1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011 Whamcloud, Inc.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/ost/ost_handler.c
41 * Author: Peter J. Braam <braam@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
46 # define EXPORT_SYMTAB
48 #define DEBUG_SUBSYSTEM S_OST
50 #include <linux/module.h>
51 #include <obd_cksum.h>
53 #include <lustre_net.h>
54 #include <lustre_dlm.h>
55 #include <lustre_export.h>
56 #include <lustre_debug.h>
57 #include <linux/init.h>
58 #include <lprocfs_status.h>
59 #include <libcfs/list.h>
60 #include <lustre_quota.h>
61 #include "ost_internal.h"
63 static int oss_num_threads;
64 CFS_MODULE_PARM(oss_num_threads, "i", int, 0444,
65 "number of OSS service threads to start");
67 static int ost_num_threads;
68 CFS_MODULE_PARM(ost_num_threads, "i", int, 0444,
69 "number of OST service threads to start (deprecated)");
71 static int oss_num_create_threads;
72 CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
73 "number of OSS create threads to start");
76 * Do not return server-side uid/gid to remote client
78 static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
80 if (exp_connect_rmtclient(exp)) {
83 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
88 * Validate oa from client.
89 * If the request comes from 2.0 clients, currently only RSVD seq and IDIF
91 * a. for single MDS seq = FID_SEQ_OST_MDT0,
92 * b. for CMD, seq = FID_SEQ_OST_MDT0, FID_SEQ_OST_MDT1 - FID_SEQ_OST_MAX
94 static int ost_validate_obdo(struct obd_export *exp, struct obdo *oa,
95 struct obd_ioobj *ioobj)
97 if (oa != NULL && !(oa->o_valid & OBD_MD_FLGROUP)) {
98 oa->o_seq = FID_SEQ_OST_MDT0;
100 ioobj->ioo_seq = FID_SEQ_OST_MDT0;
101 /* remove fid_seq_is_rsvd() after FID-on-OST allows SEQ > 9 */
102 } else if (oa == NULL ||
103 !(fid_seq_is_rsvd(oa->o_seq) || fid_seq_is_idif(oa->o_seq))) {
104 CERROR("%s: client %s sent invalid object "POSTID"\n",
105 exp->exp_obd->obd_name, obd_export_nid2str(exp),
106 oa ? oa->o_id : -1, oa ? oa->o_seq : -1);
109 obdo_from_ostid(oa, &oa->o_oi);
111 ioobj_from_obdo(ioobj, oa);
115 void oti_to_request(struct obd_trans_info *oti, struct ptlrpc_request *req)
117 struct oti_req_ack_lock *ack_lock;
123 if (req->rq_repmsg) {
124 __u64 versions[PTLRPC_NUM_VERSIONS] = { 0 };
125 lustre_msg_set_transno(req->rq_repmsg, oti->oti_transno);
126 versions[0] = oti->oti_pre_version;
127 lustre_msg_set_versions(req->rq_repmsg, versions);
129 req->rq_transno = oti->oti_transno;
131 /* XXX 4 == entries in oti_ack_locks??? */
132 for (ack_lock = oti->oti_ack_locks, i = 0; i < 4; i++, ack_lock++) {
135 /* XXX not even calling target_send_reply in some cases... */
136 ptlrpc_save_lock (req, &ack_lock->lock, ack_lock->mode, 0);
140 static int ost_destroy(struct obd_export *exp, struct ptlrpc_request *req,
141 struct obd_trans_info *oti)
143 struct ost_body *body, *repbody;
144 struct lustre_capa *capa = NULL;
148 /* Get the request body */
149 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
153 if (body->oa.o_id == 0)
156 rc = ost_validate_obdo(exp, &body->oa, NULL);
160 /* If there's a DLM request, cancel the locks mentioned in it*/
161 if (req_capsule_field_present(&req->rq_pill, &RMF_DLM_REQ, RCL_CLIENT)) {
162 struct ldlm_request *dlm;
164 dlm = req_capsule_client_get(&req->rq_pill, &RMF_DLM_REQ);
167 ldlm_request_cancel(req, dlm, 0);
170 /* If there's a capability, get it */
171 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
172 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
174 CERROR("Missing capability for OST DESTROY");
179 /* Prepare the reply */
180 rc = req_capsule_server_pack(&req->rq_pill);
184 /* Get the log cancellation cookie */
185 if (body->oa.o_valid & OBD_MD_FLCOOKIE)
186 oti->oti_logcookies = &body->oa.o_lcookie;
188 /* Finish the reply */
189 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
190 memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
192 /* Do the destroy and set the reply status accordingly */
193 req->rq_status = obd_destroy(exp, &repbody->oa, NULL, oti, NULL, capa);
198 * Helper function for getting server side [start, start+count] DLM lock
199 * if asked by client.
201 static int ost_lock_get(struct obd_export *exp, struct obdo *oa,
202 __u64 start, __u64 count, struct lustre_handle *lh,
205 struct ldlm_res_id res_id;
206 ldlm_policy_data_t policy;
207 __u64 end = start + count;
211 LASSERT(!lustre_handle_is_used(lh));
212 /* o_id and o_gr are used for localizing resource, if client miss to set
213 * them, do not trigger ASSERTION. */
214 if (unlikely((oa->o_valid & (OBD_MD_FLID | OBD_MD_FLGROUP)) !=
215 (OBD_MD_FLID | OBD_MD_FLGROUP)))
218 if (!(oa->o_valid & OBD_MD_FLFLAGS) ||
219 !(oa->o_flags & OBD_FL_SRVLOCK))
222 osc_build_res_name(oa->o_id, oa->o_seq, &res_id);
223 CDEBUG(D_INODE, "OST-side extent lock.\n");
225 policy.l_extent.start = start & CFS_PAGE_MASK;
227 /* If ->o_blocks is EOF it means "lock till the end of the
228 * file". Otherwise, it's size of a hole being punched (in bytes) */
229 if (count == OBD_OBJECT_EOF || end < start)
230 policy.l_extent.end = OBD_OBJECT_EOF;
232 policy.l_extent.end = end | ~CFS_PAGE_MASK;
234 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
235 LDLM_EXTENT, &policy, mode, &flags,
236 ldlm_blocking_ast, ldlm_completion_ast,
237 ldlm_glimpse_ast, NULL, 0, NULL, lh));
240 /* Helper function: release lock, if any. */
241 static void ost_lock_put(struct obd_export *exp,
242 struct lustre_handle *lh, int mode)
245 if (lustre_handle_is_used(lh))
246 ldlm_lock_decref(lh, mode);
250 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
252 struct ost_body *body, *repbody;
253 struct obd_info *oinfo;
254 struct lustre_handle lh = { 0 };
255 struct lustre_capa *capa = NULL;
259 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
263 rc = ost_validate_obdo(exp, &body->oa, NULL);
267 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
268 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
270 CERROR("Missing capability for OST GETATTR");
275 rc = req_capsule_server_pack(&req->rq_pill);
279 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
280 repbody->oa = body->oa;
282 rc = ost_lock_get(exp, &repbody->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
286 OBD_ALLOC_PTR(oinfo);
288 GOTO(unlock, rc = -ENOMEM);
289 oinfo->oi_oa = &repbody->oa;
290 oinfo->oi_capa = capa;
292 req->rq_status = obd_getattr(exp, oinfo);
296 ost_drop_id(exp, &repbody->oa);
299 ost_lock_put(exp, &lh, LCK_PR);
303 static int ost_statfs(struct ptlrpc_request *req)
305 struct obd_statfs *osfs;
309 rc = req_capsule_server_pack(&req->rq_pill);
313 osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
315 req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
316 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
318 if (req->rq_status != 0)
319 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
324 static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
325 struct obd_trans_info *oti)
327 struct ost_body *body, *repbody;
331 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
335 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
339 rc = req_capsule_server_pack(&req->rq_pill);
343 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
344 repbody->oa = body->oa;
345 oti->oti_logcookies = &body->oa.o_lcookie;
347 req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
348 //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
352 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
353 struct obd_trans_info *oti)
355 struct ost_body *body, *repbody;
357 struct lustre_handle lh = {0,};
360 /* check that we do support OBD_CONNECT_TRUNCLOCK. */
361 CLASSERT(OST_CONNECT_SUPPORTED & OBD_CONNECT_TRUNCLOCK);
363 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
367 rc = ost_validate_obdo(exp, &body->oa, NULL);
371 if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
372 (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
375 rc = req_capsule_server_pack(&req->rq_pill);
379 /* standard truncate optimization: if file body is completely
380 * destroyed, don't send data back to the server. */
381 if (body->oa.o_size == 0)
382 flags |= LDLM_AST_DISCARD_DATA;
384 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
385 repbody->oa = body->oa;
387 rc = ost_lock_get(exp, &repbody->oa, repbody->oa.o_size,
388 repbody->oa.o_blocks, &lh, LCK_PW, flags);
390 struct obd_info *oinfo;
391 struct lustre_capa *capa = NULL;
393 if (repbody->oa.o_valid & OBD_MD_FLFLAGS &&
394 repbody->oa.o_flags == OBD_FL_SRVLOCK)
396 * If OBD_FL_SRVLOCK is the only bit set in
397 * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
398 * through filter_setattr() to filter_iocontrol().
