req != NULL);
1213 rc = ldlm_handle_enqueue0(tsi->tsi_exp->exp_obd->obd_namespace, req,
1214 tsi->tsi_dlm_req, &tgt_dlm_cbs);
1216 RETURN(err_serious(rc));
1218 switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
1220 tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
1223 tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
1226 tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
1229 tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
1232 RETURN(req->rq_status);
1234 EXPORT_SYMBOL(tgt_enqueue);
1236 int tgt_convert(struct tgt_session_info *tsi)
1238 struct ptlrpc_request *req = tgt_ses_req(tsi);
1242 LASSERT(tsi->tsi_dlm_req);
1243 rc = ldlm_handle_convert0(req, tsi->tsi_dlm_req);
1245 RETURN(err_serious(rc));
1247 RETURN(req->rq_status);
1250 int tgt_bl_callback(struct tgt_session_info *tsi)
1252 return err_serious(-EOPNOTSUPP);
1255 int tgt_cp_callback(struct tgt_session_info *tsi)
1257 return err_serious(-EOPNOTSUPP);
1260 /* generic LDLM target handler */
1261 struct tgt_handler tgt_dlm_handlers[] = {
1262 TGT_DLM_HDL (HABEO_CLAVIS, LDLM_ENQUEUE, tgt_enqueue),
1263 TGT_DLM_HDL_VAR(HABEO_CLAVIS, LDLM_CONVERT, tgt_convert),
1264 TGT_DLM_HDL_VAR(0, LDLM_BL_CALLBACK, tgt_bl_callback),
1265 TGT_DLM_HDL_VAR(0, LDLM_CP_CALLBACK, tgt_cp_callback)
1267 EXPORT_SYMBOL(tgt_dlm_handlers);
1270 * Unified target LLOG handlers.
1272 int tgt_llog_open(struct tgt_session_info *tsi)
1278 rc = llog_origin_handle_open(tgt_ses_req(tsi));
1282 EXPORT_SYMBOL(tgt_llog_open);
1284 int tgt_llog_close(struct tgt_session_info *tsi)
1290 rc = llog_origin_handle_close(tgt_ses_req(tsi));
1294 EXPORT_SYMBOL(tgt_llog_close);
1297 int tgt_llog_destroy(struct tgt_session_info *tsi)
1303 rc = llog_origin_handle_destroy(tgt_ses_req(tsi));
1308 int tgt_llog_read_header(struct tgt_session_info *tsi)
1314 rc = llog_origin_handle_read_header(tgt_ses_req(tsi));
1318 EXPORT_SYMBOL(tgt_llog_read_header);
1320 int tgt_llog_next_block(struct tgt_session_info *tsi)
1326 rc = llog_origin_handle_next_block(tgt_ses_req(tsi));
1330 EXPORT_SYMBOL(tgt_llog_next_block);
1332 int tgt_llog_prev_block(struct tgt_session_info *tsi)
1338 rc = llog_origin_handle_prev_block(tgt_ses_req(tsi));
1342 EXPORT_SYMBOL(tgt_llog_prev_block);
1344 /* generic llog target handler */
1345 struct tgt_handler tgt_llog_handlers[] = {
1346 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_CREATE, tgt_llog_open),
1347 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_NEXT_BLOCK, tgt_llog_next_block),
1348 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_READ_HEADER, tgt_llog_read_header),
1349 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_PREV_BLOCK, tgt_llog_prev_block),
1350 TGT_LLOG_HDL (0, LLOG_ORIGIN_HANDLE_DESTROY, tgt_llog_destroy),
1351 TGT_LLOG_HDL_VAR(0, LLOG_ORIGIN_HANDLE_CLOSE, tgt_llog_close),
1353 EXPORT_SYMBOL(tgt_llog_handlers);
1356 * sec context handlers
1358 /* XXX: Implement based on mdt_sec_ctx_handle()? */
1359 static int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
1364 struct tgt_handler tgt_sec_ctx_handlers[] = {
1365 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT, tgt_sec_ctx_handle),