400 repbody->oa.o_valid &= ~OBD_MD_FLFLAGS;
402 if (repbody->oa.o_valid & OBD_MD_FLOSSCAPA) {
403 capa = req_capsule_client_get(&req->rq_pill,
406 CERROR("Missing capability for OST PUNCH");
407 GOTO(unlock, rc = -EFAULT);
411 OBD_ALLOC_PTR(oinfo);
413 GOTO(unlock, rc = -ENOMEM);
414 oinfo->oi_oa = &repbody->oa;
415 oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
416 oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
417 oinfo->oi_capa = capa;
418 oinfo->oi_flags = OBD_FL_PUNCH;
420 req->rq_status = obd_punch(exp, oinfo, oti, NULL);
423 ost_lock_put(exp, &lh, LCK_PW);
426 ost_drop_id(exp, &repbody->oa);
430 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
432 struct ost_body *body, *repbody;
433 struct obd_info *oinfo;
434 struct lustre_capa *capa = NULL;
438 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
442 rc = ost_validate_obdo(exp, &body->oa, NULL);
446 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
447 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
449 CERROR("Missing capability for OST SYNC");
454 rc = req_capsule_server_pack(&req->rq_pill);
458 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
459 repbody->oa = body->oa;
461 OBD_ALLOC_PTR(oinfo);
465 oinfo->oi_oa = &repbody->oa;
466 oinfo->oi_capa = capa;
467 req->rq_status = obd_sync(exp, oinfo, repbody->oa.o_size,
468 repbody->oa.o_blocks, NULL);
471 ost_drop_id(exp, &repbody->oa);
475 static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
476 struct obd_trans_info *oti)
478 struct ost_body *body, *repbody;
479 struct obd_info *oinfo;
480 struct lustre_capa *capa = NULL;
484 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
488 rc = ost_validate_obdo(req->rq_export, &body->oa, NULL);
492 rc = req_capsule_server_pack(&req->rq_pill);
496 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
497 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
499 CERROR("Missing capability for OST SETATTR");
504 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
505 repbody->oa = body->oa;
507 OBD_ALLOC_PTR(oinfo);
510 oinfo->oi_oa = &repbody->oa;
511 oinfo->oi_capa = capa;
513 req->rq_status = obd_setattr(exp, oinfo, oti);
517 ost_drop_id(exp, &repbody->oa);
521 static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
522 cksum_type_t cksum_type)
527 cksum = init_checksum(cksum_type);
528 for (i = 0; i < desc->bd_iov_count; i++) {
529 struct page *page = desc->bd_iov[i].kiov_page;
530 int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
531 char *ptr = kmap(page) + off;
532 int len = desc->bd_iov[i].kiov_len;
534 /* corrupt the data before we compute the checksum, to
535 * simulate a client->OST data error */
536 if (i == 0 && opc == OST_WRITE &&
537 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE))
538 memcpy(ptr, "bad3", min(4, len));
539 cksum = compute_checksum(cksum, ptr, len, cksum_type);
540 /* corrupt the data after we compute the checksum, to
541 * simulate an OST->client data error */
542 if (i == 0 && opc == OST_READ &&
543 OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
544 memcpy(ptr, "bad4", min(4, len));
545 /* nobody should use corrupted page again */
546 ClearPageUptodate(page);
551 return fini_checksum(cksum, cksum_type);
554 static int ost_brw_lock_get(int mode, struct obd_export *exp,
555 struct obd_ioobj *obj, struct niobuf_remote *nb,
556 struct lustre_handle *lh)
559 int nrbufs = obj->ioo_bufcnt;
560 struct ldlm_res_id res_id;
561 ldlm_policy_data_t policy;
565 osc_build_res_name(obj->ioo_id, obj->ioo_seq, &res_id);
566 LASSERT(mode == LCK_PR || mode == LCK_PW);
567 LASSERT(!lustre_handle_is_used(lh));
569 if (nrbufs == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
572 for (i = 1; i < nrbufs; i ++)
573 if ((nb[0].flags & OBD_BRW_SRVLOCK) !=
574 (nb[i].flags & OBD_BRW_SRVLOCK))
577 policy.l_extent.start = nb[0].offset & CFS_PAGE_MASK;
578 policy.l_extent.end = (nb[nrbufs - 1].offset +
579 nb[nrbufs - 1].len - 1) | ~CFS_PAGE_MASK;
581 RETURN(ldlm_cli_enqueue_local(exp->exp_obd->obd_namespace, &res_id,
582 LDLM_EXTENT, &policy, mode, &flags,
583 ldlm_blocking_ast, ldlm_completion_ast,
584 ldlm_glimpse_ast, NULL, 0, NULL, lh));
587 static void ost_brw_lock_put(int mode,
588 struct obd_ioobj *obj, struct niobuf_remote *niob,
589 struct lustre_handle *lh)
592 LASSERT(mode == LCK_PR || mode == LCK_PW);
593 LASSERT((obj->ioo_bufcnt > 0 && (niob[0].flags & OBD_BRW_SRVLOCK)) ==
594 lustre_handle_is_used(lh));
595 if (lustre_handle_is_used(lh))
596 ldlm_lock_decref(lh, mode);
600 /* Allocate thread local buffers if needed */
601 static struct ost_thread_local_cache *ost_tls_get(struct ptlrpc_request *r)
603 struct ost_thread_local_cache *tls =
604 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
606 /* In normal mode of operation an I/O request is serviced only
607 * by ll_ost_io threads each of them has own tls buffers allocated by
609 * During recovery, an I/O request may be queued until any of the ost
610 * service threads process it. Not necessary it should be one of
611 * ll_ost_io threads. In that case we dynamically allocating tls
612 * buffers for the request service time. */
613 if (unlikely(tls == NULL)) {
614 LASSERT(r->rq_export->exp_in_recovery);
618 r->rq_svc_thread->t_data = tls;
624 /* Free thread local buffers if they were allocated only for servicing
625 * this one request */
626 static void ost_tls_put(struct ptlrpc_request *r)
628 struct ost_thread_local_cache *tls =
629 (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
631 if (unlikely(tls->temporary)) {
633 r->rq_svc_thread->t_data = NULL;
637 static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
639 struct ptlrpc_bulk_desc *desc = NULL;
640 struct obd_export *exp = req->rq_export;
641 struct niobuf_remote *remote_nb;
642 struct niobuf_local *local_nb;
643 struct obd_ioobj *ioo;
644 struct ost_body *body, *repbody;
645 struct lustre_capa *capa = NULL;
646 struct l_wait_info lwi;
647 struct lustre_handle lockh = { 0 };
648 int niocount, npages, nob = 0, rc, i;
650 struct ost_thread_local_cache *tls;
653 req->rq_bulk_read = 1;
655 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_READ_BULK))
656 GOTO(out, rc = -EIO);
658 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
660 /* Check if there is eviction in progress, and if so, wait for it to
662 if (unlikely(cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress))) {
663 lwi = LWI_INTR(NULL, NULL); // We do not care how long it takes
664 rc = l_wait_event(exp->exp_obd->obd_evict_inprogress_waitq,
665 !cfs_atomic_read(&exp->exp_obd->obd_evict_inprogress),
669 GOTO(out, rc = -ENOTCONN);
671 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
672 * ost_rw_hpreq_check(). */
673 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
675 GOTO(out, rc = -EFAULT);
678 * A req_capsule_X_get_array(pill, field, ptr_to_element_count) function
679 * would be useful here and wherever we get &RMF_OBD_IOOBJ and
680 * &RMF_NIOBUF_REMOTE.
682 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
684 GOTO(out, rc = -EFAULT);
686 rc = ost_validate_obdo(exp, &body->oa, ioo);
690 niocount = ioo->ioo_bufcnt;
691 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
692 if (remote_nb == NULL)
693 GOTO(out, rc = -EFAULT);
695 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
696 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
698 CERROR("Missing capability for OST BRW READ");
699 GOTO(out, rc = -EFAULT);
703 rc = req_capsule_server_pack(&req->rq_pill);
707 tls = ost_tls_get(req);
709 GOTO(out_bulk, rc = -ENOMEM);
710 local_nb = tls->local;
712 rc = ost_brw_lock_get(LCK_PR, exp, ioo, remote_nb, &lockh);
717 * If getting the lock took more time than
718 * client was willing to wait, drop it. b=11330
720 if (cfs_time_current_sec() > req->rq_deadline ||
721 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
723 CERROR("Dropping timed-out read from %s because locking"
724 "object "LPX64" took %ld seconds (limit was %ld).\n",
725 libcfs_id2str(req->rq_peer), ioo->ioo_id,
726 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
727 req->rq_deadline - req->rq_arrival_time.tv_sec);
728 GOTO(out_lock, rc = -ETIMEDOUT);
731 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
732 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
734 npages = OST_THREAD_POOL_SIZE;
735 rc = obd_preprw(OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
736 remote_nb, &npages, local_nb, oti, capa);
740 desc = ptlrpc_prep_bulk_exp(req, npages,
741 BULK_PUT_SOURCE, OST_BULK_PORTAL);
743 GOTO(out_commitrw, rc = -ENOMEM);
746 for (i = 0; i < npages; i++) {
747 int page_rc = local_nb[i].rc;
749 if (page_rc < 0) { /* error */
755 if (page_rc != 0) { /* some data! */
756 LASSERT (local_nb[i].page != NULL);
757 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
758 local_nb[i].offset & ~CFS_PAGE_MASK,
762 if (page_rc != local_nb[i].len) { /* short read */
763 /* All subsequent pages should be 0 */
765 LASSERT(local_nb[i].rc == 0);
770 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
771 cksum_type_t cksum_type =
772 cksum_type_unpack(repbody->oa.o_valid & OBD_MD_FLFLAGS ?