1366 TGT_SEC_HDL_VAR(0, SEC_CTX_INIT_CONT, tgt_sec_ctx_handle),
1367 TGT_SEC_HDL_VAR(0, SEC_CTX_FINI, tgt_sec_ctx_handle),
1369 EXPORT_SYMBOL(tgt_sec_ctx_handlers);
1371 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
1372 struct dt_device *key,
1373 struct lfsck_request *lr,
1374 struct thandle *th) = NULL;
1376 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
1378 struct lfsck_request *,
1381 tgt_lfsck_in_notify = notify;
1383 EXPORT_SYMBOL(tgt_register_lfsck_in_notify);
1385 static int (*tgt_lfsck_query)(const struct lu_env *env,
1386 struct dt_device *key,
1387 struct lfsck_request *lr) = NULL;
1389 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
1391 struct lfsck_request *))
1393 tgt_lfsck_query = query;
1395 EXPORT_SYMBOL(tgt_register_lfsck_query);
1397 /* LFSCK request handlers */
1398 static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
1400 const struct lu_env *env = tsi->tsi_env;
1401 struct dt_device *key = tsi->tsi_tgt->lut_bottom;
1402 struct lfsck_request *lr;
1406 lr = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1410 rc = tgt_lfsck_in_notify(env, key, lr, NULL);
1415 static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
1417 struct lfsck_request *request;
1418 struct lfsck_reply *reply;
1422 request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
1423 if (request == NULL)
1426 reply = req_capsule_server_get(tsi->tsi_pill, &RMF_LFSCK_REPLY);
1430 rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
1431 reply->lr_status = rc;
1433 RETURN(rc < 0 ? rc : 0);
1436 struct tgt_handler tgt_lfsck_handlers[] = {
1437 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_NOTIFY, tgt_handle_lfsck_notify),
1438 TGT_LFSCK_HDL(HABEO_REFERO, LFSCK_QUERY, tgt_handle_lfsck_query),
1440 EXPORT_SYMBOL(tgt_lfsck_handlers);
1443 * initialize per-thread page pool (bug 5137).
1445 int tgt_io_thread_init(struct ptlrpc_thread *thread)
1447 struct tgt_thread_big_cache *tbc;
1451 LASSERT(thread != NULL);
1452 LASSERT(thread->t_data == NULL);
1454 OBD_ALLOC_LARGE(tbc, sizeof(*tbc));
1457 thread->t_data = tbc;
1460 EXPORT_SYMBOL(tgt_io_thread_init);
1463 * free per-thread pool created by tgt_thread_init().
1465 void tgt_io_thread_done(struct ptlrpc_thread *thread)
1467 struct tgt_thread_big_cache *tbc;
1471 LASSERT(thread != NULL);
1474 * be prepared to handle partially-initialized pools (because this is
1475 * called from ost_io_thread_init() for cleanup.
1477 tbc = thread->t_data;
1479 OBD_FREE_LARGE(tbc, sizeof(*tbc));
1480 thread->t_data = NULL;
1484 EXPORT_SYMBOL(tgt_io_thread_done);
1486 * Helper function for getting server side [start, start+count] DLM lock
1487 * if asked by client.
1489 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1490 __u64 start, __u64 end, struct lustre_handle *lh,
1491 int mode, __u64 *flags)
1493 ldlm_policy_data_t policy;
1498 LASSERT(lh != NULL);
1499 LASSERT(ns != NULL);
1500 LASSERT(!lustre_handle_is_used(lh));
1502 policy.l_extent.gid = 0;
1503 policy.l_extent.start = start & PAGE_MASK;
1506 * If ->o_blocks is EOF it means "lock till the end of the file".
1507 * Otherwise, it's size of an extent or hole being punched (in bytes).
1509 if (end == OBD_OBJECT_EOF || end < start)
1510 policy.l_extent.end = OBD_OBJECT_EOF;
1512 policy.l_extent.end = end | ~PAGE_MASK;
1514 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
1515 flags, ldlm_blocking_ast,
1516 ldlm_completion_ast, ldlm_glimpse_ast,
1517 NULL, 0, LVB_T_NONE, NULL, lh);
1518 RETURN(rc == ELDLM_OK ? 0 : -EIO);
1520 EXPORT_SYMBOL(tgt_extent_lock);
1522 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode)
1524 LASSERT(lustre_handle_is_used(lh));
1525 ldlm_lock_decref(lh, mode);
1527 EXPORT_SYMBOL(tgt_extent_unlock);
1529 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
1530 struct obd_ioobj *obj, struct niobuf_remote *nb,
1531 struct lustre_handle *lh, int mode)
1534 int nrbufs = obj->ioo_bufcnt;
1539 LASSERT(mode == LCK_PR || mode == LCK_PW);
1540 LASSERT(!lustre_handle_is_used(lh));
1542 if (nrbufs == 0 || !(nb[0].rnb_flags & OBD_BRW_SRVLOCK))
1545 for (i = 1; i < nrbufs; i++)
1546 if (!(nb[i].rnb_flags & OBD_BRW_SRVLOCK))
1549 RETURN(tgt_extent_lock(ns, res_id, nb[0].rnb_offset,
1550 nb[nrbufs - 1].rnb_offset +
1551 nb[nrbufs - 1].rnb_len - 1,
1555 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
1556 struct lustre_handle *lh, int mode)
1560 LASSERT(mode == LCK_PR || mode == LCK_PW);
1561 LASSERT((obj->ioo_bufcnt > 0 &&
1562 (niob[0].rnb_flags & OBD_BRW_SRVLOCK)) ==
1563 lustre_handle_is_used(lh));
1565 if (lustre_handle_is_used(lh))
1566 tgt_extent_unlock(lh, mode);
1570 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
1571 struct ptlrpc_bulk_desc *desc, int opc,
1572 cksum_type_t cksum_type)
1574 struct cfs_crypto_hash_desc *hdesc;
1575 unsigned int bufsize;
1577 unsigned char cfs_alg = cksum_obd2cfs(cksum_type);
1580 hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
1581 if (IS_ERR(hdesc)) {
1582 CERROR("%s: unable to initialize checksum hash %s\n",
1583 tgt_name(tgt), cfs_crypto_hash_name(cfs_alg));
1584 return PTR_ERR(hdesc);
1587 CDEBUG(D_INFO, "Checksum for algo %s\n", cfs_crypto_hash_name(cfs_alg));
1588 for (i = 0; i < desc->bd_iov_count; i++) {
1589 /* corrupt the data before we compute the checksum, to
1590 * simulate a client->OST data error */
1591 if (i == 0 && opc == OST_WRITE &&
1592 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
1593 int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1594 int len = desc->bd_iov[i].kiov_len;
1595 struct page *np = tgt_page_to_corrupt;
1596 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1599 char *ptr2 = kmap(np) + off;
1601 memcpy(ptr2, ptr, len);
1602 memcpy(ptr2, "bad3", min(4, len));
1604 desc->bd_iov[i].kiov_page = np;
1606 CERROR("%s: can't alloc page for corruption\n",
1610 cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
1611 desc->bd_iov[i].kiov_offset & ~PAGE_MASK,
1612 desc->bd_iov[i].kiov_len);
1614 /* corrupt the data after we compute the checksum, to
1615 * simulate an OST->client data error */
1616 if (i == 0 && opc == OST_READ &&
1617 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
1618 int off = desc->bd_iov[i].kiov_offset & ~PAGE_MASK;
1619 int len = desc->bd_iov[i].kiov_len;
1620 struct page *np = tgt_page_to_corrupt;
1621 char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
1624 char *ptr2 = kmap(np) + off;
1626 memcpy(ptr2, ptr, len);
1627 memcpy(ptr2, "bad4", min(4, len));
1629 desc->bd_iov[i].kiov_page = np;
1631 CERROR("%s: can't alloc page for corruption\n",
1637 bufsize = sizeof(cksum);
1638 err = cfs_crypto_hash_final(hdesc, (unsigned char *)&cksum, &bufsize);
1643 int tgt_brw_read(struct tgt_session_info *tsi)
1645 struct ptlrpc_request *req = tgt_ses_req(tsi);
1646 struct ptlrpc_bulk_desc *desc = NULL;
1647 struct obd_export *exp = tsi->tsi_exp;
1648 struct niobuf_remote *remote_nb;
1649 struct niobuf_local *local_nb;
1650 struct obd_ioobj *ioo;
1651 struct ost_body *body, *repbody;
1652 struct l_wait_info lwi;
1653 struct lustre_handle lockh = { 0 };
1654 int npages, nob = 0, rc, i, no_reply = 0;
1655 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1659 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1660 CERROR("%s: deny read request from %s to portal %u\n",
1661 tgt_name(tsi->tsi_tgt),
1662 obd_export_nid2str(req->rq_export),
1663 ptlrpc_req2svc(req)->srv_req_portal);
1667 req->rq_bulk_read = 1;
1669 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
1672 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1673 cfs_fail_val : (obd_timeout + 1) / 4);
1675 /* Check if there is eviction in progress, and if so, wait for it to
1677 if (unlikely(atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
1678 /* We do not care how long it takes */
1679 lwi = LWI_INTR(NULL, NULL);
1680 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
1681 !atomic_read(&exp->exp_obd->obd_evict_inprogress),