773 repbody->oa.o_flags : 0);
774 repbody->oa.o_flags = cksum_type_pack(cksum_type);
775 repbody->oa.o_valid = OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
776 repbody->oa.o_cksum = ost_checksum_bulk(desc, OST_READ,cksum_type);
777 CDEBUG(D_PAGE, "checksum at read origin: %x\n",
778 repbody->oa.o_cksum);
780 repbody->oa.o_valid = 0;
782 /* We're finishing using body->oa as an input variable */
784 /* Check if client was evicted while we were doing i/o before touching
787 rc = target_bulk_io(exp, desc, &lwi);
792 /* Must commit after prep above in all cases */
793 rc = obd_commitrw(OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
794 remote_nb, npages, local_nb, oti, rc);
797 ost_drop_id(exp, &repbody->oa);
800 ost_brw_lock_put(LCK_PR, ioo, remote_nb, &lockh);
805 ptlrpc_free_bulk(desc);
809 req->rq_status = nob;
810 ptlrpc_lprocfs_brw(req, nob);
811 target_committed_to_req(req);
813 } else if (!no_reply) {
814 /* Only reply if there was no comms problem with bulk */
815 target_committed_to_req(req);
819 /* reply out callback would free */
820 ptlrpc_req_drop_rs(req);
821 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
822 "client will retry\n",
823 exp->exp_obd->obd_name,
824 exp->exp_client_uuid.uuid,
825 exp->exp_connection->c_remote_uuid.uuid,
826 libcfs_id2str(req->rq_peer));
832 static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
834 struct ptlrpc_bulk_desc *desc = NULL;
835 struct obd_export *exp = req->rq_export;
836 struct niobuf_remote *remote_nb;
837 struct niobuf_local *local_nb;
838 struct obd_ioobj *ioo;
839 struct ost_body *body, *repbody;
840 struct l_wait_info lwi;
841 struct lustre_handle lockh = {0};
842 struct lustre_capa *capa = NULL;
844 int objcount, niocount, npages;
846 obd_count client_cksum = 0, server_cksum = 0;
847 cksum_type_t cksum_type = OBD_CKSUM_CRC32;
848 int no_reply = 0, mmap = 0;
849 __u32 o_uid = 0, o_gid = 0;
850 struct ost_thread_local_cache *tls;
853 req->rq_bulk_write = 1;
855 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK))
856 GOTO(out, rc = -EIO);
857 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_WRITE_BULK2))
858 GOTO(out, rc = -EFAULT);
860 /* pause before transaction has been started */
861 OBD_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_BULK, (obd_timeout + 1) / 4);
863 /* ost_body, ioobj & noibuf_remote are verified and swabbed in
864 * ost_rw_hpreq_check(). */
865 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
867 GOTO(out, rc = -EFAULT);
869 objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
870 RCL_CLIENT) / sizeof(*ioo);
871 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
873 GOTO(out, rc = -EFAULT);
875 rc = ost_validate_obdo(exp, &body->oa, ioo);
879 for (niocount = i = 0; i < objcount; i++)
880 niocount += ioo[i].ioo_bufcnt;
883 * It'd be nice to have a capsule function to indicate how many elements
884 * there were in a buffer for an RMF that's declared to be an array.
885 * It's easy enough to compute the number of elements here though.
887 remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
888 if (remote_nb == NULL || niocount != (req_capsule_get_size(&req->rq_pill,
889 &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
890 GOTO(out, rc = -EFAULT);
892 if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
893 (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
894 cfs_memory_pressure_set();
896 if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
897 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
899 CERROR("Missing capability for OST BRW WRITE");
900 GOTO(out, rc = -EFAULT);
904 req_capsule_set_size(&req->rq_pill, &RMF_RCS, RCL_SERVER,
905 niocount * sizeof(*rcs));
906 rc = req_capsule_server_pack(&req->rq_pill);
909 CFS_FAIL_TIMEOUT(OBD_FAIL_OST_BRW_PAUSE_PACK, cfs_fail_val);
910 rcs = req_capsule_server_get(&req->rq_pill, &RMF_RCS);
912 tls = ost_tls_get(req);
914 GOTO(out_bulk, rc = -ENOMEM);
915 local_nb = tls->local;
917 rc = ost_brw_lock_get(LCK_PW, exp, ioo, remote_nb, &lockh);
922 * If getting the lock took more time than
923 * client was willing to wait, drop it. b=11330
925 if (cfs_time_current_sec() > req->rq_deadline ||
926 OBD_FAIL_CHECK(OBD_FAIL_OST_DROP_REQ)) {
928 CERROR("Dropping timed-out write from %s because locking "
929 "object "LPX64" took %ld seconds (limit was %ld).\n",
930 libcfs_id2str(req->rq_peer), ioo->ioo_id,
931 cfs_time_current_sec() - req->rq_arrival_time.tv_sec,
932 req->rq_deadline - req->rq_arrival_time.tv_sec);
933 GOTO(out_lock, rc = -ETIMEDOUT);
936 /* obd_preprw clobbers oa->valid, so save what we need */
937 if (body->oa.o_valid & OBD_MD_FLCKSUM) {
938 client_cksum = body->oa.o_cksum;
939 if (body->oa.o_valid & OBD_MD_FLFLAGS)
940 cksum_type = cksum_type_unpack(body->oa.o_flags);
942 if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
945 /* Because we already sync grant info with client when reconnect,
946 * grant info will be cleared for resent req, then fed_grant and
947 * total_grant will not be modified in following preprw_write */
948 if (lustre_msg_get_flags(req->rq_reqmsg) & (MSG_RESENT | MSG_REPLAY)) {
949 DEBUG_REQ(D_CACHE, req, "clear resent/replay req grant info");
950 body->oa.o_valid &= ~OBD_MD_FLGRANT;
953 if (exp_connect_rmtclient(exp)) {
954 o_uid = body->oa.o_uid;
955 o_gid = body->oa.o_gid;
958 repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
959 memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
961 npages = OST_THREAD_POOL_SIZE;
962 rc = obd_preprw(OBD_BRW_WRITE, exp, &repbody->oa, objcount,
963 ioo, remote_nb, &npages, local_nb, oti, capa);
967 desc = ptlrpc_prep_bulk_exp(req, npages,
968 BULK_GET_SINK, OST_BULK_PORTAL);
970 GOTO(skip_transfer, rc = -ENOMEM);
972 /* NB Having prepped, we must commit... */
974 for (i = 0; i < npages; i++)
975 ptlrpc_prep_bulk_page(desc, local_nb[i].page,
976 local_nb[i].offset & ~CFS_PAGE_MASK,
979 rc = sptlrpc_svc_prep_bulk(req, desc);
983 rc = target_bulk_io(exp, desc, &lwi);
987 if (unlikely(client_cksum != 0 && rc == 0)) {
988 static int cksum_counter;
989 repbody->oa.o_valid |= OBD_MD_FLCKSUM | OBD_MD_FLFLAGS;
990 repbody->oa.o_flags &= ~OBD_FL_CKSUM_ALL;
991 repbody->oa.o_flags |= cksum_type_pack(cksum_type);
992 server_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
993 repbody->oa.o_cksum = server_cksum;
995 if (unlikely(client_cksum != server_cksum)) {
996 CDEBUG_LIMIT(mmap ? D_INFO : D_ERROR,
997 "client csum %x, server csum %x\n",
998 client_cksum, server_cksum);
1000 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
1001 CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
1002 cksum_counter, libcfs_id2str(req->rq_peer),
1007 /* Must commit after prep above in all cases */
1008 rc = obd_commitrw(OBD_BRW_WRITE, exp, &repbody->oa, objcount, ioo,
1009 remote_nb, npages, local_nb, oti, rc);
1010 if (rc == -ENOTCONN)
1011 /* quota acquire process has been given up because
1012 * either the client has been evicted or the client
1013 * has timed out the request already */
1016 if (exp_connect_rmtclient(exp)) {
1017 repbody->oa.o_uid = o_uid;
1018 repbody->oa.o_gid = o_gid;
1022 * Disable sending mtime back to the client. If the client locked the
1023 * whole object, then it has already updated the mtime on its side,
1024 * otherwise it will have to glimpse anyway (see bug 21489, comment 32)
1026 repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
1028 if (unlikely(client_cksum != server_cksum && rc == 0 && !mmap)) {
1029 int new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
1034 if (new_cksum == server_cksum)
1035 msg = "changed in transit before arrival at OST";
1036 else if (new_cksum == client_cksum)
1037 msg = "initial checksum before message complete";
1039 msg = "changed in transit AND after initial checksum";
1041 if (req->rq_peer.nid == desc->bd_sender) {
1045 router = libcfs_nid2str(desc->bd_sender);
1048 LCONSOLE_ERROR_MSG(0x168, "%s: BAD WRITE CHECKSUM: %s from "
1049 "%s%s%s inode "DFID" object "
1050 LPU64"/"LPU64" extent ["LPU64"-"LPU64"]\n",
1051 exp->exp_obd->obd_name, msg,
1052 libcfs_id2str(req->rq_peer),
1054 body->oa.o_valid & OBD_MD_FLFID ?
1055 body->oa.o_parent_seq : (__u64)0,
1056 body->oa.o_valid & OBD_MD_FLFID ?