1685 /* There must be big cache in current thread to process this request
1686 * if it is NULL then something went wrong and it wasn't allocated,
1687 * report -ENOMEM in that case */
1691 body = tsi->tsi_ost_body;
1692 LASSERT(body != NULL);
1694 ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
1695 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1697 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1698 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1700 local_nb = tbc->local;
1702 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1703 remote_nb, &lockh, LCK_PR);
1708 * If getting the lock took more time than
1709 * client was willing to wait, drop it. b=11330
1711 if (cfs_time_current_sec() > req->rq_deadline ||
1712 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1714 CERROR("Dropping timed-out read from %s because locking"
1715 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1716 libcfs_id2str(req->rq_peer), POSTID(&ioo->ioo_oid),
1717 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1718 req->rq_deadline - req->rq_arrival_time.tv_sec);
1719 GOTO(out_lock, rc = -ETIMEDOUT);
1722 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1723 repbody->oa = body->oa;
1725 npages = PTLRPC_MAX_BRW_PAGES;
1726 rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
1727 ioo, remote_nb, &npages, local_nb, NULL);
1731 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1732 BULK_PUT_SOURCE, OST_BULK_PORTAL);
1734 GOTO(out_commitrw, rc = -ENOMEM);
1737 for (i = 0; i < npages; i++) {
1738 int page_rc = local_nb[i].lnb_rc;
1746 if (page_rc != 0) { /* some data! */
1747 LASSERT(local_nb[i].lnb_page != NULL);
1748 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
1749 local_nb[i].lnb_page_offset,
1753 if (page_rc != local_nb[i].lnb_len) { /* short read */
1754 /* All subsequent pages should be 0 */
1755 while (++i < npages)
1756 LASSERT(local_nb[i].lnb_rc == 0);
1761 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
1762 cksum_type_t cksum_type =
1763 cksum_type_unpack(body->oa.o_valid & OBD_MD_FLFLAGS ?
1764 body->oa.o_flags : 0);
1765 repbody->oa.o_flags = cksum_type_pack(cksum_type);
1766 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
1767 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
1768 OST_READ, cksum_type);
1769 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
1770 repbody->oa.o_cksum);
1772 repbody->oa.o_valid = 0;
1774 /* We're finishing using body->oa as an input variable */
1776 /* Check if client was evicted while we were doing i/o before touching
1778 if (likely(rc == 0 &&
1779 !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1780 rc = target_bulk_io(exp, desc, &lwi);
1785 /* Must commit after prep above in all cases */
1786 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
1787 &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
1790 tgt_drop_id(exp, &repbody->oa);
1792 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
1794 if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
1795 ptlrpc_free_bulk_nopin(desc);
1800 ptlrpc_lprocfs_brw(req, nob);
1801 } else if (no_reply) {
1802 req->rq_no_reply = 1;
1803 /* reply out callback would free */
1804 ptlrpc_req_drop_rs(req);
1805 LCONSOLE_WARN("%s: Bulk IO read error with %s (at %s), "
1806 "client will retry: rc %d\n",
1807 exp->exp_obd->obd_name,
1808 obd_uuid2str(&exp->exp_client_uuid),
1809 obd_export_nid2str(exp), rc);
1811 /* send a bulk after reply to simulate a network delay or reordering
1813 if (unlikely(CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))) {
1814 wait_queue_head_t waitq;
1815 struct l_wait_info lwi1;
1817 CDEBUG(D_INFO, "reorder BULK\n");
1818 init_waitqueue_head(&waitq);
1820 lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
1821 l_wait_event(waitq, 0, &lwi1);
1822 target_bulk_io(exp, desc, &lwi);
1823 ptlrpc_free_bulk_nopin(desc);
1828 EXPORT_SYMBOL(tgt_brw_read);
1830 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
1831 struct ptlrpc_bulk_desc *desc,
1832 struct niobuf_local *local_nb, int npages,
1833 u32 client_cksum, u32 server_cksum,
1836 struct obd_export *exp = req->rq_export;
1837 struct ost_body *body;
1841 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1842 LASSERT(body != NULL);
1844 if (req->rq_peer.nid != desc->bd_sender) {
1846 router = libcfs_nid2str(desc->bd_sender);
1850 CDEBUG_LIMIT(D_INFO, "client csum %x, server csum %x\n",
1851 client_cksum, server_cksum);
1855 LCONSOLE_ERROR_MSG(0x168, "BAD WRITE CHECKSUM: %s from %s%s%s inode "
1856 DFID" object "DOSTID" extent ["LPU64"-"LPU64
1857 "]: client csum %x, server csum %x\n",
1858 exp->exp_obd->obd_name, libcfs_id2str(req->rq_peer),
1860 body->oa.o_valid & OBD_MD_FLFID ?
1861 body->oa.o_parent_seq : (__u64)0,
1862 body->oa.o_valid & OBD_MD_FLFID ?
1863 body->oa.o_parent_oid : 0,
1864 body->oa.o_valid & OBD_MD_FLFID ?
1865 body->oa.o_parent_ver : 0,
1866 POSTID(&body->oa.o_oi),
1867 local_nb[0].lnb_file_offset,
1868 local_nb[npages-1].lnb_file_offset +
1869 local_nb[npages - 1].lnb_len - 1,
1870 client_cksum, server_cksum);
1873 int tgt_brw_write(struct tgt_session_info *tsi)
1875 struct ptlrpc_request *req = tgt_ses_req(tsi);
1876 struct ptlrpc_bulk_desc *desc = NULL;
1877 struct obd_export *exp = req->rq_export;
1878 struct niobuf_remote *remote_nb;
1879 struct niobuf_local *local_nb;
1880 struct obd_ioobj *ioo;
1881 struct ost_body *body, *repbody;
1882 struct l_wait_info lwi;
1883 struct lustre_handle lockh = {0};
1885 int objcount, niocount, npages;
1887 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
1888 bool no_reply = false, mmap;
1889 struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
1893 if (ptlrpc_req2svc(req)->srv_req_portal != OST_IO_PORTAL) {
1894 CERROR("%s: deny write request from %s to portal %u\n",
1895 tgt_name(tsi->tsi_tgt),
1896 obd_export_nid2str(req->rq_export),
1897 ptlrpc_req2svc(req)->srv_req_portal);
1898 RETURN(err_serious(-EPROTO));
1901 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
1902 RETURN(err_serious(-ENOSPC));
1903 if (OBD_FAIL_TIMEOUT(OBD_FAIL_OST_EROFS, 1))
1904 RETURN(err_serious(-EROFS));
1906 req->rq_bulk_write = 1;
1908 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
1909 RETURN(err_serious(-EIO));
1910 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
1911 RETURN(err_serious(-EFAULT));
1913 /* pause before transaction has been started */
1914 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, cfs_fail_val > 0 ?
1915 cfs_fail_val : (obd_timeout + 1) / 4);
1917 /* There must be big cache in current thread to process this request
1918 * if it is NULL then something went wrong and it wasn't allocated,
1919 * report -ENOMEM in that case */
1923 body = tsi->tsi_ost_body;
1924 LASSERT(body != NULL);
1926 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1927 LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
1929 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
1930 RCL_CLIENT) / sizeof(*ioo);
1932 for (niocount = i = 0; i < objcount; i++)
1933 niocount += ioo[i].ioo_bufcnt;
1935 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1936 LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
1937 if (niocount != req_capsule_get_size(&req->rq_pill,
1938 &RMF_NIOBUF_REMOTE, RCL_CLIENT) /
1940 RETURN(err_serious(-EPROTO));
1942 if ((remote_nb[0].rnb_flags & OBD_BRW_MEMALLOC) &&
1943 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
1944 memory_pressure_set();
1946 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
1947 niocount * sizeof(*rcs));
1948 rc = req_capsule_server_pack(&req->rq_pill);
1950 GOTO(out, rc = err_serious(rc));
1952 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
1953 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
1955 local_nb = tbc->local;
1957 rc = tgt_brw_lock(exp->exp_obd->obd_namespace, &tsi->tsi_resid, ioo,
1958 remote_nb, &lockh, LCK_PW);
1963 * If getting the lock took more time than
1964 * client was willing to wait, drop it. b=11330
1966 if (cfs_time_current_sec() > req->rq_deadline ||
1967 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
1969 CERROR("%s: Dropping timed-out write from %s because locking "
1970 "object "DOSTID" took %ld seconds (limit was %ld).\n",
1971 tgt_name(tsi->tsi_tgt), libcfs_id2str(req->rq_peer),
1972 POSTID(&ioo->ioo_oid),