1057 body->oa.o_parent_oid : 0,
1058 body->oa.o_valid & OBD_MD_FLFID ?
1059 body->oa.o_parent_ver : 0,
1061 body->oa.o_valid & OBD_MD_FLGROUP ?
1062 body->oa.o_seq : (__u64)0,
1064 local_nb[npages-1].offset +
1065 local_nb[npages-1].len - 1 );
1066 CERROR("client csum %x, original server csum %x, "
1067 "server csum now %x\n",
1068 client_cksum, server_cksum, new_cksum);
1074 /* set per-requested niobuf return codes */
1075 for (i = j = 0; i < niocount; i++) {
1076 int len = remote_nb[i].len;
1081 LASSERT(j < npages);
1082 if (local_nb[j].rc < 0)
1083 rcs[i] = local_nb[j].rc;
1084 len -= local_nb[j].len;
1089 LASSERT(j == npages);
1090 ptlrpc_lprocfs_brw(req, nob);
1094 ost_brw_lock_put(LCK_PW, ioo, remote_nb, &lockh);
1099 ptlrpc_free_bulk(desc);
1102 oti_to_request(oti, req);
1103 target_committed_to_req(req);
1104 rc = ptlrpc_reply(req);
1105 } else if (!no_reply) {
1106 /* Only reply if there was no comms problem with bulk */
1107 target_committed_to_req(req);
1108 req->rq_status = rc;
1111 /* reply out callback would free */
1112 ptlrpc_req_drop_rs(req);
1113 CWARN("%s: ignoring bulk IO comm error with %s@%s id %s - "
1114 "client will retry\n",
1115 exp->exp_obd->obd_name,
1116 exp->exp_client_uuid.uuid,
1117 exp->exp_connection->c_remote_uuid.uuid,
1118 libcfs_id2str(req->rq_peer));
1120 cfs_memory_pressure_clr();
1125 * Implementation of OST_SET_INFO.
1127 * OST_SET_INFO is like ioctl(): heavily overloaded. Specifically, it takes a
1128 * "key" and a value RPC buffers as arguments, with the value's contents
1129 * interpreted according to the key.
1131 * Value types that need swabbing have swabbing done explicitly, either here or
1132 * in functions called from here. This should be corrected: all swabbing should
1133 * be done in the capsule abstraction, as that will then allow us to move
1134 * swabbing exclusively to the client without having to modify server code
1135 * outside the capsule abstraction's implementation itself. To correct this
1136 * will require minor changes to the capsule abstraction; see the comments for
1137 * req_capsule_extend() in layout.c.
1139 static int ost_set_info(struct obd_export *exp, struct ptlrpc_request *req)
1141 struct ost_body *body = NULL, *repbody;
1142 char *key, *val = NULL;
1143 int keylen, vallen, rc = 0;
1144 int is_grant_shrink = 0;
1147 key = req_capsule_client_get(&req->rq_pill, &RMF_SETINFO_KEY);
1149 DEBUG_REQ(D_HA, req, "no set_info key");
1152 keylen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_KEY,
1155 vallen = req_capsule_get_size(&req->rq_pill, &RMF_SETINFO_VAL,
1158 if ((is_grant_shrink = KEY_IS(KEY_GRANT_SHRINK)))
1159 /* In this case the value is actually an RMF_OST_BODY, so we
1160 * transmutate the type of this PTLRPC */
1161 req_capsule_extend(&req->rq_pill, &RQF_OST_SET_GRANT_INFO);
1163 rc = req_capsule_server_pack(&req->rq_pill);
1168 if (is_grant_shrink) {
1169 body = req_capsule_client_get(&req->rq_pill,
1174 repbody = req_capsule_server_get(&req->rq_pill,
1176 memcpy(repbody, body, sizeof(*body));
1177 val = (char*)repbody;
1179 val = req_capsule_client_get(&req->rq_pill,
1184 if (KEY_IS(KEY_EVICT_BY_NID)) {
1186 obd_export_evict_by_nid(exp->exp_obd, val);
1188 } else if (KEY_IS(KEY_MDS_CONN) && ptlrpc_req_need_swab(req)) {
1189 if (vallen < sizeof(__u32))
1191 __swab32s((__u32 *)val);
1194 /* OBD will also check if KEY_IS(KEY_GRANT_SHRINK), and will cast val to
1195 * a struct ost_body * value */
1196 rc = obd_set_info_async(exp, keylen, key, vallen, val, NULL);
1198 lustre_msg_set_status(req->rq_repmsg, 0);
1202 static int ost_get_info(struct obd_export *exp, struct ptlrpc_request *req)
1205 int keylen, replylen, rc = 0;
1206 struct req_capsule *pill = &req->rq_pill;
1209 /* this common part for get_info rpc */
1210 key = req_capsule_client_get(pill, &RMF_SETINFO_KEY);
1212 DEBUG_REQ(D_HA, req, "no get_info key");
1215 keylen = req_capsule_get_size(pill, &RMF_SETINFO_KEY, RCL_CLIENT);
1217 if (KEY_IS(KEY_FIEMAP)) {
1218 struct ll_fiemap_info_key *fm_key = key;
1221 rc = ost_validate_obdo(exp, &fm_key->oa, NULL);
1226 rc = obd_get_info(exp, keylen, key, &replylen, NULL, NULL);
1230 req_capsule_set_size(pill, &RMF_GENERIC_DATA,
1231 RCL_SERVER, replylen);
1233 rc = req_capsule_server_pack(pill);
1237 reply = req_capsule_server_get(pill, &RMF_GENERIC_DATA);
1241 /* call again to fill in the reply buffer */
1242 rc = obd_get_info(exp, keylen, key, &replylen, reply, NULL);
1244 lustre_msg_set_status(req->rq_repmsg, 0);
1248 #ifdef HAVE_QUOTA_SUPPORT
1249 static int ost_handle_quotactl(struct ptlrpc_request *req)
1251 struct obd_quotactl *oqctl, *repoqc;
1255 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1257 GOTO(out, rc = -EPROTO);
1259 rc = req_capsule_server_pack(&req->rq_pill);
1263 repoqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1264 req->rq_status = obd_quotactl(req->rq_export, oqctl);
1271 static int ost_handle_quotacheck(struct ptlrpc_request *req)
1273 struct obd_quotactl *oqctl;
1277 oqctl = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
1281 rc = req_capsule_server_pack(&req->rq_pill);
1285 req->rq_status = obd_quotacheck(req->rq_export, oqctl);
1289 static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
1291 struct quota_adjust_qunit *oqaq, *repoqa;
1292 struct lustre_quota_ctxt *qctxt;
1296 qctxt = &req->rq_export->exp_obd->u.obt.obt_qctxt;
1297 oqaq = req_capsule_client_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1299 GOTO(out, rc = -EPROTO);
1301 rc = req_capsule_server_pack(&req->rq_pill);
1305 repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
1306 req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt, NULL);
1314 static int ost_llog_handle_connect(struct obd_export *exp,
1315 struct ptlrpc_request *req)
1317 struct llogd_conn_body *body;
1321 body = req_capsule_client_get(&req->rq_pill, &RMF_LLOGD_CONN_BODY);
1322 rc = obd_llog_connect(exp, body);
1326 #define ost_init_sec_none(reply, exp) \
1328 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT | \
1329 OBD_CONNECT_RMT_CLIENT_FORCE | \
1330 OBD_CONNECT_OSS_CAPA); \
1331 cfs_spin_lock(&exp->exp_lock); \
1332 exp->exp_connect_flags = reply->ocd_connect_flags; \
1333 cfs_spin_unlock(&exp->exp_lock); \
1336 static int ost_init_sec_level(struct ptlrpc_request *req)
1338 struct obd_export *exp = req->rq_export;
1339 struct req_capsule *pill = &req->rq_pill;
1340 struct obd_device *obd = exp->exp_obd;
1341 struct filter_obd *filter = &obd->u.filter;
1342 char *client = libcfs_nid2str(req->rq_peer.nid);
1343 struct obd_connect_data *data, *reply;
1347 data = req_capsule_client_get(pill, &RMF_CONNECT_DATA);
1348 reply = req_capsule_server_get(pill, &RMF_CONNECT_DATA);
1349 if (data == NULL || reply == NULL)
1352 /* connection from MDT is always trusted */
1353 if (req->rq_auth_usr_mdt) {
1354 ost_init_sec_none(reply, exp);
1358 /* no GSS support case */
1359 if (!req->rq_auth_gss) {
1360 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1361 CWARN("client %s -> target %s does not user GSS, "
1362 "can not run under security level %d.\n",
1363 client, obd->obd_name, filter->fo_sec_level);
1366 ost_init_sec_none(reply, exp);
1371 /* old version case */
1372 if (unlikely(!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) ||
1373 !(data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA))) {
1374 if (filter->fo_sec_level > LUSTRE_SEC_NONE) {
1375 CWARN("client %s -> target %s uses old version, "
1376 "can not run under security level %d.\n",
1377 client, obd->obd_name, filter->fo_sec_level);
1380 CWARN("client %s -> target %s uses old version, "
1381 "run under security level %d.\n",
1382 client, obd->obd_name, filter->fo_sec_level);
1383 ost_init_sec_none(reply, exp);
1388 remote = data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT_FORCE;
1390 if (!req->rq_auth_remote)
1391 CDEBUG(D_SEC, "client (local realm) %s -> target %s "
1392 "asked to be remote.\n", client, obd->obd_name);
1393 } else if (req->rq_auth_remote) {
1395 CDEBUG(D_SEC, "client (remote realm) %s -> target %s is set "
1396 "as remote by default.\n", client, obd->obd_name);
1400 if (!filter->fo_fl_oss_capa) {
1401 CDEBUG(D_SEC, "client %s -> target %s is set as remote,"
1402 " but OSS capabilities are not enabled: %d.\n",
1403 client, obd->obd_name, filter->fo_fl_oss_capa);
1408 switch (filter->fo_sec_level) {
1409 case LUSTRE_SEC_NONE:
1411 ost_init_sec_none(reply, exp);
1414 CDEBUG(D_SEC, "client %s -> target %s is set as remote, "
1415 "can not run under security level %d.\n",
1416 client, obd->obd_name, filter->fo_sec_level);
1419 case LUSTRE_SEC_REMOTE:
1421 ost_init_sec_none(reply, exp);
1423 case LUSTRE_SEC_ALL:
1425 reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
1426 OBD_CONNECT_RMT_CLIENT_FORCE);
1427 if (!filter->fo_fl_oss_capa)
1428 reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
1430 cfs_spin_lock(&exp->exp_lock);
1431 exp->exp_connect_flags = reply->ocd_connect_flags;
1432 cfs_spin_unlock(&exp->exp_lock);
1444 * this should be done in filter_connect()/filter_reconnect(), but
1445 * we can't obtain information like NID, which stored in incoming
1446 * request, thus can't decide what flavor to use. so we do it here.
1448 * This hack should be removed after the OST stack be rewritten, just
1449 * like what we are doing in mdt_obd_connect()/mdt_obd_reconnect().
1451 static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
1453 struct obd_export *exp = req->rq_export;
1454 struct filter_obd *filter = &exp->exp_obd->u.filter;
1455 struct sptlrpc_flavor flvr;
1458 if (unlikely(strcmp(exp->exp_obd->obd_type->typ_name,
1459 LUSTRE_ECHO_NAME) == 0)) {
1460 exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
1464 if (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_INVALID) {
1465 cfs_read_lock(&filter->fo_sptlrpc_lock);
1466 sptlrpc_target_choose_flavor(&filter->fo_sptlrpc_rset,
1470 cfs_read_unlock(&filter->fo_sptlrpc_lock);
1472 cfs_spin_lock(&exp->exp_lock);
1474 exp->exp_sp_peer = req->rq_sp_from;
1475 exp->exp_flvr = flvr;
1477 if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
1478 exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
1479 CERROR("unauthorized rpc flavor %x from %s, "
1480 "expect %x\n", req->rq_flvr.sf_rpc,
1481 libcfs_nid2str(req->rq_peer.nid),
1482 exp->exp_flvr.sf_rpc);
1486 cfs_spin_unlock(&exp->exp_lock);
1488 if (exp->exp_sp_peer != req->rq_sp_from) {
1489 CERROR("RPC source %s doesn't match %s\n",
1490 sptlrpc_part2name(req->rq_sp_from),
1491 sptlrpc_part2name(exp->exp_sp_peer));
1494 rc = sptlrpc_target_export_check(exp, req);
1501 /* Ensure that data and metadata are synced to the disk when lock is cancelled
1503 int ost_blocking_ast(struct ldlm_lock *lock,
1504 struct ldlm_lock_desc *desc,
1505 void *data, int flag)
1507 __u32 sync_lock_cancel = 0;
1508 __u32 len = sizeof(sync_lock_cancel);
1512 rc = obd_get_info(lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
1513 KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
1515 if (!rc && flag == LDLM_CB_CANCELING &&
1516 (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
1517 (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
1518 (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
1519 lock->l_flags & LDLM_FL_CBPENDING))) {
1520 struct obd_info *oinfo;
1524 OBD_ALLOC_PTR(oinfo);
1529 OBD_FREE_PTR(oinfo);
1532 oa->o_id = lock->l_resource->lr_name.name[0];
1533 oa->o_seq = lock->l_resource->lr_name.name[1];
1534 oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
1537 rc = obd_sync(lock->l_export, oinfo,
1538 lock->l_policy_data.l_extent.start,
1539 lock->l_policy_data.l_extent.end, NULL);
1541 CERROR("Error %d syncing data on lock cancel\n", rc);
1544 OBD_FREE_PTR(oinfo);
1547 rc = ldlm_server_blocking_ast(lock, desc, data, flag);
1551 static int ost_filter_recovery_request(struct ptlrpc_request *req,
1552 struct obd_device *obd, int *process)
1554 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
1555 case OST_CONNECT: /* This will never get here, but for completeness. */
1556 case OST_DISCONNECT:
1567 case OBD_LOG_CANCEL:
1569 *process = target_queue_recovery_request(req, obd);
1573 DEBUG_REQ(D_ERROR, req, "not permitted during recovery");
1579 int ost_msg_check_version(struct lustre_msg *msg)
1583 switch(lustre_msg_get_opc(msg)) {
1585 case OST_DISCONNECT:
1588 case SEC_CTX_INIT_CONT:
1590 rc = lustre_msg_check_version(msg, LUSTRE_OBD_VERSION);
1592 CERROR("bad opc %u version %08x, expecting %08x\n",
1593 lustre_msg_get_opc(msg),
1594 lustre_msg_get_version(msg),
1595 LUSTRE_OBD_VERSION);
1608 #ifdef HAVE_QUOTA_SUPPORT
1609 case OST_QUOTACHECK:
1611 case OST_QUOTA_ADJUST_QUNIT:
1613 rc = lustre_msg_check_version(msg, LUSTRE_OST_VERSION);
1615 CERROR("bad opc %u version %08x, expecting %08x\n",
1616 lustre_msg_get_opc(msg),
1617 lustre_msg_get_version(msg),
1618 LUSTRE_OST_VERSION);
1623 case LDLM_BL_CALLBACK:
1624 case LDLM_CP_CALLBACK:
1625 rc = lustre_msg_check_version(msg, LUSTRE_DLM_VERSION);
1627 CERROR("bad opc %u version %08x, expecting %08x\n",
1628 lustre_msg_get_opc(msg),
1629 lustre_msg_get_version(msg),
1630 LUSTRE_DLM_VERSION);
1632 case LLOG_ORIGIN_CONNECT:
1633 case OBD_LOG_CANCEL:
1634 rc = lustre_msg_check_version(msg, LUSTRE_LOG_VERSION);
1636 CERROR("bad opc %u version %08x, expecting %08x\n",
1637 lustre_msg_get_opc(msg),
1638 lustre_msg_get_version(msg),
1639 LUSTRE_LOG_VERSION);
1642 CERROR("Unexpected opcode %d\n", lustre_msg_get_opc(msg));
1648 struct ost_prolong_data {
1649 struct ptlrpc_request *opd_req;
1650 struct obd_export *opd_exp;
1651 struct obdo *opd_oa;
1652 struct ldlm_res_id opd_resid;
1653 struct ldlm_extent opd_extent;
1654 ldlm_mode_t opd_mode;
1655 unsigned int opd_locks;
1659 /* prolong locks for the current service time of the corresponding
1660 * portal (= OST_IO_PORTAL)
1662 static inline int prolong_timeout(struct ptlrpc_request *req)
1664 struct ptlrpc_service *svc = req->rq_rqbd->rqbd_service;
1667 return obd_timeout / 2;
1669 return max(at_est2timeout(at_get(&svc->srv_at_estimate)), ldlm_timeout);
1672 static void ost_prolong_lock_one(struct ost_prolong_data *opd,
1673 struct ldlm_lock *lock)
1675 LASSERT(lock->l_req_mode == lock->l_granted_mode);
1676 LASSERT(lock->l_export == opd->opd_exp);
1678 /* XXX: never try to grab resource lock here because we're inside
1679 * exp_bl_list_lock; in ldlm_lockd.c to handle waiting list we take
1680 * res lock and then exp_bl_list_lock. */
1682 if (!(lock->l_flags & LDLM_FL_AST_SENT))
1683 /* ignore locks not being cancelled */
1687 "refreshed for req x"LPU64" ext("LPU64"->"LPU64") to %ds.\n",
1688 opd->opd_req->rq_xid, opd->opd_extent.start,
1689 opd->opd_extent.end, opd->opd_timeout);
1691 /* OK. this is a possible lock the user holds doing I/O
1692 * let's refresh eviction timer for it */
1693 ldlm_refresh_waiting_lock(lock, opd->opd_timeout);
1697 static void ost_prolong_locks(struct ost_prolong_data *data)
1699 struct obd_export *exp = data->opd_exp;
1700 struct obdo *oa = data->opd_oa;
1701 struct ldlm_lock *lock;
1704 if (oa->o_valid & OBD_MD_FLHANDLE) {
1705 /* mostly a request should be covered by only one lock, try
1707 lock = ldlm_handle2lock(&oa->o_handle);
1709 /* Fast path to check if the lock covers the whole IO
1710 * region exclusively. */
1711 if (lock->l_granted_mode == LCK_PW &&
1712 ldlm_extent_contain(&lock->l_policy_data.l_extent,
1713 &data->opd_extent)) {
1715 ost_prolong_lock_one(data, lock);
1716 LDLM_LOCK_PUT(lock);
1719 LDLM_LOCK_PUT(lock);
1724 cfs_spin_lock_bh(&exp->exp_bl_list_lock);
1725 cfs_list_for_each_entry(lock, &exp->exp_bl_list, l_exp_list) {
1726 LASSERT(lock->l_flags & LDLM_FL_AST_SENT);
1727 LASSERT(lock->l_resource->lr_type == LDLM_EXTENT);
1729 if (!ldlm_res_eq(&data->opd_resid, &lock->l_resource->lr_name))
1732 if (!ldlm_extent_overlap(&lock->l_policy_data.l_extent,
1736 ost_prolong_lock_one(data, lock);
1738 cfs_spin_unlock_bh(&exp->exp_bl_list_lock);
1744 * Returns 1 if the given PTLRPC matches the given LDLM locks, or 0 if it does
1747 static int ost_rw_hpreq_lock_match(struct ptlrpc_request *req,
1748 struct ldlm_lock *lock)
1750 struct niobuf_remote *nb;
1751 struct obd_ioobj *ioo;
1753 struct ldlm_extent ext;
1756 opc = lustre_msg_get_opc(req->rq_reqmsg);
1757 LASSERT(opc == OST_READ || opc == OST_WRITE);
1759 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1760 LASSERT(ioo != NULL);
1762 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1763 LASSERT(nb != NULL);
1765 ext.start = nb->offset;
1766 nb += ioo->ioo_bufcnt - 1;
1767 ext.end = nb->offset + nb->len - 1;
1769 LASSERT(lock->l_resource != NULL);
1770 if (!osc_res_name_eq(ioo->ioo_id, ioo->ioo_seq,
1771 &lock->l_resource->lr_name))
1775 if (opc == OST_READ)
1777 if (!(lock->l_granted_mode & mode))
1780 RETURN(ldlm_extent_overlap(&lock->l_policy_data.l_extent, &ext));
1784 * High-priority queue request check for whether the given PTLRPC request (\a
1785 * req) is blocking an LDLM lock cancel.