1973 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
1974 req->rq_deadline - req->rq_arrival_time.tv_sec);
1975 GOTO(out_lock, rc = -ETIMEDOUT);
1978 /* Because we already sync grant info with client when reconnect,
1979 * grant info will be cleared for resent req, then fed_grant and
1980 * total_grant will not be modified in following preprw_write */
1981 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
1982 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
1983 body->oa.o_valid &= ~OBD_MD_FLGRANT;
1986 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
1987 if (repbody == NULL)
1988 GOTO(out_lock, rc = -ENOMEM);
1989 repbody->oa = body->oa;
1991 npages = PTLRPC_MAX_BRW_PAGES;
1992 rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
1993 objcount, ioo, remote_nb, &npages, local_nb, NULL);
1997 desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
1998 BULK_GET_SINK, OST_BULK_PORTAL);
2000 GOTO(skip_transfer, rc = -ENOMEM);
2002 /* NB Having prepped, we must commit... */
2003 for (i = 0; i < npages; i++)
2004 ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
2005 local_nb[i].lnb_page_offset,
2006 local_nb[i].lnb_len);
2008 rc = sptlrpc_svc_prep_bulk(req, desc);
2010 GOTO(skip_transfer, rc);
2012 rc = target_bulk_io(exp, desc, &lwi);
2016 if (body->oa.o_valid & OBD_MD_FLCKSUM && rc == 0) {
2017 static int cksum_counter;
2019 if (body->oa.o_valid & OBD_MD_FLFLAGS)
2020 cksum_type = cksum_type_unpack(body->oa.o_flags);
2022 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
2023 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
2024 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
2025 repbody->oa.o_cksum = tgt_checksum_bulk(tsi->tsi_tgt, desc,
2026 OST_WRITE, cksum_type);
2029 if (unlikely(body->oa.o_cksum != repbody->oa.o_cksum)) {
2030 mmap = (body->oa.o_valid & OBD_MD_FLFLAGS &&
2031 body->oa.o_flags & OBD_FL_MMAP);
2033 tgt_warn_on_cksum(req, desc, local_nb, npages,
2035 repbody->oa.o_cksum, mmap);
2037 } else if ((cksum_counter & (-cksum_counter)) ==
2039 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
2040 cksum_counter, libcfs_id2str(req->rq_peer),
2041 repbody->oa.o_cksum);
2045 /* Must commit after prep above in all cases */
2046 rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
2047 objcount, ioo, remote_nb, npages, local_nb, NULL,
2049 if (rc == -ENOTCONN)
2050 /* quota acquire process has been given up because
2051 * either the client has been evicted or the client
2052 * has timed out the request already */
2056 * Disable sending mtime back to the client. If the client locked the
2057 * whole object, then it has already updated the mtime on its side,
2058 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
2060 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
2065 /* set per-requested niobuf return codes */
2066 for (i = j = 0; i < niocount; i++) {
2067 int len = remote_nb[i].rnb_len;
2072 LASSERT(j < npages);
2073 if (local_nb[j].lnb_rc < 0)
2074 rcs[i] = local_nb[j].lnb_rc;
2075 len -= local_nb[j].lnb_len;
2080 LASSERT(j == npages);
2081 ptlrpc_lprocfs_brw(req, nob);
2083 tgt_drop_id(exp, &repbody->oa);
2086 tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
2088 ptlrpc_free_bulk_nopin(desc);
2091 req->rq_no_reply = 1;
2092 /* reply out callback would free */
2093 ptlrpc_req_drop_rs(req);
2094 LCONSOLE_WARN("%s: Bulk IO write error with %s (at %s), "
2095 "client will retry: rc %d\n",
2096 exp->exp_obd->obd_name,
2097 obd_uuid2str(&exp->exp_client_uuid),
2098 obd_export_nid2str(exp), rc);
2100 memory_pressure_clr();
2103 EXPORT_SYMBOL(tgt_brw_write);