1787 * Returns 1 if the given given PTLRPC request (\a req) is blocking an LDLM lock
1788 * cancel, 0 if it is not, and -EFAULT if the request is malformed.
1790 * Only OST_READs, OST_WRITEs and OST_PUNCHes go on the h-p RPC queue. This
1791 * function looks only at OST_READs and OST_WRITEs.
1793 static int ost_rw_hpreq_check(struct ptlrpc_request *req)
1795 struct obd_device *obd = req->rq_export->exp_obd;
1796 struct ost_body *body;
1797 struct obd_ioobj *ioo;
1798 struct niobuf_remote *nb;
1799 struct ost_prolong_data opd = { 0 };
1804 * Use LASSERT to do sanity check because malformed RPCs should have
1805 * been filtered out in ost_hpreq_handler().
1807 opc = lustre_msg_get_opc(req->rq_reqmsg);
1808 LASSERT(opc == OST_READ || opc == OST_WRITE);
1810 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1811 LASSERT(body != NULL);
1813 ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
1814 LASSERT(ioo != NULL);
1816 nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
1817 LASSERT(nb != NULL);
1818 LASSERT(!(nb->flags & OBD_BRW_SRVLOCK));
1820 osc_build_res_name(ioo->ioo_id, ioo->ioo_seq, &opd.opd_resid);
1824 if (opc == OST_READ)
1826 opd.opd_mode = mode;
1827 opd.opd_exp = req->rq_export;
1828 opd.opd_oa = &body->oa;
1829 opd.opd_extent.start = nb->offset;
1830 nb += ioo->ioo_bufcnt - 1;
1831 opd.opd_extent.end = nb->offset + nb->len - 1;
1832 opd.opd_timeout = prolong_timeout(req);
1834 DEBUG_REQ(D_RPCTRACE, req,
1835 "%s %s: refresh rw locks: " LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1836 obd->obd_name, cfs_current()->comm,
1837 opd.opd_resid.name[0], opd.opd_resid.name[1],
1838 opd.opd_extent.start, opd.opd_extent.end);
1840 ost_prolong_locks(&opd);
1842 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1843 obd->obd_name, opd.opd_locks, req);
1845 RETURN(opd.opd_locks);
1848 static void ost_rw_hpreq_fini(struct ptlrpc_request *req)
1850 (void)ost_rw_hpreq_check(req);
1854 * Like ost_rw_hpreq_lock_match(), but for OST_PUNCH RPCs.
1856 static int ost_punch_hpreq_lock_match(struct ptlrpc_request *req,
1857 struct ldlm_lock *lock)
1859 struct ost_body *body;
1862 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1863 LASSERT(body != NULL);
1865 if (body->oa.o_valid & OBD_MD_FLHANDLE &&
1866 body->oa.o_handle.cookie == lock->l_handle.h_cookie)
1873 * Like ost_rw_hpreq_check(), but for OST_PUNCH RPCs.
1875 static int ost_punch_hpreq_check(struct ptlrpc_request *req)
1877 struct obd_device *obd = req->rq_export->exp_obd;
1878 struct ost_body *body;
1880 struct ost_prolong_data opd = { 0 };
1884 body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
1885 LASSERT(body != NULL);
1888 LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS) ||
1889 !(oa->o_flags & OBD_FL_SRVLOCK));
1892 end = start + oa->o_blocks;
1895 opd.opd_mode = LCK_PW;
1896 opd.opd_exp = req->rq_export;
1898 opd.opd_extent.start = start;
1899 opd.opd_extent.end = end;
1900 if (oa->o_blocks == OBD_OBJECT_EOF)
1901 opd.opd_extent.end = OBD_OBJECT_EOF;
1902 opd.opd_timeout = prolong_timeout(req);
1904 osc_build_res_name(oa->o_id, oa->o_seq, &opd.opd_resid);
1907 "%s: refresh locks: "LPU64"/"LPU64" ("LPU64"->"LPU64")\n",
1909 opd.opd_resid.name[0], opd.opd_resid.name[1],
1910 opd.opd_extent.start, opd.opd_extent.end);
1912 ost_prolong_locks(&opd);
1914 CDEBUG(D_DLMTRACE, "%s: refreshed %u locks timeout for req %p.\n",
1915 obd->obd_name, opd.opd_locks, req);
1917 RETURN(opd.opd_locks > 0);
1920 static void ost_punch_hpreq_fini(struct ptlrpc_request *req)
1922 (void)ost_punch_hpreq_check(req);
1925 struct ptlrpc_hpreq_ops ost_hpreq_rw = {
1926 .hpreq_lock_match = ost_rw_hpreq_lock_match,
1927 .hpreq_check = ost_rw_hpreq_check,
1928 .hpreq_fini = ost_rw_hpreq_fini
1931 struct ptlrpc_hpreq_ops ost_hpreq_punch = {
1932 .hpreq_lock_match = ost_punch_hpreq_lock_match,
1933 .hpreq_check = ost_punch_hpreq_check,
1934 .hpreq_fini = ost_punch_hpreq_fini
1937 /** Assign high priority operations to the request if needed. */
1938 static int ost_hpreq_handler(struct ptlrpc_request *req)
1941 if (req->rq_export) {
1942 int opc = lustre_msg_get_opc(req->rq_reqmsg);
1943 struct ost_body *body;
1945 if (opc == OST_READ || opc == OST_WRITE) {
1946 struct niobuf_remote *nb;
1947 struct obd_ioobj *ioo;
1948 int objcount, niocount;
1952 /* RPCs on the H-P queue can be inspected before
1953 * ost_handler() initializes their pills, so we
1954 * initialize that here. Capsule initialization is
1955 * idempotent, as is setting the pill's format (provided
1956 * it doesn't change).
1958 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
1959 if (opc == OST_READ)
1960 req_capsule_set(&req->rq_pill,
1963 req_capsule_set(&req->rq_pill,
1964 &RQF_OST_BRW_WRITE);
1966 body = req_capsule_client_get(&req->rq_pill,
1969 CERROR("Missing/short ost_body\n");
1973 objcount = req_capsule_get_size(&req->rq_pill,
1977 if (objcount == 0) {
1978 CERROR("Missing/short ioobj\n");
1982 CERROR("too many ioobjs (%d)\n", objcount);
1986 ioo = req_capsule_client_get(&req->rq_pill,
1989 CERROR("Missing/short ioobj\n");
1993 rc = ost_validate_obdo(req->rq_export, &body->oa, ioo);
1995 CERROR("invalid object ids\n");
1999 for (niocount = i = 0; i < objcount; i++) {
2000 if (ioo[i].ioo_bufcnt == 0) {
2001 CERROR("ioo[%d] has zero bufcnt\n", i);
2004 niocount += ioo[i].ioo_bufcnt;
2006 if (niocount > PTLRPC_MAX_BRW_PAGES) {
2007 DEBUG_REQ(D_RPCTRACE, req,
2008 "bulk has too many pages (%d)",
2013 nb = req_capsule_client_get(&req->rq_pill,
2014 &RMF_NIOBUF_REMOTE);
2016 CERROR("Missing/short niobuf\n");
2020 if (niocount == 0 || !(nb[0].flags & OBD_BRW_SRVLOCK))
2021 req->rq_ops = &ost_hpreq_rw;
2022 } else if (opc == OST_PUNCH) {
2023 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2024 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2026 body = req_capsule_client_get(&req->rq_pill,
2029 CERROR("Missing/short ost_body\n");
2033 if (!(body->oa.o_valid & OBD_MD_FLFLAGS) ||
2034 !(body->oa.o_flags & OBD_FL_SRVLOCK))
2035 req->rq_ops = &ost_hpreq_punch;
2041 /* TODO: handle requests in a similar way as MDT: see mdt_handle_common() */
2042 int ost_handle(struct ptlrpc_request *req)
2044 struct obd_trans_info trans_info = { 0, };
2045 struct obd_trans_info *oti = &trans_info;
2046 int should_process, fail = OBD_FAIL_OST_ALL_REPLY_NET, rc = 0;
2047 struct obd_device *obd = NULL;
2050 LASSERT(current->journal_info == NULL);
2052 /* primordial rpcs don't affect server recovery */
2053 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2055 case SEC_CTX_INIT_CONT:
2060 req_capsule_init(&req->rq_pill, req, RCL_SERVER);
2062 if (lustre_msg_get_opc(req->rq_reqmsg) != OST_CONNECT) {
2063 if (!class_connected_export(req->rq_export)) {
2064 CDEBUG(D_HA,"operation %d on unconnected OST from %s\n",
2065 lustre_msg_get_opc(req->rq_reqmsg),
2066 libcfs_id2str(req->rq_peer));
2067 req->rq_status = -ENOTCONN;
2068 GOTO(out, rc = -ENOTCONN);
2071 obd = req->rq_export->exp_obd;
2073 /* Check for aborted recovery. */
2074 if (obd->obd_recovering) {
2075 rc = ost_filter_recovery_request(req, obd,
2077 if (rc || !should_process)
2079 else if (should_process < 0) {
2080 req->rq_status = should_process;
2081 rc = ptlrpc_error(req);
2089 rc = ost_msg_check_version(req->rq_reqmsg);
2093 switch (lustre_msg_get_opc(req->rq_reqmsg)) {
2095 CDEBUG(D_INODE, "connect\n");
2096 req_capsule_set(&req->rq_pill, &RQF_OST_CONNECT);
2097 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET))
2099 rc = target_handle_connect(req);
2100 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CONNECT_NET2))
2103 rc = ost_init_sec_level(req);
2105 rc = ost_connect_check_sptlrpc(req);
2109 case OST_DISCONNECT:
2110 CDEBUG(D_INODE, "disconnect\n");
2111 req_capsule_set(&req->rq_pill, &RQF_OST_DISCONNECT);
2112 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DISCONNECT_NET))
2114 rc = target_handle_disconnect(req);
2117 CDEBUG(D_INODE, "create\n");
2118 req_capsule_set(&req->rq_pill, &RQF_OST_CREATE);
2119 if (OBD_FAIL_CHECK(OBD_FAIL_OST_CREATE_NET))
2121 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2122 GOTO(out, rc = -EROFS);
2123 rc = ost_create(req->rq_export, req, oti);
2126 CDEBUG(D_INODE, "destroy\n");
2127 req_capsule_set(&req->rq_pill, &RQF_OST_DESTROY);
2128 if (OBD_FAIL_CHECK(OBD_FAIL_OST_DESTROY_NET))
2130 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2131 GOTO(out, rc = -EROFS);
2132 rc = ost_destroy(req->rq_export, req, oti);
2135 CDEBUG(D_INODE, "getattr\n");
2136 req_capsule_set(&req->rq_pill, &RQF_OST_GETATTR);
2137 if (OBD_FAIL_CHECK(OBD_FAIL_OST_GETATTR_NET))
2139 rc = ost_getattr(req->rq_export, req);
2142 CDEBUG(D_INODE, "setattr\n");
2143 req_capsule_set(&req->rq_pill, &RQF_OST_SETATTR);
2144 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SETATTR_NET))
2146 rc = ost_setattr(req->rq_export, req, oti);
2149 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_WRITE);
2150 CDEBUG(D_INODE, "write\n");
2151 /* req->rq_request_portal would be nice, if it was set */
2152 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2153 CERROR("%s: deny write request from %s to portal %u\n",
2154 req->rq_export->exp_obd->obd_name,
2155 obd_export_nid2str(req->rq_export),
2156 req->rq_rqbd->rqbd_service->srv_req_portal);
2157 GOTO(out, rc = -EPROTO);
2159 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2161 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOSPC))
2162 GOTO(out, rc = -ENOSPC);
2163 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2164 GOTO(out, rc = -EROFS);
2165 rc = ost_brw_write(req, oti);
2166 LASSERT(current->journal_info == NULL);
2167 /* ost_brw_write sends its own replies */
2170 req_capsule_set(&req->rq_pill, &RQF_OST_BRW_READ);
2171 CDEBUG(D_INODE, "read\n");
2172 /* req->rq_request_portal would be nice, if it was set */
2173 if (req->rq_rqbd->rqbd_service->srv_req_portal !=OST_IO_PORTAL){
2174 CERROR("%s: deny read request from %s to portal %u\n",
2175 req->rq_export->exp_obd->obd_name,
2176 obd_export_nid2str(req->rq_export),
2177 req->rq_rqbd->rqbd_service->srv_req_portal);
2178 GOTO(out, rc = -EPROTO);
2180 if (OBD_FAIL_CHECK(OBD_FAIL_OST_BRW_NET))
2182 rc = ost_brw_read(req, oti);
2183 LASSERT(current->journal_info == NULL);
2184 /* ost_brw_read sends its own replies */
2187 CDEBUG(D_INODE, "punch\n");
2188 req_capsule_set(&req->rq_pill, &RQF_OST_PUNCH);
2189 if (OBD_FAIL_CHECK(OBD_FAIL_OST_PUNCH_NET))
2191 if (OBD_FAIL_CHECK(OBD_FAIL_OST_EROFS))
2192 GOTO(out, rc = -EROFS);
2193 rc = ost_punch(req->rq_export, req, oti);
2196 CDEBUG(D_INODE, "statfs\n");
2197 req_capsule_set(&req->rq_pill, &RQF_OST_STATFS);
2198 if (OBD_FAIL_CHECK(OBD_FAIL_OST_STATFS_NET))
2200 rc = ost_statfs(req);
2203 CDEBUG(D_INODE, "sync\n");
2204 req_capsule_set(&req->rq_pill, &RQF_OST_SYNC);
2205 if (OBD_FAIL_CHECK(OBD_FAIL_OST_SYNC_NET))
2207 rc = ost_sync(req->rq_export, req);
2210 DEBUG_REQ(D_INODE, req, "set_info");
2211 req_capsule_set(&req->rq_pill, &RQF_OBD_SET_INFO);
2212 rc = ost_set_info(req->rq_export, req);
2215 DEBUG_REQ(D_INODE, req, "get_info");
2216 req_capsule_set(&req->rq_pill, &RQF_OST_GET_INFO_GENERIC);
2217 rc = ost_get_info(req->rq_export, req);
2219 #ifdef HAVE_QUOTA_SUPPORT
2220 case OST_QUOTACHECK:
2221 CDEBUG(D_INODE, "quotacheck\n");
2222 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACHECK);
2223 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACHECK_NET))
2225 rc = ost_handle_quotacheck(req);
2228 CDEBUG(D_INODE, "quotactl\n");
2229 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTACTL);
2230 if (OBD_FAIL_CHECK(OBD_FAIL_OST_QUOTACTL_NET))
2232 rc = ost_handle_quotactl(req);
2234 case OST_QUOTA_ADJUST_QUNIT:
2235 CDEBUG(D_INODE, "quota_adjust_qunit\n");
2236 req_capsule_set(&req->rq_pill, &RQF_OST_QUOTA_ADJUST_QUNIT);
2237 rc = ost_handle_quota_adjust_qunit(req);
2241 DEBUG_REQ(D_INODE, req, "ping");
2242 req_capsule_set(&req->rq_pill, &RQF_OBD_PING);
2243 rc = target_handle_ping(req);
2245 /* FIXME - just reply status */
2246 case LLOG_ORIGIN_CONNECT:
2247 DEBUG_REQ(D_INODE, req, "log connect");
2248 req_capsule_set(&req->rq_pill, &RQF_LLOG_ORIGIN_CONNECT);
2249 rc = ost_llog_handle_connect(req->rq_export, req);
2250 req->rq_status = rc;
2251 rc = req_capsule_server_pack(&req->rq_pill);
2254 RETURN(ptlrpc_reply(req));
2255 case OBD_LOG_CANCEL:
2256 CDEBUG(D_INODE, "log cancel\n");
2257 req_capsule_set(&req->rq_pill, &RQF_LOG_CANCEL);
2258 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_NET))
2260 rc = llog_origin_handle_cancel(req);
2261 if (OBD_FAIL_CHECK(OBD_FAIL_OBD_LOG_CANCEL_REP))
2263 req->rq_status = rc;
2264 rc = req_capsule_server_pack(&req->rq_pill);
2267 RETURN(ptlrpc_reply(req));
2269 CDEBUG(D_INODE, "enqueue\n");
2270 req_capsule_set(&req->rq_pill, &RQF_LDLM_ENQUEUE);
2271 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
2273 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
2275 ldlm_server_glimpse_ast);
2276 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
2279 CDEBUG(D_INODE, "convert\n");
2280 req_capsule_set(&req->rq_pill, &RQF_LDLM_CONVERT);
2281 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CONVERT))
2283 rc = ldlm_handle_convert(req);
2286 CDEBUG(D_INODE, "cancel\n");
2287 req_capsule_set(&req->rq_pill, &RQF_LDLM_CANCEL);
2288 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CANCEL))
2290 rc = ldlm_handle_cancel(req);
2292 case LDLM_BL_CALLBACK:
2293 case LDLM_CP_CALLBACK:
2294 CDEBUG(D_INODE, "callback\n");
2295 CERROR("callbacks should not happen on OST\n");
2298 CERROR("Unexpected opcode %d\n",
2299 lustre_msg_get_opc(req->rq_reqmsg));
2300 req->rq_status = -ENOTSUPP;
2301 rc = ptlrpc_error(req);
2305 LASSERT(current->journal_info == NULL);
2308 /* If we're DISCONNECTing, the export_data is already freed */
2309 if (!rc && lustre_msg_get_opc(req->rq_reqmsg) != OST_DISCONNECT)
2310 target_committed_to_req(req);
2314 oti_to_request(oti, req);
2316 target_send_reply(req, rc, fail);
2319 EXPORT_SYMBOL(ost_handle);
2321 * free per-thread pool created by ost_thread_init().
2323 static void ost_thread_done(struct ptlrpc_thread *thread)
2325 struct ost_thread_local_cache *tls; /* TLS stands for Thread-Local
2330 LASSERT(thread != NULL);
2333 * be prepared to handle partially-initialized pools (because this is
2334 * called from ost_thread_init() for cleanup.
2336 tls = thread->t_data;
2339 thread->t_data = NULL;
2345 * initialize per-thread page pool (bug 5137).
2347 static int ost_thread_init(struct ptlrpc_thread *thread)
2349 struct ost_thread_local_cache *tls;
2353 LASSERT(thread != NULL);
2354 LASSERT(thread->t_data == NULL);
2355 LASSERTF(thread->t_id <= OSS_THREADS_MAX, "%u\n", thread->t_id);
2360 thread->t_data = tls;
2364 #define OST_WATCHDOG_TIMEOUT (obd_timeout * 1000)
2366 /* Sigh - really, this is an OSS, the _server_, not the _target_ */
2367 static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
2369 struct ost_obd *ost = &obd->u.ost;
2370 struct lprocfs_static_vars lvars;
2371 int oss_min_threads;
2372 int oss_max_threads;
2373 int oss_min_create_threads;
2374 int oss_max_create_threads;
2378 rc = cfs_cleanup_group_info();
2382 lprocfs_ost_init_vars(&lvars);
2383 lprocfs_obd_setup(obd, lvars.obd_vars);
2385 cfs_sema_init(&ost->ost_health_sem, 1);
2387 if (oss_num_threads) {
2388 /* If oss_num_threads is set, it is the min and the max. */
2389 if (oss_num_threads > OSS_THREADS_MAX)
2390 oss_num_threads = OSS_THREADS_MAX;
2391 if (oss_num_threads < OSS_THREADS_MIN)
2392 oss_num_threads = OSS_THREADS_MIN;
2393 oss_max_threads = oss_min_threads = oss_num_threads;
2395 /* Base min threads on memory and cpus */
2397 cfs_num_possible_cpus() * CFS_NUM_CACHEPAGES >>
2398 (27 - CFS_PAGE_SHIFT);
2399 if (oss_min_threads < OSS_THREADS_MIN)
2400 oss_min_threads = OSS_THREADS_MIN;
2401 /* Insure a 4x range for dynamic threads */
2402 if (oss_min_threads > OSS_THREADS_MAX / 4)
2403 oss_min_threads = OSS_THREADS_MAX / 4;
2404 oss_max_threads = min(OSS_THREADS_MAX, oss_min_threads * 4 + 1);
2408 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2409 OST_MAXREPSIZE, OST_REQUEST_PORTAL,
2410 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2411 ost_handle, LUSTRE_OSS_NAME,
2412 obd->obd_proc_entry, target_print_req,
2413 oss_min_threads, oss_max_threads,
2414 "ll_ost", LCT_DT_THREAD, NULL);
2415 if (ost->ost_service == NULL) {
2416 CERROR("failed to start service\n");
2417 GOTO(out_lprocfs, rc = -ENOMEM);
2420 rc = ptlrpc_start_threads(ost->ost_service);
2422 GOTO(out_service, rc = -EINVAL);
2424 if (oss_num_create_threads) {
2425 if (oss_num_create_threads > OSS_MAX_CREATE_THREADS)
2426 oss_num_create_threads = OSS_MAX_CREATE_THREADS;
2427 if (oss_num_create_threads < OSS_MIN_CREATE_THREADS)
2428 oss_num_create_threads = OSS_MIN_CREATE_THREADS;
2429 oss_min_create_threads = oss_max_create_threads =
2430 oss_num_create_threads;
2432 oss_min_create_threads = OSS_MIN_CREATE_THREADS;
2433 oss_max_create_threads = OSS_MAX_CREATE_THREADS;
2436 ost->ost_create_service =
2437 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2438 OST_MAXREPSIZE, OST_CREATE_PORTAL,
2439 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2440 ost_handle, "ost_create",
2441 obd->obd_proc_entry, target_print_req,
2442 oss_min_create_threads, oss_max_create_threads,
2443 "ll_ost_creat", LCT_DT_THREAD, NULL);
2444 if (ost->ost_create_service == NULL) {
2445 CERROR("failed to start OST create service\n");
2446 GOTO(out_service, rc = -ENOMEM);
2449 rc = ptlrpc_start_threads(ost->ost_create_service);
2451 GOTO(out_create, rc = -EINVAL);
2453 ost->ost_io_service =
2454 ptlrpc_init_svc(OST_NBUFS, OST_BUFSIZE, OST_MAXREQSIZE,
2455 OST_MAXREPSIZE, OST_IO_PORTAL,
2456 OSC_REPLY_PORTAL, OSS_SERVICE_WATCHDOG_FACTOR,
2457 ost_handle, "ost_io",
2458 obd->obd_proc_entry, target_print_req,
2459 oss_min_threads, oss_max_threads,
2460 "ll_ost_io", LCT_DT_THREAD, ost_hpreq_handler);
2461 if (ost->ost_io_service == NULL) {
2462 CERROR("failed to start OST I/O service\n");
2463 GOTO(out_create, rc = -ENOMEM);
2466 ost->ost_io_service->srv_init = ost_thread_init;
2467 ost->ost_io_service->srv_done = ost_thread_done;
2468 ost->ost_io_service->srv_cpu_affinity = 1;
2469 rc = ptlrpc_start_threads(ost->ost_io_service);
2471 GOTO(out_io, rc = -EINVAL);
2473 ping_evictor_start();
2478 ptlrpc_unregister_service(ost->ost_io_service);
2479 ost->ost_io_service = NULL;
2481 ptlrpc_unregister_service(ost->ost_create_service);
2482 ost->ost_create_service = NULL;
2484 ptlrpc_unregister_service(ost->ost_service);
2485 ost->ost_service = NULL;
2487 lprocfs_obd_cleanup(obd);
2491 static int ost_cleanup(struct obd_device *obd)
2493 struct ost_obd *ost = &obd->u.ost;
2497 ping_evictor_stop();
2499 /* there is no recovery for OST OBD, all recovery is controlled by
2501 LASSERT(obd->obd_recovering == 0);
2502 cfs_down(&ost->ost_health_sem);
2503 ptlrpc_unregister_service(ost->ost_service);
2504 ptlrpc_unregister_service(ost->ost_create_service);
2505 ptlrpc_unregister_service(ost->ost_io_service);
2506 ost->ost_service = NULL;
2507 ost->ost_create_service = NULL;
2508 cfs_up(&ost->ost_health_sem);
2510 lprocfs_obd_cleanup(obd);
2515 static int ost_health_check(struct obd_device *obd)
2517 struct ost_obd *ost = &obd->u.ost;
2520 cfs_down(&ost->ost_health_sem);
2521 rc |= ptlrpc_service_health_check(ost->ost_service);
2522 rc |= ptlrpc_service_health_check(ost->ost_create_service);
2523 rc |= ptlrpc_service_health_check(ost->ost_io_service);
2524 cfs_up(&ost->ost_health_sem);
2527 * health_check to return 0 on healthy
2528 * and 1 on unhealthy.
2536 struct ost_thread_local_cache *ost_tls(struct ptlrpc_request *r)
2538 return (struct ost_thread_local_cache *)(r->rq_svc_thread->t_data);
2541 /* use obd ops to offer management infrastructure */
2542 static struct obd_ops ost_obd_ops = {
2543 .o_owner = THIS_MODULE,
2544 .o_setup = ost_setup,
2545 .o_cleanup = ost_cleanup,
2546 .o_health_check = ost_health_check,
2550 static int __init ost_init(void)
2552 struct lprocfs_static_vars lvars;
2556 lprocfs_ost_init_vars(&lvars);
2557 rc = class_register_type(&ost_obd_ops, NULL, lvars.module_vars,
2558 LUSTRE_OSS_NAME, NULL);
2560 if (ost_num_threads != 0 && oss_num_threads == 0) {
2561 LCONSOLE_INFO("ost_num_threads module parameter is deprecated, "
2562 "use oss_num_threads instead or unset both for "
2563 "dynamic thread startup\n");
2564 oss_num_threads = ost_num_threads;
2570 static void /*__exit*/ ost_exit(void)
2572 class_unregister_type(LUSTRE_OSS_NAME);
2575 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2576 MODULE_DESCRIPTION("Lustre Object Storage Target (OST) v0.01");
2577 MODULE_LICENSE("GPL");
2579 module_init(ost_init);
2580 module_exit(ost_exit);